Addressing Extreme Poverty in Bangladesh - The Case of Monga 2008-2009, Round 1

Survey Areas and Survey Population

The survey has both a household level and a community level questionnaire. The survey has households belonging to the Monga-prone greater Rangpur districts (Kurigram, Gaibandha, Lalmonirhat, Nilphamari, Rangpur) with the sampling frame being representative of the diversity of the region. For example, the survey covered the various agro-ecological zones of the region, include households on river banks and away from it, include mainland households as well as char-land households. In addition to Monga area households, the survey has sample from non-Monga area households in the northwestern Bogra district. It is determined by the World Bank that the total sample size is 2,375 rural households selected from about 125 villages/PSUs. There are 95 PSUs in 5 districts of greater Rangpur (Lalmonirhat, Kurigram, Nilphamari, Gaibandha and Rangpur) and 30 PSUs in Bogra districts.

Survey Design and Sampling Strategy

It was decided by the World Bank that the sample size for the study is 2,375 households. Out of these 2,375 households, 1,805 households from 95 PSUs/villages were from Monga-prone greater Rangpur districts (Kurigram, Gaibandha, Lalmonirhat, Nilphamari, Rangpur) and rest 570 households were from non-Monga area, Bogra district, in the northwestern. Sample list of the villages/PSUs from each of the districts were provided by the World Bank.

In the first round of survey, it entailed two sub-surveys: a) Household listing operation: census of households in selected 125 villages with average size of 250 households, if a village size was more than 250 households, village was split using natural/constructed demarcation line such as by using river/canal or village road; and b) the detailed household survey and community component. For the later one it required on-site sampling of required number of households using random sampling technique from the census list and then conducted the household survey on sampled households. For the household sampling from census list a simple random sample selection protocol was developed and used for on-site sample selection. For follow-up 2 rounds of surveys same sample households were tracked and surveyed, only exception is there was no census hence no on-site sampling procedure was adopted for later rounds.

Sampling Protocol

Since village census was compulsory to list all household in the PSUs to sample required 19 households for detailed household survey, it was decided to do the census on the first day of survey in each of the PSUs and in the evening once the census is completed an on-site sampling technique was used to sample 19 households per PSU.

Below is a practical example on sample selection procedure adopted in field after the census is completed. Simple Random Selection of households was performed as follows:

a) Total number of household in the sample village/PSU; in this case it was 300.
b) Total number of sample to be selected. In this case the sample size is 19 per PSU.
c) So an "Interval" (steps) is estimate by dividing: Total Population/Sample size. In this case 300/19 = 15.79
d) A random number was generated for each of the PSU by using excel worksheet function "rand()". In this case it was 0.164
e) Then the "Interval (step)" was multiplied by random number "Interval X Random number", it gave the value for first sampling number (a non-integer number) = 2.594689447.

Now for the second sampling number the "Interval" value added with first sampling number, for third sampling number again the "Interval value" with second sampling number, and so on until we get all 19 sampling number.

The census was listed per PSU from 1 to Nth number as long as it required to number all the listed households in a PSU. As the sampling numbers are non-integer number, to get an integer that corresponds to the listing number of all households the sampling numbers were converted into integer and added "1" to get the household sample number.