Stepped-wedge study design
Given the refinements to the sampling framework described in the previous sampling update, we initiated stage 1 sampling for the stepped-wedge study so that we can begin to select villages that will be sampled either on baseline and/or at the first step of the stepped-wedge study.
To perform the stage 1 sampling, the following datasets were assembled:
- Map data for Kassala state and its localities. The map data used was the same map data used in the national S3M conducted 3 years earlier and is also the map data used to create the maps used by WFP for their programming. This map data would be used to plot the boundaries of Kassala state and its localities to create maps that can be used for the stage 1 sampling. The map data was in ESRI Shapefile (.shp) format;
- List of WFP programme sites in each of the localities of Kassala state with their corresponding longitude (x) and latitude (y) coordinates. This data would be used to plot the point locations of each of the programme sites in Kassala state. This map data was provided by WFP in Microsoft® Excel (.xlsx) format. We then converted this into comma-separated value (.csv) format to make the data universally compatible with most applications;
- List of villages/settlements in each of the localities of Kassala state with their corresponding longitude (x) and latitude (y) coordinates. This data would be used to plot the point locations of each of the villages/settlements in Kassala state. This map data was provided by UNICEF based on the current work they are doing in updating village/settlement locations in the whole of Sudan. This map data was received in Microsoft® Excel (.xlsx) format. We then converted this into comma-separated value (.csv) format to make the data universally compatible with most applications; and,
- Map data for Kassala City that delineates the various city blocks and city buildings and structures. This map data was used to plot maps of Kassala City showing the various city blocks that can be used for stage 1 sampling in Kassala City. This map data was collected from OpenStreetMap (OSM) in protocolbuffer binary (.pbf) format which is one of the standard data formats for OSM data. We then converted this into ESRI Shapefile (.shp) format.
Additional map data were collected for roads and water bodies in Sudan. These were used to plot auxiliary features on to the maps.
Stage 1 sampling
We wrote bespoke map plotting scripts in R to perform stage 1 sampling using the map data described above. The scripts performed the following functions:
- Plot maps of each locality included as a cluster in the study with the programme sites and villages overlaid;
- Create a random hexagonal grid onto the locality maps;
- Select 20 hexagonal grids that overlap with populated villages/settlements;
- Identify and list the villages/settlements that fall within the boundaries of each of the 20 hexagonal grids;
- Randomly select two villages from the full list of villages/settlements per hexagonal grid; and,
- Create a list of the selected villages per hexagonal grid
The advantage of having the map plotting and stage 1 sampling scripts in R is that the stage 1 sampling can be performed automatically before the start of every round of data collection by just re-running the script.
For the data collection in step 1 of the stepped wedge study, following are the sampling maps and list of sampling villages per locality produced by the map plotting and stage 1 sampling script.
Table 1: List of villages per hexagon per study cluster
Table 2: List of sampling villages per hexagon per study cluster
For the incidence study, a set of villages within the catchment area of programme sites in each study cluster were selected. The number of villages selected was based on an estimate of how many children under 5 years who are not undernourished and not sick who can be recruited in order to reach overall sample size required. Following are the villages selected for the incidence sub-study:
|Rural Kassala||Edd Must|
In the occasion that required sample sizes are not reached, we are planning to select between 2 to 3 additional villages per cluster as backup. We will include this on the list once selected.