Envisioning a Lower-Carbon Economy: 

An Examination of the Economic Characteristics which Decrease Emissions in Vehicles, 

Residential Buildings, and Electricity Generation 

by  

Justin Zev Strimling Szasz 

  

A dissertation accepted and approved in partial fulfillment of the   

requirements for the degree of  

Doctor of Philosophy   

in Sociology 

  

Dissertation Committee:  

Richard York, Chair  

Aaron Gullickson, Core Member  

Matthew Norton, Core Member  

Alison Gash, Institutional Representative  

 
 

University of Oregon  

 

Summer 2025 

 

 

 

 

 

 



 

2 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

© 2025 Justin Zev Strimling Szasz  

 

  



 

3 

 

 

DISSERTATION ABSTRACT 

 

Justin Zev Strimling Szasz 

 

Doctor of Philosophy in Sociology 

 

Title: Envisioning a Lower-Carbon Economy: An Examination of the Economic Characteristics 

which Decrease Emissions in Vehicles, Residential Buildings, and Electricity Generation 

 

 

Scholars, governments, and other policy analysts have made arguments about which 

characteristics in an economy will decrease fossil fuel use and resulting carbon emissions. In this 

dissertation, I examine empirical evidence from the United States to assess the actual effect of 

these characteristics on carbon emissions. In empirical chapter one, I run panel regression 

models to assess the effects of increased retail gasoline prices, increased vehicle fuel efficiency, 

an increased share of workers working from home, and an increased share of workers 

commuting on public transit on emissions from fossil fuel use in vehicles. In empirical chapter 

two, I run panel regression models to assess the effects of increased energy efficiency and 

increased residential natural gas retail prices on emissions from fossil fuel use in residential 

buildings. In empirical chapter three, I compare electricity consumption trends in Texas and 

California, over the period 2001 to 2021, to assess which characteristics of an economy (or of the 

surrounding society) decrease electricity demand – making it easier to meet demand entirely with 

non-carbon electricity supply. 

I find that increased vehicle fuel efficiency and increased retail gasoline prices decrease 

vehicle emissions. Increased residential building energy efficiency and increased residential 

natural gas retail prices decrease residential building emissions. Finally, people living in more 

temperate climates, higher retail electricity prices for commercial and industrial use, and 

(possibly) increased residential building energy efficiency decrease demand for electricity. 



 

4 

 

 

 

 

PUBLICATIONS: 

 

 Szasz, J. (2023). Which approaches to climate policy decrease carbon dioxide emissions? 

Evidence from US states, 1997–2017. Energy Research & Social Science, 97, 102969. 



 

5 

 

 

ACKNOWLEDGMENTS 

 

First and foremost, I would like to thank my graduate advisors – Richard York, Aaron 

Gullickson, and Matthew Norton – for their advice concerning my dissertation and their 

invaluable mentorship throughout my PhD. I would also like to thank the other professors in the 

Sociology Department for their support and mentorship – especially Julius McGee, Jessica 

Vasquez-Tokos, Jill Ann Harrison, Michael Dreiling, Clare Evans, John Bellamy Foster, and 

Kari Norgaard. Additionally, I would like to thank my longtime mentors Daniel Schrag of 

Harvard University and Les Guliasi of UC Berkeley and UC Santa Cruz for teaching me so much 

about climate change, energy, and climate policy and for always supporting me in my career. 

This investigation was also partly supported by a Marquina Award from the University of 

Oregon Sociology Department. 

On a personal note, I want to express my appreciation to my friends and family for 

helping me through my PhD. My close friends in Eugene included Nicholas Theis, Eid Zwicker, 

Jack Sager, Henry Corl, Julius McGee, and Jonathan Moses; outside of Eugene, Faiq Habash, 

Lorenzo Sieman, Jiayi Peng, Reece Akana, Eagon Meng, Nick Ackert, and Thomas Huling. My 

immediate family – my father Andrew, my mother Wendy, my brother Aaron, and my sister 

Emily – also supported me throughout the PhD. Thank you all; I truly could not have done it 

without you.  

  



 

6 

 

 

 

DEDICATION 

 

For humankind. We can stop climate change, together.  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 



 

7 

 

 

TABLE OF CONTENTS 

 

Introduction ................................................................................................................................... 12 

Literature on reducing emissions .............................................................................................. 13 

Reasons to reduce emissions ................................................................................................. 15 

Policy literature on emissions reduction ............................................................................... 16 

Economists’ theory of emission reduction ............................................................................ 18 

Left-wing social science theories .......................................................................................... 20 

Empirical literature ............................................................................................................... 22 

Organization of Dissertation ..................................................................................................... 23 

Approach to studying climate policies .................................................................................. 24 

Empirical Chapters 1 and 2 ................................................................................................... 28 

Empirical Chapter 3 .............................................................................................................. 29 

References Cited ....................................................................................................................... 30 

Empirical Chapter 1: Assessing Economic Determinants of Vehicle Emissions ......................... 33 

Introduction ............................................................................................................................... 33 

Approaches to reducing on-site emissions in vehicles ............................................................. 34 

Fuel efficiency ...................................................................................................................... 35 

Public transit ......................................................................................................................... 36 

Work from home ................................................................................................................... 37 

Higher gasoline prices........................................................................................................... 37 

Data and Methods ..................................................................................................................... 38 

Units of observation and dependent variable ........................................................................ 39 

Independent variables of interest .......................................................................................... 40 

Control variables ................................................................................................................... 40 

Main models.......................................................................................................................... 42 

Statistical significance indicators .......................................................................................... 45 

Robustness checks ................................................................................................................ 46 

Data sources .......................................................................................................................... 47 

Results ....................................................................................................................................... 48 

Discussion ................................................................................................................................. 52 

Conclusion ................................................................................................................................ 56 

References Cited ....................................................................................................................... 58 



 

8 

 

 

Empirical Chapter 2: Assessing Economic Determinants of Residential Building Emissions .... 62 

Introduction ............................................................................................................................... 62 

Approaches to reducing emissions............................................................................................ 63 

Energy efficiency .................................................................................................................. 64 

Carbon pricing ...................................................................................................................... 66 

Data and Methods ..................................................................................................................... 67 

Units of observation and dependent variable ........................................................................ 68 

Independent variables of interest .......................................................................................... 69 

Control variables ................................................................................................................... 69 

Main Models ......................................................................................................................... 72 

Statistical significance indicators .......................................................................................... 75 

Robustness checks ................................................................................................................ 75 

Data sources .......................................................................................................................... 76 

Results ....................................................................................................................................... 78 

Discussion ................................................................................................................................. 82 

Conclusion ................................................................................................................................ 85 

References Cited ....................................................................................................................... 86 

Empirical Chapter 3: Assessing Economic Determinants of Electricity Demand ........................ 89 

Introduction ............................................................................................................................... 89 

Existing literature ...................................................................................................................... 90 

The Texas-California comparison ......................................................................................... 90 

Population ............................................................................................................................. 92 

Climate .................................................................................................................................. 93 

Beliefs about climate change ................................................................................................ 94 

Energy efficiency measures .................................................................................................. 94 

Different economies .............................................................................................................. 95 

Electricity prices ................................................................................................................... 97 

Data and Methods ................................................................................................................... 100 

Results ..................................................................................................................................... 102 

Residential Use ................................................................................................................... 102 

Industrial Use ...................................................................................................................... 109 

Commercial Use.................................................................................................................. 113 

Discussion ............................................................................................................................... 117 

Conclusion .............................................................................................................................. 119 



 

9 

 

 

References Cited ..................................................................................................................... 121 

Conclusion .................................................................................................................................. 124 

Empirical Results .................................................................................................................... 125 

Takeaways: Energy Prices and Energy Efficiency ................................................................. 128 

Policies .................................................................................................................................... 129 

Low-Carbon Attributes Together ............................................................................................ 131 

High Fossil Fuel Prices and Lower Income People ................................................................ 132 

Literatures on Emission Reduction ......................................................................................... 134 

Conclusion .............................................................................................................................. 138 

References Cited ..................................................................................................................... 139 

Supplement 1 .............................................................................................................................. 142 

Background ............................................................................................................................. 142 

Results ..................................................................................................................................... 142 

Supplement 2 .............................................................................................................................. 146 

Background ............................................................................................................................. 146 

Results ..................................................................................................................................... 146 

 

  

 



 

10 

 

 

LIST OF FIGURES  

 

Figure Page 

 

 

1. Annual Electricity Consumption in Texas and California ..................................... 91 

 

2. Population of California and Texas Over Time .....................................................    93 

 

3. Annual ACEEE Energy Efficiency Ranking (out of 50 US states)  ...................... 95 

4. GDP in CA and TX, broken down by sector ......................................................... 96 

5. GDP per capita, broken down by sector ................................................................ 96 

6. Average Electricity Price in California and Texas  ............................................... 98 

7. Annual Electricity Consumption (kWh) by Sector ................................................ 101 

8. Residential Electricity Consumption in California and Texas ............................... 103 

9. Residential Electricity Consumption Per Capita.................................................... 104 

10. Average Residential Electricity Price in California and Texas ............................. 105 

11. Percent Who Think Climate Change Caused Mostly by Humans ......................... 106 

12. Annual electricity consumption, per capita, by residential end-use ...................... 108 

13. Industrial Electricity Consumption in California and Texas ................................. 109 

14. Electricity consumption per dollar of real GDP, national average ........................ 111 

15. Average Industrial Electricity Price in California and Texas ................................ 112 

16. Commercial Electricity Consumption in California and Texas ............................. 114 

17. Commercial Electricity Consumption Per Capita .................................................. 114 

18. Average Commercial Electricity Price in California and Texas ............................ 116 

  



 

11 

 

 

 

 

LIST OF TABLES 

 

Table Page 

 

 

1. Effect of characteristics on CO2 emissions in vehicles .......................................... 49 

 

 

2. Effect of characteristics on CO2 emissions in residential buildings ...................... 78 

 

 

S1. Effect of characteristics on CO2 emissions in vehicles  

 (incl. one variable of interest per model, using first proxy for Dem control) ........ 143 

 

S2. Effect of characteristics on CO2 emissions in vehicles  

 (incl. one variable of interest per model, using second proxy for Dem control) ... 144 

 

S3. Effect of characteristics on CO2 emissions in residential buildings  

 (incl. one variable of interest per model, using first proxy for Dem control) ........ 146 

 

S4. Effect of characteristics on CO2 emissions in residential buildings  

 (incl. one variable of interest per model, using second proxy for Dem control) ... 147 



 

12 

 

 

Introduction 
 

Society must stop emitting greenhouse gases to avoid the worst possible effects of 

climate change. Such decarbonization involves several difficult steps. Large government outlays 

may be needed to change the economy to use less fossil fuels (Aronoff et al 2019). Potential 

changes to the economy must also be beneficial to a large enough subset of the population to 

mobilize a mass coalition which is supportive of, or at least neutral towards, the changes (Ibid; 

Cullenward and Victor 2020). Before even this, however, it is necessary to determine which 

changes to the economy would decrease emissions – and which policies would result in such 

changes. 

