This paper describes the design of a multi-stage stratified
sample for the Bangladesh Household Income and Expenditure
Survey 2016/17. This survey instrument will be
used by the Government of Bangladesh to estimate reliable
poverty and welfare statistics at three different levels:
(i) annual estimates at the district level, (ii) quarterly estimates
at the national level, and (iii) annual estimates at the
division level for urban and rural areas. The sample for this survey was designed to achieve these three objectives. The
paper explains how the three objectives are prioritized and
how inconsistencies in achieving more than one objective
can be reconciled. Further, the paper modifies the standard
formulas to estimate the optimal sample size and the
allocation of the sample across strata by explicitly taking
into consideration the effect of clustering in the sample.