Marriage and Desistance From Crime: A Consideration of Gene–Environment Correlation

Overview and context

• The study reexamines the marriage–desistance link using a genetically informed design to assess whether genetic factors confound or partly explain why marriage is associated with desistance from crime.

• Previous research shows marriage often coincides with reduced offending (the “marriage effect”), but most work did not account for genetic confounding or gene–environment correlations (rGEs).

• This article integrates behavioral genetics with life-course criminology to test if active rGEs (where individuals genetically predisposed to marriage select into marital environments) help explain the marriage–desistance association.

• Key background concepts:

  • Desistance: a process and boundary concept where offending declines over time, possibly culminating in cessation (with notions of deceleration and de-escalation before termination).

  • Social bonds and life-course theory (Sampson & Laub): adult social bonds (marriage, employment, military) create stakes in conformity that facilitate desistance.

  • rGE: genetic influences can shape exposure to environments. Three types exist: passive, evocative, and active; this study focuses on active rGE (assortative mating or niche-picking effects).

• Core questions: Do genetic factors influence (a) marriage propensity, (b) desistance from crime, and (c) whether the marriage–desistance link remains after controlling for shared genetic influences?

• Data source: Add Health (National Longitudinal Study of Adolescent Health), including a sibling subsample with genetic relatedness information.

Theoretical framework: gene–environment correlations and marriage

• Active rGE in marriage:

  • Genes influence personality and preferences, which shape environmental selection (e.g., mating choices).

  • Assortative mating: genetically influenced traits lead individuals to select partners with similar traits, potentially confounding the marriage–desistance link.

  • Genetic influences on marriage have been documented (heritability estimates for marital status across ages; examples include h2 ≈ .58 in early adulthood decreasing with age).

• Mechanisms linking active rGE to marriage and desistance:

  • Personality traits shaped by genes influence marital propensity and the quality of marital relationships.

  • Marriage itself may change life trajectories by increasing routines, social support, and monitoring—mechanisms that promote desistance.

• Foundational prior work and methods to address selection:

  • Inverse probability of treatment weighting (IPTW) to account for selection into marriage (Sampson et al., 2006).

  • Propensity-score matching to balance covariates and isolate the causal effect of marriage on desistance (King, Massoglia, & Macmillan, 2007).

  • Twin-family designs showing that while marriage is associated with desistance, part of the association may be due to selection (Burt et al., 2010).

• The current study builds on these to test whether marriage–desistance is spurious due to shared genetic factors using Add Health sibling data and an ACE modeling approach.

Research questions and hypotheses

• Q1: Do genetic factors influence marital status?

  • Hypothesis: A portion of variance in marital status is due to genetic factors (heritability h2).

• Q2: Do genetic factors influence desistance from crime?

  • Hypothesis: A portion of variance in desistance is due to genetic factors (heritability h2).

• Q3: After controlling for shared genetic influences, is there still a marriage–desistance association?

  • Hypothesis: If a shared genetic pathway exists, the association will attenuate substantially when genetic factors are accounted for; if a unique causal effect remains, marriage will predict desistance even after accounting for genetics.

Data and sample

• Data source: Add Health ( Waves 1–3; 1994–1996; 2001–2002).

• Sample structure:

  • Wave 3 (ages roughly 18–27): marital status and desistance data collected.

  • Sibling subsample from Wave 1 residing in the same household; includes twin pairs, full siblings, half-siblings, and cousins.

  • Total genetic subsample: 4,568 individuals (2,284 pairs) with varying relatedness; structure allows estimation of genetic and environmental components.

• Genetic relatedness by pair type (from Table 1):

  • MZ twins: 1.00 (578 individuals; 289 pairs)

  • DZ twins and full siblings: 0.50 (2,972 individuals; 1,486 pairs)

  • Half-siblings: 0.25 (750 individuals; 375 pairs)

  • Cousins: 0.125 (268 individuals; 134 pairs)

• Key phenotype definitions:

  • Marital status: dichotomous, 0 = never married, 1 = married at least once (Wave 3).

