special article

Towards New Poverty Lines for India

Himanshu

T
This paper presents the result of an exercise prepared his paper amplifies on a suggestion made to the committee
for the Planning Commission’s Expert Group to Review constituted by the Planning Commission to review the
e­xisting methodology for official poverty estimation in
the Methodology for Estimation of Poverty to draw up
I­ndia. In brief, the suggestion was to accept the official all-India
new poverty lines and, correspondingly, new poverty urban poverty estimate of 25.7% for 2004-05, derive the all-India
estimates based on the National Sample Survey urban poverty line that corresponds to this using the multiple or
consumption data. The exercise begins by accepting the mixed reference period (MRP) rather than the uniform reference
period (URP) distribution, and to recalculate from this modified
official all-India urban poverty estimate of 25.7% for
poverty line new state-wise urban and rural poverty lines that
2004-05, then derives the all-India urban poverty line reflect actual spatial variations in cost of living during 2004-05.
that corresponds to this head count ratio by using the This suggestion is based on six considerations:
multiple rather than uniform reference period First, that in light of unnecessary past controversies on the
matter, it is essential to clarify that poverty in India is measured
distribution from the nss data. It then recalculates, based
purely on the consumption dimension and that all other dimen-
on this modified poverty line, new state-wise urban and sions, including calorie norms on which present poverty lines
rural poverty lines that reflect spatial variations in the were originally constructed, are incidental and only of historical
cost of living in 2004-05. significance.
Second, that once it is agreed that what is being measured is
The resulting estimates of the incidence of rural poverty
consumption poverty, a basic requirement for valid spatial or
show a head count ratio of 41.8% for 2004-05 as against inter-temporal comparison of this is that poverty lines used
the official estimate of 28.3%. The estimates reveal across space and time should represent equivalent purchasing
much larger rural-urban differences but less power parity (PPP) at whatever reference consumption level is
taken to be the cut-off for basic minimum needs. In particular,
concentration of either rural or urban poverty in a few
once this reference cut-off is chosen, this should apply equally
states. Although the new poverty lines preserve the and without discrimination to all locations, rural and urban,
official estimate of all-India urban poverty in 2004-05, with the only location-specific adjustment being for differences
there are significant changes at the state level. in cost of living.
Third, in order to maintain continuity of presently accepted
notions of the minimum standard of living required to avoid
a­bsolute poverty, it is desirable that the reference cut-off be
a­nchored to some aspect of present practice. There are two pos-
sible references, the present all-India urban and rural poverty
lines, of which only one can be chosen since the other must be
determined by actual cost of living differences.
This paper, which was presented at an workshop convened by the
Planning Commission’s Expert Group to Review the Methodology for
Fourth, official rural poverty estimates are widely perceived to
Estimation of Poverty, is in turn a modified version of an original paper be too low and no longer conforming to acceptable basic needs.
by Himanshu, Rinku Murgai and Abhijit Sen. Since official urban poverty estimates are less controversial, our
Unlike the original paper, which used median prices for the bottom 60% choice of reference consumption cut-off is the MRP equivalent of
of population, this paper uses median prices for the entire population.
the present official all-India urban poverty line, which leaves
Also the final cost of living indices reported here involve fewer steps
than in the original paper.
measured all-India urban poverty incidence unchanged from its
These changes were made on the suggestion of the expert group and current official estimate.
all results and diagnostics have correspondingly been recalculated. Fifth, the relatively minor matter of choosing the MRP cut-off
Comments by Suresh Tendulkar, R Radhakrishna, C Ravi, Kirit Parikh that gives the same urban poverty rate as official, rather than
and participants of the workshop are gratefully acknowledged.
taking the present official urban poverty line directly, is because
Himanshu (himanshu2@gmail.com) is at the Jawaharlal Nehru the National Sample Survey (NSS) now uses the MRP rather than
University and visiting fellow at Centre de Sciences Humaines, URP in most of its surveys. It is necessary to have poverty lines that
New Delhi.
correspond to MRP distributions.
38 january 2, 2010  vol xlv no 1   EPW   Economic & Political Weekly
 special article

Sixth, although the new poverty lines are not based on any under­estimated the cost of living increase from 1973-74 bench-
particular norm of basic needs, and are outcome of a purely marks. Unlike CPI-IW, the CPI-AL weighting diagram did not in-
t­echnical exercise to calculate cost of living indices relevant around clude e­ducation, medical services or conveyance till 1995-96, and
the present all-India urban poverty line, this choice can be de- on fuel assumed that firewood was available free. Actual pur-
fended normatively. In particular, albeit modest norms of nutrition chase prices of all of these items rose much more than of food,
and of paid-out education and health costs are adequately met at and their weight in expenditure of the rural poor also increased
the all-India level and in most, although not all, states. considerably because of lifestyle change towards urban consump-
tion patterns, inadequate provision of public services and reduced
Cost of Living Indices access to common natural resources. Even for food, as Deaton
This paper reports state-wise rural and urban poverty lines ob- (2008) notes himself, the CPI-AL may have underestimated actual
tained from a technical exercise to calculate cost of living indices price increase quite significantly in the more recent period, much
relevant around this consumption norm. Based mainly on NSS more than CPI-IW.
unit values, this is similar to Deaton (2003) and to earlier work Since original calorie norms are not met, we have compared
used by the 1993 Expert Group of the Planning Commission. actual food, education and health expenditures at reference
However, these cover more items and treat interstate and rural- p­overty lines to some other albeit rock-bottom criteria. It turns
urban price differences somewhat differently. The new poverty out that average food expenditure near the official urban
lines correlate very well with Deaton’s across states but show poverty line (and its PPP equivalent in rural) can buy an Indian
more urban-rural difference, mainly on account of wider cover- Council of Medical Research (ICMR) norm diet and also exceeds
age of non-food items. Nonetheless, we too find that existing the cut-off on 61st round distribution of persons by food expend-
o­fficial poverty lines exaggerate urban-rural price difference, iture that corresponds to National Family Health Survey-3 (NFHS-3)
spuriously place rural poverty as less than urban in as many as mal­nutrition (average of low body mass index (BMI) adults and
nine major states, and also correlate poorly with interstate varia- under­weight children). Similarly, in both urban and rural areas,
tions in cost of living. But we differ with Deaton in his suggesting actual education and health expenditures at the urban reference
upward revision of existing rural poverty lines and hence of esti- are a­dequate to send all 5-14 age children to school and meet
mated rural poverty – from 28.3% to 41.8%. normal (but not catastrophic) medical contingencies at what
This is because of the choice of the present official all-India NSS reports are age-specific probabilities of disease onset and
u­rban poverty estimate as reference and the fact that urban-rural median costs of schooling and treatment. Actual food, education
price differences are actually much less than those implicit in and medical e­xpenditure at the present official all-India rural
o­fficial poverty lines. If instead, like Deaton, we had used official poverty line (and its PPP equivalent in urban areas) do not sat-
all-India rural poverty as the reference, existing urban poverty isfy these criteria. This normative validation of the suggested
estimates would be revised down while leaving all-India rural reference is likely to be fairly robust to any NSS underestimation
poverty levels unchanged. Consequently, although pegging new that lowers both actual consumption and criteria cut-offs by
poverty lines as PPP equivalents of an existing norm avoids open- similar magnitudes.
ing a Pandora’s Box of reconsidering the cost of basic minimum However, it must be stressed that these normative criteria
needs, a normative choice different from Deaton is made here in are not the raison d’être for the suggested poverty lines. As
using existing official estimates of urban, rather than rural, made clear above, this is a purely technical exercise to obtain
p­overty as the reference. new state-specific rural and urban poverty lines that reflect
As mentioned above, the urban reference was chosen since this current spatial differentials in cost of living and yet remain
is the less controversial of two standard of living norms below rooted to a present measure of absolute consumption poverty.
which people are currently deemed officially poor. Official rural The normative criteria are used only to discriminate between
poverty estimates are widely perceived to be too low and no official all-­India urban and rural poverty lines as valid starting
longer conforming to acceptable basic needs. Critics have f­ocused points for our calculations, since only one of these can be refer-
mainly on two aspects: adequacy of poverty lines to ref­lect nutri- ence and the other must be determined by actual spatial price
tion need and validity of procedures to update poverty lines differentials. In particular, these criteria do not enter our calcu-
benchmarked far back in 1973-74. On both these there is less crit- lation of new p­overty lines; and the choice is only relative among
icism of urban poverty lines and estimates. This is justified be- two official standard of living references, both of which are
cause although both urban and rural official poverty estimates historical and need not reflect today’s norms. For example, the
are well below the incidence of persons falling short of original reference suggested is less than, although close to, the 2005 PPP
calorie norms, anthropometric estimates of urban malnutrition $1.25/day norm used by the World Bank in its latest world
are close to official urban poverty estimates while independent poverty estimates. Moreover, although we argue for official all-
estimates of rural malnutrition far exceed official rural poverty. India urban poverty line as reference, and use expenditure
Similarly, although outdated price indices have been used to shares corresponding to this in our calculations of spatial cost
update both urban and rural poverty lines, it is the Consumer of living indices, these are quite robust to moderate changes
Price Index-Agricultural Labourers (CPI-AL) rather than the Con- so that poverty lines corresponding to a range of alternative
sumer Price Index-Industrial Workers (CPI-IW) (used for official references can be obtained simply by scalar multiplication. In
rural and urban poverty lines respectively) that has definitely fact, since our main exercise is to derive spatial cost of living
Economic & Political Weekly  EPW   january 2, 2010  vol xlv no 1 39
speciAl article

