A political scientist ponders the polls
Release Date: 10/20/2016
Click to display or to hide the script from Episode 22.
Bob Long: The 2016 presidential election has been characterized by some as a choice between the lesser of two evils. What alarms some political observers is that more than half of potential voters have an unfavorable view of both Donald Trump and of Hillary Clinton. It also seems like modern elections are becoming more difficult to digest for the average voter due to the sheer number of political polls and trying to figure out, "What do they all mean?" For example, it's not unusual each week to hear something from Quinnipiac or Monmouth or the NBC Wall Street Journal or Fox News, CNN, et cetera, et cetera. It doesn't help that the cable networks and many major papers or websites like to focus on the horserace aspect of political polling, especially when you consider the margin of error within these polls. I'm Bob Long, we welcome you to another edition of Stats and Stories. It's a program where we look at the statistics behind the stories and the stories behind the statistics. And as you might have guessed, we're going to talk today about polling and politics in 2016. Well joining me for Stats and Stories is our regular panelist, Miami University Statistics Department Chair John Bailer. And our special guest today is Miami University Political Science Professor Bryan Marshall. Brian's teaching and research focuses on Congress as well as the relationship between the congressional and executive branches and quantitative methods. He also has out, his most recent book was on decision making on the modern Supreme Court, analyzing major theories of judicial decision making, as well as how the President and Congress affect the court. Bryan, we welcome you to Stats and Stories today.
Bryan Marshall: Thank you, Bob.
Long: Well as a political scientist, I'm sure you're not different from many other political scientists who are about as confused as they've ever been this year. So let's just talk a little bit about the kinds of questions as a political scientist you like to see addressed when we're talking about the whole polling process in an election year like this.
Marshall: So I certainly think that it's important for viewers and educated citizens to have a sense of the importance of polls and what goes into them. And really recognizing they're just kind of a snapshot of what's going on and that's kind of the impulse that the media kind of gives them as well, the whole horserace scenario.
John Bailer: So you have the polls that are present for just deciding what the preferences currently for particular candidates in contested seats and contested elections. Are there other kinds of questions that political scientists will often be trying to address where they are collecting data? Polls are one form of collecting of data. What are other sources that you look at for your decisions?
Marshall: Sure, absolutely. When we're talking about elections and trying to model or predict elections, some very important kinds of information can go into those models. So for example, economic indicators. So often times you'll see political scientists use GDP in the second quarter or consumer sentiment measures around the second quarter as an indicator, kind of a structural indicator, to get a sense of the way the election's going to go. Other kinds of factors, data that we collect about elections too have to do with the political environment and the political context. So for example, public approval of the President is always one big indicator that's important in terms of modeling presidential elections. In this case, even though we won't have Obama in the White House, or he's not on the ticket, public approval numbers are still going to be indicative of how voters are going to either reward his party or punish his party. So kind of capturing the political context and the political environment along with economic indicators, all of that data in conjunction with polls is very valuable in helping us understand election outcomes.
Long: I kind of mentioned in the opening, one thing that struck me through the years was how back in the day you had Gallup, for example was one of the best know polling agencies, but today there's just so many different polls. Do you think that kind of adds to some of the confusion for voters? Because as you said, a lot times they hear or see the horserace numbers and they really don't know what does into each one of those polls.
Marshall: Absolutely, and you know with the snapshot the media typically gives viewers or readers of a poll, they typically don't give readers a sense of what the margin of error is. So when you see somebody going up three percent since the last poll but this poll has a much smaller n, number of observations, for example, that margin of error gets bigger. So all of the sudden people think there's some kind of trend when really there's no difference. So that's what I think we lose often times in terms of the media when they're giving us information about polls.
Bailer: So as a quick follow-up, with the different polls that are available, one of the things that seems to have been productive in recent elections have been methods that have done some aggregation. This kind of combination across these results, this ensemble that you have of polls that are being conducted and producing predictions. That's something that we saw with FiveThirtyEight.com very successfully has taken those polls and taken the different precision and perhaps some of the different bias that might be there. That seems to have been a revolutionary idea in terms of looking at polling results. Are there other things that you've seen that have that kind of revolutionary impact?
Marshall: Yeah, just in terms of figuring out the most valuable polls, the best indicators of various poll sources. As you mentioned at the outset, there's so many different polls out there now but really FiveThirtyEight and some of the other outfits that try to model elections, it really comes down to, how do we figure out which polls are the most accurate? Which ones should we include in our model? So on and so forth. So for example, there's been lots of work that's being done in terms of using past polling information to simulate how their recent polls are doing. You could use that information to kind of penalize one poll compared to another poll or weight one poll slightly more than another poll. So there's just all different ways of using that data and that information we have now to try to come up with really precise estimates about this poll or that poll. And there's things like house effects, too. So how different polling operations identify likely voters, that's all different. It kind of goes into who's been the most accurate in the past and how we should fit that information into our models.
