 Hello, everyone. Thank you for being here. I'm Victor Chirard and very much looking forward to your thoughts on this work where I'm interested in the question of whether affirmative action can trigger sabotage. Why did I get interested in this question? Well, the idea is that if you have sabotage, that is a form of backlash as an answer to affirmative action, it actually would undermine the purpose of this affirmative action. Very broadly speaking, affirmative action. The idea is getting quite widespread, focusing on electoral quotas alone. That is a topic of my talk today. You have them in more than 100 countries worldwide. And it's a very efficient tool if you want to change visibly and immediately an outcome of interest, say the number of people of a locust that are at the head of an electoral council. The problem is that it's also controversial and in particular from the view of the people that do not benefit from it. By the very nature of affirmative action, you want to go beyond non-discrimination, so you push forward a group. The idea of sabotage is that you have two agents and one is willing to pay a bit so that the other one will lose more. So that everyone loses in the end of the day, but the agent who paid a bit improved its relative position. And there's, unfortunately, quite a bit of evidence that there may be sabotage taking place after affirmative action. It's an idea that theoretically grounded that has been proved during a lab or lab in the field experiment. One of them performed by Joseph, he's the next speaker, and also in horse races. I like that this was published in an equan journal. And the question I'm asking today is whether you can try to capture that with quote, unquote, real world later. So I'm kind of bringing this question to the nationwide policy of affirmative action for electoral councils in India. And of course what it means that I will be in a less controlled environment than what you can have during a lab or lab in the field experiment. And there's a trade-off between this and the fact that then I can use this administrative or household survey data that are also nationwide. So administrative data on caste-based crimes that span all of India and survey data, that's also national. And the conclusion is quite sad, since all results I obtained with this data are consistent with sabotage taking place. So the main contributions would be three-fold. First of all, the risk of one-size-fits-all logic. Some of you may have in mind this very nice paper by Iyer and Co-author, where they consider the case of gender electoral quotas in India and show that it increased crimes against women, but it's actually a good news. Because when you look at the way the pattern of crime increased, it's consistent with an empowerment effect. Unfortunately, in the case of my result, it's the same country, it's the same policy, but along the caste dimension, rather than the gender dimension. And point by point, my results are the opposite of theirs. So it's really like pushing forward all these ideas that in institutional work you need to be careful about what is the detail of implementation. It also confirms this possibility that affirmative action may trigger sabotage with this nationwide data. And a third methodological contribution is the idea that by combining this household and crime data, I can push forward the idea that caste-based murders in particular are a good proxy of what is the extent of untouchability practice or caste-based discrimination in India today. So that's it for the general introduction. Now you know more or less everything about the paper. Let's go a bit more into the details. First of all, the question of caste. So caste have quite a few important features very broadly for today. The first one is a sort of persistence because caste is an inherited identity and you have important segregation not only in the labor market, but also in the marriage market. So if you look at survey data, you have over 74% of respondents to a survey by Banerjee and Koso who says that they want to have a within-jati wedding. And Kassan reports that this within-jati wedding would be actually as high as 90%. What it means that of course then you have a persistence of your caste. And what I'm talking about today will be more the administrative grouping. So I will in particular focus on the scheduled caste that group all the jatties that used to be considered untouchable. So that would stand at the end of the kind of like social status ladder forecast. And this scheduled caste group alone is 16% of the Indian population that makes it 40 times the population of Finland. Unfortunately, still today this like caste distinction and even using the very broad caste categories is a marker of social economic inequalities. If you look at poverty rate among people of the SC, it's more than twice the poverty rate among people of the other caste. You also have like various accounts and survey data of exclusion from public goods in 50% villages. Members of the SC would be prevented to access a water source. In the Hindi belt, you have 44% members of the SC that cannot enter some streets in the village because of their caste. So it's still today very important question. And the Indian state is trying to acknowledge that and fight this persistence with a widespread affirmative action program. The affirmative action program I will focus on today is maybe the biggest of the latest one that is quoted in local political council. The idea is that you have this equivalent of town hall kind of and that on a