Chapter 10.5 · The counties that don't count

Suppressed counties in death data

CDC WONDER withheld the death count in 47.1% to 57.9% of county-years from 2018 to 2023, in county-years that hold 10.1% to 14.8% of residents, and in counties of 25,000 to 250,000 residents those it publishes in every year have higher death rates in CDC's injury and violence panel than those it withholds in some year.

Author: Vishal Singh
Affiliation: NYU Stern
Published: 9 October 2026

CDC WONDER, the federal database of death certificates, withheld the death count for 47.1% of county-years for drug overdose from 2018 to 2023, 57.9% for alcohol-induced deaths and 57.5% for suicide. The withheld county-years hold only 10.1%, 14.8% and 12.2% of residents, so the blanks cover many places and few people. They are also a selected set of places. WONDER publishes a county-year count when it is zero or at least 10 deaths, so a county enters the published data in the years when its count is high. A rate pooled over those years is biased, which here means that it differs from the county's true rate in a systematic direction that depends on the size of the county. It falls below the rate in CDC's modeled injury and violence panel in the smallest counties and rises above it in counties of middling small size. For suicide, the pooled rate is 0.18 (95% CI 0.00 to 0.42) times the panel's rate in counties under 10,000 residents and 1.51 (95% CI 1.47 to 1.55) times in counties of 25,000 to 50,000 residents (Figure 7).

This chapter explains the mechanism behind a rule that every chapter using WONDER follows. A withheld cell is counted and left out of both the numerator and the denominator of a rate, and the book never fills it in. It discloses only the largest share of deaths that the withheld cells could hold at the public limit of 9 deaths each, which is 13.4% of the deaths in the county files for overdose, 29.1% for alcohol-induced deaths and 28.4% for suicide. County rates in other chapters use complete counties, those whose counts are published in every year, or a modeled source. The sections below take up the rule, the counties it removes, the bias it produces and a second rule, which decides which counties WONDER names at all in its birth products. The cause maps belong to Chapter 6.3, Overdose, alcohol and suicide, the birth outcomes to Chapter 1.1, Birthweight and preterm birth, and a figure of suppressed cells by birthweight class to Chapter 10.1, Gradients, slopes and variance.

WONDER withholds small counts and publishes zeros on request

The county files of WONDER's Underlying Cause of Death database show a count above zero and below 10 as Suppressed, and the database's terms bar users from presenting a death count of 9 or fewer, or a rate built on one. A suppressed cell says only that the count is small. It is missing, and a count of zero is published as zero. Zeros appear only when a query asks for them: a default WONDER query omits county-years with no deaths, and the pulls behind this book used the option that shows zero values, so a zero here is a published count and not an absent row. Agencies withhold small counts for two reasons that are easy to confuse.[1] One is confidentiality, which sets the limit of 10 for deaths and births in subnational tables.[2] The other is reliability. Until recently NCHS suppressed or flagged any rate based on fewer than 20 events, the count at which the relative standard error of a Poisson count reaches 22.4%, and its current standard asks for at least 10 events and a 95% confidence interval no wider than 160% of the estimate.[2] The first rule protects the people counted, and the second protects the reader of the rate.

Maps in this book keep the earlier floor of 20 pooled deaths for every year, so that every map follows one rule that is stricter than the current NCHS minimum. Gradients in other chapters use no floor and weight each place by its population. Publication clears WONDER's limit and not the reliability bar. Among county-years that WONDER publishes with at least 10 deaths, 35.4% for overdose, 43.8% for alcohol-induced deaths and 44.8% for suicide have fewer than 20 deaths.

This chapter uses the county files of the Underlying Cause of Death database for 2018 to 2024 (single race), which carry a row for each of 3,142 counties and each year, for three kinds of death. They are drug poisoning of any intent, which the chapter calls overdose, alcohol-induced deaths and suicide. Over 2018 to 2023 that is 18,852 county-years for each kind of death. Rates are crude. Every year from 2018 to 2024 is final in the June 2026 pull, although North Carolina's 2023 overdose counts predate the state's amended death certificates and are low. WONDER publishes a county's population whether or not it withholds the deaths, which makes it possible to measure how many residents live in a county-year with a blank.

Figure 1

In every year from 2018 to 2024 WONDER withheld between 44.6% and 59.9% of county-year death counts, in county-years that hold between 8.6% and 17.4% of residents

Share of county-years whose death count CDC WONDER withholds, and share of residents living in a withheld county-year, by year, for drug overdose, alcohol-induced causes and suicide, all counties in the county files

County-years with the death count withheld (%)

Show data

Data for In every year from 2018 to 2024 WONDER withheld between 44.6% and 59.9% of county-year death counts, in county-years that hold between 8.6% and 17.4% of residents
cause	measure	n	value	year
Drug overdose	counties	1,541	49.05	2018
Drug overdose	residents	1,541	12.31	2018
Drug overdose	counties	1,586	50.48	2019
Drug overdose	residents	1,586	12.32	2019
Drug overdose	counties	1,504	47.87	2020
Drug overdose	residents	1,504	10.09	2020
Drug overdose	counties	1,400	44.56	2021
Drug overdose	residents	1,400	8.79	2021
Drug overdose	counties	1,413	44.97	2022
Drug overdose	residents	1,413	8.56	2022
Drug overdose	counties	1,426	45.39	2023
Drug overdose	residents	1,426	8.64	2023
Drug overdose	counties	1,535	48.85	2024
Drug overdose	residents	1,535	10.78	2024
Alcohol-induced	counties	1,863	59.29	2018
Alcohol-induced	residents	1,863	17.42	2018
Alcohol-induced	counties	1,881	59.87	2019
Alcohol-induced	residents	1,881	16.91	2019
Alcohol-induced	counties	1,805	57.45	2020
Alcohol-induced	residents	1,805	14.26	2020
Alcohol-induced	counties	1,762	56.08	2021
Alcohol-induced	residents	1,762	12.52	2021
Alcohol-induced	counties	1,801	57.32	2022
Alcohol-induced	residents	1,801	13.68	2022
Alcohol-induced	counties	1,797	57.19	2023
Alcohol-induced	residents	1,797	14.10	2023
Alcohol-induced	counties	1,820	57.92	2024
Alcohol-induced	residents	1,820	13.96	2024
Suicide	counties	1,798	57.22	2018
Suicide	residents	1,798	12.26	2018
Suicide	counties	1,820	57.92	2019
Suicide	residents	1,820	12.42	2019
Suicide	counties	1,832	58.31	2020
Suicide	residents	1,832	12.67	2020
Suicide	counties	1,801	57.32	2021
Suicide	residents	1,801	11.90	2021
Suicide	counties	1,792	57.03	2022
Suicide	residents	1,792	12.14	2022
Suicide	counties	1,792	57.03	2023
Suicide	residents	1,792	11.97	2023
Suicide	counties	1,781	56.68	2024
Suicide	residents	1,781	11.81	2024

The withheld share of county-years is large in every year, and the residents it covers are a small share of the population. Drug overdose has the smallest withheld share in every year. The first view plots the share of county-years whose count WONDER withholds, a count above zero and below 10; the second plots the share of residents living in those county-years.Note: These are full counts of the 3,142 counties in WONDER's county files, so the lines carry no sampling interval. A county-year is withheld when its count is above zero and below 10; a count of zero is published.Note: Residents come from the county population that WONDER publishes with every cell, withheld or not. Connecticut's 8 counties have no WONDER population in 2022 to 2024 and are left out of the resident shares in those years.Note: Rates are not plotted. Drug overdose is drug poisoning of any intent, so an overdose recorded as a suicide is in both the drug and the suicide counts. North Carolina's 2023 overdose counts in these files predate the state's amended death certificates and are low.Source: CDC WONDER underlying cause of death (single race), county files, 2018-2024; full counts

In every year from 2018 to 2024 the withheld share ran from 44.6% to 59.9% of county-years across the three kinds of death, and the withheld county-years held between 8.6% and 17.4% of residents. The share of residents fell for overdose, from 12.3% in 2018 to 8.6% in its lowest year, and for alcohol-induced deaths, from 17.4% to 12.5%, while for suicide it stayed between 11.8% and 12.7%. A blank is not a zero. Once a query asks for zeros, they are a large part of what WONDER publishes. Counts of zero are 30.0% of the published overdose county-years, 34.2% of the alcohol-induced ones and 24.7% of the suicide ones. A rate pooled over a county's published years keeps its years with no deaths and drops its years with a few, which is the bias that the later sections measure.

Withholding is heaviest in counties of middling small size

How often a count is withheld depends on how many deaths a county expects. Figure 2 sorts counties by their mean population over 2018 to 2023 into six classes, with boundaries at 10,000, 25,000, 50,000, 100,000 and 250,000 residents. The last two are the populations at which WONDER names a county in its birth products, taken up below.

