Skip Navigation

Understanding the Census Bureau’s Methods for Completing the 2020 Count

The government used numerous tried-and-true tools to ensure an accurate count of every person living in the United States.

Published: October 7, 2021

Last week, the Amer­ican Stat­ist­ical Asso­ci­ation released its initial report eval­u­at­ing the qual­ity of the 2020 Census.

The report made head­lines with its conclu­sions that there’s “no evid­ence” either that polit­ical inter­fer­ence influ­enced the final count, or that the numbers aren’t good enough to use for their consti­tu­tional purpose of split­ting up seats in the U.S. House of Repres­ent­at­ives. But the report also shed light on how the bureau went about its once-a-decade head count in the face of the Covid-19 pandemic. Those efforts included a vari­ety of count­ing meth­ods beyond collect­ing forms and knock­ing on doors. Let’s take a deeper look at those meth­ods and why they’re crucial to any plan for a full, fair, and accur­ate count.

How did the Census Bureau try to count the whole coun­try in 2020?

The bureau primar­ily coun­ted people by collect­ing answers sent by mail, on the inter­net, or over the phone. If a house­hold didn’t respond by late summer, the bureau sent a census worker to knock on their door — up to six times. Some house­holds didn’t answer, so the bureau asked other people who might know about them, such as neigh­bors, land­lords, or real estate agents (sources known as “prox­ies”). Together, these meth­ods helped the bureau account for 95.5 percent of all occu­pied hous­ing units in the coun­try. 

The bureau coun­ted another 3.8 percent of occu­pied hous­ing units using inform­a­tion it found in admin­is­trat­ive records either from the federal govern­ment or the states. Those records included previ­ous census forms, IRS tax returns, and home-owner­ship docu­ments. The bureau tested this proced­ure widely prior to 2020, and it helped alle­vi­ate the heavy work­load on census takers by allow­ing them to knock on some doors only once.

To count the remain­ing 0.7 percent of occu­pied house­holds, the bureau turned to a data-processing tech­nique called imputa­tion. Imputa­tion fills in gaps by using stat­ist­ics to predict what the miss­ing inform­a­tion is. The bureau bases its stat­ist­ical guesses on data it already has about a house­hold or similar house­holds. The bureau can use imputa­tion to figure out how many people are in a house­hold (known as “count imputa­tion”) or to fill in miss­ing demo­graph­ics about people, such as race or age (known as “char­ac­ter­istic imputa­tion”). This method is stat­ist­ic­ally sound, glob­ally accep­ted by social scient­ists, demo­graph­ers, and stat­ist­ical research­ers, and has been part of the bureau’s toolkit for the census since the 1960s.

Why are admin­is­trat­ive records and imputa­tion so import­ant?

As we’ve seen, admin­is­trat­ive records and imputa­tion typic­ally count only a small percent­age of the total popu­la­tion. But that small percent­age is crucial both to increas­ing the accur­acy of the count and to fight­ing the systemic under­counts that have led to unequal repres­ent­a­tion and lack of fund­ing for histor­ic­ally under­coun­ted popu­la­tions, partic­u­larly communit­ies of color.

No census has ever produced a perfect count. And while the jury is still out on the qual­ity of all 2020 Census data sets, it is unlikely that this census was perfect, either. Indeed, it may still contain signi­fic­ant under­counts and other data inac­curacies. Regard­less, it is essen­tial that the Census Bureau attempt to produce the most accur­ate enumer­a­tion possible. The bureau cannot just ignore house­holds it knows people live in — that’s the oppos­ite of what it means to take a popu­la­tion census. Imputa­tion and admin­is­trat­ive records help the bureau account for people whom it would other­wise miss or skip, and thus get closer to fulfilling its basic, consti­tu­tional respons­ib­il­ity to count every­one.

Moreover, the house­holds the bureau fails to reach aren’t a repres­ent­at­ive cross-sample of the coun­try. So, these gap-filling tech­niques aren’t just about accur­acy, they’re about equity — trying to ensure all groups of people are fairly repres­en­ted in the final numbers.

Stud­ies of past censuses have clearly shown that certain racial and ethnic minor­it­ies face persist­ent under­counts and are under­coun­ted at rates higher than their white coun­ter­parts. The 2000 and 2010 Censuses, for example, both under­coun­ted the Black popu­la­tion by about 2 percent while overcount­ing the white popu­la­tion. The poten­tial barri­ers to census parti­cip­a­tion vary by community, ranging from language barri­ers and hous­ing instabil­ity to fears that the govern­ment will use census data for harm­ful purposes.

The bureau makes efforts to over­come these barri­ers. For example, the bureau now part­ners with local govern­ments and civil soci­ety organ­iz­a­tions to support Get Out the Count campaigns that boost census parti­cip­a­tion at the grass­roots level. But many of those efforts, like the bureau’s massive door knock­ing oper­a­tion, can be extremely expens­ive and time-consum­ing. And no matter what the bureau does, there will always be people it fails to find or house­holds that refuse to open their doors to enumer­at­ors.

High-qual­ity admin­is­trat­ive records and imputa­tion can help fill in the gaps. These meth­ods provide a clearer, more accur­ate picture of the popu­la­tion. Were the bureau to stop using them, or to turn to less stat­ist­ic­ally sound meth­ods of enumer­a­tion, the census would simply miss more people, and minor­ity popu­la­tions would dispro­por­tion­ately feel the harm­ful effects. That would be a signi­fic­ant blow to the accur­acy, equity, and legit­im­acy of the final count.

