Does the Choice of Well-Being Measures Matter in Respecting Human Dignity? Empirical Results from an Indian Slum Survey

Abstract

This study compares and analyzes four leading well-being measures: standard monetary measures (income and expenditure), subjective well-being (life satisfaction), equivalent income, and the Alkire-Foster multidimensional poverty index (MPI). To consistently compare these measures, we conducted a field survey of Delhi’s low- and middle-income respondents in India. Using our dataset, we find that the equivalent income approach tends to ignore the economic value of education because of an adaptive preference problem caused by standard measures of life satisfaction. The MPI approach shows that educational deprivation is prominent in our survey. The results suggest that the MPI may complement traditional income comparisons by appropriately setting components, weights, and deprivation cut-offs. Additionally, we find many violations of the dominance principle in these measures, except for MPI.

The authors would like to appreciate kind encouragements and helpful comments from Professors Kotaro Suzumura, Reiko Gotoh (the editor of this book), and John Weymark in developing practical methods of social welfare function approach. We thank the participants of the 15th Annual Meeting International Society for Quality-of-Life Studies, the JADE Conference, and the ANPOSS/ENPOSS/POSS-RT 2021 Joint Conference who heard earlier versions of this paper. Especially, we are grateful to Takeshi Aida for thoughtful comments and suggestions. We also offer our special thanks for the grants from Grant-in-Aid for Scientific Research (C) (grant number: 19K01683) and the Joint Research Center for this research. This study is based on the results of our original survey conducted in India under the approve of ethical review (ethical review number: 2017C006, Hitotsubashi University).

1 Introduction

How can human dignity be defined? The history of human dignity reveals that the term has been used in more ambiguous and polysemic ways than expected. For example, in the traditions of ancient monarchies and Catholics, dignity simply meant respect for those of higher social positions and their privileges. According to Kant (1785), dignity is inalienable, invaluable, and non-transferable worth granted to autonomous and rational humans, who should be treated as an end in themselves, not a means to some other end. Under the Kantian view, human dignity should have intrinsic value that cannot be compromised by any quantity of resources and money. In bioethics, human dignity is often interpreted as a simple aspect of self-determination (Macklin, 2003). Recently, Nussbaum (2006, 2011) stated that a life with human dignity requires a list of basic capabilities to be higher than some threshold levels. However, a major problem with these concepts of dignity is that persons with severe intellectual or mental disabilities or dementia who need daily care cannot be the object of dignity because they seem to lack autonomy, reason, or some basic capabilities. Trying to reconstruct the idea of dignity in terms of care ethics, Kittay (2005) made some reasoning in a justification for the dignity of persons with disabilities.

Even in the context of evaluating individual well-being, the implications of respect for human dignity are ambiguous. Consider the indexing dilemma shown in the context of Rawlsian primary goods (Gibbard, 1979). This indexing dilemma shows that interpersonal comparisons of individual well-being that satisfy two plausible requirements must lead to a cycle. Formally, it states that no acyclic well-being ranking exists that satisfies the principles of individual preference and dominance. The individual preference principle requires that each intrapersonal comparison of one’s well-being reflects only one’s preference. The dominance principle requires that for all interpersonal comparisons of well-being, individual i’s well-being is at least as good as j’s well-being whenever i’s consumption bundle (or functioning vector) is weakly greater than that of j. Acyclicity requires that no cycle exists for all interpersonal comparison rankings.¹ It is well known that these principles are incompatible in various contexts of social choice theory (Brun & Tungodden, 2004; Fleurbaey, 2007; Pattanaik & Xu, 2007; Weymark, 2017). This means that any well-being measure faces difficulties in comparing individual well-being whenever there are differences among individual preferences.²

It seems plausible that respecting each individual’s preference in assessing their living standards should be part of dignity regarding self-determination. Suppose an individual’s evaluation of their living standard is not respected, and their situation is simply evaluated by social value judgment. In that case, it may be too paternalistic and perfectionistic to take dignity as self-determination seriously. On the other hand, it also seems plausible that respecting a dominance relation in assessing living standards among different individuals is consistent with the view that a life with dignity should be full of sufficient basic capabilities. As Adam Smith (1776) pointed out, life with human dignity requires a certain material basis. It is valid to claim that the living standard of a wealthier person is strictly better than that of a poorer person. An individual who appears in public feeling shameful (i.e., having inadequate food, clothing, shelter, etc.) can hardly be said to be in full respect for human dignity. This study does not offer a definitive argument as to which principles should be respected in the debate over the indexing dilemma and human dignity. Instead, we review the key measures that have received attention as important indicators of individual well-being and investigate their application issues by focusing on the indexing dilemma.

