Accessory Dwelling Unit (ADU) Permits & Neighborhood Income

Our research aims to explore if there is a correlation between the median income of Los Angeles's neighborhoods at the census tract level and their number of permitted ADUs. Our hypothesis is that the permitting of ADUs will be negatively associated with income because neighborhoods with more wealth will be more inclined to maintain their exclusivity by resisting expanded access to housing opportunity.

The Los Angeles Department of Building and Safety (LADBS) defines an ADU as “an attached or detached residential dwelling unit that provides complete independent living facilities for one or more persons and is located on a lot with a proposed or existing primary residence.”

The permitting of ADUs has increased drastically in California following state legislation enacted in early 2017. “SB 1069” authorized cities to create ordinances to allow for the development of ADUs in single-family or multifamily residential zones. This legislation is especially relevant for a city like Los Angeles as nearly half of the land it encompasses is zoned for single family use. Further, SB 1069 replaced any city’s existing ADU building, parking and use standards and imposed ADU standards on cities that had not yet adopted their own.

Since SB 1069 was enacted, Los Angeles has passed subsequent ordinances to ease restrictions for the development of ADUs, as well as developed programs to provide support to homeowners who are interested in renting to older adults facing housing insecurity. This research project focuses on Los Angeles’s recent ADU development patterns, in particular, because the city has demonstrated a clear interest in facilitating an increased supply of this particular housing typology.

Additionally, the development of ADUs offers one way to increase housing opportunity through “gentle density,” in a region like Los Angeles that has been historically averse to new mid- to high-rise housing developments. ADUs also have the potential to enhance affordability for renters, as well as to provide an additional source of income to middle-income homeowners or lower-income homeowners that may be cash poor but asset rich in that they own property.

This section introduces the data sources we used in this research project and briefly describes our analysis process.

• Our first primary data source was LADBS's permit database, which provides information about all permit applications filed with the department since early 2013. See the data source  here .

• Our second primary data type was income data. We pulled median income data for each census tract in the City of Los Angeles from the American Community Survey. To do so, we used censusreporter.org. See the data source  here .

• As we’ll explain further below, we determined the number of permitted ADUs in every census tract and normalized that data on a per capita basis. Our median income dataset, which contained the census tract geographies, did not have population data. So, we downloaded a supporting dataset from Census Reporter that contained population data. See the data source  here .

1. Cleaned data: The LADBS permit dataset is quite robust and includes over 1,200,000 entries. We focused on the column titled "# of Accessory Dwelling Units" to filter for ADU permits only. We then filtered to only include rows that provided geospatial data, which narrowed our final permit dataset to 8,668 entries.
2. Imported relevant libraries: pandas: reading and wrangling datasets; geopandas: spatializing data; contextily: creating base maps; folium: creating interactive choropleth maps; plotly express, matplotlib.pyplot: creating graphics and scatterplots; esda, splot, libpysal: creating spatial statistics.
3. Analyzed permit data with bar charts.
4. Mapped permit and income data separately through provided lat/long columns in datasets.
5. Spatially joined permit and income datasets and normalizing to account for ADU permits per 1000 people.
6. Plotted joined dataset to analyze relevant patterns.
7. Conducted significance testing through spatial lag, p-value and moran’s plotting.
8. Created final scatterplots and bar charts to show which income levels are filing the most ADU permits.
9. Contextualized findings with relevant research articles.

We started by considering our two datasets separately. First, we analyzed the permit dataset. We began with a series of bar charts to better understand its contents.

Looking at the chart above, we see that 2019 was a high performing year for ADU permits. It’s also worth noting that there was a significant decrease in the number of permits filed in 2020 compared to 2019. Our guess is that the Covid-19 pandemic caused homeowners to rethink their financial plans and put home addition projects on pause. Additionally, the shift to remote working likely caused the development process to slow down quite a bit.

The status of a permit can range from "Approved" to "Issued" to "Certificate of Occupancy Issued." The latter is largely considered to be one of the final steps in the construction process as the CofO certifies that the recently constructed or converted property complies with all building codes and is fit to be inhabited by an individual or family. In 2020, the number of "Issued" permits was much larger in comparison to the CofOs issued. This deviates from the historical patterns that we see in 2017, 2018, and 2019, where more CofOs are issued in comparison to "Issued" permits. This is likely due to Covid. It's unfortunate to see that not as many completed ADUs are receiving the necessary documents for residents to occupy them.

