Early-life Exposure to Punitive School Discipline and Midlife Cognitive Outcomes

Abstract

Introduction:

Suspension and expulsion are common in US schools and disproportionately target structurally marginalized children. No research has examined whether these punitive practices may have long-term cognitive-aging implications.

Methods:

In the prospective National Longitudinal Survey of Youth 1979 data (N = 8021), we used confounder-adjusted linear models to investigate associations between early-life suspension or expulsion and global cognition, memory, and attention z-scores at age 50. Using interaction terms, we tested for additive scale effect modification by race, gender, and at their intersections.

Results:

In all, 21.5% of participants had been suspended and 4.4% expelled, with Black men and women overrepresented among these early-life experiences. Punitive school discipline was associated with lower midlife global cognition, memory, and attention z-scores (eg, global cognition, suspension:β: −0.21; 95% CI: −0.26, −0.15; expulsion: β:−0.27; 95% CI: −0.38, −0.16). The expulsion-cognition association was stronger for men than women.

Discussion:

Punitive school discipline—an exposure modifiable at multiple policy levels—is associated with lower cognitive performance in midlife, decades after school completion.

Dementia is characterized by cognitive decline with a loss of functional independence following various pathophysiological processes, including Alzheimer disease.^1,2 Experienced by 15% of US adults aged 68 years or older, dementia prevalence is inequitably distributed in the population, with a greater burden among racially minoritized older adults.^1,3,4 Projections into 2060 suggest that Black and Latine/x adults will continue to bear a disproportionate burden of dementia diagnoses, overall and at younger ages of onset.⁵ Notably, underlying disease processes may start decades before diagnosis, indicating modifiable exposures earlier in the lifecourse may prevent or mitigate disease progression.³ Hence, urgent calls have been put forth to document potential modifiable population-level drivers of dementia and dementia inequity.^4,6

Socially patterned experiences in childhood and adolescence, like exposure to adversity, likely contribute to inequity in cognitive aging.⁷ Researchers posit that direct and indirect mechanisms may underlie such associations. On one pathway, adverse exposures during sensitive developmental stages may serve as a stressor, triggering repeated stress responses resulting in physiological dysregulation and accelerated aging.^8–11 On another pathway, early-life adversity may shift lifecourse trajectories toward other known risk factors for dementia, including lower educational attainment.^2,3,10,12 Few studies, however, have examined the effects of early-life school-based adversity, which may have specific implications for such risk factors over time.

Punitive, exclusionary school discipline, which removes students from their classrooms via retributive processes like suspension or expulsion, is a prevalent experience in the United States (US) that may influence lifecourse health.^9,13 Whereas restorative or transformative approaches focus on repairing harm and addressing root causes, punitive school discipline is often justified as a “zero tolerance” consequence for, and therefore deterrent of, student misbehavior.^13,14 Mounting evidence, however, suggests it largely fails to achieve its prevention goals, instead perpetuating interpersonal and structural harms.^9,14–16 Indeed, it has been conceptualized as a mechanism of structural racism disproportionately shaping the educational experiences of racially minoritized children in US schools.^{9,13,15,17–19} As early as preschool, Black students in particular are monitored more closely for perceived misbehavior,^20,21 targeted more often for discipline,^14,16 and punished more severely^14,16 than White students, with inequities persisting across grade levels, independent of student behavior.^14,20 Studies document how Black boys bear the greatest burden of these practices.¹⁶ However, emerging research centering Black girls similarly finds higher rates of punitive school discipline (eg, compared with White boys and girls).^9,15–18 Disproportionality in application is compounded by inequity in short-term and long-term consequences. For example, racially minoritized youth are more likely than White youth to be placed in alternative education programs or pushed out of school into pathways of confinement (eg, “school-to-prison pipeline”), within which they are more harshly adjudicated.²² To date, punitive school discipline has been linked to several adverse health outcomes in adolescence, including initiation of tobacco use and depressive symptoms.⁹ Though limited, evidence suggests health implications could extend into older ages.⁹

We evaluate if punitive school discipline—a modifiable, school-based adversity—is associated with midlife cognition, a predictor of later-life dementia risk.²³ We then test for effect modification by race and ethnicity, gender, and their intersections. We hypothesize that compared with no punitive school discipline, experiencing early-life suspension or expulsion will be associated with poorer midlife cognitive outcomes. Further, owing to structural racism’s role in shaping the distribution and compounding consequences of punitive school discipline,⁹ we anticipate associations will be stronger among racially minoritized students.

