ENGEL CURVE METHOD FOR MEASURING POVERTY By T. Krishna Kumar, Jayarama Holla, and Puja Guha January 8, 2008 P.V. Sukhatme Memorial Lecture by T. Krishna Kumar at the Annual Conference of the Indian Society of Probability and Statistics, Nagpur, India, January 10, 2008 Economics and Social Science Area Indian Institute of Management Bannerghatta Road Bangalore-560076. INDIA - 1
36
Embed
ENGEL CURVE METHOD FOR MEASURING POVERTY · 4 Engel curve for a specific commodity is the relation between the consumption expenditure on that commodity as a function of income (total
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
ENGEL CURVE METHOD FOR MEASURING POVERTY
By
T. Krishna Kumar, Jayarama Holla, and Puja Guha
January 8, 2008
P.V. Sukhatme Memorial Lecture by T. Krishna Kumar at the Annual Conference of the Indian Society of Probability and Statistics, Nagpur, India,
January 10, 2008
Economics and Social Science Area
Indian Institute of Management
Bannerghatta Road
Bangalore-560076. INDIA -
1
2
ENGEL CURVE METHOD FOR MEASURING POVERTY1
By
T. Krishna Kumar2, Jayarama Holla3, and Puja Guha3
I INTRODUCTION
1.1 Professor Sukhatme, as the Head of Statistics Division of the Food and Agricultural Organization of the United Nations, emphasized the importance of augmenting world food production, and improving its distribution so as to reach the deprived sections of human community across the world. He also brought that experience to address the hunger situation in India. It is in this connection that he made his pioneering contribution to the statistical measurement of the extent of hunger (or under-nutrition) and malnutrition (or protein deficiency) (Sukhatme (1961, 1965, 1974)). He stressed the importance of examining the total food requirements on the one hand and its requirements at the individual level distinguished by age, sex, and physical activity on the other. He went deeper into the nutrition science to argue that calorie deficiency is much more important than protein deficiency, as utilization of proteins requires a minimum quantity of calories. His measure of incidence of hunger was based on the integration of the joint distribution of the requirements and availability over the set where the availability at the individual level is less than what is required. Parts of this method were borrowed, with due acknowledgements to Sukhatme, by Dandekar and Rath (1971) in providing a scientific basis for defining a poverty line. Indian literature on measurement of poverty has set the trends in poverty measurement elsewhere in the world. Thus Professor Sukhatme’s contributions to estimation of hunger have a lasting impact on measurement of poverty and its alleviation.
1.2 Around 1991 Sitaramam showed the first author the draft of a paper he and Anil Gore, wrote. That was the first draft of an entirely new approach to the measurement of poverty without a poverty-line, and it was based on cereal consumption deprivation. While discussing
1 P.V. Sukhatme Memorial Lecture by the first author at the Annual Conference of the Indian Society of Probability and Statistics, Nagpur, January 10, 2008.
2 Guest Faculty, Indian Institute of Management, Bangalore and Retired Professor, Economic Analysis Unit, Indian Statistical Institute, Bangalore. The work reported here is based on the work the first author did with V. Sitaramam and Anil Gore in the early nineties. The authors thank N. Krishnaji, Federico Perali, and S. Subramanian for discussions and comments on a working draft of this paper. The authors thank NSSO for providing the data on CDs with excellent documentation.
3 Doctoral students, Indian Institute of Management, Bangalore
3
that draft Sitaramam explained to him the work he and his colleagues were doing on “Enzymology” or “Catalysis”. It was then that he saw a similarity between the saturation curves they were using in catalysis (as well as in his draft paper with Anil Gore) and the Engel curves for necessities (or essential commodities) of economic analysis4. He joined them in revising that draft to provide the economic content (Kumar, Gore and Sitaramam (1996))5. An empirical application of those ideas were provided by establishing a hierarchy of needs using National Sample Survey Organization (NSSO) data and using cereals, the first in that hierarchy, as the commodity to measure the commodity-specific consumption deprivation (Sitaramam, Kumar, Gore, Paranjpe, and Sastry (1996)). That study used grouped data of NSSO for various earlier rounds (16th to 46th rounds covering the period 1960-1990) and estimated saturating Engel curves for cereals using per capita monthly expenditure on cereals as a function of monthly per capita total expenditure.
