Segregation, Choices, and Inequality in Educational Outcomes: Evidence From Revealed Choice Data Open Access

Abstract

Introduction

Inequality in college degree attainment among segregated social groups is a central concern for population scientists and policymakers because it underlies disparities in occupational attainment, income, health, family structure, and civic engagement (Cheng et al. 2019; Ciocca Eller and DiPrete 2018; Hargrove et al. 2022; Hout 2012; Zhou and Pan 2023). Ample evidence links residential, educational, and social segregation to unequal opportunity structures and educational disparities (Bischoff and Owens 2019; Lichter and Johnson 2021; Malmberg et al. 2018; Marteleto 2012; Owens 2016, 2020; Rich et al. 2021). Underprivileged minority groups often reside in racially, ethnically, and socioeconomically homogeneous neighborhoods, with limited access to good schools, qualified teachers, and successful role models with accurate information about academic options (Bischoff and Tach 2020; Bruch and Swait 2019; Dynarski et al. 2023; Lichter et al. 2016; Massey 1990; Owens 2020; Reardon 2016). The financial constraints members of disadvantaged groups face further increase their sensitivity to geographic proximity, tuition costs, and overly complex financial aid policies (Burland et al. 2023; Dynarski et al. 2021; Niu and Tienda 2008). As a result, the academic options they might consider are more restricted, increasing their probability of applying to/attending programs incompatible with their academic credentials—a phenomenon known as “academic mismatch”—or forgoing college altogether (Alon and Tienda 2005; Black et al. 2015; Black and Sufi 2002; Bowen and Bok 1998; Cortes and Lincove 2016; Dyer and Román-Torres 2022; Hoxby and Turner 2013).

Notwithstanding the large body of evidence documenting the implications of segregation on the academic outcomes of minority groups, little attention has been given to how the unequal circumstances generated by segregation influence the decision-making processes underlying choices and lead to inequality in outcomes. Particularly absent is evidence of how and why individuals from segregated social groups make different life-changing choices and how these variations augment disparities in life chances and reproduce inequality.

This lacuna in demographic research is both theoretical and empirical. Theoretically, decision-making processes are critical for understanding demographic patterns, given that they link unequal macro-level conditions of segregation to stratified micro-level choices that further perpetuate inequality. Isolating the role of stratified decision-making processes in amplifying inequality from that of resources is an important goal of population science (Thomas and Frankenberg 2015). Yet, without evidence capturing inequality in decision-making, we might attribute aggregate inequality in many realms (education, health, employment, or residence) to other factors. A deeper understanding of the structural roots of the decision-making process can sharpen the demographic perspective on diverging life course trajectories.

Empirically, the data necessary to accurately model choices are scarce in demographic research. Most studies on between-group disparities analyze information on destinations (i.e., colleges attended and neighborhoods lived in) rather than decision-making processes (which options individuals consider, how they rank them, and what considerations guide them). To identify the unique effect of choices on outcomes, demographers should seek data on individuals’ revealed choices, including all aspects of individuals’ choice architecture: (1) the choice set, the full array of options available and their characteristics; (2) the consideration set, the short list of selected options and their ranking so we can identify the first choice; and (3) individuals’ characteristics and aptitude.

This study demonstrates how inequality in opportunity structure due to segregation stratifies decision-making processes, resulting in inequality in outcomes. We focus on the choices and outcomes of applicants to universities from the majority and minority groups in Israel: Jews and Palestinian Arabs (hereafter, “Arabs”). This case is particularly effective for evaluating the implications of segregation on decision-making because of the high level of residential, educational, and social segregation, which allows for little between-group contact and nearly complete homogamy (Haberfeld and Cohen 2007; Lewin-Epstein and Semyonov 1986, 1994; Okun and Friedlander 2005; Semyonov 1988; Shwed et al. 2014). This case of extreme segregation is characterized by sharp disparities in the opportunity structure, with the Arab minority disadvantaged in almost every socioeconomic aspect. In this context, the value of education is likely to vary across groups, yielding detectable differences in decision-making (Alon and DiPrete 2015; Holland and DeLuca 2016; Wilkerson 2020). This context also distills the role of segregation in shaping decision-making because of the blurring of other stratifying engines, such as skin color/race, national origin, immigration status, culture, language, religion, and legal status (Adler et al. 2005; Bar-Haim and Semyonov 2015; Haberfeld and Cohen 2007; Hummer 2023; Martinez et al. 2023). Thus, this case transcends traditional demographic distinctions of ethnic or racial groupings, allowing a clearer understanding of the effect of segregation on inequality in life outcomes.

What makes this case study unique is that, in sharp contrast to the segregation characterizing Arab and Jewish relations before college entrance (including the K–12 system), the higher education system is a beacon of meritocracy. Israeli universities’ admissions are mechanistic and transparent. In this early specialization system, applicants rank and apply directly to field-of-study (FOS) programs within a university. Admission is based solely on test scores and high school matriculation GPA, and information on admission thresholds is readily available to all applicants. Moreover, programs differ in their admission selectivity but are similar in their geographic location (same university) and tuition costs (same low tuition across universities). This setting naturally mitigates choice constraints related to costs, proximity, and information and eliminates the enigmatic effect of holistic admissions. Thus, focusing on differences in choices and outcomes of Jewish and Arab university applicants in Israel provides a unique opportunity to examine two segregated groups embedded in sharply different opportunity structures before and after college but who enjoy equal opportunity (a similar choice set) in higher education.

