Journal of Economic Literature 2018, 56(2), 657–672 https://doi.org/10.1257/jel.20161454 On Measuring Multidimensional Deprivation† Prasanta K. Pattanaik and Yongsheng Xu* This essay presents a critical review of the recent book by Alkire et al. entitled Multidimensional Poverty Measurement and Analysis, and, in the course of doing so, it also discusses some general issues that come up in this context. We outline the basic structure of the problem of measuring multidimensional deprivation and critically evaluate the methodology adopted by Alkire et al. (2015). In particular, we discuss some problems associated with the methods used by them to identify the deprived and to aggregate individual deprivations so as to derive an index of social deprivation. We examine the interpretation in terms of unfreedoms of individuals, which Alkire et al. put on one of their measures of social deprivation. We also suggest a variant of their methodology for measuring multidimensional deprivation.( JEL C38, E02, I32, Z13) 1. Introduction Measurement and Analysis by Alkire et al. (2015) fills this gap admirably. We believe that the publication of this very lucidly written volume is an important event in the literature on multidimensional deprivation. The book can be roughly divided into three parts. The first part, which consists of chapters 1, 2, and 3, presents a background of the problem of measurement of multidimensional deprivation and an overview of the history of and motivation for research in this area. It also provides a systematic discussion of the various axioms used in the analytical literature on multidimensional deprivation. The second part of the book, consisting of chapters 4, 5, and 6, focuses on the “counting approaches,” especially the methodology developed by Alkire and Foster (2011), which has found fairly wide a­cceptance in recent years. In many ways, this part, which deals with the Alkire–Foster methodology T he analytical, as well as empirical, literature on the measurement of multidimensional deprivation1 has expanded very rapidly over the last two decades or so. What, however, was lacking was a booklength treatment of the subject, providing a systematic account of the literature, a careful examination of the formal assumptions figuring in various contributions and their intuitive content, and an assessment of the major contributions. Multidimensional Poverty * Pattanaik: Department of Economics, University of California. Xu: Department of Economics, Andrew Young School of Policy Studies, Georgia State University. † Go to https://doi.org/10.1257/jel.20161454 to visit the article page and view author disclosure statement(s). 1 We use the terms “poverty” and “poor” in the context of income poverty, while we use the terms “deprivation” and “deprived” when talking about deficiencies in terms of “real” attributes, such as nutrition, health, etc. 657 658 Journal of Economic Literature, Vol. LVI (June 2018) (the AF methodology) and several important conceptual issues relating to the measurements of multidimensional deprivation, constitutes the heart of the volume. This review essay will focus on this part. The third part consists of the last four chapters. These chapters consider how the analytical framework introduced earlier can be applied in various settings and discuss many statistical issues that come up in the course of such application. This review essay is organized as follows. In section 2, we lay down some essential notation and outline the basic structure of the problem of measuring multidimensional deprivation; in this section, we also briefly present Alkire and Foster’s approach to the problem, which occupies a central place in Alkire et al. (2015). In section 3, we discuss certain difficulties with the approach of Alkire and Foster and, in section 4, we consider a variant of the AF methodology that avoids many of the difficulties discussed in section 3. Section 5 discusses the interpretation, which Alkire et al. seek to give to one of their measures of overall social deprivation, in terms of individual unfreedoms. Section 6 deals with three general issues, namely, the exclusion of the non-deprived individuals’ dimensional deprivations from the measure of social deprivation, the treatment of an individual’s achievements over and above the deprivation benchmarks in some dimensions, and interpersonal comparisons of overall deprivations of individuals. We conclude in section 7. 2. Notation, Some Basic Concepts, and an Outline of the AF Methodology There are n ≥ 2individuals in the society, to be indexed by 1, … , n, and m ≥ 2 dimensions or attributes, to be indexed by 1, … , m, in terms of which of an individual’s achievements are to be measured. Let Nand M denote the sets {1, 2, … , n} and {1, 2, … , m}, respectively. In this essay, we shall use the terms “dimensions” and “attributes” interchangeably. Let the set of m dimensions }. For each be denoted by F = { f1 , … , fm j, of valdimension fj ∈ F, there is a set, X ues or levels that dimension f j can take. Some dimensions, such as health status, may be only ordinally measurable so that we can compare different levels of health, such as “excellent health,” “good health,” “poor health,” but no meaning can be attached to statements like “the difference between being in good health and being in poor health is greater than the difference between being in excellent health and being in good health.” When an attribute f j is ordinally measurable, X j can be a discrete set containing several real numbers. Sometimes, some dimensions such as schooling may be cardinally measurable so that comparing the difference between, say, nine-year schooling and twelve-year schooling with the difference between ten-year schooling and fourteen-year schooling makes sense. When a dimension f j is cardinally measurable, we assume that X jis a nondegenerate real interval. Conventionally, for any dimension fj ∈ F, a higher real number in X jindicates a higher level of achievement in dimension fj . For each individual i ∈ N, let x i = (x i1, … , x im) be i’s achievement vector in which x ij is i’s achievement in dimension fj . The achievements of all individuals in the society can thus be summarized by an n × m , where the i th row represents matrix, (x ij)n×m individual i’s achievement vector x i and the jth column represents the achievements in dimension fj by all the individuals. For each j ∈ M, let x jbe the cutoff level ¯ for dimension f jsuch that, for any individual i ∈ N, iis considered to be deprived in dimension f j if and only if i’s achievement, x ij, in the dimension fj , falls below x j. ¯ More formally, we define the “deprivation Pattanaik and Xu: On Measuring Multidimensional Deprivation s­ tatus,” to be denoted by d ij0 , of an individual i ∈ Nalong a dimension f j ∈ F below: 1 if x ij < x j ¯ . d ij0 = {0 if x ij ≥ x j ¯ When a dimension f j is cardinally measurable, for each individual i ∈ N, the normalized deprivation gap of iin dimension f j, to be denoted by d ij, is defined as follows: x j − x ij ______ if x ij < x j ¯ x j ¯ . ¯ di j = {0 if x ij ≥ x j ¯ When all attributes are cardinally measurable, the state of the dimensional deprivations of all individuals in the society can be summarized by a deprivation matrix, , where the ijth entry, c ij, of C is C = (c ij)n×m i’s normalized deprivation gap in dimension f j. In this case, the class, , of all permissible matrices is then the class of all n × m matrices Csuch that each entry of C belongs to the interval [0, 1]. It is not obvious how one should construct a deprivation matrix for the society when attributes are only ordinally measurable. When all attributes are only ordinarily measurable, one possibility is to assume that for each individual iand each attribute fj , the deprivation, c ij, of iin dimension jcan take exactly one of only two values, namely, 0 (i is not deprived in dimension f j) and 1 (iis deprived in dimension f j). In this case, the deprivation of every individual in every dimension