Each decade we are asked, “Why don’t you just use the information the government already has about me for the census? Why ask me again?”

In some ways, we do. We regularly work with information from other government agencies to make our statistics more accurate. For example, we have used information from federal, state, and local government agencies for decades to improve our census address list and to create population estimates.

For the 2020 Census, we accepted the challenge from the public and Congress to use existing records even more to streamline census operations, reduce the burden on people who respond, and save taxpayer money.

These existing records are often called “administrative records” because they are created as an agency “administers” or does its work. For example, the Internal Revenue Service (IRS) has information about who lives at an address because people share that information on their tax returns.

A few examples of how we used administrative records for the 2020 Census include:

• To improve our address list.
• To validate census takers’ work as part of our quality checks.
• To confirm vacant or nonexistent addresses.

Most notably though, for the first time, we used administrative records to count people who otherwise hadn’t responded.

We went to great lengths to get a response directly from households. When we didn’t receive one, administrative records enabled us to count people with information they had already provided to the government.

We went to great lengths to encourage people to respond online, by phone, or by mail. If a household didn’t respond, a census taker visited to try to collect their information in person during our Nonresponse Followup (NRFU) operation.

If a household didn’t respond after one census taker visit, we checked to see if other high-quality records could provide a count of the people living at the address along with their demographic characteristics. Or if a household responded on its own but didn’t answer all the questions, we checked to see if administrative records could provide the missing information.

For example, we used records from:

• Other census and survey data maintained within the U.S. Census Bureau, including responses from the 2010 Census and American Community Survey.
• The IRS.
• The Medicare enrollment database.
• The Indian Health Service.
• The U.S. Postal Service.
• The Social Security Administration.

We used these existing data sources only if we were confident that the data accurately reflected the number and characteristics of the people living in the household around April 1, 2020. In most cases, this means that we had multiple sources for a household that corroborated the information.

Otherwise, we continued to visit the household and, if necessary, tried to get information about the address from a neighbor.

As our NRFU operation neared completion in an area, some addresses were still missing responses, despite several possible attempts to obtain one from the household or a neighbor. It was at this point that we again looked to administrative records to fill in the missing information.

As I described above, we required the highest levels of confidence in the administrative records we used after the first visit because we still had plenty of opportunity to contact the household for an interview. Toward the end of data collection, the records must still be reliable.

From our research, the administrative records available at this point would likely be more reliable or complete than just a population count from a neighbor. For example, past census responses or administrative records might also give us race, Hispanic origin, age, sex, and other characteristics.

If we had reliable information from administrative records for a household at this point in the operation, we used the records to enumerate the household and closed the case. This enabled census takers to focus on a last push to get information from every remaining household.

After data collection ended in October, we reported preliminary rates on our use of administrative records. Preliminary estimates indicate that we used administrative records to enumerate:

• Approximately 5.6% of addresses nationwide.
• About 13.9% of the total workload for our NRFU operation. This rate was significantly lower than the maximum we expected could be used. Going into the 2020 Census, we estimated we could use administrative records for up to 22.5% of cases if the first visit was not a successful enumeration or if we didn’t receive a self-response. The lower rate reflects the success of census takers resolving the household status on the first visit and some households responding on their own during NRFU.
• About 20.4% of the occupied households in NRFU. Occupied households resolved through administrative records were about 10.4% of the total NRFU workload, which was lower than the 12.9% we estimated that we might use prior to the census.

We expect these rates to change because we resolved cases and removed duplicate responses during data processing. We will provide updated rates as well as breakdowns by all 50 states and the District of Columbia in April among a variety of operational metrics from the 2020 Census.

Because this was the first time we used administrative records in this way, we do not have similar metrics from previous censuses.

We developed our plan for using administrative records in the 2020 Census over years of testing — from 2013 through 2018. We were ready to implement that plan, but the COVID-19 pandemic required some adjustments.

To help us achieve the best quality census, we modified our procedures in three areas:

• Adapting to a delay in receiving information from 2019 tax returns.
• Using administrative records from a single source when we were still missing a population count.
• Using administrative records more extensively when additional visits were not possible in certain hurricane-damaged areas in Louisiana.

I’ll explain more about each of these below.

The IRS decided to delay the deadline for filing 2019 income tax returns because of the COVID-19 pandemic from April 15, 2020, to July 15, 2020. These records are one of our main sources of administrative records.

The delayed deadline meant the bulk of the tax returns would not be available for us to use as early as we’d planned (for the start of our NRFU operation in May 2020).

However, the pandemic also delayed the start of our NRFU operation. Additionally, as planned, the IRS sent us information each month as some households filed their returns early. Between these two things, we were able to adapt our plan including:

• Updating our list of vacant addresses — As planned, we used administrative records to remove vacant addresses from the NRFU workload after a census taker visited at least once. This enabled census takers to focus on following up with addresses that were occupied. If the IRS then received a tax return from an address we had considered vacant, this new information may have made us less confident (in terms of statistical probabilities) that the address was vacant. When the new information changed our confidence, we added the address back to the workload for a census taker to visit.
• Adding to household rosters — In early 2020, the IRS sent us information on people who earned wages in 2019. As the IRS sent monthly updates, we updated the roster of who lived at an address with the other people listed on the address’ tax return such as spouses and other dependents.

Initially, where we used administrative records to enumerate a household, we made sure multiple data sources corroborated the information.

However, toward the end of the data collection period, if we were still missing a population count for an address but a count was available from an administrative record from a single source, we opted to use that population count.

For example, even if only one source, such as an IRS tax return, indicated that a family lived at an address, we believed the population count from the record provided a more reliable count for the address than leaving the count blank to later impute it. (Imputation is a statistical technique that fills in missing information with other available information. We’ll talk more about it in an upcoming blog.)

We had planned to use administrative records to count people only after census takers visited a specific number of times. However, after hurricanes prevented census takers from making the full number of visits to some households in parts of Louisiana, we used available records to complete the count in those areas.

Many areas in Allen, Beauregard, Calcasieu, and Jefferson Davis parishes were restricted because of damage from hurricanes Laura and Delta. In locations where we did have limited access, a substantial portion of the population had not returned home by the time we ended our NRFU operation.

Since it wasn’t possible to conduct additional visits, we used available administrative records to enumerate cases as occupied.

We expect that using the high-quality administrative records where available provided a more accurate count of those households than if we’d left them blank. During data processing, we would have had to impute their status and the population counts for each if occupied.

While imputation is a widely accepted statistical technique, our preference is to use information from the household whenever possible, and administrative records give us information from the household.

Using administrative records to enumerate households was a change for the 2020 Census that helped make the census more efficient and complete. By using available high-quality administrative records to count households that did not respond, we could focus on following up with the households that were the hardest to count (and the hardest to find in records).

More importantly, we believe using administrative records also helped improve the accuracy of the census because it enabled us to count people that otherwise may not have been counted.