 Rwy'n cael ei wneud y cwestiynau ar gyfer y dyfodol. Rwy'n cael ei wneud. Rwy'n cael ei wneud ar Andrew a Friedrich. Andrew, wrth gwrs, rwy'n cael ei wneud y cwestiynau yng Nghymru, ydych chi'n cyfrifiadau cyfrifiadau ar gyfer ydych chi'n cyfrifiadau. Rwy'n cael ei wneud ar gyfer y dyfodol, wrth gwrs, y cyfrifiadau ar gyfer y dyfodol, ydych chi'n cyfrifiadau ar gyfer y dyfodol, ydych chi'n cyfrifiadau ar gyfer y dyfodol, rydych chi'n cyfrifiadau ar gyfer y dyfodol, iddych chi'n gwneud y proses o'r cyfrifiadau ar ajud sy'n un suppress clicking, o unityw y fwyffrinfa yn y peirio ar gyfer y dyижwปeth Disneyland, dweud yn ymgyrch, sydd yn ymgyrch yn y busnes o ran allan. A Gweddorach, rydyn ni'n ddigon ni'n gweithio ar y cyflwyngau. Rydyn ni'n sefydlu bod chi'n gwybodóa chyflwngor o ran ddweud yn hollwch gweithio, yn rhan oherwydd mae'n gweithio bod chi'n gweithio'n gweithio'n gweithio'n gweithio. Ychydig o bwysig o'r cyflwyngau ar y gweithio'n gweithio ar y cyflwyngau oherwydd eich ddweud yma oherwydd i ddweud. I'm not sure if you've controlled for that or not, but a more general question here is about the sort of selectivity worry by choosing to look at couples and people who are currently couples. Now, the people who are currently couples might have become couples a week ago or might have been couples for years. We don't know that, but they're couples and of course there's other people out there in the population who are not, who may be quite different and because they're quite different may be exactly why they're not couples. So, I just wondered if there's a possibility to compare the couples to those that are not couples, as well as, as I said, looking at how long they've been together. Thank you. My question is for Mikael and I want to talk about the negative sign that you see on increasing taxes that you find after your treatment. And I'm concerned if this could be because people do not really view taxes as an effective measure of, as an effective means of redistribution. So maybe if you had something more direct, like say you asked people that say we increased taxes, but at the same time we had some program that gave a transfer to people, then maybe you would see a more positive impact there. I have a question for Frederick and then Andrew. Frederick, in the bargaining literature, it has been found that the partner age difference has an impact and in your case it's not significant, but I was wondering whether looking at the houses headed by males would have an impact on your disease. If we had included the houses headed by females, would you find something different? And then to Andrew, there was, I think on the proportions of a non-race in 2007, 3% in 2013, 5%. And I'm wondering whether that there is an increasing trend of, and because in some cases I found something like 74, is it an issue where people prefer not to reveal their race. And secondly, on your CDCs, the cumulative density functions, you only looked at black and white. Wouldn't you extend, and if you did, what are the findings? Because I noticed that it flattens out towards the higher levels of the capital income. So I have a question for Frederick. So I guess the one thing which worries me slightly is that your number of observations after you filter it is right down to 344. And the particular problem I have is that your test statistic is an LR test statistic. And we know that basically on this complex survey data, which NIDS is, that actually likelihood-racious statistics are really not appropriate because I assume that you've got independent sampling of all of these things, which with the clustered sampling really is a problem. So once you kind of put a cluster correction in there, it kind of worries me what you've got left kind of like with that sort of sample size and whether you kind of thought about that a little bit. Hi. My first question is for Frederick. I was just wondering about your husband's language, husband's bargaining power variable. Could you explain the, so husband's mother's education as a share of all mother's education? And the second question is for Miguel. I don't know if I missed it, but did you show us crosstabs for treatment versus control characteristics? I don't know if I missed it, but. OK, this is about a Friedrich presentation. And I have a couple of very technical points. I will not mention them. I would like to support the point made before about the fact that some of those characteristics that you call preference parameters as a matter of fact may also affect the bargaining power, the Pareto weights. And certainly the time that the people have been living together is quite important. The number of children. I mean, if you have two children, three children, presumably the bargaining power must be different. I mean, the equilibrium within the household must be different. So it's not clear the way in which you can handle that because you'd like to make a distinction between parameters which do affect the bargaining power without preferences, but certainly it is something that has to be taken into account. The other thing which I found a little disappointing at the end, we are talking about inequality. So how much inequality do you have in those households? So we know that you cannot identify the sharing rule completely because there is a constant that is missing. But at least you could do some kind of simulation analysis. It's OK, assuming that there is a kind of normalization of the data. I mean, picking up a constant. What is the kind of inequality that we're generating? Is it OK that as in the paper that you mentioned