1.3 In recent years many researchers studying poverty in India had access to ungrouped data at the household level provided by NSSO. The earlier work of Sitaramam et. al. (1996) was extended recently using this household level data for the three large surveys of 1987-88, 1993-94, and 1999-2000, and by introducing household size as an additional variable affecting the Engel curve (Kumar, Mallick, and Holla (2008)). That study observed that there could have been a substitution of cereals by some other food items and that in some states cereal substitutes, such as tapioca, are used in place of cereals. In order to take these aspects into account we extend that work further in this paper by considering deprivation in food (all food items, instead of limiting to cereals). The earlier work did not make any adjustments to different household compositions within a given household size and treated adults and children, and males and females, alike. The literature in consumer demand and living standards studies suggest that households of different compositions must be converted to comparable adult equivalent scales.6 In this paper we replace per capita consumption by per adult male equivalent consumption, and household size by number of adult equivalent units. In the previous analysis each sampled household was given equal weight ignoring the sample design that gave more opportunities for certain households to get selected. In this paper we make the necessary correction by multiplying the sample observations by a multiplier provided by NSSO to correct for this.
4 Engel curve for a specific commodity is the relation between the consumption expenditure on that commodity as a function of income (total expenditure).
5 This was perhaps one of the first instances of defining an inclusive measure of poverty, and the first without any poverty line. The first draft of the paper written by Sitaramam and Gore was privately circulated for comments in 1991. The revised paper by Kumar, Gore and Sitaramam was submitted for publication in February 1993 and presented at a conference on poverty and income inequality in March 1994 held in Bangalore.
6 We thank an anonymous referee of Journal of Development Studies who emphasized the need for such adjustment while reviewing our earlier work for that journal.
4
II DIFFERENT DIMENTIONS OF HUMAN DEPRIVATION AND TRADITIONAL MEASURES OF POVERTY
2.1 Hunger and malnutrition are the basic forms of human deprivation. Other dimensions of deprivation could be due to lack of access to basic health services, housing, primary education, drinking water, sanitation, etc. Economists label all of them together as essential goods and services and define poverty as a condition of not having adequate personal income to procure these essential services. Noting that food is the most essential of all these essential commodities, they defined poverty as not having enough personal resources to acquire adequate food, such as two square meals a day. When economists were vague about what constitute the basic food requirement such as two square meals a day Dandekar and Rath (1971) used the ideas contained in Sukhatme’s work and defined minimum calorie requirement for a reference individual and adjusted it to capture the average calorie requirements at All-India level for rural and urban sectors depending on the demographic and occupational distribution of the population. Dandekar and Rath determined, using the National Sample Survey data, at what total expenditure the households meet this average calorie requirement and called it the poverty line. The official poverty line is based primarily on this methodology developed by Dandekar and Rath, which was itself based on Sukhatme’s pioneering contribution on measuring hunger.
2.2 V.K.R.V. Rao criticized this method stating that what it measures is under-nutrition and not poverty. Sukhatme was also critical of this approach for using the average requirement. He was of the view that the capacity of a person to perform work is not limited by his current calorie intake but by the efficiency with which he converts the calorie intake into metabolisable energy over his homeostatic range of intake.TP
7PT More specifically, he stated that over a long enough period
to cover intra-individual variations that adjust storage and utilization of calories, if calorie intake is x and calorie requirement (that depends on age, sex, occupation and activity) is y and their joint distribution is f(x, y) then the extent of under-nutrition in the aggregate is given by:
In recent years there has been a considerable debate on the way poverty estimates are made in India. An excellent discussion is found in the work of Deaton and Kozel (2005). That entire
TP
7PT Although P.V. Sukhatme was a member of the Expert Group on estimation of proportion and number of poor set
up by the Government in 1989, he expressed his dissent and reiterated his point of criticism by submitting a supplementary note as Annexure 1 to the Report of the Expert Group (1993). Krishnaji (1981), however, was critical of the suggestion made by Sukhatme. He pointed out that the estimation of intra-individual variation for the same age-sex-occupation specific group of households, required for Sukhatme’s procedure, cannot be derived from non-experimental data of NSSO.
