Towards a multidimensional poverty index for Germany

Abstract

This paper proposes a more comprehensive multidimensional poverty index for an advanced economy like Germany. Drawing on the capability approach as conceptual framework, I apply the Alkire–Foster method to the German context. Special attention is paid to the conceptual integration. Specifically, I argue for including material deprivation and employment as important dimensions, but against using an additional lack-of-income indicator. The results are consistent with previous findings and also offer new insights. In particular, I find specific poverty profiles (e.g., for the elderly), but also that gaps in poverty between subpopulations change over time. Importantly, the results suggest that genuine multidimensional measures add unique insights, which neither a single indicator, nor a dashboard approach can offer. Finally, the analysis reveals multidimensional and income-poverty measures to disagree on who is poor. The subsequent analysis of this mismatch lends empirical support to abandon a lack-of-income dimension.

1 Introduction

1.1 Background

Poverty in advanced economies is returning to the agenda of both policy makers and academics. Most countries of the European Union, for instance, have adopted national action plans for social inclusion after the Nice summit in 2001 (Atkinson et al. 2002). Alternative approaches to measure human well-being more accurately also receive growing attention (e.g., OECD 2011). Indeed, the main concern of Stiglitz et al. (2009) is to close the gap between what current official statistics on social progress suggest, and what people in fact experience. Likewise, by tracing more than a dozen ‘core indicators’, poverty reports in Germany seek to go beyond what conventional income-poverty measures can capture.

Additionally, significant improvements in the methodology of multidimensional measurements have been made as well (e.g., Tsui 2002; Bourguignon and Chakravarty 2003; Alkire and Foster 2011). So far, these measures have been systematically employed to analyze poverty in the developing world; see in particular Alkire and Santos (2014) and UNDP (2011). However, applying these techniques to advanced economies requires appropriately adapted specifications and operationalizations, such as choosing the relevant dimensions, appropriate indicators, and reasonable cutoffs. Moreover, these choices are also contingent upon the concrete purpose of the poverty measure: Is the task to identify general trends across countries and to assess countries’ relative performance in fighting poverty? Or is there a need to evaluate policy measures and to examine poverty dynamics in one specific country? As these overall objectives crucially affect the response to many of the arising trade-offs during the specification, their explication is imperative. The present paper seeks to complement Germany’s official reports on poverty and wealth (RPW) with a comprehensive summary measure that also takes account of the joint distribution of deprivations. The importance of “the joint distribution” has been emphasized repeatedly (Wolff and de-Shalit 2007; Stiglitz et al. 2009).

1.2 Procedure

A clear conceptual framework is central to any measurement exercise, as it details the construct to be measured. To this end I adopt the capability approach (CA), as essentially developed by Sen (1985, 1992, 1999b). Dimensions are understood as functionings, which in turn constitute human well-being. This inherently multidimensional notion of well-being offers a comprehensive and coherent account of deprivations. Moreover, poverty is understood as capability deprivation, i.e. in addition to low achievements in dimensions, poor individuals also lack the freedom to fare better (Sen 1992, ch. 7). Finally, inevitable value judgments (e.g., choosing dimensions) need to be revealed and related to the relevant public debate (Sen 1999b, ch. 6).

Technically, I apply the dual cutoff counting approach suggested by Alkire and Foster (2011). The Alkire–Foster method (AFM) fulfills several desirable axioms that allow a sensible analysis (e.g., numerous decompositions). Changes in the index, for instance, can be traced back to changes in the indicators. Further advantages follow from having a second, the so-called poverty cutoff. First, the adjusted headcount ratio of Alkire and Foster (2011) can thereby be related to capability poverty, through being a measure of unfreedom (Alkire et al. 2015, ch. 6.1). This is relevant for the present case, since the latest RPW indeed notes difficulties in measuring capability deprivations (Bundesregierung 2013, pp. 23–24 ). Second, the poverty cutoff also allows to handle a larger number of deprivation indicators, where union and intersection approach produce impracticable results. This is important for the present case as well, since the German RPW already suggest 17 “core indicators”. Third, the poverty cutoff allows to distinguish deprivation from poverty, where individuals suffer from multiple deprivation. This aspect has not only normative force, but also offers novel insights on dynamics, since the poor are then a specific subset of the deprived (Suppa 2017). Importantly, as an “open-source technology”, the AFM reveals rather than buries the value judgments and thereby allows for a constructive exchange with the public debate.¹ The empirical analysis uses data of the German Socio-Economic Panel (SOEP). The poverty measure is calculated for three points in time, which allows a particular comprehensive specification.

1.3 Previous research

Previous attempts applying the AFM to advanced economies include Whelan et al. (2014) and Alkire and Apablaza (2016). Both studies focus on cross-country comparisons using EU-SILC data where, however, most indicators are located in resource space. Moreover, Alkire and Apablaza (2016, p. 6) note that their measure is mostly illustrative, for reasons of data availability and comparability. Using German SOEP data, Busch and Peichl (2010) apply the AFM (among other methods), whereas Nowak and Scheicher (2016) implement a modified version of the AFM. Both studies are, however, only weakly integrated into a conceptual framework. Also using SOEP data, Rippin (2016) employs a correlation-sensitive poverty index, which also reflects inequality among the poor. Alkire and Foster (2016), however, argue that no measure can be both sensitive to inequality (understood as dimensional transfer) and simultaneously satisfy dimensional breakdown and subgroup decomposability. An important intricacy with the SOEP data is that not all questions are asked in every year in order to reduce the burden for the respondents. Hence a more frequent calculation of the poverty measure necessitates a narrower specification. While Busch and Peichl (2010) and Nowak and Scheicher (2016) base most of their analyses on annual data, Rippin (2016) opts for biannual calculation.² Moreover, virtually all of the previous studies include income as a dimension, although it is conceptually and empirically unclear whether such an approach is justified. Thus, despite some attempts in this direction, more comprehensive and well-justified multidimensional poverty indexes for advanced economies are still lacking.

