 So this topic is joint work with Valerie Mueller at Arizona State, Shaya Dow, University of Maryland, and Clark Bray at the University of North Carolina in Chapel Hill. So what's the motivation for our paper? Climateologists, climate scientists expect Africa to experience warming in excess of two standard deviations in the in the coming century. And we know that heat stress has adverse effects on agricultural productivity, particularly there's a lot of evidence to support this and also perhaps in there's some emerging evidence that heat actually can adversely affect productivity even in non-agricultural sectors. And so as adaptation is a key component of UN FCCC agreements and development assistance in dealing with climate change. One key component of that is how workers will adapt to to change in temperature. And this worker response is not very well understood, and particularly in Africa. So our research is attempting to fill that gap. So what do we do? We take individual panel data from the Living Standards Measurement Survey, so it gives us 55, approximately 55,000 person years of data, working ages 15 to 65. And we look at participation in seven different occupations over the previous 12 months. So we divide these into by sector, agriculture, non-agriculture, whether they migrate, whether they were in school or whether they were unemployed. And then within agricultural and non-agricultural, we divide that between wage employment and self-employment. It's important to note that these activities are not mutually exclusive, so the surveys tell us basically what have you engaged in one of these activities in the past 12 months. So many people participate in multiple activities. Also migration that we look at is not permanent. It's basically did you migrate out of your home in the past 12 months. And our data comes from four East African countries, Malawi, Uganda, Tanzania, and Ethiopia for the years indicated up there. So we take this individual occupation data and we merge it with climate data taken from NASA's modern era retrospective analysis for research and applications. Basically, it's a pretty fine resolution climate data. So we take the mean of monthly values over a 24-month period preceding the interview month. This 24-period allows for lagged effects so that, for example, they're extremely high temperatures. A year ago that could affect the current decision, so we allowed for that to happen. Then we derived these scores, basically, the mean over the standard deviation to characterize deviations in climate relative to the historical expectations. Basically, all of their previous 24-month periods in the previous 14 years. So we do all this with our goals to see how extreme temperatures affect worker behavior. So why do we do this? So this will help us anticipate where the needs are for climate adaptation resources are likely to be highest. So we want to be able to answer questions like, do increasing temperatures lead to productivity shocks that could provoke rural outmigration? So there's considerable literature looking at whether we expect people to leave the countryside to go to the cities. We'll cause a shift from agricultural to non-agricultural economic activity. We'll cause a shift from self-employment to wage employment. So, for example, if extreme temperatures lead to loss in livestock or farmers need to basically eat their seed, will they be able to, will they be less able to be self-employed and force them into wage employment? Will these climate shocks cause rural unemployment? These are the type of questions we want to be able to answer. So our theoretical framework is based on a standard utility maximization problem. So workers allocate their time across various activities to maximize their utility from consumption and leisure. And I won't go through the math here, but basically this enables to get us some disorder conditions which we then are able to use in our econometric analysis. And one result from the theoretical analysis that's new in this literature is that we find that relative, not absolute climate productivity impacts determine the time allocated to each activity. So what do I mean by that? So suppose that a temperature shock adversely affects both agricultural activity and a non-agricultural activity. That doesn't mean that the amount of time a worker employs in those activities will necessarily decline. Why? Because it matters which activities hit hardest. So if the agricultural activities hit more adversely than the non-agricultural activity, you might see an increase in working in the non-ag activity even if climate adversely affects it because of this substitution between activities. And so that also has some other implications that you may have nonlinear effects of temperature shocks on participation decisions. So a small increase in temperature could lead to a decline in say agricultural employment, whereas a larger increase in temperature could cause it to go back up. And that's what this graph here is basically showing is due to this substitution between the two activities of the straight lines represent the marginal impact of temperature say on two activities. It shows that at a certain point you'll switch from one to another. So the basic point here is that you need to allow for nonlinear impacts of climate on these activities. So you need to allow for nonlinear impacts of climate on participation. And another conclusion that we're able to show is that so that conversely means when you're looking at participation data and you're trying to infer whether climate is having a negative impact or not, you can't take as evidence that climate is having adverse impact in a particular sector in isolation. Only changes in overall unemployment give you concrete evidence of an impact. And this is a bit different from what we've been seeing in previous literature where there has been literature showing a nonlinear impact of climate of temperature say on productivity. But basically it's looking at dividing the world into relatively cool countries and relatively warm countries. So the idea in this previous literature is that if you have an increase in temperature that can be beneficial for colder countries like in Europe, but there's an adverse impact on warmer countries, typically in the south. And our results are showing that you can have nonlinear impacts like this, even in warm countries that you would expect from this previous literature to have a monotonic impact. And then our final result, which I won't dwell