Let us provide a short history of the measurement of well-being. Traditional methods based on GDP per capita have failed to adequately reflect distributional considerations (especially the lives of the low and middle classes). Moreover, GDP per capita could not consider any deterioration of important components of social welfare (e.g., declining social bonds, worsening inter-group conflicts, degradation of the natural environment, and rising inequality of opportunity). Since the well-known report by Stiglitz et al. (2009), many governments and international organizations have started to consider and use various policy evaluation methods, such as the Alkire-Foster multidimensional poverty index (Alkire-Foster MPI), equivalent income, OECD better life index, and some subjective well-being indices, instead of the traditional GDP per capita. These novel methods are expected to shed light on various individual and social well-being aspects. Specifically, the Alkire-Foster MPI has become a common method for policy evaluation by capturing the deprivations in health, education, and living standards that a person faces simultaneously. (Alkire & Foster, 2011; Alkire et al., 2015). On the other hand, the equivalent income approach, which is a revival version of classical money-metric utility (Samuelson, 1974; 1977) and Pazner-Schmeidler’s egalitarian equivalent approach (Pazner & Schmeidler, 1978), has been now on some trials by applying standard econometric methods to measuring individual well-being (Decancq et al., 2015a). Subjective well-being is a very popular method in many happiness studies and has been considered a surrogate for well-being, reflecting non-monetary life dimensions, such as social relationships (OECD, 2013, 2019). These approaches have several reasonable properties and are useful for measuring various aspects of well-being.

However, each measure used in this study is known to have some defects. For example, the standard money-based approach, which focuses on income or consumption levels, cannot reflect the various disadvantages of discriminated or vulnerable groups that suffer from education and employment opportunities. It also fails to consider the unequal treatment of women in their ownership of household assets, which is often observed in developing countries where gender discrimination is so strong. Furthermore, even if individuals are given the same goods and purchasing power, there are non-negligible individual differences in what they can do due to disabilities and other factors.³ The subjective well-being approach has the problem of adaptive preferences, where people often adapt to their circumstances and are likely to report that their subjective well-being is not so bad, regardless of whether they are in a favorable or unfavorable circumstance.⁴ Generally, neither the income nor life satisfaction approach satisfies the dominance and individual preference principles.⁵ On the other hand, the MPI approach satisfies the dominance principle among the poor identified by some MPI criteria, but it violates the individual preference principle.⁶ In contrast, the equivalent income approach satisfies the individual preference principle but violates the dominance principle.

While each well-being measure has some limitations, there are surprisingly few comparative and comprehensive analyses of these indices.⁷ This is mainly because no database allows simultaneous comparison of these indices. Hence, this study compares and analyzes four leading well-being measures (simple money-based indices, happiness as life satisfaction, the Alkire-Foster MPI, and equivalent income) using original survey data from Delhi, India. As a result, we show how often theoretical flaws occur and the extent to which informational loss can be supplemented by using another measure.

This study contributes to the literature on well-being measurements. First, we show that in our dataset from India, the problem of adaptive preference is so serious that the subjective well-being approach cannot work well for measuring individual well-being.⁸ This result further strengthens the conventional perception that we should not only evaluate social welfare in terms of subjective well-being. Second, the problem of adaptive preferences implies that one method of the equivalent income approach, that is based on subjective well-being estimation, fails to reflect some important aspects of well-being, such as education, which does not matter regarding subjective well-being. The fact that subjective well-being among the poor and the low caste is often high also implies an endogeneity problem in previous empirical estimation methods of equivalent income. Third, by comparing leading measures with income and consumption per capita, we show that the Alkire-Foster MPI has a low correlation coefficient with the money-based measurements. This fact suggests that MPI is more sensitive to several aspects of well-being than other measures because it can consider material and social information such as housing, health, job status, and education. Finally, our sample suggests that violations of the dominance principle are the rule rather than the exception in all measures except the Alkire-Foster MPI.

The remainder of this study is organized as follows. Section 2 explains the survey design and methods of calculating the measures. Additionally, the basic statistics and properties of the well-being measures are reported and discussed. Section 3 compares the results of four measures by using pairwise rank correlation between them and observing some characteristics of the bottom 10%. Finally, Sect. 4 summarizes the results and discusses the remaining issues.

