As voters are inundated with campaign advertising and news coverage, can they learn key information about politicians through all the noise? Joshua Kalla tests hundreds of messages on thousands of voters in the 2020 presidential election, finding that the right information can persuade voters, especially with specific information about Biden. Kevin DeLuca finds that more high quality candidates, like those endorsed by newspapers, are still winning at high rates. Despite strong partisanship, voters do learn from campaigns and are willing to shift their votes in response to what they learn.

Guest: Joshua Kalla, Yale; Kevin DeLuca, Harvard

Study: “When and Why Are Campaigns’ Persuasive Effects Small?” and “Newspaper Endorsements, Candidate Quality, and Election Outcomes in the United States

Transcript

Matt Grossmann: When information about candidates persuades voters, this week on the Science of Politics. For the Niskanen Center, I’m Matt Grossman.

As voters are inundated with campaign advertising and news coverage of varying emphasis in quality, can they learn key information about politicians through all of the noise? Or does partisanship trump all? Meaning voters can’t make other distinctions based on the candidate’s qualities and positions. Despite the obstacles, voters do learn from campaign messaging and are willing to shift their votes in response to what they learn.

This week, I talk with Joshua Calla of Yale University about his New American Journal of Political Science article with David Brockman, When and Why are Campaigns Persuasive Effects Small? Testing hundreds of messages on thousands of voters in the 2020 presidential election, they find that voters can be persuaded, especially with specific information about Joe Biden.

I also talk to Kevin DeLuca of Harvard University about his new paper, Newspaper Endorsements, Candidate Quality and Election Outcomes in the United States. He finds that more high quality candidates like those endorsed by newspapers are still winning at high rates and getting more votes, but we don’t see it because of mostly uncompetitive elections. This week, we’re going to listen to each conversation in turn, starting with Kalla’s description of what they found.

Joshua Kalla: At a higher level what we were hoping to do is to understand why is it that sometimes persuasion works to persuade voters in partisan elections in American politics, but other times it does not? So David Brockman and I have a 2018 paper where we find, generally speaking, there are minimal effects of persuasion these types of partisan general elections with two exceptions. One is that in nonpartisan races, we see larger effects than in partisan races. And the other exception is that early in a campaign cycle, we tend to see partisan effects, or persuasive effects. But that these persuasive effects decay closer to election day. And can’t be replicated closer to election day even if you use the same exact persuasive messaging, the same exact type of outreach.

So, in this paper, what we’re trying to do is to unpack those two observations, those two puzzles to see why is it that sometimes you see these persuasive effects, but other times you don’t? And the kinds of context that you see them in seems fairly consistent. So, what we’re trying to do is to distinguish between one explanation that we offer in that paper and that other scholars have offered as well, which we refer to as partisan intoxication, which is just that voters in partisan elections resist information about politicians contrary to their partisan loyalties. And they’re not motivated to form accurate beliefs about candidates because of their partisan loyalties.

And that as you get closer to election day people’s partisan loyalties get more and more activated so you see an increase in this partisan intoxication so that’s one theory that we’re trying to distinguish from another, which is just a pure informational account, which is that voters have some set of priors about candidates. They know some set of facts about candidates. And as you get closer to an election, they know more and more about those candidates so any marginal piece of information is going to diminish. It’s going to swamped by people’s preexisting priors.

So both of those are explanations that fit the pattern of data that we see in that 2018 paper. But we’re trying to distinguish between those. So, specifically what we do in this study is we look at the 2020 presidential election and we conduct survey experiments with close to 100,000 people. And we randomly assign them to either receive pro or anti Trump and/or Biden messaging. So, there are four conditions where they can get pro-Biden messaging, pro-Trump messaging, anti-Biden messaging, or anti-Trump messaging, or a control condition with no messaging.

And what we know from our survey data and other survey data is just that voters going into the 2020 election knew less about Joe Biden than they did about Donald Trump. So, this is a nice case where we can vary the priors that people have coming into to this study. And we can also vary the informational content of these messages to try to distinguish between the partisan intoxication and the informational accounts.

And broadly, what we find is that persuasive messages are much more persuasive about Biden as opposed to Trump. And we find that there is partisan defection that’s taking place. So Republicans exposed to pro-Biden messages move towards Biden. It’s not only some sort of coming home cheer partisan base effect. And we also find that messages with more informational content, messages that are more specific tend to be more persuasive. So, putting together these patterns of findings, we argue that there is at least some place for this informational effect in American politics. We’re not saying that it’s everything, it’s not the only explanation for persuasion in American politics, but it’s definitely occurring. And it’s something that future research should continue to grapple with.

Matt Grossmann: So give us a sense of, I guess, the size of those effects and just the indicators that you’re using to find effects of those messaging experiments.Does the average message move voters and on what?

