\AIPS .2o1I POLtICY RESEARCH WORKING PAPER 2011 Measuring Poverty Using Subjective poverty lines- based on the self-assessed Qualitative Perceptions adequacy of a family's food, of W elfare housing, and clothing - accord closely on average with independent "objective" Menno Pradban poverty lines. There are Martin Ravallion notable differences, however, when geographic and demographic poverty profiles are constructed. The World Bank Development Research Group Poverty and Human Resources H November 1998 I POLICY RESEARCH WORKING PAPER 2011 Summary findings Pradhan and Ravallion show how subjective poverty lines The implied subjective poverty lines are robust to can be derived using simple qualitative assessments of alternative methods of dealing with other components of perceived consumption adequacy, based on a household consumptiorn, for which the subjective "adequacy" survey. Respondents were asked whether their question was not asked. consumption of food, housing, and clothing was The aggreg,ate poverty rates based on subjective adequate for their family's needs. poverty lines come close to those based on independent Pradhan and Ravallion's approach, by identifying the "objective" poverty lines. subjective poverty line without the usual "minimum- There are notable differences, however, when income question," offers wide applications in developing geographic and demographic poverty profiles are country settings. They implement it using survey data for constructed. Jamaica and Nepal. This paper - a product of Poverty and Human Resources, Development Research Group - is part of a larger effort in the department to improve methods of poverty measurement. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Patricia Sader, room MC3-632, telephone 202-473-3902, fax 202- 522-1153, Internet address psader@worldbank.org. Martin Ravallion may be contacted at mravallion@worldbank.org. November 1998. (38 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than folly polished. The papers carry the names of the authors and should be cited accordingly. The findings, inte1pretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Policy Research Dissemination Center Measuring Poverty Using Qualitative Perceptions of Welfare Menno Pradhan and Martin Ravallion' Correspondence: Martin Ravallion, World Bank, 1818 It Street NW, Washington DC, 20433, USA. Ravallion is with the Development Research Group, World Bank. Pradhan is with the Economic and Social Institute, Free University Amsterdam. These are the views of the authors, and need not reflect those of the World Bank. For their helpful comments, the authors are grateful to Kees Burger, Klaas de Vos, Francisco Ferreira, Arie Kapteyn, Peter Lanjouw, Remco Oostendorp, Lant Pritchett, Dominique van de Walle, Bernard van Praag, the Review's three anonymous referees, and participants at the 1998 Conference of the Association of Income and Wealth, Cambridge, England. 1 Introduction The most common practice in drawing a poverty line starts with "objective" pre- determnined nutritional requirements for good health and an active life. The poverty line is then defined as the value of a monetary measure of individual economic welfare, such as expenditure on all goods and services (with imputed values when necessary), at which these nutritional requirements are met given prices and reference tastes. People are deemed to be poor if and only if their welfare indicator is below this line, and a poverty measure is estimated on the censored distribution (such as the "headcount index" given by the proportion below the line). Methodological differences within this approach are known to yield different poverty measures.2 However (as has been noted before), there is an inhe:rent subjectivity and social specificity to any notion of "basic needs", including nutritional requirements. For example, psychologists, sociologists and others have argued that the circumstances of the individual relative to others in some reference group influence perceptions of well-being at any given level of individual command over commodities.' By this view, "t:he dividing line ... between necessities and luxuries turns out to be not objective and imnnutable, but socially determined and ever changing" (Scitovsky, 1978, p.108). Some have taken this view so far as to abandon any attempt to rigorously quantify "poverty". Poverty analysis (particularly, but not only, for developing countries) has become polarized between the "olbjective-quantitative" schools and 2 Ravallion (1994) gives examples. For a critical overview of alternative methods of setting poverty lines found in practice in both developing and developed countries see Ravallion (1998). 3 Runciman (1966) provided an influential exposition, and supportive evidence. Also see van de Stadt et al., (1985) and Easterlin (1995). 2 "subjective-qualitative" schools, with rather little effort at cross-fertilization. An intermediate approach has emerged in a segment of the developed country literature on poverty. "Subjective poverty lines" have been based on answers to the "minimum income question" (MIQ), such as the following (paraphrased from Kapteyn et al 1988): "What income level doyou personally consider to be absolutely minimal? That is to say that with less you could not make ends meet". One might define everyone whose income is less than the amount they give as an answer to this question as poor. However, this would almost certainly lead to inconsistencies in the resulting poverty measures, in that people with the same income, or some other agreed measure of economic welfare, will be treated differently. Clearly an allowance must be made for heterogeneity, such that people at the same level of living may well give different answers to the MIQ, but must be considered equally "poor" for consistency. Past empirical work has found that the expected value of the answer to the MIQ conditional on income tends to be an increasing function of income.4 Past studies have tended to find a relationship such as that depicted in Figure 1, which gives a stylized representation of the regression function on income for answers to the MIQ. The point z* in the figure is an obvious candidate for a poverty line; people with income above z* tend to feel that their income is adequate, while those below z tend to feel that it is not. In keeping with the literature, we term z* the "subjective poverty line" (SPL).5 4 Contributions include Groedhart et al., (1977), Colastanto et al., (1984), Danziger et al., (1985), Kapteyn et al., (1985, 1988), Stanovnik (1992) and Kapteyn (1994). 