We investigate the link between segregation, decision-making, and outcomes using unique administrative records of applicants’ revealed choices, academic aptitude, and admission decisions. These records capture 70% of all applicants to research universities in Israel during the study years (1999–2008) and half of applicants to all higher education institutions. The data on applicants’ revealed choices allow us to juxtapose the two groups’ choice architecture and evaluate their respective causal effects on outcomes. The data contain the full list of options (the choice set), the short list of options they apply to (the consideration set), the order of these alternatives (the ranked preferences), and the characteristics of each option (the alternatives’ attributes). Finally, the data contain records on admission decisions, allowing us to link choice variations to a tangible outcome with stratifying implications.

Results from conditional logit (choice) models yield three novel findings. First, we find substantial group disparities in applicants’ choices, with a higher prevalence of academic mismatch among Arab applicants, even in a setting free of geographic, financial, and informational variations across options and net of academic aptitude. Second, disparities are driven by between-group variations in the weight placed on programs’ characteristics. When choosing a program, Jewish applicants prioritize academic fit, whereas Arab applicants prioritize future employment security and (to a lesser extent) a welcoming social environment and graduation chances. Third, these differences in decision-making are consequential for inequality. They account for the bulk of the gap in admission to Israeli universities among Jewish and Arab applicants, more than the share explained by differences in academic preparation. These results illuminate the interplay between segregation, decision-making, and inequality in life trajectories. They reveal that segregation shapes variations in choices, leading to significant disparities in educational attainment and suboptimal outcomes for minority groups. We conclude by discussing the theoretical and policy implications of our findings for research in population science.

Segregation and Decision-making Processes

Decision-making processes are an essential component of human behavior, and differences in them can generate aggregate patterns of socioeconomic, racial, and gender inequality. Nonetheless, questions about whether and how unequal circumstances of segregation influence decision-making processes receive limited attention in population research and classical decision-making literature, where decision-making is viewed as a cognitive process shaped by personal aptitude, preferences, information, perceptions, and biases. Decision-makers are portrayed as rational actors who assess all available alternatives, weigh relevant information and consequences, rank their options, and select the one maximizing utility (Becker 2009; March 1994). Yet, individuals’ rationality is bounded owing to restricted information, cognitive capacity, heuristics, biases, and judgmental errors (Kahneman and Tversky 2013; Simon 1955; Tversky and Kahneman 1974). Although acknowledged, the role of social context is often vague and limited to biases and information constraints (Akerlof 1997).

Sociological approaches, particularly those focused on education, treat the association between social circumstances and decision-making more straightforwardly. Boudon (1974) introduced the concept of the primary and secondary effects of family background on educational outcomes. Primary effects are generated by disparities in academic achievements between social groups; secondary effects are generated by disparities in choices made, given prior academic achievements (Jackson 2013; Jonsson and Erikson 2000). Variations in opportunity structure are understood to create varying costs and benefits associated with different options, leading to different choices even if information or resources are unconstrained. For instance, avoiding downward mobility is likely a greater concern for privileged than underprivileged individuals, giving the former stronger incentives to pursue higher education (Breen and Goldthorpe 1997). Yet, this theoretical framework has rarely been tested with data on revealed choices (but see Hällsten 2010).

Social psychological models offer insights into the significance of social circumstances for decision-making processes. The Wisconsin model reveals the influence of significant others on occupational aspirations (Sewell et al. 1969; Zimmermann 2020), and psychological models show that heuristics can vary across social groups (Eccles 1994; Tversky and Kahneman 1974). Social settings shape variations in perceptions and beliefs about “what people like me can do” (Correll 2001, 2004), which are especially important for educational choices in advanced nations, where norms of self-expressive behavior tend to be stronger (Charles 2017; Charles and Bradley 2009). Here, individuals from different social groups might employ different guidelines to choose between options (Quadlin 2020). Social networks characterized by high homophily exacerbate these disparities and strengthen the association between demographic characteristics and choices (Centola 2015; Centola and Macy 2007; Christakis and Fowler 2007; Smith et al. 2020).

Building on these perspectives and findings, we suggest that decision-making processes are anchored in individuals’ opportunity structures, encompassing the conditions and institutions shaping life experiences and trajectories, including family, friends, neighborhoods, the education system, and the labor market. The conditions vary along spatial, gender, race, and class lines and are directly linked to the extent of social, residential, educational, and occupational segregation across groups (Bischoff and Owens 2019; Bruch and Swait 2019; Owens 2020; Zheng and Weeden 2023). Consequently, we argue that segregation yields differences in opportunity structures that generate between-group disparities in the availability of choice options, as prior studies showed, and generate discrepancies in the decision-making process used to select among available options.

Choices and Utility Weights: The Case of College Application

We study the stratified decision-making process in college application behavior, a critical juncture in young adults’ life course and a focal point in stratification in educational attainment. Applicants from disadvantaged groups are more likely than their privileged counterparts to apply to colleges academically mismatched with their credentials, leading to diverging educational and employment trajectories (Black et al. 2020; Bowen et al. 2009; Cortes and Lincove 2016). Applicants who overreach by applying to programs or colleges with admission criteria substantially above their academic qualifications risk not meeting the admission bar, interrupting their transition to higher education (Gelbgiser and Alon 2024; Tessler 2022). Applicants who apply to overly safe programs with admission criteria substantially below their academic credentials risk placement in an unchallenging academic environment that lowers satisfaction, diminishes graduation chances, and limits labor market returns (Dillon and Smith 2020; Gelbgiser and Alon 2024; Jagesic et al. 2022; Light and Strayer 2000; Ovink et al. 2018; Wolniak and Muskens 2021).