Joshua Kalla: Yeah so, the main thing that we’re looking at is vote margin. So, we ask a question to the effect of if the election were held today, would you vote for Democrat Joe Biden or Republican Donald Trump? And we’re looking to see how each of those messages, the pro-Trump/pro-Biden and anti-Trump/anti-Biden messages move respondents on that vote margin question relative to a control group. And what we’re seeing is that the Biden messages, both the pro and the anti, move voters by about four points compared to the Trump messages that move voters by a much smaller one and a half points on that vote margin question.

I don’t think the magnitude of effects are necessarily meaningful ’cause I think translating from that survey context to real world effects there’s probably some diminishment, some sort of adjustment that needs to take place there, but the relative magnitude of the effects is what I think is important where we’re finding that both positive and negative Biden messaging moves voters, more than positive and negative Trump messaging.

Matt Grossmann: So, you found that specific information was more influential and that Biden information was more influential. But this was one presidential election. How do you think that generalizes across different offices and candidates? Maybe on the one hand, this should be the least likely place to find effects compared to the average US election but, obviously, there’s lots of other differences between Biden and Trump and other candidates.

Joshua Kalla: Yeah, I think it’s a great question. It’s probably the question.

In the 2020 election voters just knew much more about Donald Trump coming into it than about Joe Biden. So, even though Joe Biden had been vice president for eight years under Barack Obama, he was still relatively unknown to many, many voters in the American electorate relative to Trump coming into the 2020 election, who had been in the media day in and day out for four years as president. So, it’s a nice context to test this idea of differential priors, or differential information about candidates.

And yes, it’s only one election. Trump and Biden are two unique characters. So, I think future work would be wise to continue investigating this in other elections, other offices, other countries to see whether or not these patterns of results hold up.

One thing that’s unique about our study is it’s being done with real world candidates. It’s often hard to find these types of settings with real world candidates. But I think to do this well, you want to focus on real world candidates where respondents and survey have actual priors about those candidates, where there is meaningful information about those candidates rather than purely hypothetical work.

Matt Grossmann: And how about the specific versus general finding? Obviously, this was… Well, I don’t think we’ve said you attributed the statements to the supporters of the opposing party. And so, that might be a particular kind of situation where people might expect more specifics. How do you think that finding would apply more broadly to what campaigns are doing?

Joshua Kalla: Yeah so, one thing that motivated this study was the observation leading into the 2020 election that a lot of the Democratic messaging tend to be anti-Trump and potentially very vague anti-Trump messaging. Talk about how Donald Trump might be bad for democracy without saying how he’s bad for democracy. What is he doing to violate a democratic norm? It was a interesting observation about the real world, that there was a big focus on this type of fairly vague attacks on Donald Trump.

So, we wanted to see whether or not those vague attacks were similarly effective as a more specific informational attack. And we found both in the Trump case, but even more strongly in the Biden case that specific information was more persuasive than vague information, which is very consistent within informational account of voter persuasion. And then, how it would generalize to other settings? Again, I think replication is the king here. But I think it seems like an argument that would hold up that if you’re trying to persuade someone to do X, the more specific your argument is about doing X, probably the more persuasive it’ll be. But I think further research in the political domain would be valuable.

Matt Grossmann: So, give us some more concrete examples of these messages that worked on Biden, either to increase or decrease support, and maybe anything that might be of next steps to differentiate beyond just specific general that might have stuck out among the most effective messages.

Joshua Kalla: Yeah so, the top three most persuasive pro-Biden messaging talked about COVID 19, his record on bipartisanship and his personal background. So, to read them, feel free to cut this out if it ends up being too long, but the most persuasive message was, “The coronavirus pandemic is ravaging America. Joe Biden has defended our country from dangerous viruses in the past. When he was vice president, he acted quickly to contain the Ebola virus and screen travelers from infected countries who were trying to enter America. This made sure that Ebola virus never spread in America like coronavirus has. This year, Joe Biden has also proposed fast action against coronavirus. He joined our national security leaders in calling for President Trump to use his authority as commander-in-chief to start manufacturing, medical supplies for doctors and nurses. Biden has also said that we need to seek out the recommendations of public health experts and allow them to speak honestly to the nation about the threat we face.”

And the second one was, “Joe Biden has a strong record of being able to work successfully with both Republicans and Democrats. As vice president, he helped pass banking reforms, criminal justice reforms and a weapons treaty with Russia, all of which passed with Republican votes. As president, he would have the skills and relationships to work with leaders of both parties and move America past divisive, partisan politics.”

And the third most effective was, “Joe Biden knows what it’s like for working class Americans because he was born and raised in a working class, American family. His family struggled when he was a child and, for several years, they had to live with their grandparents. Growing up in Scranton, Pennsylvania his father had a hard time finding work as factories closed and coal mines shuttered. His family eventually moved to Delaware for new opportunities where Biden now lives. He ran for US Senate for the first time to expand access to affordable healthcare to improve the lives of working class families like his. He understands what it’s like for hardworking Americans, because that’s who he is and promises to advocate for them every day.”

