Page 32 - IJPS-2-1

Monitoring adult mortality by type of residence in the absence of death registration: a perspective from Burkina Faso

five-year age groups whose mother (father) is still alive into survivorship ratios using a set of coeffi-
cients (obtained from simulations). The mean age at childbearing of women (men) is used to control
for variations in the fertility schedule, which affects the exposure time. Survivorship ratios are
then converted into summary indices of adult mortality using one parameter of a relational logit
model table (Moultrie, Dorrington, Hill et al., 2013). Finally, estimates are located in time under the
assumption that mortality trends have been linear. Coefficients used in this paper to convert propor-
tions into survivorship ratios are those proposed by Timæus (1992), and the time location procedure
is the method developed by Brass and Bamgboye (1981). In addition, since DHS and EMUIB data
are samples of the entire population, confidence intervals are required to statistically compare levels
of adult mortality between urban and rural areas. Orphanhood estimates are seldom presented
with confidence intervals, and there is no standard way to obtain them. Here, I computed 95% boot-
strap-based confidence intervals of estimates in each place of residence (2000 replicates). The boot-
strap technique has proven useful over the years to estimate robust confidence interval without mak-
ing strong assumptions about the distribution of estimates (Efron and Tibshirani, 1993). To the best
of my knowledge, this method has never been used to compute confidence intervals based on or-
phanhood data.
The estimates derived from the DHS and EMUIB data were obtained from the survival of parents
of young children (5–9, 10–14 years old), and young adults (15–19, 20–24, 25–29 years old) respec-
tively. I also applied the orphanhood method on data collected during the 2006 census and the mul-
tiple indicator cluster survey (MICS) conducted in 2006. The corresponding results are not com-
mented in the main text because there are in line with estimations derived from EMUIB and DHS
data (Appendix, Table A1 and Figure A1).
The orphanhood method does not assume that the population is closed to migration, but a major
issue in applying the method with data disaggregated by place of residence is the lack of information
on the urban/rural status of parents. In DHS surveys, only the place of residence of children at the
time of the survey is known. I used this information as a proxy for the parents’ place of residence at
the time of the survey or at the time of death. By contrast, in the EMUIB survey, it is possible
to correct for this and assess the impact of misclassification of parent’s place of residence on differ-
ences in urban/rural adult mortality. For this survey particularly, to investigate the variation in the
quality of data by place of residence that may affect the estimates, I compared by urban/rural loca-
tion and for each age group, the proportion of surviving parents, reported by men and women (Ap-
pendix, Table A2).
2.3.3 Estimating Adult Mortality from Sibling Survival Data
Unlike the two methods presented above, sibling survival data provide an opportunity to “directly”
estimate adult mortality rates. With the information provided by each interviewed woman (15 to 49
years) on her siblings, it is possible to compute mortality rates by dividing the number of deaths by
the population at risk for a given period and age group. However, because adult mortality is a rela-
tively rare event, and sample sizes in DHS are too small to derive age- and period-specific estimates
without introducing some smoothing, mortality rates were derived from a quasi-Poisson model for
this analysis. The data file was reshaped in person-periods and the dependent variable was the num-
ber of deaths. The age group, sex, and place of residence (urban/rural) were used as explanatory va-
riables. This approach, introduced by Timæus and Jassey (2004), also generates confidence intervals.
As suggested by Masquelier (2013), no attempt was made to weigh the data to account for selec-
tion biases, and I assume that mortality does not vary with the number of adult siblings.
Although the sibling survival method does not rely on many assumptions, the quality of data is an
issue, particularly the underreporting of deaths due to recall biases. Evidence abounds of decay in
the completeness of death reporting among siblings when the time interval between the death and the
survey increases (Masquelier, Reniers, and Pison, 2014; Obermeyer, Rajaratnam, Park et al., 2010).
To account for this, mortality estimates were restricted to the 6 years prior to each survey. The choice

26 International Journal of Population Studies | 2016, Volume 2, Issue 1