ESSAYS ON INDUSTRIAL ORGANIZATION AND HEALTH ECONOMICS
by
SICHAO JIANG
A DISSERTATION
Presented to the Department of Economics
and the Division of Graduate Studies of the University of Oregon
in partial fulfillment of the requirements
for the degree of
Doctor of Philosophy
March 2022
DISSERTATION APPROVAL PAGE
Student: Sichao Jiang
Title: Essays on Industrial Organization and Health Economics
This dissertation has been accepted and approved in partial fulfillment of the
requirements for the Doctor of Philosophy degree in the Department of Economics
by:
Keaton Miller Co-Chairperson
Wesley W. Wilson Co-Chairperson
Edward Rubin Core Member
N. Narasimha Murthy Institutional Representative
and
Krista Chronister Vice Provost of Graduate Studies
Original approval signatures are on file with the University of Oregon Division of
Graduate Studies.
Degree awarded March 2022
ii
© 2022 Sichao Jiang
All rights reserved.
iii
DISSERTATION ABSTRACT
Sichao Jiang
Doctor of Philosophy
Department of Economics
March 2022
Title: Essays on Industrial Organization and Health Economics
This thesis is composed of three essays and provides empirical contributions
to the Industrial Organization literature, especially in the field of transportation
and health economics. It aims to understand different issues related to economics
by applying various empirical methods.
The first essay (chapter 2) examines firm exit in Canadian markets,
specifically the grain elevator market. There is a long line of previous literature
that finds capacity, vintage, multi-plant ownership affect exit. In this paper, a
choice model is used to examine a firm’s decision to shut down a grain elevator
in terms of these variables, but also develops measures of spatial competition, local
economic conditions and linkages to the transportation markets. In all cases, these
variables are statistically important and point to results that reinforce previous
studies, but also direct to new explanations on the determinants of plant exit.
The second essay (chapter 3) examines the effects of marijuana legislation
change on the agricultural labor market. The paper uses differences-in-differences
with a synthetic control methodology to identify the effects of labor market
outcomes from marijuana legalization. This method aims to avoid substantial labor
market spillovers in neighboring states and to construct a decent parallel trend for
the pre-treatment time period with pretty varied agricultural markets in the U.S.
iv
The results show that cannabis legalization is associated with an increase in overall
employments that people are flushing into the industry, but no increase in per-
employee wages in both the retailer and agricultural labor market.
The third essay (chapter 4) looks into the accuracy of firms’ prediction
errors in the context of Medicare Advantage, where insurers receive subsidies from
the government and compete to provide health insurance to seniors. The results
show that on average firms overestimate future costs. Overestimation in forecast
error decreases with the experience of the firm. Firms in more competitive markets
(as measured by the number of other firms present) form more accurate estimates.
Firms with higher costs than expected generally offer plans that feature greater
patient cost sharing (i.e. higher deductibles and copays).
This dissertation includes both previously published/unpublished and co-
authored material.
v
CURRICULUM VITAE
NAME OF AUTHOR: Sichao Jiang
GRADUATE AND UNDERGRADUATE SCHOOLS ATTENDED:
University of Oregon, Eugene, OR, USA
Texas Tech University, Lubbock, TX, USA
DEGREES AWARDED:
Doctor of Philosophy, Economics, 2022, University of Oregon
Master of Science, Economics, 2018, University of Oregon
Bachelor of Science, Economics & Mathematics, 2017, Texas Tech University
AREAS OF SPECIAL INTEREST:
Industrial Organization
Applied Econometrics
Applied Microeconomics
PROFESSIONAL EXPERIENCE:
Graduate Employee, Department of Economics, University of Oregon, 2018-
2022
Summer Associate, Analysis Group, 2021
Research Consultant, Western Economic Association International,2020-2021
GRANTS, AWARDS AND HONORS:
Graduate Teaching Fellowship, University of Oregon, 2018-2022
Edward G. Daniel Scholarship, University of Oregon, 2018
Kleinsorge Fellowship Award and Distinctive Scholar Award, University of
Oregon, 2017
Graduate School First-Year Fellowship, University of Oregon, 2017
vi
PUBLICATIONS:
Jiang, S., Nolan, J. & Wilson, W.W. (2021). Exit Decisions in the Canadian
Grain Elevator. Journal of Industry, Competition and Trade, 1-19.
vii
TABLE OF CONTENTS
Chapter Page
I INTRODUCTION 1
II EXIT DECISIONS IN THE CANADIAN GRAIN
ELEVATOR INDUSTRY 4
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Econometric Specification and Results . . . . . . . . . . . . . . . . . 20
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
III WATCHING THE GRASS GROW: DOES
RECREATIONAL CANNABIS LEGALIZATION
AFFECT RETAIL AND AGRICULTURAL WAGES? 31
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2 Labor in the Cannabis Industry . . . . . . . . . . . . . . . . . . . . . 36
3.3 Data and Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4.1 Narrowly-defined industries . . . . . . . . . . . . . . . . . . . 48
3.4.2 Broadly-defined industries . . . . . . . . . . . . . . . . . . . . 55
3.4.3 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
IV ARE FIRMS’ COST PREDICTIONS ACCURATE?
EVIDENCE FROM MEDICARE ADVANTAGE 62
viii
Chapter Page
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2 Bidding in Medicare Advantage . . . . . . . . . . . . . . . . . . . . . 65
4.3 Data and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.4.1 The relationship between prediction errors and
plan-level observables . . . . . . . . . . . . . . . . . . . . . . . 71
4.4.2 The relationship between prediction errors and
product characteristics . . . . . . . . . . . . . . . . . . . . . . 72
4.5 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . 73
APPENDICES
A CHAPTER 2 APPENDIX 75
B CHAPTER 3 APPENDIX 76
B.1 Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
B.2 Additional tables and figures . . . . . . . . . . . . . . . . . . . . . . . 76
B.3 Tables of covariate balance . . . . . . . . . . . . . . . . . . . . . . . . 95
C CHAPTER 4 APPENDIX 115
C.1 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
REFERENCES CITED 124
ix
LIST OF FIGURES
Figure Page
1 Number of Primary Elevators Over Time . . . . . . . . . . . . . . 15
2 Entrant, Exits and HHI . . . . . . . . . . . . . . . . . . . . . . . . 18
3 Employment and wages for “narrowly defined”
agricultural firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4 Employment and wages for “narrowly defined” retail firms . . . . . 45
5 Comparing “narrowly defined” agricultural labor
market outcomes in Colorado and its synthetic control . . . . . . . 50
6 Comparing “narrowly defined” agriculture labor
market outcomes in Washington and its synthetic control . . . . . 51
7 Comparing “narrowly defined” retail labor market
outcomes in Colorado and its synthetic control . . . . . . . . . . . 52
8 Comparing “narrowly defined” retail labor market
outcomes in Washington and its synthetic control . . . . . . . . . 53
B.2.1 Employment and wages for “broadly defined”
agricultural firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
B.2.2 Employment and wages for “broadly defined” retail firms . . . . . 82
B.2.3 Comparing broadly-defined agriculture labor market
outcomes in Colorado and its synthetic control . . . . . . . . . . . 83
B.2.4 Comparing broadly-defined agriculture labor market
outcomes in Washington and its synthetic control . . . . . . . . . . 84
B.2.5 Comparing broadly-defined retailer labor market
outcomes in Colorado and its synthetic control . . . . . . . . . . . 85
x
Figure Page
B.2.6 Comparing broadly-defined retailer labor market
outcomes in Washington and its synthetic control . . . . . . . . . . 86
B.2.7 Placebo tests for narrowly-defined agriculture labor
market outcomes in Colorado . . . . . . . . . . . . . . . . . . . . . 87
B.2.8 Placebo tests for narrowly-defined agriculture labor
market outcomes in Washington . . . . . . . . . . . . . . . . . . . 88
B.2.9 Placebo tests for narrowly-defined retailer labor
market outcomes in Colorado . . . . . . . . . . . . . . . . . . . . . 89
B.2.10 Placebo tests for narrowly-defined retailer labor
market outcomes in Washington . . . . . . . . . . . . . . . . . . . 90
B.2.11 Placebo tests for broadly-defined agriculture labor
market outcomes in Colorado . . . . . . . . . . . . . . . . . . . . . 91
B.2.12 Placebo tests for broadly-defined agriculture labor
market outcomes in Washington . . . . . . . . . . . . . . . . . . . 92
B.2.13 Placebo tests for broadly-defined retailer labor market
outcomes in Colorado . . . . . . . . . . . . . . . . . . . . . . . . . 93
B.2.14 Placebo tests for broadly-defined retailer labor market
outcomes in Washington . . . . . . . . . . . . . . . . . . . . . . . . 94
C.1 Example bid cost reporting . . . . . . . . . . . . . . . . . . . . . . 115
C.2 The distribution of the overall prediction error over time . . . . . . 116
xi
LIST OF TABLES
Table Page
1 Descriptive Data of Primary Elevators . . . . . . . . . . . . . . . . 20
2 Basic Model Results . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3 Vertical Linkages and Spatial Competition Model Results . . . . . 25
4 Basic Model with Age Included . . . . . . . . . . . . . . . . . . . . 27
5 Vertical Linkages and Spatial Competition Model
Results with Age included . . . . . . . . . . . . . . . . . . . . . . . 28
6 Synthetic control estimates of the effect of
recreational cannabis legalization on narrowly-defined
labor market outcomes . . . . . . . . . . . . . . . . . . . . . . . . 54
7 Synthetic control estimates of the effect of
recreational cannabis legalization on broadly-defined
labor market outcomes . . . . . . . . . . . . . . . . . . . . . . . . 56
8 Results from alternative specifications of weekly wages
per worker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
A.1 Basic Model Results After 2002 . . . . . . . . . . . . . . . . . . . . 75
B.1 List of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . 76
B.2.1 Synthetic control weights assigned to each state for
narrowly-defined agriculture labor market outcomes . . . . . . . . 77
B.2.2 Synthetic control weights assigned to each state for
narrowly-defined retail labor market outcomes . . . . . . . . . . . 78
B.2.3 Synthetic control weights assigned to each state for
broadly-defined agriculture labor market outcomes . . . . . . . . . 79
xii
Table Page
B.2.4 Synthetic control weights assigned to each state for
broadly-defined retail labor market outcomes . . . . . . . . . . . . 80
B.3.1 CO broadly-defined agriculture average weekly wage
per worker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
B.3.2 CO broadly-defined agriculture total quarterly wages . . . . . . . . 96
B.3.3 CO broadly-defined agriculture average employment . . . . . . . . 97
B.3.4 CO broadly-defined agriculture number of establishments . . . . . 98
B.3.5 WA broadly-defined agriculture average weekly wage
per worker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
B.3.6 WA broadly-defined agriculture total quarterly wages . . . . . . . 100
B.3.7 WA broadly-defined agriculture average employment . . . . . . . . 101
B.3.8 WA broadly-defined agriculture number of establishments . . . . . 102
B.3.9 CO narrowly-defined agriculture average weekly wage
per worker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
B.3.10 CO narrowly-defined agriculture total quarterly wages . . . . . . . 104
B.3.11 CO narrowly-defined agriculture average employment . . . . . . . 105
B.3.12 CO narrowly-defined agriculture number of establishments . . . . . 106
B.3.13 WA narrowly-defined agriculture average weekly wage
per worker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
B.3.14 WA narrowly-defined agriculture total quarterly wages . . . . . . . 108
B.3.15 WA narrowly-defined agriculture average employment . . . . . . . 109
B.3.16 WA narrowly-defined agriculture number of establishments . . . . 110
B.3.17 CO narrow retail average weekly wage per worker . . . . . . . . . . 110
B.3.18 CO narrow retail total quarterly wages . . . . . . . . . . . . . . . . 111
B.3.19 CO narrow retail average employment . . . . . . . . . . . . . . . . 111
xiii
Table Page
B.3.20 CO narrow retail number of establishments . . . . . . . . . . . . . 111
B.3.21 WA narrow retail average weekly wage per worker . . . . . . . . . 111
B.3.22 WA narrow retail total quarterly wages . . . . . . . . . . . . . . . 112
B.3.23 WA narrow retail average employment . . . . . . . . . . . . . . . . 112
B.3.24 WA narrow retail number of establishments . . . . . . . . . . . . . 112
B.3.25 CO broad retail average weekly wage per worker . . . . . . . . . . 112
B.3.26 CO broad retail total quarterly wages . . . . . . . . . . . . . . . . 113
B.3.27 CO broad retail average employment . . . . . . . . . . . . . . . . . 113
B.3.28 CO broad retail number of establishments . . . . . . . . . . . . . . 113
B.3.29 WA broad retail average weekly wage per worker . . . . . . . . . . 113
B.3.30 WA broad retail total quarterly wages . . . . . . . . . . . . . . . . 114
B.3.31 WA broad retail average employment . . . . . . . . . . . . . . . . 114
B.3.32 WA broad retail number of establishments . . . . . . . . . . . . . . 114
C.1 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
C.2 Correlations between the overall forecast error and its
major components . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
C.3 The distribution of the overall prediction error across
quartiles of measures of experience and competition . . . . . . . . 119
C.4 The association between prediction errors and plan observables . . 120
C.5 The association between absolute prediction errors
and plan observables . . . . . . . . . . . . . . . . . . . . . . . . . . 121
C.6 The association between plan benefits and the overall
prediction error . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
C.7 The association between plan benefits and the
absolute overall prediction error . . . . . . . . . . . . . . . . . . . 123
xiv
CHAPTER I
INTRODUCTION
This thesis is composed by three essays and provides empirical contributions
to the Industrial Organization literature, specially in the field of transportation and
health economics. It aims to understand different issues related to economics by
applying various empirical methods which include models of exit, policy effects on
labor market, and prediction errors in the context of Medicare Advantage settings.
The empirical models include conventional techniques logit modeling and piecewise
regression models, causal inference methods i.e., different in different and synthetic
control, and some machine learning applications like LASSO, MICE, regression
trees and etc.
In chapter 2, jointly with Dr. James Nolan and Dr. Wesley W. Wilson, we
examined firm exit in Canadian markets, specifically the grain elevator market.
Grain elevators play a central role in the movement of grain to market and to
rural economies in terms of employment and investment. Over the last three
decades, the grain elevator industry in Canada has experienced a major decline
in the number of elevators as older and technologically obsolete elevators have
been replaced by larger and more technologically advanced elevators. We develop
a model of exit in the Canadian grain elevation industry using data from 1999 to
2016 collected at the individual elevator level. Our specification explains elevator
exit based on traditional variables used in the industrial organization literature
such as capacity, multi-plant ownership, and vintage. But, we also include a
measure of vertical linkages in the industry (i.e., the effects of vertical investments
in transportation infrastructure) as well as spatial measures to account for local
demand, supply and competition. The results provide strong evidence that exit
1
in this key agricultural and trade industry is affected by whether an elevator is
a recent entrant (vintage), its size, vertical linkages, local demand and supply
conditions, and spatial competition.
In chapter 3, jointly with Dr. Keaton Miller, we examined the effects of
marijuana legislation change on the agricultural labor market. Over the past
several years, cannabis has become legal for recreational use in several U.S.
states and jurisdictions around the world. The opening of these markets has led
to the establishment of hundreds of cannabis production and retail firms with
accompanying demand for labor, leading to concerns about spillover effects on
wages from incumbents. We study the markets for agricultural and retail labor
in Washington and Colorado, early legalizers with now-established cannabis
markets. Using a synthetic control technique to account for the possibility of
border-state spillover effects and machine learning techniques for data imputation
and variable selection, we find no evidence that cannabis legalization is associated
with increases in per-employee wages, neither within industries most similar to
cannabis production or retail, nor in more broad industry categories. We conclude
that cannabis legalization is unlikely to negatively impact incumbent firms through
the labor market channel.
In chapter 4, jointly with Dr. Keaton Miller we investigated the accuracy
of the predictions in the context of Medicare Advantage, where insurers receive
subsidies from the government and compete to provide health insurance to seniors.
As part of the “bidding” process, firms must submit forecasts of their costs.
Insurers have incentives to report accurately, as these predictions are used to
determine both the level of the subsidy and (implicitly) the degree to which that
subsidy can be spent on various plan features. We collect data on predictions
2
and realized expenses per member per month at the plan-service-category level
from 2008-2015, and document three stylized facts. Our results show that first, on
average firms overestimate future costs. Second, this overestimation decreases with
the experience of the firm. Firms in more competitive markets (as measured by the
number of other firms present) form more accurate estimates. We show that firms
with higher costs than expected generally offer plans that feature greater patient
cost sharing (i.e. higher deductibles and copays).
3
CHAPTER II
EXIT DECISIONS IN THE CANADIAN GRAIN ELEVATOR
INDUSTRY
From Jiang, S., Nolan, J. & Wilson, W.W. (2021). Exit Decisions in the
Canadian Grain Elevator. Journal of Industry, Competition and Trade, 1-19.
2.1 Introduction
Grain elevators have long been a visible and important component in rural
communities. They not only provide employment opportunities and invest in
the local areas, they are also the central local gathering point for agricultural
products and link the local markets to destinations. Elevators not only receive
grain, they also store, blend and/or treat grains, ultimately loading the stored
grain for shipment to terminals and processing plants. In Canada, since the first
Prairie elevator was built in 1881 in Gretna, Manitoba there has been enormous
investment in rural elevators to accommodate the long-term growth of the Prairie
grain industry.1
Over the last few decades, the industry has transitioned as older (mostly
built of wood) and smaller elevators have given way to more modern (mostly
built of concrete) larger elevators. In this paper, we examine this transition
by developing and estimating a model of exit that captures traditionally used
determinants such as scale, multi-plant firms, plant vintage, but also develop
and include a measure of vertical linkages as well as local measures of demand,
supply, and spatial competition. We develop a panel dataset of Western Canadian
elevators in operation during 1999-2016 time period. These data provide a
comprehensive unbalanced panel that allow patterns of exit to be described and
1https://www.thecanadianencyclopedia.ca/en/article/grain-elevators.
4
allow the estimation of a unique model of exit to be estimated. We first estimate
a model motivated by the industrial organization literature on exit. The model
includes variables such as capacity, multi-plant ownership, and whether or not an
elevator was a recent entrant to the market to capture elevator vintage. We then
develop and include in the model, measures of the vertical relationship between the
elevators and the transportation markets as well as a set of variables to reflect the
spatial setting of the elevator. These latter include local measures such as demand,
supply and spatial competition. The results provide strong evidence consistent with
the literature that capacity and time of entry (i.e. more recent, newer technology)
have negative effects on exit. We also find that vertical relationships as well as
spatial measures of demand, supply and spatial competition have expected and
strong effects on the probability that an elevator in the sample exited the market.
The current industry in Canada consists of four types of elevators. These
are: primary, forwarding, process, and terminal elevators. Primary elevators receive
grain directly from producers for storage and forwarding. Process elevators receive
and store grain for direct manufacture or processing into other grain products.
Terminal elevators receive grain after official inspection and weighing, cleaning
and storing grain before moving it along the supply chain; and transfer elevators
transfer grain that has been officially inspected and weighed at another elevator.
Transfer elevators can also receive, clean, and store domestic or foreign grain.2
Primary elevators still dominate the grain elevator industry in Canada. In 1999,
there were 976 primary elevators, and a total of 57 other types of elevators. The
number of primary elevators generally decreases throughout the range of the data
(1999-2016). Yet since 2004, total capacity of primary grain elevators has increased
2https://en.wikipedia.org/wiki/Grain elevator
5
in most Canadian provinces. For example, Alberta has seen total primary elevator
capacity increases from 1,685,250 to 1,834,160 tonnes within the time frame of 1999
to 2016. British Columbia is the only province where inland elevator capacity fell,
from 46,030 in 1999 to 41,130 tonnes in 2016.3 Overall, the industry now has fewer
but larger elevators, while these remain mostly primary elevators. Concurrently,
the number of process and terminal elevators have increased while their average
capacity has dropped. And since 2013, no transfer elevators have been operational
in this market. We focus the analysis on primary elevators, given their dominance
in the industry.
There is considerable research on industrial entry and exit.4 Generally, this
literature finds that inefficient firms/plants tend to exit the market either due to
the lack of scale economies or due to inherent inefficiencies.5 Other research points
to strategic motivations for exit e.g., Ghemawat and Nalebuff (1985), Ghemawat
and Nalebuff (1990), and others. In addition, recent research provides evidence that
firm characteristics at time of entry have an important effect on exit (T. Dunne,
Klimek, and Roberts (2005), while other research theoretically points to the role
of multiplant ownership e.g., Reynolds (1988) and Ghemawat and Nalebuff (1990)
and and empirically (e.g., Audretsch and Mahmood (1995), Mata et al. (1995),
3The data comes from the Canadian Grain Commission, https://www.grainscanada.gc.ca/
wa-aw/geic-sgc/summary-sommaire-eng.asp
4There is a related literature that examines the effects of financial information on firm
failure e.g., Altman (1968, 1973), Zingales (1998). Zingales (1998) finds, for example, that
highly leverages trucking firms are less likely to survive following deregulation. In the case of
the Canadian grain elevator industry, elevator success in Western Canada is tied to several
factors, including financial/operational elements (e.g. turnover, capacity, pricing, etc.) as well
as transportation (i.e. rail and road) connectivity (Lawrence, Nolan, & Schoney, 2016).
5See, for example, Franklin (1974), Jovanovic (1982),T. Dunne, Roberts, and Samuelson (1988,
1989), Lieberman (1990),Audretsch (1991, 1995), P. Dunne and Hughes (1994), Mata, Portugal,
and Guimaraes (1995), Gibson and Harris (1996), Audretsch, Houweling, and Thurik (2000),
Segarra and Callejón (2002), Elston and Agarwal (2004), K. S. Miller and Wilson (2018)and etc.
6
and K. S. Miller and Wilson (2018)).6 Previous research has shown that the effects
of scale can have a positive effect on the likelihood of exit. That is, in declining
demand markets. Ghemawat and Nalebuff (1985) show theoretically that a small
single plant firm can profitably “hang on” longer than a large firm, with the result
that the larger firm exits first. In a subsequent paper, Ghemawat and Nalebuff
(1990) allow partial capacity adjustments, but similarly find that large firms
reduce capacity before small firms and then both small and large firms reduce
capacity until their plants are equally sized, subsequently reducing capacity at
the same rate. When firms operate multiple plants, Whinston (1988) find that
the large plant can improve a multi-plant firm’s strategic position in the survival
game, with the result that this firm will not necessarily be the first to exit or
cut capacity. Fudenberg and Tirole (1986) analyze a model in which two firms
possess asymmetric information about each other’s fixed costs, but hold symmetric
expectations. They identify a unique subgame perfect equilibrium where high-cost
firms exit earlier than low-cost firms. We conclude that overall, the implications
of plant capacity on exit decisions are mixed. That is small and/or inefficient
firms can be “shaken” out earlier, but in declining markets it may be that small
firms “stakeout” the market with the result that larger firms exit first. (Lieberman
(1990), and Blonigen, Liebman, and Wilson (2007))
In this paper, we control for what have become standard exit related
variables such as capacity, multi-plant ownership, and whether the plant (individual
elevator) is a recent entrant (to capture vintage effects). But given the nature of
the industry, we also introduce industry specific variables that may also have an
6There are other studies that reinforce these findings, covering a broad range of countries.
These include Italy(Colombo & Delmastro, 2000), the United States (Bernard & Jensen, 2007),
Belgium (Van Beveren, 2007), Sweden (Bandick, 2007), Japan (Kimura & Kiyota, 2006), New
Zealand (Gibson & Harris, 1996), Chile (Alvarez & Görg, 2009), etc.
7
effect on elevator exit. Since elevators are part of an extensive grain supply chain
in Canada, their individual relationships with the grain transportation market
can also influence their long term viability. In this sense, rail transportation
in particular represents a vertical linkage connecting individual grain elevators
to final markets. Plants/elevators supported by well developed transportation
infrastructure would seem less likely to exit the market. Some elevators now also
have considerable capacity to load hopper cars, which means they can ship large
quantities, leading to lower negotiated freight rates than elevators with smaller
capacities. We capture this effect with a rail carload capacity variable and find that
it has a significant negative effect on the probability of a given elevator exiting the
market. In addition, we also introduce a set of variables to account for differences
in local demand and supply conditions, as well as spatial competition from other
proximate elevators. The latter measure is a measure of capacity of other proximate
elevators (calculated using weights inversely related to distance to the neighboring
elevators).
There is a limited amount of research that applies to vertical relations in
firms’ exit decisions. As an example, Chen (2002) use a duration model to find that
vertical integration reduces the likelihood of survival for US petroleum plants. de
Figueiredo and Silverman (2012) examine exit rates in the US laser printer and
manufacturing industry, finding that the density of a vertically related population
has an adverse effect on exit rate. In our market, a clear vertical linkage exists
between grain elevators and rail transportation. While some elevators can load only
a few cars at a time, others can load dozens of rail cars in short order. Railroads
will typically offer rate discounts for multiple rail car shipments over smaller rail
car movements, a situation that places elevators having high car loading capacity
8
with a substantial competitive advantage.7 Due to this, car loading capacity is our
measure of vertical linkage. In the case of grain elevators, Sarmiento and Wilson
(2005) show that large elevators have a greater tendency to adopt a shuttle train
technology than smaller ones, while the size of a rival has a negative impact on
adoption decisions. They also find shuttle train technologies tend to be adopted in
regions with high production and less competition.
The remainder of this paper is organized as follows: Section 2 contains some
general background of the grain elevation industry in Canada and provides a review
of related academic literature; Section 3 describes the data on Western Canadian
grain elevators, providing more details about firm entry and exit during the study
period; Section 4 presents the results of various econometric specifications and then
discusses our findings; Section 5 concludes. At the very end of the paper, we have
added an Appendix of the basic model estimates as Appendix A.
2.2 Background
Grain elevators have long been crucial to Canadian rural agricultural
communities, not only in terms of the services provided but also in terms of
employment, investment, local purchases, etc. Their primary role is to provide
a convenient collection point for local grain, including storage and processing,
as well as providing a connection with rail transportation that allows access to
both domestic and international markets. In Canada, most grain operations on
7While there remains limited regulation on grain rates in Canada, the importance of rate
discounts for larger shippers was buttressed with the so-called Great Northern Grain (GNG) case
of 2007 (filed with the Canadian Transportation Agency (2007)) whereby a relatively small grain
shipper (GNG) filed a rate complaint against a major railway. The case concerned the level of
rate discounts being offered by railways at that time to elevators that had high loading capacities.
GNG argued this was effectively undue discrimination against smaller grain shippers like itself
who could not easily improve their loading situation. GNG won the case with the regulator, with
the consequence that the major railways were forced to revise their grain rate schedules to render
them less discriminatory against smaller grain shippers.
9
the Prairies are linked to the Class 1 railroad network. An interesting feature of
Canadian elevators is their geographic dispersion. As discussed by Selyem (2000),
Canadian Prairie towns were historically located approximately 6-10 miles apart,
a distance based mostly on the limitation of transportation modes at the time, as
well as the availability of farm inputs such as water and fuel. While much fewer
in number today, grain elevators were and remain a significant business activity in
many rural communities, especially in Western Canada. While rooted in history,
change now characterizes the modern grain elevator industry in Canada. As an
example, up until 2012, Canadian farmers sold their grain through the Canadian
Wheat Board (CWB), the quasi-governmental agency acting as the sole marketer
of Prairie grains destined for export from Canada.8 While the CWB no longer
exists, farmers still need to deliver and sell their grain through a licensed grain
elevator company. In turn, grain companies gain comparative advantages through
procuring grain in local markets, coupled with other factors such as elevator
capacity, ownership, and rail carload capacity.
