SATISFIED WITH THE SAFETY SCHOOL: RANK, CHOICE, AND COMPETITION WITHIN THE COLLEGE ADMISSIONS MANIA by KEITH FRAZEE A DISSERTATION Presented to the Department of Educational Methodology, Policy, and Leadership and the Graduate School of the University of Oregon in partial fulfillment of the requirements for the degree of Doctor of Philosophy December 2019 DISSERTATION APPROVAL PAGE Student: Keith Frazee Title: Satisfied with the Safety School: Rank, Choice, and Competition within the College Admissions Mania This dissertation has been accepted and approved in partial fulfillment of the requirements for the Doctor of Philosophy degree in the Department of Educational Methodology, Policy, and Leadership by: Dr. Gerald Tindal Chairperson Dr. Michael Bullis Core Member Dr. Roger J. Thompson Core Member Dr. John Seeley Institutional Representative and Kate Mondloch Interim Vice Provost and Dean of the Graduate School Original approval signatures are on file with the University of Oregon Graduate School. Degree awarded December 2019. ii © 2019 Keith Frazee ii i DISSERTATION ABSTRACT Keith Frazee Doctor of Philosophy Department of Educational Methodology, Policy, and Leadership December 2019 Title: Satisfied with the Safety School: Rank, Choice, and Competition within the College Admissions Mania Over 16% of entering college students attend more than one university’s new student orientation program. How does attending multiple orientations affect the likelihood of students’ enrollment a university? Similarly, do students always attend their top-ranked college when admitted? This manuscript presents results from binary logistic regressions attempting to better understand why some students may attend a university’s orientation but not arrive for the first day of classes. Independent variables include orientation attendance and the rank students assign each college in their choice set. Additional variables investigated include cohort, estimated household income, high school GPA, SAT/ACT score, residency, proximity of the college from home, gender, first-generation status, and waitlist status. Among the results, rank choice, residency, and high school GPA provide statistically significant results though with limited effect. iv CURRICULUM VITAE NAME OF AUTHOR: Keith Frazee GRADUATE AND UNDERGRADUATE SCHOOLS ATTENDED: University of Oregon, Eugene University of Missouri, Columbia Baylor University, Waco, Texas DEGREES AWARDED: Doctor of Philosophy, Educational Leadership, 2019, University of Oregon Master of Arts, Educational Leadership, 2008, University of Missouri Bachelor of Science, Secondary Education, 2005, Baylor University AREAS OF SPECIAL INTEREST: College Choice and Higher Education Enrollment Student Affairs Administration Decision-Making and Management PROFESSIONAL EXPERIENCE: Assistant Vice President and Chief of Staff, Student Services and Enrollment Management, University of Oregon, November 2019–present Assistant Director of Orientation Programs, University of Oregon, June 2013– November 2019 Assistant Director of Campus Life, Westmont College, June 2012–May 2013 Assistant Director of Student Activities, Baylor University, June 2008–June 2012 Coordinator of Peer Education, University of Missouri, August 2006–May 2008 PRESENTATIONS AND PUBLICATIONS: Smith, J., Beach, P., & Frazee, K. (2016). Teaching across boundaries: An evaluation of technology use in a doctoral education program. In 2017 Yearbook of v Teacher Education (pp. 275–288). Brno: International Council on Education for Teaching. More Than a Logo: Funding First-Year Student Programs through Sponsorships NASPA Annual Conference presentation, March 11, 2019, Los Angeles, California We Didn’t Plan for That: Crisis Management during Campus Events NASPA Annual Conference presentation, March 12, 2018, Philadelphia, Pennsylvania Your Student Staff Can Help Prevent Sexual Violence NODA Annual Conference presentation, October 25, 2015, Denver, Colorado v i ACKNOWLEDGMENTS I wish to express my deepest appreciation for Dr. Gerald Tindal and Dr. Michael Bullis for their guidance, mentorship, and wisdom throughout the preparation of this manuscript. They taught me to say what I mean and to not use 10 words when two will suffice. Special thanks are also due to Jonathan Jacobs and Dr. Shawn Sorenson for their consultation on data collection and their openness to my persistent inquiries. I also thank my colleagues in Student Services and Enrollment Management at the University of Oregon for their professional and personal encouragement. It is an honor to work alongside such an outstanding team. As the leader of that team, Dr. Roger J. Thompson has played a special role in my professional and academic journey. My deepest gratitude is due to him for his belief in all students’ ability to succeed, including this student. Much of the success of this manuscript and my time in graduate school is due to Kerry, Addilyn, and Beau. They are my motivation, my joy, and my hope for the future. vi i I dedicate this manuscript to Alan and Stephanie Frazee, whose expectations of me to always do my best have led me here. vi ii TABLE OF CONTENTS Chapter Page I. INTRODUCTION .................................................................................................... 1 Problem of Practice ................................................................................................ 3 Theoretical Frameworks ........................................................................................ 5 II. LITERATURE REVIEW ........................................................................................ 17 Digital Search Procedure ....................................................................................... 17 Synthesis of Empirical Literature .......................................................................... 24 Research Questions ................................................................................................ 35 III. METHODS ............................................................................................................ 39 Research Design..................................................................................................... 39 Student Sample ...................................................................................................... 40 Setting .................................................................................................................... 41 Instruments ............................................................................................................. 42 Procedure ............................................................................................................... 47 Model Specification ............................................................................................... 50 Analysis.................................................................................................................. 59 IV. RESULTS .............................................................................................................. 64 Descriptive Statistics .............................................................................................. 64 Research Question Results ..................................................................................... 80 V. DISCUSSION ......................................................................................................... 88 ix Chapter Page Statistically Significant Results ............................................................................. 88 Statistically Non-Significant Results ..................................................................... 90 Limitations ............................................................................................................. 90 VI. CONCLUSIONS AND IMPLICATIONS ............................................................ 93 Implications for Practitioners ................................................................................. 93 Suggested Future Research .................................................................................... 95 Changing Landscape of College Admissions in the United States ........................ 96 APPENDICES ............................................................................................................. 101 PORTIONS OF UNIVERSITY OF OREGON’S APPLICATION FOR ADMISSION ......................................................................................................... 101 2014 CIRP FRESHMAN SURVEY INSTRUMENT ........................................... 104 2018 UO FRESHMAN SURVEY INSTRUMENT .............................................. 110 DATA USE AND NON-DISCLOSURE AGREEMENT ..................................... 112 REFERENCES CITED ................................................................................................ 115 x LIST OF FIGURES Figure Page 1. Funnel of theoretical framework ............................................................................ 6 2. Tinto’s Theory of Student Departure. .................................................................... 14 3. Path diagram of the Theory of Work Adjustment ................................................. 15 4. Results narrowed from the third literature search to the final pool of articles ...... 23 5. Path diagram of models for research questions 1, 2, and 3.................................... 36 6. Path diagram of models for research question 4 .................................................... 37 7. Path diagram of models for research questions 5 and 6......................................... 38 8. Frequency distribution of student responses by estimated annual family income . 72 x i LIST OF TABLES Table Page 1. Databases, Searches, Search Terms, and Results .................................................. 19 2. Research Methodology of a Sample of Articles within the Literature Review ..... 25 3. Shadish, Cook, and Campbell (2002) Threats to Internal Validity........................ 26 4. Settings of a Sample of Articles within the Literature Review ............................. 27 5. Participants of a Sample of Articles within the Literature Review ....................... 29 6. Variables and the Instruments from Which They Came by Year .......................... 43 7. Variable Names by Type and Their Role in the Analysis ..................................... 51 8. Family Income Response Choice Recoding .......................................................... 55 9. Frequency Distribution of Freshman Class Size, Response Rate, and Instrument Used by Cohort ...................................................................................................... 64 10. Descriptive Statistics for All Variables by Cohort ................................................ 66 11. Frequency Distribution of Student’s Rank of UO as Their College Choice.......... 68 12. Frequency Distribution of How Many Universities’ Orientation Sessions Students Attend ...................................................................................................... 69 13. Estimated Family Total Income Last Year by Enrollment on Census Date .......... 71 14. High School GPA and SAT/ACT Score by Enrollment Decision ......................... 73 15. Residency by Enrollment on Census Date ............................................................. 74 16. Miles from Home by Residency and Enrollment on Census Date ........................ 76 17. Gender by Enrollment on Census Date .................................................................. 77 18. Generational-Status by Enrollment on Census Date .............................................. 78 19. Waitlisted Students by Cohort and Enrollment on Census Date ........................... 79 20. Correlations of Rank and Orientation Attendance with Enrollment Status ........... 80 xi i Table Page 21. Combined Independent Logistic Regression Results Predicting Likelihood of Enrollment based on Rank, then Likelihood of Enrollment based on Orientation Attendance ............................................................................................................. 82 22. Correlation of Rank with Orientation Attendance ................................................. 82 23. Logistic Regression for RQ4 Predicting Likelihood of Enrollment based on High School GPA, SAT-ACT Score, Cohort, Family Income, Residency, Proximity to Campus, Gender, Generational Status, and Waitlist Status ................................... 84 24. Logistic Regression for RQ5 Predicting Likelihood of Enrollment based on Rank Choice, High School GPA, SAT-ACT Score, Cohort, Family Income, Residency, Proximity to Campus, Gender, Generational Status, and Waitlist Status .............. 85 25. Logistic Regression for RQ6 Predicting Likelihood of Enrollment based on Orientations Attended, High School GPA, SAT-ACT Score, Cohort, Family Income, Residency, Proximity to Campus, Gender, Generational Status, and Waitlist Status ........................................................................................................ 87 xi ii CHAPTER I INTRODUCTION Choosing which college to attend typically affects major elements of an individual’s life trajectory, including their economic, professional, and social wellbeing (Pascarella, Terenzini, & Wolfle, 1986). Consequently, students would ideally make the high stakes decision of where to attend college following a process of deep research, self- reflection, and a rational cost-benefit analysis. However, the choice of which college to attend is, to some extent, unscientific as students work with incomplete data, tight timelines, and many external influences. Students may feel pressure from family, friends, guidance counselors, and college ranking publications. Students may feel pressure through a desire to attend college where their best friends or significant other are attending. The sources of information and misinformation seem endless, yet the time it takes to consider such information is not. A comprehensive and systematic decision- making process including careful analysis and discernment is just not possible in today’s world of admissions mania (Bruni, 2015). Thus, students choose their college with limited and bounded rationality (Simon, 1955, 1959). The choice is a stressful one even among the best of circumstances, such as when students receive multiple admission offers including admission to their top-choice school (Galotti, 1995). An emerging national trend suggests high school seniors apply to a steadily increasing number of universities each year in the United States (M. Clinedinst, Koranteng, & Nicola, 2015). When a student chooses which universities to apply to, their options seem endless. They could apply to in-state or out-of-state universities, local regional colleges or highly selective, globally reputable universities. According to the 1 National Association of College Admission Counselors (M. E. Clinedinst & Patel, 2018), "between the Fall 2016 and Fall 2017 admission cycles, the number of applications from first-time freshmen increased four percent; applications from prospective transfer students increased by three percent; and international student applications increased by eight percent" (p. 3). This increase does not represent an increase in applicants but in applications. Students are applying to more institutions during their application and admission process. This trend is further demonstrated over time. In 1990 just nine percent of students applied to seven or more colleges, and by 2015, that group of applicants rose to 36 percent (Eagan, Stolzenberg, Ramirez, Aragon, Suchard, & Rios-Aguilar, 2016). The increase in applications can likely be contributed in part to the growing ease of applying. The Common Application represents one popular method for students to submit multiple college applications online. The Common Application (2017) is a consortium of universities sharing a single application for admission. The Common Application offers students the opportunity to explore nearly 700 colleges and apply to up to 20 through a single online application process. Submitting applications online is nearly universal with Clinedinst et al., (2015) finding that in 2014 “four-year colleges and universities received an average of 94 percent of applications online, up from 68 percent in Fall 2007 and only 49 percent in Fall 2005" (p. 3). The ease of submitting a college application online may seem like a helpful step in the complex process, but as the number of applications increase, many universities’ selectivity rates decrease. After students select which colleges will be in their choice set, colleges are then able to consider which applicants to admit through a selection process. A university’s selectivity rate is the ratio of students who are offered admission to those who applied, 2 expressed as a percentage. Higher selectivity equates to fewer applicants admitted. For fall 2016 the national average selectivity rate was 65.4%, meaning colleges and universities in the U.S. admitted on average 65.4% of their applicants (M. E. Clinedinst & Patel, 2018). With greater numbers of applicants, universities are forced to admit more students and thereby sacrifice their selectivity rate or risk a sharp decline in enrollment because not all students admitted at a university will enroll at that university. Yield defines the measure of students who actually attend a university after considering other admissions offers. For fall 2016 the national average yield rate was 33.6%, a decline remaining consistent from the previous two years when the national average yield was 35.1% in fall 2015 and 36.2% in fall 2014 (M. E. Clinedinst & Patel, 2018). As an example, for fall 2016 if the national averages applied to a single university, by admitting 65.4% of applicants, the example university would expect 33.6% of those to actually arrive for the first day of classes. In sum, nationally the percentage of applicants admitted is rising while the percent of yield is falling. If a university admits more students yet yields fewer, the predictability of which admitted students will ultimately enroll weakens. To use a real estate term, the increased applications and decrease in selectivity rate is tantamount to the admissions process transitioning from a seller’s market to a buyer’s market for all but the most selective universities. I will next describe how the current state of college admissions presents a problem for all involved. Problem of Practice The existing admissions mania presents a problem for both students and universities. Problems center around the themes of indecision and unpredictability, where student indecision (or delayed decision) may contribute to college enrollment 3 unpredictability. I next describe the problems of practice for both students and universities. Problem of practice for students. With a limited window of time during which to consider likely one of the most consequential decisions thus far in a student’s life, applying to multiple universities then being admitted to multiple universities counterintuitively becomes a burden. Adequate data-gathering and time for reflection may compromise the decision-making process. By applying to several colleges, students may think they increase the likelihood of admission to their top choice school, but they simply increase the likelihood of admission to a school. Some students will earn admission to multiple schools, none of which are their top choice. For students with multiple admissions options, more choice is not always better. Loewenstein (2000) described the cost of more choice by stating, “expanded choices can impose costs on decision-makers. It can absorb scarce time that people would prefer to spend on other activities, result in decision errors, and produce anxiety and regret” (p. 1). Considering multiple admission offers limits the time to build institutional commitment prior to students’ decisions, risking accepting admission from a school they may not be particularly excited about when compared to their top choice. One high school counselor, Lisa Sohmer, director of college counseling at the Garden School in Jackson Heights, in Queens, New York, described this in a New York Times article by stating, “when students file 20 or more applications, they’ve loaded on lots of ultra-competitive schools, so their list becomes disproportionately top-heavy. Or they throw in lots of schools at the end where they’re overqualified.” (Kaminer, 2017). What is the impact on students making a high-stakes decision with low data and limited 4 time? They may choose to attend a school that is not a strong fit, a school wherein they overmatch or undermatch with the university’s standards, thereby challenging their institutional commitment and risking drop-out or transfer. Problems within the college choice process affect not only students navigating through the process but colleges and universities also experience difficulties due to admissions mania. Problem of practice for universities. The increase in the number of applications submitted by prospective students creates uncertainty in university enrollment because many universities admit more students than will ultimately enroll (M. Clinedinst & Koranteng, 2017). As previously stated, for the fall 2017 term, the number of applications to college for first-time freshmen rose 4% from the previous year (M. E. Clinedinst & Patel, 2018, p. 3). Students submitting multiple applications increase the likelihood of receiving multiple admissions. However, if universities know or suspect the students they admit will also receive admission from other universities, enrollment becomes unpredictable. Universities depend on the predictability of enrollment to set budgets, plan personnel, and draft capital projects among other university operations. If students delay their decision, universities delay their planning process for the upcoming year. I next describe the theoretical frameworks through which I explore these problems of practice and the associated existing literature of college choice. Theoretical Frameworks Theories of student development in higher education, economics, and psychology provide insight into student decision-making related to college choice. Student development theories take into account many, often competing, influences on the student interests, such as their pre-college attributes and preparedness for college academic rigor. 