Lower Inequality and Lower Development: A Contradiction Sensitivity Analysis of Jammu and Kashmir

1 Introduction

Of late there has been a rising trend witnessed during the past few decades pertaining to the need to know (more) about the nature and characteristics of households situated in the developing part of the world. As such a remarkable rise in the generation and collection of household datasets pertaining to income and expenditure has been registered. However, most of the studies analyzing poverty have employed per capita income or expenditure as a measure to evaluate and compare the standard of living of people across world. Based on the requirements of this method, the construction of income/expenditure and poverty profiles with a different set of poverty lines attached to each respective country is needed. However, this estimation procedure faces two methodological issues. First, the per capita method of constructing poverty profiles is essentially flawed in nature since it does not take into consideration neither allows for household composition and economies of scale, or simply different equivalence scales. At the same time, there exists a considerable theoretical and empirical literature that recognizes the need of examining the sensitivity of poverty and inequality estimates to equivalence scales. These include Peichl, Pestel, and Schneider (2012) in Germany, Bönke and Schröder (2008) in European countries, Bishop, Luo, and Pan (2006) in China, Meenakshi and Ray (2002) in India, Regier et al. (2019) in Ghana and Newhouse, Becerra, and Evans (2017) in sub-Saharan Africa among many others. The notion behind the adult equivalence scale is that the members of a particular household do not share identical consumption needs. Instead, these needs are subjected to demographic composition and characteristics (such as age and sex). For instance, the consumption requirement; the children of a specific household consume lesser than the adults from the same household. As such, the consumption needs of the former group can be met moderately at lower costs than that of the latter group (Drèze and Srinivasan 1997; Short, Garner, and Johnson 1999). Likewise, the caloric intake required of women happens to be less than that of the male adults (White and Masset 2002). Departures of this kind from simple per capita are important because demographic characteristics differ substantially at various levels especially in developing countries. Households size in the Sub-Continent tends to be larger within the joint families and children live with their parents much longer than in developed countries (Meenakshi and Ray 2002). Identical outcomes are returned in higher poverty estimates for households that have a high proportion of children than adults (Cooke, Hague, and McKay 2016). As such, if the changes in household composition are not taken care of, the poverty estimates might be overestimated and therefore, would not be robust.

One of the standard methods of resolving an issue of this type is by substituting the use of per capita income/expenditure method by the ‘adult equivalent scale’. Instead of per capita, adult equivalent per capita is employed, where weight is given to each household member based on her/his age and gender.^[1] For example, if a household consists of three members – one adult male, one adult female, and one child (aged 14 or below), the unweighted size of the household will be 3. However, if the weight of 1 is given to adult males, 0.7 to adult females, and 0.5 to a child, then the household size will be only 2.2 per adult (male) equivalent.

It is also assumed in the per capita method that all household goods are private, and therefore ignores presence of any scale economies that emanate from joint consumption of many household goods (Deaton 2003). The economies of scale in household consumption relates to the size of households. Commonly, it has been observed that larger household reflect a tendency to be poorer on average as compared to the households of smaller sizes (Kaufman et al. 1997). However, in the presence of economies of scale this may not be true. For instance, in an earlier pioneering study, Lanjouw and Ravallion (1995), allowed for economies of scale and reported lower poverty rates among larger households in Pakistan. Drèze and Srinivasan (1997) found sensitivity of poverty rates to different from the economies of scale in India. Using Family Expenditure Survey of UK, Coulter, Cowell, and Jenkins (1992) report that household scale economies alter both poverty and inequality estimates in consumption as well as in welfare comparisons. The scale economies imply that a larger household is able to achieve higher level of economic well-being than, say, a single-member household in per capita expenditure terms. These benefits generally accrue from sharing a number of fixed-cost components, which includes items like, house rent, and increasing returns in domestic technology as from cooking fuel etc. At the same time, the large households also benefit from bulk discounts, say, on cereals. Therefore, not allowing for scale economies in analysis is determined to exaggerate poverty estimates in larger households as compared to smaller ones. Hence, while undertaking empirical analysis, one needs to adjust for it while assessing welfare levels, as it has a direct bearing on the policy implications.