1.4 Contribution

The present paper complements the previous literature in two directions. Conceptually, I propose how to operationalize a multidimensional poverty index for advanced economies using the example of Germany. Focussing on a single country allows me to use the most comprehensive data available for that country. More specifically, I argue for including material deprivation and employment as important dimensions, as they contribute extra information on otherwise ignored functionings. In addition to education, health, and housing, I also propose an operationalization of social participation. However, I abandon a lack-of-income dimension for both conceptual and empirical grounds. Specifically, I show that income-poverty is largely captured by material deprivation indicators. Exploiting comprehensive wealth information, I also demonstrate that income inaccurately reflects material well-being for a significant share of the income-poor. The role of income in multidimensional poverty measures has not been systematically addressed so far.

Empirically, several results are shown to be consistent with earlier findings (e.g., migrants suffer more poverty), but also new insights ensue, such as a specific profile for old-age poverty. Over the decade under consideration, multidimensional poverty exhibits a hump shape and both employment and material deprivation indicators are shown to figure prominently in this development. Moreover, I find gaps in poverty between subgroups to change over time. East-Germany, for instance, registers first the higher increase but then experiences the smaller decrease in multidimensional poverty. In contrast, both gender and migrant gap in poverty are shrinking. Importantly, the results also suggest that genuine multidimensional measures may add unique insights, which neither a single indicator, nor a dashboard approach can offer. Both approaches ignore information of the joint distribution of deprivations. Finally, I also document multidimensional- and income-poverty measures to substantially disagree on who is poor. This contrast in targeting gives reason to expect different policy implications.

1.5 Outline

Section 2 briefly introduces the underlying methods; Sect. 3 presents both data and specification. Section 4 contains some empirical results, whereas Sect. 5 scrutinizes the role of income in multidimensional poverty. Finally, Sect. 6 offers some concluding remarks.

2 Methodology

The Alkire–Foster method offers numerous benefits for the evaluation of both poverty-relevant developments and policy measures. The exposition here is restricted to those aspects used in the subsequent empirical analysis. Further aspects are found in Alkire and Foster (2011), Alkire et al. (2015).

2.1 Identification

The matrix y contains the available data, is of size $N\times D$, and describes for each individual the achievement in each dimension deemed relevant. Specifically, $y_{id}\ge 0$ represents the achievement of individual $i=1,\ldots ,N$ in dimension $d=1,\ldots ,D$. The row vector z, with $z_d>0$, describes the deprivation cutoffs, i.e., the achievements necessary for not being considered as deprived in the respective dimension. Using this information, we obtain the deprivation vector c by counting individual deprivations, i.e., the column vector’s elements are $c_i=\sum _{d=1}^D {\mathbb{1}}(y_{id}<z_d)$. Following Bourguignon and Chakravarty (2003), the discrimination between poor and non-poor individuals depends critically on dimensional achievements and the respective cutoffs. Thus identification can be described by a function $\rho (y_{i},z)$ . Several approaches have been suggested so far. While the union approach is characterized by $\rho (y_i,z)={\mathbb{1}}(c_i \ge 1)$, the intersection approach requires $c_i = D$. The key idea of Alkire and Foster (2011) is to define $\rho _k(y_i,z)={\mathbb{1}}(c_i\ge k)$ for $k \in \left[ 1,D\right]$. Since $\rho _k$ depends on both the dimension-specific cutoffs $z_j$ and the overall cutoff k, it is called the dual cutoff approach. The union and intersection approaches are included as special cases ($k=1$ and $k=D$).

2.2 Aggregation

A simple form of aggregation is the calculation of the headcount ratio, which is defined as $H=q/N$, where $q=\sum _{i=1}^N {\mathbb{1}}(c_i>k)$ is the number of the poor. Additionally, to take account of the breadth of poverty we first censor the counting vector of deprivations for non-poor and define c(k) with elements $c_i(k)={\mathbb{1}}(c_i \ge k)c_i$ for all $i=1,\ldots ,N$. As $c_i(k)/D$ is the share of all possible deprivation suffered by i, $A=\sum ^N_{i=1}{c_i(k)}/(qD)$ represents the average deprivation suffered by the poor, which is also called the intensity. Alkire and Foster (2011) then define the adjusted headcount ratio as $M_0=\frac{1}{N}\sum _{i=1}^N c_i = HA$, which is sensitive to both changes in incidence and breadth of poverty. In principle other members of the FGT class of measures (see Foster et al. 1984) can be applied as well—their discussion is however beyond the scope of this paper.

2.3 Weights

So far we have assumed equal weights for all dimensions. To allow for different weights, we introduce a weighting vector w with $\sum _{d=1}^D w_d = 1$. Then the weighted deprivation count becomes $c_i = \sum ^D_{D=1} w_d {\mathbb{1}}(y_{id}\le z_d)$, and $M_0=\frac{D}{N} \sum _{i=1}^N c_i(k)$.

