Data
Methodology
How metro values are made from county data, vintages and the latest value, and known coverage gaps.
Cadastre reports measures for metropolitan statistical areas (metros), and micropolitan areas where a publisher covers them, in the Census Bureau's current delineation in the TIGER boundaries. Some publishers publish metro figures; many publish only counties or states. This page explains how a metro value is made in each case, which vintage is read, and where values are missing and why.
Where a metro value comes from
A measure reaches a metro in one of two ways:
- Stored for the metro. The publisher's own metro figure (the American Community Survey's metro estimates, BLS's metro unemployment series), or a figure Cadastre computes for the metro itself (transmission-line miles inside the metro outline, ten-year trends from county data the import reads at both ends of the window). This is used whenever metro rows cover at least 50% of metros.
- Assembled from the metro's counties, by the rule its dataset declares for that measure, when there is no adequate metro figure. Counties belong to metros by the Census Bureau's delineation in the current TIGER vintage.
A measure without a rule is never assembled from counties: a median, for example, cannot be averaged, and a ratio has to be recomputed from its parts. Such a measure has a metro value only where one is stored.
Rollup rules
The rule for each measure is on its dataset's page. These are the rules, in the words the app itself uses, with how many of the available measures use each:
| Rule | What it does | Measures |
|---|---|---|
sum | sum over member counties | 35 |
population_weighted_mean | population-weighted county mean | 40 |
pooled_growth | growth of member counties combined | 1 |
max | highest value among member counties | 10 |
sum_disclosed | sum over member counties that disclose it, a lower bound (unavailable where suppression hides much of the industry) | 20 |
ratio_of_sums | member counties' combined job share against the U.S. share (unavailable where suppression hides much of the industry) | 19 |
weighted_mean | mean of member counties weighted by a declared measure (named on each measure), over the counties with a value (unavailable where those hold under 80% of it) | 18 |
| None | Not assembled from counties: a metro has a value only where one is stored for the metro itself | 41 |
Population weighting. A population-weighted mean weights each county by its total population (the American Community Survey's county population). For a rate whose denominator is not everyone this is an approximation, which is why CDC's adult rates are weighted by the adults CDC counts instead (the weighted_mean rule).
Disclosure and coverage. Publishers withhold some county figures: BLS suppresses a county's employment where a few employers would be identifiable, and CDC suppresses small counts. A withheld figure is stored as unknown, never as zero. The rules that can meet one (sum_disclosed, ratio_of_sums, weighted_mean) use only the counties that disclose, and give the metro no value when those counties hold less than 80% of the metro's coverage basis (its establishments in the industry, or its weight), because the number would then describe a minority of the metro under the name of the whole. At or above that, a sum is a lower bound.
Metro values are Cadastre's computation. Where a metro value is assembled from counties, it is Cadastre's figure, not the publisher's, and the measure's unit says so where the publisher requires it (CDC's measures end "metro value is Cadastre's ... average of CDC county estimates").
Vintages and the latest value
Every stored value carries its vintage: the year (or period) it describes. A dataset can hold several vintages for the same place and measure; the screens and reports read the latest one for each place and measure, which is the newest vintage (and, where two sources hold the same measure for the same year, the first by source name, so the choice is deterministic). Each dataset page states the period its values describe.
Some measures are trends over a window, such as labor-force growth from one year to another. Their labels name the window, so a value is never read as more recent than it is, and each refresh re-derives all of them from one release so revisions are never mixed.
A saved snapshot or report keeps the values it was computed from. A later import does not change it.
Coverage gaps
A missing value is shown as missing, never as zero. In the app, a requirement eliminates a metro that has no value for it and says so (for example "No population data"), so a requirement never passes a metro on data it does not have.
- Puerto Rico. Puerto Rico's metros are in the TIGER boundaries and in the screens. Coverage of Puerto Rico depends on each publisher: where a publisher has no figure for Puerto Rico, Cadastre has none either. The other U.S. Island Areas are not imported.
- Connecticut. Connecticut replaced its eight counties with nine planning regions, which the current Census geography uses. Publishers that still report the legacy counties cannot be matched for every measure: CDC's overdose rates have no Connecticut value, QCEW history has no Connecticut growth values, and the EPA Green Book's county designations are mapped onto the planning regions by overlap. Each dataset page says what applies.
- Suppression. Withheld county figures, described above, leave some metros without a value, mostly small metros and narrow industries.
- Partial surveys. Some measures are not collected everywhere: CDC PLACES has no estimates for Kentucky and Pennsylvania in its current release, and two of its measures come from a module only some states asked.
Measures are described, not judged
Cadastre does not decide whether a higher value is better. Whether a high unemployment rate counts for or against a metro depends on the question being asked, so the direction is set by each criterion in the app, not by the measure.