 OK, so I'm supposed to be appraising these appraisals. And the task here, there are more of these appraisals. In fact, I even got one of the wrong hip. So I'll be discussing Andrea's Francoise and Stephen's work, which look at the World Top Income Database, the Seppelstatt and Sedlac databases, and the Swede and the Swede, as we've just all heard. I, Martin Ravallian, who's not here, did an appraisal of Lis, and in fact, as I'll explain in a minute, I was kind of involved in, I read that very carefully for a reason I'll look at. Some of the issues he raised I'll kind of briefly refer to, but I won't be discussing his appraisal here since he's not here to, he's neither presented it nor discussed it. But I should give a couple of disclaimers, essentially to say that I'm, to some extent, I'm a guilty party here. I've been the research director of the Luxembourg Income Study, Lis, which appears in many of these appraisals for the past nine years, including two major revisions of the database. So whatever's wrong with those should be partly blamed on me. When I was here, as Tony just said, I, in fact, advised the person doing the WID revision, and I also provided the data for the Lis data points for the WID. In fact, I re-estimated them. That was part of the thing we did in cleaning them. And I also, Steven makes a point about finished data being kind of a little over the place. I provided many of the series for the original WID, and I also did the estimates for the top incomes database for Finland. So I've kind of been both a data provider and in a sense of a designer of some of these databases, much involved of this, so many of the criticisms that have appeared kind of apply to my contributions. Now appraisals, such as the ones we've heard of today, and indeed the others here, are quite useful. They provide a kind of careful scrutiny in a systematic way of the databases they're looking at. The problem is this, that once you have an appraisal of a database, you know a lot about especially what's wrong with those databases. But then the databases that aren't scrutinized, people will tend to think, well, there's not much wrong with them. And one of the things that François's appraisal raises is, although François actually makes this point quite clearly, is that while we may want to benchmark household survey data against national accounts, it doesn't follow that there's nothing wrong with national accounts data. So you have kind of two moving targets, if you will, that are being appraised. And in fact, I'd like to see somebody write really detailed reports about particular sets of national accounts to have a discussion of the ways in which, for instance, calculation methods and data sources have changed across time and this and that. I used to work at Statistics Finland, the Statistical Office here, and the kind of stories you hear from national accountants who do not lend you to believe that any given statistic from the national accounts is God's final truth on a particular matter. I'm going to be relatively brief, and I will at the end actually have one major criticism of all of these appraisals, so I won't only be praising them, so to speak. But it's useful to think about what we actually do with these inequality databases. Why do we have them? They're different uses. One is that we want to assess distributional information at the country level. We may want to compare trends within countries that may even want to compare levels across countries. And one of the things that's kind of actually lacking or wasn't much discussed in these appraisals was, well, suppose you want to do something else and look at inequality and poverty, suppose you want to do real income comparisons of some sort, then you get a totally new level of complexity again about comparing real incomes across countries and so on. So, and first of all, those need to be appraised and evaluated for usefulness, but also you may, there may be benefits to providing that information in the same go. And I saw a little discussion about these things. Now, we use these databases also to explain distribution, so essentially have distributional information on the left-hand side. And then we use the distributional information in two different ways to explain other things. One of them is to explain stuff like economic growth or some other aggregate outcome. But as Stephen points out, it's also sometimes used in combination with lower level data, in particular unit record data to explain things like people's political attitudes as based on inequality in their countries and so on. And that's a, these kind of using distributional information on the right-hand side raises other kinds of problems relative to their quality than using them on the left-hand side. And Stephen, I think quite nicely in his discussion of SWID, kind of makes a point about how biases, that is differences in what is supposed to be measured and the thing that you have to measure it affect these different kinds of analyses and how impreciseness, variability in the thing that's being measured can have an impact. Both of these, all of these papers together give a very nice, well, a nice but quite brief summary of the kinds of dimensions along which inequality statistics vary. There are issues to do with the definition of the income distribution, an important one being the difference between income and consumption. But there are lots of other things as well, reference periods, units of analysis and so on. There are big differences, although especially Stephen's a little vaguer on this about data sources. And then there's a lot about the processing of the data that are used to produce the single number, the inequality index that you're gonna be using. And anybody who's ever used any microdata to produce these numbers knows that from this multitude of choices arises a great range of different estimates. And there's really no very good reason to think that there is a single correct choice on any of this. But of course, if you use aggregate statistics that what you end up with. I will give my one criticism and then I'll kind of end with some conclusion. I won't go through the separate papers. I do think they're all very worthwhile to read. Stephen's paper runs across 63 pages of text, tables and figures, so you may want to stock up with coffee to plow through it. But it has a lot to commend it, for instance. It has an extremely clear and brief discussion about what multiple implementation actually does. So if you ever want to read a two-page summary of that, I recommend this paper for you. However, the single substantial criticism of these appraisals is that not a single one of them actually refers to any international standards on the measurement of income or distribution. These are five reasonably well-known, it's not an exhaustive list of reports which actually try to figure out that for the, what kind of income and consumption definitions should we be using in order to make distributional comparisons. Starting with the UN provisional guidelines from 1977, the two Canberra reports, and finally, the OECE had a number of working groups, one report from which is a framework on forced statistics on the distribution of household income consumption and wealth. These are cited in none of these papers and one kind of wonders why. Because there is a lot of work on, and indeed I think Martin Revaliant does cite these, but he doesn't seem to think very highly of them because he kind of does very much his own thing when it comes to what Liss should have been doing. I'm not bitter, but, but. Now, this brings me to my last concluding comment and it's the following. One of the impressions I think you should have gotten from the three presentations, and in fact the four presentations we just heard is that there's lots of information about inequality and income distribution out there. There are big gaps in what we know, in particular we know far too much about rich countries and far too little about less rich countries, but also the things we tend to know vary a lot by particular configurations of datasets that you're looking at and so on. The thing that's always bewildered me is that why do we collectively in fact in the world but also within countries use tremendous amounts of money to produce national account statistics highly regularly, but distributional statistics are mainly produced as almost an afterthought in general in research projects as kind of in a highly non-institutionalized way with the result that they're essentially all over the place. So if there actually is some kind of revealed preference which somehow you might argue there isn't, but if there is a, if there is a preference for actually having accurate distributional information the question is why is it actually not being provided in an institutionalized way as in fact our national accounts statistics. Okay, well thank you very much.