2.4 Decompositions

The adjusted headcount $M_0$ and both its single components and its changes over time have been shown to be decomposable in numerous ways. For instance, the subgroup decomposition of $M_0$ means that the overall adjusted headcount ratio can be written as a population-weighted average of the subpopulation-specific adjusted headcount ratios. Formally,

$$\begin{aligned} M_0 (y;z) = \sum _{g=1}^G \psi _g M_0(y_g;z), \end{aligned}$$

(1)

where $g=1,\ldots ,G$ denotes a particular subpopulation with $\sum _g^G N_g = N$ and $\psi _g=\frac{N_g}{N}$. The adjusted headcount ratio can also be decomposed into the contributions of each dimension. First, let $h_d(k) = \frac{1}{N} \sum _{i=1}^{N} {\mathbb{1}}({c_i \ge k \wedge y_{id} \le z_d})$ be the dimension-specific censored headcount, which allows us to rewrite the adjusted headcount as

$$\begin{aligned} M_0 = \sum _{d=1}^D \frac{w_d}{D} h_d(k). \end{aligned}$$

(2)

Then, the contribution of dimension d to overall poverty is $\frac{w_d}{D} \frac{h_d(k)}{M_0}$. If data on more than one point of time is available, we also can calculate and decompose changes of aggregate measures. Often, however, the decomposition of changes relies on questionable assumptions (Alkire et al. 2015, ch. 9.2). In analysing the dimensions behind changes I, therefore, rely on censored and uncensored headcount ratios simultaneously.

3 Data and specification

3.1 Sample

For the analysis I use data of the German Socio-Economic Panel (SOEP) and calculate a multidimensional poverty index for three periods of time (2001–2002, 2006–2007, 2011–2012).³ In particular, the SOEP provides information on various aspects of a respondent’s life. However, to avoid an overload of the respondents, some questions are only asked every other year (or less frequently), whereas other items are only collected in between these years. Consequently, a comprehensive poverty index can only be calculated for selected years. Moreover, for using the best-suited items simultaneously, I merge two consecutive years into one period. Naturally, this comes at the cost of losing those observations not observed in both years of a period.

A useful target population for the present study are the adults living in Germany in the respective year. Consequently, I treat the SOEP as repeated cross-sectional data.⁴ To account for the complex survey design of the SOEP, the subsequent analyses use sampling weights, which are basically the inverse sampling probabilities (see Goebel et al. 2008).

3.2 Operationalization

The basic prerequisite for any measurement exercise is a clear understanding of the theoretical construct to be measured. The operationalization of the present study draws on the capability approach (Sen 1985, 1992, 1999b).⁵ The capability view not only considers human well-being as inherently multidimensional; moreover, it assigns intrinsic importance to functionings and capabilities. While functionings are the doings and beings individuals have reason to value (e.g., being healthy or participating in the social life), capabilities represent the set of all functionings an individual can actually choose from. Note that intrinsic importance naturally leaves room for instrumental importance as well, as being able to read and write or being healthy illustrate (e.g., Sen 1999b). Poverty, then, is understood as capability deprivation, implying both a shortfall in one or several of the functionings deemed relevant and their infeasibility for the individual in question.⁶ Consequently, indicators of deprivation (1) need to be located in the functioning space and (2) ideally take account of the functioning’s infeasibility.

Moreover, the CA requires value judgments to be exposed rather than concealed, and in addition they must be subjected to public debate. Only with clear presentation of the normative problem, the public debate about these issues can be expected to fulfill its constructive role; see Sen (1999a, p. 10). Value judgments are needed for (1) the selection of functionings included in the index, (2) the respective deprivation cutoffs, (3) the assigned weights, and (4) the poverty cutoff. The official RPW provide a first set of indicators, which aim at measuring important functionings. Specifically, so-called core indicators are to be regularly reported, and their selection is based on scientific advice (Arndt and Volkert 2007).⁷ This selection is thus reasoned and transparent, and yet open to criticism and modification. Hence, the choice of dimensions is subjected to public debate (see also Sen 2004, on this). Subjecting the deprivation cutoffs to public debate, however, further constrains the choice of a functioning’s indicators. Specifically, indicators should allow for deprivation cutoffs that are similar and meaningful across individuals, such that a public debate can study the pros and cons and eventually agree upon those cutoffs. Limitations of available data, however, prompt us to draw on imperfect indicators as well. In some cases a functioning may be only captured incompletely; in others, measurement remains within the resources space.

Finally, the CA assigns goods, income, and other resources only an instrumental role, as they are means, which can be converted into functionings. Remarkably, this conceptual structure reconciles merits of both the “absolute” and the “relative” approach to poverty, since relative deprivation in resources may well translate into absolute capability deprivation (see Sen 1983).

Table 1 Potential dimensions

3.3 Dimensions

The increasing interest in alternative measures of well-being motivated numerous measurement initiatives in various directions. Additionally, a consensus on relevant dimensions is emerging. Table 1 provides an (non-exhaustive) overview of dimensions frequently suggested. While Nussbaum (2001) approaches the question philosophically, the other studies survey and organize available indicators. Nonetheless, they agree on certain dimensions such as education, health, or social participation.

