 Good afternoon, so Firstly I wanted to thank Finn and others at wider for organizing such a wonderful meeting and Special thanks to those of you who are in the audience It's been a long couple days in a positive way, but if like if you're like me your minds are already kind of filled with ideas and great thoughts and Have learned more than your carrying capacity, so I will try to At least end on a good note and talk about a paper by my colleague Stephen Younger and and and myself that we've been working on over the past couple months and The title is a bit is about as bad as the Terms that we use that we made up at the very last minute in the course of writing the paper but it's the incidence of recent health improvements and Let me begin by talking about the Motivation for the paper and let's see if I can I'll just do it this way. Okay So Broadly speaking, I think there's a general consensus, especially in Africa where growth has lagged or had lagged quite seriously in the last millennia or last Prior to The past ten years that growth has finally picked up and that's the case of course in many poor regions of the world But as this conference is motivated by the fact that Still there's a great deal of concern that the fruits of this growth are not equivalent equitably distributed and this has led to a Kind of burgeoning literature on not just inequality per se but issues of poor growth or sometimes called inclusive growth But the vast majority of that literature Almost all of that literature Focuses on one fruit as we call it income or expenditures some money metric of well-being and In this paper We're going to look at a at something else. We're want to look at a different measure of well-being child health and Thus the paper is kind of motivated or tries to bring together two broad literatures That both Steve and I have worked on quite a bit over the years together One is the literature on broadly speaking on improving living standards and poverty reduction, but In the context of thinking about that in terms of distribution of growth but the other is the literature on multi-dimensional poverty and And in this case the dimension that will focus on is Health although this could be done for other dimensions or be done in a multi-dimensional framework in a way that Steve Joan Eve DuClo and I have done in some other work so The purpose or the question that we are trying to Answer is a simple one and it's whether and to what extent improvements in children's health Have been distributionally progressive or pro-poor however you want to Whatever terminology you want to use and Thus we kind of address three questions In that regard the first is We ask whether there are Inter-temporal changes or whether the inter-temporal changes in health status and in terms of the distribution of health status are similar or Do they look like the inter-temporal changes in expenditure distributions? and In order to answer that question we take two Approaches one is to ask the question of whether the health improvements Look at health improvements across the income distribution basically the analogy is kind of the Gradient approach to looking at health inequalities and the second approach that we take to looking at this question is We look at how or the extent to which health improvements are Distributed across the health distribution where instead of ordering well-being along The dimension of income we actually order melt well being across the dimension of health And that's kind of analogous to what's referred to as the univariate Measures of health inequality, okay So how do we go about doing that? Well, let me start off with just talking a little about the data. We're going to use here So the health indicators that we're using in this data all come from the demographic health surveys The reason that we rely on the demographic health surveys is are to the first is that they Span a long period of time. We're really not interested in short spells of two or three years Primarily because like income there's a large stochastic element to health status and You know whether there's some An epidemic of diarrhea high rates of prevalence of malaria or whatever the issue is in a community will result in very Small but important changes from one year to the next so our interest is covering a long span of time and The only data sets out there for the most part that are comparable and that collect health data in a rigorous fashion Are the demographic and health surveys demographic health surveys? Okay, and for many of the countries that we're talking about here This spans a 20-year period of time or so we have looked at shorter spells in this paper We're just going to present the longest spells of each country and a Second reason for focusing on the Demographic well, let me say the second reason focusing on demographic health surveys as I started to intimate is that while there are other Surveys that collect health data The quality of that data, especially when people do things like impute infant mortality rates or look at even Anthropometric data and most of the LSMS type surveys the quality of that data is is very variable to put it nicely Some surveys do a really good job But for the most part the training and the effort to collect that data was of secondary Import and I can tell you that for sure having spent 10 years in the LSMS unit in its early years So the DHS data really does focus in on that But one of the problems is the DHS data do not have expenditure data or income data So how do we solve that problem? Well, we really don't fully solve it but basically what we do is for every demographic health survey we find a comparable household expenditure survey in that country for usually within a two or three year range and Simply what we do is we predict or a level of household per capita expenditures Based on a it's a projection based on a set of household characteristics that are both available in the LSMS type of survey or a household consumption expenditure survey and the DHS survey So this works out pretty