Model Extensions
After completing the data sparsity check, users can access the Advanced Options panel. By clicking to expand it, users can customize model settings such as enabling the nested model or incorporating covariates.
Nested Admin-2 Model
The spatial models we use smooth the raw data both locally and nationally. Oversmoothing is a concern when data are sparse.
Recall that surveys are powered to produce reliable estimates for some indicators at Admin-1 and, hence, we would like to recover the weighted estimates, rather than distort away from those.
To prevent overshrinkage at the Admin-1 level, we have developed a nested model in which we have
where \(a[s_c]\) is the Admin-1 area within which cluster \(c\) with location \(s_c\) is contained and the \(\alpha_a\) are a set of Admin-1 level fixed effects which are included to effectively recover the weighted estimates.
The \(e_{i[s_c]}+S_{i[s_c]}\) terms then smooth within each Admin-1 area. This model can be viewed as being consistent with the sampling scheme in which the Admin-1 areas are sampling strata. Currenlty this option only affects unit-level models at or beyond Admin-2 level.
Default: Disabled (non-nested)
Covariate Incorporation
Users can incorporate area-level external covariates into both area-level and unit-level models to improve estimation. To do so, first select the desired Admin level. A template .csv file can be downloaded, containing the list of regions at the selected admin level. Users should add new column(s) corresponding to the covariates and fill in the covariate values for each region. Once completed, the file should be uploaded back into the app. The covariates are automatically incorporated into the model fitting process after clicing the apply button.
Default: No covariates.