5 The term "social subjective poverty line" might be preferable, to distinguish it from the individual subjective poverty lines. However, the meaning will be clear from the context. 3 It is also recognized in the literature that there are other determinanits of economic welfare which will shift the SPL, such as family size and demographic composition, Indeed, the answers to the MIQ are sometimes interpreted as points on the consumer's cost function (giving the minimum expenditure needed to assure a given level of utility) at a point of "minum utility", interpreted as the poverty line in utility space. Under this interpretation, subjective welfare assessments provide a means of overcoming the well-known problem of identifying utility from demand behavior alone when household attributes vary.6 Our main aim in this paper is to develop and implement a q_alitative model of perceived consumption needs which allows us to identify the subjective poverty line without the minimum income question. We believe that our approach has marked advantages, particularly for applications in developing countries. While the MIQ has been applied in a number of OECD countries', we know of no attempts to apply it in a developing country. There are a number of potential pitfalls in doing so. "Income" is not a well-defined concept in most developing countries, particularly (but not only) in rural areas. It is not at all clear whether or not one could get sensible answers to the MIQ. The qualitative idea of the "adequacy" of consumption is a more promising one in a developing country setting. We will demonstrate that one can still estimate the SPL without the MIQ; less demanding qualitative questions suffice,o 6 On this identification problem see Pollak and Wales (: 979), Deaton and Muellbauer (1980), Pollak (1991), and Browning (1992). On the use of subjective welfare assessments to identify cost and/or utility functions see van Praag (1991) and Kapteyn (1994). 7 See for example Hagenaars (1986) for a cross-European comparisons and De Vos and Garner (1991) for a US-Dutch comparison. 8 Other problems might be anticipated in applying the subjective approach in developing countries. It was suggested by some of those we spoke to in discussions leading up to this study that we 4 We also aim to extend the range of variables that one deems relevant to explaining the variance in perceptions of poverty. It is important, we believe, to test whether objectively measured income or consumption has power in explaining subjective measures of welfare in a developing country context; if it does not, then many of the policies that are typically promoted in the name of "economic development" may bring disappointing outcomes in terms of human satisfaction. It is of interest to consider other possible determinants of perceived poverty. An obvious (although by no means sole) source of peer-group effects on subjective assessments of minimum consumption needs is the geographic neighborhood. We will test for effects of neighborhood living standards on subjective assessments of individual welfare in developing countries. The following section outlines our qualitative model of the subjective poverty line. In section 3 we present our results for two (quite different) developing countries, namely Jamaica and Nepal. Section 4 concludes. 2 A qualitative model of subjective poverty lines We assume that each individual has his or her own reasonably well-defined consumption norms at the time of being surveyed. At the prevailing incomes and prices, there can be no presumption that these needs will be met at the consumer's utility maximizing consumption vector. Let the consumption vector of a given individual be denoted y, and let z denote the matching vector of consumption norms for that individual. The subjective basic need for good k may well find that almost everyone thinks their consumption is inadequate in a low-income country. Later we will see that there is little truth to this view for the two developing countries in our study. 5 and household i is given by: Zki (Pk(y, Xi) + sk (k=l,..,m;i=l,..,n) (1) kk P X i + ,ki where (pk (k= 1 ,..,m) are continuous functions, and x is a vector of indicators of economic welfare at a given consumption vector. We assume that each 4k has a positive lower bound as actual consumptions approach zero, and that the function is also bounded above as consumptions approach infinity. The error terms, 6ki" are assumed to have zero mean, and be independently and identically normally distributed for all i with variance i . The distribution functions of the standard normal error terms (& Jak/) are denoted Fk (k 1,..,m). We define the subjective poverty line as the expenditure level at which the subjective minimums for all k are reached in expectation, for a given x. A household is poor if and only if its total expenditure is less than the appropriate SPL for a ]household with its characteristics. Thus the SPL satisfies: m z(X) = E Zk(X (2) k=1 where zk*(x) is defined implicitly by the fixed point relationship: 6(X), X) 1,..,m) (3) 6 A solution of this equation will exist as long as the functions (p are continuous for all k.9 This provides a multidimensional extension to the one dimensional case based on the MIQ, as illustrated in Figure 1. The SPL is the level of total spending above which respondents say (on average) that their expenditures are adequate for their needs. However, we do not assume that the MIQ is answerable, and so we cannot observe Zki directly. Rather we know from a purely qualitative survey question whether actual expenditure on good k by the i'th sampled household (yki) is below Zki. The probability that the i'th household will respond that actual consutmption of the k'th good is adequate will then be given by: Prob ykj> zki) Fk[ykil/k - wk(yi' X)f Ok] (4) As long as the specific parameterizations of the function (pkare linear in parameters (though possibly nonlinear in variables) one can estimate the model as a standard probit. Let us follow the literature on the MIQ and assume a log linear specification for the individual subjective poverty lines. Equation (1) is then: lfnz akPy "> /X. + Łk (k=l,..,m; i=l,..,n) (5) where y- (lny Inylf)Y . 9 This follows from the Brouwer fixed point theorem given our boundedness assumptions. Stronger assumptions are needed to rule out multiple solutions. 