However, surveying the empirical literature, there is mixed evidence on equivalence scales. Studies including Burkhauser, Smeeding, and Merz (1996); Streak, Yu, and Van der Berg (2009); Visaria (1980) determine that poverty estimates are insensitive to equivalence scales. Short, Garner, and Johnson (1999) and Haughton and Khandker (2009) argue that the results from per capita (PC) are no different from the per capita adult equivalent (PAE). The opposing evidence is provided by Buhmann et al. (1988) by analyzing 10 high-income countries, Éltető and Havasi (2002) for Hungary, Lanjouw (2009) for Brazil, Coulter, Cowell, and Jenkins (1992) for United Kingdom among others have reported substantial impact of equivalence scales on poverty and inequality. Lancaster, Ray, and Valenzuela (1999) on the other hand takes a case of multiple nations and reports sensitivity of inequality and poverty parameters on the choice of scales, though differing across countries. Drèze and Srinivasan (1997) found out that the headcount ratios in India are indifferent to the choice of adult equivalent scales but are sensitive to scale economies; similar to findings of Meenakshi and Ray (2002) who discover poverty levels declining sharply across many states of India.

Lately novel empirical and theoretical constructs focusing on the relation between equivalence scales and poverty, inequality and welfare ranking on ‘within’ or ‘sub-group’ of the population have also been growing. White and Masset (2002) in one of their recent studies have focused on the child poverty, Meenakshi and Ray (2002) on regional aspect and Drèze and Srinivasan (1997) have been analyzing the female (widow) households. Streak, Yu, and Van der Berg (2009) has studied children and report insensitivity of poverty estimates to equivalence scales but altering welfare ranking in many provinces of South Africa. Hunter et al. (2003) highlight the prevalence of higher poverty among indigenous population of Australia in comparison to non-indigenous populace while employing adult equivalent scales. Batana, Bussolo, and Cockburn (2013) also report scale sensitiveness to poverty estimates among different age groups across 73 different countries.

Scale adjustment as a phenomenon also untangles another dilemma which is especially encountered in developing nations: a higher per capita expenditure and subsequently a lower poverty among female-headed households (FHH). In countries like India, it has been observed that FHH despite being a vulnerable lot (see Visaria and Visaria (1985) and Agarwal (1986)) have reported overall lower poverty levels (Fuwa 2000). One of the reasons behind the same is lower households’ sizes which limits there capability to exploit economies of scale, as such, returning underestimated poverty rates. Once consumption expenditure is adjusted for economies of scale it raises poverty estimates as has been aptly reported by Drèze and Srinivasan (1997) and Meenakshi and Ray (2002). Therefore, standard poverty and inequality estimates need to be validated before making any welfare conclusions and comparisons. Consequently, equivalence scales are considered as one of the most important methods to check the sensitivity/robustness of these poverty and inequality estimates.

Second methodological issue pertains to poverty reduction overtime or across groups or whether robust comparisons can be drawn. This is mainly due to issues concerning poverty lines and poverty measures (Atkinson 1987; Foster and Shorrocks 1988; Foster, Greer, and Thorbecke 1984). Firstly, poverty figures are sensitive to poverty lines. Comparison on arbitrary poverty lines is not appropriate, more so when the same poverty line is made to represent both regions who may differ in characteristics. Second, poverty itself depends on the choice of poverty measure; different measures of poverty may not uphold the same ordering across or over time. In order to overcome these issues and to make an unambiguous conclusion about poverty ordering, stochastic dominance approach is employed. This approach compares and ranks the distribution of income or expenditure over the entire/unrestricted or possible/restricted poverty lines and using different indices of poverty. The idea is to ensure that poverty figures or the ranking of distributions hold throughout various ranges of poverty lines and measures.

The current study is a novel attempt to evaluate and analyze how sensitive and robust poverty and inequality estimates are to different choices of equivalence scales, poverty lines and measures in the region of Jammu and Kashmir (J&K, henceforth). J&K is one of the most industrially backward and politically unstable regions in India (Khan, Majeed, and Mushtaq 2021). Despite this fact, it has been one of the top performing regions in terms of low poverty and inequality and high consumption growth, even better than India’s most well-to-do states. From the academic point of view the region of J&K has remained under-researched and unexplored, thus, warranting an isolated analysis (Majeed, Khan, and Mushtaq 2021). The paper extends the existing literature on equivalent scales and contributes to its regional facet. The study controls for different household composition and economies of scale and analyzes how poverty and inequality behaves for different sub-groups like rural-urban, regions, and gender and also explores the question, ‘whether welfare ranking in J&K is altered or not?’.

The paper is structure as: a brief overview of poverty in J&K is presented in Section 2. The data and methodology used to understand the research problem is presented in Section 3. Following the analysis, Section 4 reports the results and discussions. The study is concluded in Section 5.

