When Georgia earlier this month held its statewide primary elections, a report in The New York Times noted that these were marked by a “full-scale meltdown of new voting systems.” Serpentine lines appeared at several polling places, and many people left without voting because of the anticipated length of waiting times. How does the length of voting time affect the effective exercise of democratic rights? Does the relationship between resource disparity and voting behavior depend on the racial composition of voters or on party affiliation? Can the application of the principles of operations research help explain how to improve the voting process and strengthen democracy?
Gerard Cachon, vice dean of strategic initiatives and a professor of operations, information and decisions at Wharton, has investigated these questions and more in his research. In this interview, he discusses two papers about voting behavior in Florida and Georgia that he co-authored with Dawson Kaaua, a Ph.D. student at Wharton. An edited transcript of the conversation appears below.
Knowledge at Wharton: What sparked your interest in voting behavior?
Gerard Cachon: I normally study operations management, which to me is about resource allocation. Mostly I study it in the context of companies, dealing with questions such as where they put their production facilities; how many people they hire; whom they train, and so on. But as a citizen of the U.S., I have always been curious about what happens with voting. It is very much an operations question in addition to being an important political question. I thought, “Well, it’s time to take some of the thinking behind operations and apply it to this political domain.”
Knowledge at Wharton: Let us turn to the papers and start first with the Florida paper. Its title is, “Serving Democracy: Evidence of Voting Resource Disparity in Florida.” What were the questions you were trying to investigate, and why did you choose Florida as the place to tackle them?
Cachon: We tackled Florida because it has been in the news over the last 15 years as a state where long queues for voting have been an issue. There have been many reports that the queues have been concentrated in districts that predominantly have minority, non-white voters. There has been other political science research that has looked at waiting times and other biases in the voting process. Researchers have found in various cases, using different kinds of data, that non-whites end up waiting longer to vote than whites do. Let us begin with the premise that in a democracy, we would like to have voting time be roughly equal across all populations. I do not know of any person who would say, “A certain group of people should spend less time to vote than others.”
“Researchers have found in various cases, using different kinds of data, that non-whites end up waiting longer to vote than whites do.” –Gerard Cachon
If we assume that this is not something that should be as advantageous to one group or the other, finding evidence of racial disparity in voting times is concerning. Most studies in the past have just looked at race. The advantage of Florida was, it is a state that provides data both on party affiliation and racial composition — and they have been providing this data over a long period of time.
Dawson and I were curious, first, if we could confirm over a long period of time — rather than just one election – if there are biases in voting patterns. And are those biases racial, or are they due to party affiliation? In Florida, the state has been trending more towards the Republican party over time. So, if you are the party in power, you can imagine that somebody might say, “Well, I want to maintain my power. If I give more resources to districts that are more favorable to me and my party, they will get better service, and therefore are more likely to vote, and my opponent might suffer.” So, we were curious: Is it a racial effect, or is it a party power effect? Florida was the unique state that gave us the opportunity to look at that.
Knowledge at Wharton: These are critical questions in any democracy. Could you take us through some of your main findings? Also, what sets your research apart from the other studies you mentioned?
Cachon: For our Florida study, we had good data on the actual number of voting resources that were given by each county at each election. This is both an advantage and a disadvantage. The advantage is that we looked at the explicit decisions that parties were making, such as how many people to staff [a polling station], how many poll workers, and so on.
The downside is that we did not actually have voting time data — about how long people waited to vote. Many other studies looked at voting time data via surveys. There are always questions about whether people truly report the correct time on a survey, so there are limitations to survey data. The advantage of having the resource data is that it is specifically tied to the actual decisions made. How many resources does a Democratic county get? How many resources does a Republican county get? Do the resources go up or down as the county becomes more Democratic or more Republican? We were able to tackle that question. That is how our research is different than previous studies. We looked at devoting resources, and we looked at both party and race.
Knowledge at Wharton: You also used a simulation, didn’t you?
Cachon: Yes, we did. The simulation part is the following. Some scholars in political science have developed a pretty elaborate simulation that said, “If a precinct has this many registered voters, and we expect this is what the turnout is going to be, and it’s going to be a district that has a pattern of voting where people come in the morning and the evening at a certain percentage.” So, [there are] various characterizations of the district. They have developed a simulation that said, “This is our prediction for the waiting time of the voters if you have two poll workers versus three, or versus four poll workers.”
