 Thank you very much. In this presentation, I am going to talk a lot about different methodological challenges of estimating intergenerational mobility and from the economics point of view. So this is a bit, you know, limited in scope and I am also going to follow a lot the outline of the chapter that we had done for the book but by this time while doing this presentation, I was thinking that it's almost time to write another one because literature really exploded. So the starting point of the chapter was that if you look at any development textbook, every textbook has a chapter on inequality but very few of those chapters talks about inequality of opportunity or inequality at the intergenerational immobility. There is not a whole lot of interest in it before, you know, this huge rise in income inequality that we had seen after globalization and following that, you know, there is quite a bit of explosion of literature on this. I should also mention here that much of the literature actually closely follow the literature that's being developed in the context of developing countries. So there are three issues that I am going to focus on for a very first one would be that what would be a good indicator of economic status. Of course in the context of developed countries permanent income is considered to be the most informative of all economic status. Chico in the morning told you why, you know, we can't really go down that road because most service in developing countries don't have income data, particularly for parents generation when parents are mostly employed in agriculture and informal sector. And then, you know, there are studies which relies on, you know, one round of survey or so on. These things leads to quite, you know, quite a bit of attenuation bias. So it would actually look like income ability is much better than it actually is. So here are some of the examples that I had put in there. I am just going to hand wave and go to the next. So in the case of developing countries, we had to come up with some indicators of economic status and what is better than education that is more or less, you know, widely accepted. You can also think of parents occupation. That's another and in fact, you would see that there are two parallel literature in developing countries, one focusing on education, one focusing on occupation. Here also the question is that, you know, you would think that both of these indicators are important as Chico pointed out in the, you know, keynote in the morning. Then the question is how do you combine them? There are ways to combine them. For example, DHS does this wealth index. This kind of, the problem with this kind of combining is that you can't really interpret the weights. I mean, it does not really have economic interpretation and you can't also link it to some, you know, economic theories and makes it difficult to interpret results. There is also the other approach that Luebowski and Wittenberg had suggested and there are a couple of papers on that. I am just going to skip that quickly also, you know, if you want, you can look at the details of that in the chapter. So next, I am coming to the measurement of it. So there are three, four different standard measurements that's being used in economics literature. The very fast one is that you regress children's education on parents' education and the slope coefficient gives you what is known as relative persistence or it would be like relative immobility. You can standardize both the children's education and parents' education run the same regression. You're going to get the Pearson correlation, which is the wrong one. And with the, you know, with the popularity of Rashid's work, there is also this rank-rank regression, which we are just calling it IRC. Basically, you take the rank of children's distribution and rank of parents' distribution and you just run one regression, which is given there. Now, remember that in developing countries, we are dealing with a very limited data. So what kind of problem you are going to be expecting there? First of all, you know, the measurement error. If there is measurement error, of course, we are going to have attenuation bias and this is going to look, you know, make things look rosier than actually. And there are some evidence of that in the context of developed countries. But for developing countries, particularly when we work on education, we kind of wave out the measurement error issues. I think maybe waving out for the children's generation is fine. I do have a bit of doubt whether that's fine for parents' generation. Okay. The second bias, and this is also going to cause attenuation bias, is in most context, actually, I should qualify as co-residency. So the idea here is that you are going to be using some sort of household survey data. And typical household survey data will have people who are resident in the same dwelling, which means a child who migrated for some reason, work, study, whatever, you're not going to see them there. That causes attenuation bias, which is known as co-residency bias. And we had done some work on this and basically we find that among different measures, the three measures that I had pointed out, IGRC performs the worst when you're looking at the estimate of the slope coefficient. Okay. The IGCI, IRC tends to do a bit better job. There are, you know, the opposite evidence, particularly for income. In the case of rural India, here is the paper for that. Now, given this, next question I am going to ask is that if we see that relative persistent in a country, one country is more than other, can we just say that one country has better opportunity for its children than other? Here I am going to take the example of China and Indonesia. The persistence estimate, that better estimate for China is .34, Indonesia is .78. From the face of it, it would seem like that children in China are having a lot more mobility. And here I grafted. So what you see is that for each of father's education, you see the expected son's schooling. And here from the, once I give you the picture, figure, you can see that actually in terms of absolute mobility, Indonesia is doing much better than China, right? There are a couple of other examples in this graph. There could be situation that in certain part of distribution one country is doing better than the other and so on. So, up short of all this discussion is that you can just look at relative persistence. You do need to look at the absolute mobility too. In other words, you also need to look at the intercept estimate. And there is a lot less evidence on, you know, how intercept estimate performs under different types of data constraints. But overall, it does look like the rank-rank regression performs much better. Okay. Now, here I'm going to qualify a bit. And this is not in the chapter. This is from the, sorry. Or maybe this is my old, okay. Before I go to the, so I have a slide missing in here. So let me just tell you what the idea there. So what measure, when are you going to use what measure? It depends on the context. What the rank-rank regression or IGC regressions are doing is that it is standardizing the distribution, parents and children's distribution in some way. In case of IGC, you are using normal distribution. In the case of rank-rank, you are using uniform distribution. And advantage of that is that you are actually, you know, kind of wiping out all the shifts in the marginal distribution for this. So if you are interested to know what happened over time in a country in terms of persistence, yes, those works better. But if you're looking at government policies, which actually the purpose is to move the marginal distribution. Think of improving school access. That's going to improve the marginal distribution for the children's generation. For that, you want to stay with the IGRC model. The other advantage of IGRC model is that it is, it can be linked to the economic theory very readily. And you can explore different mechanism through which policies changes absolute and relative persistence. Now finally, I know that I'm running out of time. There is the final concept that I talked a little bit this morning. So maybe I will just skip this and go into the interpretation of the parameters. Now we know that, you know, this is a classic case where there is omitted variable, which is that children's ability is not known. And we do know that parents and children's ability might, you know, correlate positively. And because of that, we are going to have, you know, some indigeneity biases in the estimate of the better one. And in other words, you know, these are not causal parameters. First of all, I should point out that if you're doing intergroup comparison, I would not worry about it. Is there a reason why, you know, ability of distribution for girls at birth would be different from boys? No, actually. So if we're doing that kind of comparison, we are fine. Now for the causal interpretation, you actually need some exogenous variation in parents generation. And there are a couple of new papers which looks at intervention like school construction that shifted mother's education and looked at, you know, how that affected children's education. So that's one approach. The second approach is basically doing some sensitivity analysis. And actually, Chico has a paper on this. Idea there is that if we know what is the correlation between parents and children in terms of ability, we can use it to bound the estimates and tell you, you know, how much it matters. Okay. Last slide. So that's the paper by Chico and his co-authors. I should also tell you that when you're doing policy impact, I think maybe you don't need to worry as much about this ability correlations. Because, you know, the distribution of the ability is not going to change because of the policy. I mean, because of the construction of a road or construction of a school. So that's another way that people have gone into. Finally, there is also an emerging literature right now which focuses using more of the structural models to give a lot more structure on different mechanisms. And those models also have a lot of degrees of freedom to, in fact, figure out what are these ability parameters are going to be. And on the basis of that, doing policy simulation. And that is my last point. Thank you.