Adult Child Schooling and Older Parents’ Cognitive Outcomes in the Survey of Health, Aging and Retirement in Europe (SHARE): A Quasi-Experimental Study

Abstract

A growing body of research suggests that adult child educational attainment benefits older parents’ cognitive outcomes via financial (e.g., direct monetary transfers) and nonfinancial (e.g., psychosocial) mechanisms. Quasi-experimental studies are needed to circumvent confounding bias. No such quasi-experimental studies have been completed in higher-income countries, where financial transfers from adult children to aging parents are rare. Using data on 8,159 adults aged ≥50 years in the Survey for Health, Aging and Retirement in Europe (2004/2005), we leveraged changes in compulsory schooling laws as quasi-experiments. Each year of increased schooling among respondents’ oldest children was associated with better verbal fluency (β = 0.07, 95% CI: 0.02, 0.12) scores; overall associations with verbal memory scores were null, with mixed and imprecise evidence of association in models stratified by parent gender. We also evaluated associations with psychosocial outcomes as potentially important mechanisms. Increased schooling among respondents’ oldest children was associated with higher quality-of-life scores and fewer depressive symptoms. Our findings present modest albeit inconsistent evidence that increases in schooling may have an “upward” influence on older parents’ cognitive performance even in settings where financial transfers from adult children to their parents are uncommon. Associations with parents’ psychosocial outcomes were more robust.

Abbreviations

CASP-12

Control, Autonomy, Self-Realization and Pleasure 12-item scale

CI

confidence interval

CSL

compulsory schooling law

IV

instrumental variable

OLS

ordinary least squares

SD

standard deviation

SHARE

Survey of Health, Aging and Retirement in Europe

 

The fast pace of global aging has lent urgency to efforts to identify scalable, population-level factors that may reduce the burden of age-related diseases, including Alzheimer’s disease and related disorders (1, 2). Much scholarship has focused on educational attainment as a population-level and modifiable determinant of cognitive aging amenable to policy interventions (1, 3, 4). This research has focused largely on the influence of one’s own educational attainment, with some studies also considering the “downward” intergenerational influences of parental educational attainment on their children’s cognition (5, 6). More recently, social scientists have considered that the education of adult children may have “upward” intergenerational impacts on the cognitive aging of their older parents (7–9). This nascent area of inquiry has the potential to inform new population-level targets related to the socioeconomic resources of working-age adults that may in turn benefit older parents’ cognitive outcomes.

Adult child socioeconomic status may influence parents’ late-life health, including their cognitive performance, via a range of interlinked economic and noneconomic mechanisms (7). Economic mechanisms may be in the form of direct financial transfers from children to their parents or via indirect financial transfers that occur in the form of resource sharing (e.g., coresiding in a household) or payments for older parents’ health and long-term care (9–11). Such transfers might improve older parents’ psychological well-being via reduced poverty and family financial strain, which could contribute to better cognitive performance (12). More indirect resources related to health care may contribute to better prevention or more timely treatment of chronic health conditions (e.g., hypertension, diabetes) (13), which could lead to improved cognitive outcomes. Even in the absence of financial transfers, higher education among adult children may be linked to higher levels of social engagement with older parents (9) or an improved sense of social standing, which may also have a positive impact on parents’ psychosocial well-being (14) and, consequently, their cognitive outcomes (9, 10, 15).

Residual confounding is a substantial concern in observational studies of the association between adult child socioeconomic status and parents’ health (7, 16, 17). Unmeasured dimensions of parents’ (respondents’) socioeconomic status may influence both the education of their children, given the strong intergenerational transmission of socioeconomic status (18, 19), and their own late-life cognitive performance (see directed acyclic graph in Web Figure 1, available at https://doi.org/10.1093/aje/kwac151) (3, 6). To address this potential confounding bias, 2 prior studies (10, 15), based in China and Mexico, leveraged changes in compulsory schooling laws (CSLs) that would have affected the children of current older adult cohorts as quasi-experiments. Both studies found that longer schooling of adult children was associated with large improvements in cognitive performance scores among older parents (10, 15).

