CoChat
Sign Up
Cited by 0Higher Education Research StudiesSchool Choice and PerformanceMedical Education and Admissions

An Omitted Variables Approach to Disentangling Differential Prediction in College Admissions

Open access
Demeke, Saron, Sackett, Paul R., Kuncel, Nathan R., Lee, Samuel D.·2026·Educational Measurement: Issues and Practice
DOI 10.1111/emip.70048
Open this paper in CoChat — free accountOpen PDF

Abstract

Standardized tests used in college admissions have raised fairness concerns, with critics arguing bias against underrepresented groups. Recent research suggests that apparent differential prediction may reflect omitted variables (OVs), which are unmeasured factors that systematically differ across groups and influence performance. We examined 499,622 Black and White SAT takers across 142 institutions to test whether OVs help account for prediction differences in forecasting first‐year GPA using SAT scores, high school GPA, and both. Black students’ college grades were overpredicted by .11–.24 GPA points, while SES showed smaller effects. Adding omitted variables reduced Black overprediction by 54%–72%. High school SES and racial composition showed the largest effects, suggesting that similar admission scores signal different preparation levels depending on school context. Our findings highlight that typically unmeasured differences that shape academic performance, rather than flaws in traditional predictors, account for a substantial portion of differential prediction.

AI snapshot

Written from the paper’s full text.

TL;DR

Using nearly 500,000 Black and White SAT takers across 142 institutions, this study finds that first-year college GPA is overpredicted for Black students and, more modestly, for lower-SES students when admissions predictors are used. Adding omitted variables—especially high school SES and racial composition—substantially reduces the overprediction, suggesting differential prediction partly reflects unmeasured school-context differences rather than test bias.

Key findings

  • Black students’ first-year GPA was overpredicted across models: mean residuals were −.24 GPA points for HSGPA-only, −.16 for SAT, and −.11 for the combined HSGPA+SAT model; the HSGPA-only race coefficient was −.23.
  • Adding all candidate omitted variables reduced Black overprediction by 72% (HSGPA), 54% (SAT), and 72% (combined); the HSGPA residual fell from −.24 to −.07, and the combined-model residual fell to −.03.
  • The largest individual omitted variable for Black–White differential prediction was high school racial composition: it reduced the baseline mean residual from −.24 to −.09 (HSGPA), −.16 to −.07 (SAT), and −.11 to −.04 (combined).
  • SES-based overprediction was smaller (baseline coefficients .04–.11); adding omitted variables reduced Low-SES prediction error by 50%–76%, with high school SES producing the largest individual attenuation. Including SES also modestly reduced Black overprediction (e.g., −.24 to −.18 for HSGPA).

Methods

  • Archival College Board records for freshman cohorts entering college between 2008 and 2012; final sample of 499,622 Black and White SAT takers (Black N = 54,071) from 142 institutions, with multiple imputation across 40 datasets for missing data.
  • Predictors were HSGPA, SAT composite (Critical Reading, Math, Writing summed; 60–240), and a unit-weighted HSGPA+SAT composite; outcome was first-year college GPA recoded above 4.0 to 4.0. Candidate omitted variables included Academic Rigor Index, extracurricular activities, sports, major certainty, self-rated abilities, language, job intentions, degree goal, living arrangement, and high-school-level ARI, SES, and racial composition.
  • Analysis used Cleary-based step-down nested regression models (predictor only, group main effect, group × predictor interaction), with intercept-difference models retained because slope differences were not significant; attenuation was tested by adding all omitted variables simultaneously and each individually, using subgroup mean residuals and dmod effect sizes with cluster-robust standard errors by institution.

