Table 1 Summary of estimation steps
From: A method for small-area estimation of population mortality in settings affected by crises
Step | Description | Sub-steps | Data requirements | Depends on |
|---|---|---|---|---|
Data collection and management steps | ||||
1 | Identify existing ground mortality data and prepare them for analysis | Identify all available estimates Extract meta-data for each estimate Clean and re-analyse datasets Grade estimate quality Describe data coverage and crude patterns in key demographic indicators | Raw datasets of surveys or other estimation exercises Survey reports Official administrative data, shape files for geographic boundaries | |
2 | Reconstruct population denominators [not presented in this paper] | Identify and curate alternative population datasets. Grade their robustness Identify and curate displacement data Make appropriate assumptions on population and displacement dynamics Reconstruct population for each \(kt\) stratum as an average of alternative estimates | Population datasets Remote sensing estimates Internal and refugee displacement data Explanatory accompanying documents and reports | |
3 | Capture predictor variable data and prepare them for analysis | Identify possible sources of data based on a conceptual framework Capture and curate predictor datasets Ascertain missingness and perform any appropriate imputation Convert absolute figures into population rates, smooth time series and create lags if appropriate | Predictor datasets Explanation of variable meanings/variable dictionaries | Steps 1–2 |
Analysis steps | ||||
4 | Fit a statistical model to predict the death rate as a function of the predictors | Explore correlation among predictors Do univariate analysis Fit alternative multivariate models and select the most appropriate one | Steps 1–3 | |
5 | Apply the model to estimate excess mortality while propagating known sources of error | Specify a set of counterfactual scenarios: Agree on what key deviations from normal define the crisis being analysed Arbitrarily define alternative (e.g. most likely, best-case, worst-case) scenarios for what values the model predictors would have taken in the absence of a crisis Construct counterfactual predictor datasets accordingly Apply counterfactual death rates and assumptions on displacement to reconstruct corresponding counterfactual population denominators Set up statistical simulation that implements Eq. (1) for each \(kt\) stratum while drawing from known error distributions of each parameter Compute excess death toll estimates overall and for sub-populations/periods of interest | Extensive contextual knowledge Mortality and predictor data for periods as long as possible before the crisis (recommended) | Steps 2–4 |
6 | Conduct sensitivity analyses of interest | Explore how possible bias or uncertainty in key parameters affect the estimates, by running the analysis with alternative data or assumptions | Step 5 | |