Skip to contents

Validate TPR Proxy Quality Using Leave-One-Out Cross-Validation

Source: R/calc_tpr.R
validate_tpr_proxies.Rd

Evaluates the accuracy of TPR proxy estimates by comparing them against actual facility-level TPR values. Uses leave-one-out cross-validation to calculate what each proxy level (adm2, adm1, prev_year, rolling, adm0) would have estimated for facility-months with valid raw TPR data.

Usage

validate_tpr_proxies(
  data,
  hf_var = "hf_uid",
  adm0_var = "adm0",
  adm1_var = "adm1",
  adm2_var = "adm2",
  year_var = "year",
  month_var = "month",
  conf_var = "conf",
  test_var = "test",
  min_facilities = 2,
  generate_plots = TRUE
)

Arguments

data

Output from calc_tpr() with include_flags = TRUE.

hf_var

Column name for health facility unique identifier. Should match the value used in calc_tpr(). Default is "hf_uid".

adm0_var

Column name for national/country level. Default is "adm0".

adm1_var

Column name for first administrative level/region. Default is "adm1".

adm2_var

Column name for second administrative level/district. Default is "adm2".

year_var

Column name for year. Default is "year".

month_var

Column name for month. Default is "month".

conf_var

Column name for confirmed cases. Default is "conf".

test_var

Column name for tests. Default is "test".

min_facilities

Minimum number of facilities in an admin unit to include in validation. Default is 2.

generate_plots

Logical; if TRUE (default), generates diagnostic plots. Set to FALSE for metrics only.

Value

A list containing:

  • metrics: Data frame with validation metrics for each proxy level

  • validation_data: Data frame with actual vs proxy comparisons

  • plots: List of ggplot objects (if generate_plots = TRUE):

    • scatter: Actual vs proxy scatterplots

    • error_dist: Error distribution by proxy level

    • mae_by_tests: MAE vs number of tests

    • stability_map: Error patterns by test count and TPR value

  • summary: Character vector with key findings

Examples

# result <- calc_tpr(facility_data, include_flags = TRUE)
# validation <- validate_tpr_proxies(result)
# validation$metrics
# validation$plots$scatter