Most studies (and subsequent policy interventions) have anchored disadvantaged applicants’ high rate of academic mismatch in resources and information constraints. Because of financial constraints, disadvantaged applicants in the United States are more likely to apply to universities with a lower price tag or those closer to home, even if those schools are below their academic capabilities (Bowen et al. 2009; Ovink et al. 2018). Similarly, limited access to accurate and comprehensible information on financial aid and admission policies can sway talented, disadvantaged applicants from applying to academically competitive institutions (Black et al. 2015; Cortes and Lincove 2016, 2019; Dynarski et al. 2021; Hoxby and Turner 2013, 2015). The implicit assumption is that providing more information or alleviating material concerns with financial aid will mitigate group variations in the rate of academic mismatch.

The emphasis on constraints preventing applicants from choosing matched choices, however, is an oversimplified portrayal of college application choices. From a decision-making perspective, applicants weigh options’ characteristics, attempting to maximize rewards and minimize risks. When selecting a college or academic program—one of the most complex, uncertain, and consequential decisions in life—applicants consider the likelihood of admission, potential academic success, the program's social makeup, and expected employment prospects and assign weights according to their preferences (Alon and DiPrete 2015; Alon and Gelbgiser 2011; Black et al. 2020; Borgen and Hermansen 2023; Finger 2021; Gelbgiser and Alon 2016; Hällsten 2010). The possibility of academic mismatch can be elevated if applicants prioritize other factors over academic fit. Thus, systematic between-group variations in the weights applicants assign to alternatives’ characteristics will yield variations in choices unrelated to costs, information, or admission criteria.

There are compelling reasons to expect group differences in utility weights in college applications. Economic deprivation has been linked to risk aversion, particularly in uncertain choices, with underprivileged individuals exhibiting higher risk aversion (Breen et al. 2014). Halaby (2003) suggested that underprivileged students might prioritize educational tracks leading to secure but lower reward jobs, reflecting “bureaucratic” job values; privileged students might opt for less secure but potentially higher reward tracks, reflecting “entrepreneurial” job values (see Hällsten 2010). Indeed, Holland and DeLuca (2016) showed that low-income minority students attempt to reduce long-term employment and economic uncertainty by selecting expensive for-profit colleges promising clear career pathways rather than affordable local community colleges. These findings imply that the trade-off between expected rewards and associated risks varies along demographic characteristics, including race, family background, educational experiences, and gender (Alon and DiPrete 2015; Halaby 2003; Hällsten 2010). We expect greater segregation and exposure to discrimination will lead applicants from disadvantaged groups to anticipate restricted employment opportunities and to prefer programs leading to stable employment, even at the risk of lower potential pecuniary returns or academic mismatch.

Variations in the institutional arrangements of colleges and programs—including curriculum structure, grading norms, faculty, and student support and composition—can yield variations in utility weights. To avoid a futile and costly investment (i.e., dropout), disadvantaged applicants might prioritize arrangements associated with a higher likelihood of graduation (Alon and Gelbgiser 2011; Gabay-Egozi et al. 2010; Gelbgiser and Alon 2016). Similarly, they might place more weight on demographic composition, perceiving programs with similar peers as more welcoming and inclusive (Alvarado and López Turley 2012; Banerjee 1992; Black et al. 2020; Seymour and Hunter 2019; Winston and Zimmerman 2013). This consideration is especially important for disadvantaged groups in selective colleges and programs, where their risk of isolation is higher (Bourdieu 1984; Jack 2019). Indeed, evidence shows that women, minority, and low-socioeconomic-status students are more likely to enroll in universities and programs with a higher proportion of demographically similar students, even after accounting for prior academic achievements and available resources (Alon and DiPrete 2015; Brown and Hirschman 2006; Cortes and Lincove 2019; Gelbgiser 2021; Thompson et al. 2019). In contrast, privileged applicants are incentivized to maintain class succession, making them reluctant to risk their chances of admission (Bowen et al. 2009; Mullen and Goyette 2019).¹ Their educational choices are more attuned to academic mismatch because it could disrupt their transition to college and lead to generational decline (Breen and Goldthorpe 1997; Finger 2021; Zimmermann 2020). Moreover, as a majority group, they are less concerned about their program's social environment or employment discrimination.

We argue that the segregation of socially constructed groups can lead to differences in educational choices, even when applicants have equal access to information and options do not vary by geography or cost. These differences are related to concerns about social isolation, academic success, labor market discrimination, and employment security. Hence, differences in the weight placed on the characteristics of an option are rooted in groups’ opportunity structure. Yet, as reasonable as these concerns are, variations in choices can be an important mechanism for inequality because they lead to suboptimal educational outcomes for disadvantaged groups, further stratifying academic trajectories and reproducing inequality and segregation.

Taking advantage of the high level of segregation among Arabs and Jews in Israel, the unique structure of university admission, and the availability of rich data on revealed choices, we empirically evaluate variations in educational decision-making and their stratifying consequences with three research questions:

• Research Question 1 (RQ1): How does the likelihood of academic mismatch in university application choices differ across segregated groups—in this case, Jewish and Arab university applicants?

• Research Question 2 (RQ2): What considerations underlie group disparities in the likelihood of academic mismatch in choices?

• Research Question 3 (RQ3): What are the consequences of choice disparities for inequality in outcomes—in this case, university admission?