Limitations

  • The paper states that the sample was restricted to SAT takers who enrolled in college, that FGPA is an imperfect criterion aggregating heterogeneous courses and grading practices, and that multicollinearity/suppression complicates interpreting individual omitted variables.
  • The paper also states that the 2008–2012 cohorts limit generalizability to current admissions landscapes, including test-optional policies, and that the observational design cannot establish that omitted variables caused the observed differential prediction.
  • Evident from the text: only students self-identifying as Black or White were included, with no multiracial/multiethnic response option, and the authors call for future work on other groups such as Hispanic–White comparisons.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1111/emip.70048

Cites

39 of 39 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

3 of 46 cited sources checked · 0 supported · 1 flagged · accuracy not yet measured

  • Premisecitation

    Most research does not support the claim that test scores and grades are biased against minority individuals.

    “Apart from findings of underprediction for women (Fischer et al., 2013;Keiser et al., 2016), most research does not support the claim that test scores and grades are biased against minority individuals (Culpepper & Davenport, 2009;Mattern & Patterson, 2013).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Culpepper & Davenport, 2009 source

      The Literature Review (page unavailable) reports that prior studies often found minority students’ college grades were overpredicted rather than underpredicted, supporting the narrower point that these predictors did not generally disadvantage minority students in that setting. But the claim extends beyond prediction of college grades to test scores and grades generally; the Discussion (page unavailable) also reports differential prediction for African American students, so the paper does not establish the broader absence of bias. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
    • cites Mattern & Patterson, 2013

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Findingown data

    For race, adding all candidate omitted variables reduced overprediction by 72%, 54%, and 72% across the three baseline predictor systems.

    “For race, adding all candidate omitted variables reduced overprediction by 72%, 54%, and 72%.”
    Quote in the paperSubstantial claim
  • Findingown data

    The measure with the largest individual effect for SES attenuation is high school SES.

    “The measure with the largest individual effect is high school SES (Table 3).”
    Quote in the paperSubstantial claim
  • Findingown data

    For HSGPA, adding all omitted variables reduced Black overprediction from -.24 to -.07 (a 72% reduction).

    “Starting with HS-GPA, the magnitude of prediction error is reduced from -.24 to -.07 (a 72% reduction) (Table 3).”
    Quote in the paperSubstantial claim
  • Findingown data

    High school racial composition also has a smaller but consistent effect across models for SES attenuation.

    “The high school racial composition measure also has a smaller but consistent effect across models (Table S4).”
    Quote in the paperNot a substantial claim
  • May combine two claims. 35
  • Classified as probably not a substantial claim. 25
  • The quote only partly states this claim. 1

  • May combine two claims.

    Premisecitation

    Examinations for differential prediction by socioeconomic status are rare, and the few available studies suggesting overprediction have relied on narrow SES components.

    “While overprediction for Black students has been consistently observed, examinations for differential prediction by socioeconomic status (SES) are rare and the few available studies which suggest overprediction have relied on narrow SES components (Rothstein, 2004;Wittman, 2024).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Rothstein, 2004

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Wittman, 2024 source

      In “Admissions Based on SAT Scores Are Biased in Favor of Students Who Are SED” (page unavailable), Wittman describes a “long and varied empirical literature” using socioeconomic characteristics, contrary to the claim that such examinations are rare. He also says Rothstein used a long list of student and high-school variables, including parental income, rather than only narrow SES components. The paper does report overprediction for disadvantaged students, but its account of the literature contradicts the claim’s main qualifications. (LLM, conf=0.86)

      Checked by a reasoning model — may be wrong.

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    These reductions are consistent with prior work highlighting differences in goals and perceived preparedness across subgroups.

    “consistent with prior work highlighting differences in goals and perceived preparedness across subgroups (Holtzman et al., 2021).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Holtzman et al., 2021 source

      The Diversity in Graduate School Applications section reports subgroup differences in degree aspirations, and the Discussion describes differences in applicants’ choice of major. Neither passage establishes differences in perceived preparedness; a search for preparedness and related terms found no such finding in our copy. (LLM, conf=0.83)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingown data

    Black students reported greater certainty regarding their intended college major (d = -.51) and were more likely to report seeking a doctoral degree (34% vs. 19%).

