 V tem paredu sem tudi zafrečen, da nekaj radačnega vedenčno glasba neko nekaj rov, nekaj nekaj nekaj rov, basi etniciti, religijno, predaj so povihne grav, vzveči. Zato je to vseto preko vsega etniciti in srečnega vzvečna. This is part of a larger agenda in research about differences by erasing ethnicity in several countries in Brazil, other Latin American countries, the US, China and in South Africa. In terms of differences in distribution mostly on the poverty levels and outcomes in the labor market. In one of these papers I previously analyzed the case of the huge white black differential in South Africa showing how much of this gap could be explained by Africans having lower endowments or lower access to education living in most rural and poorest areas of the country, etc. In this case I moved to analyze differences within Africans, so blacks in the country taking into account the different ethnolinguistic groups. So the aim is to explain if there are significant differences between poverty among black ethnic groups and I advance that the answer to my view there is significant differentials. Comparing with differences in other countries they are very similar or even larger than the black white poverty gap for example in the US or Brazil. There is not only differences in poverty rates but also in the probability of entering into poverty, falling into poverty. The second point is trying to say something about these differentials where they come from. As to what extent these differences can be associated with the different characteristics of the groups, so this kind of a compositional effect depending on the access to the different endowments. Is this because some groups have more children because they live in rural areas or townships or they have lower access to education, lower employment rates, etc. And the answer is yes, most of the gap is explained especially when we concentrate on persistent poverty, much less when we look at temporary poverty. And the third objective is to identify what of these characteristics are the most important and I will show that this is the accumulation of several disadvantages across these groups but with some differences between each of the ethnic groups. The data is the first wave of the national income dynamic study produced by the University of Cape Town. So I use a cross section of wave 1 in 2008 and a balance panel of waves 1 and 2 when they were reinterviewed again in 2010, 2011. I concentrate on differences in poverty rates using the standard poverty line using other studies in South Africa. And I will look also at the dynamics of poverty separating those who were poor in 2008 and see if they were poor again in the second interview or not. And then I call the first one persistent poor and the second one is temporary poor even if this is an abuse of language because I don't know what happened in the middle. These are the ethnic groups I'm going to analyze. Kosa, Sulu and I joined Soto and Suana so they in order to have significant groups, large groups, you can see that there is some geographical disparity of these groups. They tend to be concentrated in their traditional areas even if there are people of all groups everywhere. Looking at the per capita monthly household income in rents, you can see that there are different income distributions when we partition the group of Africans by ethnicity with Kosa being the poorest groups, looking very close to Kosa and Soto, Suana being the relatively affluent groups among Africans. Then other groups that I'm not going to analyze lying in the middle of between, in between both other groups. Well, this is the summary indicators, the average income, the percentage of the population. You can see that there is a significant difference in the median income and what I'm going to concentrate, there is a significant gap in poverty rates. Poverty rate for Sulu is 62% for Soto, Suana 41%. This is 50% higher, larger than the many ethnic differences in other countries. Obviously, the difference within African groups are very small if you compare with the differences with other racial groups in South Africa, whites and color or Asian, Indian. This was somehow the analysis of the previous paper, but I think that also the differences within Africans are very important in some explanation. The methodology I'm going to use is to explain this differential in proportions in head counts using a blind or a hacker type of approach, particularly in extension to analyze the differences in proportions. The original was to analyze the difference in means of continuous variables. So I estimate for each group a poverty regression, a logit regression of probability of being poor based on household characteristics. Then I know that the head count ratio is just the predicted value, the average predicted value of the probability of being poor. And then I use this to decompose the differential in the poverty of one group and the other. It's the same as the difference in the predicted values and then I use a counterfactual distribution in which I give the characteristics of the affluent group, the reference group, soto suana, to the other group, either sulu or cost, while keeping the around beta, the around coefficient effects. So this is how the counterfactual would happen if one group had the characteristics of the other group, but keeping the impact of these characteristics on the probability of being poor. And using this counterfactual, subtracting and adding this counterfactual, we generate this one-hackabinder, this composition, one is the characteristics effect, that is the effect of switching the characteristics while keeping the coefficients, and the coefficients effect, that is what happens when you keep the characteristics and switch the coefficients. So the first is what I'm concentrating because it's the part that we really explain, that is how much of the differential can be explained by some groups having lower endowments. Then I produce the detail, the composition, that is to attribute what is the contribution of each factor, and I use for that even in McPherson and June approach, because the fact that the logic is nonlinear generates some problems and they make a linearized approximation. The characteristics I'm controlling for are basically locations, so urban formal or urban informal, they live in tribal authority area or rural formal. Some demographic variables, so head, mortal status, migration, age, sex, the number of children, the number of adults, education, not only education of the head but also education of other members of the household, labor status of the head but also the dependency ratio to take into account what happens with the other members of the household, and other variables to take into account the specific structure of this data, taking into account the date of interviews, so the quarter and the time span between both interviews in the case of the panel, and also for analyzing entry into poverty, the distance from the poverty line. And these are the results for 2008, so for the cross-section. This is the result for COSA, this is the result for SULU. So this is the poverty rate in each group, COSA and SULU, the poverty rate of the reference group, in this case SOTO-SUANA, 41%, and this is the differential that I want to