• Facebook
• Twitter
• More

Description

The experience with a number of analytic household sample surveys is recalled to explain why the analysis of such surveys often takes too long to complete, and with much higher sampling errors than expected. A major reason for this is hasty choice of sampling procedure with little or no consideration being given to ease of data processing and analysis. The result usually is a complex sample, the proper statistical analysis of which would have to be correspondingly complicated. The irony is that in the end the researcher often is compelled to proceed as if the sample were "simple random", an assumption which can potentially lead to serious inferential mistakes. Moreover, sampling procedures that look good in theory can perform disappointingly in developing countries. One example of this is the use OX., stratification on the basis of imperfect prior information which causes misclassification of units. Another is stratification in terms of a dynamic variable in time series surveys in which units migrate in and out of strata. These errors and changes in classification complicate the analysis and render the estimates less precise. Two of the cases discussed are project benefit monitoring and evaluation (PBME) surveys of ADB-assisted irrigation and drainage projects. A third PBME survey is used to illustrate a sampling design approach that simplifies data analysis and reduces the sampling error of estimates. Its main features include use of extensive numerical experimentation with available background information as the basis for choosing an efficient sampling procedure, and use of replicated systematic sampling to draw a self-weighing sample.