Episode 55: Abstract art or modern congressional districts- statistical thinking to fix gerrymandering Release Date: 4/26/2018
Guest: Gary King
Gary King (@kinggary) is the Albert J. Weatherhead III University Professor and Director of the Institute for Quantitative Social Science at Harvard University . King focuses on innovations that span the range from statistical theory to practical application. His methods are used extensively in many fields of academia, government, consulting, and private industry. He is a founder, and inventor of the original technology for, Learning Catalytics, Crimson Hexagon, Perusall, and other firms. He is a founder, and inventor of the original technology for, Learning Catalytics, Crimson Hexagon, Perusall, and OpenScholar
GARY KING: Thanks for having me.
Before we dive too deeply into the specifics of gerrymandering, could
you maybe explain what that is?
KING : So, the way elections work in the United States and many other countries is we first take the population, we divide them into legislative districts which are coherent, electoral units in which we have separate elections, so for legislators, members of Congress and the Senate and State House and state senate, and plenty of other legislatures in the United States, there's an election in each district. But if you redraw the district boundaries, if you change where the elections are held, you can change the outcomes of the elections in some cases and you can produce fair results and unfair results. And since you're changing the very nature of the definition of what democracy is in the United States, this technical thing of where the district boundaries are can have an enormous impact on the outcome of the elections.
John Bailer: So, when did gerrymandering first arise in the US?
KING: Oh, it probably first arises the first time we ever drew a district. The word gerrymandering comes from Governor Gary of Massachusetts. His house actually is the house that the president of Harvard lives in and and there was a district that looked like a salamander and a creative cartoonist called it a Garymander. We now call it a gerrymander.
BAILER: Great. Well, one aspect of trying to study something like this is trying to measure it, you know, how, how do you get some kind of quantification or some kind of signal that the Gerrymandering is occurring and what its impact might be?
KING: So, we have to figure out first of all why we should do this. And the reason that we're all talking about is not only because you and I think that this is a sometimes fair, sometimes unfair and we want to know which is which, but it's the case that the, that the United States Supreme Court declared in 1986 actually, that political gerrymandering is justiciable. Justiciable just means that it is allowed for the courts to look at a gerrymandering plan or excuse me, to look at redistricting plan and decide whether it unfairly favors one party or another. And the problem is that even though they declared political gerrymandering justiciable in 1986, it's never been "justished". That's, that's my word, by which I mean that, that the court has never actually thrown out a plan and at least this, the, the Supreme Court has never actually thrown out a plan because of partisan gerrymandering. Other courts have, other courts having to call it other things. It's litigated all the time. Pretty much in every state, every time there's this redistricting, but the Court has never done it formally. It's never. They've never adopted a standard. They've never said what the standard is that never said what the metrics are by which we could measure the standard. So, that's basically what we're talking about.
PENNINGTON: So, the Pennsylvania case that just has been in the news recently, where the State Supreme Court asked law makers to create fair legislative districts. There was dickering about it and so the Supreme Court had someone else draw these districts that they've handed to Pennsylvania. And now Pennsylvania, there are some GOP lawmakers who are appealing that and trying to get, I believe Alito to take a look and evaluate that. What do you expect to sort of happen in the case of this Pennsylvania redistricting and I cannot, I was going to parrot back your word back to you, but I've completely forgotten what that was. Justiciability? I've, I can't even say it.
KING: It's justiciability. It's justiciable. Yeah. So, in Pennsylvania, the State Supreme Court of Pennsylvania said that they're throwing out the Pennsylvania plan on the basis of the state constitution. It extremely clear that it was the state constitution. So, and they did that so that the United States Supreme Court would have no jurisdiction whatsoever. And so, so that basically is, is going the way it's going and the federal courts are staying out of the way because the states have a prerogative to make their own decisions and within constraints anyway run the elections the way they want. The real question is federal, federal law and what the Supreme Court will do, so we could talk about, for example, the Wisconsin case which has been brought to the court, has been argued in the court. The oral arguments have been held, but we haven't heard that decision yet if you like.
