Measurement: Sex Differences in g

Latent and observed estimates of sex differences in general cognitive ability across three NLSY cohorts, with bootstrap confidence intervals from 499 family-clustered replicates.

Latent dg (Bootstrap)

dg is the latent mean difference in Cohen's d units, estimated via multi-group confirmatory factor analysis. Of the three cohorts, only NLSY79 passes both metric and scalar measurement-invariance gates. NLSY97 fails scalar invariance (delta CFI exceeds the cutoff), and CNLSY fails scalar invariance on both delta CFI and delta RMSEA. Results for NLSY97 and CNLSY are therefore excluded from confirmatory reporting and shown here for descriptive purposes only.

Latent dg by Cohort (Bootstrap)

Horizontal bars show point estimates; whiskers show 95% bootstrap CI

Note: NLSY97 and CNLSY are gated/excluded from confirmatory reporting due to measurement-invariance failures.

Cohort dg SE 95% CI Invariance Status
NLSY79 0.305 0.190 [−0.11, 0.36] Passed (confirmatory)
NLSY97 0.017 0.023 [−0.05, 0.04] Failed scalar (ΔCFI)
CNLSY 0.122 0.149 [−0.25, 0.28] Failed scalar (ΔCFI, RMSEA)

Observed g-proxy (Bootstrap)

g-proxy is a standardized observed composite score. Because it does not depend on latent-variable modeling, it is available for all three cohorts regardless of invariance status.

Observed g-proxy Sex Difference

Cohen's d (male − female), with 95% bootstrap CI

Cohort d SE 95% CI
NLSY79 0.263 0.017 [0.178, 0.247]
NLSY97 0.062 0.026 [0.018, 0.114]
CNLSY 0.409 0.117 [0.076, 0.540]

Variance Ratios (Bootstrap)

VRg is the ratio of male to female variance on the g-proxy composite. Values greater than 1.0 indicate greater male variability.

Observed g-proxy Variance Ratio

VR = Var(male) / Var(female); dashed line at 1.0 marks equal variance

Cohort VRg-proxy SE(log VR) 95% CI
NLSY79 1.358 [1.286, 1.426]
NLSY97 1.317 [1.263, 1.376]
CNLSY 1.356 [0.890, 1.962]

Latent Variance Ratios

Latent VRg

VR from multi-group CFA; dashed line at 1.0 marks equal variance

Greater male variability is consistent across all three cohorts for both latent and observed measures, with VR ranging from 1.12 to 1.36.