coincides with her deprivation status in that dimension, and the class, , of all permissible matrices is then the class of all n × mmatrices Csuch that each entry of Cis either 0 or 1. As we shall see, Alkire et al. take this route when all attributes are only ordinally measurable. The problem of measuring social deprivation is viewed as finding a function P : → [0, 1]with the ­following interpretation: for every deprivation matrix C ∈ , 659 P(C)denotes the level of social deprivation corresponding to Cwith a higher level of P indicating greater social deprivation. Thus, the measure of social deprivation P aggregates each deprivation matrix D ∈ to a real number indicating the level of deprivation in the society. In the literature on measuring multidimensional deprivation, two types of procedures have emerged as possible ways of aggregating deprivation matrices: row-first procedures and column-first procedures. Chapter 3 of Alkire et al. (2015) provides an overview and critical evaluation of the major existing methods in the literature on measuring multidimensional deprivation based on the two procedures. A row-first procedure first aggregates each individual’s deprivation vector to an overall deprivation index for this individual and then aggregates the individuals’ overall deprivations to a social deprivation. A column-first procedure, on the other hand, first aggregates all the individuals’ deprivations in each dimension to a degree of social deprivation in that dimension and then aggregates these degrees of dimensional social deprivations to an index of overall social deprivation. Column-first procedures are very useful and are the only viable methods to compute social deprivation if there are only aggregate data (from different sources) available. A well-known application of such a procedure is the United Nations Development Programme’s (UNDP’s) Human Poverty Index. Row-first procedures have a solid foundation in welfare economics and demand micro-level data in which information on each dimension is available for each individual (so that the data and information about joint distribution are known). It may be remarked that, in general, the two procedures can yield different rankings of social deprivation (and therefore different index numbers) for the society; see Dutta, Pattanaik, and Xu (2003) for further discussion on this aspect of the two procedures. 660 Journal of Economic Literature, Vol. LVI (June 2018) Alkire et al. (2015) focus on row-first procedures in aggregating deprivation matrices. We now outline the approach to multidimensional deprivation adopted in Alkire et al. (2015). Let w = (w 1, … , w m) be the vector of the positive weights attached to w j = 1. the different attributes, with ∑j∈M For each individual iand given the vector, ) , of her dimendi 0 = (di 01 , … , di 0m sional deprivation statuses, her depriva w j di 0j . Let k tion score is defined as ∑ j∈M (1 ≥ k > 0)be the cutoff value for the deprivation score such that a person is deprived overall if and only if her deprivation score is at least as great as the cutoff w j dij 0 ≥ k. The set, N ∗, value k, i.e., ∑j∈M of deprived individuals, is defined below: { } N ∗ = i ∈ N : ∑ w jdij 0 ≥ k . j∈M It may be noted that this approach to the identification problem in a multidimensional deprivation framework is an extension of a historically important measure for multidimensional deprivation—the “counting” method—which has been developed and used in empirical research in two regions, Latin America and Europe, since the 1970s. However, the motivations and applications are different. In Latin America, it was developed in the context of basic-needs approach, and in Europe, it was developed in the context of social exclusion (see chapter 4 of Alkire et al. 2015 for more details). It can be easily checked that 1 (1) if w 1 = w 2 = ⋯ = w m = __ m , then for every k (1 ≥ k > 0), there exists a unique integer t(k) ∈ M, such that i ∈ N ∗if and only if i is deprived in at least t (k) dimensions. Once the set of individuals who are deprived is determined, the AF methodology aggregates the overall deprivations of these individuals to reach an index of social ­deprivation. For any α ≥ 0and any deprivation matrix D ∈ , define the social deprivation, Pα (D), associated with D, as follows: (2) if α > 0, then 1 ∑ w j (dij ) α Pα (D) = __ n ∑ ∗ i∈N j∈M and (3) if α = 0, then P0 (D), the “adjusted head count ratio,” is given as follows: 1 ∑ P 0 (D) = __ w j dij 0 . n ∑ ∗ i∈N j∈M Note that, given kand the weights, 1, w 2, … , w m,if every entry of Dis either 0 w or 1, then, for every α > 0, Pα (D) = P0 (D). 3. An Evaluation of the AF Methodology In evaluating the AF methodology for measuring multidimensional deprivation, it is important to distinguish two alternative frameworks based on different assumptions about the measurability of individual deprivations in terms of the different attributes. Consider the following two alternative assumptions. ASSUMPTION 1 (binarily ordinal measurement of normalized individual deprivation in terms of every attribute): The deprivation of every individual in terms of every attribute is either 0 (the individual is not deprived in terms of the attribute) or 1 (the individual is deprived in terms of the attribute) so that the class of permissible deprivation matrices is the class, 1, of all n × m matrices C such that each element of C is either 0 or 1. ASSUMPTION 2 (cardinal measurement of normalized individual deprivation in terms of each attribute): The normalized deprivation gap of an individual in terms of every Pattanaik and Xu: On Measuring Multidimensional Deprivation a­ ttribute is cardinally measurable along the [0, 1] interval, so that the class of permissible deprivation matrices is the class, 2, of all n × mmatricesCsuch that each element of C is a number in the interval [0, 1]. Clearly, these two assumptions refer to polar cases. One can think of cases where an individual’s deprivation in a particular dimension is ordinally measurable but can have more than two levels (e.g., no deprivation, moderate deprivation, serious deprivation, and extreme deprivation) rather than just two levels, namely, 0 and 1 (see Pattanaik and Xu 2018 for an exploration of the more general case). Also, an individual’s deprivations in terms of different attributes may have different types of measurability. Much of the literature, however, deals with these two extreme cases and we still do not have any satisfactory way of handling more realistic but complex analytical structures where different attributes may be measurable in different ways. The choice between a framework based on assumption 1 and a framework based on assumption 2 will, of course, depend on the type of information about dimensional deprivations, which may be available. 3.1. The AF Method for Identifying the Deprived An important contribution of the AF “counting method” for identifying the deprived is to draw attention to the limitations of the criteria for identifying the deprived in what have been called the “union approach” and the “intersection approach” to the problem of identifying the deprived (see Atkinson 2003). Under the union approach, a necessary and sufficient condition for a person to be deprived overall is that she be deprived in terms of at least one attribute. While the condition is plausible as a necessary condition for a person to be deprived overall, it lacks plausibility as a sufficient condition: in a framework, where we have 661 ten different attributes that are all c­ ardinally measurable, we may not be willing to say that a person is deprived overall if she is not deprived in terms of any attribute other than calorie consumption and her normalized calorie deprivation is only 0.005. Even when we are working with assumption 1, we may not be prepared to say that a person is deprived overall if she is deprived in terms of only one of fifteen different attributes, all of which we consider equally important. Under the intersection approach, a person is deprived