However, Table 1 also reveals two further aspects. First, for some of the more complex functionings of human well-being, such as self-respect, practical reason, or agency, there are no accepted indicators so far. Second, several frequently proposed dimensions, are from a conceptual point of view not functionings. Hence, by themselves they are not dimensions of well-being. Leading examples are housing, material deprivation, and income. Conceptually, all of them provide resource information. In some cases, resource-indicators can be clearly related to a single key functionging, as e.g. housing indicators (even though conversion factors are then ignored). In constrast, many other indicators are likely to affect several functionings and, moreover, in an a priori unclear way (e.g., employment). Thus, they are best considered as multipurpose means.⁸ The vital point is whether we should incorporate or ignore information provided by material deprivation indicators, income and other resource-based indicators. Ignoring crucial information about the lives the poor experience, poses a serious flaw of any poverty measure, just as adding redundant information. The present paper proposes to incorporate resource dimensions, if their indicators—argumentatively or evidentially—contribute important information on otherwise ignored functionings. For instance, indicators of material deprivation may be well-suited to signal a shortfall in both practical reason and economic security. Recent insights from behavioral economics (introduced later) lend support to this nexus. In contrast, a shortfall in income, is not included, since social participation, a key functioning income helps to achieve, is already explicitly modelled. This argument is reinforced by including material deprivation indicators which tend to better reflect material well-being than income. Consequently, adding an income dimension is likely to cause redundancy—given that material deprivation and social participation indicators are already included.

Table 1 finally points to some unresolved issues: Is it better to consider security as a dimension on its own (e.g, Stiglitz et al. 2009, p. 194) or to introduce risky functionings (e.g., Wolff and de-Shalit 2007)? Likewise, how to account best for employment-related aspects requires still more investigation and debate.

Table 2 Functionings, indicators, and weights

3.4 Specification

Due to a lack of space, most indicators are only briefly introduced. The dimensions material deprivation and employment are, however, discussed in more detail. Table 2 shows the selected functionings, their indicators, and the weights. Almost all indicators are either already core indicators of or analyzed within the RPW.⁹

3.5 Education

Education is meant to capture not only achievements in reading and writing, but also the abilities to use one’s senses, to imagine, think, and reason (see Nussbaum 2001). The first indicator ($dep\_educ$) switches to deprivation if a respondent failed to complete elementary education or completed elementary education but later failed to obtain a vocational qualification. Elementary education refers to the graduation after Germany’s 10 years of compulsory education. Beyond formal education, I also consider the number of books within the household. Members of a household owning less than 10 books are considered deprived ($dep\_Nbooks$). This information proxies both the educational climate within the household and effective literacy.¹⁰ However, as a proxy located in the resource space, it suffers the usual limitations (potentially important conversion factors are ignored).

3.6 Health

Deprivation in health, which is multidimensional itself, is signalled by three indicators. First, respondents are deprived in bodily integrity if they are partially or severely disabled ($dep\_disability$). Second, I compile a sub-index, which allows for substitutability among several medical conditions. Two out of four health problems must be reported for being deprived. The four health issues are (1) a strong limitation when climbing stairs, (2) a strong limitation for tiring activities, (3) physical pain occurred always or often during the last 4 weeks, and (4) the health condition limited always or often socially. Finally, a BMI larger than 30 ($dep\_obesity$) indicates obesity (WHO 2000, p. 242) and thus is medically critical. Note that for these indicators the deprivation cutoffs are similar and meaningful across individuals—avoiding a common drawback of indicators like subjectively assessed health state or health satisfaction.

3.7 Housing

Housing indicators are to capture the functionings of being sheltered and enjoying privacy. To measure housing, I resort to resource indicators. Specifically, I consider a person to be deprived of adequate shelter and privacy if any of bath, kitchen or toilet is missing in her accommodation ($dep\_hhfacilities$) or if the respondent reports that her house either “requires major renovation” or is “ready for demolition” ($dep\_housecond$). Finally, I use a simple overcrowding index ($dep\_overcrowded$), which indicates deprivation if there is less than 1 room per person in the household (see Bundesregierung 2013, p. 243). Drawing on resource indicators, however, ignores relevant conversion factors (e.g., the power relations within the family). Moreover, a decent housing may also facilitate more health, self-respect and social participation.

3.8 Social participation

The measurement of social participation exploits information on the frequency with which certain activities are reported to be performed. These activities represent common forms of social life. Respondents may report at least once a week, at least once a month, less often, or never. Table 11 contains the exact wording of the questions. While meeting friends or relatives, the social activity par excellence, is of central importance, many other activities also facilitate relatedness and social interaction. To emphasize the importance of meeting one’s friends (for its own sake), I consider a person deprived if she reports to never meet her friends. The remaining seven items form an activity index. Specifically, the activity index considers an individual deprived if she reports never performing six or seven activities or, alternatively, never performing five activities and, additionally, performing one or two activities less often.

3.9 Material deprivation

Inspired by the work of Townsend (1979) and others, previous poverty measures also used indicators for consumption or ownership on selected goods. Conceptually goods, like income, are resources. Notwithstanding, resource indicators may provide extra information. Material and wealth deprivation are best considered as a shortfall in a multipurpose means. Lacking multipurpose means may affect several distinct functionings simultaneously and, moreover, in an a priori unclear way. This paper proposes to use resource dimensions, if their indicators argumentatively or evidentially contribute extra information on otherwise ignored functionings. More specifically, I argue that indicators of material and wealth deprivation are well-suited to infer a shortfall in both practical reason and economic security.

Nussbaum (2001) suggests the functioning practical reason, referring to an individuals’ capacity to act and to plan one’s life, including the ability to perform deliberate and reasoned actions.¹¹ In economic choice theory this corresponds to the activity of balancing costs and benefits. The proposed justification for material deprivation draws on recent research from behavioral economics. Specifically, Mullainathan and Shafir (2013) argue economic conditions to systematically distort decision-making via the so-called scarcity mindset. Important implications are both focus dividend and tunneling.¹² The authors conclude (p. 119), “When we focus so intensely on making ends meet now, we plan less effectively for the future”. Later (pp. 120–121), they continue, “myopia is not a personal failure. Tunneling is not a personal trait. [...] rather, it is the context of scarcity that makes us all act that way”.¹³ Individuals struggling hard to make ends meet are fully occupied with monitoring every penny spent and earned. Consequently, long-run effects (be it costs or benefits) and goals are located outside the tunnel, and hence ignored. Since it is these economic conditions that induce (inter alia) myopia, decision making is systematically distorted.