So rereading these and reflecting on these in advance of this interview, I think what stood out to me is all of these focus on Biden’s background and experience. Trying to paint a picture of him as, I think, a high quality candidate who has dealt with serious issues in the past, and has the experience to deal with them again in the future, and as someone who values and cares about bipartisanship. So, I think continuing to investigate the role of, I think, bipartisanship, and moderation and just candidate quality more broadly, I think, would be an interesting next step to see if there’s specific type of messaging, specific type of arguments or appeals that may be more persuasive than others.

Matt Grossmann: But to clarify, there were also negative messages that were quite effective as well against Biden. And they [inaudible 00:13:16]. Okay. [inaudible 00:13:20]. No, it’s no problem. But it doesn’t about necessarily what the top issue is, or anything along those lines. It just seems to be about the credibility of the message, is that…?

Joshua Kalla: Yeah, I mean, at the time, obviously COVID-19 was a highly salient issue. But bipartisanship not necessarily a salient issue, I think it’s a thing many Americans say in survey data they value. But it’s not one of those most important issues of the day that would show up in a Gallup type survey. And same with Biden’s background of growing up in Scranton Pennsylvania that’s like a pure candidate valence, candidate quality type appeal that really has minimal information about his policies, but tells you about who he is and what he values.

Matt Grossmann: So, you also found effects among Republicans that you say are contrary to kind of a partisan story. But, of course, there might be a specific context to Republicans who had concerns about Donald Trump, and might be willing to consider Joe Biden. Were there any other patterns as to who was influenced and what it might tell you about how this would translate into the real world?

Joshua Kalla: Yeah. So, one of the predictions of a partisan intoxication type argument would be that if you’re making a pro-Democratic that should mainly just move maybe Independents or Democrats to return home to the party base. It probably shouldn’t move Republicans to cross party basins. That would be contrary to someone’s partisan affiliation. So, I think it’s quite meaningful for our informational argument that we find that this pro-Biden appeal is able to move Republicans to vote for Joe Biden rather than to vote for Donald Trump.

Now, the 2020 election is somewhat unique ’cause there are large numbers of anti-Trump Republicans. But one thing we did find in further heterogeneity tests is that this pattern of finding holds across the strength of partisan identity. So, if we look at weak Republicans or lean Republicans compared to strong Republicans, we find fairly consistent effects of where that movement is coming from. So, it’s not just the weak or lean Republicans who might be the Republicans more likely to be anti-Trump where we find this effect. We find it across the strength of party affiliation.

I don’t think we had a specific pre-treatment measure of Trump support. Or, if we did, I just haven’t been able to look at it. But I think there could still be that concern that just because you’re a Republican, you might be anti-Trump, but it would be a strange case where someone is an anti-Trump Republican who would still have voted for Trump, but for this information appeal about Biden. So, maybe those people exist. But more likely than not those people already were not voting for Trump, or were already voting for Biden. So, I don’t think that’s the whole story that’s taking place here.

Matt Grossmann: And is there an implication about the effect of partisanship as well? That is that if information can move even counter partisans, maybe some of what looks like a partisan effect is actually kind of an accumulated information effect that voters have rather than just a social identity. What do you think?

Joshua Kalla: Yeah, I think that’s right. I think partisanship is a very important informational cue. So, knowing that Joe Biden is a Democrat and knowing that Donald Trump is a Republican tells you a lot about those two candidates. So, I think party affiliation, party identification, certainly, makes its way into an informational argument in the sense that it’s a cue, it’s a heuristic. And as the parties become more homogenized, both within Congress and in a kind of at cross levels as American politics becomes more nationalized among candidates, among political elites, I think that the informational content of that partisan cue probably becomes more valuable for voters across the board.

Matt Grossmann: So I feel like if Josh Kalla was a reviewer of this paper, rather than an author, you would ask about the sort of, well you’re interested in advertising and media communications, but what you’ve done is put something on a survey and then immediately ask someone who they’re gonna vote for. How much is that really going to translate into an advertising effect, or a media effect that matters on election day? What do you think, would you expect these to last or appear in the world and what kind of message volume or environment would be required for this kind of messaging to have an effect on actual voting?

Joshua Kalla: Yeah so, I wouldn’t literally interpret the point estimates in this paper as what we would expect to occur in the real world. I think what’s interesting is that we find this pattern of effects where some persuasive appeals are larger than others. So there is this relative difference across candidates and across specific versus vague informational appeal. So I think that that relative difference in this very artificial survey experimental context, I think is important learning from a political science, from a social science perspective.

It’s not that these survey experiments are just able to move everyone across the board. Where if you just teach someone a fact and ask them 10 seconds later, how they plan on voting, we’re not just seeing these massive effects from all informational appeals from all candidates. If we saw that, then maybe I’d be skeptical. But we do see these relative differences that are nicely predicted by this informational account of persuasion.

And then, second David and I are running field experiments that are trying to look at this informational argument where we’re trying to provide information about candidates, specifically about candidates’ policy positions to see how very neutral information changes how people vote in congressional elections. So we don’t have any findings we can share yet, but hopefully come springtime we’ll have a working paper. And so far things look fairly consistent with what we’re finding in this paper.