In this market, historical smaller capacity elevators have gradually given
way to fewer but larger and more modern successors. Average elevator capacity
has nearly tripled in recent years; in 1999 capacity was 6,558.34 tonnes but grew
to 20,568.66 tonnes by 2016. Modern elevators offer higher-speed loading and
unloading facilities, fast grain cleaning capabilities, unit train loading ability, and
substantial storage space.
The history of the ownership of Canadian grain elevators is of interest and
is important to understand the transition to the modern era. In early pioneer days
individuals living in Western Canada’s prairie towns often built their own grain
8https://www.thecanadianencyclopedia.ca/en/article/canadian-wheat-board
10
elevators (as co-operatives), and this process gradually brought in private grain
companies as competition.9 While much of the 20th century was dominated by
the provincial (Alberta, Saskatchewan and Manitoba) pool elevator companies,
by the mid-1990s falling costs of grain processing led to increased consolidation
through mergers.10 Within our data set, in fact, several major mergers occurred.
These include Agricore United taking over United Grain Growers in 2001, and
subsequently Agricore United was taken over by the Saskatchewan Wheat Pool in
2007. The latter merger created the largest grain handler in Canada, re-named as
Viterra Inc.11 Mergers in this industry have been mostly approved by Canadian
competition authorities, but the latter merger was subject to some regulatory
intervention due to competitive concerns (i.e. creation of more market power) in
several areas, including at the Port of Vancouver. In summary, over the last 25
years, the grain elevator industry in Canada has been characterized by relatively
small number of multi-plant (elevator) firms.
Bulk transportation has been a driving force in agriculture and in particular
the grain industry in Canada. When the Canadian Pacific Railway (CPR) linked
the Pacific province of British Columbia to the rest of Canada in 1871, one goal
of the Federal government at that time was that this would help Canada expand
its nascent grain export markets (Larson & James, 2007). Railroads were an
important part of Western Canadian expansion, with farming and rail access going
hand in hand for new immigrants looking to settle the West. To this end, grain
transportation rates were regulated by the Canadian government ever since the
subsidy given to help CPR build the final rail linkage to British Columbia. But
9https://en.wikipedia.org/wiki/Grain elevator
10https://en.wikipedia.org/wiki/Grain elevator
11https://en.wikipedia.org/wiki/Viterra
11
as a mature industry, by the 1970s the Canadian government had also started
to subsidize various mainline rail upgrades to support on-going grain shipments.
However, around this time the government also began to allow the two Canadian
Class 1 railways to sell or abandon rail lines that were deemed to be uneconomic
(Vachal & Bitzan, 1997).
Many of the abandoned lines were in fact so-called “grain dependent”
branch lines, track that in some cases served long lines of individual town
grain elevators through distant parts of the Prairies. The peak period of this
abandonment process was between 1984 to 1996, where the total length of so-
called grain-dependent branch lines in Western Canada dropped by about 14
percent (Thraves, 2007). These abandonments hastened the demise of the old
and small wooden elevators. The trend was reinforced with the repeal of Crow
rate regulations in 1996 (J. F. Nolan & Kerr, 2012) and then slowed when new
and stricter rate regulations were introduced in 2000 as the railways and grain
companies had streamlined into more efficient grain handling networks (Brewin,
Schmitz, Nolan, & Gray, 2017).
Concurrently, the ownership of one major Class 1 railroad underwent
dramatic change with the privatization of the formerly publicly owned Canadian
National Railway in 1995. Prior to this, the operations of the multi-modal
Canadian Pacific Limited had been devolved into five independent companies.
This ultimately left two private railroads carrying Canadian grain at regulated
rates. Change was visible in other ways. The average number of rail cars that
could be loaded by remaining grain elevators increased with the consolidation and
modernization of the grain handling system. For example, over the duration of the
12
data set we found average car load capacity for all elevators nearly tripled from
22.97 to 62.84.
In comparing the industry in Canada and the U.S., it is worth noting that
Canada is a considerably smaller grain producer, with a much greater focus on
export markets, whereas the U.S. grain industry splits between domestic and
export markets, with domestic grain markets dominating elevator operations in
most states. To this end, previous research on the grain elevator industry and
industry evolution focuses on the U.S. market. Relevant works include Frittelli
(2005) who find that between 1980 to 1998, the number of farms decreased by 15%,
but farm size increased by 11%; concurrently, the number of terminal elevators
increased, but the total number of grain elevators dramatically fell, mostly due
to country elevators exiting the market. Other research has looked into vertical
relations within the grain industry in the U.S. Schmiesing, Blank, and Gunn (1985)
find that increases in the use of (large) unit grain trains in turn gives elevators
access to larger and in some cases more distant markets, improving their price
efficiency. Huang (2003) determined factors affecting shuttle12 adoption, and as
of the late 1980s with changes to how railroads marketed to grain shippers, shuttle
trains have been increasingly adopted by the elevator industry. Prater, Sparger,
Bahizi, and O’Neil (2013) highlighted the importance of grain train shuttles to
railroad efficiencies. Local grain elevators that have been unable to accommodate
shuttle-train shipments (for example, because they had small siding or loading
capacity) have mostly gone out of business. In essence, many believe that with
the demise of the CWB, the Canadian grain elevator industry is becoming similar
12A type of bulk train movement. A shuttle is a dedicated set of 75 to 110 covered grain hopper
cars that carry just grain from one destination to another. https://www.up.com/customers/
ag-prod/shuttle/index.htm
13
to that in the US, but “is approximately 20 years behind the grain movement
system now operating in the United States.(D. Wallace, 1997).” Finally, Vachal
and Bitzan (1997) examine survey data of Canadian grain elevators. At that time,
they concluded that industry parties were in fact expecting a declining number
of elevators in Canada in the short run. However, respondents also expected an
increase in production as well as in overall elevator capacity.
2.3 Data
The data we use contains information on all grain handling facilities in
Western Canada that were licensed from 1999 to 2016 (through the Canadian
Grain Commission). The data was downloaded directly from the Canadian Grain
Commission website.13 In total, 1346 elevators operated over this time period.
Many elevators are observed over time which allows us to ascertain whether an
individual facility exited the market (or not) and when it exited. For each elevator,
the data contains details such as storage capacity, ownership, the town closest
to the facility, the geographical coordinates of the town, the type of elevator, the
railroad(s) that serve the elevator, as well as the type of elevator. As we shall see,
due to the importance of proximate grain production to an elevator, we also added
data on regional agricultural production.
Initial examination of the data highlights the major changes in the industry,
especially at the beginning of the sample. As alluded to earlier, the number of
primary grain elevators decreased markedly through the 1990s as a result of rail
line abandonment and the repeal of Crow rates (discussed previously). Some
of the decline of the 1990s is reflected in our data from 1999 to 2002 (Figure 1)
13https://www.grainscanada.gc.ca/en/grain-research/statistics/grain-deliveries/
14
and indicates the magnitude of the declines. But since 2002, the total number of
elevators in the region has been relatively stable.14
Figure 1. Number of Primary Elevators Over Time
The dramatic decline in elevators was the result of more cost efficient
transportation options for grain movement as well as industry technological change.
Ceteris paribus railroads prefer higher loading and shipment volume, creating the
so called “unit trains”.15 These lead to lower unit costs. As a result of increased
14In our later analysis, we model the data using both the whole sample period as well as data
after 2002, and the results are qualitatively equivalent as well as numerically similar.
15A unit train, also called a block train or a trainload service, is a train in which all cars
(wagons) carry the same commodity and are shipped from the same origin to the same
destination, without being split up or stored en route. https://en.wikipedia.org/wiki/
Unit train
15
freedoms to conduct line abandonments at that time, Canadian railroads chose
to abandon numerous low-density branch lines on the Prairies. Concurrently,
small-town grain elevators were gradually replaced by more dispersed but larger
and more efficient terminals, which effectively caused farmers to truck their grain
over considerably longer distances (J. Nolan, 2007). Since older grain elevators
remain a nostalgic symbol for many western Canadians, some towns have succeeded
in preserving elevators by switching them into museums or art galleries. But
preservation is not the norm and most have been deteriorated or dismantled.
Over the last 20 years, newer grain elevators have tended to have much larger
capacities and are more efficient and durable than their predecessors.16 These high-
efficiency grain elevators not only facilitate loading/unloading grain more quickly
than previously, but they also help maintain higher grain quality on a large scale
(Simmins, 2004).
The dependent variable in our analysis is a discrete variable that reflects
whether or not an elevator exits in the subsequent time period. This variable
reflects a determination by the owning grain company that the long-run profit
of a given individual elevator does not support keeping it open. In effect, these
profits are assumed to be a function of the elevator capacity, whether the elevator
is owned by a company that owns other elevators, whether the elevator entered
the data after the first year of the data (i.e., new entrant17), rail car loading/siding
capacities, and local demand and supply conditions, including the degree of spatial
competition.
16https://www.farmprogress.com/grain-handling/new-innovations-grain-storage
-systems-higher-capacities-and-better-grain-quality
17In our case the first year of data is 1999. We do not observe when the plants that existed in
1999 first appeared. But, we are able to identify plants that first appeared after 1999 which is the
basis for the entrant dummy and follows an approach used by T. Dunne et al. (2005).
16
The size of a grain elevator plays an essential role on exit behavior. Over
time the average capacity has increased. In 1999, the average capacity was 6,558
tonnes, increasing to 16,723 tonnes in 2009, and to 20,568 tonnes in 2016 (Table 1).
Elevator ownership is a discrete variable that takes a value of one if it is the
single elevator owned by a firm or zero if it is owned by a firm operating multiple
elevators. The number of single ownership elevators is relatively small, but has
increased modestly through the time period (Table 1). We also observe from the
data that the status of grain elevator ownership (i.e self-owned-plant or a multi-
plant firm) usually does not change given that the elevator remains in the market.
Statistically, we only have a single observation that switched its (plant) ownership.
In fact, this elevator was owned by a multi-plant firm at the beginning of the
sample period, but later became a self-owned elevator, and went out of business
in 2006.
Since we have the capacity of each licensed primary elevator throughout the
sample, if an elevator is not part of the 1999 data then we record it as an entrant
and, as such, these are more likely to be more technologically advanced than
elevators that have been in the market longer. In principle, such elevators tend
not to exit (at least immediately after entry), which is consistent with T. Dunne et
al. (2005), who found that entry barriers are also exit barriers. The total number
of new entrants since 1999 increased to 247 by 2016, as shown in Table 1. Figure
2 ((a)) illustrates the number of entrants and exits throughout the data. As noted
earlier, there was dramatic exit from 1999-2003. Since then the number of exits has
been relatively stable and remain about the same over time.
The number of firms in the industry has also changed over time. But,
the change varies over time. In 1999, there were a total of 32 firms in our data
17
(a) Entrants and Exits (b) HHI
Figure 2. Entrant, Exits and HHI
increasing to 68 by 2011 and remaining somewhat constant since then. In 1999, the
top three elevators owners in our data are Saskatchewan Wheat Pool (303 elevators
with 1,523,880 tonnes capacity), Agricore Cooperative Ltd. (258 elevators and
1,371,140 tonnes capacity), and United Grain Growers (126 elevators and 820820
tonnes capacity). The number of elevators owned by top firms within the industry
dropped largely while the total capacity increased. In 2016, the largest three
elevator owners were Viterra Inc. (72 elevators and 1,884,570 tonnes capacity),
Richardson Pioneer (58 elevators and 153,640 tonnes capacity), and Patterson
Grain (28 elevators and 710,750 tonnes capacity) leading the market.
The number of firms is changing over time and has generally increased over
the time span of the data, but the average capacity held by firms has also changed.
The average capacity per firm, was initially large in 1999, fell dramatically over
the next five years and then increased, leaving the concentration in the market
somewhat mixed. As illustrated in Figure 2(b), the Hirshmann-Herfindahl Index
(HHI) for the industry generally increased from 1999 to 2007, but has fallen
somewhat since then.
Vertical linkage of a grain elevator to the transportation sector is measured
through car loading capacities (in rail cars) associated with each elevator. Elevators
18
that can load greater numbers of rail cars tend not to leave the industry. The
average car loading number for each of the elevator has increased from 22.96 in
1999 to 62.83 in 2016 as shown in Table 1. From this table, it is also clear that
surviving elevators enlarge their grain and car capacities.
Total agricultural production data18 in the area was collected at the
elevator station level.19 We also have data on the third level of census division,
or subdivision of Canada, and we aggregate total production in the subdivision
to provide a measure of local demand for elevator services.20 Additionally, we
aggregate the total amount of elevator capacity in each of the subdivisions as a
measure of the supply of elevator services. From Table 1, we find that agricultural
production per subdivision, per elevator, and per unit capacity are all increasing
throughout the sample period.
The data contain the geographical coordinates of each elevator. We use the
geographic coordinates to first calculate distances between each of the facilities.
Then using the distances between facilities, we construct an inverse distance
weighted mean value of elevator capacity (Liu et al., 2021). The idea is that nearby
elevators have a larger influence than elevators located further away. The general
formula for inverse weighted means is given below. Different exponent values
relate to the decay associated with distance. To this end, we considered 1 , 1 and
2
2. Further, we restricted the range of a fixed distance. In our case, the calculations
were based on a distance of 20 miles, but we also considered distances that ranged
18https://www150.statcan.gc.ca/n1/en/type/data?MM=1
19The original data has the station geographic coordinates in the smallest administrative
division in Canada, such as cities, towns, villages, townships, and parishes and etc. https://
en.m.wikipedia.org/wiki/Administrative divisions of Canada
20https://en.wikipedia.org/wiki/Census geographic units of Canada
19
from 10 to 150 miles. The formula ∑is given by:n ∑n
ẑ (x0) = z (xi) · d−r −rij / dij
i=1 i=1
Our empirical findings are consistent across varying exponents and distances.
The reported results are based on an exponent of 1 and we have included all
2
elevators in the sample within 20 miles of the reference point.21
Table 1. Descriptive Data of Primary Elevators
Variables 1999 2009 2016
Number of Elevators 976 314 345
Exit 178 12 0
Capacity (tonnes) 6558 16723 20569
Single Elevators 18 23 39
Entrant 0 151 247
Car Loading Capacities 22.96 54.87 62.83
Agriculture Production (thousands of tonnes) 63.5 210 280
Weighted Capacity of 20 mile range 7093.88 16705.81 20221.44
Agriculture Production Per Elevator (in subdivision) 3170 25852 34549
Agriculture Production Per Unit of Capacity (in subdivision) 0.486 1.79 1.78
2.4 Econometric Specification and Results
In this section, we examine exit behavior under a logit choice specification.
In particular, we know that conceptually firms will exit if the long-run profits of the
firm with exit are greater than the long-run profits of the firm if they do not exit.
Our approach is common in the literature e.g., Blonigen et al. (2007); T. Dunne et
al. (2005); Lieberman (1990), and K. S. Miller and Wilson (2018). As discussed in
K. S. Miller and Wilson (2018), the approach can emerge naturally from common
dynamic models of entry and exit e.g., Ericson and Pakes (1995) where firms
make exit decisions based on expected future profits which are a function of their
individual state, the state of their competitors as well as market conditions. In our
21We also explored different exponents (r=1 and 2) and examined different distances
(d=10,20,· · · ,150 miles). These results are generally consistent with those presented through
the rest of the paper.
20
model, we capture individual states with measures of capacity, whether the elevator
is recent entrant, whether it is part of a multi-plant firm, the state of competitors is
captured in inverse weighted capacity measures (spatial competition), and market
conditions are captured in the local demand and supply measures. The specification
allows us to identify characteristics that contribute to elevator longevity in this
market. Effectively, for the ith elevator, we define the latent variable as Y ∗
Y ∗i = β ×Xi + ε. (2.1)
As discussed earlier, we do not observe profits, but we do observe whether the
elevator or firm exits, which is represented by Yi.
22 The dependent variable equals
one if the elevator exits the market at the end of the year, and has a value of zero if
it did not.
The explanatory variables are represented by Xi, while β is a vector of
parameters to be estimated. The variables considered include grain capacity, if
the elevator is part of a multi-plant firm, if the elevator is an entrant, car loading
amounts, agricultural production at the subdivision level, weighted capacity,
subdivision capacity, as well as agricultural production per unit of capacity within
the subdivision.23 The detailed explanations are:
22Another approach that could be considered is based on hurdle rates which reflect the
minimum rate of return to gauge where or not to pursue (or maintain) a project e.g., Brigham
(1975), Liesch and Knight (1999), These are, of course, very comparable in concept in that
decisions can be interpreted as the result of long-run profits. In our case, we simply do not have
access to the cost and returns necessary to implement a hurdle approach, and instead follow the
bulk of the literature to use the decision to reflect the long-run profits.
23The pairwise correlations amongst the right-hand side variable are generally quite small and
the empirical results are generally robust across a wide range of specifications suggesting the any
issues related to multicollinearity are non-existent or quite small. Further, a referee suggested that
we consider the possible endogeneity of capacity. However, given that the market shares defined
over sub-divisions are quite small (These average about 12 percent overall, and only 6 percent for
existing plants, with only about a 2.6 percent median value). Given the relatively small market
shares and the stability of coefficients, endogeneity does not appear to be a major issue.
21
LOG CAP = Logged capacity of the elevator
ELEV OWNERSHIP = One if the grain elevator is owned by a
single plant firm; zero if the elevator is
owned by multiplant firms.
ENTRANT = One if the grain elevator is not operating
at the start of the observation year but
enter the market later on; zero if the firm
is operating at the beginning
LOG CAR = Logged elevator car loading capacities
AG Production = Total agriculture production within each
subdivision level.
Weighted CAP 20 Mile = Weighted average capacity with inverse
distance of all elevators from the sample
excluding the reference point, center
elevator, within 20 miles
Subdiv CAP = Total grain elevator capacity within a
subdivision area
Ag Production/Subdiv Capacity = Average Agriculture Production per total
elevator capacity in a subdivision area
We present two sets of results. Table 2 contains a base model with elevator,
capacity, ownership, and whether the elevator is a new entrant (after 1999) with
different fixed treatments (by time and subdivision). Column 1 does not contain
any fixed effects and is the base specification; Column 2 includes time fixed effects;
Column 3 includes subdivision fixed effects; and Column 4 contains both time
and subdivision fixed effects. Table 3 adds measures of rail car-loading capacity
(vertical linkages) and measures of regional demand, supply and local competition.
All of the specifications in this model include time fixed effects.
The results reported in Table 2 are largely consistent with the previous
literature.24 That is, the effects of capacity and whether the elevator entered since
the beginning of the data are both negative and statistically significant. This
24Given the large degree of exit in the first few time periods, we estimated the models excluding
the initial three time periods where a lot of the exits were observed. The results are quite
comparable with similar results and is attached in the appendix, Table A.1
22
means that larger and more modern elevators are less likely to exit. The effect on
elevator ownership (a binary variable) has a negative coefficient, but is statistically
significant only in Column 1 (which does not contain any fixed effects). A negative
value here means that single ownership elevators are less likely to exit the market.
While the overall results are mixed in terms of significance, as discussed earlier, the
industry is still dominated by grain companies owning multiple elevators.
Table 3 incrementally adds rail car loading capacity, regional demand,
regional capacity, and spatial competition. In all cases, the results related to
capacity and whether or not the elevator is an entrant are quite similar to those
found in Table 2. Both capacity and being an entrant have a negative effect on the
likelihood of exit. In column 1, we add car loading capacity as a measure of vertical
linkages. In this specification as well as the ensuing specifications this coefficient
is negative and statistically significant. In column 2, we add total agricultural
production of the subdivision in which the elevator is located, while in column
3, we add the total elevator capacity in the region. The former reflects demand
conditions and the latter supply conditions. The coefficients on the demand
measure are negative, while the coefficients on the supply measure are positive,
which provide strong evidence that elevator in areas with strong demand and
weak supply are less likely to exit the market. In Column 5, we add the ratio
of the demand measure to the supply measure, and we find that the coefficient
is negative, reinforcing our findings. Finally in Columns 4 and 5 we include our
measure of spatial competition, that is the inverse distance weighted measure
of rival elevator capacities. These, as discussed, capture the presence of rivals
in the nearby area. This estimated coefficient in both specifications is positive
23
and statistically significant, providing more evidence that the likelihood of exit is
positively affected by the presence of spatial competition.
Table 2. Basic Model Results
(1) (2) (3) (4)
exit exit exit exit
LOG CAP −0.941∗∗∗ −0.889∗∗∗ −1.017∗∗∗ −0.986∗∗∗
(0.0474) (0.0506) (0.0516) (0.0553)
Elev Ownership −0.364∗ −0.223 −0.308 −0.208
(0.203) (0.209) (0.216) (0.221)
Entrant −1.307∗∗∗ −0.799∗∗∗ −1.390∗∗∗ −0.931∗∗∗
(0.110) (0.125) (0.114) (0.131)
Time Fixed Effect ✗ ✓ ✗ ✓
Subdivision Fixed Effect ✗ ✗ ✓ ✓
Constant 6.623∗∗∗ 5.928∗∗∗ 6.980∗∗∗ 6.419∗∗∗
(0.406) (0.427) (0.563) (0.587)
N 7662 7317 7660 7315
Log Likelihood −2552.8261 −2375.8799 −2489.3525 −2324.9201
standard errors in parentheses
∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Overall, multiple specifications are presented in Tables 2 and 3, yet there
is a remarkable similarity among the various models. In all models (both Table
2 and Table 3), capacity has a negative effect on the likelihood of exit which is
consistent with the notion that small country elevators are disappearing. Second,
elevators that entered since the beginning of the data are less likely to exit. Our
interpretation is that these elevators have been built with increasingly modern
technology, so our results highlight the importance of elevator modernization as
necessary in order to compete in this market. The results with respect to ownership
are somewhat mixed. The estimated effects are negative in all specifications, but
are mixed in terms of statistical significance. This may be the result, as mentioned,
that the industry in Canada is dominated by multi-elevator firms.
24
Table 3. Vertical Linkages and Spatial Competition Model Results
(1) (2) (3) (4) (5)
exit exit exit exit exit
LOG CAP −0.660∗∗∗ −0.461∗∗∗ −0.483∗∗∗ −0.468∗∗∗ −0.328∗∗∗
(0.061) (0.061) (0.062) (0.062) (0.068)
Elev Ownership −0.138 −0.413∗ −0.447∗∗ −0.371∗ −0.474∗∗
(0.220) (0.222) (0.224) (0.224) (0.235)
Entrant −0.821∗∗∗ −0.849∗∗∗ −0.887∗∗∗ −0.899∗∗∗ −0.847∗∗∗
(0.127) (0.129) (0.131) (0.131) (0.134)
LOG CAR −0.371∗∗∗ −0.386∗∗∗ −0.391∗∗∗ −0.380∗∗∗ −0.267∗∗∗
(0.062) (0.062) (0.063) (0.063) (0.067)
LOG(AG Production) −0.108∗∗∗ −0.113∗∗∗ −0.115∗∗∗ −0.274∗∗∗
(0.008) (0.008) (0.008) (0.019)
LOG(Subdiv CAP) 0.271∗∗∗ 0.231∗∗∗ 0.219∗∗∗
(0.052) (0.053) (0.054)
LOG(Weighted CAP 20 Mile) 1.876∗∗∗ 1.650∗∗∗
(0.457) (0.475)
LOG(Ag Production/Subdiv Capacity) −0.255∗∗∗
(0.027)
Constant 5.034∗∗∗ 4.818∗∗∗ 1.855∗∗∗ −14.450∗∗∗ −13.741∗∗∗
(0.435) (0.430) (0.715) (4.037) (4.186)
Time Fixed Effect ✓ ✓ ✓ ✓ ✓
N 7297 7297 7297 7297 7231
Log Likelihood -2340.7356 -2244.6586 -2230.1823 -2221.7229 -2138.394
standard errors in parentheses
∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Rail car loading capacity has a negative effect on the likelihood of exit in
all specifications. This means that elevators set up for large scale rail shipment
are less likely to exit and suggests that another avenue for elevator survival would
be to invest in rail car capacity. The regional demand and supply measures are as
expected. Elevators in regions with strong demand (large agricultural production)
are less likely to exit, while elevators in regions with considerable capacity are more
likely to exit. Finally, we introduced a measure of spatial competition and found it
has a positive effect on the likelihood of exit. Hence, elevators in areas with large
nearby competitors are more likely to exit the market.
The results in Table 3, Column 5 appear to be have the best overall results
based on the log-likelihood. In addition, generally, the coefficient estimates are
remarkably stable across specifications and consistent with prior expectations. The
25
overriding conclusion for the collective results is that the industry has transitioned
from one of small capacities and dated technologies to one of large elevator sizes
and more modern technologies. Elevators with a strong tie to transportation
(rail car capacity), in areas with strong demands (agricultural production) and
little capacity are more likely to remain in the market than those with limited
car loading capacity, those located in low demand, and high capacity regions. In
all models, elevators that existed at the beginning of the data (1999) were more
likely to exit than elevators that entered later. This is consistent with the findings
of T. Dunne et al. (2005). Finally, the role of spatial competitors in elevator exit
decisions makes intuitive sense and is underscored by the empirical results. In all
pertinent specifications, the effect of the inverse distance measure of competitor
capacity is statistically important and has a strong positive effect on the probability
of exit.
It is clear from the results in tables 2 and 3 whether an entrant has a strong
negative effect on exit. As discussed in section 3, entrants are more likely to be
more technologically advanced than non-entrants. In tables 2 and 3, we simply
have entrant or not. In tables 4 and 5, we replicate the results in table 2 and 3, but
introduce a measure of entrant age. All qualitative results remain the same and are
numerically similar. The effects of age are positive (as expected) and significant in
all specifications, suggesting that plants that enter later are less likely to exit the
market.
26
Table 4. Basic Model with Age Included
(1) (2) (3) (4)
exit exit exit exit
LOG CAP −0.942∗∗∗ −0.895∗∗∗ −1.023∗∗∗ −0.997∗∗∗
(0.048) (0.051) (0.052) (0.056)
Elev Ownership −0.358∗ −0.185 −0.293 −0.163
(0.204) (0.210) (0.216) (0.222)
Entrant −1.352∗∗∗ −1.017∗∗∗ −1.512∗∗∗ −1.931∗∗∗
(0.178) (0.182) (0.182) (0.187)
Age 0.00836 0.0478∗ 0.0235 0.0584∗∗
(0.026) (0.028) (0.027) (0.028)
Time Fixed Effect ✗ ✓ ✗ ✓
Subdivision Fixed Effect ✗ ✗ ✓ ✓
Constant 6.635∗∗∗ 5.978∗∗∗ 7.028∗∗∗ 6.523∗∗∗
(0.408) (0.428) (0.566) (0.590)
N 7662 7317 7660 7315
Log Likelihood −2552.77 −2374.41 −2488.97 −2322.83
standard errors in parentheses
∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
2.5 Conclusion
In Western Canada, elevators remain a key part of the modern grain
supply chain. But over the last few decades, the number of grain elevators on the
Prairies has fallen dramatically. In this paper, we develop and estimate a model of
elevator exit that encompasses relationships from the extant industrial organization
literature, but also measures of vertical linkages to transport markets and the
spatial setting of the elevator.
