If Voter Turnout Spikes, Our Voting Infrastructure Won’t Hold Up.

I love data spelunking, and two years ago on my last vacation, I found myself exploring the American Community Survey (ACS). I was fascinated with the level of information I was able to see, and I became a bit addicted to answering questions I never knew had publicly available answers.

“How old is the median hispanic male in North Hollywood?”

“What’s the average income of a renter in Santa Monica, west of Lincoln?”

“What cities most look like my home town, Granville, OH, based on race, income, travel patterns, education, and weather?”

To keep up my new found tradition of answering questions perhaps nobody else cares about, I decided to dig into US election data this year on my recent trip to Mexico. I wanted to know, once a voter was registered, how hard was it to vote?

I only began researching this question about 10 days ago, which is a woefully inadequate amount of time to really analyze the data, so I decided to narrow in my focus on Georgia. Not only was Georgia at the center of a lot of news, but it also had a lot of interesting characteristics.

• Georgia is a purple state, with deep blues in major metros like Atlanta, and deep reds in the suburbs and rural areas.
• Georgia has a large African American population, with over 52% of Atlanta being African American.
• Georgia uses the Accuvote system by Diebold, which is a touchscreen voting machine, so voters are bottlenecked behind the total number of voting machines.
• Georgia provides detailed information to the Election Administration Voting Survey, which allows me to dig into their statistics.

Waiting Lines

While I can’t answer the full scope of what makes it easy / hard to vote in Georgia, I did see an opportunity to analyze the capacity of each county in Georgia. Through the EAVS, we’re provided with a few key pieces of information about the 2016 elections in Georgia:

1. We get access to the number of registered voters by county
2. We can see voting patterns in each county based on things like early voting, absentee, mail-in, provisional ballots, and election day voting.
3. We can see sample ballots in different locations (thank you google!)

With all that data, we can start to paint a picture of what it’s like to vote in Georgia on November 6th, depending on three major factors:

1. How many people decide to vote (overall turnout)
2. How many people vote early / outside of a polling place on election day
3. How long it takes for each person to fill out their ballot

While these are not all the variables that go into creating waiting lines (machine malfunction, inadequate staffing, confusion, etc.), each of these variables when turned up, or down, can have massive impacts. Given that 2018 is looking to be a heavy midterm election, things may get dicey!

Running Simulations

To analyze the different permutations and combinations, I wrote a simple application, like one would in a college CS course, to simulate a massive polling location in each county. Each “election day simulation” for each county would look to measure the average time a voter took to complete their vote from the time they “arrive” at their polling location. If you read this before you go vote, you could start a timer from the moment you step into line, to the moment you submit your ballot.

Each simulation looks at different combination of voting length (how long it takes a person to fill out their ballot) mixed with voting turnout. I also assume that:

1. Voters randomly show up throughout the day, between 7AM and 7PM (12 hour voting window). There is some data that is public around voting patterns (people tend to vote early in the morning, at lunch, and after work), but I’m not sure how accurate it is. This even distribution is optimistic, so I’m actually under-calculating potential wait times with this methodology. My guess is, in reality, it is a multi-modal distribution.
2. Any voter in line at 7PM still gets to vote. I verified this against Georgia voting law.
3. There are a fixed number of voting machines for voters to use, based on EAVS data.
4. If a ballot takes N number of minutes, I assume that voters randomly can take up to +/- 20 % of that amount of time to vote. So, for example, if the average person takes 10 minutes to fill out their ballot, the fastest someone will go is 8 minutes, and the slowest is 12 minutes.

I looked at each combination from an average ballot time of 60 seconds, up to 20 minutes, at 30 second intervals. I also examined each of these voting durations against voter turnout numbers, in .25% increments from .25% up to 90%.

In total, I’ve run 1,051,476 simulations, and plan to run more to increase the fidelity of the data.

What it Looks Like

There are 159 counties in Georgia in total. Each county is unique in terms of the proportions of total voters, how many voting machines they operate, and how many voters actually vote on election day.

Chatham County Georgia

Let’s look at Chatham County Georgia, which has the fewest number of polling machines per election day voter, in the state, at 107.89 voters per machine.

Some quick stats on Chatham:

1. Savannah Georgia is the major city in Chatham County.
2. Chatham is a democratic leaning county, with Hillary Clinton receiving 57% of the vote in 2016.
3. Chatham is a majority minority district (just barely!)

We know that in 2016, this county had roughly 36.75% registered voter turnout on election day. We can then compare that against ballot length, to see how that impacts average voter time.

What we see above is that at the 7 minute mark, we see a significant uptick in queueing time, whereas at 6.5 minutes, we basically had none. At 7.5 per ballot cast, we’re looking at nearly an hour spent at the polls.

Fortunately for Georgia voters, there aren’t a ton of ballot questions, so 6–7 minutes seems doable. However, it’s easy to see how a year with a large influx of ballot questions or candidates could tip the scales dramatically and trigger a huge slow down.

We can also see what happens if voting patterns shift and voters stop voting early and start voting on election day. Below shows how wait times increase from 35% same day voters, to 45% same day voters, and finally 50% same day voters in Chatham County.

Now, these two scenarios aren’t likely. The peak same day turnout in Georgia in 2016 was only 41% of registered voters for a single county, but you can imagine how subtle changes to early voting access could dramatically tip the scales in a way election coordinators are not prepared for. As we read about polling centers closing, we can imagine a scenario where voters shift out from underneath the projections of planners, triggering a cascade of lines and disenfranchisement.

Finally, if we look at all of the counties in Georgia which have tipping points under 20 minutes (where a tipping point is the estimated time where voters have to start waiting 2 minutes or longer on average to vote), we see the following distribution of counties.

On the plus side, 95 counties could withstand long ballots without any issue. These 95 counties account for 3.8M of the registered voters in Georgia. On the downside, 2.5M registered voters could be a near term risk of increased polling times.

Counties I’ll Be Watching on Election Day 2018

I’ll be paying special attention to reports of long lines for Chatham, Cobb, and Gwinnett counties in Georgia. They account for 1.2M registered voters in Georgia, and all three could suffer from delays if election day turnout is higher than 2016, or voters take a particular long time on the voting machines.

TODOs

I’ve only scratched the surface so far on this research. I’ll be investigating more states which release this data (not everyone does), with a particular eye for states with longer ballots (for example, California’s ballot this year took me 30 minutes to complete). As states migrate to digital devices for voting, these types of long-ballots may need to go away as well; otherwise, states may be on the hook for significant purchases to appropriately distribute the machines throughout the state.

I’ll also be publishing the results of the simulations in an easy to use dashboard so you can play around with ballot length / voter turnout and see how it might impact your own county! I’ll also distribute the raw results so others can do their own spelunking.

Happy voting!