 Gwyddo nhw ddim yn edrych bod cyd-feyddwch. Ond they're all very interesting. They all use novel methods and ask novel questions, and then come to some original conclusions. So I'm not going to be sort of critiquing them from the point of view of the basis of what they do, but I'm going to be thinking in some more general ways about the papers. As you've seen, there are two on ethnicity and two on gender, and I shall treat them separately. So I shall start with the ethnicity. I think there are three interesting issues... many others, but there are three I want to comment on. One is which groups do you choose when you do horizontal inequality analysis? Second, trends in horizontal inequality and what売 USAIR and third, something about politics. In the papers that we've got the Indian paper chooses cast categories ond mae'r ystyriedau bod y wneud yn ysgrifennu'r teulu. A'r c Bemolwyrach Cyf想ol. mae'r ddoni nhw'n golygu'n ei ddwy flynedd ac'r ddweud oherwydd hwn yn ddedig iawn o'r gwmhladau hyn yn ddod, ac mae'n gweithio gennymau, mae'r ddau, oherwydd i gael Wrangol erbyn'r zeulu ac'r bwrthion rym ni'n gweld bod y gweithio'r gwahanol o'i gweithio i'r gweithi'r ysgrifennu. y gallu gwneud cyflwyno hwnnedd yn yr ourman i ddiwrnod o ddiwrnod, a ddegonod o ddiwrnod cyflwyno a ddegonod. Yn gyfo'r gwahanol, dalais yn ddegonod o y ddiwrnod o'r oran â'r gwahanol o'r ddafyn lleol o ddegonod o ddegonod o'r gwahanol yn awdurdod ychydig yn y ddegonod o'r gwahanol. Felly raddyn ynno chweithio'n gwahanol, ac yn nodd yn drwy ffyrdd o'r gwahanol a'r gwahanol os yw ddegonod o'r gwahanol. fun doing that within the ladino category. There are those people who would self identified as mixed and others who would not. And then of course you can do self identification observers identification primary language identification. They all give you different results I think the first thing to say is there's no right way of doing it and we should recognise that someone said in their presentation not sure I got the groups right. I don't think there is a right way I think you have to think about what the reason is ddim yn sefydlu cwysbeth, ac yn ychwanegau arall? Felly nawr, yn ddechrau sy'n gwyntaf, felly ond y gwynebu bwynd. Felly, sef ydych, felly rydych masiaia, mae'r dwylai meddwl er mwyn fwynt gweithgambd i'r pryd y muslim a maes ac pethau acrhyw ar hwn. Mae'n meddwl yn berthyniaid, a mae'n meddwl yn ei amser. Ond ei meddwl i'r amser afriwg, ac mae'n fathbwynt yn eu amser, Mae'r gweld i'r cyfnod i'r gael y dyfodol i'r llunio a'r llunio i'r llunio. Yn ddau'r llunio, yw'r ddysynchion? A'n gweinio ar gyfer y gallwn i'r maewn cyfnod i'r Milaysia, yna'r categoriadau i'r cyfnod i'r cyfnod i'r cyffredinol, yn y ffynllwch yw yng Nghymru, yw'r cyffredinol i'r cyffredinol, yw'r cyffredinol i'r cyffredinol, oedd yn cydweithio'r cyllid yng Nghymru, a fyddai'n gynllun, a gynnyddio'r ymchwil, felly mae'r ddysgu'r cyfrifio yn gwybod o ysgolion oherwydd ei fod yn ei hyn oedd o'r gwybod ac mae'n golygu i'r wych yn ni'r cyfrifio. Dyna, mae'n golygu i'r ddweud o'r ddweud o'r ddweud o'r ddweud o'r cyfrifio, ond ydych chi fod yn ei ddechrau, oedd yn ymgyrch, oedd ymgyrch yn ei ddweud o'r ddweud o'r gwybod. Gwylwch ein gwylwch ngoswch yn Garnaf, ac ydych chi yn ymelloddodd y byddwch gwaith c cỉau mewn gwirionedd. Felly mae'r cofyddu ddim, wrth gwrs, mae'n ceisio'n cael ei yma. Felly mae'r hoff chi'n cael ei rhan o'r ffordd yn ddiwrnod. Wedi'n ceisio, mae'n cael ei wneud symud bod yn hynny'n gilydd. Mae yw y cwestiynau hyn rydw i ar ardeddurau. I mean, I know that it was once a group is categorised by others as the enemy group, that group become coalesces that it did, for example, in Bosnia Herzegovina, in Nazi Germany, in wherever. So quite often it's what other people think and then that becomes the salient group. Then there's the issue of behavioural differences. There's a brilliant book by Blau about groups and he actually defines the difference between groups. A group is a group if it has more social interactions within the group than it does between. I guess it has to be quite a sharp difference. But that's quite an interesting way of categorising, but you need a lot of information. Then there's political salience. I talked a little bit about the voting, but political salience really depends. I mean, political leaders are quite canny and they want to get support which is going to get them voted into government and so on. So if there are a lot of little groups, they'll mass them together and redefine them in such a way that they become one big group and that's happened quite a lot in many places in the world. So what seemed to be a lot of groups suddenly becomes a single group. And it's interesting in the case of Guatemala that these individual groups haven't apparently coalesced in that way, in a way that would give them political power. But that's another way. And sometimes it's a question of money. I mean, there was some groups in, I think it was Indonesia actually who were trying to raise money. And if you want to raise money for your group, it's good to appeal to American Pentecostals. So you might as well define yourself as a Pentecostal rather than defining yourself as some indigenous group in Indonesia because the Americans are not going to be interested in you in that case. So there are lots of different reasons. So the choice is obviously going to depend on the question you're trying to answer. So on public goods provision, you might be particularly interested in political mobilisation and also in the perceptions of others because, you know, there's a theory about social psychologists have about the scope of justice. And that people think that one should be fair but only fair within certain defined groups. So your own group is, yes, we ought to be fair across these. We shouldn't let those people starve. We should give them public goods. But you