Show The Graduate Center Menu
 
 

Census Self-Response Rates Mapped and Analyzed: 2000, 2010, and (soon) 2020

March 16, 2020

PDF version

Overview

The Center for Urban Research (CUR) at the CUNY Graduate Center has updated its Census 2020 Hard to Count (HTC 2020) map with self-response trends from the past two decennial censuses, to provide historical context as stakeholders fine-tune their Get Out the Count (GOTC) plans and participate in the U.S. Census Bureau’s Response Rate Challenge.

The analysis below provides background for this information and the importance of self-responding to the decennial census, especially as the count takes place during the coronavirus pandemic.

Self-responding is the best and least-expensive way for the Census Bureau to collect information. But self-responding has taken on heightened prominence during the coronavirus outbreak. Self-responding means you will avoid a knock at your door from a census enumerator. And because of the options the Census Bureau has provided for self-responding — online, by phone, or by mail — every household should strive to respond to the 2020 Census on their own.

Comparing earlier self-response rates as an inspiration to improve in 2020

The Center has analyzed self-response rate trends to show which states improved from 2000 to 2010 and therefore can continue their upward trend, and which states had lower self-response rates in 2010 and therefore can reverse the trend in 2020. The section below titled State Trends from 2000 to 2010 presents the findings of this analysis.

When the Census Bureau begins publishing 2020 Census self-response rates on March 20, CUR will integrate these rates each day into its map to supplement the information already accessible, such as updated local population characteristics, areas with poor internet access, public library locations, and more.

"By examining whether response rates for states, localities, and neighborhoods improved in 2010 over 2000, we can help civic leaders target outreach activities for the 2020 Census and create momentum towards higher levels of participation in areas that might have lagged in the past," Steven Romalewski, director of the CUNY Graduate Center's Mapping Service, said. "A fair and accurate 2020 census depends on it."

Historical self-response rates provide context for 2020, with qualifiers

Romalewski urged stakeholders to take into account any demographic shifts within a given area when comparing upcoming 2020 Census self-response rates with earlier rates, especially at the census tract level, where changing population and household characteristics could affect likelihood of participation by different response methods during the self-response operation.

Also, there have been important differences in census operations for the 2000, 2010, and 2020 counts. For example, in 2000, most households received a “short form” (comparable to the questionnaire every household will receive in 2020), and a smaller number of households received a “long form” with additional demographic and socio-economic questions. (The ongoing American Community Survey — delivered to households separately from the decennial questionnaire — replaced the long form starting in 2005).

In 2010, there was no online response option, and the Census Bureau mailed all census packets at the same time. In 2020, the Census Bureau will mail census packets in four “waves” from March 12–20, and households will have several options to respond: online, by phone, or using a paper questionnaire. In 2010, households received only two mailings; in 2020, they will receive up to five mailings.

Some of these differences may have helped boost response rates in prior censuses (such as “replacement” mailings in 2010); other factors may have dampened response rates (such as the long form in 2000). In 2020, the online response option may help increase rates in some areas, but also introduce challenges to self-responding in communities without reliable internet access.

And of course, neither the 2000 nor the 2010 censuses took place during a pandemic.

Romalewski emphasized that despite these operational and demographic differences across decades, it can help to see the historical rates for your community. “Self-response rates from 2000 and 2010 measure how well the households in a given area responded during those censuses and in light of the environment at the time,” he emphasized. “Now, current residents of those areas – regardless of how your community may have changed or how each census was conducted – will see how their community did in 2010 (and 2000) and can use those data points as an inspiration to do better in 2020.”

Low self-response correlates with poorer census quality

In the section below titled Why Self-Response Rates Are Important, CUR’s analysis shows that census accuracy tends to decline when self-response rates are low and to improve when rates are high.

"Low self-response rates are correlated with people being missed in the census. If states and localities want to help ensure a fair and accurate census, they need to do all they can to boost self-response in 2020,” explained Dr. William O’Hare. Dr. O’Hare is the author of Differential Undercounts in the U.S. Census: Who is Missed? and a nationally recognized expert on the undercount of young children in the census.

