Moving the Needle for Ever-ELs?: Advanced Math Course Taking and College Enrollment

Abstract

STEM preparation—especially high school math course-taking—is a key predictor of college entrance. Previous research suggests that high school English learners (ELs) not only take fewer advanced math courses but also enroll in college at much lower rates than non-ELs—a group that includes former ELs. In the present study, we alter the analytic lens to examine whether ever-EL status, i.e., ever being identified for and receiving EL services, moderates the relationship between advanced math and college enrollment. Essentially, do ever-EL students experience the same boost to college enrollment from advanced math as their peers? We employ multilevel models to analyze statewide, longitudinal, administrative K–12 and higher education data to examine how ever-EL status and advanced math—and the interaction between the two—predict high school graduation, college application and enrollment, and level of college attended. Results show that both measures are associated with a greater likelihood of graduating from high school, applying to a four-year college, and enrolling in any college. We also find that ever-EL status moderates the relationship between advanced math and college enrollment, with important implications for students’ access to four-year colleges. Ultimately, ever-EL students experience different returns on advanced math relative to never-ELs.

Research on students who are English learner (EL)-identified at the time of study (current ELs) suggests that they have limited access to STEM preparation (National Academies of Science Engineering and Medicine (NASEM), 2018), particularly advanced high school math courses, compared with their non-EL peers (Callahan et al., 2010, 2021; Mosqueda, 2010; Mosqueda & Maldonado, 2013). High school advanced math is a key indicator of college readiness and college entrance (Adelman, 2006; Aughinbaugh, 2012) Although math is often framed as a universal language—a content area that supersedes language and needs no translation—Morita-Mullaney et al. (2021) show how this misconception of a linguistically complex field may feed an EL math gap, as educators pay relatively less attention to EL students’ math preparation and education. Despite evidence that students’ math preparation shapes college-going, to date we know little about how ever-EL status—whether students were ever EL-identified—moderates the relationship between advanced math and college entrance.

In fact, the bulk of the research on ELs’ college-going focuses on high school ELs, omitting former ELs, those who exited EL status prior to high school, from the sample. For instance, compared to non-ELs, high school EL are less likely to take college-preparatory coursework—including advanced math—and less likely to enroll in college after high school (Callahan & Humphries, 2016; Kanno & Kangas, 2014). These studies, however, include prior ELs in the non-EL group. The question thus remains, whether ever-being-EL-identified is associated with disparities in academic exposure via course taking. We suggest that research focused on current-ELs only may alternately miss the achievements and/or struggles of former ELs. By excluding former EL students from empirical focus, research (our prior work included) may inadvertently contribute to a deficit framing of our focal population by overestimating the gap between current ELs and non-ELs (comprised of both never-ELs and former ELs).

In the present study, we expand our empirical lens to include both current (high school) and former ELs in the ever-EL category, comparing them to their peers who have never been identified for or received EL services. Specifically, we use statewide administrative data to examine how ever-EL status and high school advanced math combine to inform students’ college-going outcomes. In our approach, we hope to disentangle whether and how advanced math is associated with ever-EL college-going.

Ever-EL Status, Advanced Math Course Taking, and College Entrance

EL and Ever-EL Status and College Enrollment

Prior research on EL students’ secondary and college outcomes tends to focus on differences between secondary EL students and their non-EL peers, often including former ELs in the comparison group. Studies of this type suggest that secondary EL students graduate high school at lower rates than their non-EL peers (Kanno & Cromley, 2013); nationally, 66% of secondary ELs graduate, compared to 85% of non-ELs (National Center for Education Statistics, 2021). Similarly, high school ELs apply to four-year institutions at lower rates than their non-EL peers, even when they hope to go to college and are qualified to attend (Kanno & Cromley, 2013; Authors, Under Review). While some suggest that that ELs’ discordant behavior may be the result of self-censorship, wherein students elect not to apply to college because they (often erroneously) believe that they will not be admitted (Kanno, 2021; Kanno & Varghese, 2010), others suggest that individual, financial, or school factors may account for some of this dissonance. Our society’s persistent conflation of low socioeconomic status (SES) with limited academic capacity (Achinstein, 2012) only further troubles the question for ever-ELs.

Given lower rates of college application among secondary EL students compared to non-ELs, it is not surprising that extant evidence similarly finds that the former appear less likely to enter college. Estimates from nationally representative data suggest that only 19% of secondary ELs were enrolled in a four-year college two years after graduating from high school (Kanno & Cromley, 2015). When ELs do enter college, they appear more likely to attend two-year colleges. Núñez and Sparks (2012) found that 61% of language-minority (LM) students—identified based on home language, rather than K–12 EL status—who went to college enrolled in a two-year institution, compared to 56% of native English speakers. After controlling for pre-college academic characteristics, the authors found that LM status was no longer predictive of institution type, which suggests that differences in secondary academic preparation drive these disparate outcomes. In this study, we examined college-enrollment outcomes for students who were ever EL-identified during their K–12 careers. Ever-EL students alternately comprise a broader population than the subset of students who are EL-identified during secondary, studied by Kanno and colleagues, and a subset of the LM college-going population at the core of Núñez and Sparks’s inquiry. In addition to ever-ELs’ college enrollment, we also explore how ever-EL status moderates the relationship between advanced math course taking and college enrollment.

College Enrollment and Secondary Math

Secondary academic experiences are arguably the strongest predictors of college entrance (Bowen et al., 2009; Jackson, 2010); among those, math course taking has long been upheld as an indicator of academic merit and rigor (Rose & Betts, 2001; Schiller & Muller, 2003). Using propensity score matching to account for myriad social and academic covariates with state administrative data in Florida, Long and colleagues (Long et al., 2012) identified a positive relationship between rigorous coursework and students’ college-going. Taking advanced math (Algebra II or higher) meant that a student was more likely, by 26 percentage points, to graduate from high school than peers who did not take advanced math (Long et al., 2012). Students from traditionally underrepresented groups, a population that includes EL students, experienced greater returns on advanced course taking than their peers (Long et al., 2012).

Evidence from nationally representative data extends the positive outcomes of advanced math to college enrollment. Using the National Longitudinal Survey of Youth 1997 (NLSY97) data, (Aughinbaugh, 2012) found that students who took advanced math (again Algebra II or higher) were not only 17 percentage points more likely to go to college than their peers who did not, but they were also 20 percentage points more likely to enroll in a four-year college or university, meaning they had the potential to attain a baccalaureate degree. In a study using propensity score matching and data from the Education Longitudinal Survey of 2002 (ELS:2002–2006), Byun et al. (Byun et al., 2015) found that secondary students who took advanced math courses were markedly more likely to enroll in a four-year college after graduation than their peers who did not take advanced math. In this case, the authors’ initial models suggested that students with low SES might benefit more from advanced math course taking; however, results from subsequent matched models found no differences by SES or by race/ethnicity. These mixed findings suggest that there is a need to explore whether advanced math course taking might mitigate disparities in EL college-going as identified in the extant literature (e.g., (Kanno & Cromley, 2013; National Center for Education Statistics, 2021; Núñez & Sparks, 2012). In the following section, we review the research on access to advanced math course taking.