 People who are intent on decarbonizing the economy often skip over this first step – 

perhaps because, as I discuss below, most thinkers who are interested in decarbonizing the 

economy believe in one of a few paradigms concerning the causes of fossil fuel use. Each 

paradigm points to certain conclusions about which changes to the economy would reduce 

emissions, and about which policies would affect such changes. Adherents of any of the 

paradigms thus implicitly believe the policies necessary to decarbonize the economy have 

already been logically deduced -- even without extensive empirical evidence that these policies 

actually decarbonize the economy.  

 However, the fact that a story is logical, according to certain premises, does not make it 

true. Economies are complex and hard to fully predict (York and Bell 2019); thus, the effects of 

policies on the economy and, ultimately, emissions are hard to predict. Real-world evidence is 

necessary to determine the effect that particular policies actually have on the economy and 

emissions. Yet, with some notable exceptions (for example, Herring 2006; Prasad and Munch 



 

13 

 

 

2012; Greene 2012; York and McGee 2016; Szasz 2023), there has not been enough work done 

determining the actual effect of different policies, and/or changes to the economy, on emissions. 

 In this dissertation, I conduct new empirical research on the effect of certain 

characteristics of an economy, which policies might aim to induce, on emissions. My first 

empirical chapter uses panel regression models to test the effects of different characteristics of 

the vehicle transportation system on carbon emissions from fossil fuel consumption in vehicles. 

My second empirical chapter uses regressions models to test the effects of different 

characteristics of residential buildings on carbon dioxide emissions from fossil fuel use in 

buildings. My third empirical chapter compares the systems of electricity provision in Texas and 

California to determine which characteristics of California’s system are limiting electricity 

demand in the state and, by extension, which characteristics of any economy will decrease 

electricity demand; this is an important question because electricity demand must be limited to 

levels which can be met entirely with supply from low-carbon sources. I hope my findings 

contribute to the empirical literature examining which policies, or which changes to an economy, 

actually result in lower emissions. 

 In the following pages, I discuss the existing literature on emission reduction in more 

detail and argue that more empirical literature is needed examining the effects of different 

policies on emissions. I then outline the studies undertaken in the dissertation’s empirical 

chapters. 

Literature on reducing emissions 

 

 Literature on policies to reduce emissions can be separated into four categories. First, 

government agencies and policymakers have divided the task of decarbonizing into a series of 

specific technological or fiscal barriers that must be overcome and have theorized policies to 



 

14 

 

 

overcome these barriers. Second, economists have conceptualized economic activity as utility-

maximizing behavior of rational actors on a market and have argued that climate change results 

from a “market failure” in which fossil fuel producers and consumers do not pay the full cost of 

their fossil fuel use. These economists argue that the ideal solution to decrease carbon use is to 

implement a tax or emission trading scheme which forces those using carbon to pay the full costs 

of its use. Third, left-wing social scientists have argued that the capitalist economic system, or 

other interlocking social systems, are to blame for climate change and other ecological problems. 

Some of these social scientists believe that these social systems will need to be abolished and 

replaced wholesale to stop climate change or other ecological problems. However, other left-

wing social scientists believe that policies may be effective – if policymakers consider which 

behaviors are incentivized under current market conditions, within the capitalist economic 

system, and change market conditions to incentivize different behavior. 

 Fourth, there is a scholarly literature which empirically evaluates the effect of previously 

adopted climate policies on energy use or emissions. This literature finds that carbon taxes and 

emission trading schemes decrease emissions, if only marginally. The efficacy of other climate 

policies varies. This literature is essentially the empirical check on the three theoretical 

literatures above – checking whether their theoretical claims about which policies will decrease 

emissions are borne out. Given its importance, this literature is currently too small: too few 

studies have examined existing empirical evidence to evaluate which types of policies actually 

reduce emissions, in practice. My dissertation helps to rectify this problem by empirically 

examining the effect of different characteristics of an economy, which policies could aim to 

induce, on emissions. Below, I briefly discuss the reasons many policymakers and academics are 



 

15 

 

 

interested in reducing carbon emissions, the four literatures concerning emission reduction 

policies, and the need for more scholarship empirically testing such policies. 

Reasons to reduce emissions 

 

Burning fossil fuels for energy produces gaseous carbon molecules as a biproduct; as 

global society has burned ever more fossil fuels for energy, ever more carbon molecules have 

been released into the atmosphere. This has increased the concentration of carbon molecules in 

the atmosphere. Gaseous carbon molecules in the atmosphere increase “radiative forcing” – the 

amount of solar radiation hitting any spot on Earth at any moment in time – increasing Earth’s 

surface temperature. The greater the concentration of carbon molecules in the atmosphere, the 

greater the “radiative forcing”, and the more surface temperature increases. Thus, as global 

society has burned steadily more carbon emissions, Earth’s surface temperature has risen – and 

the temperature will continue to rise as long as we continue to burn fossil fuels and put more 

gaseous carbon molecules into the atmosphere (Pachauri et al 2014). 

 Human-induced increases in average surface temperatures – generally called “global 

warming” or “climate change” – will have potentially dire effects on global society. If society is 

sensitive to changes in ecological conditions, as a large literature suggests (Hsiang et al 2013, 

Burke et al 2015), then we would expect that the more climate change happens, the larger the 

effects will be on society. Already, climate change may have caused natural disasters 

(Klinenberg et al 2020), which in turn may have played a role in instances of social collapse and 

conflict such as the Syrian Civil War (Kelley et al 2015). Projections suggest that enough 

warming, relative to present, could make it impossible to grow staple crops such as wheat, rice, 

and corn in traditional breadbaskets and could leave areas such as southeastern China, the 

southeastern United States, and northern India seasonally too hot for human habitation. Such 



 

16 

 

 

changes could cause conflict (King et al 2015) or mass migration. One study finds that, without 

significant reductions in greenhouse gas emissions, “accounting for expected demographic 

developments… ~3.5 billion people (roughly 30% of the projected global population…) would 

have to move to other areas” in the next 50 years to continue living in humans’ usual ecological 

niches (Xu et al 2020: 11352). 

 What would be the effect on society of global disruptions in the production of staple 

crops, of increased extreme weather events, of increased conflicts, and of mass migrations of 

billions of people? It is hard to imagine anything but catastrophic results. Many scientists, 

governments, and social movements have come to similar conclusions, leading them to advocate 

for massive reductions in fossil fuel use. If society reduces fossil fuel use enough, they hope, we 

can avoid catastrophic climate change. 

Many policymakers and academics view targeted economic and energy policies as a 

potential avenue to reduce fossil fuel emissions – though, as we will see, policymakers and 

academics disagree about which policies are effective at decreasing fossil fuel use. In the 

sections below, I discuss the policymaker and academic literatures on policy options to reduce 

carbon emissions. 

Policy literature on emissions reduction 

 

Some city, regional, and state governments have created emission reduction goals for 

their jurisdictions. These goals are generally declarations that the jurisdiction will reduce 

emissions x amount, by y year, relative to emissions in some reference year. For example, a city 

might aim to reduce emissions 90% by 2035, relative to a 2005 baseline. These jurisdictions will 

sometimes produce documents which outline policies which could be adopted to decrease fossil 



 

17 

 

 

fuel use, and resulting carbon emissions, in their jurisdictions. This is the first literature about 

policies to reduce carbon emissions. 

 When discussing decarbonization, this first literature often separates energy generation 

and consumption into several “sectors”: 

• “Vehicle transportation” – energy use in vehicles. 

• “Buildings” – energy use in buildings. This category is sometimes further separated into 

energy use in “residential buildings,” where people live, and “commercial buildings,” 

where businesses operate. 

• “Industry” – energy use in industrial processes, such as the production of plastic, cement, 

steel, and glass. 

• “Electricity generation” – the various processes which produce electricity, including 

renewable generation and the burning of fossil fuels to generate electricity. 

For an example of this categorization system, see the Obama White House’s “Vision for 2050”, 

chapter four (US White House 2016). 

The policymaker literature considers how each sector must change if our economy is to 

stop burning fossil fuels for energy. Policy specialists generally believe emissions from on-

location energy use in vehicles, residential and commercial buildings, and industrial processes 

must be reduced to zero – a goal which, they hypothesize, might be achieved by electrifying 

energy uses which can be powered with electricity (Ibid), and by finding creative workarounds 

for other energy uses. 

In addition to electrifying energy uses, climate policy specialists also aim for all 

electricity to be generated by low-carbon sources, such as wind and solar power (Ibid). 



 

18 

 

 

Electrification is only a means of decarbonizing the economy if few emissions are created while 

generating electricity. 

Although policymakers in states like California are aiming to electrify energy use as fast 

as possible, in the meantime they are also trying to reduce on-location fossil fuel use in 

buildings, vehicles, and industry through other means, such as making cars and appliances more 

energy efficient, and capping fossil fuel use or increasing its price (for examples, see policies 

discussed in Taylor 2017, Taylor 2018). 

 This literature makes comprehensive lists of the technical or fiscal problems which 

(policymakers believe) must be solved, in each sector, to achieve the sectoral goals explained 

above. For example, to eliminate on-site emissions in buildings, buildings must be transitioned 

from using fossil fuels to using electricity for heating, and this goal must be achieved using 

available government funds. Potential policies are seen as “solutions” to one or more of these 

“problems” (for examples, see Lovins 2004 or Massachusetts 2022). To return to our previous 

example, instituting building codes which require new buildings to only have electric building 

appliances might induce building electrification, and scaling electrification through these means, 

rather than government spending, fits within government’s limited budget.  

Economists’ theory of emission reduction 

 

 Academics have also written about which policies might facilitate decarbonization, but 

they think about the problem very differently. Academics generally think about the people or 

institutions that consume fossil fuels: companies seeking to maximize profits and individuals 

seeking maximum comfort at minimum cost. Academics have models of how companies and 

people behave, and therefore, how the same will respond to different types of policies. Therefore, 



 

19 

 

 

academics often debate which types of policies will induce companies and people to burn fewer 

fossil fuels. 

The second literature concerning options for emission reduction comes from a subset of 

academics: economists using the neo-classical model of an economy. These economists believe 

the most efficient, and therefore optimal, outcomes are achieved when the firms producing a 

good have to bear the full cost of production, and then are allowed to sell the good at any price. 

