 Thank you so much for inviting me to speak today. This is joint work with Gabriel Aboyadana, and the title is Temperature Trust, Political Instability in Africa. And I would like to stress that this is work in progress. So if you have any questions, comments that don't come up in the Q&A, please do come and find me. Don't be shy, I'm always, like every academic, I love talking about my own work. So you come up and find me if you like. As a motivation, I mean, I'm sure most of you know this, but maybe for those of you who are not, kind of who are new to the climate conflict literature, there's a very, very strong correlation between temperature, which I've pictured on the left, and kind of incidences of conflict. This on purpose, I picked the first maps that Google came up with. So this is something that you can just by Googling in a few seconds, you can kind of see the very correlational evidence of this relationship. Now I'm an economist and the way that a large number of economists think about it is that the link between climate and conflict goes via agricultural incomes. So you've got a climate shock on the left, so for example, a drought. That climate shock affects your agricultural output, so if there's a drought, your agricultural output goes down, and that, again, affects conflict, maybe via opportunity cost via the income channel. So this is kind of the standard way and there are loads of amazing papers that kind of look at it through that lens. However, there is kind of a growing body of evidence that shows that heat or high temperatures have a direct effect on the behavior of people. So in other words, in effect, that does not go via income. So what a lot of studies had found is that when it's hotter, individuals are more aggressive, individuals behave more aggressively, individuals think more aggressively, and there's mainly experimental evidence, but not only, that shows there's an experiment that's pretty easy to do, right? You make people do some sort of interaction and then you crank up the temperature randomly in one room and not in the other. So for example, people find that retaliation is much stronger with high temperatures, road rage, gun use in training and baseball. There's so many different kind of areas of life where they found that people are more aggressive when it's hot. There's actually a really cool study on carried out in Finland that kind of tries to identify the biological link why this is the case, and they found that when it's hot, people are more aggressive and it decreases the level of serotonin in the body. So although there is still a bit of debate, they see that some neurotransmitters react to heat and affect the way that we kind of respond. And this is where our paper comes in. We document a direct effect of temperatures, so independently of income on three determinants of instability and unrest. One is political trust. So this is actually trust in institutions, what Tim said in the keynote today. So trust in institutions, protests and riots and even voting behavior. So we hopefully convince you that we find a causal effect. I talked through the research design, a causal effect of high temperatures on trust, on protest and on voting behavior. Our research design is relatively simple. We merge high frequency climatic data, so data that are very precise, both in the geographic level, so we know exactly where it's hot and where it isn't hot, and at the temporal level, so we know which days are hot and which days are not hot. We match that to individuals, regions, and to countries in Africa. We measure temperature via a heat index, which kind of combines humidity and heat. On your phone, probably, if you look at the weather, usually it says the degrees, and then at the bottom it says feels like, and then another degree, that's what we use, because we kind of think that that's a better measure of the thermal stress that people are under. And then we're looking at daily or monthly deviation from long-term means, which are also called anomalies. That's kind of a standard practice to identify causal effects. Just to walk you through the findings, first, we focus on attitudes. So what do people think? We find that on days that are hotter than the average for that area, so if by chance you interviewed on a day that's hotter than average for that area, you're less likely to trust your government. You're more likely to report that you vote against the government, and you're also, the interviewer, so the person interviews you, is more likely to state yes, the person I interviewed yesterday was aggressive. So we find these kind of things. Very importantly, kind of a key policy finding, because there isn't much you can do about the weather, but a key policy finding is that temperature itself doesn't cause mistrust. Rather, what it does is that if people are already mistrusting, if people are already dissatisfied, it magnifies these kind of initials level of distrust, which is what's really in line with the psychological literature. It's kind of a catalyst for being upset. So what we find is that this effect that I showed you here only occurs in poor countries, so countries where there's a lot of poverty, and within countries for people who are dissatisfied already, which we identify with a machine learning algorithm. Then second part of the study, we're looking at actions. So we're looking at protests. If an area is particularly high, hotter than the average for that month, you are more likely to observe a protest or riot going on in that area. And also if it's hotter on election day, if it's hotter than the average for that