2 Poverty in J&K Region

Jammu and Kashmir^[2] is an administered part of India. It is predominantly rural in nature with 72.62% of its population residing outside the city domains (Census 2011). Agriculture has been the mainstay of (rural) J&K economy. It plays a significant economic role both in terms of income and employability. About 70% of the skilled and unskilled workforce is involved directly or indirectly in agriculture (Government of Jammu & Kashmir 2016). Within a geographical area of 222.00 thousand square kilometers, J&K hosts diverse agro-climatic variations-cold arid, temperate, intermediate and sub-tropical zones. The region is divided into broad four administrative blocks (22 districts) based on geographic conditions-Mountainous region (Jammu), Jehlum Valley (Kashmir), Outer Hills (Chenab) and Ladakh.^[3] Kashmir region has been grappling with violence and political turmoil since early 1990s. Official figures in J&K put the death toll of the conflict at about 47,000 (MHA 2008). While as NGO’s and other organizations claim the death toll to be at least 70,000 (Imroz et al. 2012). Since then a lot of resources and manpower has gone into controlling and subduing the conflict situation denting most of the developmental needs of masses.

Despite this fact, the region has been exhibiting much better welfare indicators than some well-to-do states of India (Majeed, Mushtaq, and Khan 2022; Owais, Khan, and Majeed 2022). As per the official poverty estimates (Table 1), J&K exhibits lower poverty rates than most of Indian states. From 1973–74 to 2004–05,^[4] there is an almost 87% decline in poverty in J&K which is highest than any other state and national average. The poverty rate in J&K is mere 3.5% and 5.4% for the years 1999–00 and 2004–05 respectively; which is miniscule as compared to national average of 26.1% and 27.5%. Kerala, which is considered most well-to-do state in India in terms of economic indicators, has more than twice number of poor people than rural J&K in these years. However, the poverty estimates are a slightly higher as compared to some states but still more than half percentage lower than the national average.

Based on the historic evidence, one of the reasons that J&K is better off in terms of lower poverty levels is the radical land reforms that were initiated in J&K as early as starting 1940s. Landownership was transferred to poor tenants at a mass scale. Land holdings aggregating to 450 thousand of as many as 9000 landlords were confiscated and transferred to the cultivating peasants (Aslam 1977). With redistribution of land titles, Kashmir is believed to report an absence of landlessness, individual level self-sufficiency of food and hence better nutritional positioning. It has been argued that land reforms have been a major source of prosperity and have better placed an average Kashmiri than an average Indian on measurable developmental indicators (Drabu 2019). J&K is also the highest beneficiary of central government funds. It gets a disproportionate share of central funds out of all special category states in India (Raghavan 2016). In terms of per capita, for the period of 17 years (2000–2016), central funds stand at INR 91,300 as compared to meager INR 4300 in Uttar Pradesh, which is one of the most backward states in India. Anti-poverty measures have also been implanted from time to time which has helped in further reduction of poverty in the region. Despite these facts the region still faces a huge developmental challenge especially in terms of being an industrially backward and politically unstable state.

3 Data and Methodology

The present study is based on secondary level cross-sectional data sets. Data waves from consumption expenditure surveys conducted by National Sample Survey Organization (NSSO, henceforth), Ministry of Statistical and Planning Implementation (MoSPI), Government of India are used as the basic sources of data in the present analysis. The data on NSSO waves is collected and disseminated on a regular basis with a gap of (approx.) five years. The data is predominantly used for official poverty estimates across the Indian states, and as such, for the formulation of the poverty alleviation programs and policies. We employ post 2000 AD waves of data or latest three waves of NSS – 61st (2004–05), 66th (2009–10) and 68th (2011–12). 68th wave was taken earlier than usual five-year gap on account of 66th round be less reliable due to recession at the global level and the drought that was experienced at the national level in many Indian states. However, researchers frequently have been using the 66th round for analysis. We also use the 66th round, though, estimates have to be taken cautiously. The sampling in NSS consumption rounds are based on probability. More than 350 items are surveyed; which includes broad categories of food, fuel, health and education among others. The NSS waves are based on two recent recall periods – Uniform with 30 days recall period for all items, and Mixed recalls with 30 days for high frequency purchased items including food and 365 for low frequency items like consumer durable goods. Mixed recall period was officially employed post-2011, therefore, in the resent study use is made of the mixed recall period for all rounds in order to maintain consistency throughout. Households’ weights (multipliers) given by NSS are also put to use in estimation.