    “Conversely, Black students reported greater certainty regarding their intended college major (d = -.51) and were more likely to report seeking a doctoral degree (34% vs. 19%) (Table 1).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The reduction in absolute mean residuals is similar for the combined HSGPA and SAT model (mean residual = -.03), whereas the effect for SAT is less pronounced (54% reduction).

    “The reduction in the absolute value of mean residuals is similar for the combined HSGPA and SAT model (mean residual = -.03),whereas the effect for SAT is less pronounced (54% reduction).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The magnitude of prediction error for Low SES students is reduced by 50%-76% across models, with larger effects from the baseline with HSGPA.

    “The magnitude of prediction error for Low SES students is reduced by 50%-76% across models, with larger effects from the baseline with HS-GPA (Table 3; Figure 1).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    High school racial composition (%White) is positively associated with both SES and college grades.

    “High school racial composition (%White) is positively associated with both SES and college grades (Table 2).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    White students were more likely to have attended high schools characterized by greater academic rigor, average SES, and a higher proportion of White students.

    “White students are also more likely to have attended high schools characterized by greater academic rigor, average SES, and a higher proportion of White students (Table 2).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    SES contributes significant variance in predicting FGPA regardless of whether HSGPA, SAT scores, or both are included, with a smaller effect when included alongside SAT.

    “Regardless of whether HSGPA, SAT scores, or both are included, SES contributes significant variance in predicting FGPA, with a smaller effect when included alongside SAT.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Baseline mean residuals indicate overprediction for lower SES, with larger magnitudes for Low SES compared to Middle SES groups.

    “Similarly, baseline mean residuals indicate overprediction, with larger magnitudes for Low SES compared to Middle SES groups (Table 3; Table S4).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    When all potential omitted variables are added, the race coefficient is reduced but remains significant.

    “When all potential omitted variables are added, the coefficient is reduced, although it remains significant (Table S2).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Including school racial composition reduces the baseline mean residual from -.24 to -.09 for the HSGPA-only model, -.16 to -.07 for the SAT model, and -.11 to -.04 for the combined model.

    “Including school racial composition reduces the baseline mean residual from -.24 to -.09 for the HSGPA-only model, -.16 to -.07 for the SAT model, and -.11 to -.04 for the combined model.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the sample, high school racial composition and SES are associated with race (rs = -.49 and -.19) and positively related to FGPA.

    “in our sample, these measures are associated with race (rs = -.49and -.19) and positively related to FGPA (Table 2).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    On average, White students achieved higher HSGPAs, FGPAs, SAT scores, and higher average SES than Black students (d's ranging from .53 for SES to 1.01 for SAT composite scores).

    “On average, White students achieved higher HSGPAs, FGPAs, SAT scores, and have higher average SES (d's ranging from .53 for SES to 1.01 for SAT composite scores) (Table 1).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    High school SES is substantially correlated with individual SES and with higher SAT scores and FGPA, despite a small negative correlation with HSGPA.

    “High school SES is substantially correlated with individual SES and with higher SAT scores and FGPA, despite a small negative correlation with HSGPA.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Black and lower SES students' first-year GPAs were systematically overpredicted when models included only HSGPA or SAT scores, and to a smaller extent when both were included.

    “We found that Black and lower SES students' first-year GPAs were systematically overpredicted when models included only HSGPA or SAT scores, and to a smaller extent when both were included.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Both coefficient estimates and mean residuals were virtually identical to OLS estimates on the same sample, suggesting the primary findings are not attributable to the bounded outcome structure.

    “Both coefficient estimates and mean residuals were virtually identical to OLS estimates on the same sample, suggesting the primary findings are not attributable to the bounded outcome structure (see Table S12).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For Black-White comparisons, the single omitted variable producing the largest attenuation in mean residuals was HS Racial Composition; for Low-High SES comparisons, the largest attenuation was produced by HS SES.

