Mapping NIMBY voting in San Francisco

Go to the full map

I wanted to know who votes for NIMBY ballot measures in SF, and whether that has changed over time. So I analyzed local election data since 1996 along with census data and made this map.

The fewer people vote, the greater chance NIMBYist ballot measures have of passing. In predicting whether anti-development measures will win at the ballot box, turnout is more important than income, race, or housing tenure, although those factors play a role as well.

Why is this important?

There’s a strong argument that San Francisco’s highly restrictive development regulations contribute to the city’s housing crisis [1]. San Franciscans have over and over supported measures that control development, both through their support for elected officials and directly through the ballot box [2]. Seemingly small measures like 2014’s Proposition B or 2004’s Proposition J have, in aggregate, created significant barriers to new development [3].

Who votes for these NIMBYist ballot measures? In theory, homeowners have a strong incentive to vote against new development, especially new housing, that might erode their home values. Homeowners are only about 45% of San Francisco’s population, but the need to protect home values gives them a stronger motivation to vote [4]. We’d expect renters to vote for measures that would reduce housing costs. But of course ballot measures are often complex and people vote for many reasons [5]. What do the data actually tell us about how people vote?

Preparing the data

I analyzed data from all local elections in San Francisco going back to 1996. Election results data are available from the SFDOE at the precinct level for all elections going back to 1995. (Annoyingly, they are Excel files and all in different formats with weird inconsistencies, so this was a lot of data wrangling.)

To know what data to extract, I had to decide whether or not each ballot proposition was about development and, if it was, which vote signified a “NIMBY” outcome. This was unavoidably a subjective task. Some of the propositions were clear-cut NIMBY cases, like last November’s (2015) Prop I, which would have placed a moratorium on all new market-rate development in the Mission. Voting ‘Yes’ would have been a clear NIMBY vote. Some were fuzzier: Prop G of November 2006, for example, expanded restrictions on chain stores. This one was difficult since one could be in favor of more independent stores, but against chain stores. However, since it fundamentally increased restrictions on development, in this case I classified a ‘Yes’ vote as a NIMBY vote.

In the end, I settled on 22 NIMBYist ballot measures.

Next, I extracted relevant Census data at the block group level through the Census API. I chose variables expected to be associated with pro- or anti-development voting: housing tenure, median house value, years the household has lived in the housing unit, housing unit type, median income, age, household size, and race and ethnicity. I matched the precinct election data with Census block group data, making sure to match each with the appropriate year. The block group boundaries do not match exactly with the precinct boundaries, so I used an area-weighted proportion to adjust the census data to precinct boundaries.

A map of NIMBY voting

My Leaflet map shows the election results for the NIMBY ballot measures, along with the most important variables from the regression below. We can see there is a noticeable set of precincts in the inner neighborhoods that often votes anti-development, at least for some ballot measures. Perhaps not surprisingly, the pattern seems to correlate with the Progressive Voter Index developed by Richard DeLeon and David Latterman. Interestingly, the most heavily home-owning neighborhoods, those in the southwest corner of the city, do not typically support NIMBY measures.

Regression Results

To find which features were most important, I ran OLS linear regressions, of which the best fit had an adjusted R2=0.515. (Not great, but it’ll do for purposes here. The Y values and residuals are normally distributed; no major collinearity problems.)

The model shows turnout has the strongest association with NIMBY voting. The lower the turnout in a precinct, the higher the percentage of NIMBY votes, even after accounting for home tenure race, income, and population density. This is what we’d expect: opponents of development usually feel very strongly about the issue and make it a point to vote, whereas those in favor of development usually have less incentive to vote. When turnout is low, those voters have disproportionate influence.

The magnitude of effects is show in the plot below. (An elasticity of -0.285 means that an increase in precinct turnout of 10 percentage points would be expected to correspond with a 2.85-percentage-point decrease in NIMBY votes.)

Other variables had smaller effects but were still significant. As expected, precincts with a higher percentage of homeowners were significantly more likely to vote against development. Interestingly, median income had a significantly negative effect, after controlling for homeownership.

I had expected median home value and the number of years a household has lived in their house to be positively associated with NIMBY voting, but these variables were not significant [6]. The median housing unit age was significant at the 95% level, though, indicating precincts with older housing stock tend to vote against development. Precincts with a higher proportion of housing units in large apartment buildings (‘hu_50+’) were significantly more likely to vote pro-development, although the size of the effect is small.