overall if and only if she is deprived in terms of all attributes. While the condition may have appeal as a sufficient condition for a person to be deprived overall, it lacks plausibility as a necessary condition to be deprived. Given assumption 2, if there are ten attributes and a person is severely deprived in terms of nine out of ten attributes, all of which are considered to be equally important, but is not deprived in terms of the remaining attribute, one can hardly regard the person as ­non-deprived overall. A similar objection can be raised against the intersection criterion for identifying the deprived, when one is working with assumption 1. The AF method for identifying the deprived provides a generalization of both the union and intersection approaches by permitting the possibility that a person may be deprived overall even when she is not deprived in all dimensions, as well as the possibility that a person may be non-deprived overall even when she is deprived in some dimensions. But the exact way in which this generalization is achieved causes some intuitive problems in the case where the (normalized) deprivations of an individual in terms of the different attributes are cardinally measurable. The root of the problem is this: when the dimensional deprivations happen to be cardinally measurable and the relevant information regarding such cardinal measurement is available to us, our intuition about whether someone is deprived overall often depends 662 Journal of Economic Literature, Vol. LVI (June 2018) on the depths of the ­individual’s d ­ eprivation along different dimensions. The AF method for identifying the deprived, however, completely ignores such information about the depths of dimensional deprivations. Consider an example, where 2 is the class of all permissible matrices and we have five attributes, all of which are cardinally measurable and considered to be equally important, so that the weights attached to the attributes in the AF counting criterion for identifying the deprived are given by w 5 = 1/5. Suppose w 1 = w 2 = ⋯ = the cutoff value kis 3/5 so that a person is deprived if and only if she is deprived in at least three out of five dimensions. Now consider two individuals, 1 and 2, whose normalized deprivation vectors are, respectively, d1 = (ϵ, ϵ, ϵ, 0, 0) with ϵ > 0 and d2 = (0, 0, 0, 1, 1). Then under the AF counting method, given our simplifying assumptions, individual 1 will be considered deprived overall while 2 will not be considered deprived. This will be so no matter how small ϵis so long as ϵ is positive. We find this counterintuitive. It seems to us that, if assumption 2 is satisfied and we have information about the depths of dimensional deprivations, then in assessing whether a person is deprived, not only do we take into account the list of the dimensions in which the person is deprived, but we also consider the extent to which the person is deprived in terms of each attribute in that list. The AF criterion for identifying the deprived when assumption 2 holds is not cognizant of the information that may be available about the depths of an individual’s dimensional deprivations. We now discuss a second feature of the AF method of identifying the deprived in the case where assumption 2 holds. We shall illustrate our point with reference to the class of multidimensional social deprivation > 0, measures Pα , such that, for some α n1 ∑ ∑ wij(dij ) α Pα (D) = __ i∈N ∗j∈M for all D ∈ 2. Recall that, in the above expression, N ∗ is the set of all deprived individuals defined in section 2 and w 1, … , w m are the positive weights attached to the different attributes, w j = 1.Given i ∈ N ∗, it seems with ∑j∈M plausible to interpret ∑ j∈M w j(di j) αas a measure of individual i’s “real deprivation.” Now consider an individual i′ ∈ (N − N ∗). Note that since social deprivation is being visualized as an aggregate measure of the overall deprivations of all deprived individuals in the society ,2 and since the deprivation score of i′ is not large enough for i′to be considered a deprived individual, the overall deprivation of i′is excluded from the measure of social deprivation given above. But that does not mean that i′does not have any dimensional deprivations; nor does it mean that, if one asks the question, “How should one aggregate the dimensional deprivations of i′ ,” then one will be asking a meaningless question. Further, if the overall deprivation of every , is given by ∑j∈M w j(di j) α, then it i ∈ N ∗ is not clear why the overall deprivation of any individual i′who is deprived in terms of some attributes, but who is not deprived in terms of a sufficient number of attributes to be considered to be deprived overall, should not be measured in an analogous fashion. In saying this, we are not suggesting that, if the overall deprivation of every individual i in w j(dij ) α, then it logically N ∗is given by ∑j∈M follows that, for every i′in N − N ∗, such that i′is deprived in terms of at least one attribute, the overall deprivation of i′must be given by ∑j∈M w j(di′j ) α. All that we are ­suggesting is that there is no intuitive reason why the overall deprivation of any such i′must be computed differently from the way in which the overall deprivation of every i in N ∗ is computed, though the dimensional deprivations 2 See section 6.1. Pattanaik and Xu: On Measuring Multidimensional Deprivation of i′may not be sufficiently widespread for her to be regarded as a deprived ­individual. If this intuitive position is accepted, then the AF methodology runs into a problem. Consider again our earlier example, where we have five attributes with equal weights; the cutoff value kis 3/5 (so that, given the assumption of equal weights for all attributes, an individual is deprived if and only if she is deprived in at least three dimensions); and there are two individuals, 1 and 2, such that 1’s deprivation vector is d 1 = (ϵ, ϵ, ϵ, 0, 0), with ϵ > 0 , and 2’s deprivation vector is d2 = (0, 0, 0, 1, 1). Then calculating the two individual’s overall deprivations as suggested above, it can be easily seen that the overall deprivation of 1, who is deprived under the AF rule for identifying the deprived, is given by (3/5) (ϵ) α, while the overall deprivation of 2, who is not deprived under the AF criterion, is given by 2 /5.If ϵ (ϵ > 0) is sufficiently small, then the overall deprivation of 2, who is not deprived overall under the AF criterion, is higher than the overall deprivation of 1, who is deprived overall under that criterion. This would seem to be intuitively odd. At the risk of emphasizing the obvious, we would like to clarify again that, when we use the term “odd” here, we are not referring to any logical contradiction. Clearly, there is nothing logically contradictory if: (i) one agrees to use a criterion, γ , to identify the deprived individuals in a society; (ii) one agrees to the use of a principle, δ, in assessing the overall deprivation of only those individuals who have already been identified as being deprived under criterion γ ; and (iii) it turns out that, had one used the same principle δ to assess the overall deprivation of some individual i′, who is not deprived under criterion γ, then the deprivation of i′ under principle δ would have been greater than the deprivation, under principle δ, of some individual iwhom criterion γ has declared to be deprived. We do, however, believe that the conjunction of (i), (ii), and (iii) indicates an 663 intuitive tension between criterion γ and principle δ. What happens when assumption 1 holds and there are only two levels of an individual’s deprivation in terms of any attribute, namely, 1 (the individual is deprived in terms of the attribute under consideration) and 0 (the individual is not deprived in terms of the attribute)? Given our simplifying assumption that all attributes are given the same weight, in this case of 0–1 measurement, the AF procedure would specify an integer t ∈ {1, 2, … , m}, such that an individual is regarded as being deprived overall if and only if she is deprived in at least t dimensions and, for every D ∈ 1, the society’s overall 1 1 __ deprivation is given by __ N ∗ ∑ j∈M n ∑ i∈ m d ij, ∗ where N is the set of all individuals who are judged to be deprived overall. It is clear that, in this case of 0–1 measurement of dimensional deprivations, there cannot be any counterpart of the first difficulty that we discussed when the attributes are cardinally measurable. Similarly, in the case of 0–1 measurement of an individual’s deprivation in terms of every attribute, there cannot be any counterpart of the second intuitive difficulty discussed above. 