 Hello I'm curious if you might have first of all I didn't understand what your notion of median was in the multi-dimensional world and I'd like to know what it is in your in your concept so why was that upper right-hand corner the median as opposed to something else and secondly it seems that in many cases you do have three observations a recent paper with Indranil Dutta was that case in fact using GSS survey data on life satisfaction happiness where we got some really nice you know uses of the same methodology but with three and is really quite interesting for the single dimension case and I hadn't thought about multi-dimensional in that I'd like to hear if you've thought about multi-dimension to the next case of three as opposed to two thank you okay yes maybe should I answer this first or should I click okay so let me just illustrate the median concept again so the multi-dimensional median here is is the assembly the the tuple or the vector of the regions of the partial marginals so here in this case we have the marginals here is free they are free here and that's seven healthy so it means that and here there are six poor and they have four rich so it's it's the simply the median along each dimension that we combine into the multi-dimensional and it's known and it has been argued in mathematics literature that that is the only natural multi-dimensional median concept for ordinary data and then as to your to your second question you mentioned you mentioned some examples with the free free levels yeah I mean so strictly speaking our procedure that we provide in the paper it only applies to two by two cases with two binary indicators if you have like if you have a two dimensions with three levels then it should also be possible to to to to to work out a checking procedure how it would be more it would be more complicated but but you'll be able to do it what we want to look for is is a completed general checking procedure that applies to to general bivariate data there you will need another approach than we we're taking this this paper here but for those very small small cases like two by two or three by three it should be possible to to work out some ad hoc ad hoc workout some checking procedures that that that work any other question yes please thank you so much very interesting presentation quick observation that when you do bivariate case okay it's intuitive when you go to three-dimensional case becomes a bit difficult and as you move on it of course becomes much more critical so of course you will analyze that what this particular framework can tell you whether inequality is higher or lower in terms of dominance and when you do not have faster or dominance holds then you cannot conclude anything and that's the most difficult thing when it comes to applicability whether you talk about the close on and younger or boogie and chakraborty all this word so it may be interesting to develop as I along with professor Foster has tried to develop sort of idea of robustness so you know if your dominance holds fine you know one group has higher inequality than the other what happens when dominance does not hold can you say to what extent you can make the comparison whether it is not at all possible or to what extent what kind of you know so it's sort of a robustness idea that comes underneath of dominance maybe you can think about it just a comment but it's a very valid comment because it indeed also in this application in many of the cases we don't have any relations we don't have first order dominance we don't have inequality this quality dominance and what we can do so essentially that is a part of the concept it's I mean it's the idea is to make a robust concept and then hence it has to be demanding in a sense however one possible approach could be to make since these are this is sample data to make bootstrap replications of the data and then make the comparisons again and again with the bootstrap resampled data and actually this is what we do in in this paper when we have a dominance observe the dominance we bootstrap the data resample the data and then check how many how many times out of a thousand say we actually get the dominance relation again so that's a sense and indication of the sort of statistical robustness of the of the of the dominance relation you could also do it if you don't have dominance we could resample the data and then check if at least in some cases we actually get the dominance that would be an indicator that will be close to dominance some some in some sense so that's one one way to to to to address this taking a bootstrap approach any other question okay if not then thank you have two questions one is the weights you have you said you're not explaining how the weights were calculated but nevertheless I'm just asking whether the weights are subjective or is there a method for it secondly are they different for different countries and the other question is that especially for India we know that if you're calculating the index from the DHS and only mix then DHS is last was it was done in 2006 and so I wonder why the numbers of destitute has increased when the nutrition data hasn't changed and if you're giving highway to nutrition I'm presuming because it was the top so I because you know if you schooling data is of 2010 or 11 nutrition data is of 2006 so do you make any adjustments for all that thanks may answer one by one questions and then so there was another question in the back yes I like this idea of looking at more information more variables but I wonder how do you really know you are identifying the poor or the destitute is there an external validation of these or you are just providing another index and then with this index you have another criteria to classify people in one category or the other because I don't see any external validation of this index or any other index that uses many dimensions whatever thank you James you also had a question or so you're talking about destitute versus poor the poor definition is already pretty acute and I was wondering if there was any theoretical or in fact any other intuition that you could be providing us as to why you would choose the particular cutoff levels that were chosen in defining the destitute is there a conceptual framework that guided you in doing that or was that sort of just based on what cutoffs were available in the data and you had to go with what you had that was one question I'm wondering if there's another one yeah the second one was the conceptual trade-off between changing depth and changing breath can you offer any insight as to what that does when using a measure of this type which is based on breath but nonetheless you're changing depth and the trade-off between those two could be quite interesting as to who is captured in one case versus another thank you thank you so the first question on weights so when this index was constructed this is an international index and of course it has gone through various drops you know to understand how the weights weights are same across all countries they are not different okay and given that your two