explain. Here are the standard errors. As you can see, I can explain in both cases most of the differential. So most of the differential can be attributed to differences in the basic characteristics of households. It's like more than 80% and near 90% in the case of SULU. And what factors are most important in both groups? Well, they have something in common but there are some things that are specific of each group. They have in common, for example, the important role of education. So COSA and SULU have lower levels of education, and this explains about 4% points of the gap, which is a large differential, about 20%, more than 20% of the observed differential. So this is similar in both groups. What is more specific of SULU is the huge importance of the number of children that explains about 8% points of the differential. So what is 40% of the whole gap that is observed. This is the combination of having more children and having a higher impact of the children on the probability of being poor. In the case of COSA, this is also important, explaining about 3% points, but much less than in the case of SULU. Also in the case of SULU locations, so the fact that they live in larger proportions in tribal authority areas and in informal, well, basically here is in tribal authority areas, explains about another 4% points differential, or 20%. In the case of COSA, they also have higher proportions in tribal authority areas and in informal urban areas, explains about 2%. What is more important for COSA compared to SULU is the labor attachment of household members that is explained about 5% points here and higher than in the case of SULU. Other demographic factors are important, but to a lower stand. Well, then I want to look also at the dynamics of poverty, so to take into account the fact that we have this second interview so we know who remain poor in the second survey and who exit and who entry poverty during that time. As you can see, this is the differential in poverty rates that we saw, so poor in 2008. This is what I call persistence, so poor in both periods. This is the temporary poverty rate, so those who were poor but only in the first period, this is the entry rate and the exit rate into poverty. As you can see, there is the gap in persistent poverty similar to the gap in poverty and there is an important thing here that there is also a huge differential in the entry rates. There are no substantial differences between exit rates of these routes but the differential comes from the entry rate. So COSA and SULU, they have a much larger probability of entering poverty than Soto Suana and that might help to explain their higher poverty rate. So I do the same type of analysis but looking at this persistent temporary poverty temporary poverty and entry rate. So these are the values, again for each group, for the reference and this is the differential I want to explain. In this case, I produced here, this is the result we saw earlier so it's the cross-section 2008. This is the same but using only the balance panel so I expect to have the same results. The difference comes from the attrition that is not corrected by using attrition weights but I think the results are quite similar and this is the result for persistent poverty, temporary poverty and for the entry rate. And as you can see, the unexplained part is very similar in the case of the persistent poverty rate so most of what we explained is actually the difference between persistent poor and in the case of temporary poverty we explained something in the case of Sulu but nothing basically in the case of Kosa. So it's what really remains to be explained. And in the case of the entry rates that there are also significant gaps we explained all the gap in the case of Sulu and nothing or only a third of the entry rate in the case of Kosa. So in the case of Kosa it's more difficult to know what is happening in the case of temporary poverty or people entering poverty. What are the factors, the relevant factors as we saw, I mean in the case of persistent poverty the factors are very similar to those we saw later, so earlier. There are some things that, there was some difference before that now disappear when you concentrate on persistent poverty. For example in the case of location that they look much more similar now than they did before. Education is still the most important factor. Labor is also very important and the difference in the number of children. In the case of temporary poverty for Kosa we don't explain anything in the case of Sulu we see that there are some, I mean we explain about 3.7 percentage points and most is explained by education. So education explains higher temporary poverty for Sulu but not for Kosa and this is basically the main explanation and there are some demographics like the age of the head that also is somehow explicative. And regarding the entry rate in the case of Kosa we don't explain anything in the case of Sulu we can say that it's basically a compositional effect and some part comes from the fact from the design of the survey some part also comes from the fact that Sulu are closer to the poverty rate so it's easier for them to fall into poverty than for Soto and Suana and also the demographics are especially important here you have basically the number of children so more than one and a half percentage points differential but also the different family structure seems to be very important here but basically demography is also the location I also look at the differential of trends in time but I cannot explain much I compare in 2008 with 1993 the poverty studies with another similar survey, the PSLD that is very similar but with a different income definition so I have to homogenize the definition of income for the reason the values for 2008 are different from before and what we see is that the reduction in poverty for Kosa is very similar to the reduction in poverty for Soto and Suana however the reduction in poverty for Sulu is much smaller but the difference is not statistically significant so we cannot be very sure that the gap really increase so just concluding, I think that understanding the inequalities across ethnic groups might be important to better know what is happening or what is going on in the income distribution of South Africa and even if most attention is concentrated on the black-white differential I think that there is room also to analyze what is going on within the different ethnic black groups difference in poverty rates are to a large stand associated with Kosa and Sulu having an accumulation of disadvantages in location, demographic structure, education and labour market and with some distinctive features the higher importance of the labour market for Kosa because there is more urban group and the higher importance of the demographics of the number of children in the case of Sulu I also think that it is important to take into account the different time profile of the poor because we explain better differences in persistent poverty than in temporary poverty and also that it is important to analyze the increasing gap even if the gap increase for Sulu we are not sure if because this is not statistically significant and there are also some changes in the variable which is playing in both gears so thank you very much