Based on the most recent changes and the current design planned in response to these changes, Figure 1 illustrates the proposed scheme of the stepped wedge cluster-randomised controlled trial for the assessment of the targeted FBPM and its associated components of SBCC and home fortification. In this scheme, baseline is dropped altogether and the start of interventions is in the three clusters needing to be implemented in by WFP to May 2016. We would then start with incidence data collection by the first week of May 2016 for 2 weeks and then continue with the second round of incidence data collection for June and then have the first round of stepped wedge data collection by the end of June 2016. Keep the number of steps to 4, this would mean that the final stepped wedge data collection will be on the last 2 weeks of December 2016. This option will roughly maintain the amount of time previously allocated for data analysis and will ensure deliver out outputs to 3ie by the March 2016 deadline. However, dropping baseline has a sample size implication. It should be remembered that a baseline round has two benefits to the study. It reduces the overall study sample size requirement and per cluster sample size requirement. Also, it increases the power of the study to detect variances and differences. In general, a baseline makes the study so much stronger and better. Losing baseline would require a relevant increase in sample size to make up for the variance lost by giving up baseline.
 

Limitations of current design

Though this current re-design is the most plausible option given the various issues faced with the starting up of this study, it should also be noted that it has limitations and there are still other issues that will not be solved by this current re-design.

  1. Given that WFP needs to implement in three specific localities in Kassala (i.e., Telkuk, Aroma and North Delta) first, the randomisation of rollout into the different clusters is still not going to be possible. So, this study loses its experimental nature (though it is still of a higher order of evaluation approach given that it is as close to experimental as it can be). However, this was something that the investigators knew to be difficult in a practical project implementation setting to begin with.
  2.  

  3. The number of clusters in each step is highly uneven. The current design intends to rollout in 3 clusters in the first step, 1 cluster in the second step, 1 cluster in the third step and 2 clusters in the fourth step. Whilst we tried to address this issue in the most recent re-design, the implication has been that we had to reduce the number of steps to 3 so that we can balance out the number of clusters in each step without having to increase the number of clusters to include in the study. However, as already noted earlier, reducing the study to three steps instead of four would reduce the number of data collection rounds for the incidence study to four from the six rounds that we originally planned for. Reducing to four rounds of data collection for the incidence study has its own sample size implications that at this point we will be unable to account for given resource limitations. Balancing everything else, we can deal with the unevenness in the cluster allocation for each step much reasonably as compared to having to deal with the resource implications of having to either add more clusters to the study or having to reduce the number of steps to three. The main setback with uneven cluster allocation is that the samples obtained in each step will most likely be very uneven as well. This will be most pronounced in the samples for steps 2 and 3 where there will only be 1 cluster each. We can most likely account for this in the statistical model that will be used for data analysis.
  4.  

  5. Related to point 2 above and after review of the number of villages within the study clusters in steps 2 and 3 (El Girba and River Atbara localities) while performing stage 1 spatial sampling, it will be likely that we will have to exhaustively sample both localities to ensure adequate sample size for each. Originally, we have had these two localities form a single study cluster as we have anticipated the relatively low number of potential sampling villages in each. The implication of this would be a relatively higher data collection overhead in each of these localities.
  6.  

  7. Because baseline data collection will not be done anymore, questions for specific indicator sets will need to be reconsidered as they were planned to have been collected only during baseline and during the last data collection step (step 4)1. These questions are for the multi-dimensional poverty index (MPI). These question sets will most likely have to be taken out from the data collection. The potential benefit of this is that the stepped wedge questionnaire will be a lot quicker to administer as a significant number of questions2 will not be included anymore.

Despite these limitations, this current re-design is our best option given the current circumstances and is still the most suited approach to evaluating the impact of the WFP programme. As such, we think this is the best way forward.
 

Figure 1: Schematics of adapted study design based on the current design

Cluster Preparation Step 1 Step 2 Step 3 Step 4
Urban Kassala
Rural Kassala
Rural El Girba
River Atbara
North Delta
Rural Aroma
Telkuk
Data Collection
Month Apr May Jun Jul Aug Sep Oct Nov Dec Jan
  Control   Intervention   Stepped Wedge data collection   Incidence data collection

 
 
 

Endnotes

1 These indicators are not seasonal and it takes some time before any of these indicators show any significant change if any. Measuring these indicators frequently over a period of time would therefore not be useful.

2 These questions include the question set for household roster which includes education attained by each household member, household assets, house structure characteristics and short birth history.