Poverty Dynamics in Rural India: An Asset-Based Approach

Abstract

To make directed policy for different categories of identified rural poor in India, the current focus on static income poverty as a measure is insufficient. This article uses the asset-based approach to quantify poverty dynamics for the years 1992, 1998 and 2005, for the entire country and for various states. We find that at all India level 35 per cent of the rural households are asset poor, of which 22 per cent are chronically poor and 13 per cent are transient poor. An additional 9 per cent of rural non-poor households live under the threat of poverty. This article also analyses trends in asset poverty in various states, with policy recommendations based on their asset poverty status.

JEL Classification: I32, I38, C21, C25

Keywords

Poverty Vulnerability Asset Multiple Correspondence Analyses India

1. INTRODUCTION

Historical evidence on the Indian economy points to a reduction in relative poverty since the country’s Independence in 1947, but not in absolute poverty. Two questions arise from this: has poverty been reduced by policy and, more importantly, is the measure of poverty taking into account the definition of poverty correctly? Most often poverty measurement relates to the recent poor through consumer expenditure surveys, however, while this takes into account income of the present poor, it overlooks their future state. In other words, we are shortsighted in understanding the structural reasons that underpin poverty’s persistence.

Poverty monitoring in India since the 1960s has been based mainly on household consumer expenditure surveys done as part of the National Sample Surveys (NSS). Various methods have been used to measure poverty using the NSS (Datt and Ravillion, 2002). The money-metric poverty measure is very useful in classifying the population into poor and non-poor (Foster et al., 1984). However, if over a period of time we want to compare the extent and depth of poverty, namely the vulnerability of the poor, the above measure is static and is not of much use. Even the dynamic income/expenditure measures can only capture stochastic changes like remittances, gifts, lotteries, etc., but not structural changes in people’s lives. Based on these shortcomings, the one-dimensional money-metric approaches have been questioned and alternative multidimensional approaches have been put forward. Multidimensional methods allow researchers to consider various non-monetary aspects in explaining poverty and standard of living. Multidimensional poverty is made up of several factors that constitute poor people’s experience of deprivation, such as poor health, lack of education, low quality of living, poor quality of work, disempowerment and threat of violence.

According to recent literature, poverty is widely conceptualised in the asset space (Carter and May, 2001). An asset-based poverty approach can help identify and understand poverty in different regions and is effective because the varied and complementary sets of assets address the different and interactive causes of poverty. Additionally, the approach allows for interventions that address the particular needs of the poor and focuses on household in the micro context. Reducing risk and vulnerability and fostering resilience have become important concerns in recent poverty reduction programmes. Hence, directed pro-poor policy is as effective as policies targeted at keeping children at school, reforming property rights, increasing access to drinking water or other social benefits, which could be income-enhancing or asset- enhancing. This research will contribute towards understanding how the assets of the poor give a better understanding of their vulnerability. Hopefully, in this understanding, pointed policies can be devised to decrease persistent poverty in the states. Assessment of a household’s vulnerability to poverty is more than justified to understand who is likely to be poor, how poor are they likely to be and why they are vulnerable to poverty. The rest of the article is as follows. Section 2 is a relevant literature review and Section 3 discusses the methodology of the paper. Section 4 highlights the data of our analysis. Section 5 analyses the results and Section 6 presents validation exercises on the vulnerability estimates. Section 7 concludes the article.

2. LITERATURE REVIEW

The most common approach to poverty measurement relies on household expenditure/consumption or income data from a single point in time (Foster et al., 1984). Once the money-metric poverty line is defined, the population can be divided into poor and non-poor and the standard measure of headcount ratio and the standard Foster, Greer and Thorbecke (FGT) measure can be calculated to determine the extent and depth of poverty within an economy (Foster et al., 1984). The Baulch and Hoddinott (2000) report detailed studies of poverty dynamics based on panel data from ten African countries. Updating that effort, Hoddinott (2003) finds that the number of panel studies of African poverty has risen substantially. A common finding across all these studies is that transitory poverty comprises a rather large share of overall poverty. The large share of transitory poverty is based on income or expenditure data, which highlights the inherent stochasticity of flow-based measures of welfare. People seem to be better off in one period compared to another without any significant change in their underlying circumstances, particularly in the stock of their productive assets basically due to random price and yield fluctuations and irregular and stochastic earnings from remittances, gifts and lotteries. Booysen et al. (2008) analyse the trend in poverty in seven African countries using an asset index and find that overall poverty has declined in five of the seven countries. The main finding is that though the accumulation of private assets has increased in African countries, access to public services has deteriorated.

Rutstein and Johnson (2004) discussed the advantages and disadvantages of using a wealth index rather than an income and expenditure measure. Based on demographic and health survey (DHS) data they found that the wealth index explains the degree of difference in health outcomes across various countries. The paper also discusses the importance of the wealth index in relation to the needs of the poor.

Hulme and Shepherd (2003) explain the term ‘chronic poverty’ and explore the concepts of poverty, vulnerability and poverty dynamics. Their paper reviews who is chronically poor, why they stay poor and what policies could reduce chronic poverty. They conclude that it is as important to specify the set of capability deprivations that are used to identify chronic poverty, rather than relying solely on measures for income and consumption expenditure. This is especially so where poverty is persistent as a multidimensional deprivation. As variables for poverty assessment, income and consumption are much more likely to fluctuate over short periods of time than are measures such as literacy or tangible assets. According to the authors, the adaptation of the asset-based analytical framework can be more useful for an in-depth study of chronic poverty. Many studies of poor people find that vulnerability to ill-health is a particular problem (Pryer, 1993). His paper also shows that a common cause of chronic poverty in many parts of the world is the chronic illness of a household’s main income earner. This lowers household human assets and reduces the family’s income drastically, and minimum consumption needs of the family are met initially by selling off the natural and physical assets initially. Later, the family resorts to using their financial savings, taking on even unacceptable debt, pulling their children out of school to enter the labour market and mobilising support from social networks.