Matt Grossmann: What about going back to your previous work? Has this caused you to… You did a lot of communication messages via direct voter contact and advertising in the past. So, did it cause you to say the messages weren’t right before? Or do you still believe those findings? And do you think sort of everything would just go down in the real world?

Joshua Kalla: Yeah so, I mean, when we wrote that paper in 2018 on the minimal persuasive effects, one thing we wrote about in that paper and especially when political practitioners would ask us about this is it’s totally possible that the persuasive appeals that campaigns are making are not the best appeals out there. And I think there’s a really nice parallel to the voter turnout literature. So, if you look at Don Green and Alan Gerber’s first experiment from 1998 in New Haven on voter turnout mail, they found mail had essentially no effect on voter turnout. But it was only after they incorporated social pressure and other findings from social psychology and behavioral economics that they started to find that mail is an incredibly effective and cost effective way to increase voter turnout. So I think if you were to say in 2000, when they published their paper, mail is totally ineffective, never run mail, don’t do mail anymore you would’ve missed this whole universe of untested messaging, untested tactics that ended up being quite effective.

So, I think that’s what we said back in 2018. And that’s sort of what I continue to believe today that there are certainly untested arguments, untested informational appeals, untested types of persuasion that campaigns could try and could see if they are effective or not. I think it’s hard. I think persuasion remains difficult in American politics because partisan affiliation, partisan identity is just it’s a really strong informational cue. But I think it’s worth testing and it’s something David and I are still working on and we’ll have more to share soon.

Matt Grossmann: So although we have limited evidence of persuasion effects, we do have evidence that candidate characteristics do seem to matter, especially candidate quality, and to some extent moderation. And presumably, those have to be communicated to voters somehow, and they might be tied to this kind of research on what information can be communicated to voters. Obviously, you also did research on moderation affects in the 2020 election and comparing the democratic primary candidates. So, I guess, tie those together for me. And tell me if you think part of what the effects of moderate candidates that you found, or candidate quality that others have found might be related to actually communicating that information.

Joshua Kalla: Yeah I mean, I think voters and surveys tend to tell us that they favor high quality candidates, and they tend to favor moderate candidates, whatever moderation might mean, and it’s a fraught term. But whether that means just ideologically moderate, or holding popular policy positions, whatever it means voters tend to value candidates with experience or high quality and they tend to value candidate who have some type of moderation. But voters don’t necessarily know that information about any given candidate A, who is running in election. So I think there’s an important role for the candidate to teach voters that information about how they are high quality, or how they are moderate in whatever dimension they might be moderate.

And I think there’s an important role for candidates to work with the media to try to get that narrative out there and to do direct persuasion, direct outreach like what we’re looking at more in this paper. But I think both are important mechanisms because candidates might be the most high quality, the most moderate, the perfect candidate, but unless they teach that information to voters in some way, unless voters learn that information voters aren’t going to be able to act on it. I think it’s very consistent with what Lynn [inaudible 00:24:28] writes about in her book and what Andy Gelman and Gary King write about in their research about how a big function of campaigns is in taking this information that’s out there in the world and bringing that to voters so that voters have that information to use come election day.

Matt Grossmann: So we’re seeing a rise of amateur candidates without previous political experience, especially on the Republican side and trying to apply your work to 2022. I noticed that there are three Senate candidates who are known for football, a bestselling book and TV. So, their voters may know something about them, but it’s not really a political set of characteristics. So, how do you think that would apply? Those sound like candidates where persuasion could matter a lot, but voters might have a preexisting opinion from something like TV that they might have had with Trump as well. So, what do you think?

Joshua Kalla: Yeah, I haven’t been following the races too closely. I grew up in Pennsylvania, so I have a special affinity for the Pennsylvania Senate race. And I think it’s interesting that the Fetterman campaign has focused a lot of their effort on teaching voters one fact, which is that Dr. Oz does not live in Pennsylvania. That seems to be their thesis of the race is that if we can teach voters that we will be able to win. And I think it’ll be interesting to see.

And I just don’t know if there’s polling data on this, but as more and more Pennsylvanians get exposed to that argument, you would expect that the marginal effect of continuing to talk about that argument should diminish ’cause you’re no longer teaching voters new information and you would expect at a certain point for the Fetterman campaign to pivot to some other type of argument to focus some other factor about Oz or pro-Fetterman beyond the fact that Oz does not live in Pennsylvania. So, I think it’ll be interesting to see how that plays out as we get closer to election day. And as that fact might become more and more saturated within the electorate.

But yeah, I mean, think these candidates do need to give affirmative case for why they are running, and why they’re qualified because voters just know less about them. So, I think we would expect also to see a lot of room for pro new candidate persuasion in these types of races. I’m not sure what data is out there, but I think these are interesting races to look at because you have candidates who voters just know very little about.