To our knowledge, no such quasi-experimental studies of parents’ cognitive outcomes or relevant cognitive aging risk factors have been conducted in high-income countries. There may be important differences in these settings, where older parents average higher socioeconomic status (e.g., lower late-life poverty) (20), receive few direct financial transfers from their adult children (21, 22), and tend to have more access to formal long-term care and other social protections (20, 23) compared with older parents in low- and middle-income countries. Given these contextual differences, older European adults may rely much less on the economic and noneconomic resources of their family members to support health and long-term care and other needs in late life. Resource substitution theory (24) suggests that for individuals who are more highly resourced due to higher individual socioeconomic status or other structural advantages, there may be fewer returns to socioeconomic mobility in the subsequent generation.

In the present study, we leveraged the substantial changes in compulsory schooling enacted throughout Europe in the 20th century as quasi-experiments to evaluate the effect of increased schooling among adult children on parents’ cognitive performance. While these CSLs have been utilized in quasi-experimental studies of the association between own education and late-life economic and health outcomes in Europe (25–28), to our knowledge, they have yet to be employed to evaluate the effect of adult child education on parents’ late-life health, including late-life cognitive performance. We also evaluated the effect of increased schooling among adult children on selected psychosocial outcomes for older parents, which may serve as potential mechanisms underlying any associations with older parents’ cognitive outcomes.

Given contextual differences (e.g., a low prevalence of child-to-parent financial transfers), we expected that associations might be of smaller magnitude in Europe compared with those reported in the 2 prior studies based in in low- and middle-income countries. Alternatively, associations of similar magnitude to those reported in other settings might suggest that nonfinancial mechanisms are at play or that the upward intergenerational influence of increased schooling is similar for parents whose children stand to benefit from changes in compulsory schooling laws across settings.

METHODS

We used data from the Survey of Health and Retirement in Europe (SHARE) study. The SHARE data have been described in detail elsewhere (29). Briefly, at the first wave (2004/2005) the SHARE surveyed community-dwelling adults 50 years or older and their spouses of any age from 11 European countries plus Israel. Interviews were done via face-to-face interviews, with interviewers using computer-assisted personal interviewing (CAPI) The overall household response rate was 61.8% (n = 30,421).

We first limited the analytical sample to 22,963 respondents 50 years or older in the 9 countries that passed national-level compulsory schooling laws (CSLs) in years that would have plausibly affected the schooling duration of their adult children (see Web Table 1 and Web Appendix 1 for a summary of the CSLs; see Web Figure 2 for the derivation of the analytical sample).

We further limited the analytical sample to respondents whose oldest child was born in the first 10 birth cohorts (i.e., birth years) to be affected by the CSL, which we considered to be the “treated” group, or in the 10 birth cohorts that would have just missed being influenced by the CSL (n = 8,539). Finally, we excluded those missing data on cognitive outcomes, adult child schooling, or confounder variables, for a final analytical sample of 8,159 for analyses of cognitive outcomes. Our analytical samples for analyses of psychosocial outcomes (n = 8,091 for depressive symptoms; n = 5,097 for quality of life) were smaller due to additional nonresponse on these outcome measures. Quality of life was assessed via a leave-behind pencil-and-paper questionnaire, resulting in a higher rate of nonresponse.

Cognitive performance

Cognitive performance was assessed at baseline. Immediate verbal memory was assessed by asking respondents to listen to a list of 10 words and immediately recall the list. Delayed verbal memory was assessed by asking respondents to repeat the same list of words after a delay of approximately 5 minutes.

Verbal fluency was assessed by asking respondents to list as many unique animals as they could within a 60-second time frame (range: 0–63 for the analytical sample). For each measure of cognitive performance, we calculated z scores using the analytical sample means and standard deviations. We additionally calculated an overall score equal to the sum of z scores for each cognitive performance task divided by the number of tasks.