Our findings are largely consistent with prior literature in that we find
the likelihood of exit is negatively affected by elevator size and also whether a
given elevator was an entrant over the duration of the sample. But unlike prior
literature, we find mixed evidence that multi-plant (elevator) ownership have
a significant effect on exit decisions. To this basic specification, we also added
27
Table 5. Vertical Linkages and Spatial Competition Model Results with
Age included
(1) (2) (3) (4) (5)
exit exit exit exit exit
LOG CAP −0.668∗∗∗ −0.469∗∗∗ −0.491∗∗∗ −0.476∗∗∗ −0.333∗∗∗
(0.061) (0.061) (0.062) (0.062) (0.067)
Elev Ownership −0.0982 −0.363 −0.393∗ −0.317 −0.430∗
(0.221) (0.224) (0.226) (0.225) (0.236)
Entrant −1.061∗∗∗ −1.227∗∗∗ −1.280∗∗∗ −1.293∗∗∗ −1.129∗∗∗
(0.184) (0.188) (0.190) (0.191) (0.193)
LOG CAR −0.370∗∗∗ −0.387∗∗∗ −0.391∗∗∗ −0.380∗∗∗ −0.266∗∗∗
(0.062) (0.063) (0.063) (0.063) (0.067)
LOG(AG Production) −0.110∗∗∗ −0.116∗∗∗ −0.117∗∗∗ −0.275∗∗∗
(0.008) (0.008) (0.008) (0.019)
LOG(Subdiv CAP) 0.274∗∗∗ 0.234∗∗∗ 0.220∗∗∗
(0.052) (0.053) (0.054)
LOG(Weighted CAP 20 Mile) 1.883∗∗∗ 1.654∗∗∗
(0.458) (0.476)
LOG(Ag Production/Subdiv Capacity) −0.254∗∗∗
(0.027)
Age 0.0523∗ 0.0836∗∗∗ 0.0866∗∗∗ 0.0868∗∗∗ 0.0625∗∗
(0.028) (0.028) (0.028) (0.028) (0.029)
Constant 5.092∗∗∗ 4.910∗∗∗ 1.910∗∗∗ −14.46∗∗∗ −13.72∗∗∗
(0.437) (0.431) (0.717) (4.045) (4.194)
Time Fixed Effect ✓ ✓ ✓ ✓ ✓
N 7297 7297 7297 7297 7231
Log Likelihood -2339.00 -2240.39 -2225.65 -2217.17 -2136.15
standard errors in parentheses
∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
28
variables intended to capture vertical linkages to the freight transportation market
and to accommodate local demand and supply conditions (including spatial
competition). We find that our measure of vertical linkages has a strong negative
effect on the likelihood of exit, indicating this is still a factor critical for elevator
investment decisions. We also find that local measures of demand, supply and
spatial competition matter in the exit decision. Simply put, as local agricultural
production levels increase, spatial competition falls, and production relative to
capacity increases, we find elevators were less likely to exit.
As an important industry to the economy of Western Canada, grain
elevators have a long history, characterized by a series of ongoing changes to its
organization. Based upon relatively recent industry data, the focus of this analysis
was on obtaining a better understanding of grain company decisions about opening
or closing individual grain elevators. In spite of various factors unique to this
industry that have affected its evolution, our basic findings still strongly accord
with prior work on industrial exit. To this end and due to the inherently spatial
nature of grain elevator markets, we found that conditioning on market factors the
degree of localized spatial elevator competition was a significant determinant of
an exit decision. The latter raises concerns as per the continued evolution of the
industry with respect to potential mergers/consolidation and the degree of local
market power in various key production regions.
One factor that could not be accounted for in this analysis is the overall
effect that elevator closures have had on primary agriculture in Canada. For
example, in moving grain to their “local” elevator, Prairie farmers have seen this
average distance grow considerably over the past 30 years (J. Nolan, 2007). Aside
from the social costs of rural road damage (Larson & Nolan, 2007) and considering
29
the benefits of economies of scale, elevator exit and consolidation have shifted
some logistics costs over to individual farmers and has thus affected the long-term
sustainability and structure of Prairie agriculture.
Unlike its close counterpart in the U.S., grain elevators in Canada have
always been characterized by just a few co-operatives or private firms. With the
demise of the Canadian Wheat Board in 2012, along with climate change, the
lengthening of the crop season, more arable land and increased crop diversity in
the region, the Canadian Prairie elevator industry is poised to see major changes in
the near future. Our findings on exit decisions support the need for vigilance on the
part of competition authorities in shaping how the industry moves forward. This
must be done in order to avoid the monopolizing effects of economies of scale in
elevation coupled with the identified objective of minimizing spatial competition.
30
CHAPTER III
WATCHING THE GRASS GROW: DOES RECREATIONAL
CANNABIS LEGALIZATION AFFECT RETAIL AND
AGRICULTURAL WAGES?
The synthetic procedure described in this chapter was a collaborative work
where I did the data collection, curation, analysis, and visualization and Dr. Miller
and I wrote and edited the manuscript together.
3.1 Introduction
The long-standing landscape of cannabis prohibition is rapidly changing. In
the past decade, the median American voter moved from opposing to supporting
legalization (Motel, 2015), more than a dozen U.S. states legalized the substance for
adult use, and jurisdictions around the world loosened restrictions. One argument
employed by supporters of legalization is the assertion that policy liberalization
would lead to the creation of new jobs across multiple sectors (see e.g. Keys, 2020;
A. Wallace, 2020). Indeed, according to Statistics Canada, the industry generated
over 10,000 jobs within a year of Canada’s federal-level legalization, with average
hourly wages above the national average, and Barcott and Whitney (2019) estimate
that the U.S. cannabis industry (including both medical and adult-use cannabis)
directly employed more than 200,000 workers in 2019.
Cannabis, however, does not exist in a vacuum – the labor involved in
cannabis production and retail is similar to that involved in other agricultural
and retail markets and so cannabis legalization may induce workers to substitute
between employers. Indeed, farmers of other crops in many areas have expressed
concerns about the potential for upward pressure on agricultural labor wages as
a consequence of adult-use cannabis laws (RCLs) (Smith, Powell, Mungeam, &
31
Emmons, 2019; Stoicheff, 2018; Valachovic, Quinn-Davidson, Stackhouse, Butsic,
et al., 2019; Washburn, 2020). In this paper, we investigate these concerns by
measuring the impact of recreational cannabis legalization on wages using data
collected from the U.S. Census Bureau. We focus on Washington and Colorado
due to their early adoption of legalization policies and therefore the longest post-
legalization period during which to measure any changes in labor markets. We
focus on agricultural and retail labor markets as those are plausibly the most likely
to be affected by the opening of adult-use cannabis markets.
While this policy change may seem like a relatively clean quasi-experiment—
both Washington and Colorado legalized adult-use through ballot initiatives and
while the opportunity to generate tax revenue likely played a role in the success
of these efforts, it is unlikely that the timing of these ballot initiatives or their
implementation was driven by labor market conditions—and an opportunity for
a differences-in-differences approach, we must overcome a number of challenges.
The first is data-related: cannabis is not separately categorized by the North
American Industry Classification System (NAICS) and so we cannot measure the
level of employment in the cannabis industry directly, but must instead infer it
from changes in some larger category. Using data from the Quarterly Census of
Employment, we identify NAICS categories that experience changes in the number
of firms and employees that match state regulator data on cannabis firms. These
categories differ across states as a consequence of differing regulatory frameworks.
These data limitations create potential limitations in our ability to answer
questions: if we observe a large increase in wages in the NAICS categories which
contain cannabis firms, we cannot be certain that those higher wages are being
paid to other workers in those categories without either additional assumptions or
32
additional data. We address this in part by defining broader categories of retail
and agriculture firms over which cannabis firms play a small role; if we observe an
increase in wages in these broader categories, we can more reasonably conclude that
incumbent firms are paying higher wage bills.
Second, given the spillover effects of legalization efforts both in terms of
geography (Hansen, Miller, & Weber, 2020b) and in product space (K. Miller &
Seo, 2021), as well as the mobility of (particularly agricultural) labor (Holmes,
2013; Thomas-Lycklama-Nijeholt, 2012), it is difficult to choose an appropriate
control group a priori. We therefore follow Hansen, Miller, and Weber (2020a), who
study the impact of cannabis legalization on traffic fatalities, and use a synthetic
control approach. We create a control group by choosing weights for states without
legal cannabis markets to match moments characterizing each state in the pre-
legalization period. By comparing post-legalization employment and wages in
the treated states to their synthetic controls, we can estimate the causal impact
of legalization on these outcomes of interest.
Implementing this approach for the retail sector is relatively straightforward
– the elements of retail sectors which drive labor market outcomes (i.e. household
income and population density) do so in a consistent way across states (Blakely &
Leigh, 2013; Neumark, Zhang, & Ciccarella, 2008). Agricultural sectors in different
states, however, are significantly different due to variation in growing conditions
and the characteristics of arable land. While many detailed industry measures are
available, the set of measures changes frequently and often are not available for all
states. Faced with a need to both select variables and impute certain values, we
follow the approach of White, Reiter, and Petrin (2018) and implement machine
learning techniques to accomplish these tasks algorithmically. In particular, we
33
use LASSO for variable selection and classification and regression trees (CART) to
impute missing values.
Our primary finding is a null result: we find little evidence of a significant
difference in weekly wages per worker in the most directly substitutable NAICS
categories. Furthermore, though our estimates are noisier, we do not find evidence
of changes in weekly wages per worker in our broader definitions of the retail and
agricultural sectors.
This paper adds to the growing literature investigating the legalization of
cannabis for adult (recreational) use and its effects on outcomes thought to be
related to cannabis consumption. Smart and Pacula (2019) summarizes many of
the policy implications of cannabis legalization. Specific examples include studies
on student performance (A. M. Miller, Rosenman, & Cowan, 2017), traffic fatalities
(Aydelotte et al., 2017; Hansen, Miller, & Weber, 2020a), crime (Dragone, Prarolo,
Vanin, & Zanella, 2019; Hao & Cowan, 2020; Hughes, Schaible, & Jimmerson,
2020) and the consumption of other “sin” goods and cannabis substitutes (Baggio,
Chong, & Kwon, 2018; Chan, Burkhardt, & Flyr, 2020; Hansen, Miller, Seo, &
Weber, 2020; Kerr, Bae, Phibbs, & Kern, 2017; K. Miller & Seo, 2021).
Our analysis hinges on the assumption that labor supply conditions are
largely unaffected by cannabis legalization. Since all states which have legalized
cannabis for adult use have previously legalized cannabis for medical use, the
effects of both policies are relevant to our study. Ullman (2017) finds that medical
cannabis laws (MCLs) reduce the number of absences due to sickness, while Sabia
and Nguyen (2018) employ a synthetic control approach and find “no evidence that
[MCLs] affect employment, hours, or wages among working-age adults”. Nicholas
and Maclean (2019) find evidence that MCLs “lead to increases in older adult
34
labor supply, with effects concentrated on the intensive margin” and Ghimire
and Maclean (2020) provide evidence that workers’ compensation claims fall
following the adoption of MCLs. On the adult-use side, Maclean, Ghimire, and
Nicholas (2021) argue that RCLs increase Social Security disability claims, while
Abouk, Ghimire, Maclean, and Powell (2021) find that workers’ compensation
benefits decline after RCL adoption. Taken together, these results suggest that
our assumption is reasonable to a first-order approximation, though we discuss the
way in which increases in labor supply driven by RCL adoption would influence our
results in our conclusion.
More recently, the literature has begun to examine the cannabis industry
as an economic entity of interest in and of itself and as a tool to investigate long-
standing questions in industrial organization and policy design: Hansen, Miller, and
Weber (2017) investigate the impact of a change in Washington’s tax structure
throughout the cannabis supply chain, Thomas (2018) considers the effect of
Washington’s licensing quota system, Hollenbeck and Uetake (2019) estimate the
level and effects of market power in the industry, and Berger and Seegert (2020) use
the cannabis industry to analyze the effects of financial exclusion on firms.
Within the literature, the closest effort to that of our own is that of
Chakraborty, Doremus, and Stith (2020), who study the effects of Colorado’s
legalization on labor market outcomes at the county level exploiting the timing
of retail entry across counties. Ultimately, they find, as we do, that while the entry
of legal cannabis employers leads to increases in the number of employees in the
relevant sectors, the impact on equilibrium wages is approximately zero. Relative
to that work, we aggregate to the state level to avoid concerns about intra-state
labor mobility, use states without legal cannabis markets as the bases for synthetic
35
controls to avoid inter-state spillover effects, and add an additional treated unit
(Washington).
We proceed in Section 2 by describing labor in the cannabis industry relative
to other agricultural and retail industries. In Section 3, we describe our data
on labor market outcomes and our methodology. In Section 4, we present our
findings. We conclude in Section 5 with a discussion of the policy implications and
suggestions for future research.
3.2 Labor in the Cannabis Industry
Relative to many commodity agriculture crops such as corn and wheat,
cannabis production is labor intensive owing in large part of the dioecious nature
of plants in genus Cannabis. Buds with high concentrations of the psychoactive
cannabinoids tetrahydrocannabinol (THC) and cannabidiol (CBD) (among others)
are only produced by female plants prior to pollination (Chandra, Lata, Khan,
& ElSohly, 2017). Thus, in contrast to other dioecious agriculture operations,
such as fruiting trees where males are necessary for fruit production, cannabis
growers must identify and remove male cannabis plants from growing areas as even
a small number of male plants can provide pollen for an entire crop, triggering seed
production in females, a diminished set of flowers, and a corresponding reduction
in cannabinoid production. This labor is necessary even when farmers plant
“feminized” seeds or clones of female plants as the costs of a single male plant are
high enough that growers use labor resources to identify and destroy male buds
(see e.g. Schaneman, 2019). A relevant analogy in traditionally-legal agricultural
products is hops (Humulus lupulus); producers of hops remove male plants to
prevent pollination (Shepard, Parker, Darby, & Ainsworth, 1999).
36
The prevalence of indoor growing facilities complicates direct comparisons
between cannabis and other plants. According to an industry report, 60% of legal
producers operate indoor facilities, and 41% operate greenhouses – only 12% of
firms grow cannabis in the outdoors alone (Cannabis Business Times, 2020). The
use of indoor and greenhouse spaces allows for more precise control of the growing
environment, leading to more potent output (Aizpurua-Olaizola et al., 2016), and
enables production regardless of the outdoor agricultural season. However, the
amount of labor hours needed per pound produced is likely higher for indoor and
greenhouse operations than for outdoor operations (Caulkins, 2010).
After budding, plants must be harvested and trimmed of buds – a process
which takes four to six hours per pound manually (Cervantes, 2006). While
mechanized trimmers are available, hand-trimmers are able to extract higher
quality buds from plants which can command higher prices from consumers; the
majority of products sold to consumers (by revenue) consists of dried and cured
buds and thus the visual appearance of the buds is directly relevant to demand
(K. Miller & Seo, 2021). The remaining plant material undergoes extraction
processes to produce concentrate and edible products which are generally sold at
a lower price per weight of plant input. As a consequence, skilled trimmers can
earn more than twice the average hourly wage of other laborers in crop, nursery,
and greenhouse operations (Krissman, 2017).
These features of the cannabis industry imply that it is at least plausible
that a small number of cannabis producers (relative to the number of other
agricultural producers using greenhouses) could sufficiently impact the aggregate
demand for agricultural labor to significantly change equilibrium wages. However,
relative to other agricultural products, the market for cannabis labor is tightly
37
regulated. In each state with an operating recreational market, individuals must
pass a background check before working for a cannabis producer – and to pass that
check, the worker must have legal immigration status and (in most states) must not
have recent felony convictions related to Schedule I or Schedule II drugs. According
to the U.S. Department of Labor, approximately 47% of the U.S. agricultural labor
industry are undocumented immigrants, though agricultural industry sources
estimate the share is closer to 75% (Jordan, 2020). If the labor markets are
bifurcated due to immigration status, the effects of legalization on wages may be
minimal at best. Furthermore, as the highest wages available within the cannabis
industry are paid to workers with cannabis-specific skills, the substitutability of
that labor (and therefore the upwards pressure on equilibrium wages) may be
limited.
The process of retail sales of cannabis products also differ from most retail
businesses. In most jurisdictions, psychoactive cannabis inventory must be strictly
and securely separated from the sales floor, which is often required to be separated
from pedestrian access through secure doors so that customer ages can be verified
before entry. Inventory must be tracked in real-time for compliance with federal
guidelines and state seed-to-sale traceability regulations. Audits are frequent and
penalities for non-compliance include civil and criminal liability for firm owners and
managers (Hansen, Miller, & Weber, 2018). These additional layers of security and
related regulations imply that, relative to other retailers with similar footprints,
cannabis retailers may demand additional labor hours.
Finally, though Colorado and Washington set up recreational markets
in the same time period, the regulatory structures vary in ways relevant to our
analyses; see Hansen, Miller, and Weber (2021a) for more details about the
38
regulatory structures in the various states which have legalized cannabis for adult
use. First, while Washington required vertical separation between production and
retail, Colorado initially required retailers to produce 70% of the products they
sell through vertically integrated production facilities, often located close to the
retailer (Hansen, Miller, & Weber, 2021b). As a consequence, while firms in both
Washington and Colorado set up production operations, production facilities in
Washington, which were both more geographically dispersed and more specialized,
arguably competed more directly with other greenhouse agricultural facilities for
labor. Second, Colorado initially limited adult-use licenses to existing medical
dispensaries, which may limit the number of new establishments entering at the
time Colorado’s market opened. Finally, Colorado allows home cultivation, which
Washington bans. While this may affect demand for cannabis on the margin, we
note that that to-date, the cannabis industry in Colorado has generated more
revenue per resident than Washington’s industry.
3.3 Data and Methodology
We begin our analysis of the relationship between cannabis legalization and
labor market outcomes by obtaining labor market data from the Quarterly Census
of Employment and Wages compiled by the U.S. Bureau of Labor Statistics (BLS).
BLS categorizes employers according to the North American Industry Classification
System (NAICS) – a system of 2-6 digit codes which classifies employers in
narrowing groups according to their output or primary business activity. Our
outcomes of interest include the number of establishments, the total number of
workers, the total real wages, and the average weekly real wage per worker. We
collect these outcomes at the NAICS-state-quarter level from 2000-2019, aggregate
to the annual level, and deflate to 2019 dollars using the Consumer Price Index.
39
To capture time-varying characteristics of labor markets which may
influence outcomes, we collect demographic data from the U.S. Census Bureau and
Department of Education including state-level high school and college graduation
rates, population density, the aggregate unemployment rate, and per-capita GDP.
Agricultural labor markets differ widely from state to state due to differences in the
characteristics of arable land and growing seasons and therefore to capture other
time-varying characteristics of agricultural markets which may influence relevant
labor market outcomes, we additionally collect state-year-level survey data from
the National Agricultural Statistics Service from 2000-2015 and state-level data
from the U.S. Censuses of Agriculture for 2002, 2007, and 2012 (i.e. pre-treatment
covariates). A challenge we face in using this data is the prevalence of missing
values which stem in part from changes in the survey questions from year to year.
To create a panel data set for analysis, we focus on variables for which there are
at least 30 state-level observations per year. These variables largely sort into clear
topic areas: demographics, land statistics including rental prices, counts of farm
establishments, and variables capturing output for corn, wheat, hay, and fruits and
vegetables.
Despite this restriction, the data still contain many missing values
complicating any analysis effort. Following White, Reiter, and Petrin (2012); White
et al. (2018), we use the Van Buuren, Brand, Groothuis-Oudshoorn, and Rubin
(2006) modification of the Classification and Regression Trees (CART) algorithm to
impute missing values. The algorithm uses a Gibbs sampling procedure to generate
a plausible value for each missing value. Key to our application, the algorithm
uses “chained” imputation: for each unit of observation (i.e. each state-year
observation), the most recent generated imputation for each column is used as a
40
predictor for the next column to minimize bias (Michalowsky, Hoffmann, Kennedy,
& Xie, 2020; Murray & Reiter, 2016; Van Buuren & Groothuis-Oudshoorn, 2010).
In other words, suppose the vector of independent variables for observation t is
XT = [x1t, x2t, · · · ]. Suppose x1t is known for some t but x2t is missing. The
algorithm uses a Gibbs sampler to draw a value from x2t using the empirical
distribution of x2 conditional on x1t. Now suppose x3t is also missing for t. The
algorithm uses both the observed value x1t and the imputed x2t to draw a value
of the x3 distribution conditional on both x1 and x2. Ultimately, in our primary
specification, we impute 11% of the observation-variables for the agricultural
analysis and none of the observation-variables for the retail analysis. We have re-
estimated our models excluding imputed data and found similar results.
We next turn to the issue of variable selection. The number of potential
control units (i.e. states other than Washington and Colorado) is less than the
number of potential covariates. Instead of manually choosing covariates based on
some prior hypothesis, which may be considered “cherry picking” (Ferman, Pinto,
& Possebom, 2020), we use the LASSO algorithm to select appropriate covariates
(Duncan, Ross, & Mikesell, 2019; Tibshirani, 1996). For each outcome variable,
we fit prediction models to the pre-legalization data (i.e. data from 2000-2012)
using the glmnet method of Friedman, Hastie, and Tibshirani (2010) and select the
covariates with the highest frequency for each of the outcome variables.
The final covariate matrix X for our agricultural analysis includes “Barley
for grain (acres)”; “Land in orchards (acres)”; “Snap beans harvested for sale
(acres)”; “Cherries (acres)”; “Pears (acres)”; “Commercial fertilizer, lime, & soil
conditioners (acres treated)”; “2000 Resident population 65 years & over, percent”;
“2000 Savings institutions (FDIC-insured)-total deposits”; “2000 Civilian labor
41
force unemployment rate”; “Federal Government expenditure-grants FY 2000”;
“Federal Government insurance FY 2000”; “2000 Resident population: Black alone,
percent”; “2000 Resident population: Two or more races, percent”; “2000 Resident
population: Hispanic or Latino Origin, percent”; “2000 Resident population: total
females, percent”; “Social security: retired workers-benefit recipients (Dec.) 2000”;
“Corn grain production”; “Farm operations”; “Hay production”; “Labor hired
wage rate ($ per hour)”; “Rent cash cropland expense ($ per acre)”; “Vegetable
total production”; and “Wheat production”. For our retail analyses, the covariate
matrix includes “College Graduation Rate (percent)”; “High School Graduation
Rate (percent)”; “Population Density (people per square mile)”; “Unemployment
Rate (percent)”; and “GDP per capita”. We also include the relevant outcome for
stores in NAICS 453991 (Tobacco stores).
The agricultural census data is collected every five years – the last collection
was in 2017. At the time of the last collection, only four states – Alaska, Colorado,
Oregon, and Washington – had legalization cannabis for recreational use, and
within those states, Colorado and Washington legalized earliest (voting in 2012,
markets opening in 2014). To focus on the longest post-legalization period
possible, we follow Hansen, Miller, and Weber (2020a) and focus on Colorado
and Washington as the treated states. We further note that both Oregon and
Alaska experienced significant supply issues in months immediately post market-
opening (Andrews, 2017; Sacirbey, 2016) and thus any impact on agricultural
labor is potentially more difficult to observe and/or interpret from the short post-
legalization period available.
Figure 3 plots outcomes by year for Colorado, Washington, and the average
of other states for the “greenhouse, nursery, and floriculture production” category
42
(NAICS 1114, the category containing cannabis production firms). Notably, the
establishment count for Washington increased by roughly 500 between legalization
and a peak in late 2015, which is similar to the count of cannabis production
licenses issued by the state around the same time period as reported by Hansen
et al. (2017). Washington experienced a similarly-shaped increase in the number
of workers in the sector and the total wages paid, but those outcomes in Colorado
and other states remained largely constant. Despite the increase in labor quantity
observed in Washington, the real average weekly wage per week increased after
legalization relatively uniformly everywhere.
Figure 4 reports analogous outcomes in the “store retailers not specified
elsewhere” category (NAICS 453998, the category containing cannabis retailers).
As with the agricultural sector, the establishment count in Washington increased
by several hundred immediately post-legalization corresponding to descriptive
statistics found in the literature (Thomas, 2018). Colorado also experienced an
increase of roughly 200 establishments over the same time period. Increases of
similar magnitude occured for worker counts and total wages paid in conjunction
with the opening of these establishments. As in the agricultural sector, however,
there are no clear patterns in the average weekly wage per worker; while the mean
post-reform wage in Colorado is above the mean pre-reform wage, wages had begun
increasing in the years prior to the passage of the ballot measure.
While the raw data suggest that the legalization of cannabis led to
significant changes in employment in each state corresponding to their different
regulatory structures, it is not clear that cannabis legalization caused these
changes. Estimating a causal effect requires identifying an appropriate set of
control units. While neighboring states might seem like a natural control group,
43
Figure 3. Employment and wages for “narrowly defined” agricultural
firms
A B
Number of Establishments Number of Workers
10000
750 8000
6000
500
4000
250
2000
2000 2005 2010 2015 2000 2005 2010 2015
Year Year
C D
Real Total Quarterly Wage Real Average Weekly Wage Per Worker
700
60 650
600
40
550
20
2000 2005 2010 2015 2000 2005 2010 2015
Year Year
Colorado Washington Other States
Notes: Data come from the Quarterly Census of Employment and Wages. We define “narrowly
defined” agricultural firms as those within North American Industry Classification System
category 1114 (“Greenhouse and Nursery Production”), which includes cannabis production
firms.
44
Dollar (in Millions) Count
Dollar Count
Figure 4. Employment and wages for “narrowly defined” retail firms
A B
Number of Establishments Number of Workers
8000
800
700
6000
600
500 4000
400
2000
300
2000 2005 2010 2015 2000 2005 2010 2015
Year Year
C D
Real Total Quarterly Wage Real Average Weekly Wage Per Worker
680
60
50
640
40
30 600
20
10 560
2000 2005 2010 2015 2000 2005 2010 2015
Year Year
Colorado Washington Other States
Notes: Data come from the Quarterly Census of Employment and Wages. We define “narrowly
defined” retail firms as those within North American Industry Classification System category
453998 (“Store retailers not specified elsewhere”), which includes cannabis retailers.
45
Dollar (in Millions) Count
Dollar Count
Hansen, Miller, and Weber (2020b) find evidence of substantial inter-state cannabis
demand, and it is reasonable to believe that laborers may also move across state
lines in response to cannabis legalization, particularly if cannabis producers are
indeed offering higher wages. This is a particular concern for Washington, where
many retailers are located close to the Oregon and Idaho borders.
To address this concern, we apply the synthetic control approach of Abadie,
Diamond, and Hainmueller (2010, 2015); Abadie and Gardeazabal (2003). We
construct synthetic control units separately for Washington and Colorado based
on pre-legalization data (i.e. the covariates listed above plus the lagged value of
the outcome variable) and then estimate the effect of cannabis legalization on
our outcomes of interest by calculating the post-legalization difference between
the outcomes for our treated states and for our synthetic controls. Our synthetic
control units are convex combinations of non-treated states selected in such a way
to match the pre-legalization outcomes. In addition to previous work on cannabis
legalization and traffic fatalities (Hansen, Miller, & Weber, 2020a), the synthetic
control approach has been used to analyze the effects of policy changes across
a variety of domains, including economic liberalization (Billmeier & Nannicini,
2013), pediatric health (Bauhoff, 2014), tropical deforestation (Sills et al., 2015),
foreign exchange rates (Chamon, Garcia, & Souza, 2017), tobacco policies (Chelwa,
van Walbeek, & Blecher, 2017), and the effects of medical cannabis laws on labor
market outcomes (Sabia & Nguyen, 2018) among many others.
We first select a “donor pool” of control units (i.e. states) which may be
used to construct the synthetic control units. We start with all U.S. states and
exclude any states which legalized cannabis and opened adult-use markets after
2012. We include Michigan as its first dispensary opened in December 2019, and
46
thus any labor market effects are unlikely to be observed in annualized 2019 data.
We also exclude states which are adjacent to the treated states to avoid spillover
effects. While we present results using a donor pool which includes both states with
and without legal medical cannabis markets, we have estimated separate models
using only states with or states without these markets and found similar results.
