Panel data are valuable for answering questions about change over time, but remain relatively scarce
in most developing countries, especially sub-Saharan Africa (SSA). Where there exists a series of
repeated cross-sectional data, ‘pseudo-panels’ provide a promising alternative. The use of ‘pseudo
panels’ has received considerable attention in econometrics, but application in Demography remains
rare. This paper explores the potential for using ‘pseudo panels’ to investigate causal link between
HIV and fertility in SSA. The relationship between HIV and fertility is a complex one, partly
because causality can run in either direction. We focus primarily on fertility as the outcomes of
interest and HIV as a contributing factor. Repeated cross-sectional Demographic and Health survey
(DHS) data from 20 countries in SSA are used to construct “pseudo panels” based on birth cohorts
by country. The pseudo panels allow an investigation of possible causal link between HIV in an
earlier survey and fertility behaviour of similar cohorts in a subsequent survey. Measures of HIV are
based on risk perception and HIV status, while fertility is based on births within the last five years
and future fertility intention. A total of 140 cohorts (7 age groups * 20 countries) were constructed,
with 120 cohorts having data for at least two time periods. The analysis used two alternative
approaches: (i) ‘Conditional’ models of HIV in an earlier survey and fertility behaviour in
subsequent surveys; and (ii) Repeated measures multilevel analysis, with cohort as Level-2, and
measurement occasion as level-1. An evaluation /assessment of the analysis involved a comparison
of findings from ‘pseudo cohort’ with individual-level analysis, and a multi-level estimation of intracohort
correlation coefficients to assess the degree of similarity of individuals in the same cohort. A
multivariate analysis based on fixed effects models was used to determine the extent to which
observed patterns may be attributable to key demographic/ socio-economic differences or infer
possible causal links. Preliminary analysis shows promising results on application of ‘pseudo panels’
in investigation of demographic causal links in settings with limited panel data such as sub-Saharan
Africa. However, further analysis is necessary for conclusive results. In particular, advanced
modelling using Multiprocess modelling or Structural equation modelling will be used to address
possible endogeneity in the relationships observed.