questions are related let me just tell you how the multi-dimensional poverty index is different from the idea of composite indices in compo as you said for schooling you have recent data for health we have risk old data so you are thinking in terms of the human poverty index or human development index where you have index for each dimension and you add them up this is not how the MPI is computed the for MPI all data has to be available from one source so that for each household you are able to see the multiple deprivations and then identify the household accordingly and then you agree it across the household so that is the first thing I think it answers also your partial of the second question so all data for India came from 2005 6 DHS dataset destitution increased or not I have not presented in this one but destitution has gone down in India when we analyze between 1999 and 2006 in fact in the Indian context reduction in destitution was much faster than the reduction of MPI so that's the positive story and that is what we presented in the 2013 paper with with Sabina Alkaya subjective weight or different weight weights were chosen in a way finally to represent sort of the idea of the human development index where all three dimensions were weighted equally here also we have three pillars standard of living education and health and as professor Tony Atkinson argues that in this kind of situations where you really do not know how to decide choose the dimensions in such a way that they have equal importance so that was that one factor that that played behind this though dimensions were chosen with sort of equal importance as was done in the HDI and then indicators within each dimensions they were just equally weighted okay so that's I hope that answers your questions how do we do validation what we have done in the past since 2010 that we have tried to implement these indicators to fields so you'd go and talk to people on ground and also collect information if they are really and we try to put those stories in our policy brief policy research briefs and yes so they are valley validated and it has been found that some people whose conditions whether they are really really deprived for the situation we haven't done it yet because for the situation this is just the concept here what we try to do we try to given that the MPA poor suffer in the in those dimensions and we have validated that we try to understand what if we put further threshold further put it deeper down whether or how many people do we identify but I completely agree with you this requires going to this requires validation going to field we haven't for the solution we have not done that but I take your point it should be done from the fosters questions particular cutoff level and conceptual cutoffs so for depth approach of course it was driven by both because electricity for example is a prime example where we wanted to know whether access yes or no is not enough I have been raised in a family where I had ten hour power cut I enjoyed the time I didn't have to study or whatever I had fun time but that was deprivation so of course I did not do my education properly so just having access to electricity is not enough and we wanted to and when we have worked with different country national governments we have tried to select cutoffs based on senior arc for example when we are working with the Iraqi government and even if you were there we try to select the cutoff that at least 12 hours of electricity has to be there which we we could not do here we are working on a urban sample slum project where we find that the household has access to water but it's between 3 o'clock and 4 o'clock in the morning that is the only time we have access to water whether that is sufficient whether that is enough so those are the things that we cannot address from the data because simply in the data we do not have have things so so we set the deeper cutoff but at some places where we can get the maximum possibility for example sanitation another example open defecation that's the minimum possible thing we got and it's a huge bunch of people we could not go beyond that so by part yes it was affected okay universalization of primary education so that actually motivated the MPI cutoff and then we went to a deeper cutoff where we try to say and try to be you know as much how should I say we could debate so we tried to be less controversial so for education we said if no child in that household is attending school then the child is identified so probably it gives you a legitimate cutoff so those kind of mix of conceptualizations and and availability of data what we had played the role in defining defining the cutoff trade-off so for the Indian context as I was telling what we did we implemented the cutoff say instead of 33 percent we used 50 percent the United Nation definition identifying severe poverty and then we have the set of deeper cutoff but keeping K same this poverty cutoff same and we tried to understand the interplay whether we are identifying the same number of people we did not so out of the 56 percent of the population in 1999 we identified nearly 24 percent as deprived in both and there were people in the other two segments as in the previous section was presented when you and we look at the marginal distributions there are people they were mismatch so those were the trade-off and trade-offs high it does not identify the same group of people honestly so so that is what I can tell you at this moment and particularly we have not tried to compute any index sort of that gives you you know the amount of trade-off but this is where we stand at this moment okay thank you very much then we have to move to the third presentation perfect thank you so much hi that's that's very interesting I didn't follow everything because it was presented a bit quickly so I'm not entirely sure what you're measuring in in the inequality section but I just wanted to ask you a question whether you had thought of and I'm not sure if you can do this without having pure diads but if you had thought of using something like the gender parity index to look within households suppose you had households with just one boy and one girl so make it simple what the gender parity index does is it takes whoever is higher and uses that person's as the poverty cutoff if you will and then uses a poverty gap to measure the shortfall from that of the other persons and then you just divide up you first have a headcount right of the percentage of people who were girls are above boys and percentage boys but now you can go into a gap which measures well if so then how far are they percentage wise from their from the person with the higher of the two so it'd be natural to explore not just percentages but to look at how big the gap is and I'm reminded of this because if you look at Rwanda for instance it's educational gap between men and women in married pairs in this case is is virtually nothing it's a half a year but if you subdivide it to those where women have higher education or the spouse that's male has a higher education in both directions