Home

Results

We found that kernel density estimation provided the easiest way to see if there was any spatial correlation between travel times and malnutrition indicators. This method of representation also showed the most correlation between the variables.

Hotspot/LISA methods were an effective means of confirming the statistical significance of clusters but were more difficult to interpret as our outputs of hotspots were points. It would have been more useful to have small polygons to represent hotspots, but arbitrarily defining polygon size would reduce the significance of our output.

Overall, there appeared to be a reasonable correlation between inaccessibility and malnutrition indicators when using the kernel density. This supports the wider assumptions that health outcomes and usage of health facilities are contingent on care facilities being nearby. It was more problematic to discern the relationships using hotspot analysis, due to the nature of the outputs, which, had this also shown a correlation, would have solidified our results considerably.

Limitations

There were also inherent limitations to the way in which we conducted our network analysis.

Uganda was challenging as a study area in that data availability was a limitation. We attempted to digitize road data through satellite imagery, but in some areas the quality of the imagery was low and it became difficult to differentiate dirt roads from irrigation cuts.

We had planned on doing Ordinary Least Squares (OLS) Regression to derive an R-squared value between travel time and malnutrition. Unfortunately, our data outputs did not end up being individual travel times, and instead, catchment areas. These were separate shapefiles and OLS in can only be run when the variables are within the same shapefile. However, it was not possible to merge these datasets and keep the spatial rigorousness of the dataset, because the spatial extents and shapes of kernel density and network accessibility were different. Thus, we could only do qualitative analyses.

Lastly, we must acknowledge that correlation is not causation. Although, qualitatively, our kernel density estimation outputs indicated a potential correlation between underweight as well as stunting and increased travel times, it is not possible to say this is what caused these negative health outcomes in our study area.