We were able to use the simulation that was developed by others to look at our results and say, “If a county actually did become, for example, 5% more Democratic, which would lead to a certain number of extra poll workers, or fewer poll workers in that case, what would be the implication in terms of waiting times?” We were able to provide an estimate of those waiting times as a function of the resource allocation.
“We discovered that small changes in resource allocation can have a big impact.” –Gerard Cachon
We discovered that small changes in resource allocation can have a big impact. Because a precinct might go from three poll workers to one poll workers or two poll workers – and that makes a substantial difference in the amount of time that voters have to wait in that district.
Knowledge at Wharton: What conclusions did you come to because of your study?
Cachon: We looked at each county over time. The advantage of doing our estimates within each county over time is that we are essentially controlling for differences in counties that we cannot observe. County A might have some unique attributes relative to County B. So, if we made comparisons between the two of them directly, we would not know if what we observed was because of the differences we see, or the differences that we do not see. The advantage is that we measured these things at County A across time, assuming that the unique characteristics of County A are constant across time.
A substantial amount of research shows us that there is a huge advantage to doing it that way. What we show is that as the county becomes more Democratic, as the percentage and proportion of Democratic voters in the county increases, they get fewer resources relative to a county that becomes more Republican over time. It seems that counties that are trending more Democratic in Florida are indeed getting worse service, relative to counties that are trending Republican at a faster rate.
Knowledge at Wharton: That is a critical finding. Based on your research, do you think that it is important to redesign the voting process in some way or to figure out differences in the way the voting process is staffed? What should be done so that democracy is better served?
Cachon: There is a lot of interest in things like gerrymandering and other manipulations of the voting system. And those are all great. What is interesting here, I think, is that little attention has been paid to voting resource allocation and how that differs across counties.
If we are willing to agree that all voters within a state should experience roughly the same amount of waiting time, you can imagine a state putting that into law. And second, [it may be important to] establish an independent commission that would monitor waiting times across the state and make appropriate resource adjustments if they notice that certain counties are getting unbalanced, or certain districts are getting unbalanced relative to others. I think there needs to be an explicit acknowledgement that this should not be a tool that a political party can use against opponents, along with the establishment of an independent commission to make sure that there is actual follow through on this issue.
Knowledge at Wharton: Let us turn now to the paper on Georgia. The title of that paper is, “Democracy on the Line: Polling Place Closures and the Impact on Wait Times in the 2016 Presidential Election in Georgia.” What was your research objective here? And again, why Georgia?
Cachon: A famous Supreme Court case, Shelby v. Holder, essentially eliminated the bite on the Voting Rights Act. The bite is that regions of the country that have had a history of voting discrimination had to apply to the Justice Department for any changes to their voting procedures. Georgia was entirely covered in this Voting Rights Act. The Supreme Court struck down that portion of the act, essentially claiming that the rule for how a region was included is no longer valid; as a result, until Congress came up with a new rule, the Voting Rights Act could no longer be applied to these regions. That gave Georgia the right or the ability to make changes to the state’s voting process without seeking the Justice Department’s approval.
The Director of Elections in Georgia sent out a letter to all the counties and districts of the state, which said effectively, “We no longer have to get Justice Department approval. We recommend that you save some money and close down some polling places.” So, Georgia became a state that immediately responded to the Shelby v. Holder decision by advocating polling place closures.
“It seems that counties that are trending more Democratic in Florida are indeed getting worse service, relative to counties that are trending Republican at a faster rate.” –Gerard Cachon
It became an obvious state to look at the impact of polling place closures on waiting time. And we could do this, because Shelby v. Holder happened in 2013-14 and before the 2016 election, they made the change. We could look at Georgia both before and just after the decision to look at the differences in waiting times in Georgia.
