#14 - Dr. Matilde Bombardini - U.S. Climate Politics
Today’s episode talks about what shapes voting behaviour for environmental policies. I had the pleasure of interviewing professor Professor Matilde Bombardini on her working paper “Climate Politics in the United States”. Below is the transcript of our interview.
Transcript
Arvid Viaene: We talk a lot about the “right” climate policies, such as carbon pricing. But there’s a step before all of that: politics. Because you need politicians to carry out these policies. An who gets elected is decided by voters and their preferences as well as the positions of politicians. And these are likely to change when the climate gets hotter and the economy starts to transition.
So today’s episode will tackle these in the context of climate policy in the U.S.. In particular, you will be answer the following questions at the end of the episode
When a place experiences unusually extreme heat, does it measurably shift votes?
How do local green and brown jobs shape climate politics?
And how do those forces influence the probability of future climate legislation being enacted?
My guest is Professor Matilde Bombardini, and we’re discussing her working paper “Climate Politics in the United States.” What makes this research stand out is the data: precinct-level election results—so we can compare neighborhoods within the same congressional district—and detailed measures of candidates’ environmental policy positions.
Matilde holds the Oliver E. and Dolores W. Williamson Chair in the Economics of Organizations and is Professor of Business and Public Policy at UC Berkeley Haas, affiliated with NBER, the BFI’s IOG group, CEPR, and CESifo.
So let’s dive in. Matilde, welcome to the podcast.
Matilde Bombardini: Thank you so much for having me.
Arvid Viaene: I’m very excited, as I said. In previous episodes we’ve covered the impact of climate change and how to address it cost-effectively through tools like cap-and-trade. But this is our first episode that really analyzes climate policy from the perspective of what drives it through voter behavior.
So maybe just to start off: what got you interested in this project, and what were you trying to do?
Matilde Bombardini: Sure. I should start by saying that the idea that voters react to climate shocks is not new. We’re not the first to think about this. There are other papers that look at, for example, the effect of climate shocks on beliefs and use survey data to measure that.
What’s really unique about this paper is that we have very comprehensive data on voting. So it’s not just about how people’s beliefs respond to climate change; it’s about their actions – how they actually vote. At the end of the day, that’s what determines who gets elected.
We wanted to understand how different aspects of climate adaption and climate affect both voters’ behavior and politicians’ behavior. You can think of voters as the demand side of a political market, and politicians as the supply side: they “supply” policies.
The occasion came when one of my co-authors, Nicola Longuet-Marx, had completed a large data collection effort for the U.S. I’ll go into more detail later, but essentially this gave us very granular data in the U.S. So we could ask questions like: how does climate change itself – weather shocks – matter, and how does the climate transition matter – the changes in labor markets associated with the transition? Because we can measure these things at a very granular level, we can be satisfied with the answers we get, as we’ll discuss.
Arvid Viaene: Exactly. I think that’s what’s so interesting about your paper: it looks not only at voter preferences, but also at how politicians react to them. It really treats this as a demand-and-supply market, which I don’t think people always fully appreciate.
Matilde Bombardini: Yes. The important point is that if you only look at voters’ reactions, you’re very constrained to an inevitably short-run effect. Voters react given the policy positions currently on offer and the politicians who are already out there.
But we know that politicians will also adjust. If we want to go beyond the immediate short run and think about the future, we need to model how policy itself is likely to evolve – and for that we need both sides of the market.
Our broad inspiration comes from integrated assessment models. I’m sure you’ve talked about these on your podcast: they close the loop between the economy and climate. Firms decide how much to produce, that generates emissions, emissions cause more climate change, which feeds back into the economy, and so on.
What’s missing in many of these models is the political block. Emissions lead to higher temperatures or different precipitation patterns; people experience those changes; but how does that translate into political reactions that move policy in a particular direction?
That political reaction block is missing at the moment in these IAM dmoels, and our hope is that this paper gives a way of modeling that part.
Arvid Viaene: That’s a very good point. In the paper you focus on two main measures of shocks. One is extreme temperature – these very hot (or otherwise unusual) weather events. The other is related to the part of the economy people work in: whether their jobs are “green” or “brown.”
For example, in Europe we had some really hot weeks this year, and immediately you see the costs. People become very aware of climate change when they’ve just lived through a summer like that. It becomes a lot more salient. So could you explain the measures you use for extreme temperature, and why they might influence voting behavior?