Empirical Investigation

The Setting

Arab–Jewish Stratification in Israel

Of the 6.5 million citizens of Israel in the study years (1999–2008), most (77%) are descendants of Jews who migrated from the Jewish Diaspora or lived in the area before the establishment of Israel; roughly 20% are Arabs, mostly descendants of Palestinian villagers who stayed under Israel's sovereignty after the 1948 war (Israel Central Bureau of Statistics 2010: table 2.2; Semyonov and Lewin-Epstein 2004:1–13). Several features of this case make it useful to evaluate how the segregation of socially constructed groups can stratify decision-making. First, the opportunity structure of Arabs and Jews, as defined by the social institutions shaping their life outcomes, is markedly different. They are segregated socially, educationally, and residentially, with little between-group contact or marriage and sharp socioeconomic disparities. Most Arabs (85%) live in exclusively Arab villages or towns, and dissimilarity indices in mixed cities vary between 40% and 65% (Miaari and Khattab 2013). The K–12 schooling system is segregated, with only 6% of students attending mixed schools (Shwed et al. 2014), preventing the formation of crosscutting social links (Schnell et al. 2015), increasing labor market discrimination (Miaari and Khattab 2013; Semyonov 1988), and restricting Arabs’ employment opportunities (Lewin-Epstein and Semyonov 1994; Schnell and Shdema 2016).

Second, this case allows a clearer evaluation of the implications of segregation across socially constructed groups rather than racial or ethnic hierarchies. The distinctions between Jews and Arabs in Israel do not easily fit traditional definitions of ethnic or racial groupings (Hummer 2023; Martinez et al. 2023). They share many physical and cultural similarities and exhibit substantial within-group heterogeneity in skin color, ethnicity, geonational origin, religion, and immigration status. For example, Jews migrated from a plethora of countries that differ in cultural orientation and economic development, including many Arab and Muslim countries (Bar-Haim and Semyonov 2015; Lewin-Epstein and Cohen 2019). Moreover, skin color varies substantially within populations of Jews and Arabs, including darker skinned Jews from Ethiopia and North Africa and lighter skinned ones from Europe and North America. Similarly, the Arab minority consists of multiple ethnoreligious groups—including Muslims (80%), Christians (10%), and Druze (10%)—who differ in culture, skin color, physical traits, residential patterns, and civic participation. Although these dimensions matter for socioeconomic patterns (Haberfeld and Cohen 2007), the disparities between the Arab minority and Jewish majority in all socioeconomic outcomes far exceed them (Bar-Haim and Semyonov 2015; Semyonov and Lewin-Epstein 2004:1–13), indicating a social system rooted in the specific sociohistorical context of Israel (Hummer 2023). Thus, this case allows a clearer evaluation of the role of segregation and the restrictions it imposes on groups’ opportunity structures in shaping choices.

Structure of Application and Admission to Israeli Universities

Israeli universities are integrated, transparent, and mechanistic, and their characteristics of admission make them an ideal setting to evaluate the decision-making processes underlying college applications.² First, because the country has an early specialization system, all application and admission processes are program-specific. Each university offers various FOS programs differing in admission requirements (more and less selective), curriculum structure, student body, and potential labor market payoff, thus delineating a clear choice set within each university (Alon 2015). At the same time, academic programs are similar in their geographic location (same university), stage (bachelor's level), and cost (relatively low and government-controlled). These unique arrangements allow a focus on within-institution application patterns, neutralizing any pecuniary and geographic considerations driving academic mismatch. Moreover, program characteristics (expected labor market returns, social composition, and graduation likelihood) are more easily identified than institutional characteristics, allowing precise measurement of choices and utility weights.

Second, admissions to each FOS program are mechanistic, based on academic composite scores (a combination of matriculation GPA and standardized psychometric test scores), making it possible to gauge the extent of academic mismatch at the application stage. Applicants can easily access information on the previous year's admission thresholds on the institution's website (previously provided in packets). This access mitigates the impact of unequal access to information, typically mentioned as a reason for mismatch, and eliminates heterogeneity associated with holistic admissions (Karabel 2006; Warren 2013).

Third, the choice set, consideration set, and preference ranking are easily identifiable in administrative records. Prospective students submit a ranked list of chosen programs (up to four single or double FOS programs). Applications are considered sequentially based on applicants’ ranked preferences. Admission to the first-choice program is evaluated first. The second-choice program is considered only if the applicant is not admitted to the top choice, and so on. Although applicants can apply to FOS programs at multiple institutions, most (72%) apply to only one institution yearly (Israel Central Bureau of Statistics 2019: table 1.2). Thus, application and admission are institution-specific, accurately revealing the choice architecture.

Data

We have access to administrative data for all applicants, admits, and enrolled students at the four most selective Israeli universities³ (out of six universities): Tel Aviv University (TAU), Hebrew University of Jerusalem (HUJI), Ben Gurion University (BGU), and Technion (TECH). The data cover 70% of all university applications for 1999–2008 (Israel Central Bureau of Statistics 2008: table 3), with complete information on applicants’ characteristics and aptitude, including their high school GPA, standardized test scores, academic composite scores (the criteria for admission), and demographics. The data contain full information to assess the choice architecture: the choice set (all available FOS programs each year), admission requirements for these programs, and the number of programs to which applicants can apply; and the consideration set (the shortlisted programs to which applicants actually apply) and the ranking of programs within it.

