A one-stop source about accessory dwelling units, multigenerational homes, laneway houses, ADUs, granny flats, in-law units…
Accessory dwelling units (ADUs) in California are “a revolution in progress,” in the words of researchers at the University of California’s Center for Community Innovation. A series of laws enacted over the past five years have systematically eliminated most explicit and implicit legal barriers to ADU construction across the entire state. Using data reported to the California Department of Housing and Community Development (HCD), cleaned by Kat Gordiienko of Builty, and published by Kol Peterson, I investigated likely economic and regulatory determinants of ADU growth across California cities.
Digging into the data, I found three key facts:
Fact 1: Los Angeles is dominant. Large cities are rarely at statistical extremes, because their diverse neighborhoods tend to average each other out. So it’s quite a surprise that Los Angeles is ranked third out of 457 California cities with ADU permit data. (See Table 1).
Table 1: California’s ADU Leaders
|City||Permits (2017-2019)||Permits per 1,000 SFH|
|California cities in sample||23,476||3.7|
|California, statewide||26,606||> 2.9|
A huge city producing ADUs at a high rate is a recipe for statewide dominance. The next-highest major cities, San Francisco and Oakland, produce at little over half of LA’s rate and have – combined – a quarter as many single-family homes (SFH).
Table 2: ADU Permits in California’s Biggest Cities
|City||Permits (2017-2019)||Single-Family Homes||Permits per 1,000 SFH|
A statewide map of permitting rates shows that ADUs are clustered in cities along the coast, both in major metro areas and rural counties. Figure 1 shows an inset of the Bay Area.
For statistical analysis, I divide the state into five regions. The ADU production rate is highest in the Southland – but only because of the City of Los Angeles, as Table 3 shows.
|Table 3: Regional ADU production|
|Region||# of cities||Permits per 1,000 SFH|
|— without LA||160||1.4|
Fact 2: ADU production is random but persistent
Most city statistics vary across space. If you know the median household income of a city’s neighbors, for instance, you can predict the city’s own median household income with reasonable accuracy. ADU permit rates, however, are significantly more random.
Computing “spatial autocorrelation” several different ways, I find that permit rates have less than half the autocorrelation of income, rent, and white population share.
Across time it’s a different story. For each year from 2017 to 2019, the correlation with the previous year’s ADU permit rate is at least 0.5, implying reasonably strong persistence.
Fact 3: Access to jobs is the most important predictor
Although ADU production is pretty random, city characteristics can explain about a quarter of the variation among cities. The most important factor is access to jobs – more so than rent, income, or region. It is important to note that these are descriptive, not necessarily causal, characterizations.
I measure job access two ways: average travel time to work and distance to the nearest of the state’s four largest job hubs. Some places (say, San Luis Obispo) have relatively few jobs, but all within an easy commute. Others (Antioch) have access to far more jobs, but at the cost of a long commute. In metropolitan areas, commutes increase with distance from the city center, but quite gradually.
Other variables also have statistically significant predictive power:
Karen Chapple and her colleagues at the Center for Community Innovation graded cities that had updated their ADU ordinances by early 2020. The cities that updated are likely to keep better records of ADU permits and submit higher-quality data to the state, so the statistical relationship may reflect data quality.
I expected that cities that received higher grades would also produce more ADUs. But that relationship is statistically weak. In another state – one without statewide harmonization of the key ADU rules – one would expect regulations to have a much bigger impact on ADU permits. I did not have data on ADU incentive programs, which might have a big impact.
Finally, two key economic variables did not have the effect I expected. Higher rent presumably indicates high demand for ADUs, and higher homeowner incomes may be necessary to fund supply. Neither effect shows up.
More research is necessary to determine why higher-rent locales failed to produce more ADUs.
In some California cities, notably Los Angeles, an ADU revolution is underway. If production continues at 2019 levels, one in ten LA houses will have a modern ADU by 2026. But in other cities, there is barely a blip; Fresno hasn’t built enough ADUs to field a baseball team.
Some city characteristics, like job access and being in the Coastal region, help explain high ADU permitting rates. But the majority of the variation across cities cannot be easily explained, even among those for which ordinances have been carefully analyzed. The differences may result from city outreach efforts, regulatory practices, or be an artifact of city reporting practices. Alternately, there could be strong “contagion” effects, with ADUs catching on in some places a few years earlier than others nearby.
Appendix: Data and Methods
Permits. The author thanks Kat Gordiienko for collating, checking, cleaning, and sharing HCD’s data on ADU permits. I measure the ADU permitting rate as total permits, 2017 through 2019, divided by the number of single-family homes (SFH), in thousands. Kol Peterson, who published city-by-city ADU data, notes that HCD “cannot independently verify” cities’ reported permit rates. It is likely that cities underreport by accident or by mis-attributing ADU permits to other classes. Some permits may reflect legalization of existing illegal ADUs, but Kat Gordiienko reports that fewer than 1 percent of Los Angeles permits contain “legalize” or similar words. I include an indicator for whether a city has updated its ADU ordinance to capture some of the systematic reporting error.
Scope. The data covers 457 cities that reported ADU permits to HCD. In unincorporated areas, counties issue building permits. However, data is not readily available for these areas, so I have excluded them from the analysis.
Geography. Using centroids, I compute the distance from each city to the downtowns of San Francisco, Los Angeles, and San Diego and Silicon Valley’s Googleplex, each representing a major statewide job hub. I take the natural logarithm of the shortest distance for each city. I define the Bay Area as cities within 50 miles of either downtown SF or the Googleplex. Southland is any city within 80 miles of downtown LA, with the exception of those in San Diego County, which is its own region. The Coastal region includes cities in coastal counties which are not in one of the metropolitan regions, plus one city in Napa County. The remainder is Inland.
City data. Travel time to work, demographics, household income, and median rent – are drawn from the 2015-2019 ACS and accessed via Census Reporter. I report results at a standard deviation’s change.
ADU Ordinance Grade. Chapple et al assigned letter grades, from A- to F, to updated ADU ordinances. I converted these to Grade Point Averages (3.67 to 0.0) for statistical use. I find that the grades have a spatial autocorrelation about one-third as large as income, rent, and race. Grades tend to be higher in cities that are larger, whiter, have lower rent, and are in the Bay Area and Coastal regions. However, these factors together explain only 20 percent of the variation across cities.
Spatial autocorrelation. I computed Moran’s I statistic using several power and exponential specifications, comparing the autocorrelation of permitting rate to that of rent, income, and white population share.
Regression. I used STATA’s spregress command to implement Generalized Spatial Two Stage Least Squares, with heteroskedasticity error correction and an inverse difference weight matrix. Data and code are available upon request.