A proposed rule would require the Census Bureau to use tax records to verify where people report living for the 2030 Census. Tax data are valuable administrative records, but using them as a residence test could result in people being missed or counted in the wrong place: Tax data do not cover everyone; the addresses can be stale or nonresidential; and tax processing schedules do not align neatly with the 2030 Census calendar.
What would the proposed rule do?
The Constitution requires an “actual Enumeration” of the population every 10 years. The count determines how seats in the US House of Representatives are apportioned among states and guides the distribution of federal funds (more than $2.8 trillion across 353 federal assistance programs in fiscal year 2021). In addition, Census data underpin many national and sub-national surveys and other statistical programs.
The proposed rule would redefine a person’s “usual residence” on Census Day (April 1) as the residence where the person “lawfully spent the greatest number of days” from January 3 through April 1. For comparison, in the 2020 Census usual residence generally meant the place where a person “lived and slept most of the time.”
The proposal also says a person’s “usual residence should be consistent with, and evidenced by” tax records such as tax returns and information documents such as W-2s and 1099s. Separately, the residence proposal would change which foreign citizens are included in the apportionment count—an issue beyond the scope of this discussion.
Tax records do not cover everyone or always show where people live
Tax records do represent a vast majority of the US population, but there are gaps. A 1998 IRS study using 1993 tax records, including information returns such as W-2s, identified about 95 percent of the US population in an adjusted 1990 Census estimate.
But the same research noted that some tax return addresses were not residential addresses at all. They could belong to a tax preparer, business, post office box, military post address, or otherwise fail to identify where the taxpayer actually lived.
The research also noted that information documents can provide one or multiple residential addresses to supplement return information. Such inconsistencies arise when taxpayers move after filing their annual tax return.
Movers create an additional timing problem
IRS migration data show substantial year-to-year address changes among tax filers. Almost 15 million taxpayers changed addresses between tax years 2022 and 2023, and about 8 million movers stayed with in the same state but may have changed locality (county and/or Congressional district). The remaining almost 7 million moved between states. The figure below shows that interstate movers skew younger: Filers under age 35 accounted for a much larger share of movers.
For movers, it is essential that the Bureau use the most recent IRS information for verification, but IRS processing takes time. For taxpayers who request extensions to file by October 15, updated IRS address information may not be available to the Bureau until after December 31, 2030 when it must provide apportionment data to the president.
The timing problem is not hypothetical. In the Bureau's 2020 administrative-record simulation, 3.7 million 2019 tax returns covering 5.2 million people were not available in time to produce state-level population estimates by the December 31 deadline. And the problem persists—the Treasury Inspector General for Tax Administration reported that in February, IRS had a back log of 2.4 million unprocessed returns.
Matching Census and tax documents is complex and not always successful
Even when the IRS provides current addresses, the Bureau must accurately connect tax records to its files. Its linkage systems are sophisticated, but they are not perfect. In a 2023 report on COVID era Emergency Impact Payments (EIP), the Bureau was unable to link its files to 2.7 percent or almost 6 million IRS provided records. No statistics on incorrect linkages among those who were matched were reported, but 2014 research suggests they are relatively rare.
Use tax data as one of many sources to improve Census accuracy
The Bureau’s own research points toward a more robust approach to validating 2030 Census responses: Use tax records alongside multiple administrative sources.
In the 2020 simulation, the Bureau combined 31 types of administrative-records and third-party data to produce population estimates on the same timetable as the decennial census, and concluded that expanding the number and timeliness data sources could improve administrative-record population estimates.
The simulation also illustrated the value and limitations of IRS address data. IRS data were the dominant source: 88.5 percent of people appeared in at least one IRS source, and 36.9 percent were found only in IRS data. But about 8.4 percent of IRS 1040 records lacked a residential address and could not be linked to Bureau files.
The proposed January-to-April definition of usual residence raises a separate concern for seasonal residents. Someone who spends most of January through March in a winter residence could therefore be counted there, even if they live and sleep elsewhere for most of the year. The Bureau should explain how it would handle such cases and resolve conflicting records.
Whatever approach it adopts, the Census Bureau should clearly document its record-linkage plan, validation rules, and results, consistent with its data-quality standards, and make that documentation publicly available.
The public has an opportunity to provide feedback on the proposed rule by November 2.