We aggregate program-specific annual admission requirements, student body composition, and graduation rates. We link administrative information from the Israel Census Bureau on the program's expected labor market payoff (i.e., aggregated earnings and graduates’ employment rates). The resulting dataset provides unparalleled information on applicants’ revealed choices and the alternatives’ characteristics, allowing us to accurately gauge academic mismatch, assess which program characteristics are associated with the choice, and link choices to admission outcomes. We account for variations in choice architecture by institution and time, ensuring that patterns remain robust across choice settings.⁴

We analyze the revealed choices of roughly 211,000 first-time applicants from 1999 to 2008.⁵ On average, applicants apply to 2.7 FOS programs, yielding a person–program file of 550,000 observations.⁶ By considering all the potential program options for each applicant (by year of application and institution), we derive a dataset of 6 million applications for the first-choice program and 17 million for the entire consideration set (all ranked programs to which they applied).

Measures

Applicants’ Academic Qualifications

We use applicants’ academic composite scores, a weighted mean of their matriculation diploma grades (weighted by type and level of courses) and psychometric test scores (equivalent to the SAT/ACT), the sole criteria for admission to university programs. Because the scale of composite scores might differ across universities, we convert each university-year distribution into percentiles to characterize applicants’ academic qualifications.⁷

Applicant–Program Academic Match

We measure applicant–program academic match by the distance between the applicant's academic composite score percentile rank and the FOS program's admission threshold in the previous year (set as the 25th percentile of the program's academic composite score among admitted students the previous year).⁸ A negative value indicates that the applicant's academic composite score is below the threshold for program j, and vice versa. We allow for nonlinearity by collapsing this variable into a set of 20 categorical indicators based on the size of the difference.

Table 1 presents the categories of applicant–program academic match and the mean distance between the applicants’ composite score and the programs’ admission threshold. The largest negative gap (−69 percentile, on average; category 1) captures the boldest “reach” choices. These applicants are underqualified for program j by a large margin. The eighth and ninth categories contain applicants whose academic aptitude nearly or fully matches program j's admission requirements (category 9 is the reference category). The largest positive gap (80 percentiles, on average; category 20) implies overqualification and the safest application choice.⁹ In some specifications, we collapse categories 17–20 because of the small number of applications in this very safe zone.

Consistent with other work in the college choice literature (Alon and DiPrete 2015; Niu and Tienda 2008; Stephenson et al. 2016), we characterize applicant–program academic match in applicants’ (1) choice set, (2) consideration set, and (3) first choice (program ranked first on the application form).¹⁰

Characteristics of Academic Programs

We focus on four program characteristics capturing the key motivations outlined in the literature for academic choices and fleshing out group-based variations in the decision-making process, with the caveat that they do not entirely reflect motivations for college application. Expected employability and expected salary are measured using information from government tax records on graduates’ program- and institution-specific employability (number of employed months) and monthly salary during their first years in the labor market (Israel Central Bureau of Statistics 2012).¹¹Graduation likelihood is measured using the graduation rate in each program (by program, institution, and year) in the previous year obtained from our administrative data. Program composition is measured by the annual share of Arab students in each program in the previous year (by program, institution, and year).

Admission

We evaluate the consequences of disparities in choices for inequality in two admission outcomes. Admission to the university is coded as 1 if applicants were admitted to any programs in their consideration set and 0 otherwise. Admission to the first choice is coded 1 if applicants were admitted to their top-ranked program (or programs if they selected a dual major) and 0 otherwise (even if admitted to lower ranked programs).

Sociodemographic Factors

We stratify the analyses by applicants’ group membership, coded as 1 for Arab and 0 for Jewish. Arabs represent 11% of the applicant pool in our data, lower than their share of the total population (20%). We account for standard sociodemographic variables: gender, age, and the applicant's locality socioeconomic cluster. The latter is a 10-category classification created by the Central Bureau of Statistics and based on several socioeconomic indicators, including the share of unemployed adults, education levels, and average wages in the locality.

Descriptive statistics in Table 2 demonstrate the vast socioeconomic differences between Arab and Jewish applicants’ localities. Arab applicants are, on average, more likely to come from socioeconomically weak localities where only 3% of the adult population earn high wages (11% for Jews), 50% earn minimum wage (38% for Jews), 6% are unemployed (3% for Jews), 35% have a high school diploma (52% for Jews), and only 8% of the 20- to 29-year-old population are college students (17% for Jews). Table 2 also shows applicants’ academic preparation—a common explanation of disparities in college attendance. Jewish applicants have higher average academic composite scores than Arabs (52nd vs. 42nd percentile). This gap reflects the Jewish applicants’ significant advantage in standardized test scores (642 for Jews and 558 for Arabs, on a scale of 400–800), which is not counterbalanced by Arab applicants’ slightly higher GPA in the matriculation diploma (105 and 100 for Arabs and Jews, respectively).¹²

Analytic Strategy

To net out the underlying effect of program size and popularity on the likelihood of choice, our main models suppress the constant term, forcing the coefficients for academic match categories to capture underlying variances in popularity and size.¹³ In the absence of strong theoretical assumptions about the order in which applicants form their consideration set and first choice, we fit this model twice: first to predict programs included in the consideration set, where each program included in the application is assessed relative to all other unselected options, yielding nearly 17 million person–program–institution–year alternatives; and then to predict the first-choice program out of all alternatives, yielding 6 million person–program–institution–year alternatives. To closely observe the choice for each population of applicants, especially given size disparities, we fit all models separately to the populations of Arab and Jewish applicants. We use coarsened exact matching to match populations by academic composite score, institution, and year (Iacus et al. 2012).¹⁴ The models also include the set of adjustment factors described earlier.

