Page 58 - IJPS-3-2

Zewdie S A and Adjiwanou V

All the socioeconomic variables considered at level-one of the model are significant at p<0.05 levels and are in
agreement with the results from other researches (Mosley and Chen, 1984; Hobcraft, 1993; Sastry, 1996; Kabir, Islam
et al., 2001) as well as our child mortality estimation results presented above. For instance, children of black African
mothers have a higher risk of death as compared to other population groups, while those who are from better educated
mother have much lower risk of death. As expected, the living standard (LS) index, which is used as a measure of poverty,
is highly significant too in that the higher the LS index of the household where the child lives the less likely the risk of
dying. Note that because of multicollinearity issue the household income poverty indicator variable is not included in
the model–the LS index and income poverty have a strong correlation coefficient of 0.85. Hence, in relation to poverty
the result could be interpreted as, for example, children in the least poor and the second least poor household have
more than 24% and 14% chance of dying respectively as compared to those living in the poorest households. Children
living in municipalities where there is higher level of income poverty and inequality have greater likelihoods of death.
Similarly, average years of schooling of women at municipal-level significantly affects child survival positively, whereas
higher women HIV prevalence rate of provinces is highly related with higher risk of death of children as one expects.
Poverty, inequality and women’s education of the municipalities affected the odds of infant death by 8% 13% and -21%
respectively, while the impact of province level HIV/AIDS is estimated to be 26. The results of the regression model also
indicate small but statistically significant residuals which can convey province-level and municipal-level random effects
on the risk of dying, even after controlling for a range of child-level, municipal-level, and province-level variables.
The study examined a comprehensive array of multilevel risk factors for infant mortality in South Africa by giving
more attention to provincial and municipal variation. The modelling and estimation strategies utilized are appropriate
and supported by large number of observation from the latest available census data. The use of multilevel analysis helped
to understand the impact of some community level infant mortality risk factors of infant. Developmental indicators are
likely to still vary significantly across the municipalities and also their effects on infant mortality. Hence, for further
improvements, there is a need to focus on municipal level planning as well instead of only at the national or province
level. Taking into account the findings under the present study, for a data involving hierarchical structure, there is a
need to emphasize the use of the possible highest levels in hierarchical models. To further emphasize, such optimal
considerations may provide additional important clues to policy planners leading to optimal use of available resources
regarding public health programs.
Although the study achieved its objectives, there were some unavoidable limitations. First, our analyses were primarily
based on the 2011 South African census data. Therefore, the significance and reliability of the results depends on the
quality of the census data which includes the quality of enumeration and data processing. Any defect in the census data
might seriously impact the results in the research. Second, the cross-sectional nature of this study design cannot us
determine the causal relationship between independent variables and the outcome variable. The use of longitudinal data
would have been much better. Third, the assumption that under reports of birth and deaths in the census data are the
same. Fourth, the unavailability of municipal level HIV prevalence rates, and hence the assumption that these prevalence
rates are the same as the rates at the respective provinces. Ignoring the HIV prevalence variation within provinces might
especially impact the results of the regression model to some extent.
5. Conclusions

The main objective of the research was to investigate infant mortality risk factors with special emphasis on the impact of
poverty and inequality. The results from the multilevel logistic regression model suggest that most of the demographic
and socioeconomic factors as well as the province and municipal level random effects are significant. The significant
predictors at individual-level include birth order and sex of the child, education, employment status, race, and marital
status of the mother, and living standard of the family. These factors can bring from -24% to 15% change on the odds of
death of infants. The changes in odds of infant death due to municipal and province level effects are estimated to be 12%,
-11%, 7%, and 3% respectively for the HIV prevalence women’s education, inequality and LS poverty. This implies that
communities with better living standard and women education are associated with lower infant mortality rates, while
higher income inequality and HIV prevalence in the communities depict higher levels of infant mortality. In addition,
unobservable municipal and province level random effects significantly affect the level of infant mortality rates. The
study helps to analyze infant mortality in the country by taking the hierarchical nature of the theme into account and to
investigate the with-in country variation of infant mortality.
Authors’ Contribution

SA Zewdie designed the study, prepared the data, performed the analysis, drafted and revised the manuscript, and

International Journal of Population Studies | 2017, Volume 3, Issue 2 51