For each treated unit s ∈ {Washington, Colorado}, w∑e then select weights
wj for each of the control units j (with 0 ≤ wj ≤ 1 and wj = 1) to
minimize the weighted difference between the synthetic control and the treated
unit on the pre-treatment covariates identified above. The weight matrix V used
to form the distance measure is chosen such that the mean square prediction
error is minimized for the pre-intervention period following Abadie et al. (2010).
We report the weights W ∗ chosen for each treated unit and outcome variable in
Appendix A. Tables of covariate balance are available in Appendix B. We then
obtain point estimates of the effect of recreational cannabis legalization with a
standard differences-in-differences estimating equation. For outcome y for unit s
(either a treated state or the synthetic control for that state) in year t, we estimate
the parameters of
yst = β0 + β1 ∗ Legalt + β2 ∗ Treatedt + β3 ∗ Legalt ∗ Treatedt + ϵst. (3.1)
To perform hypothesis testing, we use the “in-space” placebo tests described
in Abadie et al. (2015). In particular, we apply the synthetic control model to each
of our potential control units and interpret the results as placebos. We remove a
small number of control states with particularly poor pre-treatment fit, though this
does not affect our qualitative results. Plots of these placebos are available in the
Appendix. For each outcome Y (and corresponding sequence of state-year outcome
observations Yjt), we then calculate the empirical distribution of the ratio of the
47
mean squared prediction errors(RMSPE∑(
) where
∑ ) 2 1/2T0 J+11
RMSPE =  Y − w∗ 1t
T j
Yjt (3.2)
0 t=1 j=2
and T0 is the positive number of pre-intervention periods. The p-value is then
simply the fraction of placebo effect estimates which are greater than or equal to
the effect estimated for the tr∑eated unit (Firpo & Possebom, 2016):J+1
j=1 1 [RMSPEj ⩾ RMSPE1]
p :=
J + 1
Finally, it is plausible that, from the perspective of workers, jobs in the
cannabis industry are substitutes for jobs beyond the narrowly-defined NAICS
categories described above. We repeat this analysis for a broader set of categories
taking advantage of the hierarchical nature of the NAICS inclusive of cannabis
firms; for agriculture, we use “agriculture, forestry, fishing, and hunting” (NAICS
11) and for retail, we aggregate the “health and personal care stores” (NAICS
446), “general merchandise stores” (NAICS 452) and “miscellaneous store retailers”
(NAICS 453) categories.
3.4 Results
3.4.1 Narrowly-defined industries. Figure 5 illustrates agricultural
labor market outcome measures in Colorado and its synthetic control unit (control
weights are reported in Table B.2.1) for the “greenhouse, nursery, and floriculture
production” NAICS category. Following Figure 3, Panel (a) illustrates the log
of the number of establishments, Panel (b) illustrates the log of the number of
worker, Panel (c) illustrates the log of the real total quarterly wage, and Panel (d)
illustrates the log of the real average weekly wage. In general, the synthetic control
closely follows both the trends and the level of Colorado’s outcomes over the pre-
legalization period. In the post-legalization period, the number of establishments
48
temporarily grows relative to its synthetic control, but the number of workers
tracks closely with its synthetic control, as do wages.
Figure 6 illustrates the analogous comparisons for Washington. As in
Colorado, the synthetic control tracks closely with the Washington data in the pre-
legalization period. However, the number of establishments increases significantly
immediately after legalization, as does the number of works and (as a consequence),
the total quarterly wages paid. Though the average weekly wage in Washington
does increase post-legalization, the increase is also seen in the synthetic control.
Figures 7 and 8 repeat the exercise for outcomes for the “store retailers not
specified elsewhere” NAICS category in Colorado and Washington, respectively.
For Colorado, the synthetic control approach struggles to match the full volatility
of the pre-reform data for the number of establishments and the number of
workers. However, the method performs better (in a mean-squared-error sense)
when matching per-reform average weekly wages per worker. Across outcomes,
the synthetic control generally moves in the same direction as the Colorado data
post reform, suggesting that other trends in Colorado contributed to the increase
in establishments and workers seen in Figure 4. The synthetic control approach
performs better for Washington, where pre-trends are closely matched for most
outcomes.
Point estimates of the effects seen in these Figures (i.e. estimates of β3 in
Equation (3.1)) are reported in Table 6. Several of the changes in the number of
establishments, employees, and total wages are significant according to our placebo
test at the 10% and 5% levels. However, the change in average weekly wage is
either imprecisely estimated or negative for both sectors in both states.
49
Figure 5. Comparing “narrowly defined” agricultural labor market
outcomes in Colorado and its synthetic control
A B
Log Number of Establishments Log Number of Workers
8.1
5.10
8.0
5.05
7.9
5.00
7.8
4.95
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
6.55
17.1
17.0 6.50
16.9
6.45
16.8
6.40
16.7
2005 2010 2015 2005 2010 2015
Year Year
Colorado Synthetic
Notes: This figure depicts wage and employment outcomes for “narrowly defined” agricultural
firms for Colorado and its synthetic control. We define “narrowly defined” agricultural firms as
those within North American Industry Classification System category 1114 (“Greenhouse and
Nursery Production”), which includes cannabis production firms.
50
Figure 6. Comparing “narrowly defined” agriculture labor market
outcomes in Washington and its synthetic control
A B
Log Number of Establishments Log Number of Workers
6.6 9.00
6.3
8.75
6.0
8.50
5.7
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
18.00 6.40
17.75 6.35
6.30
17.50
6.25
17.25
2005 2010 2015 2005 2010 2015
Year Year
Washington Synthetic
Notes: This figure depicts wage and employment outcomes for “narrowly defined” agricultural
firms for Washington and its synthetic control. We define “narrowly defined” agricultural firms as
those within North American Industry Classification System category 1114 (“Greenhouse and
Nursery Production”), which includes cannabis production firms.
51
Figure 7. Comparing “narrowly defined” retail labor market outcomes
in Colorado and its synthetic control
A B
Log Number of Establishments Log Number of Workers
6.3
8.25
6.2
6.1 8.00
6.0
7.75
5.9
7.50
5.8
5.7
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
6.48
17.25
6.44
17.00
6.40
16.75
16.50 6.36
2005 2010 2015 2005 2010 2015
Year Year
Colorado Synthetic
Notes: This figure depicts wage and employment outcomes for “narrowly defined” retail firms
for Colorado and its synthetic control. We define “narrowly defined” retail firms as those within
North American Industry Classification System category 453998 (“Store retailers not specified
elsewhere”), which includes cannabis retailers.
52
Figure 8. Comparing “narrowly defined” retail labor market outcomes
in Washington and its synthetic control
A B
Log Number of Establishments Log Number of Workers
9.00
6.7
8.75
6.5
8.50
6.3
8.25
6.1 8.00
7.75
5.9
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
18.0 6.52
17.7 6.48
17.4 6.44
17.1 6.40
16.8 6.36
2005 2010 2015 2005 2010 2015
Year Year
Washington Synthetic
Notes: This figure depicts wage and employment outcomes for “narrowly defined” retail firms for
Washington and its synthetic control. We define “narrowly defined” retail firms as those within
North American Industry Classification System category 453998 (“Store retailers not specified
elsewhere”), which includes cannabis retailers.
53
Table 6. Synthetic control estimates of the effect of recreational cannabis
legalization on narrowly-defined labor market outcomes
Colorado
Log number Log number Log total Log
establishments of employees quarterly wages weekly wage
Narrowly-defined Agriculture
RCL 0.056** -0.042 -0.113 -0.007
P-value [0.030] [0.303] [0.303] [0.576]
Narrowly-defined Retail
RCL 0.000 0.220 0.306* 0.050
P-value [0.818] [0.152] [0.091] [0.212]
Washington
Log number Log number Log total Log
establishments of employees quarterly wages weekly wage
Narrowly-defined Agriculture
RCL 0.783* 0.516* 0.513** -0.013*
P-value [0.061] [0.061] [0.030] [0.091]
Narrowly-defined Retail
RCL 0.063 0.535** 0.525** 0.059
P-value [0.152] [0.030] [0.030] [0.606]
Notes: This table reports difference-in-difference estimates of the effect of recreational cannabis
legalization (RCL) on labor market outcomes using synthetic controls for the treated units.
Agriculture is the “Greenhouse and Nursery Production” (NAICS 1114) industry. Retail is the
“Store retailers not specified elsewhere” category (NAICS 453998). P-values are calculated via a
placebo test. Stars indicate standard significance levels: ∗10%, ∗∗5%, ∗∗∗1%.
54
3.4.2 Broadly-defined industries. While the above results verify
that the legalization of cannabis led to changes in the number of establishments
and employees working in the categories which contain cannabis firms, they provide
no evidence that legalization led to wage spillovers. Indeed, there is little evidence
that legalization affected the Colorado labor market at all. One possibility is that
although cannabis production facilities are coded as members of the green house
and nursery sector, cannabis facilities do not compete with other members of that
sector for labor. To explore this possibility, we first repeat the analysis for NAICS
11, which includes all “agriculture, forestry, fishing, and hunting” firms.
The results are reported in Table 7 under the headings for “Broadly-defined
Agriculture” – the relevant Figures are available in the Appendix. It is important
to note that the pre-treatment fit for Washington is generally poor. Ferman (2021)
shows that the synthetic control model can be asymptotically unbiased even
when the pre-treatment fit is imperfect. Relative to Table 6, the estimates for
Washington are generally attenuated and more noisily estimated. For Colorado, the
estimates indicate small and marginally significant increases in employees and total
wages, though once again for both states there is no increase in average weekly
wages.
The difference in results between Colorado and Washington is potentially
driven by the vertical integration requirement in Colorado and the vertical dis-
integration requirement in Washington. In particular, firms in Colorado may
classify themselves completely as cannabis retailers, as opposed to cannabis
producers. While it is unlikely that these firms would compete with other
agriculture firms for labor (and indeed even if firms are classified in this way, we see
no effect on agricultural wages in Tables 6 and 7), it is possible that firms organized
55
Table 7. Synthetic control estimates of the effect of recreational cannabis
legalization on broadly-defined labor market outcomes
Colorado
Log number Log number Log total Log
establishments of employees quarterly wages weekly wage
Broadly-defined Agriculture
RCL 0.008* 0.108** 0.064** 0.007
P-value [0.091] [0.030] [0.029] [0.242]
Broadly-defined Retail
RCL -0.044 0.035** 0.055** 0.015*
P-value [0.576] [0.030] [0.030] [0.061]
Washington
Log number Log number Log total Log
establishments of employees quarterly wages weekly wage
Broadly-defined Agriculture
RCL -0.021 0.312 0.369 -0.154
P-value [0.091] [0.333] [0.242] [0.667]
Broadly-defined Retail
RCL 0.013 0.112 0.147* 0.014
P-value [0.121] [0.121] [0.061] [0.333]
Notes: This table reports difference-in-difference estimates of the effect of recreational cannabis
legalization (RCL) on labor market outcomes using synthetic controls for the treated units.
Broadly-defined Agriculture is the “Agriculture, Forestry, Fishing, and Hunting” (NAICS 11)
industry. Broadly-defined Retail is the combination of ‘NAICS 446 Health and personal care
stores’, ‘NAICS 452 General merchandise stores’, and ‘NAICS 453 Miscellaneous store retailers’.
P-values are calculated via a placebo test. Stars indicate standard significance levels: ∗10%, ∗∗5%,
∗∗∗1%.
56
in this way have an effect on wages paid in the retail sector. We thus repeat the
analysis once more for firms in related NAICS retail categories 446, 452, and 453.
The results are reported in Table 7 under the heading “Broadly-defined Retail.” As
expected, the estimates are attenuated from the more narrowly defined category.
We find limited evidence to support the hypothesis that weekly per-worker wages
increased in Colorado (the point estimate of a 1.5% increase is significant at the
10% level) and no evidence to support such a hypothesis in Washington.
3.4.3 Robustness. In Table 8 we explore three alternative
specifications, focusing on our primary outcome of average weekly wages per
worker. In Column (2), we include only states with medical cannabis systems in our
donor pool. In Column (3), we include only states with full prohibition of cannabis
throughout our study period in our donor pool; the small number of states in this
category limits the available inference. In Column (4) we follow the suggestion of
Ferman and Pinto (2021) and repeat the analysis in levels while demeaning the
outcomes. We do find potential evidence of a small increase in wages per worker in
Washington in the broad retail category, though in context of the remainder of our
estimates this is likely spurious.
3.5 Conclusion
Over the past decade, U.S. voters have undergone a rapid shift towards
supporting the legalization of cannabis in some form and policy has changed to
follow this support. These changes, however, have not come without frictions
generated by broad society-wide concerns about (among other issues) public
health and safety (Hall & Lynskey, 2016; Kilmer, 2019), educational outcomes (van
Ours & Williams, 2015), and interactions with other substances (K. Miller & Seo,
2021). Other frictions have been caused by more immediate financial concerns:
57
Table 8. Results from alternative specifications of weekly wages per
worker
(1) (2) (3) (4)
Baseline Med. cannabis Illegal In levels,
controls only controls only demeaned
Colorado
Narrow Agriculture -0.007 0.094* 0.077 14.22*
[0.576] [0.069] [0.2] [0.091]
Narrow Retail 0.050 -0.094 -0.044 -42.62
[0.212] [0.897] [0.8] [0.879]
Broad Agriculture 0.007 -0.010 0.055 -27.22
[0.242] [0.548] [0.4] [0.697]
Broad Retail 0.015* -0.033 -0.063 -291.63
[0.061] [0.586] [0.4] [0.818]
Washington
Narrow Agriculture -0.013* 0.748* 0.695 397.8*
[0.091] [0.069] [0.4] [0.091]
Narrow Retail 0.059 0.158 0.198 -178.2
[0.606] [0.103] [0.2] [0.121]
Broad Agriculture -0.154 -0.050 -0.020 -1200.23
[0.667] [0.419] [0.6] [0.697]
Broad Retail 0.014 0.025 0.012 52.10**
[0.333] [0.103] [0.4] [0.030]
Notes: Narrow agriculture is NAICS 1114, narrow retail is NAICS 453998, broad agriculture is
NAICS 11, broad retail is NAICS 446, 452, and 453. P-values in brackets are calculated via a
placebo test. Column (1) repeats results from Tables 6 and 7. In Column (2) we restrict the set of
potential donor states to those with medical cannabis regimes. In Column (3) we restrict the set
of potential donor states to those with full prohibition of cannabis throughout our study period.
In Column (4) we use the level of average wages per worker per week (as opposed to the log wage)
and demean the outcomes. Stars indicate significance levels: *10%, **5%, ***1
58
agricultural firms in areas with legal cannabis production have expressed concerns
about upward wage pressures leading to reduced international competitiveness and
domestic agricultural output. Indeed, Bampasidou and Salassi (2019) identify a
number of instances of labor shortages in particular U.S. agricultural industries and
regions around the time of the first successful cannabis legalization campaigns. At
the same time, supporters of legalization have pointed to substantial employment
within the nascent industry as a sign of success. Taken together, it is natural to
suggest that cannabis legalization may be contributing to a highly competitive
labor market from the perspective of agricultural employers.
We investigate the relationship between cannabis legalization and labor
market outcomes across both the agricultural and retail sectors. Using a synthetic
control approach paired with machine learning techniques including LASSO to
select appropriate covariates on which to generate synthetic control units and
CART for chained imputation of missing values, we ask whether equilibrium wages
increased after legalization in Washington and Colorado, the first states to legalize.
We find limited evidence to support this assertion; while the number of workers
in the relevant sectors increased following the entry of cannabis producers and
retailers, the wage per worker remained effectively constant.
Our results indicate that cannabis is not likely to be responsible for the
broader changes in the agricultural or retail labor markets experienced during
our study period. Indeed, others have pointed to changes in immigration policy
including an increase in the intensity of enforcement (Escalante & Luo, 2017) and
frictions in the H-2A guest worker program (Luckstead & Devadoss, 2019) as key
contributing factors to changes in agricultural labor markets. On the retail side,
aggregation in brick-and-mortar retailers (Neumark et al., 2008) and the increase
59
in online shopping (Bram & Gorton, 2017) have been identified as key drivers of
changes in retail employment outcomes. Relative to these broader labor market
trends, cannabis legalization may well be the proverbial “drop in the bucket”.
At the same time, results from studies of MCLs suggest that increasing cannabis
access may increase labor supply, though results from RCLs to this point have
been mixed. If RCLs do increase labor supply, our null result could be explained
by offsetting changes on the demand and supply side of the labor market. It is also
possible that our results could be explained by the conversion of illegal production
to legal production with minimal changes in the labor force (i.e. those who were
engaged in illegal production became those employed by legal producers). More
generally, if cannabis employment is particularly attractive to individuals who were
not previously engaged in the labor market (including those who were unemployed
or self-employed), our null result may well be expected.
These results are subject to a number of limitations which may be addressed
by future work. While we have focused on the labor market motivated by anecdotal
reports and popular press accounts, it is possible that the entry of adult-use
cannabis firms may affect incumbent firms in the agricultural and retail sectors
through other channels, such as competition for desirable real estate or within
the product market. Our work is necessarily limited to a relatively short post-
legalization period, and as cannabis production continues to grow, it is possible
that other agricultural and retail firms may face competition from cannabis firms
that differs from past experience. While many states have adopted regulatory
frameworks similar to either Colorado’s or Washington’s, the details vary widely
across dimensions including the number of licensed establishments, tax rates and
licensing fees, quantity and potency limits, and out-of-state investment rules,
60
amongst others (Hansen et al., 2021a). These differences may affect the cannabis
industry’s aggregate demand for labor across states and therefore the experience of
agricultural and retail incumbents. Indeed, both Colorado and Washington allow
counties and municipalities to ban entry by cannabis firms, and so there may be
within-state heterogeneity. Finally, both Colorado and Washington had existing
medical cannabis systems before opening their recreational markets. Our results
therefore speak only to the incremental effect of recreational legalization; a state
moving from full prohibition to a fully-legal regime may experience a larger effect.
Our study may give policymakers currently considering cannabis
liberalization some indication that such a policy change is unlikely to significantly
increase wage bills for existing retailers and agricultural firms in the short term.
Indeed, legalization is likely to improve labor market outcomes for job-seekers, if
only by slightly increasing demand for labor—though long-term cannabis use may
affect labor market outcomes at the individual level (Sabia & Nguyen, 2018).
61
CHAPTER IV
ARE FIRMS’ COST PREDICTIONS ACCURATE? EVIDENCE FROM
MEDICARE ADVANTAGE
The synthetic procedure described in this chapter was a collaborative work
where I did the data collection, curation, analysis, and visualization and Dr. Miller
and I wrote and edited the manuscript together.
4.1 Introduction
How well can firms predict their costs? As costs play a key role in decisions
about prices, benefits, and the adoption of innovative technologies (Baicker &
Goldman, 2011; Dranove & Satterthwaite, 2000; Jena & Philipson, 2008), firms
which consistently forecast their own costs inaccurately may make decisions
that are consistently suboptimal ex post for both themselves and for their
customers. While it is common knowledge that firms use the data they collect
on their customers and specific market environment to make forecasts, it is
difficult to evaluate performance on this dimension as forecasts and resulting
performance evaluations are generally considered highly valuable trade secrets.
As a consequence, previous empirical work on firm learning has largely focused
on macroeconomic conditions. For instance, Coibion, Gorodnichenko, and Kumar
(2018) use survey data from New Zealand and find that firms share a pervasive
inattention to macroeconomics and have a wide set of beliefs especially regarding
inflation; competition provides incentives to collect more accurate macroeconomic
information.
We examine forecasts made by insurers participating in the Medicare
Advantage (MA) market in the United States. MA is a “managed competition”
(K. S. Miller, Petrin, Town, & Chernew, 2021) environment in which the
government provides subsidies to firms who then compete with each other to
provide health insurance plans to seniors that replace traditional fee-for-service
Medicare benefits (TM) with (generally speaking) privately-administered managed
62
care benefits. MA is substantial in its own right; the federal government spends
more than $125 billion annually on payments to MA insurers in addition to the
premiums paid by enrollees. As part of the regulatory “bidding” process detailed in
Section 4.2, firms report detailed data on realized past medical costs and forecasted
future costs across a number of categories. Firms then choose cost-sharing product
characteristics relevant to those categories. Crucially, firms are incentivized to
report accurate forecasts (i.e. the best forecasts they have) by the bidding system,
as the forecasts constrain the set of possible products the firm may choose to offer.
If a firm reports that it expects costs to be higher than it truly believes, it may be
forced to offer an inferior product (i.e. higher copays and fewer benefits) than is
optimal given its true beliefs.
We begin by combining the reported costs and forecasts of future costs to
construct forecast errors. We construct the error both in total medical expenses
and in the three largest individual categories: inpatient facility, surgery, and
professional services. We show that forecast errors in these categories are smaller
than the overall error but are positively correlated with each other. Thus, for a
given plan-year, firms tend to “miss” in the same direction across service categories.
We then document changes in the average forecast error as firms gain experience in
the market and explore heterogeneity across firm experience (as measured by the
number of enrollees in the firm’s plans) and competitive environment. Using simple
models of the prediction error, we find that the magnitude of the forecast error
decreases as the firm gains experience and competitors, consistent with theories of
learning.
Finally, we examine the relationship between these forecast errors
and product characteristics. We show that, after accounting for time- and
location-specific factors, firms which under-predict their costs offer plans with
greater patient cost-sharing—i.e. products that are ‘worse’ from the consumer’s
perspective.
63
We contribute to an extensive literature studying firm learning and the
evolution of equilibria over time (Börgers & Sarin, 1997; Fudenberg & Kreps,
1993; Hart & Mas-Colell, 2000; Milgrom & Roberts, 1991). This literature has
branched in several directions, including using lab experiments to test and verify
theoretical models (see e.g. Erev & Roth, 1998) and increasing the sophistication
of the modelled learning process (see e.g. Camerer, Ho, & Chong, 2002; Crawford,
2003). Such refinements include adding a belief elicitation process into the model;
Rutström and Wilcox (2009) find that eliciting beliefs can change games and
that there is a significant difference between the strong-elicitation treatment, but
not between the no-elicitation treatment and the less intrusive weak-elicitation
procedure. In addition to the empirical work citied above, others have also studied
firm learning and adaptation. Graham, Harvey, and Rajgopal (2005) find that
upper managers main focus lies on meeting or beating earning benchmarks;
specifically the quarterly earnings for the same quarter last year and the analyst
consensus estimate. Most managers prefer to have smooth earnings and to maintain
the stability of future predictions. Their information disclosure coincides with the
effort to stability where they strive for clarity but also select news in a strategical
manner. Doraszelski and Markovich (2007) use frequency response data from the
new UK electricity system to see how firms compete and respond with prices and
how the price converge to its equilibrium using an adaptive learning model. They
find that especially during the middle phase, the best-fitting models are those in
which firms more heavily weight recent rival behavior in forming beliefs about their
rivals’ bids and adaptively learn about the price elasticity of demand.
This work also builds upon a literature examining the behavior of firms
and seniors within the MA system. Much of the work on the bidding system has
focused on how changes in the subsidy rates offered by the government passes
through to benefits (see e.g. Cabral, Geruso, & Mahoney, 2018; Duggan, Starc,
& Vabson, 2016; Song, Landrum, & Chernew, 2013). We complement that work
64
by investigating other potential drivers of firm behavior. Other work has focused
on patient outcomes with a particular focus on comparisons between MA and the
traditional Medicare system (see e.g. Figueroa et al., 2020; Meyers, Trivedi, Wilson,
Mor, & Rahman, 2021; Park, White, Fishman, Larson, & Coe, 2020). We shed
additional light on the way in which MA firms play a role in seniors’ health.
The remainder of this paper is organized as follows: Section 2 contains some
general background of the MA market and its bidding system; Section 3 describes
the data and methodology of our practice; Section 4 presents the results in two
different settings and then discusses our findings; Section 5 concludes.
4.2 Bidding in Medicare Advantage
Medicare Advantage (MA) was developed as a response to rising costs in
the Traditional Medicare (TM) system.1Under TM, the government pays service
providers according to a fee-for-service (FFS) schedule. Under MA, the government
pays insurers a per-enrollee subsidy that is adjusted for observable risk factors (i.e.
demographic characteristics and diagnoses) but that does not vary by realized
medical expenditures. The enrollee experience is substantially different: TM
enrollees may generally choose any provider for any service, whereas MA enrollees
generally face restricted provider networks and referral requirements for specialists.
Firms offer plans on an annual basis. The bidding process for plans which
will cover medical expenses2 in year t begins in the spring of the previous year.
First, the Centers for Medicare and Medicaid Services (CMS) releases ‘benchmark’
subsidy rates—the per-enrollee subsidy that CMS will pay firms for an ‘average’
1One of the biggest goals to implement MA is to increase efficiencies of managed care and to
save money: specifically to minimize the inefficiencies induced by the inevitable errors in TM’s
administered price system, by allowing the health plans and providers to negotiate prices or, in
some cases, to integrate the finance and delivery functions (McGuire, Newhouse, & Sinaiko, 2011).
2These bids are based on the plans’ estimates of the cost of providing required Medicare Part
A, which covers most medically necessary hospital; skilled nursing facility; home health and
hospice care; and Part B services, which cover medically necessary services by providers and other
services deemed medically necessary, to cover an average beneficiary (Berenson, Sunshine, Helms,
& Lawton, 2015).
65
risk enrollee—that vary at the county level. Firms then submit detailed proposals
to provide MA plans. These ‘bids’ begin with detailed information about the firm’s
realized medical expenses per-member-per-month in year t − 1 across more than
15 categories. Firms then report their projected expenses for each category for
year t using the ‘Bid Pricing Tool’ released by CMS.3 These estimated costs are to
provide the minimal number of required benefits for an average mix of risks against
the county-level benchmark. If a plan’s bid was greater than the benchmark, it was
required to collect the difference from its enrollees through a monthly premium. If
it was lower, 75 percent of the difference was to be returned to enrollees in the form
of supplemental coverage or lower premiums, and in an effort to reduce Medicare’s
obligations, the remaining 25 percent of the savings was to be returned to the
Medicare program (McGuire et al., 2011). Thus, it is important for plan providers
to have a reasonable estimate to plan ahead and hedge against any uncertainties for
MA enrollees.
Figure C.1 illustrates an example of these submissions for a plan offered
for the 2015 benefit year as entered into the Bid Pricing Tool in 2014. The left-
hand columns represent the plan’s realized costs for the 2013 plan year, and the
right-hand columns represent the firm’s projections of their costs for the 2015 plan
year. This plan added dental coverage between 2013 and 2015, and so the firm did
not have previous data on dental costs for this particular plan with which to form
projections. Ultimately, this firm projected that the ‘per-member per-month’ cost
of this plan would increase by $39.65 from 2013 to 2015, or 7.4% of the realized
costs in 2013. In dollars, this increase is driven mostly by projected increases in
inpatient and skilled nursing facility costs, as well as professional service costs.
A key question with respect to any reported firm projections, particularly
those which the firm knows will eventually be released to the public, concerns the
extent to which these reports reflect the actual beliefs of the firm i.e. the beliefs
3While firms are required to provide projections based on their past experience, these
projections may be manually overridden.
66
which underlie the firm’s actions in the market (see e.g. Coibion et al., 2018;
Graham et al., 2005; Rutström & Wilcox, 2009). In this environment, incentives for
accurate reporting come from the payment system. The projected costs are use to
form a final ‘bid amount’ which is then compared to the benchmark subsidy rates.
