Loading summary
Transcript107 lines
- [00:00]
A
Hey there, listeners. Galen here. I just want to start off by saying that usually our Thursday episode is our paid episode, but today I have taken away the paywall. So this episode is fully free for everyone to listen to. I think it's just to the benefit of public discourse, public debate, all of that good civic stuff that more people can listen to it because it's well been the topic of debate on Twitter and substack and all of that, and I want to make it available. Nonetheless, I will give a plug if you are so inclined. If you'd like to support the GD Politics podcast adventure, go to GDPolitics.com and become a paid subscriber. I would so appreciate it. You'll get twice the number of episodes and you'll be able to join the paid subscriber chat and all that stuff. But most importantly, you will keep this podcast going. So with that, here's today's episode. Okay, couple things. First of all, are either of you offended by the term nerd?
- [00:59]
B
No, that's fine.
- [01:00]
A
I'm allowed to refer to you as nerds.
- [01:02]
B
That's fine.
- [01:03]
C
Yeah, of course it's accurate.
- [01:05]
A
Okay. All right. I'm getting a more enthusiastic endorsement of the word nerd from Laksha than I am from Elliot.
- [01:12]
B
Oh, it's just no to me. It's like obvious for me.
- [01:15]
A
Oh, why did I even ask?
- [01:17]
B
Oh, yeah.
- [01:29]
A
Hello and welcome to the GD Politics podcast. I'm Galen Drake. Do moderate candidates do better in elections? It's a question that has rocked the online world of election data nerds. In recent days. There's been hair pulling, locker stuffing, swirly giving. Sorry, I read that wrong. There has been online snark, substack posts and replies, competing Twitter and Blue sky threads, academic credential waiving, and accusations of bias. What started this whole thing is a little metric called war, which is oftentimes used in sports and means wins above replacement. Basically, how well does a particular politician perform in an election compared to how a generic candidate from their own party would have done? The folks at Split Ticket, helmed by Laksha Jain, have been using this metric to analyze electoral politics for a while and have found that the benefit to being a moderate is notable. From 2018 to 2024, according to their data, Blue Dog Democrats did about 5 percentage points better than progressive Democrats in House elections. The folks at Strength in Numbers, helmed by Elliot Morris, recently published their own version of war, showing a smaller benefit to political moderation, about a 1 to 1.5 percentage point benefit with significant uncertainty bans. Around those numbers, Elliot concluded in an article that moderation is overrated in electoral politics. This initial disagreement sparked a broader debate between other substackers, academics, and election wonks who took one side or another. Today, for the first time since this debate began, the two sides are sitting down together to hash it out. Here on the GD Politics podcast with me is Elliot Morris of Strength in Numbers and a former colleague of mine. One of the podcast Elliot.
- [03:15]
B
Hey, what's up, Galen?
- [03:16]
A
Also here with us is Laksha Jain, the director of political data at the Argument, which is the new home as of this week for Split Ticket. Laksha, welcome to the podcast.
- [03:26]
C
Thanks for having me.
- [03:27]
A
It's great to have both of you. And I just want to lay out some ground rules before we begin. So we're all on the same page. First of all, you are allowed to swear, maybe even encouraged to swear if you. If it moves you. There you go. While this will likely get wonky, where possible, please try to explain yourself in plain language. And lastly, if you are going to be mean, please also be funny. Although I. I assume that both of you will. Well, you model many things. I assume you will also model civil disagreement just perfectly. Are there any other stipulations that either of you would like to make before we begin?
- [04:04]
B
No. I like these ground rules. These are good.
- [04:05]
C
I like these.
- [04:06]
A
All right, Fabulous. Okay, let's start with laying out the stakes. Why do we care about the answer to this question? The value of moderation? This wins above replacement metric in general. And. And maybe if you want to elaborate. Are you coming to this question from the perspective of it's worthwhile to try to elect Democrats? It's more of an academic exercise. Do you want to elect Republicans? Do you want to elect a certain type of Democrat or Republican? Is this just a good governance argument? So, I guess. Launcher. Let's start with you, and you can lay out the stakes as you see them, and then Elliot will go to you.
- [04:44]
C
You know, I think that the importance of the debate on does moderation matter? Really matters, because there are a lot of people within the Democratic Party trying to chart out a future for a party that I think is probably at its nadir in the last 20, 30 years at minimum. And whether or not being moderate helps in elections is of immense consequence in deciding who to nominate, what rhetoric to take, what policy platforms to push, and everything related under the sun. And so, you know, whether the party says no, moderation is important. We need a big tent. Let's encourage all of these different people to run and let's also see if we can get closer to the median voter to try and win more votes. And is meaningfully, I think, an important question to consider because at the end of the day, elections are really close nowadays. And a difference of 1 to 2% is the difference between Kamala Harris being president and Donald Trump.
- [05:42]
A
Okay, Elliot?
- [05:43]
B
Yeah. I think we agree on the stakes, right? Like Laksh and I are both coming at this from a center left point of view. I think we both want Democrats broadly to win elections. We think that, like liberal democracy is good and that one side of the aisle is bad at preserving liberal democracies. So I think we can agree on the stakes. I would maybe come from a different angle here at Strength in Numbers. It's a collaborative war estimate that we launched. So I'm going to be saying we. We in this case includes myself and Mark Rica, who made the model for our war. He needs some credit, so I'm giving that to him. And we made our war estimate not because we wanted to enter the fray of the moderation wars, which is like, I guess we're here now. Well, here you are, Elliot, but because for entirely different reasons. And that was Split Ticket has good history of point estimates for their war. But we thought that we should introduce a quantification of uncertainty into the argument. So the only reason we ever really published this was because we wanted to put out a public, open source, transparent model that said, not here's the value of moderation, but here's what you can do with that information. Here's how much weight you can put on these conclusions. From the Bayesian statistical point of view. Elections are really complex. There's lots of reasons that elections might be close. Here's one reason why it might be close. And then people can take that information and very importantly, our uncertainty about how skilled politicians are, which is huge, by the way, the uncertainty is giant. And then do what they will with that information. So I think that's some necessary context, at least from our point of view. And I think Mark deserves that context.
- [07:23]
A
All right, I'm going to ask you to explain more how you modeled this in a second, but I just want to double check with you, Laksha, first. Elliot described the stakes for the both of you as coming from a center left position, wanting Democrats to get elected. Would you agree that that's sort of where your position in all of this is?
- [07:41]
C
Yeah, I mean, I think to be clear, Split Ticket is actually nonpartisan. We have Republicans on as well as Democrats. So our motivation in making the model was not just what helps Democrats win elections. My motivation in talking about it as extensively as I have is because I think understanding voter behavior is important. I am personally a Democrat. I want Democrats to understand this. But the motivation behind the model was actually primarily just to understand what wins elections on both sides. And hopefully, you know, selfishly, I was hoping maybe Democrats would want to pay attention to that and look at it. But that wasn't a motivation behind making the model.
- [08:21]
A
Okay, so let's get into making the model. You created yours first. Laksha. So let's start with you. Can you explain what you did and ultimately what you found?
