 I guess I'll just comment a quick one on the last one. When you say that complementarities are desirable and wonderful, of course, that's a great thing, but how much? How big are the complementarities and what are you assuming? That's always a big question and one which surprises me that if we have a neutral case that allows, you know, at least a starting point that that would be a natural way of going and saying that they're complementary but they also could be substitutes. No one knows what's happening until you get empirical data. It's probably better to deal with it as if it were neutral. That's my approach and I was wondering if you could comment on that. Secondly, in all the assumptions, the results that you have, I mean, how are you scaling things at all and to what extent does that influence what kinds of results are obtained if there are a kind of, you know, robustness that you need to do? Could you tell me what types of robustness you do and what you don't do because I'm not sure from your presentation exactly what was going on there? So if you could just help me out with that. Thanks. Okay, so just to say that firstly there's a very first thing on that second issue with causality. I'm not trying to, I'm absolutely not trying to tell a causal story. There's no doubt that there's, for sure, there's reverse causality going on between, for example, the life satisfaction and some of the capability variables. I suppose what I think is, why I say it has been interesting is I think by controlling for the different capabilities, it's given perhaps a nice picture of transmission mechanisms. So I'm thinking of it more in terms of, I suppose, conditional expectations, if you like, rather than anything which is causal. I'm really not trying to tell a causal story, but I think it is interesting that some of them are so highly significant. And I think in a way, it could even be related to the fact that whenever people are answering questions about life satisfaction, that some of these deep things, meaning that, for example, in our home capabilities, to me, and one of the ones which comes out as really strong is the opportunities to feel valued and loved. Another one actually is being able to make ends meet, which has partly a financial aspect, so partly picking up an income thing, but it's also picking up the ability perhaps to manage one's income. So yeah, so that's supposed to, the first I'm not sure if I answered your question. We've certainly done other things such as reset tests, link tests, and so on, looked at the information criteria. But yeah, that's, we probably haven't gone into, absolutely, as I say, we haven't tried to tease our causality particularly. So that's the first thing. Regarding the complementarity thing, I mean, absolutely, point taken, how much complementarity should one have. I suppose really, what I think is nice about that, as I said, not directly involved in the paper, but I think it's a nice contribution is that it gives the possibility of, it's an added thing to one's toolkit if one thinks that there has, if one has reason to suspect that there might be strong complementarities or even any complementarities, it's a nice thing to have. But the other, the added of separability and so on is still there, and it may be a reasonable assumption depending on what the dimensions are. Great, any other questions? Thank you. I also have a question for you. Actually, if you could tell a little bit more about the data, how you collected it, and you're saying it's small, I think maybe I missed it, like how small and where it comes from if you did the study and how you chose. Well, we collected it actually using the market research company, Ugov. I don't know if people can hear me or not. Oh yeah, thank you. We collected it via the market research company, Ugov, who have an online panel. So covering, they can pick out representative samples from the UK and the US. The sample sizes are fairly small, so as I say, it is really just a pilot study. It's supposed to be illustrative, more than really informing specific policies in these two countries. So we've got just over 1,000 individuals in the US, about 1,700 in the UK. The reason actually it's higher in the UK was that we later collected a second web. We wanted to see if something interesting was gonna come out of that, so to build in room for attrition and still end up with about 1,000 people who did that. But we haven't analyzed that in this paper. One of the interesting things, actually, which comes out of that second wave is how stable capabilities are over time, like over a relatively short time, such as a year. But that's just a comment. James, another question. Hello, on the chronic poverty paper, multi-dimensional paper, did you use the same data set that we used, Maria Masantos? Yes. Okay, so one of the worries we had when we used it in our other approach to looking at chronic poverty, the one that's more like rebellion, but with imperfect substitution. Yes. Cross periods rather than looking just mean income across all the periods. Was that, the data set was very short in duration and therefore chronic was sort of a misnomer. We were using it just as an illustration, to be honest, because you're talking a year and a half, to what extent are you really finding chronic poverty or just variations in when they received income or wherever they might be consuming and so forth. I'd just like for you to answer that sort of a thing or variations in what other things might be happening in the short term. So I'm just curious about that. That was one of the things that vexed us and I'd like to back to you with it. Actually, I saw your paper on chronic poverty and using that data set and I also had illustration purposes with that data set. And using rotating penal data sets, we saw that maybe these analysis is more useful for maybe not a framework of chronic poverty, but now we're adjusting this to the study of long-term unemployment. And we're doing an index on labor unemployment and we're not dealing too much with chronic poverty. We were changing towards longer, longer unemployment. Okay, thank you very much. Daniel, it's just