In such a situation, there will be some price at which the amount the firm supplies exactly 

matches the demand for the good. At this price point, the firm produces some number of goods 

and sells them for more than the full cost of production, making a return on investment. The 

consumers, collectively, decide it is worthwhile to trade some amount of money for the good, 

taking into account the full costs of production. There is neither a shortage of goods, relative to 

demand, nor a surplus that cannot be sold to anyone. In such a situation, the economists believe, 

optimal efficiency has been achieved, leading to the best possible outcome for the market and 

society as a whole. 

The issue with fossil fuel use, in this view, is that the producers do not have to pay the 

full cost of burning fossil fuels, or of selling the consumer products which burn fossil fuels. 

Firms can make and sell goods at below their full cost, allowing them to sell more goods and 

profit more. Meanwhile, consumers pay less for the good, which, again, incentivizes more 

consumption. Instead, many of the “costs” of producing the good – the negative effects of 

climate change – will be borne by other people who neither made nor bought the good. This is 

what economists call an “externality.” (Nordhaus 2008; Stern 2007). 

Economists’ solution to this problem is to “internalize” the “externality” by increasing 

the market price of fossil fuel use to match the “true” cost of fossil fuel use. Economists believe 



 

20 

 

 

any climate policy will necessarily increase or decrease the cost of using fossil fuels to some 

extent, but they generally believe that the best policy is one that directly changes the price of 

fossil fuel use – either through a tax on fossil fuel use or through an emissions trading scheme 

(colloquially called “cap and trade”; Tietenberg, 2013). Of course, economists debate both the 

ideal stipulations of an actual carbon pricing policy (Ibid) and the externalized “cost” of emitting 

a unit of carbon (Nordhaus 2008; Stern 2007). 

Left-wing social science theories 

 

The third literature concerning emission reduction comes from other academics: left-wing 

social scientists who blame the economic system of capitalism – or another, interlocking social 

system such as “patriarchy” or “settler-colonialism” –for environmental problems. Marxist 

scholars have argued for decades that, in a capitalist society, decisions about production of goods 

are mostly made by companies whose goal is to maximize profit. These companies will cause 

environmental harm if doing so increases their short-term profit margin – which it often does. 

Thus, environmental harms in our society are not incidental, but rather, are often a result of the 

capitalist economic system through which we produce and distribute goods (Schnaiberg and 

Gould 1994; O’Connor 1988, 1998; Foster 2000; Clark and York 2008; Foster, Clark, and York 

2011). 

Other left-wing academics argue that capitalism is tied up with other social systems 

which, they believe, are just as much to blame for environmental harm in our society. 

Ecofeminists argue that modern capitalist societies’ destruction of the environment is intimately 

tied to the oppression of women (Merchant 1980; Bell et al 2020). Environmental humanities 

scholars have argued that the social institutions of the slave trade and the plantation were crucial 

to the creation of capitalism and the subsequent environmental harm capitalism created (Murphy 



 

21 

 

 

and Schroering 2020). Theorists of settler-colonialism argue that settlers in the Americas and 

elsewhere, beyond being capitalists, also believed in settler-colonialist ideology – which 

included a willingness to destroy nature (McKay et al 2020). 

Many of these left-wing scholars appear to believe that the economic system of 

capitalism – and possibly the social systems connected to it – will need to be completely 

abolished and replaced with new systems to stop environmental problems (e.g. Foster 1999; Bell 

et al 2020). For example, Bell et al (Ibid: 4) state it is an 

“[assumption] that [has] not thus far been empirically demonstrated: that 

authoritarian-, imperial-, and/or capital-led forces exist with sufficient 

motivation to lead a global decarbonization; that such pathways would in fact 

be easier and faster than more democratic and inclusive ones; and that a 

decarbonization effort brought about via hegemonic political styles could ever 

be truly sustainable from an ecological and social perspective.” 

 

However, some left-wing scholars are open to the possibility that, absent the total 

abolition of the capitalist economic system or associated social systems, policies may be 

effective at decreasing environmental harm, including fossil fuel use (see, for example, 

O’Connor 1988, 1998). These scholars generally believe that, to decrease emissions, 

policymakers will have to identify the mechanisms through which existing social systems cause 

emissions – and design interventions which affect these mechanisms to produce lower emissions 

(York 2012; York and Bell 2019). 

For example, York and Bell (2019: 43) argue that as long as fossil fuel use is profitable, 

within a capitalist economy, companies and consumers will continue to use fossil fuels. They 

suggest that “Increasing the price of extracting and importing fossil fuels through a carbon fee 

and dividend”, or capping total energy use (thus increasing prices “due to scarcity”), may be 

effective at decreasing fossil fuel use – because, within a capitalist economic system, higher 



 

22 

 

 

fossil fuel prices may induce lower fossil fuel use. Alternatively, they suggest, restricting fossil 

fuel extraction may be effective, as this would necessarily limit fossil fuel use. 

Empirical literature 

 

The fourth and final literature on policies to reduce emissions is the scholarly literature 

empirically evaluating the effects of different climate policies. This literature looks at particular 

cases and measures the effect of policies, implemented in those cases, on energy use or 

emissions (see, for example, Greene 2012; Prasad and Munch 2012; Murray and Maniloff 2015). 

This literature generally finds that carbon taxes and emission trading schemes are associated with 

emission reductions, while empirical results for other climate policies are more mixed (Sterner 

2007; Tietenberg 2013; Szasz 2023). 

There has also been scholarship concerning the effects of particular characteristics of an 

economy on emissions. For example, there are multiple works examining the effect of energy 

efficiency on aggregate energy use. Some of these works have noted that, over time, increases in 

energy efficiency coincide with increasing total energy use (Herring 2006; York and McGee 

2016) – suggesting that energy efficiency may not be effective as a way to decrease aggregate 

energy use and resulting emissions.  

The issue is that the above literature is still too small. As I have explained, above, 

policymakers and scholars have presented various theoretical arguments about which climate 

policies will decarbonize the economy. The task now is to bring the full weight of empirical 

evidence to assess which, if any, of these arguments are correct. Such a task requires that as 

many instances of climate policy adoption as possible be studied, to check, across many cases, 

which policies actually seem to reduce emissions. Yet, so far, only a small number of cases of 

climate policy adoption have been studied to see how effective these policies were at reducing 



 

23 

 

 

carbon emissions – and, therefore, to see which theories about climate policy are borne out 

empirically. 

My dissertation research helps to address this gap. In the empirical chapters that follow, I 

leverage empirical evidence to evaluate the effect of different characteristics of the economy in 

reducing on-site emissions in vehicles, reducing on-site emissions in residential buildings, and 

limiting electricity demand.  

Organization of Dissertation 

 

 Below, I discuss how I evaluate the effect of economic characteristics on emissions or 

energy use in each empirical chapter. Empirical chapter one uses panel regression models to 

examine the effect higher retail gasoline prices, increased vehicle fuel economy, an increased 

share of workers who work from home, and an increased share of workers who commute on 

public transit, each have on emissions from fossil fuel use in vehicles. Empirical chapter two 

uses panel regression models to measure the effect of increased retail natural gas prices and 

increased building energy efficiency on emissions from fossil fuel use in residential buildings. In 

both empirical chapters one and two, the units of observation are state-years, for the fifty United 

States, for the years 2008-2019. Empirical chapter three is a comparative case study of electricity 

consumption in California and Texas over the period 2001-2021. Chapter three leverages this 

comparison to identify characteristics of an economy, and of the surrounding society, which can 

limit demand for electricity. 

In the sections that follow, I first explain why I take this approach to evaluating the 

effects of climate policies. I then describe each of the empirical projects I undertake in more 

detail. 



 

24 

 

 

Approach to studying climate policies 

 

The motivation for this dissertation was my previous research project evaluating the 

effects of climate policies (Szasz 2023). In that previous research, I sought to leverage empirical 

data to study which climate policies could significantly decarbonize a modern, carbon-intensive 

economy like the United States. To answer this question, I ran fixed effect panel regression 

models. The units of observation were state-years, for the fifty United States, for the years 1997-

2017. The dependent variable was carbon dioxide emissions from the consumption of fossil fuels 

for energy. The independent variables of interest were ten variables which each recorded whether 

a certain climate policy was in implementation in that state-year. The model included other 

independent variables to control for a variety of demographic factors which could affect energy 

use or emissions. 

The logic of the regressions was to control for the effects of the various demographic 

factors on emissions, leaving some amount of the emissions which could not be explained by 

these factors. The computer is left with a math problem: the amount of emissions that cannot be 

explained by demographic factors, and information about which of the ten policies was in 

implementation, for each state-year. The computer solves the math problem to determine how 

much implementing each policy increased or decreased emissions, on average. 

When I ran these regressions, I found that the most effective of the ten tested policies – 

cap and trade – was associated with about a 6-8% reduction in emissions, controlling for other 

factors. Others of the ten policies were associated with smaller emission reductions, on average, 

or even associated with increases in emissions (Ibid). In other words, none of the tested policies 

had caused significant decarbonization in any of the United States. 



 

25 

 

 

In retrospect, this result made sense. How could any of the tested policies cause 

significant decarbonization in the United States, when we know for a fact that all fifty states still 

have carbon intensive economies? 

However, this line of thinking led me to a new problem. If no single climate policy has 

significantly decarbonized a US state’s economy, how can we leverage empirical evidence to 

evaluate which policies could decarbonize an economy? 

One possible approach is to measure the effects of characteristics of the economy on 

emissions, rather measuring than the effects of policies on emissions. No US state has largely 

decarbonized after adopting any particular climate policy. However, the goal of many climate 

policies is to change the economy to have various features – more energy efficient appliances, 

higher retail prices for fossil fuels, etc. – which, policymakers hope, will induce people to use 

fewer fossil fuels. We know that states vary widely on these structural characteristics and that 

states’ emissions profiles vary widely as well. It is possible that differences in these structural 

characteristics are largely responsible for differences in states’ emissions. By comparing 

different US states on structural characteristics which might affect emissions, controlling for 

other possible causes of emissions, and then comparing the states’ emissions data, we can test 

this proposition ourselves. 

Intuitively, this approach has a better chance of producing notable results than did my 

previous models. The problem we have identified with the previous models is that there is no 

instance in the US in which the passage of a single policy led to a drastic, sudden decrease in 

emissions; therefore, it makes sense that a model checking whether any policy had such an effect 

would come up empty-handed. However, as previously stated, emissions do vary a great deal, 

overall, across time and place. There is enough variance in structural characteristics of the 



 

26 

 

 

economy across time and space, as well as enough variance in emissions, that it is possible we 

could find that some structural characteristic or other is associated with much lower emissions, 

controlling for other factors. (If our models do not find any such association, that too is 

informative.) 