- [08:30]
C
Yeah. So there are a few things that people use to evaluate over performance in general, when you're trying to evaluate whether a congressional candidate did better or worse than you would expect. People like to see, like how much did you overperform the President by now? The problem is that there are a few structural things that cause this over performance across the board. You know, regardless of who you are or what you do, we tend to notice certain boosts. So, for example, incumbents do better. This is well established in the political science literature that incumbents generally do better. Andrew Gelman has studied it, there's a lot of papers on it, so I won't rehash the literature. But generally, if you're an incumbent, you tend to do about 3 to 4% better than in an election than if you're running in an open seat. So right off the top of the bat, looking at whether or not someone did better or worse than Harris is slightly misleading because you haven't controlled for the fact that if they're an incumbent, they're actually just likely to do better simply because of the fact that, you know, voters don't like changing things up as much as people say they do. And this is an effect we've seen for decades. It's declined in importance, but it's still there. The second thing that we control for is this concept of lag partisanship, which you can think of as like a realignment type of variable almost. It's a bit simplistic to call it that, but I think it'll get the point across to the non wonky audiences. Here's what that really means. Not all Trump voters are created equal when you consider their probability of voting for the Democrat. Not all Harris voters are created equal when you consider their probability for voting for the Republican. It's not true that every Harris voter is going to vote for the Democrat down ballot and that their odds of voting Republican are 0. Percent. It's not true that every Trump voter in 24 is voting Republican down the ballot. Their odds of voting for the Democrat are zero percent. Here's why. If you voted for Donald Trump in 2016, 2020 and 2024, you think of yourself as a Republican, you are going to vote for Republicans up and down the ballot. But if you voted for Joe Biden in 2020 and then you voted for Donald Trump in 2024, you're actually more likely to consider yourself either a swing voter or really a Democrat for Trump. You think, hey, you know, I'm a Democrat, but I kind of like Donald Trump. I don't like my own party's nominee, but like, I'm a Democrat, I'm going to vote for Henry Cuellar, down ballot. I'm going to vote for aoc, down ballot. I'm going for Richie Torres because I know who these people are. I'm going to vote for them, but I'm voting for Trump. It turns out that whether or not you voted for The Democrat in 2020 is a very strong predictor of whether you're going to vote for the Democrat congressional nominee in 2024. And so we control for this by saying if you're in a district that has a lot of these ancestral Democrats, these new Trump voters, you should be doing better than Kamala Harris to begin with because you have a lot more Democrats in your district, even if those Democrats recently started voting for Trump. And we control for that as well as incumbency. And that along with like a couple of other minor factors which are small enough to where I don't, I won't get into them today, their demographics and spending, but they're very minor. But those are the two big things we control for in evaluating how well did someone really do relative to the presidential race and what was their over performance.
- [12:11]
A
Right. And so can you give us an example of somebody who overperforms or underperforms according to this metric?
- [12:18]
C
Yeah, sure. You know, people like Jared Goldin, you already know they're strong candidates. Our model says they're strong candidates. Our model says Jared golden did six points above replacement compared to a replacement level incumbent in Maine. Second, our model says that, you know, someone like Adam Gray did 5% better than replacement in margin over a challenger to John Duarte, over a replacement level challenger to Republican incumbent John Duarte. But our model says some things that people may find counterintuitive. For example, our model says that both AOC and Richie Taurus were war underperformers despite outrunning Kamala Harris by 7 and 8%, respectively. This is because both of their districts are two of the most rightward trending districts in the entire country. They swung right from 2016 by over 30 points. They swung right from 2020 by over 20 points. And so what happens is our model says, hey, hang on, you're incumbents here and you're in a super down ballot Democrat district. You should be outrunning Kamala Harris by much more than 6 or 7%. You're one point underperformers or something like that. And that's an example of what our model says where it controls for those factors. And it says, actually after looking at these factors, your overperformance wasn't as impressive as you think it was.
- [13:46]
A
Got it. Okay. So, and the conclusion of this model is that when you compare the Progressive caucus, so self identified progressives in the House with self identified moderates. So Blue Dog Democrats is the marker here for moderation. Those Blue Dog Democrats do about 5 percentage points better than the progressives.
- [14:06]
C
Yeah.
- [14:07]
A
Okay, Elliot, your turn. How did you calculate war? How did you do things maybe differently in particular in. In trying to figure out what. Whether moderation is valuable.
- [14:18]
B
Sure. So the way that our model works is going to be different than the way that Lux models work. Lux tries to predict the difference between a House candidate's performance in a given year, like 2024, and the President's performance in that year. And I think that the variables he picks are good. They're straightforward, they're founded in political science. They make sense in the findings in the context of that choice, the explanation of the difference between a House candidate and a presidential candidate's share of the vote or margin of victory. The conclusions from that make sense, but we're trying to predict something different. The Strength in Numbers model tries to predict the Democratic share or margin of the vote in an election, say 2024, with predictors that we know ahead of time and we can measure before the impact of the election are known. So that's stuff like the traditional Democratic lean in a district in the presidential election, specifically the past two cycles. Stuff like the demographics of a seat. So like what share of it is Hispanic or black, what's the average income, how old is it? And a couple of other factors specific to the election that we update over time. So that's stuff like whether or not a candidate is experienced or not, if they've held elected office before, if they're an incumbent, if their challenger is an incumbent or inexperienced. And very importantly, in our model, the share of the individual contributions for that whole district that are going to the Democratic candidate, We think that's a really strong signal. So this is very similar to the way that 538 and other predictive models for elections, actual election forecasts work, and that we're trying to design a model that predicts Democratic vote share. And the reason this is important to us is because in our war metric, we're not trying to explain the difference between a president and a House candidate. We're trying to explain the difference between a House candidate and a true replacement level House candidate. So we want to be able to make those predictions with the correlation between all these variables and the actual election result for the House candidate in those districts, for the candidate who actually won, and for some hypothetical replacement candidate. And the math on how you establish the hypothetical replacement, maybe we can get into. But for now, I think this big difference I've laid out is the big one in terms of the outcome variable, the thing we're trying to predict. There's just one more thing though, which is that our model is fully Bayesian. We wrote this model in a programming language called Stan. This is important to us because what this model does is it treats the parameters as uncertain. So when our model says the past presidential election is worth 1 percentage point of vote share for every 1 percentage point of difference in the House or the lacked presidential election, maybe for the next election it'll be only worth 0.9 or 0.95, or maybe it's worth even more. Maybe the relationship between how black a district is and the outcome in that district changes over time. And we just wanted to be like very, very careful about the certainty that we're making our claims with. So that's another big difference in how the models actually run.
- [17:34]
A
And so this ultimately gets you to a place where being a moderate is worth a point to a point and a half better than being a replacement level candidate and about two to two and a half points better than being, you know, like a progressive squad member or the likes. Right?
- [17:52]
B
Yeah. And this is the first point where you see some differences prop up. I would expect that in districts with more moderates, so those are districts around 50, 50 in typical presidential election, that voters in those districts would be more willing to vote for a moderate Democratic House candidate than a liberal presidential candidate such as Kamala Harris. And so I expect that that's why our estimate for the returns to moderation is lower, because we're not trying to explain the difference between moderate candidates at the House level and liberal candidates. Including Joe Biden was perceived as pretty liberal by the voters too. Not as liberal as Kamala Harris, but more liberal than the average Democratic presidential candidate. You know, the difference between that and predicting the traditional Democratic candidate that might win from that district is probably always going to be a bit, a bit more moderate. I think that's, I think that's what's going on under the hood.