In India, the major focus of poverty research has been on the pattern of regional variation and the incidence of rural poverty and its determinants. Datt and Ravallion (2002) showed that India has reduced poverty, but large differences exist across states. Poverty fell during the 1990s but not as the growth rate would have predicted. According to Datt and Ravallion, growth did not occur in states where poverty was highest in accordance with convergence. Certain types of initial inequalities impede the prospects for growth-mediated poverty reduction, such as asset inequality (land) and education. States with low levels of human capital and low farm productivity have lower capacity to reduce poverty in response to growth. This indicates that growth has not been sufficient to reduce poverty in the states, as the causality is shaped by inequalities in human capital and between the rural and urban areas. As noted by Deaton (2004), economic development has been increasingly conceived as poverty ‘reducing’ rather than ‘economic growth orienting’. Mehta (2003) found spatial estimates at various levels of disaggregation, which reflect the convergence of deprivation in multiple dimensions or multidimensional poverty across all India districts. Some states are performing extremely well in terms of indicators such as health, education and infrastructure, but there may be districts within these that are most deprived in the country also. Hence, he concludes, no single indicator can capture the complexities of development in the country particularly for a complex phenomenon like poverty. Mohanty (2011) examined the relationship between multidimensional poverty and child survival in India using National Family Health Survey (NFHS) data. He found that there was higher poverty among female-headed households, large households and households with no education or little education. He also found that the estimated infant mortality rate and the under-five mortality rate are substantially higher among the abject poor (those who are deprived in multiple dimensions) compared to the poor and non-poor across states.

Alkire and Seth (2009) focus on the methodology and various problems associated with identification of the poor and the aggregation problem in India’s 2002 below poverty line (BPL) census survey. The article helps identify the decomposable multidimensional poverty measures. An interesting finding is that multidimensional poverty in Jharkhand, for example, is mainly due to asset deprivation, poor air quality, poor quality of work, nutritional deficits and disempowerment of women. In Gujarat, nutrition deficit is the leading contributor to poverty, followed by deprivations in women’s empowerment and air quality. This methodology, too, shows the superiority of a multidimensional measure over a unidirectional measure both in terms of understanding the nature of poverty and for pointed policy prescription.

Recent literature on poverty has tended to focus on defining poverty in the asset space, and prominent early asset-poverty measures were offered by Oliver and Shapiro (1990) and Sherraden (1991). They define a household as asset-poor if its financial asset value is zero or negative. A review of current asset-based approaches shows there is no consensus on a single analytical framework for its measure. The asset-based approach has generated a wide range of studies that can be classified into three broad categories. The first deals with asset vulnerability (Moser, 1998), which highlights the relationship between vulnerability, risks and asset ownership, identifying not only the risks but also resilience in resisting or recovering from the negative effects of a changing environment (CPRC, 2004; Siegel, 2005; Zimmerman and Carter, 2003). The second category of studies helps explain the asset-based approach to poverty reduction (Adato et al., 2006; Barret and Carter, 2005; Carter and Barrett, 2006; Carter and May, 1999 and 2001; Carter et al., 2007). Asset-based approaches have been developed to address the causes and dynamics of longer-term persistent structural poverty primarily in rural Africa and Asia (Carter et al., 2006; Francisca and David, 2007; Naschold, 2005, 2009; Sahn and Stifel, 2000). This approach helps us understand the root causes of chronic and transient poverty. The third category focuses on community-based asset-building strategies for poverty reduction (Hulme, 2006).

Asset-based poverty evaluation is an important issue in the Indian context and little rigorous work has been done using nationally representative household data. There are some studies which discuss wealth inequality in Indian states and its measurement issues, which include under-representation, under-reporting and mis-valuation of wealth and ways to overcome these by using large samples, appropriate survey techniques and proper price deflators (Jayadev et al., 2007a, 2007b; Subramanian and Jayaraj, 2006). Other studies have found that the wealth gap is strong in the enrollment of children across Indian states (Flimer and Pritchett, 2001). However, these studies do not discuss the role of assets or asset build-up in poverty reduction.

In India, the government has tried to reduce poverty by increasing economic growth which is solely measured by per capita gross domestic product (GDP). However, to improve the effectiveness of measures to reduce poverty, we need to focus on various deprivation measures instead of only on income poverty. This article aims to include a new area of poverty measurement, based on rural households’ access to basic assets. To be poor means not only to have a low income but also to lack assets. An asset-based approach to poverty reduction focuses on assisting the poor to develop their stock of wealth and use it effectively to achieve sustainable improvements in their lives. This index, the ‘asset-based poverty’ index, will help measure poverty in various states both at a point of time and over a period of time. The study also makes a contribution to the literature on poverty by exploring both the chronic and transient components of poverty and vulnerability to poverty between 1992 and 2005. The distinction between chronically poor and transient poor is rarely made in the literature on poverty in India, as most empirical analyses are based either on NSS estimates or on village surveys for specific years (Gaiha, 1989).

The objectives of this study are to analyse trends in poverty in rural India based on the asset-based approach; quantify vulnerability to poverty in the rural areas; and uncover the determinants of poverty in India.

3. METHODOLOGY

3.1 Construction of a Composite Asset Poverty Index

When poverty is conceptualised in a multidimensional framework, it should be measured through the aggregation of the different deprivation variables experienced by individuals/households.The aggregation of variables to construct a multidimensional poverty index can be done in many ways such as principal component analysis (PCA) and multiple correspondence analyses (MCA), among others. The standard PCA can be applied only if all the variables are numeric, that is, the variables are either quantitative or continuous and the relationships between the variables are assumed to be linear (Kamanou, 2005). However, variables in our dataset are categorical and measured at the nominal or ordinal levels. Further, the ordinal variables do not have an origin or unit of measurement and, therefore, the mean, variance and co-variance have no real meaning. As PCA relies on estimating the co-variance matrix, the classical PCA model is not appropriate for our study.

MCA, in contrast, allows us to analyse the pattern of relationships for several categorical dependent variables (Asselin, 2002, 2005). There are several studies which have used MCA scores to generate a composite poverty index (Filmer and Scott, 2008; Moser and Felton, 2007). This article applied MCA to create an asset index for all the states based on data from the DHS for 1992, 1998 and 2005.

The following equation was used to calculate a composite asset index for each household for each state:

C P I_{i} = \frac{1}{2} (I_{i 1} W_{1} + I_{i 2} W_{2} .... .... .... .... + I_{i j} W_{j})

(1)

where CPI_i is the i^th household’s composite poverty indicative score. I_ij is the response of household i to category j and W_j is the weight which we will derive from MCA. K is the total number of primary indicators.

In using the asset indices to consider the evolution of assets over time, it is necessary to construct asset indices that are comparable over time. We have constructed the asset index by using pooled weights obtained from the application of MCA to all three surveys for the common states. The CPI constructed using MCA has a tendency to be negative in populations of the lowest deciles. To obtain the positive asset values required for further analysis, a value equal to the greatest negative value is added to each asset index value, so that the lowest observed values become zero (Asselin, 2002). We have added a small magnitude to make the lowest value non-zero. We use this composite asset index to estimate poverty for each state using the appropriate household survey weights.