Long: One other thing I wanted to kind of go off of that, when I go back to the 2016 primaries, you had this huge decision that had to be made that never had to be made before where you had seventeen or eighteen people running and who was going to get to be on the stage. And the same thing kind of applies in the presidential once you get through the primaries when you have independent candidates like you do this year that are polling, to some extent, eight, nine, ten percent like Libertarian candidates, but they're not quite good enough to be in with the Republican and Democratic candidates. Let's talk a little bit about that whole issue, because again I think it depends on what polls are you using to decide who gets actually to appear on the debate stage.
Marshall: No, exactly. And at the outset, I had mentioned other data we use and actually those things like economic indicators and indicators of the political environment. A lot of those folks that have used those kinds of data in the past kind of assumed away third parties. They're really predicting the two party vote as opposed to even incorporating an independent party, in this race especially. But we've had races over time where independents have played an important role but certainly this time around when you have voters that seem so frustrated with their respective party nominee, you're going to be looking at those potential choices that they're going to make and you have those independent candidates out there.
Long: You're listening to Stats and Stories where we talk about the statistics behind the stories and the stories behind the statistics. And our topic today, understanding the numerous political polls and everything that goes into the politics of major presidential election. I'm Bob Long and I'm joined by our regular panelist Miami University Statistics Department Chair John Bailer and our special guest today, Miami University Political Science Professor Bryan Marshall. I'll go to John Bailer for the next question.
Bailer: You know, the comment that you made earlier about the idea of past polling being informative and maybe being used to weight some of the results that we have from some present polls. That seems based on such a strong assumption that the past is going to be predictive of the present and we're living in an election time where it's been interesting that if you were to go back a year in the past, you may not have predicted that we would have this particular set of major party candidates running. So I'm just wondering about the validity of extrapolation of some of this past data to present circumstance. And can you think of other times in history where there's been this disconnect between the immediate past or recent past and this kind of change in terms of the current race.
Marshall: That's a really good question, John. In terms of my political memory, I don't remember a race like this because as you suggest here, both candidates are polling so negatively even for members of their own party that identify with a respective party. So we can have good data, we can have good polling data, we can have good indicators that have worked in the past, but those models also assume that voters are at least relatively happy, if not perfectly happy with their party's nominee. In this case, neither one of those things holds true. Neither for the Democrats or the Republicans.
Bailer: We you think about predictive models, predictive models are based on certain conditions and the application of these models in the future are assuming that the conditions are similar. If the conditions are broken then all bets are off and your faith in the models is gone.
Long: I think one thing that I'm kind of curious about because I think a lot of people would wonder about this. In the past it seems to me, correct me if I'm wrong, that a lot of polling was done because everyone had a phone in their house. Now today so many people have done away with their home phone or they can look on their and say, "Oh, who's calling me? I don't recognize this call. This looks like a political call. I want nothing to do with this." Does that make it more difficult to really try to find people because I know supposedly you're not supposed to call people on their cell phone but it seems to make the polling job so more difficult today?
Marshall: Absolutely Bob, I think we have really a saturated market now, especially in the United States when you think about all the polls. Before you go into this news article online, you have to click to answer this or that thing and the last thing we want to do when we get home from work is answer another poll about how we're going to vote. So I certainly think you have that saturation issue and if we look at the polls we're talking about here, the national and the state polls, those response rates are just terrible in terms of getting likely voters to respond to them. It's very difficult and when we do get a poll with a reasonable amount of respondents, what we try to do, because of those populations of people that we don't have very much information on, we try to map that out with census data. And then we have to worry about weighting and each outfit is going to weigh this differently. So we have to worry about those kinds of things. So there's lots of problems when you think about getting people to respond to a poll. The other thing too, is that it's almost an act of faith when we think about well we have this survey of a thousand respondents, what about the non-respondents? What about the people that turned us down? It's an act of faith that the people that turned us down look like the people that responded and we know that's not true. I mean we know, for example, at least in politics when one candidate's up for a week or has a really good week, the polls that are done, the likely voters of the opposite party are much less likely to respond to a poll because they're kind of down. Their candidate is going under water so they're much less likely to respond so we kind of have a cumulative affect with those polls. And that injects uncertainty about the polling results.