Secondary Math Course Taking: Access Patterns by Race and EL Status

Historically, scholars have documented disparities in access to advanced secondary course taking along racial, ethnic, and socioeconomic lines (Muller et al., 2010; Palardy et al., 2015). For Latinx students in particular, English proficiency level largely determines whether they gain access to advanced math courses (Mosqueda & Maldonado, 2013). Our present study builds on decades of research examining how EL status relates to the quality and quantity of students’ academic content area exposure, particularly via secondary college-preparatory course taking (NASEM, 2018). Being of current-EL status during the secondary grades appears to preclude students’ taking college-preparatory courses (Callahan & Humphries, 2016; Callahan et al., 2010; Thompson, 2015). Umansky (2016) further nuances our understanding of secondary ELs’ limited STEM academic exposure (NASEM, 2018), describing two main mechanisms by which current-ELs are denied full curricular access. Specifically, she identifies leveled tracking, in which they are disproportionately placed in low-level courses, and exclusionary tracking, in which by virtue of their current EL status they are denied access to (i.e., placement in) core academic content (Umansky, 2016). Additional research confirms patterns of both exclusionary and leveled tracking through which secondary ELs are systematically placed into below-grade-level and nonacademic coursework, respectively (Callahan et al., 2021; Estrada, 2014).

Perhaps most interesting in this area, Morita-Mullaney and colleagues (2020) identify a specific type of exclusionary tracking embedded in the infrastructure of a dual-language program designed specifically to support academic equity for ELs. In their study, the authors found that ELs who participated in the dual language bilingual education (DLBE) program had notably less flexibility in their schedules. Specifically, for ELs DLBE program placement ‘crowded out’ their access to math and science electives, placing them instead in Spanish as well as cohort-specific classes with their non-EL DLBE peers. ELs in the DLBE program were unable to access the electives needed to later enroll in advanced high school math and science coursework. Both exclusionary and leveled tracking patterns limit current ELs’ academic preparation; further research is needed to understand to what degree these patterns apply to former ELs.

Notably, EL students appear less likely to take advanced math courses than their peers (Callahan et al., 2021; Thompson, 2017). Using ELS:2002 data, Mosqueda and Maldonado (2013) argued that any attempts to improve linguistic minority students’ mathematics achievement will require improving their access to advanced math course taking. In a subsequent study also using a subset of the ELS:2002 data, Maldonado and colleagues (2018) found that advanced math course taking by minoritized student groups, such as ever-EL students, has the potential to temper existing racial and linguistic gaps in math achievement, offering hope for its potential impact on EL students’ college matriculation. In fact, in a recent report for the Education Trust West, (Ruffalo, 2018) argued that Math in particular has the potential to ameliorate existing educational achievement and attainment gaps related to students’ EL status.

In the present study, we hypothesized that math course taking might provide a tangible lever to improve college entrance among youth who were ever EL-identified during their K–12 careers. To extend the previous literature and understand the relationship between ever-EL status and college-going, we expanded our empirical lens on ELs to include students who, at any point in their K–12 career, were EL-identified (ever-ELs). We then considered whether and how math course taking might shape the relationship between ever-EL status and college-going. Most available postsecondary data lack measures of students’ K–12 EL status (Núñez et al., 2016) leaving the bulk of research on EL college access to focus on secondary EL students, set in contrast to their non-EL peers (Callahan & Humphries, 2016; Kanno & Cromley, 2015). Our present inquiry addresses this gap. Using comprehensive state administrative data that tracks students from kindergarten through college, we explore whether and how advanced math course taking shapes the relationship between ever-EL status and college entrance.

Current Study and Its Contributions

We used statewide longitudinal administrative data from Texas to examine whether and how ever-EL status is associated with postsecondary engagement and the extent to which math course taking closes the gap in college enrollment between ever-EL students and their never-EL peers. Specifically, we posed the following research question:

• 1How do ever-EL students’ background characteristics and academic experiences, including advanced math course taking, differ from those of students who were never identified for EL services (never-ELs)?

We then asked, controlling for variation in student and school characteristics:

• 2How do ever-EL status and advanced math course taking—our two independent variables of interest—shape students’ college-enrollment outcomes (i.e., high school graduation, college application and enrollment, and college type)?
• 3How, if at all, does ever-EL status moderate the relationship between advanced math course taking and students’ college-enrollment outcomes?

We made several important contributions to the literature; first, we used population-level data from one of the largest and most diverse states in the country, Texas, which educates 11% of the nation’s school children, nearly 20% of whom are EL-identified at any given time (National Center for Education Statistics, 2021). Second, we built a unique data set that follows students from their entrance into the K–12 system in kindergarten through high school, and, for those who enroll, college. Finally, these data allow us to capture high school transcript information along with a variety of student and school measures across students’ K–12 educational journey, key among these being ever-EL status. Because of the unique nature of the data, we were able to capture a broader swath of EL-identified students than previous researchers, i.e., students who were ever EL-identified during their K–12 experience.

Methods

To answer our research questions, we used statewide administrative data provided through a restricted-use agreement with the Texas Education Research Center (ERC), a research center and data clearinghouse at the University of Texas. We tracked students from kindergarten into college, with an eye on the factors identified in the literature as levers of postsecondary access. We used descriptive statistics to answer RQ1, focusing on differences in students’ advanced math course taking by ever-EL status. To address RQ2, we developed a series of comprehensive multilevel statistical models to examine predictors of high school graduation (any time prior to 2020, our last year of data), application to four-year colleges, and college enrollment and type. Finally, to address RQ3, we modeled interactions with advanced math course taking to better understand the moderating role of ever-EL status on students’ college-going.

Data

We used student-level K–12 data collected by the Texas Education Agency (TEA) and higher education data collected by the Texas Higher Education Coordinating Board (THECB) to develop a longitudinal data set that follows Texas students from K–12 schooling into college. In contrast to other data that included only current (i.e., secondary) EL status, longitudinal population-level data with recurrent annual indicators of EL identification enabled us to capture students who were EL-identified during their K-12 experiences, but who later exited EL status (former ELs). We can then consider end of high school and beginning postsecondary outcomes for the comprehensive ever-EL population, both former- and secondary EL students.

To capture student experiences in K–12, we used attendance and demographic data, course enrollment and completion records, and test scores from the TEA, along with school characteristics obtained from the Public Education Information Management System (PEIMS). To capture college application and enrollment information, we used student application and college enrollment files obtained from the THECB.

Our analytic sample pooled students who entered kindergarten in Texas in 1999 and 2000 (N = 525,762). We chose to combine two cohorts to maximize our sample size and strengthen the validity of our findings (Adhikari et al., 2021). We restricted our sample to students who were younger than 21 by the end of high school (N = 525,292) and who enrolled in ninth grade in Texas (N = 478,375)¹ in order to ensure adequate background information for statistical controls in our multilevel models (e.g., high school course taking milestones, 9–12^th grade state standardized test scores, and characteristics of last high school attended)—though students who were missing academic information in the years between kindergarten entrance and high school are otherwise still included in the data. We also removed students who transferred to another state before graduation (N = 36,079), as well as those who were missing course-taking data (N = 5,696), testing data (N = 14,491), and high school data (N = 7,481), resulting in a final analytic sample of 414,628² students from the combined cohorts. Given our large sample size, in all analyses, we use a higher threshold for statistical significance for coefficients than is typical in the literature, with an upper bound p-value of 0.01 to signify statistical significance, rather than a p-value of 0.05.