indices, robustness of these indices is intrinsically more impor- the expert group used calorie norms only as a peg to derive MPCE
tant than the reference standard of living to which these are ap- norms with which to measure consumption poverty. It neither
plied. For this reason, we also subject our estimates to some tests applied calorie norms to obtain 1973-74 state-specific poverty
of plausibility. lines nor suggested their use to track poverty changes over time.
The paper is organised as follows: In Section 1 we discuss the Once all-India poverty lines were fixed using calorie norms for
present official poverty lines, and assess the validity of criticisms 1973-74, all further adjustments were only for variations in cost
made against these. Our main conclusion is that while the of living across states and over time.
all-India urban poverty estimates are fairly robust to these However, the expert group’s cost of living adjustments were
criticisms, the rural estimates are not. In Section 2 we present partial and hybrid. Use of RDA to obtain 1973-74 all-India poverty
our suggested new poverty lines, and the corresponding poverty lines separately for rural and urban meant that these did not
estimates, along with details of the methodology used. In Sec- r­eflect actual rural-urban price differentials. The rural and urban
tion 3 we normatively evaluate our new poverty lines with re- interstate price indices were from different sources and for
spect to actual expenditures on food and on health and educa- d­ifferent years, which made state poverty lines prone to further
tion near these poverty lines. We also subject our new p­overty error. The method to update poverty lines with fixed 1973-74
lines and estimates to some tests of robustness. These involve weights did not allow for changes in consumption patterns. Also,
tests of whether our cost of living indices are really comparing some well-known data gaps were ignored, e g, that the 1960-61
like with like, and of the characteristics of those people whose series of CPI-AL, while including firewood in its weighting dia-
poverty status is changed by the shift to new poverty lines. The gram, treated this as available free. In addition to normative
paper concludes with some suggestions on how these new pov- i­ssues, particularly growing divergence from original calorie
erty lines may be carried forward and backward over time. norms, these weaknesses are source of criticism of existing
p­overty lines.
1 Current Official Poverty Lines Criticism of existing poverty lines has been both normative
The current poverty lines used by the Planning Commission and technical, with the most important normative criticism be-
are based on recommendations of the Expert Group on Estima- ing that existing poverty lines fail to preserve original calorie
tion of Proportion and Number of Poor (1993) chaired by norms. Also, it has been argued that these provide inadequately
D T Lakdawala.1 The approach of this expert group involved four for health and education because it was originally assumed that
major steps. The first was to identify a norm for food expendi- these services would be provided by the state (Dev and Ravi
ture. On this, the expert group relied on Recommended Dietary 2008). Technical criticism has focused mainly on problems of
Allowances (RDA) from the Report of the Task Force on Projec- cost of living adjustment both over time and across space. Al-
tions of Minimum Needs and Effective Consumption Demand though the technical literature has largely bypassed the norma-
(1979): 2,400 calories per day per capita for rural areas and tive content of poverty lines, a major concern expressed is that
2,100 calories per day per capita for urban areas. The second the ratios of official urban to rural poverty lines are unrealisti-
step was to apply these to data from the 28th round (1973-74) NSS cally large in recent years, compared to both the initial 1973-74
Consumption Expenditure survey in order to obtain all-India ratio and ratios obtained from recent NSS unit level data. This
poverty lines. A­ctual all-India urban and rural distributions of means that official poverty lines, which do not directly compare
calorie intake of households by their monthly per capita con- rural and urban prices, may be discriminating against the rural
sumption expenditure (MPCE), including non-food expenditure, poor in the sense that a rural person could be classified as non-
were used to identify MPCE levels corresponding to the respec- poor while an identical urban person consuming the same
tive RDA, and all-India poverty lines were fixed at these levels. b­undle of goods and services is classified as poor.
Thus, all-India poverty lines were obtained separately for rural As far as the original calorie norms are concerned, it is a fact
and urban areas pegged to RDA norms for food expenditure, that there is now a huge difference between the NSS estimates of
with an implicit allowance for non-food expenditure at actual proportion of people with less calorie intake than these norms
consumption levels of people whose food consumption was near (79.8% rural and 63.9% urban in 2004-05) and the official pov-
the respective RDA. By this procedure, the expert group arrived erty headcount (28.3% rural and 25.7% urban). However, this di-
at Rs 49 per capita per month as the all-India rural poverty line vergence was already evident when the expert group had submit-
and Rs 57 per capita per month as the all-India urban poverty ted its report stating that “use of calorie norm in measuring
line for 1973-74. p­overty amounts only to a first order approximation to what may
In the third step, state-wise poverty lines for 1973-74 were cal- be considered to be an acceptable level of minimum need”, and
culated from these all-India poverty lines by applying Fisher deciding against using direct calorie norms to measure poverty.3
I­ndices of state prices relative to all-India, reported by Chatterjee The reasons for this are available in the report.4 But, since recent
and Bhattacharya (1974) for rural areas and Minhas et al (1988) studies continue to highlight this issue, it is important to stress
for urban areas. For the fourth step, i e, updating these state-wise again that calorie norms were used only to peg reference 1973-74
poverty lines, the expert group suggested use of CPI-AL for rural all-India MPCEs from which state-wise poverty lines have been
areas and CPI-IW for urban,2 with components of these reweighted derived, then and subsequently, by applying cost of living adjust-
to actual 1973-74 consumption around the all-India poverty lines ments. What was being measured was consumption poverty, not
and these fixed weights applied uniformly to every state. Thus, calorie deficiency.
40 january 2, 2010  vol xlv no 1  EPW   Economic & Political Weekly
 special article