To calculate the asset poverty ratio or asset headcount ratio we use the FGT measure. The major challenge in using this measure is an appropriate definition of the poverty line in the Indian context. In the case of the traditional money-metric poverty line, the poverty line is derived from per capita consumption expenditure based on a minimum calorie intake requirement. For the asset index, however, there is no indication of an appropriate asset-poverty line. The poverty line for an individual has been chosen arbitrarily in place of an official one. The official poverty line has been under debate and various alternative poverty lines have been proposed (Deaton and Kozel, 2005; Schreiner, 2006). In 2005, the Planning Commission’s official poverty line showed that the poverty head count ratio (HCR) in India was 27 per cent. However, the World Bank uses USD1.25 and USD2 as its poverty line standards for comparison, which yield HCRs of 41.6 per cent and 75.6 per cent, respectively. The Asian Development Bank’s standard poverty line of USD1.35 estimated the HCR at around 60 per cent (Chen and Ravallion, 2008; Himanshu, 2009). To tackle the uncertainty on the poverty line, the 26th percentile is taken as the poverty line for the purpose of this article; the justification for this could also be that it is close to the average of the Planning Commission’s poverty line (Mohanty and Ram, 2011) and equivalent to the World Bank’s USD1.02 poverty line.

For comparison, we use the common poverty line, constant across time and states. We apply the FGT measure of poverty to the static asset poverty line, say $\bar{A},$ to calculate the asset head count ratio (HCR).

P_{α} = \frac{1}{N} \sum_{i = 1}^{N} I^{\bar{A}} {(\frac{\bar{A} - A_{i}}{\bar{A}})}^{a}

(2)

where A_i is the asset stock of household i and the binary indicator variable $I^{\bar{A}}$ = 1 if A_i_{<
$\bar{A},$
which indicates whether household i’s asset stock is below the static poverty line. Here α = 0 gives the head count measure of the poor.}

3.2 Estimating the Vulnerability of a Household to Poverty

The concepts of poverty and vulnerability are not the same: poverty is an ex-post state while vulnerability is generally ex-anti. To estimate a household’s vulnerability to poverty we need to obtain an estimate of the mean and variance of the household’s asset-holdings over time. This calls for panel data collected over a long period of time, but in India most household surveys are conducted for a single time period and cannot be used for this purpose. In this study, we use the vulnerability to poverty measure proposed by Chaudhuri et al. (2002) developed particularly for cross-sectional data, in which the vulnerability level of a household ‘h’ at time ‘t’ is defined as the probability that the household will be asset-poor in period t + 1.

{\hat{v}}_{h t} = Pr (A_{h, t + 1} < \bar{A}

(3)

where A_ht+1 is the household’s access to assets in period t + 1, X_h includes household characteristics and $\bar{A}$ is the poverty line.

Although we do not know what the household’s level of assets will be next year, it may be possible to arrive at a reasonable estimate by first building a model of the determinants of the household’s assets and then using the model to predict next year’s asset holding.

_ht_h_ht

where, X_h is a vector of observable household characteristics, such as years of education, age and sex of the head of the household, number of household members, number of dependent members (between 5–14 years and over 60 years), etc.

ε_ht is the mean zero disturbance term that measures ‘idiosyncratic’ factors, such as shocks that might affect one household but not another. The variance of this error could vary substantially from one household to another.

If we can estimate this relationship, including the variance of the expected asset holding, we can measure the vulnerability model as given in Chaudhuri et al. (2002) as in(4):

{\hat{v}}_{h t} = Pr (A_{h, t + 1} = {a(X}_{h}, ε_{ht + 1}) < \bar{A} | X_{h,} ε_{ht}

(4)

Next, we estimate the asset determination equation for the household as in(5):

A_{h} = X_{h} {β + ε}_{h}

(5)

The variance of ε_h is allowed to depend on observable household characteristics. Thus,

σ_{ε, h}^{2} = X_{h} θ

(6)

The estimates of β and θ are obtained through a three-step-feasible generalised least-square technique. The estimates of β and θ are used to obtain estimates of expected household’s access to assets (X_h $\hat{β}$ ) and the variance of the household’s access to assets for each household. The estimates of the household’s access to assets and variance of the household’s access to assets are used to form an estimate of the probability that a household, with characteristic X_h, will be poor, that is, the household’s vulnerability level as in(7):

{\hat{v}}_{h} {= Pr (A}_{h} < \bar{A} | X_{h}) = \emptyset (\frac{\bar{A} - X_{h} \hat{β}}{\sqrt{X_{h} \hat{θ}}} \frac{\bar{A} - X_{h} \hat{β}}{\sqrt{X_{h} \hat{θ}}})

(7)

where, ∅ (.) denotes the cumulative density of the standard normal distribution.

By following this estimation process, each household in the sample can be assigned an estimated degree of vulnerability to poverty, that is, the probability of it falling into poverty in the near future. For highly vulnerable households, a threshold of 50 per cent vulnerability to poverty is assigned. The logic behind taking 0.5 as a threshold is that, it is reasonable to say that a household is vulnerable if it faces a 50 per cent or higher probability of falling into poverty in the near future. It also means that if a household is just at the poverty line and faces mean zero shocks, it continues to be both poor and vulnerable in the next period. It also implies that if time goes to zero, then being currently poor and being currently vulnerable to being poor coincide (Prichett et al., 2000).

Using all the estimated results from above, we categorise households under various types of poverty:

The household is chronically poor (CP) if its expected assets value is below the poverty line, i.e., A_h < $\bar{A}$ and E[A_h] < $\bar{A}$ .

The transient poor (TP) are those poor households which have expected asset values above the poverty line, i.e., A_h < $\bar{A}$ and E[A_h] > $\bar{A}$ .

The non-poor households are classified into highly vulnerable non-poor (HVNP) and low-vulnerable non-poor (LVNP). The HVNP are those households for which A_h > $\bar{A}$ , but v ≥ 0.5; the LVNP are households for which A_h > $\bar{A}$ and v < 0.5.

Total vulnerable group (TV) consist of the total poor (CP + TP) and the HVNP.

The idea behind this categorisation is to recognise that the poor and vulnerable categories are not the same—they are distinct groups, even though they may not be mutually exclusive. The total vulnerable group thus includes all those who are currently poor plus those who are currently non-poor but have a relatively strong chance of falling into poverty in the near future.