LADBS has five different permitting offices throughout the city: Metro, Van Nuys, West LA, South LA, and San Pedro. From the chart above, we learn that, over the last 5 years, the majority of permits have been filed in the Van Nuys Initiating Office.

For our final bar chart, we considered the number of ADU permits for each year by zip code. This chart allows us to focus on specific geographic areas within Los Angeles. We found that a few zip codes filed a much larger number of ADU permits in comparison to others. The standouts were 91331, 91335, 91342 and 91605, all of which are located in the San Fernando Valley. This came as little surprise to us, as we knew the Van Nuys Permitting Office has processed a disproportionate number of ADU permit applications in comparison to the city's other offices.

The following subsection describes the results of our neighborhood median income analysis. The figure below shows the median income of each neighborhood in bar plot form.

The index, from 0 to approximately 1,000, presents each census tract in Los Angeles. While this information doesn’t clearly reveal spatial information, it is nevertheless interesting to see how segregated high- and low- income neighborhoods are. Neighborhoods with similar median incomes tend to have similar indexes, denoting their spatial proximity.

The histogram above shows that the largest number of neighborhoods have median incomes around the $50,000 range, while some outlier neighborhoods have median incomes as high as $250,000. Our data has a right-skew, as income data usually does. A larger portion of neighborhoods have lower incomes, while a smaller portion of neighborhoods have higher incomes.

The figure below shows neighborhood income throughout the city in map form. With this figure, we can more clearly see the spatial trends that we predicted from our first bar plot for neighborhood median income.

As anyone who has studied or spent time in Los Angeles could predict, the city exhibits clear spatial segregation of income levels. Neighborhoods like Bel Air, Beverly Glen and Laurel Canyon have the highest income levels, South LA has some of the lowest income levels, and the San Fernando Valley, parts of Mid City, and West LA have a middle range of incomes. 

After analyzing the two datasets separately, we then conducted a spatial join to create a single geodataframe containing each permit, along with its census tract, and that tract's median income. The figure below shows the 20 census tracts with the most ADU permits filed since 2017, all but a few of which are located in the San Fernando Valley.

We then normalized our joined datasets to account for ADU permits filed per 1000 people. If a census tract has twice the population of another, we did not want to give it credit for having more ADU permits. We then plotted the top performing 20 census tracts. Even after normalizing the data, the vast majority of tracts are still located in the San Fernando Valley.

Once we mapped our spatially joined datasets to show the location of permits per 1000 residents by census tract, we began to notice spatial clusters of permit activity throughout the city. To further evaluate these clusters, we conducted spatial autocorrelation, a process to determine to what degree an existing pattern is or is not random. We used the Global Moran's I statistic to help us quantify the degree to which similar geographies are clustered.

Spatial Weights / Spatial Lag: We began by assigning spatial weights to each tract to compare its ADU permit performance to its 8 nearest neighbors. From here, we then used these weights to calculate the spatial lag, a single number per census tract that is calculated by taking the average of its 8 nearest neighbors' spatial weights.

Comparing the "permits_per_1000" and "permits_per_1000_lag" columns, we see that when the former value is higher, the census tract is closer to being a "diamond" - more permits than its neighbors. When the latter value is higher, it's closer to a "donut" - less permits than its neighbors.

Moran's Local Scatterplot: Our Moran's I value is 0.047. This positive value indicated that we had a positive spatial autocorrelation and that our results are not occurring at random, in a global sense. We then conducted a local spatial autocorrelation to determine exactly where clusters of activity with statistical significance are happening in the city. We used the "Local Indicators of Spatial Association (LISA)" to classify these clusters of activity into the following four groups:

• HH (red): high permit rate geographies near other high permit rate neighbors
• LL (blue): low permit rate geographies near other low permit rate neighbors
• LH, donuts (light blue): low permit rate geographies surrounded by high permit rate neighbors
• HL, diamonds (yellow): high permit rate geographies surrounded by low permit rate neighbors

As seen below, we then mapped these four categories to show the statistically significant census tracts. As mentioned previously, significant clusters of high permit rate activity are predominantly seen throughout the San Fernando Valley. However, this map also highlights that significant clusters of low permit rate activity can be seen throughout many parts of East LA, South LA, Echo Park, East Hollywood, Brentwood, Venice, Playa Del Rey, as well as communities located along the southernmost parts of the 110.