Bailer: Well it certainly makes you wonder what reported margin of error means because that's based on assuming that everything is tied up in the sampling variability, not in some of the non-sampling errors. And the non-representativeness of non-responders versus responders, that's a critical question in any kind of survey research. You can probably do some work, you can say okay, the average age of responders was this, non-responders was this. The percentage of certain demographic characteristics that you might compare. At some point, like you said, you just throw your hands up.
Marshall: Right, how do you capture the non-respondents excitement about politics? Or some of those key variables that really tell us if they are going to turn out. And that's the other question, going from likely voters and trying to predict who's going to turn out. That's huge, especially nowadays with such a polarized, very close competitive elections that we've had the last twenty years.
Long: Yeah, and I think the other thing that complicated this year is just the number of people who are, for example die hard Republican or die hard Democrat, who don't like their party's candidates and don't want to vote for either one of them. And that makes it hard, they may go to the polls but will they even bother to pull the lever, so to speak, for the presidential candidate?
Marshall: You know, absolutely Bob, and that's why you have, I certainly think this is true with the national Republican Party but also the Democratic Party to some extent, they try to give their base reasons to show up. You have lots of money being spent in various states in state campaigns trying to focus on the top of the ticket, other than the presidential seat. So make sure you vote for the Senate incumbent, the governor, what have you, reasons to show up for your party. On the other side of the equation, you have candidates trying to distance themselves, to some extent, from the presidential candidates. So again, giving voters a reason to show up other than the top of the ticket.
Long: You're listening to Stats and Stories and again we're talking about, as you can guess, presidential politics. Hopefully, we'll get into some other related things. You mentioned some of the other down ballot races which also to the parties are very critical this year as to who's going to control Congress and things like that. Our special guest today is Miami University Political Science Professor Bryan Marshall. He's done a lot of research that focuses on Congress and relationships between the executive and legislative branches and also has looked at the judicial branch and the decision making of the modern Supreme Court. Also joining me today, our regular panelist Miami University Statistics Department Chair John Bailer. John I'll go back to you for our next question.
Bailer: I was curious, do other countries have the same kind of passion about polling as us?
Marshall: You know, that's a really good question, John. I'm not a comparative politics expert, but what I do know is that the typical election season in other countries is far shorter and so there's just not as much emphasis on who's ahead and who's behind like we have in this country. So it is quite a bit different, I think.
Long: I went back and looked at some stories about presidential races, say in the early 1900s where back then the incumbents didn't even want to go out and campaign. They thought that was not the right thing to do.
Marshall: It wasn't statesmanlike behavior.
Long: Exactly, so obviously things have changed dramatically. I think one of the things that we haven't talked about today, we may have touched on it a little bit, but in today's world I think another thing that makes things so polarizing is the fact that there are so many different kinds of media outlets. Say if you're a conservative person, you're going to get your news from this website, this television network, et cetera. And if you're more liberal, you're going to do the opposite. How much more difficult do you think that makes it in this election cycle, not just for president but for other races, the fact that the people only hear what they want to hear? They're no longer getting an impartial view, if there ever was such a thing. It's so much more dramatic, I think today.
Marshall: Yeah. I think that's a really good point, Bob. If you look at past elections over time, post-WWII and on, there used to be a sizeable number of the electorate that would split ticket. That is they'd vote for president of one party and down the ballot maybe for a US Senator or their House member maybe vote for the other party. But in the last couple of cycles, we've seen that ticket splitting really, really shrank in terms of the frequency that it occurs, to less than ten percent now. It used to be well over twenty percent of the electorate would split ticket but now that has really, really shrunk. So a big explanation for that is partisanship, people have sorted themselves out more, ideologically and with respect to political parties. It used to be the case that you'd see lots of very conservative Democratic voters, but those folks have switched parties over time and the same thing with Republicans. There used to be a very strong base in the northeast. Northeastern Republican, for example, they tended to be more liberal or moderate voters. We used to have this kind of regional impact of ideology and party that we don't see much of anymore.
Bailer: So that would suggest that future elections and maybe even the current elections are going to be decided by smaller and smaller percentages of the population.
Marshall: Yeah, I think that's exactly right, John. Because when we look, the two parties are relatively evenly distributed for the most part and even people that say they're independents, what we know is they're really not independents. So the thirty-five or so percent of people that say they're independent in polls, they act like partisans. They're independent but they lean with one party or the other party so what we actually have in terms of true independents is just a very, very small percentage of people. That can be important, right? If those folks stay home, then what really matters is that you turn out your base. So you have these different campaign strategies during midterm election year versus presidential election year that can vary for that reason. Well because the independents are going to stay home anyway, well that means that we've got to turn out our base. So you have campaigns totally predicated on that goal, trying to make sure your base turns out.