3.3 Determinants of Poverty

As the poverty outcome can only take four distinct values in our framework, it is necessary to use a discrete choice model to analyse the determinants of poverty and vulnerability. We use a multinomial logit model for this purpose. The dependent variable can take one of four discrete values indicating the poverty status of a household (LVNP, HVNP, TP and CP). The explanatory variables are household size, sex, age, years of education and caste of the head of household, proportion of individuals under 15 or over 59 (dependents) and the proportions of adults in the household who work in the agricultural sector, in the non-agricultural sector and who have livestock and dairy business.

The probability (Pr) that a household ‘h’ is in a particular poverty state ‘j’ is modelled as a function of the explanatory variables X_h given by:

Pr (poverty = j) = \frac{\exp (x_{h} β_{j})}{1 + \sum_{j = 1}^{3} \exp (x_{h} β_{j})} for j = 0, 1, 2, 3

J = 0 (if LVNP), 1 (CP), 2 (TP), and 3 (HVNP);

J = 0 is considered as the base category in the regression based on the above equation.

4. DATA

The article uses secondary information on household assets mainly from DHS data for the years 1992, 1998 and 2005. The DHS in India, known as the National and Family and Health Survey, was first conducted in 1992–93 and the second and the third rounds were conducted in 1998–99 and 2005–06, respectively. Household assets are defined as the stock of financial, physical, human, natural or social resources that can be acquired, developed, improved and transferred across generations (Ford, 2004).

Current poverty-related development studies include tangible and intangible assets, broadly identified as natural, physical, financial, human and social assets. However, in this study we have not incorporated social assets as DHS data do not report these. Natural assets include agricultural land and livestock which help maintain livelihoods of rural people. In this study, physical assets include various types of consumer durables or household amenities and housing quality. Housing is the most important component of physical assets. DHS presents data on quality of housing based on materials used for the construction of walls and roof separately. If both the walls and roofs are made of pucca material, a house is classified as pucca; and if the walls and roof are made of kutcha material the house is classified as kutcha. In all other cases the house is classified as semi-pucca.¹

As a proxy for standard of living we use quality of drinking water facility, toilet facility, type of cooking fuel and possession of various household amenities such as electricity, television, radio, bicycle, watch, fan, water pump and kitchen facility within a household. A financial or productive asset comprises savings, credit, jobs and employment opportunities and non-earned income used to achieve livelihood objectives and invest in new livelihood assets. However, DHS data has limited information on these. In our study, productive assets count as financial assets because they represent a current or potential income stream. In rural states, sewing machines, tractors, threshers and animal-drawn carts are key examples of productive assets.

Human assets include skills, knowledge, labour and capacity to work, which vary according to household size, skill levels, education, leadership potential, health status, etc. Human assets are a prerequisite for using other types of livelihood assets. Education is the main indicator of human assets. In this article, we have considered the education level of the male members, highest average education levels of the women and children (aged 5–14 years) and school enrolment. The body mass index (BMI) directly represents the nutritional state of a household and the health status of the household is measured by women aged 15–49 years with low BMI (a BMI lower than 18.5 is classified as underweight). BMI data are available only for females in the DHS data, which is not sufficient, but can be taken because of the importance of women’s health. The study also includes information on any family member suffering from a severe disease (for example, tuberculosis, heart disease) to arrive at the health status of the household.

5. RESULTS

5.1 Analysis of Progress in Trend in Assets

Before assessing the extent of asset-based poverty in rural India, we report trends in asset ownership between 1992 and 2005 in rural India (Appendix Table A1). The proportion of households with electricity has increased from 30 per cent to 65 per cent in this period. It is interesting to note that although household ownership of radios has declined from 34 per cent in to 28 per cent, the proportion of household with televisions has drastically increased from 8 per cent to 30 per cent in this time period. Also, while the proportion of pucca houses has increased from 7 per cent to 40 per cent, 19 per cent of households still live in kutcha houses in 2005. In the case of drinking water, most people in rural areas depend on public water sources. In 1992, 44 per cent of households had public water facilities which increased to 70 per cent in 2005, but only 12 per cent of the households were able to manage having a piped water facility in 2005, which indicates very little improvement from 7 per cent in1992 to 9 per cent in 1998. Although the proportion of households with a flush toilet increased from 1 per cent in 1992, to 11 per cent in 1998 and 23 per cent in 2005—it is disheartening that in 2005, 72 per cent of the households did not have a toilet facility within their houses.

Importantly, while household access to agricultural land decreased from 64 per cent in 1992 to 58 per cent in 2005, their ownership of livestock increased from 29 per cent to 66 per cent in the same period. Under human assets, school enrollment saw an increase from 34 per cent in 1992, to 65 per cent in 1998 and 86 per cent in 2005. There is also a significant improvement in female education within households: secondary education of women increased from 2 per cent in 1992 to 16 per cent in 2005. Thus the data shows an improvement in household asset allocation from 1992 to 2005, indicating that household well-being has improved.

In the light of the above findings, it would be interesting to see how household well-being helps reduce poverty in the different states.

5.2 Poverty Analysis Using FGT Index

To compare asset poverty across states consistently, all three surveys (1992, 1998 and 2005) are pooled to estimate asset weights by using MCA and constructing a household asset index. The MCA based on the 23 variables (primary indicators as in Appendix Table A2) and 55 categories (as in Appendix Table A2) demonstrates that the first factorial axis explains 76.27 per cent of the observed inertia (that is, the eigen value) while the second axis accounts for only 6.28 per cent of the observed inertia. To construct the CPI for each household, we use the functional form of the CPI expressed in equation 1. The weights (factorial scores on the first axis) attributed to the variable categories are presented in Appendix Table A2. Weights with smaller or negative numbers indicate lower welfare, that is, higher poverty; the larger numbers indicate higher welfare and lower poverty. To use these weights, the monotonicity axiom must be fulfilled, meaning that the CPI must be monotonically increasing for each primary indicator (Asselin, 2002). The axiom means that if a household improves its situation for a given primary variable, then its CPI value increases so that its poverty level decreases (larger values mean less poverty or, equivalently, welfare improvement). The largest positive scores are observed to be associated with goods and services comfort, whose access is limited to well-off households. The better-off the household, the more access it has to these goods and services, which include television, pucca house, piped water facility, flush toilet facility, modern source of cooking fuel such as LPG, sewing machine and literacy of household members. The categories associated with the largest negative scores on the first axis are the most accessible goods and services. The poorer the households, the less they possess such goods and services. These households may lack a bicycle, have no access to safe drinking water or a hygienic toilet, have an illiterate household member and a child not enrolled in school.