  • Desistance: a boundary/directional measure constructed from delinquency/criminal behavior across waves; desisters are those with delinquency in adolescence (Wave 1 or 2) but no crime in adulthood (Wave 3).

• Sample sizes for main analyses:

  • Marital status analysis: N = 3,745

  • Desistance analysis: N = 3,008

• Desistance measurement details:

  • Wave 1 and Wave 2 delinquency indices built from 17 items (α ≈ .85 and .82 respectively).

  • Wave 3 delinquency index built from 12 items (α ≈ .69).

  • Abstainers (no delinquency across all waves) removed from desistance analysis.

• Control variables in main analyses:

  • Age (in years), race (0 = non-Black, 1 = Black), gender (0 = female, 1 = male).

• Descriptive statistics:

  • About 18.34% of Wave 3 respondents reported having been married at least once.

  • Approximately 72% of the analyzed sample were desisters (n ≈ 2,151 of 3,008).

Measures in detail

• Marital status (Wave 3):

  • Binary indicator: married at least once vs never married.

• Desistance ( Waves 1–3):

  • Step 1: Create Wave 1 and Wave 2 delinquency indices by summing responses to 17 items (higher scores = more delinquency).

  • Step 2: Create Wave 3 delinquency index by summing 12 items (higher scores = more delinquency).

  • Step 3: Define desister as having at least one delinquent act in adolescence (Wave 1 or Wave 2) and zero delinquent acts in adulthood (Wave 3).

  • Excluded: abstainers (0 on all waves) due to lack of opportunity to desist.

• Controls:

  • Age (continuous), race (Black vs non-Black), gender (Male vs Female).

Analytic strategy and models

• Part 1: Genetic and environmental contributions to variation in marriage and desistance

  • Probandwise concordance rates: use the formula PC = rac{2C}{2C+D} where C = concordant pairs and D = discordant pairs.

  • Cross-sibling tetrachoric correlations to account for thresholded/dichotomous outcomes and population incidence.

  • ACE modeling (A = additive genetics, C = shared environment, E = nonshared environment) for dichotomous outcomes (thresholds), to estimate h2, c2, e2.

  • Genotype-related structure: set genetic correlation between siblings according to relatedness (e.g., a1, a2 paths in ACE; correlations of A factors scale with relatedness: 1.00 for MZ, 0.50 for DZ, 0.25 for HS, 0.125 for cousins).

• Part 2: Baseline association between marriage and desistance (without genetic controls)

  • Logistic regression: model the effect of marital status on desistance controlling for age, race, and gender (no genetic controls).

  • Outcome: desistance status; predictor: marital status; report odds ratio (OR) for marriage.

• Part 3: Genetic controls via DeFries–Fulker approach (DF) to test for confounding by shared genetics

  • Logistic DeFries–Fulker (DF) model specification:

  • ext{logit}ig(P(K1)ig) = eta0 + eta1(K2) + eta2(R imes [K2]) + eta_3( ext{ENVDIF}) + e

  • Here:

  • $K1$ = Sibling 1's desistance score; $K2$ = Sibling 2's desistance score; $R$ = genetic relatedness between siblings (e.g., 1.00 for MZ; 0.50 for DZ).

  • $R imes [K_2]$ = multiplicative interaction term capturing genetic similarity effects on the outcome.

  • ENVDIF = measure of difference in marital status and other covariates (age, race differences) between siblings.

  • Interpretation: if the marriage effect remains after accounting for genetics (β2), there is a potential unique causal effect of marriage on desistance; if the effect attenuates substantially, it suggests genetic confounding or selection via active rGE.

• Outcome interpretation from results:

  • Compare Model 1 (no genetic controls) vs Model 2 (genetic controls) in terms of the marriage coefficient (log-odds or odds ratio).