Richard CAMPBELL: So are there, are there states that have drawn on your work significantly and other states that have ignored it? How, how has this worked in terms of the symmetry that, that I think this, you've done this work 30 years ago, right?
KING: Yeah. So, I mentioned in 1986, the court declared political gerrymandering justiciable, actually the next year in 1987 that was my job talk at Harvard. And my job was the most important thing that happened. And my job talk was proposing a standard for fairness and legislative redistricting. It was from a paper that Robert Browning and I wrote, he was a buddy from graduate school now at Purdue. And we proposed the standard, the standard is partisan symmetry. I could describe how about I describe what partisan symmetry is and see whether your audience agrees with it. And we can, I can do it in the context of the Wisconsin case, which is up for discussion right now.
CAMPBELL : Perfect. That's great.
KING: OK, so, so in Wisconsin, the Republicans got 48 percent of the votes statewide. So, so of the two party vote they got 48 percent. OK. So, they've got 48 percent of the vote. How many seats did they get? They got 60 percent of the seats. So, with a minority of the votes that got a majority of the seats. Now wait a second, is that fair? Is that unfair? Well, seems a little weird. OK? But according to our standard, it's not necessarily fair or unfair. Here is what makes it unfair. OK. So, the results in which the, the Republicans got 48 percent of the vote in 60 percent of the seats. That was in 2012. In 2014, the Democrats got 48 percent of the vote. How many seats did they get? They got 36 percent of the seats. So, it's asymmetric, right? In different elections, under the same redistricting plan, each of the parties took a turn by coincidence at getting 48 percent of the votes and one party got 60 percent of the seats and the other got 36 percent of the seats. Moreover, we political scientists, know perfectly well that no matter, no matter how many more elections are held under that redistricting plan, even if the Democrats keep getting 48 percent of the seats, it's almost surely, that is likely, but not, not necessarily the case, but almost surely the case that they will continue to get way fewer than 60 percent of the seats. So that's partisan asymmetry. Symmetry would be if one party gets a certain percentage of the votes and they turn that into a particular percentage of the seats and the other party, if they were to get the same percentage of votes, also gets the same percentage of the seats. So that's, that was our idea of partisan symmetry. And you can see it very vividly in Wisconsin because nature, that is the voters of Wisconsin happened to give us these two examples of elections in which the vote for one party switched and again gave the stay basically the same voting percentage for the other party, and we could directly examine symmetry. It's often more difficult to examine symmetry because elections are not as nice to us political scientists, but we can now estimate whether a plan is symmetric or asymmetric, by looking at several elections by looking at one election or even by looking at the redistricting plan before any election has ever been held.
PENNINGTON: You're listening to Stats and Stories where we discuss the statistics behind the stories and the stories behind the statistics. Our guest today is Harvard University's Gary King talking about gerrymandering. Gary, when I was reading your bio, there is mention of something called ecological inference methods. So, could you take a moment to explain what ecological inference is in relation to the work that you're doing?
KING: Sure, in fact, we were talking about gerrymandering and it's actually used in gerrymandering cases in litigation, deviations from partisan symmetry that we were just talking about are used in almost every case by experts on both sides, and also ecological inference is used as well. So, what's ecological inference? It's using information about, from aggregate data to try to learn about individuals. So, in order to apply the voting rights act, which is an act in the United States, designed to protect minorities. When, when litigants go into court and they say, "it's not fair" this redistricting plan is not fair because it's been gerrymandered to hurt minorities. Then you can go to the courts and its long been justiciable that the courts could throw out a plan if it hurts minorities. However, how do you know whether minorities have been hurt? How do you know whether the voting rights act even applies? Well, it must be the case in order for the voting rights act to apply for minorities to vote in different ways than majorities, if African Americans are voting the same way as whites and Asians and Hispanics, then it doesn't really matter how you draw the districts, but if they're voting differently, then it may be a problem if they're redistricted in a way that it's impossible for them to get elected. So how do you figure out how people are elected in the United States where we have this rule, which is the secret ballot? We have these two rules, a secret ballot - you may not know how different races vote and we have the second role, which is in order to apply the voting rights act, you must know how people vote. So, the court is basically requiring statisticians to figure out how people vote, not individuals, but groups anyway, on the basis of aggregate data. So, we know in every legislative district and even every individual precinct the number of African Americans and the number of Whites and the number of Hispanics. And we know from electoral data in every individual precinct, the number of Democrats and the number of people voting Democratic and the number of people voting Republican. What we don't know, what no data to tell us for sure, is the number of African Americans who are voting for the Democrats and the number of Whites that are voting for the Democrats. And so what ecological inference can do is use information about the aggregate information, which is the percentage of Blacks voting and the percentage of African Americans in order to figure out the percentage of African Americans voting for the Democrats. And, in most redistricting cases the methods of ecological inference that I've come up with and some others have as well to they use this in order to figure out whether minorities are voting in similar ways as the majority.
