Regional Estimates of Multidimensional Poverty in India

This paper estimates and decomposes multidimensional poverty in 82 natural regions in India using unit data from the Indian Human Development Survey (IHDS), 2011–12. Multidimensional poverty is measured in the dimensions of health, education, living standard and household environment using eight indicators and Alkire-Foster methodology. The unique contributions of the paper are inclusion of a direct economic variable (consumption expenditure, work and employment) to quantify the living standard dimension, decomposition of MPI across the dimensions and the indicators, and estimates of multidimensional poverty at the sub-national level. Results indicate that 43% of India's population are multidimensional poor with large regional variations. The average intensity of poverty was 45.5% with a MPI value of 19.3. Six states in India—Bihar, Chhattisgarh, Jharkhand, Madhya Pradesh, Odisha and West Bengal who have a share of 45% of the total population—account for 58% of the multidimensional poor. Across regions, more than 70% of the population are multidimensional poor in the southern region of Chhattisgarh and the Ranchi plateau, while they comprise less than 10% in the regions of Manipur, Mizoram and Chandigarh. The economic poor have a weak association with health and household environment dimensions. The decomposition of MPI indicates that the economic dimension accounts for 22%, the health dimension accounts for 36%, the education dimension accounts for 11% and the household environment accounts for 31% of the deprivation. Based on these analyses, the authors suggest target based interventions in the poor regions to reduce poverty and inequality in India. JEL I J Z


Introduction
During the first four decades of development studies , poverty was primarily measured in money metric form, either from household income or consumption expenditure.
The limitation of money-metric poverty to capture the multiple deprivations of human life and the development of the capability approach (Sen, 1985) led to growing interest to measure poverty in a multidimensional space. The evolution of the human development paradigm in 1990 led to a strong theoretical foundation to measure multidimensional poverty.
The United Nations Development Programme (UNDP) in its annual publications devised a set of composite indices, the Capability Poverty Measure (CPM), the Human Poverty Index 1 (HPI 1) and the Human Poverty Index 2 (HPI 2) to measure multidimensional poverty (UNDP 1996(UNDP , 1997 using aggregate data. The Millennium Declaration has outlined eradication of poverty in its all forms -hunger, ill health, illiteracy. The goals, targets and indicators of the Millennium Development Goals (MDGs) are included in national and local planning (United Nations, 2000). In recent years, the UNDP has disseminated the multidimensional poverty index (MPI) for 104 countries (UNDP, 2010). While the HPI measures poverty at the macro level, the MPI is unique as it identifies individuals (at the micro level) deprived in overlapping multiple dimensions and captures both the extent and intensity of poverty (Alkire and Santos, 2010).
Following the UNDP's work, several researchers have contributed towards measurement of multidimensional poverty (Anand and Sen, 1997;Chiappero-Martinetti, 2000;Bourguignon and Chakravarty, 2003;Gordon et al., 2003;Qizilbash, 2004;Alkire and Foster, 2007;Antony and Rao, 2007;Calvo, 2008;Wagle, 2008;Alkire and Santos, 2010;Alkire and Foster, 2011;Mohanty, 2011). Most of these studies used the dimensions of education, health and standard of living and a few studies included subjective well-being such as fear of facing hardship (Calvo, 2008) in defining multidimensional poverty. However, these studies differed in measuring multidimensional poverty, for instance in fixing the poverty cut-off point of each dimension, weighting the dimensions, deprivation cut-off point in separating the poor from the non-poor and so on. With respect to measurement, some researchers considered the union (poor in any dimension) approach (Bourguignon and Chakravarty, 2003) while others have used the intersection approach (poor in two or more dimension) (Gordon et al., 2003) or relative approach (Wagle, 2008) in defining the poverty line.
While earlier studies used aggregate data, recent studies estimated multidimensional poverty using micro level data. Based on the counting approach, Alkire and Foster (2007;2011) developed a new methodology in estimating multidimensional poverty. Following the Alkire-Foster method, some studies estimated multidimensional poverty (Alkire and Santos, 2010;Coromaldi and Zoli, 2012;Alkire et al., 2013;Batana, 2013;Battiston et al., 2013;Santos, 2013;Yu, 2013). Alkire and Santos (2010) provided estimates of multidimensional poverty for many developing countries using Demographic and Health Survey (DHS) and other large scale survey data. However, their analysis was restricted to three dimensions and had data constraints. Santos (2013)  Consumption expenditure with other indicators was used in measuring multidimensional poverty. The reduction in multidimensional poverty was observed irrespective of indicators weights, deprivation cut-off and identification criterion of the poor. A significant poverty reduction was found due to reduction in the proportion of poor accompanied by the intensity of poverty among those who were less intense poor. Batana (2013) measured multidimensional poverty among the women in Sub-Saharan countries using four dimensions -assets, health, schooling and empowerment.  Mohanty (2011;2012), using the unit data from NFHS 3, linked multidimensional poverty with health and health care utilisation. Children belonging to multidimensional poor households are more likely to be deprived of health care and lower survaival. Alkire and Seth (2013b) suggested a new method using binary scoring method, which can be updated periodically, to target BPL households in India.