- [18:43]
A
I have some follow up questions for both of you, but I want to take a step back for a second. Laksha, I assume that you have some disagreements with this process. Talk about it.
- [18:54]
C
Yeah, I think the biggest disagreement I have is that Elliot doesn't consider the presidential vote of 2024 in actually evaluating over or underperform. And to me that is really critical because everything is kind of like everything nowadays when you talk about under overperformance really does need to be considered in the context of what actually happened. Like what were the results, what they're trying to do, in my opinion, what Elliot and Mark are trying to do. It's basically like, you know, try to infer and analyze what. And Elliot, correct me if I'm wrong, what you guys are really trying to do is you're trying to consider like, and evaluate what treatment effects would really be by looking at pre election predictors specifically.
- [19:45]
B
Right?
- [19:45]
C
You're looking primarily and predominantly at pre election predictors. You're not considering the in sample data. You're not considering what actually happened in 2024. And so the example that I give is let's say that their model has a district where Biden got 82% in 2020. And so it believes that, you know, hey, in this district in 24, we would predict the incumbent House Democrat to get 83% in 2024 and we think a replacement would get 81%. So in this case their war would be D + 2. Now let's say in reality what happened was that Harris got 79% in 2024. That's 3% worse than Biden in vote share while the House Democrat gets 78%. So the strength in numbers approach doesn't know this. So it still thinks that the House Democrat has a positive or because in the simulations he outperformed a hypothetical replacement. But you don't know how to calibrate what actually happened against the real presidential number. And to me that's a mistake because you have to look at what really happened here and calibrate it against the actual results and do some type of post stratification there to actually figure out, hey, did they overperform did they underperform? And I, I think this is why we get such different results. Right. Like Elliot's model says Pramila Jayapal was an above replacement incumbent. And mine says that she's possibly one of the worst incumbents in the entire House when it comes to electoral over performance. And so. Or like Marionette Miller Meeks, where I think she was a really bad underperformer, Elliot says she was a slight over performer. And Adam Gray, same thing. I thought he was an over performer. Elliot's model thinks that he underperformed substantially. So these are just, you know, some of the differences we have, that's all.
- [21:31]
B
Yeah, let me respond to this. So it's not the case that our model is not trained on 2024 data. So our model is trained to predict the House election outcome in 2024. The split ticket approach, the characterization here is correct. Our model is not trained to predict the outcome of the presidential race in 2024. That's because what we care about in our approach, from our point of view, is House election candidates. Now, our model does know about the traditional systematic differences in districts between a presidential candidate and a House candidate. It does know, as I mentioned, that in moderate districts, according to the presidential race, you should tend to see outperformance among Democrats. That's all baked in into the historical relationships between these two variables. So just in terms of walking through why this calculation matters, for example, if you have a district that's traditionally 50, 50 in terms of presidential vote share, and Democratic candidates tend to do well in that district because voters are moderate and they like Democrats, whatever is a historically, you know, Democratic district somewhere in like the middle of Iowa or like Kentucky, whatever you want and you predict Democratic vote share, you're going to get the expected result. Democrats do well if instead you control for the fact that voters in those districts don't like Kamala Harris, and then you start treating the replacement level candidate in those districts as Kamala Harris, you are artificially inflating the difference between the actual Democrats running and the replacement level candidates in those districts. And then just the final thing here, while we're looking at differences between candidates, I think there's justifiable reasons for all of the differences here. I'm not really going to go through every single one because I think the most important point is that our estimates of war come with margins of error, huge ones. So for Marinette Miller Meeks, which Lux mentions, our estimate is that she does about two points better than a replacement level candidate, mainly because this is a district where traditionally Democrats do better than the presidential results indicate, and because her opponent Bohannon had a huge fundraising advantage that would have gone to the other candidates that would have won anyway. And we don't think that she would have done as well as Trump did in that district. The most important thing is that plus 2, our low point estimate is minus 13 and our high point estimate is plus 17. We really don't know how well she did because we don't have a whole lot of elections where she's running. And there's so much uncertainty in this whole process. So our approach here is that even differences of two or three points on war are not predictive over the long term. If you consider the uncertainty and in election outcomes and the processes that generate those election outcomes. The biggest source of uncertainty being how well Democrats do in the national popular vote.
- [24:37]
C
I think the thing here that there's two points. Number one, our metric does actually correlate cycle to cycle and decently so like about like a.05ish slope, which means basically half of the over performance of a candidate carries over from cycle to cycle, at least for incumbents. And this is important to me because we're firstly, war for us is a two way delta, right? It's the gap in candidate quality between the Republican and the Democrat in a race. With the case of an incumbent, usually you're facing a different opponent every year. So it makes sense that about half of your over performance carries over. But I think even more than that, Elliot, the thing, the issue that I have here is by that logic, if your point estimates are so wide, then you would say, well, we can't really infer anything, period about candidate over or underperformance. And that's not really true. Like, you know, you can see the types of candidates, the types of districts that the House Majority PAC targets, that the NRCC targets. You can see people have been making these models in private for a lot longer to figure out who's good and who's bad and who to target. You know, you can see, for instance, cycle over cycle, things like, hey, Jared Golden, Brian Fitzpatrick are two pretty good candidates. You can see, for example, that every single cycle Pramila Jayapal tends to underperform the Democratic presidential nominee. And so I do feel like if you just cite the high margins of error, it's basically just saying it's, it's just akin to throwing your hands up and saying like, oh, we don't know anything, therefore we can't conclude anything and so forth.
- [26:22]
B
I'll just like Crystallize two things here. You know, I think we fundamentally disagree about the baseline and we're just going to continue to do that. But from our perspective, a estimate of candidate skill at the House level should not be pulled around by whether or not voters in that district don't like Kamala Harris. There's systematic reasons in many of these districts where voters would not like Kamala Harris. They're moderates, she's not a moderate. But they would vote for a Democratic candidate in that district that's a moderate. And just I think we disagree about the right way to construct the model. And I think we've laid out the argument there, you know, on the uncertainty point. I think this is exactly the point that in political practitionery people have been making predictive models with like similar tools of linear regression for 30 years when we now have a lot better tools available to us that can quantify the uncertainty of the things we know about elections. And I think political practitionery is just ripe for over interpretation of small sample sizes. We only have 400 elections at the House level every year, and we only have 14, really in the training sample at the national level. And again, the national intercept is really what matters here. So yeah, that's exactly right. We think that the people that have been drawing conclusions from war estimates have been doing so with much too much certainty. And that's why we made the model.