Plans with a bid amount that is higher than the benchmark must charge premiums
to enrollees; supplemental premiums may also be charged if plans include benefits
beyond those offered by TM. Firms that bid below the benchmark receive a portion
of the difference as a ‘rebate’ that must be passed on the consumers through plan
benefits. The rebate payment varies across firms and over time based on the CMS
‘star rating’ measure of insurer quality, which is a summary of multiple measures
of service quality such as the fraction of members receiving influenza vaccinations,
the 30-day hospital readmittance rate, and enrollee qualitative assessments of care
quality. After taking into account risk adjustment, we can write the rebate payment
as a function of the bid bjt and the plan-level benchmark Bjt = Bmt × ϕft with
reb(bjt;Bjt, λft) =  λft(Bjt − bjt) if bjt < Bjt , (4.1)0 if bjt ≥ Bjt
where λft is the rebate percentage.
Crucially, any reductions in patient cost-sharing, relative to TM, must
be ‘paid for’ with rebate funds. Cost-sharing plan features such as copays and
deductibles have previously been shown to be determinants of demand for MA
plans (see e.g. Curto, Einav, Levin, & Bhattacharya, 2021; K. S. Miller et al.,
2021). As a consequence, the projections made by firms directly constrain the
set of possible product characteristics offered to consumers along dimensions
that are relevant for their (the firms’) success in the market. If a firm reports
that it expects to have much higher costs than it truly believes it will face,
then it may be constrained to offer an less-than-optimally generous plan (i.e.
higher copays/deductibles and fewer additional benefit categories) and therefore
lose market share. If a firm reports that it expects to have lower costs than it
67
truly believes it will have, it may not receive enough revenue (in the form of the
government subsidy plus premium payments from enrollees) to pay for the care
that it covers under the plan.
One of the major differences between MA and TM is that MA plans must
provide all of the mandated insurance benefits of TM in exchange for a capitated
monthly payment (Abaluck, Caceres Bravo, Hull, & Starc, 2021). In general,
MA plans typically offer more generous benefits and lower cost-sharing than
TM, whereas MA plans tend to have limited physician networks and require
higher cost-sharing for costly services, perhaps because of these risk adjustment
practices (Meyers & Trivedi, 2021). According to the Medicare Payment Advisory
Commission (MedPAC), MA payers plans to continue to increase enrollment by
offering extra benefits that beneficiaries find attractive. According to their 2021
report, bids slightly decreased to 87 percent of TM, a record low (Commission et
al., 2021).
4.3 Data and Methods
We collect data on all bids and plans offered from 2007 to 2015 from
public CMS files, including data generated by submissions made through the Bid
Pricing Tool and (separately) detailed data on the provision of benefits. Following
K. S. Miller et al. (2021), we focus on the market for individual insurance. We drop
plans sponsored by employers and plans designed for individuals who are “dual-
eligible” for Medicare and Medicaid.
For each plan, we collect enrollment, the average per-capita payment, the
star rating, the deductible, the out of pocket limit, and copays for primary care
visits, specialist visits, and 7-day hospital stays.4 We construct the forecast error
for each plan j offered in year t by comparing the projected costs reported at t − 1
to the actual costs reported at t + 1: Errjt = actualj,t+1 − projectedj,t−1. In other
4A small fraction of plans use coinsurance cost-sharing mechanisms. We convert these to
copayments using the Medicare Physician Fee Schedule and the American Hospital Association
Annual Survey. Details and code are available upon request.
68
words, positive prediction errors indicate that firms underestimated their costs,
while negative prediction errors indicate that firms overestimated their costs.
We construct the overall forecast error (i.e. the forecast error in the Total
Medical Expenses line in Figure C.1), as well as forecast errors in the Inpatient
Facility, Professional, and OP Facility - Surgery service categories. We choose
these three categories as the largest components of plans’ medical expenses. We
also construct the percent forecast error as Errjt/projectedj,t−1.
We construct four additional covariates from these data. First, we identify
the state in which the plan has the most enrollees. Second, we construct the total
previous enrollment at the contract-level to capture variation in the experience
of the firm submitting the bid.5 Finally, we construct two variables designed to
capture the extent of competition: the share-weighted average number of contracts
and plans offered by competitors in each plan’s service area. That is, we weight
the number of competitor contracts/plans offered in each county in which the plan
offered by the share of the plan in that county.
Table C.1 provides summary statistics on our 11, 670 plan-year observations
offered from 2008-2013 (i.e. after calculating forecast errors). On average, firms
over-estimate their costs by $270 per-member-per-month (PMPM), though the
interquartile range include firms who underestimate their costs. The average
prediction error in inpatient facility costs is $103 PMPM, 38% of the average
overall prediction error. Professional and outpatient surgery facility costs are
overestimated on average by $77 and $16 PMPM, respectively. Over 75% of
the plan-years in our data feature zero deductible and a limit on out-of-pocket
expenses. A few plans feature zero copays; most feature positive copays for primary
5Firms sign contracts with CMS to offer potentially several plans; plans under the same
contract generally offer similar provider networks and are available in similar geographies. Large
firms (e.g. Aetna, Blue Cross Blue Shield, etc.) may have multiple contracts with CMS to offer
plans in different areas; we calculate enrollment at the contract-level instead of the firm level to
capture the possibility that the firm’s experience may vary locally in ways that affect their ability
to accurately forecast costs.
69
care visits, specialist visits, and hospital stays. Roughly one quarter of plan-
years did not receive a star rating from CMS, either because they were too new
or because not enough data was available for CMS to evaluate the plans. The
median plan is in a competitive environment featuring more than 10 firms offering
approximately two plans each.
Table C.2 reports correlations between the overall forecast error and the
three components discussed above. All correlations are positive and greater than
0.7, indicating that errors in the prediction process tend to compound across cost
categories, rather than offsetting.
Table C.3 details the distribution of the overall forecast error across
quartiles of our measure of experience in the top panel and our contract measure
of competition in the bottom panel. With respect to experience, the biggest
difference is between the first (lowest experience) quartile and other quartiles; the
difference between (for example) the third and fourth (highest experience) quartile
is minimal. This suggests that while there may be some benefit to experience, the
benefit is quickly realized. This is perhaps reasonable if firms are using regression
techniques to estimate future costs, as the power of such techniques does not
scale linearly. With respect to competition, variation across quartiles is minimal
– if anything, this cut of the data suggests firms’ prediction errors increase in
magnitude as the number of competitors increases. While one might expect firms
to invest more in accurate predictions when the competitive environment is more
saturated, the presence of other firms may increase the variance in costs as insurers
compete to add providers to their network (Gaynor, Ho, & Town, 2015).
Figure C.2 illustrates the evolution of the distribution of forecast errors
over time. While the interquartile range shrinks dramatically from 2008 to 2011,
along with a decrease in the absolute value of the mean and median, these trends
reverse in 2012 and 2013. We note that several provisions of the Affordable Care
70
Act affecting MA went into effect in around this time, which may have increased
the aggregate uncertainty experienced by MA firms.
Our analyses broadly fall into two categories. First, we study the
relationship between the prediction error and covariates of interest including our
measures of experience and competition, taking into account the possibility that
prediction errors may be serially correlated. We model prediction error outcomes
yjt as
yjt = αyy
′
j,t−1 +Xjtαx + FXt + FXf + FXs + ϵy,jt, (4.2)
where αx is the vector of coefficients of interest (where Xjt includes a constant),
FX are fixed effects (t indicates time, f indicates the insurer sponsoring plan J ,
and s indicates the primary state in which j is offered), and ϵy,jt are unobservable
factors influencing the forecast error.
Second, we study the relationship between the last-period prediction
error and the product characteristics chosen by the firm. We model product
characteristic xjt as
xjt = βyy
′
j,t−1 + Zjtβz + FXt + FXf + FXs + ϵx,jt, (4.3)
where βy is the coefficient of interest, Z
′
jt includes the star rating and the average
per-enrollee payment from the government, FX are the fixed effects described
above, and ϵx,jt are unobservable factors influencing product characteristics.
We estimate the parameters of Equations (4.2) and (4.3) using OLS and
report heteroskedasticity-robust standard errors.
4.4 Results
In this table we report the results of our analyses. We begin by exploring
the relationship between prediction errors and plan-level observables and then
continue by examining the relationship between product characteristics and prior
prediction errors.
4.4.1 The relationship between prediction errors and plan-level
observables. Table C.4 reports estimates of the parameters of Equation (4.2)
71
when the prediction error is measured in levels. Column (1) reports estimates for
the overall prediction error. The error is increasing in the previous forecast error,
the number of competitors, and the age of the plan (measured as the number of
years the plan has been in the market since 2007, the plan year that Medicare
Advantage implemented risk adjustment). The error is decreasing in the number of
competing plans, though this is estimated with slightly more noise. These patterns
generally continue across the major cost components reported in Columns (2)-(4),
though the sign on the estimated relationship between the prediction errors and the
log of the total previous contract enrollment varies.
In isolation, these results suggest that prediction errors may worsen
over time, due to the positive coefficients on both the past prediction error
and the age of the plan. However, we note that the mean prediction error is
significantly negative; thus these results may instead indicate that firms in our
data are becoming more accurate. To investigate this possibility, we re-estimate
Equation (4.2) where yjt is the absolute value of the prediction error. Table C.5
reports the results. We estimate in Column (1) that last period’s error, previous
enrollment, the number of competitors, and the age of the plan are all associated
with a decrease in the magnitude of the overall prediction error. This pattern is
largely consistent across the major cost components in Columns (2)-(4). While the
number of competing plans enters positively for several regressions, the effect size is
smaller than that of the number of competitors, indicating that, overall, increased
competition is associated with smaller forecast errors (in magnitude).
4.4.2 The relationship between prediction errors and product
characteristics. Table C.6 reports estimates of the parameters of Equation (4.3)
when the prediction error is measured in levels. Each column represents a distinct
product characteristic: the deductible, the out-of-pocket spending limit, and copays
for primary care, specialist, and hospital visits. We include the star rating (and
relevant indicators) and risk-adjusted payments as covariates to control for plan
72
quality and observable differences in mean costs across geographies (K. S. Miller
et al., 2021). At the point estimates, increases in the last-period prediction error
(i.e. as firms actual costs are more and more above their prediction costs), the
deductible, out-of-pocket expenditure limit, specialist visit copay, and hospital
visit copay all increase (though the hospital visit copay is estimated imprecisely).
In other words, as firms realize costs are higher than they believed, they seek to
pass-through a portion of those costs onto consumers (see e.g. Butters, Sacks, &
Seo, 2020; Kim, 2021). Indeed, the negative coefficient on primary care copays
(indicating that first realizing higher costs than expected decrease the patient-
facing cost of primary care visits) is consistent with this hypothesis, as increasing
primary care service utilization is thought to reduce health care expenses overall
(see e.g. Starfield, Shi, & Macinko, 2005).
As above, while these results are informative about the direction of the
relationship, they are less information about the impact of the magnitude of the
prediction error. We therefore re-estimate the parameters of Equation (4.3) with
the absolute overall prediction error as an independent variable. The results are
reported in Table C.7. Across product characteristics, the estimated coefficients on
the absolute prediction error are the opposite sign of those reported in Table C.6.
This is perhaps best interpreted in light of the results reported in Table C.4. Firms
with large prediction errors are likely to more accurately predict their costs in
the future. If uncertainty had pushed firms to offer plans with high levels of cost
sharing (i.e. large deductibles, etc.), a reduction in that uncertainty should push
firms to offer more desirable plans (see e.g. De Vany & Saving, 1977).
4.5 Discussion and conclusion
The idea that firms can learn about and adjust to changes in market
environments is at the heart of many common models of equilibrium (see e.g. Berry,
Levinsohn, & Pakes, 1995; Ifrach & Weintraub, 2016). While there is an extensive
literature developing the theory of firm learning, the empirical literature has largely
73
been limited by a lack of revealed preference data on firm beliefs. We fill that gap
by studying the MA market, in which health insurance firms are incentivized to
accurately disclose their best estimates of their own future medical costs as part of
a mandatory regulatory process. These disclosures are then released publicly some
years after the plan year.
We document the relationship between these estimates along two directions.
First, we consider the “input” side: what factors influence the accuracy of the
forecasts? We find that as the experience of the firm increases, as the age of the
plan increases, and as the competitive environment intensifies, the magnitude of
forecast errors decreases. In other words, those factors which we might reasonably
expect to positively affect firms’ ability to forecast their own costs do so in the
direction consistent with reasonable priors.
We then turn to the “output” side: How do accurate and inaccurate
forecasts influence product characteristic decisions. We document two stylized facts.
First, the more firms underestimate their own costs, the more they seek to pass
their costs on to consumers in the future through cost-sharing benefit structures.
Second, the larger the magnitude of past forecast errors (in either direction), the
more firms seek to improve their plans by offering reduced cost sharing.
Taken together, our results paint a picture of firms behaving as is often
implicitly assumed in dynamic models of imperfect competition. Even though
the firms in our data consistently overestimate costs on average, at the plan level
the predictions increase in accuracy, allowing firms to offer better products (i.e.
insurance plans with lower out-of-pocket costs) to consumers.
We conclude by pointing out that significant gaps in our understanding of
firm expectations remain for future research to fill. For example, firms generally
must predict not just supply (cost) conditions but also demand conditions. These
predictions may behave differently than cost predictions, particularly as they are
made in the context of competitors who are also making similar predictions.
74
APPENDIX A
CHAPTER 2 APPENDIX
Table A.1. Basic Model Results After 2002
(1) (2) (3) (4)
exit exit exit exit
LOG Cap −0.522∗∗∗ −0.525∗∗∗ −0.592∗∗∗ −0.598∗∗∗
(0.0545) (0.0569) (0.0609) (0.0640)
Elev Ownership 0.155 0.208 0.229 0.263
(0.210) (0.214) (0.228) (0.232)
Entrant −0.795∗∗∗ −0.632∗∗∗ −0.947∗∗∗ −0.799∗∗∗
(0.124) (0.133) (0.135) (0.146)
Time Fixed Effect ✗ ✓ ✗ ✓
Subdivision Fixed Effect ✗ ✗ ✓ ✓
Constant 2.397∗∗∗ 3.118∗∗∗ 2.740∗∗∗ 3.469∗∗∗
(0.489) (0.517) (0.752) (0.780)
N 5214 4869 5153 4816
Log Likelihood −1221.539 −1163.7757 −1180.4597 −1125.3187
standard errors in parentheses
∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
75
APPENDIX B
CHAPTER 3 APPENDIX
B.1 Abbreviations
Table B.1. List of Abbreviations
Abbrev. Meaning
BLS Bureau of Labor Statistics
CART classification and regression trees
CBD cannabidiol
GDP gross domestic product
LASSO least absolute shrinkage and selection operator
NAICS North American Industry Classification System
RMSPE ratio of the mean squared prediction errors
THC tetrahydrocannbinol
US United States
B.2 Additional tables and figures
76
Table B.2.1. Synthetic control weights assigned to each state for
narrowly-defined agriculture labor market outcomes
Log Real Average
Log Number of Log Number of Log Real Total
Weekly Wage
Establishments Workers Quarterly Wage
Per Worker
Colorado
Arizona 0.00 0.00 0.00 0.04
Georgia 0.19 0.00 0.38 0.00
Hawaii 0.00 0.25 0.00 0.00
Maryland 0.00 0.00 0.00 0.33
Minnesota 0.00 0.00 0.00 0.25
Montana 0.00 0.22 0.20 0.11
New Hampshire 0.19 0.00 0.00 0.00
South Carolina 0.00 0.00 0.00 0.23
Texas 0.45 0.53 0.41 0.04
Vermont 0.17 0.00 0.00 0.00
Washington
Arizona 0.00 0.03 0.11 0.00
Connecticut 0.00 0.02 0.00 0.07
Florida 0.10 0.04 0.00 0.00
Georgia 0.00 0.00 0.00 0.37
Hawaii 0.00 0.05 0.01 0.07
Illinois 0.08 0.00 0.00 0.00
Kentucky 0.00 0.00 0.08 0.00
Michigan 0.54 0.40 0.05 0.00
Minnesota 0.28 0.37 0.00 0.00
Montana 0.00 0.00 0.00 0.14
South Dakota 0.00 0.00 0.00 0.15
Texas 0.00 0.08 0.66 0.19
West Virginia 0.00 0.00 0.09 0.00
Notes: The table provides the weights assigned to states for the synthetic controls used to
estimate the “narrowly-defined agriculture” models in Table 6. All states except those which
legalized cannabis during our study period and those bordering either Washington or Colorado
were included in the pool of potential control units. Only states which received positive weight for
at least one outcome are included in the table.
77
Table B.2.2. Synthetic control weights assigned to each state for
narrowly-defined retail labor market outcomes
Log Real Average
Log Number of Log Number of Log Real Total
Weekly Wage
Establishments Workers Quarterly Wage
Per Worker
Colorado
Georgia 0.07 0.11 0.19 0.00
Iowa 0.41 0.26 0.18 0.00
Kentucky 0.10 0.00 0.01 0.33
Louisiana 0.20 0.28 0.21 0.00
Minnesota 0.00 0.28 0.18 0.00
Mississippi 0.00 0.00 0.00 0.11
Missouri 0.00 0.00 0.01 0.00
New Hampshire 0.00 0.00 0.00 0.14
Pennsylvania 0.00 0.00 0.00 0.22
South Dakota 0.00 0.00 0.00 0.19
Texas 0.22 0.06 0.05 0.00
Wisconsin 0.00 0.00 0.18 0.00
Washington
Connecticut 0.00 0.27 0.31 0.00
Illinois 0.00 0.38 0.36 0.00
Iowa 0.35 0.00 0.00 0.21
Michigan 0.33 0.04 0.05 0.07
Mississippi 0.05 0.00 0.00 0.00
New York 0.00 0.00 0.02 0.00
North Carolina 0.12 0.18 0.00 0.53
Pennsylvania 0.15 0.00 0.00 0.08
South Carolina 0.00 0.13 0.26 0.11
Notes: The table provides the weights assigned to states for the synthetic controls used to
estimate the “narrowly-defined retail” models in Table 6. All states except those which legalized
cannabis during our study period and those bordering either Washington or Colorado were
included in the pool of potential control units. Only states which received positive weight for at
least one outcome are included in the table.
78
Table B.2.3. Synthetic control weights assigned to each state for
broadly-defined agriculture labor market outcomes
Log Real Average
Log Number of Log Number of Log Real Total
Weekly Wage
Establishments Workers Quarterly Wage
Per Worker
Colorado
Arizona 0.19 0.26 0.12 0.00
Georgia 0.52 0.00 0.10 0.02
Hawaii 0.01 0.02 0.01 0.00
Kentucky 0.00 0.00 0.00 0.16
Minnesota 0.00 0.24 0.10 0.43
Montana 0.07 0.27 0.08 0.00
New Hampshire 0.00 0.05 0.22 0.00
South Dakota 0.21 0.00 0.00 0.00
Texas 0.00 0.15 0.29 0.02
Virginia 0.00 0.00 0.08 0.37
Washington
Connecticut 0.00 0.00 0.00 0.07
Florida 0.00 0.86 0.74 0.18
Michigan 0.00 0.00 0.00 0.69
Minnesota 0.00 0.00 0.00 0.05
Montana 0.07 0.00 0.00 0.00
Texas 0.93 0.12 0.24 0.00
Notes: The table provides the weights assigned to states for the synthetic controls used to
estimate the “broadly-defined agriculture” models in Table 7. All states except those which
legalized cannabis during our study period and those bordering either Washington or Colorado
were included in the pool of potential control units; Alaska and California were added to the pool
for Washington due to the similarity in their agriculture, forestry, and fishing industries. Only
states which received positive weight for at least one outcome are included in the table.
79
Table B.2.4. Synthetic control weights assigned to each state for
broadly-defined retail labor market outcomes
Log Real Average
Log Number of Log Number of Log Real Total
Weekly Wage
Establishments Workers Quarterly Wage
Per Worker
Colorado
Alabama 0.02 0.01 0.02 0.02
Arizona 0.02 0.18 0.11 0.02
Arkansas 0.02 0.01 0.03 0.03
Connecticut 0.23 0.00 0.00 0.01
Florida 0.00 0.00 0.00 0.01
Georgia 0.02 0.00 0.01 0.02
Hawaii 0.04 0.00 0.01 0.00
Illinois 0.01 0.00 0.01 0.03
Indiana 0.02 0.00 0.01 0.03
Iowa 0.02 0.01 0.02 0.05
Kentucky 0.02 0.01 0.01 0.02
Louisiana 0.03 0.01 0.02 0.05
Maryland 0.02 0.00 0.00 0.01
Michigan 0.01 0.00 0.01 0.02
Minnesota 0.02 0.01 0.03 0.07
Mississippi 0.02 0.01 0.03 0.03
Missouri 0.02 0.01 0.02 0.03
Montana 0.03 0.35 0.29 0.03
New Hampshire 0.03 0.00 0.01 0.03
New Jersey 0.07 0.00 0.00 0.01
New York 0.12 0.00 0.00 0.00
North Carolina 0.01 0.00 0.01 0.02
Ohio 0.01 0.00 0.00 0.02
Pennsylvania 0.01 0.00 0.00 0.02
South Carolina 0.02 0.00 0.01 0.02
South Dakota 0.03 0.04 0.04 0.06
Tennessee 0.02 0.00 0.01 0.01
Texas 0.01 0.34 0.28 0.19
Vermont 0.03 0.00 0.00 0.02
Virginia 0.02 0.00 0.01 0.04
West Virginia 0.02 0.01 0.01 0.03
Wisconsin 0.02 0.01 0.01 0.04
Washington
Alabama 0.00 0.00 0.00 0.01
Connecticut 0.38 0.00 0.00 0.00
Georgia 0.00 0.00 0.00 0.01
Hawaii 0.00 0.00 0.00 0.09
Illinois 0.28 0.00 0.00 0.01
Iowa 0.00 0.24 0.01 0.00
Michigan 0.00 0.00 0.00 0.01
Minnesota 0.00 0.00 0.00 0.01
Missouri 0.00 0.00 0.00 0.01
New York 0.02 0.01 0.00 0.49
North Carolina 0.01 0.31 0.01 0.01
Pennsylvania 0.01 0.00 0.00 0.00
South Carolina 0.27 0.22 0.23 0.00
Texas 0.00 0.00 0.00 0.31
Vermont 0.00 0.00 0.00 0.01
Virginia 0.00 0.21 0.74 0.00
Notes: The table provides the weights assigned to states for the synthetic controls used to
estimate the “broadly-defined retail” models in Table 7. All states except those which legalized
cannabis during our study period and those bordering either Washington or Colorado were
included in the pool of potential control units. O8n0ly states which received positive weight for at
least one outcome are included in the table.
Figure B.2.1. Employment and wages for “broadly defined” agricultural
firms
A B
Number of Establishments Number of Workers
10000
90000
7500
60000
5000
30000
2500
2000 2005 2010 2015 2000 2005 2010 2015
Year Year
C D
Real Total Quarterly Wage Real Average Weekly Wage Per Worker
750
750
700
650
500
600
250
550
2000 2005 2010 2015 2000 2005 2010 2015
Year Year
Colorado Washington Other States
Notes: Data come from the Quarterly Census of Employment and Wages. We define ”broadly
defined” agricultural firms as those within NAICS 11 (“Agriculture, Forestry, Fishing, and
Hunting”).
81
Dollar (in Millions) Count
Dollar Count
Figure B.2.2. Employment and wages for “broadly defined” retail firms
A B
Number of Establishments Number of Workers
5500 110000
100000
5000
90000
4500
80000
4000
70000
2000 2005 2010 2015 2000 2005 2010 2015
Year Year
C D
Real Total Quarterly Wage Real Average Weekly Wage Per Worker
650
900
800 600
700
550
600
500
500
2000 2005 2010 2015 2000 2005 2010 2015
Year Year
Colorado Washington Other States
Notes: Data come from the Quarterly Census of Employment and Wages. We define “broadly
defined” retail firms as those within NAICS 446, 452, and 453 (“Health and personal care stores”,
“General merchandise stores”, and “Miscellaneous stores”, respectively).
82
Dollar (in Millions) Count
Dollar Count
Figure B.2.3. Comparing broadly-defined agriculture labor market
outcomes in Colorado and its synthetic control
A B
Log Number of Establishments Log Number of Workers
9.9
7.4
9.8
7.3 9.7
9.6
7.2
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
6.60
19.0
18.9
6.55
18.8
6.50
18.7
6.45
18.6
2005 2010 2015 2005 2010 2015
Year Year
Colorado Synthetic
Notes: This figure depicts wage and employment outcomes for “broadly defined” agricultural firms
for Colorado and its synthetic control. We define ”broadly defined” agricultural firms as those
within NAICS 11 (“Agriculture, Forestry, Fishing, and Hunting”).
83
Figure B.2.4. Comparing broadly-defined agriculture labor market
outcomes in Washington and its synthetic control
A B
Log Number of Establishments Log Number of Workers
11.6
9.1
11.4
9.0
11.2
8.9
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
20.6
20.5 6.45
20.4
6.40
20.3
6.35
20.2
20.1 6.30
2005 2010 2015 2005 2010 2015
Year Year
Washington Synthetic
Notes: This figure depicts wage and employment outcomes for “broadly defined” agricultural firms
for Washington and its synthetic control. We define ”broadly defined” agricultural firms as those
within NAICS 11 (“Agriculture, Forestry, Fishing, and Hunting”).
84
Figure B.2.5. Comparing broadly-defined retailer labor market
outcomes in Colorado and its synthetic control
A B
Log Number of Establishments Log Number of Workers
11.35
8.35
11.30
11.25
8.30
11.20
8.25
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
6.30
20.20
6.28
20.15
20.10 6.26
20.05
6.24
20.00
6.22
2005 2010 2015 2005 2010 2015
Year Year
Colorado Synthetic
Notes: This figure depicts wage and employment outcomes for “broadly defined” retail firms
for Colorado and its synthetic control. We define ”broadly defined” retail firms as those within
NAICS 446, 452, and 453 (“Health and personal care stores”, “General merchandise stores”, and
“Miscellaneous stores”, respectively).
85
Figure B.2.6. Comparing broadly-defined retailer labor market
outcomes in Washington and its synthetic control
A B
Log Number of Establishments Log Number of Workers
8.60
11.6
8.55
11.5
8.50
11.4
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
6.450
20.6
6.425
20.5
6.400
20.4
6.375
20.3
2005 2010 2015 2005 2010 2015
Year Year
Washington Synthetic
Notes: This figure depicts wage and employment outcomes for “broadly defined” retail firms for
Washington and its synthetic control. We define ”broadly defined” retail firms as those within
NAICS 446, 452, and 453 (“Health and personal care stores”, “General merchandise stores”, and
“Miscellaneous stores”, respectively).
86
Figure B.2.7. Placebo tests for narrowly-defined agriculture labor
market outcomes in Colorado
A B
Log Number of Establishments Log Number of Workers
0.6 0.50
0.3
0.25
0.0
0.00
−0.3
−0.25
−0.6
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
0.50
0.25 0.2
0.00
0.0
−0.25
−0.2
−0.50
2005 2010 2015 2005 2010 2015
Year Year
Synthetic Colorado
Notes: This figure depicts the placebo tests for “narrowly defined” agricultural firms for Colorado.
We define ”narrowly defined” agricultural firms as those within the “Greenhouse and Nursery
Production” (NAICS 1114) industry.
87
Figure B.2.8. Placebo tests for narrowly-defined agriculture labor
market outcomes in Washington
A B
Log Number of Establishments Log Number of Workers
0.8
1.0
0.5 0.4
0.0
0.0
−0.5
−0.4
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
0.75
0.50 0.2
0.25
0.0
0.00
−0.2
−0.25
−0.50 −0.4
2005 2010 2015 2005 2010 2015
Year Year
Synthetic Washington
Notes: This figure depicts the placebo tests for “narrowly defined” agricultural firms for
Washington. We define ”narrowly defined” agricultural firms as those within the “Greenhouse
and Nursery Production” (NAICS 1114) industry.