2 Data and Methodology

2.1 Survey

We surveyed low- and middle-income individuals in Delhi, India, in November 2017 to consistently compare various well-being measures.⁹ To generate our sample, we first randomly select five assembly constituencies. The results of this selection included Jahangirpuri, Lajpat Nagar, Hauz Khas/Malviya Nagar, Raghubir Nagar, and Okhla. Next, to obtain a sample representing low-income individuals, we randomly selected jhuggi-jhopdi (JJ) colonies, which are slum resettlement locations, using the list of JJ colonies available on the Delhi Municipality website. For the middle-income group, areas adjacent to the JJ colonies were selected for the survey. After selecting low- and middle-income localities, we randomly selected households for the survey.¹⁰ Ultimately, 510 respondents completed the survey. The average age of respondents was 35.36, and men were 46.08%. The number of respondents who belong to households holding the ration card (below the poverty line type, BPL)¹¹ is 226, which is almost half the sample. Additionally, 28.63% of respondents receive no official education.

Our survey covered the following information: (1) demographics (gender, age, family configuration, caste, religion); (2) happiness as life satisfaction; (3) education; (4) occupation and employment status; (5) income and consumption level; (6) health status; (7) housing facilities and household assets; (8) social capital; and (9) security and environment. The following subsections explain the four well-being measures used in this study.

2.2 Income and Expenditure

The first well-being measure is based on the income/expenditure approach, which has traditionally been mainstream in poverty and inequality analysis. Income level has often been interpreted as a surrogate for individual well-being and plays an important role in policy goals and evaluation. However, income is just one of the various tools through which an individual achieves well-being. Moreover, there are difficult problems with the income approach, such as life cycle (e.g., a tendency to have low income in one’s youth and old age and high income in middle age) and diversity among human abilities to transform goods into functionings (i.e., differences among individual living levels due to diversity of abilities and social discrimination in gender, class, occupation, and race). Hence, using income level as a surrogate for human well-being requires significant research. Note that, as is widely recognized in recent years, consumption expenditure is a more suitable surrogate for human well-being than income.¹²

In the survey, annual household income and expenditure brackets included 50 or less, 50–100, 100–200, 200–300, 300–400, 400–500, 500–600, and 600 or above, denominated in thousands of rupees. Therefore, to calculate individual income, household income was divided by the square root of the number of persons in the household.¹³ Figure 1 depicts the histograms of individual income and expenditure. Using a poverty line (Rs. 13,608) for urban areas in Delhi, only 2.6% of respondents were below the poverty line in the case of individual incomes, while 16.3% were below the poverty line in the case of individual expenditures. Generally, individual expenditures seem to be a good surrogate for one’s living standard. They are suited to identify the poor since consumption directly generates utility and is smoothed over one’s life cycle. However, previous studies have used individual income because of its availability and simplicity, and we use both income and expenditure variables in the following sections.

Fig. 1

Distribution of individual income and expenditure

2.3 Alkire-Foster MPI

The second well-being measure is the Alkire-Foster MPI, a multidimensional poverty index proposed by Alkire and Foster (2011).¹⁴ The Alkire-Foster MPI satisfies some desirable properties, such as monotonicity, poverty focus, and decomposability. MPI is defined as a generalization of the weighted average deprivation rate,¹⁵ which is a natural extension of the well-known single-dimensional income inequality measure (Foster et al., 1984). Therefore, it is expected to be more adequate in analyzing the details of poverty at individual or national levels compared to traditional money-based methods, such as average income and consumption expenditure. The global MPI evaluates national poverty in three dimensions: health (infant mortality and malnutrition), education (school attendance and years of schooling), and living standards (housing facilities and assets), and it succeeds in showing multifaceted deprivations in developing countries.¹⁶ However, there are certain problems with the MPI approach. For example, any MPIs fail to respect unanimous judgments of interpersonal well-being comparisons.¹⁷ Additionally, since empirical results strongly depend on a specification of the MPI, such as choosing dimensions, setting each poverty line in each dimension, selecting the poor, and deciding on weights, the rationale of the MPI is still unclear and has to remain arbitrary.¹⁸