- [27:55]
C
So I do think that we can draw inferences over consistent patterns that we see over the long term. Right. Like we see every single cycle, the Blue Dogs do better and they do better than the progressives. And it's not like there were just like only like four blue dogs. Like yes, there's 10 right now. There were 20 in 2018. But every single cycle they tend to do better. And if you look to see the progressives, they always actually underperform the New Democrats. The closest they got was in 2022 when they basically did the same. The progressives only underperformed by like 0.2%. But like, in general, it's just something we see on both sides of the aisle. We just consistently see that the more moderate members tend to do better. And no one seems to have a problem accepting this conclusion. On the Republican when it comes to discussing Republicans. Like no one here has challenged that Blake Masters was a bad candidate or to pick incumbent Marjorie Taylor Greene and Matt Gaetz are bad incumbents. And no one seems to challenge that Marjorie Taylor Greene is viewed as a bad incumbent precisely because most of America thinks she's way too extreme. So why are we having trouble accepting that, hey, the rules of politics, as they apply to one side, also do tend to apply to the other. And for all the people that see Marjorie's Taylor Greene is crazy. And there is a smaller but still substantial chunk of people that think Alexandria Ocasio Cortez is way too far to the left and therefore don't vote for her even though they're voting for Kamala Harris. Right. Or there's a lot of people that look to Pramila Jayapal, just as one example that I brought up again, and say, that's too much for me. Right. Person who objected to the 2016 results, person who repeatedly positions herself as like the forefront of the left movement in America in Congress. I think, like, it just seems to me inconsistent that there's so much controversy over the finding that Blue Dogs do better and people are like, oh, well, we can't conclude anything from that because the point estimates are so wide. And it's like, well, no one seems to have any problem concluding that the more extreme Republicans do worse. Why is it all of a sudden that we're concluding, oh, those rules don't work for Democrats?
- [30:01]
B
Yeah, I mean, I didn't state that those rules don't work for Democrats. In fact, our model does find a small over perform or, you know, small benefit to being moderate, about a point and a half. It's just that our model is saying if you ran another Democrat in that district, that Democrat, regarding controlling for all the other factors, would probably be the type of candidate that is similar to a Blue Dog because of the structure of that district, who the primary voters in that district would favor, the types of advantages that they would get from their state party, whatever the selection mechanism for that candidate would probably tend to push them towards a Blue Dog. So the WAR estimate that we have for current Blue Dogs is lower because we're not trying to explain the difference between a very moderate Blue Dog in that district and a presidential candidate running in the district. We're trying to explain the difference between what would it be basically to two Blue Dogs.
- [30:50]
A
I'm going to jump in here so I can bring in some critiques to both of your models. So, Elliot, a question that I have pertaining to a couple of points that you've made is one, if you're comparing the incumbent Blue Dog to a replacement Blue Dog, then it's already in some ways acknowledging that moderation helps because the districts that determine who wins control of the House are going to be those frontline districts that are more 50, 50 and so it seems like ultimately, whether it's the difference between Kamala Harris and a Blue Dog or the difference between one Blue dog and another Blue Dog, the conclusion is kind of the same, that moderation is helping here.
- [31:39]
B
Yeah. And like I said, we do find an effect for moderation. We agree on the direction of the effect. I just think, and maybe to get into a little bit of the interpretation here rather than the model differences, I think that the uncertainty here is really important. And I guess I can give you an example of this, which is, you know, we can construct some hypotheticals using the model and I can tell you why the uncertainty matters here. So, for example, if you are replacing all Democrats in the House with hypothetical moderates, in our case, we use Jared Goldin. He's not as moderate as Henry Cuellar, but we think he's closer to the moderate that Democrats would get everywhere than Henry Cuellar. So just like in this hypothetical world, you replace all Democrats and all their districts with Jared Goldin and you give Jared Goldin the benefit, or you give the little Jared Goldins the benefit of real Jared Goldin, which is about nine points of war. If you do that and then you ask yourself how many seats would flip in this scenario according to the uncertainty of doing a hypothetical exercise like this, then you increase the number of Democrats Democratic seats in the house by about 4, which, yes, in 2024, flips the outcome of the House. But because there's so much uncertainty, we're not able to say that this would definitively save the Democratic Party or what have you, because there's a lot of extra uncertainty that goes into these predictions. So, you know, according to our model, there'd still be a 95% or 90% confidence interval between Democrats winning 210 seats and 228. So, yeah, they could do better. They could also lose by five seats if they replace all their incumbents. For example, if the Jared golden doesn't do as well in fundraising in your weird hypothetical. So we consider this to be a best case scenario. And even that best case scenario is not all that convincing. So if people want to make their decisions from a like, purely point estimate, probabilistic maximization point of view, then they would switch to moderates for sure. Not debating that. But we think that the extra uncertainty here is worth taking into account in your simulations in how you make your campaign decisions. So you wouldn't pick a moderate Democrat that isn't as good as fundraising, for example, or isn't as good as a fit for the district, which is Sort of nebulous or doesn't have experience. And we think that those caveats are important.
- [34:20]
A
I want to dig a little bit deeper into those things in a second, but I think maybe one critique here would be there's a difference between sort of being humble in terms of what the data can tell us and incorporating so much uncertainty. Maybe because in an academic setting, that uncertainty makes sense. In a Bayesian setting, that uncertainty makes sense to the point where, like in actual practice, you sort of throw your hands up and say, oh, well, you know, the benefit of being a moderate could be a 10 point advantage or it could be an 8 point disadvantage. Is that too much uncertainty in a sense?
- [34:56]
C
No.
- [34:56]
B
I mean, we're not constructing the uncertainty. That's just the uncertainty. I mean, we've measured the historical ability for these variables that we've selected. Again, the partisanship of a seat, what the national environment is for the Democrats, whether one candidate is an incumbent, one is experienced or not, and the amount of money the Democratic candidate is running in those districts, and some demographic information, the ability of those variables to predict House election outcomes. And the uncertainty is just the uncertainty that exists conditional on all those variables. In our predictive model, we're not adding uncertainty. These models do predict election results very well. We get about a 2 percentage point standard deviation or uncertainty interval in our predictions. And our war metric is correlated at like 0.7 year to year with the difference in our predictions for the next year. So we think that this is a pretty strong model. So I think there's a difference between injecting uncertainty in our understanding of the world and shifting from a paradigm where basically everyone's overconfident to one where you acknowledge uncertainty. And I do acknowledge that there's a lot of uncertainty here, but I just think that that updating of our Bayesian priors, let's say, about uncertainty, is, is worth doing.
- [36:12]
A
Laksh, I've got some questions for you, but do you want to weigh in on any of that before I do to this point?
- [36:18]
C
Right. Like maybe you can say, yeah, that's the uncertainty that exists. I tend to think that in a lot of academia and a lot of probabilistic modeling settings, I call this paralysis by uncertainty, where people are afraid of looking at the data as it stands in front of them and saying, well, the uncertainty is so great that it could mean anything. And you kind of use that to hide behind, you know, hide behind conclusions that you don't necessarily want to acknowledge. Now, I'm not saying that's what Elliot is doing. But I am saying this is what I have seen a lot of people use Elliot's model to do. They've basically decided to conclude that, oh, well, this means there's no benefit to moderation. We should ignore it. And anyone who pushes, anyone who says that the directional inference points in favor of moderation is to be ignored. And I mean, I do think that, again, the benefit of moderation has been fading with time. This is like, undeniably true. You can just look at 2006 versus 2024. Not only are the people that are in the House in 2024 less moderate, like Jared Goldin wouldn't even be a centrist in the 2006 House Democratic caucus. Right. Not only are the House Democrats that are there in 2024 more liberal, the House Republicans more conservative, but also the benefits that they are getting for that moderation are also diminished. And this is not something that anyone disagrees on. And yet in that process, that diminishment, plus the bands of uncertainty that exist I think have been used in rhetoric to basically throw our hands up and say, well, yeah, these patterns may exist, but the uncertainty is so large we shouldn't even consider anything about it. And it's like, no, that, that this lines up very well with, like, decades of political science. It lines up very well with within survey evidence that we see. And I think we do ourselves a disservice by choosing to say, well, right now the uncertainty point estimates are so high, it could be anything. Well, okay, but the median outcome is not nothing.
