Under the Hood: Multifactor, ZIP code-level targeting for ADU outreach
Accessory dwelling units (ADUs) have the potential to add to the affordable housing stock in developed single-family neighborhoods. The Bay Area is in desperate need of affordable housing, and communities are turning to ADUs to pick up the slack. Unfortunately, ADUs can be expensive to build in the Bay Area (due in part to large permit fees and high costs of labor). This is often a deterrent for low-income households, who would actually benefit the most from a monthly stream of rent income that an ADU would provide.
San Mateo County’s Office of Sustainability, which notably champions ADUs as a sustainable communities strategy, reached out to us for advice on how to geo-target their ADU outreach in the coming months. Their goal is to focus limited Facebook ad dollars on neighborhoods that do not already have a high ADU production rate. Additionally, they wanted a method to weigh various socioeconomic characteristics to align with their equity goals. Prompted by this need, we set out to build a communicative weighting tool that would allow the county to target ADU outreach opportunities at the ZIP code level. The datasets, processing, and methodology of the tool creation is described below.
Assessor Data
We were able to acquire San Mateo County’s Assessor Roll, which contains a list of parcels and descriptive characteristics of buildings located within the county. For the purpose of our analysis, we were only interested in considering single-family residential parcels for potential ADU construction (although the state law does allow for ADU construction on multi-family and some commercially zoned lots). As a result, we narrowed down the pool to only include those with a property use code of 1 (single-family residential). In order to quickly assign each parcel to a ZIP code, we spatially joined the parcel geometry object from the assessor roll to a ZIP code tabulation area (from the tigris package in R). We also used the building characteristics data to obtain a simplified version of a parcel’s yard area — the difference between lot size and the first floor area of the residential structure. Although there are technically other spatial constraints to consider, such as setbacks, we determined that the simplified buildable area was a good enough proxy to obtain a median buildable area at the ZIP code level.
Annual Progress Report (APR) Data
A jurisdiction must fill out an Annual Progress Report to notify the State of California of the number of building projects that received entitlement permits, building permits, and certificates of occupancy in a given year. We obtained a California APR state summary through our partnership with 21 Elements and San Mateo County. We used the building permit counts as a proxy for the rate of ADU production in each year (instead of using certificate of occupancy, since we’re mainly interested in a jurisdiction’s intent to build ADUs).
The APR summary contains building permit data for the years 2018, 2019, and 2020. When these counts are totaled at the jurisdictional level, they can provide some insight on the success of a city’s ADU outreach over time (after accounting for confounding variables). For the purposes of our weighting tool, we averaged the counts from these three years together to represent the recent average rate of ADU production at the ZIP code level. On their own, these average counts do not provide much insight; some ZIP codes may have more residential parcels than others, so we need a way to even the playing field for comparison. As a result, we calculated the average ADU production rate per 1000 eligible residential lots to represent a “normalized” value.
“Effective” Parcels
Furthermore, an “eligible” lot (mentioned above) does not necessarily have the best building conditions for an ADU. A more accurate variable would be “effective lots”, on which the physical conditions of the parcel are favorable for ADU construction. Lot size and steepness are two such factors that can impact ADU feasibility.
Effective Lot Size
A parcel’s buildable yard area is the make-or-break factor for a potential ADU. We needed a way to determine a general threshold to eliminate lots with small yard sizes from ADU consideration. Using a cumulative distribution plot, we identified the buildable area threshold that 5% of residential lots in SMC are beneath. This corresponds to a buildable area of approximately 2,500 square feet, so we classified parcels with buildable areas above this size as “effective”, or practical, for an ADU.
Effective Lot Steepness
Lots in hilly regions will also have limitations for ADU construction. In order to calculate the number of residential lots that are too steep for an ADU in San Mateo County, we made use of the publicly available SMC Digital Elevation Model (DEM) from 2017. Using R, we applied geospatial and raster extraction techniques to obtain the minimum and maximum elevations for each residential parcel. To estimate slope, we took the difference between the two elevations and divided it by the square root of the parcel’s area. Like for buildable area, we plotted the slopes in a cumulative distribution plot to identify the 95th percentile of parcel slopes. This corresponds to a slope value of 0.47, so any lot that has a higher slope than this value is considered too steep and removed from the “effective” pool.
Census Data
Socioeconomic variables also have an effect on ADU construction expectations (and equity goals) in a given neighborhood. The County wanted to be able to, for example, focus a greater share of their outreach in neighborhoods with lower-income and more diverse households. Using the Census API to pull 2019 5-year American Community Survey data, we obtained the median income and racial make-up of each ZIP code in San Mateo County. Since we had this data handy, we decided to examine the relationship between median income and ADU permits per 1000 effective lots.
There is an apparent positive correlation between income and ADU production, which suggests that all else being equal, we might expect a more affluent city to naturally have more ADU production. So it’s not enough for a more affluent city to have slightly higher production than a less affluent city. In fact, it could be that the less affluent city is doing better on ADUs relative to its population.
In the map linked below, you can also observe an inverse relationship between median income and % non-white population at the ZIP code level.
Ranking ZIP Codes for ADU Outreach
The final step was to communicate the results of our data processing to the County and create a tool that allowed them to fine-tune the weights of individual variables to create a ranking of ZIP codes for outreach. The following steps outline the process of creating a ZIP code ranking tool:
- Use a Google sheet as the tool’s database platform to allow for straightforward collaboration
- Normalize all of the variables described above to eliminate the impact of differing data ranges
- Create a “scoring inputs” table, where the county can change the weight that the normalized variables have on the final ZIP code rank. One can enter any value between -1 and 1 in the score cells. A negative sign indicates that you would like to “flip” the impact of lower scoring items. For example, if you want to target areas with the fewest number of ADUs, then adding a negative sign to the permit data weights will result in ZIP Codes with fewer ADUs scoring higher.
- The final score is the sum of all weighted, normalized factors. The ZIP codes with the highest score can be considered as the prime locations for ADU outreach.
To illustrate a potential weighting scheme, imagine you decide to set all weights associated with ADU building permits -1 (so ZIP codes with fewer ADU permits will have a higher score). The weight for % White population is maybe set to 0, so we are removing the impact of that % on the score. The weight for % Asian population is maybe set to 0.5, so we are still considering that % but lessening its impact on the final score. All other weights are maybe set to 1, so they have a full impact on the score.
Bigger Picture
Beyond the specific outreach efforts of San Mateo County, this methodology has broader implications on how to encourage more ADU production in jurisdictions. Using the APRs, and controlling for confounding factors as we’ve done, regional and state leaders can identify which cities are exceeding in ADU building permit distribution, and which are lagging behind. All cities have been provided a Regional Housing Needs Allocation (RHNA) through a state-wide process, and have until January 2023 to submit a Housing Element that describes their plan for how to produce the allocated number of housing units across different affordability tiers. Cities are eager to use ADUs as a way to meet their affordable housing requirements, and we would want that too — but we wouldn’t want a city to exaggerate its ADU production projections so as to get away with actually planning for affordable housing. So we can use the data analysis we’ve presented here as a ground-truth mechanism to create a reasonable range of ADU production estimates for each city that is based on their actual history of production. Furthermore, we can create incentives for cities to be able to incrementally increase their projections, only if they proactively implement certain kinds of ADU-friendly policies and streamlining processes. In any case, tying the RHNA process to actual data may create a virtuous feedback loop where cities are rewarded for immediately getting to work on increasing their ADU numbers this year — and that’s good for all of our Bay Area neighborhoods.