Matilde Bombardini: Absolutely. I’d also reinforce what you just said: when you’re experiencing that heat, if you ask people right afterward, there’s strong evidence that they say, “Oh my God, climate change is the worst problem we’re facing.” We wanted to check whether that kind of reaction persists when people actually go to vote. We experiment with different measures, but the goal is to capture variation in both temperature and precipitation. We look at absolute levels – is it a particularly hot or particularly cold day – but our main measure focuses on an extraordinarily hot day relative to the baseline, so how unusual the weather is relative to local history.
As you know, some places are always hot. We want to capture unusually hot days, so we define a day as “extremely hot” if it is more than two standard deviations above the historical average for that specific day of the year in that location. We then count, for each year, how many days cross this threshold – and we do the same for extremely high precipitation.
Arvid Viaene: Right, because it’s really hot in summer and then we get to winter. By the time you go vote, does the effect? You mentioned earlier that you weren’t sure whether the effect would persist. Did the paper convince you that it does – at least to some extent?
Matilde Bombardini: At the end of the day, we do find effects. We’re working with about the data that as good as it is going to get for the U.S. in terms of quality and spatial detail, and that’s important for being confident in the results.
When I talk about “granular data”, I mean that our unit of observation is the precinct – essentially the polling place where you go to vote. In the U.S., that’s roughly 1,200 people per precinct, and there are about 400 precincts in a typical congressional district. A congressional district is a single-member district that elects a member of the House of Representatives.
A lot of previous work, in the U.S. and Europe, is at a much more aggregate level – at the district or county level. If we only had data at the congressional-district level (or higher, like counties), it would be very hard to do the joint demand-and-supply analysis we do.
Why? With precinct-level data, we can compare different locations within the same congressional district that happen to experience different numbers of extremely hot days, holding constant who the politician is – their party, ideology, and other characteristics. So we’re observing at the smallest level and isolate the effect of the weather shocks themselves. There is some individual voting data out there, but only on turn-out basis.
Arvid Viaene: Yes, I really want to emphasize how granular your data are. I was very impressed reading the paper: you have this long period of time and very micro-level voting data. Politicians are complex: they have stances on a wide variety of issues, and you can now filter out a lot of that and zoom in on their climate positions. The way you do that in the paper is really impressive.
Matilde Bombardini: I should take zero credit for that part. This is all our young co-author, Nicolas Languet-Marx, who is going to join Berkeley next year and is currently a postdoc at Stanford. In his job-market paper he collected and assembled this voting dataset. You might think it shouldn’t be so hard to get, but in the U.S. elections are run by states and counties, and the data are stored locally. So he had to collect precinct-level results from hundreds of different sources and then harmonize them.
The other major data source he built is on policy positions. He uses both surveys of candidates and information from their websites and campaign platforms. Combining those, in a way that’s detailed in his paper, he constructs measures of each candidate’s positions along several dimensions: cultural, economic, and environmental. That is really what he did in his job market paper.
What’s amazing is that he gets positions for both candidates in a race. We already have good ways to measure the positions of incumbents – people who are elected and cast roll-call votes – but those are typically only for politicians after they are in office. To get the positions of both candidates before the election, you need this kind of campaign-based data.
Arvid Viaene: Exactly. And you also mention in the paper that the way candidates market themselves during the campaign is a good predictor of how they actually vote once in office. It is a credible signal. It’s not just cheap talk.
Matilde Bombardini: I’m glad you noticed that. Some people immediately think, “Well, this is just what they say; of course they’ll behave differently once elected.” But in the data, these pre-election positions are actually very good predictors of roll-call votes. So we take them as pretty good measures of policy stances.
Arvid Viaene: The other important measure you look at is green jobs versus brown jobs. Could you explain how you measure those?
Matilde Bombardini: Yes. At the same level of granularity, we construct measures of “green” and “brown” employment. You can think about the costs and opportunities of the green transition in many ways, but we chose a very intuitive one: how many people are likely to lose jobs, and how many are likely to gain jobs.
Measuring brown jobs is relatively straightforward. We focus on employment in oil, gas, and coal. Those are very easy to identify and measure. Green jobs are trickier. We could spend half an hour talking about different definitions, but in the end we try to capture jobs related to the green transition, electrification, and renewables.
So it’s not just people working directly in wind and solar. It’s also jobs related to replacing power plants, upgrading the power grid, improving infrastructure, and a variety of ancillary activities. We follow definitions from the U.S. Bureau of Labor Statistics for green industries and occupations.
