Table 4.

Model 2 (dynamic): Results of a multilevel multinomial logistic regression analysis predicting student roles at wave 2 by change in teacher interventions while controlling for several demographic and bullying-related factors on the student- and class level

Variable	T2: Bully vs. non-participant				T2 Victim vs. non-participant				T2 Bully-victim vs. non-participant				T2 Defender vs. non-participant
	Est	SE	p	OR	Est	SE	p	OR	Est	SE	p	OR	Est	SE	p	OR
Student level
Gender (1 = female)	−0.557**	0.195	0.004	0.57	0.098	0.232	0.674	1.10	−0.285	0.217	0.188	0.75	0.585*	0.250	0.019	1.80
Ethnicity (1 = non-Czech)	0.314	0.580	0.589	1.37	−0.400	0.770	0.603	0.67	−0.334	0.725	0.645	0.72	0.460	0.606	0.448	1.58
Role (reference: non-participant)
- Bully	3.308***	0.451	<0.001	27.34	0.437	0.797	0.583	1.55	2.338***	0.533	<0.001	10.36	−6.505***	0.542	<0.001	0.001
- Victim	1.972*	0.800	0.014	7.18	4.707***	0.601	<0.001	110.67	3.285***	0.934	<0.001	26.71	3.535***	0.837	<0.001	34.30
- Bully-victim	3.788***	0.588	<0.001	44.17	2.449***	0.700	<0.001	11.57	4.891***	0.789	<0.001	133.10	2.151	1.150	0.061	8.60
- Defender	−0.363	1.257	0.773	0.70	−0.780	0.997	0.434	0.46	−0.905	1.007	0.369	0.41	3.317***	0.374	<0.001	27.59
CLASS LEVEL
Gender (% female)	−0.282	0.187	0.132	0.75	−0.075	0.245	0.759	0.93	0.085	0.199	0.668	1.09	−0.060	0.200	0.763	0.94
Ethnicity (% non-Czech)	−0.304	0.444	0.494	0.74	0.308	0.353	0.383	1.36	−0.431	0.468	0.358	0.65	−0.456	0.399	0.253	0.63
Class role percentage (T1)
- Bullies (%)	0.532	0.355	0.134	1.70	0.838*	0.353	0.018	2.31	1.105*	0.459	0.016	3.02	1.383**	0.463	0.003	3.99
- Victims (%)	0.634	0.543	0.243	1.89	−0.055	0.497	0.912	0.95	0.322	0.460	0.484	1.38	−1.169*	0.559	0.036	0.31
- Bully-victims (%)	0.186	0.149	0.213	1.20	0.410**	0.145	0.005	1.51	0.513**	0.198	0.010	1.67	0.630***	0.131	<0.001	1.88
- Defenders (%)	−0.047	0.178	0.790	0.95	−0.022	0.255	0.932	0.98	0.668***	0.167	<0.001	1.95	0.289	0.303	0.340	1.34
Δ Teacher interventions (T2 − T1)
- Δ Non-intervention	−1.163	1.279	0.363	0.31	2.388*	1.144	0.037	10.89	−1.245	1.252	0.320	0.29	−2.269*	1.027	0.027	0.10
- Δ Disciplinary sanction	−0.062	1.091	0.954	0.94	1.545	1.096	0.159	4.69	−1.424	0.919	0.121	0.24	−1.253	1.834	0.495	0.29
- Δ Group discussion	−0.171	0.610	0.779	0.84	0.345	0.734	0.638	1.41	1.560	0.813	0.055	4.76	1.546*	0.688	0.025	4.69
- Δ Mediation/victim support	0.449	1.342	0.738	1.57	0.116	1.144	0.919	1.12	0.642	1.586	0.686	1.90	2.302	2.190	0.293	9.99
Intercept	−4.015***	0.433	<0.001	0.02	−3.619***	0.35	<0.001	0.03	−4.462***	0.54	<0.001	0.01	−4.622***	0.515	<0.001	0.01
Residual variance	0.503	0.396	0.204	–	0.266	0.341	0.435	–	0.577	0.441	0.191	–	1.016*	0.483	0.036	–
Intraclass correlation (ICC)	0.133	0.091	0.143	–	0.075	0.089	0.399	–	0.149	0.097	0.124	–	0.236**	0.086	0.006	–

Student level N = 750; class level N = 39; AIC = 2777.382; BIC = 3230.149, Loglikelihood H₀ = − 1290.691, H₀ Scaling correction factor for robust maximum likelihood estimation (MLR) = 1.6880. At the student level where the outcome variable is multinomial (student roles), raw estimates (Est) represent log odds. At the class level where the outcome variable is continuous (latent random intercept), estimates represent linear regression slopes. These can be interpreted as log-odds when the log of the odds of student role adoption at student level is considered to be the dependent variable). On the student-level, gender is effect-coded (−1 = male, 1 = female), all other variables are dummy coded; ethnicity has been group-centered; the reference category for student roles is non-participant. On the class-level, the first six variables refer to class composition, and represent class percentages, whereas 1 unit represent 10% (possible range: 0–10); the remaining variables refer to class-aggregated teacher intervention change scores and have been grand-mean centered

*p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001