Beyond Calorie Norms appear to be too low, but the all-India urban poverty count of
In other words, while deficiencies in calorie intake are certainly 25.7% is again defensible.
important, this was not the dimension that either the expert Our main conclusion from this discussion of criticisms of the
group or previous Indian literature had used to measure pov- nutritional adequacy of present official poverty lines is that it is
erty. On the dimension of consumption expenditure that was confusing and also probably misleading to continue to rely on the
chosen, actual calorie intake is immaterial so long as norm original calorie norms.10 Critics are correct in questioning the
intakes are affordable. Poverty lines, even in 1973-74, assured at present official estimates of rural poverty since there is strong
best the affordability of a consumption bundle that contained prima facie evidence to suggest that this underestimates nutri-
food items with norm calorie content and did not classify as poor tional inadequacy, but this evidence also lends some support to
all those with less than norm calorie intake. Whether this re- official estimates of all-India urban poverty. Since calorie intakes
mains true is part of the technical issues concerning subsequent correlate poorly with other indicators, and nutritional adequacy
cost of living adjustment. But as Pronab Sen (2005) showed, al- needs to be judged in a holistic manner, it is probably better to
though actual calorie intake around the poverty line was well move directly to a consumption expenditure norm and judge the
below the norm in 1999-2000, calorie norms could have been nutritional adequacy of this norm against data from nutrition out-
met in most states without reducing non-food expenditures if come indicators. On this basis, the present all-India official urban
those around the poverty line had spent the same amount on poverty line is certainly preferable as the consumption poverty
food but consumed the more cereal-intensive diet of the average cut-off than the present official all-India rural poverty line.
poor. Moreover, recent research shows that a similar conclusion This conclusion is supported by three other normative
continues to hold in u­rban areas, although not in rural, even if c­onsiderations:
intake norms are expressed in terms of culturally palatable
minimum cost diets that adequately meet ICMR norms of all (1) Education: In 2004-05, NSS reports that as against only about
nutrients, not just calories.5 75% of 5-14 year rural children in the official rural poverty line
The implication of the above is that the actual calorie shortfall class attending school, nearly 90% of urban 5-14 year old chil-
now observed at the poverty line may be because of preference dren in the urban official poverty line class were attending
for a more varied and expensive diet and, at least in case of urban school. In terms of expenditure, average actual expenditure on
areas, is not because of lack of affordability. Nonetheless, since education was adequate to send all these children to school at the
adequate nutrition is the most basic minimum need that poverty median cost of schooling per child.
lines should satisfy, mere affordability of norms cannot be a con-
vincing defence of official poverty counts if actual hunger and (2) Health: In terms of expenditure on health in 2004-05, aver-
malnutrition at present were anything like the incidence of peo- age actual expenditure on non-institutional medical care in-
ple with calorie intake below these norms.6 But this is unlikely, curred by those at the urban poverty line class was slightly more
not only because the NSS reports negligible incidence of subjec- than required to meet a probability of disease onset of 9% per
tively felt hunger, but also because the original Task Force calorie month at the median cost of treatment per disease episode as re-
norms were almost certainly too high. For example, FAO’s mini- ported in NSS 60th round. As against this, actual average ex-
mum calorie norm for India is currently 1,770 calories per capita penditure on non-institutional medical care by those around the
per day,7 which is very close to average 61st round calorie intake official rural poverty line was lower than required. For institutional
of those near official poverty lines in both rural and urban areas. medical care, actual expenditure by both the urban and rural
Although the sedentary activity levels assumed are questionable p­overty classes was less than required to meet a 2.5% probability
for rural areas, the fact that actual intake at the cut-off is close to of hospitalisation at the median cost of hospitalisation. However,
the FAO norm does provide a defence at least for the official taking both institutional and non-institutional medical care, a­ctual
all-India urban poverty line. expenditure of the urban poverty line class was adequate while
However, the real lesson from all this is that calorie needs vary that of the rural poverty line class was substantially less required
widely and all-India calorie norms are not the best way to define than to meet the above probabilities at median cost.
minimum nutritional need.8 The best indication of this is that the
incidence of calorie deficiency from any all-India norm, whether (3) Treatment of Similar Households: Considering two groups
original or FAO, correlate very poorly with direct anthropometric of poorest households, those in the casual labour household type
measures of malnutrition from the National Nutrition Monitoring and those where the household head is illiterate, the official u­rban
Bureau (NNMB) and the NFHS. This is true over time (while calo- poverty line classified 58% and 53% of persons in such u­rban
rie intake has declined, nutrition outcome indicators have im- households as poor in 2004-05. However, the official rural pov-
proved); across states (e g, Kerala and Tamil Nadu have both erty line classified as poor only 46%, 34% and 38% respectively
lower calorie intake and lower malnutrition than Bihar and Uttar of those in rural agricultural labour households, non-­agricultural
Pradesh); and also between rural and urban. If incidence of inad- labour households and households with illiterate heads. Clearly,
equate nutrition is judged by the average of low BMI adults and similar household types in urban and rural areas are being classi-
underweight children, this was about 25% in urban areas and fied differently by the two poverty lines. The u­rban poverty lines
40% in rural during 2005-06 (NFHS-3).9 On this basis, the appear more appropriate since it classifies as poor the majority of
2004-05 official all-India rural poverty count of 28.3% does people in what are acknowledged to be the poorest groups. By
Economic & Political Weekly  EPW   january 2, 2010  vol xlv no 1 41
speciAl article

the same token, the present rural official p­overty lines are some- Most of these issues with the existing poverty lines are already
what suspect especially since it is difficult to believe that poverty in the public domain and there have been efforts to correct the
is higher among urban casual labourers than their rural counter- poverty line of these mistakes. Two important efforts in this re-
parts, most of whom are currently classified as non-poor. gard are by Martin Ravallion and Gaurav Datt at the World Bank
and Angus Deaton. However, these corrections differ in how they
Technical Criticisms correct the existing poverty lines. The Ravallion-Datt corrections
These observations lead into the set of technical criticisms that are limited to (1) correction of some distortions created due to the
have primarily focused on the problem with price deflators, at use of interstate price differential for wrong years by the 1993
interstate level, rural-urban and over time. The primary reason expert group, and (2) correction for the index of fuel and fire-
for the criticism has been some obvious anomalies in poverty wood in the CPI-AL using price data from labour bureau. In par-
e­stimates using existing poverty lines. For majority of the states, ticular, Ravallion-Datt follow the expert group in starting with
urban poverty in 2004-05 turns out to be higher than rural pov- the same two different all-India poverty lines for 1973-74 and use
erty. This is the case in some of the major states such as Andhra the same price indices, i e, the CPI-AL and CPI-IW.
Pradesh, Haryana, Karnataka, Kerala, Chhattisgarh, Jammu and
Kashmir, Madhya Pradesh, Maharashtra, Rajasthan and Tamil Deaton Approach
Nadu by URP poverty estimates. Orissa, Uttar Pradesh and Uttara­ Deaton’s approach to correcting the poverty line is, on the other
khand get added to the list of these states if MRP estimates are hand, substantially different from the methodology adopted by
used. That is, among major states only Assam, Bihar, Jharkhand, the expert group. Not only does Deaton use price data from the
Gujarat and West Bengal remain as states where rural poverty is consumption expenditure surveys of NSSO but he also uses Fish-
higher than urban poverty. Even for all-India estimate of poverty er’s index for both the interstate price differential and the rural-
by MRP, rural HCR is 21.8 compared to 21.7 in urban areas. While urban price differential. The notable point with Deaton’s exercise
this in itself may not be problematic, it appears so given the fact is that he uses the rural-urban price differential to correct imbal-
that urban growth per capita has outpaced rural growth per ances in poverty lines across rural and urban areas of each state.
c­apita, particularly in the post-liberalisation period. Similar His starting point of this exercise is the 1987-88 consumption
anomalies appear as far as interstate poverty estimates are e­xpenditure survey. But with any such exercise which uses rural-
c­oncerned (for example, urban Assam and rural Andhra Pradesh). urban indices as well as interstate indices, his starting poverty line is
While some of these anomalies have their genesis in the way pegged to any one poverty line. In the case of Deaton’s e­xercise,
that the expert group of 1993 used the then available indices of he takes the 1987-88 all-India rural poverty line as the starting
interstate price differentials, these have become acute because of point of his calculations. This is a departure from the expert group
the deflators used to update these poverty lines. This was pointed method which uses normative bundles separately for rural and
out comprehensively by Deaton and Tarrozi (2000) and Deaton urban areas. Since his method involves use of rural-urban indices
(2008). As Deaton points out, part of the reason that the CPI indi- along with interstate price indices, there can be only one norma-
ces have failed to capture the true extent of inflation has been the tive bundle in this kind of exercise. However, the real problem
use of outdated weights. While the share of food in consumption with Deaton’s exercise is that his indices are based on unit values
basket of the people around the poverty line has fallen to less than from the NSSO consumption expenditure surveys and therefore
65% in 2004-05, the poverty lines deflators still use the weights limited to only food items. This limitation is a crucial factor espe-
implicit in the 1973-74 distribution which is around 80%; and while cially when the share of food in total expenditure has been declin-
75% of rural children in the poverty line class went to school in ing secularly over the years. One of the important findings from
2004-05, this proportion in the 1970s was only around 40%. The Deatons’s exercise was that rural-urban price differential obtained
second problem has been the movement of CPI indices for certain using unit values from NSS consumption e­xpenditure surveys was
groups over time, e g, firewood in CPI-AL in rural areas, and the fact only 15% (higher for urban areas) compared to roughly 40% im-
that health and education costs which have increased faster than plicit in the existing poverty lines (in 1999-2000).
all other costs according to private f­inal consumption expenditure However, both these corrections have avoided the issue of MRP-
(PFCE) deflators from the National Accounts Statistics (NAS) were URP differences in consumption basket. At the same time, both
not even monitored by the CPI-AL till the mid-1990s. Even within these exercises are limited to correcting for the anomalies in price
food, cereals prices have increased much less than non-cereals indices without looking at the normative aspect of the p­overty
prices between 1973-74 and 2004-05, so that it is not surprising line, particularly its anchoring with the calorie norm. While Raval-
that Deaton (2008) finds underestimation also in the inflation rate lion’s starting point is the normative poverty line taken by the ex-
of food, particularly in the rural poverty line calculations, com- pert group for 1973-74, Deaton’s starting point is the rural all-India
pared to inflation rates obtained using unit values from the NSSO official poverty line. In that sense, both these corrections do not
consumption expenditure surveys. The combination of these errors reject the normative poverty line suggested by the expert group.
which crept in over time appears to have led to serious distortions
in official poverty lines. These distortions are, moreover, much 2  New Poverty Lines for India
more serious in the rural poverty lines than urban, essentially be- In light of the above criticisms and the apparent problems with
cause the CPI-AL had more gaps and also because rural consump- the existing poverty lines, this paper proposes a new set of
tion patterns have shifted more than those of the urban poor. p­overty lines. The basic approach of setting the new poverty line
42 january 2, 2010  vol xlv no 1  EPW   Economic & Political Weekly
 special article