5 Economic theories allow researchers to consider cost-benefit analysis as a critical sway for students in their decision-making process. Psychological theories provide general insight into how decisions are considered, formed, and reinforced. Taken together, theories within multiple disciplines provide a foundation for how I frame the exploration of college choice. I next describe the theoretical frameworks I drew from as I explored the literature related to decision-making and college choice. I begin with broad decision- making theories and narrow to college choice theories. Figure 1 displays how each theory narrows toward my focus on the construct of interest—institutional commitment. Figure 1. Funnel of theoretical framework 6 Decision-making theories. Generally, decision-making represents a problem- solving activity with the decision serving as the resolution. However, the decision- making process may include rationality and irrationality; it may include empirical and explicit knowledge as well as implied and tacit beliefs about the world. I next describe three decision-making theories and how they scaffold with one another to form a general framework prior to describing theories directly related to college choice. Rational choice theory. To help understand the nature of rational decisions, Homans (1961) developed rational choice theory, also known as rational action theory, which assumes perfect information. If all available information about a choice were taken into account, including probabilities and costs as well as benefits, a person would select the option that they determined was best. Rational choice theory assumes the sum of social behavior stems from a collection of individuals making decisions rationally toward their preferences (Oliveira, 2007). If applying the rational choice model to college choice, the student would be able to rank their admission alternatives without uncertainty due to incomplete information and without uncertainty of possible outcomes, and they would choose the college they determined was best. Yet such an omniscient circumstance is not possible; students are unable to obtain complete information, comprehensive probabilities, or thorough cost- benefit analysis about every college in their choice set. Students’ decide their college of choice with limitations on their ability to obtain all possible information, and so they decide their college of choice with bounded rationality. 7 Bounded rationality. Without perfect information people still maintain the ability to make consequential decisions rationally. Bounded rationality adds to my theoretical framework beginning with rational choice theory by accounting for lack of complete information or computational resources (Simon, 1955, 1959). Bounded rationality suggests decision-makers settle on a satisfactory solution rather than an optimal one due to the complexity of time limitations and cognitive constraints required in determining the optimal solution. Bounded rationality provides a helpful addition to the theoretical framework because the time students have between admission and the start of classes is just not enough to learn all available information about a university, much less multiple universities. Furthermore, students are unable to compute all possible probabilities of their decision or the complete cost-benefit of choosing one university over another. So they take the information they have and compensate for missing information with mental shortcuts or heuristics. However not accounting for emotion or intuition presents a critique of bounded rationality (Hanoch, Wood, & Rice, 2007). Selecting a university to attend may include intuitive responses not accounted for in the rational decision-making model of bounded rationality. A student’s first impression of a university can create a lasting impression whether accurate and rational or not. Bounded rationality does not take into account the effect of intuition on the decision-making process because intuition may not be rational. Prospect theory. Heuristics supplement a decision-making process that may not be purely rational, and Prospect Theory adds dimension and understanding of such mental shortcuts in the decision-making process. As a critique of rational choice theory, Prospect Theory presents an alternative idea of decision-making under risk using 8 heuristics (Kahneman & Tversky, 1979; Kahneman, 2014). Prospect Theory attempts to model real-life decision-making with heuristics, instead of decision-making under optimal conditions. Heuristics are generally understood as the building blocks of an educated guess or intuitive judgements, often colloquially described as “common sense.” Heuristics help ease the cognitive load in the decision-making process by filling in missing information (Gigerenzer & Selten, 2002). Kahneman and Tversky (1979) theorized that people, when deciding among options, set a reference point based on a heuristic that considers outcomes equivalent. After the reference point is set, people see lesser outcomes as losses and greater outcomes as gains. For example, when applied to a student’s college choice, after a student receives admission from a university, they rank any subsequent admissions from other universities as better or worse by comparison. Such ranking behavior poses an interesting question related to college choice — does a student’s ranking behavior of the colleges in their choice set affect their ultimate decision of which college to attend? I intend to explore students’ ranking behavior in my first and fourth research questions, which I list below. The authors also noted that people tend to remove characteristics when shared by all options, zeroing out similarities. When applied to college decision-making, Prospect Theory would suggest when students compare colleges, they dismiss similarities during the decision-making process. For example, if a student receives the same amount in scholarship awards from two colleges, in deciding which of the two to select, the student would likely dismiss scholarship awards in the decision-making process, instead focusing on other factors that differ between the two colleges. 9 College choice and retention theories. Following general decision-making theories, I next add to the theoretical framework with theories specific to college choice and retention. The general psychology of complex decision-making, while critical context, does not provide enough specificity to fully understand how students choose their college. To that end, I looked to theoretical models specific to higher education. What influences students to choose a college? Presumably, all students choose among a common, finite set of factors and variables of influence. While that set of factors and variables is certainly vast, it is doubtfully limitless. It likely includes cost, living arrangements, social circumstances, academic programs, and many more, though not infinitely more. The Model of College Choice (Chapman, 1981) identifies common, measurable influences, and it adds to the theoretical framework. Notably, I also included a theory of college retention (Tinto, 1987) that focuses chronologically after students chose their college because I believe some inferences may arise retrospectively. Why a student chooses to stay at a university may provide insight into why they chose the university in the first place. The theory of students’ decisions to stay or depart from college are therefore included in my theoretical framework. Finally, I included a theory of change related to workers adjusting to new jobs. Some inferences may be drawn from a person’s choice of job, so I included a career-related theory to build on the theoretical framework. Literature in each research area can provide insight to how students make decisions about college selection. I next describe theories of college choice, student retention, and work adjustment. Model of student college choice. When looking specifically at the decision- making of students choosing their college, Chapman (1981) identified factors that affect 10 college choice and grouped those factors into student characteristics and external influences, which included significant persons, fixed college characteristics, and college marketing efforts. Chapman suggests fixed student characteristics of influence include family income, aptitude test results, high school grade-point average, and level of educational aspiration. Missing from Chapman’s list of student characteristics is first- generation status, which may correlate closely with level of educational aspiration, though Chapman does not specifically address generational status in their Model of College Choice. The second level of Chapman’s model includes influences external to the student, such as significant persons, college characteristics, and college marketing efforts. Of the significant persons Chapman lists in the model, parents hold the greatest influence over students college choice while other significant persons include guidance counselors, peers, teachers, and lastly, college admission officers (Tillery, 1973). College characteristics include cost, financial aid, location, and availability of academic programs (Chapman, 1981). And finally, Chapman includes the college’s marketing efforts in the Model for College Choice. Of all the factors within the model, the marketing efforts seem to represent the factor least resistant to change. In other words, a college can add or reduce its marketing efforts with greater ease than adding or reducing its academic program offerings. Such possible volatility in marketing efforts may present a challenge in measurement over time because if college administrators change their marketing efforts year-to-year, it would be difficult to analyze any lasting effect on such marketing efforts. 11 The Model of College Choice seems to not account for student behavior which may indicate institutional commitment. For example, events for prospective students may influence the college choice, such as attending a campus tour or a new student orientation session, yet the model seems not to account for such events. The lack of inclusion of orientation attendance represents a gap in the research which is worth exploring. I intend to utilize the Model of College Choice to guide which factors I include in my model, while building on the model to include orientation attendance as a factor of interest, specifically in research questions two and five. The Model of College Choice (Chapman, 1981) pairs well with an additional higher education model, the Model of Student Departure (Tinto, 1987). Together, both models could add insight to my theoretical framework by considering why students choose their college but also why students retain or depart from college. Theory of student departure. Tinto (1987) presented a theoretical framework regarding student persistence in higher education, which represents a decision-making process downstream of the college choice process yet with applicable concepts. Figure 2 shows Tinto’s Theory of Student Departure. A students’ academic skill and prior schooling (or lack thereof) play an important role in their academic goals for higher education, and those goals and institutional commitment then influence academic performance. The theory considers pre-college attributes, such as academic skill and ability, as well as institutional commitment, the primary construct I explore for explained variance in the college choice process. Tinto’s theoretical framework tries to help explain why students choose to leave or stay at college. When applied upstream and considered for students choosing which college to attend, Tinto’s framework may provide unique 12 insight into students’ ultimate decision for where to attend by measuring their commitment to the college prior to the start of classes. A student’s decision to stay enrolled at a university may provide evidence for why they chose that university in the first place, though such linkage may not apply to all students in all circumstances. Theory of work adjustment. The matching of students and universities could also be explained by looking to theories of career choice. The Theory of Work Adjustment (Dawis & Lofquist, 1964) attempts to explain the reciprocal match of people and their career environment. A person seeks work with organizations that match their needs. Similarly, work organizations seek people with the capability of meeting the needs of the organization. The Theory of Work Adjustment uses indications of satisfaction and satisfactoriness, where individuals seek satisfaction with their work environment, and the workplace assesses a person’s abilities by a degree of satisfactoriness. A person’s satisfaction and a workplace’s satisfactoriness would jointly predict the person’s tenure in that workplace. As a critique of the theory, the Theory of Work Adjustment does not consider between-group differences by diverse populations such as under-represented minorities. Bias unrelated to a person’s abilities may play a factor in the degree of congruence of the satisfaction and satisfactoriness between a person and their work environment. Furthermore, Krumboltz (1992) points out some employees are “chronically undecided” about the fit of their career and goes further to suggest such indecision should not be viewed negatively but instead can be interpreted as a “profound philosophical perspective” that leads to health and happiness (p. 244). 13 Figure 2. Tinto’s Theory of Student Departure (Tinto, 1988), red box added for emphasis on the construct of interest 14 If applied to the relationship between students and universities, the Theory of Work Adjustment describes how a student seeks belonging that would match what they need, and universities in turn seek students who have the abilities to match the need of the university. The closer a student’s abilities, skills, attitude, and behavior align with the requirements of the university, the more likely they will fulfill the “job” of being a student well and be perceived as satisfactory by others. Figure 3 represents the theory model. However, special consideration should be given to students who persist in a state of indecision because, as Krumbotlz (1992) suggests, societal pressures may give way to anxiety and unhappiness during the decision-making process. Figure 3. Path diagram of the Theory of Work Adjustment (Dawis & Lofquist, 1964). 15 Summary. Taken together, the theories of decision-making and college choice suggest choosing a college to attend is a complex, multi-stage process. In an optimal scenario, students could select the college of best fit because they have all possible information and omniscience of probabilities of outcomes. However, students can not obtain perfect information prior to the decision, so they prioritize characteristics of the options, setting a reference point with the first option, and comparing subsequent options. Throughout the process institutional commitment, student characteristics, significant persons, college attributes, and college marketing efforts, all play influential roles through students’ college choice process toward the goal of finding a match. Such influences construct my theoretical framework within which I explore college choice. 16 CHAPTER II LITERATURE REVIEW The search for empirical research about college choice proceeded iteratively and included multiple databases and multiple attempts to refine search terms. In the following section, I describe my literature search process to thoroughly account for the process I followed in establishing the literature pool. I then synthesize the literature to suggest gaps that I work to fill. Digital Search Procedure Articles about the topic of college choices and decisions may not use a precise and universal key term for the college decision. My intent was to therefore search a series of college choice-related terms alongside the term “decision making.” My hope was to obtain a robust body of literature on the topic of college choice from which I could review and synthesize in order to locate a gap in the research. I searched the University of Oregon LibrarySearch using the Boolean phrase provided by the ERIC thesaurus, “(‘College Choice’ OR ‘College Admission*’ OR ‘College Applicant*’ OR ‘College Bound Student*’ OR ‘College Freshm*’ NOT ‘Community College’) AND (‘Decision Making’).” The search yielded 3,951 results after filtering for peer-reviewed articles. The search attempt indicates the precisicion of the Boolean search phrase helped cast a wide search of articles from which to review. I then considered narrowing the search parameters by date range. I considered two years as important in the history of the college 17 choice process in American higher education. First, the Higher Education Act of 1965 established the Pell Grant, a program of federal funding that opened college access to students from lower socio-economic status to postsecondary education (Higher Education Act, 1965). Although 1965 was a landmark year in higher education to consider as the start of a date range limit, narrowing the date range of my search to 1965–present excluded only 38 of the 3,951 results. The next year I considered of landmark importance to the college choice process was 1998, the year internet search engines were introduced to the public (Van Couvering, 2008). Internet searches democratized the college search process by providing detailed, searchable information on colleges to any prospective student with minimal effort and time. By narrowing the literature search date range to 1998–present, only 844 of the 3,951 results were excluded. Despite using a search phrase with several filters, the remaining 3,000 search results were unmanageable for an in-depth inspection of the literature for review. I then reviewed article titles across several pages of search results and could see that my search process identified articles unrelated to my specific topic of interest. For example, one title was, “Track Placement and the Motivational Predictors of Math Course Enrollment” (Reyes & Thurston, 2017), which did not fit my ultimate goal because it was specific to math course enrollment, not general college enrollment. I determined to edit my search phrase with new search terms in the hope of yielding results more precise to the topic of college choice and to reduce the volume of results that I would ultimately read. The final search terms are displayed in Error! Reference source not found.. 