rotating basis, the administration says that equivalent of the mayor needs to be a member of a given caste. And I'm in particular here interested in the members of the scheduled cast. And this policy was constitutional in 1993 but implemented in a staggered manner across the different states. The four states implemented before 1992, eight states between 1993 and 1995 and five states after 1995. And this staggered implementation would be interesting for me to see what is the effect of the implementation of the police. The other interesting aspect of the policy is that so it rotates within the states such that if your village this time doesn't have this quota for the mayor equivalent or members of the council actually, it can have it at the next election. And that's not up to your village or the caste relationship in your village that will decide it. And the last feature that makes it important is that this form of quota is particularly visible because it's for a political representative. So everyone in the village knows what is the identity of the political representative and in particular the guy who used to be a mayor or his friends. And the question then is what does it change? To try to assess that, I turn to different sets of data. And first of all, this administrative crime data, it's collected at the state level yearly since 1992. And it makes a distinction in recording since 1992, that is really interesting because you have a separate file to register cases of crimes when the victim is a member of a scheduled caste, the person who filed a complaint with the police because that's the data I'm using. The victim is a member of a scheduled caste and the perpetrator, the suspected person who committed the crime, is a member of a higher caste. So if you control for the share in the population, it can be a nice indicator of what is caste-based targeted crimes. And there's actually quite a bit of evidence that these caste-based crimes evolve with things that are related to caste conservation such as change in relative wealth or access to some water sources that are both having symbolic weight. However, this data doesn't have any advantages. It also has one issue, which is that because it's this administrative data, you need to have had not only a crime taking place, but also a person going to the police to report this crime and the police willing to record that crime. And to acknowledge for that, what I will do is actually reasonably standard in the literature that is putting a lot of emphasis on the merger statistic, because it is the one that is the least likely to suffer from the reporting and recording bias. Because if you have a burden, it's harder to hide. And formally, the way I will study this question is taking the law of the crime rates given the number of S.C. Households. And the way I am identifying this alpha one coefficient is with the standard implementation, so by including states and year fixed effects. The special crimes, it's offenses that are by themselves also symbolically loaded. So for example, preventing access to water. If you're in Finland and someone prevents you to access water, it may be like, yeah, whatever. But if you're in India, it actually has a symbolic sense. And all of these sort of offense have been considered by the Indian state in a specific record, that is the special crimes. And this during the S.C. Court that would increase by 300%. That various interpretation for this, so I displayed, I don't want to push it too much because it can be an improvement and increase in perpetration in record in report. It can also be a shift from penal crimes to the special crimes. There's quite a number of crimes that actually you can consider as both. So if you have some sort of improvement during court, that's as a shift. What I'm more concerned and not to push here is the question of murders because murders, there's like no margin of interpretation if there's a dead body. And this is then very consistent with a backlash effect or sabotage taking place. All the more is that results are actually inconsistent with misrecords. So you can say, okay, there's a body, but you can actually record this body under different heading. It could be a suicide. It could be a general murder, but they actually can show that this is not the case. Suicide or general murder do not react to the court police. It's only murder of S.C. Household. The same is true of saying, oh, yeah, but it's then a general increase in violence. This is inconsistent with the fact that you have no increase in the general heading for murders or the general crimes. And it's also unfortunately consistent with quite a bit of qualitative evidence. Maybe the most famous case is one where it's newly elected head of the no-call electoral council got murdered along with five other members of the Dalits or members of the scheduled cast. So that's for the state level evidence as an answer to this electoral protest. Then because this administrative data is still very aggregated, what I'm doing is turning to household level survey. Here you have the Indian Women and Development Survey that was performed in 2012. So now I'm in a cross-sectional setting, but with this like huge household level survey that I will focus on rural areas that are the ones that have the implementation of the quota policy. And it asks many questions on attitudes and perception. And what is really interesting is that then I can disentangle the sample between members of the scheduled cast and members of the non-scheduled cast and tribes to see how they