Page 391 - Contributed Paper Session (CPS) - Volume 2
P. 391
CPS1899 Ira Joy H.M.
E. Media
F. Friends and peers
G. Health and lifestyle
H. Marriage
I. Puberty, dating, and sex
J. Fertility and contraception
K. Knowledge and attitude towards marriage, sex and related issues
L. Reproductive health
For each block except blocks for marriage and reproductive health, several
data items are included in the questionnaire that are analyzed to determine
the factors that significantly affect the behavior of drug use, these data Items
are listed in Table 1 of the Appendix. Using the weights that was calculated by
the UPPI and DRDF statistical division, the YAFS 4 data set was weighted to
consider the multi-stage sampling design of the survey. The weighted
percentage distributions were used in describing the demographic and socio-
economic status of the respondents and then the response variable (ever used
drugs) was cross-tabulated with urbanity, region, respondent type, and age to
further describe the data set.
The Rao-Scott Chi-square test, an adjusted version of the Pearson Chi-
square test for association between two variables that deals with the
correction of the survey design, was used to find significant variables that
could be associated with drug use to account for the complex design of the
YAFS survey. Significant variables were then included in the construction of
the model. In addition to this, Cramer’s V was computed using the computed
Rao-Scott Chi-square to measure the strength of association among the
weighted variables.
To generate the final model binary logistic regression analysis, which is
considered to be an important tool used to analyze the relationship between
several explanatory variables and the qualitative response variables, was fitted
for the likelihood to engage in drug use. The stepwise regression model was
used in generating the possible best model.
Several tests were performed to assess the model. First, the Likelihood
ratio test was used to assess if the generated model is better than the model
without the predictors. Second, the receiver operating characteristic (ROC)
curve was used to asses if the model has a good predictive power or if the
model can predict new cases given the predictors. Third, the Hosmer-
Lemeshow goodnessof- fit test was used to asses if model is well-fitted. Lastly,
classification tests were used to assess if the model can correctly identify those
who actually use drugs (sensitivity) and if the model can correctly identify
those who actually do not use drugs (specificity). In addition to this,
multicollinearity and parsimony were also considered. The odds was estimated
380 | I S I W S C 2 0 1 9