Longitudinal study finds preparation statistically explains ADHD-linked STEM persistence gap

College students with attention-deficit/hyperactivity disorder were as likely as their peers without ADHD to begin college intending to pursue a science, technology, engineering or mathematics major, but they were less likely to remain in STEM through their senior year. A longitudinal analysis of matched first-year and senior-year surveys found that 38.51% of students with ADHD who initially intended to major in STEM switched to a non-STEM field, compared with 28.34% of students without ADHD. Generalized structural equation modeling indicated that academic preparation and disengaged academic behaviors together accounted for 98.88% of the estimated association between ADHD status and STEM persistence, with preparation contributing the dominant share.

Researchers analyzed de-identified responses from 26,199 students at four-year U.S. institutions who completed the Higher Education Research Institute’s Freshman Survey and College Senior Survey. The matched cohorts completed the freshman surveys in 2010 and 2014 and the corresponding senior surveys in 2014 and 2018. Students had to complete both surveys at the same institution, enabling comparisons between first-year intended majors and senior-year primary majors. The full sample was 64.03% female, 72.06% White, and 70.34% enrolled at four-year colleges. These demographic and institutional distributions limit how broadly the results represent the wider U.S. student population.

Chi-square tests found no statistically significant association between ADHD status and a student’s first-year intended major. By the senior-year survey, however, students with ADHD were significantly underrepresented among STEM majors. The analysis of major trajectories included 16,135 students with complete information: students with ADHD were overrepresented in the transition from an intended STEM major to a non-STEM major and underrepresented among those who stayed in STEM. These results separate initial interest in STEM from the later decision to continue in a STEM primary major, a distinction because a student can leave STEM while remaining enrolled in college.

For the formal persistence model, the researchers created a subset of 4,644 students who reported a STEM primary major in their first year and had senior-year major information. Of them, 3,309 remained in STEM and 1,335 switched to a non-STEM field. The ADHD subgroup contained 148 students: 91 persisted in STEM and 57 switched. Although the overall longitudinal dataset was large, the number of students with ADHD in the modeling subset was comparatively small, limiting precision for subgroup estimates. The 38.51% and 28.34% switching rates therefore describe the analyzed cohorts rather than all students entering STEM with or without ADHD.

The investigators fitted three generalized structural equation models to compare a model containing academic and motivational predictors, one that also included a direct path from ADHD status to persistence, and a fuller model linking ADHD status to academic preparation, academic behaviors and STEM persistence. Academic preparation was a latent construct indicated by high school grade point average, standardized test scores and self-rated mathematical ability. Academic behaviors formed a second latent construct based on reported frequency of arriving late to class, missing class for other reasons and failing to complete homework on time. Drive to achieve was an observed measure of motivation, and primary-major grade point average served as a measure of academic performance. The binary persistence outcome was modeled with a Bernoulli distribution and logit link.

Model 3 produced the most favorable balance of fit and parsimony, with an Akaike Information Criterion of 50,509.17 and a Bayesian Information Criterion of 50,694.90, compared with 52,533.18 and 52,707.15 for Model 1. In this model, ADHD status was associated with lower academic preparation, with an unstandardized coefficient of −0.498, and with more frequent disengaged academic behaviors, with a coefficient of 0.370; both associations had p-values below 0.001. The direct coefficient from ADHD status to STEM persistence was not statistically significant. Academic preparation was positively associated with persistence, with a coefficient of 1.317, whereas disengaged behaviors were negatively associated with persistence, with a coefficient of −0.137.

Mediation analysis quantified the indirect statistical pathways. The association through academic preparation was −0.655, with p < 0.001, while the association through academic behaviors was −0.051, with p < 0.05. Academic preparation accounted for 91.74% of the mediated association, and academic behaviors accounted for 7.14%, together explaining 98.88% of the estimated total association. These percentages describe the proportions assigned by the fitted model; they are not causal effect estimates. The observational design, self-reported measures and potential unmeasured influences prevent the analysis from establishing that inadequate preparation causes students with ADHD to leave STEM.

The result offers a possible explanation for a divergence between early intentions and later academic paths. The study’s framework treats persistence as a dynamic process shaped by academic preparation, engagement with coursework, motivation and academic experience. Students with ADHD may bring creativity, divergent thinking, high energy and sustained interest in open-ended problem solving to STEM. At the same time, ADHD-related difficulties in attention regulation, time management, task initiation and sustained effort can make classroom participation, assignment completion and timed assessment more demanding. The model connects ADHD status with these measured academic pathways rather than treating diagnosis as a direct explanation for switching out of STEM.

Course environments may also shape how those pathways operate, but the study did not directly measure their features. STEM programs can include demanding mathematics sequences, lecture-based instruction, rigid pacing and exam-intensive assessment. The authors note that some students with ADHD may be particularly challenged by these structures, while hands-on and exploratory research environments can draw on reported strengths. Because course structure, pedagogy and assessment design were not modeled as measured variables, the findings cannot show that a particular instructional format caused major switching. Instead, the results identify academic factors for further investigation and indicate where additional research could connect student pathways to specific learning-environment features.

Several limitations also narrow the conclusions. The dataset excluded transfer students and students who left their institution, so the outcome applies to students who remained at the same institution and completed the relevant surveys, not to all students who entered STEM. Senior-year major information was missing for approximately one-third of respondents, and complete-case chi-square results may be biased if missingness was systematic. ADHD diagnoses, grades and behaviors were self-reported; students at large research universities and students from underrepresented racial and ethnic backgrounds were underrepresented, while female students were overrepresented. The model omitted race, ethnicity, gender and institution to preserve degrees of freedom and statistical power for the 148-student ADHD subgroup.

Measurement limitations affected the interpretation of additional model components. Motivation was represented by a single item, which may introduce measurement error. Primary-major grade point average was also difficult to interpret because students who switched fields reported grades in their new major, and grading scales differ across disciplines; its negative coefficient may therefore reflect major switching or cross-disciplinary grading rather than a straightforward achievement effect. The generalized structural equation modeling approach accommodated binary and ordinal outcomes but provided relative model comparison rather than conventional absolute fit indices. A split-sample analysis produced similar fit statistics, suggesting the model was not highly sensitive to random sampling variability. Taken together, the evidence supports a focused conclusion: among students initially intending to study STEM, measured academic preparation was the dominant statistical explanation for the lower persistence observed among students with ADHD, while disengaged academic behaviors made a smaller contribution.

Subject of Research: Longitudinal analysis of STEM persistence and academic pathways among U.S. college students with ADHD

Article Title: Modeling STEM persistence in college: the mediating roles of academic preparation and academic behaviors for students with attention-deficit/hyperactivity disorder

Article References: Li, X., Oquendo-Colón, N., Finelli, C. J., & Carroll, L. J. (2026). Modeling STEM persistence in college: the mediating roles of academic preparation and academic behaviors for students with attention-deficit/hyperactivity disorder. International Journal of STEM Education, 13(1), Article 64. https://doi.org/10.1186/s40594-026-00639-6

Image Credits: AI Generated

DOI: 10.1186/s40594-026-00639-6

Keywords: ADHD, STEM persistence, academic preparation, academic behaviors, college students, major switching, longitudinal survey, structural equation modeling, higher education, academic engagement, Modeling, STEM