This modeling approach can also produce results which are meaningful for policymakers. 

If the results suggest a particular characteristic is associated with much lower emissions, 

controlling for other factors, this indicates that policies which change an economy to have that 

characteristic would reduce emissions. For example, if the analysis finds that higher retail fossil 

fuel prices are associated with lower emissions, then policymakers should enact policies which 

aim to raise energy prices. In this way, the results can inform future decisions by policymakers. 

In this dissertation, I take the above approach to the empirical study of emissions 

reduction. Empirical chapters one and two use panel regression models, similar to my previous 

research (Szasz 2023), to examine which factors can decrease emissions from consumption of 

direct fossil fuels for energy in vehicles and in residential buildings. However, instead of markers 

of which policies were implemented in which state-year, the independent variables of interest in 

these regression models are measurements of the structural characteristics of the economy which 

might affect fossil fuel use, such as the level of energy efficiency of fuel-consuming 

technologies, or the retail price of fossil fuels. 

Empirical chapter three is a comparative case study of electricity consumption in 

California and Texas over the period 2001-2021. To decarbonize the economy, modern society 

will most likely need to switch out many technologies which currently burn fossil fuels for 

electric equivalents, and society will need to source that electricity entirely from low-carbon 

sources. Currently, a large share of electricity in the United States is produced by burning fossil 



 

27 

 

 

fuels. To switch electricity entirely to low-carbon sources, society will need to: generate large 

amounts of electricity from low-carbon sources, such as wind and solar power; limit electricity 

demand to a level that can be met entirely with available electricity from low-carbon sources; 

and ensure that electricity from low-carbon sources (rather than electricity from burning fossil 

fuels) is used to meet this demand. 

The Texas-California comparison can be used to gain empirical leverage into the second 

of these three tasks: limiting electricity demand to a level that can be met entirely with electricity 

from low-carbon sources. Over the period 2001 to 2021, both Texas and California saw fairly 

consistent increases in population and GDP – both of which existing research (e.g. York 2007) 

has found to be positively correlated with energy consumption. (California’s population 

increases consistently until about 2017, when it levels out, then starts to decrease in 2021.) 

Furthermore, both states saw large increases in renewable electricity generation over this period 

and, again, existing research (York 2012) indicates that increased electricity generation from 

non-carbon sources generally results in increased overall electricity demand. Thus, we might 

expect that both Texas and California would see an increase in electricity consumption over the 

study period, and, indeed, in Texas electricity consumption steadily rose between 2001 and 2021. 

In California, too, electricity consumption increased between 2002 and about 2008. However, 

from 2008 to 2021, while GDP, population, and renewable generation in California all increased 

– which we might expect to cause more consumption – electricity consumption declined. 

The purpose of the Texas-California comparison is to identify reasons why electricity 

consumption did not follow a similar trend in California, over the period 2001-2021, as in Texas. 

I may discover that the different consumption trends can be partly attributed to features of the 

states’ economies. If so, other jurisdictions could attempt to change their own economies to have 



 

28 

 

 

those features as well, to limit electricity demand. Limiting electricity demand, in turn, may 

make it easier for the jurisdictions to decarbonize electricity supply. 

Empirical Chapters 1 and 2 

 

 Empirical chapters one and two use fixed effect panel regression models to evaluate the 

effect of structural characteristics in energy-use “sectors” on those sectors’ on-location CO2 

emissions. I am specifically interested in characteristics of these sectors that climate policies aim 

to change, as an avenue to emissions reduction, such as the retail price of fossil fuels or the 

energy efficiency of buildings/vehicles. The dependent variables for these models are CO2 

emissions from the on-site consumption of fossil fuels for energy in vehicles and residential 

buildings, respectively. The models include independent variables to measure the different 

characteristics of these sectors which climate policies aim to affect; the variables test the effect 

of changes to these characteristics on sectoral emissions. The models also include independent 

variables which control for a variety of other characteristics of states which may affect 

emissions. I include these other variables so I can more accurately pinpoint the effects of the 

variables of interest on emissions. 

 The units of observation are state-years, for the fifty United States, for the years 2008-

2019. The key independent variables for the vehicle regression models are the real average retail 

gasoline price in a state-year, vehicles’ average miles per gallon when traveling on highways in 

the state-year (as a proxy for vehicle fuel efficiency), the percentage of workers who take public 

transit to work (as a proxy for overall public transit use), and the percentage of workers who 

work from home. The key independent variables for the residential building models are the real, 

average retail natural gas price in residential buildings in the state-year and the percentage of a 



 

29 

 

 

state’s housing units which were built before the year 2000 (as a proxy for the residential 

building stock’s overall energy efficiency). 

Both regressions include independent variables to control for various demographic 

factors that may affect on-site energy use in these sectors, including the size of the population, 

the share of the population who are dependents, the median household income, the percentage of 

state residents who live in urban areas, the number of cooling and heating degree days, the 

percent of residents who believe in man-made climate change, and the cumulative amount of 

time Democrats have controlled state government since the turn of the century. The vehicle 

regression models include an independent variable to control for the amount of economic activity 

in the transportation and shipping sectors in a state-year, while the residential building models 

include independent variables to control for electricity’s share of building energy use, the 

average number of people per occupied housing unit, and the median number of rooms per 

housing unit.  

Empirical Chapter 3 

 

My comparative case study of electricity consumption in Texas and California has two 

components. First, I examine the existing literature to compile a list of reasons other 

commentators have suggested as to why the states have different electricity consumption trends.  

Second, I subdivide electricity consumption in each state into consumption for 

residential, commercial, and industrial uses. For each category of electricity consumption, I 

visualize data concerning each characteristic that some commentators attribute the disparate 

consumption trends to, such as different industries’ size over time in the two states, beliefs about 

man-made climate change, and retail electricity prices. I compare the trends in each of these 

characteristics to consumption trends to see which characteristics appear to be correlated with 



 

30 

 

 

consumption trends. If some characteristic is consistently correlated with consumption – for 

example, if higher electricity prices are consistently associated with lower electricity 

consumption – this is evidence that characteristic is driving consumption trends.  

 

References Cited 

 

Aronoff, K., Battistoni, A., Cohen, D. A., and Riofrancos, T. (2019). A planet to win: why we 

need a Green New Deal. Verso Books. 

 

Bell, S. E., Daggett, C., & Labuski, C. (2020). Toward feminist energy systems: Why adding 

women and solar panels is not enough. Energy research & social science, 68, 101557. 

 

Burke, M., Hsiang, S. M., & Miguel, E. (2015). Climate and conflict. Annu. Rev. Econ., 7(1), 

577-617. 

 

Clark, B., & York, R. (2008). Rifts and shifts. Monthly Review, 60(6), 13-24. 

 

Cullenward, D., and Victor, D.G. (2020). Making climate policy work. John Wiley & Sons. 

Foster, J. B. (1999). Marx's theory of metabolic rift: Classical foundations for environmental 

sociology. American journal of sociology, 105(2), 366-405. 

 

Foster, J. B. (2000). Marx’s Ecology: Materialism and Nature. Monthly Review Press. 

 

Foster, J. B., Clark, B., & York, R. (2011). The ecological rift: Capitalism’s war on the earth. 

NYU Press. 

 

Greene, D. L. (2012). Rebound 2007: Analysis of US light-duty vehicle travel statistics. Energy 

Policy, 41, 14-28. 

 

Herring, H. (2006). Energy efficiency—a critical view. Energy, 31(1), 10-20. 

 

Hsiang, S. M., Burke, M., & Miguel, E. (2013). Quantifying the influence of climate on human 

conflict. Science, 341(6151), 1235367. 

 

Kelley, C. P., Mohtadi, S., Cane, M. A., Seager, R., & Kushnir, Y. (2015). Climate change in the 

Fertile Crescent and implications of the recent Syrian drought. Proceedings of the 

National Academy of Sciences, 112(11), 3241-3246. 

 

King, D., Schrag, D., Dadi, Z., Ye, Q., & Ghosh, A. (2015). Climate change: A risk assessment. 

Harvard University Center for the Environment. 

 



 

31 

 

 

Klinenberg, E., Araos, M., & Koslov, L. (2020). Sociology and the climate crisis. Annual Review 

of Sociology, 46(1), 649-669. 

 

Lovins, A. B. (2004). Energy efficiency, taxonomic overview. Encyclopedia of energy, 2(283-

402), 6. 

 

Massachusetts. (2022). Clean Energy and Climate Plan for 2025 and 2030. Massachusetts. 

 

McKay, D. L., Vinyeta, K., & Norgaard, K. M. (2020). Theorizing race and settler colonialism 

within US sociology. Sociology Compass, 14(9), e12821. 

 

Merchant, C. (1980). The death of nature: Women, ecology, and the scientific revolution. 

 

Murphy, M. W., & Schroering, C. (2020). Refiguring the Plantationocene: racial capitalism, 

world-systems analysis, and global socioecological transformation. Journal of World-

Systems Research, 26(2), 400-415. 

 

Murray, B. C., & Maniloff, P. T. (2015). Why have greenhouse emissions in RGGI states 

declined? An econometric attribution to economic, energy market, and policy factors. 

Energy Economics, 51, 581-589. 

 

Nordhaus, W. (2008). A question of balance: Weighing the options on global warming policies. 

Yale University Press. 

 

O'Connor, J. (1988). Capitalism, Nature, Socialism: A Theoretical Introduction. Capitalism, 

Nature, Socialism. A Journal of Socialist Ecology, Santa Cruz (Capitalism, Nature, 

Socialism) No. 1, Fall 1988, pp. 11-38. 

 

O'Connor, J. R. (Ed.). (1998). Natural causes: Essays in ecological Marxism. Guilford Press. 

 

Pachauri, R. K., Allen, M. R., Barros, V. R., Broome, J., Cramer, W., Christ, R., ... & van 

Ypserle, J. P. (2014). Climate change 2014: synthesis report. Contribution of Working 

Groups I, II and III to the fifth assessment report of the Intergovernmental Panel on 

Climate Change (p. 151). IPCC. 

 

Prasad, M., & Munch, S. (2012). State-level renewable electricity policies and reductions in 

carbon emissions. Energy Policy, 45, 237-242. 

 

Schnaiberg, A., and Gould, K. A. (1994). Environment and society: The enduring conflict. 

 

Stern, N. (2007). The economics of climate change: the Stern review. HM Treasury. 

 

Sterner, T. (2007). Fuel taxes: An important instrument for climate policy. Energy policy, 35(6), 

3194-3202. 

 



 

32 

 

 

Szasz, J. (2023). Which approaches to climate policy decrease carbon dioxide emissions? 