Figure 2

WONDER withholds 82.3% of overdose county-years in counties of 10,000 to 25,000 residents and 45.2% in smaller counties, where 54.7% are published as zero

Share of county-years withheld, published as zero and published with at least 10 deaths, by size of county (mean residents 2018 to 2023), drug overdose, alcohol-induced causes and suicide

Withheld
Published, zero deaths
Published, 10 or more deaths

Show data

Data for WONDER withholds 82.3% of overdose county-years in counties of 10,000 to 25,000 residents and 45.2% in smaller counties, where 54.7% are published as zero
cause	counties	county_years	segment	share	size
Drug overdose	729	4,374	Withheld	45.18	Under 10,000
Drug overdose	729	4,374	Published, zero deaths	54.66	Under 10,000
Drug overdose	729	4,374	Published, 10 or more deaths	0.160	Under 10,000
Drug overdose	799	4,794	Withheld	82.29	10,000 to 24,999
Drug overdose	799	4,794	Published, zero deaths	11.22	10,000 to 24,999
Drug overdose	799	4,794	Published, 10 or more deaths	6.49	10,000 to 24,999
Drug overdose	619	3,714	Withheld	64.38	25,000 to 49,999
Drug overdose	619	3,714	Published, zero deaths	1.72	25,000 to 49,999
Drug overdose	619	3,714	Published, 10 or more deaths	33.90	25,000 to 49,999
Drug overdose	389	2,334	Withheld	21.77	50,000 to 99,999
Drug overdose	389	2,334	Published, zero deaths	0.090	50,000 to 99,999
Drug overdose	389	2,334	Published, 10 or more deaths	78.15	50,000 to 99,999
Drug overdose	329	1,974	Withheld	2.53	100,000 to 249,999
Drug overdose	329	1,974	Published, zero deaths	0	100,000 to 249,999
Drug overdose	329	1,974	Published, 10 or more deaths	97.47	100,000 to 249,999
Drug overdose	277	1,662	Withheld	0	250,000 or more
Drug overdose	277	1,662	Published, zero deaths	0	250,000 or more
Drug overdose	277	1,662	Published, 10 or more deaths	100	250,000 or more
Alcohol-induced	729	4,374	Withheld	54.94	Under 10,000
Alcohol-induced	729	4,374	Published, zero deaths	44.79	Under 10,000
Alcohol-induced	729	4,374	Published, 10 or more deaths	0.270	Under 10,000
Alcohol-induced	799	4,794	Withheld	84.27	10,000 to 24,999
Alcohol-induced	799	4,794	Published, zero deaths	14.12	10,000 to 24,999
Alcohol-induced	799	4,794	Published, 10 or more deaths	1.61	10,000 to 24,999
Alcohol-induced	619	3,714	Withheld	84.17	25,000 to 49,999
Alcohol-induced	619	3,714	Published, zero deaths	1.99	25,000 to 49,999
Alcohol-induced	619	3,714	Published, 10 or more deaths	13.84	25,000 to 49,999
Alcohol-induced	389	2,334	Withheld	49.14	50,000 to 99,999
Alcohol-induced	389	2,334	Published, zero deaths	0.390	50,000 to 99,999
Alcohol-induced	389	2,334	Published, 10 or more deaths	50.47	50,000 to 99,999
Alcohol-induced	329	1,974	Withheld	9.73	100,000 to 249,999
Alcohol-induced	329	1,974	Published, zero deaths	0	100,000 to 249,999
Alcohol-induced	329	1,974	Published, 10 or more deaths	90.27	100,000 to 249,999
Alcohol-induced	277	1,662	Withheld	0.060	250,000 or more
Alcohol-induced	277	1,662	Published, zero deaths	0	250,000 or more
Alcohol-induced	277	1,662	Published, 10 or more deaths	99.94	250,000 or more
Suicide	729	4,374	Withheld	61.34	Under 10,000
Suicide	729	4,374	Published, zero deaths	38.45	Under 10,000
Suicide	729	4,374	Published, 10 or more deaths	0.210	Under 10,000
Suicide	799	4,794	Withheld	93.12	10,000 to 24,999
Suicide	799	4,794	Published, zero deaths	5.74	10,000 to 24,999
Suicide	799	4,794	Published, 10 or more deaths	1.15	10,000 to 24,999
Suicide	619	3,714	Withheld	80.67	25,000 to 49,999
Suicide	619	3,714	Published, zero deaths	0.590	25,000 to 49,999
Suicide	619	3,714	Published, 10 or more deaths	18.74	25,000 to 49,999
Suicide	389	2,334	Withheld	28.36	50,000 to 99,999
Suicide	389	2,334	Published, zero deaths	0.040	50,000 to 99,999
Suicide	389	2,334	Published, 10 or more deaths	71.59	50,000 to 99,999
Suicide	329	1,974	Withheld	1.52	100,000 to 249,999
Suicide	329	1,974	Published, zero deaths	0	100,000 to 249,999
Suicide	329	1,974	Published, 10 or more deaths	98.48	100,000 to 249,999
Suicide	277	1,662	Withheld	0	250,000 or more
Suicide	277	1,662	Published, zero deaths	0	250,000 or more
Suicide	277	1,662	Published, 10 or more deaths	100	250,000 or more

The withheld segment is longest for counties of middling size and short at both ends: in the smallest counties most county-years have no deaths and WONDER publishes the zero, and in the largest counties nearly every count reaches 10. Counties of 100,000 or more residents are withheld in few county-years, and counties of 250,000 or more in almost none. Each bar splits the county-years of one size of county. The buttons switch the kind of death.Note: Size is a county's mean WONDER population over 2018 to 2023. The boundaries are 10,000, 25,000, 50,000, 100,000 and 250,000 residents; the last two are the populations at which WONDER names a county in its birth products, although a county's class here and its being named do not coincide exactly.Note: These are shares of all county-years in each class, including Connecticut's 8 counties, so the bars are full counts with no interval. A published zero is a real zero; a withheld count is missing.Source: CDC WONDER underlying cause of death (single race), county files, 2018-2023; full counts

Counties of 10,000 to 25,000 residents are withheld most often, in 82.3% of overdose county-years, 84.3% of alcohol-induced county-years and 93.1% of suicide county-years. Counties under 10,000 residents are withheld less, in 45.2%, 54.9% and 61.3%, because many of their county-years have no deaths and WONDER publishes the zero. Published zeros are 54.7%, 44.8% and 38.5% of their county-years. At the large end, counties of 100,000 to 250,000 residents are withheld in 2.5% of overdose county-years, 9.7% of alcohol-induced county-years and 1.5% of suicide county-years, and counties of 250,000 or more are withheld in 0.0% of overdose, 0.1% of alcohol-induced and 0.0% of suicide county-years. A county that expects more than a handful of deaths in a year but fewer than the limit is withheld in most years, and the section on bias below works out how much that selects.

Withheld years are fewest in the Northeast and the Pacific states

Figure 3 counts, for each county, the years from 2018 to 2023 in which its count is withheld. A county published in every year has a count, possibly zero, and a population in each of those years. That is a complete county, and other than Connecticut's counties there are 942 for overdose, 644 for alcohol-induced deaths and 745 for suicide. At the other end, 591 counties for overdose, 782 for alcohol-induced deaths and 919 for suicide are withheld in every year. The overdose map shades 950 counties at no withheld year, which are the 942 complete counties and Connecticut's 8 counties, which have deaths but no population in 2022 and 2023 and so are not complete. The alcohol-induced and suicide maps shade 652 and 752 counties at no withheld year, again the complete counties and the Connecticut counties with no withheld year.

Figure 3

WONDER withheld the overdose count in every year from 2018 to 2023 in 591 counties and in no year in 950

Number of years from 2018 to 2023 in which CDC WONDER withholds a county's death count, by county, for drug overdose, alcohol-induced causes and suicide

Loading county values

Years the drug overdose count is withheld, 2018 to 2023

Published every yearWithheld every year

No data (3,143 counties)

Select a class to show only those places.

Find a county

Hover or tap a county for its value, or search by name. Zoom with the buttons on the map, then drag to move it.

Counties toward the low end of the key have a published count in every year and counties toward the high end have a published count in none, so every county has a value. Counties in New England, the Middle Atlantic and the Pacific states average fewer withheld years than counties in any other Census division, for each kind of death. The buttons switch the kind of death; tooltips give the years published as zero and with 10 or more deaths.Note: A count is withheld when it is above zero and below 10, and the key counts the years it is withheld out of the years from 2018 to 2023. A county with no withheld year can still have a count of zero in some years.Note: The overdose view shades 950 counties at no withheld year: the 942 complete counties and Connecticut's 8, which have deaths but no population in 2022 and 2023.Note: WONDER codes Valdez-Cordova AK as one county, so its counts are drawn on both successor areas, Chugach and Copper River; Kusilvak AK and Oglala Lakota SD are drawn under their current codes. Connecticut's 8 counties are drawn from their legacy codes, and WONDER has no population for them in 2022 and 2023. Puerto Rico is not drawn.Source: CDC WONDER underlying cause of death (single race), county files, 2018-2023; full counts

Counties in New England, the Middle Atlantic and the Pacific division average fewer withheld years than counties in any other Census division, for each kind of death. For suicide counts, 32.7% of counties in the South and 31.1% in the Midwest are withheld in every year, against 22.5% in the West and 12.0% in the Northeast. The map shows where blanks occur, and it says nothing about whether the deaths behind a blank are many or few.

Residents of withheld counties live in smaller, more rural and more deprived places

The counties fall into three groups by how often WONDER withholds their count: in every year, in some years or in no year. Take overdose. The 942 counties published in every year hold 83.8% of the residents of the 3,134 counties other than Connecticut's. The 1,601 counties withheld in some years are 51.1% of counties and hold 12.1% of residents, and the 591 counties withheld in every year hold 4.1% (Figure 4).