on too much here, is how our data is participation decisions. The theory is in terms of continuous hours in a particular activity, and we show how to basically translate the theory from a continuous decision of how many hours you're going to work in each activity to something that better matches our data, which is simply, did you engage in this activity in the previous 12 months? So our basic data, so important thing to notice here is that wage labor is not very common either in rural or urban areas in the agricultural sector. Second thing to notice is there's a lot of agricultural self-employment, obviously in rural areas, but surprisingly, well maybe not surprisingly, but 51% of respondents in urban areas are engaged in self-employed agriculture. Nonagricultural activity much lower in rural areas and urban areas, but still it's dominated by self-employment. Migration is roughly consistent around 10% across rural and urban areas, and schooling is a little bit less. Also these years that we're looking at, so these climate variables indicate that the years in our study were slightly warmer than previous years and with less rainfall. So our main regressions look at a linear probability model where the left-hand side is an indicator of whether or not the respondent engaged in the given activity, and then the right-hand side variables are basically our climate variables temperature and rainfall, and basically the square terms are the way we capture this non-linearity that I mentioned in the theoretical discussion. We also have individual fixed effects. We have time trends. We run various robustness checks with common linear time trends, linear country time trends, common quadratic, a whole host of different time trends, and this doesn't really affect our results are robust to these variations. We cluster our standard errors by the enumeration areas of the surveys, and we use sampling and inverse probability weights to account for attrition, basically for people that drop out of the survey over time, and all our results are robust to this. Finally, since we have seven outcome, basically these are seven different regressions, statistically that increases your probability of getting a significant result just by luck, because the more regressions you run, the more likely you are just to have something. We adjust our inference for these false discovery rates. Don't squint too hard here, I'm going to zoom in on these, but just to give you a flavor of where the results are strongest, and I guess it's just too blurry to see anyway. The important thing here is that our big results are in the middle, which is on the non-agricultural self-employment. We see a strong decline both in urban and rural areas. The other thing to note is the other big effect is on the far right, which is unemployment, which we have a strong positive result in urban areas, and migration also is strong. Anyway, I'll zoom in on these. Our first set of results is that high temperatures cause a decline in agricultural wage labor, but as I mentioned earlier from the descriptive statistics, although this is statistically significant, it's not very important just because so few people are involved in this. We do see a decline in agricultural wage employment both in rural and urban areas. We see at high temperatures a decline in urban out migration. Sorry, with these graphs, on the horizontal axis we have our temperatures, these scores, and on the vertical axis is the participation rate. Going back to the wage employment, basically as temperature increases, we see a decline. It's roughly linear in urban areas, and it's quadratic in rural areas. And these are both statistically significant. For migration, the box indicates results that are statistically significant. So in urban areas, we do have a quadratic shape, which means that at high temperatures we do see a decline in urban migration. Whereas in rural areas, there's a slight increase at high temperatures, but it's not statistically significant. Breaking down the migration by gender, we find that the migration results hold basically for male migrants, for male respondents, but they're not significant for female. And we think this could be the fact that a lot of female migration might be due to non-economic reasons for going to marriage or dealing with family, or visiting family in other regions, whereas this might not be so prevalent for the male. Okay, our big result is on non-agricultural self employment, where we see, this is a surprise to us that with extreme temperatures we see a pretty sharp decline, significant both in rural and urban areas of non-agricultural self employment with temperature. And then the other is that in urban areas, we see a pretty big increase in unemployment at high temperatures, but not in rural. So putting this together, this is a bit puzzling for us. First of all, our priors going into this was we expected to see the big impacts in rural areas, not in urban areas, and the big impacts in agriculture, not in non- agriculture. If anything, we expected to see a shift from agriculture to non-agriculture, whereas we actually see a bit of the opposite. So why do we see unemployment in urban areas? So in our theoretical model, we do a bit of an extension showing how, since we have multiple activities, there can be a backstop activity. So basically where the idea is if you have, say, some plot of land, you have access to for agricultural production, even if the temperature adversely affects your productivity, you may stop selling agricultural goods for cash, but maybe still work your farm for subsistence. In which case, despite the adverse productivity impact, you're always, if you have access to land, you're always going to work even in extremely high temperatures, at least a little bit, because remember our surveys are did you participate in this activity at all? Not how much did you participate? So if in rural areas, you have people have more access to land and they're able to work on subsistence level, but in urban areas you don't, this could be a reason why we see the upshot in unemployment in urban areas, because they don't have access to this backstop activity. Okay, and so that's our hypothesis. So again, just to substantiate that, now looking at the data, we see that in urban areas, in general, both urban and rural areas, agricultural self-employment is not sensitive to temperature. So you see these are pretty straight lines and they're not statistically significant. So what this tells us is this hypothesis