By , Chief Economist and , Director of Economic Research聽
The 40% drop in multifamily unit starts posted by the Census Bureau last week was a shocker and on its face seems deeply troubling. Dodge鈥檚 measure for the same month came in down as well, but only 12% when compared to April. Both negative, one a slowdown, the other a collapse.
The Census Bureau and 51黑料 have a long history monitoring the US construction industry, which made us wonder why our estimates of multifamily construction were diverging. As many close followers of this data will know, the definitions between Census and Dodge differ slightly, but usually when we harmonize the definitions the broad levels and trends are closely aligned. Every once in a while there may be short lived deviations, but eventually the data reconverge (see Chart 1). With the recent divergence stretching into its fourth year, we dug into the project level data and found that a dramatic rise in mixed use buildings over the past decade appears to be the missing category.
To fully understand, we recommend reading the next couple of sections, which highlight the methodological differences between Census鈥� and Dodge鈥檚 data. However, the reader who is short on time can skip to the fourth section to get to the meat of the issue.
How Census Measures Housing Starts
The Census Bureau’s monthly housing starts figures are statistical estimates derived from the Survey of Construction (SOC), a joint program with the Department of Housing and Urban Development. Rather than counting every home broken ground, Census samples roughly 900 permit-issuing offices across the country, selecting all permits for buildings with five or more units and a 1-in-50 sample of permits for smaller structures. Field representatives then follow up with builders and owners to determine when each unit broke ground 鈥� defined as when excavation begins for the footings or foundation.
How Dodge Measures Single-Family and Two-Family Starts
Dodge collects single-family and two-family housing starts through a monthly survey of permit-issuing offices, mirroring the Census Bureau’s approach but with some important differences. Starting from the most recent Census Annual data, Dodge ranks all permit offices by single-family activity and targets the top 5,000 most active offices as participants. Offices that have asked not to be contacted, cannot report in a usable format, or require a FOIA request are removed and replaced with lower-ranked offices. One notable difference from Census: Dodge tracks only single-family detached units, while Census combines detached and attached (e.g., townhouses and rowhouses) into a single category. Once these definitional differences are accounted for Dodge and Census track reasonably well (see Chart 2).
Dodge’s multifamily methodology diverges significantly from its single-family approach and from Census. Rather than relying on permit office surveys, Dodge collects multifamily starts of three or more units through the Dodge Reporter Network 鈥� a network of staffers in ongoing contact with owners, general contractors, and others with projects in planning. Individual project-level data is gathered for all multifamily projects across the U.S., consistent with all other building types. A multifamily project is recorded as a start, defined as within 60 days of groundbreaking. This project-level approach is why there is less consistency between Dodge and Census generally, but has grown over the last decade (see Chart 3).
Prevalence of Mixed-Use
Much of the divergence between Dodge and Census multifamily figures stems from differences in what each series captures rather than errors in measurement. When we highlight contributions of private housing, public housing, mixed-use developments, condos, and townhomes it鈥檚 clear that the bulk of the issue comes from the mixed-use developments (see Chart 4). The correlation between Dodge鈥檚 private apartment starts and Census multifamily starts was 88% between 2009 and May 2026. For context, the correlation between Dodge single family starts (with harmonized definition) with Census single family starts was 98% over the same time period.
The broader housing market is in decline, but perhaps not as much as suggested by the recent Census Bureau data. If they are missing the mixed-use category, an increasingly important category according to Dodge data, then we have good reason to believe the market may not be in as much trouble as one might think.
Related Articles:
Construction聽Job聽Growth Tilts Nonresidential,聽Housing聽Jobs聽Quietly Retreat聽
Producer Prices Surge in April. Construction Feels The Heat.
January Update on Census Housing Starts and Construction Put-in-Place Spending