- [38:18]
B
Yeah, I mean, we're not saying it could. It could be anything. We're saying it's probably something. But you want to acknowledge that there's a probability here. The reason you want to do that is because I think acknowledging the uncertainty gets us a lot closer to how campaigns are actually operating, which is not that they would always switch to a moderate because there's, according to the split ticket model, a 5 percentage point advantage of being a moderate. I don't think you guys would say that that's what campaigns do.
- [38:45]
C
No, that's also. That's not what the model really says. It says, on average, there's a difference between those. But it's not saying that if you just switch out someone with that policy, with that policy view of the Blue Dog, that they would do 5 percentage points better. It's saying, how much better would they do? Well, that's the thing. Okay, so you can make inferences that moderation helps. We can say that, on balance, over the course of the Last six years, the moderates have tended to do better. And I think there is value in acknowledging that you don't need to put a precise point estimate on the, on the degree of moderation when it's so hard to quantify ideology in a numerical sense beyond like caucus positionings. You don't necessarily need to say this is exactly how much better they would do to conclude that, hey, unbalanced, the evidence we see suggests it helps.
- [39:37]
B
Right, but campaigns are making prescriptions, right? So if you were making a prescription that switching to a moderate, a more moderate candidate, and Iowa one, whatever, would give a benefit to the Democrats in that candidate, what would you say that that benefit would be to, to the person? I mean, if you were making a prediction right now, would you say it would be five points or whatever? Or would you say it'd be five points, but it could be more and it could be less based on other factors?
- [40:01]
C
No, but that's not what the war model says, though. Firstly, if you're trying to decide whether or not to take on different positions, campaigns do issue polling, and not just issue polling, but also message testing for this exact reason. They see, like when I give a message, how much does it move a candidate's support up or down? And you evaluate based on that. You're not looking at war models for that. War models are generally use to evaluate after the fact. Okay, how well did these people do? What are the traits that the winners have in common? What lessons can we learn from it? And it helps in recruitment, it helps in analysis. You can say that if I, if I recruit someone like a Blue Dog, maybe you can say like the mean estimate is 5 is like that they would do 5% better in margin than a progressive. You can say the mean estimate is that, well, what types of districts would that put into play? You get a new district map and you say, okay, if I get that, what would my targets look like? That's the type of thing you would do. So if you want to say that like, well, on balance, these types of candidates tend to do better, you can do that while also acknowledging that, hey, there's a lot more that goes into this than just ideology. But ideology tends to help.
- [41:12]
B
Yeah, look, if you were making a prediction about how a candidate would do and you shift their ideology by five points, you would say, you know, they would do X percent better, but you wouldn't tell your client they're going to win for sure. Even if it was a marginal district, you'd say, oh, on average, historically, they've gained X Percentage points. But there's other sources of uncertainty in this election. That's what we're saying. We're saying you want to acknowledge the uncertainty because it unlocks a more nuanced view of the campaign. Instead of, for example, DCCC always prioritizing moderate candidates, they might prioritize a moderate candidate half the time, but other times they might pick someone who's a little more liberal but a much, much stronger fundraiser. By building a more comprehensive model looking at all the parameters that impact House election outcomes, looking at the uncertainty, then you unlock the ability to have a more nuanced campaign strategy. And that's basically our argument.
- [42:09]
A
Okay, I want to ask some questions based on critiques of locksha, your model, and you can incorporate responses to anything Eliot just said as well. So Adam Bonica, a political science professor at Stanford, and Jake Grumback, a public policy professor at UC Berkeley, wrote a post on Substack accusing you laksha of being biased in favor of moderates. In how you calculate wins above replacement, they take issue with how you calculate the baseline against which you compare any given candidate. We got into that a little bit, but you're welcome to respond here to their critiques as well. They also say, you know, correlation is not causation, that we don't know if these lawmakers are doing better because they're moderates or if it's for some other reason. The floor is yours to respond to. And if you'd like to flesh out any of their critiques further in your response, please feel free. But, yeah, I'm curious for your response to them.
- [43:05]
C
Look, I have a lot less patience for Bonica and Grumbach than I do for Elliot. Elliot, I think at the heart of it at least, understands mathematics. Now, I disagree with his inferences. I disagree with the way that the model was constructed. Those are fundamental disagreements we have. I disagree with some of the findings, and I happen to think ours are better. But functionally, what Elliot says about, like, the findings in moderation are not actually all that different from what I say. Will disagree a bit on the importance on just the way it should be used. But functionally, if you ask Both of us 10 questions about the importance of moderation, I don't think you'd get too many different answers out of those 10.
- [43:46]
A
Okay. Those are coming next, by the way.
- [43:48]
C
Okay, yeah, I don't think we disagree that much with Grumbach and Bonica. To me, it is a fundamental misunderstanding of basic elementary political science. Reading that critique, it starts by trying to assert a call to Authority. The authors of this piece are political science. Who gives a.
- [44:10]
A
All right, we got one swear word.
- [44:13]
C
Who cares if you are.
- [44:14]
B
No, I swear. I swear at the opening.
- [44:16]
A
Okay, all right, all right.
- [44:17]
C
Argue for your piece on the merits, ok? Don't give this bull appeal to authority background. The second thing is the entire piece is basically just one way of saying we don't like the fact that moderates do better in split tickets work, therefore they are hacks and you shouldn't listen to them. But like the variables that we control for. And we were very detailed in laying out what we did in an email to Jake, which then we ended up publishing all of that fleshed out in an article really we were very thorough about and in explaining what we controlled for and what we didn't. Their argument is, is basically that the variables we chose which were really just lagged partisanship and incumbency. They're saying those variables are uncharitable to progressives and that a really good model would be unbiased in ideology. That's what they said. They said a good model should not have any correlation between ideology and war. That is preposterous. I mean that is something where you were trying to isolate the effects of. You really should be like you should be trying to understand what types of candidates do better. You shouldn't be controlling for ideology in a model, nor should you try to hack your model to make candidates of a certain ideological wing look better. That is basically just reverse engineering to fit the conclusions you want. And in not accepting that you need a control for incumbency or lack partisanship, they're basically defying decades of political science research, including by the way, Adam Bonica's own work, which where he saw a study that said moderates did better and he said this isn't valid. You didn't control for live partisanship. And hey, you know what? That was actually correct. Bonica was right that you need to control for lag partisanship back then. Why does he object to it now? Now that the metric we use in lack partisanship suggests that progressives are underperforming. So this to me is just crazy.
- [46:14]
A
Elliot, do you agree with any of Bonica and Grumbach's assessment?
- [46:20]
B
So I have not seen the methods spec for their model. Maybe it's in their article and I.
- [46:27]
A
You haven't seen the tweet, Elliot?
