ISPS Faculty Fellow Kevin DeLuca on Redistricting, Voting Rights, and Media

Authored By 
Rick Harrison
August 28, 2023

Kevin DeLuca on the ISPS patio in New Haven

Elections reveal more than who wins and who loses. They can tell us about who voted and why — which segments of society care about which issues and how leaders might respond in office or when campaigning in the next race.

But even though public records can reveal who registers and shows up to vote, the ballot remains secret. There is no simple way to determine how individuals or different categories of people voted without using surveys and extrapolating from public census data. Similar analyses are needed to draw voting districts that empower residents to elect representatives of their choice.

That’s where political scientists like Kevin DeLuca come in.

As a new resident faculty fellow with the Institution for Social and Policy Studies (ISPS) and a faculty affiliate in the Center for the Study of American Politics (CSAP), DeLuca focuses his research and teaching on American politics, elections and election law, media in politics, local newspapers, and political economy. He received his Ph.D. in political economy and government from the John F. Kennedy School of Government at Harvard University and his M.A. and B.A. in economics from the University of Texas at Austin.

We spoke recently about the latest statistical advancements in redistricting efforts, how political science can protect voting rights, and what archives of local newspapers can teach us about the role of media in politics today.

ISPS: You and your colleagues recently collected, cleaned, and standardized precinct-level election results from every available race above the very local level in almost every state for the last three national elections. This data include nearly every candidate for president, U.S. Congress, governor, or state legislator, and hundreds of thousands of precinct-level results for judicial races, other statewide races, and even local races and ballot initiatives. Why did you do this? What can this data be used for?

Kevin DeLuca: In the United States, statewide and electoral district results are easy to obtain. But data from the precincts — the smaller areas that basically define specific polling places — are reported differently by different states and without the same quality assurance. The data we gathered encompass 180,000 precincts across the country, amounting to 36 million unique candidate-precinct combinations. By collecting, cleaning, and standardizing precinct-level results and making them openly available to everyone, we can empower researchers to dig down in a more granular way and answer questions needed for better redistricting practices, to track the efficacy of public health initiatives, evaluate policing, model labor markets, analyze municipal spending, and more.

ISPS: You mentioned redistricting. Another recent project involved evaluating a relatively new statistical method to identify voter race, called Bayesian Improved Surname Geocoding (BISG). What is that, and why is this useful for redistricting?

KD: Basically, BISG is a way to predict someone’s race using their surname and where they live. For example, I might know John Smith lives in a particular zip code. So, I know his name, where he lives, and let’s say that 90% of people there are white. Census data can also tell us how many people with a specific name break into different racial categories. We learn that Smith is a common white surname. Altogether, the probability John Smith is white is pretty high in this situation, and by taking the information from his surname and zip code, we are more confident in our prediction than we would be if we had only used either his surname or zip code alone. We can apply this method to all the people in a geographic area — to all the voters in a voting precinct, for example — to give us a prediction of the overall racial composition of the electorate in that area.

ISPS: Why would you want to know this?

KD: In the context of redistricting and drawing majority-minority districts, state officials need to know the racial composition of a proposed voting precinct in order to estimate vote preferences across different racial groups. To abide by the Voting Rights Act when drawing such districts, you first have to show that the preferences of non-white voters are different than those of white voters in a particular area. But since we don’t have individual-level records of who people voted for, such preferences need to be inferred from precinct-level election results and the racial composition of precincts.

ISPS: That sounds difficult.

KD: There are complicated methods to do this, but the basic idea is that you look at voting behavior in precincts with mostly white residents and compare it to precincts that are mostly-non-white. If you see that, for example, Republicans get most of the votes in precincts that are mostly white and Democrats get most of the votes in precincts that are mostly Black or mostly Hispanic, then you can say that white voters there seem to prefer Republicans and that non-white voters seem to prefer Democrats.

ISPS: So where does BISG come in? Why not just use census data to figure out the racial composition of precincts?

KD: Census data is a good start, but it’s not perfect. The issue we are addressing by using BISG on voter files is that the overall census demographics of an area may not reflect the true on-the-ground voting power of different racial groups. Ultimately, we are trying to make sure that districts that are majority-minority are what we call “performing districts” under the Voting Rights Act. We want to make sure that residents who live there can elect a representative of their choice and not have their voting power diluted.

ISPS: How might that work? Diluted how?