Matt Grossmann: So, if I’m not mistaken, you don’t find an overall kind of positivity versus negativity effect, which is kind of an old question, I guess, as to what campaigns should do. But it seems to me that your practical advice this year would be that often both campaigns should be talking about the same candidate, even if that’s positive for one side and negative for the other. So for example, in Georgia, Herschel Walker is unknown politically. And so, both the Republican and Democratic campaign should be focused on trying to provide information about the one candidate. So, is that the true advice? And if so, why don’t we always seem to see candidates following it? That is the 2020 campaign that your study was about, it seems like both candidates were actually talking more about Trump.

Joshua Kalla: So, one observation and I don’t want to claim a causal effect of our paper, but when we released a working draft of our paper in June 2020, we did actually notice that the Democratic side started to do more pro-Biden argumentation after our paper came out than they had before. So, there was a bit of a shift on the Democratic side to maybe not exclusively anti-Trump argument to less anti-Trump argumentation and more pro-Biden argumentation. So that was an interesting thing to observe at the time.

And then, with the Georgia Senate race, I mean, I’m not sure how much voters know about Warnock given he’s only been in the Senate for two years and he hadn’t held public office before then, I believe. So he’s still a relative newcomer. But, relatively speaking, yes I assume voters know much, much more about Warnock than they do about Herschel Walker. So, given just those relative comparisons, yeah I think you would expect to see more anti-Walker argument from the Democratic side and more pro-Walker argument from the Republican side and very little focusing on Raphael Warnock.

I’m not sure if that’s the case in Georgia. I just haven’t been following what’s happening there. I think part of the challenge is in 2020, a lot of Democrats were lukewarm about Biden. So, I think there was a hesitancy to be so strongly pro-Biden. And I think a lot of candidate campaigns like talking about themselves. So, I think you probably do have a bias from campaigns themselves too, to run more campaign or candidate messaging than anti-opponent messaging even if our paper might argue they should be doing that.

Matt Grossmann: Can information lead voters to select high quality candidates? Kevin DeLuca says yes. And he began by explaining his project.

Kevin DeLuca: Well, I ask how much does candidate quality matter for election outcomes? I also examine whether the effects of quality have declined over time, given the rise of political polarization in recent times. To do this, I use newspaper endorsements as a measure of differences in candidate quality. And I have a large set of endorsements and I estimate the quality differences for thousands of elections between 1950 and 2020.

Kevin DeLuca: I found that on average, a one standard deviation increase in relative candidate quality will increase that candidate’s vote shares by about four percentage points. So, that kind of gives you a first pass estimate of the size of the effect here. So, one standard deviation, four point effect. I also, showed that contrary to this popular belief about the effect of polarization on quality. The effect of quality on vote shares does not decrease over this period. If anything, it increases a little bit.

However, I also show that over the same time period, the number of competitive elections is declining. And specifically the number of elections where this candidate quality effect could affect the final outcome that has declined. And so, I think this popular perception that quality doesn’t matter anymore, I think, that’s sort of right in the sense that it doesn’t determine the final outcome in a lot of elections. But the nuance here is that voters themselves still care about quality. It doesn’t seem to be an effect of the voters not voting for high quality candidates. It’s more of a systemic explanation.

Matt Grossmann: So let’s make it more concrete with some specific examples. So, people talk about Trump versus Clinton as being an inexperienced candidate and a non. But, obviously, you have all these newspaper endorsements to get a sort more precise estimate. So, talk through some examples that we might have heard of and what your measure shows versus others.

Kevin DeLuca: Yeah well, I can talk about the Trump and Clinton race. That’s one, I think, example that people want to hear about. So Trump hardly got any newspaper endorsements. He was the worst candidate since George McGovern in 1972, at least according to my measure. I do have him at a 0.63 standard deviation quality disadvantage, which predicts that he would run about 2.6 points behind whatever the national environment is that year.

In 2016, I have here that Republicans won 50.5% of the two party vote chair in the House while Trump won 48.9% of the popular vote, which is a 1.6 difference. So, actually the model in this case kind of overestimates how bad Trump is. And I think this is partly due to, we can talk about this in a little bit, but the newspapers themselves have, I think, a little bit of a pro kind of establishment bias. And Trump is very anti-establishment, anti-press in particular. And I think they couldn’t bear to endorse him, even if they’re very conservative.

It’s a bit of a trick answer, but anything within this one standard deviation effect, which is four points, I think is fair to say that quality could potentially affect that outcome.

One benchmark that I like to use in my head is that the average quality difference of people who win their elections is actually equivalent to a 1.6% percentage point boost in their vote shares. It’s one way to think of it is anything less than 1.6% is very easily flipped by candidate quality differences.

I think an important example is the Georgia Senate races, sorry, in 2020 and the runoffs in 2021 for both Warnock and Ossoff to win in essentially a 50/50 environment that kind of indicates to me that candidate quality tipped the scales here. In particular, this is also a case where the previous measures of candidate quality like [inaudible 00:34:43] and previous experience would’ve done really bad explaining the result because both Warnock and Ossoff were facing incumbents, neither of them had previous experience. And yet, it looks like candidate quality is what determined the final election outcome. So, I think that’s a good example of where this newspaper measure is kind of better than the previous measures.