- [46:29]
B
I have not seen the regression, so I don't know how much weight to put on that. I don't know what they're controlling for or not controlling for in Other words, I have seen that their output correlates well with candidate over performance in an out of sample test, which is great. I'm a big proponent of testing your model's predictions on unseen data. And so I would give it some credit, but I'm not going to defend them in the war debate. The strength in numbers WAR estimate did come out before all of the mean tweets. And again, I think we, we published this from the point of view of like wanting to move the WAR conversation to an open source, one that acknowledges uncertainty. So what I'll say on that front is to the extent that the war debate mid split ticket release their model spec publicly, I think that's a positive movement. And like I said, I have been saying that I disagree with the baseline. If your baseline is between a moderate House candidate and a liberal presidential candidate, you're gonna observe a difference that would not necessarily exist between a replacement level moderate candidate because candidates in whatever district you're looking at tend to be more moderate. So I do think that's better.
- [47:41]
C
Right, but going back and forth, I'm not gonna get to be clear. Grumbach and Bonica, the structure that you were specifying, the specification that you were, you are talking about is I see the value in it. I disagree fundamentally, but I see the value in it. But I want to specify, Galen, that what Grumbach and Bonica do is not what Elliot is saying. What they are saying is that ideology just doesn't matter, period. They're not actually saying like, oh, Elliot's model evaluates say Jared Goldin or another Blue Dog against another Blue Dog esque candidate that would come from that region. They are saying that these types of scores should be used to conclude ideology is null. Even though Elliot's model, really a closer description for the types of inferences he's describing is Blue Dog versus Blue Dog because he says you shouldn't be comparing Blue Dog to liberal presidential candidate and doing an over performance metric there. But that's not what Grumbach and Bonnik are doing. They're just taking it to spin something completely different. And in the process it's just a whole bunch in my opinion, of random political science errors that would get you laughed out of any real political practitioners meetings where they would say, what the hell is this? Like, how can you conclude this? I mean, that's why Nate Silver wrote an article about this exact thing and he's like, none of these critiques add up.
- [49:05]
B
Yeah, I haven't seen Nate Silver's article. I'm a Trained political scientist and pollster. So I would say that if you're going to Write a whole 10,000 word academic paper on this thing that gets published in the peer review process, then there's going to be some merit to it. And I do think that there's merit to the conversation about a bad baseline or a biased baseline. I didn't mean bad biased, but again, I have not. I'm not defending them. I haven't read their paper to the extent that they are making critiques, that I'm making a critique, the same critique. It's about the baseline not speaking to the merits of their prediction. Because I haven't read it. I have read yours, I have not read theirs.
- [49:45]
A
Okay, So I want to try to tie some threads together and maybe ask some other related questions before we end here. Is it fair to say that you both find that moderation helps? The question is just degree?
- [49:58]
B
Yeah, this question is degree and implications.
- [50:01]
A
But yeah, degree and implications. I mean, I guess that's like another level of analysis, like what should you do with this information? Which we'll get to in a second. But when you talk about uncertainty, Elliot, you're saying that there are other factors that go into over performance, like fit for district or fundraising ability or what have you. But isn't.
- [50:24]
B
Yeah, just like, I mean, we're saying that the observed presidential result is a bad baseline, a biased baseline, sorry. And that the uncertainty that you can attach, the certainty that you can attach to that by using the observed presidential result biases you towards large effects. Whereas if you created a model like we did to predict hypothetical candidate performance for a true replacement level candidate, not a presidential candidate, you have a lot more uncertainty because we don't know who those candidates would be. And there's so many other factors that go into their skill.
- [51:00]
A
I'm not, I don't have a model in this fight, so I'm just speaking off of my own intuition. But like, my sense here is that if you use the presidential candidate, you have like a similar baseline across all districts. Whereas if you say that, well, the replacement for Jared Goldin wouldn't be someone similar to Kamala Harris, it would be someone more similar to Jared Goldin. Isn't that already baking in the assumption that moderates do better? Because the districts that decide who control Congress, which is kind of what the whole point of this debate is, are going to be those frontline districts where you have a bigger mix of Democrats and Republicans than say, you know, the district that I'm sitting in in New York City, right? Now. And so yes, on an academic level, if you say, well, the person needs to be better than a replacement level moderate, then sure, they need to be even better than a moderate in that district. You're already acknowledging that the moderation helps in the districts that matter most.
- [52:02]
B
Yeah, I mean that would be a good critique, except for the fact that we also do this analysis not with WAR as the dependent variable, but completely separately with the actual result of the election as the dependent variable, with your ideological score as a control variable. And you get the same result about a one and a half percentage point boost in recent elections. So I think the fundamental difference here is that if you are selecting candidates who are part of some caucus, they identify with that caucus. In this case it's Blue Dogs and the squad as your, as your big polarized caucuses groups, then you are selecting for something other than pure ideology. And we are making a statement about the pure ideology. I think this could get us into a really great other conversation. Perhaps we don't have time for it, about what moderation is for people listening. That's in air quotes. And I think we would probably come to about the same conclusion, which is it's part ideology and part skill. But we have, when you, when split ticket groups these candidates together and looks at the differences between them, they are already selecting on the dependent variable, which is in this case skill, because they're, you know, like blue dogs traditionally that they're grouping together have some other weird quality that make them better. That's not moderation. That's what we would call candidate skill.
- [53:37]
A
Well, I actually have written in front of me, how do you both define what moderation is and what's the difference between moderation and a voter's perception of moderation? So I think we do have to talk about that before, before we end here, which is. What do you guys mean? So Lautra, what do you mean?
- [53:55]
C
It means in practice rather than in theory. In practice it means being willing to, to cut against your own party. And more often than not, that includes working with the other side. It means having policies that appeal to a broader, broader swath of voters across parties. And it means if you have like a spectrum of Democrat and Republican that, you know, you're, you may have some policies that are like super left and some policies that are super right. Some people call that hetero, heterodox or incoherent. That's really like most voters. But on the whole it means that you don't just exclusively draw your ideas from one side and anchor them there. It means that you have either a healthy spread of ideas across the spectrum or a lot of ideas that are in between both extremes when it comes to actual policy. And it does matter how voters perceive it, because if voters don't buy it, then even if you've moderated, like Kamala Harris tried to do, you're gonna find that they just don't believe it because they're skeptical of what Nate Silver calls Etch A Sketch candidates.
- [55:09]
A
Elliot, how do you define moderation in this debate?
- [55:14]
B
I would say moderate is typically used to describe someone with a moderate ideology. And to the extent that we are talking about something else, we're talking about vibes. The title for my first article was just gonna be Moderation is just vi based on the lack of a relationship between empirical moderation and election results and the fact that if you construct these very contrived groupings of candidates, you can select into higher over performance by them. Again, I think it's exaggerated because of the baseline difference here, but even in our model, if you just look at Blue Dogs, they do a little bit better than moderates than the actual empirical moderates do. So I do think, to answer the direct question that you posed initially, that perception of moderation matters more. And to the extent that candidates can establish different perceptions of themselves, that's a product of candidate skill, not of ideology.
