FAQ & Glossary

Common questions and key terms.

Frequently Asked Questions

Latent mean difference in general cognitive ability between males and females, expressed in Cohen’s d units (standard deviations). Extracted from a multi-group confirmatory factor analysis. Positive values indicate higher male mean.

Variance ratio: male variance divided by female variance. Values above 1 indicate greater male variability. This is computed from the latent factor variances in the CFA model.

An observed (not latent) composite score: the mean of all age-residualized, standardized subtests for a given respondent. Unlike dg, g_proxy does not require measurement invariance to hold. It is used for all exploratory analyses.

The confirmatory pipeline requires scalar measurement invariance before reporting latent mean differences. NLSY97 fails the scalar gate (ΔCFI too large). CNLSY fails on both ΔCFI and ΔRMSEA. For these cohorts, only the observed g_proxy difference is reported.

The family bootstrap resamples entire families (not individuals) to account for within-family correlation, then refits the full SEM on each replicate. 499 replicates produce percentile confidence intervals.

Three: NLSY79 (born 1957–64, tested 1980), NLSY97 (born 1980–84, tested 1999), and CNLSY (children of NLSY79 women, tested 1986–2014 on PIAT).

Not directly. NLSY79 uses 10 ASVAB subtests, NLSY97 uses 12 CAT-ASVAB subtests, and CNLSY uses 3 PIAT subtests. Cross-cohort comparisons use g_proxy (observed composite) rather than latent dg.

All outcome association analyses (education, earnings, health, etc.) use the observed g_proxy composite and are classified as exploratory. Only the measurement invariance–gated dg and VRg are confirmatory.

Sex differences are broken out by parental education tercile (low/mid/high) to test whether the magnitude of sex differences varies with socioeconomic background.

Yes. Full pipeline under MIT License. Raw NLSY microdata is not redistributed; researchers access it through the NLS Investigator.

Glossary

Term Definition
g General cognitive ability — the common variance across cognitive tests
dg Latent sex difference in g, Cohen’s d units
VRg Variance ratio (male/female) of latent g
g_proxy Observed composite: mean of standardized, age-residualized subtests
ASVAB Armed Services Vocational Aptitude Battery
PIAT Peabody Individual Achievement Test
CFA Confirmatory factor analysis
SEM Structural equation model
FIML Full information maximum likelihood (handles missing data)
MLR Maximum likelihood with robust standard errors
Measurement invariance Whether the test measures the same construct equivalently across groups
Metric invariance Equal factor loadings across groups
Scalar invariance Equal factor loadings AND intercepts across groups
Bootstrap Resampling with replacement to estimate sampling distributions