7 If we observed the values of zki(analogously to the answers to the MIQ) then a unique solution for the subjective poverty line could be obtained by directly estimating equation (5) and solving, assuming that the following matrix is non-singular (in obvious notation): mm ~~~m1 ~~~~~1 mm The (unique) solution for z - (Inz;,.., Inz* ) is then given by (in obvious notation): z * B='(a+flx) (6) However, the parameters B, Ol and x are not identified when we only have qualitative data on consumption adequacy relative to latent norms. Equation (4) becomes: Prob (yk, > Z,) = F XY(I J(7) Prob (ky > zk*) = Fk[(lnyk)/ak - (ak + jkyi + Ikxi)/ak( As in any probit, we do not identify the parameters of the umderlying model generating the latent continuous variable (equation 5), but only their values normalized by ak. Thus, armed with only the qualitative welfare assessments (telling us Prob (y k > Zk')), we cannot identify the 8 parameters of the model determining the individual basic needs. That fact does not, however, limit our ability to identify the SPL. To see why, consider first the special case of one good with Inz - a + Piny + , . The SPL is a/(I -3). The probability of reporting that actual consumption is adequate is F[lny(l -3)/a-a/al] which only allows us to identify (I -Py)Iand a/a. Nonetheless a/(1 -{) is still identified. This property caries over to the more general model with more than one good, and other sources of heterogeneity in welfare, as in (5). In this case, define the estimable normalized matrix B0, obtained by post-multiplying the B matrix by the column vector formed by CaI (k=1,.,m). Similarly define the normalized vector a, and parameter matrix II, (so, for k example, the k'th element of a. is ak/ok.) It is clear that we can always re-write the solution for the SPLs given by (6) in terms of the observed (normalized) parameters: z * =B'(ci + fI x) (8) Thus we can solve for the subjective poverty line without the MIQ as long as we have the qualitative data to determine Prob yki > zh ) for all i, k. Instead of asking for the minimum income, we simply ask whether current consumptions are adequate. 3 Results For the purpose of this paper, qualitative questions on perceptions of consumption adequacy were added to both the Jamaica Living Conditions survey of 1993 and the Nepal 9 Living Standards Survey of 1995/96. The questions asked are given in Table 1. (For Jamaica a similar question was also included for access to transport.) [n the survey schedule, these questions came after a detailed consumption module. For house owners a rent is imputed based on the quality of the house, facilities and location of the residence. Consumption in kind (including from home production) is valued at local market prices and included in the consumption aggregate. Other information was also collected on a wide range of household characteristics. Aside from the addition of the "consumption adequacy" questions, the surveys followed the reasonably standard practices of the surveys done under the auspices of the World Bank's Living Standards Measurement Study.10 Table 2 summarizes the answers to the questions in 'rable 1. In all categories that can be compared, a higher percentage of respondents in Nepal than Jamaica said that their consumption was less than adequate. For Nepal, the percentages range from 42 to 59, while in Jamaica they range from 20 to 42, with schooling the lowest and housing the highest in both countries (though other categories are ranked differently)."1 Relatively few respondents in either country deemed their consumptions "more than adequate" in either country. Nonetheless, we considered it preferable to keep the information in his category, and use an ordered probit estimator. Table 3 gives sumnmary statistics on the variables we will use in attempting to explain the differences in self-rated consumption adequacy. 'o For further information on these surveys see Grosh and Glewwe (1995). Deaton and Zaidi (1998) provide further details on the construction of the consumption measure.. " In Nepal, the survey also asked about the adequacy of "income"; 69% said their income was less than adequate, appreciably higher than for any consumption components. We will not use these answers, however, since it is implausible that respondents will have similar ideas about what "income" means; no doubt, many were answering about their cash income only. 10 In deriving subjective poverty lines from these data we consider three methods, each motivated by the model described in the previous section, but dealing in different ways with unobserved variables. Method 1 anchors the subjective poverty lines to the perceived adequacy of food consumption alone. We ignore the answers given to the other questions in Table 1. This method is of interest because it corresponds closely to a widely used practice in constructing objective poverty lines in which the poverty line is a level of total consumption or income at which food spending is sufficient to assure that food consumption is deemed nutritionally adequate by pre- determined "objective" criteria of requirements for good health and normal activity levels (for a discussion of this method see Ravallion, 1998). The difference here is that we abandon nutritional requirements in favor of the information contained in the subjective qualitative assessments of food adequacy."2 Method 2 uses the answers on perceived adequacy of other non-food consumptions, as described in section 2. We did not use health care and schooling because these are to a large extent public goods for which the perception of adequacy is not necessarily related to private consumption. (We will be analyzing these data in future work.) All consumption which does not fall under the headings in Table 1 was lumped into a remainder, which we deal with in Method 2 12 Blaylock and Salmwood (1986) also use a food adequacy question in deriving poverty measures, though their approach is quite different to our Method 1. Blaylock and Salmwood use an ordered probit model of survey responses on food adequacy to predict the probabilities of inadequate food consumption at given poverty lines, which are chosen to correspond to predetermined food shares (by inverting an estimated Engel Curve for food, at the given food share). So in their method, the food share defines the underlying reference welfare level to which the poverty line is anchored. Our Method 1, by contrast, derives a poverty line in the consumption space which assures food adequacy in expectation. This is the more natural analogue of the idea of the "subjective poverty line", as discussed in the Introduction. 