don't have a scope of justice between the groups. I'm not totally convinced by that because we have done some collected evidence in Africa where a lot of people support a cross-group redistribution. But there is that. And that's one of the explanations of why public goods are lower in fragmented societies that you don't want to give goods to other people, raise your taxes and give goods to other people. So that's another reason why you would take a particular attitude. On labour market outcomes, the relevant groupings were presumably, you know, the source of discrimination. And to discover that, you might have to do the sort of experimental research that we've just heard about for men and women and so on. So basically, I think what's a very interesting question when you're starting work on this is to think, why are we choosing this group? Is it the right group? Why don't we try the same thing with different groups? Would we get different results and so on? And of course, in some societies, there may not be so many potential categorisations than in others. Second question is trends in horizontal inequality. And there's very little data on this over time. So it's very interesting. And I think one does need these case studies to put it together. And for India, you've found that there's been some increase in horizontal inequalities over the last 20 years. But in more recent years, different trends at the national and the district and the state level, which is, again, interesting. And now, in Guatemala, you've found decreasing horizontal inequalities, which was interesting to me because earlier work we did, we found that it was decreasing on the social side, like education, but increasing on the economic side, because of regional differentiation. But that might be because we chose the groups so that we used region as the basis. So it could be the way we chose the groups. So the point to note really is that the trends are going to be sensitive to choice of group. They're going to be sensitive to the choice of measure you adopt. So if you take education, and you're nearly at 100%, as you go on, you're obviously going to get equality. At the beginning, it's like a cousinous curve, you're going to get widening inequality. So if you take intersture education, you probably find widening inequality. But at primary education, you find declining. So it's sensitive to the measure. Then any aggregate measure, and none of you actually used an aggregate measure. What I mean by aggregate measure is you take society and you aggregate all the groups together and see whether horizontal inequality is increasing or decreasing. And that may not be very useful because some groups may be decreasing and others increasing. And I was thinking of, say, the United States, where if you took blacks versus others, you would probably find it staying about the same. But if you took Asians versus others, you'd find it was closing up completely. And finally, I was just going to say that I like the analysis of why there's a trend decrease in labour market outcomes in Guatemala because there's much too little work in general on explaining change. Then coming to politics, and my original interest in horizontal inequality was and remains deeply political, why people go into conflict. And then your work on public goods is obviously political too. On conflict, the one question is how do horizontal inequalities lead to mobilisation and tracing it? And what we find, and you didn't really go into the political inequalities at all, but we find political inequalities very relevant. I think the political inequalities are also relevant to the public goods issue because not only the quantity of public goods but the distribution of public goods. So you're assuming that if you had more public goods, they would be distributed in such a way that horizontal inequality would decline. There are many cases where if you have more public goods and one good dominates, and in that you can show that in Kenya that's happened a lot of times, certainly happened in white South Africa, more public expenditure fine, but actually horizontal inequalities worsen. Let me see. So I think that's really basically what I want to say about the ethnic papers, and I did find them extremely interesting and stimulating. Let me come to the gender papers, and they have some interesting things in common. First of all, they both find an adverse outcome, not very surprising, and they both try to explain the adverse outcome, obviously in education, in response to weather change via meningitis in the case of Niger, and in wage gaps in general in the second case. In wage gaps in that case too, and in the second case clearly why men have different views on women. By the way, I found the second paper very interesting, and I'm going to tell my granddaughter it, because I was just talking to her and she's in a physics group in which everyone is boys except for her, and she said that they got different results from her and she was absolutely inclined to think that hers must be wrong. We were saying, why should yours be wrong? Why not theirs? Tell her the paper, and then she'll believe that she's right. Now the second thing about the gender papers, which is so interesting, is that they both use very innovative methods to get at results, and I always like using quasi-experiments or experiments, and you can find innovative results. One paper uses a type of natural experiment, and the other one uses more. It's more explicit experiments. They both have innovative policy conclusions. On the sexism sort of issue, did you ask, did you give the sexism questions to the respondents as well as to the advisers? Because I wonder if that would influence the extent to which they... Yes, do, why not? The way the sexism questions happened was that we had a post-experiment survey in which we asked lots of things, and included