- [56:12]
C
But my objection to this is, would you say that Marjorie Taylor Greene is doing badly simply because she has terrible vibes? Or that the freedom cock is consistently underperforms election after election because it's just vibes like what you say Mike Lawler is a strong congressman because of vibes. Like you can concretely point to liberal policies that Mike Lawler and Brian Fitzpatrick has signaled openness to compare, like Ukraine aid, for example, compared to what Marjorie Taylor Greene does. And I think it does a disservice to say that this is just vibes when you can concretely point to things that are actually there.
- [56:50]
B
Yeah, I mean, if candidates are taking policy positions in their districts that are better matches for the district, but that do not systematically affect the empirical estimates of these candidates ideology, then by definition they are not ideological positions. They're positions that match these candidates better with their district. And that is what elections, at least in the view of our model, really hinge on. Those effects are larger than the effects of ideological moderation.
- [57:21]
A
Isn't that moderation by a different name, though? Like, it's calling it like a fit to your district or candidate skill, making yourself be perceived as a moderate because you are in a More moderate district.
- [57:35]
B
No, I mean in this case, the positions that these candidates are taking are not actually making them empirically like more moderate. They are taking a mix of positions evidently that don't push their ideological score into the very moderate zone, like Henry Cuellar, for example, but that get them some extra over performance. And from our point of view, we wouldn't call it ideological moderation because they don't have moderate scores in their ideology. They're doing something else.
- [58:10]
A
So you're saying these are like good candidates, lived out candidates who just have like one sister soldier moment and that's how they win their district or something like that?
- [58:19]
B
No, I would say that there's a small boost to being a moderate. To the extent that they're moderate, they also might be winning their districts because they're like strong fundraisers. Maybe they have some sort of viral sista soldier moment that helps them raise money or convince more voters that they should vote for them. But by and large, candidates that are successful know how to position themselves relative to their districts. David Mayhew called this the electoral connection. They know how to juice themselves on the electoral connection and they win elections. This is just much closer to like the traditional understanding of how candidates win. That's not just, let's call it ideological moderation.
- [58:57]
A
Would anyone make the case that ideological. There's a benefit to ideological extremism?
- [59:02]
B
I wouldn't.
- [59:03]
C
I wouldn't.
- [59:04]
A
All right, case closed. We're. No, I'm kidding. You know, I want to also approach this question from another direction, which is like we do have lots of data that aren't just war models. And is there anything that we can use from whether it's public policy, opinion polling, issue polling or anything else to try to get to a better answer as to whether moderation is helpful or not?
- [59:30]
C
I tend to think issue pulling is a lot of it's very badly done. That's actually one thing we're trying to fix at the argument. We're trying to do robust and good.
- [59:37]
A
Issue pulling promo in.
- [59:39]
C
I hope, I hope we'll succeed with that. Yeah. But you know, there's an interesting article by Jesse Richardson. He's a forecaster I like on Twitter. He goes by the name Political Kiwi. But he posted something about how moderation measures that people encode. Like for example, do you support a 15 week abortion ban or are you in favor. Are you largely in favor of pro choice legislation? If you encode things like that into your model for forecasting, it actually tends to help when it comes to forecasting the Real results of the model or of like of an election model. And so to me that's actually really interesting. Nate Silver again, yesterday in our substack live he said the same thing. He's like, you know, we've been encoding stuff like ideology into our models for a while because it actually concretely does help in performance. So I think beyond war models and again we found our war model also helped us when it came to predicting the 2024 elections, which I thought we did pretty well in. And Elliot, so did you. I remember you guys got like six house seats wrong. It was pretty good. But like, you know, these things help. These, you can look at the impact of moderation on things like forecasting models. You know, Jesse Richardson has a public piece out about it and it really does help and I think that's useful.
- [61:00]
B
I will take an example straight from the horse's mouth in this case. Matthew Iglesias in his response to the headline of my first article, seemingly just that says Jared Goldin is a high performing candidate in 2024 because he says that Trump won't be bad for democracy and that convinces voters in his moderate main to to vote for him. But there's a problem with this, which is that Jared Goldin does worse in 2024 than he does in 2022 or 2020. He does even worse than you'd expect based on the national swing in the election. So if the examples that we're using to describe moderates aren't even the things aren't even correlated with their over performance sometimes, then what we're engaging in is some like post hoc rationalizations of why good candidates do well. And that's just like borderline tautological. And that's kind of the point if true ideological modernization empirically measured by things that correlate well with donor networks, with votes in Congress. Other factors about district results don't predict over performance, but the things that do are like examples from the campaign trail that is so much closer to the definition of this person is good, they're a good politician than they have a benefit from moderation. And Jesse Rischan, by the way, is someone we've published at strength in numbers. I think his analysis is good, but it's also a one off example of why they would have been better at predicting something and not an empirical long range study of that thing. Also we are directionally aligned. Moderation slightly helps. And in cases where a forecasting model is uncertain about something, you can probably find some examples in the qualitative, not quantitative world that improve your performance in your forecasting model.
- [63:11]
A
So I think this is an interesting sticking point, which is if a candidate comes onto the scene and says, I want to build a wall at the southern border and I want Medicare for all, to me that reads as moderate by way of being, you know, heterodox. But once they get to Congress, they may not ever have the chance to vote on either of those things. And so that specific moderation or triangulation would never be recorded in their ideology. But you could still make the case that if they won, it was in part because they sort of pieced together two, maybe at that moment in time, popular positions, you know, better, more affordable healthcare and better border security. But that might not necessarily show up in their ideology. And so I kind of, I don't want to like, be the person here who's like, you guys are saying the same thing. It's like, watcha, you're just saying all of that is moderation. And Elliot, you're saying like, to the extent that they vote as a moderate once they get to Congress, that's moderation. And all the rest of that is just skill.
- [64:19]
B
Well, but it's, but it has nothing to do in these examples with how people are voting. It has everything to do with how they're campaigning. And that's closer to a candidate skill parameter. Just like on the, on the congressional votes thing. The two examples you cite are things that were voted on in Congress. But I get your point. And votes in Congress are not perfectly captured by some measurement scales. We try to get around this by using ideological, empirical measurements of ideology that take into account multiple things and that do predict House votes better than something like dwnominate, which has a really hard time predicting how the squad votes and calls many of them moderates for like, weird statistical reasons that make the metric biased. But, you know, one of the reasons that we published the model was to be open source and transparent about how the models, how these things work, and about how much uncertainty there is. So we will be updating them over time with other metrics as we get data from those authors and we'll reassess the conclusions. But like, I just think that we are so dominated by, in these discussions of Democratic strategy by people who see candidate skill parameters as moderation that aren't moderation. And the people actually making decisions about where to put money in the dccc, for example, are probably looking at these closer to candidate skill parameters than everything. Everything that someone who I like does is moderation. Not to characterize laksha, but to Characterize some other people making these arguments.
- [65:58]
A
Laksha, do you agree or disagree with anything that I've said or that Elliot has said there?