Figure 4

Counties whose overdose count WONDER publishes in every year hold 83.8% of residents, and counties withheld in every year hold 4.1% and are smaller, more rural and more deprived

Counties by how many years from 2018 to 2023 WONDER withholds the death count: share of counties and residents, median size, nonmetro residents, Area Deprivation Index rank and race and ethnicity of residents, other than Connecticut's counties

Share of counties (%)

Share of residents (%)

Median county residents (thousands)

Residents in nonmetro counties (%)

Mean Area Deprivation Index rank (1 least, 100 most deprived)

Residents who are Hispanic (%)

Residents who are non-Hispanic Black (%)

Residents who are non-Hispanic white (%)

Residents who are American Indian or Alaska Native (%)

Value in the unit of each panel title

Show data

Data for Counties whose overdose count WONDER publishes in every year hold 83.8% of residents, and counties withheld in every year hold 4.1% and are smaller, more rural and more deprived
cause	estimate	group	measure	n_units
Drug overdose	30.06	Published every year	Share of counties (%)	942
Drug overdose	83.81	Published every year	Share of residents (%)	942
Drug overdose	134.4	Published every year	Median county residents (thousands)	942
Drug overdose	4.16	Published every year	Residents in nonmetro counties (%)	942
Drug overdose	43.04	Published every year	Mean Area Deprivation Index rank (1 least, 100 most deprived)	942
Drug overdose	20.97	Published every year	Residents who are Hispanic (%)	942
Drug overdose	13.37	Published every year	Residents who are non-Hispanic Black (%)	942
Drug overdose	55.34	Published every year	Residents who are non-Hispanic white (%)	942
Drug overdose	1.18	Published every year	Residents who are American Indian or Alaska Native (%)	942
Drug overdose	51.08	Withheld in some years	Share of counties (%)	1,601
Drug overdose	12.09	Withheld in some years	Share of residents (%)	1,601
Drug overdose	16.08	Withheld in some years	Median county residents (thousands)	1,601
Drug overdose	61.06	Withheld in some years	Residents in nonmetro counties (%)	1,601
Drug overdose	67.71	Withheld in some years	Mean Area Deprivation Index rank (1 least, 100 most deprived)	1,601
Drug overdose	9.66	Withheld in some years	Residents who are Hispanic (%)	1,601
Drug overdose	9.38	Withheld in some years	Residents who are non-Hispanic Black (%)	1,601
Drug overdose	75.91	Withheld in some years	Residents who are non-Hispanic white (%)	1,601
Drug overdose	2.12	Withheld in some years	Residents who are American Indian or Alaska Native (%)	1,601
Drug overdose	18.86	Withheld in every year	Share of counties (%)	591
Drug overdose	4.10	Withheld in every year	Share of residents (%)	591
Drug overdose	19.76	Withheld in every year	Median county residents (thousands)	591
Drug overdose	75.52	Withheld in every year	Residents in nonmetro counties (%)	591
Drug overdose	70.17	Withheld in every year	Mean Area Deprivation Index rank (1 least, 100 most deprived)	591
Drug overdose	10.00	Withheld in every year	Residents who are Hispanic (%)	591
Drug overdose	8.11	Withheld in every year	Residents who are non-Hispanic Black (%)	591
Drug overdose	77.24	Withheld in every year	Residents who are non-Hispanic white (%)	591
Drug overdose	2.13	Withheld in every year	Residents who are American Indian or Alaska Native (%)	591
Alcohol-induced	20.55	Published every year	Share of counties (%)	644
Alcohol-induced	77.87	Published every year	Share of residents (%)	644
Alcohol-induced	207.1	Published every year	Median county residents (thousands)	644
Alcohol-induced	2.07	Published every year	Residents in nonmetro counties (%)	644
Alcohol-induced	41.45	Published every year	Mean Area Deprivation Index rank (1 least, 100 most deprived)	644
Alcohol-induced	22.08	Published every year	Residents who are Hispanic (%)	644
Alcohol-induced	13.34	Published every year	Residents who are non-Hispanic Black (%)	644
Alcohol-induced	53.77	Published every year	Residents who are non-Hispanic white (%)	644
Alcohol-induced	1.26	Published every year	Residents who are American Indian or Alaska Native (%)	644
Alcohol-induced	54.50	Withheld in some years	Share of counties (%)	1,708
Alcohol-induced	15.76	Withheld in some years	Share of residents (%)	1,708
Alcohol-induced	15.46	Withheld in some years	Median county residents (thousands)	1,708
Alcohol-induced	50.04	Withheld in some years	Residents in nonmetro counties (%)	1,708
Alcohol-induced	65.48	Withheld in some years	Mean Area Deprivation Index rank (1 least, 100 most deprived)	1,708
Alcohol-induced	8.85	Withheld in some years	Residents who are Hispanic (%)	1,708
Alcohol-induced	10.62	Withheld in some years	Residents who are non-Hispanic Black (%)	1,708
Alcohol-induced	75.66	Withheld in some years	Residents who are non-Hispanic white (%)	1,708
Alcohol-induced	1.54	Withheld in some years	Residents who are American Indian or Alaska Native (%)	1,708
Alcohol-induced	24.95	Withheld in every year	Share of counties (%)	782
Alcohol-induced	6.37	Withheld in every year	Share of residents (%)	782
Alcohol-induced	23.70	Withheld in every year	Median county residents (thousands)	782
Alcohol-induced	70.16	Withheld in every year	Residents in nonmetro counties (%)	782
Alcohol-induced	71.23	Withheld in every year	Mean Area Deprivation Index rank (1 least, 100 most deprived)	782
Alcohol-induced	8.84	Withheld in every year	Residents who are Hispanic (%)	782
Alcohol-induced	9.60	Withheld in every year	Residents who are non-Hispanic Black (%)	782
Alcohol-induced	77.47	Withheld in every year	Residents who are non-Hispanic white (%)	782
Alcohol-induced	1.62	Withheld in every year	Residents who are American Indian or Alaska Native (%)	782
Suicide	23.77	Published every year	Share of counties (%)	745
Suicide	81.97	Published every year	Share of residents (%)	745
Suicide	176.5	Published every year	Median county residents (thousands)	745
Suicide	2.89	Published every year	Residents in nonmetro counties (%)	745
Suicide	42.37	Published every year	Mean Area Deprivation Index rank (1 least, 100 most deprived)	745
Suicide	21.38	Published every year	Residents who are Hispanic (%)	745
Suicide	13.36	Published every year	Residents who are non-Hispanic Black (%)	745
Suicide	54.77	Published every year	Residents who are non-Hispanic white (%)	745
Suicide	1.22	Published every year	Residents who are American Indian or Alaska Native (%)	745
Suicide	46.90	Withheld in some years	Share of counties (%)	1,470
Suicide	12.23	Withheld in some years	Share of residents (%)	1,470
Suicide	16.53	Withheld in some years	Median county residents (thousands)	1,470
Suicide	60.19	Withheld in some years	Residents in nonmetro counties (%)	1,470
Suicide	67.54	Withheld in some years	Mean Area Deprivation Index rank (1 least, 100 most deprived)	1,470
Suicide	9.02	Withheld in some years	Residents who are Hispanic (%)	1,470
Suicide	9.58	Withheld in some years	Residents who are non-Hispanic Black (%)	1,470
Suicide	76.38	Withheld in some years	Residents who are non-Hispanic white (%)	1,470
Suicide	1.89	Withheld in some years	Residents who are American Indian or Alaska Native (%)	1,470
Suicide	29.32	Withheld in every year	Share of counties (%)	919
Suicide	5.80	Withheld in every year	Share of residents (%)	919
Suicide	18.89	Withheld in every year	Median county residents (thousands)	919
Suicide	73.05	Withheld in every year	Residents in nonmetro counties (%)	919
Suicide	71.44	Withheld in every year	Mean Area Deprivation Index rank (1 least, 100 most deprived)	919
Suicide	9.12	Withheld in every year	Residents who are Hispanic (%)	919
Suicide	9.38	Withheld in every year	Residents who are non-Hispanic Black (%)	919
Suicide	77.41	Withheld in every year	Residents who are non-Hispanic white (%)	919
Suicide	1.72	Withheld in every year	Residents who are American Indian or Alaska Native (%)	919

Within each panel, compare the dots for the three groups. For overdose, counties withheld in every year have a median of 19,761 residents against 134,390 in counties published every year, and 75.5% of their residents live in nonmetro counties against 4.2%. Their mean deprivation rank is 70.2 against 43.0. They are less often Hispanic or non-Hispanic Black and more often non-Hispanic white or American Indian or Alaska Native. Panels of means plot the population-weighted mean over counties with a 95% interval from a bootstrap over counties. The buttons switch the kind of death.Note: Shares of counties and residents and the median are full counts of the 3,134 counties other than Connecticut's 8 and carry no interval. Residents are the mean WONDER population over 2018 to 2023. The other panels are population-weighted means across counties, so an interval describes the spread of county values.Note: Deprivation is the mean of the Area Deprivation Index national ranks of a county's tracts, weighted by 2019-2023 tract population, and the index understates deprivation in high-cost counties. Nonmetro counties have a Rural-Urban Continuum Code of 4 or higher.Note: Race and ethnicity shares come from County Health Rankings 2024 (Census Population Estimates) and use that source's categories: Hispanic of any race, and non-Hispanic Black and white. They describe a county's residents, not the people who died.Source: CDC WONDER underlying cause of death, county files, 2018-2023, Area Deprivation Index 2023, Rural-Urban Continuum Codes 2023 and County Health Rankings 2024; full counts and population-weighted means

