 What I'm presenting today is our contribution to the volume that Jean-Philippe talked about. This is joint work with Nishta Kochir at Georgetown and Biju Rao at the World Bank. I want to begin in a conceptual place. You know, there's a lot of discussion of the Indian caste system. It's a well-known system of social stratification. But there's actually no universally agreed-upon definition of the caste system itself. The word caste is actually a Portuguese word that began to be used in the 16th century to describe what Portuguese travelers saw in India. Right? So in Hindu texts, ancient Hindu texts, you see the word caste described as Varna. The Varna categorization was a division of society into four groups where you have the Pramins, where the scholars, if you will, the Shatriyas, which is the ruling class, the Veshas and the Shudras and those outside the system. Today, and for the past 80 or more years, what you have is the caste system defined in very broad strokes by the government of India. Since about 1935, you have terms like forward caste, backward caste, scheduled caste, or scheduled caste, as we say in India, or scheduled tribe. Right? And what you have is these are broad groups whose in these groups changes. So when the in the Government of India Act of 1935, you see 429 groups included as SC. Today, you have more than 1200. Right? So what most of us are used to doing as economists is running our regressions controlling for SC and ST. And what we're going to try and do today is tell you in the context of gender inequality, why this is not entirely okay. Right? We don't really have a solution. But the point is most sample surveys of India, most household surveys, restrict information to caste to these very broad government defined categories. And so our understanding of caste is limited to these categories. But if you go to India, caste is experienced in everyday life as something very different. Right? So what you have is essentially the operational concept is jati. Right? A jati is a hereditary group, endogamous. Right? So it's an endogamous hereditary group or hereditary occupation within groups. Right? So what you have is essentially a group that has a common occupation that marries within the group. They have, you know, rituals that are shared within the group, common diets and common lifestyles. Within a particular, there's several thousand jatis. There's no agreed upon way of ranking them. They are specific to regions and sub regions and they affect all aspects of life. Right? So marriage is well known, but also political mobilization, access to public services, credit, employment. I mean, there's a big literature on the Indian caste system that has used jati as an organizing principle or definition in at least some recent work. Okay. Now there's a large empirical literature that argues very persuasively that the caste system is an enduring form of inequality. That it stratifies Indian society and that some groups, and again, because government surveys use SC and ST or these broad categories as definitions, our understanding of caste is that people within these groups continue to be disadvantaged relative to the broader population. Right? And what you see is a relationship also between caste and gender, where higher caste or higher caste as used in these broad group, broad categorizations is associated with lower status. So what you see is probably typical of many settled agricultural societies, which is the higher up you go in the caste hierarchy, the less likely you are to see a woman employed outside the home or experiencing independent mobility or having decision-making autonomy. So an inverse relationship between caste status and, you know, between gender empowerment, if you will. The problem is a lot of this literature focuses on these very broad characterizations. Right? So what we want to do, and what we do in this work is basically look at three states of India. Today I'll just talk about one to save on time. We actually look at detailed data on the Jati level, right? Which is Jati is oftentimes called subcast. That's not quite right. I think it's fair to call it a subcast, but it's a lot more. It's not easy to rank. It's easy to rank broadcast. It's not easy to rank subcast. So we collect data at the Jati level and we try to look at the relationship between Jati affiliation and, you know, female empowerment. And we actually argue that depending on how you define caste, you get very different results. So when you use broad categories, you see patterns emerge that are very standard in the literature, which I just told you an inverse relationship between caste status and female empowerment. But when you look at the Jati level, there's a lot more going on and it's much more nuanced than that. Right? Our methods are relatively simple. And lastly, we're going to show this affects public policy. It affects take up of public programs. Right? So, I mean, at full disclosure, our work has some limitations. Our data is going to be baseline surveys for large anti-poverty programs in India, which are targeted to rural areas. Our data is representative of poor rural populations, not state level populations. And lastly, we do this is descriptive work. We do not claim any causal relationships. We are purely looking at associations between gender and caste, right? Hopefully robust associations. So as I said, I'm going to focus on one state, the state of Bihar. In the paper, we have three states. This was a baseline survey and a follow-up survey. In the baseline survey, we have about 8,000 households, and then there was a follow-up, which had about the same. 180 grand panchayats from 16 blocks in seven districts, right? Over samples of SC and ST populations, because that was part of the research design, and households were randomly selected from within Hamlets. Okay, this is our distribution. Sorry, this still has the three state data, but you can look at the left column for Bihar. We basically have, sorry, I'm going to... You basically have the districts that we sampled, and you can see there are seven districts over here. And then you have the breakdown by SC, ST, OBC as other backward caste, EBC extremely backward caste, Muslims and forward caste. I will not expect anybody to look at our breakdown of Jati. I'll talk more about these specific groups in a couple of minutes. So very quickly, I'll skip this. This is our summary stats of some basic household characteristics and our female respondents. I can come back to this. And what I want to emphasize for now is we looked at three different types of indicators for women's status, right? So the first one we looked at was measures of intra-household bargaining. So women were asked if they are able to provide inputs into purchasing household durables, kids' education, their own livelihoods, and then voting. We also look at female mobility. But the one I'm going to focus on today, again, in the interest of time between three states and so many different outcomes, I made a hard choice of what to show you. I'm going to focus on female labor force participation, which we define as a dummy variable that takes value one if a woman participated in any labor outside the household in any season, at least one season of the year. So that was our definition. And we ran reduced form regressions of