The aggregation method proposed by the Alkire-Foster MPI considers both the incidence and intensity of poverty. Generally, an individual or group i’s Alkire-Foster MPI is calculated as follows:

$${MPI}_{i}=\left\{\begin{array}{ll}\sum_{d\in D}{w}_{d}{\left[\frac{max \{{y}_{d}^{*}-{y}_{id}, 0\}}{{y}_{d}^{*}}\right]}^{\alpha } & if \, individual\, i\, is\, a \,poor,\\ \\ 0 & otherwise,\end{array}\right.$$

where d is a component of the MPI, ${y}_{id}$ is i’s value of component d, ${y}_{d}^{*}$ is the poverty line of d, and w_d is the weight of d with $\sum_{d\in D}{w}_{d}=1$. Let $max\{{y}_{d}^{*}-{y}_{id}, 0\}/{y}_{d}^{*}$ be the deprivation rate of d. If α = 0, MPI measures the percentage of components that are deprived within the poor. If α = 1, it is equivalent to the weighted average of the deviation rates. We calculate the Alkier-Foster MPI for both cases in which α = 0 and 1, but the case where α is zero shall not be listed in our results to save space.

The dimensions, weights, and components are presented in Table 1. We calculated three types of Alkire–Foster MPI by considering three (AF-MPI-3D), four (AF-MPI-4D), or five dimensions (AF-MPI-5D). For these three dimensions, we consider health, education, and living standards, which have the same dimensions as the global MPI. Additionally, we add the employment dimension for the AF-MPI-4D, and the environmental dimension for the AF-MPI-5D since the dimensions of employment and environment are also considered for estimating the equivalent income based on life satisfaction.

Table 1 Dimensions, components, and weights for calculating MPI

Table 2 summarizes the questions in our questionnaire for measuring each component and the cut-off points for deprivation. Note that our survey is not based on the household level but at the individual level. Since we asked one respondent within each household in our survey due to budget constraints, we have only information about the respondent’s health, education, and employment status.

Table 2 Components, related questions, and deprivation cut-offs

Health is measured by the so-called subjective health index. Previous studies show that the subjective health index can represent the respondent’s actual health conditions (See van Doorslaer & Jones, 2003).¹⁹ We asked respondents to rate and evaluate their health on a 5-point scale: very good, good, fair, bad, and very bad. We assume that a person is deprived if their answer is “bad” or “very bad.” Education is measured by years of schooling. A person is considered deprived if the number of years of schooling is less than ten, which is the lower secondary education level. Living standards are measured in almost the same way as the global MPI: cooking fuel, sanitation, drinking water, electricity, roofs and floors, and assets. Drinking water was divided into two categories: quality of drinking water and location of the water source. Employment was measured by two categories: unemployment and working hours. In urban areas, employment matters for well-being. Therefore, a person is considered deprived when they are unemployed. Additionally, a person is considered deprived when they work over 60 h per week since such long working hours seem to affect physical and mental conditions. The environment was measured in two categories: safety and air quality. Respondents were asked to rate and evaluate their area’s safety and air quality on a 5-point scale. A person is considered deprived when the answer is “bad” or “very bad.”

Table 3 shows the proportion of respondents considered deprived for each dimension. The share of deprivation of education is close to 40%, which is the highest among all dimensions. Figure 2 presents the distribution of the weighted averages of the deprivation rates related to the five dimensions. Although about 50% of respondents in our survey were from BPL households, the most frequent deprivation rate was 0 among all dimensions. Regarding living standards, no individual is simultaneously deprived of all categories, while about 30% of individuals are deprived of at least one category. Note that the deprivation rates for education and employment can range from zero to one.

Table 3 The share of deprived respondents in each component

Fig. 2

Distribution of deprivation score related to five dimensions

Finally, we calculated the weighted average deprivation rates of all respondents using three types of Alkire-Foster MPI with α = 1²⁰: AF-MPI-3D, AF-MPI-4D, and AF-MPI-5D. Figure 3 shows the distribution of deprivation rates for all the respondents. The global MPI identifies the poor when at least 33% of the weighted components are zero. The solid line in Fig. 3 is the global MPI threshold of the poor (0.33). The poverty ratios identified by AF-MPI-3D, AF-MPI-4D, and AF-MPI-5D were 36%, 18%, and 16%, respectively. While education is a significant dimension that increases deprivation scores, the weights for education decrease as the dimension increases. As a result, the poverty ratio decreases as the number of dimensions considered in the AF-MPI increases.

