Show The Graduate Center Menu

Census 2020 Response Rate Analysis: Week 4

Mapping "Self-Response" for a Fair and Accurate 2020 Census (Week 4)

Prepared April 17, 2020

HTC 2020 map news

Links to analysis from prior weeks:

WEEK 4 (April 10 through April 16)

Summary materials:

National Response Rate Trends

The nationwide 2020 Census self-response rate as of Thursday, April 16, was 49.4%. In the past week, the U.S. rate increased by just over 2 points. By comparison, in the second week of response rate reporting (March 27 through April 2) the U.S. rate rose by an average of 1.5 points per day.

The trend is shown in the chart in Figure 1 below. The chart is annotated to show that as of April 8, the Census Bureau has been mailing paper questionnaires and reminder letters to non-responding households. Due to the COVID-19 crisis, the Census Bureau has extended the timeframe of these mailings: paper questionnaires (and related reminder letters) will continue to be mailed through April 30, and reminder postcards will be mailed from April 27 to May 9.

The chart visualizes the day-to-day response rates in 2010 (the long red line) compared with the rates so far in 2020 (the shorter red line noted with the arrow). The other lines represent state-by-state response rates in 2020. (The Census Bureau did not publish daily response rates from 2010 at the state level.)


NB: Keep in mind the differences in census operations between 2010 and now when comparing response rate trends. Census operations were different in 2010, which may impact the difference in the national trend. For example, the 2010 Census was largely paper-based (the Bureau didn’t advertise the online option, which was used minimally); there were only two initial mailings (rather than staggered waves of initial mailings and multiple follow up mailings in 2020); there was a targeted replacement mailing in 2010; and of course, the 2010 Census did not face a pandemic that delayed key census operations and upended Get Out the Count (GOTC) campaigns that were based significantly on in-person outreach. Nonetheless, stakeholders will measure the success of the 2020 self-response operation by whether the nation, states, and localities meet and surpass their 2010 rates. Our map allows you to compare not only the 2020 rates with final 2010 “goalpost” rates, but also assess how your area is doing compared with nearby communities or similar areas across the state, and for larger regions such as counties, the state, and nation as a whole.

The following graphs show the latest 2020 response rates by state and the percentage point change by state from March 20 (when the Census Bureau first reported response rates) until now.

FIGURE 2 States ranked by April 16 census response rate (the final 2010 U.S. rate and the current U.S. 2020 rate are included for comparison). Click to enlarge.

FIGURE 3 States ordered left-to-right by lowest response rate to highest on April 16 (including the current national rate and final 2010 response rate for comparison). Brown bars represent the beginning response rate (March 20); the blue bars represent the increase between 3/20 and 4/16. Click to enlarge.

The pace of change in the response rate has been uneven on a state-by-state basis. The U.S. increased in Week 4 on average by 0.33 percentage points per day (Florida and Illinois also had the same daily increase). West Virginia had the fastest increase, of 0.5 points per day. The three states following West Virginia were Iowa, California, and Nebraska, tied at 0.44 points per day. The slowest increase was in Wisconsin, whose response rate rose on average by only 0.21 points per day during Week 4.

You can visualize the pace of change for any geographic location on our HTC 2020/Response Rate map. The following examples highlight a new feature added to the map: a trendline chart plotting day-to-day response rates for each of the 84,000 census tracts, roughly 34,000 cities and towns, 3,100 counties, 50 states and the District of Columbia, and the hundreds of tribal areas and congressional districts on the map (described further here).

In West Virginia, for example, Figure 4 below shows how the statewide response rate has increased from 9.8% on March 20 to 37.8% on April 16. The trendline is color-coded to match the legend on the right side of the map. The green line just below the color-coded trendline represents the rate based on internet responses. In West Virginia, where almost a third of the state’s housing units received the “Internet Choice” mailing — which included a paper questionnaire, the internet-only response rate of 28.7% on April 16 is almost 10 points lower than the overall statewide rate.