The problem is, of course, that other things could have changed between the pre-2016 elections and 2016 election. We needed a state that would provide a control. And then conveniently [we could look at] South Carolina, right next to it, which was also subject to this act. But South Carolina has a law that says, “You can’t just close down polling places without explicit approval.” There are explicit laws in South Carolina that prevent closures. South Carolina did not close polling places, but Georgia did. What we could do is what is called a difference in difference. We could look at the difference between what happened in South Carolina, and the difference that happened in Georgia. Any difference in those differences could be attributed to the actions that were taken in Georgia. Georgia was very convenient from an empirical point of view to get a nice test of this act’s effect on waiting times.
Knowledge at Wharton: Were there any similarities between the situations in Florida and Georgia?
Cachon: There were more differences between them than similarities. They both had to do with voting, and waiting times, and resources. But in Florida, we did it over multiple elections, whereas in Georgia we were really focused in on just before the decision and just after the decision. In Florida, we were looking at voting resource allocation across counties, across time. In Georgia, we were looking at the closing of polling places and what happens when you do that. It is a very different mechanism that influences waiting times in Florida and in Georgia.
Knowledge at Wharton: What did your research show about the impact on wait times when the polling places closed? Could you take us through some of your other key findings?
Cachon: In Georgia, when you close a polling place, what would you expect to happen to waiting times? There are a couple of possibilities. Because you close polling places, you are now making it less convenient for people to vote. So, you would think that fewer voters would go to the polls. If fewer voters go to the polls, you will have shorter queues, all else being equal. So, you may have the same amount of capacity and the same ability to process voters, but if you have fewer voters, you’re going to have fewer lines. I think that is intuitive.
So one of the effects is that it could suppress voter turnout, and therefore reduce waiting times. Now, the suppressing of voter turnout is not a desirable thing. It is mitigated or softened if you have shorter waiting times. However, there is another possibility. If you close many polling places and you just mothball their capacity, now that means all the people who were going to those voting places have to go to the more centralized ones. The total demand at the ones that remain goes up. If they do not have any more resources, now you have just expanded demand without expanding supply. You are going to end up with queues that go higher.
It really is an empirical puzzle as to whether this would lower waiting times or would raise them. It lowers them if the capacity of the polling places is just moved to the ones that remain open, which seems logical. It would increase waiting times if that capacity at the polling places that are closed is just lost.
“Compared to South Carolina, waiting times in Georgia went up substantially, relative to what we would have expected them to do if we didn’t have polling place closures.” –Gerard Cachon
So, what did we find? Compared to South Carolina, waiting times in Georgia went up substantially, relative to what we would have expected them to do if we didn’t have polling place closures. The specific numbers were like a 70% increase in the waiting time of voters in Georgia. So clearly, this action had an impact of making everybody wait longer to vote in Georgia. This is, of course, not necessarily desirable.
Knowledge at Wharton: One of the things that I found so interesting about your research is that voter suppression has been alleged in so many elections, even beyond Florida and Georgia. And this issue so timely because of what has happened in various state primary elections. CNN reported a few weeks ago that in states like Texas, people waited to vote for more than five hours. Did your research turn up evidence about the criticism of voter suppression being valid even beyond the states you studied? If so, what should be done?
Cachon: Well, I tip my hat to the people in Texas who were willing to wait five hours to vote. Because that is a commitment, and it is admirable. But there is no way in this society we should condone a system that requires people to wait that long. I can’t speak to exactly why that happened in Texas. I don’t have the data. Although that is clearly an opportunity for another study, to see what happened in Texas that caused that. And was that just a coincidence, or was it more systematic?
My research in Georgia was looking at one particular action, which was announced to the entire public. We know that this action was taken. And what were the specific implications of that action? Somebody who defended that action might say, “Well, they’re just trying to save money,” which is true. If you do not spend as much money on voting, you save money. Maybe people in Georgia do not really care that they have to wait 76% longer. I am skeptical of that. But if people feel like they would rather save the money and just wait, well that is fine as a society.