Before analysing the poverty index, it is useful to start with the descriptive statistics of the asset index score (presented in Appendix Table A3). Once the asset index is constructed, the poverty line is chosen as the 26th percentile of the pooled distribution of the indices (see Appendix Table A3). FGT measures of poverty (equation 2) are then applied to each state for the years 1992, 1998 and 2005 to calculate the asset poverty ratio for each state and for the three periods (Appendix Table A4).

Table 1

Ranks of States in Terms of Asset Poverty

State	Rank 2005	State	Rank 1998	State	Rank 1992
Jammu & Kashmir	1	Jammu & Kashmir	1	Jammu & Kashmir	1
Kerala	2	Delhi	2	Delhi	2
Goa	3	Goa	3	Goa	3
Delhi	4	Kerala	4	Punjab	4
Punjab	5	Punjab	5	Kerala	5
Rajasthan	6	Himachal Pradesh	6	Himachal Pradesh	6
Himachal Pradesh	7	Gujarat	7	Tamil Nadu	7
Madhya Pradesh	8	Maharashtra	8	Maharashtra	8
Gujarat	9	Rajasthan	9	Rajasthan	9
Karnataka	10	Tamil Nadu	10	West Bengal	10
Haryana	11	Haryana	11	Gujarat	11
Maharashtra	12	Karnataka	12	Haryana	12
Tamil Nadu	13	West Bengal	13	Andhra Pradesh	13
West Bengal	14	Andhra Pradesh	14	Assam	14
Assam	15	Assam	15	Karnataka	15
Andhra Pradesh	16	Madhya Pradesh	16	Madhya Pradesh	16
Bihar	17	Bihar	17	Orissa	17
Uttar Pradesh	18	Uttar Pradesh	18	Bihar	18
Orissa	19	Orissa	19	Uttar Pradesh	19

Source: Authors’ calculation.

It is seen that irrespective of the time period, asset poverty declined in all states except Orissa (Figure 1).

Figure 1

Source: Authors’ calculation.

However, the rankings of the states are different in different periods (Table 1). Jammu and Kashmir, Kerala, Goa, Delhi and Punjab have ranked highest among the 19 states in all three years, 1992, 1998 and 2005. One notable change is the movement of Kerala to the second position from the fifth position. Uttar Pradesh, Bihar, Orissa, Madhya Pradesh and Karnataka ranked at the bottom in terms of asset-based poverty in 1992. However, Madhya Pradesh and Karnataka have shown positive movements over the years and moved to eighth and tenth position, respectively, in 2005. Notable among other better performing states are Gujarat and Rajasthan. Between 1992 and 2005 we, however, find that Tamil Nadu, Maharashtra, West Bengal and Andhra Pradesh have become worse off in terms of asset-based poverty. All other states remain almost the same in their relative rankings.

5.3 Vulnerability to Poverty Measurement of Households

The analysis based on the above static, asset-based poverty, suffers from two conceptual weaknesses. First, like standard flow-based measures, its definition depends on an arbitrary living standard (here the 26th percentile). Second, the analysis based on an arbitrary poverty line does not account for any predictable future changes in the assets of the poor or any predictable changes in future returns to these assets. The analysis based on static asset poverty, therefore, cannot reliably indicate whether structurally poor households are likely to remain so in the foreseeable future, whether they are headed in the right direction, or whether structurally non-poor households can be expected to remain non-poor indefinitely.

To tackle these questions, the article has calculated vulnerability to poverty by using equations(3) to(7). We use these to categorise households as CP, TP, HVNP and LVNP (Appendix Tables A5 to A7). The poverty figures for 2005 indicate that despite the increased effort aimed at reducing poverty over the years, it persists in rural India. Around 35 per cent of the rural households are below the asset-poverty line, while 22 per cent of them probably will remain there for a few more years to come, that is, they are chronic poor. About 13 per cent of rural households are identified as transient poor, some of which may escape poverty in future, while 9 per cent of the rural non-poor households are living under the threat of becoming poor in the future (Appendix Table A7).

Figures 2, 3 and 4 highlight the position of the states in terms of TP, CP and HVNP for 1992, 1998 and 2005. Figure 2 indicates that in Madhya Pradesh and Orissa, TP increased in 2005 over the previous two time periods. The important point is that at the all-India level the transient poor was high in 1998 compared to 1992 and 2005. Also, Haryana, Kerala, Punjab, Rajasthan and West Bengal faced higher transient poverty in 1998, compared to 1992 and 2005. Another finding is that chronic poverty declined in 2005 compared to 1992 in all states except Orissa (Figure 3). Further, the HVNP people increased in 2005 compared to 1992 in Madhya Pradesh and Maharashtra (Figure 4).

Figure 2

Source: Authors’ calculation.

Figure 3

Source: Authors’ calculation.

Figure 4

Source: Authors’ calculation.

For a comprehensible depiction of the Indian states’ position in terms of poverty status, the states are classified as severely chronically poor, severely transient poor and severely high vulnerable non-poor (Table 2) to help prescribe policy measures to address the various aspects of poverty. When a state’s average poverty is above the national average in any of the three poverty indicators, that is, chronically poor, transient poor and high vulnerable non-poor, we categorise it as severely chronically poor, severely transient poor or severely high vulnerable non-poor, respectively. The table shows that Bihar, Orissa, Assam, Uttar Pradesh, Karnataka and Maharashtra are chronically poor for all time periods. The poor here lack basic assets and hence need help building these before they can take up income-generating activity. The major policy implication for these states is that the government needs to focus on building basic requirements for these people. Therefore, priority should be given to building assets through a combination of financial and community programmes, possibly in the form of free quality school education, pucca house construction, free health facility access, etc.

Table 2

Comparative Performance of States in Terms of Poverty Categories

Year	Severely Chronic Poor	Severely Transient Poor	Severely High Vulnerable Non Poor
1992	Bihar, Orissa, Assam, Uttar Pradesh, Madhya Pradesh, West Bengal, Karnataka, Tamil Nadu, Maharashtra	Haryana, Himachal Pradesh, Gujarat, Andhra Pradesh, Rajasthan	Andhra Pradesh, Gujarat, Punjab, Haryana, Himachal Pradesh, Bihar, Orissa
1998	Bihar, Orissa, Assam, Uttar Pradesh, Madhya Pradesh, West Bengal, Karnataka, Tamil Nadu, Maharashtra	Punjab, Haryana, Himachal Pradesh, Gujarat, Andhra Pradesh, Rajasthan	Haryana, Gujarat, Punjab, Himachal Pradesh, Maharashtra, Bihar, Orissa, Andhra Pradesh
2005	Bihar, Orissa, Assam, Uttar Pradesh, Karnataka, Maharashtra	Punjab, Haryana, Himachal Pradesh, Orissa, West Bengal, Madhya Pradesh, Gujarat, Andhra Pradesh, Tamil Nadu, Rajasthan	Gujarat, Punjab, Himachal Pradesh, Bihar, Orissa, Madhya Pradesh, Maharashtra, Andhra Pradesh

Source: Authors’ calculation.