3.2. Aggregation of Individual Deprivations in the AF Methodology Under the AF methodology, the measure of social deprivation takes the form of 1 j∈M w j(dij ) αfor some α ≥ 0. P α = __ N ∗∑ n ∑ i∈ It may be noted that, for each individual i w j(di j) α who is identified as deprived, ∑j∈M can be regarded as i’s “real overall deprivation.” Consequently, the AF measure of social deprivation can be regarded as summing up the “real overall deprivations” of deprived individuals and then dividing this sum by the population size n . Function Pα has several attractive properties as discussed in chapter 5 of the book. For example, P α is separable across dimensions and across 664 Journal of Economic Literature, Vol. LVI (June 2018) deprived individuals, making it easy and simple to be implemented empirically. On the other hand, ­separability across dimensions comes at a “cost,” as it has some undesirable implications illustrated below. For brevity and simplicity, we focus on 0–1 deprivation levels, though our discussion can easily be formulated for cardinally measurable deprivation levels in the different dimensions. Also, we assume that: (i) there are exactly five dimensions; (ii) all dimensions have the same weight, namely, 1/5; and (iii) the cutoff value kis 2/5. We believe that most people will agree with Stiglitz, Sen, and Fitoussi’s Commission on the Measurement of Economic Performance and Social Progress (2009) report when it says, “the consequences for quality of life having multiple disadvantages far exceed the sum of their individual effects” (emphasis added). In fact, Alkire et al. also seem to agree with this statement, which they cite (see Alkire et al. 2015, p. 21). Consider, however, the implications of the measure of a deprived individual’s overall “real deprivation” embedded in an AF measure of social deprivation. Given a deprivation vector d i of individual i, let the “real overall deprivation” of ibe denoted by h(di ). Consider the AF measure of multidimensional deprivation P α where α > 0. Implicit in this measure is the measure of a deprived individual’s real overall deprivation given ∑j∈M w j (dij ) α, where w j is the by h (di ) ≡ weight attached to attribute f j. In the case, where we have five attributes with equal weights and k = 2/5 , take five vectors, (1, 1, 0, 0, 0), (1, 1, 1, 0, 0), (1, 1, 0, 1, 0), (1, 1, 0, 0, 1), and (1, 1, 1, 1, 1). Given the separability of hacross dimensional deprivations, it is clear that the addition to i’s real overall deprivation when her deprivation vector changes from (1, 1, 0, 0, 0) to (1, 1, 1, 1, 1) is exactly equal to the sum of: (i) the addition to i’s real overall deprivation when her deprivation vector changes from (1, 1, 0, 0, 0) to (1, 1, 1, 0, 0), (ii) the addition to i ’s real overall deprivation when her deprivation vector changes from (1, 1, 0, 0, 0) to (1, 1, 0, 1, 0), and (iii) the addition to i’s real overall deprivation when her deprivation vector changes from (1, 1, 0, 0, 0) to (1, 1, 0, 0, 1). This goes counter to the intuition underlying Stiglitz, Sen, and Fitoussi’s (2009) statement. 4. A Variant of the AF Methodology It seems to us that a variant of the AF methodology will avoid many of the difficulties that we discussed in section 3. Assume that the class, , of permissible deprivation matrices is either 1 or 2 and, for every k (1 ≥ k > 0)and for every α > 1, define a class of social deprivation measures Q α as follows: (4) for all D ∈ , Qα (D) α ∗∗ ∑ w ( d ) , where = ∑i∈ N j∈M j ij w j w 1 > 0, … , w m > 0 with ∑j∈M = 1are the weights attached to the different attributes and w j d ij ≥ k} is N ∗∗ ≡ { i ∈ N : ∑j∈M the set of deprived individuals. It may be checked that, given α > 1, this simple variant of the AF measures avoids the problems of identification, as well as the problem of aggregation that we discussed in sections 3.1 and 3.2. In fact, Alkire et al. (2015) very briefly mention this measure (when α = 2 and 1is the set of permissible deprivation matrices) in the penultimate paragraph on page 191 of their book.3 the case of ­ cardinally When = 2( 3 See also Dhongde et al. (2016) and Pattanaik and Xu (2018) for an axiomatic analysis and discussion of a broader class of measures that subsume the aggregation rule (4) as a specific example. Pattanaik and Xu: On Measuring Multidimensional Deprivation measurable normalized dimensional deprivations), the rule for identification of the deprived embedded in (4) is a little more complex than its AF counterpart, since it takes into account the depths of dimensional ­deprivations. However, if we are to avoid the ­intuitive problems d ­ iscussed in section 3.1, then there does not seem to be any way of avoiding the necessity of taking into account the depths of dimensional deprivations when information about the depths of dimensional deprivations are available.4 The aggregation rule implicit in (4) is also fairly straightfor w j dij ) α ward. But since the expression ( ∑j∈M is not separable across the attributes, we cannot any longer calculate how much the individuals’ deprivations in a specific dimension contribute to the total overall deprivation of the society, as we can do under the AF methodology. But separability of an individual’s overall deprivation across the different attributes is precisely what contradicts Stiglitz, Sen, and Fitoussi’s (2009) intuition, which we have discussed in section 3.2 and which we find rather compelling. 5. The Adjusted Head-Count Ratio and Individual Freedom In a framework where assumption 1 holds so that every individual’s deprivation in each dimension takes exactly two values, 0 (the individual is not deprived in terms of the attribute) and 1 (the individual is deprived in terms of the attribute), Alkire et al. (2015) have sought to link their adjusted headcount ratio (P 0) to the literature on the measurement of freedom in welfare economics. There are two distinct levels at which these links are established. First, at a purely formal level, they demonstrate that, given 0–1 measurement of dimensional deprivations 4 It may be noted that when = 1, for each given vector w = (w 1, …, w m )and cutoff value k, we have N ∗ = N ∗∗. 665 of individuals, an extension of the axiomatic analysis in Pattanaik and Xu’s (1990) paper on the measurement of freedom yields the results that: there exists t ∈ {1, 2, … , m}, such that a person is deprived if and only if she is deprived in at least tdimensions (this is exactly the criterion the AF methodology uses to identify the deprived when assumption 1 is satisfied and the same weight is attached to each attribute); and for any two deprived persons, iand i′, with deprivation vectors di and di′ , respectively, iis at least as deprived as i′if and only if the number of dimensional deprivations in d i (i.e., the number of 1s in the vector d i) is at least as great as the number of dimensional deprivations in d i′. Second, at an intuitive level, Alkire et al. claim that the numbers of deprivations figuring in the deprivation vectors of individuals give us a certain type of information about these individuals’ “unfreedoms.” The demonstration by Alkire et al. (2015) that a modified version of the set of axioms in Pattanaik and Xu’s (1990) contribution on the measurement of freedom can be used to highlight the formal structure of the AF methodology is ingenious and elegant. This demonstration of similarity