The final analytic dataset represents the population of 1999 and 2000 kindergarten entrants who entered high school in Texas and had high school course and enrollment data. The longitudinal nature of our data allowed us to explore the relationship between ever-EL status and our outcomes of interest: high school graduation and college application and enrollment. In addition, the timing of these two cohorts (i.e., kindergarten entrants in 1999 and 2000) enabled us to follow students for three years after high school graduation. At the time our study began, TEA and THECB data were available through 2020, allowing us to follow the students up to seven years after they were projected to finish high school. Our longitudinal data offer an analytic window similar to that provided by national datasets collected by the National Center for Education Statistics (NCES), that is from high school through college. However, our data offer three distinct advantages: (a) access to population-level data for the entire state of Texas, (b) comprehensive K-12 data on educational enrollment and experiences linked to higher education data, and, most importantly, (c) the ability to identify ever-EL status, which is not possible with existing NCES postsecondary datasets.

Key Analytic Variable

Independent variables of interest.

Our key independent variable of interest is ever-EL status, which captures whether students were identified in kindergarten to receive linguistic support or EL services from a Texas school. We created an indicator of ever-EL status using measures of: (a) students’ home language (non-native English speaker); (b) school system EL-identification;³ and (c) EL program placement (enrollment in a bilingual or English as a second language [ESL] program) obtained from the TEA enrollment files. In short, to determine whether students had experienced EL services in kindergarten, we first identified nonnative English speakers and then used both the EL-identification and EL program placement measures to specify our population of interest. Of the total sample of 414,632 students who entered kindergarten in either 1999 or 2000, nearly one-quarter (23%) were identified as ever-EL; never-EL students comprised the remaining 77%of the population.

Our second independent variable of interest in our models is a measure of advanced math course taking, which we operationalize as a continuous measure that captures the number of math courses a student took beyond Algebra II⁴, motivated by research documenting its relationship with students’ enrollment in a four-year college or university (Adelman, 2006; Aughinbaugh, 2012; Rose & Betts, 2001). In addition to including the continuous measure, we also captured the type of advanced math course students took in our descriptive tables to examine descriptive differences in math course taking by ever-EL status.

Additional student characteristics and academic measures.

We captured measures of student demographics including gender, race/ethnicity, and eligibility for free lunch—a proxy for SES. We included several additional measures of students’ academic exposure and experiences, including indicators of whether a student ever received special education services or was ever retained; the percentage of classes missed in 9th grade (absenteeism); and the number of high schools attended (mobility), courses failed, and dual enrollment (college) courses taken, in addition to students’ 9th grade standardized state test scores in math and reading. These elements serve as additional statistical controls, capturing students’ academic background beyond our focus on math course taking.

School characteristics.

To align our statistical model with the structure of the data, in which students are nested within high schools, we needed to first identify students’ high schools. We used TEA attendance data to identify students’ last known high school. We merged TEA high school ID with data from the PEIMS, which allowed us to capture school-campus-level variables, including the percentage of students on campus who were White, EL-identified, in special education, and taking advanced (i.e., Advanced Placement [AP], International Baccalaureate [IB], or Dual Enrollment [DE] college preparatory) coursework. In addition, we captured indicators of share of teachers with five or fewer years of experience, distance to the nearest four-year institution of higher education, and K–12 charter status.

We focus on five separate outcome measures at two points in the educational pipeline that capture the transition from high school to college: end of high school and three years after graduation. End of high school outcomes of interest include: (a) high school graduation and (b) application to a public four-year university. For those students who graduate from high school, we capture college enrollment within three years of graduation, in: (c) any college, (d) a four-year institution, and (e) a two-year college.

Analytic Strategy

To address RQ1, we relied on descriptive statistics to compare the backgrounds and academic exposure and experiences of ever-EL and never-EL students. In turn, to address RQ2, we employed multilevel mixed-effects logistic regression models to account for the nesting of students in schools and to allow for estimating binary outcomes. Initially, we used the following two-level model to predict our five dichotomous outcomes:

$$\mathit{\text{logit}}\left(Y_{ij}\right)={\beta }_{0j}+{\beta }_{1j}{X}_{1ij}^{*}+{\beta }_{2j}{X}_{2ij}^{*}+\cdots +{\beta }_{kj}{X}_{kij}^{*}+e_{ij}$$

In the equation, the outcome $Y_{ij}$ is a dichotomous indicator of whether student i in high school j achieved a specific educational outcome (e.g., graduating from high school, applying to a university). $X_{1ij}^{*}- X_{kj}^{*}$ are the group-mean centered level 1 predictors for student i in school j, including our key independent variable of interest, ever-EL status; ${\beta }_2-{\beta }_{kj}$ are the slopes for the group-mean centered predictors in school – calculated with the sum of the level 2 fixed effects and school-level variables; and $e_{ij}$ is the residuals.

To address RQ3, we added an interaction between ever-EL status and our key independent variable of interest, advanced math course taking, to examine variation in returns to math course taking by ever-EL status. We used the following equation to model this relationship – wherein ${\beta }_{1j}{X}_{1ij}^{*}*{\beta }_{10j}{X}_{10ij}^{*}$ represents the interaction between ever-EL status (${\beta }_{1j}{X}_{1ij}^{*}$) and advanced math course taking (${\beta }_{10j}{X}_{10ij}^{*}$):

$$\mathit{\text{logit}}\left(Y_{ij}\right)={\beta }_{0j}+{\beta }_{1j}{X}_{1ij}^{*}+{\beta }_{2j}{X}_{2ij}^{*}+\cdots +{\beta }_{kj}{X}_{kij}^{*}+{\beta }_{1j}{X}_{1ij}^{*}*{\beta }_{10j}{X}_{10ij}^{*}+e_{ij}$$

A mixed-effects multilevel model addresses the nested nature of the data, where students are nested in high schools. The model helps us understand how much of the variation in a given outcome is explained by the high school attended, as opposed to the variation explained by individual-level variables (Raudenbush & Bryk, 2002). We first performed an unconditional mean model, an empty model with only school IDs at level 2 and calculated the intraclass correlation coefficient (ICC) to determine the variation of the outcome across campuses. Because we took this step to establish whether multilevel modeling was an appropriate methodological approach, no individual characteristics were included. Results confirmed the need for multilevel modeling; the high school a student last attended explained a considerable amount of the overall variation in the outcomes of interest, notably between 6% and 44% of the variation in outcomes (enrollment in a two-year institution and high school graduation, respectively). These findings indicated a need for multilevel models.

Given that level-1 variables are our primary effects of interest, we chose to group-mean center our individual level variables, thereby removing between-school variance and providing estimates of within-cluster effects (Enders & Tofighi, 2007; Raudenbush, 2009; Yaremych et al., 2023). Group-mean centering also allows us to capture how many advanced math courses students completed beyond the average of their school—probably the most meaningful comparison group as advanced course offerings frequently vary by school context, as schools with higher numbers of economically disadvantaged students tend to offer fewer advanced STEM courses (Garland & Rapaport, 2017). We then ran preliminary analyses to assess model fit, leveraging a host of student- and school-level variables specified in the literature as predictors of our outcomes after first checking for correlation to protect against multicollinearity in our models. Measures of race/ethnicity and home language were highly correlated; 99%⁵ of students who spoke Spanish as a home language also identified as Hispanic/Latino (although the reverse is not true). Given the high correlation between home language and race/ethnicity, we dropped student home language in order to retain the race/ethnicity measure. We provide a table in Appendix A that includes definitions, means, and standard deviations (SDs) for all variables used in our analyses.

