This research reviewed the "Smoking, Drinking and Drug Use among Young
People in England" 2010 survey (the Year 2010 Survey) study in terms of its
data collection, processing and analysis. The research aim was to gain increased
understanding of young people’s drug-trying behaviour in England through
appropriate handling of missing data, as well as, to build upon the previous
work done, developing and applying statistical methodologies for analysis of
multivariate categorical data collected by the Year 2010 Survey study.

The main work done in this research included: (1) modifying the original data
set to arrive the useful working data set; (2) conducting exploratory data analysis
with the working data set to identify direction for further empirical investigation;
(3) properly handling the missing data problem in the working data set
and (4) developing and applying advanced statistical methodologies to further
analyse the working data set.

Apart from supporting the main findings of the Year 2010 Survey study that
smoking, drinking and some drug-related socio-demographic covariates were positively associated with the students’ drug-trying behaviour, additional significant results found by the univariate logistic regression models, log-linear
analysis models, two-parameter item response theory models and latent class
analysis models reported that (1) the 15 drugs were highly and positively associated with each other and each drug exerted different extent of influences
on the students’ drug-trying behaviour and (2) generally, students’ drug-trying
behaviour could be further explained by numerous smoking, drinking and drug related socio-demographic factors at different extent.

These additional findings contributed to a deeper understanding of the drug use
problem, added evidence to the drug related research literature and provided
helpful guidance on formulating policies to combat against drug use problem in
England. Another contribution of this research was the development of a new
methodology for backward elimination of latent class analysis models which
provided a more thorough evaluation of the optimal number of latent class and
covariate elimination from saturated model.