about some, at least some, subsistence being non-sensitive, subsistence work in agriculture not being sensitive to temperature is held up by the data, and also that in rural areas you have much higher access to agricultural self-employment than in urban areas. So to further delve into this puzzle a little bit deeper, then we wonder, is there some sort of barrier to entry to agricultural self-employment in urban areas? We can't observe this directly, instead we divide a sample into two groups. So in one group is respondents that report having worked in agricultural self-employment at the same time as working in some other activity, and we call those people, these people by definition have access to agricultural self-employment, and then people who never participate in agricultural self-employment. And so our question then is if we think that access to, or lack of access to agricultural self-employment is what's causing the unemployment rates, then we would expect a differential response to temperature between these two groups. If on the other hand access to agricultural self-employment was not a barrier, then both groups should have an equal probability of becoming unemployed. So once we divide our sample into two groups, we find pretty stark results that the probability of not being employed is a function of temperature for, even for urban areas, for people that have access to agricultural self-employment, unemployment is not sensitive to temperature. But for the people that do not have access to agricultural self-employment, we see a very significant uptick in unemployment as temperature rises. For rural areas, we get a similar outcome. For the rural people that have access to ag self-employment, again there's almost no impact of temperature. These are not significant. There is a slight uptick, but that's not significant either. The only one that's significant is the urban people that don't have access to ag self-employment. Okay, so then the question is, so that's our puzzle about why unemployment goes up in urban areas, but not rural areas. Then the other puzzle is, why do we see a decrease in non-agricultural self-employment? So again, we expect there to be an adverse productivity shock of high temperatures on agriculture, on plant output. So again, our prior would be that would mean people would switch to non-agriculture activity, but in fact we see a decline. So in order to delve deeper in this puzzle, what we do is we divide non-agricultural self-employment into two subcategories. One that uses agricultural inputs intensively and one that doesn't. So think of maybe vendors, people that buy agricultural products from the countryside and resell it in the city. And so our thought here is that for these type of activities, the agricultural inputs are highly complementary to labor. You can't substitute more labor for fewer vegetables if you're a vendor. And so when we divide up the non- agricultural sector in these two subgroups, depending on how they use, how intensively they use agricultural inputs, this bears our hypothesis in that, so on the left hand side is urban area, the right hand side is rural area in both rural and urban areas. The solid line is non-agricultural self-employment that does use agricultural inputs and the dash line is ones that don't. So we see as temperature goes up, a very striking decline in self-employment and those activities that use agricultural inputs and actually the opposite, an increase in non-ag that does not use agricultural inputs. So there appears to be some degree of substitution there within the non-agricultural sector but not enough. So on the whole, once you put these two groups together, non-agricultural self-employment declines with temperature increases, extreme temperatures. So in conclusion, at high, so our story is at high temperatures we see an increase in urban unemployment. And part of, oh sorry, I didn't dwell too much on the migration which I should have given this a migration conference, but our migration result was that in urban areas we actually see an increase in migration at moderate temperatures. So and a decrease as temperatures get extremely high. So what's going on here? We believe this is consistent with other literature that's found that actually a lot of, instead of inducing, high temperatures inducing migration from the countryside to the urban areas, we find that at moderate temperatures, at good growing years, you see a temporary out-migration from cities to the countryside for seasonal work, working in agriculture and then they return. So when you have extreme temperatures that causes a drop in agricultural productivity, which means that there are fewer opportunities for urban people to go out and work in the countryside and that contributes again along with this drop in non-agricultural self- employment, this lack of migration opportunities increases unemployment in urban areas. So we see this reduced migration and that in combination with this non-agricultural self-employment reliant on agricultural inputs, together these two factors cause an increase in urban unemployment. But we find rural unemployment is unaffected by extreme temperatures and that we believe is due to this fact, the presence or the availability of this backstop activity of subsistence agriculture. Although even in rural areas we see that there's a decline in employment in non-agricultural self-employment, but again, since people are also they're employed in multiple activities, they can still remain working in this agricultural self-employment sector. So again, our empirical results are consistent with this narrative of agricultural self-employment being a backstop application regardless of temperature. So we have reduced demand for agricultural wage labor and temporary urban migrants, reduced demand for labor in sectors which is a combination for agricultural inputs and reduced employment in urban areas since relatively little access to ag self-employment. So our policy implications at its temperatures in East Africa increase the global climate change. We may not see this expected migration from rural to urban areas, but actually reduced migration from urban to rural areas. We may not see a shift from ag to non-ag self-employment and we see an increase in unemployment and the attendant social disruption primarily in urban areas. So there may be a greater need for adaptation resources in urban areas that we would have thought looking at the agronomic literature of climate impacts on agriculture.