NIMBY voters are often stereotyped as higher-income whites, but in SF strong neighborhood attachments exist among all racial and ethnic groups. It turned out precincts with higher Asian and black populations were significantly less likely support NIMBY measures.

This isn’t a great model, though; it only explains about half the variation in the data (R2=0.52), and a lot of what is explained is wrapped up in the dummy variables that represent unobserved factors in each year.

Types of development-related ballot measures?

I suspect this regression doesn’t adequately capture how turnout works for different types of ballot proposals. For example, in the map there seem to be two kinds of ballot measures; some had strong opposition from the traditionally progressive inner neighborhoods like the Mission and other had strong opposition from outer home-owning neighborhoods like the Sunset. It’s worth taking take a closer look, at least qualitatively. The plots below simply show the correlation between turnout and percent nimby, with a point for each ballot proposal. (It’s not a very strong correlation; like our model said, many other factors are at play.)

Classic NIMBY measures get passed in off-year elections with low turnout

For example:

  • In November 2013, Proposal B and C let voters decide on the fate of a proposed condo project at 8 Washington Street.
  • June 2014, Prop B: a yes vote meant that voters would need to approve any future height limit increases for SF Port (Waterfront) development.

But classic nimby measure tend to fare worse in high-turnout elections

  • November 2000 Prop L was an anti-growth measure that would have preserved caps on office space and would have placed many more controls on the amount of live/work space, among other things. It was very narrowly defeated (49.8% voted in favor) by only 1272 votes. It’s very easy to imagine that this would have passed in an off-year election, rather than a presidential election.

Development outside of traditional residential neighborhoods gains broader support

  • November 1999 Prop H, which approved a downtown Caltrain station, received support from almost all precincts in an election with middling turnout. It probably helped that the station would be downtown and not disrupt residential neighborhoods.
  • March 1996 Prop B: allowed approval of the (now) AT&T ballpark. Despite relatively low turnout, the development received support: the few NIMBY precincts were in lightly populated adjacent neighborhoods. (As we can see in the map).

If I wanted to build a better model, I would include a feature for proximity to the proposed development–this would truly represent the not-in-my-backyard factor. Of course it would also help to have individual voting data, instead of relying on precinct aggregation.

In any case, if the goal is building more housing, increasing turnout would be a good place to start! 

[1] Glaeser, E.L., Gyourko, J., Saks, R., 2003. Why is Manhattan So Expensive? Regulation and the Rise in House Prices (Working Paper No. 10124). National Bureau of Economic Research; Saiz, A., 2008. On Local Housing Supply Elasticity (SSRN Scholarly Paper No. ID 1193422). Social Science Research Network, Rochester, NY; Quigley, J., Raphael, S., 2005. Regulation and the High Cost of Housing in California. The American Economic Review 95, 323–328.; Malpezzi, S., 1996. Housing Prices, Externalities, and Regulation in U.S. Metropolitan Areas. Journal of Housing Research 7, 209–241. Glaeser, E.L., Ward, B. a., 2009. The causes and consequences of land use regulation: Evidence from Greater Boston. Journal of Urban Economics 65, 265–278; Glaeser, E.L., Gyourko, J., 2002. The Impact of Zoning on Housing Affordability; Mayer, C.J., Somerville, C.T., 2000. Land use regulation and new construction. Regional Science and Urban Economics 30, 639–662; Robinson, T., 1995. Gentrification and Grassroots Resistance in San Francisco’s Tenderloin. Urban Affairs Review 30, 483–513.
[2] For a good history of SF politics and their influence on development regulations, see DeLeon, R., 1992. Left Coast City: Progressive Politics in San Francisco, 1975-1991. University Press of Kansas, Lawrence, KS.
[3] High housing demand also contributes to the housing crunch, of course, but if we’re interested in acting to solve the problem, targeting supply-restricting regulations is more effective strategy, because regulations can be changed through policy.
[4] Fischel elaborated on this phenomenon he called the “homevoter hypothesis.” Fischel, W.A., 2009. The Homevoter Hypothesis. Harvard University Press, Cambridge, MA.
[5] There’s some evidence that liberal cities place more limits on development. Dubin, J.A., Kiewiet, D.R., Noussair, C., 1992. Voting on Growth Control Measures: Preferences and Strategies*. Economics & Politics 4, 191–213.
[6] One reason median home value was not significant might be that the ACS truncates housing unit values at 1 million, so any value over $1 million is just reported as $1 million. Obviously in San Francisco the scale needs to be higher.