Weekly analysis of the 2020 rates

The Center will also work closely with other census experts to analyze the self-response rate trends on a weekly basis, to provide context and interpretation for census stakeholders who can use this information to re-prioritize their Get Out the Count strategies as needed, and for philanthropic funders, journalists, elected officials, and others who will be following the progress of the 2020 Census.

State Trends from 2000 to 2010

The U.S. Census Bureau’s Response Rate Map displays self-response rates from the 2010 Census as an incentive for communities across the country to surpass those rates in 2020. But to provide context for the 2010 rates and show which areas improved or not from 2000 to 2010, the Center for Urban Research has mapped and analyzed the 2000 and 2010 self-response rates by state, county, and census tract. On a statewide level, the discussion below highlights which states improved or fell below their 2000 self-response rates in 2010, providing a data-driven incentive to do even better in 2020.

Nationwide, the final mail response rate in 2010 was almost 1 percentage point less than in 2000, decreasing from 67.4% in 2000 to 66.5% in 2010.

2000 mail response rate from “Census 2000 Mail Response Rates: FINAL REPORT” (Census 2000 Evaluation A.7.a; January 30, 2003). 2010 mail response rate from “Mail Response/Return Rates Assessment: FINAL REPORT” (2010 Census Assessments; May 30, 2012). 

Note also that in 2000, most households received a short form questionnaire and some households received a long form, a practice which ended in 2005 when the long form was replaced by the ongoing American Community Survey (which is sent to a much smaller share of households). Households that received the short form in 2000 – comparable in length to the 2010 Census form – had a mail response rate of 68.7%.  Therefore, the decrease from the 2000 self-response rate for short form households only to all households in 2010 was a larger drop (just over two percentage points).  The Census Bureau’s statewide data on 2000 self-responses is not limited to short form-only responses, however, so our analysis uses the overall statewide 2000 self-response rates.
Statewide mail response rates from the 2000 Census are available on the Census Bureau’s website at https://www.census.gov/dmd/www/response/2000response.html.  Statewide mail response rates from the 2010 Census are available from https://www2.census.gov/dssd/.

But changes in mail response rates between 2000 and 2010 were uneven across the country. At the state level, 17 states plus the District of Columbia improved over the nationwide trend and increased their self-response rates between 2000 and 2010. South Carolina and Washington, DC led the increase, increasing their self-response rates by 6 percentage points or more:

  • South Carolina increased by 6.7 points, from 58% in 2000 to 64.7% in 2010; and
  • Washington, DC increased by 6 points, from 60% in 2000 to 66% in 2010.

Although the other 16 states improved by smaller amounts, several of these states exceeded the national 2010 self-response rate, such as Illinois (70.5%), Pennsylvania (70.2%), and Indiana (69.6%). One state – Florida – matched its 2000 rate in 2010 of 63%.

The remaining 32 states had lower self-response rates in 2010 compared to 2000, and most of these rates fell by a greater amount than the national decrease from 2000 to 2010 of almost one percentage point. For example, South Dakota's 2010 rate of 67.1% was almost 7 points less than its 2000 rate of 74%. West Virginia’s rate fell by almost 5 points, from 64% in 2000 to 59.1% in 2010. (See Table 1, below.)

TABLE 1. Statewide Census Mail Response Rates in 2000 and 2010, sorted by largest to smallest change from 2000 to 2010

