Overlap methodology

Method and purpose

New work uses method v4 and recipe lookout.convergence-count.v2. Active overlap links without an overlayMethod parameter and saved v3 work retain method v3 and recipe lookout.convergence-count.v1. Work saved under v3 retains its recorded method and recipe and is not silently recomputed under v4. The v3 method card offers “Use the current method (v4)”; choosing it recomputes the overlap under v4 and records overlayMethod=v4 in the link. An artefact names the method it was computed with. The method card records the computed recipe, version and method hash. This is a composite ordering by an equal-weight count of selected criteria, not a published index, a statistical finding or a probability.

Choose up to five tract measures available in the selected Washington view. County-only resilience is excluded. The UI reports counted measures, excluded measures and any rule that prevents a measure from counting. A method change requires a new version; a documentation edit must not quietly redefine the implementation.

Reference set and directions

The reference set depends on the current view. A tract’s raw published value can stay the same while its local percentile decision changes when the comparison geography changes. Record the county, view, selected measures, directions and thresholds with an interpretation.

Choose higher or lower as the direction of concern where allowed. FEMA rating measures have a fixed rating direction. Measures with a choose-direction requirement do not count until a direction is selected. Zero-population tracts are excluded from the resident base.

Population mixes and related measures can strongly affect the result. Preventive-health crude rates can vary with age; disability’s default direction in this overlay differs from its neutral county-ranking treatment elsewhere.

Numeric thresholds and FEMA rating floors

Numeric measures use an exact mid-rank test on served values, after aligning the direction. Available top-percent settings are 10, 20, 25 and 33, with a default of 20. Ties are all in or all out; the selected fraction therefore need not equal the requested percentage exactly. Low reliability does not remove a tract from the ranking.

FEMA measures use national rating classes, not the local numeric-percentile threshold. Available floors are Relatively Moderate, Relatively High and Very High; the default is Relatively Moderate. Very Low and Relatively Low never qualify.

Not Applicable, No Rating and No Expected Annual Losses are assessed but do not meet the FEMA criterion. Insufficient Data and absent FEMA tracts are not assessed. This method treatment does not convert a null publisher value into a numeric score of zero.

Single-measure income maps, tables and profiles show median household incomes published as "$250,000 or more" as tied highest with the publisher’s text, regardless of the overlap method.

Example (illustrative, not source data):
Selected measures: 3
Tract A: met, not met, missing → 1 of 2 assessable
Tract B: not met, not met, not met → 0 of 3 assessable
Tract C: missing, missing, missing → not assessable

Uncertainty markers

For eligible ACS and PLACES measures, a met tick can be marked confident or borderline using one-sided 95% bounds. The comparison is with the middle (median) of the view, not with the chosen percentile cut. Markers never change the map fill, count, totals or order.

PLACES interval conversion assumes each side is approximately normal and the interval equal-tailed; this is not verified by the source package. Shared county-level model uncertainty can overstate uncertainty for within-county tract comparisons; that is Lookout’s inference.

Derived ACS share ranges are approximate. A missing published interval means the marker cannot be assessed, not that the value is certain. FEMA NRI does not publish a range for these scores.

Current method v4 rules

The following rules apply to new overlap work under method v4. Under v3 these top-coded cells remain outside the overlap ranking base.

A top-coded cell is a cell of tract median household income (t:median_household_income_tract) that the publisher reports only as its highest, open-ended bracket ("$250,000 or more"). It is served with no number, reason code TC, and no range. The v4 rule applies to that measure and that reason only.

All eligible top-coded cells form one tied group above every numeric value of the measure before direction is applied. Lower = more concern reverses this order, as for any other value (§3). A top-coded cell is an order token, never 250001, 250000, infinity or any other number in comparison, storage, display or export. Its publisher text is retained as text.

If the central position, or either central position for even m, is top-coded, the median is not determinable and G6 refuses the measure for counting in this view. No numeric-only median is substituted.

When G6 fires, qualification cells remain visible for reference, but uncertainty cannot be assessed for any cell of that measure because the base median is unavailable. Numeric met cells retain their qualification symbol and show "Base median cannot be determined; uncertainty can't be assessed"; they receive neither a confident nor an ≈ marker. The measure contributes nothing to the confident-ticks summary. This treatment applies with markers on or off.

A top-coded cell has no published range. It can meet, but is never confident and never "≈". It shows qualification and "no published range; can't be assessed".

The G4 “because of ties” label is a separate disclosure. In v4 it remains visible even when a counting guard refuses the measure; saved captures, county renders and mitigation narratives and CSV preserve it.

Chance reference and dependence

The chance numbers are a reference point, not a statistical test. They use the qualifying proportions of the counted measures for the displayed class comparison. The observed and chance counts balance across all classes.

Related measures can increase the number of tracts meeting none or many of the criteria. A class above the chance reference may reflect one shared factor rather than several independent concerns. Read the pair/dependence information alongside the count; do not label it causal evidence or statistical significance.

For the method’s correlation calculation only, FEMA Not Applicable, No Rating and No Expected Annual Losses occupy the lowest value. This is a method convention separate from the publisher’s missing-value display.

Reporting an exploration

1. Record recipe/version, comparison view, selected measures and their periods, direction choices, thresholds and any exclusions.
2. State assessable denominators, missingness and whether uncertainty markers were shown. Preserve relevant source attribution and limitations.
3. Inspect tract-level inputs and correlated measures before describing a spatial pattern. Use local expertise to frame follow-up questions.
4. Label the result calculated by Lookout. A permitted U3b Thurston annex capture includes its exact inputs, directions, missingness and limitations. Do not call it a published score, validated risk estimate or diagnosis.

Clinic cross-source matching

Candidate matches require the same county and meaningful name or alias agreement, together with the same normalized street address or a nearby point. Strong candidates require at least 80% name-token agreement and matching street, with no published unit or ZIP conflict; approximate points more than one kilometre apart cannot qualify as strong. Nearby-only candidates remain possible and use a 300-metre search ceiling, tightened to 150 metres for weaker name agreement.

Candidates show their reasons and any specialty overlap, disagreement or missing information. Multiple counterpart candidates remain ambiguous. No records are automatically merged, and records without a candidate are unmatched only within the currently filtered results. Source records, not unique buildings or verified operating clinics, are counted.

Provider-directory specialties are separate from CMS taxonomy descriptions. Search covers published aliases. Directory additions cover nine reviewed locations and do not establish complete outpatient coverage. OSM-only downloads contain the currently filtered OSM records and attribution, without inferred links to registry identities.