react maybe differently to these quotas. Just to show you what this means, so here we have the repetition of murders in 2012 for members of the SCL assault. And here you have how members of the non-SCL assault, so higher cast, say that they practice intelligibility or that they don't want that a member of the SCST enters the speech, which is consistent with intelligibility practice in rural areas. So this is anti-constitutional for more than half a century. Still you have states where up to 60 to 70 percent people report that, yeah, yeah, they do it. So it's just completely forbidden. And you can see that that's some sort of correlation between these like state level average of the household answers. The correlation is actually 44 percent insignificant. The small matter is just to show that it does not really follow the share of AC household in the states. More formally, to check how this household answer reacts to the fact that at a given moment during the survey in that village, the Maya equivalent to Pradhan is a member of the scheduled cast due to a scheduled cast quota. I use this specification where for trying to pin down what I use is the fact that the administration typically allocates quota either at random or by using the share of AC household in that village compared to an area of reference. So it goes down the list according to the share of AC household in that village. So what I'm doing is controlling for district fixed effect and the share of AC household in the village. And then checking how household answer evolved depending on the fact that they remember of the scheduled cast or not. And results are here. So here you have for AC household and then non-SCST household. And interestingly, on the first two column on conflict and caste conflicts, they clearly don't have the same reaction to the AC quota. So non-SCST household that used to be more than one likely to have the political power in the village. When there is this scheduled cast quota, I think there's much more conflict and caste conflict. Well, also they're fine with it. And what is even more interesting is when you turn to this question of enterability being a team of members of the SC or practicing it for member of the SCST. And here they also say, oh yeah, I practice it more. So now you have this scheduled cast member who's the head of the local political council and say, oh yeah, yeah, I do more of this forbidden stuff of discriminating of the members of these groups. And here the number is not as precisely estimated, it's a coefficient, but the number is actually exactly the same. So I find it very interesting that you have this joint evolution. Then I can check for, oh yes, sorry. And this is of course entirely inconsistent with any empowerment effect, reporting effect, recording effect. If anything, if there's a change in social norm, this should have been negative. So it's like completely inconsistent with anything you do. You can then check what's happening in terms of trust, trust in institution, in the police, whether the maybe members of the SCST were still more likely to complain more to the police because they trust them more. This is not the case. The coefficient is zero or negative on everything. You can check still whether there's an increase in crimes in general that for some reason was not picked up in the administrative data, but in terms of how the household perceives things, there's more crime. It's not the case. So it's really only on this caste-based interaction that things change. A last extension of things that check is whether this backlash effect is due to the implementation of the quota by itself, or if it's like changes with the detail of how the quota is implemented. So is it the case that the moment of the election is particularly crime-prone? And it's not. Is it the case that with the variation in the size of the quota, because the size of the quota is supposed to mirror the population in states, even if the implementation is maybe not ideal? But for what I can read from the text and pin down the data, that doesn't matter. And also with the implementation of exclusive special courts. So you have these ideas to ease the access to the judiciary system in some places that are particularly tense. There's special courts that are designed such that it's easier for members of the scheduled cast to go and complain. And then the case should be handled faster. And that doesn't matter either. So the increase in murders really and in general actually also in the special crimes is really coming from the quota implementation by itself. That's it to conclude. Well the study is consistent with SCCOTA being met with kind of backlash, sabotage, to have this increase in murders, increase in declaration of practicing untouchability by members of the non-schedule cast in tribes. And so it's clearly impossible to extend straightforwardly to cast what are the results of year and course for gender. And Aflamatia election is at risk of being undermined by sabotage. To try to open up on a more positive note, affirmative action is a positive tool and a powerful tool in many dimensions. You have quite a bit of evidence on how it can change the action of the leaders, the minority members, the majority members in terms of norms, role models in many dimensions. I guess what I have in mind for future work on the topic is that it would be interesting to still keep in mind the risk of this backlash sabotage effect and whether how it's possible to then design affirmative action to try to limit it. Thank you for your attention.