18 Table 1 Databases, Searches, Search Terms, and Results Search 1 Search 2 Search 3 “prospective college student” AND ("College Admission*" OR "College (“College Applicant*" OR "College choice AND admission Freshm*" OR "College Applicant*" Bound Student*”) OR "College Bound Student*" NOT AND "Community College") (“College Choice" OR “Decision AND Making” OR "College Selection”) ("Decision Making”) NOT (“Community College” OR Career) Database Results Database Results Database Results UO LibrarySearch 242 UO L ibrarySearch 4,471 UO L ibrarySearch 151 ERIC 1 ERIC 214 ERIC 160 Academic Search Premier 0 Acad emic Search Premier 74 Acad emic Search Premier 24 Total 243 Total 4,759 Total 335 Notes. All search results were narrowed to only include peer-reviewed journals, and the date range was narrowed to 1998–2017. Quotation marks = exact phrase, ALLCAPS = Boolean search operator 19 Revised search attempt. I decided to delete the term “college freshm*” from the search phrase because the phrase was pulling too many articles unrelated to the focus of my search about college choice. Such articles included emphasis on the freshman year of college but were not relevant to the college choice process preceding the start of the freshman year. The third search phrase was, (“College Applicant*" OR "College Bound Student*”) AND (“College Choice" OR “Decision Making” OR "College Selection”) NOT (“Community College” OR Career) Including filters for peer-reviewed journals and a date range of 1998–present, this new search phrase provided 151 search results from the University of Oregon LibrarySearch. I added to the results by using the same search phrase in the ERIC database and Academic Search Premiere database. Removing all duplicates among the databases left 152 articles for the initial review. I later added seminal studies related to decision-making and college choice that were outside the date range. Initial review and exclusion criteria. I initially reviewed the 152 results by reading the article titles, which enabled me to create a categorization system based on article theme. That is, I determined the literature in my initial review could be categorized based on the titles by the following themes: influence of financial aid on college choice, equity in the college admission process, influence of college rankings, higher education 20 marketing strategies, decision processes for college, and international case studies. I tagged each article with one or more category. Exclusion criteria. I decided to exclude articles that focused on college access instead of college choice. Generally, the term college choice refers to a process by which students decide whether and where to attend college, yet this term assumes students have access to college. The term college access, however, refers to who gets to attend college. Many articles include or conflate topics of college choice and college access. The differentiation in terms is necessary to fully understand the nature of college-going behavior for students, yet I excluded articles about college access to gain a discrete understanding of the choice process. For example, articles including both topics of college access and college choice focused generally on a critique of the entire admission process, such as, “The Admission Industrial Complex: Examining the Entrepreneurial Impact of College Access” (Liu, 2011). The literature is replete with articles placing emphasis on the limitations to college access as a bind on the college choice set. In order to focus on college choice more generally, I excluded articles focused on college access, in order to not draw conclusions more appropriately related to access instead of choice. I also excluded articles about college decision-making in countries outside of the United States that have substantially different college enrollment processes. For instance China holds a national standardized test for all students. The test is known as the National College Entrance Examination, or gaokao, and the result of this single test determines students’ admission to the highly stratified university system in China (Gu & Magaziner, 2016). I excluded articles focused on college admission in China, India, or Thailand 21 because their university systems, and centralized admissions processes utilize testing as the placement determinant, and student choice is not taken into account. I excluded “community college” from the search terms to focus on the decision- making of students who consider four-year universities for their baccalaureate degree. The weight of evidence indicates beginning at community college after high school tends to suppress degree attainment in part due to the barrier of getting admitted into a four- year university (Dougherty, 1992, 1994). I therefore excluded articles focused on community college. After excluding articles about college access, community college, or the countries of China, India, and Thailand, 63 results remained in the pool of articles. I then reviewed article abstracts to determine the final selection of research articles that are synthesized in search of gaps in the literature. Final review and selection of research literature. Of the 63 articles in the next- to-final literature pool, I searched for indicators of empirical research in the article abstracts. I read for terminology such as, “investigated,” “measured,” “participants,” “sample” as well as descriptions of statistical methodology. For example, one article abstract mentioned that the researchers examined, “variables on post graduating high school choices using multinomial logistic regression analysis” (Lee, Jara Almonte, & Youn, 2013). I sorted the 63 articles into categories as empirical or theoretical. A total of 25 article abstracts met my preliminary criteria for empirical research as displayed in Figure 4. Another 11 articles required further reading, as the abstracts did not provide an indication of methodology. I reviewed those 11 articles to determine their inclusion or not based on the aforementioned criteria for terminology of empiricism. I determined 6 of the 22 11 were empirical studies, and thus I added them to literature I review further. Specifically, I identified 31 articles that met my search parameters and that described the college choice process through empirical research methods. Figure 4. Article results narrowed from the third literature search to the final pool of articles. 23 I next read the collection of 31 articles to summarize the findings and conclusions of the empirical research. From that reading, I continued gathering relevant literature through ancestral searches from the references in the pool of articles. In the next section I synthesize the literature in search of gaps in the body of research. Following a synthesis of the empirical literature, I synthesize the theoretical articles regarding decision-making theory and college choice. Synthesis of Empirical Literature I synthesized the literature by reviewing the methods, settings, participants, instruments, findings, and conclusions, of articles in the literature pool. Methods and quality of the empirical literature. From the 31 empirical articles, I selected during my literature review search process, I list the methodological tradition, method description, and research quality of a sample of the articles in Table 2. Methodological tradition is the broad categorization of quantitative methods, qualitative methods, or mixed methods. I determined the methodological tradition of each article based on either the authors’ direct statement within the methods section or from implications that I discerned from the method description. Method description is what I gathered from each article’s methods section and relates to the specific research study design. For example, Elliott’s (2016) study of self-efficacy on student retention used a quantitative methodological tradition and a logistic regression, and David, Ball, Davies, & Reay’s (2003) study of gender and parental involvement in 24 the college search process used a qualitative methodological tradition and ethnography as their methods. Table 2 Research Methodology of a Sample of Articles within the Literature Review Method Research Citation Author(s) Method Description Tradition Quality Quantitative Logistic regression, High 1 Chang, L. data-mining quality Christiansen, D. L., Quantitative One-way ANOVA Low Davidson, C. J., Roper, quality 2 C. D., Sprinkles, M. C., & Thomas, J. C. David, M. E., Ball, S. J., Qualitative Ethnography Moderate 3 Davies, J., & Reay, D. quality Quantitative Unstated Low 4 Dawes, P. and Brown, J. quality Quantitative Conjoint analysis Moderate 5 Dillon, E, and Smith J. (market research) quality Dunnett, A., Moorhouse, Mixed Multivariate, High 6 J., Wash, C., et al Methods interviews quality Qualitative Case study; focus Low 7 Smith, M. groups quality Quantitative Nested logistic High 8 Elliott, D. regression quality Fletcher, J. and Tienda, Variables-fixed- Moderate 9 Quantitative M. effects estimation quality Gonzalez, J. and Artificial neural Moderate 10 Quantitative DesJardins, S. network quality 25 Note. Ratings of research quality was assessed based on the depth and description provided in the articles’ methods sections and how the authors addressed threats to internal validity. I rated research quality on a three-level summative index scale of high, moderate, or low quality. I rated each study’s research quality based on the depth and description the author provided of the methods they used and whether the author’s chosen research design was appropriate for the research question and variables. In assessing the overall research quality of a study, I analyzed how each article addressed or failed to address six common threats to internal validity displayed in Table 3 (Shadish, Cook, and Campbell, 2002). For example, Chang’s (2006) study of data mining on college admissions addressed several threats to internal validity and was therefore rated high quality. Dillon and Smith’s (2017) study of market economic factors on students’ college choice utilized market research without addressing all or most threats to internal validity. I therefore rated their study as moderate quality. Smith’s (2012) study of parents’ perceptions of the college choice process addressed no threats to validity and was rated low quality. Table 3 Shadish, Cook, and Campbell (2002) Threats to Internal Validity Internal Validity Threat Definition Ambiguous temporal Lack of clarity about which variable occurred first may precedence yield confusion about which variable is the cause and which is the effect Selection Systematic differences over conditions in respondent characteristics that could also cause the observed effect History Events occurring concurrently with treatment could cause the observed effect Maturation Naturally occurring changes over time could be confused with a treatment effect Attrition Loss of respondents to treatment or to measurement 26 can produce artifactual effects if that loss systematically correlated with conditions. Instrumentation The nature of a measure may change over time or conditions in a way that could be confused with a treatment effect Settings of the empirical literature. Studies within the literature pool comprised a wide range of settings as shown in Table 4. Timing of the study played an influential role in determining setting because researchers study the topic of college choice before and after students make their choice. Settings within the literature pool therefore include high schools and universities. For example, Smith (2012) conducted a case study of one inner-city high school in Los Angeles, California, and Elliott (2016) studied sophomore university students across 14 states. Studying students’ choice of college before the decision may yield quite different inferences than studying students’ choice post hoc when confirmation bias by affect students’ account of their college choice. Table 4 Settings of a Sample of Articles within the Literature Review Citation Author(s) Setting 1 university 1 Chang, L. *insufficient description Christiansen, D. L., Davidson, C. J., Roper, 1 research university in the 2 C. D., Sprinkles, M. C., & Thomas, J. C. Midwest David, M. E., Ball, S. J., Davies, J., & Reay, 1 high school in the United 3 D. Kingdom 4 Dawes, P. and Brown, J. United Kingdom 5 Dillon, E, and Smith J. *insufficient description 6 Dunnett, A., Moorhouse, J., Wash, C., et al United Kingdom 27 1 inner-city public high 7 Smith, M. school in Los Angeles 8 Elliott, D. 14 states in USA 1 research I university in 9 Fletcher, J. and Tienda, M. Iowa 1 research I university in 10 Gonzalez, J. and DesJardins, S. Texas Note. Studies with no or little description of the setting of the study are indicated with “insufficient description.” Of the 31 studies within the pool, three studies were conducted in the United Kingdom. The articles set within the United Kingdom addressed discrete topics within college choice and therefore met the inclusion criteria. For example, David, Ball, Davies, and Reay’s (2003) study on gender differences in parental involvement during the college choice process revealed transferable findings for the roles mothers and fathers play during their student’s college choice process. Participants of the empirical literature. The literature pool comprised of studies with a wide range of participants as shown in Table 5. As with setting, timing of the study played an influential role in determining participants because researchers study the topic of college choice before and after students make their choice. Participants in studies within the literature pool therefore include current high school students, parents of high school students, and current university students reflecting on their college choice process. For example, Dawes and Brown (2002) studied a convenience sample of 266 freshman college students 28 in a single introductory business course who were asked about their past college choice process, and Christiansen, et al (2003) studied 406 high school juniors and seniors who attended a college visit day as prospective students. Table 5 Participants of a Sample of Articles within the Literature Review Surveyed Before Citation Author(s) Participants or After Decision 26,611 prospective 1 Chang, L. Before college students Christiansen, D. L., Davidson, 185 HS juniors and 221 2 C. J., Roper, C. D., Sprinkles, Before HS seniors M. C., & Thomas, J. C. 120 students, 17–20 years David, M. E., Ball, S. J., 3 old, across six After Davies, J., & Reay, D. universities 266 freshman students in 4 Dawes, P. and Brown, J. a single intro Business After course 5 Dillon, E, and Smith J. 2,406 university students After Dunnett, A., Moorhouse, J., 6 400 university students After Wash, C., et al 8 mothers of students 7 Smith, M. Before from a single HS school 8 Elliott, D. 2,358 freshman students After Colleges and universities across the United States annually submit data to the federal government on their enrollment, which provides the opportunity for very large 29 datasets. Some of the studies within the literature pool reflect studies of very large sample sizes, such as Fletcher and Tienda (2009) who studied 66,654 files of prospective students from a single university. Studies in the literature pool also included small sample sizes, such as the focus group conducted by Smith (2012) with a sample of eight mothers of high school students in the college search process. Instruments of the empirical literature. Studies in the literature pool, in most cases, collected and analyzed extant data from individual institutional datasets or nationally available datasets. For instance, in their study of student sorting by academic ability, Dillon and Smith (2017) analyzed data from several extant data sources, including the Integrated Postsecondary Education Data System (IPEDS), a national dataset provided through the National Center for Education Statistics (2018) and managed by the U.S. Department of Education. Such a dataset is publicly available, and colleges and universities are obligated to submit annual updates on a series of metrics. Other studies involved collecting data using program evaluation surveys, such as the campus tour surveys administered by Christiansen, et al (2003). No detail on instrumentation was provided for that specific campus tour survey, and such was the case for many studies in the literature pool. A smaller number of studies in the pool utilized unique instruments intended to measure more complex latent constructs. One example of the use of unique measures was Elliott’s (2016) study of self- efficacy on student persistence. Elliott utilized the Freshman Survey from the Cooperative Institutional Research Program (CIRP), which has a history of over 15 million respondents across 1,900 institutions since 1966 (Higher Education Research Institute, 2011). The Freshman Survey provides a technical report alongside the 30 instrument that includes exploratory factor analysis, parameter estimation, and scoring detail using item response theory to measure the latent traits. Such instrument validation was rare in the literature pool, suggesting a gap in the literature that could be filled by the validation of some other existing instrument or the creation of a new robust instrument that could measure latent constructs of college belonging and decision-making. Findings of the empirical literature. The articles within the literature pool found multiple associations among students’ choices of university. For example, Dunnett, Moorhouse, and Wash (2012) suggest that college reputation was by far the strongest association to a student’s college decision, whereas David, et al, (2003) reported student identities, such as gender, social class, and ethnicity, played a critical role in their decision. Some authors noted discrepancies in the college choice among students of different socio-economic backgrounds. For example, students from lower socio-economic backgrounds are academically capable to succeed at selective colleges, but many do not attend those colleges according to Carnevale and Van Der Werf (2017) who identified that about, “86,000 students receiving Pell Grants who scored 1120 or higher on the SAT [above the median] are not attending selective colleges” (p. 13). They also discovered that students who are recipients of Pell Grants are more likely to attend open-access colleges that average far lower graduation rates (49%) than selective colleges (82%). Graduation rates of Pell Grant recipients follow the average graduation trend at those two types of institutions with only 48 percent graduating from open-access colleges, and about 78 percent of Pell Grant recipients graduate from selective colleges and universities (Carnevale & Van Der Werf, 2017, p. 9). The type of university would seem to have an 31 effect on the ultimate graduation rates of its students that heightens the necessity to eliminate any barrier in the decision-making process during the admissions process. Inconsistent with the aforementioned findings, Elliott (2016) reported academic and social self-efficacy are the most influential factors on student persistence in college. Elliott’s study is unique because it was conducted with a high quality of research methodology relative to many studies in the literature pool by utilizing a nested logistic regression, and yet Elliott’s focus was not precisely on college choice but on college persistence from freshman-to-sophomore year. The lack of high-quality research methodology on college choice presents a gap in the literature that I intend to help fill with my study. The use of logistic regression in Elliott’s study presents an opportunity to test the replication of results on her hypothesis with a slightly different student sample, admitted students instead of continuing students. College ranking publications. Publications such as U.S. News & World Report’s Annual Guide to America’s Best Colleges serves as a source for many prospective college students (Griffith & Rask, 2007). Ample literature exists regarding the influence of ranking publications on students’ choice of college, and such literature presents mixed results. For example, Griffith and Rask (2007) noted that sensitivity to rank diminished as students considered lower ranked colleges, whereas sensitivity to rank was highest among students who considered higer ranked colleges. Contrary to the findings of Griffith and Rask, Soo (2013) detected no statistical significance of college ranking publications on the ultimate enrollment choice of students, however rankings did have a meaningful effect on the perceptions of high school teachers, who, as Chapman (1981) observed, serve as significant persons in the college choice process of students (Tillery, 1973). 