Evidence from US states, 1997–2017. Energy Research & Social Science, 97, 102969. 

 

Taylor, M. (2017). The 2017-18 Budget: Cap-and-Trade. California Legislative Analyst’s Office. 

 

Taylor, M. (2018). Assessing California’s Climate Policies – Transportation. California 

Legislative Analyst’s Office. 

 

Tietenberg, T. H. (2013). Reflections—carbon pricing in practice. Review of Environmental 

Economics and Policy. 

 

White House. (2016). United States Mid-Century Strategy for Deep Decarbonization. White 

House. Washington DC. 

 

Xu, C., Kohler, T. A., Lenton, T. M., Svenning, J. C., & Scheffer, M. (2020). Future of the 

human climate niche. Proceedings of the National Academy of Sciences, 117(21), 11350-

11355. 

 

York, R. (2007). Demographic trends and energy consumption in European Union Nations, 

1960–2025. Social science research, 36(3), 855-872. 

 

York, R. (2012). Do alternative energy sources displace fossil fuels?. Nature climate change, 

2(6), 441-443. 

 

York, R., & Bell, S. E. (2019). Energy transitions or additions?: Why a transition from fossil 

fuels requires more than the growth of renewable energy. Energy Research & Social 

Science, 51, 40-43. 

 

York, R., & McGee, J. A. (2016). Understanding the Jevons paradox. Environmental Sociology, 

2(1), 77-87. 

  



 

33 

 

 

Empirical Chapter 1: Assessing Economic Determinants of Vehicle 

Emissions 

Introduction 

Currently, most vehicles in the United States burn fossil fuels for energy; this reaction 

releases fossil fuels as a biproduct. Fossil fuel use in vehicles is one of the major sources of 

greenhouse gas emissions globally; in 2022, CO2 emissions from fossil fuel use in vehicles 

accounted for roughly 34.66% of all carbon dioxide emissions in the United States (US EPA 

n.d.). Thus, if policymakers wish to greatly reduce total carbon dioxide emissions in the United 

States, they must find a way to decrease aggregate carbon dioxide emissions from fossil fuel use 

in vehicles. 

Policymakers and scholars have proposed various changes to vehicle transportation which, 

advocates contend, may reduce fossil fuel combustion and resulting emissions from vehicle use. 

Four commonly proposed changes are increasing the price of gasoline, increasing the prevalence of 

working from home, increasing average vehicle miles per gallon (“vehicle fuel economy”), and 

increasing the prevalence of public transit use. There are theoretical reasons why each of these 

changes might, or might not, be effective in reducing vehicle emissions. Thus, empirical research is 

needed to determine which of these changes actually reduce emissions, in practice. 

In this chapter, I conduct just such empirical research. I leverage the fact that these four 

characteristics have varied in the US across different states and years to measure the effect of 

changes to each of the four characteristics on vehicle emissions. More specifically, I run panel 

regression models, using data from all fifty United States, from the years 2008-2019, to measure 

the effects of changing each of the four characteristics on emissions, controlling for other factors. I 

find that increasing vehicle “fuel economy” (miles per gallon) reduces emissions more than 

increasing gasoline price, the share of people who work from home, or the share of people who 



 

34 

 

 

commute on public transit. That said, higher gasoline prices also reduce emissions, partly by 

inducing people to drive more fuel efficient vehicles, and partly through other means. On the other 

hand, my results indicate that increased public transit use for travel to and from work, and possibly 

increased transit use overall, have little impact on emissions. 

Approaches to reducing on-site emissions in vehicles 

Multiple potential approaches to reducing on-site emissions in vehicle transit have been 

proposed. Proposed approaches include: increasing vehicles’ average miles per gallon of 

gasoline, increasing public transit use, increasing the prevalence of remote work (“work from 

home”), raising the price of gasoline, and electrifying personal vehicles. 

I will not examine the effect of electrifying a greater percentage of vehicles here. If an 

increasing share of the vehicle fleet is electrified, while the number and fuel efficiency of cars 

remains the same – and they are driven the same amount – on-location vehicle emissions will 

necessarily decrease. This is because all electricity to fuel the vehicles would be produced from a 

remote source, and any emissions associated with generating that electricity would occur remotely, 

where the electricity is produced. Therefore, electrifying a greater share of vehicles would of 

course be effective at reducing on-location emissions. 

However, jurisdictions are interested in reducing emissions from vehicle transportation in 

other ways, on top of just electrifying vehicles. Among the proposed methods of decreasing 

emissions are the other four approaches listed above: increasing fuel efficiency, increasing public 

transit use, having more people work from home, and increasing the price of gasoline. 

Below, I will explain the scholarly arguments for why each approach might or might not be 

reduce vehicle emissions. Then, in later sections of the chapter, I use empirical methods to test the 

actual effect of each approach on emissions. 



 

35 

 

 

Fuel efficiency 

One proposed method of decreasing on-site vehicle emissions is to increase vehicles’ fuel 

efficiency: their average miles per gallon (US White House 2016; Taylor 2018). The argument 

for fuel efficiency is simple. Most cars burn gasoline to create energy, which moves the car. If 

people continue to drive the same distance in cars, but cars need less energy to drive that 

distance, aggregate energy use in cars will decrease – and, with energy, gasoline use and vehicle 

emissions. Therefore, increasing vehicles’ average miles per gallon may decrease emissions. 

However, there is reason to question whether increasing average vehicle miles per gallon 

would actually decrease gasoline use and emissions. The above reasoning assumes that the amount 

people drive is constant, while the amount of energy used to drive that distance is potentially 

variable. What if the amount people drive is not constant, but is instead a trade-off between the 

amount of time and money people have to spend driving versus the benefits of a trip? 

Economists and Sociologists have noted that increasing the energy efficiency of an 

appliance – the amount it can do with any given energy input – decreases the energy costs 

associated with one use of the appliance. Because the cost of any one use of the appliance has 

decreased, people and companies may choose to use more of the product. This has been termed the 

“rebound effect” or the “Jevons paradox” (Herring 2006, York 2006, Saunders 2013, York and 

McGee 2016). 

According to this logic, increasing vehicles’ miles per gallon decreases the amount of fuel 

needed to make a trip – decreasing the monetary cost of driving. If this is the case, then increasing 

vehicle fuel efficiency may change people’s calculus about the costs and benefits of a trip, pushing 

people to drive more. This, in turn, would cause fuel efficiency to reduce emissions less than we 

would expect, based on per-mile energy savings. 



 

36 

 

 

Public transit 

A second proposed method of decreasing on-site vehicle emissions is to increase public 

transit use (see, for example, Aronoff et al 2019). Vehicle travel in personal vehicles, with (at 

most) a few people inside, is a highly inefficient energy use. Theoretically, energy use from 

vehicle transportation could be decreased if people used public transportation such as subways, 

trains, or buses for most vehicle trips. For example, the same bus might be able to transport thirty 

people, rather than all thirty making the same trip in their own personal vehicles. Their personal 

vehicles will use less fuel to make that trip than a bus, but not thirty times less – so all thirty 

people traveling together on a bus rather than driving their own cars will decrease net energy use. 

The potential issue with this logic is that public transit use might be added on top of 

personal vehicle use, rather than replacing it. York (2006; 2012; 2017) has discussed the 

phenomenon in which the provision of a new energy source does not replace an old energy source; 

instead, overall energy consumption increases, as people continue to use the old energy source at 

the previous rate and also begin using the new energy source. The idea is that, within a capitalist 

market economy, firms’ production decisions are made not to minimize energy use, but rather, to 

maximize profit (Clark and York 2008). Similarly, individual consumers may not make 

consumption decisions with the aim of minimizing energy consumption, but instead, with some 

other aim such as maximizing comfort. York (2021: 766) calls these phenomena the “displacement 

paradox.” 

This logic may apply to public transit use. The proponents of public transit use see it as a 

substitute for personal vehicle use. However, it may be that people who gain access to public 

transit continue to use personal vehicles for the trips they were previously taking and use public 

transit to do additional trips. Alternatively, they might switch over to public transportation for 

some trips, such as going to school or work, but if transit is cheap, this may save the traveler 



 

37 

 

 

money – money they spend on additional trips in a personal vehicle. In these ways, it is possible 

that public transit use will not “displace” personal vehicle use, but rather, be added on top of it. 

Work from home 

A third proposed method to decrease on-site emissions from personal vehicle use is to 

have a larger percentage of the workforce work from home. Travel to and from work accounts 

for a large proportion of all personal vehicle trips in the United States – about 17% as of 2017 

(US Bureau of Transportation Statistics, n.d.). Therefore, theoretically, if a large portion of 

workers switched from in-person to remote work modalities, total vehicle use in the United 

States could be reduced significantly. 

This potential approach to emission reduction was discussed during the early Covid 

pandemic, when many people switched to work from home arrangements for the first time. Some 

of those who made the switch wondered if people being switched to remote work might be 

decreasing vehicle emissions, and if so, whether maintaining remote work arrangements post-

pandemic might be a way to decrease carbon emissions from vehicle transportation long-term 

(Shreedhar et al 2022). 

The worry, once again, is that a switch to remote work might change people’s behavior 

such that they take more trips for other reasons. This could lead to smaller decreases in the total 

amount of driving than we would expect, based on the number of people switching to remote work 

(Ibid). For example, workers who are “working” remotely may take advantage of the remote 

modality to stop working early and drive somewhere else for some other reason. 

Higher gasoline prices 

A fourth proposed method to decrease on-site emissions from personal vehicles is to 

increase the price of the fossil fuels these vehicles run on. The logic of this method is simple: all 

else being equal, if the price of a good increases, consumers buy less of it. If we increase the 



 

38 

 

 

price of gasoline, people will buy less gasoline and, consequently, use less gasoline in their cars. 

Carbon emissions from personal vehicle use will decrease (Sterner 2007; FHWA 2008; Szasz 

2023). 

There is, however, a reason that increasing the price of fossil fuels consumed in vehicles 

might not decrease vehicle use and emissions as much as expected, at least in the US: the country 

is built out in a way that often forces residents to drive. Metropolitan areas are often divided into 

residential neighborhoods, where people live, and commercial or business districts, where people 

work and shop for goods. Walking from the residential neighborhoods to the business 

neighborhoods is often not feasible, and many metropolitan areas have insufficient public 

transportation for travel between neighborhoods. Public transit is often even sparser between cities, 

making it difficult to reach many cities without a car. 

In this context, it is not clear how much increasing gasoline prices will decrease personal 

vehicle use. Presumably, residents would respond to higher prices by buying more fuel efficient 

cars (FHWA 2008) and by decreasing vehicle use when they can. However, if the built 

environment makes it difficult for residents to shop, get to school or work, or see friends or family 

without driving, residents may be forced to continue driving even if gasoline becomes more much 

expensive. 