Different weights for adult equivalent (household composition) scales and economies of scale are used in the current analysis. As mentioned in Section 1, the standard methods for poverty and inequality estimation are standard and un-weighted per capita. For a household, HH _i , it is calculated as:

where PE _i refers to the per capita expenditure and TE _i presents the total expenditure and N _i in the reference period is the household size (HH _i ). Subsequently the PE _i is compared to the given poverty line in order to determine if at all the household is poor or not. This way of analysis however gives equal weight to all members of the household. Consequently for adult equivalent scale, equation (1) becomes:

PAE _i represents the per adult equivalent expenditure of the ith household, number of males, females and children in household are represented by M, F and C respectively while a, b and d are respective the weights so employed. Four different weights for households are used in the analysis, (1, 1, 1), (1, 1, 0.6), (1, 0.8, 0.6) and (1, 0.6, 0.4) respectively. The weight of (1, 1, 1) for a, b and d implies that a weight of 1 is given to each males, females and children whereas (1, 0.8, 0.6) implies that 1 is the weight for males, 0.8 if for each female and 0.6 is for each child under study. This is equivalent to the fact that a household member with 1 male, 2 females and 2 children equals 5 standard weight of (1, 1, 1) are given. When weights are changed to (1, 0.8, 0.6), the household size becomes 3.8 (1*1 + 0.8*2 + 0.6*2). Likewise, as we go one decreasing the weight for females and children, the adult equivalent household size subsequently goes further down. Since the denominator becomes small in equation (2) as compared to 1 as weight increase, the adult per capita expenditure rises for the same household; accordingly, poverty estimates become lower. Although these weights differ across studies, it may be useful to check the sensitivity of poverty rates over these four approximate weights employed by a number of researchers including Drèze and Srinivasan (1997) in case of India (at the countrywide level). However, in this case poverty line is normalized for each equivalence scale. The modification is basically taking the poverty line of per-adult equivalent of household (of average composition) with per-capita expenditure equal to the unadjusted poverty line [see Drèze and Srinivasan (1997); Meenakshi and Ray (2002) for details]. This keeps the average composition household below or above poverty line regardless of the choice made about the adult equivalence scales.

Eventually for the economies of scale, equation (1) becomes:

P E i ∗ in equation (3) represents the scale-adjusted per capita expenditure, while θ denotes the extent of economies of scale. The value of θ ranges between 0 and 1; where the non-existence of scale economies is implied by 1. To uphold a certain level of welfare with an additional household a proportionate expenditure is accordingly required. In case the value of θ is 0 it represents a situation where given the total expenditure, welfare remains equivalent/constant for all the households regardless of the size of household (White and Masset 2002). In other words, consumption takes the form of a pure public goods. The values falling between 0 and 1 are more viable in practice. In a study, Weaver and Deolalikar (2004) contends that the economies of scale are not greater than 0.75 for developing countries. In yet another important study on sensitivity analysis of poverty in India, Meenakshi and Ray (2002), using Engel approach, estimates economies of scale to be around 0.85 for the region of J&K. For the current analysis, the values taken are; 0.7, 0.8, 0.9 and 1. Given the values of θ, P E i ∗ is estimated and compared to the poverty line in order to determine the poverty status of the households. However, a normalization rule is followed here, to fix the poverty line (z) for different values of θ as:

In the above equation, m is the average size of the households under study. This rule implies that a household (of an average size) will be poor, regardless of the value of θ, if its per capita consumption expenditure size is less than z(1) (Drèze and Srinivasan 1997). Following this, we present poverty estimates both adjusted and unadjusted to adult equivalent and economies of scale by making use of the headcount ratio’s.^[5]

For inequality estimation, we make use of the Gini Coefficient and the Atkinson index to observe any changes after the adjustments for various groups. Gini index is an extensively used inequality measure. It is grounded in the Lorenz curve, an observed cumulative distribution that compares, (say) income or consumption distribution, with the perfectly equal distribution (Haughton and Khandker 2009). Its value varies between 0 (perfect inequality) and 1 (perfect equality). Atkinson Index is a class of inequality proposed by Atkinson (1970). It apprises us about which part of the distribution (lower or upper) contributes highest to inequality. In other words, its value tells us how much society is prepared to pay in order to have equal income. Atkinson Index A(e) equals

y _EDE in equation (5), is the equally distributed equivalent (EDE, henceforth) income which happens to be the cornerstone of Atkinson index. EDE is the standard of living of households that if achieved will enable the society to reach the same social welfare or level of income as actual distribution/income (Creedy and Sleeman 2005). y _EDE equals;

where ε refers to the subjective inequality aversion parameter in the model. With this parameter, the index can be transformed into a normative index depending on the level of aversion or weight (usually 0, 1 and 2) given to inequality by the society. If ε = 0, y _EDE is going to be equal to average level of income. Greater the value of ε, lower the y _EDE, and higher the value of A(e), higher will the inequality be. In others words, the more the society as a whole has to give up in order to have equal size of cake in terms of income/expenditure. In the present study we attempt to measure this index using different aversion parameters and eventually observe how equivalence scales respond to such index.