Our final multilevel models include student demographic characteristics (e.g., ever-EL status, gender, race/ethnicity) and indicators of academic exposure and experiences (e.g., special education status, grade retention, advanced math and dual enrollment course taking) at level 1 (student). At level 2, models include campus-level variables, such as proportion of EL students, White students, students in AP or IB coursework, and teachers with 5 or fewer years of experience, along with distance to four-year college and an indicator of whether the school is a charter school. We used school IDs to group students by last school attended. Including school measures enabled us to capture variation at the school level at a time in the life course when peers and the school context may matter more than other factors (Muller, 2015).

Limitations

We leverage the rich and informative student and school-level data to build comprehensive statistical models with many covariates, but, as with any study, limitations remain. Perhaps first and foremost is our inability to control accurately on students’ initial English proficiency upon entry into kindergarten. Our models capture ever-EL status and compare ever- and never-ELs; including an indicator of initial English proficiency, while ideal (Cook & Linquanti, 2015; Linquanti & Cook, 2013), would preclude any comparison to never-ELs’ experiences and outcomes (as those students have no initial proficiency measures). In addition, although the state’s guidelines for EL-identification and exit remained relatively constant during our sample’s analytic window (1999–2013)⁶, there may be variation in the rate of exit from EL status that we are unable to capture in our data. We turn now to the question of heterogeneity within the EL population and the plethora of studies describing the depressed academic outcomes of long-term ELs (e.g., Biernacki et al., 2023; Umansky & Avelar, 2023). In an attempt to address this challenge, we performed sensitivity analyses to examine whether the observed differences between ever-ELs and never-ELs might be driven by long-term ELs in the sample (tables available upon request). The results are remarkably similar, almost all coefficients retain the same significance and directionality, and remain for the most part, comparable to the original results. However, we clearly acknowledge that other sources of variance within the ever-EL population certainly remain unaddressed. And, finally, although we focus on advanced math course taking as our primary lever of interest in its relationship with college-going (Adelman, 2006; Aughinbaugh, 2012), we are quite aware that it may in fact be the selection bias inherent in students’ decisions to enroll in advanced math that motivate the observed relationship. As such, we include a variety of measures of course taking beyond advanced math, which should be correlated with their motivation toward college entrance, as an attempt to further isolate the relationship of interest.

Results

Differences by Ever-EL Status

We first examine how ever-EL students’ background characteristics and academic experiences, especially advanced math course taking, differ from those of never-EL students (RQ1). Table 1 presents descriptive statistics capturing how students’ individual background, academic experiences, and high school contexts vary by ever-EL status.

Table 1.

Descriptive Statistics by ever-EL status

	Ever-English Learners		Never-English Learners
	(N = 99,599)		(N = 315,029)
	Mean	SD	Mean	SD
Demographics
Female	0.495	0.500	0.492	0.500
Race/Ethnicity
Hispanic	0.924	0.265	0.302	0.459
White	0.011	0.106	0.498	0.500
Black	0.006	0.076	0.182	0.386
Asian	0.058	0.234	0.015	0.120
Other	0.001	0.024	0.004	0.059
Home Language
English	0.005	0.070	0.957	0.203
Spanish	0.921	0.269	0.033	0.179
Other Language	0.074	0.261	0.010	0.100
Eco. Dis. (Ever Free Lunch)	0.847	0.360	0.571	0.495
Academic Experiences
Participated in Special Education	0.165	0.372	0.215	0.411
Ever Repeated a Grade	0.297	0.457	0.198	0.399
Percent Days Absent in 9th Grade	0.052	0.082	0.052	0.082
Number of High Schools Attended	1.322	0.573	1.322	0.573
Course Taking – The number of:
Courses failed in 9th grade	0.439	1.121	0.273	0.852
Math courses higher than Algebra II	0.733	0.764	0.792	0.791
Dual Enrollment courses	0.768	2.196	0.904	2.194
State Standardized 9th Grade Test Scores
TAKS Math Z-score	−0.018	0.912	0.106	0.962
TAKS Reading Z-score	−0.090	0.866	0.109	0.952
Advanced Math Course Passed
Precalculus	0.435	0.496	0.453	0.498
AP Calculus AB	0.071	0.257	0.080	0.271
AP Calculus BC	0.016	0.127	0.020	0.140
AP Statistics	0.039	0.194	0.051	0.220
IB Math	0.005	0.068	0.005	0.068
Advanced Quantitative	0.055	0.228	0.049	0.215
Other Math	0.097	0.296	0.121	0.326
High School Characteristics
Percent English Learner Students	0.098	0.088	0.042	0.050
Percent White Students	0.154	0.195	0.388	0.265
Percent in Special Education	0.092	0.037	0.097	0.040
Percent in Advanced Coursework	0.300	0.137	0.293	0.129
Percent Beginning Teachers	0.347	0.132	0.312	0.120
Charter Status	0.058	0.234	0.034	0.180
Distance to Four-Year College (Log)	2.316	1.223	2.273	1.119
Outcomes
High School Graduation Rate	0.901	0.299	0.914	0.280
Applied to a Four-year Institutiona	0.217	0.412	0.252	0.434
College Enrollment within Three Yearsa
Any College	0.632	0.482	0.642	0.480
Four-year	0.200	0.400	0.259	0.438
Two-year	0.432	0.495	0.383	0.486

Note. Means and standard deviations calculated using uncentered variables.

^aMeans and standard deviations for college outcomes calculated using subsample of Ever-ELs/Never-ELs who graduated from high school (N = 89,696 and N = 288,037, respectively).

A number of differences emerge across the two student groups by ever-EL status. As shown in the first panel of Table 1, about half of ever-EL students were female, and nearly all identified as Hispanic (92%), with Asians comprising the next largest subgroup, and 92% spoke Spanish at home, reflecting the prevalence of Spanish as a home language in Texas households. To compare, most never-EL students identified as non-Hispanic White (49.8%) or Hispanic (30.2%) and spoke primarily English at home (95.7%). At this point, it is important to note that not all students who speak a language other than English in the home are identified as needing EL services and support upon entry into the K–12 school system (Cook & Linquanti, 2015; Lopez et al., 2016). In addition, whereas the majority of ever-ELs (84.7%) qualified for free or reduced-price lunch for at least one year during their K-12 tenure, just over half (57.1%) of their never-EL peers qualified, suggesting another notable socioeconomic disparity.

We also found variation in students’ academic exposure and experiences by ever-EL status. Overall, a higher share of ever-ELs than of never-ELs repeated at least one grade during K–12 (30% vs. 20%, respectively), but a smaller share experienced special education services (17% vs. 22%). Both groups show similar patterns of absenteeism, each missing on average 5% of school days in ninth grade. Likewise, both groups attended on average 1.3 high schools.