Material and wealth deprivation are also suited to signal a lack of economic security. Goods not consumed for financial reasons already indicate difficulties to make ends meet, and thus a threatened level of consumption. The role of wealth (and borrowing) in consumption smoothing is theoretically supported by the permanent income hypothesis. Finally, depending on the specific goods used, material deprivation indicators may also indicate shortfalls in even other functionings (e.g., respecting oneself).

The dimension of material deprivation is operationalized using two sub-indices, which allows a certain substitutability. First, $dep\_wealth$ equals one if none of the following wealth items is owned: life insurance, pension, house or apartment, financial assets, commercial enterprise, tangible assets.¹⁴ Second, $dep\_matdep$ equals one if two or more items of the following are missing for financial reasons (1) a warm meal, (2) friends are invited for dinner, (3) money is put aside for emergencies, and (4) worn out furniture is replaced. Both indicators are suited to detect shortfalls in practical reason and both indicators capture important aspects of economic insecurity. Consequently, extra information on otherwise ignored functionings is added.

3.10 Employment

Previous studies either include an employment dimension or explicitly advocate an employment capability (e.g., Leßmann and Bonvin 2011; Alkire and Apablaza 2016). In fact, by now there is widespread agreement about the importance of employment for human well-being (e.g., Stiglitz et al. 2009; Bundesregierung 2013). Employment may indeed help to do things which are intrinsically important (e.g., contributing one’s share for the good of all). However, an array of effects of labour on other dimensions of well-being has been documented as well. In fact, most information collected for the labour-well-being nexus usually pertains to its instrumental relevance (e.g., occupational diseases and risks for accidents, various security schemes, workers’ participation in various processes, exposure to adverse conditions, etc.). Suppa (2015) argues labour to be a crucial device for achieving numerous functionings, such as being healthy, agency, self-respect, practical reason, appearing in public without shame, etc.

Unemployment, for instance, was found to decrease life satisfaction (Kassenböhmer and Haisken-DeNew 2009) and public social participation (Kunze and Suppa 2017). Hetschko et al. (2013) find identity utility to be important, which from a capability view may indicate an effect on being ashamed or respecting oneself. If, however, perfect measures for all relevant functionings were available, there would be no need to rely on an unemployment indicator. As it stands, however, accepted measures for many of the more complex functionings are lacking and existing ones might be incomplete. Thus, similar to the material deprivation indicators, employment-related indicators may provide important extra information on otherwise ignored functionings.

The current specification draws on three employment-related indicators. First, if an individual reports to be registered unemployment $dep\_unemp$ equals one. As outlined above deprivation in numerous functionings a likely to accompany unemployment. Moreover, $dep\_underemp$ equals one if a person reports to involuntarily work less than 30 h a week. This may be associated by shortfalls in similar functionings, although to a lesser extend. Moreover, part-time jobs are often found to provide lower job quality. Restricting deprivation to the involuntary is important, since for many households part-time work may, in fact, be desirable for improving the work-life balance. Finally, $dep\_precemp$ equals one for persons who are marginally employed or in temporary employment. Associated deprivations are in social and economic security and practical reason.

3.11 Weights

The main specification assigns equal weights to each dimension and, within a dimension, equal weights to each indicator. Consequently, most indicators receive a weight of $\frac {1}{18}$, whereas education and social participation indicators receive $\frac {1}{12}$ each. Finally, note that full deprivation in employment is only achieved by unemployed (weighted with $\frac {1}{6}$). Assigning the other two indicators a weight of $\frac {1}{18}$ each, implies an improvement for a formerly unemployed, who finds a precarious part-time job.

Table 3 Deprivation headcount ratios

3.12 Deprivation headcounts

Table 3 provides first information about deprivation indicators. The uncensored deprivation headcount is simply the share of individuals deprived in a given indicator. Uncensored headcounts for the whole population (total) indicate different levels of prevalence for different dimensions. Housing indicators, for instance, vary from 1 to 5%. Similarly, employment indicators vary from 4 to 7%, whereas deprivations in wealth or social participation amount up to 20% each. The so-called dashboard approach exclusively relies on these headcount ratios along with their changes (i.e. the marginal distributions). However, the uncensored headcount ratio can also be calculated for certain subgroups, e.g., by poverty status. While only 1% of the non-poor is found unemployed, 32% of the multidimensional poor are. Table 3 also reveals that the most prevalent deprivations among the poor are material deprivations (67–79%) and social participation (54–62%). Similarly, 47% of all poor are deprived in education. Given the counting approach to poverty, higher prevalences for the poor are to be expected. In fact, prevalences are substantially larger for the poor—often by a triple or more. This finding simply mirrors the fact, that the AFM exploits the joint distribution of deprivation in the identification step of poverty measurement. Put differently, the AFM uses the joint distribution to distinguish more important (i.e. coupled) from less important (i.e. occasional) deprivations.

