We extended the biosocial model of problem behavior tested by Udry (19
90) to a panel design, following a sample of over one hundred boys in
adolescence for three years. We found the expected results for sociolo
gical variables, but weaker effects for testosterone than Udry found o
n cross-sectional data. Using panel models with lagged hormone effects
, we identified relationships between Time-1 testosterone and problem
behavior one year or more later. The relationship between testosterone
and problem behavior was not present for subsequent measures of testo
sterone, either in cross-section or with time-lagged models. Therefore
we cannot interpret the results as showing testosterone effects on pr
oblem behavior. Rather it appears that testosterone level in early ado
lescence is a marker for a more general growth trajectory of early dev
elopment.