month in that country, that decreases votes cast for the government, which really fits with the attitudinal data that we found. So as I said, the heat index combines air temperature and humidity in kind of this formula here. We took this from an old Australian study that was kind of the first one of a heat index because we wanted to use kind of very established stuff. So this heat index is very similar to other heat indices that you use, and it combines air temperature and humidity. Climate data, we use the era five data set, which I think, Grazia, you use very similar one. It's basically a map that every point on the planet, you know exactly on what it's by hour, you know exactly how hot and humid it is for every point. We merged that to the afrobarometer, which is a big attitudinal data set for Africa. It's about 50,000, a bit more than 50,000 respondents, and these are the GPS coordinates of these respondents. So we have a pretty good sample of Africa, not every country, unfortunately, but pretty good thing. These are the measures of trust that we use. So probably this is too small, sorry, I'm used to zoom where you can kind of amplify stuff. So it does not trust parliament, does not trust the ruling party, doesn't trust the president, doesn't trust the police and doesn't trust the courts. So about 50% don't trust the government. This is people who want to vote for a party that's not the party of the president, so about 60%, and about 33% of individuals were perceived to be aggressive. These in the international context, I live in the UK and at the moment because we are having party gates, and so trust in the prime minister is ridiculously low at the moment, but before party gate, it was around this level as well. So these are not fit in pretty well with kind of international measures. The research designer wanted to walk through the intuition rather than focus on the equations. So one, I'm sorry, one problem you might say is that if you're picking up kind of heat, what you're actually measuring is some sort of underlying institution. So Tim, when he had these three groups, I kind of tried to read the countries. So all the countries in the kind of weak states there were countries predominantly that are hot. So you might be picking up hot countries. So colder countries have, on average, more efficient institutions. So we want to not do that. We want to get a causal effect only off temperature. So what we do is we divide the whole of Africa into 2,500 squares. It's more or less 100 per 100 kilometers, a little bit more. And then we calculate the average temperature for each month in that square. And what we're doing is, if this is person A and person B, so we are calculating is the difference on the day of interview in which person A was interviewed to the average for that month in that quadrangle. So to give you an example, let's say the quadrangle of Helsinki. So in Helsinki, the average temperature in May is 14 degrees. Today, I did this yesterday, it's supposed to be 11 degrees today. So what we are using is the difference between the two. So whilst you might argue that, okay, 14 is pretty cold, that might be correlated with institutions. So that is endogenous. The fact that today it's three degrees colder than the average, that is clearly random. So if it's hotter or colder compared to the average for that cell, that is something that is clearly random because we are counting for the fact that different countries have different temperatures. In terms of regression analysis, the key thing, the key variable that we're using is the C bar. That's the average temperature in that month in that cell. So because of that, we can interpret the coefficients as causal. So to some results, just the kind of overview. So these are just correlations. So we plot our heat index, which goes from 9.5 degrees to 40 degrees on the left for all the afrobarometer responders on here. And on the right, we plot some sort of index for trust where red is less trusting. And you can already see just by eyeballing there's a pretty strong correlation. Now if we use that regression model, which kind of identifies the causal effect rather than just the correlation, we find pretty consistent positive effects on mistrust. So positive meaning trusting less. So for example, if it's one degree hotter, you are 0.5 percentage points more likely to distrust the parliament. We got pretty, pretty similar results for the ruling party, president, police and courts. In terms of magnitudes, they are relatively small but not very small. There's only one study I found that does something similar and we got extremely similar results to them. The study looks at 4th of July rain in the US. So totally different context, but in terms of the size of our estimates are pretty comparable to that study. If we unpack the effect, this kind of linear effect to look at the shape it is, so if you look on the left one, zero, that's the base. So this is if the temperature on interview day is exactly what it usually is. So if today was 14 degrees in Helsinki, and if you go left, this is what happens to trust. If it's one degree colder than the average, two degrees colder and so on. So one thing you can see is that if it's colder, you don't find any effect on trust. If it gets hotter, for the first two degrees, you don't find an effect, but once it gets to three or four degrees hotter than usual, so this is always hotter than usual, we find kind of pretty consistent effects. And we kind of were pretty reassured by this because I personally find it almost impossible to distinguish one or two degrees difference and I