    “For Black-White comparisons, the single omitted variable (OV) producing the largest attenuation in mean residuals was HS Racial Composition; for Low-High SES comparisons, the largest attenuation was produced by HS SES.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Missingness is at random (MAR) conditional on auxiliary variables.

  • May combine two claims.

    Findingown data

    Slope differences tests for each predictor system indicated no significant improvement in model fit by adding interaction terms, which were near zero and insignificant.

    “The slope differences tests for each predictor system indicated no significant improvement in model fit by adding the interaction terms which were near zero and insignificant”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Missingness is at random (MAR) conditional on auxiliary variables.

  • May combine two claims.

    Findingown data

    School SES and racial composition had the largest effect in the HSGPA-only models.

    “These variables had the largest effect in the HSGPA-only models”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationcitation

    Admission based on test scores and/or high school grades alone is not biased against Black applicants under the predictive bias definition; overprediction results in relatively favorable predictions.

    “Admission based on test scores and/or high school grades alone is not biased against Black applicants under the predictive bias definition and, in fact, overprediction results in relatively favorable predictions.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Overprediction is evident even when high school grades are the sole predictor, and is often larger than when test scores are included, indicating that it is not specific to standardized tests.

    “In fact, overprediction is evident even when high school grades are the sole predictor, and is often larger than when test scores are included, indicating that it is not specific to standardized tests (Mattern & Patterson, 2013).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Mattern & Patterson, 2013

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Premisecitation

    The Academic Rigor Index was developed and validated by the College Board to quantify the rigor of students' high school coursework.

    “The ARI was developed and validated by the College Board to quantify the rigor of students' high school coursework (Wiley et al., 2010).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Wiley et al., 2010

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductcitation

    Nested baseline models were compared using a pooled Wald-type test for multiply imputed data alongside changes in R2.

    “Nested baseline models were compared using a pooled Wald-type test for multiply imputed data (Grund et al., 2021) alongside changes in R 2 .”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Grund et al., 2021

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation

    Standardized tests and high school GPA show substantial validity for predicting first-year college GPA and degree completion.

    “These predictors show substantial validity for predicting various relevant outcomes, including first-year college GPA and degree completion (Berry & Sackett, 2009;Kuncel & Hezlett, 2010).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Berry & Sackett, 2009 source
    • cites Kuncel & Hezlett, 2010 source

  • May combine two claims.

    Premisecitation

    Intercept differences do not unequivocally demonstrate test bias; they may arise from omitted variables bias, measurement artifacts, or deficiencies in how the predictor or criterion is defined.

    “However, intercept differences do not unequivocally demonstrate test bias.Instead, they may arise from a range of other explanations, including omitted variables bias (OVB), measurement artifacts, or deficiencies in how the predictor or criterion is defined (Kuncel & Klieger, 2012).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Kuncel & Klieger, 2012 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Some prior work demonstrates the utility of considering multiple social dimensions together when assessing bias.

    “Some prior work demonstrates the utility of considering multiple social dimensions together when assessing bias (e.g., Shewach et al., 2017).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Shewach et al., 2017 source

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Prior work suggests individual and school SES are not merely at different levels of the same construct and may exert independent effects on educational outcomes.

    “Prior work suggests individual and school SES are not merely at different levels of the same construct and may exert inde-pendent effects on educational outcomes (Tan et al., 2025;Zwick & Himelfarb, 2011).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Tan et al., 2025

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Zwick & Himelfarb, 2011 source

  • Probably not a substantial claim: may not pass S3 (load-bearing). The quote only partly states this claim.

    Study design and conductreasoning

    Mean residuals are not strictly comparable across studies given their dependence on sample-specific regression equations and criterion scaling.

    “we caution that mean residuals are not strictly comparable given their dependence on sample-specific regression”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Bias in admissions testing is more appropriately examined through differential prediction, which addresses whether, given the same predictor scores, focal group individuals are predicted to have the same outcome scores as the referent group.