In addition to this, Table 3 also shows the censored headcount ratios, i.e. the share of the population who is poor and deprived in the given indicator. While the censored headcount must be smaller or equal the uncensored headcount, it is important to note that none of the censored headcounts is really close to its uncensored headcount. Thus, virtually no indicator directly implies poverty (i.e. multiple deprivation). The final column contains the share of a given deprivation borne by the non-poor. For most indicators the non-poor account for 50% or more of a deprivation. The only exception is unemployment, where only 20% of the unemployed are non-poor (which results from the higher weight). Note that remarkable shares of deprivations are deliberately ignored in the subsequent analysis, as they are not coupled with other deprivations. More importantly, to infer from a declining uncensored headcount ratio what happens to the multiply-deprived becomes a doubtful exercise for the data at hand (see Suppa 2017). In sum, these findings suggests that neither a single indicator, nor a dashboard approach can replace the multidimensional approach.

Fig. 1

Aggregate measures over time. Notes Data from SOEP v29.1

Fig. 2

$M_0$ by subgroups. Notes Data from SOEP v29.1. calculations for 2011/2012

4 Empirical analysis

4.1 Aggregate measures

Figure 1 depicts the multidimensional poverty measure $M_0$, the incidence H, and the intensity A—each for all three periods and for poverty cutoffs $k \in [25,50]$. Figure 1 suggests for both $M_0$ and H an increase from period 1 to 2 and a decrease from period 2 to 3—independent of k. Average intensity seems to be lowest in 2001/02. Setting $k=33$ implies 10.7% to be poor, whereas a more conservative cutoff, say $k=38$, implies a headcount ratio of 6.8%.

To obtain a more detailed account of multidimensional poverty Fig. 2 shows $M_0$ for different subpopulations. Apparently, individuals with a background of migration exhibit a larger $M_0$, irrespective the poverty cutoff chosen. Likewise, East-Germany and women experience more multidimensional poverty, although with less pronounced differences. The results are less clear-cut for different age groups, since they depend on the chosen poverty cutoff k. People aged 45–65, however, tend to experience most multidimensional poverty.

Figure 2 also uncovers that persons in single households tend to experience more poverty than individuals in households of couples, regardless of eventual children in the household. Finally, the importance of the father’s education on an individual’s deprivation is salient. Three groups need to be distinguished: First, persons with fathers completely lacking education or where education is unknown are associated with the highest $M_0$. The second group consists of individuals whose father completed Hauptschule or other schools, while the third contains those whose fathers completed Realschule and Abitur. Differences in average intensity vary less by subgroup (results available upon request).

Note that these insights, generated by the adjusted headcount ratio, are consistent with previous findings. The systematic discrimination of individuals with migration backgrounds is just as well documented as the influence of the family background on the offspring’s educational achievements (e.g., Bundesregierung 2008, ch. IX and III.5).

Table 4 Contributions to $M_0$ and H by Education of Father

4.2 Contributions of subpopulations

In addition to an analysis of multidimensional poverty by socio-demographic groups, overall multidimensional poverty can also be decomposed into the contributions of each group to account for relative population sizes (see Eq. 1). Setting to $k=33$, Table 4, shows such a decomposition for the father’s educational background. Specifically, respondents reporting their fathers to have completed Hauptschule alone make up approximately $\frac {2}{3}$. Including individuals reporting their father’s education to be unknown, uncompleted, or absent, the share of multidimensional poverty associated with a handicapped education of the father climbs to ca. 85%. Even though the corresponding population share is 69%, this finding clearly underlines the role of the educational background of the father in multidimensional poverty.

Table 5 Dimensional breakdowns

4.3 Dimensional breakdown

Multidimensional poverty ($M_0$) can also be decomposed into each indicator’s contribution using Eq. (2). The contribution of indicator d can be expressed absolutely, $\frac{w_d}{D} h_d(k)$, summing to $M_0$, or relatively, $\frac{w_d}{D} \frac{h_d(k)}{M_0}$, summing to 100%. Table 5 (a) reveals social participation, material deprivation and unemployment to contribute most to $M_0$, whereas housing indicators contribute least. Even though, dimensional contributions are broadly similar over the years, Table 5 (a) also suggests employment and material deprivation indicators to become more important over the decade, both absolutely and relatively. Apparently, these indicators also account for the peak of $M_0$ in 2006/07.

To study different profiles of poverty, dimensional breakdowns can be calculated for subgroups. Table 5 (b) suggests absolute and relative contributions of social participation and health to increase with age. Material deprivation indicators contribute slightly less in old age. For persons with a background of migration, material deprivation and housing contribute relatively more to multidimensional poverty, whereas health indicators contribute relatively less. Absolute contributions, however, reveal virtually every indicator to contribute to the higher $M_0$ for persons with migration background.

Table 6 Changes in the adjusted headcount ratio

4.4 Dynamics

Indeed, the period of investigation covers, among other things, a major labour market reform, which was implemented in several steps over the decade (along with some modifications), and the financial crisis starting in 2007. Unfortunately, a more careful evaluation of these events is, however, difficult with data for only three points of time. Nonetheless, a natural starting point for studying poverty dynamics is to document changes over time. Table 6 (a) contains absolute and relative changes in $M_0$ for several k. Irrespective of k, multidimensional poverty is growing during the first and falling during the second part of the decade. Moreover, Table 6 indicates that over the whole decade, multidimensional poverty remains approximately unchanged for $k=33$, but would have increased for $k\ge 44$ and decreased for $k<33$. In order to obtain a deeper insight into changes of multidimensional poverty one can compare changes in censored and uncensored headcounts, which are both depicted in Table 6 (b). Specifically, the three employment indicators and the material deprivation indicator exhibit relatively high changes in both censored and uncensored headcounts during the first half of the decade. Apparently, these four indicators drive the overall increase in multidimensional poverty observed from 01/02 to 06/07. Similarly, indicators for education and unemployment play a crucial role for reducing $M_0$ during the second half of the decade.