at the beginning, these insights, you are getting 50% of the total inequality is due to inter-housel inequality in South Africa. I think it's very important to make these. And my second question is about it to Miquel. It's a little related to the issue of what about the control. But because this information on international inequality is important, could we know how many people in the population initially knew something about inequality in Namibia as compared to inequality in South Africa or as compared to inequality in the US? Because there is always an issue with this kind of experiment on information, which is that there must be already some information in the population. And the problem may be more on the way in which the information is spreading among people. So it seems to me that it would be important to have some complementary information about these. So we'll start with Andrew and then go in the same. Thanks for the questions. Yes, there are differences in the types of programs that people apply to. I haven't investigated that systematically. But I do know, for example, that the differences between black and white students in the commerce faculty are much smaller than they are in the science faculty. Which would suggest that better quality black applicants are applying to commerce rather than to science. So I can definitely take that a little bit, do a little bit more thinking about that. So thanks for that suggestion and comment. Does the process actually work differently? The university process, for the most part, no, it doesn't. Different faculties and programs have different requirements, but they all advertise those requirements and they're not really taking stuff into account other than basically your secondary education results. And in some cases an extra test that's now come about the national benchmark test because the universities are a bit worried that maybe the secondary education is not quite a good predictor of how people will do in universities. So there's this kind of university generated test that's now also being used. In medicine, for example, they used to ask for CVs and references and things like that, but I think in the data that we're using, that's not the case for these years. Trends in not applicable? Yes, I think that's an important issue. So many, I think, are white students who think that if they say not applicable, they're going to somehow be treated better than if they say that they're white. That's not the case. The university puts everyone who says I don't want to say or whatever into the open group along with the rest of the white students. And so I would agree, yes, that probably there is some sort of sense. I went and actually had a look at the data and there was definitely some white, what looked like white candidates from the very sort of higher end of the schooling. So they went to that elite private schools look like they're probably white students. But so there is that trend. And in the CDFs now, I haven't drawn it for the other groups, but that's a helpful suggestion. Thanks. I'm going to answer the question in the order in which it would also. In terms of the relationship length, the selection. So I don't think we observe relationship lengths in the net data, but obviously this is an important factor. If it was observed, we would have included it as a control variable. And then a related question was just the generalisation. So the idea was to restrict the sample to the persons we have as I don't really think that just putting in a dummy variable for just for certain aspects, just for a four person household. Or adult member household, for example, is sufficient to account for the different level of bargaining power that we would observe. We would need to do a far more complex econometric analysis. And at this stage, I'm not exactly sure how we would do that. And then in terms of the age differences and the male-headed household's heads, this is also somewhat related to Professor Benion's question. So we wanted to separate effects that we expect only work through the sharing rule on household demands from effects that we expect maybe also due to preferences. We do not have a price variation in the same way so that when we see changes in, for example, the age difference, we don't know if it's maybe just because the older household members in general prefer higher expenditure on medical, for example. And that's not exactly clear. And similarly for male-headed households, there may have been some pre, well, some marriage market interactions or something like that that happened in female-headed households. So a household that identifies female-headed households that are not necessarily able to control, but if I recall correctly, our results weren't that sensitive to it. We just didn't want to do it. In terms of the number of observations in LRTC, that is a very, very valid remark. Rulof and I are redoing the data on the upgraded NETSET to Wave 3. We're attempting to incorporate far more observations. And we're also looking to a few more methods to test NETSET and NESB because, yes, this is a very big issue. Currently we're looking at a sample that's quite a bit more, but it's getting there. In terms of the husband-husband maternal education share interpretation, the idea is in South Africa because this data set is a 2008 data set and our sample is of people that are between the ages of 25 and 65. The idea is that the husband's maternal education share gives an indication of potential wealth or access of the partner's pre-marriage. So we would expect that if one spouse's