    “Bias in this context is more appropriately examined through differential prediction (or predictive bias), 1 which addresses whether, given the same predictor scores, individuals in a focal group, typically the minority group, are predicted to have the same outcome scores as the referent or majority group (AERA et al., 2014).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites AERA et al., 2014

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    Multiple imputation was employed to address missing data.

    “To address missing data (mean missingness across measures = 11% [SD = 11%, maximum = 38%]), we employed multiple imputation (van Buuren, 2018).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites van Buuren, 2018 source

  • May combine two claims.

    Premisecitation

    Mattern and Patterson (2013) found that regression models including SAT scores and high school GPA significantly overpredicted first-year college GPA for Black students by approximately .18 GPA points, with greater magnitude when using HSGPA alone.

    “In a meta-analysis of nearly 200 university samples, Mattern and Patterson (2013) found that regression models including SAT scores and high school GPA significantly overpredicted first-year college GPA for Black students by approximately .18GPA points, with a greater magnitude of overprediction evident when using HSGPA alone.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Mattern and Patterson (2013)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    All baseline and omitted variables models were fit with cluster-robust standard errors to account for nonindependence of students nested within institutions.

    “All baseline and omitted variables models were fit with cluster-robust standard errors to account for nonindependence of students nested within institutions (McNeish, 2026).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites McNeish, 2026

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    Institution-clustered standard errors were used rather than random-effects multilevel modeling because the interest lies in fixed-effect estimates.

    “We address this through institution-clustered standard errors rather than random-effects multilevel modeling, as our interest lies in the fixed-effect estimates (McNeish, 2026).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites McNeish, 2026

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • The authors state this as an assumption: Missingness is at random (MAR) conditional on auxiliary variables.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Multilevel multiple imputation was applied to account for between-institution effects in the missingness mechanisms.

    “Multilevel multiple imputation was also applied to account for between-institution effects in the missingness mechanisms.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Missingness is at random (MAR) conditional on auxiliary variables.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    Some of these additional determinants may lie at the college level, such as systematic differences in course difficulty.

    “Some of these may lie at the college level, such as systematic differences in course difficulty (Young, 1991).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Young, 1991 source

  • May combine two claims.

    Interpretationcitation

    This sample restriction mirrors the context in which admissions decisions are made, making uncorrected estimates more relevant.

    “While this reduces variability in both predictors and criteria, this mirrors the context in which admissions decisions are made, making uncorrected estimates more relevant (AERA et al., 2014;Kuncel & Klieger, 2012).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites AERA et al., 2014

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Kuncel & Klieger, 2012 source

  • May combine two claims.

    Interpretationcitation

    The limited impact of the Academic Rigor Index may reflect overlap with HSGPA and limitations in how the index is operationalized.

    “Its limited impact may reflect both overlap with HSGPA, which can incorporate weighted grading for honors or advanced courses, and limitations in how the index is operationalized (e.g., assuming equal weights across subject domains) (Shewach et al., 2019).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Shewach et al., 2019

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • The authors state this as an assumption: The Academic Rigor Index assumes equal weights across subject domains.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationnone given

    Considerable variability exists within examined groups, and differential prediction effects should not be interpreted as uniform across all individuals.

    “Moreover, considerable variability exists within examined groups, and differential prediction effects should not be interpreted as uniform across all individuals.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Understanding why differential prediction occurs under traditional admissions systems can provide a baseline against which future work can evaluate whether similar mechanisms operate under test-optional policies.

    “and understanding why differential prediction occurs under traditional admissions systems can provide a baseline against which future work can evaluate whether similar mechanisms operate under test-optional policies.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Interpretationnone given

    School SES and racial composition may proxy for additional unmeasured factors related to college preparation, such as study habits, attitudes, and familiarity with different types of academic evaluation.