Figure 4 also highlights the one-day boost in response rates that likely reflects “Census Day” outreach activities on April 1.


In Figure 5 below, the map is zoomed in on West Virginia, and McDowell County is highlighted, along the Virginia border. In McDowell County, almost all (99.6%) of housing units are part of the now-suspended Update/Leave operation, and the county’s overall self-response rate is only 3.7%. Most (if not all) homes in the county haven’t yet received their initial census packets.


Figure 6 shows another county in West Virginia with a different response rate pattern: Wood County along the Ohio border. In Wood County, all housing units received a mailed invitation from the Census Bureau. The response rate for Wood County is much higher than McDowell County: 52%.


In Figure 7, the map is zoomed in and highlights Tract 108 in Wood County. In this tract, an estimated 9 homes out of more than 1,400 were to have received a hand-delivered census packet; the rest received an “Internet Choice” mailing. Consistent with most homes receiving a paper questionnaire in the mailing, most of the responses have been by mail or phone and fewer have been online (the internet-only response rate is just 22%).

Nonetheless, the tract’s 58% response rate is relatively high. Below in this analysis we show that nationwide, tracts that received Internet Choice mailings tend to have response rates lower than both the U.S. response rate and for tracts receiving Internet First mailings. This tract is an exception.


Key areas of concern analyzed for Week 4

In addition to the nationwide response rate trends, we focus our Week 4 analysis on the following issues:

  1. New extended census timeframe (and a new approach to assessing self-response/nonresponse follow-up implications for historically undercounted populations)
  2. Update/Leave and internet access
  3. Internet First compared with Internet Choice census tracts


The Census Bureau’s original plan for 2020 was to publish response rates seven days per week from March 20 through May 31. The 2020 Census Nonresponse Follow-up (NRFU) operation was to have started in mid-May. Householders could still self-respond through July 31, so the Bureau planned to continue publishing response rates until then, but only on weekdays starting June 1.

Due to the COVID-19 crisis, after modest initial changes to the census schedule, the Bureau on April 13 announced significant adjustments to many of its operations. Now, householders will be able to self-respond through October 31, 2020 – an additional three months of self-response opportunity. NRFU will begin on August 11 (three months after it was originally scheduled to being) and will continue through the end of October.

In prior censuses, self-response rates typically increase only marginally once the formal self-response operation ends and NRFU begins. In 2010, for example, the U.S. response rate on April 30, 2010 (the end of the self-response operation) was 65.3%. Some households still responded on their own after NRFU began, but the nationwide response rate increased only slightly more than a percentage point during this time, peaking at 66.5% on June 22 (this represents an average daily increase from May 1 to June 22 of only .02 percentage points per day). See the daily 2010 response rates in Appendix H in this 2012 Census Bureau memo. [PDF]

Nonetheless, we have no historic context for how self-response rates will change over the course of such a longer timeframe.

Who will be in the NRFU universe by August, and how will that have changed during the extended 2020 period for self-response?

When our team at the CUNY Graduate Center was asked to develop an online map leading up to the 2020 Census to help stakeholders focus their education and outreach on the hardest to count communities and populations, the main question was how to define “hard-to-count.” In consultation with several census experts, we used an approach that identified the census tracts nationwide with the lowest shares of households that filled out the census questionnaire on their own during the last decennial census. These tracts had the most homes that needed to be counted in-person through NRFU. The NRFU door-knocking operation is challenging, expensive, and adds substantially to the risk that people will be missed or counted inaccurately. In other words, areas with the greatest share of homes that needed to be counted in-person were considered the hardest to count.

In order to identify tracts with the lowest shares of self-responding households, we used a threshold of the tracts in the bottom 20% of self-response rates nationwide. These tracts were then color-coded on the HTC 2020 map in orange-to-dark red as the “hardest to count” tracts.

