TABLE 5.

Estimates of the Odds Ratios Describing the Association Between Disruptive Behaviors at 41 Months of Age, and the Same Behaviors at 17 Months of Age, and the Child’s Gender

Disruptive Behavior	Child’s Gender		Disruptive Behavior at 17 Months of Age
	Odds Ratio† [99% CI]	%of Variance Explained/Effect Size/PAF	Odds Ratio‡ [99% CI]	%of Variance Explained/Effect Size
Opposition-defiance
1. Defiant	1.42 [1.06, 1.90]b*	0.7/.07/.14	1.43 [1.18, 1.74]abd	2.2/.19
2. Didn’t feel guilty	1.34 [1.09, 1.65]a*b*	1.2/.09/.22	1.28 [1.10, 1.49]acd	1.8/.13
3. Didn’t change behavior	1.60 [0.97, 2.65]b*	0.7/.06/.25	1.65 [1.32, 2.07]ad	3.3/.13
Inattention
4. Inattentive	1.17 [0.92, 1.48]a*b*	0.2/.06/.13	2.15 [1.80, 2.56]ad	3.2/.21
5. Easily distracted	1.36 [1.10, 1.67]a*b*	1.5/.10/.22	1.76 [1.37, 2.27]ad	3.4/.20
Hyperactivity
6. Restless or hyperactive	1.68 [1.21, 2.34]b*	2.0/.14/.27	Boy: 2.85 [1.76, 4.62]a	Boy: 8.6/.31
			Girl: 1.76 [1.40, 2.21]abcd	Girl: 7.6/.35
7. Fidgets	1.44 [1.003, 2.06]b*	0.9/.10/.23	1.41 [1.18, 1.68]abd	3.4/.26
8. Difficulty waiting	1.75 [1.24, 2.45]b*	2.2/.11/.25	1.57 [1.30, 1.88]acd	5.2/.21
Physical Aggression
9. Fights	1.51 [1.17, 1.95]a*	1.3/.11/.19	2.89 [2.07, 4.03]acd	6.2/.31
10. Attacks	1.30 [1.02, 1.67]a*b*	0.4/.08/.26	2.26 [1.69, 3.03]abd	3.1/.27

Note. The percentage of variance due to gender after taking into account gender differences in the same behavior at 17 and 29 months of age was estimated as [L²(BD, CD) − L²(BD, CD, GD)]/L²(equiprobability model). Similarly, the percentage of variance due to the same behavior at 17 months of age after taking into account the association between the behavior in question at 29 and 41 months of age was estimated as [L²(GD, CD) − L²(BD, CD, GD)]/L²(equiprobability model). In both formulas, the equiprobability model was used as a benchmark. Under this model, the response categories—never, sometimes and often—are equiprobable. Effect size was estimated using Cohen’s (1988) w statistic. PAF refers to the population attributable fraction for the child’s gender. It was estimated as the proportion of male children who exhibit a particular disruptive behavior on a frequent basis, minus the proportion of female children who do so, over the proportion in question among all children in the population.

^a*Refers to the boy/girl ratio of the odds of exhibiting a particular disruptive behavior sometimes rather than never;

^b*Refers to the boy/girl ratio of the odds of exhibiting a particular disruptive behavior often rather than sometimes;

^aRefers to the odds of exhibiting a particular disruptive behavior at 41 months of age sometimes rather than never for children who did exhibit the same behavior sometimes rather than never at 17 months of age;

^bRefers to the odds of exhibiting a particular disruptive behavior at 41 months of age often rather than sometimes for children who did exhibit the same behavior sometimes rather than never at 17 months of age;

^cRefers to the odds of exhibiting a particular disruptive behavior at 41 months of age sometimes rather than never for children who did exhibit the same behavior often rather than sometimes at 17 months of age;

^dRefers to the odds of exhibiting a particular disruptive behavior at 41 months of age often rather than sometimes for children who did exhibit the same behavior often rather than sometimes at 17 months of age.

^†These estimates were obtained from a restricted version of the selected baseline hierarchical loglinear model. Six restricted versions were considered using a coding scheme (Galindo-Garre, Vermunt, & Croon, 2002; see also Galindo-Garre & Vermunt, 2005) wherein equality restrictions were imposed between the three local log-odds ratios a*, b*, and c* in the 2 × 2 subtables formed by considering, respectively, the never and sometimes, the sometimes and often, and the never and often rating categories. More specifically, three restricted models were obtained by imposing equality restrictions between pairs of log-odds ratios (i.e., a* = b*; a* = c*, equivalent to b* = 0; b* = c*, equivalent to a* = 0) and three other models were obtained by imposing equality restrictions between one log-odds ratio and the inverse of another log-odds ratio (i.e., a* = 1/b*, equivalent to c* = 0; a* = 1/c*; b* = 1/c*). [Note that these models included, as special cases, the loglinear models (i.e., uniform and row-effect) proposed by L.A. Goodman (1979) for the analysis of association in cross-classifications having ordered categories.] The restricted model with the smallest likelihood-ratio chi-square statistic was chosen. Note that this restricted model did not represent a statistically significant decrease in fit over the selected baseline model.

^‡These estimates were obtained from a restricted version of the selected baseline hierarchical loglinear model. Many restricted versions were considered using the same coding scheme described earlier wherein equality (including equality to 0) restrictions were imposed between the four local log-odds ratios, a, b, c, and d in the 2 × 2 subtables formed from adjacent rating categories (Clogg & Shihadeh, 1994). The log-odds ratio a involved never and sometimes for both time points. Similarly, the log-odds ratio d involved sometimes and often for both time points. The log-odds ratio b involved never and sometimes at 17 months of age, and sometimes and often at 41 months of age. The contrary was true for the log-odds ratio c. More specifically, 10 restricted models were obtained by imposing one equality restriction (i.e., a = b; a = c; a = d; b = c; b = d; c = d; a = 0; b = 0; c = 0; d = 0). Twenty-five restricted models were obtained by imposing two equality restrictions (i.e., a = b = 0; a = c = 0; a = d = 0; b = c = 0; b = d = 0; c = d = 0; a = b & c = d; a = c & b = d; a = d & b = c; a = b = c; a = b = d; a = c = d; b = c = d; a = 0 & b = c; a = 0 & b = d; a = 0 & c = d; b = 0 & a = c; b = 0 & a = d; b = 0 & c = d; c = 0 & a = b; c = 0 & a = d; c = 0 & b = d; d = 0 & a = b; d = 0 & a = c; d = 0 & b = c). Finally, 15 other models were obtained by imposing three equality restrictions (i.e., a = b & c = d = 0; a = c & b = d = 0; a = d & b = c = 0; b = c & a = d = 0; b = d & a = c = 0; c = d & a = b = 0; a = b = c & d = 0; a = b = d & c = 0; a = c = d & b = 0; b = c = d & a = 0; a = b = c = d; a = b = c = 0; a = b = d = 0; a = c = d = 0; b = c = d = 0) [Note that these models included, as special cases, some of the loglinear association models (i.e., uniform, row-effect, and column-effect) proposed by L.A. Goodman, 1979.] For each set of models, the restricted model with the smallest L² was chosen, and the resulting three models were compared among themselves to find the most parsimonious model for the data. Note that this restricted model did not represent a statistically significant decrease in fit over the selected baseline model.