    “At the same time, school SES and racial composition may proxy for additional unmeasured factors related to college preparation, such as study habits, attitudes, and familiarity with different types of academic evaluation.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    The omitted variables bias approach examined is not specific to a testing era or policy regime.

    “However, the omitted variables bias approach we examine (e.g., that school-level SES and racial composition contribute to inflated differential prediction effects when omitted) is not specific to a testing era or policy regime”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Study design and conductnone given

    Interpreting the contribution of individual omitted variables is complicated by possible multicollinearity and suppression effects.

    “Third, interpreting the contribution of individual omitted variables is complicated by possible multicollinearity and suppression effects.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    All analyses reestimated with hierarchical linear modeling (with random intercepts) are reported in the Supplementary Materials; results are substantively unchanged.

    “All analyses reestimated with hierarchical linear modeling (with random intercepts) are reported in the Supplementary Materials; results are substantively unchanged.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Missingness is at random (MAR) conditional on auxiliary variables.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    Total N ranges from 482,819 to 499,622 students and 119-142 schools due to varying levels of remaining missingness and differences in within-institution variance for some measures.

    “Total N ranges from 482,819 to 499,622 students and 119-142 schools due to varying levels of remaining missingness and differences in within-institution variance for some measures.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Missingness is at random (MAR) conditional on auxiliary variables.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    Imputation models included a large set of auxiliary variables beyond those used in the primary analyses.

    “Our imputation models included a large set of auxiliary variables beyond those used in the primary analyses (e.g., number of years of individual high school courses).”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Missingness is at random (MAR) conditional on auxiliary variables.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    Correlation and regression estimates were pooled across 40 imputed datasets.

    “Correlation and regression estimates were pooled across 40 imputed datasets.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Missingness is at random (MAR) conditional on auxiliary variables.

  • May combine two claims.

    Interpretationown data

    Black overprediction declined modestly when SES was included, suggesting some shared contextual influences.

    “Further, Black overprediction declined modestly when SES was included, suggesting some shared contextual influences.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductcitation

    Including auxiliary variables related to the incomplete variables or the missingness process can improve imputation quality and increase the plausibility of the MAR assumption.

    “including auxiliary variables related to either the incomplete variables themselves or the missingness process can improve imputation quality and increase the plausibility of the missingness-at-random (MAR) assumption (Enders, 2022;van Buuren, 2018).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Enders, 2022

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites van Buuren, 2018 source
    • The authors state this as an assumption: Missingness is at random (MAR) conditional on auxiliary variables.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Correlations are computed using imputed data pooled across 40 imputations.

    “Correlations are computed using imputed data (pooled across 40 imputations).”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Missingness is at random (MAR) conditional on auxiliary variables.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    All analyses were conducted on the complete non-imputed dataset for comparison.

    “For comparison, all analyses were conducted on the complete non-imputed dataset (N = 3,144; Black N = 345).”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Missingness is at random (MAR) conditional on auxiliary variables.

  • May combine two claims.

    Study design and conductcitation

    The criterion (FGPA) is an imperfect measure of academic success; it aggregates performance across heterogeneous majors, courses, and instructors, reflecting not only students' knowledge and skills but also course composition and grading practices.

    “Second, the criterion is an imperfect measure of academic success; FGPA aggregates performance across heterogeneous majors, courses, and instructors, reflecting not only students' knowledge and skills but also course composition and grading practices (Berry & Sackett, 2009).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Berry & Sackett, 2009 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    This interpretation is broadly consistent with recent arguments that relying on grades alone may increase the influence of differences in high school context and local grading standards on admissions decisions.

    “This interpretation is also broadly consistent with recent arguments regarding the reinstatement of standardized test requirements, namely that relying on grades alone may increase the influence of differences in high school context and local grading standards on admissions decisions (Camara et al., 2026).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Camara et al., 2026 source

  • May combine two claims.

    Interpretationown data

    Degree aspirations, major certainty, and self-reported academic ability appear to operate as proxies for motivational or self-concept factors.