a comment more than a question. It's on Shapley decomposition. We spoke about it a bit. The problem with decomposing using Shapley, using say between H and A, the problem is that even if you're A, the intensity that you're saying, even if it does not change, it does not mean it has not contributed. So even when you are decomposing between H and A and suppose you see no change in average intensity, then you will conclude 100% of the change has been contributed by a reduction in headcount. The problem is that when A remains same, that means those who have gone out of poverty, their intensity has changed. Also, it remains same, which means that somebody who are mostly deprived, their intensity has improved. So the overall change in poverty has caused by intensity, but in this way, the way you are doing Shapley, you will be putting more weight on headcount, although it does not deserve that much weight. So that is one issue. When you are doing Shapley, you are making that kind of assumptions that you are going to be ignoring and sort of overemphasize on the multi-dimensional headcount ratio. I don't know to point that out. Yep, any more questions? Please. Thank you very much. Just curious about the force of paper. You kind of presented the anonymous approach versus the panel data regression master. I wonder whether you can, are you able to conduct a test for one against another? That's the first point. Second point, in the empirical exercise, I know you're not focusing on the empirical part. It turns, for both approach, a larger chunk of the decomposition, the share, say, more than 30%, it goes to the residual. You only identify, you have a list of variables, the only peak of one, then, which accounts for the 7%, then all 30% it goes to the residual. There's a lot of going on there. I wonder whether you can comment on that. Thanks. So let me see if I understood the first question was, whether there was a test. Well, actually, the way we think about it is no, because it's not that you are preferring one to the other. It's basically two different ways of cutting the cake, if you want, of analyzing this. You have, imagine you have a barrier data on income, income today and income tomorrow, and you can match individuals over time. What we are doing, so the anonymous approach looks at that data in one way, the panel approach looks at data in another way, so unless you clarify a little bit more what you mean by test, I wouldn't think that there is a preferred way, so you say, okay, I test it, hence I go with the panel, I test it, hence I go with the anonymous one. I can tell you, of course, if there are no positional changes, then, yeah, you should draw very similar conclusions with one and the other, that for sure. Just let me, since you give me a little bit the opportunity to explain myself a little bit more, when do I think, for instance, that it is important to use the panel approach? So many times people are interested in well-being and well-being over time, so I think that there it's very useful to count with panel data and break the anonymity and see how the destinies of the people that it started initially back fared over time. There the panel approach, I think it has an advantage. One instance that it's very popular right now, and it's why inequality is in the headlines now, where I think the anonymous approach is actually more relevant, is the relationship between inequality and political power. So all the discussion from Stiglitz and Piketty and so on and so forth that they claim, well, the super wealthy are in a better position to accumulate power and to influence policy. Well, in that case, you probably want to stick to the anonymous approach because it doesn't matter pretty much how you accumulated your wealth if you came from racks or you came from riches. If you are at the top 1%, you're probably gonna defend your 1% share. You're gonna defend your position very strongly. So in that case, I would say that the relevant tool to use would be the anonymous approach and that's why we think that it's not that one is better than the other, it really depends on what question you want to answer. You can use one or the other. And then about your second question on the residuals. Well, yes, I totally agree on these. Residuals have, you complain about residuals having a 30% share. If I showed you the results, there is another decomposition I didn't present that doesn't decompose the gap, but decomposes the level of average inequality or initial inequality. And in those regressions, the share of the residuals is doubled the size of that. So it's much worse. Is this bad? I think it is, but I think this is where we are in all of the labor literature, I would say. And I would say even the income literature. If you think about it, we are just using a set of standard observables and there is so much heterogeneity out there that doesn't get captured by your occupation, your industry, your education, your gender and so on and so forth that I guess is not surprising. In a Mincerian regression, R squares are 20% and you're jumping, right, of happiness. So it's not, and this method builds upon that. So it's not surprising, I guess, to me that the method leaves so much unexplained. I'm not saying it's satisfactory. Of course it is not satisfactory, but I think that any regression-based method with those limited observables as the ones that are collected in any standard survey is gonna leave a lot of the heterogeneity hidden. In fact, we were actually quite happy that the residuals only accounted for 30% of the equalization gap because by any account, that's small in the literature. But no, I do agree that that's a problem with these methods, yes. Please, yes, we have two and a... Right, thanks. Question for Lawrence, or maybe it's a little bit more of a comment. And it's like well-being and happiness and progress. The way you formulate the questions and the cultural sensitivity of that. Now, you did it easy for yourself by only comparing the UK and the US. But even, and there was some curious things in the results because the US score was always higher than the UK by roughly about 0.5. So I was just