is similar to the exercise attempted by Deaton. The last year for the PDS and non-PDS items together. For, clothing and footwear
which Deaton’s exercise is reported is 1999-2000. However, for we have used the prices obtained from the 365-day estimates of
all calculations in this paper we have used the consumption these commodities. Some items, which are insignificant in terms
e­xpenditure survey of 2004-05. Apart from the fact that the 61st of their consumption share, have been dropped. The cut-off for
round is free from all the problems of contamination (which even selecting these items is Rs 0.15 per month (Rs 1.8 per year for
though does not affect unit values, is not the right survey year 365-day items). If these items have lower than this consumption
because of the problems inherent in budget shares), it is also the in six or more states out of a possible combination of 70 values (35
most recent quinquennial round for which the consumption states and union territories for rural and urban areas each) they
e­xpenditure data is available. have been dropped. Implementation of this procedure has led to
The second point of departure is that the interstate and rural- 11 items being dropped. For those states and sector where the
urban price indices used in the calculation of the new price indices item has been consumed by less than five households, it has been
are much more comprehensive than the indices used by Deaton treated as not being consumed in that state.
which were primarily based on food prices. However, we follow The all-India prices are not calculated directly from the unit
the essential suggestion of Deaton of using price data from NSS data but are derived from the state prices. Using state quantity
surveys rather than the CPI price data. This is possible for food, shares and prices, the all-India price of each item is the weighted-
fuel, clothing and footwear, which together account 75-80% of average of state prices. The weights used in this case are the pop-
the consumption of poor. Among other items, we have used the ulation weights from the census. These state-wise price and
Employment-Unemployment Survey (EUS) of NSS 61st round to quantity data are used to obtain a Paasche index and Laspeyre
create a price index for education expenditure and for the index index for each state relative to all-India. The geometric mean of
of health expenditure we have used the 60th round NSSO survey these gives us the Fisher’s index for each state relative to all-
which was focused on health expenditure. CPI indices from India. We repeat the same procedure to obtain urban to rural
Saluja-­Yadav’s calculations for the expert group have been used Fisher index for each state. These indices were obtained for all
only for remaining goods and services, other than rent and states including the north-eastern states. However, we have
conveyance. Our NSS unit-based indices thus cover 90% of the dropped union territories from the calculation at this stage.
consumption basket of the poor excluding rent and conveyance, This procedure has been used for food, fuel, clothing and foot-
and final p­overty lines also use actual NSS expenditure on rent wear. For all these items and item groups, the unit values can be
and c­onveyance. obtained from the NSS consumption expenditure surveys. For the
The third point of departure is that we use the all-India urban education, there are no such unit values that can be used. There-
poverty line of 2004-05 as the starting point of other state pov- fore, we have used the EUS of NSSO for obtaining the index of cost
erty lines in urban areas and also in rural areas. However, this is of education. For this purpose, we have calculated the cost of
not an arbitrary choice, but is based on evidence which we present education per schoolgoing child in each state for rural and urban
in the following section. While there is no attempt in this paper areas for children in the age-group of 5-15 from the 61st round
to decide new normative poverty lines either rooted in a nutri- (2004-05) of EUS. For the index of health expenditure we have
tional norm or any other objective criterion and we like others used the 60th round (January-June 2004) which focuses on
take the official poverty line as our starting point, we do cross- health expenditure. For construction of the index, we have calcu-
check the suitability of the existing poverty line in capturing the lated health expenditure per treatment in case of non-institu-
poor using external indicators. tional medical care and health   expenditure per case of hospitali-
The fourth point of departure is that the budget shares used sation for institutional medical care.
in our calculation of price indices are based on mixed recall While these items together (food, fuel, clothing, footwear, ed-
period. This is justified in light of the fact that NSSO has already ucation and health) cover around 85% of the consumption basket
moved to the MRP reference period in its annual rounds after of the poor, we still require an index for the remaining items of
1999-2000. This is likely to continue in the future and therefore expenditure. Major heads of expenditure for which the index
any new poverty line should be based on MRP estimates of con- cannot be computed from the NSS surveys are conveyance, rent,
sumption expenditure rather than URP estimates which has been durables, entertainment and miscellaneous goods and services.
the practice so far. However, no firm price data is available for these item groups.
Although price data for some of these items is collected for the
NSSO Price Data purpose of construction of CPI indices, a closer look at the data
As with Deaton’s exercise, our poverty line relies on the indices suggests that these may not be free from errors on account of
created using price data from the NSSO consumption expenditure non-standardisation of the items of expenditure. Nonetheless, in
survey, but with the 2004-05 round of consumption expenditure the absence of alternative price data we use the price indices cal-
data. We use the price data obtained from the consumption ex- culated using the CPI price data for these items except rent and
penditure survey for food, intoxicants, fuel, clothing and foot- conveyance. Using these Saluja-Yadav indices and our unit value-
wear. For all these items price data can be obtained from the con- based indices, we create an aggregate index using all items ex-
sumption survey. However, unlike Deaton, we use median prices cept rent and conveyance for the states relative to all-India, as
of each item in each state and sector. For PDS items (rice, wheat, well as urban relative to rural for each state. In all, we use 23
sugar and kerosene), the prices are obtained after aggregating commodity group indices and these are aggregated using the
Economic & Political Weekly  EPW   january 2, 2010  vol xlv no 1 43
speciAl article