between the mathematical structures of the two very different problems provides valuable insights but by itself, it does not establish any direct intuitive connection between the measure of an individual’s overall deprivation in their analysis and the intuitive conception of an individual’s freedom or unfreedom. Alkire et al. (2015) seek to establish this intuitive connection through a different route. We consider their argument below. For the sake of simplicity, in the rest of this section (i.e., section 5), we assume that all the attributes are considered to be of equal importance so that the weights attached to them in calculating the deprivation score of an individual are the same. Alkire et al. (pp. 188–89) write, “A higher value of [the adjusted head-count ratio] 666 Journal of Economic Literature, Vol. LVI (June 2018) represents more unfreedom, and a lower value, less. Given that the set of indicators will be unlikely to represent everything that ­constitutes poverty, if each element is widely valued, and if people who are poor and are deprived in a dimension would value being non-deprived in it, then we anticipate that deprivation among the poor could be interpreted as showing that poor people do not have the capability to achieve the associated functionings.” Suppose we have seven attributes, all of which are considered equally valuable by everybody (so that each attribute is given the weight of 1/7 in the adjusted head-count ratio). Suppose k = 3/7so that an individual is deprived if and only if she is deprived in at least three dimensions (see (1)). Assume that the observed deprivation vector of an individual iis (0, 0, 0, 0, 1, 1, 1). In the context of this specific example, how do we interpret the statement that “deprivation among the poor could be interpreted as showing that poor people do not have the capability to achieve the associated functionings”? We consider some possible interpretations and their plausibility. First, can our observation that (0, 0, 0, 0, 1, 1, 1) is the actual deprivation vector of individual ijustify the inference that i could not possibly have attained a deprivation vector where she would not have been deprived in any of the attributes, f 5, f6 , and f7 ? It is difficult to think of a set of reasonable assumptions under which the observation of the deprivation vector (0, 0, 0, 0, 1, 1, 1) can lead us to conclude that the individual did not have the freedom to achieve a deprivation vector,5 say, (1, 1, 1, 0, 0, 0, 0), where she would have been non-deprived in terms 5 The terminology of “achieving” or “choosing” a deprivation vector is slightly awkward, and it would be more natural to talk about choosing a vector of attribute levels, which implies the deprivation vector under consideration. But, for the sake of brevity, we shall continue to use the terminology of achieving or choosing a particular deprivation vector. of each of the attributes f 5, f6 , and f7 . There is no compelling reason why we would never see the individual choosing to attain the deprivation vector (0, 0, 0, 0, 1, 1, 1), when (1, 1, 1, 0, 0, 0, 0) is available to her even if we are prepared to make the following rather strong assumptions: (i) the set of attributes under consideration in our analysis coincides with the set of all attributes that i values, (ii) the individual’s values attach the same weight to all attributes, and (iii) the individual never chooses to have deprivation in terms of a larger number of attributes if she has the option of being deprived in terms of a smaller number of attributes. Given the above assumptions, and given the observed deprivation vector (0, 0, 0, 0, 1, 1, 1), it is, however, reasonable to infer that, when iachieved the deprivation vector (0, 0, 0, 0, 1, 1, 1), no achievement vector involving deprivation in less than three dimensions was available to her. In that sense, one can say that the observation of the deprivation vector (0, 0, 0, 0, 1, 1, 1) of igives us some specific type of i’s unfreedom, namely, the information that the opportunities available to her did not include the possibility of having deprivation in less than three attributes. The number, 3, then becomes a measure of the extent to which ifalls short of complete freedom from deprivations. What the adjusted head-count ratio does is to sum up these numbers across all deprived individuals identified by the “counting criterion” and to take the average for the entire society. But it is important to note that such inference is crucially dependent on assumptions (i) and (iii), which, as empirical assumptions, are Pattanaik and Xu: On Measuring Multidimensional Deprivation far from being innocuous. First, consider (i). The list of all attributes that an individual belonging to a society values may be very different from the list of a­ttributes, which the social scientist/­philosopher ­studying the overall deprivation of that society6 considers to be reasonable. On the other hand, if the list of attributes figuring in a study is the list that has emerged from democratic deliberation and decisions in the society under consideration (cf. Sen 1999), again there is no reason to expect that the list will coincide with the list of all attributes that some specific individual belonging to the society considers valuable7. It is obvious that if, for whatever reason, (i) does not hold, then very little can be inferred about an individual’s unfreedoms (in the sense explained above) from that individual’s deprivation vector as observed by the social scientist. Assumption (iii) is also suspicious as an empirical assumption. It is worth recalling that the functioning and capability approach to living standards and deprivation, which has provided considerable inspiration for the measurement of multidimensional deprivation in recent decades, often emphasizes that there may be large gaps between what people value and what they prefer or choose (see Sen 1985, 1987). Thus, from what nutritionists tell us, it seems that a significant number of people in many societies suffer from serious health problems (e.g., obesity, high cholesterol, and high blood pressure) 6 In particular, the set of attributes chosen by the social scientist/philosopher is likely to miss out many of the vast number of attributes that an individual may consider to be valuable in her life. 7 The difference between the list of all attributes that the individual values and the list of all relevant attributes that emerges from democratic deliberation and voting is somewhat analogous to the intuitive difference, discussed in welfare economics, between an individual’s value judgment about social states and the ranking of social states emerging from a social decision process (see, among others, Little 1952, Bergson 1954, Arrow 1963, Sen 1970, and Pattanaik 2005). 667 because of lifestyles (e.g., heavy dependence on unhealthy fast food, lack of exercise, etc.), which they can easily avoid but which they adopt despite knowing that those lifestyles are harmful for their health. If it is accepted that, as an empirical assumption, (iii) is of doubtful validity and that often there may not be any tight link between what people value in their lives and what they choose to have, observed deprivation vectors of people can hardly bear the burden of the interpretation in terms of unfreedoms of the type discussed above. Alkire et al. (2015) are aware of problems involved in trying to infer people’s unfreedoms from the observation of their actual deprivation vectors. Referring to cases where people may choose not to achieve the minimum satisfactory level of an attribute even if they could do so and, presumably, having in mind Sen’s (1999) well-known example of a person who chooses to fast though she can afford to eat, Alkire et al. write: . . . suppose that we can anticipate what percentage of people would refrain from such achievements in certain functionings—those who might be fasting to the point of malnutrition, at any given time, for example. . . . Assume that in identifying who is poor, the calibration of poverty cut off kreflects these predictions of voluntary abstinence, as well as anticipated data inaccuracies. . . . Applying such a poverty cut off reduces errors in identification—for example, by permitting people who would voluntarily abstain not to be identified as poor. . . . We are puzzled by the method suggested by Alkire et al.