METHODS

Sample

We used National Longitudinal Survey of Youth 1979 cohort (NLSY79) data. In 1979, 12,686 participants aged 14 to 22 years were recruited and prospectively followed on an annual basis through 1994, then biennially thereafter. Eligible participants for our analysis comprised those invited to complete the midlife “Cognition Module” between 2006, when first administered, and 2016 (N = 8021; assessed once per participant during that time period).²⁴ Missing data were imputed, as detailed below.

Exposure

School discipline in childhood and adolescence was assessed in the 1980 survey, with respondents queried on experiences of suspension and, separately, expulsion (Have you ever been suspended? Have you ever been expelled? Yes/no). We operationalized the exposure as a 3-level, ordinal variable: no punitive discipline (“no” to both; referent); suspension (“yes” to suspension, “no” to expulsion); and expulsion (“yes” to expulsion, regardless of suspension). Among eligible participants, 3.2% (n = 257) were missing exposure data.

Outcomes

The primary outcome was global cognition, and we performed exploratory analyses of memory and attention separately to assess whether punitive school discipline has cognitive domain-specific implications. Outcome data, assessed via the Cognition Module, comprised questions from the modified Telephone Interview for Cognitive Status.²⁴ Participants become eligible for the module in the survey year during which they turn at least 48 years (mean age: 48.9 y; age range: 46 to 59 y).²⁴ Outcomes were operationalized as follows:

Memory (Immediate/Delayed Word Recall)

Immediate word recall was assessed at the start of the cognition battery by presenting participants with 10 words, then summing the number of words correctly remembered.²⁴ Following a delay, during which participants completed 3 interim cognition tests, delayed word recall was assessed by summing the number of those original words correctly remembered.²⁴ For both tests, higher scores (range: 0 to 10) indicate better cognitive performance. Among eligible participants, 4.8% (N = 387) were missing immediate word recall. Largely owing to a 2008 survey malfunction that resulted in no respondents being administered the second set of recall questions, 32.6% (N = 2617) were missing delayed word recall.²⁴ For the complete case analysis, if a participant was missing one measure and not the other, we considered the data missing, resulting in n = 5353 eligible responses for the memory domain (66.7% of eligible participants). For the imputed analysis, we imputed missing values before combining the measures into a single memory score. To create the memory score, we averaged and then z-scored the immediate and delayed recall measures.

Attention (Serial 7 Subtraction/Backwards Counting)

For the serial 7 subtraction test, participants repeatedly subtracted 7 starting from 100 over 5 intervals (eg, 93, 86, 79, 72, 65); possible scores ranged from 0 to 5 with higher scores reflecting a greater number of correct subtractions.²⁴ There were 2 backwards counting tests: one where participants counted backward from 20 and another from 86; we dichotomized scores from each test (0 = incorrect on first attempt; 1 = correct on first attempt), with higher scores indicating better cognitive performance.²⁴ Among eligible participants, 16.5% (N = 1324) were missing serial 7 subtraction, 1.5% (N = 120) missing backwards-counting-from-20, and 1.8% (N = 147) missing backwards-counting-from-86. Higher missingness for the serial 7 subtraction test was largely due to participants discontinuing the assessment before completion. For the complete case analysis, if a participant was missing at least one measure but had others, we considered the data missing resulting in n = 6680 eligible responses for the attention domain (83.3% of eligible participants). For the imputed analysis, we imputed missing values before combining the measures into a single attention score. To create the attention score, we z-scored each indicator, averaged them, and then z-scored the resulting composite.

Global Cognition Measure

To operationalize global cognition, we averaged and then z-scored the memory and attention z-scores.² For the complete case analysis, this resulted in n = 4584 eligible responses for the overall global cognition measure (57.1% of eligible participants).

See eMethods for outcome cleaning instructions specific to NLSY79.