Psychosocial outcomes

Past-month depressive symptoms were assessed at SHARE baseline with the 12-item EURO-D scale (30). The scale asks respondents to report whether or not they experienced 12 symptoms, including depressed mood, fatigue, and irritability, with scores ranging from 0 (no symptoms) to 12. The EURO-D scale has been shown to be a reliable measure of late-life depressive symptoms, and it is strongly correlated with other depressive symptoms measures such as the Center for Epidemiologic Studies Depression Scale (CES-D) (31).

Subjective quality of life was measured at baseline with the Control, Autonomy, Self-Realization and Pleasure 12-item scale (CASP-12) (32). This abbreviated version of the original CASP-19 (19 items) scale (33) was developed for SHARE. Respondents were presented with a number of statements that measured the 4 CASP domains (e.g., “look forward to each day,” “feel full of energy”). Statements were measured with a 4-item Likert scale of whether the respondents’ experiences corresponded with a given statement often (scored as 1), sometimes, rarely, or never (scored as 4). Items were reverse-coded such that higher scores reflected higher quality of life, with a range between 12 and 48. Internal reliability for the scale ranged from 0.56 to 0.76 across the 4 domains (32). Because prior psychometric studies have suggested that the CASP-12 assesses a single latent factor, our study focused on associations with the overall scale (32).

Adult child schooling

At baseline, SHARE respondents provided information on the highest degree of schooling attained for up to 4 living children aged 16 years and up (94.3% had ≤4 children). This information was reclassified into years of schooling following the United Nations Educational, Scientific and Cultural Organization (UNESCO) International Standard Classification of Education (ISCED) (34). We considered our primary exposure as the years of schooling attained by the oldest child. We considered the schooling duration of oldest daughters and oldest sons separately, given prior evidence of gender-specific pathways of impact (9). As an alternative, we considered the highest-educated child as the index child, following prior literature (15).

Instrumental variables

We followed 2 complementary approaches to generate our instrumental variables.

First, we used information on respondents’ country of residence and their oldest child’s year of birth to create a binary variable indicating whether the respondents’ oldest child was born in the 10 birth cohorts first affected by the CSL vs. the 10 birth cohorts that would have just missed being affected by the same law. Birth cohort bounds are commonly applied in analyses that leverage CSLs in the European context given that other CSLs and related social policies were adopted frequently throughout the 20th century (27, 28). We chose the oldest child as our primary index child because, conceptually, if oldest children benefitted from a given CSL, the remaining children should have also benefitted from the same CSL.

Second, we followed Brunello et al. (28) by using a continuous instrumental variable (IV) indicating the distance between the oldest child’s birth year and the first birth cohort affected by the CSL in a given country. The IV, z, was calculated as , where b is the birth year of the respondents’ oldest child and is the birth year of the first birth cohort affected by the CSL in country k. We applied the same 10-year birth cohort bounds such that values on this continuous IV ranged from −9 to 9. This latter specification allowed for greater precision.

Confounders

We controlled for respondents’ demographic characteristics and life-course socioeconomic status measures as well as characteristics of respondents’ spouses, if relevant. All models included country fixed effects as well as a term for interaction between respondents’ country of residence and their birth year in order to account for country-specific time trends. Further details on the confounder variables and a directed acyclic graph are included in Web Appendix 2 and Web Figure 3.

Statistical analyses

We used 2-stage least squares (2SLS) (35) estimation to evaluate the quasi-experimental association between increased schooling among oldest children and parents’ scores on each of the cognitive tests. We adopt the monotonicity assumption: that increases in mandatory schooling would not lead any children to have less schooling than they otherwise would complete. Under this assumption (and conventional IV conditions), the IV estimate is interpretable as the effect of additional schooling completed by the oldest child due to the policy among those whose children complied with the CSL change. We compared these estimates with those generated via conventional ordinary least squares regression; we additionally provide nonparametric estimates using targeted maximum likelihood estimation (36).

