 This paper sort of came out of conversations between myself, my co-author, and the National Statistical Agency. We were sort of interested in the context of a highly unequal society, what would be the differences between people who we measure as poor and people who perceive themselves to be poor. It seems like a fairly sort of simple question, but luckily the answers are a bit more complex, which allowed us to write what we hope is an interesting paper and to look at some of the differences. When we started both in our discussions and looking with the data, we discovered what you might expect in a country like South Africa, that many households that were very, very wealthy in terms of having several cars, several high earners in the household were also identifying themselves as subjectively poor. There are many reasons we could think of this, and I think some of those were covered in the previous presentation. But it got us to thinking more about what we were both really interested in, and that's the sort of overlap between objective and subjective measures of poverty at the lower end of the income distribution. And in particular, what does it mean when these two measures of poverty don't overlap? What can it tell us? I can see this is a fairly non-South African crowd. So I'll start with some of the context of South Africa. It's widely documented that objective rates of income poverty in South Africa have remained stubbornly high in the post-apartheid period. We see somewhat of a modest decline in income poverty in the 2000s after massive investment in social expenditure by the government, but there's a strong sense that the dip in poverty or the progress in poverty reduction is not in line with the amount of expenditure, and in particular, pro-poor expenditure. Not surprisingly then, the post-apartheid government has been sort of displeased and very surprised by our findings in terms of income poverty. It's very hard to reconcile the types of spending, particularly on primary health care, education, sanitation, and housing, which we grew broadly under the term social wage. It's very difficult to explain why this type of investment, and in particular, the expansion of a fairly comprehensive social grant system, hasn't yielded better results in terms of poverty reduction. In particular, government points to during the period where much of the debate about income poverty was happening really incredible expansion of electrification to households that didn't have it before, access to piped water, and in the period in which we tend to debate trends in poverty in South Africa, six million South Africans receive government subsidized housing. So that's the one side government is sort of incredulous at our find means that the reduction in poverty has been marginal, but on the research and academic side, recent poverty analysts have also noted problems that would suggest we're underestimating households resources when we measure income and expenditure. In particular, that we have large numbers of missing data in some of our income and expenditure type surveys, implausible zero income households, even in households where there are employed members, as well as the sort of usual concerns you have with disclosing sensitive information, all of which has contributed to the debate amongst academics about the possible underestimation of economic resources in the household. So the objective objectives of this study are to explore an alternative way of measuring poverty where we use respondents own perceptions of economic well-being in their household. And then we then want to learn to see what we can learn about these objective poverty measures when they don't overlap with subjective assessments of poverty. Given the sort of state of the debate in South Africa, we're then interested in seeing whether the components of this government expenditure broadly that termed the social wage affects perceptions of poverty. And then what differences between subjective and objective poverty measures, whether they're consistent with the underestimation of resources and households, which we suspect from the data that we have. I don't need to go into too much detail here. The previous presentation sort of set the stage for subjective literature. And I think this crowd in particular is very aware of the steps that are involved, the assumptions that you're involved when deriving objective poverty measures. Among other things, it requires whether you decide to use income and expenditure information, how to adjust for non-response measurement error, differences in cost of living across regions and groups, and differences in types of households by size and the type of people who live in them. Not to mention the decisions that go around identifying a poverty threshold. The advantage then of subjective assessments is they don't require these types of assumptions about how to adjust for household resources and some of the things I just mentioned. And critically, they don't depend on some sort of predetermined or expert-derived threshold of poverty. Some of the literature has also suggested that it's easier to report than income, somewhat less sensitive, and there's been an argument that there's no obvious reason why people would not be willing to assess their own economic status. Additional advantages we might expect from a subjective measure of poverty include the capturing of longer-term measures of welfare, things that aren't sort of a part of current or aren't captured adequately by current income or expenditure, as well as a wider range of welfare components, which we would expect would include things like state-subsidized housing, access to basic services, education, and health. There has been a growing body of work on subjective poverty internationally, and it seemed to us that the focus was largely on not replacing objective measures with subjective measures, but seeing if there are ways to combine the two. And in