Multivariate Logistic Regression Models
I presented bivariate analyses for understanding the individual relationship between
child mortality and each of independent variables in multivariate analyses. Some findings
in terms of rural-urban differences, marital status, and educational attainment did not
support previous research. Simultaneously, there are paradoxes when the associations
between ethnic background and other mothers' characteristics were employed. To
examine the net of these various factors' influence on the probability of child mortality,
the relationships between child mortality and indicators by means of the multivariate
logistic regression analysis, which allows us to obtain more definite comparisons between
the two ethnic groups in terms of child mortality is explored next.
Ethnic Influence on Child Mortality
Table 4-3 presents five nested logistic regression analyses. Model 1 includes only
ethnicity. The coefficient for ethnicity is .428. By taking the antilog of the coefficient of
ethnicity, the probability that African mothers have lost at least one child is 1.534 times
more than East Indian mothers. Demographic factors and socioeconomic factors are
introduced into Model 2 and Model 3 respectively. The coefficients for ethnicity in the
two models indicate that the East Indian division consistently has a lower probability of
having experienced a child loss than the African division after controlling for
demographic factors in Model 2, and after controlling for socioeconomic factors in
Model 3. Model 4 includes all explanatory variables. The odds ratio of Africans increases
to 2.177.
The non-significant factors are removed from the model to produce the best fit with
the fewest variables, and this final model is presented in the last column. Differences
between the chi-squares reported for Model 4 and Model 5 are not significant (/2=2.727