Effect Modifiers

We tested for effect modification by gender [an indicator of the socially stratifying effects of gender as a social construct; assessed during 1979 household screening; interviewer assigned: female (referent)/male] and, separately, by race and ethnicity [an indicator of the socially stratifying effects of systemic racism²⁵; assessed during 1979 household screening interviews; self-report: “from this listing of more than 20 categories, what is/are your racial/ethnic origins?”; re-coded: Black, Latine/x (NLSY collects these data using the response option “Hispanic,” which includes Mexican American, Chicano, Mexican, Mexicano, Cuban, Cubano, Puerto Rican, Puertorriqueño, Boricua/Boriqua/Boriken, Latino, Other Latin American, Hispano, or Spanish descent),^26,27 White (referent), Other/Missing]. Due to small numbers and measurement concerns (e.g., Counts for participants who identified as “Indian American or Native American are unusually high. About 5 percent of respondents reported this racial/ethnic origin, compared to Census estimates of approximately 0.5 percent of the population. This may have resulted from some respondents’ misinterpretation of the term Native American”)²⁶ the “other race or missing” group comprised participants who reported one of the remaining racial identities (Asian, Hawaiian or Pacific Islander, American Indian), more than one race, or for whom race or ethnicity data were missing; given the heterogeneity of experiences therein, we do not present analysis results for this group. We also tested for effect modification by race, ethnicity, and gender together by generating an intersectional variable [Black women, Black men, Latine/x women, Latine/x men, White women (referent), White men].

Confounders

We adjusted for birth year (range: 1957 to 1964, centered at 1960), birth in a southern state (yes/no),^28,29 county-level rural residence at age 14 (yes/no),^13,30 birth in a country other than the United States (yes/no),^31,32 race, ethnicity,^3,25,33 and gender^33,34 (see the Effect modifiers section for conceptualization/operationalization), and measures of childhood socioeconomic status (cSES).^13,35,36 We operationalized cSES using: each parent/caregiver’s education (centered at 12 y), a quadratic term for parent/caregiver’s education, and whether each parent/caregiver worked for pay [did not work (referent)/worked in low Duncan Index occupation/worked in high Duncan Index occupation, dichotomized at 45 using 2-digit DI score]³⁷ (see Supplemental Material eTable 1 for proportion missing by covariate, Supplemental Digital Content 1, http://links.lww.com/WAD/A530).

Main Statistical Analysis

We accounted for missing exposure, covariate, and outcome data via multiple imputation using all variables considered in main and sensitivity analyses with 200 imputed data sets (R package: Multiple Imputation by Chained Equations, MICE), which produces unbiased estimates under a missing at random (MAR) assumption.³⁸ To estimate the exposure-outcome association, we used confounder-adjusted linear regression models then tested for additive scale effect modification using interaction terms (P < 0.05). To calculate standard errors, analyses were performed separately in each imputed data set and resulting estimates pooled into single parameters.³⁸ We used Stata statistical software version 17 (StataCorp. 2022. Stata Statistical Software: Release 17. College Station, TX: StataCorp LLC.) for analyses with code reviewed by an independent analyst.

Robustness Checks

We adjusted for additional potential confounders of the primary association that were measured at baseline for which temporal ordering with the exposure was unclear. We ran 3 models, serially adding: (1) parental/caregiver death (each parent/caregiver; yes/no/don’t know); (2) school-based behaviors (attitude toward current K-12 school: dissatisfied/satisfied; skipped school day in past year: never/ever; got into a fight at school or work in past year: never/ever) and substance use indicators (alcohol use before age 18: yes/no; past year marijuana use: never/ever; and past year other drug use: never/ever); and (3) cognitive test performance in early life [Armed Forces Qualification Test (AFQT) score].^2,12,16 Measured between 1979 and 1982, we consider these in sensitivity analyses because of the ambiguity around whether they are confounders (temporally precede the exposure) or potential mediators (temporally follow the exposure).³⁹ Lastly, to evaluate our assumptions around the missing data (eg, delayed word recall are largely missing completely at random, serial 7 subtractions are missing at random) we performed a complete case analysis for comparison with the imputed analysis.

RESULTS

Sample Characteristics

The mean birth year among participants was 1961, with 37.9% of births in the US South and 6.9% outside of the United States. At baseline, 20.3% resided in a rural county. The mean age at recruitment was 17.6 years (SD: 2.25). Just over half of the participants were women (50.9%). Participants primarily identified as White (48.8%), Black (30.1%), or Latine/x (16.5%). Nearly 1 in 4 participants (21.5%) experienced suspension, while 4.4% experienced expulsion. Key demographic characteristics varied by exposure (Table 1; eTable 2, Supplemental Digital Content 1, http://links.lww.com/WAD/A530).