To assess the utility weights underlying group disparities in academic mismatch (RQ2), we estimate a series of conditional logit models predicting the applicant–program academic match category of an applicant's chosen program as a function of the program's characteristics (i.e., labor market, academic, and social considerations). Formally, this specification is similar to that in Eq. (1). However, Y_ij is the academic compatibility category of each program (estimated from 1 to 17) relative to the applicant's credentials (e.g., mathematics at TAU in 2003 is category 9 for a strong applicant and category 2 for a weak applicant). The choice is modeled as a function of individual characteristics (x_iβ_j) and vector z_ji containing the characteristics of each alternative j for individual i (i.e., labor market outcomes, social composition, and expected graduation likelihood), along with a corresponding vector α denoting the effect of each characteristic on the odds of choosing a program. This specification enables us to assess which program characteristics are directly associated with variations in academic match. We treat academic match categories (the outcome variables) as nonlinear and independent to avoid imposing parametric assumptions on the distribution of applicant–program match.

Finally, we assess the impact of differences in application choices on the gap in university admission (RQ3). We fit a logistic regression model to applicants’ likelihood of admission (to their top-ranked program and the university) as a function of group affiliation, adjusting for institution and application year. To isolate the effect of application preferences on the admission likelihood from compositional and individual heterogeneity, we fit this specification to a general sample (Model 1) and a sample matched on the applicant–program academic match (Model 2). Differences between Models 1 and 2 in the expected admission gap yield the impact of choice differences on admission likelihood. To contextualize these results, we repeat the analyses with a third sample matched on academic composite scores (Model 3), which yields the effect of differences in academic preparation on admission likelihood.¹⁵ A comparison of Models 2 and 3 yields insights into the effect of choices and composition on inequality in university admission.

Results

RQ1: Group Differences in Academic Mismatch in Application Choices

Does the likelihood of academic mismatch in university application choices differ across segregated groups—in this case, Jewish and Arab applicants? If choice constraints are the main driver of disparities in academic mismatch, group differences will be small in the Israeli case because of the constant choice set across applicants and the minimal cost, geography, and information constraints. Alternatively, if the unequal opportunity structure drives group disparities in utility weights, differences in the academic mismatch rate will be evident even in this context. To evaluate these possibilities, Figure 1 plots the estimated coefficients (in odds ratios) from a series of conditional logit choice models predicting applicants’ consideration set (panel a) and first choice (panel b) as a function of the applicant–program academic match category. The estimated coefficients can be interpreted as the adjusted change in the odds of an applicant selecting major j associated with the program's academic match category relative to the reference category (category 9) representing an academic match.¹⁶

The results are consistent with the utility weights framework. Arab applicants are more likely to include reach programs (categories 3–5) in their consideration set than academically matched programs (category 9) and less likely to apply to safe ones. Jewish applicants are more likely to apply to matched programs (categories 6–8). Similar patterns are evident in applicants’ first-choice programs: Arab applicants are more likely to overreach (categories 1–3) in their first choice, whereas Jewish applicants gravitate toward more academically matched programs (categories 6–8). Sensitivity analyses confirm that these results are robust across universities (Figure A2) and the specification of the admission threshold (Figures A3 and A4) and are not driven by ceiling and floor effects (Figure A5; all figures and tables designated with an “A” are available in the online appendix).¹⁷

Importantly, Arab applicants’ ambitious first choices are not the sole drivers of their higher probability of including reach programs in their consideration set. Figure 2 plots the estimated coefficients from a conditional logit choice model predicting the programs selected in lower ranked options (2nd–4th) in applicants’ consideration set by their academic match category, calculated only for the subset of applicants who applied to a reach program in their first choice. Arab applicants with reach programs as their top choice are more likely than similar Jewish applicants to include reach programs also in their lower ranked options. Jewish applicants are more likely to balance their reach choices with safe ones.

Figures 1 and 2 reveal a key finding: group disparities in application choices resulting in academic mismatch emerge even when often-cited constraints (information, costs, and proximity) are minimal and differences in prior academic qualifications are accounted for. Moreover, although differences in academic mismatch are largest in the first-choice program, they characterize the entire consideration set: Arab applicants are more likely than Jewish applicants to include reach programs in their consideration set, not just as their top choices.

The higher level of academic mismatch in Arabs’ application choices echoes the findings of other studies (e.g., Bowen et al. 2009; Hoxby and Avery 2013). However, unlike studies in the United States showing minorities’ tendency to apply to overly safe programs, we find that Arab applicants are more likely to apply to reach programs. There are several plausible explanations for this disparity. First, research in the United States has focused on applications to institutions, so the alternatives vary in geography, cost, information, financial aid, and admission policies—all of which constrain choices (Dynarski et al. 2021). Moreover, because aiming high in college applications in the U.S. context is considered positive, most studies focus on applications to overly safe rather than reach programs.¹⁸ The few studies considering reach applications suggest that segregated minority applicants are more likely to apply to reach colleges than majority applicants (An 2010; Ciocca Eller and DiPrete 2018; Mullen and Goyette 2019; Tessler 2022), which is consistent with our findings.