those indicators on a whole bunch of controls, which include expenditure, expenditure squared, land, household size, gender of the head, et cetera. And some individual controls, and we put in panchayat level fixed effects. First, we run our regressions using government categories. It's a lot easier to look at this visually. So these are regression coefficients on the box, the blue box in this slide. And what you see is in the top left-hand side figure, this is again the state of Bihar. All the controls are in there. I'm just showing you the coefficients on the caste variables. So on the left-hand side, SC scheduled caste is the excluded category. And what you have is ST, OBC, EBC, Muslim and forward caste presented to you on the slide. Roughly, if we rank casts in this area of Bihar, SC would be at the bottom, right? If you accept a kind of ranking of broadcast groups, SC would be at the bottom. And all of these would be ranked higher. ST would probably be at the bottom as well. So what you see is the way we interpret this is that zero is the middle red line over here. All our coefficients are to the left. So as you go up the caste hierarchy, you can see that female labour force participation is falling. This is that classic relationship. And Bhojrab wrote about this. A lot of scholars of India have written about this. The evidences comes from not just empirical surveys, but from religious texts written a long time ago. That as female status improves, their participation in the paid labour force goes down. This is also consistent with a backward bending supply curve for women, right? If you look, this is Orissa and Tamil Nadu, which I will skip for today. Next, what we do is we look at the same relationship at the level of jati, right? And remember the conclusion I gave away that you'll see a lot more nuance when we look at the jati level. First, we're going to exclude, we're going to treat the higher ranked caste as the omitted category in our OLS regression. And in the bottom table, I'm going to treat SC as the excluded category. And again, I'm going to try and show you this visually because you can't read 10 point font even when it's blown up on a screen. What you see over here is on the left hand side, now this is all Bihar, right? All of this is Bihar. What you see is these are the different jatis that are classified as SC. So in the state of Bihar, you have the Chamar Jati, the Doba Jati, the Dushads, the Musahars. These are distinct jatis. They live in, you know, they are not just confined to the state of Bihar. The Chamars, for example, are in this broad area of India. Then Bihar is only one region where they're found. So these particular groups could go into other states. But within the state of Bihar, they are represented and they're in our sample. And you can see that some of them, like the Musahars, the Dushads and the Chamars, the women from these groups are more likely to participate in the labor force than, say, other SC group women. Right? So there's a fair amount of variation within SCs. When you go to the right hand side figure, now we have SC as the omitted category. So we're looking at non-SCs and ST jatis. And again, you can see there's a fair amount of variation. Brahmins are about, you know, 34% less likely to participate in the labor force than their women counterparts in the SC population. And you can see that for others, it's lower, right? So EBC, extremely backward caste jatis, it's nowhere near that amount. So again, these are statistically significant effects, but they show considerably more nuance than the literature would predict. All right, so I've said this verbally, so I'm just going to skip it. We can come back to it. And in the last five minutes I have, I want to show you why this affects or how this might affect policy, right? So India has some very large scale poverty alleviation programs that are probably well known in this group. So we take one of them, you know, the MG Norega program, and we look at our dependent variable being the possession of a job card for access to this program, right? As a second outcome, we look at participation in a female livelihoods program for which this survey was a baseline, right? So this is the World Bank livelihoods program broadly defined in India under the umbrella of the national rural livelihoods mission, right? So we're trying to look at is take up selectively different among these different groups. And again, our list of controls. And by the way, I should have said this before in the paper we show our results are robust to the exclusion of certain controls or the inclusion of different controls, right? So, okay. So first and Rega, and again, this is all Bihar. So what you see is that ST. So on the left over here is again, SC is the excluded group. And you can see that all other groups are less likely to possess a Norega job card, right? And again, this is the woman's sample. So women are less likely to possess a Norega job card. If you look at the second panel, this is presenting the SC Jatis. So again, the non SCs are excluded group. And you can see even within SCs, Dolbha's, for example, are less likely to have a job card than the Musahars. And we're controlling for things like Panchayat fixed effects and which might affect rollout of the program, right? And when you go to the upper costs, you can see even there, there's a fair amount of variation. Women's, for example, are much less likely to participate than other forward costs. If you go to the Jeevika program, which is again the base, the program that this was a baseline for, you see very similar results, right? That SCs overall are the ones participating within SCs. You see the Chamars and the Dushads are more likely to participate. Now you see Musahar women are a little less likely to participate than they were, say, in the NG Norega program, right? And you see similar variation over here. Rajput women are really not participating relative to, say, Yadav women, right? Okay. So I'll conclude what very broadly speaking we can conclude from this. And remember, our samples are not representative of the state. They're represented of rural populations that are most likely to benefit from these large-scale programs. What we think is going on here is, or we can conclude, is when we use broad government categories, you see simple patterns that have been documented in the literature, but these hide many details of the reality of how cost matters in day-to-day life, right? When you focus now doing this right is very difficult because most surveys do not gather JATI level identifiers. It's not really legal to question people on JATI when you're doing large national surveys. But we did it, right, using our data. And what we find is that there's a much more nuanced relationship between cost and gender than you see using these very broad categories, right? So that's about all we want to say. Remember, these are not causal claims. In the paper, we also do other things like we interact cost with expenditure to see, you know, is there a creamy layer as cost, you know, are wealthier women within groups more likely to benefit, and those results are all in the paper. And those results also are very nuanced, right? We don't, we see considerable variation in those patterns as well. So I'll stop and turn it over to my next colleague.