On the other hand, Stochastic dominance has several set of orders – first, second and third orders of dominance are mostly employed. Initially, pioneered by Atkinson (1987), who postulated that if dominance of first order is achieved in a given year the over previous for a range of poverty lines, the poverty is said to be decreased unambiguously for such index whose income/expenditure profiles are continuous. Later Foster and Shorrocks (1988) associated the higher orders in stochastic dominance with different poverty indices of their earlier Foster-Greer-Thorbecke (FGT) index (Foster, Greer, and Thorbecke 1984). The advantage of using FGT index is that it suffices all basic four principles/axioms of poverty measure: Anonymity principle, Population Independence principle, Monotonicity principle and Transfer principle (see Ray 1998; Todaro and Smith 2012 for details).

These orders are specified by way of cumulative distribution functions (CDFs). Suppose G(x) and F(x) represent two CDFs. Stochastic dominance of first order of distribution G(x) over distribution F(x) is possible only if

where a(x) = monotone and non-decreasing function and the integrals are to be taken over the entire set of x. Since valuation function is monotone, non-decreasing (more is better), a has an average value that is at least as large in G distribution as in F Distribution. Therefore, distribution G dominates (stochastically) distribution F. Put in a different way^[6]

For all x, meaning CDF of G(x) is at least greater as in F(x); F(x) having more mass in lower segment of distributions (Madden and Smith 2000). If two curves cross, 1st order dominance fails to hold. Therefore, reduction in poverty is not robust.

Stochastically dominance of Second order exists if we following condition is fulfilled

compares the integrals below the CDFs, instead of CDFs themselves. From the perspective of poverty measures, it means the dominance of Second-order will exist when the poverty measure is decreasing (strictly) and weakly convex in incomes/expenditure of the poor. This yields a measure which is sensitive to depth of poverty, i.e. a poverty gap measure (Madden and Smith 2000).

However, if we are still unable to verify the presence of Second-Order dominance, we have to further add restrictions for function a(x) or the poverty indices, and move to stochastically higher order dominance. For instance, if we restraint function a(x) so to the extent that third-derivative is non-negative (a‴ ≥ 0).

We further add not only concavity, as in Second-order, but also with the restriction of its third derivative being non-negative which equals

This is equivalent to confining the poverty index to a measure that puts more weight to poorer households: holding dominance for an index that is sensitive to the severity among poor households. Validation of First-order dominance implies higher dominances a well (Davidson 2006; García-Gómez, Pérez, and Prieto-Alaiz 2019; Madden and Smith 2000). There is no need to check Second and Third orders if First-order is confirmed, however, the vice-versa is not implied.

For testing these orders over time, and see whether poverty has decreases or not, we first deflate the rounds with respective base round. For instance, while ranking of 61st and 66th is concerned the later year is deflated to former by the ratio of official poverty lines as in Araar (2006) and others using similar stochastic dominance approach. Similarly, when comparing 66th and 68th rounds.

4 Results and Discussions

4.1 Equivalence Scales and Impact on Poverty Estimates

The poverty estimates of three years (NSS waves) using different measures are presented in Table 2. As presented in table, poverty in J&K has declined from 12.9% in 2004 (61st wave) to 9.2% in 2009–10 (66th wave). For the time period 2011–12 (68th round) it has again increased by 1.3% points to 10.5%. It can be deduced that the apprehension about the fact that at national level about 66th round pertaining to drought and recession did not appear to have impacted; as poverty rate has declined by 3.7% during the said time period.

Table 2:

Poverty rates in J&K.

