The Crisis Of Voter Turnout In Public Housing
By John Ray (@johnlray)
Using the New York State voterfile made available by the New York State Board of Elections, Data for Progress geocoded active voters to understand the role that public housing status plays in election turnout. Here, we focus on turnout in the November 8, 2016 general election. While statewide turnout in 2016 was about 57 percent of the vote-eligible population, here we report turnout as a percent of those who are active on the NYS voterfile. Using the active voterfile as the denominator, our reported numbers will appear higher than this 57 percent baseline. This is because we report turnout among active voters, not among the entire vote-eligible population.The NYS voterfile contains a variety of geographic information for each active voter, including address, town, county, and ZIP-9 for each voter ID. We concatenated those fields into an address string, which we then passed to the Census geocoding API. [1] Passing an address to the API returns a lat/lon coordinate associated with that address.
Then, using a publicly available shapefile of public housing buildings in New York City, we conducted a spatial overlay of those address lat/lons on top of the public housing shapefile to determine whether an address successfully geolocated to a public housing building, or not. In total, we located about 6,000 New York City voters to a public housing unit, among whom about 3,500 were active voters.
We then computed mean turnout for New York State voters grouped into the following populations:
New York State voters as a whole,
New York State, not in New York City [2]
New York City
In a New York City zip code that has at least one public housing unit in it
In public housing
Because of the large sample sizes involved, each of the results reported below are statistically if not always substantively significant.
There were significant disparities in 2016 turnout across New York State. While 57 percent of the vote-eligible population in New York State voted, that same population as a share of the active New York State voterfile is 68 percent. Sixty-nine percent of New York State voters not residing in New York City voted in 2016, compared with 65 percent of New York City voters active on the voterfile.
While voter turnout in ZIP codes that contained at least one public unit was very similar to citywide voter turnout at about 65 percent as well, voting by residents living in a public housing unit was significantly lower. Only 54 percent of New York State voters active on the voterfile and geolocated to a public housing unit turned out to vote. In other words, turnout among public housing voters is about 14 percentage points lower than the statewide average.
Indeed, compared to the demographic information available on the New York State voterfile, living in public housing is a stronger predictor of lower turnout than voters’ age, sex, or party identification. The following plots the regression coefficients for a simple linear model where a 0/1 term for “voted in 2016” is the dependent variable, and voters’ age in years, sex, party identification, and public housing status are the independent variables. The coefficients are scaled so that their magnitudes are comparable.
Indeed, the correlation between voter turnout in 2016 and living in public housing is about six times the magnitude of the party identification effect. Here, party identification is measured as a 0/1 variable for whether or not the respondent was registered as a Democrat.
There are clear voter turnout deficits in public housing in New York City. Compared to the rest of the city and the rest of the state, voters in public housing are significantly less likely to have voted in 2016. If we value high turnout across the electorate, there is a clear need to improve turnout in this subset of the active voter population.
The endpoint for this API is https://geocoding.geo.census.gov/geocoder/locations/addressbatch/. The API takes in addresses stored as well-formatted strings, plus an argument for which Census “vintage” to use for the geocoder. Here, we used the 2010 Census vintage, the most recently available for the address locator API.
We counted voters as living in New York City if their address was located in any of the towns listed on the New York City Department of Planning list of “community names” or “neighborhood districts” in the city, of which there are fifty-nine.