BAILER: This idea of a partisan symmetry seems just really beautiful. I mean, it's a really clever idea. Very, very neat. How, how has it been? It sounds like it's being used a lot based on what you just described as it hits the courts.
KING: It's been, it's pretty much a, this is pretty much a consensus about partisan symmetry in the academic literature. The experts on both sides in most cases actually use it. The Supreme Court about a dozen years ago literally said, "hey, you know," they use different words, but they said roughly speaking, "hey, you know, if there was some academic who had some standard of partisan fairness out there that everybody agreed with, we'd like to hear about it." And so, you know, we sort of woke up and said, "hey, wait a minute, that's us." So, there was, there was another case that came a couple years later and so we filed an amicus brief in that case, the experts on both sides were using our partisan symmetry measure. As have, you know, as, as most of the most academics when they analyze the fairness of electoral plans. There have been a, it's a great example of statistics as well because this has been a long sequence of methods that have been improving from over the last 30 years. In which we can use whatever data is available to estimate partisan symmetry, very accurately, or deviations from partisan symmetry, very accurately.
BAILER : So, have you done a comparison across the states in terms, could you order the states in terms of their asymmetry in terms of their partisan asymmetry of their current districts?
KING: Yes, absolutely. We certainly could. I mean, I haven't done it right this second, but I've done it in the past and published articles about it. Yes, absolutely is, it depends who controls the redistricting process. I can give you, overall, we know if the Democrats control the redistricting process, it, it will probably benefit them more than if the Republicans control it. We know that if nobody controls it and we just let the districts go, and between the censuses from one census to the next, there's typically no redistricting. Although sometimes there are exceptions, and although the district lines remain the same, people move and die and become old enough to vote, which is basically equivalent to moving the lines. And sometimes the districts are gerrymandered by the people moving, and sometimes not. And so, this is, this is a basic feature of democracy that in electoral districts that you really need to pay attention to measure and correct.
CAMPBELL: So, I'm going to switch subjects here a little bit because we always like to ask our guest about their training and background. One of the things as the sort of humanities person in the group, you know, we often like to think that we're the most broadly trained and open to all kinds of different stuff, but looking at your record and the wide range of interests you have. And it does remind me of John. I mean, you get to, you get to work on a lot of projects that are very different as a, you know, with your statistical background. Can you talk a little bit about like how you choose the work you do today, which things are most interesting to you? And talk to a little bit about your training. If somebody, if one of our students wants to be you, how would somebody go about doing that? What would be the path?
KING: I really appreciate that, but in academia, if you did exactly what I did, you wouldn't do anywhere near as well because I've already done it. Well, the interesting thing is that statistics has had such an influence on the world that it is a subfield of almost all the others have almost all the other fields. So, my particular training is, is in political methodology, which is a sub field of political science, akin to psychometrics within psychology, econometrics within economics and on and on, there's a, probably 100 of these fields. So, when you have training in one of them and all, and each of these fields now talk to the other field. Sometimes that's called data science, sometimes it's machine learning. One day I wake up and I realized, Oh, I'm now doing big data according to the media. Lately I think we're doing AI is what is apparently my mom thinks that we do.