Poverty measure	Total (in %)			Rural (in %)			Urban (in %)
	61st	66th	68th	61st	66th	68th	61st	66th	68th
Headcount ratio	12.9	9.2	10.5	14.2	8.1	11.5	8.9	12.7	7.2
Poverty gap ratio	2	1.4	1.7	2	1.2	1.9	1.7	1.9	0.9
Squared poverty gap ratio	0.5	0.4	0.4	0.5	0.3	0.5	0.5	0.4	0.2

1. Source: Author’s Calculation from NSSO Rounds.

Table 3 (in appendix) and Figure 1 reports the outcomes on poverty estimates to different choices of adult equivalence scales for different households’ groups under study. The first column (of headcount ratios) from each round is given a normal weight (1, 1, 1), consequently, it reflects the simple headcount ratio. Marching forth to the right of each round, subsequent low weight is eventually given to the women and children. The main feature that can be identified and deduced from Table 3 is that the poverty rate shrinks in most of the household groups in each round as we move towards lower weight of adult equivalence scales, whereas in a few cases it increases (but) marginally. The households with single member (single HH) observe a continuous increase in poverty rate as lower weights are given to females and children. The poverty rate increases more than 6% points from unadjusted or equal weights (1, 1, 1) to a weight of (1, 0.7, 0.4) in the 61st round. While for the same equivalence, poverty rate increases to 2% and 0.8% in subsequent rounds. Since more than 85% households in the region of J&K are male-headed, it implies adult equivalent weight remain the same. However, the adjusted poverty line (normalized for average composition) increases with the low weights, the result is obviously higher poverty. It is also revealed that the female-headed households do not face worse welfare conditions as argued earlier. Therefore, adult equivalent scale provides a robust estimate that the female-headed households. These households do not face higher incidence of poverty and are a more vulnerable group compared to the male-headed group.

Figure 1:

Adult equivalence and poverty in selected groups. Source: Table 3.

From Table 3 an important insight drawn is, that there are no evidence of rank reversals taking place among comparable groups in any NSS rounds over equivalence scales. The figures of selected groups remain almost flat in nearly all the cases. To prove the said point, for instance, rural households have higher poverty than urban households irrespective of the choice of AE scales. Similarly classifying by regions, household sizes and gender categories pointing to the fact that poverty estimates are insensitive to different AE scales under this classification. So to say, AE does not affect relative welfare positions of the household groups.

The headcount ratios using different economies of scale from 1 to 0.7 are reported in Table 4 and reproduced in Figure 2. The results derived from poverty estimates are more sensitive as there is a continued decline in total headcount ratios for both rural and urban households across all the years under study or the subsequent NSS rounds. This implies that in the absence of controlling for the scale economies, the poverty rates returned are inflated. Similarly for different regions, except in the 61st and 66th round for Jammu region, the poverty rates decline. As argued before, some studies [like, Meenakshi and Ray (2002)] find out that the impact of these scales have the tendency to vary across regions within a state. One of the important groups here is the household size and the characteristic of how far households of different sizes are sensitivity to different households’ scale economies (or the value of θ). From the outcomes of Table 4, it can be well observed that lower household size groups report higher estimates of poverty when θ decreases or there is higher presence of economies of scale. Taking for instance, the case of a single member household (single HH) there is an increase in the headcount ratios from 3.4 to 13.9%, 0.5 to 2.6%, 4.3 to 19.2% in each round of 61st, 66th and 68th respectively. The case on the other hand turns contrary in the case of larger households, like the households with 8 or above members where poverty estimate decreases with higher presence of scale economies. These findings are in line with majority of the studies quoted in Section 1 emphasizing the fact that larger households tend to exploit scale economies well in comparison to the smaller households. The smaller households on the other hand, fail to take advantage of the benefits that accrue from these economies. Since the size of female-headed households is also lower, the poverty rates also increase with lower value of θ except for 68th round. Thus, we find that, on an average, the official poverty estimates are overstated in the absence of a consideration for the economies of scale.

Figure 2:

Economies of scale and poverty in selected groups. Source: Table 4.

The scale economies on the other hand are highly sensitive with respect to the welfare ranking among comparable groups, in light of the household size and gender groups (except in 66th round) As the weightage or presence of scale economies is increased to 0.7, the poverty rate of single member HHs shoots up to 13.9% while concurrently lager households come down to 9.4 (for HH up to 8 members) and 12.2% (for HH above 8 members); making a typical case of ‘rank reversal’. Similarly, female-headed HHs become poorer than male-headed HHs proving the case of higher incidence caused by the disadvantages position of women in exploiting scale economies. The data also provides a robust estimate that poverty has indeed declined in 2009 and increased in 2011–12 using different scale economies in line with AE scales, presented in Table 3.