The counties withheld in every year are small and rural. Their median county has 19,761 residents, against 134,390 in counties published every year, and 75.5% (95% CI 71.2% to 79.9%) of their residents live in nonmetro counties, those with a Rural-Urban Continuum Code of 4 or higher, against 4.2% (95% CI 3.4% to 5.1%) in counties published every year. They are more deprived: the population-weighted mean Area Deprivation Index rank is 70.2 (95% CI 68.5 to 71.8) against 43.0 (95% CI 39.9 to 46.0), where a higher rank is more deprived. By County Health Rankings' figures their residents are less often Hispanic or non-Hispanic Black and more often non-Hispanic white or American Indian or Alaska Native: 10.0% (95% CI 8.4% to 11.7%) are Hispanic against 21.0% (95% CI 18.2% to 24.0%), 8.1% (95% CI 7.1% to 9.2%) are non-Hispanic Black against 13.4% (95% CI 12.1% to 14.7%), 77.2% (95% CI 75.3% to 79.1%) are non-Hispanic white against 55.3% (95% CI 52.1% to 58.5%) and 2.1% (95% CI 1.7% to 2.6%) are American Indian or Alaska Native against 1.2% (95% CI 1.0% to 1.3%). The alcohol-induced and suicide views show differences between counties published every year and counties withheld in every year in the same direction on every measure. For alcohol-induced deaths the counties withheld in every year number 782 and hold 6.4% of residents, and for suicide they number 919 and hold 5.8%, with 1,708 and 1,470 counties withheld in some years.

These are properties of counties and of their residents, and not of the people who died (ecological inference). For a comparison of counties, the consequence is that published counts describe the larger, more urban and less deprived counties, with their higher shares of Hispanic and Black residents, and leave out the rural counties where deaths are few.

Counties published every year have higher death rates than counties of their size withheld at least once, except among the smallest

A county's count is published when it is high, so the counties published in every year are the ones with high counts. Testing that needs a rate that does not depend on WONDER's rule. CDC's Mapping Injury, Overdose, and Violence panel supplies one for overdose and suicide. It gives modeled estimates of death rates for every county except Connecticut's, and where a count is under 10 a model that borrows from neighboring counties replaces it. A county's rate here is the mean of its annual rates for 2019 to 2023, kept when CDC publishes a stable rate in each year. This comparison uses the same years for WONDER counts, so a county counts as published every year when its count is published in each year from 2019 to 2023.

Figure 5 compares the panel's rate in counties that WONDER publishes in every year with the rate in counties where it withholds the count at least once, which include counties it never publishes, within a size of county. Among counties of 25,000 to 50,000 residents the overdose rate is 52.3 (95% CI 47.3 to 57.8) deaths per hundred thousand residents where WONDER publishes every year and 22.3 (95% CI 21.5 to 23.2) where it withholds the count at least once. Among counties of 50,000 to 100,000 residents the rates are 35.8 (95% CI 33.5 to 38.5) and 17.5 (95% CI 16.5 to 18.5), and among counties of 100,000 to 250,000 residents they are 27.6 (95% CI 26.2 to 29.2) and 11.3 (95% CI 10.2 to 12.6). For suicide the rates are 30.9 (95% CI 27.5 to 34.9) and 19.4 (95% CI 19.0 to 19.8) among counties of 25,000 to 50,000 residents, 21.6 (95% CI 20.7 to 22.5) and 17.0 (95% CI 16.6 to 17.4) among counties of 50,000 to 100,000 residents, and 17.0 (95% CI 16.4 to 17.5) and 12.4 (95% CI 11.3 to 13.7) among counties of 100,000 to 250,000 residents. A group needs at least 10 counties to be plotted, and few counties of the middle sizes are published in every year, which is itself a result of the rule.

Figure 5

Among counties of 25,000 to 50,000 residents, the panel's overdose rate is 52.3 where WONDER publishes every year and 22.3 where it withholds the count in at least one year, and the order reverses under 10,000 residents

CDC modeled death rate per 100,000 residents, mean of annual rates for 2019 to 2023, in counties whose WONDER count is published in each year 2019 to 2023 and in counties where it is withheld in at least one of those years, including counties never published, by size of county, population-weighted

Published in every year
Withheld in at least one year

Show data

Data for Among counties of 25,000 to 50,000 residents, the panel's overdose rate is 52.3 where WONDER publishes every year and 22.3 where it withholds the count in at least one year, and the order reverses under 10,000 residents
cause	estimate	n_units	size	status
Drug overdose	0	128	Under 10,000	Published in every year
Drug overdose	12.97	326	Under 10,000	Withheld in at least one year
Drug overdose	78.76	11	10,000 to 24,999	Published in every year
Drug overdose	20.98	662	10,000 to 24,999	Withheld in at least one year
Drug overdose	52.28	66	25,000 to 49,999	Published in every year
Drug overdose	22.33	536	25,000 to 49,999	Withheld in at least one year
Drug overdose	35.83	227	50,000 to 99,999	Published in every year
Drug overdose	17.54	160	50,000 to 99,999	Withheld in at least one year
Drug overdose	27.63	306	100,000 to 249,999	Published in every year
Drug overdose	11.26	19	100,000 to 249,999	Withheld in at least one year
Drug overdose	28.41	273	250,000 or more	Published in every year
Suicide	0	41	Under 10,000	Published in every year
Suicide	14.23	365	Under 10,000	Withheld in at least one year
Suicide	18.27	406	10,000 to 24,999	Withheld in at least one year
Suicide	30.91	12	25,000 to 49,999	Published in every year
Suicide	19.41	472	25,000 to 49,999	Withheld in at least one year
Suicide	21.57	145	50,000 to 99,999	Published in every year
Suicide	17.03	234	50,000 to 99,999	Withheld in at least one year
Suicide	17.01	304	100,000 to 249,999	Published in every year
Suicide	12.40	21	100,000 to 249,999	Withheld in at least one year
Suicide	12.54	273	250,000 or more	Published in every year

Within a size of county, compare the two dots: counties whose count WONDER publishes in every year have higher death rates than counties of the same size where it withholds the count at least once, except in the smallest counties, where a county is published every year only if it has no deaths at all. Each dot is the population-weighted mean of the panel's county rates with a 95% interval from a bootstrap over counties. The second view shows suicide, where the rate is 21.6 in counties of 50,000 to 100,000 residents published every year and 17.0 in counties withheld at least once.Note: The comparison uses CDC's injury and violence panel because its rates do not depend on WONDER's publication rule. They are modeled estimates, built partly from neighboring counties where deaths are few. Where WONDER withholds a county-year, 70.2% of the panel's overdose county-years are modeled, against 1.1% in counties published every year, so part of each gap is a direct rate against a model.Note: A group of fewer than 10 counties is not plotted, and an interval is absent where every county has a rate of zero. Overdose in the panel is unintentional or undetermined-intent drug overdose. Connecticut is not in the panel. WONDER counts and panel rates both cover 2019 to 2023, and the group withheld in at least one year includes counties that WONDER never publishes in those years.Note: In the smallest counties, counties published every year are the counties with no deaths in any year, because a count of 10 is almost out of reach there.Source: CDC WONDER underlying cause of death, county files, 2019-2023, and CDC Mapping Injury, Overdose, and Violence county file, 2019-2023; crude rates, modeled estimates, population-weighted

The order reverses in the smallest counties. A county under 10,000 residents is published in every year only if its count is zero in every year, because it almost never reaches 10, so the panel's rate for those counties is 0.0 for overdose and 0.0 for suicide, against 13.0 (95% CI 11.7 to 14.3) and 14.2 (95% CI 13.2 to 15.4) where it is withheld at least once. The rule selects on the outcome in both directions. A county of middling size has to have a high count in every year to be complete, and a very small county has to have none.

Part of each gap is a direct rate set against a model. In counties published every year the panel's overdose rates are modeled in 1.1% of county-years and its suicide rates in 0.0%, against 70.2% and 72.5% in counties where WONDER withholds the count at least once. A modeled rate borrows from neighbors, and the rates of small counties are noisy before any suppression, so a map of them can mislead on its own.[3] The comparison shows the direction and rough size of the selection. It does not give the true rate of any withheld county.

The bias of a pooled rate follows from the rule

The rule alone fixes the sign of the bias, and the sign depends on how many deaths a county expects. Suppose a county has a constant true rate and its deaths in a year are a Poisson count, the standard model for deaths in a small population.[4][2] WONDER publishes the year when the count is zero or at least 10. Figure 6 plots, for counties that share an expected number of deaths a year, the rate pooled over their published years divided by their true rate. The ratio is far below one when a county expects only a few deaths, because nearly every published county-year is then a zero. It reaches one near 4.2 expected deaths and peaks at 1.8 at 5.4 expected deaths, where only 5.3% of county-years are published and nearly all of them have a high count. Above 14.2 expected deaths it is within 5% of one.