Academic exposure through course placement comes into greater focus at the secondary level. As shown in the second panel of Table 1, ever-ELs completed fewer advanced math and college preparatory courses (including dual credit, AP, and IB) than never-ELs. Placed in lower-level classes, ever-ELs also failed more courses on average than never-ELs (1.25 vs. 0.82) and earned lower average ninth grade reading and math exam scores. In sum, ever-ELs experience less academic rigor than never-ELs and, not surprisingly, demonstrate lower academic outcomes (at least when other demographic characteristics are not controlled for).

Given our STEM focus with an eye to math as key to college entry (Adelman, 2006; Aughinbaugh, 2012), we take a moment to explore the specific advanced math courses taken by high school graduates (i.e., students eligible for college entrance). Here, the reader will note that a smaller share of ever-EL high school graduates takes nearly every type of advanced math from precalculus on, with one notable exception: advanced quantitative reasoning. Ever-EL graduates take this course more frequently than their never-EL peers, a pattern that suggests some substantive difference may exist between this and other advanced math courses. According to the TEA (Texas Education Agency, 2012), advanced quantitative reasoning includes course content focused on “develop[ing] and apply[ing] reasoning, planning, and communication to make decisions and solve problems in applied situations involving numerical reasoning, probability, statistical analysis, finance, mathematical selection, and modeling” (p. 1). Although this course could lead to other advanced math options like AP statistics (at least based on described course content), it does not seem to do so for ever-ELs, who are underrepresented in all other advanced math courses, including AP statistics.

We move now to the high schools our students attend, contexts that have been shown to shape how students think about college and college-going (Jarsky et al., 2009). On average, ever-ELs attended high schools in which one in 10 students were EL-identified, whereas never-ELs attended schools where, on average, less than one in 20 of their peers were EL-identified. Never-ELs also attended high schools with larger White populations, where on average 39% of students identified as White; in contrast, ever-ELs attended high schools where the share of White peers was less than half that (15%). Interestingly, the share of students receiving special education services (9.7% and 9.2%) and enrolled in advanced course taking (29% and 30%) did not differ notably by students’ EL status. That said, ever-ELs attended high schools in which a larger share of the teachers had been teaching for five or fewer years (35% v. 31%), a pattern reflective of previous research finding that novice teacher distribution frequently disadvantages minoritized student populations and hinders their achievement (Clotfelter et al., 2005; Clotfelter et al., 2007). In addition, ever-EL students’ high schools were located farther from four-year institutions on average and were more often charter-identified than the high schools attended by never-ELs.

Now, with a sense of the background, academic experiences, and high school context of ever-EL students, we turn to their early postsecondary outcomes. The bottommost panel of Table 1 displays students’ postsecondary outcomes, beginning with high school graduation, by ever-EL status. Although ever-ELs and never-ELs demonstrate relatively similar high school graduation rates (90% v. 91%), their trajectories quickly diverge. The rate of four-year college application is lower for ever-ELs (22%) than for never-ELs (25%), as is that of college enrollment within three years (63% and 64%, respectively). Here we note that ever-EL high school graduates enroll in two-year colleges at higher rates (43% v. 38%) and four-year colleges at lower rates (20% v. 26%) than never-ELs. These bivariate trends in postsecondary application and enrollment portend very different academic and professional trajectories over time.

Understanding Ever-EL Status as It Relates to College-Enrollment Outcomes

We turn now to our models designed to examine the link between ever-EL status and college enrollment (RQ2). Specifically, we explore how ever-EL status and math course taking—our two independent variables of interest—shape students’ high school graduation, college application and enrollment, and college type, controlling for variation in student and school characteristics. For each outcome, we ran two separate multilevel models; in the first, we included only ever-EL status (level 1) and school ID (level 2), and in the second, we added all remaining student and school variables. Given that all level 1 variables are group-mean centered, student background coefficients should be interpreted as the within-cluster effect—the average change in odds within a school given a one-unit change in the response variable (Raudenbush, 2009; Yaremych et al., 2023).

Graduating from high school and applying to college.

Table 2 shows results from multilevel models predicting students’ end of high school outcomes: high school graduation and application to a four-year college. Results from Table 2, Model 1 show that, within a given school, the odds of high school graduation for both groups do not differ significantly when we do not account for differences in their background characteristics or the schools they attend (OR = 1.020, SE = 0.016, p > 0.01). However, Model 2 results suggest that, after we include student- and school-level variables, ever-ELs’ odds of graduating from high school are 17.2% higher than those of their never-EL peers (OR = 1.172, SE = 0.026, p < 0.0001). Table 2, Columns 3 and 4 show results predicting application to a four-year institution. At the baseline, ever-ELs’ odds of applying to a four-year institution were 28% lower than those of never-ELs (OR = 0.723, SE = 0.007, p < 0.0001); however, after adding student- and school-level characteristics, including advanced math course taking, the estimated effect of ever-EL status was rendered moot (OR = 0.998, SE = 0.013, p > 0.01). In other words, despite ever-ELs’ significantly greater likelihood of graduating from high school, we found no statistical difference between the two groups in the odds of applying to a four-year college. As expected from the literature, students who took advanced math were significantly more likely to graduate from high school and apply to college.

Table 2.

Multilevel Logistic Regression Models Predicting Student High School Graduation and College Application

	High School Graduation		College Application
	Model 1	Model 2	Model 1	Model 2
Student Background
Ever-EL Status	1.020	1.172***	0.723***	0.998
	(0.016)	(0.026)	(0.007)	(0.013)
Female		1.092***		1.389***
		(0.016)		(0.011)
Hispanic		1.422***		0.946**
		(0.034)		(0.012)
Black		1.527***		2.084***
		(0.040)		(0.032)
Asian		0.938		2.123***
		(0.075)		(0.064)
Other Race		0.833		0.833
		(0.106)		(0.064)
Ever Eligible for Free Lunch		0.738***		0.630***
		(0.017)		(0.007)
Academic Experiences
Ever in Special Education		1.373***		0.663***
		(0.023)		(0.007)
Ever Repeated a Grade		0.291***		0.460***
		(0.005)		(0.006)
% Days Absent in 9th Grade		0.001***		0.026***
		(0.000)		(0.003)
# of High Schools Attended		0.760***		0.804***
		(0.009)		(0.008)
# of Courses Failed		0.785***		0.729***
		(0.003)		(0.003)
# of Courses Higher than Algebra II		6.185***		2.208***
		(0.140)		(0.016)
# of Dual Enrollment Courses		1.337***		1.250***
		(0.019)		(0.003)
9th Grade TAKS Test Scores
Math Z-score		0.994		1.275***
		(0.009)		(0.008)
Reading Z-score		1.033**		1.162***
		(0.009)		(0.006)
School Context
% of EL Students in HS		0.214**		0.215***
		(0.075)		(0.075)
% of White Students in HS		1.023		1.147
		(0.153)		(0.138)
% of Students in Special Education in HS		0.001***		0.010***
		(0.000)		(0.004)
% of Students in Advanced Courses Students in HS		528.755***		5.992***
		(86.470)		(0.743)
Charter High School		0.255***		0.364***
		(0.031)		(0.039)
Distance to 4-Year College		1.211***		1.041
		(0.042)		(0.030)
% of Teachers with 5 or Fewer Years of Experience		0.144***		1.112
		(0.020)		(0.133)
Model Statistics
Constant		11.452***		5.673***
Log Likelihood		−74780.8		−197712
Intraclass Correlation		0.440		0.338

N = 414,628

Note. Student background and academic characteristics group-mean centered.

