Starts in five-plus-unit buildings accounted for about 97% of June's housing-starts increase

Privately ownedPrivately ownedPrivately owned housing excludes structures owned by federal, state or local governments. https://www.census.gov/construction/soc/definitions.html housing starts rose from a revised annualized rate of 1.199 million in May to a preliminary rate of 1.427 million in June. The 19.0% increase was statistically significantstatistically significantIn this release, a change is statistically significant when its 90% confidence interval does not include zero. https://www.census.gov/construction/nrc/pdf/newresconst.pdf under Census's 90% standard. It was not broad across building types: single-familysingle-familySingle-family includes detached homes and attached homes such as townhouses when each unit has a ground-to-roof dividing wall and does not share utilities or heating systems. https://www.census.gov/construction/soc/definitions.html starts slipped from 897,000 to 895,000, while starts in buildings containing at least five units jumped from 291,000 to 513,000. [Census and HUDHUDHUD is the U.S. Department of Housing and Urban Development, which jointly publishes the New Residential Construction release with the Census Bureau. https://www.census.gov/construction/nrc/pdf/newresconst.pdf release](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

The 228,000-unit difference needs translation. It compares two seasonally adjusted annual rates, not the physical number of additional homes started during June. Think of SAAR as a speedometer rather than an odometer: it describes June's adjusted pace as though that pace continued for a year. Census says explicitly that SAAR is neither a forecast nor a projection. [Census definitions](https://www.census.gov/construction/soc/definitions.html)

Using the rounded published rates, five-plus-unit starts contributed 222,000 of the 228,000 increase, or about 97.4%. Single-family starts contributed negative 2,000. That is an arithmetic decomposition, not evidence that five-plus-unit construction caused another category to move. [Census Table 3a](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

The component figures do not add neatly to the total because Census suppresses the seasonally adjusted estimate for buildings with two to four units. The agency says that series does not meet its publication standards for stable, identifiable seasonality. Rounding can create additional small differences. The five-plus-unit category also should not be read as a rental-apartment count: it includes qualifying condominium buildings and identifies structure size, not tenure. [Census Table 3a](https://www.census.gov/construction/nrc/pdf/newresconst.pdf) [Census definitions](https://www.census.gov/construction/soc/definitions.html)

Census placed a margin of plus or minus 15.9 percentage points around the reported 19.0% increase. Subtracting and adding that margin gives a 90% interval from 3.1% to 34.9%. Because the entire interval is above zero, the agency classifies the overall rise as statistically significant. [Census explanatory notes and Table 3a](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

The five-plus-unit estimate was much less precise: a 76.3% increase with an interval from 9.5% to 143.1%. The single-family estimate was negative 0.2%, with an interval from negative 10.4% to positive 10.0%. Because that single-family interval crosses zero, the survey cannot establish whether the underlying month-to-month movement was an increase or a decrease. [Census Table 3a](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

These intervals cover sampling variabilitysampling variabilitySampling variability is uncertainty that arises because an estimate is calculated from a sample rather than from every construction project. https://www.census.gov/construction/nrc/pdf/newresconst.pdf only. They do not cover nonresponse, undercoverage, reporting mistakes or other nonsampling errors. The release reports a 75.6% total quantity response ratequantity response rateA quantity response rate summarizes how much of the requested survey quantity is represented by responding cases under Census procedures. https://www.census.gov/construction/nrc/pdf/newresconst.pdf and says month-to-month movements may be irregular. Census advises that it can take six months to establish an underlying trend in total starts. [Census explanatory notes](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

Revisions add another layer of uncertainty. Census says preliminary seasonally adjusted estimates of total permits, starts and completions are revised by 2.9% or less on average. That average is not a limit on how much any particular monthly estimate can change. [Census explanatory notes](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

June did not merely return to May's depressed rate. Total starts were 0.9% above April, and five-plus-unit starts were 3.8% above April. That is the strongest descriptive evidence that June was more than a perfect reversal. Single-family starts, however, remained 2.1% below April. [Cumulant Research calculations from Census Table 3a](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

The three-month comparison is still too short to demonstrate a trend. Census itself recommends substantially more time for judging total starts, and April and May were revised in the June release. The chart therefore answers a limited question: where June stood relative to the two preceding published months. It does not show that a new cycle has begun. [Census explanatory notes and Table 3a](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

Geography supplies a second counterargument to the narrow-building-type result. The point estimatepoint estimateA point estimate is the survey's single best numerical estimate before its uncertainty is considered. https://www.census.gov/construction/nrc/pdf/newresconst.pdf for total starts rose in all four Census regions: 10.3% in the Northeast, 33.3% in the Midwest, 15.2% in the South and 22.1% in the West. [Census Table 3a](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