It gets a little devious if you are in charge of a state and you think your voters are less time-sensitive than your opposition voters. In other words, if there is an increase of let’s say 20 minutes to vote, your voters are so motivated, they’ll still come and vote, but your opponent’s voters, because they have day jobs or they can’t sacrifice the hour that they’re working, they just stop voting. You can say you are going off to save money. And you are. You are making everybody wait longer. But you know without explicitly stating it that this is giving you a political advantage. Again, I just don’t think that as a democratic society, we should be giving political parties advantages based on waiting times. We should be giving them advantages off the quality of their ideas, and their ability to motivate voters on their proposals. It is hard to argue that they should be getting advantages by making their opponents wait longer.
“I just don’t think that as a democratic society, we should be giving political parties advantages based on waiting times.” –Gerard Cachon
Knowledge at Wharton: What are the implications of your research for voters and citizens on the one hand, and political parties on the other hand?
Cachon: Well, from the voters’ point of view, it is critical that we establish confidence that there is some equity in the cost of voting across all citizens within a state. If you are in a county that shades either to one kind of group or another, however you want to define that group, and you discover that you all have to wait longer to vote than others, that is just not fair.
Providing information to the voters about the disparities in voting and making them salient to everybody, I think would motivate the political parties to say, “We are going to take this potential tool of competition off the table. Let us explicitly write legislation that says we are not going to accept big differences in waiting times across different groups, across our different states or across our country.” I think that increases the credibility of the process. And with democracies, credibility is so critical. You want people, after they vote, to feel that everybody has had equal opportunity to express their opinion through their vote. That is an essential part of our democracy.
Knowledge at Wharton: What surprised you most about your findings?
Cachon: In the Florida paper, what surprised me was that we found it was a party effect and not a racial effect. It is not that counties that are more non-white have to wait longer because they are non-white; it is because they tend to vote more Democratic in Florida. Another county that let us say is trending more Democratic because it has young, white professionals with education–another group that’s moving more in the Democratic party–they also have to wait longer because they’re Democratic.
I should also emphasize that that finding does not mean that there is no racial bias in voting. I am not making that claim. I want to be cautious about someone assuming I am saying there is no racial bias anywhere. All I am saying is that in this specific Florida study, what surprised me is that we were able to distinguish between a party effect and a racial effect — and it is a party effect more so than a racial effect. However, because non-whites are trending more towards the Democratic party, it can look as if it is a racial effect.
In Georgia, the thing that surprised me most was that we looked at it from the point of view of the entire state. We did not look at the specific areas that closed polls. We just looked at what happened to the entire state. And yet, we still found the effect. So, in the areas that did close polls, I suspect the effect is even stronger, mitigated by the fact that we are lumping them in with other areas that did not close. Just the fact that we found the result was surprising to me.
Knowledge at Wharton: What are the most important policy implications that follow from your research?
Cachon: There needs to be more consistency and monitoring of the actual voting methods at least within the state. You know, states have adopted different policies — voter ID, et cetera — that are controversial themselves. But even given those policies within the state, there should be more consistency within that state, across the different regions of the state. And I will just keep coming back to that. I think that there needs to be some kind of independent commission that says whatever decision we are going to make about how much people are going to wait, for example, in Georgia, we can invest more in resources and they’ll wait five minutes. We will invest less resources and they will wait 30 minutes. Whatever that decision is, it should be 30 minutes across everybody in the state, or five minutes across everybody in the state — and not that some get five minutes and others get 30 minutes or longer, as the case may be. My hope is that this research gets publicized enough that it gets people to think, there needs to be more oversight on this issue.
Knowledge at Wharton: Based on your research about voting behavior, what future questions would it be important to investigate?
Cachon: The data we are working with on voter waiting times is not as good as it could possibly be. The government now has a survey involving a couple of thousand people across the country, and they ask them how long it took for them to vote. We need to have, as a society, much better data on how long each voter waits at a particular precinct at a particular time of the day. With that kind of data, we would be able to do better planning on capacity allocation throughout the day. Like, there are some regions that have more spikes in demand at morning, noon, and evening, than others. We need to know that.
We need to be able to understand what waiting times are, without using self-reported data. Self-reported data is better than nothing. But people are notoriously not great at estimating the actual time it takes them to vote. And we now can collect data at a much finer grain. With that finer-grained data, we would be able to answer questions with much greater precision. And we could probably ask new questions. That is my hope — that some more resources go into better data collection so that we can tackle this problem with as good knowledge as possible.