Table presents odds ratios (exponentiated coefficients) with standard errors in parentheses.

^*p<0.01

^**p<0.001

^***p<0.0001

College enrollment within three years of graduation.

In Table 3, we examine predictors of college enrollment among high school graduates within three years of graduation. When considering ever-EL status alone, we found ever-ELs’ odds of any college enrollment to be 8.6% lower than those of never-ELs (OR = 0.914, SE = 0.009, p < 0.0001); however, after accounting for student and school characteristics, ever-EL students’ odds of enrolling in any college are significantly higher than those of never-ELs by nearly 12% (OR = 1.119, SE = 0.013, p < 0.0001). However, those patterns appear to be driven primarily by two-year college enrollment, as shown in the subsequent columns of Table 3. The odds of ever-ELs’ enrolling in a four-year institution are initially 37.8% lower than for never-ELs. After controlling for the full range of student characteristics, including advanced math, the gap in four-year college enrollment is smaller, but the odds of ever-ELs’ enrolling in a four-year institution are still 16.8% lower than those for never-ELs (OR = 0.832, SE = 0.012, p < 0.0001). Our models predicting two-year college enrollment find that, even after controlling for student background characteristics, ever-EL students’ odds of enrolling in a two-year college are 22% higher than for their never-EL peers (OR = 1.215, SE = 0.013, p < 0.0001).

Table 3.

Multilevel Logistic Regression Models Predicting College Enrollment Within Three Years of High School Graduation

	Any College		4-Year College		2-Year College
	Model 1	Model 2	Model 1	Model 2	Model 1	Model 2
Student Background
Ever-EL Status	0.914***	1.119***	0.622***	0.832***	1.284***	1.215***
	(0.009)	(0.013)	(0.007)	(0.012)	(0.012)	(0.013)
Female		1.326***		1.156***		1.161***
		(0.010)		(0.011)		(0.008)
Hispanic		1.135***		0.853***		1.194***
		(0.014)		(0.013)		(0.013)
Black		1.540***		2.065***		0.933***
		(0.022)		(0.036)		(0.013)
Asian		1.791***		1.650***		0.856***
		(0.056)		(0.045)		(0.021)
Other Race		0.746**		0.684**		0.963
		(0.052)		(0.063)		(0.064)
Ever Eligible for Free Lunch		0.635***		0.703***		0.882***
		(0.006)		(0.008)		(0.008)
Academic Experiences
Ever in Special Education		0.736***		0.695***		0.821***
		(0.007)		(0.010)		(0.008)
Ever Repeated a Grade		0.591***		0.433***		0.743***
		(0.006)		(0.008)		(0.008)
% Days Absent in 9th Grade		0.153***		0.012***		0.441***
		(0.015)		(0.002)		(0.039)
# of High Schools Attended		0.846***		0.806***		0.937***
		(0.007)		(0.011)		(0.008)
# of Courses Failed		0.856***		0.722***		0.929***
		(0.003)		(0.005)		(0.003)
# of Courses Higher than Algebra II		1.535***		1.991***		0.812***
		(0.011)		(0.015)		(0.005)
# of Dual Enrollment Courses		1.214***		1.116***		1.009***
		(0.004)		(0.003)		(0.002)
9th Grade TAKS Test Scores
Math Z-score		1.040***		1.468***		0.805***
		(0.006)		(0.010)		(0.004)
Reading Z-score		1.087***		1.188***		0.975***
		(0.005)		(0.007)		(0.004)
School Context
% of EL Students		0.386**		0.302*		1.059
		(0.085)		(0.108)		(0.185)
% of White Students		1.403***		1.366		1.400***
		(0.096)		(0.144)		(0.071)
% of Students in Special Education		0.075***		0.035***		0.405**
		(0.019)		(0.016)		(0.085)
% of Students in Advanced Courses		6.806***		9.472***		1.512***
		(0.605)		(1.165)		(0.103)
Charter Status		0.567***		0.305***		0.757***
		(0.036)		(0.031)		(0.037)
Distance to 4-Year College		1.016		1.016		1.022
		(0.016)		(0.024)		(0.011)
% of Teachers with 5 or Fewer Years Experience		0.827		1.440		0.894
		(0.074)		(0.186)		(0.066)
Model Statistics
Constant		1.568***		2.935***		1.240***
Log Likelihood		−213136.56		−163144.85		−244451.86
Intraclass Correlation		0.144		0.229		0.063

N = 377,733

Note. Student background and academic characteristics group-mean centered.

Table presents odds ratios (exponentiated coefficients) with standard errors in parentheses.

^*p<0.01

^**p<0.001

^***p<0.0001

Our models also suggest a significant positive relationship between each additional advanced math course and any college enrollment and, more specifically, four-year college enrollment. However, on average, each additional advanced math course was negatively related to two-year college enrollment, suggesting that advanced math coursework—at least on average—shapes the type of institution students enroll in. Our next set of analyses enables us to understand whether those relationships vary across ever-EL status.

Advanced Math Course Taking and Ever-EL Status

Given the differences in advanced math course taking by ever-EL status we noted earlier, we move to address RQ3. Results in Table 4 explore how ever-EL status might moderate the existing relationship between students’ advanced math course taking and college enrollment. Models include all variables previously discussed as well as an interaction between advanced math course taking and ever-EL status. For ease of interpretation, we provide Figures 1–3 to graphically illustrate the relationships for the statistically significant interaction terms (all three college enrollment outcomes), showing changes in the predicted probability of the outcomes across number of advanced math courses and ever-EL status⁷. As we group mean centered math course taking in our models, the x-axis can be interpreted as the number of advanced math courses a student completes above the average number taken in her school, with the average represented by “0” in these figures. This is especially noteworthy given the vast differences between the schools ever-ELs attend, on average, and the schools never-ELs attend.

Table 4.

Multilevel Logistic Regression Models Interacting Ever-EL Status and Advanced Math Course Taking