32 Although the literature on college choice includes ample research on the effect of publications ranking colleges, I found no related research on the effect of students ranking of colleges within their choice set, which presents one important gap in the literature among others. Gaps in the empirical literature. From a thorough review of the pool of articles in my literature search, blank spots were revealed that I intend to address with my proposed study. Paramount among the gaps in the literature is the absence of investigation of students’ rank choice as well as multiple orientation attendance, which I investigate as independent variables. Ranking behavior, as previously indicated in the theoretical framework, plays an integral role in decision-making (Kahneman & Tversky, 1979). Consequently, how students rank their choice set of colleges may prove worthy of exploration empirically. Although considerable research has been conducted on the relationship between college rankings publications such as U.S. News & World Report and college selection, I found no empirical research on the relationship between students’ ranking of their college options and their ultimate decision-making in my search. Another notable gap in the empirical literature was the absence of research on new student orientation attendance. As described previously Chapman’s Theory of College Choice (1981) does not include certain student behaviors which may influence decision-making, such as taking a campus tour or attending new student orientation. Similarly, I discovered no existing, empirical research on such behavior as the relationship between students attending multiple universities’ orientation programs and their decision-making. These gaps reveal an opportunity to investigate such phenomena 33 to see if any variance may be explained by student ranking of their college options and by students attending multiple universities’ orientation programs. Although student ranking of their college options and multiple orientation sessions serves as my main predictor variables, another gap in the literature suggests the need for further exploration. I intend to include the variables of waitlist status and first- generation status as predictor and moderator variables in my models. I found no existing literature directly investigating variance in college selection that can be explained by either of these two variables, yet the literature does include other critical identity-based factors. In terms of methods, much of the research in the literature about college choice include qualitative methods, such as David, et al’s (2003) study of the association of gender and the decision process. Such studies provide interesting examples of the nuances in students’ decision-making process, yet few studies provided large sample sizes and generalizable results. I intend to help fill a gap in the literature by providing quantitative methodology using logistic regression, to create a predictive model of college choice by addressing nuances in the complexity of the decision-making process yet with a large, generalizable sample size. Conclusions of the empirical literature. The articles reviewed in the literature pool seemed to conclude the college choice process is a complex, multi-stage process for high school students, one with multiple latent and manifest factors. No seminal study provides a comprehensive, contemporary model for college choice. The body of literature on college choice spans multiple academic disciplines, including sociology, economics, psychology, and marketing communications. Among those disciplines, the body of 34 literature includes multiple research methodologies, including quantitative, qualitative, mixed-methods, and newer methodologies, such as neural network analysis. With a broad and disparate body of literature, I intend to add to the body of research on college choice with quantitative methods, testing the relationship between ranking choice sets, orientation attendance, and college choice. Research Questions RQ1: Does the rank of incoming students’ choice of university relate to their enrollment decision? RQ2: Does the number of universities’ freshman orientations that students attend affect their enrollment decision? RQ3: Does the rank of incoming students’ choice of university relate to the number of universties’ freshman orientations that students attend? RQ4: Do demographic variables (listed below) affect incoming students’ decision to enroll at the University of Oregon? • IV: Cohort • IV: High school GPA • IV: SAT/ACT composite score • IV: Family Income • IV: Residency • IV: Proximity to home • IV: Gender • IV: First-gen status • IV: Waitlist status RQ5: Do demographic variables (listed above) moderate the effect of rank on students’ decision to enroll at the University of Oregon? RQ6: Do demographic variables (listed above) moderate the effect of orientations attended on students’ decision to enroll at the University of Oregon? 35 Path diagrams for RQ1–RQ6 are displayed in Figures 5–7. Figure 5. Path diagram of models for research questions 1, 2, and 3. 36 Figure 6. Path diagram of models for research question 4. 37 Figure 7. Path diagram of models for research questions 5 and 6. 38 CHAPTER III METHODS I next describe the methodology I used to answer my research questions. I describe the research design chosen and its justification as well as the variables, instrument, participants, setting, and procedure. Research Design I conducted a non-experimental research design using logistic regression of secondary data on college choice. Logistic regression is an appropriate statistical analysis because it allows for the analysis of issues with binary outcomes and multiple dichotomous and/or continuous predictor variables (Huang & Moon, 2013). Students make a binary decision about attending a university or not attending, thus the enrollment decision makes a fitting dependent variable (DV) to explore with logistic regression. Ordinary least squares (OLS) regression was at one time an acceptable method for testing a binary outcome, yet most researchers today acknowledge OLS has limitations measuring binary outcomes due to affecting model parameter estimates and standard errors (Long, 1997; Peng, Lee, & Ingersoll, n.d.). OLS regression may also produce results less than zero and more than one, which would not make sense in the context of a student’s decision to attend college. Logistic regression allows researchers to predict probabilities based on the maximum likelihood of an outcome given a set of characteristics. For understanding student decision-making about college enrollment, predicting the likelihood of enrollment based on a series of dichotomous and continuous variables presents a practical and advantageous opportunity to address the enrollment unpredictability that I presented in 39 the problems of practice and the review of literature. Logistic regression also enables interpreting results based on odds and odds ratio, which is a familiar format for practitioners to understand otherwise complex analysis. For example, an enrollment professional not steeped in statistical analysis can interpret when a specific composition of student characteristics suggests an 80% greater likelihood that the student will attend the university. Student Sample Students enroll at the University of Oregon from a variety of circumstances and conditions, including via transferring from another institution, from another country, or after taking time off from school and beginning university study in the winter or spring terms. For the purposes of this manuscript, I used the term freshman to refer to first-time, full-time, domestic, undergraduate students enrolling for fall term only. This definition represents a common definition of incoming cohorts of new college students shared among higher education institutions (M. E. Clinedinst & Patel, 2018; “Undergraduate Retention and Graduation Rates,” 2018). I included UO freshman students only in the sample. Students at the UO come from all 50 states and over 100 countries, and 34% of undergraduate students identify as a domestic ethnic or racial minority (“About the UO,” 2018). Approximately 29% of freshman identify as first-generation students, and approximately 37% of freshmen qualify for federal Pell Grants due to financial aid status. Because I analyzed data post-hoc, with a sample size of 9,266 participants, achieving adequate power from the sample did not cause an issue when interpreting results. However, of the students sampled, only approximately 4% chose not to enroll 40 after attending their orientation session, which created a visible imbalance in the S-curve once data were plotted and established a possible limitation to the study. Setting The University of Oregon is a public, tier-one research institution in the Pacific Northwest. The UO enrolls slightly more than 20,000 undergraduate students, with over 4,500 new freshman students and over 1,000 new transfer students annually (“About the UO,” 2018). One of the primary events designed to support freshmen through their college transition process is new student orientation. Nearly all incoming freshman students attend an orientation session as one of their required tasks prior to matriculation. The UO offers freshman students 14 orientation options throughout the months of June, July, August, and September. Of the 14 session options, four are offered off-campus for out-of- state students whose hometown is a great distance from campus. One of the 14 orientation options takes place in September immediately prior to the start of fall term. In this manuscript I include only data from the survey administered to students who attended the 10 on-campus summer orientation sessions. Programming for the off- campus IntroDUCKtion sessions has changed year-to-year, so including the survey administration at the off-campus sessions may have posed a validity risk. Excluding off- campus IntroDUCKtion participants removes approximately 150 students from the retained sample of 9,266 participants. Timing. The UO Freshman Survey, described below, was administered on the second day of students’ two-day orientation session. Students completed the survey immediately following their academic advising appointment and immediately prior to 41 course registration. Students took the survey at a computer lab in the UO library, the same site as where they registered for classes. As students waited for an available computer to use for course registration, staff requested their participation in the UO Freshman Survey. The purpose of timing the survey near the timing of course registration was to encourage a high response rate, because nearly all students at orientation visited the computer lab to register for classes. The survey took approximately 7–10 minutes for students to complete. Instruments I used data collected from three instruments, the UO application for admission, the CIRP Freshman Survey, and the UO Freshman Survey. I next describe each instrument. Table 6 displays the variables and the instruments from which data were collected by year. Each instrument links participants with the keyed variable of their student ID number, and data were collected based on the variables listed previously in RQ3 and described in detail below. 42 Table 6 Variables and the Instruments from Which They Came by Year Variable Name 2014 2015 2016 2017 2018 Rank choice CIRP CIRP CIRP UOFS UOFS Orientations attended CIRP CIRP CIRP UOFS UOFS Family income CIRP CIRP CIRP UOFS UOFS High School GPA App App App App App Proximity to home App* App* App* App* App* (residency) Gender App* App* App* App* App* SAT/ACT composite score App App App App App Waitlist status App* App* App* App* App* First-generation status App* App* App* App* App* Note. CIRP = CIRP Freshman Survey; UOFS = UO Freshman Survey; App = the application for admission. * = data are dichotomous. Application for admission. The University of Oregon annually collects data from students when they apply for admission to the university. The application for admission collects demographic data and previous academic performance. Students provide information in detail about their characteristics and circumstances as applicants, including demographic details. Error! Reference source not found. displays a sample of the UO application for admission. I utilized data from the application for admission to explore RQ4, RQ5, and RQ6 by including the variables of high school GPA, SAT/ACT composite score, first-generation status, gender, residency, waitlist status, and family income. Each of the aforementioned variables are collected from the application for 43 admission except the waitlist status, which is determined as an outcome of the application for admission. Following students’ application for admission, Admissions officers determine the students’ application statuses as admitted, denied, or waitlisted. Staff in the UO Office of Admissions then enter data on the determination of students’ application to the Banner data system, the data repository where all student demographic characteristics are stored at the University of Oregon. I will later describe in detail the procedure for how data were delievered. Application format. Students complete the application for admission via one of three formats, via the Common Application, the online UO proprietary application, or the paper UO proprietary application. As described previously in the introduction, the Common Application provides students the option to complete a single application for admission that is then submitted to up to 20 universities (Rickard, 2017). The UO application asks the same questions yet the look and feel of the application is slightly different from the Common Application. The UO paper application mirrors the questions asked by both online versions and is available for the students who do not have access to submit their application online. Although the instrument format is different, the seven variables I included from the application for admission are universal items without between-format variation. For example, a student’s high school GPA remains the same whether they submit it via the Common Application or the UO’s proprietary application. All formats of the application for admission require students agree to the truthfulness of their responses under penalty of admission denial should students submit falsehoods. Confidence in the fidelity of the data should therefore remain high despite the multiple formats of the application. 44 For RQ4, I use the variables taken from the application for admission as the predictor variables. For RQ5 and RQ6, I use the variables taken from the application for admission as the moderator variables. CIRP freshman survey. The CIRP Freshman Survey is administered nationally via the Higher Education Research Institute (HERI) at the University of California, Los Angeles. The CIRP Freshman Survey has been administered at colleges and universities across the United States since 1971 (Higher Education Research Institute, 2011). Over 1,900 colleges and universities have participated in the CIRP Freshman Survey, and HERI tabulate, report, and share results with participating institutions (“CIRP Freshman Survey,” 2018). The UO participated in the CIRP Freshman Survey program for three years, 2014–2016. I analyzed data for the predictor variables of rank choice of the university in RQ1 and RQ5 taken from the CIRP Freshman Survey, and I analyze data for the predictor variables of orientations attended in RQ2 and RQ6 taken from the CIRP Freshman Survey. The number of orientations students attended was collected as a supplemental, custom question to the full CIRP Freshman Survey, as that item was not a part of the original item bank provided by HERI. The CIRP Freshman Survey was administered as a paper survey along with a page describing informed consent in compliance with requirements of the UO Institutional Research Board. Error! Reference source not found. displays the CIRP Freshman Survey instrument, and questions 15 and 59 display the items of interest. CIRP freshman survey reliability and validity. The Higher Education Research Institute provides reliability testing of items because it has been administered annually 45 since 1971 (Higher Education Research Institute, 2011). By administering repeatedly HERI mitigates random fluctuations of observations, and nearly 90 percent of participating institutions in the CIRP Freshman Survey are repeat participants (Stolzenbert, E. B., personal communication, November 14, 2018). To examine the validity of the CIRP Freshman Survey in measuring certain constructs, previous literature found CIRP factors held together via measurement of Cronbach’s Alpha with most coefficients in excess of .70 (Astin, 1993; Luo & Jamieson-Drake, 2005). UO freshman survey. To understand characteristics of the entering class through the new student orientation and transition process, the University of Oregon administers a survey to students who enroll at the UO each year. The CIRP Freshman Survey surved as the tool to collect data on those student characteristics in 2014–2016 at which time the Office of Enrollment Research switched to the UO Freshman Survey for 2017 and 2018. The UO Freshman Survey was developed in collaboration among the Office of Student Orientation Programs and the Office of Enrollment Research (OER) in the Division of Student Services and Enrollment Management. Staff from the two departments modeled the UOFS after the CIRP Freshman Survey by omitting and revising survey items. Two survey items were not revised from the CIRP Freshman Survey. The rank choice and orientations attended survey items remained unchanged from the CIRP format and wording. Error! Reference source not found. displays the UOFS, and questions 1 and 2 display the survey items of interest. The UO Freshman Survey (UOFS) aids UO enrollment professionals toward a critical goal—understanding students’ commitment to the university. For example, the survey item that collects student’s rank choice of the university illustrates that the student 46 sees the UO as their top choice or less than top choice for college. As referenced in my theoretical framework, the Theory of Student Departure illustrates how students’ institutional commitment fits in their decision-making for staying or leaving a college (Tinto, 1988). Figure 2 illustrates Tinto’s theory. The UOFS measures the construct of institutional commitment. Similar to the CIRP Freshman Survey, data include key items related to college choice, such as rank choice and orientations attended. The UO Freshman Survey was administered as a paper survey along with a page describing informed consent in compliance with requirements of the UO Institutional Research Board. I worked with staff in the Office of Enrollment Research as we consulted each survey item. Due to many demanding activities during the program schedule of orientation and due to the timing of the survey in the program schedule, I was particularly sensitive to the risk of survey fatigue, so I scrutinized each survey item to limit the duration of the survey. Once OER staff and I agreed on the survey items and the predicted survey duration, staff in OER submitted the instrument for review to the UO Institutional Research Board and obtained approval for use on June 20, 2018. The instrument includes nine numbered items, with a total of 31 sub-scale items as indicated in Error! Reference source not found.. I used only data from items 1 and 2 in the analysis. These items from the UOFS mirror items 15 and 59 on the CIRP Freshman Survey. Procedure I next describe the procedure by which data were collected and delivered to me for use in this manuscript. I describe the data delivery procedures for each instrument— the application for admission, the UO Freshman Survey, and CIRP Survey. Data were 47 collected from each instrument and matched to the student-level by the keyed variable of the student ID number. Application for admission procedure. Students submit their application for admission to the University of Oregon online via the Common Application (Rickard, 2017) or the UO’s proprietary application, as shown in Appendix 1. The UO Office of Admission processes data from each student’s application for admission by reviewing the responses on each application and storing the data in Banner, the UO’s student information system (“Banner Guide: Display a Student’s Admission Records,” n.d.). Staff in the Office of Enrollment Research hold clearance to access student-level admission data via the UO Student Data Warehouse, which syncs with Banner data in order to provide ad-hoc queries of student data (“Office of the Registrar: Faculty & Staff,” 2019). Further sharing of data requires a data-share agreement, reviewed and approved by the Office of the Registrar, the Office of Admissions, and the Office of General Counsel. The data-share agreement for this manuscript can be reviewed in Appendix 4. Staff in the Office of Enrollment Research downloaded the data from the UO Student Data Warehouse as a .csv file and uploaded it as an Excel file via Microsoft OneDrive. UO freshman survey procedure. Students completed the UO Freshman Survey in-person, via paper survey, while attending new student orientation. The procedure for administration of the CIRP Freshman Survey mirrored that of the UO Freshman Survey. As students prepared to register for their fall term courses, staff in the Office of Student Orientation Programs presented the students with the survey and briefly described its purpose and informed consent parameters. I next describe the training I provided to staff who directly administered the survey. 