    “These variables appear to operate as proxies for motivational or self-concept factors.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationown data

    The null effects may reflect reduced variability in these indicators within the sample or limitations in how constructs are operationalized, rather than a true absence of explanatory value.

    “This may reflect reduced variability in these indicators within our sample (e.g., 87% reported intentions to work part-time) or limitations in how constructs are operationalized, rather than a true absence of explanatory value.”
    Quote in the paperNot a substantial claim
54 of 54

  • The paper relies on this without support.

    Students from higher SES schools, who are disproportionately White, typically have access to greater instructional resources, teacher capacity, and other advantages.

    “Students from higher SES schools, who are disproportionately White, typically have access to greater instructional resources, teacher capac-”

  • The paper relies on this without support.

    Black students are more likely to come from lower socioeconomic backgrounds than their White peers.

    “Black students are more likely to come from lower socioeconomic backgrounds than their White peers, raising the possibility that SES may account for part of the race-based differential prediction effect”

  • The paper relies on this without support.

    No prior study has systematically examined a broad set of candidate variables together, including individual-level psychological and motivational factors alongside high school-level contextual measures, nor evaluated their joint and independent contributions.

    “First, to our knowledge, no study has systematically examined a broad set of candidate variables together, including individual-level psychological and motivational factors alongside high schoollevel contextual measures, nor evaluated their joint and independent contributions.”

  • The paper relies on this without support.

    No prior study has systematically examined differential prediction on the basis of both race and SES.

    “Second, to our knowledge, no studies have systematically examined differential prediction on the basis of both race and SES.”

  • The authors state this as an assumption.

    Missingness is at random (MAR) conditional on auxiliary variables.

    “increase the plausibility of the missingness-at-random (MAR) assumption”

  • The authors state this as an assumption.

    The Academic Rigor Index assumes equal weights across subject domains.

    “limitations in how the index is operationalized (e.g., assuming equal weights across subject domains)”
6 of 6

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

  • Doctoral degree subgroup counts do not match the overall count

    Minor

    The White and Black doctoral-degree counts sum to 92,518, not the reported overall count of 95,895.

    Reported 95895 · recomputed 92518

1 of 1

Cites

Flagged first, then by citations

Research works printed but not linked: 4

  1. Aboko‐Cole Diakité R., 2015, A comparison of the school experiences of low‐scoring and high‐scoring African American SAT test takers
  2. American Educational Research Association, American Psychological Association, & National Council on Measurement in Education, 2014, Standards for educational and psychological testing
  3. Enders C. K., 2022, Applied missing data analysis
  4. Reardon S. F., 2011, Whither opportunity?, 91

Related

Authors
Demeke, S., Sackett, P. R., Kuncel, N. R., & Lee, S. D.
Affiliations
  • University of Minnesota (US)
  • St. Catherine University (US)
Corresponding author
Demeke, Saron
ORCID
Venue
Educational Measurement: Issues and Practice
Publisher
Wiley
Published
2026-09-29
Volume
45
Issue
4
ISSN
  • 0731-1745 (ISSN-L)
  • 1745-3992
DOI
10.1111/emip.70048
OpenAlex
W7214780106
Semantic Scholar
2283e9c81c653678420473a438add2f079f8d4c9
Cited by
0
References
33
FWCI
0.00
Type
journal article
Language
en
Open access
Licence
Locations
Educational Measurement Issues and Practice (publishedVersion, cc-by)
Funders
University of Minnesota
Topics
  • Social Sciences › Education › Higher Education Research Studies
  • Social Sciences › Education › School Choice and Performance
  • Health Sciences › Medicine › Public Health, Environmental and Occupational Health › Medical Education and Admissions
Keywords
university admission, differential prediction, omitted variables, SAT, standardized testing, racial bias, Black students, first-year GPA, socioeconomic status, high school GPA, school racial composition, underrepresented groups

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-29.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-29.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review