Scatterplot - Median Income vs. ADU Permits per 1,000 Residents: The figure below shows the relationship between each census tract’s median income and the number of ADUs it has permitted along with a best-fit line for this relationship. As seen by this line's slope, this relationship is positive – the higher a census tract’s median income, the more ADUs it is likely to have permitted. However, this doesn’t tell the whole story.

Bar Chart - Median Income vs. ADU Permits per 1,000 Residents: The figure below shows the average number of ADUs permitted per 1,000 people for each given median income range. This sheds more light on the relationship between neighborhood median income and ADUs permitted per 1,000 people. We can see that the trendline from the scatter plot is a bit misleading; ADU permitting is highest for the "middle" income neighborhoods (median incomes greater than $50k and less than $175k). Neighborhoods below or above this range have less ADUs permitted per person.

Side-by-side Map - Median Income vs. ADUs Permitted per 1,000 Residents: The figures below show choropleth maps of ADUs permitted per person and census tract median income side-by-side. We can apply the findings from the previous histogram to this map. Some of LA's highest income neighborhoods (Bel Air, Laurel Canyon, etc.) and its lowest income neighborhoods (South LA) have very few ADUs permitted per 1,000 people. The area with the most ADUs permitted per 1,000 people is, as the histogram would suggest, "middle" income areas, many of which can be seen throughout the San Fernando Valley.

Our research hypothesis was that the permitting of ADUs will be negatively associated with neighborhood income because neighborhoods with more wealth will be more inclined to maintain their exclusivity by resisting expanded access to housing opportunity. The results above partially validate this hypothesis, but also add more nuance. It's true that the wealthiest neighborhoods in Los Angeles permit very few ADUs per 1,000 people. However, the poorest neighborhoods also permit very few ADUs per 1,000 people. Perhaps there is a lower threshold of income needed for a neighborhood to construct a large number of ADUs and also a higher threshold of income where neighborhoods discourage ADU construction even though they can afford them. This is one possibility; existing literature on ADU production can help us understand our results.

• UC Berkeley: Terner Center for Housing Innovation, " Reaching California's ADU Potential: Progress to Date and the Need for ADU Finance ":

"...given how ADU production in Los Angeles, California’s most diverse region, dominates the rest of the state. Looking at ADU patterns by tract median rent quartile, Figure 10 confirms that ADU permitting and completions have lagged in the neighborhoods in the lowest quartile of median rent for the state, but shows that the two middle quartiles have outperformed the highest quartile in ADU production."

"...Los Angeles and Orange County experience most of their building in low resource areas, while moderate and higher resource areas see most ADU construction in other regions. In other words, though all types of communities are embracing ADUs, exclusive areas in Southern California are less likely to produce ADUs, while elsewhere, low resource areas lag behind."

• UC Berkeley: Center for Community Innovation, " ADUs in California: A Revolution in Progress ":

"Homeowners in high home value areas across the state are more likely to construct ADUs, but those in lower-income, lower-rent areas of Los Angeles are also more likely to build. The other significant variables in the Los Angeles case reveal further nuance. Here, neighborhoods with higher proportions of non-Latinx White, Latinx, and/or Black populations are all more likely to build ADUs, controlling for all else, as are neighborhoods with high rates of overcrowding (people per room), smaller lots, and more recently purchased homes."

This project's final results prompt many exciting future research inquiries, a couple of which are listed below:

• Is there a relationship between a census tract's population by housing tenure and the number of ADU permit applications submitted to the City of Los Angeles?
• Is there a relationship between neighborhood income by housing tenure and the number of ADU permit applications submitted to the City of Los Angeles?
• Is there a relationship between a neighborhood's ethno-racial makeup and the number of ADU permit applications submitted to the City of Los Angeles?
• Are certificates of occupancy being issued at higher rates in LA neighborhoods with higher incomes?

When we first started this project, one of our initial research interests was to explore rental prices of LA's ADUs constructed since the passage of SB 1069. We were curious to see which kinds of renters this building typology has primarily served in recent years. However, because the city denies access to any publicly available rental registry, we were unable to pursue this research endeavor.

We would like to thank Yoh, Chris and Eleanor for creating an incredibly rewarding learning experience this past quarter! We are grateful for your support and guidance.

Bryan Graveline: Focused on analyzing and plotting the median income and population data, performing the spatial join, and creating the final scatterplots and bar charts.

Nathan Keibler: Focused on cleaning, analyzing and plotting the permit data, conducting the local spatial autocorrelation of the joined dataset, and creating the StoryMap.