For the 2SLS approach, our first-stage estimates were generated with a linear model regressing the years of schooling completed by the oldest child (or oldest daughter) on the birth cohort–based IV and covariates (see Web Appendix 3 for equations). We evaluated the strength of our instrument via the Kleibergen-Paap Wald F-statistic test (37); all primary analyses had a first-stage F-statistic above the conventional cutoff of 10 (Web Table 2). Our second-stage model regressed respondents’ cognitive z scores on the predicted value of schooling for their oldest child, generated with the first-stage regression, and the same set of covariates as the first-stage model. In the models for the overall sample, we clustered standard errors at the household level.

We conducted these same analyses using our second instrument, the continuous measure of distance from the CSL across the 20 eligible birth cohorts. In order to indirectly shed light on potential mediators of our main associations of interest, we replicated the same procedures for our 2 psychosocial outcome measures. We additionally estimated parent gender–stratified models, given prior evidence of heterogeneity in the association between adult child schooling duration and late-life health for older mothers and fathers (15, 38). In order to understand whether our observational results generalized to respondents and European countries excluded from our primary analyses, we generated ordinary least squares (OLS) estimates for the respondents in all European countries included in the baseline wave of SHARE. Finally, given the fact that the Swedish schooling reform was rolled out in some regions prior to national-level implementation (see Web Appendix 1), we repeated analyses without Swedish respondents.

RESULTS

Descriptive

Respondents were an average of 68.7 (standard deviation (SD), 8.7) years old (range, 51–99), majority women (58%), and reported a mean of 2.6 (SD, 1.4) living children (Table 1). Respondents reported that they had achieved an average of 8.6 (SD, 4.4) years of education, but that their oldest children had completed a mean of 12.3 (SD, 3.6) years of schooling (with similar average years completed by oldest daughters and sons). The average age of respondents’ oldest children was 43.6 (SD, 7.9) years (range, 26–55).

Table 1.

Descriptive Characteristics, Older European Adults Aged ≥50 Years Whose Oldest Child Was in the First Birth Cohorts to Just Benefit From or Just Miss Benefitting From a Change in Compulsory Schooling Laws (n = 8,159), Survey on Health, Aging and Retirement in Europe, 2004/2005

Characteristic	No.	Mean (SD)
Respondent demographic characteristics, baseline
Age, years		68.71 (8.68)
Female sex	4,731
Born in country of current residence	7,720
Marital status
Married	5,882
Never married	92
Divorced	456
Widowed	1,729
Spouse age, yearsa		67.48 (8.97)
Total no. of living children		2.59 (1.39)
Respondent and family life-course socioeconomic status
Respondent years of education		8.58 (4.37)
Spouse years of educationa		8.71 (4.36)
Maternal occupational prestige score		9.29 (16.52)
Maternal occupational prestige score missing	6,018
Paternal occupational prestige score		32.98 (16.19)
Paternal occupational prestige score missing	1,032
Respondent ever worked	7,154
Respondents’ occupational prestige score for current/last job		31.42 (20.14)
Respondents’ occupational prestige score for first job missing	1,701
Spouses’ occupational prestige score		28.33 (21.22)
Spouses’ occupational prestige score missing	2,354
Adult child educational attainment
Years of educational attainment, oldest child		12.30 (3.55)
Years of educational attainment, oldest daughterb		12.22 (3.53)
Years of educational attainment, oldest sonc		12.28 (3.54)
Respondent cognitive scores
Immediate verbal recall (range: 0–10)		4.50 (1.79)
Delayed verbal recall (range: 0–10)		3.01 (1.94)
Verbal fluency (range: 0–63)		17.58 (6.82)
Respondent psychosocial outcomes
Past-month depressive symptoms, EURO-Dd (range: 0–12)		2.45 (2.31)
Quality of life, CASP-12e (range: 12–48)		36.74 (6.26)

Abbreviations: CASP-12, Control, Autonomy, Self-Realization and Pleasure 12-item scale; EURO-D, European depression scale; SD, standard deviation.