11 by estimating a reduced form Engel curve for this component as a function of all other spending and the demographic and regional variables. The Engel curve is thus used to make an allowance for the remaining components of spending which is an estimate of the expected value for someone consuming the subjective poverty line levels of the other components. Method 3 is the samne as Method 2, except that we do not use the Engel curve allowance for the remaining consumption. Instead, we simply exclude the remaining consumption from both the poverty lines and from the welfare indicator. As regressors we use log actual household consumption (in total for Method 1, and by component for Methods 2 and 3), log household size, demographic composition variables, log mean consumption in the primary sampling unit, and regional dummy variables. A practical problem arose in the case of transport spending in Jamaica and clothing in Nepal, namely that the relatively large number of zero entries in the data created a very weak relation between actual consumption and perceived adequacy. In the case of clothing in Nepal the underlying reason may be that clothing is a durable, bought only infrequently because of the large travel distances to markets (especially in the hills and mountains). In Jamaica, the transport question was phrased as perceived adequacy of access to transport which could be sufficient even for those who do not use it. The result was a considerable instability in the poverty lines, whereby the allowances for these components could fluctuate wildly according to other household variables. We decided not to include these comtponents in the subjective poverty line, although they are included in the consumption remainder imder Method 2. Table 4 gives the ordered probit estimates of the parameters of the model for food adequacy as a function of total consumption spending, log household size, demographic 12 variables, the (log) mean consumption of the primary sampling unit, and regional dummy variables. For implementing Methods 2 and 3, Table 5 gives the results for the perceived adequacy of food, housing and clothing in Jamaica, while Table 6 gives the corresponding results for food and housing in Nepal. Notice that in these regressions we separately identify the corresponding consumption components. For Method 2 we also require the Engel curves for remaining consumption, as given in Table 7. The regressions in Tables 4-7 are self-explanatory and there are few surprises. Actual measures of consumption tend to be highly significant predictors of perceived consumption adequacy. The perceived adequacy of food and housing tends to respond more elastically to actual spending on each component than on other components (Tables 5 and 6). Clothing in Jamaica, however, tends to respond more elastically to actual housing consumption than clothing; the lack of imputations for clothing services may be the reason. Larger households tend to perceive their consumptions as less adequate holding other variables constant. Holding per capita consumption constant, we find no significant economies of in Jamaica but we do for Nepal. From Table 4, the estimated elasticity of the SPL based on food adequacy in Nepal with respect to family size equals 0.47 (=0.37/0.79).'3 The demographic compositional effects tend not to be significant. Regional effects are stronger in Nepal, which is unsurprising given the country's much greater geographic diversity. There is also a strong negative effect of 13 It is widely assumed that poor households in low income, countries do not face significant economies of scale in consumption since the share of their consumption going to "private" goods within the household is high. However, this assumption is questionable, and a quite wide range of elasticity values might be defended in such settings (Lanjouw and Ravallion, 1995). Nonetheless, we do find this size elasticity for Nepal to be surprisingly low. Household size might well be picking up so other factor influencing subjective perceptions of welfare, though what that factor might be is unclear. We hope to investigate this finding further in future work. 13 neighborhood consumption on perceived adequacy in Nepal, but not in Jamaica. The implied elasticity of the SPL for Nepal with respect to mean consumption of the cluster is 0.29 (=0.23/0.79). The region-specific SPLs are given in Table 8 for both countries and each method. We give the poverty lines at mean points of other variables. However, the calculation of poverty measures (to follow) naturally uses household-specific poverty lines rather than the averages in Table 8. The last column gives previously established "objective" poverty lines for both countries, which will be discussed later. Method 2 requires the more prior estimation than either of the other methods; it requires both the ordered probits by category of consumption and the Engel curve for the remainder. It is to be expected that this creates imprecision in the resulting estimates. (Most methods of calculating poverty lines require prior estimations, although we have not seen prior attempts to calculate standard errors.'4) How much so can be seen from Table 9, which gives standard errors for the SPLs in Table 8, calculated by the Delta method. Standard errors increase substantially as one moves from Method 1 to Method 3, and are highest for Method 2. The aggregate poverty measures are given in Table 10; we give the popular headcount index as well as the poverty gap index and the squared poverty gap index (introduced by Foster et al., 1984) which penalizes inequality among the poor.'5 The three methods are in close 14 This is sometimes done for the poverty measures, though treating the poverty line as non- stochastic; see Kakwani (1993). 15 Notice that, when comparing method 3 with the other two, the poverty measures do not necessarily follow the same raking as the poverty lines from Table 8. This is because the poverty lines are being applied to a differenet consumption aggregate under Method 3, in that the consumption remainder is excluded. 