4.2 Equivalence Scales and Impact on Inequality Estimates

Figure 3 and Table 5 (in appendix) report the inequality ratios and Gini coefficient for all households over the years across J&K. It presents the basic scenario on inequality in the region of J&K and the impact of various equivalence scales. The inequality is found to be on the lower side; whether we analyze proportion of consumption of top 90% to 10% (p90/p10) or p90/p50 or Gini Coefficient. The results so derived are very similar to poverty estimates as far as sensitivities to adult equivalence scales are concerned whether within each or over various NSS rounds. For instance, the inequality is found to be declining in the 66th round and observed increasing again in the 68th round, using both inequality proportion measures. The impact of AE scales is found to remain quite robust to these ranking over the years of study period. The impact of different AE scales either decreases or increases inequality (slightly) but remains closer to original unadjusted estimates through each round. Identical results are provided by Atkinson index as reported in Table 6 for AE scales with different inequality aversion parameter. Though, within each round, inequality increases consistently, as the inequality aversion parameter is amplified. Using different scale economies, inequality estimates also found to be declining in the 02nd round and then estimated to be increasing. However, when more weight is given to presence of economies of scale, it declines consistently within each round.

Figure 3:

Adult equivalence and economies of scale adjusted Gini coefficient and Atkinson index. Source: Tables 5 and 6.

Remarkably, the estimates from Gini coefficient report continuous increase in the inequality over the years which are also robust with the AE and economies of scales outcomes. It can be concluded based on the Gini coefficient that inequality in the region of J&K has increased over time. The use of Atkinson index (Table 7 and Figure 3) also manifests an increase in the consumption inequality over the years. Quite interesting is the fact that the region of Kashmir has shown lowest inequality in all the years; substantiated by both the AE and economies of scale but at the same time it reflects a higher poverty rate as compared to Jammu in all the years under consideration. The female headed HHs represent higher inequality rates. No rank reversal is however found in any household group while employing any equivalence scales in inequality measure.

4.3 Stochastic Dominance and Poverty Ordering

Stochastic Dominance results are reported in Figures 4 and 5. In Panel A of Figure 4, the round 61st and 66th are compared. It is clear that the respective FGT curves cross each other at very levels of poverty line (x-axis). This implies 1st order stochastic dominance does not hold. However, this is for unrestricted dominance or range of poverty lines. For restricted Stochastic Dominance or over a range of potential poverty lines.^[7] The dominance of order one is achieved. In other words, poverty reduction over these rounds, 61st and 66th (2004–05 to 2009–10) is robust. Hence, robust ranking or ordering can be made-, i.e. poverty has been lower in 2009–10 as compared to 2004–05 round ranking can, thus, be made. Second graph in Panel A produces 1st order dominance for 66th and 68th rounds but with convincing rejection – both restricted and unrestricted, where it crosses midway around mean value. Hence no poverty ordering is possible.^[8] Nor robust poverty reduction during the mentioned period has taken place. This is in line with the results of scale equivalence reported above, that poverty has declined in 2009 but it has increased in 2011. This may be the reason that the FGT curves are crossing in the 66th and 68th round. Panel B and C produces 1st order dominance for Kashmir and Jammu regions over the period of 61st–66th round and 66th–68th rounds respectively. The results show that between the period of 61st and 68th, poverty reduction is robust only in Jammu region. While as Kashmir region, the FGT curves crosses around INR 400 and INR 900. Between 66th and 68th round the poverty reduction is robust only in Kashmir region, not in Jammu region (see Panel C).

Figure 4:

Ist order stochastic dominance.

Figure 5:

Ist order stochastic dominance and bottom 50% population.

