Causal Fairness Analysis of ADHD Status and High School STEM Outcomes

Shuhan (Alice) Ai · 2026

Read Causal Fairness Analysis of ADHD Status and High School STEM Outcomes by Shuhan (Alice) Ai (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-10824.txt.

Introduction

Ai splits the gap in high school STEM results tied to ADHD status into direct, indirect, and spurious parts, using national data from the High School Longitudinal Study of 2009. For overall STEM grades, the ADHD penalty is large: total variation of minus 0.670, with 63.3 percent of it direct, acting outside seen social or school factors. For science identity, the gap is small and not significant, at minus 0.068. The direct effect also shifts by race and ethnicity, and checks show it holds up under modest hidden bias.

Catalog record

CREATORShuhan (Alice) Ai
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44267
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.10824

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Causal Fairness Analysis of ADHD Status and High School STEM Outcomes Shuhan (Alice) Ai Abstract 1 Study Motivation and Research Questions Science, technology, engineering, and mathematics (STEM) careers represent an important pathway for upward mobility among people with disabilities ( Meyer 2017 ) , but the STEM potential of youth with cognitive disabilities is dismissed by the assumption that these students are inherently unable to succeed. This deficit framing is misleading: an Attention-Deficit / Hyperactivity Disorder (ADHD) diagnosis is not based on low IQ ( Matson et al. 2010 ) , and the lower achievement observed among these students reflects not individual factors alone, but the joint product of individual differences and social processes, including stigma, reduced access to advanced coursework, and lower expectations. Beyond achievement, disability labels may also shape how teachers perceive a child’s potential ( Owens 2020 ) and can lead to curricular marginalization ( Shifrer et al. 2013 ) . Moreover, because cognitive disabilities lack objective biomarkers, these labels are disproportionately assigned to youth from lower-SES and racial minority backgrounds ( Saatcioglu & Skrtic 2019 ) . A recent study conducted by Shifrer & Mackin Freeman 2021 found that youth with medicated ADHD, who tend to come from more advantaged backgrounds, show the highest STEM achievement among all disability groups, yet still lag behind non-disabled peers in science identity. However, existing studies rely on descriptive or regression-based methods that cannot unravel the causal mechanisms of these disparities. This study aims to address this gap by applying the Causal Fairness Analysis (CFA) framework ( Plecko & Bareinboim 2024 ) , which is grounded in Pearl’s Structural Causal Model ( Pearl 2009 ; Pearl et al. 2021 ) , to decompose the observed disparity in science identity and high school STEM GPA between students with ADHD and non-ADHD into three components: confounded (spurious), indirect, and direct effects. In doing so, this study provides a causal perspective on how a neurodevelopmental health condition translates into educational inequality. Three research questions guided this study: 1. To what extent does ADHD status contribute to disparities in high school STEM outcomes (science identity and STEM GPA), and through what causal pathways (confounding, indirect, direct) does this disparity operate? 2. How does the total causal effect of ADHD on STEM outcomes vary across student demographic subgroups defined by race/ethnicity, socioeconomic status, and educational expectations? 3. To what extent does the direct causal effect of ADHD on STEM outcomes, after removing the influence of mediating academic pathways, differ across students’ demographic backgrounds? The first question addresses the total variation decomposition, decomposing the overall ADHD outcome gap into its constituent causal channels. The second question investigates heterogeneity in the conditional average treatment effect (CATE) across demographic subgroups. The third question examines heterogeneity in the counterfactual direct effect (ctf-DE), isolating the portion of the ADHD penalty that bypasses the measured mediators. 