14 agreement, with headcount indices for Jamaica of 32-34% and 44-46% in Nepal. The proportionate divergence between the three methods is somewhat greater for the squared poverty gap. As an aside, it may be noted that the headcount index for Method 1 in Table 10 is not the same as the percentage of people who say that their food is inadequate, as given in Table 2. This is in keeping with the SPL approach, which (as noted in the introduction) identifies the poor as those for whom total income or spending is less than the level which, on average, is deemed to be adequate "to make ends meet". Given latent heterogeneity and measurement error there will be people above this point who still feel that their level of living is inadequate, and people below this point who feel that it is adequate. It is striking how close these aggregate poverty rates are to the results obtained by two independent studies of poverty in these countries which have been based on objective poverty lines. The Planning Institute of Jamaica (the statistics office of the government of Jamaica) estimated the incidence of poverty at 31.5% (Social Policy Development Unit, 1994). As part of the World Bank's Poverty Assessment for Nepal, Bank staff estimated the headcount index in Nepal to be 42% (Lanjouw, Prennushi and Zaidi, 1996). Both estimates are based on the same survey but use per capita poverty lines based on a food basket yielding minimum nutritional requirements (2245 calories per person per day for Jamaica and 2124 for Nepal).'6 The resulting poverty lines are given in the last column of Table 8. 16 The Nepal Study employed the same measure of consumption as this study. The Jamaican Planning Institute constructed their own consumption measure based on the same survey which was not available to the authors. The results quoted in table 8, 10 and 11 are directly taken from (Social Policy Development Unit, 1994). Figure 2 is based on the authors' calculations. 15 The regional poverty profiles vary more depending on the method used. Regional poverty profiles can be found in Table 11 for Jamaica and T'able 12 for Nepal. The strongest differences are between Methods 1 and 2. This was to be expected since Method 1 only controls for differences in food adequacy by region. In Nepal for instance, housing conditions - holding everything else constant - are perceived to be less adequate in the westem hills than in the eastern hills while for food adequacy the opposite holds. As a result method 2 yields a higher headcount index than method 1 for the western hills while the opposite holds for the eastern hills. The urban versus rural poverty comparisons are of special interest in a developing country setting. Poverty comparisons between the two "sectors" have often been controversial, with different measurement methods giving very different results, including rank reversals (Ravallion and Baden, 1994). It has been argued that by ignoring relative welfare considerations, conventional approaches based on (objective) absolute poverty lines (which attempt to fix the real value of the poverty line between the two sectors) will tend to underestimate poverty in urban areas versus rural areas. The previous estimates we have quoted for both countries follow the conventional approach, and so they could also be criticized from this point of view. Yet, our subjective poverty lines tend to show even larger differences between rural and urban poverty measures than do the more conventional methods. Our resualts do not suggest that the conventional approach has underestimated urban versus rural poverty when compared to subjective poverty lines incorporating relative welfare effects, consistently with welfare perceptions. Next we examine differences in the demographic poverty profile. Standard methods of setting poverty lines typically find that larger households are poorer in developing countries 16 (Lipton and Ravallion, 1995). The relationship between poverty and household size is known, however, to be sensitive to measurement assumptions even within the class of standard "objective" methods (Lanjouw and Ravallion, 1995). The previous objective poverty lines for both Jamaica and Nepal followed the common practice in developing countries of having a constant per capita value, i.e., without any allowance for economies of scale in household consumption. In Table 13 we give our subjective poverty lines for various demographic groups, and each of the three methods described above. The SPL is found to increase less than proportionately with household size, with somewhat stronger economies of scale indicated for Nepal than Jamaica. For example, the poverty line for a family of four is (depending on the method) 2.3 to 2.4 times that for a single adult in Nepal, versus 3.1 to 3.9 in Jamaica. Given the sizable scale economy in the Nepal SPL it should not be surprising that this greatly changes the demographic poverty profile when compared to poverty lines which do not incorporate scale economies. That is confnrmed in Figure 2, which compares the poverty rates by household size implied by the previous objective poverty lines (Table 8) with those based on our SPL.'7 The per capita "objective" poverty line suggests that larger households tend to be poorer in both countries. This is also the case for the Jamaican poverty measures based on subjective poverty lines. However, for Nepal the poverty measures based on our SPL tend to fall as household size increases, though not monotonically. The objective poverty lines indicate that '7 Using Method 2; this choice made little difference. The relationship with household size was also similar for the poverty gap and squared poverty gap. The Social Policy Development Unit (1994) does not quote poverty measures by household size for Jamaica. We have calculated our own measures for Figure 2. However, since the precise definition of their consumption aggregate is not given in Social Policy Development Unit (1994), our consumption aggregate gives a slightly different (higher) aggregate poverty measure. Full details are available from the authors. 17 single person families are the least poor, while the subjective poverty lines for Nepal indicate that they are the poorest. As an aside on methodology, we also estimated our models using a probit estimator combining the "more than adequate" responses with the 'lust adequate" ones. The results were similar. For example, for Jamaica, the headcount varies from 34% for Method 3 to 35% for Method 2. For Nepal the headcount varies from 45% for Method 1 to 48% for Method 3. This was to be expected given that very few households reported their consumption to be more than adequate and the fact that the derivation of the poverty line is based on the first threshold only. These results suggest that in future surveys which are augmented to include a module on subjective welfare perceptions, it would be sufficient to include a simple yes/no answer on the question whether consumption is adequate for the household. 