A regression model is used to analyse the determinants of excess child mortality among Aboriginal Australians. The independent variables examined in this study are:

The age of the mother is also introduced as a control variable to indirectly capture the effect of duration of exposure on mortality of children. In the regression, with the exception of maternal age (which is treated as continuous), for each of the remaining explanatory variables, one of the categories is taken as a reference category.

The model was fitted using the Poisson regression technique with the number of children who have died as the dependent variable and the number of live births to the mother as the exposure variable. This method is appropriate for the available data given the fact that the dependent variable has only non-negative integers (Cameron & Trivedi 2001; Gurmu & Trivedi 1996; Long 1997). In the course of the analysis, other variants of the count data model (such as the negative binomial regression model and the zero and hurdle variants of Poisson and negative binomial regression) were also tested, but as these did not produce a statistically superior outcome, the results are not reproduced here.

Table 6.2 presents the effects of socioeconomic and other background variables on child mortality estimated using the Poisson model. While interpreting these results, it is important to note that as death is a biological process, factors affecting child mortality in the most direct manner are bio-medical in nature; background variables such as those identified in the present study impact on child mortality only in an indirect manner. Another important limitation of the present analysis is that almost all the explanatory variables are contemporary variables relating to the time of the survey and these are used to ‘predict’ an event (the loss of a child) that happened in some distant past. In addition, some of the variables (such as drinking, smoking, labour force status, marital status and presence of stress) may indeed have been shaped by the negative shock to one’s life that can occur through the death of a child. Ideally, these processes entail a dynamic analysis, but this is limited in the present study due to the nature of the data. However, from the point of view of future data collection, one way of addressing this problem in a cross-sectional survey such as the NATSISS is by collecting birth history data. This enables tracing the timing of the death of children which could then be linked with the characteristics of the mother at the time of the event.

The multi-variate regression presented in Table 6.2 shows that better environmental quality and home ownership have positive and statistically significant association with the risk of dying in childhood. The result shows that the odds of child mortality increase by almost 53 per cent for children in community housing and by some 72 per cent for children in private rentals. These differentials, which persist when the effects of other factors are controlled, underscore the importance of a family’s economic standing in determining the probability of survival of its children. Moreover, as both size and quality of housing facilities are often correlated with household income/wealth level, those in privately-owned premises are likely to enjoy better and well maintained facilities, and hence be able to minimise or eliminate the chance of environmental exposure of their children to infectious agents. This is also confirmed by the result of the regression result which shows a statistically significant association between child mortality and adequacy of lavatory facilities. These results are consistent with the UN (1985) study which showed that, in general, old housing and deficient sanitary conditions constitute risk factors for child survival.

Table 6.2. (continued)

a. The model includes children ever-born as an exposure variable. R refers to reference category.

Source: Author’s calculations from the 2002 NATSISS MURF conducted at the ABS

Neighbourhood problems, existence of stress and high-risk drinking problems among parents increase the likelihood of child mortality. The odds of child mortality are 61 percent higher for children born and raised in a neighbourhood that has problems, while a high-risk drinking problem and the existence of stress within a household each increase the odds of child mortality by some 70 per cent. Similarly, children of never-married mothers and mothers who have difficulty in dealing with service providers show respectively a 35 and 12 per cent elevated risk of mortality compared to their counterparts who live with both parents or live with a mother who has no difficulty in accessing services. A strong and statistically significant association is also evident between maternal age, child mortality and degree of attachment to and residence on their homeland.

However, maternal education, place of residence and labour force participation, which are known to have a strong association with mortality in the literature do not show significant association with child mortality in the present analysis. Other factors that were also found to be not significant were composition of household and smoking behaviour of the mother. These findings may suggest that the relationships between residential pattern and socio-occupational status and child mortality are not direct but operate through other variables.