don't feel that. But three or four or five degrees, that is really something that you do perceive. So it kind of makes sense that for small deviations, you don't have an effect whereas you have them for large ones. We also look at different days. So this is the effect of temperature anomalies on two days before the interview, that's one day before the interview, that's interview two days after. One day after, two days after. So we only find effect of temperature on interview day, which kind of makes sense because definitely not the day after or two days after, that should have no effect and also two days before doesn't have any effect whatsoever. One really important finding for policy is this, which kind of fits with the psychological literature is that temperature itself doesn't cause aggression or mistrust, which I think is not very intuitive, rather it magnifies already existing mistrust. So we do that in two ways. We look at poor versus rich countries, so very easy, how poor the country is, we distinguish countries. Within countries then we use almost 180 predictors in a machine learning algorithm to predict your, how trusting you are. We use a lot of questions from the after barometer, so stuff like, have you ever experienced corruption? Have you missed any meals? Do you have electricity in your area? Do you have a school in your area? Plus we merge it to historical slave trade data from this paper here, which is one of the most cited ones I think in economics that they show that if you ethnicity experienced slave trade 100 years ago, that decreases trust. And here we plot the real distribution with the algorithm and apart from the very trusting people who are probably just weird, we can kind of identify, it's pretty good, our machine learning algorithms are pretty good in identifying them. And the key finding here is that if we interact heat index times poor, so this is the difference in the effect between poor and rich countries, then we find that most of the findings are driven by poor countries, thank you. With the machine learning algorithm exactly the same, so within countries the effect is more, is about double the size for less trusting and for more trusting individuals. Second, we look at actions, so the research design is exactly the same, so we can go pretty fast here. We divide the country into 2,757 squares, and for each square we calculate the number of protests and riots per month, and we regress that on monthly anomalies. So if the month is hotter than the usual for that month in that square, and these are the maps, so again we plot the heat index on the left and protest and riots on the right, and we can see a pretty strong correlation, and the results seem to show that we don't have time to go through this, but we find exactly the same pattern as with the kind of daily data, so what seems to matter only is temperature in that month. Temperature the month before, two months before, three months before, and three months after does not matter, which is kind of, we found really reassuring. The last set of findings is we're looking at whether temperature affects actual and actual outcomes, so we digitized all presidential elections between 1985 and 2019, then we overlaid the GPS coordinates of the country with our climate data, and we calculate temperature in the country on election days, on the very day of election. We kick out, we only select countries that are either free or partly free by Freedom House because it wouldn't make much sense to do this in Zimbabwe, for example, where the elections are clearly rigged, so we kind of try to identify elections that are relatively fair. We get 96 elections, and we estimate the effect of a daily anomaly on votes cast for the government. We find that if the election day is hotter than the average for that month, that decreases votes by 3.8 percentage points, and in a weird way, it increases election turnout, so you might be thinking that, okay, these people don't want to go on vote because it's too hot, but if anything, it's not statistically significant, but if anything, we find it pretty strong, so it seems to mobilize people. Now, 4 percentage points seems very, very large, but in our sample, it's not, so one standard deviation is about 6.5 percentage points, and in only 6% of African elections was the distance between winner and loser less than 6%. So although it might seem a very large point estimate is actually in the African context, it's not that large, which I think is intuitive. So to wrap up, we believe we're the first study to show that higher temperatures will trust increased protests and even affects electoral voting outcomes, so we looked at attitudes, we looked at actions, and I think it really fits with what Grazia said before that although we know that there's a strong link between climate and conflict, we don't really know the mechanisms through which that works, and maybe attitude could be one mechanism that people haven't thought about so far in which we kind of think is intuitive. And again, I really want to stress this. It's very important from a policy point of view that if you are either in a rich country or if you're satisfied, temperature doesn't affect anything, no effect whatsoever. If you're dissatisfied or live in a poor country where people are likely to be dissatisfied, then that magnifies, and that's very important for, it seems that climate change is gonna have a disproportionate effect on disadvantaged societies. Like always, the disadvantaged ones always suffer more and climate change doesn't seem to be the exception. I hope I was on time. Thank you so much.