Matt Grossmann: So, talk through that a little bit more. So, I know we usually use just have you previously held an elected office, maybe which office at the most. So your measures are related to those measures, I think, but what else are they capturing and how different are they?

Kevin DeLuca: Right so, in general, I’d say the newspaper based measure so it’s a pretty similar story, but it’s not exactly the same as using these previous measures.

So, one thing I do is I do a sort of simple text analysis where I just count the most common words used in newspaper endorsement articles. And I find that the top two things are experience and issues. So, in my mind, that kind of validates this model that I have in the back of my head, which is newspapers consider the issues. This is the partisan dimension of the election like what’s at stake in terms of policy. But they also consider experience, which is an indicator of candidate quality.

There are other interesting words, I think that show up. Things like leadership. Served is also a word that comes up a lot. And record talking about the record of each candidate. And these are things that are even harder to get objective measures of. There’s no objective measure of leadership, or of serving your constituency well. But newspapers, it’s indicated that they do consider these factors and that weighs into their endorsement. And so, that’s a little bit more of a sophisticated nuanced measure of quality. Whereas, these binary kind of indicator variables won’t be able to catch those nuances.

Matt Grossmann: So, help us understand how you could get an increasing effective candidate quality when we have elections that are just overwhelmingly partisan now and nationalized, and we get just very similar differences… Very, very similar partisan results across different offices. How can we be seeing an increasing candidate quality effect? What’s the logic there?

Kevin DeLuca: Well, the logic is that, I mean, I hypothesize in my paper, I can talk about this a little bit, that it’s totally possible that while most of the electorate and while elites are polarizing, while the parties are polarizing, there’s still a group of moderate, or cross pressured voters like you like policy A of the Democrats, but policy B of the Republicans, where candidate quality will ultimately kind of flip your vote. And that can be true even as polarization is increasing.

So, it’s not necessarily that the effects of polarization that everyone else finds are wrong. It’s just that the people who vote based on quality, there still might be enough of them for you to find this kind of four point effect. At the same time, I think this decrease in the number of competitive elections, you might call that part of the effects of polarization, but it’s more of a geographic sorting story where states just aren’t as competitive as they used to be. There’s a redistricting angle here where politicians don’t like to draw competitive districts. And so, that will also reduce the number of elections where quality might actually affect the final outcome. But all that can be true even while these kind of moderate, or cross pressure voters are still responding to candidate quality and voting more for the higher quality candidate.

Matt Grossmann: And how does this size compare to incumbency and is incumbency a signal of candidate quality?

Kevin DeLuca: Yeah, so incumbency is a signal of candidate quality. I find that in 80 to 85% of cases, the incumbent is the higher quality candidate. So, again, it’s not all the previous work on incumbency or previous experience also, I find that they’re a higher quality it’s not all that work is wrong. It’s just that the measure of incumbency, that can’t tell you whether in these 15% of cases where the incumbency is the worst candidate… Or the incumbent is the worst candidate incumbency won’t pick that up. It also won’t tell you whether an incumbent is a really, really good incumbent or just slightly better than their challenger. And so, using incumbency as a proxy for quality, I mean, it is a proxy for quality, but it just misses this kind of nuance in precision that the newspaper endorsements allow you to estimate.

Matt Grossmann: So, you also acknowledge that newspapers don’t just endorse on quality. One of the things they endorse is just the partisan or ideological spectrum. So, tell us about how you separated those two and what kind of a new measure of partisan bias of the newspapers that developed.

Kevin DeLuca: Right so in my model, I estimate the bias of the papers too, since you have to control for the bias of the papers ’cause that reveals information about the magnitude of the quality differences. So, the really basic idea is that higher quality candidates will get more endorsements. And if you see a very Republican leaning paper endorse a Democrat that’s a strong sign that Democrat is kind of significantly higher quality than the Republican in that race. And that’s kind of the logic of the model.

This is not really the focus of the paper, but I do find some interesting trends with the newspapers’ bias. So, over time, starting in the ’40s, ’50s into the ’60s, local newspapers are heavily pro-Republican. They are very likely to endorse Republican candidates. This has been documented in some other papers as well, but I find the same trends here where they start off Republican in that era. And then, by the 2000s, they’re pretty neutral maybe with a slight Democratic bias. So there’s a big kind of shift in the local news environment, at least in terms of their sort of partisanship.

There’s also an interesting kind of pro-incumbency bias, which rises and falls over time. So in the ’50s it’s sort of low, they don’t have a strong preference for incumbents. And that rises starting around the ’70s and then it kind of stays constant over time where they have a much stronger pro-incumbent bias.

And I think my… Sorry, my data’s a little noisy, but starting around 2008, 2010, there seems to be a decline in this pro-incumbent bias and this mirrors the general trends of incumbency on electoral performance over time. So, I think that’s interesting, something I should look into more across the sample and give some estimate of effect sizes here. So, across the sample being an incumbent makes you about 30 percentage points, more likely to get endorsed. Having previous and relevant experience also increases your probability of being endorsed by a similar magnitude. As does having a longer tenure of experience.

