Compare Counties

Every statistic below is derived at page load from the live POST /v1/policy/query distribution of all counties for the same indicator and vintage. Nothing is hardcoded.

Direction (who is "better")

Taken verbatim from the API's direction field on each indicator (lower_is_better, higher_is_better, or neutral). Indicators marked neutral — or with no direction at all — are shown as context and are never colored or scored, because there is no defensible better/worse reading for them.

Statewide rank

Each county's ordinal position among all counties in its state on that indicator, ordered so that rank 1 is the better end per the indicator's direction. The accompanying percentile uses the mean rank for ties over n − 1, matching pkg/stats.PercentileRank.

Gap in state SD

(value_A − value_B) / SD_state, where SD_state is the population standard deviation of all county values for that indicator, matching pkg/stats.ZScore. Expressing the gap in standard deviations is what makes a poverty-rate gap and an income gap comparable on one scale. Sign is re-oriented so a positive bar always means County A is better off.

Counts normalized per 10,000 residents

Indicators with unit: count are population-confounded — a larger county has more of nearly everything. For rank and gap only, counts are divided by the county's population from the same API response and scaled to a rate per 10,000 residents. Rows treated this way are flagged per 10k and show the rate beside the rank. The value columns still show the raw count the API returned, and on these rows the raw Δ columns are deliberately left uncolored: the raw count difference is a true statement, but it is not the basis on which this page decides who is better off. In this comparison its verdict between the raw-count and per-capita bases. Note that for housing measures the more appropriate denominator would be housing units, which the API does not currently pair to each variable.

Materiality

Two independent checks. Margin of error: where the API publishes margin_of_error for both counties, the difference is tested against √(MOE_A² + MOE_B²) — the Census Bureau's formula for the margin of error of a difference. ACS margins are 90% confidence, so a gap smaller than that is reported as not distinguishable from noise. Distributional size: where no margin of error is published, the gap is described by its size in state SD using Cohen's conventional benchmarks (< 0.2 negligible, < 0.5 small, < 0.8 moderate, otherwise large), applied here to a gap scaled by the statewide county standard deviation. This is a descriptive convention, not a significance test.

Counties between

The count of other counties in the state whose value falls strictly between the two selected counties. This is exact and assumption-free: it answers "how much of the state sits in this gap?" A large SD gap with zero counties between means both counties are outliers at the same end of the distribution.

Reliability dots

Where a margin of error is published, the coefficient of variation (MOE / 1.645) / |estimate| is classified using the Census Bureau thresholds in pkg/stats.ReliabilityLevel: high < 0.15, moderate < 0.30, otherwise low.

Vintage: . Margins of error are published for of indicators in this comparison.