Sampling

Sampling Procedure

The primary objective of the sample design for the Bangladesh Multiple Indicator Cluster Survey was to produce statistically reliable estimates of most indicators, at the national level, for urban and rural areas, and for the six divisions of the country, municipal areas, city corporation's slum areas of two big cities and tribal areas. Rural areas, municipal areas, city corporation areas, slum areas and tribal areas were defined as the sampling domain.

A multi-stage, stratified cluster sampling approach was used for the selection of the survey sample. 


Sample Size and Sample Allocation

The target sample size for the Bangladesh MICS was calculated as 68247 households. For the calculation of the sample size, the key indicator used was the DPT immunization (3+doses) prevalence among children aged 12-23 months. The following formula was used to estimate the required sample size for these indicators:

n	=	[ 4 (r) (1-r) (f) (1.1) ]	
		[ (0.12r)2 (p) (nh) ]	

where
·	n is the required sample size, expressed as number of households
·	4 is a factor to achieve the 95 per cent level of confidence
·	r is the predicted or anticipated prevalence (coverage rate) of the indicator 
·	1.1 is the factor necessary to raise the sample size by 10 per cent for non-response
·	f is the shortened symbol for deff (design effect) 
·	0.12r is the margin of error to be tolerated at the 95 per cent level of confidence, defined as 12 per cent of r (relative sampling error of r)
·	p is the proportion of the total population upon which the indicator, r, is based
·	nh is the average household size.

For the calculation, r (DPT immunization 3+doses prevalence) was assumed to be 39.7 percent in the Rangamati districts. The value of deff (design effect) was taken as 1.5 based on estimates from previous surveys, p (percentage of children aged 12-23 months in the total population) was taken as 2.3 percent, and nh (average household size) was taken as 4.9 households. 
For the sub national level, the margin of error should be high which was also acknowledged in the MICS manual. Therefore, for sub national estimates the margin of error need to be relaxed considerably. If a rate of 30% of r is used this would give a margin of error ± 0.06 for prevalence rates of 0.20, ± 0.12 for prevalence rates of 0.40, and so on.  Considering this phenomenon, in case of Rangamati 30% of r has been used.

The resulting number of households from this exercise was about 900 households which is the sample size needed in each district - thus yielding about 68250 in total. The average cluster size in the Bangladesh MICS was determined as 35 households, based on a number of considerations, including the budget available, and the time that would be needed per team to complete one cluster. Dividing the total number of households by the number of households per cluster, it was calculated that the selection of a total number of 26 clusters would be needed in each district. 

Equal allocation of the total sample size to the 75 domains was targeted. Therefore, 26 clusters were allocated to each district with the final sample size calculated at 68250 households (1950 cluster X  35 households per cluster). In each stratum, the clusters (primary sampling units) were distributed to rural, municipal, city corporations, slum and tribal areas on PPS method.

 
Sampling Frame and Selection of Clusters

The 2001 census frame was used for the selection of clusters.  Census enumeration areas were defined as primary sampling units (PSUs), and were selected from each of the sampling domains by using systematic pps (probability proportional to size) sampling procedures, based on the estimated sizes of the enumeration areas from the 2001 Population Census.  The first stage of sampling was thus completed by selecting the required number of enumeration areas from each of the 5 strata namely rural, municipal, city corporations, slum and tribal areas.


Listing Activities

Since the sample frame of the 2001 Population Census was not up to date, household lists in all selected enumeration areas were updated prior to the selection of households. For this purpose, listing teams were formed, who visited each enumeration area, and listed the occupied households. The BBS officials working in the upazila were responsible for the listing of all households in the respective PSUs.

Selection of Households

Lists of households were prepared by the Upazila officials of BBS. The households were sequentially numbered from 1 to 100 (or more) households in each enumeration area at the  where selection of 35 households in each enumeration area was carried out using systematic selection procedures.  


(Information extracted from the final report: BBS and UNICEF. 2007. Bangladesh Multiple Indicator Cluster Survey 2006, Final Report. Dhaka, Bangladesh: BBS and UNICEF)

Deviation from Sample Design

No major deviations from the original sample design were made.  All sample enumeration areas were accessed and successfully interviewed with good response rates.

Response Rates

Of the 68,247 of households selected for the sample, 67,540 were found to be occupied. Of these, 62,463 households were successfully interviewed for a household response rate of 92.5 percent. In the interviewed households, 78,260 of eligible women (age 15-49) were identified. Of these, 69,860 of women were successfully interviewed, yielding a response rate of 89.3 percent. In addition, 34,710 of children under 5 were listed in HH questionnaire. Of these, questionnaires were completed for 31,566 under-five children which correspond to a response rate of 90.9 percent. Overall response rates of 82.6 were for women's questionnaire and 84.1 for under-5 questionnaire.

Weighting

Sample weights were calculated for each of the datafiles.  

Sample weights for the household data were computed as the inverse of the probability of selection of the household, computed at the sampling domain level (urban/rural within each region).  The household weights were adjusted for non-response at the domain level, and were then normalized by a constant factor so that the total weighted number of households equals the total unweighted number of households.  The household weight variable is called HHWEIGHT and is used with the HH data and the HL data.

Sample weights for the women's data used the un-normalized household weights, adjusted for non-response for the women's questionnaire, and were then normalized by a constant factor so that the total weighted number of women's cases equals the total unweighted number of women's cases.

Sample weights for the children's data followed the same approach as the women's and used the un-normalized household weights, adjusted for non-response for the children's questionnaire, and were then normalized by a constant factor so that the total weighted number of children's cases equals the total unweighted number of children's cases.
Generated: FEB-14-2008 using the IHSN Microdata Management Toolkit