Table 2.

Negative binomial regression results of baseline alcohol frequency, quantity, and problems as a function of baseline relationship status and covariates

	Frequency (n = 1,462)			Quantity (n = 1,420)			Problems (n = 1,634)
Variable	B (IRR)	SE	z	B (IRR)	SE	z	B (IRR)	SE	z
Intercept	0.51 (1.67)	0.13	4.00	0.46 (1.58)	0.13	3.41	-1.59 (0.20)	0.18	-8.92
Sex (1 = female)	-0.12 (0.89)	0.09	-1.34	-0.23 (0.79)	0.09	-2.53	-0.09 (0.91)	0.11	-0.79
When took survey (1 = after school started)	0.11 (1.12)	0.11	1.05	0.22 (1.25)	0.11	1.98	0.31 (1.36)	0.15	2.05
Parental alcohol problems (1 = yes)	0.18 (1.20)	0.09	1.92	0.04 (1.04)	0.10	0.46	0.23 (1.26)	0.11	2.07
High school conduct problems	0.05 (1.05)	0.01	3.24	0.02 (1.02)	0.01	1.31	0.08 (1.08)	0.02	4.91
Peer deviance (past year)	0.09 (1.09)	0.01	15.62	0.09 (1.09)	0.01	14.43	0.08 (1.08)	0.01	10.51
Extraversion	0.05 (1.05)	0.01	6.65	0.04 (1.04)	0.01	5.68	0.04 (1.04)	0.01	4.08
Baseline relationship status: Exclusive	Set as reference
Baseline relationship status: Single	-0.09 (0.91)	0.09	-1.01	-0.13 (0.88)	0.09	-1.37	-0.09 (0.91)	0.11	-0.78
Baseline relationship status: Dating several	0.17 (1.19)	0.20	0.86	-0.03 (0.97)	0.22	-0.15	0.07 (1.07)	0.22	0.31

Notes: Bold type indicates p < .05. Bold italic type indicates p < .01. B = unstandardized regression coefficient; IRR = incidence rate ratio; SE = standard error of unstandardized regression coefficient; z = z score associated with the coefficient. The intercept for negative binomial models is equivalent to the log of the expected count for the outcome variable when all other predictors in the model are set to zero. The regression coefficients for negative binomial models indicate the degree and direction of the difference in the logs of the expected counts for the outcome variable for each one-unit change in the predictor variable. Thus, positive coefficients denote that a one-unit change in a predictor variable is associated with an increased expected count, and negative coefficients denote that a one-unit change in a predictor variable is associated with a decreased expected count. The IRR indicates the factor by which a one-unit change in a predictor variable increases (for IRR > 1) or decreases (for IRR < 1) the expected count. The IRR for the intercept is the expected count for the outcome variable when all other predictors in the model are set to zero.