^a Limited to those who were married at baseline.

^b Limited to n = 6,227 with at least 1 adult daughter.

^c Limited to n = 6,218 with at least 1 adult son.

^d The EURO-D scale is a measure of past-month depressive symptoms, with scores ranging from 0 (no symptoms) to 12.

^e Items were reverse-coded such that higher scores reflected higher quality of life, with scores ranging from 12 and 48.

We also present descriptive characteristics by treatment status on the binary IV (Web Table 3); there were differences across demographic characteristics, including respondents’ own age and education, which we controlled for in our analyses. The descriptive characteristics for the subsample included in the analysis of quality of life (Web Table 4) were similar to those for the analytical sample in our primary analysis.

First-stage regression results

Our first-stage estimates (Web Table 5) indicated that the binary IV based on exposure to CSL reforms was associated with an additional 0.64 years of schooling among oldest children overall (95% confidence interval (CI): 0.41, 0.86; for oldest daughters, β = 0.86, 95% CI: 0.61, 1.11; for oldest sons, β = 0.47, 95% CI: 0.21, 0.73). We show the discontinuity in average years of adult child schooling duration before and after CSL changes in Web Figure 4–6. First-stage point estimates for the highest-educated child were of smaller magnitude, but 95% CIs were overlapping with those generated for our primary analyses (Web Table 5).

Oldest child’s schooling duration and parents’ cognitive scores

OLS regression estimates (not based on instrumental variables) suggested that each additional year of schooling for an oldest child was associated with better cognition for their parents, with small but relatively consistent estimates across all cognitive outcomes (e.g., for the overall cognitive performance z score, β = 0.02, 95% CI: 0.01, 0.02; Table 2).

Table 2.

Comparison of β Coefficients From Regression Analysis Evaluating the Association Between Oldest Child’s Educational Attainment and Cognitive Performance z Scores for Older Parents (n = 8,159), Aged ≥50 Years, Survey on Health, Aging and Retirement in Europe, 2004/2005

	OLS a		2SLS, Binary IV a		2SLS, Continuous IV a
Outcome Variable	β	95% CI	β	95% CI	β	95% CI
Immediate verbal recall z score	0.017	0.011, 0.023	−0.027	−0.104, 0.050	−0.010	−0.058, 0.038
Delayed verbal recall z score	0.013	0.007, 0.019	0.019	−0.062, 0.099	0.018	−0.034, 0.070
Verbal fluency z score	0.018	0.012, 0.024	0.045	−0.030, 0.119	0.066	0.017, 0.115
Overall cognitive z score	0.016	0.011, 0.021	0.012	−0.048, 0.072	0.025	−0.014, 0.063

Abbreviations: 2SLS, 2-stage least squares regression; CI, confidence interval; IV, instrumental variable; OLS, ordinary least squares estimation.

^a Controls: age, gender, marital status, own education, spouse age (if married), spouse’s education (if married), total number of kids, whether or not the respondent was born in country of current residence, paternal and maternal occupational prestige, spouse’s occupational prestige, respondent’s occupational prestige, whether or not respondent ever worked outside the home, country of residence, and a term for interaction between country of residence and respondents’ birth year. Models cluster standard errors at the household level.

In IV models, we found that each 1-year increase in schooling for SHARE respondents’ oldest children was associated with improved verbal fluency z scores (β = 0.07, 95% CI: 0.02, 0.12) when using the continuous IV; estimates were similar but less precise when using the binary IV (Table 2, Figure 1C). Associations with immediate and delayed verbal recall and overall cognitive performance scores were null in overall models.

Figure 1.

Two-stage least squares and ordinary least squares (OLS) regression estimates, oldest child's schooling duration, and older parents’ cognitive performance z scores in the Survey of Health, Aging and Retirement in Europe, 2004/2005. A) Immediate recall z scores; b) delayed recall z scores; c) verbal fluency z scores; D) overall cognition z scores. IV, instrumental variable.