TABLE 1.

Demographic Characteristics of Study Participants, Total, and Stratified by Punitive School Discipline Exposure

	Total sample (N = 8021)	No suspension or expulsion (n = 5941)	Suspension only (n = 1728)	Expulsion with or without suspension (n = 352)
Outcomes
Global cognition z-score (mean ± SD)	0.00 ± 1.00 (range: −5.17 to 2.02)	0.10 ± 0.95	−0.25 ± 1.05	−0.41 ± 1.17
Memory z-score (mean ± SD)	0.00 ± 1.00 (range: −3.02 to 2.48)	0.08 ± 0.99	−0.21 ± 1.00	−0.30 ± 1.06
Attention z-score (mean ± SD)	0.00 ± 1.00 (range: −5.21 to 0.71)	0.08 ± 0.93	−0.19 ± 1.12	−0.35 ± 1.24
Confounders
Birth year (mean ± SD)	1961 ± 2.20	1961 ± 2.24	1961 ± 2.15	1960 ± 2.17
Women	50.9%	55.9%	38.7%	27.8%
Race*
White, non-Latine/x	48.8%	52.8%	38.5%	33.1%
Black, non-Latine/x	30.1%	24.8%	44.2%	50.6%
Latine/x	16.5%	17.4%	14.2%	12.8%
Other or missing race	4.6%	5.0%	3.1%	3.5%
Race and gender*
White men	24.0%	23.7%	25.2%	23.6%
White women	24.8%	29.0%	13.4%	9.5%
Black men	14.6%	10.2%	25.3%	36.8%
Black women	15.4%	14.5%	18.9%	13.8%
Latine/x men	8.1%	7.7%	8.9%	8.7%
Latine/x women	8.5%	9.7%	5.4%	4.2%
Men identifying as other race or missing	2.4%	2.4%	1.9%	3.2%
Women identifying as other race or missing	2.2%	2.6%	1.1%	0.3%
Southern birth	37.9%	36.1%	41.5%	51.4%
Born outside US	6.9%	7.8%	4.5%	2.7%
Rural residence	20.3%	21.2%	17.6%	18.4%
Childhood SES
Mother’s education (mean yrs ± SD)	10.73 ± 3.35	10.87 ± 3.43	10.40 ± 3.14	9.92 ± 3.31
Father’s education (mean yrs ± SD)	10.64 ± 4.27	10.87 ± 4.32	10.11 ± 4.07	9.39 ± 4.18
Mother worked for pay at respondent age 14
Yes, high DI	14.8%	16.2%	11.6%	7.7%
Yes, low DI	37.4%	35.8%	41.4%	46.3%
No	47.7%	48.0%	47.1%	46.0%
Father worked for pay at respondent age 14
Yes, high DI	29.4%	33.0%	20.5%	12.9%
Yes, low DI	60.8%	58.3%	67.3%	71.4%
No	9.8%	8.7%	12.2%	15.7%

Interpretation: no suspension or expulsion, participants who did not report experiences of suspension or expulsion; suspension only, participants who reported experiences of suspension but no expulsion; expulsion with or without suspension, participants who reported experiencing expulsion and either did or did not experience suspension as well.

Note. This table presents grand means, proportions, SDs, counts, and ranges estimated across the multiply imputed data sets (200 imputations). Counts may not sum to 8021.

^*Race and ethnicity data served as measures of the socially stratifying effects of systemic racism. Assessed during 1979 household screening interviews via self-report: “from this listing of more than 20 categories, what is/are your racial/ethnic origins?”; re-coded: Black, Latine/x, White, Other/Missing). Note that NLSY collects data on ethnicity (Latine/x) using the response option “Hispanic,” which includes Mexican American, Chicano, Mexican, Mexicano, Cuban, Cubano, Puerto Rican, Puertorriqueño, Boricua/Boriqua/Boriken, Latino, Other Latin American, Hispano, or Spanish descent. Due to small numbers and some mismeasurement concerns, the “other race or missing” category comprised participants who reported at least one of the remaining racial identities (Asian, Hawaiian or Pacific Islander, and American Indian) or for whom race or ethnicity data were missing.^26,27

DI indicates Duncan Index; N, number of participants in sample; SD, standard deviation; SES, socioeconomic status.