4 Conclusions Methods of poverty analysis have differed radically between the "objective-quantitative" and "subjective-qualitative" schools, with little effort to learn from both. We have suggested a hybrid approach, building on past methods of subjective-welfare measurement, but adapted to a developing country setting. It is difficult to believe one could get sensible answers to the minimum income question in most developing countries. For this and other reasons (including priors that almost everyone will think they are poor in a poor country) the subjective poverty line approach found in some of the developed-country literature has attracted little interest in developing countries. The method we have proposed allows one to retrieve the SPL from simple qualitative questions on perceived 18 consumption adequacy added to an integrated household survey of the type favored in objective- quantitative welfare measurement. We have implemented the approach using surveys for Jamaica and Nepal. The results seem encouraging. The reasonably close correspondence we have found between various methods suggests that even a single question on the perceived adequacy of food consumption will give poverty measures which accord closely with subjective poverty lines based on a fuller set of consumption components. The aggregate poverty measures obtained accord quite closely with more conventional "objective" methods. However, more notable differences emerge in the geographic and demographic poverty profiles. The poverty measures by region are more sensitive than are the aggregates to the choice of method, though there is still considerable agreement on rankings. Interestingly, our subjective qualitative approach, incorporating effects of relative deprivation, does not tend to narrow the differentials in poverty measures between "poor" and "rich" areas. For example, our results suggest a larger difference in poverty measures between urban and rural areas than found by more conventional objective approaches based on a concept of basic and absolute consumption needs. People in poor areas perceive themselves to be even poorer than objective comparisons suggest. So our results do not suggest the SPL behaves more like a "relative poverty line" (which rises with average income) than an "absolute poverty line" (which does not). Other differences emerge in the demographic poverty profile. Our SPL indicate sizable scale economies in consumption, particularly for Nepal. Indeed, the scale economy in our SPL for Nepal is strong enough to reverse the tendency for larger households to appear to be poorer when this assessed by commonly used "objective" methods. 19 References Blaylock, James R. and David M. Salmwood, "An Alternative Approach to Defining and Assessing Poverty Thresholds", Western Journal of Agricultural Economics 11 (1986): 100-105 Browning, Martin, "Children and household economic behavior", Journal of Economic Literature 30 (1992): 1434-1475. Colastanto, D., Arie Kapteyn and Jacques van der Gaag, "Two subjective definitions of poverty: Results from the Wisconsin Basic Needs Study", Journal of Human Resources 19 (1984): 127-138. Danziger, S., J. van der Gaag, E. Smolensky and M. Taussig, "The direct measurement of welfare levels: How much does it take to make ends meet", Review of Economics and Statistics 66 (1984): 500-505. Deaton, Angus, and Muellbauer John, Economics and Consumer Behavior (Cambridge: Cambridge University Press, 1980). Deaton, Angus and Salman Zaidi, "How to construct a consumption aggregate from household survey data", Development Research Group, World Bank, 1998. De Vos, Klaas and Thesia I. Garner, "An Evaluation of Subjective Poverty Definitions: Comparing Results from the U.S. and the Netherlands", Review of Income and Wealth 37 (1991): 267-285. Easterlin, Richard A., "Will raising the incomes of all increase the happiness of all?", Journal of Economic Behavior and Organization 27 (1995): 35-47. Foster, J., Greer, J., and Thorbecke E., "A class of decomposable poverty measures ", 20 Econometrica 52 (1984): 761-765. Groedhart, Theo, Victor Halberstadt, Arie Kaptein and Bernard van Praag, "The poverty line: Concept and Measurement", Journal of Human Resources 12 (1977): 503-520. Grosh, Margaret E., and Paul Glewwe, "A guide to living standards measurement study surveys and their data sets", Living Standards Measurement Study Working Paper 120, World Bank, Washington DC.. 1995. Hagenaars, Aldi J.M., The Perception of Poverty, (North Holland, Amsterdam 1986) Hagenaars, Aldi J.M., and Bernard M.S. van Praag, "A synthesis of poverty line definitions", Review ofIncome and Wealth, 1985: 139-154. Kakwani, Nanak, "Statistical Inference in the Measurement of Poverty", Review of Economics and Statistics 75 (1993): 632-63 9. Kapteyn, Arie., "The measurement of household cost functions. Revealed preference versus subjective measures", Journal of Population Economics 7 (1994): 333-350. Kapteyn, Arie., Sara van de Geer, and Huib van de Stadt, "The impact of changes in income and family composition on subjective measures of well-being", in Martin David and Timothy Smeeding (eds) Horizontal Equity, Uncertainty, and Economic Well-Being, Chicago, University of Chicago Press, 1985. Kapteyn, Arie, Peter Kooreman, and Rob Willemse, "Some methodological issues in the implementation of subjective poverty definitions", The Journal of Human Resources 23 (1988): 222-242. Lanjouw, Peter, Giovanna Prennuhi and Salman Zaidi, "Building Blocks for a Consumption- Based Analysis of Poverty in Nepal", mimeo, World Bank, Washington DC., 1996. 21 Lanjouw, Peter, and Martin Ravallion, "Poverty and Household Size", Economic Journal 105 (1995): 1415-1434. Lipton, Michael and Martin Ravallion, "Poverty and Policy", in Jere Behrman and T.N. Srinivasan (eds) Handbook of Development Economics Volume 3 Amsterdam: North- Holland, 1995. Pollak, Robert, "Welfare Comparisons and Situation Comparisons", Journal of Econometrics 50 (1991): 31-48. Pollak, Robert and T.J. Wales, "Welfare Comparisons and Equivalence Scales". American Economic Review 69 (1979): 216-221. Ravallion, Martin, Poverty Comparisons Fundamentals of Puire and Applied Economics Volume 56, Harwood Academic Press, Chur, Switzerland, 1994. ., "Poverty Lines in Theory and Practice", LSMS Working Paper 133, World Bank, Washington DC., 1998. Ravallion, M. and Bidani B., "How Robust is a Poverty Profile?", World Bank Economic Review, 8 (1994): 75-102. Runciman, W.G., Relative Deprivation and Social Justice. London: Routledge and Kegan Paul, 1966. Scitovsky, Tibor, The Joyless Economy, Oxford: Oxford University Press, 1978. Social Policy Development Unit, "Estimates of Poverty in Jamaica: for the years 1992 and 1993", Planning Institute of Jamaica, Kingston, Jamaica, 1994. Stanovnik, Tine, "Perception of poverty and income satisfaction. An empirical analysis of Slovene households", Journal of Economic Psychology 13 (1992): 57-69. 