in them were these 25 items on the sexism inventory. Now this being my first experiment, I didn't realise that that could be a problem, in the sense that the treatment itself could have affected the answers that you got on the sexism score. So what we've done is we've run some regressions, and those regressions are telling us that if we run the sexism score on the treatment, it's telling us that there's no difference between the treatment, the control treatment and the gender treatment. What we need to do is go back to the lab and add the sexism inventory questions before we make them do the experiment and to see if that has an effect. But for now our regressions are telling us that we are fine, but that is indeed a design flaw. But you did make the subjects as well as the advisers take the questions. The advisers are actually people who've done the experiment, so they were subjects in the same experiment, and then we took their answers, so they did exactly the same thing. The other thing, which is the sexism question seemed to me to be open to people, not answering them truthfully because they're obviously politically incorrect, but I don't know if they've allowed for that. That's quite right. That's why I presented our scores next to the other countries, which shows that we are similar, so it seems that we are fine. The other thing is that the social psychologists have a measure of internal consistency. What they do is they switch the scales of certain questions, such that some questions become more sexist if you answer one, and some questions are more sexist if you answer five. That is an attempt to get at the internal consistency of the measure itself. We ran those tests, and one of them is the benevolent sexism test is borderline. It has a score of 0.7, which is marginally acceptable. The other is 0.83, which is fairly acceptable. That we are essentially taking from the social psychologists, they are telling us that here is a measure of sexism, and we've done like 2,000 papers use this measure of sexism. We're just taking their word for it, really. Fair enough. Both these papers have rather innovative policy conclusions. The paper on health suggests that you should use health policy and education policy. I was interested that you didn't suggest that you could use financial compensation as a policy mechanism, and I think that might be a more direct policy given that your theory is that the policy, that it's money that's the problem. Your paper suggests that you need anonymity in promotions, which is absolutely right, but you also suggest talk of the need to reform attitudes. That's a big issue, and I think you need to explain what you mean by that, how you do it, and so on. Maybe the very fact that this sort of results are there should be widely disseminated and might change attitudes, so I think that might in itself be helpful. I felt that both the two papers made a bit of a leap between the findings, the strictly econometric findings, and the conclusions in a sense. The first one is a little bit speculative about the constraints leading to the reduced age of female marriage being the cause, though you do have evidence for it. You show that there is an increase in that, and I felt it was probably right, and I felt when you gave the talk you sort of enlightened me because you said both authors lived there and we've talked to people and we know what happens, and that's what I felt was missing in the paper, a sort of anthropological dimension which explores these things, which would then be more convincing when you went to test them, and then just reading it, I thought you were outside as we came with this theory and he wasn't sort of grounded, and so I thought that was good and interesting, but you should put more of it in the paper. Then I felt your paper jumped from the attitudes issue to this being a cause of discrimination and wage gaps, which again is probably right, but of course there are many other reasons why there are wage gaps. There's sexism itself, there's all sorts of things like there's pregnancy among women which might make employers rationally discriminate against them, and so on, so you can't quite jump to the conclusion. So it's a little bit of a leap, and I felt there was maybe a bit of a leap in some of the ethnic papers into the final conclusion as opposed to the findings. So in a sense what I'm saying, that's always good if you end research by saying you need more research, is that having come to these tentative conclusions you say there's more research. On the climate change conclusion I felt that that needed more explanation for me to be convinced that your findings were relevant to climate change, so you need to say in more specificity how that would be translated. Then of course both the papers and indeed the earlier papers raised the question of replicability and that's always a question. For example in particular on the bride price thing, in Asia where you don't have a bride price, you have a dairy, do we find that a health incident like that actually leads to the men marrying? You can imagine, it would get the opposite conclusion. So it'd be quite interesting to try and replicate it. In many cases where there's neither a bride price nor a dairy do we then have a zero conclusion? So that would be interesting to try that. And then your students, your elite students from Lums, can we jump from that to what happens everywhere in the world? I mean that is a problem about experimental evidence. So much of it is based, a lot of it is based on students in Oxford. They all make a lot of money through these experiments and then all around the world the policies have changed because of these student attitudes. So that is a problem about experiments. I don't know what one does about it. But anyway despite these thoughts that the papers raised in me, I thought they were great papers.