48 Survey administration training. I trained 25–28 staff members (depending on the cohort year) in the Office of Student Orientation Procedures during their staff training three days prior to the first orientation session in which the survey was administered. I trained each staff annually. During the training, I presented a sample of the survey instrument for their review as well as the informed consent cover sheet. I made clear during the training that survey completion should be presented to participants as a voluntary activity and that participation should not be coerced or compulsory. I instructed the staff that pencils would be provided to participants yet blue or black ink was permissible for use on the paper survey instrument. I displayed a sample key-locked black metal box and instructed the staff that all survey participants should deposit their completed survey into the slot in the box once complete. I then gave instructions for the staff on where to physically deliver the locked metal box upon the conclusion of each orientation session. Students submitted their completed UO Freshman Surveys by placing them in the key-locked metal box located in the library where survey administration occured. At the conclusion of the orientation session, staff carried the metal box to staff in the Office of Enrollment Research who had the key to unlock the box and retrieve the surveys. Staff in OER then scanned the paper surveys to .pdf format. Survey responses were automatically coded using Remark Office OMR software (“Remark Office OMR,” 2019), manually reviewed for accuracy by OER staff, and exported to a Microsoft Excel spreadsheet file and shared via secure link with me. Student ID and name were used to combine survey responses with selected student data. Electronic data files were subsequently stored on 49 the Enrollment Management server, the same server used by the Registrar’s Office for securely storing all student records. The director of Enrollment Research then de-identified the data and assigned a randomized identification number which is not linked to the students’ original student identification number. He then transferred data to me via a secure online link through Microsoft OneDrive. I kept all data in a password-protected file behind a secondary password vault on a university-owned laptop, which required a username and password or fingerprint verification to login. I kept the laptop in a locked office which is behind two sets of locked doors when outside of regular business hours. In accordance with the data share agreement, as outlined in APPENDIX , I will return all data within 30 days of the completion of my dissertation, and I will retain no copies. As all data were collected based on administrative rules and standardization with the application for admission and the UO Freshman Survey, fidelity of the data in this study should be considered reliable. I next describe the model specification and each variable in the models. Model Specification All factors on students’ decisions to enroll or retain at a university may not be observable and therefore measurable. However, the literature is replete with suggested influences measured as predictors for college decision-making. I chose a paramorphic model structure that allows an analysis of input variables and their relationship to the outcome in predicting a decision. Paramorphic models focus on the output, i.e., which action a person will choose, not how they chose the action (Skořepa, 2011). I chose to focus on a paramorphic model structure to limit the scope of the complexity preceding 50 college decision-making. I do not investigate the process by which students navigate toward their college decision but the decision itself and which variables relate to the decision. Figures 5–7 display the path diagrams for the models of each research question. Variables. I next describe the variables grouped into predictors, moderators, and outcomes. The control variables represent the independent variables of interest, and the moderator variables represent unchanging variables that are primarily demographic in nature. The outcome variables represent the dependent variable of interest. Table 7 displays the variable names and groups by variable type. I next describe each variable in detail. Table 7 Variable Names by Type and Their Role in the Analysis Variable Type Variable Name Predictor Moderator Outcome Rank choice Continuous Orientations attended Continuous Family income Nominal High School GPA Continuous Residency Nominal Proximity to home Continuous Gender Nominal SAT/ACT composite score Continuous Waitlist status Nominal First-generation status Nominal 51 Decision to attend the UO Nominal Predictor variables. The independent variables of interest were rank choice and number of orientations attended. The rank students assign to universities in their choice set may help explain the variance in their ultimate decision to attend a university. As stated in the theoretical framework section regarding Prospect Theory, ranking behavior plays an important role in general decision-making (Kahneman & Tversky, 1979). If a student ranks the University of Oregon as their top choice or as their third choice, does that rank predict ultimate attendance. I represented rank choice on a four-point ordinal scale where 1 = the UO is the student’s first choice college to attend, 2 = the UO is the student’s second choice college to attend, 3 = the UO is the student’s third choice college to attend, and 4 = the UO is the student’s less-than-third choice college to attend. The number of universities’ orientations a student attends may also help explain additional variance in their ultimate decision. As noted in the theoretical framework, Tinto (1987) theorized that institutional commitment determines whether a student chooses to stay or withdraw from a specific college, and attending a college’s new student orientation and registering for classes plays a signifier of that commitment. If a student attends two or three universities’ orientations, they have also placed deposits to secure their spot in the freshman class, travelled to the campuses, and enrolled in classes at multiple universities. Does such behavior help to explain any variance in the enrollment for freshmen at the University of Oregon? As previously described, the Model of College Choice (Chapman, 1981) lists multiple factors though it does not account for student behavior such as orientation attendance, which may help predict enrollment. I represent the number of university orientations attended on a five-point scale where 1 = a 52 single orientation attended (the UO’s), 2 = two orientations attended, 3 = three orientations attended, 4 = four orientations attended, and 5 = five or more orientations attended. Moderator variables. Student characteristics such as demographic variables may provide important understanding of group differences. I analyzed if any of the moderator variables explain a statistically significant portion of variance in student decision-making. Moderator variables include cohort, family income, high school grade-point average, standardized test scores, residency, gender, first-generation college student condition, and waitlist status. Chapman (1981) includes the first five moderator variables in the Model of College Choice, so I include them as well. Waitlist status and first-generation status are not included in the Model of College Choice, though are worth exploring further as predictors of college choice. I next describe each moderator variable. Cohort. Data included five cohorts of students, each cohort defined by an incoming class of first-year college students. By analyzing data by cohort I accounted for possible variation due to systematic differences related to the year of admission. For example, if the university altered its programming of new student orientation during one specific year by adding sessions and thereby increasing the options students and their families have to visit campus, such altered orientation programming may explain a difference in student behavior regarding one of the independent variables. Cohort was represented as an ordinal variable, where 2014 = the year the student was admitted. Cohorts include 2014, 2015, 2016, 2017, and 2018. Family income. I represented the students’ socio-economic status (SES) by the estimated annual family income, which was self-reported by the students on both the 53 CIRP Freshman Survey and the UO Freshman Survey. Students provided an estimate of their family income via a 14-point ratio scale with “Less than $10,000” listed as the lowest response choice, and “$250,000 or more” listed as the highest response choice. The 14-item response options between the lowest and highest varied incrementally. For example, one item response choice was “$10,000–$14,999” while a higher choice was “$150,000–$199,999.” The lower item response options offered students a $5,000 scale of response choices, while the higher response options scaled by $50,000. To produce a scale with equal differences, I collapsed the 14 categories to six with each category representing an equal difference of $50,000. I collapsed the categories by creating a new variable in SPSS and recoding response options, and Table 8 displays the original income options and how I recoded. Collapsing 14 categories to fewer provides convenient groupings for a frequency distribution (Scalise, 2016). 54 Table 8 Family Income Response Choice Recoding Original Family Income Item Responses Recoded Family Income Responses Less than $10,000 Less than $50,000 $10,000–14,999 Less than $50,000 $15,000–19,999 Less than $50,000 $20,000–24,999 Less than $50,000 $25,000–29,999 Less than $50,000 $30,000–39,999 Less than $50,000 $40,000–49,999 Less than $50,000 $50,000–59,999 $50,000–99,999 $60,000–74,999 $50,000–99,999 $75,000–99,999 $50,000–99,999 $100,000–149,999 $100,000–149,999 $150,000–199,999 $150,000–199,999 $200,000–249,999 $200,000–249,999 $250,000 or more $250,000 or more By collapsing item response groups into fewer groupings, I no doubt forfeited a degree of precision that may provide clearer explanation of variance. Collapsing the choice categories, however, provided parsimony of the data and ability to create equally scaled response options. Any aggregate of a variable risks high multicollinearity so results should be interpreted cautiously. High school grade-point average. College admissions professionals often utilize a measure of students’ academic performance in high school as admission criteria. Many 55 factors affect a student’s high school GPA, and ample literature exists on the topic as a valid (or not-so valid) measure of student’s actual ability. For example, Smith, et al. (2013) suggets high school GPA as a strong predictor of college access though only at selective universities (p. 251). However, due to the ubiquity, however, of high school GPA in admissions criteria throughout the United States, I included it in the model. High school GPA is a continuous variable converted to a range from 0.0 to 4.0. Test scores. Like high school grade-point average, standardized test scores play a common role in the admissions criteria of many universities in the United States. Those standardized scores are generally from the SAT or ACT. The University of Oregon Admissions Office creates a composite score to account for scoring differences between the SAT and ACT, in the event students take one or the other or both standardized tests. I represented the SAT/ACT composite score as a continuous variable. Residency. The proximity of students’ home to the university may explain variance in decision-making for many possible reasons. For example, out-of-state status determines a substantial increase in the cost of tuition as well as more complex travel arrangements to get to and from campus. In-state students pay substantially less in tuition and likely grew up with imagery of the University of Oregon present in their schools via teacher influence or admissions counselor visits. Some students want to attend a university in their home state, while others want to venture out farther from home despite non-resident tuition costs. I represent residency as a dichotomous variable where 1 = Oregon resident and 0 = out-of-state student, otherwise known as a non-resident. To address further detail on the influence of proximity, I also included a continuous variable, 56 proximity, to explore if any variance could be explained by the distance students travel from home to attend the University of Oregon. Gender. Decision-making for college by gender may explain variance in the decision-making through an important social construct. For example, one study of the role gender plays in college decision-making revealed a critical difference in the way parents supported or guided their students based on the student’s gender (David et al., 2003). Such a difference suggests including gender in the model could explain a portion of variance. I represented gender in the way the data are represented via the application for admission on the Common App, which are dichotomous in nature. A dichotomous representation of the social construct of gender would be more aptly defined as sex, not gender, however I chose not to analyze data through a physiological characteristic. Doing so would be tantamount to attempting to understand college decision-making by students’ height or eye color. No literature suggests that generally students’ decision-making for college is influenced by physiological characteristics. However, the physiological designation of sex is closely tied to the social construct of gender, which may indeed explain some variance in student decision-making. Ideally, the Common App would collect students’ gender identity instead of students’ sex, which would allow for more accurate analysis of gender as a moderator. As an important social construct, I understand a person may identify beyond the binary constraints of man or woman. However, for parsimony in this manuscript I adhere to the category options presented by the instrument from which the data were collected, which are as a dichotomous variable where 1 = female and 0 = male. 57 First-generation status. Students who are the first in their families to attend college may experience the college choice process differently than those students whose family members did attend college. As an example, one study revealed that first- generation college students experience higher levels of “achievement guilt” that likely affects their student experience unlike their continuing-generation peers (Covarrubias & Fryberg, 2015). To measure if and how first-generation status explains variance in student enrollment decisions, I investigated if first-generation status affects or moderates an effect on enrollment decisions. First-generation status is a dichotomous variable, where 0 = the student is continuing-generation, (i.e., the student’s family member(s) attended college) and 1 = the student is a first-generation college student. Waitlist status. After a student applies for admission to the University of Oregon, the UO notifies the student they have been admitted, denied, or added to the waitlist. After time on the waitlist, some students eventually receive admission. Students’ status as formerly waitlisted may explain some of the variance in their ultimate decision to attend the University of Oregon. I represented the waitlist status as a dichotomous variable, where 0 = the student was directly admitted (not waitlisted) and 1 = the student was admitted after time on the waitlist. Outcome variables. The research questions involve a proximal outcome with the ability to add distal outcomes at a later date for future research, and the proximal outcome presents as a dichotomous variable, whether or not the student enrolled. The students’ decision to attend a particular university defines the proximal outcome for five of the six RQs. I represent the outcome as 0 = student chose not to attend the UO, and 1 = student attended the UO. The national best practice for measuring attendance is to conduct a 58 university census on a predetermined census date during the fall term, which for the University of Oregon takes place on the Friday of the fourth week of classes (“2018– 2019 Survey Materials Frequently Asked Questions,” 2018). Therefore, the proximal dependent variable for five of the six research questions is whether or not the student currently attends the University of Oregon through the Friday of the fourth week of classes during their freshman year. Analysis I analyzed data with descriptive statistics, a test for statistical significance using likelihood ratio test, Spearman’s correlation, and binomial logistic regression. I next describe the analyses chosen. I provide descriptive statistics of the sample in Tables 9–19. Following an analysis of the descriptive statistics, I tested assumptions for the propriety of using logistic regression, including a test for linearity using the Box-Tidwell (1962) procedure. I then tested for statistical significance using likelihood ratio and then analyzed data using binomial logistic regression. I next describe the justification for use of Spearman’s correlation and logistic regression as the appropriate statistical tests. Spearman’s correlation I tested if there’s a relationship between student’s rank order of college choices and the number of orientation attended in RQ3. I used Spearman’s correlation to determine a coefficient, rs, which measures the strength and direction of an association between one continuous and one ordinal variable (Laerd Statistics, 2017). The data must pass three assumptions to use Spearman’s correlation. Assumptions. Spearman’s correlation requires two variables that measure as continuous and/or ordinal in scale. In the data I represented the IV of rank order and the 59 IV of orientations attended as continuous, so the data pass the first assumption. The second assumption is that the two variables represent paired observations. Both observations pair on the same student, so the variables pass the second assumption. The third assumption for use of Spearman’s correlation is the need for a monotonic relationship between the two variables. Based on a visual inspection of the scatterplot, the data passes the third assumption for monotonicity. I next describe logistic regression, the analysis used for RQs one, two, four, five, and six. Logistic regression Binomial logistic regression, also simply known as logistic regression, allows for the analysis of data with a binary outcome variable and multiple predictors that can be dichotomous or continuous. The basic requirements of a logistic regression include seven assumptions (Laerd Statistics, 2017), four of which relate to study design and three of which can be statistically tested. I next describe the seven assumptions for a logistic regression. Assumptions. Logistic regression first assumes one dichotomous dependent variable. Second, logistic regression assumes one or more independent variables measured as continuous or nominal. Table 6 displays a description of variables demonstrating how data pass the first two assumptions necessary for logistic regression. Third, logistic regression assumes independence of observations. The dependent variable represents a mutually exclusive category; students are enrolled on the census date or they are not. Students can not simultaneously be enrolled and not enrolled, therefore the dependent variable passes the assumption of independent obervations. Likewise, the independent variables also maintain mutually exclusive categories. For example, a 60 student may not have ranked the university as both their first and third choice college. Data therefore pass the third assumption for logistic regression. Fourth, logistic regression assumes a minimum of 15 cases per independent variable, and with over 9,000 cases data pass the fourth assumption. Assumptions five, six, and seven are statistically tested, and I next describe each. Fifth, logistic regression includes the need for a linear relationship between any continuous independent variables and the logit transformation of the dependent variable (Laerd Statistics, 2017). I test for linearity using the two step Box-Tidwell (1962) procedure, which first requires a transformation of continuous IVs to their natural log. I computed a new variable in SPSS for each of the three continuous IVs. I next computed the interaction terms between the continuous IVs and their respective logits. I describe results for the test of linearity in the results section. As the sixth assumption for logistic regression, data should not show multicollinearity, and I inspect multicollinearity using correlation coefficients and tolerance/VIF values, which I provide in the results section. Seventh, data should not include meaningful outliers, which I detect using casewise diagnostics and describe in the results section. The general logistic regression model is: where k is the number of independent variables,  is the probability of the outcome of interest (i.e., a student deciding to attend the UO),  is the Y-intercept, and  is the regression coefficient (Huang & Moon, 2013). After running the binary logistic procedure using SPSS (IBM Corp. Released, 2017), I converted the log odds into probabilities to ease interpretation. I converted the log odds to probabilities using the following formula: 61  = (e/(1- e)), where e is the fixed constant raised to the power of the log odds and  is the intercept (Huang & Moon, 2013, p. 198). For RQ5 and RQ6, I conducted independent tests for moderation based on the results from RQ4. Alpha-levels. I used an alpha-level of 0.05, which is standard in education and social science research. However, I used a Bonferroni correction to establish an adjusted alpha based on the number of individual tests I ran with each independent variable for RQ4–6 (m = 11). The Bonferroni correction compensates for the increased chance of incorrectly rejecting the null hypothesis (Type I error) due to many tests within a model. With 11 independent tests, the Bonferroni correction establishes an adjusted alpha of 0.0045 for the models in RQ5 and RQ6. Model evaluation. R2 provides the usefulness of the model by measuring its predictive accuracy (Huang & Moon, 2013). I will measure the proportion of the variance in the dependent variable predicted by the independent variables using Nagelkerke’s pseudo R2, which acts similarly to calculating R2 in linear regression. The Nagelkerke pseudo R2 is: The overall probability will be calculated by dividing the total number of students who enrolled by the number who did not. The overall probability, however, would be a poor predictor of likelihood, so instead I took the log of the likelihood of the data and projected it onto the log of the overall probability. The log of the likelihood of the overall probability is the log-likelihood of the many student enrollment decisions based on their choice rank projected onto the overall probability of enrolling. 