Methods

This page defines the terms and the rules that every chapter shares. Each chapter's methods note, at the end of the chapter, gives the sample, weights, years, estimator and interval method for its own figures, and a chapter that departs from a rule here says so.

Area Deprivation Index

The Area Deprivation IndexArea Deprivation Index (ADI)A score that ranks neighborhoods by income, education, employment and housing conditions drawn from the American Community Survey. In this book a higher national percentile means a more deprived neighborhood. (ADI) ranks neighborhoods by how deprived they are, using census measures of income, education, employment and housing. The book uses its national percentileNational percentileA neighborhood's position in the national ordering, given as the share of all neighborhoods that fall below it. A higher percentile means a higher value than more of the rest of the country.: every neighborhood in the country is ranked against all the others, from one, the least deprived, to one hundred, the most deprived. The publisher's state deciles are not used. The index is published for census block groups, and the book uses it at the tract grain unless a chapter says otherwise. A chart that groups neighborhoods by deprivation groups them in bins of this percentile and names the bins.

The index has a known weakness. It includes home values and rents measured on a national scale, so a neighborhood in a high-cost city where many households are poor can rank as less deprived than its poverty alone would suggest. Read the gradients for those cities with this in mind. Chapter 10.2, The anatomy of a deprivation index shows what goes into the index and how much weight each component carries, and Chapter 10.3, A deprivation index at two zooms shows how the index changes when it is measured for counties instead of tracts.

Population weighting

A headline estimate is population-weightedPopulation-weightedAveraged so that each place counts in proportion to its number of residents, which lets a large county move the result more than a small one. An unweighted average counts every place equally.: it counts each unit in proportion to the people it describes. A tract with many residents counts for more than a tract with few, and a county counts in proportion to its residents or its births, so a small county does not carry the weight of a large one. The source line under a figure says "population-weighted" where this applies, and "unweighted" where every unit counts equally. Where the weighted and unweighted results differ enough to change a reading, the chapter reports both.

A survey estimate uses the survey's own weightsSurvey weightsNumbers attached to survey respondents that say how many people in the population each one represents, so that results describe the whole population and not only the people who answered. where the survey provides them. "Unweighted" under a survey figure means that every respondent counts equally.

Suppressed counts

Public health agencies suppressSuppressionThe withholding of a number because the count behind it is too small to publish without risking the identification of a person. A suppressed value is missing, not zero. small counts to protect privacy. CDC WONDER withholds any count above zero and below ten, whether the count is of deaths or of births, and publishes a count of zero. A withheld count says that the number is small and nothing more, so the book treats it as missing. On a map the unit is drawn as no data, gray with a hatch and the reason in the tooltip, and no chart or table fills it in with a zero or an estimate.

Withholding is not random. It falls on the places with the fewest events, which are often the smallest places, so an average over the units that remain describes a selected group of places. A rate pooled over the published years of a county keeps its years with no deaths and drops its years with a few, so county death rates from WONDER use complete counties, those with a count published in every year of the window, and a comparison that needs every county uses a modeled source such as CDC's county panel. Complete counties are themselves selected on the outcome, and the chapters that use them say so. A map draws a county only if the deaths pooled across the years shown reach a floor of twenty, and the map's notes say so. Chapter 10.5, Suppressed counties in death data asks how much the withholding biases comparisons between counties. Survey cells with too few respondents are likewise left out, and the chapter's methods note gives the minimum.

Modeled estimates

Some measures are modeled estimatesModeled estimateA figure produced by a statistical model instead of a direct count or measurement, usually for small areas where direct data are thin, borrowing strength from surveys, neighboring areas or other data. Neighboring areas can share errors, so single extremes call for caution.: a statistical model produces them, and no one counted them place by place. CDC PLACES gives health measures for every tract by combining a national health survey with census characteristics of each area. The injury and violence panel gives modeled death rates from injury, and its tract file reports estimates for clusters of neighboring tracts, which cannot be split into single tracts. A source line says "modeled estimates" wherever these appear. Where the source publishes an interval for a modeled estimate, the figure draws it, and that interval describes the model's uncertainty.

The model for a PLACES measure uses census characteristics of the same kind that an index such as the ADI is built from. A gradient of a PLACES measure against the ADI is therefore partly built in by the model, and a chapter that plots one against the other says so.

Ecological inference

A pattern across places can differ from the pattern among the people who live in them, in size and even in direction. This is the problem of ecological inferenceEcological inferenceDrawing conclusions about individuals from data about places or groups. A pattern across counties need not hold for the people inside them.. A county with higher deprivation and a higher death rate does not show that the people who died were the deprived ones, and a county average can hide the neighborhoods inside it.

A chapter that works with tracts, counties or states describes places. It reports a difference between places with descriptive verbs such as "is higher where" and "moves with", and it uses a causal verb only inside a box that cites a study whose design supports it. A ZIP code summarizes the exposures and opportunities that gather in a place and is not a cause. Chapters that ask one question of places and of people point to each other, for example Chapter 5.1, Health behaviors and the supply of care and Chapter 5.2, Health behaviors, person by person. The reading guide explains what grain means.

Boundary rules on maps

Exclusions, code mappings and rows not drawn are collapsed under "Map notes" beneath each map. Three places need rules of their own.

Connecticut

The county map draws Connecticut's eight historical counties. Since 2022 the Census Bureau reports Connecticut data for planning regions instead. Data keyed to planning regions are converted to the historical counties before they reach a map, and a map note says so. Connecticut's ZIP code areas are missing from the book's geography file, so where ZIP-area data are shown, a note says that Connecticut is absent.

Alaska

Alaska is drawn, at the grain of its boroughs and census areas, because health and mortality sources report for them. A source that predates a boundary change, such as the 2019 split of the Valdez-Cordova census area or the renaming of the Kusilvak census area, is matched to the current areas, and the map notes say how the codes were matched.

Puerto Rico

Puerto Rico is in some of the book's sources but not on the county map. A map of data that include Puerto Rico says "Puerto Rico not drawn", and a chapter that leaves Puerto Rico out of its estimates says that too.

Data vintages

A number describes a place and a period, and the sources behind one figure often describe different periods. Each figure's source line gives the years. The rules below apply wherever these sources appear, and a figure that joins sources on different tract boundaries says how it matched them.

Where the numbers come from

The data sources page lists each dataset with its years, weights and license. The glossary defines the terms of art that the text marks. The figure gallery lists every figure in the book.