Note: States which are above the national average of chronic poor, transient poor and high vulnerable non-poor are categorised as severely chronically poor, severely transient poor and severely high vulnerable non-poor, respectively.

Although Madhya Pradesh, Tamil Nadu and West Bengal were seen as chronically poor in 1992 and 1998, in 2005 the major problem in these states is transient poverty. The substantial reduction in the severity of chronic poverty in these states is because of initiatives by the local government, either the panchayat or the municipality, to build tube wells and provide pump sets and loans for transforming kutcha houses into semi-pucca or pucca houses. Governments also have made efforts to make primary healthcare centres more effective. Haryana, Himachal Pradesh, Gujarat and Andhra Pradesh face the problem of transient poor in all three time periods. Punjab was free of transient poverty in 1992 but has been facing this issue since 1998. Therefore, there is a good chance that a good proportion of households in these states can come out of poverty in the near future. Greater efforts should be made to help them increase their returns from their activities and diversify their risks. One reason for this form of poverty is that these states are highly agriculture-dependent; agriculture-based rural farm income fluctuates, and this affects consumption patterns. Hence households in these states require support prices as well as diversification of livelihoods to diversify their risk, which could help move the rural population from being transiently poor. Microfinance loans are one way to diversify risk in these states. States like Gujarat and Andhra Pradesh have had deep rural microfinance penetration. A microfinance loan can help build animal husbandry and other small-scale businesses which could have a smoothening effect on income and consumption. It is also noticed that some proportion of households in states like Punjab, Haryana, Andhra Pradesh, Gujarat, Bihar and Orissa are high vulnerable non-poor and are facing the threat of becoming poorer in the near future. Vulnerable households will benefit from a combination of prevention, protection and promotion aid which would give them a more secure base to diversify their activities for higher returns.

5.4 Determinants of Poverty

While hardly any literature seems to be available on the determinants of chronic and transient poverty in India, past research indicates that factors typically contributing to static poverty in India are agricultural production, land ownership, land quality, agricultural wages, etc. (Fan et al., 1999; Himanshu, 2006). In this article we have used the multinomial logistic model to isolate the determinants of chronic and transient poor and the high vulnerable non-poor.

Appendix Table A8 represents the multinomial logit regression results for the determinants of chronic poor, transient poor and high vulnerable non-poor groups of the population, considering the low vulnerable non-poor group as a base period group. Results from the model indicate that there is no association between the gender of the household head and poverty status. Chronic poverty was higher among scheduled caste and other backward caste households and in homes where a larger proportion of adults were engaged in the agricultural sector. In contrast, the higher the proportion of adults engaged in the non-agricultural sector, or in the livestock and dairy business and the greater the years of education of the household head, the lower was chronic poverty.

We also find that the proportion of adults with livestock and dairy business tends to decline with the proportion of transient poor. Although years of education of the household head and proportion of household members who can read and write can significantly reduce the status of poverty, the level of education can also strongly influence chronic poverty. We also notice that if household members are engaged in the agricultural sector, transient poverty increases significantly. The study also finds that households with aged members and a higher proportion of dependent members are likely to be more vulnerable to poverty. Further, high vulnerability has increased significantly among the other backward castes (OBCs). The negative significance of the year dummies indicates that over time there has been a decline in the poverty status of all households.

6. HOW RELIABLE ARE THE VULNERABILITY ESTIMATES?

This section presents a series of validation exercises of the vulnerability estimates presented in the previous section. The interpretation of these results comes from variability in future welfare levels and depends crucially on the assumptions in the model, which by definition cannot be tested. However, it is possible to establish the extent to which cross-sectional data can capture ex-ante vulnerability in specific settings where longitudinal data is available. While these exercises can only test the internal validity of the estimates for the specific country and data under study, positive validation results could hint at the robustness of cross-sectional vulnerability measurement with the same methodology. The following exercise was performed in comparing the predicted levels of vulnerability with future outcomes. More specifically, following Chaudhuri et al. (2002), measures of vulnerability in one period are compared to the actual poverty outcome in the following period.

The findings presented in this section include a series of contributions with respect to the performance of vulnerability as an estimate of expected poverty. The discussion below also focuses on how well estimates actually predict whether a specific household will be poor in future, quantifying misclassifications—poor households classified as not vulnerable in the previous period (b), and non-poor households originally classified as vulnerable (c). Therefore, the total misclassification can be calculated as (b+c)/N, where N is the total population. Such misclassification can be termed Type I and Type II errors. A Type I error is defined as the proportion of currently poor households which were classified as not vulnerable in the previous period. Type II error, on the other hand, defines non-vulnerable households as vulnerable—they were expected to become poor, but did not. From the policy maker’s perspective Type I errors seem more serious than Type II errors. This allows counting overall error at the household level, which consists of the sum of those households estimated as vulnerable but did not become poor and the non-vulnerable households which actually became poor.

The series of exercises presented here, thus, allows an extensive robustness check to the cross-sectional approach to vulnerability. The discussion will determine whether the framework of expected poverty gives a better indicator of the aggregate predictor of poverty (at the country level), or whether it is more efficient in identifying household-specific risks at the micro-level (and is thus more useful in targeting beneficiaries for social programmes). The validation exercise presented here is based on longitudinal data for major Indian states given by the India Human Development Survey (IHDS)² for 1993–94 and 2004–05, where the same 13,000 households were surveyed in both years. Finally, it should also be stressed that the proportion of vulnerable households depends on the arbitrary probability threshold chosen by the researcher. The results showed that at the national (rural) level, 90 per cent of the households are classified correctly. This total consists of 70 per cent of non-poor households and 20 per cent of poor households. The remaining 10 per cent of households were classified wrongly, with 3 per cent classified as non-poor households in 2004–05, but vulnerable in 1993–94 and 7 per cent as poor households in 2004–05 but non-vulnerable in 1993–94. The calculation of these errors indicates that, irrespective of the state, Type I and Type II errors range from 1 per cent to 7 per cent (Appendix Table A9). The findings show that cross-sectional vulnerability estimates identify most households correctly, suggesting that the estimates could provide useful information for analysts and policy makers, but that the results from the methodology need to be complemented with state-specific information. For instance, the vulnerability profiles could help distinguish which of the poor households classified as not vulnerable are only experiencing a temporary poverty spell and which are true classification errors. Moreover, the estimates can benefit greatly from information on overall economic conditions, or on aggregate or group-specific shocks.

