Assessing the usefulness of sample survey designs and census data with respect to poverty mapping in Ethiopia

Abstract

We investigated the effect of different probability sampling designs (simple random sampling, stratified, cluster and systematic sampling designs) on estimates of age, gender proportions, and proportions of the population with different education levels, and access to water. To investigate the effect of the sample designs we used as a sampling frame the census 1994 data obtained from Central Statistical Agency of Ethiopia (CSA). In order to compare results obtained from different sampling designs, estimates of population means and proportions for selected variables at household levels using 1, 5 and 10% sample size is considered. The performance of these sampling techniques is observed on the following dimensions: relative accuracy, standard error, confidence intervals, precision and also we compared the effect of the different methods of calculating variances like Taylor series and jack-knife, the comparison of this two methods is done by drawing 100 different samples of 1% for stratified and cluster sampling. It was found that in general the relative accuracy for the selected variable obtained from the stratified sampling design is better than other estimates. We also find evidence that the 95% confidence interval constructed for each method contains the true population value, but the confidence interval for cluster sample is wider than the others that indicate less precision than others. The result of the comparison of the different methods of estimating the variance from the simulated population variance indicates that, for cluster sampling, both of the Taylor linearization and jack-knife methods tend to overestimate the population variance, whereas for stratified sampling the population standard error is in between of the standard error of the both methods indicating when one underestimates the other overestimates. Design effects were also compared, and it was found that the design effect for cluster sampling was larger that for the other methods, as expected. Survey cost information is needed to further inform discussion of the usefulness of the different methods. We assessed the extent of poverty by calculating the poverty index for the Tigray region. To calculate the poverty index we used the method proposed by the United Nations Development Programme (UNDP). Using the Welfare/Poverty Monitoring Survey year 2000 (WMS2000) data obtained from CSA, we found that the region is highly affected by each of the dimensions investigated. Further questions need to be added to questionnaires in order to allow use of the census data in calculating poverty maps.