  • A substantial attenuation (e.g., ~60% reduction) supports genetic selection as a confounder; a remaining significant effect would support a causal role for marriage in promoting desistance.

Results: key findings and numbers

• Marriage status: genetic and environmental influences

  • ACE model results (Table 2): h^2 (A) = 0.56, c^2 (C) = 0.00, e^2 (E) = 0.44

  • Probandwise concordance and tetrachoric correlations follow the expected pattern for genetic influence:

  • MZ concordance: 0.48; tetrachoric r ≈ 0.60

  • DZ/FS concordance: 0.29; tetrachoric r ≈ 0.26

  • Half-siblings: 0.22; tetrachoric r ≈ 0.06

  • Cousins: 0.21; tetrachoric r ≈ 0.23 (note: the cousin tetrachoric correlation was not statistically different from zero)

• Desistance: genetic and environmental influences

  • ACE model results (Table 3): h^2 (A) = 0.49, c^2 (C) = 0.00, e^2 (E) = 0.51

  • Twin-pair concordances and tetrachoric correlations support genetic and nonshared environmental contributions:

  • MZ concordance: 0.82; tetrachoric r ≈ 0.53

  • DZ/FS concordance: 0.75; tetrachoric r ≈ 0.22

  • Half-siblings: 0.69; tetrachoric r ≈ 0.20

  • Cousins: 0.76; tetrachoric r ≈ 0.28

• Association between marriage and desistance with and without genetic controls (Table 4)

  • Model 1 (no genetic controls): marital status associated with desistance with OR ≈ 2.33 (p < .05); age, gender, race included as covariates.

  • Model 2 (genetic controls via DF): marital status associated with desistance with OR ≈ 1.37 (p < .05).

  • Quantitative attenuation: the effect of marriage on desistance was about 60% smaller after controlling for genetic influences (βModel1 = 0.84; βModel2 = 0.31; (0.84 − 0.31)/0.84 ≈ 0.63, i.e., ~63%, rounded to ~60%).

• Supplemental notes on model interpretation:

  • The DF model in Model 2 tests sibling differences, so a significant, positive marriage coefficient implies the married sibling desists more than the unmarried sibling, even after accounting for genetic similarity.

  • The attenuation pattern suggests a substantial share of the marriage–desistance association may be explained by shared genetic factors (active rGE) rather than a purely causal effect of marriage.

Interpretations and implications

• Main takeaway:

  • There is genetic influence on both marriage propensity and desistance from crime.

  • A substantial portion of the marriage–desistance association can be explained by shared genetic factors; once these are controlled, the association remains but is greatly attenuated (about 60% reduction).

• Theoretical implications:

  • Supports an active rGE account: genetic propensities contribute to the environments individuals select (e.g., marriage), which in turn relate to desistance trajectories.

  • Suggests human agency and life-course turning points (e.g., entering marriage) operate within a genetic–environment framework; genes shape the likelihood of selecting into environments that facilitate desistance, but these environments still matter for behavioral change.

• Practical implications:

  • Interventions aimed at promoting desistance may benefit from considering selection into social bonds and the contexts that support marriages, rather than assuming a purely causal effect of marriage.

• Limitations noted by authors (and relevant for interpretation):

  • Sample restricted to young adults; limited variation in marital status.

  • Wave 3 data reflect early adulthood; desistance measurement may entail measurement error or under-ascertainment of longer-term desistance trajectories.

  • Desistance operationalization as a boundary event vs. process; results robust to alternative coding in supplemental analyses (Wave 3 delinquency index yielded similar findings).

  • Despite genetic controls, omitted variable bias can never be completely ruled out; requires larger covariate sets and possibly different designs.

Methodological notes: measurement and design details

• Add Health sampling design and sibling subsample (Table 1):

  • Inclusion of multiple types of related individuals within households to enable ACE modeling across relatedness levels.

  • Balancing within households to avoid autocorrelation and clustering biases.

• Key measurement specifics:

  • Marital status: Wave 3 dichotomous variable (married: yes/no).