Other patterns, however, are more difficult to rationalize and require a more careful analysis. The uncensored headcount of education, like several other indicators, falls during the first part of the decade. In contrast, the respective censored headcount of education (at $k=33$) hardly changes at all. This finding simply means that the same share of the population is still multidimensionally poor and deprived in education—despite the decrease in the uncensored headcount. Likewise, over the decade unemployment among the poor is unaffected by the substantial reduction of the unemployment rate. Moreover, censored headcounts also suggest that precarious employment and underemployment increase among the poor during the second half of the decade, despite the overall decrease in $M_0$ for this period.

A complementary analysis is to study changes by subpopulations. A simple and yet instructive exercise compares absolute changes in $M_0$ by selected subgroups. Note, however, that the poor people at the beginning of the period may not be the same as the poor people at the end—even though they may exhibit the same socio-demographic characteristics. Table 6 (c) shows both absolute and relative changes by four different subgroups. East-Germany, for instance, experiences both a larger rise during the first half and a smaller fall during the second half. Notably, this implies the multidimensional poverty gap between both regions to grow during the period investigated. Moreover, while most age groups first experience a higher $M_0$, only youngest (<25) and oldest (65+) people are finally better off. Even though the group of people with a background in migration experienced a stronger increase during the first half, they undergo an even stronger relief during the second half of the decade. Taken together, the gap between migrants and non-migrants is shrinking. Likewise, I find a slightly decreasing gender gap for multidimensional poverty.

4.5 Interim conclusions

The empirical analysis reveals distinct profiles of poverty (e.g. for the elderly or migrants). Moreover, the findings also highlight the role of the education of the father, evidently an important factor for multidimensional poverty. Independent of the the poverty cutoff, multidimensional poverty rose during the first and fell during the second half of the decade. While remaining approximately unchanged for $k=33$, more conservative cutoffs imply an increase for the decade. A more detailed analysis suggests unemployment and material deprivation as critical factors behind changes in multidimensional poverty. However, as the poor might be hit by very different trends simultaneously, a more complex picture emerges. Precarious and underemployment, for instance, rise throughout the decade, whereas unemployment and material deprivation both peak in 06/07. While unemployment later falls even below its initial level, material deprivation remains above. The dynamics behind changing gaps between subpopulations in multidimensional poverty may add even more depth to explanations of poverty.

Table 7 Income and multidimensional poverty

5 Multidimensional poverty and income

5.1 Multidimensional and income poverty

Income-poverty is both an alternative to multidimensional poverty measures and a potential dimension. Table 7 (a) contrasts the headcount ratios for income and multidimensional poverty. Note that $k=33$ and 60%-of-median-income imply similar incidences (ca. 11–13%) during the first years. Monetary poverty, however, also rises from 2006/07 to 2011/12, for which multidimensional poverty is falling. Suppa (2016) studies changes in both measures more carefully.

An important question is whether both measures identify the same individuals as poor. Naturally, such a comparison depends on the poverty cutoffs. Table 7 (b) shows the population shares of individuals who are considered poor (1) by both measures (both-poor), (2) by income poverty only (IO-poor), and (3) by multidimensional poverty only (MDO-poor). For $k=33$ and an income poverty cutoff of 60% only 5% of the population is identified as poor by both measures. Moreover, 8% are IO-poor and 5% are MDO-poor. Neither other cutoffs nor years essentially affect this finding (further results available upon request). Generally less than 50% of the income-poor are also multidimensionally poor. As both measures substantially disagree on who is poor, different policy implications are to be expected.

5.2 Income as a dimension?

Previous studies frequently used income as a dimension. Indeed this paper, proposes to use resource dimensions as well—however only if their indicators, argumentatively or evidentialy, contribute extra information on otherwise ignored functionings. Abandoning a lack-of-income dimension may result in ignoring multiply deprived, whereas including such a dimension introduces the risk of double-counting. Rather than counting a novel deprivation, the income-driven lack of, say, health or social participation is counted once again.

Thus, a lack-of-income dimension necessitates missing functionings. Frequently social participation is a reasonable candidate. As social participation is shaped by customs, organization and endowment of a society, it is also often used to justify a relative income-poverty cutoff. Consequently, adding a lack-of-income dimension is sensible if social participation is unavailable as a separate dimension. Similar arguments may apply for other functionings like practical reason or economic security. However, the present study conceptually prefers material deprivation indicators to an lack-of-income dimension: First, material deprivation indicators can be linked more closely to specific functionings, as practical reason and economic security (see Sect. 3). Moreover, consumption is often argued to better proxy material well-being than income (e.g., Slesnick 2001).¹⁵

Table 8 Deprivation rates and socio-economic characteristics by poverty status

5.3 Who are the IO-poor?

The individuals potentially added or ignored as poor are the IO-poor. To assess whether IO-poor should, in fact, be considered multidimensionally poor as well, one can compare their deprivations with those of the non- and both-poor. Table 8 provides information about the incidence of deprivations and the (uncensored) deprivation count by poverty status. Indeed, the IO-poor exhibit a slightly higher average deprivation count of 0.17 compared with the non-poor (0.1). Their deprivation count is, however, much lower than that of the MDO- or both-poor (0.4 and 0.45) and it is also not just barely below the poverty cutoff. While most deprivation indicators are similar in size for IO-poor and non-poor, the outstanding exceptions are both material deprivation indicators, which are substantially higher for the IO-poor and also can explain their higher deprivation count. This finding points to a sizeable overlap of material deprivation indicators with income-poverty and, moreover, weakens the concern that IO-poor are actually multiply deprived.