Only the Midwest's 90% interval remained wholly above zero. The intervals for the Northeast, South and West included both declines and increases. The correct reading is therefore two-part: the published regional point estimates were geographically broad, but the survey established a positive change only in the Midwest at its stated confidence standard. [Census Table 3a](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

A housing starthousing startA home is counted as started when excavation begins for the building's footings or foundation, and every unit in a multifamily building is counted at that time. https://www.census.gov/construction/soc/definitions.html is one moment in a longer construction process. Census separately reports permits, permitted units that have not started, units under constructionunder constructionA housing unit is under construction after ground has been broken but before the unit is complete. https://www.census.gov/construction/soc/definitions.html and completions. Looking at those stages cannot predict the next starts release, but it can test whether June's jump appeared throughout the same building-size category. [Census definitions](https://www.census.gov/construction/soc/definitions.html)

It did not. Five-plus-unit permits fell from an annualized 468,000 in May to 445,000 in June, a 4.9% decline. The seasonally adjusted count of authorized-but-not-started units fell from 120,000 to 110,000, a statistically significant decline of 8.3% under Census's 90% standard. [Census Tables 1a and 2a](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

Five-plus-unit homes under construction edged from 660,000 to 665,000, a 0.8% increase whose interval crossed zero. Completions fell from an annualized 428,000 to 413,000, but the completioncompletionA single-family home is generally counted as complete when finished flooring is installed, while Census applies an occupancy or availability rule to buildings with multiple units. https://www.census.gov/construction/soc/definitions.html estimate's interval was so wide that it also included a substantial increase. [Census Tables 4a and 5a](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

Those figures do not prove that five-plus-unit starts will fall. Permits can take varying amounts of time to become starts, and projects already in the pipeline can move between stages unevenly. They do show that June's surge at the start stage was not accompanied by similar increases in the adjacent national measures. Description is warranted; a causal story is not.

We tested whether similar jumps usually continued. Using the FREDFREDFRED is the Federal Reserve Bank of St. Louis database that republishes economic series from agencies including Census and HUD. https://fred.stlouisfed.org/series/HOUST total-starts series and its five-plus-unit series, we screened every comparable month from August 1963 through May 2026. A month qualified if total starts rose at least 15% and the increase in five-plus-unit starts equaled at least 75% of the total increase. We began in August 1963 because FRED says the earlier five-plus-unit series also included buildings with two to four units. [FRED HOUST](https://fred.stlouisfed.org/data/HOUST) [FRED HOUST5F](https://fred.stlouisfed.org/data/HOUST5F)

The screen found four months: July 1982, February 2009, January 2011 and November 2011. The share attributed to five-plus-unit starts ranged from 79.2% to 98.9%. These are mechanical matches to analyst-selected thresholds, not evidence that the four periods had the same economic causes. [Cumulant Research calculations from FRED HOUST and HOUST5F](https://fred.stlouisfed.org/data/HOUST)

Total starts fell in the following month after all four episodes. The medianmedianThe median is the middle value after observations are placed in order, or the average of the two middle values when there is an even number of observations. one-month change was negative 11.8%. Yet the six-month outcomes ranged from negative 1.1% to positive 36.0%, and the median was positive 0.1%. That dispersion is the warning: an apparent short-run pattern does not become a useful forecast when it rests on four selected observations. [FRED HOUST](https://fred.stlouisfed.org/data/HOUST)

The historical figures are also revised values, not necessarily the numbers known to investors, builders or policymakers at the time. FRED describes a vintage date as a date when a series received a new or revised value. A real-time forecasting test would need the contemporaneous vintage for every episode. [FRED vintage-date documentation](https://fred.stlouisfed.org/docs/api/fred/series_vintagedates.html)

June establishes that the estimated national pace of starts rose sharply and that the increase was statistically significant under Census's stated 90% standard. It also establishes that the arithmetic was dominated by units in buildings containing at least five homes. It does not establish a broad building-type recovery because single-family starts did not rise, and it does not establish durability because one month is not a trend. [Census release and explanatory notes](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

• Evidence for a broader turn would include sustained gains across more than one building type.
• Evidence for durability would include later months remaining elevated after revisions.
• A firmer five-plus-unit pipeline would require permits and authorized-but-not-started units to stop contracting or begin rising.
• Narrower confidence intervals would make it easier to separate genuine movement from sampling noise.

Census scheduled the July construction report for August 18, 2026. That release will provide the first direct test of whether June's increase persisted, while also revising June's preliminary figures. [Census and HUD release](https://www.census.gov/construction/nrc/pdf/newresconst.pdf)

This analysis concerns measured construction activity. It does not measure a financial-market reaction and does not estimate effects on rents, home prices, employment or economic growth. A change in the reported pace of starts is not, by itself, evidence that any of those outcomes changed because of June's construction figures.