Disproportionalities in Punitive School Discipline

Participants reporting early-life suspension and/or expulsion were more likely than those with no such experience to be men, identify as Black, have been born in the US South, and have had lower cSES (Table 1). For example, Black men comprised 14.6% of the sample but 25.3% of those who experienced suspension and 36.8% of those who experienced expulsion. Similarly, Black women comprised 15.4% of the sample, but 18.9% of suspensions. By contrast, White women comprised 24.8% of the sample, but 13.4% of suspensions and 9.5% of expulsions.

Early-life Punitive School Discipline Associated With Midlife Cognition

After adjustment for primary confounders, both suspension—and to a greater extent—expulsion were associated with lower midlife global cognition (suspension: β: −0.21, 95% CI: −0.26, −0.15; expulsion: β: −0.27, 95% CI: −0.38, −0.16). In exploratory domain-specific analyses, we again observed suspension and expulsion were associated with lower midlife memory (suspension: β: −0.15, 95% CI: −0.21, −0.09; expulsion: β: −0.15, 95% CI: −0.26, −0.04), and attention scores (suspension: β: −0.18, 95% CI: −0.23, −0.12; expulsion: β: −0.28, 95% CI: −0.39, −0.16) (Fig. 1; eTables 3–5, Supplemental Digital Content 1, http://links.lww.com/WAD/A530).

FIGURE 1.

Mean differences in midlife cognition z-scores and 95% CIs associated with suspension and expulsion, N = 8021. Adjusted for: race, gender, birth year, southern birth, foreign birth, rural residence before age 14, mother’s education, mother worked for pay, father’s education, father worked for pay (see the Methods section for operationalization). The exposure referent group comprised participants who reported no early life experience of suspension or expulsion. Procedure: linear regression (outcomes: global cognition, memory, and attention z-scores). Interpretation: compared with no punitive school discipline, suspension and expulsion were associated with lower midlife global cognition (suspension: β: −0.21, 95% CI: −0.26, −0.15; expulsion: β: −0.27, 95% CI: −0.38, −0.16), memory (suspension: β: −0.15, 95% CI: −0.21, −0.09; expulsion: β: −0.15, 95% CI: −0.26, −0.04), and attention z-scores (suspension: β: −0.18, 95% CI: −0.23, −0.12; expulsion: β: −0.28, 95% CI: −0.39, −0.16).

Expulsion-midlife Cognition Association Varied by Gender

Associations between punitive school discipline and midlife cognition varied by gender for expulsion but not for suspension. Specifically, expulsion was associated with larger decrements among men than women for global cognition (interaction term β: −0.32; 95% CI: −0.56, −0.08; P < 0.05), memory (interaction term β: −0.24; 95% CI: −0.49, −0.002; P < 0.05), and attention (interaction term β: −0.26; 95% CI: −0.51, −0.01; P < 0.05) (Fig. 2). Results suggested no statistically significant effect modification by race alone (eTables 3–5, Supplemental Digital Content 1, http://links.lww.com/WAD/A530). At the intersections of race, ethnicity, and gender, expulsion was associated with larger decrements in global cognition for White men compared with White women (interaction term β: −0.44; 95% CI: −0.86, −0.03; P < 0.05); estimates for other outcomes and across other racial groups did not meaningfully differ from White women (Fig. 3; eTables 3–5, Supplemental Digital Content 1, http://links.lww.com/ WAD/A530).

FIGURE 2.

Mean differences in midlife cognition z-scores and 95% CIs associated with suspension and expulsion stratified by gender, N = 8021. Note. For interpretability, the figure presents stratified results; however, estimates reported in the text reflect interaction models used to test for statistical significance. Adjusted for: race, gender, birth year, southern birth, foreign birth, rural residence before age 14, mother’s education, mother worked for pay, father’s education, father worked for pay (see the Methods section for operationalization). Procedure: linear regression (outcomes: global cognition, memory, and attention z-scores); tested for effect modification using interaction terms (P < 0.05) with referent group for effect modification analysis by gender comprising participants who reported no early life experience of suspension or expulsion, comprising participants who identified as women. Interpretation: associations between suspension and midlife cognitive performance did not appear to be modified by gender, however, expulsion was associated with larger decrements among men than women for global cognition (interaction term β: −0.32; 95% CI: −0.56, −0.08; P = 0.009), memory (interaction term β: −0.24; 95% CI: −0.49, −0.002; P = 0.048), and attention z-scores (interaction term β: −0.26; 95% CI: −0.51, −0.01; P = 0.043).