particular, among this research, one of the main themes seems to be calculating a subjective poverty line based on subjective assessments combined with objective data. However, a few studies are doing something which is more closely in line with our main interest, which is the comparing subjective and objective measures and profiles and seeing if there are actually systematic differences across a range of characteristics and what these differences might say about our traditional measures of objective poverty. So in terms of this specific body of work which is looking at the differences between objective and subjective, there seem to be several main conclusions. One is that it's possible that objective poverty, particularly when using the per capita measure of income or expenditure, are assigning the wrong weights and in particular to economies of scale in household spending, as well as the different consumption needs of different types of household numbers. One of the typical findings across a range of fairly diverse countries is that objective and subjective poverty rates diverge particularly with household size. Another reason that's put forward in the literature is the low dimensionality of the objective economic measure, simply meaning that income is only one component of well-being and subjective poverty assessments, therefore widen that beyond income or expenditure. And finally, returning to the point that there's like the underestimation of objective economic welfare, income and expenditure and a number of the surveys that we tend to use and further suggestions that this is often exacerbated with small scale activities such as subsistence farming. Turning to our methods, we have a recently released data set in South Africa. It's a large scale nationally representative survey and for the first time in South Africa, it combines comprehensive income and expenditure information with a number of subjective well-being and subjective poverty questions. The question we use, arguable how it sort of compares with something like subjective well-being is the direct subjective poverty question, where one member in the household is asked to assess the poverty status of her or his household with options ranging from wealthy, very comfortable, reasonably be comfortable just getting along poor or very poor. We then classify following the work of others, poor as those who identify themselves as poor or very poor. In terms of objective poverty, we include a lot more detail in this in the paper, but we use the national poverty line of 577 rands per month per person. Very, very roughly, if you divided this by 10, it would give you a loose but very fuzzy concept of what it is in dollars. And the data that's in 2000 prices and the data are in 2000 prices, sorry, 2008. So our goal here is to offer descriptive analysis of the ways that objective and subjective poverty measures differ by characteristics. And then we conclude with an econometric analysis where holding expenditure constant, trying to identify what other characteristics predicts a household's assessment of their subjective poverty status. This table would sort of suggest that there's a very high degree of overlap between subjective and objective measures. I just hold it down. So about 49% of households are neither subjectively poor nor objectively poor, and about 20% of households are both objectively and subjectively poor, which means that about 69% of all households have the same objective and subjective poverty status. As we expect, there seems to be some sort of linear sort of relationship between average income, or in this case, expenditure, based on these sort of categories. So households that are both objectively and subjectively poor are far below the poverty line of 577. Households that are objectively poor but don't perceive themselves as poor, slightly better off, and then the relationship with expenditure increases beyond that. So among all households that are identified as objectively poor, 60% self-assess themselves as poor. I don't know if there's too much points in comparing with other contexts, which use different questions, but it seems that this overlap is slightly greater than in other papers that have looked at this. Of those 40% who we measure as poor but don't assess themselves as poor, most of them identify themselves as just getting along, which is, I think, what we would expect. This table then, I would suggest, would represent our headline finding. So if people asked us what's the subjective poverty rate versus the objective poverty rate in South Africa, we would say that 33.8% of households in South Africa are below our poverty line, and there's a big jump in terms of proportion of individuals. 47% of individuals live in poor households, which sort of follows the conventionalism that poor households tend to be larger. If we look at subjective poverty, it's significantly higher overall in terms of households. 37% of households are subjectively poor, and a slightly smaller percentage, 39.5% of individuals are subjectively poor, suggesting, really, that there's a big difference between household size and objective and subjective poverty measures. So if the question is asked, are South Africans subjectively poor? More likely to be subjectively poor than objectively? It would certainly differ depending on whether we're talking about households or individuals. So this graph sort of gives a representation of what happens as household size increases in terms of the incidence of poverty. So with very small households, the red line is subjective poverty. The level of the incidence of subjective poverty is much higher in small households. It declines, so there's some sort of convergence between households with three or four members. Thereafter, for both objective and subjective poverty, the risk of poverty increases, but far more dramatically, the line is steeper for objective poverty. So we can see what we've found in