Fig. 3

Distribution of AF-MPI-3D, AF-MPI-4D, and AF-MPI-5D. Note The solid line represents the poverty line for the MPI, that is, 0.33

2.4 Subjective Well-Being as Life Satisfaction

The third well-being measure is happiness as life satisfaction, a key measure of subjective well-being. Generally, subjective well-being, such as life satisfaction, measures the cognitive aspect of subjective well-being by focusing on reflective and objective evaluations of well-being rather than the emotional factors that lead to a feeling of happiness. The biggest problem within the subjective well-being approach is an adaptive preference problem due to both the hedonic and aspiration treadmill. This leads to many paradoxical situations. For example, because individuals with disabilities or low incomes also change their preferences and aspirations to suit their circumstances, they often evaluate their lives at normal levels. As a result, their subjective well-being levels are the same as (or higher than) those of a healthier or wealthier person (Loewenstein & Ubel, 2008; Oswald & Powdthavee, 2008). Our analysis also suggests that an adaptive preference problem exists in our survey.

Following Diener et al. (1985) and OECD (2013), our survey asked five questions on subjective well-being. Respondents were asked about their degree of agreement with the five statements on a scale of 1–7, where 1 meant “strongly disagree,” and 7 meant “strongly agree.” The five-question statements were as follows:

• 1. “In most ways, my life is close to my ideal.”

• 2. “The conditions of my life are excellent.”

• 3. “I am satisfied with my life.”

• 4. “So far, I have gotten the important things I want in life.”

• 5. “If I could live my life over, I would change almost nothing.”

Using the above standard measure for life satisfaction, we summed up the five scores (Fig. 4). While the distribution of life satisfaction was skewed to the right, the scores showed sufficient variation. For example, the highest score for life satisfaction was 30 on the histogram. This suggests that most people are satisfied with their lives, regardless of their actual living standards (i.e., the adaptive preference problem seems to occur).

Fig. 4

Distribution of life satisfaction

To investigate the factors affecting life satisfaction, we used an ordered logit model to estimate the correlation between life satisfaction and various factors.²¹ The dependent variable is life satisfaction, and the explanatory variables are a log of income; components of MPI such as subjective health index, years of schooling, living standards, employment, and environment²²; other health-related variables such as nutrition, disability score, fatigue, pain, and mental conditions; social capital such as the number of friends, relationship with family, and relationship with neighbors²³; demographic variables such as gender, age, and square of age; and dummy variables for caste groups, religious affiliation,²⁴ and assembly constituencies. The nutrition and disability scores were used as alternative measures for the health dimension of the Alkire-Foster MPI, and the detailed definitions are explained in footnote 18. Additionally, we used fatigue, pain, and mental conditions to control health conditions. For caste groups, scheduled castes (SCs), scheduled tribes (STs), and other backward classes (OBCs) were used.²⁵ Others represented higher-caste individuals. To explore the different functional forms of life satisfaction between social categories such as castes and religions, we investigated the specifications between SCs and non-SC individuals separately.

Table 4 presents the results. As shown in Column 1, the coefficients of individual income and living standards representing material wealth are statistically insignificant. Additionally, the coefficient of years of schooling was statistically insignificant. On the other hand, the coefficients of subjective health index, environment, and relationship with family are larger than other coefficients and are statistically significant.

Table 4 Life-satisfaction regression

Columns 2 and 3 display the results for SCs and non-SCs, respectively. For example, while the coefficient of individual income is positive and statistically significant among SCs, it is small and statistically insignificant among non-SCs. Additionally, while the coefficient of the relationship with family is insignificant among SCs, it is significant among non-SCs. There are other different results, such as environment, age, and disability. These different coefficients between SCs and non-SCs suggest that functional forms of life satisfaction differ among social groups.

As shown in previous studies, our regression results are consistent with the claim that non-monetary factors are important for subjective well-being. Additionally, because many individuals are materially poor but sufficiently satisfied, there seems to be an adaptive preference problem.

2.5 Equivalent Income

The fourth well-being measure is based on the equivalent income approach proposed by Fleurbaey (2005) and many studies (Fleurbaey, 2007; Fleurbaey et al., 2013; Decancq and Neumann, 2016; Decancq & Schokkaert, 2016; Decancq et al., 2015a, 2015b, 2017). A person’s equivalent income is defined as an income level that would make the person indifferent between her actual situation and the hypothetical reference situation, where she would be at the reference values for all non-income dimensions. This approach satisfies the individual preference principle but violates the dominance principle. Although this approach has many problems, it is considered acceptable for measuring individual well-being.²⁶