88
Figure B.2.9. Placebo tests for narrowly-defined retailer labor market
outcomes in Colorado
A B
Log Number of Establishments Log Number of Workers
1.0
0.5
0.5
0.0
0.0
−0.5
−0.5
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
1.0
0.50
0.5
0.25
0.0
0.00
−0.5
−0.25
−1.0
2005 2010 2015 2005 2010 2015
Year Year
Synthetic Colorado
Notes: This figure depicts the placebo tests for wage and employment outcomes for “narrowly
defined” retail firms for Colorado. We define ”narrowly defined” retail firms as those within the
“Store retailers not specified elsewhere” category (NAICS 453998).
89
Figure B.2.10. Placebo tests for narrowly-defined retailer labor market
outcomes in Washington
A B
Log Number of Establishments Log Number of Workers
1.0
0.25
0.5
0.00
0.0
−0.25
−0.5
−0.50
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
1.0
0.4
0.5
0.2
0.0
0.0
−0.5
−0.2
−1.0
2005 2010 2015 2005 2010 2015
Year Year
Synthetic Washington
Notes: This figure depicts the placebo tests for wage and employment outcomes for “narrowly
defined” retail firms for Washington. We define ”narrowly defined” retail firms as those within the
“Store retailers not specified elsewhere” category (NAICS 453998).
90
Figure B.2.11. Placebo tests for broadly-defined agriculture labor
market outcomes in Colorado
A B
Log Number of Establishments Log Number of Workers
0.2
0.1
0.1
0.0
0.0
−0.1 −0.1
−0.2
−0.2
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
0.3
0.1
0.2
0.1
0.0
0.0
−0.1
−0.1
−0.2
2005 2010 2015 2005 2010 2015
Year Year
Synthetic Colorado
Notes: This figure depicts the placebo tests for “broadly defined” agricultural firms for Colorado.
We define ”broadly defined” agricultural firms as those within NAICS 11 (“Agriculture, Forestry,
Fishing, and Hunting”).
91
Figure B.2.12. Placebo tests for broadly-defined agriculture labor
market outcomes in Washington
A B
Log Number of Establishments Log Number of Workers
0.4
0.2
0.2
0.0
0.0
−0.2
−0.2
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
0.50
0.1
0.25
0.0
0.00
−0.1
−0.25
2005 2010 2015 2005 2010 2015
Year Year
Synthetic Washington
Notes: This figure depicts the placebo tests for “broadly defined” agricultural firms for
Washington. We define ”broadly defined” agricultural firms as those within NAICS 11
(“Agriculture, Forestry, Fishing, and Hunting”).
92
Figure B.2.13. Placebo tests for broadly-defined retailer labor market
outcomes in Colorado
A B
Log Number of Establishments Log Number of Workers
0.2 0.2
0.1 0.1
0.0 0.0
−0.1
−0.1
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
0.10
0.3
0.2 0.05
0.1
0.00
0.0
−0.05
−0.1
−0.10
2005 2010 2015 2005 2010 2015
Year Year
Synthetic Colorado
Notes: This figure depicts the placebo tests for wage and employment outcomes for “broadly
defined” retail firms for Colorado. We define ”broadly defined” retail firms as those within
NAICS 446, 452, and 453 (“Health and personal care stores”, “General merchandise stores”,
and “Miscellaneous stores”, respectively).
93
Figure B.2.14. Placebo tests for broadly-defined retailer labor market
outcomes in Washington
A B
Log Number of Establishments Log Number of Workers
0.2 0.2
0.1 0.1
0.0 0.0
−0.1
−0.1
2005 2010 2015 2005 2010 2015
Year Year
C D
Log Real Total Quarterly Wage Log Real Average Weekly Wage Per Worker
0.10
0.3
0.2 0.05
0.1
0.00
0.0
−0.05
−0.1
−0.10
2005 2010 2015 2005 2010 2015
Year Year
Synthetic Washington
Notes: This figure depicts the placebo tests for wage and employment outcomes for “broadly
defined” retail firms for Washington. We define ”broadly defined” retail firms as those within
NAICS 446, 452, and 453 (“Health and personal care stores”, “General merchandise stores”, and
“Miscellaneous stores”, respectively).
94
B.3 Tables of covariate balance
Table B.3.1. CO broadly-defined agriculture average weekly wage per worker
Treated Synthetic Sample Mean
Lagged outcome 6.46 6.45 6.45
Barley for grain (acres) 65547.33 64022.64 41592.84
Land in orchards (acres) 6444.00 18069.70 51262.40
Snap beans harvested for sale, harvested (acres) 590.67 4442.38 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 159.67 28.15 2208.25
Fruits & nuts, pears, all, total acres (acres) 313.67 128.15 227.10
Comm. soil conds. (thousands of treated acres) 4130.86 8059.39 5357.77
Resident population 65 years & over (percent) 10.29 11.95 13.03
Savings institutions - total deposits (thousands) 1210.38 1377.93 2692.87
Civilian labor force unemployment rate (percent) 5.36 4.93 5.53
Federal Government expenditure-grants (millions) 6.04 8.5 10.89
Federal Government insurance (millions) 3.89 12.04 25.58
Resident population: Black alone (percent) 4.25 11.00 13.53
Resident population: Two or more races (percent) 1.79 1.42 1.71
Resident population: Hispanic or Latino Origin (percent) 17.58 5.92 7.41
Resident population: total females (percent) 49.70 50.64 50.93
Social security: retired workers-benefit recipients (thousands) 386.55 612.67 716.18
Corn Grain Production (dollar, millions) 492.99 1678.67 1045.22
Hay production (dollar, millions) 493.11 412.94 260.35
Farm operations (acres, millions) 62.65 39.43 34.27
Labor hired wage (per hour) 8.50 8.64 11.20
Rent cash cropland expense (acres) 60.00 77.05 75.20
Vegetable totals (dollars, millions) 110.31 56.15 140.21
Wheat production (dollars, millions) 352.67 233.34 138.08
95
Table B.3.2. CO broadly-defined agriculture total quarterly wages
Treated Synthetic Sample Mean
Lagged outcome 18.59 18.58 18.37
Barley for grain (acres) 65547.33 87647.33 41592.84
Land in orchards (acres) 6444.00 86064.13 51262.40
Snap beans harvested for sale, harvested (acres) 590.67 4755.89 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 159.67 105.70 2208.25
Fruits & nuts, pears, all, total acres (acres) 313.67 247.04 227.10
Comm. soil conds. (thousands of treated acres) 4130.86 8048.08 5357.77
Resident population 65 years & over (percent) 10.29 11.79 13.03
Savings institutions - total deposits (thousands) 1210.38 2651.68 2692.87
Civilian labor force unemployment rate (percent) 5.36 5.10 5.53
Federal Government expenditure-grants (millions) 6.04 13.57 10.89
Federal Government insurance (millions) 3.89 36.76 25.58
Resident population: Black alone (percent) 4.25 9.76 13.53
Resident population: Two or more races (percent) 1.79 1.44 1.71
Resident population: Hispanic or Latino Origin (percent) 17.58 15.23 7.41
Resident population: total females (percent) 49.70 50.44 50.93
Social security: retired workers-benefit recipients (thousands) 386.55 825.32 716.18
Corn Grain Production (dollar, millions) 492.99 728.86 1045.22
Hay production (dollar, millions) 493.11 439.95 260.35
Farm operations (acres, millions) 62.65 99.11 34.27
Labor hired wage (per hour) 8.50 10.15 11.20
Rent cash cropland expense (acres) 60.00 65.61 75.20
Vegetable totals (dollars, millions) 110.31 194.72 140.21
Wheat production (dollars, millions) 352.67 216.10 138.08
96
Table B.3.3. CO broadly-defined agriculture average employment
Treated Synthetic Sample Mean
Lagged outcome 9.57 9.57 9.36
Barley for grain (acres) 65547.33 258337.20 41592.84
Land in orchards (acres) 6444.00 47636.03 51262.40
Snap beans harvested for sale, harvested (acres) 590.67 2589.34 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 159.67 89.21 2208.25
Fruits & nuts, pears, all, total acres (acres) 313.67 118.67 227.10
Comm. soil conds. (thousands of treated acres) 4130.86 8445.85 5357.77
Resident population 65 years & over (percent) 10.29 12.61 13.03
Savings institutions - total deposits (thousands) 1210.38 1339.26 2692.87
Civilian labor force unemployment rate (percent) 5.36 5.15 5.53
Federal Government expenditure-grants (millions) 6.04 9.72 10.89
Federal Government insurance (millions) 3.89 19.85 25.58
Resident population: Black alone (percent) 4.25 5.00 13.53
Resident population: Two or more races (percent) 1.79 1.70 1.71
Resident population: Hispanic or Latino Origin (percent) 17.58 14.14 7.41
Resident population: total females (percent) 49.70 50.23 50.93
Social security: retired workers-benefit recipients (thousands) 386.55 600.74 716.18
Corn Grain Production (dollar, millions) 492.99 1036.97 1045.22
Hay production (dollar, millions) 493.11 451.59 260.35
Farm operations (acres, millions) 62.65 92.69 34.27
Labor hired wage (per hour) 8.50 12.82 11.20
Rent cash cropland expense (acres) 60.00 71.83 75.20
Vegetable totals (dollars, millions) 110.31 246.95 140.21
Wheat production (dollars, millions) 352.67 396.12 138.08
97
Table B.3.4. CO broadly-defined agriculture number of establishments
Treated Synthetic Sample Mean
Lagged outcome 7.22 7.21 7.15
Barley for grain (acres) 65547.33 72538.36 41592.84
Land in orchards (acres) 6444.00 81145.72 51262.40
Snap beans harvested for sale, harvested (acres) 590.67 7659.43 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 159.67 13.79 2208.25
Fruits & nuts, pears, all, total acres (acres) 313.67 161.75 227.10
Comm. soil conds. (thousands of treated acres) 4130.86 4692.31 5357.77
Resident population 65 years & over (percent) 10.29 11.78 13.03
Savings institutions - total deposits (thousands) 1210.38 576.47 2692.87
Civilian labor force unemployment rate (percent) 5.36 5.16 5.53
Federal Government expenditure-grants (millions) 6.04 8.54 10.89
Federal Government insurance (millions) 3.89 10.31 25.58
Resident population: Black alone (percent) 4.25 16.11 13.53
Resident population: Two or more races (percent) 1.79 1.47 1.71
Resident population: Hispanic or Latino Origin (percent) 17.58 9.66 7.41
Resident population: total females (percent) 49.70 50.52 50.93
Social security: retired workers-benefit recipients (thousands) 386.55 533.48 716.18
Corn Grain Production (dollar, millions) 492.99 427.44 1045.22
Hay production (dollar, millions) 493.11 253.37 260.35
Farm operations (acres, millions) 62.65 42.90 34.27
Labor hired wage (per hour) 8.50 11.43 11.20
Rent cash cropland expense (acres) 60.00 70.24 75.20
Vegetable totals (dollars, millions) 110.31 282.20 140.21
Wheat production (dollars, millions) 352.67 214.24 138.08
98
Table B.3.5. WA broadly-defined agriculture average weekly wage per worker
Treated Synthetic Sample Mean
Lagged outcome 6.31 6.33 6.45
Barley for grain (acres) 245385.00 17117.25 41592.84
Land in orchards (acres) 308608.00 209756.76 51262.40
Snap beans harvested for sale, harvested (acres) 3418.67 19386.12 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 1976.33 30113.96 2208.25
Fruits & nuts, pears, all, total acres (acres) 26240.67 751.24 227.10
Comm. soil conds. (thousands of treated acres) 3959.26 5087.90 5357.77
Resident population 65 years & over (percent) 11.55 13.50 13.03
Savings institutions - total deposits (thousands) 3693.15 4339.17 2692.87
Civilian labor force unemployment rate (percent) 6.50 6.98 5.53
Federal Government expenditure-grants (millions) 9.92 14.84 10.89
Federal Government insurance (millions) 7.33 69.71 25.58
Resident population: Black alone (percent) 4.45 13.37 13.53
Resident population: Two or more races (percent) 2.78 1.40 1.71
Resident population: Hispanic or Latino Origin (percent) 9.38 7.01 7.41
Resident population: total females (percent) 50.23 50.91 50.93
Social security: retired workers-benefit recipients (thousands) 622.15 1242.64 716.18
Corn Grain Production (dollar, millions) 78.87 911.96 1045.22
Hay production (dollar, millions) 445.95 247.26 260.35
Farm operations (acres, millions) 30.02 20.39 34.27
Labor hired wage (per hour) 9.50 14.85 11.20
Rent cash cropland expense (acres) 136.50 81.52 75.20
Vegetable totals (dollars, millions) 182.97 362.61 140.21
Wheat production (dollars, millions) 782.89 168.52 138.08
99
Table B.3.6. WA broadly-defined agriculture total quarterly wages
Treated Synthetic Sample Mean
Lagged outcome 20.20 20.16 18.37
Barley for grain (acres) 245385.00 8933.69 41592.84
Land in orchards (acres) 308608.00 586488.64 51262.40
Snap beans harvested for sale, harvested (acres) 3418.67 28444.94 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 1976.33 20969.07 2208.25
Fruits & nuts, pears, all, total acres (acres) 26240.67 236.82 227.10
Comm. soil conds. (thousands of treated acres) 3959.26 6550.59 5357.77
Resident population 65 years & over (percent) 11.55 15.42 13.03
Savings institutions - total deposits (thousands) 3693.15 6468.79 2692.87
Civilian labor force unemployment rate (percent) 6.50 5.76 5.53
Federal Government expenditure-grants (millions) 9.92 23.07 10.89
Federal Government insurance (millions) 7.33 296.41 25.58
Resident population: Black alone (percent) 4.45 14.89 13.53
Resident population: Two or more races (percent) 2.78 1.28 1.71
Resident population: Hispanic or Latino Origin (percent) 9.38 21.37 7.41
Resident population: total females (percent) 50.23 50.86 50.93
Social security: retired workers-benefit recipients (thousands) 622.15 2187.51 716.18
Corn Grain Production (dollar, millions) 78.87 214.88 1045.22
Hay production (dollar, millions) 445.95 301.58 260.35
Farm operations (acres, millions) 30.02 78.23 34.27
Labor hired wage (per hour) 9.50 32.52 11.20
Rent cash cropland expense (acres) 136.50 77.57 75.20
Vegetable totals (dollars, millions) 182.97 1073.11 140.21
Wheat production (dollars, millions) 782.89 65.42 138.08
100
Table B.3.7. WA broadly-defined agriculture average employment
Treated Synthetic Sample Mean
Lagged outcome 11.34 11.33 9.36
Barley for grain (acres) 245385.00 10088.92 41592.84
Land in orchards (acres) 308608.00 642996.75 51262.40
Snap beans harvested for sale, harvested (acres) 3418.67 31630.43 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 1976.33 24223.85 2208.25
Fruits & nuts, pears, all, total acres (acres) 26240.67 171.85 227.10
Comm. soil conds. (thousands of treated acres) 3959.26 4600.35 5357.77
Resident population 65 years & over (percent) 11.55 16.18 13.03
Savings institutions - total deposits (thousands) 3693.15 6420.39 2692.87
Civilian labor force unemployment rate (percent) 6.50 5.75 5.53
Federal Government expenditure-grants (millions) 9.92 21.60 10.89
Federal Government insurance (millions) 7.33 325.02 25.58
Resident population: Black alone (percent) 4.45 15.18 13.53
Resident population: Two or more races (percent) 2.78 1.29 1.71
Resident population: Hispanic or Latino Origin (percent) 9.38 19.48 7.41
Resident population: total females (percent) 50.23 50.94 50.93
Social security: retired workers-benefit recipients (thousands) 622.15 2241.95 716.18
Corn Grain Production (dollar, millions) 78.87 134.56 1045.22
Hay production (dollar, millions) 445.95 191.14 260.35
Farm operations (acres, millions) 30.02 47.49 34.27
Labor hired wage (per hour) 9.50 36.43 11.20
Rent cash cropland expense (acres) 136.50 85.01 75.20
Vegetable totals (dollars, millions) 182.97 1198.46 140.21
Wheat production (dollars, millions) 782.89 34.54 138.08
101
Table B.3.8. WA broadly-defined agriculture number of establishments
Treated Synthetic Sample Mean
Lagged outcome 9.00 8.99 7.15
Barley for grain (acres) 245385.00 62094.10 41592.84
Land in orchards (acres) 308608.00 199298.35 51262.40
Snap beans harvested for sale, harvested (acres) 3418.67 7564.68 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 1976.33 292.81 2208.25
Fruits & nuts, pears, all, total acres (acres) 26240.67 586.85 227.10
Comm. soil conds. (thousands of treated acres) 3959.26 17524.58 5357.77
Resident population 65 years & over (percent) 11.55 10.91 13.03
Savings institutions - total deposits (thousands) 3693.15 6096.23 2692.87
Civilian labor force unemployment rate (percent) 6.50 5.72 5.53
Federal Government expenditure-grants (millions) 9.92 29.17 10.89
Federal Government insurance (millions) 7.33 100.50 25.58
Resident population: Black alone (percent) 4.45 12.36 13.53
Resident population: Two or more races (percent) 2.78 1.25 1.71
Resident population: Hispanic or Latino Origin (percent) 9.38 29.70 7.41
Resident population: total females (percent) 50.23 50.33 50.93
Social security: retired workers-benefit recipients (thousands) 622.15 1658.19 716.18
Corn Grain Production (dollar, millions) 78.87 718.13 1045.22
Hay production (dollar, millions) 445.95 926.67 260.35
Farm operations (acres, millions) 30.02 250.92 34.27
Labor hired wage (per hour) 9.50 7.59 11.20
Rent cash cropland expense (acres) 136.50 32.10 75.20
Vegetable totals (dollars, millions) 182.97 249.71 140.21
Wheat production (dollars, millions) 782.89 302.52 138.08
102
Table B.3.9. CO narrowly-defined agriculture average weekly wage per worker
Treated Synthetic Sample Mean
Lagged outcome 6.46 6.45 6.34
Barley for grain (acres) 65547.33 120968.85 41592.84
Land in orchards (acres) 6444.00 19911.85 51262.40
Snap beans harvested for sale, harvested (acres) 590.67 4764.35 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 159.67 50.67 2208.25
Fruits & nuts, pears, all, total acres (acres) 313.67 91.63 227.10
Comm. soil conds. (thousands of treated acres) 4130.86 7547.31 5357.77
Resident population 65 years & over (percent) 10.29 11.98 13.03
Savings institutions - total deposits (thousands) 1210.38 1257.63 2692.87
Civilian labor force unemployment rate (percent) 5.36 4.97 5.53
Federal Government expenditure-grants (millions) 6.04 8.89 10.89
Federal Government insurance (millions) 3.89 9.88 25.58
Resident population: Black alone (percent) 4.25 13.61 13.53
Resident population: Two or more races (percent) 1.79 1.44 1.71
Resident population: Hispanic or Latino Origin (percent) 17.58 7.24 7.41
Resident population: total females (percent) 49.70 50.75 50.93
Social security: retired workers-benefit recipients (thousands) 386.55 542.95 716.18
Corn Grain Production (dollar, millions) 492.99 1575.22 1045.22
Hay production (dollar, millions) 493.11 311.07 260.35
Farm operations (acres, millions) 62.65 41.48 34.27
Labor hired wage (per hour) 8.50 10.21 11.20
Rent cash cropland expense (acres) 60.00 81.00 75.20
Vegetable totals (dollars, millions) 110.31 111.04 140.21
Wheat production (dollars, millions) 352.67 268.18 138.08
103
Table B.3.10. CO narrowly-defined agriculture total quarterly wages
Treated Synthetic Sample Mean
Lagged outcome 17.01 17.00 16.39
Barley for grain (acres) 65547.33 155321.91 41592.84
Land in orchards (acres) 6444.00 140953.22 51262.40
Snap beans harvested for sale, harvested (acres) 590.67 9375.06 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 159.67 133.44 2208.25
Fruits & nuts, pears, all, total acres (acres) 313.67 370.37 227.10
Comm. soil conds. (thousands of treated acres) 4130.86 9932.14 5357.77
Resident population 65 years & over (percent) 10.29 11.01 13.03
Savings institutions - total deposits (thousands) 1210.38 2910.26 2692.87
Civilian labor force unemployment rate (percent) 5.36 5.52 5.53
Federal Government expenditure-grants (millions) 6.04 17.65 10.89
Federal Government insurance (millions) 3.89 48.94 25.58
Resident population: Black alone (percent) 4.25 17.28 13.53
Resident population: Two or more races (percent) 1.79 1.26 1.71
Resident population: Hispanic or Latino Origin (percent) 17.58 15.83 7.41
Resident population: total females (percent) 49.70 50.51 50.93
Social security: retired workers-benefit recipients (thousands) 386.55 1026.54 716.18
Corn Grain Production (dollar, millions) 492.99 384.11 1045.22
Hay production (dollar, millions) 493.11 499.44 260.35
Farm operations (acres, millions) 62.65 133.07 34.27
Labor hired wage (per hour) 8.50 7.91 11.20
Rent cash cropland expense (acres) 60.00 46.64 75.20
Vegetable totals (dollars, millions) 110.31 233.26 140.21
Wheat production (dollars, millions) 352.67 291.41 138.08
104
Table B.3.11. CO narrowly-defined agriculture average employment
Treated Synthetic Sample Mean
Lagged outcome 8.00 7.98 7.48
Barley for grain (acres) 65547.33 201800.53 41592.84
Land in orchards (acres) 6444.00 132432.44 51262.40
Snap beans harvested for sale, harvested (acres) 590.67 7592.61 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 159.67 155.41 2208.25
Fruits & nuts, pears, all, total acres (acres) 313.67 358.83 227.10
Comm. soil conds. (thousands of treated acres) 4130.86 10667.28 5357.77
Resident population 65 years & over (percent) 10.29 11.41 13.03
Savings institutions - total deposits (thousands) 1210.38 3147.63 2692.87
Civilian labor force unemployment rate (percent) 5.36 5.39 5.53
Federal Government expenditure-grants (millions) 6.04 17.55 10.89
Federal Government insurance (millions) 3.89 52.10 25.58
Resident population: Black alone (percent) 4.25 14.02 13.53
Resident population: Two or more races (percent) 1.79 1.98 1.71
Resident population: Hispanic or Latino Origin (percent) 17.58 16.85 7.41
Resident population: total females (percent) 49.70 50.40 50.93
Social security: retired workers-benefit recipients (thousands) 386.55 1011.87 716.18
Corn Grain Production (dollar, millions) 492.99 435.23 1045.22
Hay production (dollar, millions) 493.11 553.07 260.35
Farm operations (acres, millions) 62.65 149.25 34.27
Labor hired wage (per hour) 8.50 9.84 11.20
Rent cash cropland expense (acres) 60.00 45.00 75.20
Vegetable totals (dollars, millions) 110.31 219.30 140.21
Wheat production (dollars, millions) 352.67 346.31 138.08
105
Table B.3.12. CO narrowly-defined agriculture number of establishments
Treated Synthetic Sample Mean
Lagged outcome 5.05 5.03 4.82
Barley for grain (acres) 65547.33 197582.96 41592.84
Land in orchards (acres) 6444.00 63451.92 51262.40
Snap beans harvested for sale, harvested (acres) 590.67 5331.53 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 159.67 300.51 2208.25
Fruits & nuts, pears, all, total acres (acres) 313.67 194.82 227.10
Comm. soil conds. (thousands of treated acres) 4130.86 8550.48 5357.77
Resident population 65 years & over (percent) 10.29 11.99 13.03
Savings institutions - total deposits (thousands) 1210.38 1793.17 2692.87
Civilian labor force unemployment rate (percent) 5.36 4.97 5.53
Federal Government expenditure-grants (millions) 6.04 10.85 10.89
Federal Government insurance (millions) 3.89 24.73 25.58
Resident population: Black alone (percent) 4.25 10.88 13.53
Resident population: Two or more races (percent) 1.79 2.48 1.71
Resident population: Hispanic or Latino Origin (percent) 17.58 10.44 7.41
Resident population: total females (percent) 49.70 50.41 50.93
Social security: retired workers-benefit recipients (thousands) 386.55 686.91 716.18
Corn Grain Production (dollar, millions) 492.99 1068.64 1045.22
Hay production (dollar, millions) 493.11 425.24 260.35
Farm operations (acres, millions) 62.65 85.34 34.27
Labor hired wage (per hour) 8.50 12.01 11.20
Rent cash cropland expense (acres) 60.00 64.08 75.20
Vegetable totals (dollars, millions) 110.31 158.05 140.21
Wheat production (dollars, millions) 352.67 338.87 138.08
106
Table B.3.13. WA narrowly-defined agriculture average weekly wage per worker
Treated Synthetic Sample Mean
Lagged outcome 6.27 6.30 6.34
Barley for grain (acres) 245385.00 143746.92 41592.84
Land in orchards (acres) 308608.00 167305.66 51262.40
Snap beans harvested for sale, harvested (acres) 3418.67 12799.15 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 1976.33 3479.40 2208.25
Fruits & nuts, pears, all, total acres (acres) 26240.67 183.71 227.10
Comm. soil conds. (thousands of treated acres) 3959.26 4957.35 5357.77
Resident population 65 years & over (percent) 11.55 12.11 13.03
Savings institutions - total deposits (thousands) 3693.15 1313.45 2692.87
Civilian labor force unemployment rate (percent) 6.50 5.21 5.53
Federal Government expenditure-grants (millions) 9.92 10.07 10.89
Federal Government insurance (millions) 7.33 54.63 25.58
Resident population: Black alone (percent) 4.45 18.64 13.53
Resident population: Two or more races (percent) 2.78 1.27 1.71
Resident population: Hispanic or Latino Origin (percent) 9.38 7.05 7.41
Resident population: total females (percent) 50.23 50.67 50.93
Social security: retired workers-benefit recipients (thousands) 622.15 748.42 716.18
Corn Grain Production (dollar, millions) 78.87 305.53 1045.22
Hay production (dollar, millions) 445.95 215.55 260.35
Farm operations (acres, millions) 30.02 46.90 34.27
Labor hired wage (per hour) 9.50 12.33 11.20
Rent cash cropland expense (acres) 136.50 62.55 75.20
Vegetable totals (dollars, millions) 182.97 350.80 140.21
Wheat production (dollars, millions) 782.89 255.36 138.08
107
Table B.3.14. WA narrowly-defined agriculture total quarterly wages
Treated Synthetic Sample Mean
Lagged outcome 17.36 17.35 16.39
Barley for grain (acres) 245385.00 8869.55 41592.84
Land in orchards (acres) 308608.00 207300.29 51262.40
Snap beans harvested for sale, harvested (acres) 3418.67 8706.87 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 1976.33 3410.88 2208.25
