World Health Survey Plus Nepal 2023

Wave 1

The WHS+ Nepal design used a stratified sample of enumeration areas (EAs) from which households and residents were sampled for interviewing during the field work.
A sampling frame containing data from the Nepal 2021 population census was provided, containing 36,229 enumeration areas. Out of these, 897 had less than 10 households, according to the census count. These were not considered eligible for sampling in the WHS+.
Therefore, the final list of EAs from which a sample was obtained comprised 35,332 EAs.

In the absence of relevant information that could be used to stratify the EAs according to socio-economic characteristics, the stratification adopted was by Province and District, the two major administrative units identified in the sampling frame.
Nepal is divided into 7 provinces and 77 districts, which were used as the sampling strata.
The WHS+ sample was intended to be spread in all the country's administrative regions defined by these districts.

Given the requirement that the overall sample size for the WHS+ should be 6,000 households and considering that 15 households are to be selected from each sampled EA after updating the corresponding list of households, a target sample of 400 EAs was defined.
The final total sample size was 402 EAs. This slightly larger sample of EAs results from rounding allocations of target numbers of EAs to sample per district, and from specifying a minimum sample size of 2 EAs per district.

The allocation of the sample of EAs to the districts was performed in two steps.
First, the overall sample of 400 EAs was distributed between the provinces proportionally to the number of households in each province. Then the target sample of EAs in each province was distributed between its districts proportionally to a size measure defined as 'cen_hhld ^0.8' (i.e., the power 4/5 of the number of households in each district). The use of the power 0.8 aimed at smoothing the allocation to avoid extreme values of the sample size in each district.

After the initial calculation of the district sample sizes, a few districts that had target sample sizes less than 2 were allocated samples of at least 2 EAs, and the largest district that had an initial sample allocation of 25 EAs had its allocation reduced to 20 EAs.
As already mentioned, the rounding and these allocation changes produced a total sample size of 402 EAs across all the 77 districts in Nepal.

Sampling of eligible EAs within districts was carried out using Pareto PPS Sampling.
The size measure considered for sampling the EAs was defined as: where Hk denotes the number of households in EA k. The idea of using truncation of the EA sizes in the range [50; 1000] aims to avoid having extreme weights, which would increase the variability of the resulting sample estimates.

After the initial sample of 402 EAs was selected, the list of sampled EAs was provided to the Nepal Public Health Research and Development Center (PHRD Nepal), the local agency in charge of data collection in Nepal.
They tried to secure access to maps and descriptions of the sampled EAs but were not successful.
Instead, the WHO GIS Centre for Health (GISC) was only able to secure access to the maps and descriptions of wards that contained at least one sampled EA.

In addition, the high cost of data collection meant that the survey team would not be able to visit all 77 districts within Nepal.
Therefore, a sample of districts was selected and only wards in the sampled districts were maintained in the final area sample.

The sampling of districts was carried out according to the following steps.

1. Three districts were included in the sample with certainty: Bhaktapur, Kathmandu and Lalitpur.
2. The remaining 74 districts were stratified into four strata, using a composite index derived as the arithmetic average calculated from three socioeconomic indicators for each district ranging from 0-100, and arranged such that larger values represent improved socioeconomic conditions: 100 (Human Development Index) + (100 Female Literacy Rate) + 100 * (Per Capita Income) / 3166.
3. The stratum boundaries and allocation were obtained using an optimal stratification approach proposed by (Brito et al., 2019) a implemented in the stratbr package in R.
4. 8 districts in stratum 'st4' were included in the sample with certainty, given the optimal allocation assigned to this stratum equaled the total number of districts in it. This happened because this stratum had larger variance of the composite index when compared to the others.
5. The other 27 districts were sampled from within the 3 strata defined using a Pareto Probability Proportional to Size sampling (Pareto PPS) method (Rosén, 2000) available from the sampling package in R (Matei & Tillé, 2006).

In the end, 233 of the wards previously sampled located in the 27+3+8 = 38 districts sampled as described above were retained for the WHS+ survey.
Since the target sample size for the survey was a total of 6,000 households, a total of 26 households were sampled per ward, yielding an expected sample of 6,058 households.
Given that the local survey team would not be able to obtain maps for the enumeration areas sampled within those 233 wards, and the fact that most wards are too large for conducting direct enumeration of households across the whole ward, an alternative approach was implemented.
The approach involved work from the GISC team using satellite imagery and GIS technology to digitize buildings in each one of the 233 wards retained in the sample. Once these building lists within each ward were compiled, three sets of 26 building coordinates were sampled within each ward by simple random sampling without replacement (SRSWOR). Such lists were labelled "Initial", "First reserve" and "Second reserve", respectively.
The idea was that the first set of 26 building coordinates (labelled "Initial") would be allocated to the field team for visiting and interviewing. It is likely that several such buildings contain no eligible households for the WHS+. After the initial set of buildings was approached and the number of interviews missing from the ward was known, the second set of building coordinates (labelled "First reserve") were allocated to the field team to conduct the visits and attempt to obtain interviews. If necessary, the third set of building coordinates (labelled "Second reserve") were also allocated to the field team to try to achieve the target of 26 interviews per sampled ward.

For individual interviews, one person aged 18 or older per household was eligible to participate.
The eligible individual was selected with equal probabilities using the CAPI program from a list of eligible household members.

Stages of selection
Strata: Province, District = 7 x 38 = 266
PSU: Wards=231 surveyed
SSU: Households=6110 surveyed
TSU: Individuals=6110 surveyed