mother was better educated than the other spouses, we would expect that person to have better access to education opportunities and eventually labour market opportunities or income. The bargaining power difference. We do not attempt to identify the sharing rule in the paper presented. We were, once again, mainly focused on factors that we do know are not then correlated to other preference factors. But, yes, this is a valid critique and I'm not exactly sure how we would control for this other than perhaps interacting those preference factors with the distribution factors inside that lambda equation. This is definitely something to look at in terms of how much inequality. So, yes, maybe I've gone through that a bit far. So what we see is that female bargaining power, at least as is defined, is far, is on average way less than male bargaining power in the households. But it's definitely worthwhile for us to go further and look at maybe a better accurate measure of inequality inside the household when we actually observe these things. Because we only observe effects of changes in the sharing rule and not the sharing rule itself. Thank you. OK, so first about this balance test, like whether for the controls there's different between the treatment and control stuff. So, yes, I didn't show it. So the survey was done with mobile phones and so the randomization was sort of inbuilt in the thing. So I mean it should be fine. It's fine. There's like I think for some treatments like some people are older but in ways that are sort of stuff where even if the randomization works like sometimes things are imbalanced. And with all the results we put a bunch of controls of demographic and pretreatment stuff and everything is the same. So there's no problem with that. The issue of whether people identify this taxation with redistribution. So actually we do have a question on I think general taxation and to spend it on, I don't know, on stuff. And from what I recall actually there's not much action there. My feeling is that so the one that we have here about the top tax we were careful to emphasize the sort of that we are talking about the rich taxes for the rich. And that we were saying something like, so now I don't want to get the numbers wrong. I don't recall what were the numbers, but something like 60,000 grants per month. A person earning 60,000 times per month has to pay say on average 20,000 in taxes and keeps 40. You think that person should pay more of it. So I feel that this one is the one that maybe is like most clearly linked to emphasizing someone that earns 60,000, which for the incomes of the people we are talking about is something massive. Whereas the other one might be that is interpreted in a rest actually less redistributed way, just like okay you tax a bit more and then you offer more public services. And the last question I have to say I didn't understand it exactly. So we do have information on what they thought inequality in South Africa was. So we could check if the proportion of people that said something lower than what we actually showed them. Is that what you are saying? The knowledge of inequality in other countries in this chart that you have shown, how many people knew that there were countries where inequality was much lower than in South Africa? So I guess that the best would be that. So obviously we don't know. We cannot know what they knew about the others, but then I guess that by checking what they thought about South Africa relative to the others like we can see. I guess the point here is to make a distinction between knowledge and sensitivity by showing those chart to people. Maybe they had some knowledge about the fact that South Africa was definitely much more unequal than other countries. But simply because you are showing this, this is triggering some reaction from them. So it is not an issue of knowledge of information. It is something else. This is the point I want to make. So I mean in some sense, yes we wanted to be something else. We wanted to be something that makes people like shocks this inevitability thing. So we wouldn't now advocate a policy of like now we have to distribute information and that's what will work. No, we wanted to explicitly shock the perceptions of people and whether this happens to the information channel or through the sensibility. We would have to think about whether it matters a lot. Right, so we have time for a few more questions. Thank you. So I have a question for Laura. You are showing that the poorer under-report ill health, but then your conclusions are all about, so you say then that the inequality as measured in self-reported health is underestimated and that's true. But what are the implications of that? It seems to me that turns a lot on why the poor under-report ill health. Like is it you implying at points that it's endogenous that because they can't afford to be ill they under-report ill health. But it could just be information, right? It could be the links between education and knowing how healthy you actually are and it seems you somehow need to benchmark, I don't know if you can in the data. You mentioned that you've got sort of objective measures of health and then you've got subjective measures of health. And it seems to me that the relationship between those two is very, very important before one extrapolates to saying that this has got huge policy implications for the delivery of health services rather than just the understanding of people about how healthy they are. OK, and then I had a point for Fredw. I think some of the comments have been around