Figure 6

A rate pooled over published county-years is 1.8 times the true rate at 5.4 expected deaths a year, below it under 4.2 and within 5% of it above 14.2

Expected pooled rate over the county-years WONDER publishes (counts of zero and of at least 10) divided by the true rate, against the expected number of deaths in a county-year, Poisson counts and a constant rate

Pooled rate over published county-years relative to the true rate (1 = unbiased)

Show data

Data for A rate pooled over published county-years is 1.8 times the true rate at 5.4 expected deaths a year, below it under 4.2 and within 5% of it above 14.2
lam	published	ratio
0.100	90.48	0
0.200	81.87	0
0.300	74.08	0
0.400	67.03	0
0.500	60.65	0
0.600	54.88	0
0.700	49.66	0
0.800	44.93	0
0.900	40.66	0
1	36.79	0
1.10	33.29	0
1.20	30.12	0
1.30	27.25	0
1.40	24.66	0.000
1.50	22.31	0.000
1.60	20.19	0.000
1.70	18.27	0.000
1.80	16.53	0.001
1.90	14.96	0.001
2	13.54	0.002
2.10	12.25	0.003
2.20	11.09	0.004
2.30	10.04	0.006
2.40	9.09	0.009
2.50	8.24	0.014
2.60	7.46	0.020
2.70	6.77	0.028
2.80	6.15	0.040
2.90	5.59	0.055
3	5.09	0.075
3.10	4.64	0.101
3.20	4.25	0.134
3.30	3.91	0.177
3.40	3.61	0.230
3.50	3.35	0.295
3.60	3.13	0.372
3.70	2.96	0.463
3.80	2.82	0.567
3.90	2.71	0.683
4	2.65	0.808
4.10	2.61	0.938
4.20	2.61	1.07
4.30	2.65	1.20
4.40	2.72	1.32
4.50	2.82	1.43
4.60	2.96	1.52
4.70	3.13	1.61
4.80	3.34	1.67
4.90	3.58	1.73
5	3.86	1.77
5.10	4.17	1.79
5.20	4.52	1.81
5.30	4.91	1.82
5.40	5.33	1.83
5.50	5.79	1.83
5.60	6.28	1.82
5.70	6.82	1.81
5.80	7.39	1.80
5.90	8.00	1.78
6	8.64	1.77
6.10	9.32	1.75
6.20	10.04	1.73
6.30	10.80	1.72
6.40	11.59	1.70
6.50	12.41	1.68
6.60	13.27	1.66
6.70	14.17	1.64
6.80	15.09	1.62
6.90	16.05	1.61
7	17.04	1.59
7.10	18.06	1.57
7.20	19.11	1.56
7.30	20.19	1.54
7.40	21.29	1.52
7.50	22.41	1.51
7.60	23.57	1.49
7.70	24.74	1.48
7.80	25.93	1.46
7.90	27.14	1.45
8	28.37	1.44
8.10	29.62	1.42
8.20	30.88	1.41
8.30	32.15	1.40
8.40	33.43	1.39
8.50	34.72	1.37
8.60	36.02	1.36
8.70	37.33	1.35
8.80	38.64	1.34
8.90	39.96	1.33
9	41.27	1.32
9.10	42.59	1.31
9.20	43.90	1.30
9.30	45.22	1.29
9.40	46.52	1.28
9.50	47.83	1.27
9.60	49.12	1.26
9.70	50.41	1.25
9.80	51.69	1.25
9.90	52.95	1.24
10	54.21	1.23
10.10	55.46	1.22
10.20	56.69	1.22
10.30	57.90	1.21
10.40	59.10	1.20
10.50	60.29	1.20
10.60	61.46	1.19
10.70	62.61	1.18
10.80	63.74	1.18
10.90	64.86	1.17
11	65.95	1.16
11.10	67.03	1.16
11.20	68.08	1.15
11.30	69.11	1.15
11.40	70.13	1.14
11.50	71.12	1.14
11.60	72.09	1.13
11.70	73.04	1.13
11.80	73.97	1.12
11.90	74.88	1.12
12	75.76	1.12
12.10	76.62	1.11
12.20	77.47	1.11
12.30	78.28	1.10
12.40	79.08	1.10
12.50	79.86	1.10
12.60	80.61	1.09
12.70	81.34	1.09
12.80	82.06	1.09
12.90	82.75	1.08
13	83.42	1.08
13.10	84.07	1.08
13.20	84.70	1.07
13.30	85.31	1.07
13.40	85.90	1.07
13.50	86.47	1.07
13.60	87.03	1.06
13.70	87.56	1.06
13.80	88.08	1.06
13.90	88.58	1.06
14	89.06	1.05
14.10	89.53	1.05
14.20	89.97	1.05
14.30	90.41	1.05
14.40	90.82	1.04
14.50	91.22	1.04
14.60	91.61	1.04
14.70	91.98	1.04
14.80	92.34	1.04
14.90	92.68	1.04
15	93.02	1.03
15.10	93.33	1.03
15.20	93.64	1.03
15.30	93.93	1.03
15.40	94.21	1.03
15.50	94.48	1.03
15.60	94.74	1.03
15.70	94.99	1.03
15.80	95.22	1.02
15.90	95.45	1.02
16	95.67	1.02
16.10	95.88	1.02
16.20	96.08	1.02
16.30	96.27	1.02
16.40	96.45	1.02
16.50	96.63	1.02
16.60	96.79	1.02
16.70	96.95	1.02
16.80	97.10	1.02
16.90	97.25	1.01
17	97.39	1.01
17.10	97.52	1.01
17.20	97.65	1.01
17.30	97.77	1.01
17.40	97.88	1.01
17.50	97.99	1.01
17.60	98.09	1.01
17.70	98.19	1.01
17.80	98.29	1.01
17.90	98.38	1.01
18	98.46	1.01
18.10	98.54	1.01
18.20	98.62	1.01
18.30	98.69	1.01
18.40	98.76	1.01
18.50	98.83	1.01
18.60	98.89	1.01
18.70	98.95	1.01
18.80	99.01	1.01
18.90	99.06	1.01
19	99.11	1.00
19.10	99.16	1.00
19.20	99.21	1.00
19.30	99.25	1.00
19.40	99.29	1.00
19.50	99.33	1.00
19.60	99.37	1.00
19.70	99.41	1.00
19.80	99.44	1.00
19.90	99.47	1.00
20	99.50	1.00
20.10	99.53	1.00
20.20	99.56	1.00
20.30	99.58	1.00
20.40	99.61	1.00
20.50	99.63	1.00
20.60	99.65	1.00
20.70	99.67	1.00
20.80	99.69	1.00
20.90	99.71	1.00
21	99.72	1.00
21.10	99.74	1.00
21.20	99.75	1.00
21.30	99.77	1.00
21.40	99.78	1.00
21.50	99.80	1.00
21.60	99.81	1.00
21.70	99.82	1.00
21.80	99.83	1.00
21.90	99.84	1.00
22	99.85	1.00
22.10	99.86	1.00
22.20	99.87	1.00
22.30	99.88	1.00
22.40	99.88	1.00
22.50	99.89	1.00
22.60	99.90	1.00
22.70	99.90	1.00
22.80	99.91	1.00
22.90	99.91	1.00
23	99.92	1.00
23.10	99.92	1.00
23.20	99.93	1.00
23.30	99.93	1.00
23.40	99.94	1.00
23.50	99.94	1.00
23.60	99.94	1.00
23.70	99.95	1.00
23.80	99.95	1.00
23.90	99.95	1.00
24	99.96	1.00
24.10	99.96	1.00
24.20	99.96	1.00
24.30	99.97	1.00
24.40	99.97	1.00
24.50	99.97	1.00
24.60	99.97	1.00
24.70	99.97	1.00
24.80	99.97	1.00
24.90	99.98	1.00
25	99.98	1.00
25.10	99.98	1.00
25.20	99.98	1.00
25.30	99.98	1.00
25.40	99.98	1.00
25.50	99.98	1.00
25.60	99.98	1.00
25.70	99.99	1.00
25.80	99.99	1.00
25.90	99.99	1.00
26	99.99	1.00
26.10	99.99	1.00
26.20	99.99	1.00
26.30	99.99	1.00
26.40	99.99	1.00
26.50	99.99	1.00
26.60	99.99	1.00
26.70	99.99	1
26.80	99.99	1
26.90	99.99	1
27	99.99	1
27.10	100.00	1
27.20	100.00	1
27.30	100.00	1
27.40	100.00	1
27.50	100.00	1
27.60	100.00	1
27.70	100.00	1
27.80	100.00	1
27.90	100.00	1
28	100.00	1
28.10	100.00	1
28.20	100.00	1
28.30	100.00	1
28.40	100.00	1
28.50	100.00	1
28.60	100.00	1
28.70	100.00	1
28.80	100.00	1
28.90	100.00	1
29	100.00	1
29.10	100.00	1
29.20	100.00	1
29.30	100.00	1
29.40	100.00	1
29.50	100.00	1
29.60	100.00	1
29.70	100.00	1
29.80	100.00	1
29.90	100.00	1
30	100.00	1
30.10	100.00	1
30.20	100.00	1
30.30	100.00	1
30.40	100.00	1
30.50	100	1
30.60	100	1
30.70	100	1
30.80	100	1
30.90	100	1
31	100	1
31.10	100	1
31.20	100	1
31.30	100	1
31.40	100	1
31.50	100	1
31.60	100	1
31.70	100	1
31.80	100	1
31.90	100	1
32	100	1
32.10	100	1
32.20	100	1
32.30	100	1
32.40	100	1
32.50	100	1
32.60	100	1
32.70	100	1
32.80	100	1
32.90	100	1
33	100	1
33.10	100	1
33.20	100	1
33.30	100	1
33.40	100	1
33.50	100	1
33.60	100	1
33.70	100	1
33.80	100	1
33.90	100	1
34	100	1
34.10	100	1
34.20	100	1
34.30	100	1
34.40	100	1
34.50	100	1
34.60	100	1
34.70	100	1
34.80	100	1
34.90	100	1
35	100	1
35.10	100	1
35.20	100	1
35.30	100	1
35.40	100	1
35.50	100	1
35.60	100	1
35.70	100	1
35.80	100	1
35.90	100	1
36	100	1
36.10	100	1
36.20	100	1
36.30	100	1
36.40	100	1
36.50	100	1
36.60	100	1
36.70	100	1
36.80	100	1
36.90	100	1
37	100	1
37.10	100	1
37.20	100	1
37.30	100	1
37.40	100	1
37.50	100	1
37.60	100	1
37.70	100	1
37.80	100	1
37.90	100	1
38	100	1
38.10	100	1
38.20	100	1
38.30	100	1
38.40	100	1
38.50	100	1
38.60	100	1
38.70	100	1
38.80	100	1
38.90	100	1
39	100	1
39.10	100	1
39.20	100	1
39.30	100	1
39.40	100	1
39.50	100	1
39.60	100	1
39.70	100	1
39.80	100	1
39.90	100	1
40	100	1