(2015) to handle the problem of “voluntary abstinence” by adjusting (presumably upwards) the deprivation cutoff level k > 0. As we have noted earlier (see (1)), given any given cutoff value kfor the deprivation score, there exists a unique integer t (k) ∈ {1, 2, … , m}such that the set of all i ∈ N, for > kholds, coincides with whom ∑m j=1 w j dij the set of all i ∈ N,such that iis deprived in at least t(k) dimensions. So, instead of defining deprived ­individuals in terms of k , we can 668 Journal of Economic Literature, Vol. LVI (June 2018) equivalently define deprived individuals in terms of t(k). Suppose we have seven attributes and a person is considered deprived if and only if she is deprived in at least three dimensions. Assume that (i) 30 percent of the population have deprivation in at least three dimensions; (ii) 5 percent of the population are actually deprived in exactly three dimensions, though they had the opportunity of being deprived in exactly two dimensions (these individuals being the “voluntary abstainers”); (iii) 7 percent of the population are deprived in exactly three dimensions and did not have the opportunity of having deprivations in fewer than three dimensions; and (iv) 18 percent of the population are deprived in four attributes and did not have the option of being deprived in fewer than four attributes. Given this configuration, the only way of eliminating the “voluntary abstainers” from the group of the deprived is to raise the cutoff number of dimensional deprivations from three to at least four. But, if we do this, not only will the voluntary abstainers be eliminated from the newly specified deprived group, but 7 percent of the population, who have exactly three deprivations and did not have any opportunity to have fewer than three deprivations, will also be thrown out from the set of deprived individuals specified under the new, stricter criterion. Thus, it may not always be possible to adjust upward the deprivation benchmark kso as to throw out the “voluntary abstainers” from the set of deprived individuals without simultaneously throwing out from the set of deprived people a chunk of the population who are deprived overall not because of voluntary abstinence, but because the individuals concerned did not have the option of having fewer deprivations than they actually have in their observed deprivation vectors. 6. Some Broader Issues In this section, we comment on three general issues, namely, the exclusion of the non-deprived individuals’ dimensional deprivations from the measure of the society’s overall deprivation, the relation between an individual’s extra achievements beyond some dimensional deprivation benchmarks, and interpersonal comparisons of overall deprivations of individuals. 6.1. The Dimensional Deprivations of Individuals Who Are Not Deprived Overall The AF approach makes a very useful distinction between a person being deprived and a person having some dimensional deprivations. When we have only one dimension or attribute, say, income, it seems natural to say that a person is deprived if and only if her achievement falls short of the deprivation benchmark in that single dimension, no matter how small that shortfall may be. In contrast, when there are multiple dimensions, a person may have mild deprivations in a few of these dimensions, but we may not still consider such dimensional deprivations, together, to be serious enough to regard the person as a deprived individual or, if we like, as an individual who is deprived overall, just as, in our ordinary language, we may not be prepared to say that a person is a bad person simply because she has a few minor character flaws. In view of this distinction between an individual who has some dimensional deprivations and an individual who is deprived overall, does it necessarily follow that, in measuring multidimensional social deprivation, we should take into account only the dimensional deprivations of people who are considered deprived overall and ignore the dimensional deprivations of other individuals in the society? It seems to us that the answer will depend on one’s motivation. Every dimensional deprivation of an individual damages her quality of life even if the damage caused to her quality of life by her different dimensional deprivations, Pattanaik and Xu: On Measuring Multidimensional Deprivation taken together, may not be large enough to regard her as being deprived overall. Given this, there may be two different types of concerns in seeking to measure overall deprivation at the social level. One may be exclusively ­concerned with the urgent problem of measuring (and relieving) the deprivation of people whom one considers to be deprived overall; in that case, it would be justifiable to ignore the deprivation of people who have some dimensional deprivations, but whose dimensional deprivations are not extensive or deep enough to regard them as being deprived overall. On the other hand, there is no reason why one may not be interested in devising an aggregate measure of the damages that dimensional deprivations cause to the qualities of life of all individuals in a group, irrespective of whether the damage to an individual’s quality of life, caused collectively by her dimensional deprivations, is large enough to regard her as being deprived overall.8 When Alkire et al. (2015) base their measure of social deprivation on the dimensional deprivations of only those individuals in the group who are classified as being deprived overall, they are presumably guided by the former motivation.9 But, even if one accepts that an individual having some dimensional deprivations may not necessarily be deprived, in measuring a group’s deprivation one may still decide to take into account the dimensional deprivations of all individuals in the group if one is guided by the latter concern mentioned above. 8 It is clear that when there is only one dimension, say, income, this distinction between the two motivations becomes irrelevant. 9 The distinction between the two motivations disappears if one takes the position that a person is deprived overall if and only if she is deprived in at least one dimension. 669 6.2. The Role of an Individual’s Achievements above Deprivation Benchmarks for Some Dimensions in the Determination of Her Overall Deprivation Most measures of social deprivation incorporate explicitly or implicitly some ­conception of a person’s overall deprivation. In assessing a person’s overall deprivation, it is a common practice to ignore her achievements, if any, over and above the respective deprivation benchmarks in some dimensions. (For the sake of convenience, we shall call an individual’s extra achievement beyond the deprivation benchmark for a dimension simply her extra achievement in that dimension.) Like most people, including us, who have worked in this area, Alkire et al. (2015) have followed this convention. As far as we can see, there are at least two different ways in which one can justify this convention intuitively. First, one may take the position that the damage to a person’s quality of life, caused by even a very small shortfall from the deprivation benchmark in one dimension cannot be mitigated to any extent by the beneficial contributions to the individual’s quality of life made by extra achievements, however large, in other dimensions. It is not clear how plausible this position is, especially when one is simultaneously prepared to allow trade-offs between deprivations in different dimensions. The second justification may be as follows. A social scientist or philosopher may admit the possibility that the damage to an individual’s