well I can talk more about it at some point if people have questions about it how we verify and Look at the robustness of that, but it actually works out pretty well so the other thing to mention at the beginning before we start looking at results is that The inequality measures we're looking at here are for samples of kids samples of children They're not for households, but they're for individuals. So this has certain obvious disadvantages It's for a select group of the population, right old people or Households without children are not represented here. So the upside of a course is that Most or all household based levels inequality measures assume that there's equal well-being within a household and So when we look at income inequality Everybody in the household is ascribed the same income and assumed to have the same level of well-being now Steve and I have done some earlier work of earlier papers There's also been a paper by I think when Robbie Combour was at wider I only know of our paper and his paper that actually try to get at the issue of within household inequality And decompose within versus between household inequality and we found actually that within household inequality is Greater than between household inequality. So if you believe that Your kind of would be motivated to say yeah, this is a really great idea to look at inequality at the individual level So there's an upside to it, but as I say, there's a downside to it. Okay so So much for the data actually I'm gonna skip this slide because of time but just to say that in fact If we look at health indicators broadly over the past 20 years They've increased quite dramatically and in fact more so than income, but let me go right to the methods and Go through this relatively quickly so I Presume that most of you if not everybody in this room is acquainted with kind of the rebellion and Chen Tools that they've developed for looking at growth incidents Okay, these growth instance curves that they have measured and It's basically a simple tool where they take to you can use to cross-sectional data sets Comparable data sets for examining whether economic growth is pro-poor or not and Basically the tool or the technique Generates a set of curves that ask or show How much income growth has taken place at various quantiles along? The income ordering so these curves Kind of look like this and this is one that we generated using the Methods I talked about where we map on the income expenditure data is from the income expenditure data onto the DHS data and You can come up with these growth incidents curves Okay, and this one is actually quite typical of the ones in the paper and the ones that we find and it's typical along two dimensions one is that as I said for both countries in Africa as well as the rest of the world now We see that growth has increased so this is the difference in log per capita expenditures And this is just the percentiles of household expenditures across this axis and Then we draw this growth incidence curve following The methods of a value than dot and others we estimate standard errors around it and we can say that Yes household expenditures per capita have increased throughout the expenditure Decisal across these two decades between 1988 and 2011 But the important point here for those of you who are interested in distribution is The increase in growth is much greater amongst households at the upper end of the income distribution Okay, this comes as no surprise to many of you out there, and that's not what this paper is about Now of course some people in the audience will say yeah, this is pro-poor growth because the poor witness growth Okay, others of you are going to take a Relative definition of pro-poor growth and say no this is not poor pro growth Because the growth amongst the poor is slower than the growth amongst the rich I'm not going to debate that point But the clear message here is that when we look at most growth incidence curves Not only when we do but when others do they generally tend to look like this Where there's more growth at the upper end of the distribution? I'll leave it to you to just determine whether that's good bad or otherwise And here's some other examples from our mapping exercises using the DHS And again that allows us to look at these long periods that we couldn't do with just household expenditure surveys But I'm not going to talk much more about these so I'm just going to flip through them and get to what I want to talk about Which is what I call first the gradient health improvement incidence curve and there should be a G in front of that H But this is an example of terrible nomenclature So if somebody can come up with a better name for these curves, please let us know We're not terribly creative in that regard at least So so what are these gradient health incidence curves health improvement incidence curves? Okay, so basically what we're asking here is what we want to know about is the distribution of health Across the income distribution. That's what these curves do these gradient health improvement incidence curves and The basic question is simply whether health improvement is Larger for children in income poor households or larger for rich richer households and we can answer that question Basically by relying on this term which is either the height of children or the predicted infant mortality probabilities associated with the With a with a specific quantity of the income distribution Okay, so I'm going to And as I started to mention we have to do this this requires relying on a regression which we do non-parametrically using local linear regression techniques and One advantage of this approach actually is that allows us to handle the screen indicators like infant mortality Which the next set of curves will not be able to do Because we can look at predicted probabilities using this technique. Okay, so what do these curves look like? again, I'm going to rush through this stuff in Deference to the time that we have allocated But if just think about the curves