22 van de Stadt, Huib, Arie Kapteyn, and Sara van de Geer, "The relativity of utility: Evidence from panel data", Review of Economics and Statistics 67 (1985): 179-187. van Praag, Bernard, "Ordinal and Cardinal Utility: An Integration of the Two Dimensions of the Welfare Concept", Journal of Econometrics 50 (1991): 69-89. 23 Figure 1: The subjective poverty line (z*) Subjective minimum income I ~ ~ ~ /~~~~~~~ 450 Z * Actual income Figure 2: Poverty and Household Size Jamaica 70 - 60 H - ~ 50-/ '40+ r 30- = 20 - _ Subjective - - Objective 0 j j I I i 1 2 3 4 5 6 7 8 9 10+ Household size Nepal 60 30- no 000_ 201 / 10 --- Subjective -/ - Objective 0 1 2 3 4 5 6 7 8 9 10+ Household size 25 Table 1: Questions on consumption adequacy I would like to ask your opinion of your family's It was less than adequate for your family's needs .....1 standard of living It was just adequate for your family's needs .............2 It was more than adequate for your family's needs ...3 Not applicable ...................................4 "Adequate" means no more nor less than what the respondent considers to be the minimum consumption needs of the family Concerning your family's food consumption over the past one month, which of the following is true? Concerning your family's housing, which of the following is true? Concerning your family's clothing, which of the following is true? Concerning the health care your family gets, which of the following is true? Concerning your children's schooling, which of the following is true? Table 2: Perceived adequacy of consumption in Jamaica and Nepal Percentages Less than Just More than Not adequate adequate adequate applicable Food Jamaica 39 55 6 0 Nepal 47 51 2 0 Housing Jamaica 42 50 8 0 Nepal 59 41 0 0 Clothing Jamaica 36 57 7 0 Nepal 53 47 0 0 Transport Jamaica 48 47 4 0 Health care Jamaica 41 55 4 0 Nepal 52 48 0 1 Schooling Jamaica 20 35 2 43 Nepal 42 38 0 19 Table 3: Descriptive statistics for explanatory variables used in analysis Jamaica Nepal mean std. dev. mean std. dev. Log food consumption 10.14 0.70 9.94 0.63 log housing consumption 7.98 1.20 7.87 1.37 log clothing consumption 8.58 1.00 log household size 1.11 0.72 1.60 0.53 fraction males aged < 18 0.151 0.185 0.224 0.174 fraction females aged < 18 0.151 0.187 0.205 0.176 fraction males aged [18-60] 0.290 0.316 0.232 0.167 fraction females aged [18-60] 0.244 0.241 0.268 0.157 fraction males aged > 60 0.078 0.214 0.033 0.097 fraction females aged > 60 0.086 0.212 0.038 0.124 log mean consumption of cluster 10.10 0.42 8.97 0.63 Number of observations 1954 3373 Table 4: Adequacy of food as a function of total consumption (t-ratios in parentheses) Jamaica Nepal log total consumption 0.64 0.79 (11.69) (16.01) log household size -0.54 -0.37 (-8.25) (-5.77) fraction males age < 18 -0.13 -0.35 (-0.72) (-2.04) fraction females aged < 18 -0.09 -0.45 (-0.48) (-2.60) fraction females aged [18-60] 0.33 0.11 (2.61) (0.61) fraction males aged > 60 0.12 -0.08 (0.86) (-0.34) fraction females aged > 60 -0.01 0.11 (-0.07) (0.53) log mean consumption of cluster 0.07 -0.23 (0.83) (-3.33) other urban 0.17 -0.40 (2.13) (-3.85) rural Jamaica -0.004 (-0.070 rural west hills Nepal -0.45 (-3.89) rural east hills Nepal -0.58 (-5.71) rural west Terai Nepal 0.003 (0.03) rural east Terai Nepal -0.15 (-1.34) 6.91 5.08 (8.91) (8.44) 8.92 7.58 (11.37) (12.41) McFadden's Pseudo 2 0.09 0.13 Table 5: Perceived consumption adequacy by commodity group in Jamaica Food Housing Clothing log food consumption 0.24 0.04 0.13 (4.04) (0.71) (2.11) log housing consumption 0.23 0.47 0.23 (7.51) (14.29) (7.60) log clothing consumption 0.06 -0.02 0.14 (1.64) (-0.59) (3.83) log household size -0.39 -0.18 -0.29 (-5.58) (-2.64) (-4.19) fraction males age < 18 -0.31 -0.21 -0.57 (-1.63) (-1.09) (-2.99) fraction females aged < 18 -0.19 -0.27 -0.17 (-0.99) (-1.39) (-0.90) fraction females aged [18-601 0.15 0.04 0.09 (1.09) (0.27) (0.64) fraction males aged > 60 -0.03 0.61 0.15 (-0.17) (3.66) (0.92) fraction females aged > 60 -0.26 0.58 0.35 (-1.54) (3.53) (2.12) log mean consumption of cluster 0.16 0.02 0.16 (1.89) (0.28) (1.89) other urban 0.14 0.17 0.10 (1.75) (2.07) (1.24) rural 0.08 0.33 0.08 (1.10) (4.27) (1.03) 5.69 4.05 5.17 (6.80) (4.84) (6.24) 7.73 5.98 7.21 (9.15) (7.09) (8.61) McFadden's Pseudo R2 0.09 0.12 0.08 Table 6: Perceived adequacy of food and housing in Nepal Food Housing log food consumption 0.60 0.22 (10.60) (3.81) log housing consumption 0.32 0.32 (12.57) (12.03) log household size -0.37 -0.19 (-5.660 (-2.72) fraction males age < 18 -0.32 -0.43 (-1.84) (-2.37) fraction females aged < 18 -0.43 -0.36 (-2.44) (-2.00) fraction females aged [18-60] 0.06 -0.01 (0.35) (-0.04) fraction males aged > 60 -0.07 0.18 (-0.29) (0.70) fraction females aged > 60 0.07 0.14 (0.34) (0.62) log mean consumption of cluster -0.23 -0.37 (-3.41) (-5.16) other urban -0.34 -0.10 (-3.26) (-0.90) rural west hills -0.26 -0.75 (-2.19) (-5.99) rural east hills -0.40 -0.50 (-3.62) (-4.30) rural west Terai 0.29 -0.54 (2.14) (-3.76) rural east Terai 0.03 -0.25 (0.25) (-2.06) at 5.40 0.75 (8.67) (1.16) a2 7.92 3.90 (12.55) (5.87) McFadden's Pseudo B? 0.14 0.12 Table 7: Engel curves for remaining consumption Jamaica Nepal constant -1.02 -1.78 (-3.36) (-5.24) log core consumption 1.08 1.09 (36.31) (34.05) log household size 0.08 0.10 (2.11) (2.20) fraction males age < 18 -0.48 -0.11 (-4.41) (-0.87) fraction females aged < 18 -0.34 -0.18 (-3.02) (-1.40) fraction females aged [18-60] -0.15 -0.25 (-.95) (-1.82) fraction males aged > 60 -0.43 -0.20 (-5.13) (-1.05) fraction females aged > 60 -0.53 -0.38 (-6.32) (-2.37) other urban -0.08 -0.20 (-1.74) (-2.64) rural Jamaica -0.32 (-8.35) rural west hills Nepal -0.79 (-11.07) rural east hills Nepal -0.56 (-8.48) rural west Terai Nepal -0.53 (-6.48) rural east Terai Nepal -0.53 (-7.62) R squared 0.57 0.50 Note: Core consumption is food and housing, plus clothing for Jamaica. The dependent variable is total consumption minus core consumption. Table 8: Subjective poverty lines for families with average characteristics - Jamaica and Nepal Method I Method 2 Method 3 Independent, Based on perceived Based on perceived Same as Method 2, previous estimates adequacy of food adequacy of food, but excluding of objective poverty alone housing and (for remaining lines; Cost of basic Jamaica) clothing, consumption needs poverty lines, and using an Engel anchored to pre- curve for remaining determined consumption nutritional requirements Jamaica Kingston 13110 10524 6290 14472 Other Urban 10082 7624 4743 14319 rural 13203 10980 7336 13203 Nepal Kathmandu 4129 5164 3674 6122 otherurban 6790 8851 6552 5197 rural western hills 7256 12821 10657 