62 When the model is a poor fit, the log-likelihood is a relatively large negative value in logistic regression, whereas when the model is a good fit, the log-likelihood is a value close to zero. Maximum likelihood. The process of determining maximum likelihood occurs by repeatedly estimating the size and direction of the logit coefficient until the log likelihood reaches convergence (Huang & Moon, 2013). In other words, the maximum likelihood method finds the highest likelihood of reproducing the data given the parameters. Odds. After running the binary logistic procedure using SPSS (IBM Corp. Released, 2017), I converted the log odds into probabilities to ease interpretation. I converted the log odds to probabilities using the formula,  = (e/(1- e)), where e is the fixed constant raised to the power of the log odds and  is the intercept (Huang & Moon, 2013, p. 198). Odds are ratios of probabilities of an event () to the probability the event does not occur (1 - ), or /(1 - ) (Huang & Moon, 2013). Importantly, odds and probability are not defined interchangeably. Missing data. The nature of missing data is that they “hide true values that are meaningful for analysis” (Little & Rubin, 2002, p. 8). Consequently, addressing missing data reduces the chance of losing important meaning. Because data were taken from multiple instruments over five years, I addressed missing data in the sample to reduce bias. To do so, I first identified and coded nonresponses in the data set to determine if data were missing at random or if data are missing systematically. I compared descriptive statistics to investigate patterns and to determine if missingness occurs completely at random or not. 63 CHAPTER IV RESULTS I next present the results of data analysis with descriptive statistics presented for factors and variables, a test of Spearman’s correlation including tests of assumptions, a test of statistical significance, and logistic regression. Descriptive Statistics Descriptive statistics include general results as well as an analysis of each factor and variable. I then provide descriptive statistics of each moderator variable. Over the course of the five cohort years of data collection, 2014–2018, the University of Oregon enrolled 25,130 first-time, full-time freshman students (Office of Institutional Research, 2019). Of all 25,130 freshman students, 9,266 responded to either the CIRP Freshman Survey or the UO Freshman Survey, depending upon cohort year for an overall response rate of 36.87%. Table 9 displays class size and response rate by cohort year. Table 9 Frequency Distribution of Freshman Class Size, Response Rate, and Instrument Used by Cohort Cohort Year Class Size Response n Response Rate Instrument 2014 5,022 1,960 39.03% CIRP 2015 5,220 958 18.35% CIRP 2016 5,120 2,510 49.02% CIRP 2017 4,834 1,201 24.84% UOFS 2018 4,934 2,637 53.45% UOFS TOTAL 25,130 9,266 36.87% UOFS 64 Note. CIRP = CIRP Freshman Survey; UOFS = UO Freshman Survey. Fewer student surveys were collected during the 2015 orientation sessions due to changes in the collection location mid-summer from the student union building, which at that time was undergoing renovations, to the library. There is no reason to suspect a systematic drop in response by any moderator variable; missing data from the 2015 cohort data should therefore be treated as missing at random. The 2015 cohort experienced a notable dip in response rate (18.35%) by comparison to the other four cohorts (41.59%). Without fully knowing the reason behind this dip in response rate, I suspect it is due to a change in the survey collection site when the site underwent renovation mid-summer. The change in survey collection site included a new way for students to enter the library to register for classes whereby students could enter from doors on multiple sides of the building, only one of which included the survey distribution site. Upon discovering that some students bypassed the survey distribution site, I made operational changes so that all students had the opportunity to receive and complete the survey. Nevertheless, the response rate for the 2015 cohort is notably lower. Similarly, the 2017 cohort experienced a lower response rate (24.84%) by comparison to the other cohorts. I have no rational explanation or guess for why the 2017 cohort submitted fewer surveys than other cohort years. In other words, based on the non- missing data for 2015 and 2017, I have no reason to believe the students in the 2015 and 2017 cohorts are different from the 2014, 2016, or 2018 cohorts, except that they submitted fewer surveys. Upon review of the descriptive statistics by cohort, which Table 10 displays, means and standard deviations for the 2015 and 2017 cohort seem consistent with the means and standard deviations from the other three cohorts, therefore I have no indication of a systematic difference among cohorts based on student characteristics. Consequently, 65 I chose not to exclude the 2015 nor 2017 cohorts’ data, and I analyzed data without deleting missing cases. I next present descriptive statistics by variables, beginning with the outcome variable and independent variables of interest—rank choice and orientations attended. Table 10 Descriptive Statistics for All Variables by Cohort 2014 2015 2016 2017 2018 Variable Name M (SD) M (SD) M (SD) M (SD) M (SD) Enrolled 0.96 (0.2) 0.96 (0.20) 0.97 (0.18) 0.96 (0.21) 0.95 (0.22) Rank choice 1.53 (0.84) 1.56 (0.85) 1.48 (0.75) 1.54 (0.78) 1.42 (0.75) Orientations attended 1.28 (0.70) 2.03 (1.28) 1.20 (0.56) 1.23 (0.57) 1.27 (0.67) Family income 3.01 (1.69) 3.00 (1.67) 2.92 (1.69) 3.60 (1.48) 2.56 (1.35) High school GPA 3.59 (0.33) 3.66 (0.32) 3.59 (0.33) 3.58 (0.37) 3.61 (0.36) 1197.28 1204.14 1193.59 1198.03 1,197.42 SAT/ACT score (136.98) (140.00) (142.68) (131.44) (143.79) Residency 0.52 (0.50) 0.47 (0.50) 0.51 (0.50) 0.48 (0.50) 0.56 (0.50) Proximity to 405.52 506.91 449.70 491.62 394.59 home (519.33) (661.57) (585.81) (626.32) (519.014) Gender 0.59 (0.49) 0.62 (0.49) 0.59 (0.49) 0.58 (0.49) 0.56 (0.50) First-gen status 0.30 (0.46) 0.31 (0.46) 0.31 (0.46) 0.28 (0.45) 0.36 (0.48) Waitlist status 0.03 (0.18) 0.00 (0.06) 0.04 (0.19) 0.01 (0.08) 0.00 (0.07) 66 Enrollment outcome. Of the students who submitted either a CIRP or UO Freshman Survey (n = 9,266), 8,401 arrived on campus for classes in the fall term, and 8,252 remained enrolled on the official census day, which is the Friday of the fourth week of fall term classes. For the purposes of the enrollment outcome, I compared the 8,252 students who submitted a survey and remained enrolled on the census date to the 372 students (4.01%) who submitted a survey but were not enrolled on the census date. Rank choice. From the total sample of 9,266 participants, 9,081 responded to the item of rank choice. The rank choice options ranged from one through four, where one represented the first choice of the student, two represented the second choice of the student, three represented the third choice of the student, and four represented the less- than-third choice of the student (M = 1.49, SD = 0.79). Rank choice was non-normally distributed with skewness of 1.64 (SE = 0.03) and kurtosis of 2.06 (SE = 0.05). After inspecting the distribution for skewness, I transformed the independent variable of rank choice using the base 10 log transformation. For all logistic regressions, I used the log10 transformation of the rank choice independent variable. Among the respondents, 5,941 (65.42%) of respondents ranked the University of Oregon as their top-choice college, while 2,163 (23.82%) of respondents ranked the University of Oregon as their second-choice college. Those ranking the UO as their third choice were 626 (6.89%) of respondents while 347 (3.82%) students stated the UO was less than their third-choice. Four students among the sample marked multiple ranks in their response, and 185 students did not respond to the survey item. Table 11 displays descriptive statistics by rank choice. 67 Table 11 Frequency Distribution of Student’s Rank of UO as Their College Choice Rank of UO in Choice Set N % First choice 5,941 65.42% Second choice 2,163 23.82% Third choice 626 6.89% Less than third choice 347 3.82% Missing 189 0.04% Note. M = 1.49, SD = 0.79. Rank choice was non-normally distributed with skewness of 1.64 (SE = 0.03) and kurtosis of 2.06 (SE = 0.05). Orientations attended. From the total sample of 9,266 participants, 8,288 responded to the item of orientations attended. The item response choices for orientations attended ranged from one through five, where five represented five or more orientations attended (M = 1.31, SD = 0.73). Orientations attended was non-normally distributed with skewness of 2.88 (SE = 0.03) and kurtosis of 8.65 (SE = 0.05). After inspecting the distribution for skewness, I transformed the independent variable of orientations attended using the base 10 log transformation. For all logistic regressions, I used the log10 transformation of the independent variable of orientations attended. Among the respondents, 6,523 (80.37%) stated that the University of Oregon is the only orientation they attended, while 1,060 (13.06%) respondents attended an additional university’s orientation as well as the University of Oregon’s. Those attending three universities’ orientations were 256 (3.15%) respondents, and those attending four orientations were 201 (2.48%) respondents. Seventy-six students (0.94%) stated they attended five universities’ orientation sessions. Of the sample, 1,150 students (12.41%) 68 represent missing data for orientations attended. Table 12 displays descriptive statistics by orientations attended. Table 12 Frequency Distribution of How Many Universities’ Orientation Sessions Students Attend Number of Universities’ N % Orientations Students Attended 1 6,523 80.37% 2 1,060 13.06% 3 256 3.15% 4 201 2.48% 5 76 0.94% Missing 1,150 12.41% Note. M = 1.31, SD = 0.73. Orientations attended was non-normally distributed with skewness of 2.88 (SE = 0.03) and kurtosis of 8.65 (SE = 0.05) Estimated family income. Perhaps not surprisingly, fewer students provided responses to the estimated family income survey item than any other survey item. Of the total sample, 5,579 students (60.21%) responded with an estimation of their annual family income. The income categories broke down by $50,000, where one represents an estimated family incomes less than $50,000. Two represents incomes of $50,000–$99,999. Three represents incomes of $100,000–$149,999. Four represents incomes of $150,000– $199,999. Five represents incomes of $200,000–$249,999, and six represents incomes of $250,000 or more (M = 3.06, SD = 1.67). The distribution of estimated family incomes 69 approximated normality with skewness of 0.51 (SE = 0.03) and kurtosis of -0.95 (SE = 0.07). The lowest income response choice, less than $50,000, received 1,086 answers (19.47%), and of those 952 remained enrolled on the census date, and 50 students were not enrolled on the census date. The next higher income response choice, $50,000–99,999, received 1,462 answers (26.21%), and of those 1,279 students remained enrolled at the University of Oregon on the census date while 74 were not enrolled. The two lowest income response choices combined suggest nearly half of survey respondents (45.68%) have an estimated family income less than $100,000. The middle tercile of income response choices represent an additional 31.47% of respondents. Specifically, 1,147 students (20.56% ) estimated their family income is $100,000–149,999, and of those, 1,043 remained enrolled at the UO while 21 did not on the census date. Meanwhile 609 students (10.92%) estimated their family income is $150,000–199,999, and of those 542 continued their enrollment on the census date while 22 did not. The upper tercile of income responses represents those students whose family income is $200,000 or more. With 459 students (8.23%) responding that their family income was $200,000–249,999, 417 of them were enrolled on the census date while 11 were not. The highest income choice, $250,000 or more, received 816 responses (14.63%) with 744 remaining enrolled on the census date and 19 not enrolled. Table 13 displays descriptive statistics, and Figure 8 displays the frequency distribution. 70 Table 13 Table of Estimated Family Total Income Last Year by Enrollment on Census Date Not Enrolled Estimated Family Income N % Valid % Enrolled (1) Less than $50,000 1,086 11.72% 19.47% 50 952 (2) $50,000–99,999 1,462 15.78% 26.21% 74 1,279 (3) $100,000–149,999 1,147 12.38% 20.56% 21 1,043 (4) $150,000–199,999 609 6.57% 10.90% 22 542 (5) $200,000–249,999 459 4.95% 8.23% 11 417 (6) $250,000 or more 816 8.81% 14.63% 19 744 Subtotal 5,579 60.21% 100.00% 197 4,489 Missing 3,687 39.79% Total 9,266 100.00% Note. M = 3.06, SD = 1.67. The distribution of estimated family incomes approximated normality with skewness of 0.51 (SE = 0.03) and kurtosis of -0.95 (SE = 0.07). 71 Figure 8. Frequency distribution of student responses by estimated annual family income. High school GPA. Of the 9,266 survey respondents, I was able to obtain a high school grade-point average for 8,599 students (92.80%) (M = 3.60, SD = 0.35). The distribution of high school GPAs approximated normality with skewness of -0.10 (SE = 0.03) and kurtosis of -0.48 (SE = 0.05). Among the students for whom I had high school GPA data, 8,236 students remained enrolled on the census date and had an average high school GPA of 3.61. Meanwhile, 363 students of the 8,599 were not enrolled on the census date, and their average high school GPA was 3.49, a difference of 0.11 GPA between groups. Table 14 displays descriptive statistics. 72 Table 14 High School Grade-Point Average and SAT/ACT Score by Enrollment Decision Enrollment Total SAT/ACT composite HS GPA sample (N) M SD M SD No 1,014 1155.35 149.94 3.49 0.36 Yes 8,252 1198.79 139.45 3.61 0.35 Total 9,266 1196.96 140.17 3.60 0.35 Note. For HSGPA, M = 3.60, SD = 0.35. The distribution of high school GPAs approximated normality with skewness of -0.10 (SE = 0.03) and kurtosis of -0.48 (SE = 0.05). For SAT-ACT composite, M = 1196.96, SD = 140.17). The distribution of SAT/ACT composite scores approximated normality with skewness of -0.22 (SE = 0.03) and kurtosis of -0.13 (SE = 0.05). SAT/ACT composite score. For students with an SAT/ACT score, 8,580 responded to a survey during orientation (M = 1196.96, SD = 140.17). The distribution of SAT/ACT composite scores approximated normality with skewness of -0.22 (SE = 0.03) and kurtosis of -0.13 (SE = 0.05). Among those, 8,219 students remained enrolled on the census date and had an average SAT/ACT score of 1198.79 (SD = 139.45), while 361 were not enrolled on the census date and had an average SAT/ACT score of 1155.35 (SD = 149.94). Between the two groups, students who remained enrolled on the census date had an average SAT/ACT composite score that was 43.44 points higher than those students who were not enrolled on the census date. Table 14 displays descriptive statistics. Proximity to home and residency. Geographic proximity to the University from students’ hometowns could be considered dichotomously as in-state students (residents) or out-of-state students (nonresidents). Such designation defines the students’ home states 73 as well as a their tuition rate. Students designated as residents represent 4,475 (51.89%) of survey respondents while domestic nonresidents represent 4,149 (48.11%) of survey respondents. International students did not receive the surveys. Among in-state students 4,346 respondents remained enrolled on the census date while 129 did not. Put another way, 97.12% of in-state survey respondents ultimately continued taking classes. Among out-of-state students, 3,906 respondents remained enrolled while 243 did not, in otherwords 93.78% of nonresident survey respondents continued taking classes at the university. Table 15 displays descriptive statistics. Table 15 Table of Residency by Enrollment on Census Date Not Enrolled Residency N % Enrolled Oregon residents 4,475 51.89% 129 4,346 Domestic nonresidents 4,149 48.11% 243 3,906 Total 8,624 100.00% 372 8,252 Proximity from home. While residency represents an important factor due to tuition rate implications, students’ actual proximity between the university and their home may offer more precise inferences for explaning variance in the decision-making process when layered on top of the residency factor. Students’ home address was collected from the application for admission and distance to campus is measured in miles. Among the respondents I obtained data on proximity from home for 8,375 students (M = 435.65, SD 74 = 568.80). The independent variable of proximity from home was non-normally distributed with skewness of 2.23 (SE = 0.27) and kurtosis of 5.03 (SE = 0.05). After inspecting the distribution for skewness, I transformed the independent variable of proximity using the base 10 log transformation. For all logistic regressions, I used the log10 transformation of the independent variable of proximity. The average distance from campus for in-state students who responded to the survey was 97.28 miles with a standard deviation of 169.68 miles. Among those in-state students, those who remained enrolled on the census date averaged 97.79 miles from campus, while those who did not enroll averaged 80.50 miles from campus. The average distance between campus and the homes of out-of-state students who responded to the survey was 792.61 miles with a standard deviation of 612.53 miles. Among those out-of- state students, those who remained enrolled averaged 792.19 miles from campus with a standard deviation of 612.86 miles, while those who did not remain enrolled averaged 799.43 miles from campus with a standard deviation of 608.38 miles. Table 16 displays descriptive statistics. 75 Table 16 Miles from Home by Residency and Enrollment on Census Date Enrolled Not Enrolled Total Residency n M SD n M SD n M SD Oregon residents 4,346 97.79 171.95 129 80.50 52.96 4,475 97.28 169.68 Domestic 4,105 792.61 612.53 nonresidents 3,869 792.19 612.86 236 799.43 608.38 Total 8,252 430.36 566.39 372 552.42 608.96 8,580 444.95 391.10 Note. M = 435.65, SD = 568.80). The independent variable of proximity from home was non-normally distributed with skewness of 2.23 (SE = 0.27) and kurtosis of 5.03 (SE = 0.05) 76 Gender. For this manuscript data on gender are represented dichotomously because one of the data sources, the Common App, asks students to identify their gender dichotomously as male or female. As previously described in the methods section, I maintain that gender is a more apt description of the variable than the physiological dichotomous designation of sex because the social construct may explain variance according to previous literature while physiological designations likely do not (Mansfield & Warwick, 2005). For students who identified as female, 5,024 (58.26%) responded to the survey, while 3,603 students (41.74%) who identified as male responded to the survey. Among the 5,024 female-identifed students, 4,799 remained enrolled on the census date while 225 did not. Among the 3,600 male-identifed students, 3,453 remained enrolled on the census date while 147 did not. Table 17 includes descriptive statistics. Table 17 Table of Gender by Enrollment on Census Date Gender Not Enrolled Enrolled Total Women 225 4,799 5,024 (58.26%) Men 147 3,456 3,603 (41.74%) Total 372 8,255 8,627 First-generation status. Among all survey respondents, 8,050 students provided data on their parents’ or guardians’ highest level of educational attainment. From those respondents 2,558 students (31.78%) had parents or guardians without a bachelor’s 77 degree or higher, classifying them as first-generation students. On the enrollment census date, 2,345 (91.67%) first-generation students who responded to the survey remained enrolled, while 144 (5.63%) first-generation students did not. Among their continuing- generation counterparts, which represented 5,492 (68.22%) total survey respondents, 5,299 remained enrolled on the census data representing 96.49% of continuing-generation survey respondents. Those not enrolled made up 3.51% (193) of continuing-generation respondents. Table 18 displays descriptive statistics. Table 18 Table of Generational-Status by Enrollment on Census Date Generational Status Not Enrolled Enrolled Total First-gen 144 2,345 2,489 (31.19%) Continuing-gen 193 5,299 5,492 (68.81%) Total 337 7,644 7,981 Waitlist status. Between 2014–2018, 176 students responded to the survey after spending time on the university’s waitlist before receiving full admission. The distribution of waitlisted students is imbalanced due to enrollment goals year-to-year. In 2014, for instance, 59 previously waitlisted students responded to the survey while only three previously waitlisted students responded the following year in 2015. Then the waitlist again rose in 2016 resulting in 94 previously waitlisted students responding to the survey. Consequently, no pattern exists for waitlist behavior due to enrollment changes. However, among 176 survey respondents who were waitlisted prior to admission, 168 78 remained enrolled on the census date while only eight were not enrolled. Table 19 displays descriptive statistics. Table 19 Table of Waitlisted Students by Cohort and Enrollment on Census Date Cohort n % of Respondents Not Enrolled Enrolled 2014 59 3.01% 3 56 2015 3 0.31% 1 2 2016 94 4.75% 3 91 2017 8 0.67% 0 8 2018 12 0.46% 1 11 Total 176 1.90% 8 168 Correlation results. I next examined the nonparametric correlations among the main effect independent variables and the dependent variable. For the independent variable rank choice, I ran a Spearman’s correlation to determine the relationship between students’ rank choice of university and their ultimate enrollment status. A weak, positive correlation exists, which was statistically significant (rs(8,539) = .022, p = .04). For the independent variable orientations attended, I similarly ran a Spearman’s correlation to determine the relationship between the number of universities’ orientations a student attends and their ultimate enrollment status. A weak, negative correlation exists, which was not statistically significant (rs(7,730) = -.012, p = .29). Table 20 displays correlation results. 