budget shares of population around poverty line class in an itera- the poverty line class. The all-India poverty headcount ratio is the
tive manner to obtain a Fisher aggregate index. Table 1 gives the population weighted poverty headcount ratio of all states.
final index numbers used in our calculation of poverty lines. (c) We use the final all-India urban poverty lines excluding rent
and conveyance to obtain rural poverty lines in each state using
Table 1: Poverty Lines and Final Poverty Estimates for 2004-05
Index Numbers Poverty Line Poverty HCR the rural to urban Fisher indices. These are then adjusted upwards
Urban SRAI Urban-Rural Rural Urban Rural Urban using the actual share of rent and conveyance around the poverty
Andhra Pradesh 94.3 122.9 433.4 563.2 32.3 23.4 line class to obtain the final rural poverty line for each state.
Arunachal Pradesh 109.9 111.5 547.1 618.4 33.6 23.5
Assam 104.4 121.4 478.0 600.0 36.4 21.8 Use of Fisher Index
Bihar 93.8 119.9 433.4 526.2 55.7 43.7 Note that we make extensive use of the Fisher index properties
Chhattisgarh 89.8 125.1 398.9 513.7 55.1 28.4
for obtaining final index numbers at the state level. Also, the pro-
Delhi 109.3 114.5 541.4 642.5 15.6 12.9
cedure outlined above primarily deals with construction of price
Goa 109.3 106.8 608.8 671.2 28.1 22.2
indices across rural-urban within each state and for each state
Gujarat 113.1 128.6 501.6 659.2 39.1 20.1
Haryana 107.8 115.1 529.4 626.4 24.8 22.4
relative to the all-India. It is in this sense conceptually similar to
Himachal Pradesh 107.5 117.0 520.4 605.7 25.0 4.6 the exercise attempted by Deaton. However, the major point of
Jammu and Kashmir 105.2 114.8 522.3 602.9 14.1 10.4 departure is the choice of starting poverty line which in our case
Jharkhand 93.7 128.7 404.8 531.3 51.6 23.8 is the all-India urban poverty line unlike Deaton who started
Karnataka 98.6 133.4 417.8 588.1 37.5 25.9 with the all-India rural poverty line.
Kerala 100.1 108.6 537.3 584.7 20.2 18.4
Madhya Pradesh 91.8 126.2 408.4 532.3 53.6 35.1 3 An Evaluation of the New Poverty Lines
Maharashtra 107.0 127.0 484.9 631.8 47.9 25.6 Since our methodology is substantially the same as Deaton’s, it is
Manipur 111.2 110.5 578.1 641.1 39.3 34.5
useful to begin by identifying where we differ with him and how
Meghalaya 118.5 136.3 503.3 745.7 14.0 24.7
we defend these differences. First, the purely technical point
Mizoram 122.0 106.1 639.3 699.8 23.0 7.9
that, unlike Deaton, we incorporate items for which Consumer
Nagaland 134.0 108.6 687.3 782.9 10.0 4.3
Orissa 86.9 118.0 407.8 497.3 60.8 37.6
Expenditure Survey (CES) unit values are not available. Does this
Pondicherry 86.1 123.3 385.5 506.2 22.9 9.9 make any difference? The answer is not much if separately for
Punjab 112.0 116.2 543.5 642.5 22.1 18.7 rural or urban. Across-states correlation between our poverty
Rajasthan 98.6 116.5 478.0 568.2 35.8 29.7 lines and state relative to all-India indices constructed only from
Sikkim 114.7 122.8 531.5 741.7 31.8 25.9 items with CES unit values are 0.97 rural and 0.92 urban, and
Tamil Nadu 93.7 121.7 441.7 559.8 37.5 19.7 these also correlate well with Deaton’s earlier results. However,
Tripura 99.3 123.2 450.5 555.8 44.5 22.5 our all-India urban-rural price ratio is 1.27 against 1.15 using CES
Uttar Pradesh 94.4 121.2 435.1 532.1 42.7 34.1 unit values only, and our state-level ratios correlate less well with
Uttarakhand 104.4 119.5 486.2 602.4 35.1 26.2
corresponding ratios using only unit value data (0.64). This is
West Bengal 100.7 126.0 445.4 572.5 38.2 24.4
because the cost of services (e g, education, health, conveyance
All-India 446.7 578.8 41.8 25.7
Urban SRAI: Urban state indices relative to all-India.The index numbers reported in columns 2 and 3
and rent) are relatively much higher in urban than rural areas,
are for all items excluding rent and conveyance. The poverty lines are further adjusted for state/ but sometimes (e g, health costs in Kerala and Goa) cheaper in
sector specific actual expenditure on rent and conveyance by those near the poverty lines.
richer states with better public provision than in poorer ones. We
With these final Fisher price indices across states relative to all- feel that our attempt to incorporate these in a plausible manner
India for urban areas and rural-urban indices for each state, the pov- has improved upon using only CES unit values.
erty line for each state in each sector has been arrived as follows. Second, the substantial normative point on which we differ with
(a) We start with the existing urban poverty estimate at all-India Deaton, i e, to use the urban rather than rural all-India poverty line
as our starting point. The existing urban poverty line is first ad- as the starting reference. Table 2 (p 45) compares actual food and
justed for URP-MRP difference.11 However, since our aggregate health and education expenditure near the poverty line with some
indices exclude rent and conveyance, we calculate the poverty norms. We do this not only with our suggested poverty lines, using
line corresponding to the MRP distribution by obtaining the the u­rban reference, but also with what would be obtained if we
urban poverty line, which gives us the same headcount ratio had used the rural reference like Deaton. The norm against which
(25.7%) using MRP distribution as one would obtain using the actual food expenditures are judged is the actual state-wise food
URP poverty line of Rs 538.60. The poverty line, which gives a expenditure from the 61st round NSS when persons are ranked by
poverty headcount of 25.7% in urban areas using MRP distribu- their food expenditure and cut-off taken at the state-specific mal-
tion is Rs 579. The share of rent and conveyance in total MPCE of nutrition rate from NFHS.12 For both rural and urban the actual food
this poverty line class is 5.3%. Excluding this share, the MRP expenditure is 6% higher than the norm at the all-India level at
poverty line is {579*(1-0.053)} Rs 548. our poverty lines. These would have been 8% and 6% lower in
(b) With this poverty line we arrive at the state urban poverty lines rural and urban areas if we had followed Deaton in using the
using the state relative to all-India index numbers. Since these urban rural reference. For education our norm is the expenditure
poverty lines exclude rent and conveyance, we adjust these urban required at state­-specific median cost of education per schoolgoing
state poverty lines with actual share of rent and conveyance around child aged 5-14 of sending all such children to school. For health
44 january 2, 2010  vol xlv no 1  EPW   Economic & Political Weekly
 special article
Table 2: Comparison of Actual Expenditure around Poverty Line against as used by Deaton. As far as the state-wise pattern is concerned, in
Normative Expenditure
17 out of 20 major states actual food expenditure near our pov-
Food Education and Calories Per Capita
Health Per Day erty lines is higher or equal to 95% of our norms in rural areas
Reference Poverty Line AI-Rural AI-Urban AI-Rural AI-Urban AI-Rural AI-Urban and this is so for 19 out of 20 major states in urban areas. For
Rural Andhra Pradesh 0.88 1.01 0.72 0.84 1,710 1,825 health and education, actual expenditure is 90% of norm in 12
Assam 0.92 1.03 0.58 0.81 1,806 1,977 and 10 of the 20 major states in rural and urban areas. The short-
Bihar 0.99 1.13 0.49 0.77 1,928 2,146
fall in almost all other cases is on hospitalisation, not education
Chhattisgarh 1.00 1.09 0.80 1.78 1,837 1,962
or non-­institutional treatment.
Gujarat 0.88 1.00 0.91 1.45 1,742 1,768
As far as calorie intake is concerned, the all-India averages are
Haryana 0.76 0.89 0.60 0.95 1,670 1,800
Himachal Pradesh 0.85 0.94 0.59 1.03 1,847 1,947
1999 and 1776 cals/day in rural and urban areas near our sug-
Jammu and Kashmir 0.83 0.94 0.43 0.58 1,792 2,016 gested poverty line and would be 1,840 and 1,691 cals/day with
Jharkhand 0.83 0.95 0.43 0.57 1,844 1,991 the rural reference. Although considerably less than the existing
Karnataka 0.92 1.06 0.86 0.91 1,651 1,751 norm of 2,100 cals urban and 2,400 cals rural, it is interesting that
Kerala 1.07 1.20 1.73 2.19 1,445 1,704 the intake of those near our urban poverty line is almost exactly
Madhya Pradesh 0.88 0.99 0.91 1.13 1,834 1,880 the same as the present FAO norm for India, with, as expected, a
Maharashtra 0.98 1.12 1.21 1.69 1,738 2,384 higher intake in rural areas because of higher activity. At the state
Orissa 1.11 1.27 0.99 1.30 2,052 2,167
level, actual calorie intake of those near our poverty lines exceeds
Punjab 1.02 1.14 0.61 0.80 1,709 1,868
the FAO norm in rural areas of 16 out of 20 states and in urban
Rajasthan 0.85 0.99 0.57 0.60 1,909 1,971
a­reas of 13. Moreover, we have checked that actual food expendi-
Tamil Nadu 1.00 1.16 1.01 1.45 1,589 1,748
Uttarakhand 0.85 1.01 0.54 0.73 1,811 1,947
ture of those near our suggested poverty lines is sufficient to afford
Uttar Pradesh 0.94 1.08 0.78 1.03 1,996 2,115 a nutritionally balanced diet, based on the higher ICMR norms, as
West Bengal 0.87 1.00 0.90 1.11 1,815 1,957 suggested by the National Institute of Nutrition.14 Further, the new
All-India 0.92 1.06 0.85 1.14 1,840 1,999 poverty lines presented here are much more in line with outcome
Urban Andhra Pradesh 1.00 1.09 0.62 1.05 1,628 1,627 indicators from the NFHS on malnutrition.15 The correlation across
Assam 0.88 0.95 0.42 0.72 1,756 1,931 states/sectors between NFHS malnutrition indicators and head-
Bihar 1.00 1.13 0.40 0.82 1,944 1,947 count poverty using the new poverty lines is 0.87, as against 0.5
Chhattisgarh 0.94 1.05 0.98 1.64 1,781 1,774
with the official Planning Commission poverty estimates. Scatter
Gujarat 0.85 0.95 0.67 0.80 1,575 1,644
plots of existing official and the new poverty headcounts against
Haryana 0.80 0.95 0.71 0.82 1,534 1,800
NFHS malnutrition indicators are presented in Charts 1 and 2.
Himachal Pradesh 0.68 0.86 0.94 0.47 1,422 2,119
Jammu and Kashmir 0.74 1.00 0.35 0.66 1,420 1,959 Chart 1: Scatter Plot of NFHS Malnutrition Index and Planning Commission
Poverty Estimates
Jharkhand 0.84 0.96 0.71 0.69 1,907 2,013
Karnataka 1.00 1.13 0.66 0.67 1,643 1,712 50
y = 0.6664x + 4.1086
Kerala 1.01 1.13 1.38 1.48 1,456 1,503 R2 = 0.2586
Planning Commission Poverty HCR (%)

