Sampling Framework
A questionnaire survey has been conducted among the representative panel
of households from selected five marginalised groups namely Rural Poor, Urban
Slum Dwellers, Ethnic Minority Groups, Female headed HHs and HH having Persons
with Disabilities. Considering all five categories of marginalisation we divide
the target population into three strata namely rural-poor, urban slum dwellers
and ethnic minority groups where we can find the other two groups as
sub-categories of the marginalised population which are poor women (female
headed HH) and Persons with Disabilities in each stratum as cross-cutting
groups.
As one of
the objectives of the study is to find the estimate for each of these strata. We calculated sample size for each
groups separately. For each stratum, we calculated sample size considering each
of the indicators of the expected outcome for the respective marginalised
target groups and the maximum sample size was reported for the indicator with
50% coverage. The sample size has been determined considering 95% confidence
interval, 5.6% admissible error and design effect (1.5) using the following
equation.
where:
Zα= the z-score corresponding to the
95% confidence interval of the true proportion
P =
the expected proportion of the respective indicators
d = admissible error
deff = design effect
Using the above formula, the sample size for each stratum is 460, and
with 10% non-response, it is 505.
Altogether, for the 3 target strata, the total sample size is
1,515. Sampling has been done in two
stages. A random sample of 30 clusters (a union/ward with approximately 500 HHs
in rural setting and a slum-segment/block with approximately 100 HHs in urban
slum setting) has been selected from each stratum in the first stage and a total
of 17 HHs has been selected from each cluster following a systematic sampling
scheme in the 2nd stage (Table 1). Finally, one respondent from each
HH was interviewed. Details of the sample selection for each stratum are
discussed below.
We selected the respondents in two stages. In the
first stage, we randomly selected 30 clusters from each stratum. These selected
30 clusters were then proportionately distributed to the respective districts
and upazilas of the study areas under each stratum. In the 2nd
stage, we selected 17 HHs (one respondent from each HH), systematically from
each cluster. We selected the sub-category of the marginalised population such
as women or female headed households and persons with disability across the
three main strata, we considered to interview 10% HHs
for each case from all HHs in each selected cluster.
Table: Stratum specific
sample size
|
Strata |
Areas |
Respondents |
Clusters |
Number of HHs |
|
Slum-dwellers |
City Corporation areas |
505 |
30 |
17 |
|
Ethnic/religious minority |
Both hilly and plain land area
where majority of the indigenous and religious minority people live |
505 |
30 |
17 |
|
Rural poor |
Districts with extreme poverty |
505 |
30 |
17 |
|
Total |
1515 |
90 |
1515 |
|
Sample Selection
For each study area (e.g., districts/ upazilas) of the
respective stratum, we divided the area/region into different small clusters. A
cluster is a union/ward in rural setting consisting of approximately 500 HHs
(it could be a ward of the union or a portion of the union with approximately
500 HHs in rural setting and a slum-segment/portion with approximately 100 HHs
in urban slum setting). For rural clusters, there are approximately 25-45% poor
HHs (according to the national poverty rate in HIES 2016: poor if per capita
expenditure belong to the upper poverty line of 2,268 BDT) in each cluster and
hence approximately 125 poor HHs in each cluster.
As discussed, these selected 30 clusters were
proportionately distributed to the respective districts and upazilas of the
study areas under each stratum. In the 2nd stage, we selected 17 HHs (one
respondent from each HH) systematically from each cluster.
To select the sub-category of the marginalised
population, such as women or female-headed households, and persons with
disabilities across the three main strata, we have considered interviewing 10%
HHs for each case from all HHs in each selected cluster.
In this procedure, we selected approximately 2 HHs
with Persons with Disabilities and 2 female-headed HHs in each selected
clusters, and then a total of approximately 2X90=180 respondents for Persons
with Disabilities, and 180 respondents for female-headed HHs. That is, with
this procedure, sub-categories: female-headed HHs and Persons with Disabilities
were selected across the 3 main strata (urban slum, ethnic minority and rural
poor). Due to field reality, the 1st round survey finally
interviewed 1,533 HHs instead of 1,515, and the subsequent rounds are expected
to experience a certain level of attrition (2.4% in the second round). The
distribution is given below.
Table: Conducted Interviews
under the sample categories
|
Category of sample |
Number of interviews conducted |
|
|
1st Round |
2nd Round |
|
|
Ethnic and religious community |
492 |
492 |
|
Rural poor |
531 |
524 |
|
Urban slum dwellers |
510 |
480 |
|
HH having Persons with Disabilities* |
184 |
179 |
|
Female headed HH* |
211 |
206 |
|
*These two groups are cross
cutting of the main three stratum and therefore not additive to the total
sample |
||
The sample area was well distributed geographically to capture the
selected marginalised groups under this survey. It covered all the
administrative divisions (8), where 20 districts
and 40 upazilas were brought under the survey. Under these 40 upazilas a total
of 90 Mauzas/ Clusters were covered where we have got 50 unions, and 16 wards
and 329 villages/ mohallas. The survey areas are shown in the Bangladesh Map here.