In models stratified by parent gender (Web Table 6, Figure 2), there was evidence of association between increased adult child schooling and higher delayed verbal recall z scores among older fathers (β = 0.05, 95% CI: −0.03, 0.13, Figure 2B) but lower immediate verbal recall z scores for older mothers (β = −0.04, 95% CI: −0.11, 0.02, Figure 2A); 95% CIs crossed the null in both cases.

Figure 2.

Two-stage least squares analysis of oldest child's schooling duration and older parents’ cognitive performance in the Survey of Health, Aging and Retirement in Europe, 2004/2005, according to parent gender. A) Immediate recall z scores; b) delayed recall z scores; c) verbal fluency z scores; D) overall cognition z scores.

Oldest child’s schooling duration and parents’ psychosocial outcomes

OLS estimates indicated that more years of schooling for the oldest child was associated with better quality of life and fewer depressive symptoms for their older parents (Table 3).

Table 3.

Comparison of β Coefficients From Regression Analysis Evaluating the Association Between Oldest Adult Child’s Educational Attainment and Psychosocial Outcomes for Older Parents, Aged ≥50 Years, Survey on Health, Aging and Retirement in Europe, 2004/2005

		OLS a			2SLS, Binary IV a			2SLS, Continuous IV a
Outcome Variable	No.	β	95% CI	No.	β	95% CI	No.	β	95% CI
Quality of life, (CASP-12b	5,097	0.196	0.140, 0.252	5,003	0.745	−0.024, 1.513	5,058	0.745	0.245, 1.245
Past-month depressive symptoms, EURO-Dc	8,091	−0.036	−0.053, −0.019	8,035	−0.208	−0.426, −0.015	8,035	−0.095	−0.243, 0.053

Abbreviations: 2SLS, 2-stage least squares regression; CASP-12, Control, Autonomy, Self-Realization and Pleasure 12-item scale; CI, confidence interval; EURO-D, European depression scale; IV, instrumental variable; OLS, ordinary least squares estimation.

^a Controls: age, gender, marital status, own education, spouse age (if married), spouse’s education (if married), total number of kids, whether or not the respondent was born in country of current residence, paternal and maternal occupational prestige, spouse’s occupational prestige, respondent’s occupational prestige, whether or not respondent ever worked outside the home, and country of residence. Models cluster standard errors at the household level.

^b Items were reverse-coded such that higher scores reflected higher quality of life, with scores ranging from 12 and 48.

^c The EURO-D scale is a measure of past-month depressive symptoms, with scores ranging from 0 (no symptoms) to 12.

In IV analyses, each 1-year increase in the schooling duration of SHARE respondents’ oldest children was associated with improved quality-of-life scores (using the binary IV, β = 0.75, 95% CI: −0.02, 1.51; using the continuous IV, β =0.75, 95% CI: 0.25, 1.25; Table 3, Figure 3A) and fewer depressive symptoms (using the binary IV, β = −0.21, 95% CI: −0.43, −0.02; using the continuous IV, β = −0.10, 95% CI: −0.24, 0.05; Table 3, Figure 3B). There were no differences according to parent gender (Web Table 6).

Figure 3.

Two-stage least squares and ordinary least squares (OLS) regression estimates, oldest child's schooling duration, and older parents’ psychosocial outcomes in the Survey of Health, Aging and Retirement in Europe, 2004/2005. We used 2 measures. A) The Control, Autonomy, Self-Realization and Pleasure 12-item scale (CASP-12) has a range of 12–48; B) the EURO-D depressive symptom scale has a range of 0–12. IV, instrumental variable.