40
Madhya Pradesh 0.94 1.05 0.71 1.29 1,718 1,804
Maharashtra 0.96 1.10 1.04 1.52 1,592 1,696
30
Orissa 1.03 1.20 0.76 0.78 1,909 2,056
Punjab 0.89 1.05 0.55 0.90 1,532 1,789
20
Rajasthan 0.91 1.03 0.98 1.17 1,711 1,816
Tamil Nadu 0.96 1.10 0.85 1.45 1,567 1,715 10
Uttarakhand 0.94 1.08 0.60 0.89 1,829 1,850
Uttar Pradesh 0.99 1.11 0.84 1.25 1,818 1,926 0
West Bengal 0.98 1.05 0.63 0.92 1,751 1,757 0 30 10 40 20 50 60
NFHS Malnutrition Index (%)
All-India 0.94 1.06 0.79 1.22 1,691 1,766 NFHS Malnutrition Index average of three indicators.
AI-Rural represents the actual expenditures around the poverty line class with starting
reference as the all-India rural poverty line for 2004-05. Similarly, AI-Urban represents the actual Chart 2: Scatter Plot of NFHS Malnutrition Index and New Poverty Estimates
expenditures around the poverty line class with starting reference as the all-India urban poverty
line for 2004-05. 70
y = 1.1163x – 2.3898
R2 = 0.7509
we calculate the age-weighted probabilities of disease onset and 60

hospitalisation from the a­ge-specific data for urban areas in the

New Poverty HCR (%)

50
60th round and m­ultiply these by the state-specific median cost
per treatment and per hospitalisation.13 For all the three norms, 40

the all-India norm is the population weighted average of states. 30

Taken together, the all-India actual expenditure is 14% and 22%
higher than the norm in rural and urban all-India with our pov- 20

erty lines. These would be 15% and 21% lower than the norm if we 10
had like Deaton used the rural reference. Clearly, this comparison
of actual expenditures with norms provides very strong evidence to 0
0 10 20 30 40 50 60
choose the urban reference as we do, rather than the rural reference NFHS Malnutrition Index (%)

Economic & Political Weekly  EPW   january 2, 2010  vol xlv no 1 45

 speciAl article
Chart 3: Real Food Expenditure at Real MPCE Class and simply impose 100% school attendances and the ability to
800 meet some probabilities of disease onset. Again, any under­
700 estimation in NSS would lead to an underestimation of norms. For
this reason, our estimates appear to be normatively superior to
Real Food Expenditure (in Rs)

600
Rural Deaton’s and also immune to the charge of NSS underestimation.
500
We turn now to some tests of robustness. First, when plotted
400 against per capita real MPCE, per capita real food expenditure is
300
Urban almost identical for rural and urban households over a fairly wide
range of real per capita MPCE (Chart 3). Real food expenditure
200
here is nominal food expenditure of each household deflated by
100 our indices of food prices from CES unit values. Similarly, real
Less than 250

300-350

400-450

500-550

600-650

700-750

800-900

1,200 and above

MPCE is the nominal MPCE of each household deflated by the de-
flators implicit in the final rural and urban poverty lines assum-
Real MPCE Class (in Rs)
ing all-India urban to be 100. This means that rural and urban
real food shares are similar at similar levels of real MPCE, making
Chart 4: Cumulative Distribution of Casual Labour Households by Real MPCE Class
our rural-urban comparisons quite robust within this range. Sec-
100
ond, we consider the case of poor household groups. It may be
80 seen from Charts 4 and 5 that cumulative distributions by real
Urban MPCE (nominal MPCE deflated by our cost of living indices) of two
60
clearly poor groups (those in casual labour households and those
(%)

40 in households with an illiterate head) are almost identical in

r­ural and urban areas (Charts 4 and 5). Thus, urban or rural resi-
20 dence does not significantly affect the probability of being poor
Rural
0
for these groups. This corrects the implausible implication of the
current official estimates: that poverty among these groups is
Less than 250

300-350

400-450

500-550

600-650

700-750

800-900

1,200 and above

much more in urban than in rural areas. Both observations sug-

Real MPCE (in Rs) gest that it is indeed possible to have a common yardstick to
measure consumption poverty without discriminating between
Chart 5: Cumulative Distribution of Illiterate Households Head by Real MPCE Class urban and rural, and that our cost of living indices provides this
100 much better than the present official poverty lines.
Rural Finally, since any change in poverty lines from those used
80
currently involves reclassifying the poverty status of some per-
60 sons, do the proposed poverty lines reclassify households in the
Urban
proper direction? We examine this at household level for some
(%)

40 characteristics – household type, education status of household

head, possession of luxuries, per capita calorie intake and the
20
food and fuel share in total nominal household consumption – of
0 four groups: those unambiguously poor (i e, poor by present
Less than 250