Aim and Rationale
Though eradication of multidimensional poverty has been at the centre stage of development agenda, there are only a few studies that estimated multidimensional poverty in India. This paper aims at providing estimates of multidimensional poverty at disaggregated level; in the regions of India, and decomposing multidimensional poverty dynamics across dimensions and regions. This is an improvement on existing literature as we have measured multidimensional poverty by including direct economic variables rather than economic proxies, incorporated the missing dimensions of work/employment and household environment, provided estimates for 84 regions of India, and disaggregated across dimensions, indicators and regions.
We put forward the following rationale in support of the study. First, the regions of India are classified baed on agro-climatic conditions and homogenous with respect to economic, social, cultural and demographic variabls. On the otherhand variation in the socio-economic development among regions of India are large. Regional estimates of multidimensional poverty will be helpful in identifying the backward areas for policy intervention. Second, earlier studies in India (Alkire and Seth, 2013a; Mohanty 2011) used economic proxies rather than direct economic variables in measuring living standard and were restricted to three dimensions -health, knowledge and living standard. We have included some of the key missiing dimension such as consumption expenditure, work/employment and household environmental dimensions in estimating multidimensional poverty. Third, for the first time, we provide the estimates of multidimensional poverty fat disaggregated level (for 84 regions of India) and decomposed the MPI by indicators, rural and urban, regions and states to understand the relative contribution of factors in explaining multidimensional poverty.

Data
The Indian Human Development Survey (IHDS), 2004-05, conducted by the University of Maryland and the National Council of Applied Economic Research (NCAER), New Delhi is used for the analyses. The IHDS survey interviewed 41554 households and covered 215754 individuals from 1503 villages and 971 urban blocks of India. The advantage of using the IHDS survey in estimating multidimensional poverty is that it provides comprehensive information on key dimensions of income, consumption expenditure, health, wealth and work/employment. It provides comprehensive information on income, consumption expenditure, employment, education, fertility, reproductive health, child health, morbidities, gender relations, social capital and cognitive development of children. The details of the survey design, sampling instrument, variables and constructed variables, and various codes used are available in the national report (Desai et al., 2008). Households with missing information were small and we have excluded missing values from analyses.

Dimensions and Indicators
In measuring multidimensional poverty, five dimensions have been selected, namely health, education, economic, work/employment and household environment. These five dimensions comprise a total of ten indicators. The description of dimensions, indicators and the weight to each indicator is shown in Table 1 (GOI, 2011). With respect to work/employment two indicators, occupation and employment are used. A household is said to be deprived in occupation if the household's annual per capita income is less than 5000 rupees and, either the household belongs to labour class households or low paid non-farm business or has low land holdings with less than 2.5 acres. The three indicators used in the household environment dimension are access to clean drinking water, adequate sanitation and clean cooking fuel.

Measurement of Multidimensional Poverty
We measured the multidimensional poverty index (MPI) using the dual cut-off method based on the counting approach developed by Alkire and Foster (2007;2011). This method is gaing popular and disseminated by UNDP in the Human Development Report (HDR) 2010 (UNDP, 2010). The Alkire and Foster assigns equal weight to each dimension and equal weight to each indicator within each dimension. An individual gets a weighted deprivation score according to his/her number of weighted deprivations. The total weighted deprivation score ranges 0-1 and a household is identified as multidimensional poor if the weighted deprivation score is greater than 0.33, which is one-third of the total weighted deprivation score. To derive multidimensional poverty, the Head count ratio (H) and intensity of poverty The headcount ratio is the proportion of the population who are multidimensional poor. The headcount ratio is computed as:

H=
Where, q is number of multidimensional poor, n is total population. is computed as A= Where, c is the total weighted deprivations experiences by the poor.
The multidimensional poverty index (MPI) is the product of headcount ratio (H) and the intensity of poverty (A). It is also referred as adjusted headcount ratio. The MPI is computed as: MPI= H * A

Decomposition of MPI
We have further decomposed the MPI by its component indicators. The censored headcount ratio is first identified to decompose MPI into each indicator. The censored headcount ratio is defined as the proportion of multidimensional poor deprived in the given indicator to the total population. The contribution of deprivation of a particular indicator is computed as:

Contribution of Indicator i to MPI = * 100
Where w i is the weight of i th indicator and CH i is the censored headcount ratio of i th indicator.
The contribution of each region to overall poverty is computed by using the following formula:

Contribution of region i to MPI = * 100
Where n i is the population of i th region and n is the total population. MPI i is the MPI of i th region.
We prepared state and region maps of multidimensional poverty index using ArcGIS software package (ArcMap 10) to show the spatial variation of multidimensional poverty .

Multidimensional Poverty in the States of India
Multidimensional poverty at the national level was estimated at 45% and it is close to the estimates of Alkire and Seth (49%) (Alkire and Seth, 2013a). The correlation coefficient of our estimates with Alkire-Foster estimates is 0.77. Among the bigger states of India (states with population of more than 10 million), our estimate of multidimensional poverty is maximum in Chhattisgarh (71.3%) followed by Odisha, Bihar, Jharkhand and Uttar Pradesh.
All these states are also marked red in Map 1. It is minimum in the state of Jammu and Kashmir followed by Himachal Pradesh and Punjab. Among smaller states, the variation in multidimensional poverty estimates is large, from 52% in Dadra & Nagar Haveli to less than 5% in Goa.

Poverty Estimates at the Regional Level
The multidimensional poverty index (MPI) is the product of two measures, headcount ratio (H) and intensity of poverty (A). The headcount ratio is the proportion of multidimensional poor to the total population. The intensity of poverty is the average weight of deprivations experienced by the multidimensional poor at a time.

Robustness of the Estimation
Dominance analysis is performed to check the robustness of the estimation of multidimensional poverty across deprivation cut-off (k). The headcount ratio and multidimensional poverty index are estimated using different deprivation cut-off (k) among the bigger states of India. The dominance relations among the states are shown in Figure 1.
Each curve in the figure indicates the poverty level in the states when k is varied. If a curve lies below or above another curve, we can say a dominance relation exists between two states. On the other hand, when two curves cross each other, there is no possibility of dominance. There are many dominance relations between the states as is evident from this

Decomposition of MPI by Dimensions and Component Indicators
Decomposition is an important and useful tool to understand the contribution of each dimension and indicator to multidimensional poverty. At the state and regional level, the decompositions are presented across dimensions and indicators (Table A.

Decomposition of MPI by Regions
Columns 7 and 8 in Table A.1 present the percentage of contribution to MPI and percentage of population among regions respectively. We found that Uttar Pradesh is home to the largest number of multidimensional poor, where 14.7% of the population account for more than 18% of multidimensional poor. This is also true for the states of Bihar, Chhattisgarh, Jharkhand, Madhya Pradesh, Odisha and West Bengal, where the share of poverty is higher than the share of population. These seven Indian states are home to 58% of the multidimensional poor and they account for 45% of the total population. Among the regions, Eastern Uttar Pradesh has the largest share of multidimensional poverty. It is home to more than 9% of the total multidimensional poor, though it has only 7% of the total population. It is also found that the contribution of regions to multidimensional poverty varies within the states. In Maharashtra, the coastal region contributes only 0.5% while it shares 2.2% of the total population. On the other hand, the inland central region contributes 2% while it shares only 1.7% of the total population. This shows how poverty inequality prevails within the states.

Discussion and Conclusion
Multidimensional poverty is a priority research agenda, both nationally and globally. Based on these findings, we suggest that attempts be made to provide estimates at the district level, as the district is the centre of planning and programme implementation in India. We also suggest targeted intervention in backward regions to reduce poverty and inequality, and achieve the Millennium Development Goals in India.