KD: For example, you might have a congressional district that is 50% non-white based on the census.   Seems like a majority-minority district. But we don’t know the racial breakdown of who is actually a citizen there, who is actually registered to vote there, or who actually turns out to vote there. It could be the case that even though the district is 50% non-white, based on the census, once you take into account differences in citizenship, registration, and turnout rates, the district is actually a majority white district. Using BISG on a voter file helps us address this issue because we can see who is registered and, often, who turns out to vote, which gives us a better picture of the actual electorate in a geographic area.

ISPS: Why should electoral mapmakers consider registration and turnout rates across racial groups when redistricting? Some people might wonder why the government should help to increase representation for populations that are not harnessing the political power they already have.

KD: Section 2 of the Voting Rights Act, as written and interpreted by the courts, requires fair representation regardless of the reasons why minority groups outlined in the law might be unable to elect representatives of their choice. As we have seen prior to the original passing of this law in 1965 and through today, differences in registration and turnout rates can be caused by institutional discrimination and unfair voting practices.

ISPS: Why do researchers use simulations so often in redistricting research?

KD: One problem with studying redistricting and gerrymandering is that we don’t get to see every possible map. It’s just an impossible task to create every map and evaluate them. So, the solution is to simulate a potential distribution of maps using a computer.

ISPS: And what might you learn?

KD: A lot of times it’s easy to see the bias in maps once you have a distribution of possible maps. If we simulate, say, 1 million maps, we can look at the results and see, using some objective criteria, what is the most biased version. And then we can look at the real map created by the actual government officials and compare it to the series of simulated maps. Often in cases of suspected partisan gerrymandering, the real map will be among the most biased in a series of thousands or more simulations. And so, the legal argument is that this level of bias was not produced by chance, that it must have been intentional gerrymandering on the part of the mapmakers.

ISPS: How are you using local newspapers in your research?

KD: Historically, local newspapers were how most people received information about what was happening around them, especially about local issues. So they have always played a big role in the political process. Even today, I think a large amount of information is actually generated from local newspapers or local reporters, which often gets picked up by bigger outlets and TV to get wider notice.

ISPS: But you focus more on earlier times, right?

KD: Yes. There used to be so many papers in the country. Basically every city or town had one, and the larger cities often had many. Every editorial board had their own take on the news, and there was often variation in how national or statewide news got reported in local outlets. This variation provides an opportunity for researchers like me. For example, as something was happening in a state — say a gubernatorial election — we can see how the many different newspapers across the state covered it.

ISPS: What kinds of data are you collecting?

KD: One example is a database I’ve been building of local newspaper political endorsements of candidates and ballot propositions. Let’s say a paper in Jacksonville, Fla., endorses one candidate for governor, and a paper in Tampa endorses someone else. I can use this variation in endorsements along with historical newspaper circulation data and election results to determine if the local newspapers are informing voters, what they are informing them about, and how voters are reacting to newspaper coverage and their political endorsements. Are they voting differently based on the media they are exposed to, based on endorsements, and/or based on what we can infer about the candidates from the overall set of endorsements?

ISPS: What are you hoping to learn?

KD: I want to learn about the effect of media bias on voter behavior, election results, and political representation. The endorsement dataset can be used to calculate new measures of media bias to use in studies aimed at figuring that out. Current measures of media bias are hard to calculate for a lot of outlets and over long periods of time. These alternative measures often involve research assistants reading transcripts of shows over the course of a month or reading hundreds of editorial articles for a newspaper and judging their tone — methods that are somewhat subjective and can’t be scaled up easily. But historical endorsements from a local newspaper provide an explicit revelation of partisan preferences. And with online archives, I can look up what a paper published before an election and collect these endorsements. Then we can try to figure out if media bias affects voter behavior, how much of an effect it might have, and if this bias distorts public policy.

ISPS: What do you think the average voter should understand about their role in our current political system?

KD: They should vote and be an informed citizen. I know it can be hard. Even as someone who studies politics, I am not familiar with every issue and every stance a candidate has taken. There is a duty to be informed, but people should understand that they can’t know everything, and that’s OK. Also, it’s easy to get disillusioned. We need to stay vigilant, but maybe not get obsessed with the day-to-day political battles which the media tends to sensationalize. It is good to have some perspective; learning more about politics and political history can help with that. The issue over voting rights, for example, is older than even the country’s founding. That’s not to say it isn’t important — it’s super important. But it’s not something that will be solved in one election cycle, after you vote one time. At the end of the day, we need to all do our best to be informed, participate in the process, and know there is progress.