Who is going to turn out for a municipal election held in an off-year, in an off-brand month, in the immediate aftermath of the 2020 election, and under conditions of a pandemic? That is the question we seek to answer with Turnout Modeling.
For starters, and even before accounting for the new world of Ranked Choice Voting, a turnout model is necessary to begin approximating a Win number and various objectives along the way. This is not an exercise in who will be casting the votes, but in where votes in what quantity should be found. To us, there are two processes to turnout modeling, neither is perfect, and it is best to combine both.
First, we take in turnout figures at the Election District (ED) level for numerous election contests held in the area. We build our historical trend-lines at the Assembly District (AD)-level at first because EDs periodically change from cycle to cycle. This is working with tallies of votes that were counted, from certified election results.
Next we go into the Voter File, where voter history gives us a propensity of (a voter, a group of voters, a turf of voters) to participate in a given election or cycle. Though we do not know with certainty what ED these voters resided in at any time in the past, we do know where they are now and that is all that matters. We are building up counts of voters of a given propensity to vote, and can begin tallying them by our current districting at the ED level.
And then we bring in our Confounding Features, otherwise known as the political topography. For instance, a contest may feature an incumbent or not, and have strong endorsers or heavyweight cash–but maybe not. We also check trendlines on how local legislators in other lanes are performing relative to the tops of recent tickets, such as Biden and Cuomo, because this is where we begin to squeeze out ED-level insights that are current. In some parts of NYC, you are running against the Mayor more than in other parts, even within your own party. We have a sensitivity to this, but frankly, so does the data.
Our turnout models let you get past the hype of what your coalition ought to be, and into the specifics of where it will come from. When output on a map, our turnout models can help you figure out which neighborhoods to prioritize (and not) quickly.
Here are just four features of political topography which we consider when producing a local turnout model:
1 – The “attack” and “decay” as voters enter and leave the count between Primary and General elections of the Presidential, Gubernatorial, and Municipal cycles, when controlling for interest in the contests. For instance, the 2017 municipal election was very low interest, but the presence or clumping of people who turned out for it is important to our modeling effort. This is a window into local organizing strength and interest. This election will feature much more interest, with lots of open seats and a big ballot. All of these factors tug in different directions.
2 – The “relative performance” of various incumbent Assembly and Senate members, when compared to each other and also when compared to top-of-the-ticket interest such as Biden and Cuomo (or Sanders and Nixon). Where these leaders run strong, their apparatus will contribute to turnout. These factors help us “trade-weight” how we interpret some of the other measurements and derived statistics we are making. If an Assembly Member runs well ahead of Biden when they appear on the ballot at the same time, this is a good proxy for local organizing strength.
3 – The propensity–at the ED level–for voters to use mail-in/absentee ballots. This is the COVID-special, and our analysis of where these products are used as opposed to a heavier reliance on voting at the poll-site has proven terribly interesting (to us, anyway).
4 – Roll-off–We examine how far down the ticket voters tend to go in each area. This can be vital if your campaign is not at the top of the ticket. We measure how much resistance there is to voters traversing the ballot on your turf. This can be done by counting how many votes for {President, Senator, State Assembly, State Senate… or Mayor, Public Advocate, Comptroller, BP, DA, City Council,DL,…} are historically cast in the same places at the same time.
I believe that any good campaign should actually be looking to increase turnout over its own turnout model. If the turnout model itself is sound, your ability to increase turnout would be a measure of your campaign’s effectiveness. Also, if your opponents have sound turnout models but are not engaged in increasing turnout as well, you seize some advantage when you do it.