that we saw or that one example that I put up in the big screen of Uganda Where the curves kind of were going in this direction? So here's the growth incidence of infant survival For Ghana actually, this is Uganda Like I left the label off But look at this curve. So what this curve basically says is Is that in this case? the probability of surviving to one's first birthday of a child surviving to their first birthday or in this case being taller, which is a good measure of health is First increased across the entire distribution, right and it increased quite dramatically Three centimeters of standard or two centimeters of standardized height or one centimeter standardized height Equals about 1.5 standard deviations of children's height at that level. That's a big increase now to you one centimeter sounds small But at the population level it can take countries decades to do that So in both cases infant mortality rates decline because it's above the zero mark and Heights increased all that's great, but the big story here is unlike the income data. We looked at Notice that the distribution of benefits is at the lower end of the income distribution and again We're ordering household or individuals here according to the household income distribution so the bottom line is that the largest gains in health are Concentrated at the lowest end of the income distribution and that seems to be a general finding across All almost every country we looked at what button Okay, I'm sorry. This is the Uganda case. I guess I didn't get to label them all But the only cases that you do not see that happen Is and I'm sorry some of the labels didn't come out our countries like Actually, I Like Cameroon or and I forgot which country this is my apologies For some reason it didn't come through. I think that's yeah, this is these are both Cameroon where there's been little improvement So when there's virtually no improvement the curves tend to be flat But where there is any substantial improvement The curves tend to always slope downward increasing the improvement is concentrated at the lower end of the income distribution Okay, so now there's one other type of curve that we can look at and That is simply the health improvement incident curve So what's the difference between the gradient health improvement incident curve and the simple health Improvement incident curve. Well in this case, this is analogous to the univariate approach looking at health inequality Where the ordering along the Y the X axis is Actually not in terms of income, but in terms of health status per se okay, so Basically again, we're looking at two minutes Okay, so let me just instead of trying to clarify it just show you the results These results tend to be more mixed where we order individuals across the distribution based on their health status We find that indeed in certain countries for example Peru and Bangladesh Again, the benefits are concentrated at the lower end of this the distribution in some other countries However, this is a little more ambiguous These let me find one okay Cameroon and Ghana are good examples where the benefits tend to be In the case of Ghana where there were benefits more Concentrated in the upper end of the income distribution so in the terms of the growth incidence The health improvement incidence curves where we order well being across the high distribution the results are are a little bit more mixed So let me go to the overall results of of the study of what we know so far okay, so basically The experience for more work, but mostly more important from the vast literature on this in this area suggests that Traditional expenditure-based growth incidence curve tend to be regressive or at best Distributionally neutral and that's generally what one finds in the literature in our case it applied to all the countries that we've looked at except for Peru Okay, but what we've also found is that the distribution of benefits associated with health improvements Differs dramatically from income and almost is always pro-poor meaning pro-poor in absolute terms but also pro-poor in relative terms and they're pro-poor in a relatively dramatic way and the other thing is That the countries that have that observed or witnessed the most substantial health improvements a country like Uganda or a country like Ghana are also the countries that Tend to have the most pro-poor improvements in their health status, okay So it's countries like Cameroon that hardly grew in terms of its income and which hardly Improved in terms of its health status that these growth that these health incidence curves tend to be flat but every country that witnessed a large or dramatic improvement in health also in The very progressive change in health over the time period, okay, and then The so again just skipping to the these health improvement incident curves They tend to be more mixed than where we order well-being by income and We find that in non-African countries particularly in the three or four non-African countries We've looked at less less healthy kids grow more for example Columbia and Peru were examples of that, but there are some cases for example Madagascar We're actually the healthier kids. We're growing more across the spells that we looked at So what is the bottom line of all this? It's that we cannot predict what the gradient health improvement incidence curves look like or the health incidence curves will look Like based on the traditional growth incidence curves that are used income or expenditures so basically the incidence of income growth and health improvements are not the same in a country and while we are Rightfully, I think or correctly somewhat agitated or worried about the lack of relative progressivity in terms of income growth The good news is that when we're looking at improvements in another measure of well-being particularly children's health Not only do we see quite large improvements Overall, but those improvements tend to be far more concentrated at the lower end of the income distribution scale So thanks very much