300-350

400-450

500-550

600-650

700-750

800-900

1,200 and above

o­f ficial classification and by new poverty lines), those who get
reclassified from non-poor in official to poor by new lines, those
Real MPCE (in Rs) who get reclassified from poor to non-poor, and those unambig-
uously non-poor. Of these, the change in status is only for those
Thus the choice of the all-India urban reference rather than who get classified as non-poor by the official method but are
the all-India rural reference used by Deaton is normatively poor by our method (nonpoor_poor) and those who were poor
j­ustified, except possibly that it still does not meet catastrophic by the o­f ficial method but are now characterised as non-poor by
medical contingencies leading to hospitalisation. To have used our method (poor_nonpoor). There is no change in status for the
the all-India rural poverty line as reference would have meant two remaining groups; poor_poor and nonpoor_nonpoor. If our
falling below the norm on food, health and education in the vast poverty lines reclassify the households in the right direction
majority of states. Moreover, it may be noticed that the norms we then we expect the nonpoor_poor to have similar characteristics
use are also immune to the criticism made (for example, by Surjit as that of the poor_poor while the characteristics of the poor_
Bhalla) that poverty is overestimated because of NSS underesti- nonpoor will be closer to that of the nonpoor_nonpoor. For
mation of consumption. Since, our food norm is the cut-off on household type and education of household head we expect a
NSS food distribution corresponding to NFHS malnutrition; this lower share among the non-poor compared to the poor; simi-
would itself be an underestimate if the NSS food consumption is larly, for the share of food in total consumption which is lower
underestimated. Similarly, our health and education norms for the non-poor compared to the poor. On the other hand, the
a­ccept NSS expenditure per treatment and per schoolgoing child percentage of households possessing luxury goods is higher for
46 january 2, 2010  vol xlv no 1  EPW   Economic & Political Weekly
 special article

the non-poor compared to the poor. For per capita calorie intake, mainly because our estimates are more than double in eight states
we expect non-poor households to have a higher calorie intake (Andhra Pradesh, Gujarat, Delhi, Himachal Pradesh, Jammu and
than the poor households. Kashmir, Karnataka, Punjab and Rajasthan) all with relatively
In almost all cases those whom our poverty lines reclassify low official poverty. An important consequence of this is that we
from non-poor to poor (nonpoor_poor) are on the average worse- report no case of a state having higher urban poverty i­ncidence
off (poorer) by these characteristics than those reclassified from than rural, unlike official estimates that show higher urban
Table 3: Characteristics of Households by Alternative Poverty Lines poverty than rural in a number of states. How-
Percentage of Households in Each Group ever, the broad clustering of states by high poverty
Categories Household Type Education of HH Head Possess Share of Food Calories (Bihar, Chhattisgarh, Jharkhand, Madhya Pradesh
Casual Casual Labour and Illiterates Illiterates and Luxury in Total Per Capita
Labour Self-Employed Literates Up to Goods Consumption Per Day and Orissa), low poverty (­Haryana, Himachal
in Non-Farm Primary Pradesh, Jammu and K­ashmir, Kerala and Punjab)
Rural and the rest remains unchanged.
  Poor_Poor 51.7 65.7 60.4 84.8 1.3 63.0 1,642
On how to carry these new poverty lines
  Nonpoor_Poor 46.3 61.3 53.5 79.9 3.2 62.3 1,830
  Poor_Nonpoor 27.1 39.6 44.3 75.5 5.6 46.4 1,824
f­orward, there are two alternatives consistent
  Nonpoor_Nonpoor 25.2 43.3 35.8 62.8 19.0 56.6 2,299 with the method used here. The first is to carry
Urban forward the new state/sector-specific poverty
  Poor_Poor 28.6 73.78 43.6 73.2 6.7 59.1 1,600 lines with state/sector-specific inflation indices
  Nonpoor_Poor 18.3 70.56 31.3 69.6 14.1 58.3 1,757 derived as far as possible from NSS unit values,
  Poor_Nonpoor 15.9 61.86 28.7 53.9 22.0 50.3 1,742 and using CPI-AL/CPI-IW only for commodity
  Nonpoor_Nonpoor 6.2 48.11 11.9 29.9 56.2 44.8 2,182
groups where unit values are not available. The
The category before the under-stroke (_) is the status of the household by existing official poverty line of the Planning
Commission and the category after the under-stroke is the status of the households by our poverty lines. For example, Nonpoor_ state/sector-­specific commodity group weights
Poor category of households was non-poor using the existing official poverty line but is poor by the new poverty lines.
could either be those from the actual consumption
poor to non-poor (poor_nonpoor). Moreover, on basis of these pattern near each poverty line group from the previous NSS large
characteristics, those reclassified from non-poor to poor are sample or Fisher indices could be used to combine weights from
much closer to those unambiguously poor than those unambigu- both the previous and current NSS round. The second method
ously non-poor. Interestingly, it is only for calorie intakes that the would be to carry forward only the all-India urban poverty line in
new poverty lines may reclassify incorrectly in both rural and this manner, and then derive state/sector-specific poverty lines
urban areas, suggesting that calorie intake has a very weak link corresponding to this by using current commodity group weights
with all other poverty indicators. Nonetheless, although these and spatial indices, as is done for 2004-05 in calculations for this
characteristics are not necessarily the best proxies for consumption paper. The advantage of the first method is that this would pre-
poverty and our findings less firm on some of these than others, serve the current practice of deriving all-India lines from state-
this analysis and comparison with NFHS data on malnutrition by specific lines obtained by updating these independently. The
states/sectors does provide independent and fairly strong support second method does not do this, so inflation factors could differ,
for shifting from existing official poverty lines and poverty esti- but has the advantage that spatial differences in poverty lines
mates to the alternative suggested at the outset of this article. would be based entirely on current prices and weights, and thus
be more up to date. The second method would be closer to the
Conclusions approach followed in this paper. These alternatives can also
If this suggestion is accepted, some major modifications will be be used to carry poverty lines backward, although additional
necessary both in estimates of poverty incidence across states/ complications would arise since NSS unit values are available only
sectors and on how to carry these forward and backward over from 1983 and URP and MRP measures in earlier rounds involve
time.16 As for poverty estimates, the most important difference different weights. This paper does not provide past estimates, but
with official estimates is that, like Deaton’s earlier estimate, the actual past decline in rural consumption poverty was clearly
our estimates show a much larger rural-urban difference but less than official, in part because of deficiencies of the CPI-AL and
less concentration of either rural or urban poverty in a few in part because the present method has not captured actual
states. Although construction of our new poverty lines preserves changes in consumption patterns over time.
the official estimate of all-India urban poverty incidence in
2004-05, there are significant changes at the state level. Andhra Notes
Pradesh, Chhattisgarh, Delhi, Karnataka, Kerala, Madhya 1 For a history of setting of poverty lines in India, see Srinivasan (2007).
2 For urban areas, the Expert Group had actually suggested use of both CPI-IW
Pradesh, Maha­rashtra, Orissa, Rajasthan, Tamil Nadu and and CPIUNME. However, the use of CPIUNME was not accepted by the government.
Uttarakhand (accounting for 69% of urban poor by official 3 The expert group also looked at the hunger c­riterion and food share criterion
along with c­alorie norms.
estimates) show less poverty with our poverty lines while other 4 Some of these recommendations were also r­eported in Dev (2005).
states, particularly north-east states, show more. As a result, the 5 Mahal and Karan (2007) calculate “culturally palatable” minimum cost diets
that meet ICMR RDAs on energy, proteins, fats and eight micronutrients for 16
interstate coefficient of variation in urban poverty falls from 0.75 major Indian states in 1993-94 and 1999-2000, and report the percentage of
by official estimates to 0.41 with our poverty lines. Our rural pov- population who are poor by this dietary adequacy measure if food share in to-
tal expenditure is maintained at actual levels. They find that dietary poverty by
erty estimates are higher than the official in almost every major this definition is marginally lower than official poverty at the level of all-India
state; and again interstate coefficient of variation in this is lower, urban, with states roughly equally divided on which urban poverty measure is