Sensitivity analyses

Across cognitive and psychosocial outcomes, findings were similar when separately evaluating increased years of schooling for oldest daughters and oldest sons (Web Tables 7–10). Findings were of larger magnitude but less precise when instead focusing on the highest-educated child as the index (Web Table 11). OLS estimates were very similar when including all respondents included in the baseline wave of SHARE, including all birth cohorts and European countries (Web Table 12). Finally, across all outcomes, the magnitude of some estimates differed, but the 95% CIs were overlapping when instead using a nonparametric estimator (Web Table 13) and when removing Swedish respondents from our analytical sample (Web Table 14).

DISCUSSION

In this population-based quasi-experimental study, we found that increased schooling duration among adult children contributed to improved performance on a measure of verbal fluency for older parents in 9 European countries. We additionally found associations with improved quality of life and lower depressive symptoms; these are commonly proposed as important nonfinancial mechanisms underlying the association between adult child schooling duration and older parents’ cognitive outcomes. CSL changes have long been used to evaluate the quasi-experimental association between own education and late-life health (4, 39)—and particularly in Europe, where there were substantial changes in school-leaving policies throughout the 20th Century (26, 27). Ours is, to our knowledge, the first quasi-experimental study to evaluate the association between increased adult child schooling duration and parents’ cognitive outcomes in a high-income setting.

Our estimates were more modest and inconsistent than those found in the 2 prior studies (10, 15) on this topic, which found that each additional year of schooling was associated with increased verbal memory scores among older parents by between 0.07 SD and 0.11 SD in China and Mexico and increased verbal fluency scores by an estimated 0.17 SD in Mexico. In our study, each additional year of adult child schooling was associated with a 0.07-SD increase in older parents’ verbal fluency scores, and associations with verbal memory scores were null for respondents overall and of small magnitude and inconsistent in gender-stratified models. To place these findings in context, quasi-experimental studies—including those using SHARE data—consistently report that that each additional year of one’s own schooling is associated with an increase of between 0.10 SD and 0.15 SD on late-life cognitive performance scores (40, 41). The estimates from prior studies therefore suggest that adult child schooling duration is as important for older adults’ late-life cognition as their own schooling duration—a surprisingly large estimate—whereas our largest estimates are less than half the magnitude of what would be expected for own years of schooling.

The differences between our and prior studies could have been driven by contextual differences, including the very rare occurrence of child-to-parent financial transfers, higher average socioeconomic status, and greater state-sponsored social protections among older respondents in Europe compared with their counterparts in low- and middle-income countries (20–23). These factors may have meant that increased schooling among adult children was less impactful for older parents’ outcomes in a setting where older adults were less reliant on the financial resources of their children. In addition, the CSLs we focused on were typically increases of 1 or 2 years, whereas the CSLs leveraged in the prior studies increased mandatory schooling by 3 years, which may have contributed to a smaller magnitude of association in our study.

Differences could also have been due to the choice of the “index” child for the primary analysis. In both prior studies, the highest-educated child served as the index, while estimates with alternative index children (e.g., the lowest-educated child) were of smaller magnitude or null. We also generated larger estimated associations with verbal fluency scores (e.g., 0.12 SD for verbal fluency z scores; 0.10 for delayed verbal recall z scores) when we focused on the highest-educated child. These larger point estimates may have been because children in a given family with the propensity to be the highest-educated may have already been achieving higher than minimum schooling prior to CSL changes; higher “noncompliance” among this group may have been reflected in our smaller first-stage estimates for highest-educated children compared with the oldest children. Noncompliance with the CSL changes (i.e., smaller first-stage estimates) make the IV estimate more responsive to small violations of the exclusion restriction (i.e., small violations could lead to large net biases). Further research that directly harmonizes studies from different settings is needed to disentangle contextual vs. measurement-related drivers of cross-study differences.