State 2000 Mail Response Rate* 2010 Mail Response Rate Change (in percentage points)
South Carolina 58% 64.7% 6.7
District of Columbia 60% 66.0% 6.0
Hawaii 60% 64.1% 4.1
Tennessee 65% 67.1% 2.1
New York 63% 64.6% 1.6
Alabama 61% 62.5% 1.5
Illinois 69% 70.5% 1.5
Washington 66% 67.2% 1.2
Delaware 63% 64.1% 1.1
Louisiana 60% 61.0% 1.0
North Carolina 64% 64.8% 0.8
Indiana 69% 69.6% 0.6
Utah 68% 68.6% 0.6
Maryland 69% 69.5% 0.5
Texas 64% 64.4% 0.4
Vermont 60% 60.3% 0.3
Pennsylvania 70% 70.2% 0.2
Idaho 67% 67.1% 0.1
Florida 63% 63.0% 0.0
Massachusetts 69% 68.8% -0.2
Kentucky 66% 65.7% -0.3
Alaska 56% 55.6% -0.4
New Jersey 68% 67.6% -0.4
Connecticut 70% 69.5% -0.5
Minnesota 75% 74.1% -0.9
Kansas 71% 70.0% -1.0
Oregon 68% 66.9% -1.1
Rhode Island 67% 65.7% -1.3
Missouri 69% 67.5% -1.5
Wisconsin 75% 73.5% -1.5
Oklahoma 64% 62.3% -1.7
Arizona 63% 61.3% -1.7
Mississippi 63% 61.3% -1.7
Arkansas 64% 62.3% -1.7
California 70% 68.2% -1.8
New Mexico 62% 60.0% -2.0
Georgia 65% 62.5% -2.5
Wyoming 66% 63.4% -2.6
New Hampshire 67% 64.4% -2.6
Colorado 70% 67.2% -2.8
Iowa 76% 73.0% -3.0
Ohio 72% 69.0% -3.0
Virginia 72% 69.0% -3.0
North Dakota 72% 68.8% -3.2
Michigan 71% 67.7% -3.3
Montana 68% 64.6% -3.4
Maine 61% 57.4% -3.6
Nebraska 75% 71.1% -3.9
Nevada 66% 61.4% -4.6
West Virginia 64% 59.1% -4.9
South Dakota 74% 67.1% -6.9

*NOTE: Statewide mail response rates from the 2000 Census are published by the Census Bureau only as integers with no decimal places.

Why Self-Response Rates Are Important

When households fill out the census questionnaire on their own, they help the Census Bureau save time and money and collect the highest quality population data.

Equally important, when a household self-responds, it avoids a visit from a census enumerator who will try to collect data from the household in person. The effort to count households that do not self-respond is expensive and challenging.

If an area has a low self-response rate, more of its households will need to be counted in-person, increasing the risk that people will be missed or counted inaccurately. In 2010, states with low self-response rates (even if they had improved over 2000) tended to have the greatest percentages of people missed (i.e., omissions) in the 2010 Census.

Census “omissions”

Despite the Census Bureau’s best efforts and those of local “trusted partner” organizations helping to “get out the count,” some people are missed during the census. The Bureau estimated that it missed about 16 million people nationwide in the 2010 Census. (see U.S. Census Bureau, “DSSD 2010 CENSUS COVERAGE MEASUREMENT MEMORANDUM SERIES #2010-G-01” May 22, 2012.)

On a nationwide basis, the number of people missed is offset by people who are included in the census by mistake (such as students away at college whose parents mistakenly include the students on their own census forms, or people with more than one home who fill out the census at both locations), and by counts of people added to the census using a statistical method called “imputation” – determining the count and population characteristics of similar households nearby and applying them to a household where people were living but could not be reached directly.

The Census Bureau estimated that in 2010, the number of people missed (omissions) was equivalent to the number of people counted twice, otherwise counted erroneously (such people who died before Census Day or were counted in the wrong place), or counted through imputation. This resulted in a “net overcount” in 2010 that was statistically equal to zero – in other words, an accurate nationwide count.

Nationwide net undercount/overcount masks concerns about local census quality

But Terri Ann Lowenthal, a former staff director of the Congressional census oversight subcommittee and consultant to FCCP’s Funders Census Initiative, cautioned that political representation and government resources are not allocated based on national numbers. “At the local level, the number of people missed might not be offset by the number counted twice or mistakenly, in part because the people more likely to be missed often do not live in the same communities as those who are more likely to be counted twice. This can result in an undercount locally or across population subgroups that results in unfair political representation and skewed distribution of resources.”

The problem of omissions in the decennial census has been documented by experts such as Dr. William O’Hare in several publications and the staff of the New York City Department of City Planning’s Population Division. (See Differential Undercounts in the U.S. Census: Who is Missed? as cited above, and Understanding Who Was Missed in the 2010 Census prepared for the Population Reference Bureau August 13, 2019.)