79 Table 20 Spearman’s Correlations of Rank and Orientation Attendance with Enrollment Status. Independent Variable N Correlation Coefficient Rank 8,539 .02* Orientations 7,731 -.01 *. Correlation is statistically significant at the 0.05 level (2-tailed). Research Question Results I next provide results for tests addressing RQs 1–6. I used logistic regression to address research questions one and two. I then used Spearman’s correlation to address research question three. I returned to logistic regression for research questions 4–6. I first provide results from the tests of assumptions. Tests of assumptions. Logistic regression assumes a linear relationship between the continuous independent variables and the logit transformation of the dependent variable (Laerd Statistics, 2017). Since RQs 4–6 include continuous independent variables, I testd the assumption that the model is correctly specified using the Box- Tidwell (1962) approach to testing the linear relationship, which assesses whether the continuous independent variables linearly relate to the logit of the dependent variable. The continuous independent variables in the data are high school grade-point average, SAT-ACT test scores, and proximity from home. I assessed the linearity of continuous variables with respect to the logit of the dependent variable via the Box-Tidwell procedure. I applied a Bonferroni correction using all 11 terms in the model resulting in statistical significance accepted when p 80 < .0045 (Tabachnick & Fidell, 2013). Based on this assessment, I found no interaction terms statistically significant, and therefore all continuous independent variables linearly related to the logit of the dependent variable. I next tested for outliers, highlighting participants with standardized residuals greater than ±2 standard deviations. There were 229 standardized residuals ranging from - 10.425 to -2.527 standard deviations. After review of the outliers, all 229 ultimately were not enrolled on the census date. No other apparent commonality existed among the 229 outliers. I therefore kept all 229 participants in the analysis to try to better understand their circumstances, however I transformed the independent variables using the log base 10 transformation because it is a strong transformation with a major effect on distribution shape (Jason W Osborne & Overbay, 2008). I then re-ran the regressions for the RQ1–2 and RQ4–6 with the newly computed log10 independent variables. Logistic regression results for RQ1 and RQ2. I performed a binomial logistic regression to address RQ1 and RQ2, whether students’ rank choice of college and the number of orientations attended each affect the likelihood that students ultimately enroll at the university. I used the base log10 transformations of the independent variables. The logistic regression model for RQ1 with rank choice as the independent variable was statistically significant, X2(1) = 4.601, p = .032. However, the model explained only 0.2% (Nagelkerke R2) of the variance in enrollment and correctly classified 95.7% of cases. Sensitivity was 0.00%, and specificity was 100.00%. The logistic regression model for RQ2 with orientations attended as the independent variable was not statistically significant, X2(1) = .279, p = .597. Table 21 displays logistic regression results for both RQ1 and RQ2. 81 Table 21 Combined Independent Logistic Regression Results Predicting Likelihood of Enrollment based on Rank, then Likelihood of Enrollment based on Orientation Attendance. Odds 95% CI for Ratio Odds Ratio B SE Wald df p Lower Upper Constant 3.03 .06 2,331.74 1 .000 20.66 Rank 0.64 .31 4.40 1 .036 1.90 1.04 3.45 Constant 3.36 .07 2,309.90 1 .000 28.90 Orientations 0.22 .41 0.27 1 .602 1.24 0.55 2.80 Note. The log10 transformations of rank and orientation attendance were used as IVs due to extreme skewness. Spearman’s correlation results for RQ3. To address RQ three, whether the independent variables of rank choice and orientations attended relate to one another, I ran a Spearman’s correlation test. A statistically significant, weak negative correlation existed between the rank students assigned to the university and the number of orientations attended, rs(8,057) = -.032, p = .004. Table 22 displays results. Table 22 Spearman’s Correlation of Rank with Orientation Attendance. Independent Variables N Correlation Coefficient Rank by orientations 8,059 -.03* *. Correlation is statistically significant at the 0.05 level (2-tailed). 82 Logistic regression results for RQ4. In RQ4 I asked if the moderator variables, which are demographic in nature, affect the students’ enrollment decisions. To address RQ4, I next present results from the logistic regression. The overall test of statistical significance indicates the logistic regression model fits the data well, X2(24) = 130.213, p < .0005. Further the Hosmer and Lemeshow test similarly indicates the model is not a poor fit, p = .436. The model explained 16.7% (Nagelkerke R2) of the variance in whether a student remained enrolled on the census date and correctly classified 98.11% of participants. Sensitivity was nearly 100%, specificity was nearly 0.00%, positive predictive value was 98.11%, and negative predictive value was 100%. Of the 10 moderator variables only two were statistically significant: high school GPA and residency as shown in Table 23. The area under the Receiver Operating Characteristic (ROC) curve was .690, 95% CI [.630, .750], which bordered acceptable discrimination according to Hosmer et al. (2013). 83 Table 23 Logistic Regression for RQ4 Predicting Likelihood of Enrollment based on High School GPA, SAT-ACT Score, Cohort, Family Income, Residency, Proximity to Campus, Gender, Generational Status, and Waitlist Status. B SE Wald df p Odds 95% CI for Ratio Odds Ratio Lower Upper Constant -.639 1.55 .170 1 .681 .528 Cohort 22.22 3 .507 45.500 6.10 339.47 HSGPA 1.133 .390 8.45 1 .004* 3.106 1.45 6.67 SAT-ACT .002 .001 2.66 1 .103 1.002 1.00 1.00 Income 16.53 5 .005 Residency -1.052 .334 9.92 1 .002* .349 .181 .67 Proximity .034 .211 .025 1 .873 1.034 .68 1.56 Gender .077 .239 .10 1 .747 1.080 .68 1.73 First-gen .316 .254 1.55 1 .213 1.372 .83 2.26 Waitlist -.951 .757 1.58 1 .209 .386 .09 1.70 Note. *Statistical significance at the .0045 level according to the Bonferonni adjustment. “Wald” represents the Wald’s X2, a squared standardized z-score. Logistic regression results for RQ5. In RQ5 I asked if the main effects independent variable of rank choice, along with the moderator variables, affect the students’ enrollment decisions. To address RQ5, I performed a binomial logistic regression. The overall test of statistical significance indicates the logistic regression model fit the data well, X2(16) = 136.39, p < .0005. Further, the Hosmer and Lemeshow test indicates the model was not a poor fit, p = .468. The model explained 17.6% 84 (Nagelkerke R2) of the variance in whether a student remained enrolled on the census date and correctly classified 98.13% of participants. Sensitivity was nearly 100%, specificity was nearly 0.00%, positive predictive value was 98.13%, and negative predictive value was 100%. Of the 10 moderator variables only three were statistically significant: high school GPA and residency (as shown in Table 24). Table 24 Logistic Regression for RQ5 Predicting Likelihood of Enrollment based on Rank Choice, High School GPA, SAT-ACT Score, Cohort, Family Income, Residency, Proximity to Campus, Gender, Generational Status, and Waitlist Status. B SE Wald df p Odds 95% CI for Ratio Odds Ratio Lower Upper Constant -0.63 1.56 0.17 1 .684 0.53 Rank -0.03 0.63 0.00 1 .965 0.97 0.29 3.32 Cohort 9.73 3 .051 HSGPA 1.15 0.39 8.64 1 .003* 3.17 1.47 6.84 SAT-ACT 0.001 0.001 2.17 1 .140 1.00 1.00 1.00 Income 15.83 5 .007 Residency -1.04 0.34 9.68 1 .002* 0.35 0.18 0.68 Proximity 0.05 0.21 0.05 1 .815 1.05 0.69 1.59 Gender 0.06 0.24 0.07 1 .789 1.07 0.67 1.71 First-gen 0.35 0.26 1.89 1 .169 1.42 0.86 2.34 Waitlist -0.94 0.76 1.54 1 .215 0.39 0.09 1.72 Note. *Statistical significance at the .0045 level according to the Bonferonni adjustment. “Wald” represents the Wald’s X2, a squared standardized z-score. The rank choice IV and proximity moderator IV were calculated using the base log 10 transformation due to skewness. 85 Logistic regression results for RQ6. For RQ6 I asked if the main effects independent variable of orientations attended, along with the moderator variables, affect the students’ enrollment decisions. To address RQ6, I performed a binomial logistic regression. The overall test of statistical significance indicates the logistic regression model fits the data well, X2(16) = 118.469, p < .0005. The Hosmer and Lemeshow test similarly reveals the model is not a poor fit, p = .735. The model explained 16.1% (Nagelkerke R2) of the variance in whether a student remained enrolled on the census date and correctly classified 97.97% of participants. Sensitivity was 100%, specificity was 0.00%, positive predictive value was 97.97%, and negative predictive value was 100%, though notably, the model predicted zero cases would not be enrolled. Of the 10 predictor variables only one variable was statistically significant: residency (as shown in Table 25). 86 Table 25 Logistic Regression for RQ6 Predicting Likelihood of Enrollment based on Orientations Attended, High School GPA, SAT-ACT Score, Cohort, Family Income, Residency, Proximity to Campus, Gender, Generational Status, and Waitlist Status. B SE Wald df p Odds 95% CI for Ratio Odds Ratio Lower Upper Constant -0.89 1.57 0.32 1 .573 0.41 Orientations 0.14 0.82 0.03 1 .862 1.15 0.23 5.72 Cohort 7.41 3 .060 HSGPA 1.07 0.40 7.22 1 .007 2.92 1.34 6.38 SAT-ACT 0.002 0.001 3.67 1 .055 1.00 1.00 1.00 Income 15.17 5 .010 Residency -1.01 0.34 9.06 1 .003* 0.36 0.19 0.70 Proximity 0.04 0.21 0.03 1 .857 1.04 0.68 1.58 Gender -0.01 0.24 0.001 1 .975 0.99 0.62 1.60 First-gen 0.29 0.26 1.22 1 .269 1.33 0.80 2.22 Waitlist -0.91 0.76 1.44 1 .230 0.40 0.09 1.78 Note. *Statistical significance at the .0045 level according to the Bonferonni adjustment. “Wald” represents the Wald’s X2, a squared standardized z-score. The oreintation IV and proximity moderator IV were calculated using the base log 10 transformation due to skewness. 87 CHAPTER V DISCUSSION As the literature on decision-making and college choice suggested, the choice of which college to attend embodies a complex, multi-stage process (Chapman, 1981). The results of the binomial logistic regressions seem to confirm such complexity. I next discuss the results by statistically significant and non-significant outcomes. Statistically Significant Results The logistic regressions provided statistically significant results for rank, residency, and high school GPA, though the effects varied. I next discuss the statistically significant results from RQ1 and RQ4–6. Rank affects enrollment. The results of model one indicate the rank a student places on each of the colleges in their choice set does play a role in their ultimate enrollment decision. This would seem to confirm an element of Kahneman and Tversky’s Prospect Theory (1979) wherein a reference point is set during the decision-making process, and decision-makers compare subsequent options to the reference point. In a competition among multiple decision options, Prospect Theory suggests that the advantage goes to the reference point. In the competition among colleges, the advantage goes to a student’s top ranked option. Heuristically, it makes sense that the likelihood a student attends a specific college increases when a student ranks it higher among their choice set. Results from model one suggest the odds of a student enrolling after attending an institution’s orientation session are 1.90 times greater every rank higher the university climbs in students’ choice set. 88 However, such an outcome is not always the case. Students do not always attend the university ranked highest among their choice set. Model one includes no moderator variables, and when moderators add to the model, as in model five, rank no longer plays a statistically significant predictor in the likelihood of enrollment (p = .965). The simplicity of model one should give readers caution or risk overinterpretation of the results. When looking at more complex models, the college choice process, as Chapman (1981) and Tinto (1987) theorized, becomes more complex. GPA and residency affect enrollment. Of the 10 variables in models four, five, and six, only residency and high school grade-point-average added significantly to the model prediction of the likelihood of students’ enrollment. However, despite the statistically significant findings for both residency and high school GPA, the effect appears minimal. High school GPA. For high school GPA, the odds of a student enrolling after attending an institution’s orientation session are approximately three times greater for every one grade-point increase in high school GPA, which holds true in all three logistic regression models that included high school GPA. Put another way and more precisely, for every one-tenth of a GPA increase, the odds a student enrolls after attending an institution’s orientation session increase by approximately 20%. Residency. For residency, the effect was similar. The odds of a student enrolling at the institution are approximately 0.35 times greater for in-state students than for out-of- state students. Odds for residency were similar for all three logistic regression models that included the variable of residency. In a practical application for enrollment professionals, the effect described by the results for high school GPA and residency, 89 while statistically significant, are practically little better than a coin toss at predicting a student’s likelihood of enrolling. Statistically Non-Significant Results The Model of College Choice (Chapman, 1981) does not account for student behavior which may indicate institutional commitment. I identified college orientation attendance as a student behavior that may provide such an indication. Model two, however, suggests orientation attendance does not have statistical significance in students’ enrollment decision (p = .602). When adding moderators, as in model six, the lack of statistical significance remains (p = .862). Attending an orientation session therefore should not be a predictor of the likelihood of enrollment. Beyond the main effects of orientations attended and rank choice, moderator variables of cohort, estimated family income, SAT-ACT score, proximity to home, gender, first-generation status, and waitlist status did not provide statistically significant results and therefore should not be interpreted as predicting the likelihood of enrollment for students after their orientation session. Each of the aforementioned moderator variables were not statistically significant in all three regression models, four, five, and six. Limitations All research experiences threats to validity, and this study is no different. I next acknowledge existing threats to internal validity using the framework from Shadish, Cook, and Campbell (2002). Selection. The sample of respondents may pose a selection bias threat because those students who attended an orientation and volunteer to complete a survey may not accurately reflect the response of those students who did not attend an orientation or 90 opted-out of completing a survey. The literature is unclear on the effect of orientation on matriculation, but attending an orientation session may constitute selection bias. Instrumentation. From 2014–2018 the survey items relating to the IVs of research questions did not change, however the instrument did. Table 8 displays the instruments and years of use. In 2014–2016, the instrument used was the CIRP Freshman Survey, and in 2017–2018, the instrument used was the UO Freshman Survey. The UO Freshman Survey was a shorter survey with fewer items than the CIRP. It is unknown if the change of the instrument affected responses to unchanged items. Resentful demoralization. Although usually affecting an interaction between a control group and experimental group, resentful demoralization may have taken effect in participants of the Freshman Surveys due to the survey administration’s proximity to students’ academic advising sessions. For example, if students attend an academic advising session in which they receive negative news, such as their math placement test scores not qualifying them for a certain desired major, then students may enter the survey demoralized about the prospect of the University of Oregon being their top ranked college. Contextual, environmental factors, not accounted for in the model, may have altered the results. Experimenter bias. Social acceptability may pose a threat to the validity of responses because students may have provided survey responses intended to impress the staff administering the survey. For example, if a student believed that ranking the UO as their top choice on a survey may somehow affect their admission or financial aid status during orientation, they may have inaccurately ranked the UO as their top choice. Such bias within student responses may not be likely but is certainly possible. 91 Missing data. The amount of missing data within the dataset may pose a risk to the generalization and interpretation of results, specifically among the family income variable, which students reported at lower levels than other data. Missing data may insert estimation bias into the study (Stevens, 2004). The dataset contained enough data to successfully perform the regressions, however data on self-reported family income were missing more often than other variables. More data on student income may boost the generalization of results, though it is unknown if more data would alter the fundamental conclusions of how income affects the models. 