it's three years on average and so and it's about 50 you know equal share in both of those cases it turns out and the women with more in the urban area and the men with more in the rural areas but you're hiding all the interesting disparities if you look at the average and that's what you're pointing out I'm just saying there might be another way of pointing it out by means of the gender parity index of the women's empowerment agriculture index which of course was put together by Sabina and myself and a bunch of other people so thank you please tell us which of the indicate a level of chronic under nutrition you're using minus two or minus three zed scores second question and remind us the MICS definition of household is it based on residents or is it based on kinship if the former one would be much less surprised at the extent of disparity if it's based on kinship one would expect to find much lower levels of disparity we take a second question here in the back thank you very much I'm his eye on the disability researcher and this comment is actually for all of the three presenters and if you rely your data on the existing ones that means that many of the children who are socially marginalized are not necessarily included into those and particularly so for children with disabilities and I would like to pose a question to all of you whether you have ever thought of relying on the existing data only and by that perhaps you might be reinforcing the fact that inequalities are continuing thank you thank you do we have another question okay if not yes please minus quick one of the indicators better the registration I couldn't understand how it's going to be like capturing what you want for inequality I couldn't understand really whether it's I doubt whether it's even a very good indicator sometimes to foreign household inequality I don't know I just want clarification okay thank you and maybe you start and because the question before was also to the other presenters then I just hand over to the two previous presenters in case you want to respond something okay it's the first one the gender parity index I think it's it's a good one it's I think it would be possible the one thing is that for some indicators that they're binary how would that work how the gap work so but yeah I haven't thought of that but yeah I think it would be a good way to do it this the way I do it only captures whether girls or boys in households are more or less advantage and and also it I guess it obscures whether there could be differences also across girls or across boys and this assumes that all girls within a household have a similar value and all boys within a household have a similar value under there are the differences that could be due to age for example birth order and that's not captured here this is this only captures gender differences so so that's that's important to take into account the second under nutrition minus two standard deviations and I should mention that although I know there is a newer reference from the WHO I use the old one because most of the mixed service from previous rounds only have the old standards for nutrition so for for comparing across different periods I use the old ones but it's not the best the new standards are much better and a newer rounds have that new standard that all the rounds of the service don't and about the household definition I believe it's on residents but I need to double check with the with the mixed service but yes I'll go on the birth registration and then finish with the with the overall question birth registration is is whether a child is registered at birth and that may not seem highly relevant but it actually is because birth registration gives access to services gives access to many things for children it's part of the right to have a nationality for example gives right to vote when when children grow up so it is actually a very important indicator of child well-being and and I think one of the main things I wanted to do with this is to show a wider range of indicators for child well-being as I said that the convention of rights of child has 17 these are only four so there are many other things that are not included here but I think birth registration is a it's a very important one and the final one on using existing existing data only yes I think it it it could reinforce inequalities because we're not we're not looking at other things and one of the things I I just said is that for example I would have liked to look at many other dimensions of child well-being but I only have four so I work with that and hopefully new service which have more information could be used you could apply the same methodology if you were looking at one specific country for example you could look at other sources of data I'm using this international comparable source which is the mixed service but if you would want to focus on other things you in some countries you could go and find other service in terms of disability I think the WHO has a new module on disability to including household service and I think it's a pilot but if for example that was included in standard household service that would be very useful to have a sense of other inequalities not only for children but for other groups as well I'll leave it to the other families thank you thank you would you like to respond something to this question so yes on data issues yes it's a it's a valid concern that we work with that the existing that says that we have at this moment but you also have to keep in mind that it depends on the objective of the study so here the objective of our study at this moment was to make international comparisons based on what we have and unfortunately we don't have the resources to go and collect the data by ourselves what we can do we compute and we can try to influence those who are collecting the data to include those further things honestly in the Indian context I know the the pavement dwellers they're not included in the service because this is only household service so we have no way of understanding their situation so these household surveys are meant to be nationally representative because you cannot collect a massively large sample has to be restricted by budget but the coverages can only improve over time and even if we want to go and try to collect probably we'll be able to collect data on those particular people but maybe not the others it's a complex problem but I completely take your point that there is need to go beyond what we have at this moment collect information on more dimensions as well as more people to understand and the data reflect the reality last peter anything you would like to add no okay then we are at the end thanks again to the presenters thanks for coming to this session two further announcement first the lunch is downstairs and then if you're interesting to connect a notebook or a smartphone or whatever to the internet I received these instructions and I think it would take this one sorry too much time to explain to you so I just leave it here and you can come to the desk and then check it out thank you