- [66:02]
C
No, I think it's just, again, like, fundamentally, it just comes down to a different scale. And as you said, in what we define as moderation, in a lot of ways, I think Elliot takes a much, I don't mean this derogatorily Elliot, but like, Elliot takes a much narrower view of what constitutes substantive moderation. And he says the rest is campaigning. And I, I tend to fold a lot of those into the same umbrella. And again, I want to be clear here, right. Like, a lot of what I'm saying here in my objections, these are good faith methodological disagreements that two experienced practitioners can have. Right. Like we're. I can sit here and I can talk till I'm blue in the face about why I think you need to consider the 24 presidential baseline. I think Elliot will see the value in some of what I'm saying, and I think he'll just fundamentally still disagree, and that is fine. What I think is important though is that, like, there are a lot of areas of agreement here and a lot of what we have said has been taken by, I would say, less scrupulous practitioners of political science in general and used to say something that it's not saying. Right. Like, you know, I know Elliot and Matt Iglesias don't get along with each other, but I'll give my own example just again with Grumbach and Bonica, what Elliot characterized his model as doing is explicitly very different from what they said in their own critique. And, and to me, somehow this has become a big meta commentary. I think in part because people chose to lead with ad hominem attacks and criti rather than discuss in good faith the model because they were upset at the tractions or the types of people that were using these models to make inferences. And you know what? Fair enough. Like, again, water's warm, everyone jump in. It's all free for all and debate. And I get that. I just happen to think that in the process, we shouldn't lose sight of what, you know, what the commonalities and agreements are.
- [68:02]
A
What do you think folks should do with all of the information that we've discussed? Like, like, if moderation directionally helps, what should they do? If moderation doesn't help, what should they do? Like, should this dictate what parties pursue?
- [68:20]
C
I think more than that, it just highlights the need. Like our war model number one does say, like, yes, moderation is important. I think like literally the findings say that. I think it at least tells you that this theory of winning by just going all in on your base, I think meaningfully decreases your in cycle odds of success, which I think is important to note. It decreases your in cycle odds success. As Elliot said, you can win anyways, but you will pay a price in cycle with the odds you've, you have. And you know, maybe you don't notice because you win a trifecta anyways, but again, I got to point out Republicans probably should have more than 53 seats in the Senate, right? A lot of what you're seeing with legislation, it would be even worse if they had 58, 59 seats as they've cost themselves. So whether you attribute that to ideology or not, we can go round and round in but to me it at least tells you about the types of candidates that win, the types of things that you may need to appeal to and the types of inferences you need to draw if you want to continue as a Democratic party think needs to do competing on a broad map and win swats of red voters. And for the record, my criticism doesn't stop at progressives. I've criticized the New Democrats more than anyone because I think at least with progressives I have seen a template where their type of candidate has produced someone really unique and with the potential of winning these red states. Like I want to highlight Bernie was doing better than Hillary in a lot of these red states in 2016 in polling. But like, like I've never seen the New Democrats actually do anything over the last like eight, ten years beyond just talk about the need for moderation which I call dog whistle moderation as Matt Iglesias does too without actually doing anything right. Like it's not just oh, the progress is underperformed, it's no, look, they have done worse. But there's a version here that I think could work. It's not this version, but there is a version and I think like more than anything it just highlights the need to, to pick candidates that as Elliot said, are unorthodox and can break from your party to really break the mold. Like Dan Osborne.
- [70:31]
A
Right. The irony here is that oftentimes in moderation is taking cross cutting extreme positions. Like New Democrats are never going to be. Like you should take an extreme left position and pair it with an extreme right position and that's how you're going to appeal to main second district or I mean maybe they would acknowledge it now but like that seems to be evidently true. But like the kinds of moderation that. That the New Democrats might talk about is very distinct from the kinds of moderation that we see perform the best, actually, I think in both of your war models.
- [71:01]
B
Yeah, I mean, it's just like the problem is still the examples that we're taking here, to use the term the dog whistle, moderation examples are also the things that, you know, empirically have not actually helped candidates in the own examples that some of these people cite. Again, the golden Trump won't be bad for democracy thing, which, by the way, turns out to be total bull. Right. So if the things that we are saying are moderation, and we're also saying those help people and our hypotheticals are the things that aren't actually helping them sometimes, then I think this just underscores our point about uncertainty. We don't actually know a lot of the time what is the reason that a candidate won their campaign. And that boils down the uncertainty in elections that we observe from factors like the national environment, mainly, but also skill and fundraising. And to the extent that those are going to exist in the future, they weaken our ability to make prescriptions for campaigns purely based on ideology and push us in the direction of doing a lot of due diligence to select skilled candidates.
- [72:16]
A
Well, wrapping up here, I'll say that one important lesson for me from all of this is that maybe especially in. In the world of folks who pay a lot of attention to politics and are intrigued by data, when something has the, oh, this is data label next to it, people are like, inclined to just sort of maybe take it at face value more often than not. I think this conversation is a great example of that. You can start with the same data and process it in different ways and get very different results based on sort of the models that you build. And in many ways, models are not divined by God. They're not written in the stars. They are codes that represent how we view the world or how we understand the world. And a lot of times it's based in sort of years of academic research, but it's art as much as it is science in some ways. Right? You. You make different interpretive decisions based on what kind of artist you are, and you get different results. And that doesn't mean that anyone's bad or good or completely wrong or completely right. I mean, over time, we will find out who's more right in a sense. But I think this is an example of why people should also just stop for a second and be like, famously, as this podcast frequently asks, is this good? Data is this bad data is this not data. And if two models that are trying to explain the same thing disagree, ask why. I don't want to be the final word here, though. Have either of you taken anything away from how this has all played out that you'd like the. Or do you disagree with how I've characterized this that you'd like the public to remember?
- [73:58]
B
Going forward, I will give Laksha the opportunity to have the final word. And I'll say I think that this was a much more productive conversation than the one that has been happening on Twitter, particularly on social media. And I've got a lot of respect for him and we're friends in real life, so don't take this out of context. People listening.
- [74:18]
C
Yeah, likewise. You know, I just want to say, right, Like, I respect Elliot as a person. I think his math skills are, again, very strong. He's much better at statistical inference and probabilistic reasoning than a lot of political scientists are. And I think if more people understood the uncertainty in modeling and the choices that go into these, it would only help. Our disagreements are just in design, and those are fundamental disagreements. But again, two smart people, or in, in our case, one somewhat dumb person, me and a smart person, Elliot, can disagree without going into bad faith attacks. What I was disappointed by was the way that people used Elliot's model as like a way to say that we were just pushing hackish findings that we wanted. When we have progressives, moderates, and Republicans on split ticket, like, we didn't design this thing to come out with any ideological thing. We, our first time writing about this was 22, when we said, hey, there wasn't really much of a difference between progressives and New Democrats here. It was just the blue dogs. What we did was just write the data as it seems. And I think if you lead with that and you understand people come from good places, you'll have more productive conversations.
- [75:41]
A
Well, with that, thank you for modeling war. Thank you for modeling civil disagreement. And we're gonna leave it there. Thank you, Lachla and Elliot.
- [75:49]
B
Thanks, Galen.
- [75:49]
C
Thanks, guys.
- [75:50]
A
My name is Galen Druke. Remember to become a subscriber to this podcast@gdpolitics.com and wherever you get your podcasts. Paid subscribers get about twice the number of episodes and they can also join our paid subscriber chat and pass along questions for us to discuss on the show. For this episode in particular, you can also watch the video. And most importantly, you. You ensure that we can keep making this podcast into the future. Also, if you're so inclined be a friend of the podcast and give us a five star rating wherever you listen to pods, maybe even tell a friend about us. Thanks for listening and we will see you soon.