Economic & Political Weekly  EPW   january 2, 2010  vol xlv no 1 47

speciAl article
higher. On the other hand, they report that die- 14 National Institute of Nutrition recommends Dev, Mahendra (2005): “Calorie Norms and Poverty”,
tary poverty exceeds official poverty in rural are- d­ietary intakes by age-sex and activity status for Economic & Political Weekly, 19 February.
as of almost all states in both years and the excess Indians. This diet weighted by actual age-sex Dev, Mahendra and C Ravi (2008): “Revising Esti-
averages 8-10 percentage points at all-India level. composition for a person doing moderate work is mates of Poverty”, Economic & Political Weekly,
6 See Utsa Patnaik (2007) for an influential critique affordable without cutting non-food expenditure 8 March.
of the present methodology. by those at our poverty lines for majority states as FAO (2008): The State of Food Insecurity in the
7 The norm used by FAO for India in its State of well as in all-India. World, Food and Agriculture Organisation, Rome.
Food Insecurity Report (SOFI) 2008 is 1,770 calo- 15 NFHS malnutrition indicators referred here is the Gaurav, Datt and Martin Ravallion (1998): “Why Have
ries per day, down from 1,820 calories per day simple average of three indicators; percentage of Some Indian States Done Better Than Others at
used in SOFI 2006. These current FAO norms fol- children under age 3 who are underweight, per- Reducing Rural Poverty?”, Economica, Vol 65,
low from an ongoing joint WHO/FAO consulta- centage of men with low BMI and percentage of No 257, February, pp 17-38.
tion. In 1985, FAO shifted from its then current women with low BMI. Government of India (1979): “Report of the Task Force
RDA for calories to Estimated Average Require- 16 These suggestions and the proposed methodology on Projections of Minimum Needs and Effective
ments (EAR). The earlier RDA had sought to were discussed in its meetings by the Expert Consumption Demand”, Perspective Planning Di-
e­nsure that, given normal variation across indi- Group and were subsequently accepted. The final vision, Planning Commission.
viduals, requirements were met for at least 95% of poverty line and the poverty estimates reported – (1993): “Report of the Expert Group on Estima-
population. Since this overestimates require- in this paper are the same as reported in the final tion of Proportion and Number of Poor”, Perspec-
ments for most of the population and in case of report of the expert group submitted to the tive Planning Division, Planning Commission.
calories could exceed desirable maximum levels Planning Commission. International Institute for Population Sciences
for many, EAR is based on the requirement at the (IIPS) and Macro International (2007): National
median of the distribution. In 2004, WHO/FAO Family Health Survey (NFHS-3), 2005-06, Mumbai:
reduced the EAR further based on new data. IIPS.
These revisions apply to all countries not just References
Mahal, A and A K Karan (2007): “Adequacy of Dietary
I­ndia. For example, FAO’s norm for China is now Bhalla, Surjit (2003): “Recounting the Poor: Poverty Intakes and Poverty in India: Trends in the 1990s”,
1,900 calories, which is higher than the India in India, 1983-99”, Economic & Political Weekly, 25 Economics and Human Biology (2007),
norm only because its demographic composition January. doi:10.1016/j.ehb.2007.10.001.
has more adults. Chatterjee, G S and N Bhattacharya (1974): “Between Minhas B S, L R Jain, S M Kansal and M R Saluja
8 Calorie requirement depends on age, sex, weight, State Variations in Consumer Prices and Per Capi- (1988): “Measurement of General Cost of Living
height, activity pattern, climate, water quality ta Household Consumption in Rural India” in for Urban India-All-India and Different States”,
and various other factors, and RDA norms are T N Srinivasan and Pranab Bardhan (ed.), Poverty Sarvekshana, 12.
available by the first four of these. The 1979 Task and Income Distribution in India, Statistical Pub- Patnaik, Utsa (2007): “Neoliberalism and Rural
Force took average all-India data from 1971 cen- lishing Society, Calcutta. Poverty in India”, Economic & Political Weekly,
sus, but the resultant need not be RDA for any Deaton (2003): “Prices and Poverty in India, 1987-2000”, 28 July.
state. For example, calorie intake in Kerala and Economic & Political Weekly, 25 January. Saluja M R and Bhupesh Yadav (2008): Construction of
Tamil Nadu are lowest in every NSS survey, but – (2008): “Price Trends in India and Their Implica- Rural and Urban Consumer Price Indices for India,
independent data on health status and nutrition tions for Measuring Poverty”, Economic & Political India Development Foundation.
from NFHS/NNMB repeatedly show these states Weekly, 9 February. Sen, Pronab (2005): “Of Calories and Things: Reflec-
to be among the best states. The expert group
Deaton, Angus and Jean Dreze (2009): “Food and tions on Nutritional Norms, Poverty Lines and
recognised that RDA exercise was best done at
Nutrition in India: Facts and Interpretation”, Eco- Consumption Behaviour in India”, Economic &
state level but, since they were not recommending
nomic & Political Weekly, 14 February. P­olitical Weekly, 22 October.
c­alorie-based poverty lines, accepted all-India
RDAs from the Task Force as reference for con- Deaton and Tarrozi (2000): “Prices and Poverty in Srinivasan, T N (2007): “Poverty Lines in India: R­ef­
sumption poverty.
India”, Research Programme in Development lections After the Patna Conference”, Economic &
Studies, Princeton University, processed. Political Weekly, 13 October.
9 According to NFHS-3, urban malnutrition was
36.3% in terms of underweight children, 19.8% in
terms of adult women with low BMI and 17.5% in
terms of adult men with low BMI. The correspond-
ing figures for rural areas were 49%, 38.8%
and 33.1%.

10 Deaton and Dreze (2009) have analysed the calo-
rie puzzle in detail and have arrived at a similar REVIEW OF WOMEN’S STUDIES
conclusion.
11 The poverty estimate of 25.7% for urban areas is October 31, 2009
based on the size distribution of households in
a­scending order of per capita total (consumer)
e­xpenditure (PCTE for short) on the basis of the The Law, Gender and Women – Kalpana Kannabiran
URP. The MRP equivalent poverty line is that lev- ‘Nonconformity Incarnate’: Women with Disabilities,
el of PCTE which is obtained when population is
ranked in ascending order of size of MRP-based   ‘Gendered’ Law and the Problem of Recognition – M Pavan Kumar, S E Anuradha 
PCTE and the percentage of poor equals 25.7%. Indian Muslim Women, Politics of Muslim Personal
12 Average of the proportion of underweight chil-
dren below five years of age and proportion of   Law and Struggle for Life with Dignity and Justice – Razia Patel
men and women aged 15-49 years with low body Bringing Rights Home: Review of the Campaign
mass index is taken to be an aggregate outcome
indicator of malnutrition. When estimated (rural/­
  for a Law on Domestic Violence – Indira Jaising
urban/state) population from NSS is ranked ac- Conjugality, Property, Morality and Maintenance –Flavia Agnes
cording to ascending size of food expenditure per
capita, normative food expenditure per capita is Women, Forestspaces and the Law: Transgressing the Boundaries – Sagari R Ramdas
defined by that level of food expenditure per
c­apita that corresponds to cumulative share of
Women’s Land Rights in South Asia: Struggles and Diverse Contexts – Meera Velayudhan
population from NSS that equals the index of mal- Outside the Realm of Protective Labour Legislation:
nutrition derived from NFHS-3 for that state.
13 We use the age-specific PAP (persons reporting
  Saga of Unpaid Labour in India – Padmini Swaminathan
ailment) in the last 15 days for non-institutional Judicial Meanderings in Patriarchal Thickets:
medical expenditure hospitalisation by state from   Litigating Sex Discrimination in India – Kalpana Kannabiran
the 60th round (January-June 2004). Weighted
by the age-specific population distribution these
give us the probabilities of disease onset and hos- For copies write to:
pitalisation by state and sector. To arrive at the Circulation Manager,
norm, we multiply these by the median cost of ex-
penditure assuming everybody is treated. That is,
Economic and Political Weekly,
our norm is the expected expenditure of treating 320-321, A to Z Industrial Estate, Ganpatrao Kadam Marg, Lower Parel, Mumbai 400 013.
all persons given the probabilities of disease onset email: circulation@epw.in
and hospitalisation.

48 january 2, 2010  vol xlv no 1  EPW   Economic & Political Weekly