Predicting Poverty Incidence In Statistically Underdeveloped Countries

ABSTRACT It is generally acknowledged that availability of data remains critical to effective monitoring of poverty indicators and evaluation of policies and programmes towards alleviating poverty. The paucity of data has reduced the frequency of measuring poverty indicators at the national and regional level. District level poverty indicators are yet to be estimated since sampling designed data collected can only provide national and regional level indicators. This thesis applied multiple imputation technique to predict poverty incidence from a non-expenditure household data (MICS4) based on a regression model developed from a recent household expenditure data (GLSS6). The time interval between the two data sets is one and a half years, and few variables in GLSS6 data that significantly influence poverty were not found in the MICS4 data. However, the poverty incidence predicted for national, urban and rural levels at 95 percent probability were very close to that estimated by the Ghana Statistical Service from the GLSS6 data. The study recommends application of this technique to future non-expenditure survey to improve the frequency of poverty indicators to inform policy and programmes.