Data and Methods 

In this chapter, I test the effect of the four approaches on emissions from fossil fuel use in 

vehicles. I use fixed effect panel regression models. The data are US state-years, for all fifty 

states, for the years 2008-2019. The dependent variable is a measure of emissions from the 

consumption of fossil fuels for energy in vehicles. The key independent variables of interest are 

measures of real gasoline prices, average vehicle fuel efficiency, the percentage of workers who 



 

39 

 

 

take public transit to work (a proxy for overall public transit use), and the percentage of workers 

who work from home. The models include controls for a variety of other factors that may affect 

emissions. The models also control for the the time-invariant effect of being in a certain state, 

and the location-invariant effect of being in a certain year, on emissions. 

The logic of the modeling is that the differences in emissions from fossil fuel use in 

vehicles between the fifty United States may be partly caused by differences in fuel efficiency, 

public transit use, the prevalence of working from home, and the price of gasoline, as well as by a 

variety of demographic factors. After controlling for the other demographic factors we might 

expect to have an effect on emissions, we are left with state years that have different average 

vehicle fuel efficiency, different gas prices, different rates of public transit use and working from 

home, and unexplained differences in vehicle emissions. The computer solves this math problem to 

determine the average effect of each of these four characteristics of the economy on vehicle 

emissions. 

My models include multiple independent variables of interest in the same model. If I only 

include one of independent variables of interest at a time, the models may misattribute increases or 

decreases in emissions to the wrong variable of interest. 

Below, I discuss the empirical models in more detail. 

Units of observation and dependent variable 

The units of observation are state-years, for the fifty United States, for the years 2008-

2019. The dependent variable is transportation energy-related carbon dioxide emissions: an 

estimate of the total carbon dioxide emissions in a state-year from consumption of fossil fuels for 

energy in vehicles, calculated by the US Energy Information Administration. 



 

40 

 

 

Independent variables of interest 

My models include four independent variables of interest. The first independent variable 

of interest is the average, real retail price of gasoline, in 2012 dollars per million BTU, in each 

state-year, including state and federal fuel taxes but not including sale taxes. The second 

independent variable of interest is an estimate of the percentage of workers who work from home 

in each state-year. 

The third independent variable of interest is an estimate of the percentage of workers who 

take public transit to work. I use this variable as a proxy for public transit use. I expect that, in 

states where a larger share of people commute on public transit, public transit is available and 

affordable enough that it would also be used for other trips. 

The fourth independent variable of interest is vehicles’ average miles per gallon when 

traveling on highways in the state-year. I use this metric as a proxy for average fuel efficiency of 

cars in the state-year. 

Control variables 

My models include independent variables to control for other potential determinants of 

vehicle use. I include controls for population, because total vehicle use will likely correlate with 

the size of the population, and percentage of residents living in urban areas, since personal 

vehicle use will likely vary between urban and rural areas. I control for the age structure of the 

population – the ratio of children and seniors to adults aged 18 to 64 – because working age 

people are likely to have different driving habits, compared with seniors or people under 18. I 

control for mean household income, in 2012 dollars, because households with more disposable 

income may choose to drive more. I control for heating degree days and cooling degree days 

because people might respond to very cold or hot temperatures by walking and biking less and 

driving more. 



 

41 

 

 

Another factor that could affect driving behavior is whether states’ residents believe in, and 

worry about, man-made climate change. Residents who believe in the reality of man-made climate 

change, and are concerned about it, may make lifestyle decisions to limit their carbon footprint – 

such as cutting down on the amount they drive gas-powered vehicles. To control for the effect of 

belief in climate change on driving behavior, I include an independent variable estimating the 

percentage of people who believe climate change is mostly caused by human activities. 

Over my study period, states might have passed other climate policies which reduced 

carbon emissions in vehicle transportation through some other means than pricing, public transit, 

working from home, or fuel economy. Thus, I also include a variable to control for cumulative 

Democratic control of a state’s government over time, as a proxy for a state’s likelihood to pass 

climate policies – since the Democratic party is far more concerned than the Republican party with 

reducing carbon emissions (McCright et al 2016). 

It is not clear how much control of state government Democrats must have to pass climate 

policies. It is conceivable that Democrats controlling a state’s House, Senate, or Governorship, 

even without the other two, increases the likelihood of the state passing climate policies. It is also 

conceivable that control of one house of the legislature, or of the governorship, is not enough. 

Instead, Democrats might need to control the state House, Senate, and Governorship – what is 

sometimes termed a “government trifecta” (Ballotpedia n.d.) – to pass climate policies. 

Because I do not know which of these possibilities is correct, I create two different 

variables as proxies for Democratic control of state government and consequent climate policy 

adoption. I create a continuous variable whose count increases by one for every year since 2001 

that Democrats controlled the state’s House, one for every year that Democrats controlled the 

state’s Senate, and one for every year that Democrats controlled the Governorship; this variable is 



 

42 

 

 

a good proxy for climate policy adoption if the likelihood of climate policy adoption increases each 

time Democrats control either House or the Governorship. I also create a variable whose count 

increases by one for each year since 2001 that Democrats have had a “government trifecta” in the 

state: control of the state’s House, Senate, and Governorship, all at once. This variable is a good 

proxy for climate policy adoption if Democrats need a government trifecta to pass climate policies. 

I do not know which variable is a more accurate proxy for climate policy adoption, so I run 

my models twice – once using the first variable, and once using the second variable. Ideally, the 

regression results for the variables of interest will be robust to whichever proxy I use for climate 

policy adoption. 

Finally, states in which transportation, shipping, and warehousing businesses make up a 

large portion of the economy may have more vehicle travel, all else being equal. To control for the 

prevalence of the transportation and shipping industries in a state, I include a variable measuring 

the amount of GDP (in 2012 dollars) in these industries in the state-year. 

Main models 

My main models (Table 1, Models 1-4) include variables to control, to the best of my 

ability, for all possible causes of emissions other than the four independent variables of interest. 

I take the natural log of all variables. Ln(y) = b*ln(x) can fit many different relationships 

between an independent and a dependent variable, depending on the value of b. Thus, performing a 

natural log transformation on both my independent variables and my dependent variable allows the 

computer find to a best fit line which closely approximates the relationship between the variables 

in my data. For independent variables that never take a value of zero in my data, I use the 

transformation ln(y) = bln(x). For my independent variables that take a value of zero for any unit of 

observation, I instead use the transformation ln(y) = bln(x+1) – since ln(0) is undefined. 

For the independent variables that receive the transformation ln(y) = bln(x), a 1% increase 



 

43 

 

 

in the independent variable is associated with a b% increase in the dependent variable. For the 

variables that receive the transformation ln(y) = bln(x+1), a 1% increase in the independent 

variable is associated with a b * x/(x+1) % increase in the dependent variable. Variables that 

receive this latter transformation mostly have values of 10 or more. Thus, for most units of 

observation, b * x/(x+1) will be close to b; a 1% increase in the independent variable will be 

associated with approximately a b% increase in the dependent variable. 

Table 1 Model 1, below, includes variables to control for as many other potential causes of 

emissions as possible, as well as the independent variables measuring average gasoline price and 

the percentage of workers who work from home. Table 1 Model 2 includes all four independent 

variables of interest. 

Model 1 measures the effect of a 1% increase in average gasoline price and of a 1% 

increase in the percentage of workers working from home on emissions, holding everything else 

constant. 

Model 2 also measures the effect of a 1% increase in vehicle fuel efficiency – and of a 1% 

increase in the percentage of workers commuting on public transit – on vehicle emissions. 

However, Model 2 may not capture the full effect of gasoline prices on vehicle emissions. Higher 

gasoline prices may cause people to buy more fuel-efficient cars (so driving will cost less) or to use 

public transit more often. Because Model 2 includes variables for vehicle fuel efficiency and for 

the amount of public transit use, this model’s gas price variable only captures the effect of gasoline 

prices on vehicle emissions, through other means than inducing people to buy more fuel efficient 

cars or use public transit. 

Table 1 Models 3 and 4 are the same as Models 1 and 2, respectively, except that Models 3 

and 4 operationalize cumulative Democratic control of state government differently. Models 1 and 



 

44 

 

 

2 operationalize cumulative Democratic control of state government with a counting variable 

whose count increases by one every time Democrats control the state’s House, Senate, or 

Governorship for an additional year; Model 2 uses a variable whose count increases by one each 

time the Democrats have a “government trifecta” – control of the state House, Senate, and 

Governorship – for an additional year. I do not know which measure is a more accurate proxy for 

climate policy adoption. Thus, I run the main models twice – once using each variable for 

Democratic control of state government – to see whether the results for the variables of interest are 

robust to how I operationalize cumulative Democratic control of state government. 

Table 1 Model 1 

𝑙𝑛(𝑒𝑛𝑒𝑟𝑔𝑦 − 𝑟𝑒𝑙𝑎𝑡𝑒𝑑 𝐶𝑂2 𝑒𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠 𝑖𝑛 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡𝑎𝑡𝑖𝑜𝑛𝑖𝑡)

= 𝛽1 𝑙𝑛(𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑖𝑡) + 𝛽2 𝑙𝑛(𝑎𝑔𝑒 𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑦 𝑟𝑎𝑡𝑖𝑜𝑖𝑡)

+𝛽3 𝑙𝑛(𝑢𝑟𝑏𝑎𝑛𝑖𝑧𝑎𝑡𝑖𝑜𝑛𝑖𝑡) + 𝛽4 𝑙𝑛(𝑚𝑒𝑎𝑛 ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝑖𝑛𝑐𝑜𝑚𝑒𝑖𝑡)

+𝛽5 𝑙𝑛(ℎ𝑒𝑎𝑡𝑖𝑛𝑔 𝑑𝑒𝑔𝑟𝑒𝑒 𝑑𝑎𝑦𝑠𝑖𝑡 + 1) + 𝛽6 𝑙𝑛(𝑐𝑜𝑜𝑙𝑖𝑛𝑔 𝑑𝑒𝑔𝑟𝑒𝑒 𝑑𝑎𝑦𝑠𝑖𝑡)

+𝛽7 𝑙𝑛(𝐷𝑒𝑚 𝑐𝑜𝑛𝑡𝑟𝑜𝑙 𝑜𝑓 𝑠𝑡𝑎𝑡𝑒𝑖𝑡 + 1) + 𝛽8 𝑙𝑛(𝑏𝑒𝑙𝑖𝑒𝑓 𝑖𝑛 𝑐𝑙𝑖𝑚𝑎𝑡𝑒 𝑐ℎ𝑎𝑛𝑔𝑒𝑖𝑡)