Two practical methods for estimating equivalent income have been proposed in the literature. One method is to estimate the WTP for a hypothetical reference situation by using the contingent valuation method to calculate the equivalent income. For example, Fleurbaey et al. (2013) asked respondents the highest amount they would pay to be in perfect health for one year and estimated their equivalent incomes. Another method is to estimate equivalent income using a regression of life satisfaction. For example, Decancq et al. (2015a) estimated equivalent income under the highest reference values for health, education, and housing. This study follows the latter strategy for estimating equivalent incomes since the contingent valuation method has many disadvantages in evaluating hypothetical reference situations of several non-income dimensions.²⁷

Next, we explain our estimation strategy. Following Decancq et al. (2015a, 2016), we calculate the equivalent income using life satisfaction regression. Note that their methodology requires the strong assumption that all individuals in the same group classified by some demographic properties have the same preference relationship. Therefore, we first select the life dimensions to be analyzed to estimate the preference structure using life satisfaction regression. Considering the standard variables of life satisfaction regression and the availability of our data, we selected the following six life dimensions: income, health, education, living standards, employment, and environment. All life dimensions are measured using the same components to calculate the Alkire-Foster MPI in Sect. 2.4. The definition of each variable was the same as that used in the regression analysis of happiness.

We consider the following regression specification:

$${s}_{i}=\alpha +\left(\beta +\varphi \times {d}_{i}\right)ln\left({y}_{i}\right)+{\left(\gamma +\delta \times {d}_{i}\right)}^{\mathrm{^{\prime}}}{x}_{i}+{\theta }^{\mathrm{^{\prime}}}{z}_{i}+{u}_{i},$$

where ${s}_{i}$ is life satisfaction described in Sect. 2.4, ${y}_{i}$ is an individual annual income, ${z}_{i}$ is a vector of variables reflecting individual characteristics, ${x}_{i}$ is a vector of non-income life dimensions, ${d}_{i}$ is a vector of dummy variables reflecting membership in socio-demographic groups, and ${u}_{i}$ is an error term. To consider differences in preference structures among socio-demographic groups, the coefficients on income and other functions are allowed to differ for the robustness check. We consider SCs and Muslims for socio-demographic groups and then separately run the regression using either dummy variable. Regarding ${z}_{i}$, we included dummies for male, age, the square of age, assembly constituencies, castes, and religious groups. This specification was estimated using an ordered logit model.

An equivalent income $({y}_{i}^{*})$ is defined as an income level that would make individual i indifferent between i’s actual situation and the hypothetical reference situation where i would be at maximum levels ($\tilde{x }$) for all non-income life dimensions. As shown in Decancq et al. (2016), we can measure the equivalent income as follows:

$${y}_{i}^{*}={y}_{i}\times exp\left[{\left(\frac{\gamma +\delta \times {d}_{i}}{\beta +\varphi \times {d}_{i}}\right)}^{\mathrm{^{\prime}}}\left({x}_{i}-\tilde{x }\right)\right],$$

where ${y}_{i}^{*}$ is the equivalent income and $\tilde{x }$ are the maximum levels of ${x}_{i}$.

Table 5 presents the regression results.²⁸ The results were almost the same as those of the life satisfaction regression, except for individual income. While the coefficient of individual income in Table 4 is statistically insignificant, that in Column 1 of Table 5 is positive and significant. This may be because the relationship with family matters in subjective well-being estimations. If a person with a high income tends to be highly satisfied with better social relations, then adding social relations as a variable will underestimate the effect of income. On the contrary, if better social relations tend to yield high incomes that directly contribute to improving life satisfaction, ignoring social relations will overestimate the effect of income due to omitted variable bias. Our results suggest that it is difficult to select control variables to estimate the effects of individual income on life satisfaction. Additionally, the coefficient of education is quite small and statistically insignificant since education has little impact on life satisfaction. Thus, education is rarely considered in the equivalent income, which causes quite different results in identifying the poor between the equivalent income and the Alkire-Foster MPI approaches.

Table 5 Equivalent-income regression

3 Comparison of Well-Being Measures

3.1 Comparisons to Income/Expenditure

Income/expenditure is the most popular measure to identify poor individuals. Here, we compare income/expenditure with other measures, such as the Alkire-Foster MPI, life satisfaction, and equivalent income. Figure 5 shows the relationship between the AF-MPI-5D and individual income/expenditure. As the figure shows, some people with low incomes or expenditures have low deprivation scores and are not poor in the AF-MPI-5D. On the contrary, people with an MPI score of 0.4 have various incomes widely distributed from 0 to 300. This implies that income is only one aspect of poverty and that we should identify the poor by considering other non-income dimensions.

