Table 2 Description of measures

(1) Self-control

The Brief Tangney Self-Control Scale: Consists of 13 self-report items on a 5-point Likert scale [56]. Example questions are: “I am good at resisting temptation” and “I have a hard time breaking bad habits.” A mean score was created based on the 13 items, with a higher overall mean score indicating higher self-control

The Impulsivity and Sensations Seeking Scale: We assessed impulsivity and sensation seeking with the Zuckerman–Kuhlman–Aluja Personality Questionnaire (ZKA-PW, [67]. This self-reported scale consists of 8 items measuring impulsivity and 11 items measuring sensation seeking, including items such as “I’m an impulsive person” and “I usually think about what I am going to do before doing it.” We created a mean score, with higher scores reflecting more impulsivity and sensation seeking

Risk preference: We assessed risk preference or aversiveness with one item where participants are asked to rate themselves on a 11-point scale on the following question: “In general, are you someone who is willing to take risks or do you try to avoid risks?”. We recoded the scale such that higher scores reflect more risk aversiveness [2]

The Attention Problems scale: We used the 11 items of the attention problems scale of the Child Behavior Checklist [1]. Children filled in questions such as “I fail to finish things that I start,” “I can’t sit still,” on a 3-point scale. A sumscore was created, with higher scores reflecting more attention problems

We preregistered to use the ASEBA Self-Control Scale, but the items required for this scale were not available. We therefore used the ASEBA-Attention problem scale, which overlaps in 4 items with the ASEBA-Self-Control Scale [64]

Our preregistration included the Impulsivity and Patience scale (IPS, [62], but the Cronbach alpha of this scale was not sufficient (Cronbach α = .39) unlike the Brief Tangney Self-Control scale (Cronbach α = . 76). Thus, the Impulsivity and Patience scale was dropped from analyses

Grit: We used the Short Grit Scale (SGS) which is a self-report scale assessing diligence and grit with an 8-item questionnaire developed by Duckworth & Quinn [12]. It includes self-reported items on a 5-point scale with questions such as “new ideas and projects sometimes distract me from previous ones” and “setbacks don’t discourage me.” We created an overall sum score, with higher scores indicating more grit

(2) DNAm measures of aging-related health^a,b

DunedinPACE was developed in the Dunedin Study birth cohort and is based on analyses of within-person change in 18 physiological markers measured repeatedly at age 26, 32, 38 and 45. It is an extension of the DunedinPoAm pace of aging which was based on a 12-year period, while DunedinPace is based on 20 years of follow-up [5, 13]. Briefly, DunedinPACE was developed using a subset of EPIC array probes that were also included on Illumina’s earlier 450 k array, showing to have higher test–retest reliability [55]. Subsequently, elastic-net regression analyses was applied to fit Pace of Aging to DNAm data to blood samples collected when participants were 45 years, resulting in a 173-CpG algorithm. DunedinPACE was calculated based on the algorithm published by Belsky et al. [5]

GrimAge is a DNAm measure developed on a set of physiological indicators using machine learning analyses and DNAm algorithms to predict morbidity and mortality. GrimAge signifies the age in years at which average mortality risk in the Framingham Heart Study Offspring cohort matches actually predicted mortality risk. We used DNAm principal components when computing GrimAge to increase reliability [23], and created GrimAge Acceleration by residualizing GrimAge for chronological age. We preregistered to use GrimAge version 2, but the code to calculate this score is not yet publicly available

PhenoAge is modeled based on physiological markers and chronological age and subsequently applied to a new sample modeled from DNA methylation to derive a final DNA methylation clock [30]. It represents the age in years at which average mortality risk in NHANES III matches the mortality risk predicted by the PhenoAge algorithm. We used DNAm principal components when computing PhenoAge to increase reliability [23], and created PhenoAge Acceleration by residualizing PhenoAge for chronological age

(3) Socioeconomic contexts

Family-level socioeconomic contexts were indexed by an average z-score including household income (equivalent net income) from different resources such as employment, child support, unemployment benefits, and pensions corrected for the number of people living in the household) and educational attainment (the highest degree obtained by any individual in the household in number of educational years + additional occupational training years) corrected for the number of people living in the household

Family-level socioeconomic contexts In line with earlier studies using the TTP data [14], we computed a socioeconomic composite as the average of standardized parent educational attainment and standardized household income

Initially, we preregistered a broad socioeconomic disadvantage score (e.g., including US household food security, father absence, changes in home address, family public assistance, income and education). For comparison purposes, we computed a socioeconomic composite in the same way as in the SOEP cohort instead

Our preregistration included analyses with neighborhood SES to examine gene-by-environment interactions on self-control. Given the lack of association between polygenic indices and self-control, we did not include neighborhood SES

(4) Polygenic indices^c

Polygenic Index for externalizing (PGI-EXT) has been computed in both cohorts based on the most recent genome-wide association study (GWAS) of externalizing problems [26]. This GWAS pooled data from ~ 1.5 million people, applying a multivariate GWAS approach leveraging genetic correlations among externalizing-related measures (attention-deficit/hyperactivity disorder, problematic alcohol use, lifetime cannabis use, age at first sexual intercourse, number of sexual partners, general risk tolerance & lifetime smoking initiation). The PGI-EXT is an aggregate of the effects of observed SNPs (including 1,020,283 SNPs), weighted by their estimated effect sizes, from an independent GWAS sample. This PGI is of particular interest to our study as the score includes traits highly correlated with self-control such as ADHD, risk tolerance, problematic alcohol use, and smoking

Deviating from our preregistration, the PGI for non-cognitive skills [9] was not available and therefore not included in analyses

(5) Self-reported health

Self-reported disease severity: participants were asked how they would describe their current state of health on 1 item, ranging from 1 = very good to 5 = very bad, with higher scores reflecting higher self-reported disease severity

For the analyses on Health, we focused on SOEP-G as the TTP consists of children and adolescents that are generally in good health

Self-reported health: participants were asked to rate across 5 items if they in the last 4 weeks experienced any limitations in life due to physical pain, physical problems or mental health problems, with 1 = always, to 5 = never, with higher scores reflecting more self-reported health

In our preregistration, we did not integrate health variables (see main text for motivation). We selected health variables that previously found to be associated with the PGI-EXT [28]

(6) Covariates

Body Mass Index (BMI) was computed by transforming self-reported height (in cm) and weight (in kg) in sex- and age-normed z-scores

Smoking was measured by self-reported tobacco use, grouping those who smoke, used to smoke or ever smoked into a smoking group versus a non-smoking group with participants who have never smoked

Deviating from our preregistration, we did not include substance use as a covariate as the sample sizes were too small in both samples (n < 5%)

Pubertal development was measured using children’s self-reports on the Pubertal Development Scale [44] assessing development across height, body hair growth, skin changes. Specific additional questions for girls included onset of menses, breast development and questions specifically for boys included, growth in body hair, deepening of voice. Pubertal development was residualized for age separately for each sex