Table 3.

Multilevel analysis of condom use—OLS regression vs. hierarchical linear model results

Fixed effectsa	OLS estimate (cluster effect ignored)	S.E. (OLS)	t-value	Hierarchical linear model estimate	S.E. (HLM)	t-value
Individual-level
Knowledge	0.063	0.019	3.32*	0.064	0.026	2.46*
Attitude	0.061	0.012	5.08**	0.062	0.018	3.44**
Self-efficacy	0.051	0.017	3.00*	0.05	0.030	1.67
Establishment-level
Establishment rule	0.704	0.113	6.23**	0.705	0.128	5.51**
City-level
Peer education	0.251	0.098	2.56*	0.250	0.110	2.27*
Managerial training	0.628	0.158	3.97**	0.628	0.264	2.38**
Combined	1.306	0.098	13.33**	1.304	0.278	4.69**
Cross-level interaction
Peer × establishment rule	0.421	0.124	3.40**	0.421	.147	2.86**
Managerial × estab. rule	0.563	0.176	3.20**	0.563	.209	2.69**
Combined × estab. rule	0.840	0.238	3.53**	0.840	.325	2.58**
Estimation of variance components for hierarchical models
Variation between GROs					0.79 (S.E. = 0.063)
Variation between establishments					0.31 (S.E. = 0.069)
Variation between cities					0.18 (S.E. = 0.031)*

^aModel also controlled for age, education and marital status and baseline condom use

^*P < 0.05

^**P < 0.01