5065 rural eastern hills 8620 5834 4721 5241 rural western Terai 4112 11896 9435 3964 rural eastern Terai 4973 3655 2963 4404 Note: All poverty lines are per capita. Poverty lines for Method 1,2 and 3 were calculated on the basis of country specific average household characteristics (see Table 3), and normalized by household size. Table 9: Standard errors of the subjective poverty lines Method 1 Method 2 Method 3 Based on perceived Based on perceived Same as Method 2, adequacy of food adequacy of food, but excluding alone housing and (for remaining Jamaica) clothing, consumption and using an Engel curve for remaining consumption Jamaica Kingston 1174 4906 2840 Other Urban 1141 3579 2160 rural 1011 4546 2958 Nepal Kathmandu 498 1494 1034 other urban 643 2544 1838 rural western hills 447 8174 6694 rural eastern hills 528 460 364 rural western Terai 387 11549 8994 rural eastern Terai 311 317 252 Note: Standard errors for the SPLs in Table 8, calculated using the Delta method. Table 10: Aggregate poverty measures Percentages Headcount index Poverty gap index Squared poverty gap index Jamaica Method 1 34.4 11.' 5.3 Method 2 31.5 13.2 7.7 Method 3 31.9 13.5 7.6 Previous estimate 31.5 n.a. n.a. Nepal Method 1 43.6 14.5 6.5 Method 2 43.0 16.7 8.6 Method 3 46.0 17.9 9.3 Previous estimate 42 12.1 5,0 Note: See Table 7 for description of alternative methods; see text for full details. Table 11: Poverty profile by region for Jamaica Method Headcount Poverty gap Squared poverty index index gap index Kingston 1 21.4 6.1 2.8 2 18.1 5.8 3.1 3 16.8 6.2 3.3 Previous estimate 21.8 n.a. n.a. Other urban 1 19.6 5.1 1.9 2 13.2 4.3 2.2 3 12.0 3.9 1.9 Previous estimate 28.9 n.a. n.a. Rural 1 47.8 16.7 8.2 2 46.5 21.1 12.5 3 48.6 21.5 12.3 Previous estimate 38.9 n.a. n.a. Table 12: Poverty profile by region for Nepal Percentages Method Headcount Poverty gap Squared poverty index index gap index Kathmandu 1 0.7 0.2 0.0 2 1.1 0.3 0.1 3 0.9 0.3 0.1 Previous estimate 4 0.4 0.1 Other urban 1 30.5 9.1 3.8 2 39.4 16.2 8.5 3 40.3 15.9 8.3 Previous estimate 34 10.9 4.4 Rural western hills 1 71.1 27.9 13.6 2 84.7 39.7 22.4 3 89.6 43.2 25.1 Previous estimate 57 21.0 9.9 Rural eastern hills 1 66.7 23.5 10.8 2 38.7 10.7 4.1 3 43.1 11.2 4.2 Previous estimate 33 9.1 3.6 Rural western Terai 1 22.6 4.5 1.4 2 62.7 23.0 10.7 3 68.3 24.3 11.2 Previous estimate 46 11.2 3.9 Rural eastern Terai 1 31.5 7.0 2.3 2 12.2 2.1 0.6 3 12.7 2.3 0.7 Previous estimate 39 8.7 2.9 Table 13: Household poverty lines by family composition Method I Method 2 Method 3 Jamaica family poverty index poverty index poverty index size line line line one prime age male 1 16187 1.00 8096 1.00 5065 1.00 one prime age female 1 9626 0.59 5334 0.66 3551 0.70 one prime age male plus one prime 2 22428 1.39 12680 1.57 7888 1.56 age female one prime age male plus one prime 3 36954 2.28 22063 2.73 14032 2.77 age female plus one male child one prime age male plus one prime 3 36138 2.23 22216 2.74 13878 2.74 age female plus one female child one prime age male, one prime age 4 50121 3.10 31599 3.90 19959 3.94 female, one male child, one female child Nepal one prime age male 1 11985 1.00 10129 1.00 8256 1.00 one prime age female 1 10425 0.87 9772 0.96 8307 1.01 one prime age male plus one prime 2 15397 1.28 12878 1.27 10566 1.28 age female one prime age male plus one prime 3 22018 1.84 19449 1.92 15725 1.90 age female plus one male child one prime age male plus one prime 3 22971 1.92 18574 1.83 15094 1.83 age female plus one female child one prime age male, one prime age 4 28268 2.36 23886 2.36 19198 2.33 female, one male child, one female child Note: All poverty lines are at the household level and should be compared with total household consumption. Poverty lines for Method 1,2 and 3 were calculated on the basis of country specific average household characteristics (Table 3). Policy Research Working F'aper Series Contact Title Author Date for paper WPSI 993 Financial Safety Nets and incentive Philip L. Brock October 1998 K. Labrie Structures in Latin America 38256 WPS1994 Estimating Wealth Effects without Deon Filmer October 1998 S. Fallon Expenditure Data - or Tears: Lant Pritchett 38009 with an Application to Educational Enrollments in States of India WPS1995 What Macroeconomic Policies Mansoor Dailami October 1998 B. Nedrow Are"Sound?" Nadeem ul Haque 31585 WPS1996 Namibia's Social Safety Net: Kalinidhi Subbarao October 1998 P. Lizarondo Issues and Options for Reform 87199 WPS1997 On Measuring Literacy Kaushik Basu October 1998 M. Mason James E. Foster 30809 WPS1998 The Structure and Determinants of Sudharshan Canagarajah October 1998 A. Garscadden Inequality and Poverty Reduction Dipak Mazumdar 38400 in Ghana, 1988-92 Xiao Ye WPS1999 Heterogeneity among Mexico's Wendy V. Cunningham Octobesr 1998 T. Gomez Micro-Enterprises: An Application William F. Maloney 32127 of Factor and Cluster Analysis WPS2000 GATT Experience with Safeguards: J. Michael Finger October 1998 L. Tabada Making Economic and Political 36896 Sense of the Possibilities that the GATT Allows to Restrict Imports WPS2001 Measuring the Dynamic Gains from Romain Wacziarg November 1998 S. Crow Trade 30763 WPS2002 Accounting for Toxicity Risks in Susmita Dasguptc November 1998 Y. D'Souza Pollution Control: Does It Matter? Benott Laplante 31449 Craig Meisner WPS2003 Thailand's Corporate Financing Pedro Alba Noveimber 1998 R. Vo and Governance Structures Stijn Claessens 33722 Simeon Djankov WPS2004 What Can Be Expected from African Alexander Yeats November 1998 L. Tabada Regional Trade Arrangements? 36896 Some Empirical Evidence WPS2005 Fiscal Federalism and Anwar Shah November 1998 S. Valle Macroeconomic Governance: 84493 For Better or For Worse? WPS2006 Household Welfare Measurement Jesko Hentschel November 1998 PREM and the Pricing of Basic Services Peter Lanjouw 87736 Policy Research Working Paper Series Contact Title Author Date for paper WPS2007 Regional Integration Arrangements: Dean A. DeRosa November 1998 L. Tabada Static Economic theory, Quantitative 36896 Findings, and Policy Guidelines WPS2008 Volatility and Contagion in a Pedro Alba November 1998 D. Fischer Financially Integrated World: Amar Bhattacharya 38656 Lessons from East Asia's Recent Stijn Claessens Experience Swati Ghosh Leonardo Hernandez WPS2009 Poverty and the Economic Transition: Peter Lanjouw November 1998 P. Sader How Do Changes in Economies Branko Milanovic 33902 of Scale Affect Poverty Rates for Stefano Paternostro Different Households? WPS2010 The Real Impact of Financial Shocks: liker Doma9 November 1998 M. Greaves Evidence from the Republic of Korea Giovanni Ferri 81876