We propose to use a similar approach during the 2020 Census self-response operation, to identify the census tracts that will have the greatest share of homes to be counted through NRFU starting in August 2020. But we will enhance this approach in two ways:

  • We can determine the “bottom 20%” of tracts each day during the self-response operation based on the latest 2020 Census response rates. This enables us to understand not only where the tracts are located that will eventually require the greatest share of NRFU effort. But we can identify if the geographic concentrations of these tracts are changing, or if they remain concentrated in certain parts of the country.
  • We can use the latest data from the Census Bureau’s American Community Survey (ACS) (2014-2018 “5-year estimates”) to determine the characteristics of the overall population in these “bottom 20%” census tracts. This will show not only if historically undercounted population groups are in these tracts, but if these groups are becoming more concentrated in the tracts with lowest self-response rates between now and when NRFU begins.

Of course, this approach will not identify who is being counted as the self-response operation continues. It only identifies the demographic characteristics of census tracts overall. But it provides an indication of which population groups (and which communities) are at risk of being undercounted, and whether that risk is increasing or decreasing for historically undercounted groups if they continue to occupy a large share of the population of the bottom 20% of tracts.

Initial results

In order to determine the threshold U.S. response rate for the tracts in the bottom 20% of response rates, we use the following assumptions:

  • We are omitting tracts that include homes covered by the Update/Leave (U/L) operation. Response rates so far for U/L areas are low because most homes in those areas have not yet received census packets.
  • We are also omitting tracts with very few people (any tract with fewer than 100 people as estimated by the American Community Survey is excluded from our analysis).

This provides a universe of approximately 65,000 tracts with more than 100 people each and with no U/L units. Twenty percent of this universe is approximately 13,000 tracts.

On March 20, when the Census Bureau began reporting response rates, tracts with response rates in the lowest fifth had a response rate of less than 8.6%. Based on ACS estimates for the 2014-2018 period, there are approximately 51.5 million people in these tracts, approximately 20.3 million total housing units, and approximately 18.1 million occupied housing units (i.e., “households”).

On April 16, tracts with response rates in the lowest fifth had rates less than 41.4%. This also represents approximately 13,000 tracts, but they are different tracts than on March 20. Based on ACS estimates for the 2014-2018 period, the universe of population and housing units in these tracts had changed:

  • overall population decreased about 14% to 44 million people;
  • the number of housing units decreased about 7% to 18.9 million; and
  • the number of occupied housing units (households) decreased about 13% to 15.7 million.

However, certain population groups increased their concentration in the bottom 20% of census tracts between March 20 and April 16. The following examples illustrate this pattern. A detailed table is provided at the end of this section.

Population in poverty

Based on the March 20 threshold response rate of 8.6%, there were approximately 50 million people in those 13,000 tracts for whom the poverty rate had been determined. Of these, just over 7.5 million — approximately 15% of 50 million — had incomes below the poverty line.

By April 16, the population in the “bottom 20% of tracts” (the 13,000 tracts with a response rate of less than 41.4%) for whom poverty had been determined had decreased by 16% to about 42 million people. But people with incomes below the poverty line in those tracts had increased to almost 11 million (a 45% increase). The share of the population in poverty in the lowest response tracts grew by 11 percentage points, from almost 15% to almost 26%. In other words, a greater share of the population in the bottom 20% of tracts between March 20 and April 16 was living in poverty.

Households with “limited English proficiency”

Based on the March 20 threshold response rate of 8.6%, there were approximately 1.2 million households — about 6.7% of households — considered by the Census Bureau to have “limited English proficiency” (LEP) in the bottom 20%.

Although the estimated number of households in the bottom 20% of tracts decreased from March 20 to April 16 by about 13% (from approximately 18.1 million to 15.7 million households), the estimated number of LEP households increased by almost 500,000 during that time to 1.7 million households in the bottom 20% of tracts — an increase of more than 41%.