Being in a scandal makes you much less likely to be endorsed, not surprising, I think. And I haven’t really looked at how these vary over time, but I think that the fact that newspapers are more and less likely to endorse based on these different factors that we already think are signals of candidate quality that helps to back up this idea that the newspapers’ endorsements themselves are a signal of candidate quality.

Matt Grossmann: So, I know you’re using the newspapers mostly as an indicator of something that might be apparent outside of people actually reading the newspapers. But, of course, we’re talking at a time of huge decline in newspaper subscriptions and in the role of newspapers in politics. So, kind of tell us about how you’re thinking about the actual effect of newspapers themselves versus newspapers as just an indicator. And then, if it’s still gonna be relevant going forward.

Kevin DeLuca: So I have some work in progress as the last part of my dissertation, which is looking at what kind of voters respond to newspapers and whether that can explain some of these candidate quality effects that we’ve seen like whether newspapers are informing voters about quality and whether that is affecting their votes. It could be the endorsements per se, or it could also be just the general coverage of those papers circulating among people.

And I have a couple of different ways to look at this. I do find that more moderate and independent voters seem to be responding to quality more, which is consistent with my paper and other work. But I can’t rule out other sources of information about politicians that persuades voters like campaigns or just other news sources, TV news, rather than newspapers themselves. Or nowadays, news spread through social media might also be affecting those things.

And so, I guess I can’t really say whether newspapers themselves are causing this. I do find results that are consistent with it, but I can’t rule out that advertising is the main driver of learning about quality.

Matt Grossmann: And what about the trends though? I mean, we have all this evidence that people are differentiating less in favor of their local incumbents because of a decline in local media. So, that would seem to cut against this story that people would learn more about their individual candidates versus just voting along party nationalized [inaudible 00:45:58].

Kevin DeLuca: Yeah so it’s possible that as local news declines especially in the kind of the recent decade, there’s been a even more stark decline. I think it’s possible that we would see smaller effects of quality. Maybe people just learn less about their incumbents.

I’m not totally convinced that that’s true though necessarily, particularly if you think that this is explained by a smaller group of cross pressured voters that end up voting on quality, because they’re kind of at the end of the day in the middle of the two parties. It’s possible that they still learn about quality since it actually will affect their vote. And then they vote that way. But it could be that you could see, for example, in aggregate surveys that people tend to know less about their local incumbents, and know less about the quality, even though there’s still an effect of quality on vote shares.

Matt Grossmann: So, we are having some important trends in primaries that seem to be cutting against this. Amateur candidates without previously held elected office are winning more often kind of started in Republicans and moved to the Democrats. So would you expect candidate quality effects to be declining in primaries along with that? And if so, why aren’t parties selecting candidates that would do better in the general election?

Kevin DeLuca: Yeah so, I mean the previous work that I’ve read on this suggests that at least historically higher quality candidates did well in primaries. And I don’t have newspaper endorsements for primary elections so I can’t do the same sort of model here. And I don’t have endorsements for the most kind of recent… I don’t have enough endorsements for kind of the last decade to really estimate this. But I do think you’re right. And I think people are right in general that particularly on the Republican side, maybe starting with the Tea Party movement, you have these candidates who are maybe not high quality, according to their resumes, are winning elections more.

And I think there’s a little bit of a disconnect between the Republican primary electorate and the general election electorate, where they kind of care about other things, basically electing more ideological candidates, not really thinking about the general election strategy there. And so, I do think in terms of, if I could estimate this with my quality measure, I would see that lower quality candidates would win in primaries more often nowadays.

Now, there’s a little bit of an anti-establishment streak in American politics. And I think in recent times, it’s perhaps stronger than it had been. I worry a little bit because it’s not that the newspapers necessarily are picking up the ideal kind of… It’s not like they always make the right decision. So it could be that the electorate itself realizes, “Look, this guy doesn’t have the right resume, but he has some other feature that we want to take a risk on him and give him a chance. And that might be totally rational.

So, I just would hesitate to say that the electorate is making worse choices now. But at least in terms of this quality measure that I find Republicans, I think are having a little bit of a problem recruiting good candidates to run that are gonna be good in the general election that may or may not be the right choice for them. But that’s what I would assume that I find.

Matt Grossmann: And Mitch McConnell has sort of famously in this election, said that they’re suffering from low quality candidates for the US Senate races. I know you’re trying to estimate the bias of the newspapers, you probably have year effects. So, are you able to show when one party does have an overall candidate quality say across Senate elections or house elections, are you able to replicate the conventional wisdom that Republicans had a candidate quality disadvantage in like 2010 and 2012, maybe solved it in 2014 and might be seeing it again?

Kevin DeLuca: Yeah so, in theory, I could try to do that. I haven’t done it exactly. I mean, I do actually agree with Mitch McConnell on this point. It’s kind of a small part of the paper, but in the appendix, I kind of run this model separately for different types of races. So for US Senate, governor, other statewide races, US House, et cetera. And I find that for US Senate elections quality has some of the biggest effects there both in terms of vote shares, and also probability of winning on average.