The curve is below one when a county expects only a few deaths, because nearly every published county-year is a zero, crosses one near 4.2 expected deaths, and peaks where only 5.3% of county-years are published and nearly all of them have a high count. It returns to within 5% of one beyond 14.2 expected deaths. The second view plots the share of county-years that WONDER publishes. This is a calculation from the rule, not an estimate from data.Note: A county-year with a given expected number of deaths is published when its Poisson count is zero or at least 10. The ratio is the average death count in published county-years divided by the average count in all county-years, which is the ratio of the pooled rate to the true rate when counties share a population.Note: The model assumes a constant rate in each county, independent years and Poisson counts. Real counts vary more than Poisson counts, so the curve gives direction and rough size, and it is not used to correct any estimate in this book.Source: Calculation from the publication rule of the CDC WONDER county files, 2018-2024, applied to Poisson counts; full counts of the model, no sampling error, equal weight for every county-year

The rate of a complete county is exact for that county, so complete counties escape the bias in their own rates. They do not escape it as a group. Under the same model the counties published in every year are a selected set whose pooled rate is the same multiple of the true rate as the rate pooled over published county-years, so the set overstates the typical county of its size.

Figure 7 compares the model with data. For each size of county it divides two pooled WONDER rates for 2019 to 2023 by the panel's population-weighted rate for the same years in all counties of that size with a stable panel rate: the rate pooled over published county-years, and the rate pooled over counties published every year. A county that WONDER never publishes adds nothing to the pooled WONDER rate and stays in the panel rate. The data follow the sign pattern of the model. In counties under 10,000 residents the rate pooled over published county-years is 0.18 (95% CI 0.00 to 0.42) times the panel's rate for suicide and 0.05 (95% CI 0.00 to 0.13) times for overdose. Among counties of 10,000 to 25,000 residents the ratio is 0.78 (95% CI 0.50 to 1.07) for suicide and 1.63 (95% CI 1.45 to 1.77) for overdose. Overdose is above one and suicide below it, although the suicide interval reaches one, which fits the model because a county of that size expects more overdose deaths than suicides.

Among counties of 25,000 to 50,000 residents the ratios are 1.51 (95% CI 1.47 to 1.55) and 1.63 (95% CI 1.58 to 1.68), and the overdose rate pooled over counties published every year is 2.09 (95% CI 1.92 to 2.28) times the panel's. Among counties of 50,000 to 100,000 residents the ratios are 1.09 (95% CI 1.07 to 1.10) and 1.18 (95% CI 1.16 to 1.20), with 1.30 (95% CI 1.26 to 1.36) for overdose pooled over counties published every year. Among counties of 100,000 to 250,000 residents the ratios are 1.00 (95% CI 1.00 to 1.01) and 1.07 (95% CI 1.06 to 1.08), with 1.09 (95% CI 1.07 to 1.10) for the complete counties. In counties of 250,000 or more residents, where every count is published, the overdose ratio is 1.05 (95% CI 1.05 to 1.06). That is the difference between the two definitions of overdose, since WONDER counts drug poisoning of any intent and the panel counts unintentional and undetermined-intent poisonings.

Figure 7

A suicide rate pooled over published county-years is 0.18 times the panel's rate in counties under 10,000 residents and 1.51 times in counties of 25,000 to 50,000 residents

Death rate pooled over the county-years CDC WONDER publishes, 2019 to 2023, divided by the CDC panel's population-weighted modeled rate for 2019 to 2023 in all counties of each size with a stable panel rate, suicide and drug overdose

Pooled over published county-years
Pooled over counties published every year

Show data

Data for A suicide rate pooled over published county-years is 0.18 times the panel's rate in counties under 10,000 residents and 1.51 times in counties of 25,000 to 50,000 residents
cause	conf_high	conf_low	estimate	n_units	series	size
Suicide	0.419	0	0.182	406	Pooled over published county-years	Under 10,000
Suicide	n/a	n/a	0	41	Pooled over counties published every year	Under 10,000
Suicide	1.07	0.496	0.781	406	Pooled over published county-years	10,000 to 24,999
Suicide	1.55	1.47	1.51	484	Pooled over published county-years	25,000 to 49,999
Suicide	1.76	1.39	1.56	12	Pooled over counties published every year	25,000 to 49,999
Suicide	1.10	1.07	1.08	379	Pooled over published county-years	50,000 to 99,999
Suicide	1.18	1.11	1.14	145	Pooled over counties published every year	50,000 to 99,999
Suicide	1.01	1.00	1.00	325	Pooled over published county-years	100,000 to 249,999
Suicide	1.02	1.01	1.01	304	Pooled over counties published every year	100,000 to 249,999
Suicide	1.00	1.00	1.00	273	Pooled over published county-years	250,000 or more
Suicide	1.00	1.00	1.00	273	Pooled over counties published every year	250,000 or more
Drug overdose	0.130	0	0.051	454	Pooled over published county-years	Under 10,000
Drug overdose	n/a	n/a	0	128	Pooled over counties published every year	Under 10,000
Drug overdose	1.77	1.45	1.63	673	Pooled over published county-years	10,000 to 24,999
Drug overdose	4.43	2.82	3.63	11	Pooled over counties published every year	10,000 to 24,999
Drug overdose	1.68	1.58	1.63	602	Pooled over published county-years	25,000 to 49,999
Drug overdose	2.29	1.92	2.09	66	Pooled over counties published every year	25,000 to 49,999
Drug overdose	1.20	1.16	1.18	387	Pooled over published county-years	50,000 to 99,999
Drug overdose	1.36	1.26	1.30	227	Pooled over counties published every year	50,000 to 99,999
Drug overdose	1.08	1.06	1.07	325	Pooled over published county-years	100,000 to 249,999
Drug overdose	1.10	1.07	1.09	306	Pooled over counties published every year	100,000 to 249,999
Drug overdose	1.05	1.05	1.05	273	Pooled over published county-years	250,000 or more
Drug overdose	1.05	1.05	1.05	273	Pooled over counties published every year	250,000 or more

A dot at one means the pooled rate matches the panel. The pooled suicide rate is below the panel's in counties under 10,000 residents, above it in counties of 25,000 to 50,000 residents, and near it from 100,000 residents. At 10,000 to 25,000 residents the suicide ratio is 0.78, with an interval that reaches one. For overdose, whose rate is higher than suicide in counties of that size, the ratio is above one at 10,000 to 25,000 residents (1.63), and for counties of 25,000 to 50,000 residents the rate pooled over counties published every year is 2.09 times the panel's. Each dot has a 95% interval from a bootstrap over counties. The buttons switch the kind of death.Note: The panel is a model, so these ratios compare WONDER's rule with a modeled rate and not with the true rate. The panel's rates for small counties borrow from neighboring counties. WONDER counts and panel rates both cover 2019 to 2023. In counties of 50,000 to 100,000 residents the ratios are 1.09 for suicide and 1.18 for overdose, and in counties of 100,000 to 250,000 residents 1.00 and 1.07.Note: Counties are all counties with a stable panel rate in every year 2019 to 2023, other than Connecticut's. A county that WONDER never publishes adds nothing to the pooled WONDER rate and stays in the panel rate. Restricting both sides to counties with a published county-year, the ratios for counties of 25,000 to 50,000 residents are 1.34 for suicide and 1.38 for overdose. A series is plotted for a size of county only when it has at least 10 counties; where every complete county has a rate of zero, the ratio is zero and has no interval.Note: Overdose in the panel is unintentional or undetermined-intent drug overdose and in WONDER drug poisoning of any intent, so the levels differ by definition. In counties of 250,000 or more residents, where every count is published, the overdose ratio is 1.05, which measures that difference.Source: CDC WONDER underlying cause of death, county files, 2019-2023, and CDC Mapping Injury, Overdose, and Violence county file, 2019-2023; crude rates, modeled estimates, population-weighted

The data match the sign pattern and not the sizes of the model, which assumes one rate for every county of a given expected count. The panel is itself a model, so these ratios compare WONDER's rule with a modeled rate and not with the true rate, and the panel's rates for counties that WONDER never publishes are modeled in every year. Restricting both sides to counties with at least one published county-year, the ratios among counties of 25,000 to 50,000 residents are 1.34 (95% CI 1.31 to 1.37) for suicide and 1.38 (95% CI 1.35 to 1.41) for overdose, lower than the ratios above but still over one.