Fig. 5

Individual expenditure and AF-MPI-5D

Figure 6 shows the relationship between life satisfaction and individual income/expenditure. While people with high incomes are not necessarily highly satisfied with their lives, people with low incomes are often highly satisfied, which suggests an adaptive preference problem in our survey.

Fig. 6

Individual income and life satisfaction

Finally, Fig. 7 shows the relationship between equivalent income and monetary measures (income and expenditure). Owing to the adaptive preference problem, some equivalent income is almost the same as the corresponding actual income over various income levels. Moreover, some individuals with high incomes have very low equivalent incomes, which suggests that there are violations of the dominance principle in many cases. Figure 8 displays scatter plots of income and equivalent income measured by different functional forms of life satisfaction among social groups. Panel A is the same as that on the left-hand side of Fig. 7. In Panel B, we control for the interaction term between SCs and each function, while the interaction term between Muslims and each function is controlled in Panel C. Due to the small estimates of the interaction terms, the three figures are almost similar.

Fig. 7

Individual income and equivalent income

Fig. 8

Individual income and equivalent income when SCs/muslims have different preferences

Since the coefficients of life dimensions are used as weights for calculating the willingness to achieve the reference vector of life dimensions, the relationships between life satisfaction and life dimensions play a major role in estimating one’s equivalent income. However, life satisfaction is not necessarily correlated with meaningful life dimensions that significantly impact human flourishing. As a result, the equivalent income approach fails to satisfy the dominance principle among different groups.²⁹

3.2 Ranking, the Worst-Off, and the Dominance Principle

To compare the interpersonal well-being rankings generated by these measures, we calculated Spearman rank correlation coefficients among seven measures: individual income, individual expenditure, AF-MPI-3D, AF-MPI-4D, AF-MPI-5D, life satisfaction, and equivalent income.³⁰ Table 6 presents the results. The correlations between MPI and individual income and expenditure ranged from 0.10 to 0.22. On the other hand, the correlation between MPI and equivalent income is approximately 0.6. This is partly because equivalent income is estimated using the same MPI dimensions. Life satisfaction was also used to calculate the equivalent income. However, the correlation between life satisfaction and equivalent income is lower than that between the MPI and equivalent income. Additionally, life satisfaction is weakly correlated with other measures because it is influenced by various non-income factors other than material assets and education.

Table 6 Pairwise rank correlation coefficients between well-being measures

Next, let us consider and compare the bottom 10% of respondents identified by different measures. Table 7 summarizes the means of the basic variables. The average income and years of schooling differed among the measures. The bottom 10% identified by MPI or life satisfaction included individuals with high income. In the AF-MPI-3D and AF-MPI-4D results, the average number of years of schooling was less than one year. This is because both the AF-MPI-3D and AF-MPI-4D have a heavier weight of education compared to the other measures. As shown in Table 5, subjective health had a larger impact on life satisfaction than the other variables. Consequently, a person with low subjective health tends to be identified as poor using the equivalent income approach. Additionally, because education is less correlated with life satisfaction, the equivalent income approach implies that individuals with lower education levels are not included in the worst-off group. As the effects of individual characteristics strongly depend on the specifications and definitions of well-being measures, the bottom 10% groups are quite different among the measures. Our results showed that the bottom 10% group identified by AF-MPI-3D had the largest share of women.

Table 7 Basic statistics of the bottom 10% according to different well-being measures

Finally, we consider the violations of the dominance principle. The dominance principle requires interpersonal comparisons of well-being to be consistent with resource dominance relationships. That is, individual i’s well-being is better than that of individual j whenever i’s relevant live dimensions are greater than those of j. Generally, MPI satisfies the dominance principle on the set of poor individuals,³¹ but the other measures violate it. Panel A of Table 8 shows how often these measures cannot satisfy the dominance principle. We calculate the violation ratios of the dominance principle according to several life-dimension scenarios. Generally, there are many violations of the dominance principle for each measure. Violations of the dominance principle tend to increase as the number of life dimensions is reduced, or personal attributes are increased in the equivalent income approach. Additionally, when education, which has a low correlation with life satisfaction, is included in the life dimensions, violations of the dominance principle increase. As shown in Panel A of Table 8, all measures substantially violate the dominance principle, meaning that disadvantaged individuals are often judged to be better than wealthier individuals in some life dimensions. Moreover, Panel B of Table 8 shows that the situation is the same with the bottom 10%. This invokes serious problems in policy intervention since it sometimes recommends antiegalitarian transfers from a disadvantaged to an advantaged group.