2 Data and Variables 2.1 Study Design Overview This study employed a six-step analytic pipeline grounded in the CFA framework ( Plecko & Bareinboim 2024 ) to investigate how ADHD status shapes high school STEM outcomes. The full analysis pipeline is presented in Appendix A. Drawing from the High School Longitudinal Study of 2009 (HSLS:09), I constructed two outcome-specific datasets, one for science identity and one for STEM GPA, each organized according to the Standard Fairness Model (SFM) with variables assigned to the protected attribute ( X ), confounders ( Z ), mediators ( W ), and outcomes ( Y ). Missing data ranged from 1% to 20% across variables, with an average rate less than 5%. To address missingness, I applied multiple imputation by chained equations (MICE) with 10 imputations. The analysis proceeded in three phases. First, I provided descriptive context by comparing covariate and mediator distributions between ADHD and non-ADHD students using standardized mean differences (SMD). Second, I addressed the three research questions sequentially: decomposing the total variation in each outcome into confounded, indirect, and direct effects using semiparametric debiased estimation; estimating heterogeneity in the total causal effect across demographic subgroups using generalized random forest; and examining heterogeneity in the direct causal effect using the one-step debiased estimator with cross-fitting. Third, I assessed robustness through confounder sensitivity analysis ( sensemakr ) and propensity score overlap trimming. Methodological details for each step are described in the corresponding analysis sections below. 2.2 Data Source and Sample This study used the HSLS:09, a nationally representative survey conducted by the National Center for Education Statistics (NCES) that followed approximately 23,000 U.S. students from 9th grade through post-secondary education. I drew on three waves: the base year (9th grade, 2009) for baseline covariates, the first follow-up (11th grade, 2012) for mediating variables and the science identity outcome, and the 2013 college update for the STEM GPA outcome. This longitudinal structure ensures that confounders, mediators, and outcomes are temporally ordered in accordance with the causal model. The analytic samples included all students with non-missing ADHD diagnosis status and the relevant outcome. The science identity sample comprised 13,760 students, and the STEM GPA sample comprised 14,600 students. Across both samples, students with ADHD were disproportionately male and White. These demographic patterns were consistent with national estimates that approximately 12% of U.S. adolescents have a cognitive disability ( Office of Special Education Programs 2015 ) . 2.3 Variables As shown in Figure 1 , variables were organized according to the SFM framework ( Plecko & Bareinboim 2024 ) into four groups: protected attribute ( X ), confounders ( Z ), mediators ( W ), and outcomes ( Y ). Full variable definitions and coding details are provided in Appendix B. The protected attribute X was a binary indicator of ADHD diagnosis (1 = diagnosed, 0 = non-ADHD). Confounder block Z captured baseline characteristics not causally influenced by ADHD status: race/ethnicity, sex, family socioeconomic status (SES quintile), 9th-grade science identity, and highest educational expectation. Mediator block W captured 11th-grade STEM learning experiences that ADHD may influence, including students’ enrollment in Algebra II, Geometry, and Pre-calculus, and participation in a science club or study group. The outcome Y was measured in two ways: high school STEM GPA as a performance-based measure, and 11th-grade science identity as an identity-oriented measure. Figure 1: Standard Fairness Model (SFM) 3 Analysis 3.1 Causal Modeling Using the SFM constructed in Figure 1 , I investigated how the observed ADHD-outcome disparity arises through three channels: (1) the direct channel ( X\rightarrow Y ), whereby ADHD status affects STEM outcomes holding all observed confounders and mediators constant; (2) the indirect channel ( X\rightarrow W\rightarrow Y ), capturing the effect transmitted through ADHD’s influence on educational engagement; and (3) the spurious channel ( X\leftarrow Z\rightarrow Y ), capturing non-causal variation due to demographic differences between ADHD and non-ADHD students. The first two channels represent causal mechanisms, while the third reflects confounding—group differences that would exist even in the absence of any causal effect of ADHD. 3.2 Descriptive Balance Diagnostics Before estimating the causal effects, I assessed baseline imbalance between ADHD and non-ADHD students. For each confounder in Z and mediator in W , I computed SMDs between the two groups—the difference in group means divided by the pooled standard deviation for continuous variables, and SMDs on indicator-coded levels for categorical variables.

Read the complete text