+𝛽9 𝑙𝑛(𝑠𝑡𝑎𝑡𝑒 𝐺𝐷𝑃 𝑖𝑛 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡𝑎𝑡𝑖𝑜𝑛𝑖𝑡)

+𝛽10 𝑙𝑛(𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑔𝑎𝑠𝑜𝑙𝑖𝑛𝑒 𝑝𝑟𝑖𝑐𝑒𝑖𝑡) + 𝛽11 𝑙𝑛(𝑝𝑐𝑡 𝑤𝑜𝑟𝑘 𝑓𝑟𝑜𝑚 ℎ𝑜𝑚𝑒𝑖𝑡)

+𝑢𝑖 + 𝑤𝑡 + 𝑒0𝑖𝑡

 

Table 1 Model 2 

𝑙𝑛(𝑒𝑛𝑒𝑟𝑔𝑦 − 𝑟𝑒𝑙𝑎𝑡𝑒𝑑 𝐶𝑂2 𝑒𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠 𝑖𝑛 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡𝑎𝑡𝑖𝑜𝑛𝑖𝑡)

= 𝛽1 𝑙𝑛(𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑖𝑡) + 𝛽2 𝑙𝑛(𝑎𝑔𝑒 𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑦 𝑟𝑎𝑡𝑖𝑜𝑖𝑡)

+𝛽3 𝑙𝑛(𝑢𝑟𝑏𝑎𝑛𝑖𝑧𝑎𝑡𝑖𝑜𝑛𝑖𝑡) + 𝛽4 𝑙𝑛(𝑚𝑒𝑎𝑛 ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝑖𝑛𝑐𝑜𝑚𝑒𝑖𝑡)

+𝛽5 𝑙𝑛(ℎ𝑒𝑎𝑡𝑖𝑛𝑔 𝑑𝑒𝑔𝑟𝑒𝑒 𝑑𝑎𝑦𝑠𝑖𝑡 + 1) + 𝛽6 𝑙𝑛(𝑐𝑜𝑜𝑙𝑖𝑛𝑔 𝑑𝑒𝑔𝑟𝑒𝑒 𝑑𝑎𝑦𝑠𝑖𝑡)

+𝛽7 𝑙𝑛(𝐷𝑒𝑚 𝑐𝑜𝑛𝑡𝑟𝑜𝑙 𝑜𝑓 𝑠𝑡𝑎𝑡𝑒𝑖𝑡 + 1) + 𝛽8 𝑙𝑛(𝑏𝑒𝑙𝑖𝑒𝑓 𝑖𝑛 𝑐𝑙𝑖𝑚𝑎𝑡𝑒 𝑐ℎ𝑎𝑛𝑔𝑒𝑖𝑡)

+𝛽9 𝑙𝑛(𝑠𝑡𝑎𝑡𝑒 𝐺𝐷𝑃 𝑖𝑛 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡𝑎𝑡𝑖𝑜𝑛𝑖𝑡)

+𝛽10 𝑙𝑛(𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑔𝑎𝑠𝑜𝑙𝑖𝑛𝑒 𝑝𝑟𝑖𝑐𝑒𝑖𝑡) + 𝛽11 𝑙𝑛(𝑝𝑐𝑡 𝑤𝑜𝑟𝑘 𝑓𝑟𝑜𝑚 ℎ𝑜𝑚𝑒𝑖𝑡)

+𝛽12 𝑙𝑛(𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑚𝑝𝑔 𝑜𝑛 ℎ𝑖𝑔ℎ𝑤𝑎𝑦𝑠𝑖𝑡)

+𝛽13 𝑙𝑛(𝑝𝑐𝑡 𝑝𝑢𝑏𝑙𝑖𝑐 𝑡𝑟𝑎𝑛𝑠𝑖𝑡 𝑡𝑜 𝑤𝑜𝑟𝑘𝑖𝑡)

+𝑢𝑖 + 𝑤𝑡 + 𝑒0𝑖𝑡

 

Table 1 Model 3 



 

45 

 

 

𝑙𝑛(𝑒𝑛𝑒𝑟𝑔𝑦 − 𝑟𝑒𝑙𝑎𝑡𝑒𝑑 𝐶𝑂2 𝑒𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠 𝑖𝑛 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡𝑎𝑡𝑖𝑜𝑛𝑖𝑡)

= 𝛽1 𝑙𝑛(𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑖𝑡) + 𝛽2 𝑙𝑛(𝑎𝑔𝑒 𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑦 𝑟𝑎𝑡𝑖𝑜𝑖𝑡)

+𝛽3 𝑙𝑛(𝑢𝑟𝑏𝑎𝑛𝑖𝑧𝑎𝑡𝑖𝑜𝑛𝑖𝑡) + 𝛽4 𝑙𝑛(𝑚𝑒𝑎𝑛 ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝑖𝑛𝑐𝑜𝑚𝑒𝑖𝑡)

+𝛽5 𝑙𝑛(ℎ𝑒𝑎𝑡𝑖𝑛𝑔 𝑑𝑒𝑔𝑟𝑒𝑒 𝑑𝑎𝑦𝑠𝑖𝑡 + 1) + 𝛽6 𝑙𝑛(𝑐𝑜𝑜𝑙𝑖𝑛𝑔 𝑑𝑒𝑔𝑟𝑒𝑒 𝑑𝑎𝑦𝑠𝑖𝑡)

+𝛽7 𝑙𝑛(𝑦𝑒𝑎𝑟𝑠 𝑜𝑓 𝐷𝑒𝑚 𝑡𝑟𝑖𝑓𝑒𝑐𝑡𝑎 𝑖𝑡 + 1) + 𝛽8 𝑙𝑛(𝑏𝑒𝑙𝑖𝑒𝑓 𝑖𝑛 𝑐𝑙𝑖𝑚𝑎𝑡𝑒 𝑐ℎ𝑎𝑛𝑔𝑒𝑖𝑡)

+𝛽9 𝑙𝑛(𝑠𝑡𝑎𝑡𝑒 𝐺𝐷𝑃 𝑖𝑛 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡𝑎𝑡𝑖𝑜𝑛𝑖𝑡)

+𝛽10 𝑙𝑛(𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑔𝑎𝑠𝑜𝑙𝑖𝑛𝑒 𝑝𝑟𝑖𝑐𝑒𝑖𝑡) + 𝛽11 𝑙𝑛(𝑝𝑐𝑡 𝑤𝑜𝑟𝑘 𝑓𝑟𝑜𝑚 ℎ𝑜𝑚𝑒𝑖𝑡)

+𝑢𝑖 + 𝑤𝑡 + 𝑒0𝑖𝑡

 

Table 1 Model 4  

𝑙𝑛(𝑒𝑛𝑒𝑟𝑔𝑦 − 𝑟𝑒𝑙𝑎𝑡𝑒𝑑 𝐶𝑂2 𝑒𝑚𝑖𝑠𝑠𝑖𝑜𝑛𝑠 𝑖𝑛 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡𝑎𝑡𝑖𝑜𝑛𝑖𝑡)

= 𝛽1 𝑙𝑛(𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑖𝑡) + 𝛽2 𝑙𝑛(𝑎𝑔𝑒 𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑦 𝑟𝑎𝑡𝑖𝑜𝑖𝑡)

+𝛽3 𝑙𝑛(𝑢𝑟𝑏𝑎𝑛𝑖𝑧𝑎𝑡𝑖𝑜𝑛𝑖𝑡) + 𝛽4 𝑙𝑛(𝑚𝑒𝑎𝑛 ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝑖𝑛𝑐𝑜𝑚𝑒𝑖𝑡)

+𝛽5 𝑙𝑛(ℎ𝑒𝑎𝑡𝑖𝑛𝑔 𝑑𝑒𝑔𝑟𝑒𝑒 𝑑𝑎𝑦𝑠𝑖𝑡 + 1) + 𝛽6 𝑙𝑛(𝑐𝑜𝑜𝑙𝑖𝑛𝑔 𝑑𝑒𝑔𝑟𝑒𝑒 𝑑𝑎𝑦𝑠𝑖𝑡)

+𝛽7 𝑙𝑛(𝑦𝑒𝑎𝑟𝑠 𝑜𝑓 𝐷𝑒𝑚 𝑡𝑟𝑖𝑓𝑒𝑐𝑡𝑎 𝑖𝑡 + 1) + 𝛽8 𝑙𝑛(𝑏𝑒𝑙𝑖𝑒𝑓 𝑖𝑛 𝑐𝑙𝑖𝑚𝑎𝑡𝑒 𝑐ℎ𝑎𝑛𝑔𝑒𝑖𝑡)

+𝛽9 𝑙𝑛(𝑠𝑡𝑎𝑡𝑒 𝐺𝐷𝑃 𝑖𝑛 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡𝑎𝑡𝑖𝑜𝑛𝑖𝑡)

+𝛽10 𝑙𝑛(𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑔𝑎𝑠𝑜𝑙𝑖𝑛𝑒 𝑝𝑟𝑖𝑐𝑒𝑖𝑡) + 𝛽11 𝑙𝑛(𝑝𝑐𝑡 𝑤𝑜𝑟𝑘 𝑓𝑟𝑜𝑚 ℎ𝑜𝑚𝑒𝑖𝑡)

+𝛽12 𝑙𝑛(𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑚𝑝𝑔 𝑜𝑛 ℎ𝑖𝑔ℎ𝑤𝑎𝑦𝑠𝑖𝑡)

+𝛽13 𝑙𝑛(𝑝𝑐𝑡 𝑝𝑢𝑏𝑙𝑖𝑐 𝑡𝑟𝑎𝑛𝑠𝑖𝑡 𝑡𝑜 𝑤𝑜𝑟𝑘𝑖𝑡)

+𝑢𝑖 + 𝑤𝑡 + 𝑒0𝑖𝑡

 

In the equations above, i represents the state, t represents the year, u~i represents the 

constant effect of being in a certain state, across years, and w~t represents the constant effect of 

being in a certain year, across states. 

Statistical significance indicators 

Some statistical studies sample a larger population and use statistical analysis to find 

correlations in the sample. These studies often use p-values to assess the likelihood that 

correlations in the sample hold for the larger population. 

This study uses population data. Thus, the correlations in my data are the correlations in 

the larger population. Calculating “p-values” is unnecessary to determine the correlations between 

variables in the population. 



 

46 

 

 

Thus, while I report my regression results’ associated “p-values” by convention, I do not 

use the calculated “p-values” when interpreting my results. Instead, I interpret the correlations 

between variables based on the sign and magnitude of the regression coefficients. 

