The Determinants of Infant, Child and Maternal Mortality in Sub-Saharan Africa

Abstract

In Sub-Saharan Africa, infant, child and maternal mortalities are very high compared to other regions. We estimate a cross-country empirical model of the determinants of those mortalities. We find, similar to other studies, that in addition to per capita GDP, health and education interventions can affect mortalities, however, the effect depends on the mortality rate being modelled. Importantly, the prevalence of the adult HIV/AIDS infection rate is detrimentally impacting mortality in the Sub-Saharan region.

Notes

1 The differences in the UNESCO figures and those reported by Harch (2002) and Todaro (2000) are difficult to completely reconcile, although, clearly, the data sources are different. In particular, the UNESCO literacy data is only for ages 15 to 24, while Todaro's data is for a larger segment of the population.

2 Rate calculated from Children Data Bank (http://childrendatabank.org/international/mortality/mortality1.html) as an average across countries, rather than a population weighted average.

3 The result of the 1995 Uganda demographic survey shows that infant and child mortalities were declining in the 1990s. In 1995, infant and child mortalities per 1000 live births were reported at 81,3 and 147,3, respectively. Five years before the survey, however, infant and child mortalities were 92 and 167,2.

4 Data was presented graphically in the report; therefore, the numbers are sight estimates from the graphs.

5 Though the percentage of births attended by trained health personnel was assumed to have a significant impact, it could not be included in their estimation due to missing data. However, the primary school enrolment ratio was found to be insignificant and dropped from the model.

6 Some possible explanatory variables are excluded from the model on account of missing data.

7 The countries and regions are listed in the footnote to Table 1.

8 The war dummy is equal to one if the sample country was directly or indirectly engaged in war during the last decade (1989 to 1999), otherwise it is zero.

9 Unfortunately, the completeness and accuracy of the available data cannot be verified. Clearly, if all of the variables are measured with error, then there is little that can be done to improve the quality of the estimates, other than wait for better data in the future.

10 Although the data is presented by sub-region in the table, regional dummy variables, used in the initial specifications of the model, had negligible explanatory power and were removed from the empirical models.

11 Where i = 1, 2, 3, ....38 (countries in the sample are listed at the bottom of Table 1). * HEALTHCARE is proxied by PBA_TP and PPP in IMR and CMR, respectively.

12 The problem of heteroskedasticity is more common in cross-sectional data. The presence of heteroskedasticity in the disturbance of an otherwise specified model leads to consistent but inefficient parameter estimates. This lack of efficiency makes the usual hypothesis testing procedures dubious. A heteroskedasticity-consistent covariance matrix estimator, therefore, provides correct estimates of the coefficient covariances in the presence of heteroskedasticity of unknown form

13 Jarque-Bera is a test statistic for testing whether a data series is normally distributed. Under the null hypothesis of a normal distribution, the Jarque-Bera statistic is distributed as chi-square with 2 degrees of freedom. If the p value of the computed chi-square in an application is sufficiently low, one can reject the hypotheses that the residuals are normally distributed; otherwise the normality assumption couldn't be rejected.

14 Reset stands for Regression Specification Error Test and was proposed by Ramsey (1969). Reset is a general test for the following types of specification errors: Omitted variables, incorrect functional form and correlation between the independent variables and the error terms. Under such specification errors, OLS estimators will be biased and inconsistent, and conventional inferences procedures will be invalidated. The null hypothesis in Reset test is that the model is correctly specified.

15 Where i = 1, 2, 3, ....38 (countries in the sample are listed at the bottom of Table 1). * HEALTHCARE is proxied by PBA_TP.

16 Due to the 100 000 denominator in maternal mortality, a reduction of 1000 maternal deaths is a reduction in deaths of 10 per 1000, which was the base used for the infant and child mortality results discussion, above. The maximum maternal death rate used in the analysis was only 1500, and, therefore, a reduction, by 1000, in maternal deaths would be an unlikely prospect.