FIGURE 3.

Mean differences in midlife global cognition z-scores and 95% CIs associated with suspension and expulsion stratified by race and gender, N = 8021. Note. For interpretability, the figure presents stratified results; however, estimates reported in the text reflect interaction models used to test for statistical significance. Adjusted for: race, gender, birth year, southern birth, foreign birth, rural residence before age 14, mother’s education, mother worked for pay, father’s education, father worked for pay (see the Methods section for operationalization). Procedure: linear regression (outcomes: Global Cognition z-scores); tested for effect modification using interaction terms (P < 0.05) with referent group for effect modification analysis by race and gender comprising participants who reported no early life experience of suspension or expulsion, comprising participants who identified as White women. Interpretation: associations between suspension and midlife cognitive performance did not appear to be modified by race and gender; however, expulsion was associated with larger decrements among White men when compared with White women (interaction term β: −0.44; 95% CI: −0.86, −0.03; P = 0.035). We encourage caution when interpreting these results due to (1) smaller, potentially nonrepresentative sample sizes among racially minoritized participants at the intersections of race and gender and (2) that expulsion was a rare exposure among women.

Robustness Checks

Primary findings were robust to additional adjustment for parental/caregiver death, school-based behaviors, and substance use (eTable 6, Supplemental Digital Content 1, http://links.lww.com/WAD/A530). Adjusting for AFQT attenuated estimates such that only suspension-global cognition, suspension-attention, and expulsion-attention remained significant (eTable 6, Supplemental Digital Content 1, http://links.lww.com/WAD/A530). Complete case analysis results were similar in magnitude and direction, though the expulsion-global cognition and expulsion-memory associations were no longer significant (eTable 7, Supplemental Digital Content 1, http://links.lww.com/WAD/A530).

DISCUSSION

Using prospective NLSY79 cohort data, we evaluated associations between exposure to punitive school discipline in childhood and adolescence—a modifiable school-based adversity—and midlife cognition. We found that both suspension and expulsion were associated with lower performance in midlife measures of global cognition, memory, and attention. As in prior work, our results suggested Black men and women were overrepresented among early-life experiences of punitive school discipline. Finally, we found evidence of effect modification by gender such that estimated decrements in cognition associated with expulsion was consistently larger among men than women. Together, our results suggest that experiences of punitive school discipline are negatively associated with midlife cognition, which could have implications for cognitive aging inequity.

While unique in examining an exposure modifiable at state, district, and school policy levels, our results contribute to a growing literature on early-life adversity and cognitive aging. To date, this literature has centered around experiences of parental death¹² or index adversity measures,¹⁰ generally linking early-life stressors to increased risk for poor cognitive aging outcomes in older adulthood. Fewer studies in this literature have examined school-based adversity specifically, with those that have focused on indicators like “mistreatment by schoolmates”⁴⁰ or quality of education.⁴¹ They find—as we did—higher risk for poorer cognitive outcomes among people with adverse school-based exposures in childhood and adolescence.

The expulsion-midlife cognition relationship varied by gender, such that expulsion had a stronger negative association for men than women. Similarly, at the intersections of race, ethnicity, and gender, expulsion was associated with larger decrements in global cognition for White men than White women. This could indicate that expulsion triggers more consequential shifts in lifecourse trajectories for boys with implications for cognitive aging. Studies suggest that following punitive school discipline, boys are more likely than girls to be repeatedly placed in alternative education programs or pushed out of school—both of which have adverse implications for access to supportive peer/teacher social networks, educational attainment, and involvement in the criminal legal system.^9,16 An existing literature links such compounding factors to cognitive aging, including risk for dementia, over the lifecourse.²