other studies that there seems to be some sort of difference between objective and subjective measures with household size. In addition, when we're talking about the share of household members, which are under the age of 11, households with no children are far more likely to be subjectively poor than objectively, and that it increases as the risk of poverty increases in both poverty measures as the share of children increases. But again, the increase is much steeper for objective poverty. So it seems that the relationship between the proportion of children in a household and poverty is stronger in objective measures. We were also interested in a number of other characteristics which we sort of hypothesized things, and also considered in the literature, the first of which is what type of area are households based? So subjective poverty is higher than objective poverty in all areas except for tribal authority areas where objective poverty is higher. This could be for a number of reasons. It could be due to limited horizons associated with relative deprivation, or in the South African context, this denotes households that are deeply rural and far more likely to be subsistence and agricultural based. We can explore this further. Households with land for farming have far higher objective poverty rates than households that don't. Again, this is a proxy for households in rural areas. We see the same for subjective levels of poverty, but the difference is much smaller. And then overall, households with access for land for farming are much more likely to be objectively poor than subjectively poor. Perhaps an indication that access to land for subsistence activities is a protector of subjective poverty. We do the same for the access to a dwelling and find the same results, but keeping in mind ownership of dwellings in South Africa is often representing ownership of shacks and informal types of homes, not the types of homes you would imagine in other contexts. So what we wanted to do then was to estimate two probits to where we're looking at the predictors of subjective poverty. It's a binary outcome. Our variable of interest is of course per capita household expenditure normalized by the poverty line. And we control for a group of demographic income generating asset characteristics variables grouped under the social wage, as well as the local income average expenditure in the district, as close as we could get for a reference group, as well as the Gini coefficient for the district. In the first regression, we control only for characteristics of the household. And in the second, we were able to work with the data agency to identify which household member was reporting on subjective poverty and providing the assessment. So we control for their characteristics as well. Before we get to the full regressions, we did an estimated a reduced model controlling only for household size and household composition. And we found that the significance in the model on household size and share of children disappeared with an economy of scale parameter of 0.42 and child cost ratio of 0.5 for younger children and 0.9. So we find that compared to the poverty literature in South Africa, we find a much stronger effect of economy of scale suggesting that this is underestimated in existing poverty studies. Not surprisingly then, income is a strong protector against being subjectively poor as our household size and household composition. So after controlling for expenditure, larger households and households with more children are less likely to be subjectively poor. There's also an effect for pensioners, which is very likely related to the social pension in South Africa, which provides almost $100 a month to pensioners in South Africa. But it looks like there's an effect of that grant over and above the effect of income. We find that self-reported health, access to farming land, the number of employed in the household, assets and quality of housing, as well as basic services such as piped water at the dwelling and electricity are all protective of subjective poverty over and above at similar levels of income. Carrying on in the first regression, we find that Africans, for a number of reasons we propose in the paper, are still more likely to be subjectively poor, controlling for all other factors, and we find the effect of relative deprivation. The higher the level of the people in your district, households are more likely to report being subjectively poor. In the second model, we control for these characteristics, age, education, gender, self-reported health, emotional and physical disability, and employment of the household member reporting, providing the self-assessment, and we find that these variables are also all significant, but they don't change the significance or the direction of the sign on the marginal effects from the first part of regression one. So we find that it doesn't change our conclusions. So what do we take from this? We find considerable overlap between subjective and objective measures of poverty in South Africa, but where these measures do not overlap, it seems that subjective assessments are affected by a wide range of factors in addition to current income. We think we find some evidence which suggests that expenditure and income are underestimated in our household surveys, in particular per capita measures, which don't take account scale economies, and it seems that there's some evidence that small-scale, deeply rural, agricultural activities aren't captured accurately in terms of current resources. Implications for future work on poverty in South Africa. The social wage is highly protective of subjective poverty. Social grant contributions contribute to subjective poverty reduction over and above their actual income contributions. And finally, that income, that considerations about household size and composition probably deserve more attention in the South African poverty literature. Thank you.