Second, and not mutually exclusive, most U.S. studies have analyzed survey data on application and enrollment destinations (where students apply and enroll) (Dillon and Smith 2017; Dynarski et al. 2021; Hoxby and Turner 2013; Mullen and Goyette 2019; Smith et al. 2013), complicating the identification of the choice set (Niu and Tienda 2008). This problem is evident even in studies using administrative data (Black et al. 2020; Cortes and Lincove 2016; Dynarski et al. 2021; Hoxby and Avery 2013), given that the decentralized structure of the U.S. higher education system allows only a partial portrayal of the choice set and the ranking of applications is generally absent. In contrast, the choice set and the ranking of the options are clear in our revealed choice data, and our focus on within-university choices ensures that constraints (costs, geography, financial aid, and information) do not vary across options. It is possible that net of these differences in structure and data, similar dynamics underlie mismatch patterns in both countries. We return to this point in the Discussion and Conclusion section.

RQ2: Group Differences in Utility Weights

What considerations underlie disparities in the likelihood of academic mismatch in choices across segregated groups? Figures 3 and 4 plot coefficients from a series of conditional logit models, showing which program characteristics account for differences in the prevalence and type of academic mismatch. The models’ outcome is the applicant–program match category for all alternatives in the consideration set (Figure 3) and for the first choice (Figure 4). A baseline model without the program characteristics for each choice yields the distribution of academic mismatch for Jewish and Arab applicants (comparable to Figure 1). Subsequent models add each program's characteristics: expected salary, expected employability, graduation likelihood, and program composition. In essence, the estimated coefficients from the adjusted models capture the residual distribution of academic mismatch after the effect of each program's characteristics is netted out. A comparison of these residual distributions with the baseline (without program characteristics) reveals the characteristics shaping the prevalence of academic mismatch.

Three program characteristics elevate the prevalence of reach choices among Arab applicants: expected employability, graduation likelihood, and program composition. Once we account for each of these characteristics, Arab applicants’ predisposition for reach programs drops substantially, and the gap in overmatch mostly disappears or even reverses. Consistent with our expectations, Arab applicants make reach choices because they seek programs that lead to stable employment, increase the chances of attaining a diploma, or provide a welcoming social environment. In contrast, the residual distribution of academic match categories among Jewish applicants barely changes in the adjusted models, suggesting that they prioritize academic match over all other concerns tested.

Expected employability has the most potent effect on the academic mismatch of Arab applicants’ choices. When we net out the weight placed on employability concerns, Jewish applicants are slightly more likely to make reach choices than Arabs. These results, broadly consistent with Halaby's (2003) findings, suggest that securing a job in a segregated and discriminatory labor market is a strategic priority for minorities. This preference is reflected in the high share of Arab students in fields leading to public sector jobs offering full-time, continuous, and long-term employment with predictable work hours (Kalleberg and Hewison 2013), including health care professions, medicine, social work, and education (see Figure A1). Thus, whereas Jewish applicants to elite universities prioritize fit with the program and being admitted, Arab applicants prioritize labor market security. Arab applicants’ reasoning seems sound, given the context of limited labor market opportunities.

Arabs’ higher probability of academic mismatch does not stem from financial considerations. When we consider a program's expected salary, the gap in academic compatibility expands. This expansion suggests that Arabs’ gravitation toward reach choices would be higher if their focal consideration were pecuniary, likely reflecting the relatively lower pay of the public sector jobs they covet and the employment relations such jobs entail.

These results indicate that group disparities in the weight applicants place on a program's characteristics contribute to the gap in academic mismatch, even in settings absent of information, cost, and proximity variations across options and applicants. Patterns of differences in choice architecture suggest that decision-making processes are anchored in the unequal opportunity structure of Arabs and Jews in the highly segregated Israeli context.

RQ3: The Admission Consequences of Disparities in Application Choices

What are the consequences of choice disparities for admission outcomes? Table 3 reports predicted admission probabilities of Arab and Jewish applicants from a series of logit models for overall admission (panel A) and first-choice admission (panel B).¹⁹ The first line of each panel (“unmatched”) shows predicted admission likelihoods from a baseline model that includes indicators for applicants’ group, institution, and year. This model yields an unadjusted gap of 26.1 percentage points in the overall admission rate (64.2% of Jewish applicants admitted vs. 38.1% of Arab applicants; panel A) and 28.1 percentage points in the admission rate to the first choice (45.9% and 17.8% of Jewish and Arab applicants, respectively; panel B).²⁰

Matching Jewish and Arab applicants on applicant–program academic fit reduces the overall admission gap by 41% (from 26.1 to 15.3 percentage points) and the gap for first-choice admission by 46% (from 28.1 to 15.1 percentage points).²¹ Thus, if Arab and Jewish applicants had similar levels of academic mismatch in their choices, but everything else stayed the same (including gaps in academic preparation; see Table 2), nearly half of the admission gap would disappear. Conversely, matching groups on their academic composite scores reduces only a quarter of the gap in overall admission chances (from 26.1 to 19.9 percentage points) and 14% of the gap in first-choice admission chances (from 28.1 to 24.2 percentage points). Thus, if Arab and Jewish applicants had the same distribution of academic preparation, but their choices remained the same, most of the admission gap would remain.²² These disparities in application choices undoubtedly contribute to ongoing inequality in the labor market, given that unadmitted applicants might defer enrollment; opt to study abroad at extreme financial, personal, and social costs; or forgo higher education altogether (Alon 2015; Haberfeld and Cohen 2007; Okun and Friedlander 2005).