7. CONCLUSIONS

Although there is substantial literature on poverty in India, as well as academic and policy discussions, the discourse basically focuses on the static income notion of poverty. There is an urgent need to gain a better understanding of the persistence of poverty and poverty dynamics at the household level in India. This article studies poverty dynamics in rural India by studying household access to basic assets for 1992, 1998 and 2005. Our finding is that around 35 per cent of rural households are below the asset poverty line in India, while 22 per cent of them are chronic poor and 13 per cent are transient poor. However, the real challenge is the 9 per cent of rural non-poor households living under the threat of becoming poor in the future.

Our study shows that irrespective of the time period, asset poverty has declined in all the states, except Orissa. However, the ranking of the states is different in the different periods. Between 1992 and 2005, we find that Tamil Nadu, Maharashtra, West Bengal and Andhra Pradesh have become worse off in terms of asset-based poverty. The study of poverty dynamics shows that while in 1992 and 1998 Tamil Nadu and West Bengal experienced severe chronic poverty, in 2005 their major problem was transient poverty.

The study also finds that Bihar, Orissa, Assam, Uttar Pradesh, Karnataka and Maharashtra are chronically poor across the three time periods while Punjab, Haryana, Himachal Pradesh, Gujarat and Andhra Pradesh people are mostly transiently poor. Another interesting finding is that in Gujarat, Punjab, Himachal Pradesh, Bihar, Orissa, Madhya Pradesh and Maharashtra, non-poor sections of the population are becoming vulnerable, that is, there is a probability they will fall below the poverty line in the near future.

From our multinomial logistic regression we find that households with more dependent or aged members and those headed by people with fewer years of education are more likely to be vulnerable to poverty. The outcomes of this article indicate that to reduce poverty more effectively we need a different subset of policy instruments for the different stages of poverty. For the chronically poor households we need to build assets, whereas for the transient poor households we need strategies to diversify their livelihoods to smoothen consumption. In contrast, vulnerable households tend to have more dependents, non-working members and aged members, so they need greater support from the government in terms of social security or old age pension schemes.

Footnotes

Acknowledgements

Two earlier versions of this article were presented at the following conferences and the authors are grateful for inputs from participants: The 48th Indian Econometrics Society Conference, at Pondicherry University, March 1–3, 2011; and the National Seminar on Demographic Transition and Inclusive Development organised by the International Institute for Population Science (IIPS), Mumbai, at the Indian Statistical Institute, Kolkata, March 15–17, 2012.

APPENDIX

Table A9

Error Types in Classifying Households for the Period 1993–94 and 2004–05 (per cent)

	Type I (Poor Households Estimated as Not Vulnerable)	Type II (Non-poor Households Estimated as Vulnerable)
Andhra Pradesh	7	3
Assam	4.5	5.01
Bihar	3.2	4
Gujarat	1.04	4.01
Haryana	3.01	1.08
Himachal Pradesh	2.5	6.7
Kerala	2.86	3.01
Madhya Pradesh	6	4.3
Maharashtra	5	3.8
Orissa	2.9	5
Punjab	7	6.5
Rajasthan	6.5	6.9
Uttar Pradesh	7	6
West Bengal	4	7

Source: Authors’ calculation from IHDS panel data for India.

Notes: Type I households are the fraction of poor households in 2004–05 which are classified as not vulnerable in 1993–94; Type II households are the fraction of non-poor households in 2004–05 which are classified as vulnerable in 1993–94.

Notes

References

Adato

Michael

Julian

(2006). Exploring poverty traps and social exclusion in South Africa using qualitative and quantitative data. Journal of Development Studies, 42(2), 226–47.

Alkire

Seth

(2009). Measuring multidimensional poverty in India, A new proposal, OPHI Working Paper No. 15.

Asselin

L.M.

(2002). Composite indicator of multidimensional poverty, CECI, June. Retrieved from http://www.pep-net.org/fileadmin/medias/pdf/Multi-Dim-Pov-Doc.pdf

Asselin

L.M.

T. A.

(2005). Multidimensional poverty in Vietnam 1993–1998, according to CBMS indicators. Vietnam Socio-Economic Development Review, 41, 217–252.

Barrett

Michael

(2005). Risk and asset management in the presence of poverty traps: Implications for growth and social protection. In background note prepared for World Bank/DFID Project ‘Linking social protection and growth: assessing empirical evidence, developing the future agenda’, Leuven, Belgium June 23–24.

Baulch

Hoddinott

(2000). Economic mobility and poverty dynamics in developing countries. London, Frank Cass.

Booysen

Michael

V.M.

Gideon

D.R.

(2008). Using an asset index to assess trends in poverty in seven Sub-Saharan African countries. World Development, 36(6), 1113–30.

Carter

M.R.

Little

P.D.

Mogues

Negatu

(2007). Poverty traps and natural disasters in Ethiopia and Honduras. World Development, 35(5), 835–56.

Carter

M.R.

Barrett

(2006). The economics of poverty traps and persistent poverty: An asset based approach. Journal of Development Studies, 42(1), 178–99.

10.

Carter

M.R.

Peter

D.L.

Tewodaj

Workneh

(2007). Poverty traps and natural disasters in Ethiopia and Honduras, World Development, 35(5), 835–56.

11.

Carter

M.R.

May

(1999). Poverty, livelihood and class in rural South Africa. World Development, 27(1), 1–20.

12.

Carter

M.R.

May

(2001). One kind of freedom: The dynamics of poverty in post-apartheid South Africa. World Development, 29(12), 1987–2006.

13.

Chaudhuri

Jalan

Suryahadi

(2002). Assessing household vulnerability to poverty: A methodology and estimates for Indonesia. Columbia University Department of Economics Discussion Paper No. 0102-52. New York: Columbia University.

14.

Chen

Ravallion

(2008). The developing world is poorer than we thought, but no less successful in the fight against poverty. Policy Research Working Paper 4703, Washington DC, World Bank.

15.

Chronic Poverty Research Centre (CPRC). (2004). The chronic poverty report 2004–05, University of Manchester: Institute for Development Policy and Management.

16.