quality of life, which is caused by some dimensional deprivations, can be mitigated, at least partially, by the contributions to the individual’s quality of life of extra achievements in other dimensions. But, having admitted this possibility, the social scientist or philosopher may also argue that: for the moment, she (i.e., the social scientist/ philosopher) is interested in the exercise of assessing just the damage to an individual’s quality of life made by the dimensional 670 Journal of Economic Literature, Vol. LVI (June 2018) deprivations that the individual may have; and if anyone is interested in finding out to what extent such damage is compensated or overcompensated by the individual’s extra achievements in other dimensions, that is a matter of another distinct exercise, namely, the exercise of ­assessing the ­individual’s well-being/overall quality of life/living standards. This can be a reasonable defense of the convention of ignoring an individual’s dimensional extra achievements in the assessment of the individual’s overall deprivation if one assumes that the damage caused to an individual’s quality of life by deprivations in certain dimensions is independent of the extra achievements of the individual in other dimensions. But, if one takes this line of defense, then, in the absence of further assumptions, one may have to accept, at least in principle, the possibility that a policy maker may be justified in devoting resources to increase people’s dimensional extra achievements, instead of devoting the same resources to reducing their dimensional deprivations, since that may be the most cost-effective way of increasing their well-being/overall quality of life/living standards. Our discussion in the preceding paragraph brings us to another possible approach to the notion of an individual’s overall deprivation, namely, the approach through the notion of an individual’s well-being defined with reference to the same space of attributes or indicators, which is considered relevant for the measurement of the individual’s overall deprivation. Suppose there is an agreedupon “well-being function” or “quality of life function,” W , to evaluate an individual’s overall well-being or quality of life on the basis of her dimensional achievements, and there is a given well-being threshold, W , below which ¯considered to the individual’s well-being is be “unacceptably low.” Then, one can say that an individual, i,with the achievement (x i) vector x i, is deprived if and only if W < W (see Bourguignon and Chakravarty ¯ and Tsui 2002). The magnitude of i’s 2003 shortfall from the deprivation benchmark can then be taken to be 0 if W(x i) ≥ W and it can be taken to be W(x¯ i)if W(x i) < W . The specifica W – ¯ tion of the function W is ¯ the analytical core of this approach. Once the function W is ­specified and the level of deprivation of each individual is assessed in this fashion, the remaining exercise of aggregating the individual deprivations in terms of well-being to reach an assessment of social deprivation becomes exactly analogous to the aggregation problem in the case of unidimensional deprivation. This approach, discussed by Bourguignon and Chakravarty (2003) and Tsui (2002), has the advantage of bringing together the analytical framework for measuring well-being or living standards and that for measuring deprivation. It may be thought that this approach will push the issue of dimensional deprivations of individuals to the background, but it is not clear why this should be so. Dimensional deprivations of an individual are important because of the damage they do to the individual’s well-being or quality of life. If the well-being function for an individual is defined with respect to the same space of attributes used in the analysis of multidimensional deprivation, then all our concerns relating to dimensional deprivations of an individual would come to the surface when we seek to determine what the individual’s well-being function should be and how it should take into account the individual’s dimensional deprivations. In general, it seems to us that the approach to the problem of multidimensional deprivation through the notion of an individual’s ­well-being function deserves more attention than it has received so far. 6.3. Intergroup Comparisons of Deprivation w m, Consider the weights w 1, w 2, … , which are attached to different dimensional Pattanaik and Xu: On Measuring Multidimensional Deprivation deprivations in the AF methodology (see section 2). These play a crucial role in the AF method for identifying the deprived and their method for aggregating the individual deprivations to assess the magnitude of social deprivation. Even in a given s­ ociety, however, there may be different distinct ­ culturally identifiable groups, which may have different sets of weights for the dimensions; this is true especially in the case of large and very diverse societies, such as in India. How does one determine the set of weights to be used to assess the deprivation status (deprived versus non-deprived) of an individual and the magnitude of the overall deprivation of an individual identified as being deprived? Following Sen (see, for instance, Sen 2009, 241–43), it is often argued that the set of weights to be attached to different dimensions should be determined through “public reasoning,” which takes the form of democratic deliberations and debates. While the importance of democratic deliberations and debates in the evolution of values can hardly be doubted, such deliberations and debates do not necessarily lead to unanimity of values. If after extensive public discussion, there still remain differences between the sets of weights that different cultural groups attach to different dimensions, the question arises of what should be done about the choice of weights to be used to determine whether an individual is deprived and to measure the extent of her overall deprivation if she is judged to be deprived. Should one use the same set of weights, possibly reached through some form of democratic decision procedure, for every individual in the society irrespective of what cultural group she belongs to? If one feels, as we do, somewhat uncomfortable with such “absolutism,” which does not make any allowance for cultural differences of different groups, and if one allows for some difference between the sets of weights to be used for computing the overall deprivations of individuals coming from different 671 cultural groups, then can one do so without running into problems with some very plausible principles of interpersonal comparisons of overall deprivation? In particular, can one allow the set of weights to vary between individuals while adhering to the principle that, if, for every dimension, the ­deprivation of a deprived individual, i , is at least as high as that of another deprived individual, i′, and, if, for some dimension, the deprivation of iis strictly greater than the deprivation of i′, then the overall deprivation of i must be at least as high as that of i′ ? The discussion of the exact counterpart of this issue in the context of the measurement of standard of living when the attributes are cardinally measurable (see, among others, Pattanaik and Xu 2007, 2012 and Fleurbaey 2007) has turned up some rather depressing negative results, and it is not difficult to prove the counterpart of those negative results for the problem of measuring overall deprivation of individuals when the attributes are cardinally measurable. 