We observed racial inequity in early-life punitive school discipline prevalence—with Black women overrepresented in experiences of suspension and Black men overrepresented in experiences of both suspension and expulsion. In effect modification analyses, associations between punitive school discipline and midlife cognition did not vary by race and ethnicity alone or at the intersections of race, ethnicity, and gender when comparing racially minoritized participants to White women. Racial inequity in exposure to punitive discipline—a function of structural racism—is well-documented in prior work, with Black boys and girls consistently found to be disproportionately targeted.^{14,16,17,20,21} In the absence of effect modification, differential exposure prevalence could serve as a mechanism through which punitive school discipline contributes to cognitive aging inequity. Notably, the afore-mentioned pathways posited to explain gender inequities in our results have also been shown to differentially shape experiences among racially minoritized young people. For example, compared with White children, Black children are not only targeted more frequently for punitive school discipline, but they are more likely to endure harsher consequences (eg, negative impacts on educational opportunity, lost social networks, pushout into the criminal legal system, barriers to re-entering educational trajectories) which have implications for cognitive aging.^17,22 It may therefore be that, given increasingly smaller sample sizes in our data at these intersectional axes of structural oppression, our analysis was underpowered to test for effect modification.

Limitations and Future Research

With NLSY79 school discipline data constrained to indicators of suspension and expulsion up to 1980, it is possible that participants (1) exposed to other types of school discipline (eg, school-based corporal punishment, arrest, detention) or (2) who were still enrolled in school in 1980 and only later experienced suspension and/or expulsion were classified as unexposed, biasing results toward the null. Further, prior work has suggested punitive school discipline may have adverse spillover effects on nonsuspended students¹⁶ indicating our results may underestimate the total association. Future research should evaluate whether other school-based punitive practices, like corporal punishment, are associated with cognitive aging and test for spillover effects to nonsuspended/expelled students. Our results may reflect unmeasured or residual confounding; while we accounted for measured, pre-exposure confounders and conducted sensitivity analyses adjusting for additional sets of covariates, future research leveraging spatiotemporal variation in school discipline practice to draw on the complementary strengths of quasi-experimental designs is warranted. In our AFQT-adjusted sensitivity analysis, specifically, we observed meaningful attenuation of the estimated associations. This likely reflects bias induced by conditioning on a mediator given AFQT measurement temporally followed, and was likely shaped by, the exposure.³⁹ Though these sensitivity analyses were intended to provide a conservative estimate of the association, we urge caution when interpreting these results and recommend future research in data sets—as they become available—with validated, pre-exposure measures of early-life cognitive skills. Due to substantial missingness on indicators used to construct the outcome measures, we examined results using both a complete case and an imputation approach. That these data were hypothesized to largely be missing completely at random (delayed word recall) or missing at random (serial 7 subtractions) may explain why, on empirical assessment, the complete case and imputed results were substantively similar in direction and magnitude despite differences in standard errors. Specifically, while the magnitude of, and precision around, the estimates varied in some of the complete case analysis models, point estimates were within the confidence intervals of the imputed models. We present exploratory cognitive subdomain-specific findings (ie, attention, memory) but encourage future studies to replicate analyses in data sets with additional domain-specific cognition measures.⁴² Given that gender data were assigned by NLSY interviewers and constrained to a gender binary, we interpret findings with caution as they may reflect bias due to gender misclassification or obscure differential results by aggregating across the gender spectrum. We encourage future work examining the experiences of gender minoritized participants given increased risk for punitive school discipline.⁹ Due to the likely myriad experiences collapsed into a single category, we did not interpret findings for the “other race or missing” group and urge replication of these analyses in data sets that better sample these participants. Relatedly, our effect modification analyses may have been underpowered to detect differences at the intersections of race, ethnicity, and gender. We recommend further study in data sampled for representativeness at these intersectional levels of sociopolitical experience.

CONCLUSIONS

Our results suggest punitive school discipline in childhood and adolescence—an exposure modifiable at state, district, and school levels—is negatively associated with midlife cognition and may contribute to cognitive aging inequity, with differences detectable at approximately age 50. To build on this work, we have made several recommendations for future research. Advancing this literature could facilitate considerations for health alongside the education and legal scholarship findings already informing an evolving US school discipline policy landscape.^13,16 Toward that end, prior work has recommended clinicians screen for punitive school discipline as an adverse childhood experience with potential implications for lifecourse health.¹⁹ Indeed, clinicians may serve an essential role by documenting early-life school-based adversities in clinical encounters, and both advancing public awareness of and encouraging research agendas around the social determinants of cognitive aging.^4,6

Supplementary Material

Supplemental Digital Content is available for this article. Direct URL citations are provided in the HTML and PDF versions of this article on the journal’s website, www.alzheimerjournal.com.