But the great thing is that what we do has an influence on people in so many different fields and that actually gives us a seat at the table and not anything. I mean, you know, you don't want me in the room when we're doing heart surgery but we actually have a skill that is useful in a vast array of areas. So, I love to get involved in lots of different areas and learn about different things and it's really only useful for me to be in the room if I have some comparative advantage. But it turns out statistics often gives us that. So, I've got to work on gerrymandering and real cases and seen how that works and maybe have a real influence on that part of the world. And I've figured out how to download all Chinese language social media posts before the Chinese government could censor them. And we reverse engineered what they were doing in China. We also discovered that the Chinese government was fabricating social media posts 450,000,000 of them a year and make it and writing them by hand and posting them in the name of ordinary people and we figured out what they were, who they were, what they were doing, why they were doing it. We've forecast election results and I've done things on a forecasting social security, done automated text analysis and there's no end of projects that you get to participate in or at least have the option to participate in if you have an expertise in this area. So, I highly recommend political methodology as the place to go learn statistics, but there are lots of other choices as well.
CAMPBELL: Very good
PENNINGTON: You're listening to Stats and Stories. And our discussion today focuses on the statistics of politics. Political reporting often relies on data to fuel stories. And so, Gary, I was wondering if you could talk a little bit about maybe some missteps that you've seen political reporters making in their use of data and reporting on politics, and what might they do to avoid those?
KING: Yeah. So, there's a couple of examples, like, actually I can give you lots of examples, more than you'd want to know. What often happens is journalists are exceptionally good at seeing one example of something and like, you know, other humans, we can spin an incredible theory out of one observation. My colleagues and I are good at spinning theories out of literally nothing at all. And so, paying attention to that and realizing that your theory is different than an actual conclusion is really important. So, I mentioned the censorship example. So, we had a new view of censorship because we were the first ones to be able to download tens of millions of Chinese language social media posts, each one of which we could see and read even though the Chinese people couldn't. And we had some tagged as censored and others tagged as not censored. And so, we knew the systematic system-wide patterns, but then a journalist would talk to one person and that person would say, "Hey, I wrote this post and this post was censored." And the journalists would say, "oh, that post was censored." This must be the reason and they spin up a big towel, but that's one post. If you think about the process of censorship, well, it's somebody's reading these posts, millions of posts and deciding which one stays up and which one, which one doesn't. There's error of course, in choosing which one is going to be going to be left up and which one's going to be put down. So, making a generalization on the basis of just one observation is not usually a good plan and that happens again and again, I, my guess is you will see this in the newspaper in the next few weeks, just because it was in the last few weeks ago, a few weeks before that. So the difference between looking at something at scale which requires statistics and something close up and personal, which is also a valuable important, and we need that as humans to understand things. That's a really big difference.
CAMPBELL: And this is useful because, you know, I'm teaching a journalism class right now to advanced students. And one of the things I always tell him to be careful about is if you're doing a profile on somebody and it's one person, make sure if there's sort of data involved that this person, how representative of the person you're interviewing is what's going on here and you're suggesting what I'm telling them to do, which is go start with the one person, but then talk about the context in which you're interviewing that person.
KING: Yeah, and as a journalist we have to recognize that it's not always possible for you to go do the big study, that's unreasonable. But if somebody has that can be really valuable. So, we did a, we did a study actually of journalism recently where we randomized, we got 48 media outlets to agree to let us experiment on them although we did refer to it in a different way.
PENNINGTON: That's actually very shocking, because newsrooms are very closed places.
So we ran the first large scale randomized experiment to figure out
what the effect of small media outlets was. And the effect actually is
quite enormous. It's really huge. The three small media outlets about
the size of like The Progressive with about 50,000
subscribers, publishing something at a randomly chosen time on a chosen
topic, on a topic that we choose, increases the discussion in social
media by about 63 percent. That's a, it's, that's a really big, big
amount. So, journalists that is you guys have a serious responsibility
and quite a big power because of it.