Robustness checks 

Regression models are large math problems. The researcher feeds the computer a dataset, 

specifies which variables relate to each other, and specifies many aspects of the relationship 

between the variables: which variable is the dependent variable, whether the relationship 

between the dependent variable and particular independent variables is linear. Then, given the 

dataset, the choice of which variables relate to each other, and the researcher’s determination of 

how the different variables relate to each other, the researcher asks the computer to calculate the 

sign and magnitude of the relationship between the dependent variable and each independent 

variable. The danger with a regression model is that, if the researcher feeds the computer 

inaccurate data, chooses the wrong variables to include in the model, and/or specifies the wrong 

relationship between variables in the model, the model can yield inaccurate results. 

I have been thoughtful about the data I used, the variables I included in my regression 

models, and the relationships between the variables in those models. However, it is still possible 

that my choice to include or exclude particular variables skewed my regression results, such that 

my main model results are incorrect as to the effect of my independent variables of interest on 

vehicle emissions. 

To begin testing this possibility, in Supplement 1 Tables S1 and S2, I run my vehicle 

regression models again, including only one independent variable of interest per model run. 

Ideally, these supplementary models will find a similar relationship between my dependent 

variable and each independent variable of interest, when only including that one independent 

variable of interest in the model, as my main models find when I include multiple independent 



 

47 

 

 

variables of interest in the same model. Such a finding would indicate that my results are at least 

somewhat robust to my decisions about which independent variables of interest to include in the 

model. 

Data sources 

I retrieved data on emissions from consumption of fossil fuels for energy within vehicles 

from the US Energy Information Administration (EIA 2024a). The population in each state, in 

each year, was estimated by the US Census Bureau; I retrieved these estimates from the Federal 

Reserve of St. Louis (FRED 2024). I retrieved data on the average motor gasoline price, heating 

degree days, and cooling degree days in different state-years from the Energy Information 

Administration’s State Energy Data System (EIA 2024b, EIA 2024c). I retrieved the amount of 

state GDP in transportation and shipping from the US Bureau of Economic Analysis (US BEA 

2023). 

I retrieved data on the number of vehicle-miles traveled on highways, and on total use of 

gasoline and gasohol on highways, in each state-year, from the US Federal Highway 

Administration (FHWA 2024a, 2024b). For each state-year, I divided the number of vehicle-miles 

traveled on highways by the total highway use of gasoline and gasohol to calculate vehicles’ 

average miles per gallon on highways in that state-year. 

I retrieved the number of housing units in urban areas, and the total number of housing 

units, in the years 2010 and 2020, from the US Census Bureau (Census Bureau 2010, 2020). I 

divided the number of housing units in urban areas from the total number of housing units to 

approximate the percentage of the population living in urban areas, in each state, in those years. I 

then linearly interpolated the data from years 2010 and 2020 to create annual estimates of the 

percentage of the population in urban areas for every year from 2008 to 2019. 

The Yale Climate Communications has created estimates of the percentage of people in 



 

48 

 

 

each state, in each year, who believe climate change is mostly caused by human activities; I 

retrieved the estimates for years 2010 to 2019 from the project’s website and received the 2008 

data directly from the project via email (Howe et al 2015; Marlon et al 2022). I constructed the 

proxy for Democratic control of state government using data on the composition of state 

legislatures from the National Conference of State Legislatures (NCSL 2011, 2024) and data on 

governorships from the Inter-university Consortium for Political and Social Research (Kaplan 

2021). 

A number of my independent variables – my measures of the percentage of workers 

commuting on public transportation or working from home, of the age dependency ratio, and of 

mean household income – are estimates from the American Community Survey (ACS 2024a,ACS 

2024b, ACS 2024c). The ACS is administered each year to a sample of people in each US state, 

and the Census Bureau uses survey results to estimate values for the whole population. The ACS 

calculates population estimates based on a single year of survey data and estimates based on five 

consecutive years of survey data. For my variables based on ACS data, for the years 2008 to 2017, 

I use the five-year estimate centered around the current year: I use the five-year estimate for 2006-

2010 as my value for 2008, the 2007-2011 estimate as my value for 2009, and so on. 

The covid pandemic resulted in large changes to some of the variables I am examining – 

for example, the prevalence of work-from-home arrangements – so survey data from during the 

pandemic should not be used to estimate population data from the pre-pandemic period. Thus, I 

cannot use the method described above to estimate population values for the years 2018 and 2019. 

Instead, for those two years, I use the population estimates based on just that year of survey data. 

Results 

 

Table 1: Effect of characteristics on CO2 emissions in vehicles 



 

49 

 

 

 Model 1 Model 2 Model 3 Model 4 

log(pop_people) 1.01 *** 1.00 *** 1.04 *** 1.01 *** 

 (0.15) (0.15) (0.15) (0.14) 

log(age_dep) 0.26 0.34 0.27 0.34 

 (0.19) (0.18) (0.19) (0.18) 

log(pct_urban) -0.04 0.11 -0.00 0.12 

 (0.31) (0.30) (0.31) (0.30) 

log(house_inc_2012dollars) 0.24 0.12 0.23 0.12 

 (0.14) (0.14) (0.14) (0.14) 

log(hdd + 1) 0.02 0.02 0.02 0.01 

 (0.03) (0.03) (0.03) (0.03) 

log(cdd) -0.02 -0.02 -0.02 -0.02 

 (0.01) (0.01) (0.01) (0.01) 

log(state_dem_c + 1) -0.00 -0.00   

 (0.02) (0.02)   

log(pct_clim_cause_hum) -0.06 -0.05 -0.08 -0.06 

 (0.06) (0.06) (0.06) (0.06) 

log(gdp_tran_2012dollar) 0.14 *** 0.13 *** 0.15 *** 0.13 *** 

 (0.03) (0.03) (0.03) (0.03) 

log(gas_price_2012dollars) -0.21 ** -0.13 -0.22 ** -0.13 

 (0.08) (0.07) (0.08) (0.07) 

log(pct_work_home) -0.08 * -0.06 -0.08 * -0.06 

 (0.03) (0.03) (0.03) (0.03) 

log(hw_mpg)  -0.37 ***  -0.37 *** 

  (0.05)  (0.05) 

log(pct_transit)  -0.01  -0.01 

  (0.03)  (0.03) 

log(dem_trif_c + 1)   0.00 0.00 

   (0.00) (0.00) 

R^2 0.24 0.31 0.25 0.31 

Adj. R^2 0.14 0.21 0.15 0.21 

Num. obs. 600 600 600 600 

 

Table 1 Models 1, 2, 3, and 4 all find that increases in gasoline prices are associated with 

decreased vehicle emissions – whether or not the model controls for level of vehicle fuel efficiency 



 

50 

 

 

and for amount of public transit use. Models 1 and 3 find that, holding constant the control 

variables and the percentage of workers who work from home, a 1% increase in real, average retail 

gasoline price is associated with a statistically significant 0.21-0.22% decrease in on-site vehicle 

emissions, on average. Models 2 and 4 find that, holding constant all of the above as well as the 

level of vehicle fuel efficiency and the percentage of workers commuting on public transit, a 1% 

increase in real average gas price is associated with a non-significant 0.13% decrease in emissions, 

on average. 

The models also find that an increased percentage of workers who work from home is 

associated with decreased emissions, though the emissions reductions are smaller than those 

associated with gasoline prices. Models 1 and 3 find that, holding constant the controls and the 

average gasoline price, a 1% increase in the percentage of workers working from home is 

associated with a statistically significant 0.08% decrease in vehicle emissions, on average. Models 

2 and 4 find that, controlling for the above variables, as well as the level of vehicle fuel efficiency 

and the percentage of workers who commute on public transit, a 1% increase in the percentage of 

workers working from home is associated with a non-significant 0.06% decrease in emissions, on 

average. 

Models 2 and 4 also test the effects of vehicle fuel efficiency and public transit use on 

emissions, all else being equal. Models 2 and 4 find that, all else being equal, a 1% increase in 

average vehicle miles per gallon on highways is associated with a statistically significant 0.37% 

decrease in vehicle emissions, on average. The correlations between vehicle fuel efficiency and 

vehicle emissions in Models 2 and 4 are the highest correlations between any variable of interest 

and vehicle emissions in any of my four models. 

On the other hand, Models 2 and 4 find a negligible correlation between the percentage of 



 

51 

 

 

people who take public transit to work and vehicle emissions. Models 2 and 4 find that, all else 

being equal, a 1% increase in the percentage of people who take public transit to work is associated 

with a non-significant 0.01% decrease in emissions, on average. 

Moving to the controls, all four models (Table 1, Models 1-4) find that a 1% increase in 

population is associated with approximately a 1% increase in vehicle emissions, while a 1% 

increase in the quantity of real GDP in transportation industries is associated with a 0.13-0.15% 

increase in emissions, on average. These results are statistically significant in all four models. All 

four models also find that real mean household income and the age dependency ratio are 

positively, but non-significantly, correlated with emissions, on average. 

On the other hand, the models find that a 1% increase in the percent of people who believe 

in man-made climate change is associated with a small (0.05-0.08%) decrease in vehicle 

emissions, on average. These results are not statistically significant. 

The models find that other variables – heating degree days, cooling degree days, and either 

cumulative measure of Democratic control of state government – have little correlation with 

emissions. The models find that a 1% increase in any of these four variables is associated with a 

non-significant 0.00 - 0.02% change in emissions, on average. 

In Supplement 1, I ran regression models which only include one independent variable of 

interest, each (Tables S1 and S2, Models 1-4), to see if these models had similar findings as to the 

effect of my variables of interest on emissions. The models which include real, average retail 

gasoline price but not the other independent variables of interest (Tables S1 and S2, Model 1) find 

that a 1% increase in real, average retail gasoline price is associated with a statistically significant 

0.18-0.20% decrease in vehicle emissions, on average. This is similar to the association in the main 

models which included retail gasoline price and the percentage of workers who work from home 



 

52 

 

 

but not average vehicle miles per gallon on highways or the percentage of workers who take public 

transit to work (main models 1 and 3). The supplementary models and the main models had similar 

findings concerning the effect of increased vehicle miles per gallon on highways, the effect of an 

increased percentage of workers working from home, and the effect of an increased percentage of 

workers who take public transit to work. Collectively, these supplementary model results provide 

more robustness to my main model results concerning the effect of gasoline price on vehicle 

emissions (when I do not control for vehicle fuel efficiency), the effect of workers working from 

home, the effect of vehicle fuel efficiency, and the effect of workers commuting on public transit. 

Discussion 

My models yield three major findings. The first finding is that increasing vehicle fuel 

efficiency is effective at decreasing on-site vehicle emissions. According to Table 1 Models 2 

and 4, holding gas price and other determinants of vehicle use constant, a 1% increase in average 

miles per gallon (on highways) in vehicles is associated with a 0.37% decrease in vehicle 

emissions. 

As previously noted, improved vehic