who correctly answered three items. To account for clustering of individual observations by school, a random
effects ordinal regression model (PROC GLIMMIX) was
initially applied and the COVTEST statement was used to
test for differences by school. Where the covariance
parameter was not significant, standard ordinal regression
analysis (PROC LOGISTIC) was conducted. In each case,
the proportional odds assumption was used to test whether
the ordinal regression model was appropriate. The analyses
were conducted with backward selection, initially including
all independent variables significant in bivariate analyses
(significance set at p B 0.05). Thereafter, non-significant
variables in the multivariable model were removed stepwise.
The demographic variables age, gender, migrant background
and school education of mother were maintained in all
analyses as they have been shown to be associated with
knowledge and awareness of STDs among adolescents in
previous studies.