Datt

Ravallion

(2002). Is India’s economic growth leaving the poor behind? Policy Research Working Paper Series 2846, The World Bank. Retrieved from http://wwwwds.worldbank.org/servlet/WDSContentServer/WDSP/IB/2002/07/16/000094946_02060404051087/Rendered/PDF/multi0page.pdf

17.

Deaton

(2004). Measuring poverty, Working Paper 170. Princeton University, Woodrow Wilson School of Public and International Affairs, Research Program in Development Studies. Retrieved from http://rpds.princeton.edu/rpds/papers/pdfs/deaton_povertymeasured.pdf

18.

Deaton

Kozel

(2005). The Great Indian Poverty Debate. Delhi: Macmillan.

19.

Fan

Hazell

Sukhadeo

(1999). Linkages between government spending, growth and poverty in rural India, Research Report No. 110, International Food Policy Research Institute, Washington, DC.

20.

Filmer

Scott

(2008). Assessing the asset index. Policy Research Working Paper 4605, The World Bank.

21.

Flimer

Pritchett

L.H.

(2001). Estimating wealth effects without expenditure data–or tears: An application to educational enrollments in states of India. Demography, 38(1), 115–32.

22.

Ford Foundation. (2004). Building assets to reduce poverty and injustice. New York: Ford Foundation.

23.

Foster

Greer

Thorbecke

(1984). A class of decomposable poverty measures. Econometrica, 52(3), 761–66.

24.

Francisca

David

(2007). Poverty traps and nonlinear income dynamics with measurement error and individual heterogeneity. Journal of Development Studies, 43(6), 1057–83.

25.

Gaiha

(1989). Are the chronically poor also the poorest in India? Development and Change, 20(2), 295–322.

26.

Himanshu. (2006). Agrarian crisis and wage labour: A regional perspective, Indian Journal of Labour Economics, 49(2), 827–46.

27.

Himanshu. (2009). New global poverty estimates: What do these mean? Retrieved from http://www.networkideas.org/featart/jan2009/Himanshu_EPW.pdf

28.

Hoddinott

(2003). Pathways from poverty in sub-Saharan Africa, International Food Policy Research Institute, Working Paper. Retrieved from http://www.eldis.org/vfile/upload/1/document/0708/DOC15179.pdf

29.

Hulme

(2006). Asset-based approaches to poverty reduction in Bangladesh, and BRAC’s Ultra poor Programme. Retrieved from http://www.brookings.edu/global/assets06/conference.htm

30.

Hulme

Shepherd

(2003). Conceptualizing chronic poverty. World Development, 31(3), 403–23.

31.

Jayadev

Motiram

Vakulabharanam

(2007a). Patterns of wealth disparities in India during the era of liberalization. Economic and Political Weekly, 42(39), 3853–63.

32.

Jayadev

Motiram

Vakulabharanam

(2007b). Imagined problems in computing wealth inequalities. Economic and Political Weekly, 42(51), 69–71.

33.

Kamanou

(2005). Methods of factor analysis for ordinal categorical data and application of poverty data. International Conference, The many dimension of poverty, Brazil. Retrieved from http://www.ipc-undp.org/md-poverty/papers/Gisele_.pdf

34.

Mehta

A.K.

(2003). Multidimensional poverty in India: District level estimates, CPRC-IIPA Working Paper No. 9, Chronic Poverty Research Centre (CPRC), Manchester, UK. Retrieved from http://www.chronicpoverty.org/publications/details/multidimensional-poverty-in-india-district-level-estimates

35.

Mohanty

K.S.

(2011). Multidimensional poverty and child survival in India. PLoS ONE, 6(10). Retrieved from http://www.plosone.org/article/info:doi/10.1371/journal.pone.0026857

36.

Mohanty

K.S.

Ram

(2011). Spatial pattern of poverty reduction and fertility transition in India. Population Review, 50(1), 62–78.

37.

Moser

(1998). The asset vulnerability framework: Reassessing urban poverty reduction strategies. World Development, 26(1), 1–19.

38.

Moser

Felton

(2007). The construction of an asset index: Measuring asset accumulation in Ecuador, CPRC Working Paper 87. Retrieved from http://www.dfid.gov.uk/r4d/PDF/Outputs/ChronicPoverty_RC/87Moser_Felton.pdf

39.

Naschold

(2005). Identifying asset poverty thresholds—New methods with an application to Pakistan and Ethiopia, Ithaca, Cornell University. Retrieved from http://www.dfid.gov.uk/r4d/PDF/Outputs/ChronicPoverty_RC/87Moser_Felton.pdf

40.

Naschold

(2009). Poor stays poor: Household-asset poverty traps in rural semi-arid India, Selected Paper prepared for presentation at the Agricultural & Applied Economics Association 2009 AAEA & ACCI Joint Annual Meeting, Milwaukee, Wisconsin, July 26–29, 2009.

41.

Oliver

Melvin L.

Shapiro

T.M.

(1990). Wealth of a nation: A reassessment of asset inequality in America shows at least one-third of households are asset poor. American Journal of Economics and Sociology, 49, 129–151.

42.

Pritchett

Suryahadi

Sumarto

(2000). Quantifying vulnerability to poverty: A proposed measure applied to Indonesia. Policy Research Working Paper No. 2437. Washington DC: The World Bank.

43.

Pryer

(1993). The impact of adult ill-health on household income and nutrition in Khulna, Bangladesh. Environment and Urbanisation, 5(2), 35–49.

44.

Rutstein

Johnson

(2004). The DHS wealth index in DHS comparative report. ORC Macro. Retrieved from http://www.measuredhs.com/pubs/pdf/CR6/CR6.pdf

45.

Schreiner

(2006). Is one simple poverty scorecard enough for India? Retrieved from http://www.microfinance.com/English/Papers/Scoring_Poverty_India_Segments.pdf

46.

Sherraden

(1991). Assets and the poor. A new American welfare policy, Armonk. NY: M.E. Sharpe.

47.

Siegel

(2005). Using an asset-based approach to identify drivers of sustainable rural growth and poverty reduction in Central America: A conceptual framework. World Bank ESSD Policy Research Working Paper 3475. Washington, DC: World Bank.

48.

Sahn

D.E.

Stifel

D.C.

(2000). Poverty comparisons over time and across countries. World Development, 28(12), 2123–55.

49.

Subramanian

Jayaraj

(2006). The distribution of household wealth in India. UNU WIDER Research Paper No. 116.

50.

Zimmerman

F.J.

Carter

Michael R.

(2003). Asset smoothing, consumption smoothing and the reproduction of inequality under risk and subsistence constraints. Journal of Development Economics, 71(2), 233–60.