curious, yeah, but it's also the, in a sense, the whether you could use the same questions in completely different cultural contexts because it was an extremely sort of, all the questions started with I. It was the I versus the we, the collective unit. So in a less individualistic-oriented cultures, say outside of Europe, could you ask the same questions? Or if you ask the same questions, you probably needed to be very careful how you interpret the answers because I could easily see a number of the questions where people in completely different cultural contexts would be very uncertain how they would respond when it was only the I rather than we, the family, or we, the village, we, the community. Well, I mean, I completely agree with both points, actually. So, well, to take the second one first, well, in fact, no, let me start with the first. It's maybe more logical to begin there. So, yes, absolutely. I think that the, I mean, and there is, it's been discussed in a number, by a number of people now in the life satisfaction literature, whether the fact that in the Gallup poll and so on that life satisfaction on average is much higher in some countries than it is in other countries, and whether that's a real phenomenon or whether it's a cultural reporting phenomenon. And, I mean, I guess, you know, it's very hard to determine the answer to that, but personally I think that there's a very strong cultural aspect to it. And actually, that's why, initially, you know, we thought about comparing some of our data for the UK and the US. As you rightly observed, the US scored higher in everything. I mean, including lots of other measures that I didn't present. So, I would have absolute reservations about using, so, you know, in the very brief stochastic dominance results, I showed it was all within country, which I think, you know, it's probably not also without some question marks. I mean, do, for example, males and females use the same reporting skill exactly? Do different ethnic groups use the same reporting skill? But I definitely believe that completely different cultures don't, and that's why, in fact, we didn't compare them that way. The second point, so, following on from that, I think, yes, absolutely, the questions would certainly need to be tweaked for different cultures. The I versus we thing being an important one. And, you know, we have actually begun to think about how we might roll out in slightly different contexts. These were obviously two, for two, you know, culturally not completely different developed countries. Thank you. Another question for Lawrence. You referred several times to your sample size as a shortcoming. For those of us who work intensively with long-term panels, a sample of a thousand seems like a great luxury. I wonder if you could amplify the extent, I wonder if things, particularly when you're talking about comparing the US and the UK, I wonder if things aren't getting swamped in your sample size. I wonder if you could say something about where one might go with a higher intensity of that kind of questioning with a much smaller sample size over time. Yeah, I mean, that's another good question. So, I mean, as I briefly mentioned in response to one of the other questions, we did collect a second wave of data for the UK. And I think in some respects, we were disappointed with the results from it, but I think it's always easy to justify things with hindsight when you see how they turned out. But what we observed was that the capabilities within the panel were at an individual level, were remarkably stable over time, over such a short period as one year. Now, it would be really interesting, I think, however, you know, starting off with bigger sample sizes and allowing for attrition, you know, over a decade or 20 years or whatever. I think they would change over time, but I think the point is that a lot of the capabilities that people have are things which, you know, it's essentially human capital, and it's not something that usually changes overnight. It's probably, you know, there's a big literature now, of course, with James Hackman and Kunya and so on on the early formation of capital. I think a lot of it is developed probably in early childhood, so it, you know, it could be that in adult sample sizes even with quite long panels, they might not change as much as we think. So I don't know if that really quite answers your question, but I think certainly it would be really interesting to see how these things go over long periods, but exactly what'll come out of it, I don't know. I also have a question to Lawrence. I really liked your presentation, but frankly speaking, it's still not quite clear to me what's really different in your approach to the many other well-being approaches and why do we really need another survey on well-being, and I mean, I don't know where you are really leading to or where you see the value that you add compared to other well-being approaches. I suppose really the starting point in a way was trying to implement something as closely as we could to Sam's capabilities approach, which has very widely regarded to very desirable theoretical properties. So in contrast, let's say to life satisfaction approach, for example, one of the really key things being that it isn't just how content people are with their situation, but because there's all sorts of issues over adaptability and so on. So people might be very happy, but from an objective sense, they might be pretty destitute and have proof of freedom to do things and so on. And therefore, I think focusing only on the things like how happy one feels or how satisfied one is with one's life can lead to quite perverse balls at conclusions. So really it was trying to implement Sam's framework as closely as we felt we could do empirically. And that was that's what we see as a contribution, I suppose. Great, do we have any final burning questions? Okay, well in that case, I'd like to thank very much the speakers for a very interesting session and also to the audience for some good comments and questions and thanks very much and look forward to reading more about this work in the coming years. Thanks.