Status: Past

Reichlin’s firm, CollegeApp, combined over 200,000 survey responses with commercially-available demographic, financial, employment, and lifestyle data on 241 million US adults.  CollegeApp used machine learning to score each US adult on their intent to enroll in higher education and training; their preference for which type of institution to attend; their preference for instructional delivery mode; and their motivations for enrolling.  Based on this research, Reichlin’s talk explores the geography of who intends to enroll in higher education and training programs.

Despite billions of dollars invested in STEM education and employer claims of shortages of STEM graduates, only about a third of STEM graduates work in STEM jobs. This talk explores the reasons why, and how returns on STEM education can be improved. It offers an overview of Skrentny’s new book, Wasted Education: How We Fail Our Graduates in Science, Technology, Education and Math (Chicago, 2023).

This project utilizes course enrollment data from Cornell University to assess how students from different social groups (self-identified race/ethnicity, gender, first gen, and domestic vs. international status) are segregated across courses, how much of this segregation is tied to college major, and whether students are more likely to swap out of courses where they are social isolates.

Racial and ethnic disparities in educational trajectories and outcomes continue to be central concerns for stratification scholars and policymakers worldwide. A key contributor to these disparities lies in ethnic and racial variations in college application behaviors, which lead to higher rates of academic mismatch among disadvantaged applicants. This paper delves deeper into the role of decision-making processes in generating ethnic and racial disparities in college application choices. We propose that application considerations anchored in an unequal and segregated opportunity structure can generate systematic group differences in college application choices, resulting in suboptimal outcomes for disadvantaged minorities. We evaluate this argument using unique administrative records detailing the revealed choices of Jewish and Arab applicants to universities in Israel, recognizing the high levels of ethnic segregation, education, and labor market stratification in this country. The data and context allow us to pinpoint group differences in decision-making because we can discount costs, geographic proximity, or information constraints—factors often cited as reasons for disparities in application choices. Results from conditional logit (choice) models uncover ethnic differences in how applicants weigh program characteristics. This leads to substantial variation in the rate of academic mismatch and accounts for the bulk of the ethnic gap in university admission. Results demonstrate the importance of decision-making processes in understanding ethnic-racial stratification.

Community college (CC) students who intend to transfer to baccalaureate programs often encounter complex curricular requirements. To navigate them, students activate their social and academic networks in a variety of ways. In this case study of a cohort of CC students in an urban system, we trace the the importance of those we call keystone agents — people in network positions which bridge campus ecologies. We find that keystone agents are important source of information and other supports. We illustrate how keystone agents share information across student networks and how their beliefs about curriculum navigation hold sway over students’ course-taking behaviors, even when these beliefs run counter to the design of guided pathways programs and other local campus-based interventions. Keystone agents’ information sharing aims to create organizational pathways that are intended to reduce friction within CC course sequences, but they also have a series of unintended consequences when students choose to transfer. We offer implications for the development of transfer support programs and interventions, curricular policy-making, and the design of campus environments.

In this update of work presented at the Pathways seminar in February 2023, we will present our approach to creating measures to describe and quantify complexity in major curricular requirements, which may act as a barrier to the students’ ability to navigate college. We discuss our general goals in creating the measures, the past work we draw upon, and our different analytic approaches, which were variably fruitful. We present descriptive results showing our measures of task complexity in major requirements in four departments at each of 32 colleges.

Doing good pathways research requires access to the right kinds of data. For example, studying students’ trajectories through college requires data that is longitudinal and relational; learning from the diverse experiences of students at different types of institutions requires data from multiple colleges; and understanding the long-term value of educational experiences requires data that follows students well beyond college. It is rare for one data source to have all these qualities—plus be accessible to all qualified researchers—which is why we constructed the College and Beyond II (CBII) data. The purpose of CBII is to democratize access to rich institutional data, and in doing so, produce new insights about how undergraduate education works. In this presentation we provide a general overview of CBII that highlights the many data types (e.g., administrative records, transcripts, survey outcomes, written responses) and measures (e.g., validated scales, National Student Clearinghouse enrollment and awards records, AP test scores) that are available. To illustrate the data’s potential, we will highlight preliminary work using the data. The presentation will also be conversational, allowing pathways researchers an opportunity to discuss how the data could be used to answer their own research questions and further their research agendas.

While existing research on academic pathways has typically observed progress via observation of course enrollments and major selection, there are more subtle aspects of students’ everyday experiences that comprise academic progress as well. My research explores the potential of large-scale digital trace data from learning-management systems (such as Canvas) to capture students’ longitudinal patterns of engagement, which is a precondition for development and success in higher education. By examining engagement patterns, I provide a more nuanced and comprehensive picture of student activity and experience, and better understand the development of academic pathways.

Social scientists have long recognized that students’ course grades are consequential for academic progress, yet they have devoted little attention to variation in the protocols through which instructors assign grades. I call these protocols “grading practices.” Their variation may be especially wide in college settings, where instructors often have considerable discretion over grading practices. In some practices, grades are criterion-based, wherein student performance is compared against a set of standards. In other cases, students are compared to the performance of other students in a practice known as curving. Students entering higher education face the challenge of recognizing variation in grading practices and making sense of them under conditions they may perceive as high stakes. I report preliminary findings from a longitudinal study of undergraduates moving through an admissions-selective university to demonstrate the breadth of variation grading practices students encounter. I find substantial variation in how grades are assigned even among courses utilizing curved grades. Perhaps remarkably, initial analyses of qualitative interview data with students in courses with curved grades surface little evidence that grading curves per se engender competition; rather, perceptions of grades in curved courses are highly dependent on course structure and students’ previous exposure to course content.

Postsecondary course-taking is of interest to researchers from diverse domains including economics, sociology, and policy. Transformations in digital infrastructure mean researchers increasingly have access to rich administrative transcripts on course-taking. However, administrative transcripts are seldom standardized across institutions or state systems, preventing researchers from easily examining trends in course-taking and course pathways at scale. To address this challenge, we apply machine learning and natural-language processing techniques to efficiently standardize administrative transcripts at scale. Drawing on four waves of the National Center for Education Statistics’ Postsecondary Education Transcripts Studies, we train logistic regression models to classify courses drawn from administrative transcripts into the College Course Map, a hierarchical taxonomy of course-taking. We apply these models to administrative transcripts from 18 institutions in the College and Beyond II dataset and use the standardized transcript measures to examine longitudinal trends in course-taking in the core liberal arts and professional disciplines from ten years of cohorts of baccalaureate graduates. Contrasting these trends in course-taking with those of majors, we find that the proportion of course enrollments in the core liberal arts is meaningfully higher than that of the proportion of majors in those fields. Examining course-taking trends within major, we descriptively observe that majors in three of the core liberal arts domains – the natural sciences, humanities, and social sciences – take substantially more of their coursework outside of their home discipline but within the liberal arts than majors in the professional disciplines and fine arts.