WONDER names only large counties in its birth products

The rule above acts on cells. A second rule acts on whole counties in WONDER's birth products. Natality names a county only if it has 100,000 or more residents, and linked birth and infant death only if it has 250,000 or more, as the documentation of the two products states. WONDER pools every other county in a state into one row labeled Unidentified Counties, so smaller counties have no cell at all, and there is nothing to suppress or to recover. Natality lists 580 county identifiers, 578 of them with births, and those counties hold 79.6% of births in 2018 to 2023. It pools 2,564 counties, counting as pooled the identifiers with no births, which are described below. Linked birth and infant death lists 263 identifiers, 261 with births, which hold 64.0% of births in 2018 to 2022, and pools 2,881 counties. A county's size class in this chapter is not the same as being named: 606 counties have a mean population of 100,000 or more, against 578 named in natality, and 277 have 250,000 or more, against 261 named in linked birth and infant death.

Cell suppression then applies inside the named counties: a count of births or infant deaths above zero and below 10 is withheld, as Chapter 10.1, Gradients, slopes and variance shows for birthweight classes in natality. Chapter 1.1, Birthweight and preterm birth and Chapter 1.2, Infant mortality work with those files. The mortality county files differ. They name every county and rely on the cell rule alone, so a missing county there is a blank inside a named county and not a pooled row.

The line between named and pooled moves slightly. Some identifiers are listed with no births of their own in any year. In natality 2 identifiers are such counties (Cape May County NJ and Portsmouth city VA), and in linked birth and infant death 2 are (Hinds County MS and Mahoning County OH). WONDER shows their births as missing or zero in every year because it has moved them into the state's pooled row. A zero for these counties is not a count. The book's loaders set them to missing, which is why the balanced natality panel has 578 counties and not 580.

What the book does instead

A withheld cell is counted and left out of the numerator and the denominator. It is never replaced by a value, a midpoint or the remainder of a total, because a total minus the published cells pins a withheld cell exactly wherever only one cell is withheld. Every WONDER chapter reports how many cells were withheld and the largest share of events they could hold, as Chapter 1.1, Birthweight and preterm birth, Chapter 1.2, Infant mortality and Chapter 6.3, Overdose, alcohol and suicide do.

County death rates in those chapters use complete counties. For overdose, the 942 complete counties hold 83.8% of residents, and the alcohol-induced and suicide counts give 644 counties holding 77.9% and 745 counties holding 82.0%. A rate for a complete county is exact for all its years, and the set of complete counties is selected on the outcome, so a map or gradient of complete counties says so. Comparisons across all counties use a modeled source such as the CDC panel, which has no stable overdose rate in some year for 428 counties holding 2.3% of residents.

National series are summed from the state files, which have no withheld cell. All 357 state-years of each kind of death are published, and the number of withheld state-years is 0. County Health Rankings pools several years into one rate, which brings in more counties than a single county-year does. It still leaves a county without a rate when the window holds fewer than 10 deaths. That removes 716 counties for suicide, 1,244 for overdose, with 4.0% of residents, and 1,767 for homicide, with 9.7%. For premature death the floor is 20 deaths, and 55 counties have no rate. A map of WONDER counts needs at least 20 pooled deaths in a county to draw it.

The book does not model the withheld counts. One study treats them as censored data and estimates county rates with spatial Bayesian models, and it finds that ignoring the suppression biases inference.[5] That approach produces a value for a cell that WONDER withholds, and the modeled sources this book uses are published by CDC itself.

Figure 8

In every source the counties left without a value are a larger share of counties than of residents, from 55 counties without a premature death rate to 2,881 that WONDER linked birth and infant death does not name

Counties with no value in each source and measure, with the share of counties and the share of residents in them, for WONDER mortality, County Health Rankings, the CDC panel and the counties WONDER names in its birth products

Source	Measure	Years	Rule that leaves a county out	Counties left out	Share of counties (%)	Share of residents (%)
CDC WONDER mortality, one county-year	Drug overdose	2023	Count of 1 to 9 deaths withheld	1,426	45.4	8.6
CDC WONDER mortality, one county-year	Suicide	2023	Count of 1 to 9 deaths withheld	1,792	57.0	12.0
CDC WONDER mortality, counties published every year	Drug overdose	2018 to 2023	Any county-year withheld, or no population	2,200	70.0	17.1
CDC WONDER mortality, counties published every year	Suicide	2018 to 2023	Any county-year withheld, or no population	2,397	76.3	18.9
County Health Rankings, pooled window	Drug overdose	2019 to 2021	Fewer than 10 deaths in the window	1,244	39.6	4.0
County Health Rankings, pooled window	Suicide	2017 to 2021	Fewer than 10 deaths in the window	716	22.8	1.3
County Health Rankings, pooled window	Homicide	2015 to 2021	Fewer than 10 deaths in the window	1,767	56.2	9.7
County Health Rankings, pooled window	Firearm deaths	2017 to 2021	Fewer than 10 deaths in the window	864	27.5	2.0
County Health Rankings, pooled window	Premature death	2019 to 2021	Fewer than 20 deaths in the window	55	1.7	0.0
CDC injury and violence panel, modeled	Drug overdose	2019 to 2023	No stable rate in some year	428	13.6	2.3
CDC injury and violence panel, modeled	Suicide	2019 to 2023	No stable rate in some year	869	27.7	5.1
CDC WONDER natality, counties named	Births	2018 to 2024	Counties under 100,000 residents pooled	2,564	81.6	21.0
CDC WONDER linked birth and infant death, counties named	Infant deaths	2018 to 2022	Counties under 250,000 residents pooled	2,881	91.7	37.1

Show data

Data for In every source the counties left without a value are a larger share of counties than of residents, from 55 counties without a premature death rate to 2,881 that WONDER linked birth and infant death does not name
left_out	measure	rule	share_counties	share_residents	source	window
1426	Drug overdose	Count of 1 to 9 deaths withheld	45.40	8.60	CDC WONDER mortality, one county-year	2023
1792	Suicide	Count of 1 to 9 deaths withheld	57	12	CDC WONDER mortality, one county-year	2023
2200	Drug overdose	Any county-year withheld, or no population	70	17.10	CDC WONDER mortality, counties published every year	2018 to 2023
2397	Suicide	Any county-year withheld, or no population	76.30	18.90	CDC WONDER mortality, counties published every year	2018 to 2023
1244	Drug overdose	Fewer than 10 deaths in the window	39.60	4	County Health Rankings, pooled window	2019 to 2021
716	Suicide	Fewer than 10 deaths in the window	22.80	1.30	County Health Rankings, pooled window	2017 to 2021
1767	Homicide	Fewer than 10 deaths in the window	56.20	9.70	County Health Rankings, pooled window	2015 to 2021
864	Firearm deaths	Fewer than 10 deaths in the window	27.50	2	County Health Rankings, pooled window	2017 to 2021
55	Premature death	Fewer than 20 deaths in the window	1.70	0	County Health Rankings, pooled window	2019 to 2021
428	Drug overdose	No stable rate in some year	13.60	2.30	CDC injury and violence panel, modeled	2019 to 2023
869	Suicide	No stable rate in some year	27.70	5.10	CDC injury and violence panel, modeled	2019 to 2023
2564	Births	Counties under 100,000 residents pooled	81.60	21	CDC WONDER natality, counties named	2018 to 2024
2881	Infant deaths	Counties under 250,000 residents pooled	91.70	37.10	CDC WONDER linked birth and infant death, counties named	2018 to 2022

In each row the share of residents is smaller than the share of counties, because the counties left out are small. The first four rows are WONDER mortality, for a county-year in 2023 and for a county that must be published in every year from 2018 to 2023. The next five pool several years in County Health Rankings, and the next two use CDC's modeled panel. The last two rows follow a different rule, a population threshold for naming a county, at 100,000 residents in natality and 250,000 in linked birth and infant death.Note: Counties left out are those with no value. The share of counties uses each source's own count of counties (3,142 in WONDER, one more in County Health Rankings). Residents use WONDER's population for the WONDER mortality rows, County Health Rankings' for its rows and 2019-2023 American Community Survey residents for the rest.Note: County Health Rankings windows are those of its 2024 release for each measure and the floors are the smallest published counts. The CDC panel row counts counties without a stable rate in every year, including Connecticut's 8, which the panel does not carry. Naming thresholds are 100,000 and 250,000 residents.Source: CDC WONDER mortality, natality and linked birth and infant death, County Health Rankings 2024 and the CDC injury and violence panel, 2018-2024; full counts, crude rates and modeled estimates

What these data cannot show

The comparisons of complete counties with the rest use panel rates that are modeled for the counties WONDER withholds. They show the direction of the selection and cannot give the true rate of a withheld county. The two sources also define overdose differently: the panel counts unintentional and undetermined-intent drug poisonings, and WONDER counts drug poisoning of any intent, so levels differ a little by definition.

