TABLE 3—

Logistic Regression Predicting the Odds of Early Intercourse: Northwestern Quebec, 1987–1994

	Change in F (df)	Change in R2	Odds Ratio	t
Step 1	16.61 (6)	0.04†	. . .	. . .
Gender	. . .	. . .	3.08**	2.44
Pubertal status	. . .	. . .	2.95†	4.42
Antisocial behavior	. . .	. . .	0.79	0.94
Peer rejection	. . .	. . .	0.94	0.33
Verbal abuse by teacher	. . .	. . .	1.99***	2.88
Step 2	11.83 (2)	0.07†	. . .	. . .
Delinquent behavior	. . .	. . .	2.54†	4.01
Self-esteem	. . .	. . .	0.80	1.08
Step 3a	8.66 (1)	0.02***	. . .	. . .
Interaction: Delinquent behavior × gender	. . .	. . .	0.33**	2.45
Step 3b	7.29 (1)	0.02***	. . .	. . .
Interaction: Self-esteem × gender	. . .	. . .	2.58**	2.16

Note. Instead of the χ² statistic provided in standard logistic regression, the SAS procedure Proc MIANALYZE provides an F test as a multivariate inference for all the predictors of the multiple logistic regression. As in linear regression, the F statistic indicates whether the averaged explained variance is significant. Gender is coded such that 0 = girls and 1 = boys. The no–sexual intercourse group served as the comparison group for model tests and odds ratios. Only significant interaction terms are shown. For details on how each variable was assessed, see “Methods” section.

**P < .05; ***P < .01; †P < .001. All were 2-tailed tests.