The reasons for differing associations across cognitive performance tasks in our study is unclear. Given our relatively low precision and wide, overlapping confidence intervals, we cannot rule out chance as the explanation for differences in estimates across domains. Verbal memory performance tasks—and particularly the immediate verbal recall task—appeared to have lower variability, and associations may have been biased by stronger ceiling effects. Future studies should continue to replicate analyses across data sources in order to investigate whether there is true heterogeneity in the impact of adult child schooling on parents’ cognitive outcomes by cognitive domain. We similarly are not able to draw definitive conclusions about the differences by parent (respondent) gender, given limited precision. We did note differences in first-stage estimates by child gender, suggesting that adult daughters of respondents in this sample may have benefitted more from increases in minimum schooling laws. However, our second-stage results suggest that the returns of increased schooling among daughters and sons were similar for older parents.

Our findings support the hypothesis that the modest associations between adult child education and parents’ cognitive performance might be explained at least in part by psychosocial pathways. Our findings are consistent with prior observational studies of adult child education and parents’ depressive symptoms based in Taiwan and the United States (12, 42). In the only other quasi-experimental study on this topic, Ma (10) found no association between increased adult child schooling duration and past-week depressive symptoms for older Chinese adults but a robust association with their ratings of life satisfaction. Both depressive symptoms and quality of life are important endpoints themselves, and also have important implications for cognitive aging: Late-life depression is one of the most important modifiable risk factors for dementia (1). Higher quality of life has also been linked to late-life cognition and dementia risk, as well as higher social engagement and fewer chronic health conditions, which are associated with cognitive aging (32, 43–45). However, there may be other important mechanisms at play, and future research should formally quantify the extent to which associations between adult child schooling duration and parents’ cognitive outcomes are mediated by psychosocial vs. alternative (e.g., financial, behavioral) pathways.

Limitations

Estimates generated from the IV approach are interpreted as the local average treatment effect; in our case, this would be the quasi-experimental association between increased child schooling duration and parents’ cognitive outcomes for families whose children increased their years of schooling because of the change in compulsory schooling laws. Estimates are therefore not generalizable to families whose children’s education was not affected by the compulsory schooling law, for example, because they would have achieved higher than the minimum levels of schooling regardless of the compulsory schooling laws or because they dropped out of the minimum level of schooling even with the new laws in place. In addition, our study excluded SHARE participants in countries with no CSL change during the relevant time period or whose oldest child was not born within the relevant birth cohort bounds, potentially reducing external validity. In particular, estimates might not be generalizable to other birth cohorts or historical time periods. Despite these limits to external validity inherent in our IV analyses, we found consistent evidence of association in observational estimates including those generated for all cohorts and European countries included in the baseline SHARE wave.

We also note limitations related to exposure classification and the selection of the “index” child. Respondents classified as “unexposed” in our analysis could have had younger children who were exposed to the CSL, leading to exposure misclassification and potentially biasing our estimates towards the null. However, we have already outlined potential concerns with our alternative analyses focused on the highest-educated child as the index.

Conclusion

In this study, we found that adult children’s increased schooling duration was associated with better verbal-fluency scores for older parents in 9 European countries. Associations with verbal memory scores were null for the sample overall, although there was modest and inconsistent evidence of association between increased adult child schooling and verbal memory scores according to older parents’ gender. Even our largest results were more modest than estimates from prior studies based in China and Mexico. However, evidence of protective associations with older parents’ psychosocial outcomes, which may be important pathways underlying the relationship between adult child education and parents’ late-life cognition, were more robust.

Our findings contribute rigorous quasi-experimental evidence to the growing body of research from other global settings suggesting that adult child education may have “upward” intergenerational influences on cognitive aging and related health outcomes for older parents. This nascent line of research adds to the already robust evidence of own educational attainment as a critical determinant of cognitive aging, further reinforcing the importance of education as a population-level investment in improved cognitive health across generations.

Future research should directly compare estimates across settings using harmonized measures and methods to identify contextual vs. measurement-related drivers of cross-study differences. In addition, further quasi-experimental research on this topic should be extended to a broader set of dementia risk factors across varied settings, including longitudinal cognitive decline, and health behaviors and chronic health conditions.