Arvid Viaene: I really recommend people check out the working paper, because you have some great graphs that show this visually. I might even include them when I publish this transcript. You can clearly see, for example, that Texas looks like you’d expect: there are a lot of oil and gas jobs, so more “brown” jobs that are at risk when regulation tightens. Naturally there’s more concern in those regions. I think we’ve covered the data, unless there’s something more you want to add.
Matilde Bombardini: We can always come back to it, but that’s the main idea.
Arvid Viaene: Great. Let me just sketch the structure of the paper for listeners. Roughly speaking, you first look at how extreme weather and green versus brown jobs influence voting behavior. Second, you look at how those changes in votes influence politicians’ positions. And third, you use those estimates to build projections into the future. So let’s start with the first piece: what are the impacts of extreme weather and green versus brown jobs on voter behavior?
Matilde Bombardini: We measure extreme temperatures as the number of “extremely hot” days in a year, in the sense we discussed earlier (two standard deviations above the historical local average). A one-standard-deviation increase in this measure – roughly eight additional extremely hot days in a year – increases the Democratic vote margin by about 0.66 percentage points. So if Democrats and Republicans were tied at 50–50, that shock would move the race to about 50.66–49.34.
For extremely high precipitation, we find a smaller effect – roughly half the size, about 0.35 percentage points for a one-standard-deviation increase – and those precipitation results are less robust overall.
If you think of both temperature and precipitation shocks moving together by one standard deviation, you get roughly a one-percentage-point increase in the Democratic margin. In a marginal or competitive district, one percentage point can matter a lot.
On the labor-market side, we measure the importance of green and brown jobs as employment shares. A one-standard-deviation increase in the green-job share increases the Democratic margin by about 0.2 percentage points, while a one-standard-deviation increase in the brown-job share reduces the Democratic margin by about 0.6 percentage points.
So the magnitudes are comparable and economically sizable, even if they’re not enormous. Many things affect voting, so we wouldn’t expect climate alone to explain everything.
Arvid Viaene: To summarize the temperature result: roughly one extra week of these extremely hot days – about eight days – translates into a 0.66-percentage-point shift toward Democrats. In a close election, that can really make a difference.
And on the jobs side, as you’d expect, places with more brown jobs are more worried about regulation, so that also has a sizable impact.
Matilde Bombardini: Exactly. We also tried to contextualize these magnitudes by comparing them to other known determinants of voting. One of them – and this may be less familiar outside the U.S. – is Fox News exposure, measured in minutes watched.
A one-standard-deviation increase in our extreme-temperature measure has a similar effect on Democratic vote share as roughly 2.5 extra minutes of Fox News exposure in the estimates from other papers. Those Fox News effects are very large, which is why such a small number of minutes shows up here as comparable.
We also compare our effects to those of partisan political advertising, and again, a standard-deviation change in advertising has a similar order of magnitude.
Arvid Viaene: When I first read that comparison – 2.5 minutes of Fox News – I remember thinking: that doesn’t sound like much, but given how powerful the media effects are in those studies, it’s actually huge at the margin.
Matilde Bombardini: That’s right.
Arvid Viaene: Earlier you mentioned persistence. Just to be clear: are these persistent shocks, in the sense that if people experience extreme heat this year, they’re still affected when they vote later on?
Matilde Bombardini: What I meant is that we don’t literally know whether someone remembers, say, “There were 12 extremely hot days last summer” when they vote a year later. Our estimates pick up the effect of weather shocks on voting when the election comes around. They’re not designed to track whether a person retains that memory in a psychological sense. So I wouldn’t claim strong “persistence” in that strict sense.
Supply-side respones ~19:00
Arvid Viaene: Got it. Let’s turn to one of the really interesting parts of the paper: how all of this shapes the positions of politicians. We’ve got climate change happening – and I’d assume anyone listening to this podcast accepts that it’s real. As we see more extreme days, people gradually shift toward voting more for Democrats, who, on average, have more pro-climate platforms. But you show some very interesting supply-side responses. Could you walk us through how you model that?
Matilde Bombardini: We model this in the simplest way political economists usually do: a spatial competition model in one dimension. The dimension is environmental policy: are you more or less in favor of ambitious climate policy? Candidates choose positions along that line.
Two forces pull on politicians. First, voters: each voter has an “ideal point” on that line, and candidates want to get enough votes to be elected. Second, their party or own ideology: they may have internal preferences – for example, a party platform – that pushes them toward a particular position. Politicians are solving a trade-off between appealing to voters and staying close to these internal or party preferences.