A recent report by Dr. O’Hare (see p. 17) concludes by noting that:

Obtaining an accurate count of state and local populations is important because the data affect the balance of political power across geographic areas and are widely used for state and local decision-making. A comprehensive picture of census accuracy requires assessment of the number, characteristics, and geographic locations of those who are missed in the census in addition to analysis of net undercount rates. By understanding who was omitted in the 2010 Census, community leaders, advocates, and others can better target geographic areas and population subgroups for Get-Out-the-Count efforts to reduce the number of people who are missed in the 2020 Census and improve the accuracy of the count for state and local populations.

A recent paper by NYC Department of City Planning staff discusses the problems with census omissions and describes census self-responses as “the gold standard in decennial census data collection and … the most accurate and eļ¬ƒcient source of data.” Where self-response is low, these areas are at risk of greater omissions, which lead to undercounts and poorer census data quality.

Omissions are correlated with census self-response

Table 2 below shows the relationship by state between 2010 self-response rates and 2010 omission percentages calculated by the Census Bureau.

TABLE 2. Statewide 2010 Response Rates and Omissions Rates, sorted by largest to smallest omission rate

State 2010 Mail Response Rate 2010 Omission Percentage
District of Columbia 66.0% 9.0%
Mississippi 61.3% 8.9%
Hawaii 64.1% 7.8%
Alabama 62.5% 7.7%
New Mexico 60.0% 7.7%
West Virginia 59.1% 7.7%
North Carolina 64.8% 7.6%
Florida 63.0% 7.5%
Arizona 61.3% 7.3%
Georgia 62.5% 7.3%
Nevada 61.4% 6.9%
Texas 64.4% 6.9%
Louisiana 61.0% 6.8%
Oklahoma 62.3% 6.4%
Wyoming 63.4% 6.4%
Delaware 64.1% 6.2%
Montana 64.6% 6.1%
New York 64.6% 6.1%
Maryland 69.5% 6.0%
Colorado 67.2% 5.9%
Idaho 67.1% 5.8%
Tennessee 67.1% 5.8%
Virginia 69.0% 5.8%
Massachusetts 68.8% 5.7%
Alaska 55.6% 5.5%
Kentucky 65.7% 5.5%
Arkansas 62.3% 5.4%
Vermont 60.3% 5.4%
South Carolina 64.7% 5.2%
California 68.2% 5.1%
New Hampshire 64.4% 5.0%
Rhode Island 65.7% 4.9%
South Dakota 67.1% 4.9%
Utah 68.6% 4.9%
Illinois 70.5% 4.6%
Michigan 67.7% 4.5%
Missouri 67.5% 4.5%
New Jersey 67.6% 4.5%
Pennsylvania 70.2% 4.5%
Washington 67.2% 4.5%
Minnesota 74.1% 4.4%
Maine 57.4% 4.2%
Wisconsin 73.5% 4.1%
Oregon 66.9% 4.0%
Connecticut 69.5% 3.9%
North Dakota 68.8% 3.9%
Kansas 70.0% 3.7%
Indiana 69.6% 3.6%
Ohio 69.0% 3.5%
Nebraska 71.1% 3.1%
Iowa 73.0% 2.6%

See Table 14 for the omission percentages from the U.S. Census Bureau “DSSD 2010 CENSUS COVERAGE MEASUREMENT MEMORANDUM SERIES #2010-G-01” (May 22, 2012).

A statistical analysis of these two metrics shows they have a strong, inverse relationship; the Pearson value (a commonly-used correlation statistic) for the data in Table 2 is -.611. The negative value means the two measures are inversely correlated — that is, as self-response rates decrease, omission rates tend to increase, and vice versa. The range of correlation values is between -1 (the strongest negative relationship) and 1 (the strongest positive relationship). Therefore, the -.611 value is relatively strong.

Therefore, efforts by state and local officials and census stakeholders across the country to boost self-response rates by urging householders to fill out the census questionnaire on their own — either online, by phone, or by mail — will help protect localities against the risk of being undercounted.


For more information, contact:

Center for Urban Research

at the Graduate Center, City University of New York

cunymapping@gc.cuny.edu

All work and materials are supported by a grant from the 2020 Census Project

and developed in partnership with the Leadership Conference on Civil and Human Rights.