		Enrollment within 3 Years
	High School Graduation	Any College	4-Year College	2-Year College
Student Background
Ever-EL	1.126*	1.126***	0.789***	1.215***
	(0.035)	(0.014)	(0.013)	(0.013)
Female	1.092***	1.326***	1.157***	1.162***
	(0.016)	(0.010)	(0.011)	(0.008)
Hispanic	1.421***	1.136***	0.853***	1.197***
	(0.034)	(0.014)	(0.013)	(0.013)
Black	1.526***	1.539***	2.052***	0.931***
	(0.040)	(0.022)	(0.036)	(0.013)
Asian	0.937	1.777***	1.618***	0.838***
	(0.075)	(0.056)	(0.044)	(0.020)
Other Race	0.833	0.746**	0.686**	0.963
	(0.106)	(0.052)	(0.063)	(0.064)
Ever Eligible for Free Lunch	0.738***	0.635***	0.701***	0.880***
	(0.017)	(0.006)	(0.008)	(0.008)
Academic Experiences
Ever in Special Education	1.373***	0.736***	0.694***	0.821***
	(0.023)	(0.007)	(0.010)	(0.008)
Ever Repeated a Grade	0.291***	0.591***	0.433***	0.744***
	(0.005)	(0.006)	(0.008)	(0.008)
% Days Absent in 9th Grade	0.001***	0.153***	0.012***	0.440***
	(0.000)	(0.015)	(0.002)	(0.039)
# of High Schools Attended	0.760***	0.846***	0.805***	0.936***
	(0.009)	(0.007)	(0.011)	(0.008)
# of Courses Failed	0.785***	0.856***	0.722***	0.930***
	(0.003)	(0.003)	(0.005)	(0.003)
# of Courses Higher than Algebra II	6.181***	1.535***	2.000***	0.811***
	(0.140)	(0.011)	(0.016)	(0.005)
# of Dual Enrollment Courses	1.337***	1.214***	1.116***	1.009***
	(0.019)	(0.004)	(0.003)	(0.002)
Math Z-score in 9th Grade	0.994	1.040***	1.468***	0.806***
	(0.009)	(0.006)	(0.010)	(0.004)
Reading Z-score in 9th Grade	1.033**	1.087***	1.188***	0.975***
	(0.009)	(0.005)	(0.007)	(0.004)
School Context
% of EL Students	0.215**	0.386**	0.301*	1.057
	(0.075)	(0.085)	(0.108)	(0.185)
% of White Students	1.023	1.403***	1.367	1.400***
	(0.153)	(0.096)	(0.144)	(0.071)
% of Students in Special Education	0.001***	0.075***	0.034***	0.403**
	(0.000)	(0.019)	(0.016)	(0.084)
% of Students in Advanced Courses Students	528.490***	6.800***	9.421***	1.509***
	(86.420)	(0.604)	(1.159)	(0.103)
Charter High School	0.255***	0.567***	0.306***	0.756***
	(0.031)	(0.036)	(0.031)	(0.037)
High School Distance to 4-Year College	1.210***	1.016	1.016	1.022
	(0.042)	(0.016)	(0.024)	(0.011)
% of Teachers with 5 or Fewer Years of Experience	0.144***	0.826	1.438	0.893
	(0.020)	(0.074)	(0.186)	(0.066)
Interaction: # Courses Higher than Algebra II *Ever-EL	0.924	1.069**	1.199***	1.129***
	(0.041)	(0.017)	(0.023)	(0.015)
Model Statistics
Constant	11.444***	1.568***	2.936***	1.240***
Log Likelihood	−70588.4	−211405.3	−159350.2	−243876.8
AIC	141230.9	422864.6	318754.4	487807.6

N = 414,628

Note. Student background and academic characteristics group-mean centered.

Table presents odds ratios (exponentiated coefficients) with standard errors in parentheses.

^*p<0.01

^**p<0.001

^***p<0.0001

Figure 1.

Predicted enrollment at any college by ever-EL status.

Figure 1 illustrates group-mean differences across ever-EL status in the change in students’ predicted probabilities of enrolling in any college for each additional advanced math course above the average number taken by other students in the high school, with 0 representing taking the average number as other students, drawing on the statistical model presented in Table 4.

Figure 3.

Predicted enrollment at a two-year college by ever-EL status.

This figure illustrates group-mean differences across ever-EL status in the change in students’ predicted probabilities of enrolling in a two-year college for each additional advanced math course above the average number taken by other students in the high school, with 0 representing taking the average number as other students, drawing on the statistical model presented in Table 4.

In columns 2 through 4 of Table 4, we explore how ever-EL status moderates the relationship between math course taking and college enrollment. Column 2 of Table 4—as shown in Figure 1—presents evidence that advanced math courses can help equalize ever- and never-ELs’ likelihood of college enrollment. While never-ELs have greater likelihood of enrolling when they take the average number of courses higher than Algebra II, the ever-EL likelihood gap closes once they take one or more advanced math courses, and both groups’ likelihood continues to increase with each additional math course. Column 3 and Figure 2 illustrate the moderating effect of ever-EL status for four-year college enrollment (main ever-EL term: OR = .789, SE = .013, p < .0001; interaction term: OR = 1.199, SE = 0.023, p < .0001). At the baseline, ever-EL students are significantly less likely to enroll at a four-year institution than never-EL students. As shown in Figure 2, ever-ELs remain less likely to do so until they take at least three advanced math courses more than their school average—at which point the two groups reach parity. In contrast, Column 4 and Figure 3 show that ever-EL and never-EL students are as likely to enroll in a two-year college, if they take one or two fewer courses than their school average. However, once students take the average number of advanced math courses in their school, the two paths begin to diverge and continue to widen as students take more courses above Algebra II. In other words, taking more advanced math courses appears to decrease both groups’ odds of enrolling in a 2-year college, but more so for never-ELs.

Figure 2.

Predicted enrollment at a four-year college by ever-EL status.

This figure illustrates group-mean differences across ever-EL status in the change in students’ predicted probabilities of enrolling in at a four-year institution for each additional advanced math course above the average number taken by other students in the high school, with 0 representing taking the average number as other students, drawing on the statistical model presented in Table 4.

Discussion and Implications

In some ways, our analyses capturing ever-ELs’ postsecondary outcomes challenge conventional wisdom regarding EL student achievement and attainment (Garcia, 2015). Scholars have long expressed concern about EL students’ high school graduation rate (Sugarman, 2021; Zaff et al., 2021) and, relatedly, their likelihood of dropping out of high school (Callahan, 2013; Slama et al., 2015). By using longitudinal student data, however, we show that the vast majority of ever-ELs do graduate from high school and that they appear to do so at a higher rate than their never-ELs peers. They also, when demographic and academic measures are held constant, appear more likely to enroll in college than their peers, although they prove more likely to attend a two-year rather than a four-year institution.

Descriptively, ever-EL students remain underrepresented in advanced math courses, which positively predict enrolling in college and, particularly, entering a four-year college. Ever-EL students were only overrepresented in one advanced math course: advanced quantitative reasoning. Although advanced quantitative reasoning should help students “develop and apply skills necessary for college, careers, and life,” it may be the case that advanced quantitative reasoning in high school creates a separate track for students that is unlikely to lead to calculus (Schudde & Bernell, 2019; Texas Education Agency, 2012). Morita-Mullaney et al. (2020) argue that courses of this type—that appear on the surface to be advanced, but ultimately hold students back—are “traps” that students must be made aware. This eventuality may indicate that there are qualitative differences between ever- and never-ELs’ advanced math coursework, in addition to the quantitative differences in the number of courses taken. By sorting ever-EL students into separate tracks for advanced math, schools may undermine their ability to get the biggest boost possible—in terms of college entrance qualification—from advanced math, effectively maintaining inequality between ever- and never-ELs (Lucas, 2001). This outcome is especially salient given that our group-centered models demonstrate that even when students take the same number of courses above their school average, ever-EL status still affects the relationship between advanced math and college enrollment.

Our final models, which enabled us to consider the interaction between ever-EL status and number of advanced math courses, suggest that accruing additional advanced math courses can close the gap of college enrollment between never- and ever-ELs. Given the high rates of two-year college entrance among ever-EL students, the interaction between ever-EL status and advanced math course taking was particularly interesting for the four-year college-enrollment outcome. We observed that the predicted probability of enrolling at a four-year institution for ever-EL students reaches parity with that of their never-EL peers only when they take three or more advanced math classes beyond the average of their peers. Thus, although ever-EL status does moderate the relationship between advanced math and four-year college enrollment, it does so ever so slightly and only for a very specific population of students. It is here that we take a moment to reflect on one key additional factor that may influence a student who is otherwise prepared (i.e., has completed advanced math) to opt for two-year rather than a four-year enrollment—financial status. Notably, in the descriptive section, we document significant economic disparities by ever-EL status. One way this difference may play out is through students’ beliefs about whether or not they can afford to attend a four-year institution. Prior research documents how financial insecurity makes starting at a two-year, even after applying to a four-year and likely being accepted, a more tenable option (Acton, 2021). To better understand EL college-going in general, future research will need to disentangle the role of economic status.