Long: One issue that relates to that that I've noticed in this presidential election and you've kind of written about the Supreme Court which is the third branch that we haven't really talked about here, but there's this fear with one seat open and the fact that there may be other seats open in the next four years, that that could have a huge impact. Do you think that's also something being used on both sides, by Republicans and Democrats to try to motivate people? If you don't come out and vote, this is what's going to happen.
Marshall: Oh absolutely, especially both parties use the Supreme Court to try to get their base out. I think it's true when you think of maybe the regular voter might not pay a whole lot of attention to the Supreme Court or even Supreme Court rulings. When we look at public opinion and people's knowledge of the court it's very sparse, but certainly the base of both parties cares quite a bit about it. It's really interesting too, this year with the Republican strategy in the Senate to block the Garland nomination and to basically let that seat ride until after the presidential election. I think that strategy really came from Mitch McConnell who has been a very good kind of strategizer for the party and so when they decided to do that of course they didn't know who the nominee was going to be and now with Trump being the nominee, I think that decision looks even smarter, politically. Because now McConnell and Republicans have a reason for all those folks that wanted Marco Rubio or all those people who wanted to vote for Ted Cruz. All those disaffected Republican primary voters, now the Republican Party can say, "Hey, the Supreme Court lays in the balance, we have to show up." I think it makes that strategy look even wiser.
Bailer: One thing we ask all the guests that join us is what quantitative skills people should have to be political scientists? And also a related question to this is what should journalists know about reporting political science news better? So weigh in on that.
Marshall: Absolutely, I think one of my suggestions, I wouldn't call it a pet peeve because we all have pet peeves about other people's disciplines. But if I were to give some advice to journalists, I would say a little bit of historical context goes a long, long way. So I would want journalists to report a little bit more about historical context. We're so focused on the polls, on the snapshot right now, this debate that's going to happen between Hillary Clinton and Donald Trump, how about a little historical context. Do these debates matter, for an example, in elections? That would be one takeaway for journalists. What was the other question?
Bailer: It was related to what kind of training political scientists, what kind of quantitative skills should they possess to be high functioning political scientists?
Marshall: Certainly I've got my own bias, I'm an empiricist so I look at what very smart people like John do in stats and other disciplines and say hey, how can I use this for my data and what I do? But certainly I think students have to have at least a comfort with data. That's not to say you can't do qualitative research, too, and compare because that is an incredibly strong methodology. They're both very valuable in what we do. But just the plethora of data that we are awash with nowadays makes it that if you can at least be comfortable with data, to use data then it's really going to give you a lot of opportunities as a political science student.
Long: Well as a former journalist, I guess one thing that I sometimes worry about is this fixation that we have in America that you vote every four years and you vote for President and there's so much more at stake in the state legislature, in Congress. I guess I worry about the fact that we tend to ignore, I often wonder if you went out and asked a person on the street who is running for Congress or who is running for US Senate in their particular state, how many of them would even know that. Do you think that's a problem too that we don't give enough credit to the other branches of government and levels of government that are very, very important?
Marshall: I think that's exactly right Bob, especially in midterm elections. That's when we see turnout just plummet, thirty-five percent, forty percent, we can do better than that as the leading democracy of the world, I would hope. That is something that I worry about like you say, it's a presidential election year, we're going to spend $2 billion in campaign contributions and this is going to be a highly salient thing and that's it, we're going to go home. I certainly do worry about that. I think keeping politicians accountable is something we have to do daily, not just every four years.
Long: And it's also local elections, the odd year elections that are very important too. The turnout sometimes is even under thirty percent for those. One more question, John.
Bailer: One of the things when you raised the issue of historic context and we've talked about the idea of past conditions and models built on those. The idea that there's uncertainty associated with the estimates. It seems like there's all this nuance. How can you effectively communicate the nuances of the political process in the way, how do journalists do that? What are the suggestions you might have?
Marshall: You're exactly right there, too. In terms of uncertainty, we can have these polls, we can make predictions. There's something we call the margin of error but there's all kinds of different uncertainty or aspects that can go into that poll. I think just for journalists to give a sense of the different sources of uncertainty that we have would be valuable. So again, giving a little bit more context to the readers and the average citizen so they can understand better.
Long: Miami University Political Science Professor Bryan Marshall has been our special guest on Stats and Stories. Bryan, we thank you again very much. If you'd like to share your thoughts on our program, we remind you, you can send your emails to StatsandStories@MiamiOH.edu. Be sure to listen for future editions of Stats and Stories where we always try to focus on the statistics behind the stories and the stories behind the statistics.
Click to close the script.