Fruits & nuts, pears, all, total acres (acres) 26240.67 378.24 227.10
Comm. soil conds. (thousands of treated acres) 3959.26 10768.68 5357.77
Resident population 65 years & over (percent) 11.55 12.04 13.03
Savings institutions - total deposits (thousands) 3693.15 4829.62 2692.87
Civilian labor force unemployment rate (percent) 6.50 5.40 5.53
Federal Government expenditure-grants (millions) 9.92 21.16 10.89
Federal Government insurance (millions) 7.33 103.23 25.58
Resident population: Black alone (percent) 4.45 9.97 13.53
Resident population: Two or more races (percent) 2.78 1.19 1.71
Resident population: Hispanic or Latino Origin (percent) 9.38 22.69 7.41
Resident population: total females (percent) 50.23 50.47 50.93
Social security: retired workers-benefit recipients (thousands) 622.15 1354.89 716.18
Corn Grain Production (dollar, millions) 78.87 564.14 1045.22
Hay production (dollar, millions) 445.95 581.62 260.35
Farm operations (acres, millions) 30.02 152.02 34.27
Labor hired wage (per hour) 9.50 12.47 11.20
Rent cash cropland expense (acres) 136.50 59.11 75.20
Vegetable totals (dollars, millions) 182.97 345.46 140.21
Wheat production (dollars, millions) 782.89 161.37 138.08
108
Table B.3.15. WA narrowly-defined agriculture average employment
Treated Synthetic Sample Mean
Lagged outcome 8.52 8.52 7.48
Barley for grain (acres) 245385.00 75321.52 41592.84
Land in orchards (acres) 308608.00 81773.67 51262.40
Snap beans harvested for sale, harvested (acres) 3418.67 8773.77 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 1976.33 8330.37 2208.25
Fruits & nuts, pears, all, total acres (acres) 26240.67 403.45 227.10
Comm. soil conds. (thousands of treated acres) 3959.26 13251.21 5357.77
Resident population 65 years & over (percent) 11.55 11.94 13.03
Savings institutions - total deposits (thousands) 3693.15 2691.01 2692.87
Civilian labor force unemployment rate (percent) 6.50 5.78 5.53
Federal Government expenditure-grants (millions) 9.92 15.09 10.89
Federal Government insurance (millions) 7.33 30.89 25.58
Resident population: Black alone (percent) 4.45 8.54 13.53
Resident population: Two or more races (percent) 2.78 1.37 1.71
Resident population: Hispanic or Latino Origin (percent) 9.38 10.79 7.41
Resident population: total females (percent) 50.23 50.45 50.93
Social security: retired workers-benefit recipients (thousands) 622.15 964.12 716.18
Corn Grain Production (dollar, millions) 78.87 2217.22 1045.22
Hay production (dollar, millions) 445.95 554.67 260.35
Farm operations (acres, millions) 30.02 98.75 34.27
Labor hired wage (per hour) 9.50 8.89 11.20
Rent cash cropland expense (acres) 136.50 79.54 75.20
Vegetable totals (dollars, millions) 182.97 135.63 140.21
Wheat production (dollars, millions) 782.89 339.72 138.08
109
Table B.3.16. WA narrowly-defined agriculture number of establishments
Treated Synthetic Sample Mean
Lagged outcome 5.91 5.94 4.82
Barley for grain (acres) 245385.00 12036.46 41592.84
Land in orchards (acres) 308608.00 110787.49 51262.40
Snap beans harvested for sale, harvested (acres) 3418.67 17676.70 8048.64
Fruits & nuts, cherries, tart, total acres (acres) 1976.33 35032.51 2208.25
Fruits & nuts, pears, all, total acres (acres) 26240.67 1001.38 227.10
Comm. soil conds. (thousands of treated acres) 3959.26 5728.86 5357.77
Resident population 65 years & over (percent) 11.55 12.73 13.03
Savings institutions - total deposits (thousands) 3693.15 3733.34 2692.87
Civilian labor force unemployment rate (percent) 6.50 7.50 5.53
Federal Government expenditure-grants (millions) 9.92 14.36 10.89
Federal Government insurance (millions) 7.33 4.51 25.58
Resident population: Black alone (percent) 4.45 13.49 13.53
Resident population: Two or more races (percent) 2.78 1.43 1.71
Resident population: Hispanic or Latino Origin (percent) 9.38 3.78 7.41
Resident population: total females (percent) 50.23 50.86 50.93
Social security: retired workers-benefit recipients (thousands) 622.15 1062.24 716.18
Corn Grain Production (dollar, millions) 78.87 960.76 1045.22
Hay production (dollar, millions) 445.95 305.97 260.35
Farm operations (acres, millions) 30.02 22.56 34.27
Labor hired wage (per hour) 9.50 9.01 11.20
Rent cash cropland expense (acres) 136.50 74.39 75.20
Vegetable totals (dollars, millions) 182.97 161.09 140.21
Wheat production (dollars, millions) 782.89 219.83 138.08
Table B.3.17. CO narrow retail average weekly wage per worker
Treated Synthetic Sample Mean
Lagged outcome 6.65 6.57 6.30
College Graduation Rate (percent) 52.48 52.41 53.63
High School Graduation Rate (percent) 76.25 76.77 75.35
Population Density (people per square mile) 45.97 127.83 204.23
State Unemployment Rate (percent) 5.67 5.72 5.93
GDP per capita (dollars, thousands) 68.59 55.18 59.65
Tobacco Store log average weekly wage per worker 6.12 6.11 6.09
110
Table B.3.18. CO narrow retail total quarterly wages
Treated Synthetic Sample Mean
log total quarterly wages 998 lag 17.11 17.05 16.16
College Graduation Rate (percent) 52.48 52.45 53.63
High School Graduation Rate (percent) 76.25 76.28 75.35
Population Density (people per square mile) 45.97 99.35 204.23
State Unemployment Rate (percent) 5.67 5.66 5.93
GDP per capita (dollars, thousands) 68.59 62.11 59.65
Tobacco Store log total quarterly wages 14.78 14.76 14.66
Table B.3.19. CO narrow retail average employment
Treated Synthetic Sample Mean
log average employment 998 lag 7.90 7.85 7.31
College Graduation Rate (percent) 52.48 52.44 53.63
High School Graduation Rate (percent) 76.25 76.28 75.35
Population Density (people per square mile) 45.97 87.94 204.23
State Unemployment Rate (percent) 5.67 5.48 5.93
GDP per capita (dollars, thousands) 68.59 63.53 59.65
Tobacco Store log average employment 6.10 6.10 6.02
Table B.3.20. CO narrow retail number of establishments
Treated Synthetic Sample Mean
log number of establishments 998 lag 6.28 6.27 5.73
College Graduation Rate (percent) 52.48 52.54 53.63
High School Graduation Rate (percent) 76.25 76.04 75.35
Population Density (people per square mile) 45.97 84.01 204.23
State Unemployment Rate (percent) 5.67 5.49 5.93
GDP per capita (dollars, thousands) 68.59 61.31 59.65
Tobacco Store log number of establishments 4.77 4.74 4.47
Table B.3.21. WA narrow retail average weekly wage per worker
Treated Synthetic Sample Mean
Lagged outcome 6.32 6.33 6.30
College Graduation Rate (percent) 63.07 59.27 53.63
High School Graduation Rate (percent) 73.58 73.90 75.35
Population Density (people per square mile) 95.95 158.82 204.23
State Unemployment Rate (percent) 6.90 6.43 5.93
GDP per capita (dollars, thousands) 70.29 57.73 59.65
Tobacco Store log average weekly wage per worker 6.03 6.20 6.09
111
Table B.3.22. WA narrow retail total quarterly wages
Treated Synthetic Sample Mean
log total quarterly wages 998 lag 16.09 16.07 16.16
College Graduation Rate (percent) 63.07 59.18 53.63
High School Graduation Rate (percent) 73.58 73.89 75.35
Population Density (people per square mile) 95.95 362.27 204.23
State Unemployment Rate (percent) 6.90 6.69 5.93
GDP per capita (dollars, thousands) 70.29 69.04 59.65
Tobacco Store log total quarterly wages 14.52 14.71 14.66
Table B.3.23. WA narrow retail average employment
Treated Synthetic Sample Mean
log average employment 998 lag 7.20 7.24 7.31
College Graduation Rate (percent) 63.07 59.31 53.63
High School Graduation Rate (percent) 73.58 75.02 75.35
Population Density (people per square mile) 95.95 342.54 204.23
State Unemployment Rate (percent) 6.90 6.68 5.93
GDP per capita (dollars, thousands) 70.29 68.95 59.65
Tobacco Store log average employment 5.93 6.11 6.02
Table B.3.24. WA narrow retail number of establishments
Treated Synthetic Sample Mean
log number of establishments 998 lag 5.63 6.14 5.73
College Graduation Rate (percent) 63.07 59.52 53.63
High School Graduation Rate (percent) 73.58 78.49 75.35
Population Density (people per square mile) 95.95 144.57 204.23
State Unemployment Rate (percent) 6.90 6.26 5.93
GDP per capita (dollars, thousands) 70.29 57.48 59.65
Tobacco Store log number of establishments 4.97 4.94 4.47
Table B.3.25. CO broad retail average weekly wage per worker
Treated Synthetic Sample Mean
Lagged outcome 6.25 6.25 6.26
College Graduation Rate (percent) 52.48 52.72 53.63
High School Graduation Rate (percent) 76.25 76.39 75.35
Population Density (people per square mile) 45.97 126.41 204.23
State Unemployment Rate (percent) 5.67 5.76 5.93
GDP per capita (dollars, thousands) 68.59 59.97 59.65
112
Table B.3.26. CO broad retail total quarterly wages
Treated Synthetic Sample Mean
Lagged outcome 20.04 20.04 20.14
College Graduation Rate (percent) 52.48 48.54 53.63
High School Graduation Rate (percent) 76.25 76.28 75.35
Population Density (people per square mile) 45.97 68.04 204.23
State Unemployment Rate (percent) 5.67 5.73 5.93
GDP per capita (dollars, thousands) 68.59 56.62 59.65
Table B.3.27. CO broad retail average employment
Treated Synthetic Sample Mean
Lagged outcome 11.23 11.23 11.31
College Graduation Rate (percent) 52.48 47.95 53.63
High School Graduation Rate (percent) 76.25 76.70 75.35
Population Density (people per square mile) 45.97 55.57 204.23
State Unemployment Rate (percent) 5.67 5.67 5.93
GDP per capita (dollars, thousands) 68.59 56.81 59.65
Table B.3.28. CO broad retail number of establishments
Treated Synthetic Sample Mean
Lagged outcome 8.33 8.33 8.43
College Graduation Rate (percent) 52.48 56.25 53.63
High School Graduation Rate (percent) 76.25 76.25 75.35
Population Density (people per square mile) 45.97 381.19 204.23
State Unemployment Rate (percent) 5.67 5.79 5.93
GDP per capita (dollars, thousands) 68.59 68.52 59.65
Table B.3.29. WA broad retail average weekly wage per worker
Treated Synthetic Sample Mean
Lagged outcome 6.40 6.40 6.26
College Graduation Rate (percent) 63.07 53.43 53.63
High School Graduation Rate (percent) 73.58 70.26 75.35
Population Density (people per square mile) 95.95 264.90 204.23
State Unemployment Rate (percent) 6.90 5.98 5.93
GDP per capita (dollars, thousands) 70.29 70.85 59.65
113
Table B.3.30. WA broad retail total quarterly wages
Treated Synthetic Sample Mean
Lagged outcome 20.43 20.43 20.14
College Graduation Rate (percent) 63.07 60.78 53.63
High School Graduation Rate (percent) 73.58 73.94 75.35
Population Density (people per square mile) 95.95 180.77 204.23
State Unemployment Rate (percent) 6.90 5.17 5.93
GDP per capita (dollars, thousands) 70.29 63.21 59.65
Table B.3.31. WA broad retail average employment
Treated Synthetic Sample Mean
Lagged outcome 11.47 11.47 11.31
College Graduation Rate (percent) 63.07 59.88 53.63
High School Graduation Rate (percent) 73.58 73.84 75.35
Population Density (people per square mile) 95.95 146.39 204.23
State Unemployment Rate (percent) 6.90 5.87 5.93
GDP per capita (dollars, thousands) 70.29 58.93 59.65
Table B.3.32. WA broad retail number of establishments
Treated Synthetic Sample Mean
Lagged outcome 8.48 8.48 8.43
College Graduation Rate (percent) 63.07 59.59 53.63
High School Graduation Rate (percent) 73.58 73.87 75.35
Population Density (people per square mile) 95.95 396.66 204.23
State Unemployment Rate (percent) 6.90 6.50 5.93
GDP per capita (dollars, thousands) 70.29 70.13 59.65
114
APPENDIX C
CHAPTER 4 APPENDIX
C.1 Tables and Figures
Figure C.1. Example bid cost reporting
Actual, 1/1/2013-12/31/2013 Projected, 1/1/2015-12/31/2015
Annualized Allowed Annual Total Allowed
Service Category Util/1000 Avg Cost PMPM Util/1000 Avg Cost PMPM
a. Inpatient Facility             802 $2,358.54 $157.71          813 $2,486.41 $168.47
b. Skilled Nursing Facility             672 427.90 23.97          800 473.02 31.54
c. Home Health             580 192.78 9.32          674 211.30 11.86
d. Ambulance             107 438.99 3.92          125 480.31 4.99
e. DME/Prosthetics/Supplies          9,312 14.40 11.18     10,850 15.78 14.27
f. OP Facility - Emergency             305 747.00 19.00          323 778.66 20.94
g. OP Facility - Surgery             444 2,133.11 78.91          462 2,196.53 84.57
h. OP Facility - Other          5,106 108.59 46.20       5,384 112.78 50.60
i. Professional        14,225 122.25 144.92     15,009 120.85 151.14
j. Part B Rx          1,765 216.40 31.83       1,861 201.07 31.18
k. Other Medicare Part B                 0 0.00 0.00              0 0.00 0.00
l. Transportation (Non-Covered)                 0 0.00 0.00              0 0.00 0.00
m. Dental (Non-Covered)                 0 0.00 0.00          195 88.10 1.43
n. Vision (Non-Covered)                 0 0.00 0.00              0 0.00 0.00
o. Hearing (Non-Covered)                 0 0.00 0.00              0 0.00 0.00
p. Health & Education (Non-Covered)        19,435 4.92 7.96       6,791 5.53 3.13
q. Other Non-Covered          5,347 1.17 0.52       7,256 1.64 0.99
r. COB/Subrg. (outside claim system) 0.00
s. Total Medical Expenses $535.47 $575.12
Notes: This figure is reconstructed from the 2015 bid data for plan H0028-01-0, “Humana Gold
Plus”, offered in Cedar Rapids, Iowa, and surrounding counties, using the CMS public bid data
and the 2015 Bid Pricing Tool (BPT). The ‘Actual’ columns are excerpted from the “MA Base
Period Experience” portion of the BPT. The ‘Projected’ columns are excerpted from the “MA
Projected Allowed Costs” portion.
115
Figure C.2. The distribution of the overall prediction error over time
0
−200
−400
−600
2008 2009 2010 2011 2012 2013
Year
25th percentile mean median 75th percentile
Notes: An observation is a plan-year. Statistics are unweighted. The overall prediction error
is calculated per-member-per-month (PMPM). Negative prediction errors indicate firms
overpredicted their costs.
116
Forecast error ($)
117
Table C.1. Summary statistics
Min 25th Mean Median 75th Max
Prediction errors:
Overall -1515.93 -638.00 -270.20 -89.29 5.89 2228.70
Inpatient Facility -638.93 -243.72 -103.09 -43.10 1.80 778.10
Professional -596.44 -167.62 -77.03 -37.73 1.03 2168.77
Outpatient Facility - Surgery -244.07 -38.12 -15.61 -8.42 3.27 158.45
Firm-chosen product characteristics:
Deductible 0 0 96.15 0 0 5000
Out-of-pocket limit 0 2000 3437.44 3400 5000 12000
Primary care copay 0 5 11.83 10 15 47.5
Specialist copay 0 20 26.09 30 35 60
Hospital copay 0 400 893.79 875 1400 2800
CMS-determined quality rating variables:
Star rating 0.0 0.0 2.48 3.0 3.5 5.0
‘Too new’ indicator 0 0 0.08 0 0 1
‘No data’ indicator 0 0 0.18 0 0 1
Other covariates:
Risk-adjusted payment 312.75 683.11 731.49 726.17 775.25 1134.19
Number of competitors 1.06 7.41 11.72 10.21 15.02 38
Number of competing plans 0 12.38 23.42 20.10 29.90 89
Log previous enrollment 2.48 8.34 9.47 9.69 10.73 13.30
Observations 11,670
Notes: An observation is a plan-year. Statistics are unweighted. Prediction errors and risk-adjusted payments are per-member-per-month
(PMPM). Negative prediction errors indicate firms overpredicted their costs. Hospital copays are for seven day stays. The number of
competitors is calculated at the contract level, weighted by the number of contract enrollees in each county covered by the contract. The
number of competing plans is calculated at the plan level, weighted by the number of plan enrollees in each county covered by the contract.
Table C.2. Correlations between the overall forecast error and its major
components
Overall Inpatient Professional Outpatient Facility
Facility Surgery
Overall 1
Inpatient Facility 0.97 1
Professional 0.84 0.82 1
Outpatient Facility - Surgery 0.94 0.86 0.73 1
Notes: This table presents correlations between the overall forecast error and its three largest (on
average) components. An observation is a plan-year; there are 11,670 observations total. Statistics
are unweighted.
118
Table C.3. The distribution of the overall prediction error across
quartiles of measures of experience and competition
Experience
First Second Third Last
quartile quartile quartile quartile
Outcome: Overall prediction error
Min -1515.93 -1358.61 -1247.43 -1308.29
25th percentile -667.15 -632.43 -570.92 -649.47
Mean -308.36 -257.80 -227.50 -287.14
Median -191.04 -79.68 -58.62 -86.17
75th percentile 0 12.96 9.73 0.55
Max 1276.81 924.99 792.77 2228.70
Competition
First Second Third Last
quartile quartile quartile quartile
Outcome: Overall prediction error
Min -1494.49 -1309.37 -1358.61 -1515.93
25th percentile -590.81 -639.82 -685.87 -614.90
Mean -230.48 -287.67 -307.58 -255.08
Median -56.11 -124.33 -131.33 -81.24
75th percentile 13.27 7.53 -2.87 7.77
Max 1276.81 2278.70 792.77 1023.31
Notes: An observation is a plan-year. Statistics are unweighted. The overall prediction error
is calculated per-member-per-month (PMPM). Negative prediction errors indicate firms
overpredicted their costs. Our measure of experience is the log of the total number of previous
enrollees at the contract level. Our measure of competition is the number of competitors
calculated at the contract level, weighted by the number of contract enrollees in each county
covered by the contract.
119
120
Table C.4. The association between prediction errors and plan observables
Dependent variable: Prediction errors
Overall Inpatient facility OP Facility - Surgery Professional
(1) (2) (3) (4)
Overall prediction error 0.002∗∗∗ 0.001∗∗∗t−1 0.0001
∗∗∗ 0.001∗∗∗
(0.0001) (0.00003) (0.00001) (0.00002)
Log previous enrollment 2.945 −1.907∗ −0.572∗∗∗ 3.122∗∗∗
(2.461) (1.018) (0.205) (0.756)
Number of competitors 6.413∗∗∗ 3.199∗∗∗ 0.267∗∗ 1.661∗∗∗
(1.403) (0.580) (0.117) (0.431)
Number of competing plans −0.902∗∗ −0.430∗∗∗ −0.059∗ −0.229∗∗
(0.367) (0.152) (0.031) (0.113)
Plan age 46.284∗∗∗ 17.676∗∗∗ 4.218∗∗∗ 11.924∗∗∗
(9.357) (3.869) (0.779) (2.872)
Constant −811.990∗∗∗ −295.974∗∗∗ −27.565 −292.505∗∗∗
(206.648) (85.458) (17.196) (63.439)
Year FE Yes Yes Yes Yes
State FE Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes
Observations 11,530 11,530 11,530 11,530
R2 0.257 0.238 0.239 0.265
Adjusted R2 0.232 0.213 0.214 0.241
Notes: An observation is a plan-year. Estimates are obtained using unweighted OLS. Stars indicate p-values: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
121
Table C.5. The association between absolute prediction errors and plan observables
Dependent variable: Prediction errors
|Overall| |Inpatient facility | |OP Facility - Surgery | |Professional |
(1) (2) (3) (4)
Overall prediction error −0.001∗∗∗ −0.001∗∗∗t−1 −0.0001∗∗∗ −0.0005∗∗∗
(0.0001) (0.00003) (0.00001) (0.00002)
Log previous enrollment −6.270∗∗∗ −1.580∗ −0.083 −1.883∗∗∗
(2.170) (0.855) (0.158) (0.640)
Number of competitors −5.345∗∗∗ −2.363∗∗∗ −0.281∗∗∗ −0.956∗∗∗
(1.236) (0.487) (0.090) (0.365)
Number of competing plans 0.634∗ 0.336∗∗∗ −0.023 0.316∗∗∗
(0.324) (0.128) (0.024) (0.096)
Plan age −43.944∗∗∗ −15.107∗∗∗ −4.044∗∗∗ −7.898∗∗∗
(8.248) (3.252) (0.602) (2.434)
Constant 888.563∗∗∗ 314.236∗∗∗ 68.322∗∗∗ 246.858∗∗∗
(182.162) (71.815) (13.303) (53.755)
Year FE Yes Yes Yes Yes
State FE Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes
Observations 11,530 11,530 11,530 11,530
R2 0.261 0.259 0.218 0.282
Adjusted R2 0.236 0.234 0.191 0.258
Notes: An observation is a plan-year. Estimates are obtained using unweighted OLS. Stars indicate p-values: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
122
Table C.6. The association between plan benefits and the overall prediction error
Dependent variable:
Deductible OOP limit Primary care copay Specialist copay Hospital copay
(1) (2) (3) (4) (5)
Overall prediction errort−1 0.012
∗ 0.373∗∗∗ −0.001∗∗∗ 0.001∗∗∗ 0.008
(0.007) (0.052) (0.0002) (0.0002) (0.013)
Star rating −11.060 −1.900 0.319∗ −0.882∗∗∗ −37.499∗∗∗
(6.746) (52.015) (0.182) (0.246) (12.696)
Star ‘too new’ indicator −43.882∗ 784.931∗∗∗ 0.545 −3.279∗∗∗ −67.160
(22.888) (176.471) (0.616) (0.835) (43.072)
Star ‘no data’ indicator −56.314∗∗ 388.196∗∗ 1.477∗∗ −3.496∗∗∗ −150.724∗∗∗
(21.984) (169.503) (0.592) (0.802) (41.372)
Risk-adjusted payment 0.369∗∗∗ 0.588∗ 0.009∗∗∗ −0.006∗∗∗ 0.010
(0.041) (0.314) (0.001) (0.001) (0.077)
Constant −194.165 1, 169.898 −1.320 29.584∗∗∗ 983.983∗∗∗
(141.780) (1, 093.145) (3.816) (5.174) (266.812)
Year FE Yes Yes Yes Yes Yes
State FE Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes
Mean Dependent Variable 96.14722 3437.44 11.83048 26.0924 893.788
Observations 11,670 11,670 11,670 11,670 11,670
R2 0.566 0.514 0.476 0.475 0.447
Adjusted R2 0.551 0.498 0.459 0.458 0.429
Notes: An observation is a plan-year. The overall prediction error is defined as actual costs for the year minus the projected costs at the
time of plan submission. Estimates are obtained using unweighted OLS. Stars indicate p-values: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
123
Table C.7. The association between plan benefits and the absolute overall prediction error
Dependent variable:
Deductible OOP Limit Primary care copay Specialist copay Hospital copay
(1) (2) (3) (4) (5)
|Prediction Error| 0.002 −0.483∗∗∗ 0.0004∗ −0.002∗∗∗ −0.053∗∗∗t−1
(0.008) (0.059) (0.0002) (0.0003) (0.014)
Star rating −10.551 −5.256 0.306∗ −0.917∗∗∗ −39.096∗∗∗
(6.748) (51.987) (0.182) (0.246) (12.690)
Star ‘too new’ indicator −43.005∗ 780.005∗∗∗ 0.521 −3.335∗∗∗ −69.828
(22.892) (176.355) (0.616) (0.834) (43.048)
Star ‘no data’ indicator −55.382∗∗ 390.855∗∗ 1.445∗∗ −3.528∗∗∗ −152.776∗∗∗
(21.985) (169.373) (0.592) (0.801) (41.344)
Risk-adjusted payment 0.355∗∗∗ 0.636∗∗ 0.010∗∗∗ −0.005∗∗∗ 0.050
(0.041) (0.314) (0.001) (0.001) (0.077)
Constant −191.285 1, 215.398 −1.455 29.616∗∗∗ 981.348∗∗∗
(141.791) (1, 092.351) (3.818) (5.167) (266.643)
Year FE Yes Yes Yes Yes Yes
State FE Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes
Mean Dependent Variable 96.14722 3437.44 11.83048 26.0924 893.788
Observations 11,670 11,670 11,670 11,670 11,670
R2 0.566 0.515 0.475 0.477 0.448
Adjusted R2 0.551 0.499 0.458 0.459 0.429
Notes: An observation is a plan-year. The overall prediction error is defined as actual costs for the year minus the projected costs at the
time of plan submission. Estimates are obtained using unweighted OLS. Stars indicate p-values: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
REFERENCES CITED
Abadie, A., Diamond, A., & Hainmueller, J. (2010). Synthetic control methods for
comparative case studies: Estimating the effect of california’s tobacco
control program. Journal of the American statistical Association, 105 (490),
493–505.
Abadie, A., Diamond, A., & Hainmueller, J. (2015). Comparative politics and the
synthetic control method. American Journal of Political Science, 59 (2),
495–510.
Abadie, A., & Gardeazabal, J. (2003). The economic costs of conflict: A case study
of the basque country. American economic review , 93 (1), 113–132.
Abaluck, J., Caceres Bravo, M., Hull, P., & Starc, A. (2021). Mortality effects and
choice across private health insurance plans. The quarterly journal of
economics , 136 (3), 1557–1610.
Abouk, R., Ghimire, K. M., Maclean, J. C., & Powell, D. (2021). Does marijuana
legalization affect work capacity? evidence from workers’ compensation
benefits (Tech. Rep.). National Bureau of Economic Research.
Aizpurua-Olaizola, O., Soydaner, U., Öztürk, E., Schibano, D., Simsir, Y., Navarro,
P., . . . Usobiaga, A. (2016). Evolution of the cannabinoid and terpene
content during the growth of cannabis sativa plants from different
chemotypes. Journal of Natural Products , 79 (2), 324-331. Retrieved from
https://doi.org/10.1021/acs.jnatprod.5b00949 (PMID: 26836472)
doi: 10.1021/acs.jnatprod.5b00949
Altman, E. I. (1968). Financial ratios, discriminant analysis and the prediction of
corporate bankruptcy. The journal of finance, 23 (4), 589–609.
Altman, E. I. (1973). Predicting railroad bankruptcies in america. The Bell
Journal of Economics and Management Science, 184–211.
Alvarez, R., & Görg, H. (2009, January). Multinationals and plant exit: Evidence
from Chile. International Review of Economics & Finance, 18 (1), 45-51.
Andrews, L. (2017, Jan). Legal weed is hard to come by in alaska. Anchorage Daily
News . Retrieved from https://www.adn.com/alaska-marijuana/2017/01/
04/legal-weed-is-hard-to-come-by-in-alaska/
Audretsch, D. B. (1991). New-firm survival and the technological regime. The
Review of Economics and Statistics , 73 (3), 441–450.
124
Audretsch, D. B. (1995). The propensity to exit and innovation. Review of
Industrial Organization, 10 (5), 589–605.
Audretsch, D. B., Houweling, P., & Thurik, A. R. (2000). Firm survival in the
netherlands. Review of Industrial Organization, 16 (1), 1–11.
Audretsch, D. B., & Mahmood, T. (1995). New firm survival: new results using a
hazard function. The Review of Economics and Statistics , 77 (1), 97–103.
