map-itn-cube

NOTE: This branch is under active development. To view the branch used for the 2021 research paper, switch to branch publication-2021.

The code in this repo generates estimates of insecticide treated net (ITN) ownership, access, and use in sub-Saharan Africa.

Stock and Flow (stock_and_flow)

Mechanistic model fit in rjags to estimate country-specific time series of ITN distribution and retention. Todo: script-specific descriptions.

01a_prep_hh_survey_data.r

Clean household-level survey data; save for ITN cube; aggregate to national level for stock and flow.

01b_prep_reportonly_survey_data.r

Append those surveys whose results are only available in aggregated form to the stock and flow dataset.

01c_prep_indicator_priors.r

Run a regression to find coefficients for the "proportion of households with no net" and "mean nets per household" metrics.

01d_calculate_use.r

Run a regression to find coefficients for the access-to-use conversion used in the World Malaria Report. Use is calculated overall, among children under 5, and among pregnant people.

02_prep_delivery_dist_data.r

Clean and format ITN delivery and ITN distribution data.

03_stock_and_flow.r; jags_functions.r

Run and save results for the stock and flow model; run separately for each country.

04_compare_outputs.r

View results of stock and flow compared to earlier model versions.

05_aggregate_for_cube.r

Collect national-level outputs to pass along to the itn_cube code.

06_aggregate_for_wmr.r

Calculate national and continental ITN indicators used in the World Malaria Report.

07_half_life_convergence.r

Generate plots to assess JAGS model fit based on the convergence of the half-life parameter; also save uncertainty intervals for half-life. Diagnostic purposes only.

08_analyze_sensitivity.r

For sensitivity analysis model runs, aggregate and plot results. Not usually used.

ITN "cube" (itn_cube)

Geospatial regression model fit in R-INLA that utilizes the national mean from the stock and flow outputs as a baseline for disaggregating spatially.

000_make_dsub.r:

Construct the long bash command used to collate and save the desired covariates (see covariate_key.csv) from the COVARIATE bucket.

000_extract_covariates.r

Collate and save covariates from the COVARIATE bucket using the dsub command constructed above.

No table since this should only be run via 000_make_dsub.

00_generate_cube_master.r

This is the script that gets submitted for a full run of the cube. Loads all input data and runs steps 1-3. Step 4 needs to be run separately to paralleleize correctly.

01_prep_data.r; supported by 01_data_functions.r

Load household-level survey data cleaned in the stock and flow code, calculate cluster-level access, aggregate to the 5km-by-5km pixel level.

02_prep_covariates.r

Subset full covariate set down to only those needed for model fitting; merge onto survey data.

03_regress.r; supported by 03_inla_functions.r

Run regression (including appropriate data transformations) and save outputs.

04_predict_rasters.r; 04_prediction_functions.r; 04_batch_submit_predictions.r

Recently modified to run separately for each year. Predict outputs on the monthly level; save national and continental aggregation of monthly time series; aggregate rasters to the annual level and save; calculate exceedance and relative uncertainty.

05_relative_gain.r

Calculate relative effect of increasing access vs increasing use.

Not used in a typical model run; for publication only.

view_changes.r

Compare versions of cube outputs to each other.

Data Versioning

Each of the following types of input data receives separate, dated labels:

Household survey data

This includes three components:

• DHS data-- updated when the team cleans new surveys;
• MICS -- at the moment, these are manually extracted and cleaned by the ITN person;
• MIS and other surveys that do not report microdata-- again, extracted manually from time to time.

In addition, this analysis includes some older surveys (DHS and MICS 3 and 4) extracted by Bonnie Mappin that I could not find matches for in the current DHS framework.

This data is cleaned in the 01a_prep_hh_survey_data.r and 01b_prep_reportonly_survey_data.r scripts, and would benefit substantially from streamlining.

Net stock and flow data

Updated approximately annually when new data is recevied from WHO. Cleaned using 02_prep_delivery_dist_data.r (requires customization)

Covariates

Updated relatively rarely; Extracted using 000_make_dsub.r and 000_extract_covariates.r in the itn_cube folder

Output versioning

Every time a stock and flow or ITN cube model is run, outputs are saved in a dated and uniquely labeled folder in the appropriate spot on the google cloud buckets. Older versions are not deleted to serve as comparison points.