The complication is that when shocks hit – weather shocks or labor-market shocks – they affect both sides. Voters move, and politicians want to follow voters to win elections. But the shocks may also affect politicians’ own ideal points, perhaps through pressure from donors, activists, or the party.
So we need to disentangle how these shocks affect politicians’ final policy positions through these two channels. That’s what we do in the model and empirically. We estimate what happens to voters and what happens to policy positions, and then back out what part of the change in politicians’ positions is a response to demand (voters) and what part is a change in their own supply-side preferences.
Both parties react quite strongly to voters’ behavior, but in different ways, and it also matters whether the shock is a weather shock or a jobs shock.
For Democrats, start with the weather shocks. When you get a climate-related “popularity shock” – say, more extremely hot days that increase support for Democrats – voters move toward more pro-environment positions. Through that demand channel, Democratic candidates would like to move left on the environmental dimension to track their voters.
At the same time, when we look at their own supply-side preferences, we find that weather shocks push Democrats to moderate slightly – to move a bit to the right relative to where they’d otherwise be. One possible interpretation is that as they see climate becoming a more salient issue, they anticipate real competition on that dimension and choose somewhat more moderate positions rather than very extreme ones.
So for weather shocks, you have these two forces: voters pulling them left, and their own preferences nudging them slightly right. The net effect is relatively modest. For jobs shocks, when there are more green jobs, Democrats know voters will be more favorable to environmental policy, so again the voter-demand channel pushes them to the left.
But in this case, their own supply-side preferences also move strongly to the left. They may over-interpret the importance of green jobs, thinking, “Voters are going to love this so much that I should move very far toward ambitious green policies.”
We find that this overall shift to the left can actually be counter-productive for their electoral prospects, in the sense that they overshoot what would be vote-maximizing. That’s a potential lesson for Democrats – and for politicians in general – about not overreacting to favorable structural trends.
For Republicans, we find that their own ideological or party preferences are relatively insensitive to these shocks. The supply-side channel is weak. They do respond to voters moving. But here we find something that puzzled us for a while: when there is a pro-Democratic shock – for example, extreme weather that pushes voters toward Democrats – Republican candidates tend to move further to the right on environmental policy.
Our interpretation, which is consistent with the model, is that in some contexts Republicans essentially give up. Democrats become so hard to beat on the climate dimension that Republicans focus on signaling their ideological purity instead of trying to converge. They move away rather than toward the median voter.
Arvid Viaene: That result really resonated with some conversations I’ve had. A friend of mine in California said something similar: as a Republican, you might not want to compete with Democrats on climate if you think it’s a losing battle. You might actually double down and dismiss it even more.
Matilde Bombardini: Yes, that’s very much in line with how we think about it.
Arvid Viaene: Which may also help explain why Democrats can afford to moderate a bit on environmental policy: if Republicans move further away, there’s less direct competition on that dimension. Is that a fair reading?
Matilde Bombardini: It’s a good intuition, but it’s not exactly how our empirical model is structured. For estimation reasons, we don’t allow a candidate’s policy position to respond directly to the other party’s policy position. In our equations, Democrats don’t react to Republicans’ positions per se, and vice versa. They only react indirectly, through the observed vote margin.
So what we can say is that Democrats respond to how voters are moving, and Republicans respond to how voters are moving, but we don’t model a direct “I move because you moved” interaction between the two parties.
Arvid Viaene: That makes sense. You then take all these fairly complex voter and politician responses and plug them into climate projections. Earlier you mentioned integrated assessment models: temperatures get noticeably warmer over time, and that feeds into damages. Here, you let that warming feed into voting behavior, which then influences politicians’ positions. And you use that to run some really interesting simulations.
Could you walk us through what those projections say?
Matilde Bombardini: Yes. Once we’ve estimated the responses on both the demand side (voters) and the supply side (politicians), we embed them in projections of future climate and labor-market conditions.
Of course, you could imagine a complete realignment of U.S. parties that our framework wouldn’t capture, but conditional on the current party structure we can ask: if temperatures and green employment evolve in a certain way, what happens to votes and policy positions?
We take standard 2022–2050 projections of climate change under different socio-economic scenarios – we look at a more extreme pathway and a more moderate “business-as-usual” one that doesn’t meet the Paris targets.