7. Concluding Remarks In this essay, we have tried to provide a critical evaluation of Alkire et al. (2015) and, in the course of doing so, we have also commented on certain general issues in the literature on the measurement of multidimensional deprivation. While we have some reservations about the AF method for measuring multidimensional deprivation for a society—especially when individual dimensional deprivations are cardinally measurable—and about the interpretation by Alkire et al. (2015) of certain measures of deprivation in terms of freedom/unfreedom, such reservations hardly detract from our admiration for the volume as a comprehensive, meticulous, and authoritative treatment of the problem of measuring multidimensional deprivation. The volume will be cherished by all who are interested in the measurement of 672 Journal of Economic Literature, Vol. LVI (June 2018) multidimensional deprivation, irrespective of whether they are seasoned researchers or beginning students looking for a highly readable but sophisticated account of the issue of measuring multidimensional deprivation, and it should stimulate further future research in this important area. References Alkire, Sabina, and James Foster. 2011. “Counting and Multidimensional Poverty Measurement.” Journal of Public Economics 95 (7–8): 476–87. Alkire, Sabina, James Foster, Suman Seth, Maria Emma Santos, José Manuel Roche, and Paola ­Ballon. 2015. Multidimensional Poverty Measurement and Analysis. Oxford and New York: Oxford University Press. Arrow, Kenneth J. 1963. Social Choice and Individual Value, Second edition. New Haven and London: Yale University Press, 1951. Atkinson, A. B. 2003. “Multidimensional Deprivation: Contrasting Social Welfare and Counting Approaches.” Journal of Economic Inequality 1 (1): 51–65. Bergson, Abram. 1954. “On the Concept of Social Welfare.” Quarterly Journal of Economics 68 (2): 233–52. Bourguignon, François, and Satya R. Chakravarty. 2003. “The Measurement of Multidimensional Poverty.” Journal of Economic Inequality 1 (1): 25–49. Dhongde, Shatakshee, Yi Li, Prasanta K. Pattanaik, and Yongsheng Xu. 2016. “Binary Data, Hierarchy of Attributes, and Multidimensional Deprivation.” Journal of Economic Inequality 14 (4): 363–78. Dutta, Indranil, Prasanta K. Pattanaik, and Yongsheng Xu. 2003. “On Measuring Deprivation and the Standard of Living in a Multidimensional Framework on the Basis of Aggregate Data.” Economica 70 (278): 197–221. Fleurbaey, Marc. 2007. “Social Choice and the Indexing Dilemma.” Social Choice and Welfare 29 (4): 633–48. Little, I. M. D. 1952. “Social Choice and Individual Values.” Journal of Political Economy 60 (5): 422–32. Pattanaik, Prasanta K. 2005. “Little and Bergson on Arrow’s Concept of Social Welfare.” Social Choice and Welfare 25 (2–3): 369–79. Pattanaik, Prasanta K., and Yongsheng Xu. 1990. “On Ranking Opportunity Sets in Terms of Freedom of Choice.” Recherches Économiques de Louvain 56 (3–4) 383–90. Pattanaik, Prasanta K., and Yongsheng Xu. 2007. “Minimal Relativism, Dominance, and Standard of Living Comparisons Based on Functionings.” Oxford Economic Papers 59 (2): 354–74. Pattanaik, Prasanta K., and Yongsheng Xu. 2012. “On Dominance and Context-Dependence in Decisions Involving Multiple Attributes.” Economics and Philosophy 28 (2) 117–32. Pattanaik, Prasanta K., and Yongsheng Xu. 2018. “Measuring Multi-dimensional Well-being and Deprivation with Discrete Ordinal Data.” Mimeograph. Forthcoming in Deprivation, Inequality and Polarization: Essays in Honour of Satya Ranjan ­Chakravarty, Indraneel Dasgupta and Manipushpak Mitra, eds. Singapore: Springer Nature. Sen, Amartya. 1970. Collective Choice and Social Welfare. San Francisco: Holden–Day. Sen, Amartya. 1985. Commodities and Capabilities. Amsterdam: North–Holland. Sen, Amartya. 1987. The Standard of Living. Cambridge and New York: Cambridge University Press. Sen, Amartya. 1999. Development as Freedom. Oxford and New York: Oxford University Press. Sen, Amartya. 2009. The Idea of Justice. Cambridge, MA: Harvard University Press. Stiglitz, Joseph E., Amartya Sen, and ­ Jean-Paul Fitoussi. 2009. “Report by the Commission on the Measurement of Economic Performance and Social Progress.” http://ec.europa. eu/eurostat/documents/118025/118123/ Fitoussi+Commission+report. Tsui, Kai-yuen. 2002. “Multidimensional Poverty Indices.” Social Choice and Welfare 19 (1) 69–93. This article has been cited by: 1. Shatakshee Dhongde, Xiaoyu Dong. 2022. Analyzing Racial and Ethnic Differences in the USA through the Lens of Multidimensional Poverty. Journal of Economics, Race, and Policy 5:4, 252-266. [Crossref] 2. Sabina Alkire, Usha Kanagaratnam, Ricardo Nogales, Nicolai Suppa. 2022. Revising the Global Multidimensional Poverty Index: Empirical Insights and Robustness. Review of Income and Wealth 68:S2. . [Crossref] 3. Francesco Burchi, Daniele Malerba, Claudio E. Montenegro, Nicole Rippin. 2022. Assessing Trends in Multidimensional Poverty During the MDGs. Review of Income and Wealth 68:S2. . [Crossref] 4. Reiko Gotoh, Ryo Kambayashi. 2022. What the Welfare State Left Behind—Securing the Capability to Move for the Vulnerable. Asian Economic Policy Review 35. . [Crossref] 5. Shatakshee Dhongde, Robert Haveman. 2022. Spatial and Temporal Trends in Multidimensional Poverty in the United States over the Last Decade. Social Indicators Research 163:1, 447-472. [Crossref] 6. Benoit Decerf. A Welfarist Theory Unifying Monetary and Non-Monetary Poverty Measurement 2, . [Crossref] 7. Ed Wilson, Reetu Verma, Kankesu Jayanthakumaran. 2022. Can reducing inequality reduce the disutility of the poor?. Applied Economics Letters 62, 1-4. [Crossref] 8. Jean-Marie Baland, Guilhem Cassan, Benoit Decerf. 2021. “Too Young to Die”: Deprivation Measures Combining Poverty and Premature Mortality. American Economic Journal: Applied Economics 13:4, 226-257. [Abstract] [View PDF article] [PDF with links] 9. Adrián Cabrera, Carmelo García-Pérez. 2021. Deprivation levels among people living homeless: a comparative study of Spain and France. Applied Economics 53:35, 4118-4133. [Crossref] 10. Ann Mitchell, Jimena Macció. 2021. Using Multidimensional Poverty Measures in Impact Evaluation: Emergency Housing and the “Declustering” of Disadvantage. Journal of Human Development and Capabilities 22:3, 379-402. [Crossref] 11. Sabina Alkire, Christian Oldiges, Usha Kanagaratnam. 2021. Examining multidimensional poverty reduction in India 2005/6–2015/16: Insights and oversights of the headcount ratio. World Development 142, 105454. [Crossref] 12. Hanna Dudek, Wiesław Szczesny. 2021. Multidimensional material deprivation in Poland: a focus on changes in 2015–2017. Quality & Quantity 55:2, 741-763. [Crossref] 13. Vito Peragine, Maria G. Pittau, Ernesto Savaglio, Stefano Vannucci. 2021. On multidimensional poverty rankings of binary attributes. Journal of Public Economic Theory 23:2, 248-274. [Crossref] 14. Marco J. Haenssgen, Nutcha Charoenboon, Giacomo Zanello. 2021. You’ve got a friend in me: How social networks and mobile phones facilitate healthcare access among marginalised groups in rural Thailand and Lao PDR. World Development 137, 105156. [Crossref] 15. Shatakshee Dhongde. 2020. Multidimensional economic deprivation during the coronavirus pandemic: Early evidence from the United States. PLOS ONE 15:12, e0244130. [Crossref] 16. Renuka Mahadevan, Maneka Jayasinghe. 2020. Examining Multidimensional Poverty in Sri Lanka: Transitioning Through Post War Conflict. Social Indicators Research 149:1, 15-39. [Crossref] 17. Sara Mota Cardoso, Aurora A. C. Teixeira. 2020. The Focus on Poverty in the Most Influential Journals in Economics: A Bibliometric Analysis of the “Blue Ribbon” Journals. Poverty & Public Policy 12:1, 10-42. [Crossref] 18. Roger White. A Survey of the Literature on Multidimensional Poverty 29-59. [Crossref] 19. Iñaki Permanyer. 2019. Measuring poverty in multidimensional contexts. Social Choice and Welfare 53:4, 677-708. [Crossref] 20. Shatakshee Dhongde, Prasanta K. Pattanaik, Yongsheng Xu. 2019. Well‐Being, Deprivation, and the Great Recession in the U.S.: A Study in A Multidimensional Framework. Review of Income and Wealth 65:S1. . [Crossref] 21. Prasanta Pattanaik, Yongsheng Xu. Measuring Multidimensional Well-Being and Deprivation with Discrete Ordinal Data 3-14. [Crossref] 22. Asis Kumar Banerjee. 2018. Multidimensional Indices with Data-driven Dimensional Weights: A Multidimensional Coefficient of Variation. Arthaniti: Journal of Economic Theory and Practice 17:2, 140-156. [Crossref]