92 CHAPTER VI CONCLUSIONS AND IMPLICATIONS Predicting the likelihood of enrollment is critical for university operations, but it’s not simple. Predicting any behavior for students who are typically 17–18 years old poses a challenge for researchers, and predicting their college enrollment is no different. Critically, the time immediately prior to the start of classes must receive further examination. A clear explanation remains elusive for why some students may attend their new student orientation session but not arrive on day one. Implications for Practitioners For enrollment professionals wanting to maximize their recruitment efforts as the first day of school approaches, the results suggest slightly more attention could be given to nonresident students and those with relatively mid-to-lower high school GPAs. With the odds of enrollment changing by approximately 20% by residency and by every one- tenth of a GPA point, giving more attention to non-residents and those with lower GPAs may yield only slight advantages in accomplishing enrollment goals. That said, practically, a 20% chance of rain seems little different from a 30% chance of rain when getting dressed in the morning. Similarly, when looking at all students who have attended an orientation session, focusing more on students with a 2.7 GPA than students with a 2.8 GPA because the 2.7 student is 20% less likely to enroll seems too granular of an effort for not much better odds. One element of contextual framework that may help practitioners lies within the understanding of students’ rank choice. Generally, most admissions professionals likely have a concept of choice sets, and whether their particular university is a students top 93 choice or safety school. What admissions professionals may benefit more from, however, is understanding with precision how the odds improve for every rank improvement. As stated in the results, the odds of a student enrolling at a university improve by 90% for every rank improvement among the student’s choice set. As an example, if an admissions counselor is in conversation with a prospective student, and through the course of the conversation learns that the student ranks their respective university third among the universities considered, by improving that rank from third to second in the student’s mind improves the likelihood the student will enroll by 90%. Such improvement may not prove meaningful, especially if the starting likelihood is low, but admissions practitioners may benefit from more precision when considering how to prioritize their efforts with prospective students. Practitioners could apply the knowledge learned from these results by working to identify nonresident students from resident students and applying unique marketing and experiences based on their residency classification. For example, during admissions receptions, administrators may benefit from knowing the home state of the students and families with whom they are in conversation. A simple indicator on nametags or some other indicator of residency that would be easily recognizable may provide admissions professionals the quick additional information that could help inform their conversations. Practitioners may also benefit from understanding more about specific student behaviors that may be indicators of institutional commitment. In this manuscript I only explored one student behavior, orientation attendance, however many more behaviors could be investigated for explaining variance. For example, a student attending a campus tour may indicate a higher level of institutional commitment or interest than a student 94 who does not tour. Tracking tour attendance and linking such attendance to other student behaviors in the recruitment process may help explain additional variance in the decision- making process. Explaining variance based on such student behavior requires that practitioners apply meticulous data collection methods, beginning with tracking attendance. Suggested Future Research For scholars of higher education and college choice, much remains unknown about how students make their decisions of which university to attend. Future research could help add to the literature and better explain the college choice phenomenon. Chapman (1981) suggested significant persons played an important role in the college choice process, and future researchers should consider better understanding the effect of parents, family members, and significant others on students’ college choice. As a correlary to the effect of significant persons, future research could specifically analyze the effect of legacy status on students. For continuing-generation college students, where their family members attended may (or may not) affect students’ college choice. As previously mentioned in the implications for practitioners, future research may benefit from better understanding specific student behaviors which may indicate institutional commitment, such as attending a campus tour and/or admissions events. Other student behaviors that could provide interest for future research could include interactions that are not in-person, such as engagement via social media. Such digital behaviors are tracked algorithmically across multiple industries such as retail, and digitally tracking student behavior may violate philosophical boundaries of academia, so 95 future researchers should tread carefully when considering digital tracking of students in order to explain decision-making. One consideration to better understand the decision-making about college choice may be to examin a counterfactual sample. While it may prove difficult to ascertain the decision-making of students who choose not to enroll at a university, their decision- making may reveal more than analyzing students who did enroll. In addition to studying a counterfactual sample of the population, future researchers could focus on a different setting or time as an explanation of the college choice. For example, what do the variables analyzed in this manuscript explain about the downstream effects on college retention and completion? Colleagues who study student success may benefit from understanding the pre-college variance between and among the students with whom they work. Future research could run the models with different dependent variables, such as retention to sophomore year or graduation instead of enrollment on the census date. For all future research about college choice, the inertia of change across the landscape of college admissions unfortunately belies much confidence in predicting student decision- making. Changing Landscape of College Admissions in the United States Further research must seek out more explanation, though future recruitment of college students may only get more unpredictable with recent changes to the higher education practices and perspectives nationwide. Among the changes in higher education include the rising costs that for many institutions currently outpace inflation. College is getting more and more expensive. If students identify cost and debt-aversion as important factors in their decision-making, other factors, previously ranked as important, may lose 96 influence as economic considerations crowd out social or academic considerations when students decide their college options. Among admissions practices, just as search engines revolutionized the college search process in the 1990’s, social media and algorithmic marketing push the boundaries of influence on students’ college choices. Enrollment professionals may not have the resources for data mining like those in other private industries, yet the application of such resources may not live far in the future, and enrollment professionals must grapple with the ethical boundaries of privacy, behavior predictability, and market forces of competition. Political perspectives may also influence professionals’ perspective of those boundaries. The value of higher education in America seems to adjust based on political polarization and as such, who applies to college (or specific colleges) may limit the ability for universities to diversify their student bodies, which is often listed among the goals of universities. When the pool of diverse applicants is low universities wrestle with who to admit by adjusting admissions practices. Federal courts, including the Supreme Court of the United States, have issued rulings on affirmative action as recently as the fall of 2019, suggesting the practice of deciding who to admit to college remains contested in America. Other trends in enrollment practices that may also provide unpredictability in the future of admissions include the increasing trend of eliminating standardized test scores from admissions criteria, the increasing trend of students applying to college with mental health concerns, and the decreasing trend of international students applying to college in the United States. Each of these trends represent a concern that may shake the admissions landscape in a way that substantially reduces enrollment predictability. Some of the 97 aforementioned concerns may not respond to influences from higher education professionals themselves, however those professionals may enact polices and practices to mitigate their effect. In a recent collective decision by admissions professionals in the United States, policies and practices were adjusted and the unpredictability of enrollment based on that decision is yet to be seen. Admissions code of ethics changes. On September 28, 2019, the Assembly for the National Association for College Admission Counseling (NACAC) voted to change its Code of Ethics and Professional Practice (CEPP) that establishes the agreed-upon professional conduct for admission practices at universities across the United States (“NACAC’s code of ethics and professional practices,” 2019). The Assembly voted to remove three provisions from the CEPP that the United States Justice Department claimed to stifle competition among colleges and therefore violate antitrust laws (Jaschik, 2019). One of the removed provisions prohibited the recruitment of first-year undergraduate students who have committed to another college. Without this provision, colleges and universities may continue to recruit students, offering scholarships or incentives, all the way until classes begin. As mentioned previously in the problems of practice, limited time that a student previously had to decide from which college to accept admission created a cost on the decision-maker (Loewenstein, 2000). Students have a limited time in which to build affinity and belonging with the college they choose, which is one of the necessary components of their retention (Tinto, 1987). The NACAC vote may very well provide students more time in the decision-making process so that they may make more-informed 98 and fitting decisions about their college choice. However, allowing colleges to continue to recruit students after they committed to another university may also pose an enormous challenge to the decision-making process because competing marketing efforts and influential external people may cause students to second-guess their decision (Tillery, 1973). Moreover, the problem of practice for universities magnifies as the upredictability of enrollment could increase due to delayed decision-making. The NACAC decision rewrites the expectations of conduct for the national admissions landscape. Higher education professionals may not fully understand the outcome of such policy changes for years, though I predict the students applying to colleges this year will understand swiftly as they will be the first cohort to apply to college under the new admissions practices. In this manuscript I attempted to identify indicators of student decision-making for college choice, their rank choice and the number of orientations attended among demographic factors, and one difficulty was attempting to measure why students choose not attend a college so close to the start of classes. It is tantamount to trying to understand why four-percent of people running a marathon may run the first 26 miles and choose not to complete the final 50 feet of the race. Such a research agenda is important and would make a substantial contribution to social science, however, based on the changing landscape of admissions, the research may be quite difficult. Perhaps an important addition to the metaphor of trying to understand why some college applicants choose not to complete the final 50 feet of the admissions marathon should include that the race 99 takes place during an earthquake. Enrollment researchers try to predict the decision- making and behavior of 17- and 18-year-olds while the ground shifts beneath our feet. 100 APPENDIX A PORTIONS OF UNIVERSITY OF OREGON’S APPLICATION FOR ADMISSION 101 102 103 APPENDIX B 2014 CIRP FRESHMAN SURVEY INSTRUMENT 104 105 106 107 108 109 APPENDX C 2018 UO FRESHMAN SURVEY INSTRUMENT 110 111 APPENDIX D DATA USE AND NON-DISCLOSURE AGREEMENT 112 113 114 REFERENCES CITED 2018–2019 Survey Materials Frequently Asked Questions. (2018). Retrieved October 19, 2018, from https://surveys.nces.ed.gov/ipeds/VisFaqView.aspx?mode=reg&id=3#795 About the UO. (2018). Retrieved October 27, 2018, from https://www.uoregon.edu/about Astin, A. W. (1993). What matters in college?: Four critical years revisited (1st ed.). San Francisco: Jossey-Bass. Banner Guide: Display a Student’s Admission Records. (n.d.). Retrieved June 2, 2019, from https://bg.uoregon.edu/content/display-a-students-admission-records- sqaadms Box, G. E. P., & Tidwell, P. W. (1962). Transformation of the independent variables. Technometrics, 4. doi:10.1080/00401706.1962.10490038 Bruni, F. (2015). Where you go is not who you’ll be: An antidote to the college admissions mania. New York: Grand Central Publishing. Carnevale, A. P., & Van Der Werf, M. (2017). The 20% solution: Selective colleges can afford to admit more pell grant recipients. Retrieved from https://cew.georgetown.edu/cew-reports/pell20/#full-Report Chapman, D. W. (1981). A model of student college choice. The Journal of Higher Education, 52, 490–505. https://doi.org/10.2307/1981837 CIRP Freshman Survey. (2018). Retrieved October 22, 2018, from https://heri.ucla.edu/cirp-freshman-survey/ Clinedinst, M. E., & Koranteng, A.-M. (2017). State of college admission 2017. Retrieved from https://www.nacacnet.org/globalassets/documents/publications/research/soca17fin al.pdf Clinedinst, M. E., Koranteng, A.-M., & Nicola, T. (2015). State of college admission 2015. Retrieved from https://www.nacacnet.org/news-- publications/publications/state-of-college-admission/ Clinedinst, M. E., & Patel, P. (2018). State of college admission 2018. Retrieved from https://www.nacacnet.org/globalassets/documents/publications/research/2018_soc a/soca18.pdf 115 Covarrubias, R., & Fryberg, S. A. (2015). Movin’ on up (to college): First-generation college students’ experiences with family achievement guilt. Cultural Diversity and Ethnic Minority Psychology, 21, 420–429. https://doi.org/10.1037/a0037844 David, M. E., Ball, S. J., Davies, J., & Reay, D. (2003). Gender issues in parental involvement in student choices of higher education. Gender and Education, 15(1), 21–37. Retrieved from http://libproxy.uoregon.edu/login?url=http://search.ebscohost.com/login.aspx?dire ct=true&db=eric&AN=EJ667283&site=ehost-live&scope=site Dawis, R. V., & Lofquist, L. H. (1964). A theory of work adjustment. Minnesota Studies in Vocational Rehabilitation, 15. Dunnett, A., Moorhouse, J., Walsh, C., & Barry, C. (2012). Choosing a university: A conjoint analysis of the impact of higher fees on students applying for university in 2012. Tertiary Education and Management, 18, 199–220. Eagan, M. K., Stolzenberg, E. B., Ramirez, J. J., Aragon, M. C., Suchard, M. R., & Rios- Aguilar, C. (2016). The American freshman: Fifty-year trends, 1966–2015. Retrieved from https://www.heri.ucla.edu/monographs/50YearTrendsMonograph2016.pdf Elliott, D. C. (2016). The impact of self beliefs on post-secondary transitions: The moderating effects of institutional selectivity. Higher Education: The International Journal of Higher Education Research, 71, 415–431. Galotti, K. M. (1995). A longitudinal study of real‐life decision making: Choosing a college. Applied Cognitive Psychology, 9, 459–484. https://doi.org/10.1002/acp.2350090602 Gigerenzer, G., & Selten, R. (Eds.). (2002). Bounded Rationality: The Adaptive Toolbox. Cambridge, MA: The MIT Press. Griffith, A., & Rask, K. (2007). The influence of the US News and World Report collegiate rankings on the matriculation decision of high-ability students: 1995– 2004. Economics of Education Review, 26, 244–255. Gu, M., & Magaziner, J. (2016). The Gaokao: History, reform, and rising international significance of China’s national college entrance examination. World Education News & Reviews (WENR), 1–8. Hanoch, Y., Wood, S., & Rice, T. (2007). Bounded rationality, emotions and older adult decision making: Not so fast and yet so frugal. Human Development, 50, 333–358. https://doi.org/10.1159/000109835 Higher Education Act of 1965, Pub. L. No. 89–329, § 1001 et seq. 79 Stat. 1219 (1965). 116 Higher Education Research Institute. (2011). 1971-2011 CIRP Freshman Survey Trends Item List. Los Angeles. Retrieved from https://heri.ucla.edu/instruments/ Homans, G. (1961). Social behavior: Its elementary forms. Retrieved from http://search.proquest.com/docview/37730326/ Hosmer, D. W., Lemeshow, S., & Sturdivant, R. X. (2013). Applied logistic regression. (Third ed.). Hoboken, New Jersey: Wiley. Huang, F. L., & Moon, T. R. (2013). What are the odds of that? A primer on understanding logistic regression. Gifted Child Quarterly, 57, 197–204. https://doi.org/10.1177/0016986213490022 IBM Corp. Released. (2017). IBM SPSS Statistics for Macintosh (Version 24) [software]. Armonk, NY: IBM Corp. Jaschik, S. (2019). Department of Justice probes admissions ethics code. Retrieved September 14, 2019, from https://www.insidehighered.com/news/2018/01/10/department-justice- investigating-antitrust-issues-regard-nacacs-ethics-code Kahneman, D. (2014). Thinking Fast and Slow. Igarss 2014, (1), 1–5. https://doi.org/10.1007/s13398-014-0173-7.2 Kahneman, D., & Tversky, A. (1979). Prospect theory : An analysis of decision under risk. Econometrica, 47, 263–292. Kaminer, A. (2017, November 15). Applications by the dozen, as anxious seniors hedge college bets. The New York Times. Retrieved from https://www.nytimes.com/2014/11/16/nyregion/applications-by-the-dozen-as- anxious-students-hedge-college-bets.html Krumboltz, J. D. (1992). The wisdom of indecision. Journal of Vocational Behavior, 41, 239–244. https://doi.org/10.1016/0001-8791(92)90025-U Laerd Statistics. (2017). Binomial logistic regression using SPSS Statistics. Statistical Tutorials and Software Guides., 1–14. Retrieved from https://statistics.laerd.com/ Little, R. J. A., & Rubin, D. B. (2002). Statistical analysis with missing data (2nd ed.). Hoboken, NJ: Wiley. Liu, A. (2011). The admission industrial complex: Examining the enterpreneurial impact of college access. Journal of College Admission, (210), 8–19. Retrieved from http://search.proquest.com/docview/853876708/ 117 Loewenstein, G. (2000). Is more choice always better? National Academy of Social Insurance, 1–15. Retrieved from https://www.andrew.cmu.edu/user/gl20/GeorgeLoewenstein/Papers_files/pdf/too_ much_choice.pdf Long, J. S. (1997). Regression models for categorical and limited dependent variables. Thousand Oaks: Sage Publications. Luo, J., & Jamieson-Drake, D. (2005). Linking student precollege characteristics to college development outcomes: The search for a meaningful way to inform institutional practice and policy. Association of Institutional Research, 7. Mansfield, P. M., & Warwick, J. (2005). Gender differences in students’ and parents’ evaluative criteria when selecting a college. Journal of Marketing for Higher Education, 15(2), 47–80. NACAC’s code of ethics and professional practices. (2019). Retrieved from https://www.nacacnet.org/advocacy--ethics/NACAC-Code-of-Ethics/ National Center for Education Statistics. (2018). Retrieved from https://nces.ed.gov Office of Institutional Research. (2019). Detailed enrollment. Retrieved on June 6, 2019, from https://ir.uoregon.edu/detailenroll Office of the Registrar: Faculty & Staff. (2019). Retrieved June 6, 2019, from https://registrar.uoregon.edu/faculty-staff Oliveira, A. (2007). A discussion of rational and psychological decision-making theories and models : The search for a cultural-ethical decision-making model. Journal of Business Ethics, 12, 1478–1482. https://doi.org/10.1111/j.1572- 0241.1998.00467.x Osborne, J. W., & Overbay, A. (2008). Best practices in data cleaning: How outliers and “fringeliers” can increase error rates and decrease the quality and precision of your results. In J. W. Osborne (Ed.), Best practices in quantitative methods (205– 213). Thousand Oaks, CA: SAGE Publications. https://doi.org/10.4135/9781412995627 Pascarella, E. T., Terenzini, P. T., & Wolfle, L. M. (1986). Orientation to college and freshman year persistence/withdrawal decisions. The Journal of Higher Education, 57, 155–175. https://doi.org/10.2307/1981479 Peng, C. J., Lee, K. L., & Ingersoll, G. M. (2002). An introduction to logistic regression analysis and reporting, The Journal of Educational Research 96(1), 3–14. 118 Remark Office OMR. (2019). Retrieved June 10, 2019, from https://remarksoftware.com/products/office-omr/ Rickard, J. (2017). The common application. Retrieved June 12, 2017, from https://www.commonapp.org Scalise, K. (2016). EDLD 560: Measurement and Assessment, week 3 notes [PowerPoint slides]. Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and quasi- experimental designs for generalized causal inference. Boston, MA: Houghton, Mifflin and Company. Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics, 69(1), 99–118. https://doi.org/10.2307/1884852 Simon, H. A. (1956). Rational choice and the structure of the environment. Psychological Review, 63(2), 129–138. https://doi.org/10.1037/h0042769 Simon, H. A. (1957). Models of man: Social and rational. New York: Wiley. Simon, H. A. (1959). Theories of decision-making in economics and behavioral science. American Economic Review, 49, 253–283. Skořepa, M. (2011). Decision-making : A behavioral economic approach. Basingstoke. New York: Palgrave Macmillan. Smith, J., Pender, M., & Howell, J. (2013). The full extent of student-college academic undermatch. Economics of Education Review, 32, 247–261. https://doi.org/10.1016/j.econedurev.2012.11.001 Soo, K. T. (2013). Does anyone use information from university rankings? Education Economics, 21(2), 176–190. Tabachnick, B. G., & Fidell, L. S. (2013). Using multivariate statistics (6th ed.). Boston: Pearson Education. Tillery, D. (1973). Distribution and differentiation of youth: A study of transition from school to college. Cambridge, MA: Ballinger. Tinto, V. (1987). A theory of individual departure from institutions of higher education. Leaving College: Rethinking the Causes and Cures of Student Attrition, 84–137. Tinto, V. (1988). Stages of student departure: Reflections on the longitudinal character of student leaving. Journal of Higher Education, 59, 438–455. https://doi.org/10.2307/1981920 119 Undergraduate Retention and Graduation Rates. (2018). Retrieved May 19, 2019, from https://nces.ed.gov/programs/coe/indicator_ctr.asp 120