 All right, so my name is Marika, and this is co-authored work with Brian Felt. Brian Felt is just an hour flight away at D.F. at Universidad in Medellin. Do we have a clicker? Yeah, there we go. Perfect. Thank you so much. All right, and we're going to go to the other side of the world to Indonesia for this paper. But otherwise, I think it actually follows very naturally on what we just saw presented. Across the world, and as I've learned over the last few days here in Colombia as well, there's considerable debate about migration, particularly the worries that migrants worsen labor market conditions at the destination, and that they increase crime and violence at the destination. The first one right here, that's a really large literature, especially so of migration into high-income countries, a little bit also across the world. The literature on the impact of immigration on crime is actually really small, and mostly focused again on high-income countries and on international migration. And if I summarize these papers, the overall story, broadly speaking, is that the effect of immigration on crime is really small, if at all. Some papers find no effect. Some papers find a small effect, and if they do, it's exclusively on economic crimes, not on violent crimes, but economic crimes. Okay, that said, that stands in stark contrast with the public debate around migration. I'm showing here, from a number of countries, people's agreement with the statement that migrants increase crime rates here on the vertical axis, and that migrants take jobs away from natives. And as you see here, it's a lot of variation between countries, but it's high in a lot of, in a lot of countries, very high numbers, very high worry about those issues across the world. So for this paper, we're gonna look at Indonesia, internal migration, which is quite different from the rest of the literature. Indonesia's a country with weaker institutions and higher crime rates, but also a very large informal sector that could potentially help absorb some of the displaced workers at the destination. Because we're talking about internal migration, so those moving within the borders of their country, big country, fourth largest country in the world, because of that, there's different kind of selection. You don't need a visa, you know, there's shorter distance kind of migration, so it's a bit of a different concept than we just talked about. So in this context, we use, we combine a couple of data sources, we get a really detailed, long-term panel of over 32,000 individuals on very detailed migration data, and we combine that with data from crime reports from over two million newspaper articles, local newspaper articles, as well as some survey data on crime. I'll say a lot more about the data very soon, but I also wanted to mention right away, many of you working on migration, migration is as endogenous of a choice as it gets. People choose to go at a particular time, they choose to go to a particular destination, so we wanna take that carefully into account when we try to establish a causal relationship, and we do so by following a previous paper of mine with Jeremy McRuder, whereby we instrument migrants out of an origin, migrants into a destination with weather shocks, rainfall shocks in our case, rainfall shocks in the origin areas, and that turns out quite strongly predicts who enters a destination. In this paper, that other paper is focused, indeed, on labor market impacts, and I wanna mention right away that we do indeed find some small but significant negative effects on the labor market of internal migration in Indonesia. A one percentage point increase in the share of migrants decreases income by almost a percent and employment by a 0.24 percentage point. So that made it indeed kind of mean that there's some displacement which may then spill over to crime and violence. Okay, so thinking a little bit right away about an empirical strategy, the underlying, the structural relationship that we're interested in is the one right here. We would like to regress crime rates at a destination at time T on a migrant variable, how many migrants there are at that location at the time together with some control and time and destination fixed effects. Clearly this regression, this simple OLS would be biased, right, like again people migrate to areas for certain reasons. The direction of the bias in this particular case is a little bit unknown. On the one hand you can imagine that migrants would avoid high crime areas, so they would go to areas where we have less crime, but you can also imagine that migrants just go to areas that tend to be urban areas and they have higher crime rates. So we extend T to bias on this OLS is a little bit unknown, but we gotta address it either way. So the way that we do that is by running this first stage relationship which is motivated by the earlier seminal work by Munshi that sort of, I think was the first one to combine these weather shocks in a migration literature. Case of Mexico-U.S., as I already alluded to the paper, here with Jeremy McRuder, that's the one that we follow exactly in this paper as well. So what does that mean? We, in the first stage relationship, we predict a migrant share using rainfall shocks at the origin. We do so with a bit of a lag, we try various lags but it turns out that they're sort of about a one year lag that's the best predictor on how long it takes for people to move. And then we take into account rainfall at the origin of all the origins that are part of this catchment area of a destination. So imagine that there's one destination and sort of historically, let's say there's two origins that feed migrants into that destination. And let's say