nwss_hhs_region aggregation computes `region_value = mean(region_value * population) / sum(population)` — off by a factor of the region's state count, and NA-prone

Start in R/targets/flu_data_targets.R around line 194 and compare the nwss_hhs_region aggregation with its sibling statistics; then inspect the identical block in R/targets/covid_data_targets.R around line 135. Use the two-state population-weighted repro from the issue and verify both paths handle missing populations without producing an unintended NA. Done means the regional value matches the intended sum/sum calculation in both files.

Severity (my assessment): medium
Area: data targets (R/targets/flu_data_targets.R, R/targets/covid_data_targets.R)
Commit observed: current main (ccc2ee4); also verified at 5e35f53
Related existing issue: #165 (open, "Double-check NWSS national aggregations" — overlapping concern about the NWSS aggregation paths; my report is a distinct, specific defect on the regional aggregation)

Summary

In the nwss_hhs_region aggregation, four sibling statistics use a population-weighted sum/sum, but region_value uses mean/sum:

R/targets/flu_data_targets.R:194 (identical at R/targets/covid_data_targets.R:135):

summarize(
  value = sum(value * population, na.rm = TRUE) / sum(population, na.rm = TRUE),
  activity_level = sum(activity_level * population, na.rm = TRUE) / sum(population, na.rm = TRUE),
  region_value = mean(region_value * population) / sum(population, na.rm = TRUE),
  national_value = sum(national_value * population, na.rm = TRUE) / sum(population, na.rm = TRUE),
  ...
)

mean(region_value * population) is the per-state average of the weighted numerator — off by a factor of the region's state count (e.g. ÷9 for a 9-state region... effectively multiplied by 1/9 relative to the intended sum). Also, the bare mean() (no na.rm) returns NA if any population in the group is NA, so a single missing population poisons that region's row entirely.

Maintainer-runnable repro

sed -n '188,199p' R/targets/flu_data_targets.R
sed -n '130,140p' R/targets/covid_data_targets.R

Or in R: build a two-state group with populations 1M and 3M and equal region_value — the current formula returns region_value * 2M / 4M-scaled arithmetic that diverges from the intended sum/sum by the state count.

Impact

Miscalibrated nwss_region exogenous predictor in explore sweeps. As far as I can tell prod is unaffected here (the prod path drops nwss rows upstream via drop_na(hhs)), so this is an explore/report-path correctness issue.

Suggested fix

region_value = sum(region_value * population, na.rm = TRUE) / sum(population, na.rm = TRUE),

in both files (flu + covid), matching the sibling statistics.