The calculation of the bias treats a county's deaths as Poisson counts with a constant rate and independent years. Real counts vary more than that, and rates move over time, so the curve gives direction and rough size and does not correct any estimate in this book. The sizes of county and their boundaries are choices, and the figures move with them.

Withheld counts cannot be recovered. A withheld cell is known only to lie above zero and below the limit, and the bound printed here is a ceiling set by the public limit and the published cells, and it says nothing about any one cell. The group comparisons describe counties and their residents, and County Health Rankings' shares of residents come from Census population estimates with that source's categories. Connecticut's counties are left out of every county comparison because WONDER has no population for them in 2022 and 2023. North Carolina's 2023 overdose counts predate the state's amended death certificates and are low, which enters the 2018 to 2023 window of the overdose counts. A pull that left out WONDER's option to show zero values would turn county-years with no deaths into absent rows, and the below-one ratios in the smallest counties depend on keeping them. The chapter covers three kinds of death. Kinds of death with more or fewer deaths would move the sizes of county at which the rule bites, which the Poisson calculation implies and these data do not test.

Data sources

CDC WONDER underlying cause of deathCDC WONDER underlying cause of death, county files for drug poisoning (overdose), alcohol-induced causes and suicide, 2018 to 2024, and state files for the same causesYears: 2018-2024 · Weights: none (full counts of deaths; crude rates); WONDER county population · License: Public domain (NCHS, CDC); WONDER data-use restrictions apply
CDC WONDER natalityCDC WONDER natality, county totals 2018 to 2024: births in the counties that WONDER names and in the pooled Unidentified Counties rowsYears: 2018-2024 · Weights: none (full counts of births) · License: Public domain (NCHS, CDC); WONDER data-use restrictions apply
CDC WONDER linked birth and infant deathCDC WONDER linked birth and infant death, county totals 2018 to 2022: births in the counties that WONDER names and in the pooled rowsYears: 2018-2022 · Weights: none (full counts of births) · License: Public domain (NCHS, CDC); WONDER data-use restrictions apply
CDC injury and violence mappingCDC Mapping Injury, Overdose, and Violence county file: modeled overdose and suicide death rates, annual 2019 to 2023Years: 2019-2024 for counties, 2022-2024 for tracts (data as of 2026-05-14) · Weights: mean WONDER population of each county · License: Public domain (CDC, U.S. Department of Health and Human Services)
County Health RankingsCounty Health Rankings 2024: counties with no rate for overdose, suicide, homicide, firearm deaths and premature death, and race and ethnicity shares of residents (Census Population Estimates)Years: 2024 release · Weights: county population in County Health Rankings · License: CC BY 4.0 (University of Wisconsin Population Health Institute)
Area Deprivation IndexNeighborhood Atlas Area Deprivation Index 2023, tract national rank averaged to countiesYears: 2023 release (version 4.0.1; built on ACS 2019-2023) · Weights: 2019-2023 tract population · License: Free for non-commercial use; cite the Neighborhood Atlas and Kind and Buckingham
Rural-Urban Continuum CodesUSDA Rural-Urban Continuum Codes 2023, countyYears: 2023 · Weights: none (classification) · License: Public domain (U.S. federal)
American Community SurveyTotal population of 2020 tracts, 2019-2023, summed to counties as population for deprivation and for the shares of residents in the tableYears: 2019-2023 (5-year) · Weights: none (full count) · License: Public domain (U.S. Census Bureau)

All data sources

Methods

This chapter examines which county-years CDC WONDER withholds in its death counts for drug poisoning of any intent, alcohol-induced causes and suicide, 2018 to 2024, across 3,142 counties. A withheld cell is left out of both the numerator and the denominator of a rate and nothing is imputed, and county groups are formed on whether WONDER publishes the count, which depends on the death count itself, so every comparison between groups is a comparison between selected sets of counties. The comparison rate is CDC's modeled county panel for 2019 to 2023, which is mostly modeled in the counties WONDER withholds, so a gap between a WONDER rate and the panel rate sets a direct rate against a model. Connecticut's counties are left out of the group comparisons, and rates are crude, not age-adjusted.

Which death counts and years are used

County-year counts come from the consolidated county files of CDC WONDER underlying cause of death for 2018 to 2024: drug poisoning of any intent, alcohol-induced causes and suicide, 3,142 counties by year. WONDER publishes a count of zero and counts of 10 or more and withholds counts above zero and below 10; the county population is published for every cell. The county statistics use 2018 to 2023, the window of the chapter on overdose, alcohol and suicide, and the by-year figure uses all seven years, which are final in the June 2026 pull. North Carolina's 2023 overdose counts in these files predate the state's amended death certificates and are low.

A default WONDER query omits county-years with no deaths, so zeros appear only because the pulls asked for them (Show Zero Values).

Figure explained here: Figure 1

How withheld cells and complete counties are handled

A withheld cell is counted and left out of both the numerator and the denominator of a rate. Nothing is imputed, derived from a total or bounded cell by cell; the one bound printed is the largest share of the county files' deaths that withheld cells could hold at 9 each, computed from the public limit and the published cells.

A complete county has a published count and a population in every year 2018 to 2023. The comparisons with the panel use 2019 to 2023 on both sides, the panel's years (its 2024 values are provisional), and call a county complete there when its count is published in each of those years. Connecticut's 8 counties have deaths but no WONDER population in 2022 to 2024 and are left out of the group comparisons and of every county rate.

County size, groups and characteristics

County size is the mean WONDER population over 2018 to 2023, in classes bounded at 10,000, 25,000, 50,000, 100,000 and 250,000 residents, and the results move with these boundaries.

County groups are formed on whether WONDER publishes the count, which depends on the death count itself, so every comparison between groups is a comparison between selected sets of counties.

County deprivation is the mean of tract Area Deprivation Index national ranks weighted by 2019-2023 tract population (Connecticut tracts assigned to the eight legacy counties), county type is the Rural-Urban Continuum Code (nonmetro from code 4), and race and ethnicity shares are County Health Rankings' figures from Census Population Estimates.

Figures explained here: Figure 2 and Figure 4

Counties named in the birth products

Counties named by WONDER natality and linked birth and infant death are the identifiers other than the pooled Unidentified Counties rows, and identifiers with no births of their own, which WONDER has moved into the pooled rows, count as pooled; they (2 in natality, Cape May NJ and Portsmouth VA, and 2 in linked birth and infant death, Hinds MS and Mahoning OH) are set to missing.

Figure explained here: Figure 8

The comparison rate and the bias calculation

The comparison rate is the mean of CDC's annual modeled county rates for 2019 to 2023 from the injury and violence panel, kept only when CDC marks the rate stable in every year, weighted by mean county population over the same years; the panel is modeled where counts are small, and it is mostly modeled in the counties WONDER withholds. The rate pooled over published county-years is deaths over person-years in the published cells only, and a county never published adds nothing to it while it stays in the panel rate.

Bias here means a difference from the rate in a systematic direction. The calculation of the bias of that rate treats a county's yearly deaths as Poisson counts with a constant rate and is a calculation from the rule, not an estimate.

The counts of counties without a value in other sources use County Health Rankings' numerators (a county has no rate when the window holds fewer than 10 deaths, or 20 for premature death) and the panel's stable-rate flag.

Figures explained here: Figure 5, Figure 6 and Figure 7

Uncertainty, maps and rate type

Means across counties are population-weighted with a 95% interval from a bootstrap that resamples whole counties (2,000 replicates, seed 12345); ratios and pooled rates use shared draws.

Maps of WONDER counts would need at least 20 pooled deaths; the map here counts withheld years and needs no floor. Rates are crude, not age-adjusted.

Figure explained here: Figure 3

References

Klein, Richard J.; Proctor, Suzanne E.; Boudreault, Manon A.; Turczyn, Kathleen M. (2002). Healthy People 2010 criteria for data suppression. Healthy People 2010 Statistical Notes, No. 24 (National Center for Health Statistics). www.cdc.gov/nchs/data/statnt/statnt24.pdf
Parker, Jennifer D.; Talih, Makram; Irimata, Katherine E.; Zhang, Guangyu; Branum, Amy M.; Davis, Danielle; Das, Barnali; Hamilton, Brady E.; Kochanek, Kenneth D.; McCarty, Frances; Strashny, Alexander (2023). National Center for Health Statistics data presentation standards for rates and counts. Vital and Health Statistics, Series 2, No. 200 (National Center for Health Statistics). doi:10.15620/cdc:124368
Gelman, Andrew; Price, Phillip N. (1999). All maps of parameter estimates are misleading. Statistics in Medicine 18(23): 3221-3234. doi:10.1002/(sici)1097-0258(19991215)18:23<3221::aid-sim312>3.0.co;2-m
Brillinger, David R. (1986). The natural variability of vital rates and associated statistics. Biometrics 42(4): 693-734. doi:10.2307/2530689
Quick, Harrison (2019). Estimating county-level mortality rates using highly censored data from CDC WONDER. Preventing Chronic Disease 16: 180441. doi:10.5888/pcd16.180441

How to cite

@misc{singh2026suppressedcountiesindeathdata,
  author = {Singh, Vishal},
  title = {Suppressed counties in death data},
  year = {2026},
  url = {https://vishalsingh.org/zipcode/suppressed-counties-in-death-data},
  note = {In ZIP Code Destiny. Published 9 October 2026.}
}