5.4 Income and material well-being

Conversely, one may also question that low income accurately reflects material well-being, in particular for the IO-poor. In fact, Slesnick (2001, pp. 196–197), notes that this applies in particular for young and elderly persons. One reason for income to underestimate actual consumption is that it ignores the role of wealth. Wealth can be directly consumed, but may also provide a service flow from its stock (e.g., self-occupied property or durables). Another reason is that permanent rather than current income matters for consumption, implying students, for instance, to be on a different consumption trajectory than their even-aged contemporaries.

As also shown by Table 8, the young (aged 30 and below) are not only overrepresented among the IO-poor, moreover, together with the elderly (aged 60 and above) they account for ca. 60% of the IO-poor. Table 8 also contains frequently collected wealth information. It turns out, that 24% of the IO-poor own their accommodation and 64% own a car, indicating a substantial share of this group to have indeed access to wealth.

Table 9 Wealth by income-poverty status and material deprivation

5.5 The role of wealth

To illuminate the role of wealth for the present context, I exploit data collected by a comprehensive wealth module in 2002 and 2007 (Frick et al. 2007). Table 9 provides information on wealth by income poverty and material deprivation status, the latter meaning individuals are deprived in at least one material deprivation indicators. Interestingly, if individuals report to be non-deprived according to both material deprivation indicators, almost half the respondents own their accommodation—whether income poor or not. Moreover, 72% of the not materially deprived, but income poor report to own a car, and 78% own wealth totalling 7.000 EUR or more. Indeed, their median net wealth is with around 66.000 EUR, quite substantial. If however respondents report to be materially deprived, all wealth indicators drop substantially (only car ownership falls less). These observations are important for two reasons: first they document a certain number of income-poor to own significant wealth and, second, that material deprivation indicators seem to capture this quite accurately (despite relying on considerably less questions than the comprehensive wealth module).

Table 10 Partitioning the income-poor

Finally, 10 partitions the income-poor by wealth ownership, material deprivations, and multidimensional poverty status in order to gauge the scope for incorrectly ignored IO-poor. Apparently,  60% of the income-poor do not own wealth and are materially deprived, i.e. in these cases low income and a lack of wealth manifest in material deprivation. Another ca. 15% are materially deprived despite owning some wealth, implying a total of 75% of the income-poor to be materially deprived in at least one of the indicators. Another, 20% of the income-poor, which are IO-poor in fact, report to own wealth and not to be materially deprived, suggesting income to reflect their material well-being inaccurately. The remaining 5% of the income-poor, however, lack wealth and do not report material deprivation. As the low income of this group neither manifests in material deprivation, nor is indicated to inaccurately reflect material well-being, it may point to incorrectly ignored poor persons.

5.6 Interim conclusions

The lack-of-income dimension is abandoned for both conceptual and empirical grounds. First, social participation, a key dimension income is important for, is directly implemented. This also extends to material deprivation, which is related more directly to deprivations in other functionings (economic security and practical reason). Third, there is evidence for extensive double-counting as ca. 75% of the income-poor are also deprived in at least one material deprivation indicators. Moreover, the slightly higher deprivation count of IO-poor is substantially driven by material deprivation indicators, and thus weakening the concern for the IO-poor to be truly multiply deprived. Finally, for a significant share of the IO-poor (20%), income seems to inaccurately reflect their material well-being.

6 Concluding remarks

Instead of another summary, I conclude with some final remarks. Several of the results, the present study finds, are independent of the underlying poverty cutoff k. For instance, poverty unambiguously rises during the first half of the decade, as it falls during the second half. In contrast, whether poverty ultimately stands at a higher level at the end of the decade, depends on k. As defining k is inevitably a normative exercise, it can neither be solved solely within academia, nor can it be purely chosen to obtain a “reasonable” poverty headcount. Instead, setting the poverty cutoff must draw on the public debate, for which it is crucial that the cutoff can be expressed in semantic meaningful terms (e.g., being poor requires full deprivation in at least two dimensions). Although the present study frequently uses $k=33$, this is most likely a lower bound. Targeting individuals where deprivations are most interwoven, may suggest stricter cutoffs. While stricter cutoffs are normatively more forcing, lower cutoffs may imply headcount ratios too high to be useful. However, introducing more convincing poverty measures may, on their part, also help to objectify the public debate on poverty—especially in advanced economies like Germany.

The next steps towards a multidimensional poverty index for Germany should explore the options for direct implementations of missing dimensions such as such as agency, self-respect, security, practical reason and appearance in public without shame. Currently, complex functionings like these are only captured indirectly. Research on direct implementations is under way (Alkire 2007). Additionally, a clear conceptual account of both employment and time is needed for a better integration of deprivations related to them. The integration of time-related aspects is, though complex, in progress (e.g., Merz and Rathjen 2014). Implemented dimensions might be slightly refined as well. Social participation, for instance, could be complemented with questions on Internet-based activities.

Importantly, some groups currently ignored, must be taken adequately into account. While homeless people are not covered by the underlying data, children, in contrast, are deliberately excluded, since accurately capturing their well-being requires a distinct specification. Taking these groups into due consideration is of highest significance. Moreover, given the current data, a more careful analysis of shocks (e.g., of the financial crisis) and reforms (e.g., of the labour market regulations) is difficult, since the poverty measure cannot be computed on a yearly basis. Assuming a consensus on the indicators, this is, however, straight forward to fix. Similarly, accounting for confounding factors would deepen the analysis and help to uncover the mechanism behind multidimensional poverty.

Appendix

See Table 11.

Table 11 Questions