How long has the Census Bureau been using admin­is­trat­ive records and imputa­tion to count people?

The bureau has a long track record of using admin­is­trat­ive records and imputa­tion to fill in the count. Each decade, the bureau has refined these meth­ods to include more complex proced­ures, handle larger data­sets, and stream­line processing through computers and auto­ma­tion.

As redu­cing racially differ­en­tial under­counts became cent­ral to the bureau’s work, the 1980 Census used admin­is­trat­ive records to identify members of minor­ity communit­ies it had missed. In the 1990 and 2000 Censuses, the bureau coun­ted people on parole or proba­tion as well as milit­ary and federal employ­ees living over­seas using admin­is­trat­ive records. These prac­tices continue today as the 2020 Census used admin­is­trat­ive records both to identify occu­pied house­holds that needed to be coun­ted and to supple­ment inform­a­tion for house­holds that did not respond. 

Imputa­tion also has a long pedi­gree. The 1960 Census was the first to use a modern version of imputa­tion. The proced­ure filled in the count and char­ac­ter­ist­ics of miss­ing house­holds using census data from neigh­bor­ing hous­ing units, ulti­mately imput­ing 0.5 percent of the total count. The bureau contin­ued to use imputa­tion at relat­ively low levels through the follow­ing decades: 0.44 percent of the total resid­ent popu­la­tion in 1970, 0.34 percent in 1980, and, in 1990, 0.02 percent (53,600 responses), the lowest number ever recor­ded.

The 2000 Census used imputa­tion at a some­what higher level. While the bureau origin­ally aimed to reduce the census’s rising cost by devel­op­ing data for non-respond­ing house­holds using a sampled subset of the popu­la­tion, it was forced to change plans after the Supreme Court barred it from doing so. As a result, the 2000 Census relied heav­ily on imputa­tion to supple­ment its door knock­ing efforts, even­tu­ally adding about 1.2 million people (0.43 percent) to the total count. The National Research Coun­cil concluded that had the bureau not used imputa­tion, it would “undoubtedly have under­es­tim­ated the true number of house­hold resid­ents.”

The 2010 Census had similar count imputa­tion rates as 2000, with 1,163,462 people (0.39 percent) added to the total house­hold popu­la­tion. An imputa­tion rate of less than one percent was consist­ent with previ­ous decades, and the contro­versy surround­ing count imputa­tion dimin­ished as the proced­ure became more widely accep­ted.

During the 2020 Census, the bureau contin­ued to use both count and char­ac­ter­istic imputa­tion to cover house­holds. The pandemic also led the bureau to expand count imputa­tion for resid­ences that house unre­lated people, like home­less shel­ters and college dorms, in addi­tion to doing char­ac­ter­istic imputa­tion for those kinds of places.

Are these meth­ods legal?

Admin­is­trat­ive records and imputa­tion are both legal meth­ods for count­ing.

The Consti­tu­tion requires that the decen­nial census produce a count of the “whole number of persons in each state” every decade, but it does not require or prohibit the use of any partic­u­lar count­ing method. Instead, the Enumer­a­tion Clause gives Congress the discre­tion to conduct the census “in such Manner as they shall by Law direct.” Congress has in turn deleg­ated power to the commerce secret­ary to conduct the census “in such form and content as [s]he may determ­ine.” The Consti­tu­tion’s lack of specificity — and Congress’s author­iz­ing the Commerce Depart­ment to determ­ine the best ways to count — matches the Framers’ under­stand­ing that the census requires flex­ib­il­ity in order to produce an accur­ate count and form the basis of a demo­cracy in which every­one is repres­en­ted equally regard­less of how wealthy they are.

The Supreme Court has upheld the consti­tu­tion­al­ity of imputa­tion. Utah chal­lenged its use in the 2000 Census, arguing that the Enumer­a­tion Clause does not allow the bureau to rely on a method that the Framers could not have fore­seen. The Court disagreed and upheld imputa­tion, in part because strik­ing it down would produce “a far less accur­ate assess­ment of the popu­la­tion.” The same would be true were the bureau to aban­don its use of admin­is­trat­ive records, a prac­tice which has never been chal­lenged in court.

In the Utah decision, the Court distin­guished imputa­tion from sampling, a stat­ist­ical method that involves collect­ing inform­a­tion from a subset of the popu­la­tion and extra­pol­at­ing the sample’s results to the entire popu­la­tion. The distinc­tion is import­ant because federal law currently prohib­its the Census Bureau from using sampling for purposes of divvy­ing up seats in the U.S. House (though not for collect­ing char­ac­ter­istic data for draw­ing elect­oral maps at the state level). Of course, Congress could repeal that stat­utory bar at any time, and the bureau could then determ­ine whether its use would improve the accur­acy of census data for congres­sional appor­tion­ment, too.

Accur­ate data is vital to accom­plish the consti­tu­tional purpose of equal repres­ent­a­tion, and the Consti­tu­tion accord­ingly provides flex­ib­il­ity to determ­ine how best to achieve the greatest accur­acy. Admin­is­trat­ive records and imputa­tion are crucial tools for combatting under­counts and increas­ing the equity of the count. Prevent­ing the Census Bureau from using them would not only present severe consti­tu­tional issues — it would bake inequity right into the count.