This thesis consists of three self-contained chapters on issues relating to the empirics of poverty and household analysis in Africa. The first analyses poverty at the country level, highlighting the diversity of experience in poverty reduction performance over the last few decades and assessing its causes. The other two studies are more methodological in their content, exploring how poverty and related issues can be analysed at the household level using existing survey data that pose difficulties as they do not track the same set of households over time. Chapter 2 highlights the fact that although growth has improved substantially in most African countries in recent years, poverty across the continent has fallen very little in the aggregate. Poverty reduction performance has varied across countries: there are apparently ‘two Africas’, one with an ability to reduce poverty and one without. The main argument is that one reason for this difference is rooted in colonial times. Countries with strong smallholder cash crop sectors emerged into independence with broad-based labour-intensive economies supporting a more equitable income distribution conducive to inclusive growth and poverty reduction compared to initially more inequitable mineral resource and large farm based economies. This did not necessarily determine the post-colonial path: many peasant export economies achieved no poverty reduction (often because of little growth), and some mine/plantation economies did achieve poverty reduction. The key reasons for this evolution lie in the motivation and ability of African elites to form pro-poor coalitions, which in some cases were then able to implement policies supporting a pro-poor pattern of growth. Chapter 3 focuses on pseudo-panel estimation and how it enables the estimation of panel models when only repeated cross-sections rather than panel data are available. It involves the grouping of individual into cohorts and using the cohort means as if they are observations in a genuine panel. The usual assertion is that as long as there are enough individuals within cohorts so that the cohort sample means are a good approximation of the cohort population means then pseudo-panel estimates are consistent, otherwise they may suffer from measurement error bias. We show that there can be substantial bias arising from the grouping process itself due to the loss of variation and heterogeneity as one moves from the individual to the cohort level. Thus we find many of the common methods used for grouping into cohorts produces inconsistent estimates. We develop some measures for assessing whether the cohort level data contains sufficient variation for pseudo-panel estimates to be consistent, focusing on the variation across cohort means, over time and within cohort groups. We then test the measures empirically and with Monte Carlo simulations, providing useful thresholds that can be used to accept or reject the cohort construction method. Chapter 4 assesses four different estimation methods of binary response models with individual effects where the data is a time-series of independent cross-sections. We compare Deaton’s (1985) linear fixed effects estimator, which is most often used in applied work, to three non-linear estimators. The first is a simplified version of Collado’s (1998) Minimum Distance estimator. The other two are based on the fractional response estimators developed by Wooldridge and Papke (2008) which, unlike Collado’s estimator, can be used for dynamic models as well. Results from Monte Carlo simulations and from an empirical application indicate the linear estimator is just as good as the other non-linear estimators and is generally more robust to problems arising from the process of creating the pseudo-panel.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:740644 |
Date | January 2017 |
Creators | Khan, Rumman |
Publisher | University of Nottingham |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://eprints.nottingham.ac.uk/45535/ |
Page generated in 0.0063 seconds