Table 8 The share of cases where well-being measures violate the dominance principle

Generally, the equivalent income approach is strongly affected by life dimensions that greatly impact life satisfaction.³² Moreover, individuals deprived of these dimensions are judged to be disadvantaged in interpersonal comparisons of equivalent incomes. The Alkire-Foster MPI has a fixed weight; therefore, deprivation in one life dimension disadvantages interpersonal comparisons of the MPI by that weight. If the MPI weights were determined using statistical methods such as life satisfaction regression and principal component analysis, a large part of the differences between the MPI and equivalent income approaches might disappear.³³

Although many life dimensions seem important for well-being, they do not necessarily significantly affect life satisfaction. Moreover, the relationship between life dimensions and life satisfaction could be very weak because of the aspiration/hedonic treadmill problem. In this study, we have been able to illustrate that education is such a factor and that there are still major challenges in using well-being measures to estimate equivalent incomes and determine endogenous weights.³⁴ We must develop a measure of life satisfaction that can adequately reflect a cognitive evaluation of a good life based on appropriate life dimensions. We also need to consider developing a methodology for better preference-based interpersonal comparisons.

4 Conclusions

As the indexing dilemma shows in the context of individual well-being measurement, there is a sharp conflict between the individual preference principle (requirement of dignity as a self-determinant) and the dominance principle (requirement consistent with the dignity view as sufficient basic capabilities). While philosophical considerations are important in determining which requirement of human dignity should be respected, we empirically analyze the extent to which key well-being measures conflict with the two principles. This study conducts a field survey in and around the slums of Delhi to consistently compare four leading well-being measures: individual income/expenditure, the Alkire-Foster multidimensional poverty index, happiness as life satisfaction, and equivalent income. As Sen (1985) suggests, regarding the problem of adaptive preference in India, we find that lower caste or Muslim respondents tend to have high life satisfaction. Therefore, following the subjective well-being approach based on life satisfaction, even lower-caste respondents with health problems or lower living standards are evaluated as enjoying good lives. On the other hand, due to this adaptive preference problem, the equivalent income approach based on life satisfaction regression fails to grasp various important aspects of well-being, such as education and health. Additionally, this study confirms that the MPI approach could reflect various deprivation levels of the poor while strongly depending on the functional form of MPIs. From our survey data, the correlation between MPI and individual expenditure is relatively low among our measures, suggesting that the MPI approach can be a complementary and useful tool for measuring individual well-being. Moreover, our results show that violations of the dominance principle are not rare, following both subjective well-being and equivalent income approaches. These empirical results suggest that a more sophisticated strategy is needed in the estimation methods and theoretical analysis of well-being measures to take human dignity and decent lives seriously.

The remaining issues of this study are as follows. First, our sample is small and limited because we focused on slums in India. Additionally, our survey data did not include details of household information, especially for children. More comprehensive data are required to further investigate the properties of various well-being measures. In particular, we should reexamine how often violations of the dominance principle occur in equivalent income and subjective well-being approaches by testing them with a large dataset.

Second, we cannot sufficiently reflect the differences among individual preferences to calculate equivalent income in the previous estimation method. Although the equivalent income approach emphasizes the importance of reflecting individual preferences for interpersonal comparisons of well-being, the previous method strongly assumes that individual preference structures are the same for similar demographic groups. Thus, an improvement in measuring equivalent incomes is needed, and we must consider the heterogeneity of preferences within similar demographic groups.

Third, among the individual well-being measures for evaluating human dignity and decent lives, a consensus-based method proposed by Sakamoto (2018) is consistent with a unanimous judgment and satisfies the dominance principle. Similar to the estimation strategies of the equivalent income approach, we can also use three methods: life-satisfaction-based estimation, inferring willing-to-pay to obtain one’s reference bundle, and revealed-preference-based estimation. If the dominance principle plays a dominant role in measuring individual well-being, these methodologies should be developed.

Finally, a theoretical framework for social evaluation should be developed for aggregating individual well-being. Practical exercises of the social welfare ordering approach require us to develop a class of acceptable individual well-being measures and construct a class of desirable aggregation rules for various contexts of social choice theory. It is especially important to characterize a class of ethically appealing social welfare orderings in uncertainty and risky situations.