As a result, language challenges in the bottom fifth of tracts with the lowest response rates became greater. The share of LEP households in this group of tracts grew by 4 percentage points between March 20 and April 16.

People of color

We focus our initial analysis on two groups of people of color: those who reported their race during the 2014-2018 American Community Survey as Black but not of Hispanic origin, and separately for the population that reported Hispanic origin regardless of race.

By comparison, we also examine the population who reported their race during the 2014-2018 ACS as White but not of Hispanic origin.

Based on the March 20 threshold response rate of 8.6%, there were approximately 8 million people in the bottom 20% of tracts who were non-Hispanic Black. By April 16, the estimated non-Hispanic Black population in the bottom 20% of tracts increased by almost 33% to 10.5 million. Because the overall population of the bottom 20% of tracts declined between March 20 and April 16, the non-Hispanic Black population share increased, growing by just over 8 points (from just over 15% to almost 24%).

Based on the March 20 threshold response rate of 8.6%, there were approximately 12 million people in bottom 20% of tracts of Hispanic origin. By April 16, the estimated Hispanic population in the bottom 20% of tracts increased by almost 24% to approximately 15 million. Because the overall population in the bottom 20% of tracts declined between March 20 and April 16, the Hispanic population grew in concentration, increasing by more than 10 points (from just over 23% to almost 34%).

By comparison, the estimated non-Hispanic White population in the bottom 20% of tracts decreased overall and declined as a share of the population in the low-response-rate tracts.

Based on the March 20 threshold self-response rate, there were almost 27 million non-Hispanic Whites in the bottom 20% of tracts. By April 16, the White population in these tracts fell by about 12 million people (a 45% reduction). The concentration of non-Hispanic Whites as a share of population in the bottom 20% of tracts therefore decreased by more than 18 points (from just over 52% to almost 34%).

Table 1 below presents the data described above. It also includes statistics for other population groups historically at risk of being undercounted, such as renters, people in crowded housing, and single parent households. It also presents data for people who may be at risk of being undercounted in 2020 due to greater automation of the census process, such as households with no home internet subscriptions and householders age 65 or older. Click to enlarge.


Based on feedback from census stakeholders on the value of analyzing population characteristics for tracts with the lowest self-response rates, we can build on this analysis in future weeks, such as mapping the bottom 20% of tracts to visualize if the geographic patterns are changing day-by-day or week-by-week. We can also examine other population groups, if requested.


Analysis: are there differences across states in tract-level home internet access for areas with no units in Update/Leave, compared with areas that have at least some U/L coverage?

We found that in 44 states, tracts with at least some U/L areas had a greater share of homes without internet subscriptions, compared to tracts with no U/L areas.

Virginia had the greatest difference in home internet access:

  • On average, 14% of households have no internet subscription in tracts with no U/L areas.
  • But that more than doubles for tracts covered by U/L, increasing to 31% of homes with no internet, on average.

Analysis: has the gap in response rates increased between areas that received Internet First mailings compared with Internet Choice mailings?

When we first analyzed the 2020 Census response rates, we examined tract-level rates as of March 26. On that date, in tracts where households received the Internet First mailing, the average response rate was 29.2%. This was 1 percentage point above the U.S. rate on March 26. But in tracts that received the Internet Choice mailing, the average response rate was 24.6%4.5 percentage points below the U.S. rate. Therefore, the gap between response rates for Internet First tracts compared with Internet Choice was less than 5 percentage points.

As of April 16, the average response rate in tracts where households received Internet First mailing was 53.5% — 4 percentage points above the U.S. rate In Internet Choice tracts as of April 16, the average response rate was 44.4%.

Notably, the gap had almost doubled between March 26 and April 16. Not only was the average response rate in Internet Choice tracts 5 points below the U.S. rate on April 16, but the difference between Internet First and Internet Choice tracts is now more than 9 percentage points.

For more information, contact:

Center for Urban Research

at the Graduate Center, City University of New York

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.