Then we ask: under each scenario, how does the composition of Congress change, and what does that imply for climate policy? What we find is that, if nothing else changes, the composition of Congress becomes more Democratic over time because of these climate and green-jobs shocks. That’s the primary reason why we get more pro-environment policy in our simulations.
We translate that into a concrete policy experiment: we ask how the probability that the House passes a carbon-pricing bill, similar to the cap-and-trade proposal that nearly passed in the early Obama years, changes over time. Our estimate is that, under the more extreme climate scenario, this probability is about 9 percentage points higher in 2050 than in 2020. So if the probability was in the upper 60s in 2020, it would be in the upper 70s by 2050.
You can ask questions about this size – we’re not doubling the probability – but it is meaningful. People have reacted differently to this number: some find it encouraging; others are disappointed that it isn’t larger.
It raises a natural question: voters do react to climate change and green jobs, but do they react fast enough and strongly enough to really change the trajectory? That’s something we can’t fully answer, but the estimates give one perspective.
Arvid Viaene: I chose to read it in a somewhat optimistic way. An extra 9 percentage points, combined with other shifts, can move you from, say, a 45% chance to a 54% chance of passing a bill. And as you mentioned, a cap-and-trade bill like Waxman–Markey was already very close to passing once. So these margins matter.
Matilde Bombardini: Exactly. Even in 2020, the probability of passing such a bill is not zero, and an extra nine points can be pivotal. But, yes, we’ve definitely heard people say they wish the number were higher.
Arvid Viaene: And just to be clear, that 9-percentage-point increase is by 2050, right? Is that a roughly linear increase, or does most of the action come later, when climate impacts really ramp up?
Matilde Bombardini: Yes, that 9-point increase is by 2050. In our framework the effects are essentially linear in the underlying climate and jobs projections. So the timing depends largely on when the climate models forecast temperatures to rise and when green employment grows. So if the physical climate models show more change later in the period, that’s when our political effects also become stronger.
I should add that we’re currently working on extending the analysis to incorporate a more optimistic scenario for green and brown jobs. In the current draft we rely on BLS projections that are close to business as usual. We’re now adding a scenario with more aggressive electrification. So, for anyone interested, it may be worth checking out future versions of the paper.
Arvid Viaene: When you say that the Democratic vote margin increases by 9%, do you also translate that into a number of additional Democratic seats in Congress?
Matilde Bombardini: We mostly report results in terms of vote margins. Under the worst climate scenario we analyze, we find that by 2050 the Democratic vote margin for the House increases by about 1.4 percentage points. Under the more muted scenario, it increases by about 0.8 percentage points.
Arvid Viaene: Before we wrap up, is there anything important we haven’t touched on yet?
Matilde Bombardini: Yes, there’s one concern we’ve heard a lot, and I think it’s important. Someone might say: maybe when there are temperature shocks, people vote more for Democrats, but not because of climate policy. Perhaps heat is correlated with other things – like crime – that push people toward Democrats for other reasons.
We do several checks that make us much more confident that what we’re picking up is really about environmental policy. First, we run a kind of falsification exercise. Instead of using candidates’ environmental positions, we plug in their cultural positions. If heat shocks were just making people more likely to vote Democratic for generic reasons, we’d see effects there too. But we don’t: we find no corresponding effect on cultural policy positions.
Second, we interact the climate and jobs shocks with the distance between candidates in policy space. We find that the effects of the shocks on voters are stronger when candidates are further apart on the environmental dimension – but not when they’re far apart in other dimensions. Again, that’s what you’d expect if voters are responding specifically to climate policy.
Third, we look at a more narrowly focused type of election in two states, Oklahoma and Texas. There, voters elect commissioners who are directly in charge of energy and environmental regulation. These races are much more purely about climate and energy policy. We find similar patterns there, which reassures us that the effects we see are really tied to the climate dimension, not just “voting for Democrats because it’s hot.”
Arvid Viaene: I really liked that Texas and Oklahoma commissioner example, because those races are only about energy. There’s no other big ideological dimension that can sneak in. So for me the takeaway is that there’s an abundance of evidence that the effect is real and specifically about climate policy.
I would like to thank you for this conversation. I think people will learn a lot from this work. Some of my friends were already excited when I told them this episode was coming up. Thank you so much for taking the time and for giving us insight into the drivers of voting behavior on climate policy.
Matilde Bombardini: Thank you so much. It’s been a pleasure – and thank you for the very useful feedback for the paper itself. We’ll go back to the drawing board with some of your suggestions.
Arvid Viaene: Great. We’ll leave it there.