Aydelotte, J. D., Brown, L. H., Luftman, K. M., Mardock, A. L., Teixeira, P. G.,
Coopwood, B., & Brown, C. V. (2017). Crash fatality rates after
recreational marijuana legalization in washington and colorado. American
journal of public health, 107 (8), 1329–1331.
Baggio, M., Chong, A., & Kwon, S. (2018). Marijuana and alcohol: Evidence using
border analysis and retail sales data. Canadian Journal of Economics/Revue
canadienne d’économique.
Baicker, K., & Goldman, D. (2011). Patient cost-sharing and healthcare spending
growth. Journal of Economic Perspectives , 25 (2), 47–68.
Bampasidou, M., & Salassi, M. E. (2019). Agricultural labor trends. Journal of
ASFMRA, 6–13.
Bandick, R. (2007). Multinationals and plant survival in Swedish manufacturing
(Discussion Papers No. 07/31). University of Nottingham, GEP.
Barcott, B., & Whitney, B. (2019). Special report: cannabis jobs count. Leafly.
Bauhoff, S. (2014). The effect of school district nutrition policies on dietary intake
and overweight: a synthetic control approach. Economics & Human Biology ,
12 , 45–55.
Berenson, R. A., Sunshine, J. H., Helms, D., & Lawton, E. (2015). Why medicare
advantage plans pay hospitals traditional medicare prices. Health Affairs ,
34 (8), 1289–1295.
Berger, E., & Seegert, N. (2020). Half banked: The real effects of financial
exclusion on firms (Tech. Rep.). Working paper.
Bernard, A. B., & Jensen, J. B. (2007). Firm structure, multinationals, and
manufacturing plant deaths. The Review of Economics and Statistics , 89 (2),
193–204.
Berry, S., Levinsohn, J., & Pakes, A. (1995). Automobile prices in market
equilibrium. Econometrica, 63 (4), pp. 841-890.
125
Billmeier, A., & Nannicini, T. (2013). Assessing economic liberalization episodes:
A synthetic control approach. Review of Economics and Statistics , 95 (3),
983–1001.
Blakely, E. J., & Leigh, N. G. (2013). Planning local economic development. Sage.
Blonigen, B. A., Liebman, B. H., & Wilson, W. W. (2007, December). Trade Policy
and Market Power: The Case of the US Steel Industry (NBER Working
Papers). National Bureau of Economic Research, Inc.
Börgers, T., & Sarin, R. (1997). Learning through reinforcement and replicator
dynamics. Journal of economic theory , 77 (1), 1–14.
Bram, J., & Gorton, N. (2017). How is online shopping affecting retail
employment? (Tech. Rep.). Federal Reserve Bank of New York.
Brewin, D. G., Schmitz, T. G., Nolan, J. F., & Gray, R. S. (2017). Grain
transportation policy reform in western canada. Canadian Journal of
Agricultural Economics , 65 (4), 643–665.
Brigham, E. F. (1975). Hurdle rates for screening capital expenditure proposals.
Financial Management , 17–26.
Butters, R. A., Sacks, D. W., & Seo, B. (2020). How do national firms respond to
local shocks? Available at SSRN 3535206 .
Cabral, M., Geruso, M., & Mahoney, N. (2018). Does privatized medicare benefit
patients or producers? evidence from the benefits improvement and
protection act. American Economic Review .
Camerer, C. F., Ho, T.-H., & Chong, J.-K. (2002). Sophisticated
experience-weighted attraction learning and strategic teaching in repeated
games. Journal of Economic theory , 104 (1), 137–188.
Canadian Transportation Agency. (2007). Decision No.344-R-2007
https://otc-cta.gc.ca/eng/ruling/344-r-2007.
Cannabis Business Times. (2020). State of the industry report (Tech. Rep.).
Cannabis Business Times. Retrieved from
https://www.cannabisbusinesstimes.com/article/
2020-state-of-the-industry-report-nexus-greenhouse-systems/
Caulkins, J. P. (2010). Estimated cost of production for legalized cannabis (Tech.
Rep.). RAND Drug Policy Research Center.
Cervantes, J. (2006). Marijuana horticulture: The indoor/outdoor medical grower’s
bible. Van Patten Publishing.
126
Chakraborty, A., Doremus, J., & Stith, S. (2020). The effects of recreational
cannabis access on the labor market: Evidence from colorado (Tech. Rep.).
Chamon, M., Garcia, M., & Souza, L. (2017). Fx interventions in brazil: a
synthetic control approach. Journal of International Economics , 108 ,
157–168.
Chan, N. W., Burkhardt, J., & Flyr, M. (2020). The effects of recreational
marijuana legalization and dispensing on opioid mortality. Economic
Inquiry , 58 (2), 589–606.
Chandra, S., Lata, H., Khan, I. A., & ElSohly, M. A. (2017). Cannabis sativa l.:
Botany and horticulture. In S. Chandra, H. Lata, & M. A. ElSohly (Eds.),
Cannabis sativa l. - botany and biotechnology (pp. 79–100). Cham: Springer
International Publishing. Retrieved from
https://doi.org/10.1007/978-3-319-54564-6 3 doi:
10.1007/978-3-319-54564-6 3
Chelwa, G., van Walbeek, C., & Blecher, E. (2017). Evaluating south africa’s
tobacco control policy using a synthetic control method. Tobacco Control ,
26 (5), 509–517.
Chen, M.-Y. (2002). Survival duration of plants: Evidence from the us petroleum
refining industry. International Journal of Industrial Organization, 20 (4),
517–555.
Coibion, O., Gorodnichenko, Y., & Kumar, S. (2018). How do firms form their
expectations? new survey evidence. American Economic Review , 108 (9),
2671–2713.
Colombo, M. G., & Delmastro, M. (2000). A note on the relation between size,
ownership status and plant’s closure: sunk costs vs. strategic size liability.
Economics Letters , 69 (3), 421–427.
Commission, M. P. A., et al. (2021). The medicare advantage program: status
report. In Report to the congress: Medicare payment policy.
Crawford, V. P. (2003). Lying for strategic advantage: Rational and boundedly
rational misrepresentation of intentions. American Economic Review , 93 (1),
133–149.
Curto, V., Einav, L., Levin, J., & Bhattacharya, J. (2021). Can health insurance
competition work? evidence from medicare advantage. Journal of Political
Economy , 129 (2), 570–606.
127
de Figueiredo, J. M., & Silverman, B. S. (2012). Firm survival and industry
evolution in vertically related populations. Management Science, 58 (9),
1632–1650.
De Vany, A. S., & Saving, T. R. (1977). Product quality, uncertainty, and
regulation: The trucking industry. The American Economic Review , 67 (4),
583–594.
Doraszelski, U., & Markovich, S. (2007). Advertising dynamics and competitive
advantage. The RAND Journal of Economics , 38 (3), 557–592.
Dragone, D., Prarolo, G., Vanin, P., & Zanella, G. (2019). Crime and the
legalization of recreational marijuana. Journal of economic behavior &
organization, 159 , 488–501.
Dranove, D., & Satterthwaite, M. A. (2000). The industrial organization of health
care markets. Handbook of health economics , 1 , 1093–1139.
Duggan, M., Starc, A., & Vabson, B. (2016). Who benefits when the government
pays more? pass-through in the medicare advantage program. Journal of
Public Economics , 141 , 50 - 67. Retrieved from http://
www.sciencedirect.com/science/article/pii/S0047272716300767 doi:
https://doi.org/10.1016/j.jpubeco.2016.07.003
Duncan, D., Ross, J. M., & Mikesell, J. L. (2019). Can the federal budget process
promote fiscal sustainability? evidence from synthetic control using lasso.
Evidence from Synthetic Control Using Lasso (December 4, 2019).
Dunne, P., & Hughes, A. (1994, June). Age, Size, Growth and Survival: UK
Companies in the 1980s. Journal of Industrial Economics , 42 (2), 115-140.
Dunne, T., Klimek, S. D., & Roberts, M. J. (2005). Exit from regional
manufacturing markets: The role of entrant experience. International
Journal of Industrial Organization, 23 (5-6), 399–421.
Dunne, T., Roberts, M. J., & Samuelson, L. (1988, Winter). Patterns of Firm
Entry and Exit in U.S. Manufacturing Industries. RAND Journal of
Economics , 19 (4), 495-515.
Dunne, T., Roberts, M. J., & Samuelson, L. (1989). The growth and failure of us
manufacturing plants. The Quarterly Journal of Economics , 104 (4),
671–698.
Elston, J. A., & Agarwal, R. (2004). Bank influence, firm performance and
survival: Empirical evidence from germany 1970-1986. Corporate Ownership
and Control , 1 (2), 65–71.
128
Erev, I., & Roth, A. E. (1998). Predicting how people play games: Reinforcement
learning in experimental games with unique, mixed strategy equilibria.
American economic review , 848–881.
Ericson, R., & Pakes, A. (1995). Markov-perfect industry dynamics: A framework
for empirical work. The Review of Economic Studies , 62 (1), 53–82.
Escalante, C. L., & Luo, T. (2017). Sustaining a healthy farm labor force: issues
for policy consideration. Choices , 32 (1), 1–9.
Ferman, B. (2021). On the properties of the synthetic control estimator with many
periods and many controls. Journal of the American Statistical Association,
116 (536), 1764–1772.
Ferman, B., & Pinto, C. (2021). Synthetic controls with imperfect pretreatment fit.
Quantitative Economics , 12 (4), 1197–1221.
Ferman, B., Pinto, C., & Possebom, V. (2020). Cherry picking with synthetic
controls. Journal of Policy Analysis and Management , 39 (2), 510–532.
Figueroa, J. F., Wadhera, R. K., Frakt, A. B., Fonarow, G. C., Heidenreich, P. A.,
Xu, H., . . . others (2020). Quality of care and outcomes among medicare
advantage vs fee-for-service medicare patients hospitalized with heart failure.
JAMA cardiology , 5 (12), 1349–1357.
Firpo, S., & Possebom, V. (2016). Synthetic control estimator: A generalized
inference procedure and confidence sets. Sao Paulo School of Economics,
EESP-FGV .
Franklin, P. J. (1974). Some observations on exit from the motor insurance
industry, 1966-1972. The Journal of Industrial Economics , 22 (4), 299–313.
Friedman, J., Hastie, T., & Tibshirani, R. (2010). Regularization paths for
generalized linear models via coordinate descent. Journal of statistical
software, 33 (1), 1.
Frittelli, J. (2005). Grain transport: modal trends and infrastructure implications..
Fudenberg, D., & Kreps, D. M. (1993). Learning mixed equilibria. Games and
economic behavior , 5 (3), 320–367.
Fudenberg, D., & Tirole, J. (1986, July). A Theory of Exit in Duopoly.
Econometrica, 54 (4), 943-960.
Gaynor, M., Ho, K., & Town, R. J. (2015). The industrial organization of
health-care markets. Journal of Economic Literature, 53 (2), 235–84.
129
Ghemawat, P., & Nalebuff, B. (1985, Summer). Exit. RAND Journal of
Economics , 16 (2), 184-194.
Ghemawat, P., & Nalebuff, B. (1990). The devolution of declining industries. The
Quarterly Journal of Economics , 105 (1), 167–186.
Ghimire, K. M., & Maclean, J. C. (2020). Medical marijuana and workers’
compensation claiming. Health economics , 29 (4), 419–434.
Gibson, J. K., & Harris, R. I. D. (1996, August). Trade Liberalisation and Plant
Exit in New Zealand Manufacturing. The Review of Economics and
Statistics , 78 (3), 521-529.
Graham, J. R., Harvey, C. R., & Rajgopal, S. (2005). The economic implications of
corporate financial reporting. Journal of accounting and economics , 40 (1-3),
3–73.
Hall, W., & Lynskey, M. (2016). Evaluating the public health impacts of legalizing
recreational cannabis use in the united states. Addiction, 111 (10),
1764–1773.
Hansen, B., Miller, K., Seo, B., & Weber, C. (2020). Taxing the potency of sin
goods: Evidence from recreational cannabis and liquor markets. National
Tax Journal, Forthcoming .
Hansen, B., Miller, K., & Weber, C. (2017). The taxation of recreational
marijuana: Evidence from washington state (Tech. Rep.). National Bureau
of Economic Research.
Hansen, B., Miller, K., & Weber, C. (2018). Auditing and enforcement in the
recreational marijuana industry. In 111th annual conference on taxation.
Hansen, B., Miller, K., & Weber, C. (2020a). Early evidence on recreational
marijuana legalization and traffic fatalities. Economic inquiry , 58 (2),
547–568.
Hansen, B., Miller, K., & Weber, C. (2020b). Federalism, partial prohibition, and
cross-border sales: Evidence from recreational marijuana. Journal of Public
Economics , 187 , 104159.
Hansen, B., Miller, K., & Weber, C. (2021a). Up in smoke? the market for
cannabis. In K. F. Zimmermann (Ed.), Handbook of labor, human resources
and population economics (pp. 1–29). Cham: Springer International
Publishing. Retrieved from
https://doi.org/10.1007/978-3-319-57365-6 138-1 doi:
10.1007/978-3-319-57365-6 138-1
130
Hansen, B., Miller, K. S., & Weber, C. (2021b). Vertical integration and production
inefficiency in the presence of a gross receipts tax (Tech. Rep.). National
Bureau of Economic Research.
Hao, Z., & Cowan, B. W. (2020). The cross-border spillover effects of recreational
marijuana legalization. Economic inquiry , 58 (2), 642–666.
Hart, S., & Mas-Colell, A. (2000). A simple adaptive procedure leading to
correlated equilibrium. Econometrica, 68 (5), 1127–1150.
Hollenbeck, B., & Uetake, K. (2019). Taxation and market power in the legal
marijuana industry. Available at SSRN 3237729 .
Holmes, S. (2013). Fresh fruit, broken bodies: Migrant farmworkers in the united
states (Vol. 27). Univ of California Press.
Huang, W. (2003). Shuttle train adoption strategy (Unpublished master’s thesis).
North Dakota State University, Fargo, North Dakota. (North Dakota State
University Press)
Hughes, L. A., Schaible, L. M., & Jimmerson, K. (2020). Marijuana dispensaries
and neighborhood crime and disorder in denver, colorado. Justice Quarterly ,
37 (3), 461–485.
Ifrach, B., & Weintraub, G. Y. (2016, 09). A Framework for Dynamic Oligopoly in
Concentrated Industries. The Review of Economic Studies , 84 (3),
1106-1150. Retrieved from https://doi.org/10.1093/restud/rdw047 doi:
10.1093/restud/rdw047
Jena, A. B., & Philipson, T. J. (2008). Cost-effectiveness analysis and innovation.
Journal of health economics , 27 (5), 1224–1236.
Jordan, M. (2020, Apr). Farmworkers, mostly undocumented, become ’essential’
during pandemic. The New York Times . Retrieved from
https://www.nytimes.com/2020/04/02/us/
coronavirus-undocumented-immigrant-farmworkers-agriculture.html
Jovanovic, B. (1982, May). Selection and the Evolution of Industry. Econometrica,
50 (3), 649-670.
Kerr, D. C., Bae, H., Phibbs, S., & Kern, A. C. (2017). Changes in
undergraduates’ marijuana, heavy alcohol and cigarette use following
legalization of recreational marijuana use in oregon. Addiction, 112 (11),
1992–2001.
131
Keys, A. (2020, January 17). Gov. lujan grisham unveils recreational marijuana
legalization proposal. KRQE Albequerque. Retrieved 2020/07/03, from
https://www.krqe.com/news/politics-government/gov-lujan-grisham
-unveils-recreational-marijuana-legalization-proposal/
Kilmer, B. (2019). How will cannabis legalization affect health, safety, and social
equity outcomes? it largely depends on the 14 ps. The American journal of
drug and alcohol abuse, 45 (6), 664–672.
Kim, H. (2021). Heterogeneous impacts of cost shocks, strategic bidding and
pass-through: Evidence from the new england electricity market. American
Economic Journal: Microeconomics .
Kimura, F., & Kiyota, K. (2006). Exports, fdi, and productivity: Dynamic
evidence from japanese firms. Review of World Economics , 142 (4), 695–719.
Krissman, F. (2017). America’s largest cannabis labor market. Humboldt State
University, Department of Anthropology.
Larson, E. K., & James, N. F. (2007). Assessing the Effects of Road Pricing on an
Industry: The Grain Handling and Transportation System in Saskatchewan.
Journal of the Transportation Research Forum, 46 (3), 1-19. doi:
10.22004/ag.econ.206895
Larson, E. K., & Nolan, J. F. (2007). Assessing the effects of road pricing on an
industry: the grain handling and transportation system in saskatchewan.
Journal of the Transportation Research Forum, 46 , 69–86.
Lawrence, R., Nolan, J., & Schoney, R. (2016). Simulating contestability in freight
transportation: the canadian grain handling and transportation system.
Journal of Transport Economics and Policy , 50 , 1–25.
Lieberman, M. B. (1990, Winter). Exit from declining industries: ‘shakeout’ or
‘stakeout’? RAND Journal of Economics , 21 (4), 538-554.
Liesch, P. W., & Knight, G. A. (1999). Information internalization and hurdle
rates in small and medium enterprise internationalization. Journal of
International Business Studies , 30 (2), 383–394.
Liu, Z.-N., Yu, X.-Y., Jia, L.-F., Wang, Y.-S., Song, Y.-C., & Meng, H.-D. (2021).
The influence of distance weight on the inverse distance weighted method for
ore-grade estimation. Scientific Reports , 11 (1), 1–8.
Luckstead, J., & Devadoss, S. (2019). The importance of h-2a guest workers in
agriculture. Choices , 34 (1), 1–8.
132
Maclean, J. C., Ghimire, K. M., & Nicholas, L. H. (2021). Marijuana legalization
and disability claiming. Health Economics , 30 (2), 453–469.
Mata, J., Portugal, P., & Guimaraes, P. (1995). The survival of new plants:
Start-up conditions and post-entry evolution. International Journal of
Industrial Organization, 13 (4), 459–481.
McGuire, T. G., Newhouse, J. P., & Sinaiko, A. D. (2011). An economic history of
medicare part c. The Milbank Quarterly , 89 (2), 289–332.
Meyers, D. J., & Trivedi, A. N. (2021). Medicare advantage chart reviews are
associated with billions in additional payments for some plans. Medical
Care, 59 (2), 96–100.
Meyers, D. J., Trivedi, A. N., Wilson, I. B., Mor, V., & Rahman, M. (2021). Higher
medicare advantage star ratings are associated with improvements in patient
outcomes: Study examines how medicare quality ratings impact patient
outcomes for those enrolled in medicare advantage. Health Affairs , 40 (2),
243–250.
Michalowsky, B., Hoffmann, W., Kennedy, K., & Xie, F. (2020). Is the whole larger
than the sum of its parts? impact of missing data imputation in economic
evaluation conducted alongside randomized controlled trials. The European
Journal of Health Economics , 21 (5), 717–728.
Milgrom, P., & Roberts, J. (1991). Adaptive and sophisticated learning in normal
form games. Games and economic Behavior , 3 (1), 82–100.
Miller, A. M., Rosenman, R., & Cowan, B. W. (2017). Recreational marijuana
legalization and college student use: Early evidence. SSM-Population
Health, 3 , 649–657.
Miller, K., & Seo, B. (2021). The effect of cannabis legalization on substance
demand and tax revenues. National Tax Journal , 74 (1), 107–145.
Miller, K. S., Petrin, A., Town, R., & Chernew, M. (2021). Optimal managed
competition subsidies (Tech. Rep.). National Bureau of Economic Research.
Miller, K. S., & Wilson, W. W. (2018). Governance structure and exit: Evidence
from california hospitals. Review of Industrial Organization, 53 (1), 31–55.
Motel, S. (2015). Six facts about marijuana. Pew Research Fact Tank . Retrieved
10/22/2016, from http://www.pewresearch.org/fact-tank/2015/04/14/
6-facts-about-marijuana/
133
Murray, J. S., & Reiter, J. P. (2016). Multiple imputation of missing categorical
and continuous values via bayesian mixture models with local dependence.
Journal of the American Statistical Association, 111 (516), 1466–1479.
Neumark, D., Zhang, J., & Ciccarella, S. (2008). The effects of wal-mart on local
labor markets. Journal of Urban Economics , 63 (2), 405–430.
Nicholas, L. H., & Maclean, J. C. (2019). The effect of medical marijuana laws on
the health and labor supply of older adults: Evidence from the health and
retirement study. Journal of Policy Analysis and Management , 38 (2),
455–480.
Nolan, J. (2007). Saskatchewan: Geographic perspectives. In Transportation
(chap. 19). Regina, SK: Canadian Plains Research Centre Press.
Nolan, J. F., & Kerr, W. A. (2012). The estey review’s contribution to
transportation policy. Journal of International Law and Trade Policy , 13 (2),
243–262.
Park, S., White, L., Fishman, P., Larson, E. B., & Coe, N. B. (2020). Health care
utilization, care satisfaction, and health status for medicare advantage and
traditional medicare beneficiaries with and without alzheimer disease and
related dementias. JAMA network open, 3 (3), e201809–e201809.
Prater, M. E., Sparger, A., Bahizi, P., & O’Neil, D. J. (2013). Rail Market Share of
Grain and Oilseed Transportation. Journal of the Transportation Research
Forum, 52 (2), 1-25. doi: 10.22004/ag.econ.207350
Reynolds, S. S. (1988). Plant closings and exit behaviour in declining industries.
Economica, 55 (220), 493–503.
Rutström, E. E., & Wilcox, N. T. (2009). Stated beliefs versus inferred beliefs: A
methodological inquiry and experimental test. Games and Economic
Behavior , 67 (2), 616–632.
Sabia, J. J., & Nguyen, T. T. (2018). The effect of medical marijuana laws on
labor market outcomes. The Journal of Law and Economics , 61 (3),
361–396.
Sacirbey, O. (2016, Dec). Testing, supply issues ’could be death knell’ for oregon
mj market. MJBizDaily . Retrieved from https://mjbizdaily.com/
testing-supply-issues-could-be-death-knell-for-oregon-mj-market/
Sarmiento, C., & Wilson, W. W. (2005). Spatial modeling in technology adoption
decisions: The case of shuttle train elevators. American Journal of
Agricultural Economics , 87 (4), 1034–1045.
134
Schaneman, B. (2019). Marijuana, hemp cross-pollination a budding conflict. AP
News . Retrieved from
https://apnews.com/article/wa-state-wire-ca-state-wire-jay
-inslee-cannabis-marijuana-eded257dc4de4ce19a5388a7acbf7c38
Schmiesing, B. H., Blank, S. C., & Gunn, S. P. (1985). The influence of
technological change on grain elevator pricing efficiency. North Central
Journal of Agricultural Economics , 7 (2), 95–107.
Segarra, A., & Callejón, M. (2002). New firms’ survival and market turbulence:
New evidence from spain. Review of Industrial Organization, 20 (1), 1–14.
Selyem, B. K. (2000). The legacy of country grain elevators: A photo essay.
Kansas History , 23 (1-2).
Shepard, H., Parker, J., Darby, P., & Ainsworth, C. C. (1999). Sex expression in
hop (humulus lupulus l. and h. japonicus sieb. et zucc.): floral morphology
and sex chromosomes. Sex determination in plants , 8 , 139–50.
Sills, E. O., Herrera, D., Kirkpatrick, A. J., Brandão Jr, A., Dickson, R., Hall, S.,
. . . others (2015). Estimating the impacts of local policy innovation: the
synthetic control method applied to tropical deforestation. PloS one, 10 (7),
e0132590.
Simmins, G. (2004). Prairie grain elevators: An old purpose in search of a new
form. Challenging Frontiers , 205-230.
Smart, R., & Pacula, R. L. (2019). Early evidence of the impact of cannabis
legalization on cannabis use, cannabis use disorder, and the use of other
substances: findings from state policy evaluations. The American journal of
drug and alcohol abuse, 45 (6), 644–663.
Smith, V. M., Powell, M., Mungeam, D., & Emmons, R. G. (2019). Stakeholder
perceptions of the impact of cannabis production on the southern oregon
food system. Journal of Agriculture, Food Systems, and Community
Development , 8 (4), 125–135.
Song, Z., Landrum, M. B., & Chernew, M. E. (2013). Competitive bidding in
medicare advantage: Effect of benchmark changes on plan bids. Journal of
health economics , 32 (6), 1301–1312.
Starfield, B., Shi, L., & Macinko, J. (2005). Contribution of primary care to health
systems and health. The milbank quarterly , 83 (3), 457–502.
135
Stoicheff, L. (2018, April 3). The impact of santa cruz county agriculture’s labor
shortage. Good Times Santa Cruz . Retrieved 2020/07/03, from
https://goodtimes.sc/cover-stories/
impact-santa-cruz-county-agriculture-labor-shortage/
Thomas, D. (2018). License quotas and the inefficient regulation of sin goods:
Evidence from the washington recreational marijuana market. Available at
SSRN 3312960 .
Thomas-Lycklama-Nijeholt, G. (2012). On the road for work: Migratory workers on
the east coast of the united states (Vol. 7). Springer Science & Business
Media.
Thraves, B. D. (2007). Saskatchewan: Geographic perspectives (No. 52). University
of Regina Press.
Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of
the Royal Statistical Society: Series B (Methodological), 58 (1), 267–288.
Ullman, D. F. (2017). The effect of medical marijuana on sickness absence. Health
economics , 26 (10), 1322–1327.
Vachal, K., & Bitzan, J. (1997). Implications of a north american grain marketing
system for prairie transportation and elevators (Tech. Rep.). Upper Great
Plains Transportation Institute, North Dakota State University.
Valachovic, Y., Quinn-Davidson, L., Stackhouse, J., Butsic, V., et al. (2019).
Perceptions of cannabis among humboldt county timberland and ranchland
owners. California Agriculture, 73 (3), 161–168.
Van Beveren, I. (2007). Footloose multinationals in belgium? Review of World
Economics , 143 (3), 483–507.
Van Buuren, S., Brand, J. P., Groothuis-Oudshoorn, C. G., & Rubin, D. B. (2006).
Fully conditional specification in multivariate imputation. Journal of
statistical computation and simulation, 76 (12), 1049–1064.
Van Buuren, S., & Groothuis-Oudshoorn, K. (2010). mice: Multivariate imputation
by chained equations in r. Journal of statistical software, 1–68.
van Ours, J. C., & Williams, J. (2015). Cannabis use and its effects on health,
education and labor market success. Journal of Economic Surveys , 29 (5),
993–1010.
Wallace, A. (2020, June 25). Why more states could legalize cannabis in 2021.
CNN Business . Retrieved 2020/07/03, from https://www.cnn.com/2020/
06/24/business/cannabis-legalize-2021-recession/index.html
136
Wallace, D. (1997). Positioning for the future: Canadian shipper perspective.
Customized Ag Production & Marketing Conference. (Unpublished
presentation)
Washburn, K. (2020, February 5). Wine to weed: Wine grape growers lose labor to
cannabis farms. The Sun Gazette. Retrieved 2020/07/03, from
https://thesungazette.com/article/news/2020/02/05/wine-grape
-growers-are-losing-labor-to-commercial-cannabis-farms/
Whinston, M. D. (1988). Exit with multiplant firms. The RAND Journal of
Economics , 19 (4), 568–588.
White, T. K., Reiter, J. P., & Petrin, A. (2012). Plant-level productivity and
imputation of missing data in us census manufacturing data (Tech. Rep.).
National Bureau of Economic Research.
White, T. K., Reiter, J. P., & Petrin, A. (2018). Imputation in us manufacturing
data and its implications for productivity dispersion. Review of Economics
and Statistics , 100 (3), 502–509.
Zingales, L. (1998). Survival of the fittest or the fattest? exit and financing in the
trucking industry. The Journal of Finance, 53 (3), 905–938.
137