My guess is that it’s because Senate races and other statewide races get a lot more attention than US House races. So, it’s really common for local newspapers to endorse for the governor, but they might not endorse in every House race. They might not ever even mention the state legislative races that are going on. And so, people just know more quality in these top of the ticket races.

It could also be that people care more about quality in different offices. So, maybe you care that your governor is running the government well. But you don’t really care if your state legislator is just 1 out of 100 giving votes every now and then, you don’t really care if they’re high quality or not. So it doesn’t affect your decision as much.

As of this year, the newspaper endorsements aren’t out yet. So I don’t have a estimate of the quality of those candidates. But just given what I know, I’d say, it could be a big problem for the Republicans. If you assume that in places like Arizona, Pennsylvania, Georgia, that the partisanship or the expected vote is close to 50/50 and if you add on what looks like maybe a slight edge for Republicans nationally, though maybe that’s going away a little bit then I think it’s very likely that these candidate quality differences, which are big in those races could totally flip the seats. Normally, I think Republicans shouldn’t be worried that they would win all of those seats. But I think given that they kind of ended up nominating candidates that are worse than the Democratic candidates it’s possible that they lose all those seats and lose control of the Senate because of it.

Matt Grossmann: So a paper that might be paired with this interview is about advertising effects. And the similarity that I saw was that they’re saying there’s very small effects, but that they can differentiate based on whether the advertising is actually providing information about politicians that voters did not previously know, and that the impact is gonna be on these same moderate or cross pressured voters in the middle, actually learning that new information. So, I guess, to what extent do you buy that we sort maybe underestimated voters here, and that there are still people who can be swayed and they basically are swayed by real information?

Kevin DeLuca: Right so, I think the more general point of my newspaper endorsement model, which is that voters can take into account the bias of the source of the information that they’re getting about the candidates. And they’ll kind of adjust to that. I think that kind of relates to this paper because if you see an ad by the Democratic campaign that just says, “Our guy is the best, vote for him.” Well, of course, they’re gonna say that it’s the Democratic campaign. Doesn’t really tell you new information. There’s nothing really to update as a rational voter thinking about quality.

But if the campaign, the ad actually points to something concrete that the voter didn’t know, like maybe the candidate’s previous experience doing something, that could actually change the information that the voter knows and update their beliefs about the differences in candidate quality, where you might see some of these more moderate voters change their vote because they realize, “Oh, this candidate, I didn’t realize he had all this experience and did a good job doing that thing,” or whatever, and then they could change their vote.

So, I think you’re right, that the general points are consistent with the model that I have and the findings that I have.

Matt Grossmann: So on the other side, I guess, to defend maybe the partisan voters or the other side here, I mean, are newspaper’s really right that if they generally prefer Republicans to Democrats, that they should not support Herschel Walker, just because he’s a bad candidate individually. Isn’t the most important thing who controls the Senate and therefore, everybody’s kind of hanging on to these details that don’t matter as much as the key detail of party and maybe the 80, 90% of voters are right and the newspapers are wrong?

Kevin DeLuca: Yeah. I mean that’s totally possible. And, like you said, it could depend on, is this a race where it’s gonna determine party control of the chamber, or if it’s a presidential race, then there’s a big value in having the same partisan side as your president if you care about policy. So, it’s possible that the people who are flipping their votes are just not paying attention to policy and they just vote for the charismatic guy who also happens to have experience. And so, maybe you think it’s not related to the actual performance of the candidates once they get into office.

I think newspapers, having read a lot of these and seeing what they talk about, I think they really are trying very hard to pick candidates who are gonna do a good job in the office, particularly for races, if it’s attorney general or something where it’s less about the policies that get implemented, but just being competent is really important. I think they really key in on those things. And so, even if you think that the moderate voters are not paying attention and not making the right rational kind of calculus, the fact that they still seem to be voting for the higher quality candidates, as long as you believe that newspapers are actually picking out higher quality people to govern then, I mean, that’s still good to have them around. Even if, it doesn’t fit these more ideal bottles where they’re super sophisticated and helping improve the quality of governance to just putting on charisma, or something it’s still correlated with what newspapers think is good governing, which is I think a good measure of a good prediction of how someone’s going to perform in office.

Matt Grossmann: There’s a lot more to learn.

The Science of Politics is available biweekly from the Niskanen Center. I’m your host, Matt Grossman.

If you liked this discussion, I recommend checking out these episodes: Can TV News Keep Politics Local? How Does the Public Move Right When Policy Moves Left? How Record Television Advertising is Shaping American Elections. Policymakers Follow Informed Expertise. And Does Nationalized Media Mean the Death of Local Politics?

Thanks to Joshua Kalla and Kevin DeLuca for joining me. Please check out When and Why are Campaigns Persuasive Effects Small. And Newspaper Endorsements, Candidate Quality and Election Outcomes in the United States. And then, listen in next time.

Photo Credit: iStock