The other important criteria to judge the welfare of the population is to observe the conditions of the bottom section of people. The same can be traced from Figure 5 which shows poverty ranking or the bottom 50% population the whole as well as two biggest rejoins of Kashmir and Jammu. It follows similar ordering as at J&K level. Jammu’s poverty reduction is robust to different poverty lines and measures (FGT) between 61st and 66th period while as Kashmir’s between 66th and 68th round. This implies that although poverty has declined in both time periods, the reduction is robust only within Jammu between 61st and 66th and Kashmir within 66th−68th round. However, whenever, poverty has declined in any region, it had been pro-poor or the conditions of bottom 50% population has improved. Altogether the analysis suggests Four major results. First, while poverty and inequality are on the lower side in the region, it has still managed to increase steadily over time. Inequality has been validated as a major ethical and socio-economic problem faced by the developing world (Birdsall 2001). It not only inhibits growth but at the same time slows down poverty reduction. Over time, the trust of people declines consistently on the governments. The problem becomes a major concern in the fragile and conflict-affected regions like J&K itself. In totality it signifies a concern that needs to be dealt in a longer-run perspective. Second, while adult equivalence scales do not provide a consistent estimate, and do not cause any welfare rank reversal, the economies of scale provide both consistent lower poverty estimates and welfare implication in specific HH groups like households of different sizes and female headed households. This finding validates a lower poverty level in J&K as compared to the rest of the country. The benefit of this comparative advantage could be drawn only if the inequality in the region could be decreased over time. These two parameters need to go hand-in-hand in one direction in order to influence a longer-run sustainable economic change in the region. The third important evidence found in the study is the validation of the higher inequality rates among the FHH. Being traditional in nature, the developing world has an overall patriarchal trend, where the household is headed by the (elder) men (Beechey 1979). The economic endeavors of the household are taken care of, by the male folk (Fuwa 2000). Similarly, in the region of J&K. Men have been heading the households. The conflict in the region has taken a toll on the male population and as such considerable number of households have been affected. This might have increased the number of female-headed households in the region. Although, no evidence is available for J&K, but around the globe evidence is available that mentions conflict alters traditional gender roles (see Bandarage 2010). However, lack of skill and economic resources have made these HH vulnerable to poverty and inequality. At the same time, FHH being smaller are unable to draw the benefits of economies of scale making them further vulnerable to inequality (O’Connell 2011). Therefore, the FHH in J&K are at an added disadvantage of poverty and inequality. Fourthly, the poverty reduction is robust only during 61st and 66th round, that too under restricted 1st order stochastic dominance. This poverty reduction is led by Jammu region. During the next period of 66th and 68th rounds, no robust poverty reduction with respect to both poverty lines or measures at the J&K level. However, in Kashmir region the welfare in terms of poverty has taken place.

5 Summary and Conclusion

The present study in a novel attempt intending to open the floor for empirical enquiry and to present an overview of the poverty and inequality scenarios in J&K region. Concurrently the study is an effort to analyze the sensitiveness and robustness of these poverty and inequality estimates using different equivalence scales, and over time poverty ordering using stochastic dominance technique. The paper meets this ascertained goal by using the latest available 03 ‘five-year NSS consumption expenditure rounds’, and utilizes different poverty and inequality measures. Geographically J&K is characterized to be having an extreme location, with hills and mountains on all sides. This geographic reality has constrained the J&K economy to be mostly rural over time with 73% people residing in rural areas. Subsequently, agriculture has been the dominant economic endeavor with diverse agro-climatic variations. The region has been witnessing fragility and conflict on the political front. However the region has been performing far better on a number of economic indicators as compared to the rest of the country; including poverty levels for a long time by now (from 1947 till date, changing facets and ways of expression).

The present study using the method of Adult Equivalence and Scale economies attempts to validate and update poverty levels in the region. The results suggest that adult equivalence scales provide mixed results, increasing and decreasing for various weights. The outcomes provide a robust estimate of welfare ranking within the groups and overtime among various other comparison of the household groups. In other words, it can be concluded that the adult equivalence is not so sensitive or provides alternative or counter-intuitive results. Different values of economies of scale, provide more sensitive estimates of poverty – a constant decline in poverty rates as more weight is given to scale economies. In addition to this, it’s significant rank reversals among household sizes and female-headed household takes place when higher economies of scale are controlled for. These points to the counter-intuitive case suggesting that women-headed households face higher incidence of poverty against the ordinary or unadjusted official poverty figures. Similar results are provided by the inequality measures as in both adult equivalence and economies of scale, however, no welfare changes in rankings are observed in any household group. Overall the Gini coefficient and Atkinson index provides a strong case of rise in inequality. All these negative factors weighing down the local economy, the region of J&K still performs.

Further, the results of stochastic dominance technique informs about the robustness of poverty reduction only during 61st–66th round. However, this is reflected only in Jammu region. Further, no robustness, and hence social ethical ordering can be made for 66th and 68th round except in the Kashmir region where the decline in poverty is validated.

The outcomes of the paper suggest that the adult equivalence does not provide a statistical and compelling case; however, economies of scale need to be controlled when poverty and inequality cases are presented. The policy measures need to take care of steadily increasing poverty and inequality in the region of J&K. The study also suggests that the formulation of the anti-poverty schemes (informed from research) for targeting the most vulnerable sections like female-headed households in countering poverty and inequality must be put in place effectively.