Evidently, educational choices are a key mechanism driving variations in educational, employment, and economic outcomes among Arabs and Jews in Israel. Although the motivations underlying Arabs’ university choices are compelling and reasonable for a segregated minority group facing discrimination and restricted employment options, the academic mismatch associated with their choices hinders their degree attainment and inadvertently contributes to the reproduction of inequality. These results capture the importance of choices in stratification processes and show how they contribute to the reproduction of inequality, particularly in the context of segregation.

Discussion and Conclusions

People make both trivial and consequential choices every day, deciding among multiple alternatives: what school to attend, who to mate with and who to marry, where to live or work, what to eat, and what type of health program to follow. These behavioral decisions are a critical link between the social setting and various educational, health, marriage, fertility, and residential outcomes—issues central to demographic research. To understand how these choices underlie outcomes of interest to demographers, we need to study them directly to ascertain how they are shaped and constrained by the structure of opportunity that varies by space, time, gender, class, minority status, and race and ethnicity. Doing so requires juxtaposing the choice architecture of groups, including the options available to individuals in those groups, the short list of alternatives they contemplate, and how they prioritize these choices. Notwithstanding the potential for profound insights into demographic patterns, this crucial element has not yet attracted the attention of population science to any great extent.

We demonstrate this potential by exposing the role of stratified decision-making in generating educational inequality between segregated groups. Using revealed choice data from Israel, we show the unequal opportunity structure of the Jewish majority and Arab minority yields choice variations that underlie further disparities in educational trajectories and outcomes, even if both groups have similar choice options. In Israel, Arab applicants are more likely than Jewish applicants to apply to university programs incompatible with their academic credentials. This difference stems from each group's unique set of considerations, leading to variations in utility weights. As a segregated minority group with restricted labor market opportunities, university-aspiring Arabs prioritize employability and, to a lesser extent, the social environment of their program and graduation chances over academic fit. Yet, as we show, even these compelling considerations can lead to suboptimal outcomes for a disadvantaged minority group. Choice variation is the primary source of the gap between Arab and Jewish applicants in university admission, augmenting already significant differences in university attendance. By revealing that an unequal opportunity structure yields a remarkable group-based choice polarization that ultimately perpetuates stratification, this study extends prior work on the reproduction of inequality.

The theoretical insights from this case are pertinent for understanding patterns and consequences of the segregation of socially constructed groups elsewhere. Stratification systems in which opportunity structures differ sharply between groups—even if traditional groupings of race, ethnicity, immigration, and religion do not apply—produce group variations in choices and outcomes. These findings support recent calls to move beyond predefined ethnoracial categories as the axis for inequality in outcomes and consider segregation as a social phenomenon grounded in a specific ideological context (Hummer 2023; Martinez et al. 2023; Wilkerson 2020), which allows a more constructive theoretical lens through which to view the mechanisms driving inequality. For example, we can use it to frame the preferences of U.S. underprivileged high-achieving students: they choose less selective colleges and trade schools instead of elite institutions emphasizing the value of intellectual exploration and broad education because these students prioritize vocation-oriented, shorter length programs that offer a viable career path without needing to go on to a professional school (and amass additional debt) (Holland and DeLuca 2016).

Importantly, a perspective focused on segregation and decision-making can refine policies and interventions aiming to reduce inequality in educational opportunities. Interventions intended to reduce disparities in college applications focus on providing more information on alternatives, giving application assistance, reducing costs with tax credits, or offering information on financial aid (Bird et al. 2019; Dynarski et al. 2023; Ortagus et al. 2020). Yet, the success of these policies is limited (Bergman et al. 2017; Bettinger et al. 2012; Dynarski et al. 2021; Hoxby and Avery 2013; Oreopoulos and Ford 2019).²³ Our findings suggest that policies must also address the decision-making process and contextual factors underlying choice variations, particularly segregation. A deeper understanding of disadvantaged students’ considerations would complement existing efforts, enabling colleges and policymakers to design more effective solutions. Policies that reduce dropout rates, develop career-relevant programs and stronger industry ties, provide multiyear financial aid, and offer a welcoming social context can be useful in attracting talented, disadvantaged students to selective four-year colleges. Reducing discrimination and labor market barriers across occupations and sectors might further alleviate differences in application behavior.

The theoretical foundation and empirical evidence marshaled here add a fundamental layer to the understanding of stratified decision-making processes by uncovering between-group variations in the considerations (utility weights) underlying the formation of inequality in choices and their ranking. These findings should encourage population scholars to use revealed choice data to uncover stratified decision-making processes in other demographic domains, including health, employment, housing, and fertility. For example, neighborhood choice models have proven useful in explaining the consistently high levels of residential segregation and heterogeneity in individuals’ life chances across neighborhood contexts (e.g., Bruch and Swait 2019; van Ham et al. 2018). Still, analyses of neighborhood choice rely on destination data (where people live and whether they move), overlooking important aspects of the choice architecture. The revealed choices of stayers and movers might shed light on the relatively low compliance with moving-based interventions (Bergman et al. 2019; Darrah and DeLuca 2014; Edin et al. 2012).²⁴ Can the expectation of looming discrimination in a potential destination's local job market or of social isolation in affluent neighborhoods and schools outweigh the quality of the schools and community facilities in this destination? Insights into the decision-making process underlying individuals’ choices can broaden existing theories and help design more effective policies in various domains of interest to demographers.

Acknowledgments

This research was supported by grants 200800120 and 200900169 from the Spencer Foundation and grant 7590 from the Yad Hanadiv Foundation. The content is solely the authors’ responsibility and does not necessarily represent the official views of the foundations.

Notes

References