that one of these origins is responsible for half of the migrants and the other one for just a quarter, then we would just weigh them disproportionately, a half, quarter, quarter. So it's fairly sort of similar in essence to like a shift share or a bar to kind of shock. So that's what we see explained here. And then the predicted migrant share we then use in our second stage to look at the effect on various crime variables as a destination and we cluster at the destination level. So that's briefly what we're doing. Couple of more identical identification issues to think about. There may of course be serial correlation in your weather variables. So you want to control for weather at the destination. And then as such, the identifying assumption comes down to the exclusion restrictions which states that if we control for weather at the destination, the only way that weather at those origins affects crime at the destination is through migrants going there. That would be the exclusion restriction. I have a couple of tests. I haven't actually seen a bonus site. I'd love to show more about these, but we actually feel quite comfortable about that. Exclusion restriction. We do need to know that of course we're talking about rainfall-induced migration. So that is the effect that we're estimating the effect of. These are clearly a local average treatment effect of those kinds of migrants. And also we're looking at short-term effects of them. Okay, a little bit about the data. We are using a data source that I've used for some of my other migration work as well. Very detailed migration data from the Indonesian Family Life Survey. It's been collected for over five waves now. And they're really known for very low attrition across waves. And that's really important as we study migration because those are precisely the ones who tend to retreat from a panel. So bringing it together for a period of 10 years, we create a longitudinal data set of over 32,000 individuals where we know from everybody where they are at each point in time, okay? We base it both on retrospective and current data on their location. And as such we define a migrant as somebody, we also have information on where they were born, and we then define a migrant as somebody who is in a place different from where they were born. This data set only picks up on moves that are at least six months in duration. So there's not really some of this seasonal or circular migration going on. That's not what we're picking up on here. And locations are defined at the sub-district level, Kachamatan, mostly internal migration. We will be collapsing everything up to the location level, including our crime data. So as such, for each destination year pair, we create a variable of the migrant share, which says how many migrants are there at that location in that year. So here is our data and the migration data is so very precise. This is for just one month with all the moves. Let me actually see if I can just click on this button here. Just click on it, yep. And then just okay. And that's actually my favorite way to show these data. So what you see right here is our data. This is exactly our migration data. Every line is a move. We see that the time going up here is a very long panel. Every line is a move from one person, I believe from the Ratu degree, but I'm not even sure, X, I should know that. Anyway, there's lots of moves. It's only about half of the moves presented here because for about half of the people we don't have their month of migration, but that's fine for us because we run everything at the annual level. But anyway, so that gives a bit of an idea of what the data look like. Yeah, let's go back to the slide, yep. So then let me show you a little bit about the, you should be able to just, yep. There you go. Let me then talk a little bit about the crime data. The crime data is coming from the National Violence Monitoring Survey that has collected data from over these two million newspaper articles. Really interesting, really detailed data. And we classify based on the description that they have about the crime, we classify them into 15 types of categories of the kind of crime that is taking place that we aggregate to economic crimes and more violently motivated crimes. Again, we aggregate everything to the district, to the sub-district year level and we define crime rates per 100,000 of the population. Okay, where did they did the last one? This is coming from University of Delaware, where again, we see the same map and we have little grids of about 50 by 50 kilometers that we're matching these folks to. So then a little bit of summary statistics. The top here is from the migration and rainfall data, you know, kind of what the folks who move around look like. And this is then coming, these are these crime rates. So the total crime is 14 per 100,000, which is a pretty high crime rate in the world. Economic crimes, violent crimes. And we also have detailed data on people being injured, assassinated, sexually assaulted or kidnapped in every one of these newspaper articles. So very detailed data. Okay, let's look at some of the results. The first result here that I'm showing is a simple association between various types of crime variables and how many migrants there are at that destination. This is not causal, this is just a simple correlation, the OLS, and as I already alluded to, it's kind of accentually unclear what it would, you know, how the bias would go, but we actually see that by and large the effects are positive, meaning that migrants tend to migrate to locations where there is higher crime to begin with. Okay, both total crime, economic violence, as well as some of these number of individuals injured. Again, this