PENNINGTON: That study is really interesting because so much of the communication and media research has sort of suggested that it's really hard to get people to actually do anything when they're exposed to media. And this study that you and your coauthors produced shows that these local small outlets actually did kind of get people to do something, which I think is- that should challenge. I mean as, as someone who was trained in a media program, I feel like that should challenge some of the work that, that media researchers have been doing in the past.
KING: I think so, I think media researchers have had it very difficult, because media outlets for the most part are not trying to influence people. For the most part they're business is trying to follow people. If they didn't follow people, even if they didn't want to, they have to follow people because if they didn't they would go out of business and the media outlets that we're studying would be the only ones that actually did follow people. So yet as academics we happen to be interested in the effect of the media and journalists are interested in this as well, in the effect of the media on people. And so, the only way to really do that is with either a creative observational setting or by randomizing, and the observational studies, you never know, they could be right, they might not be right. They're some of the discipline's most creative studies, but they depend upon assumptions. And we could drop the assumptions with the randomization. But to get the randomization right, to convince 48 media outlets to do something like this, something they've never done before, something no journalist had ever agreed to before we had to really understand them and they had to really understand us. That study took us 5 years.
PENNINGTON: Oh yeah, I'm not surprised.
BAILER: And how did they react when they saw the results of this work?
KING: I don't know. Did you watch the Olympics when someone wins? I think they liked it because it really shows that individual journalists, you know, individual journalists, sometimes they, they go on a little bit about how, important a job they have. You know why? they're absolutely right. This is not just another job and individual journalists has quite a remarkable power and therefore a serious responsibility. It's also the case that full-time journalism jobs in this country have dropped by about a quarter. If that has had an effect on the ecosystem of articles that are published and on the opinions of the outlets at which journalists work, then that's also going to have a big effect on the national conversation and we know that that then has an effect on public policy. And so, these are not small things.
BAILER: So how have journalists done in covering your work on partisan symmetry, sort of bringing it back to kind of where we started in this conversation?
KING: There've been a number of articles about it, which I appreciate and you know, it's gone on for, you know, I didn't think that- I remember in 1987 I was sort of hoping somebody would like my job talk, and now as we wait for the Supreme Court to make this decision, I sort of feel like I'm in the same position. And so, journalists, some journalists have told that story. I mean, not me personally, but about the importance of the court stepping in and dealing with this problem. It's actually a much bigger problem now than it has ever been because of computers and statisticians and all of the power that we have since there's much more information and we're better at predicting, we can gerrymander much better and so the court really needs to step in to really make a difference that all they need to do is to say partisan symmetry can be one of the criteria that lower courts use to make decisions about whether to throw out redistricting plans.
PENNINGTON: One last question before we wrap up Gary, what political story are you keeping an eye on that you think journalists aren't covering as well as they could be going into the mid-term elections?
I have no idea. No, I think that's a great idea. I guess my answer
would be that we tend to focus on the highest politics, right? We tend
to focus on the big spectacular stories. Of course, no surprise, that's
what we're interested in. But, you know, politics actually is a
relatively small fraction of people's lives. It affects most of their
lives, but most of the time, like 95 percent of the time, most people
are not thinking about politics. They're thinking about getting their
kids to school and their job and their spouse and all kinds of other
things going on in life. So, focusing on what people actually do and
how they're spending their time and, the other dimensions of life, I
think that's really important. And I say that even as a political
scientist who is also interested in the big spectacular stories.
PENNINGTON: Great. Well thank you so much.
BAILER: Thank you Gary
CAMPBELL: Yes, thank you.
KING: Thanks a lot. I appreciate joining you.
PENNINGTON: Well, that's all the time we have for this episode of Stats and Stories. Stats and Stories is a partnership between Miami University's Departments of Statistics, and Media, Journalism and Film, and the American Statistical Association. You can follow us on Twitter or iTunes. If you'd like to share your thoughts on the program, send your email to firstname.lastname@example.org , and be sure to listen for future editions of Stats and Stories where we discussed the statistics behind the stories and the stories behind the statistics.
Click to close the script.