On one hand, advanced math appears to act as an important lever, pushing ever-ELs to enroll in a four-year college. On the other, it appears to be a necessary but insufficient means by which to tip the scales and neutralize ever-ELs’ predilection to undermatch (Callahan & Humphries, 2016) by enrolling at two-year institutions when they appear prepared to pursue a bachelor’s degree. For the most part, our findings offer some hope—ever-ELs’ achievement disparities can be ameliorated to some degree by their academic experiences. Ever-EL status does not appear to preclude these students from earning a high school diploma, and advanced math offers a tool that can move the needle ever so slightly on four-year enrollment.

Combined, these trends suggest that ever-ELs academically prepared to attend four-year institutions may end up at public two-year colleges, where those with baccalaureate aspirations may need to navigate byzantine transfer policies to attain a bachelor’s degree (Schudde et al., 2021). Whether and how to guide the large and growing ever-EL population—which comprises nearly one in four Texas public school students—toward baccalaureate-granting institutions remains a challenge to the field. Although advanced math coursework shows promise, the boost it provides to four-year enrollment is driven primarily by the minority of students who complete Calculus and beyond. Fewer than one in 10 high school graduates complete Calculus AB, and fewer than two out of every 100 complete Calculus BC. Math remains linguistically segregated. Notably fewer ever-EL than never-EL students take nearly every advanced math course.

Schools shape the college-going habitus of ever-EL students (Jarsky et al., 2009; Kanno, 2021). Although an examination of the processes that shape ever-ELs’ advanced math access is beyond the scope of this study, our findings suggest a need for greater research on this topic. One area for future research includes how counselors discuss postsecondary options with ever-EL students—especially those who, based on advanced math course taking and college application behavior, appear interested in attending four-year colleges. Research documents educator bias based on English proficiency; Morita-Mullaney and colleagues (2020) found that even when EL students earn high test scores, educators may grade them lower- if not expect less of them- based on their current-EL status, limiting future opportunities. In terms of practice, we hope our results encourage educational leaders to (a) do an equity audit to examine the quality and quantity of their school’s advanced math course taking, including variation by ever-EL status; and to (b) examine advising practices for EL students taking advanced math to ensure they have necessary support to consider all the postsecondary pathways available to them.

Conclusions

Given our results, we urge educators and policymakers alike to carefully consider how to increase ever-EL students’ advanced math course taking—with an eye on math course type—and to contemplate what other factors may preclude ever-ELs’ pursuing a bachelor’s degree. Further research is needed to understand what prevents ever-ELs from engaging with advanced math content at the same rate and in the same ways as never-ELs. Although advanced math course taking does move the needle ever so slightly for high school graduation and college enrollment, it does not offset the patterns of institutional stratification for college goers. Regardless of why this occurs, both secondary and college advisors would do well to keep the potential of a bachelor’s degree in mind when discussing postsecondary options with ever-EL youth.

Appendix A

Variable Names and Descriptions

Variable Name	Description	Mean (SD)
Ever-English Learner	Identified as Limited English Proficient by the Language Proficiency Assessment Committee in Kindergarten; obtained from TEA demographic data	0.24 (0.43)
Demographics
Female	Identified as female; drawn from TEA demographic data, which offers dichotomous measure of gender (male or female)	0.49 (0.50)
Race	Race/ethnicity of the student, obtained from TEA demographic data
White (reference)	Identified as non-Hispanic White	0.38 (0.49)
Hispanic	Identified as Hispanic	0.45 (0.50)
Black	Identified as non-Hispanic Black	0.14 (0.35)
Asian	Identified as Asian	0.03 (0.16)
Other Race	Identified as another race, including Native Hawaiian or Other Pacific Islander, Native American, two or more races, and unknown	<0.01 (0.05)
Economically Disadvantaged	Ever been identified as eligible for free lunch; derived from TEA demographic data	0.64 (0.48)
Academic Measures
Ever in Special Education	Dichotomous indicator of whether a student ever received special education services; generated from TEA enrollment demographic data	0.20 (0.40)
Ever Repeated a Grade	Dichotomous indicator of whether a student ever repeated a grade; generated from TEA attendance data	0.22 (0.42)
Percent Days Absent (9th)	Percentage of days absent in 9th grade; generated from TEA attendance data	0.50 (0.07)
N of High Schools Attended	Cumulative number of high schools attended in 9th – 12th grade; generated from TEA attendance data	1.34 (0.59)
Course Taking: Number of Courses
Number of Courses Failed	Cumulative number of courses failed in 9th – 12th grade; generated from TEA course complete data	0.92 (1.67)
Higher than Algebra II	Cumulative number of courses taken higher than Algebra 2; generated from TEA course complete data	0.78 (0.79)
Dual Enrollment	Cumulative number of dual enrollment courses taken; derived from TEA courses data	0.87 (2.20)
TAKS Math Z-Score	Z-Score of TAKS Math scaled score from 9th grade; generated from TEA testing data	0.08 (0.95)
TAKS Reading Z-Score	Z-Score of TAKS Reading scaled score from 9th grade; generated from TEA testing data	0.06 (0.94)
Campus-Level Variables
Proportion
English Learners	Percentage of English Learners identified in each high school; obtained from public TEA Texas Academic Indicator System	0.06 (0.07)
White Students	Percentage of White students in each high school; obtained from public TEA Texas Academic Indicator System	0.33 (0.27)
Special Education	Percentage of students in Special Education in each high school; obtained from public TEA Texas Academic Indicator System	0.10 (0.04)
Advanced Coursework	Percentage of students in advanced coursework, including dual credit, in each high school; obtained from public TEA Texas Academic Indicator System	0.30 (0.13)
Experienced Teachers	Percentage of teachers in each high school with 5 or fewer years of experience; derived from public TEA Texas Academic Indicator System	0.32 (0.12)
Distance to Nearest Four-Year College	Logged distance in miles from each high school ZIP to the nearest 4-year institution; derived from Texas Education Directory Customized Reports and Data Files and the Integrated Postsecondary Education Data System	2.28 (1.15)
Charter Status	Dichotomous variable indicating if a school is identified as a charter school; derived from TEA campus data	0.04 (0.20)
Outcomes
High School Graduation	Indicates whether the student graduated from high school; derived from TEA graduation data	0.91 (0.29)
Applied to Four-Year Institution	Indicates whether the student applied to a four-year institution; derived from THECB student admissions data and enrollment data	0.24 (0.43)
College Enrollment Within 3 Years of High School Graduation
Any College	Indicates whether the student enrolled in any college within 6 long semesters after high school graduation; derived from THECB enrollment data and TEA graduation data	0.64 (0.48)
Four-Year	Identified as four-year institution; obtained from IPEDS data	0.25 (0.43)
Two-Year	Identified as two-year institution; obtained from IPEDS data	0.39 (0.49)

Note. N = 414,628.