  • Desistance: constructed from Wave 1/2 (adolescence) delinquency index and Wave 3 (adulthood) delinquency index; desisters defined as those with adolescent delinquency but no adult delinquency.

• Statistical techniques:

  • Probandwise concordance: a risk-based concordance metric for twin/sibling data.

  • Tetrachoric correlations: account for dichotomous outcomes and population incidence.

  • ACE modeling: estimates of h2, c2, e2 for dichotomous outcomes under threshold assumptions.

  • DF logistic regression: tests sibling differences while controlling for genetic similarity and environmental differences; coefficients interpreted as effects net of shared genetics.

Additional context and references (selected themes)

• Foundational ideas on marriage and desistance: Laub, Sampson, and colleagues; Sampson & Laub (1993); Laub & Sampson (2003); Laub, Nagin, & Sampson (1998).

• Active rGE and assortative mating as mechanisms linking genetics to environment: Jaffee & Price (2007); Krueger et al. (1998); Johnson et al. (2004); Spotts et al. (2004, 2005).

• Methodological precedents for behavior genetics in social science contexts: Burt et al. (2010); Burt (2009); DeFries & Fulker (1985); Neale & Maes (2004).

• Notes on Add Health design and measurement: Harris et al. (2009); Rowe & Jacobson (1998); Kelley & Peterson (1997).

• Selected empirical findings cited in the article supporting the attenuation of the marriage effect after genetic controls and the existence of genetic influences on both marriage and desistance.

Takeaway for exam preparation

• The marriage desistance link is not purely causal; genetic factors contribute to both the likelihood of marriage and desistance from crime, and active rGE (self-selection into marriage) can inflate observed associations.

• Using genetically informed designs (ACE modeling, DF regression) can disentangle genetic from environmental contributions and reveal the true extent of causal effects.

• Even after accounting for genetics, marriage may still relate to desistance, but the effect is substantially reduced, implying a combination of selection and environment in shaping desistance trajectories.

• When discussing crime prevention and life-course theory, acknowledge the interplay between genetic predispositions, choice of social bonds, and the social context that supports desistance.

Notes on key equations and numerical references (for quick review)

• Probandwise concordance rate: PC = \frac{2C}{2C + D} where C = concordant pairs, D = discordant pairs.

• ACE model components: A = additive genetics (heritability, h^2), C = shared environment (c^2), E = nonshared environment (e^2). For the outcomes in this study:

  • Marital status: h^2 = 0.56,\quad c^2 = 0.00,\quad e^2 = 0.44

  • Desistance: h^2 = 0.49,\quad c^2 = 0.00,\quad e^2 = 0.51

• DF logistic model (genetic control of desistance):

  • \log\left(\frac{P(K1)}{1 - P(K1)}\right) = b0 + b1(K2) + b2(R \times [K2]) + b3(ENVDIF) + e

  • Interpretation: b1 = shared environmental effect (c^2), b2 = genetic effect (h^2), b3 = nonshared environmental effect (e^2).

• Change in marriage effect after genetic control (illustrative values): Model 1 OR ≈ 2.33; Model 2 OR ≈ 1.37; 60% reduction in the effect when genetic factors are controlled (approximate). If using regression coefficients: βModel1 ≈ 0.84 and βModel2 ≈ 0.31; percent reduction ≈ \frac{0.84 - 0.31}{0.84} \approx 0.63 (63%).

• Wave 3 married respondents: n = 687 (18.34% of 3745).

• Desistance: n ≈ 3008 after excluding missing data; ~72% desisters (n ≈ 2151).

• Relatedness levels used in ACE modeling:

  • MZ: a1 = a2 = 1.00

  • DZ/FS: a correlation of 0.50

  • HS: a correlation of 0.25

  • Cousins: a correlation of 0.125

Title for the notes

Notes: Marriage and Desistance From Crime — Gene–Environment Correlation (Barnes & Beaver, 2012)