the fact that you clearly reduced your sample and you tried to put it into a nice structure to help you make tractable what's already quite hard to make tractable by these couples. And I think a lot of the comments are about that the reality of life in South Africa and South African families and households aren't like that. So you might not be making things tractable and the trade-offs in terms of the sample size might just be too hard to bear. So I'm not sure what you lose by opening it up to a more complicated version of the household, a South African type household. That's the question. Then I wanted to draw your attention then to the questions in the NED's data on who makes the decisions about this and that, the next thing. The so-called bargaining questions, whether those can't be useful to you somehow in these models and in separating out these different effects. So there is this paper on the South African pensions because there was a lot of speculation in a similar type of literature to you where people were just using cross-sectional data but talking about unitary versus bargaining models of the household and doing exactly what you're doing. And there's this paper from Kate Ambler out of the University of Michigan that uses those questions about who's making the decisions to try and come into that very same issue more tractably. So that might be useful to you as well. Yes, a small point to being there. I don't understand why you look at the average reaction to the politician speech. I would like to see what is the distribution of the reaction as I would expect. I don't know much about the political situation in South Africa, but I expect opposite reactions from people sharing or not sharing the view of the politician who is talking. So maybe a slide with the distribution of the change would have been interesting. Maybe you have a kind of polarization of the reaction and this negative sign you find is simply like a result of reaction in opposite directions. All right, so then you can answer questions. So I think that is a valid question about the transmission mechanism and I do a lot of theorising about it, but it's tough to control for it and even if I do control for objective health measures, I'm not sure if that would tell me anything about the actual transmission mechanism. And also the stage data is very much self-reported except for symptoms. One possibility if I did want to do something looking at objective versus subjective health is to look at the next data set, which has more objective health measures. But also I'm not sure if it really matters what the transmission mechanism is, if it's a lack of information or if it's suppressing of health, of the idea of your health, either way the health needs are going unmet and that you should have better access to meet these health needs. Okay, so on the reduced sample and what do we lose? So yeah, so when I initially started this research question, I was also far more interested in the households where we have six working age adults with children from each household. And I think the main issue for me at least conceptually is that we would expect the distribution factors to affect these persons differently depending on their relationship. So we would expect the balance of power between a married couple changing those distribution factors for those to be different than those for the brother that also lives in the household with his own children. And I'm not exactly sure yet how we would apply this, but yes, this is most certainly the case. And then the bargaining question, so the reason that we didn't use the bargaining questions as it is presented is it's one, it's ordinal, well it's ordered so it doesn't give us the same sensitivity. The other thing is like there might be a difference between what the household members that is currently answering the questions, the person thinks he is the boss or she is the boss, where this might not necessarily be the case. But we do use that data later on in a probate estimate to see whether or not the distribution factors are indeed significant in predicting the results. And we see that yes, they are in the same direction predictive of those results. Is that all the questions? So about looking at the distribution of reactions, so I guess we have to think about what indeed what I had thought about and what is going in the direction of what you said that I think would be really nice is to try to identify groups that might have been positively shocked. And so the way actually in some sense what I thought and what I was expecting is that the difference in the townships would give us this thing or differences of the videos or the way that the fact that the result seems to be so robust across different instances makes it hard to find variables. I was thinking more in terms of pretreatment variables, to group like the make interaction effects on the basis of pretreatment variables and maybe we can look at your thing. I mean it's hard because like for example in the Caillicia sample an overwhelming majority says that the ANC is their closest, so they support the ANC or at least they say that the ANC is the closest thing. So that would have been the people that you would have in principle say that should be sort of more convinced by this type of thing. And there are very few people that report like in Malema or supporting the party of Malema. So we will think about that. Any more questions? All right then, thank you everyone for being here early in the morning and thank you to the four presenters for four fascinating talks. Thank you.