is not causal, so now let's use our instruments. The first column right here shows the first stage relationship of the number of migrants, or the percentage of migrants on our summed origin level rainfall term, our instrument, which is quite predictive as you can see in a negative way, meaning that if there's a bad rainfall shock, people are more likely to migrate away, a fairly high F stats, so this is the one that we use. We test with a couple of other lags, but we see that the one year lag is most predictive and we end up working with that one. The rest of these columns from two onwards, those are the second stage results, so that's where we use our instrument, and now that we see, unlike in the OLS, the only crime variable that still shows up are these economic crimes, okay? Meaning that there would be a 0.44 additional crime, actually 44, multiply by 44 additional crime per 100,000. To give you an idea, dividing these by the mean dependent variable, this is about a one percentage point increase in the number of migrants, which is about a six percent increase, leads to a 3.9 percent increase in economic crimes reported in newspapers. So a little bit higher than what we've seen in some of the high income country, literature, but not huge, and also importantly, nothing left on these violent crimes or some of these other indicators, and even not on total crime, okay? So, I have them split up, by the way, by various categories as well, in case there's interest later on. Then we wanted to think a little bit better, and I have to admit that this is a little bit more a preliminary result, but I thought I would present it just as a little bit of food for thought. We started thinking like, well, wait a minute, what are these newspapers really picking up on? And the main thing, and maybe some of you have been thinking about this already, could there be some kind of a bias? Some kind of a bias in what people choose to write about, and of course the main bias that we're curious about is could this bias be correlated with the share of migrants at the destination? Could it just be like, hey, I see all these migrants coming in, all these new people, and people sort of become more likely to write about those crimes? So, we said like, well, Indonesia's a data rich country. Let's combine our newspaper data with data from household surveys, the National Economic Survey, Susan Us, which is nationally representative, more than two million individuals over districts, and this is a very rich data set on the potential crime victim, because it's asking people whether or not they were a victim of a crime, and then if so, yeah, they ask, did you report that crime to the police? So, let's look at the results of these household survey data. It's a little bit different. If we simply ask people, this is already our second stage, it's already our causal estimate, if we ask people, were you a victim of crime, we see nothing. People are not more likely to say that they were a victim of a crime if we see more migrants coming in. This, of course, stands in stark contrast to a fairly sizable effect that we saw in these newspaper articles. Interestingly for us, there's one other variable that's not a lot of data, but there's one other variable in these data, which is did you report a crime to the police conditional on being a crime? Again, we see no effect on being a victim of crime, but we do see that people are way more likely to report that crime to the police after more migrants are coming in. Again, this is a causal effect of migrants coming in on IO crime victim, and migrants coming in, did you report it to the police? No more likely to be a crime victim, but much more likely to report it to the police. Again, it's a little bit of food for thought, we still wanna export this a little bit further, but I do think it's actually quite interesting, and it could potentially then importantly tie back into these newspaper reports, because if we're gonna report these things to the police, journalists may pick up on these things as well, and they become a lot more salient. So in conclusion, I'm thinking a little bit about that, we estimate the causal effect of internal migration on crime using these wetter shocks as instruments in Indonesia. Here's the effect that we found, one percentage point increase leads to a 3.9% increase of these economically motivated crimes in the newspaper articles, but no effect on reporting to be a victim of crime. And indeed, a very large effect on reporting your crime to the police conditional on being a victim. So that brings us sort of to my little bit of food for thought, the importance of really understanding your data sources, like what is everything picking up on, right? Like, could there be this kind of reporting bias in newspaper reporting? And again, could that be a function of the number of migrants at the destination? Becomes a lot more salient. And if it does, tying back to my initial motivation, may actually see that the initial discrepancy that I mentioned in these public opinions on the one hand and the very small negative impacts that we find empirically, well, maybe what's going on is that, sure, there's actually a small impact, but we do see them much more likely to be reported. And guess what? That's how people feel, right? These articles, these newspaper, that's what you read. That's the salience of, hey, there's all these migrants around. Oh, there's all these crimes being committed. So maybe a lot of what's going on here is just really the salience about these newspaper articles and the potential for migrants causing crime and violence to increase at the destination. All right, thank you very much. Looking forward to your comments.