5
debate is based on accepting the basic methodology of Dandekar and Rath and focuses on several empirical shortcomings of data used and the dogma associated with it.
2.4 There are some other major directions in which the poverty research has been commented upon recently. It is noted that the calorie norm itself is inappropriate. Behrman and Deolalikar (1987) found that the poor substitute, even at low levels of income, luxury food items for food items with higher calories when their income rises, implying that the poor’s preferred need is not necessarily that of meeting the calorie requirement. Rath (1996) and Rath (2003) and Sen (2005) demonstrate that the actual calorie intake has gone down below the norm over the years even at moderate levels of income. Meenakshi and Viswanathan (2003) demonstrate that while poverty line-based poverty has decreased the calorie deprivation has increased. The main theme of most of these papers is that the poverty line is measured for a reference group for a reference year and it is adjusted for other non-reference groups and non-reference years using price deflators. The changing poverty lines so determined do not conform to the calorie norm used in defining the reference poverty line.
2.5 It is also suggested that instead of the research economist suggesting what should be the poverty line the responding household or individual must answer a minimum income question stating at what level of income he or she would consider that his or her both ends are met-thus going back to the vague definition of poverty line that prevailed before the work of Sukhatme and Dandekar and Rath (Pradhan and Ravallion (2000)). Kumar, Mallick, and Holla (2008) show that cereal consumption deprivation, a major component of poverty in India, has little correlation with the various traditional measures of poverty, thus questioning the practical relevance of the traditional measures of poverty if consumption deprivation of essential commodities is the focus of any study on poverty. Lipton (1997) picks a few holes in the prevailing traditional methodology of poverty measurement. In particular he comments that the severity of poverty measured by traditional Foster-Greer, Thorbecke (1984) type of measure lacks intuitive economic meaning. Atkinson commented, nearly two decades ago, that there is a need to bring about a vertical integration between poverty measurement and welfare economics based on consumption (Atkinson (1987)).
2.5 Some of the essential goods and services, such as food, are private goods supplied mostly through market mechanism, while some others such as primary health services, drinking water, and sanitation are public goods provided by the government. A few other essential goods and services, such as education and non-primary health services, are quasi-public goods provided by non-governmental organizations. The economic access to these basic goods and services is determined not only by the resources at the command of each individual or individual household, but also by the quantity and quality of public and quasi-public goods provided by the
6
government and non-governmental organizations.8 Thus personal income is not the major determinant of poverty. Poverty is associated with unemployment or irregular employment thereby making measured or observed income being very volatile. In fact it is that volatile low income that is a major characteristic feature of poverty. Thus, the identification of the poor by a volatile measure seems to be inappropriate. The traditional methods of poverty measurement based on income or total expenditure and its distribution thus seem to be only an indirect way to study poverty. In view of all these deficiencies of the traditional measures of poverty direct measures of poverty, based on consumption deprivation of essential commodities, will be much more useful.
III A NEW MEASURE OF POVERTY BASED ON THE ENGEL CURVE
3.1 The expert group (Government of India, 1993) observed various limitations of the traditional method of estimating the proportion and the number of the poor. Some of them are worth reporting here:
(i) One must note that poverty concept and poverty line are not absolute but relative, and in fact it is the relative poverty that determines the absolute poverty at any income level
(ii) Poverty line determined from consumption pattern does not take into account items of social consumption provided by the government
(iii) The poverty line combines normative and behavioral elements as while the calorie requirement is determined normatively for an essential item food, the requirements for non-food items are set as determined by the observed relation between the food consumption and non-food consumption
(iv) The head-count measure does not measure the severity of poverty
3.2 We follow the route taken by Atkinson and try to integrate poverty measurement with basic microeconomic analysis of demand for an essential commodity. Our approach can be explained easily by identifying two different ways of viewing poverty. One may view poverty at an individual level and then aggregating it over the set of all those who are poor in a given community. This requires associating poverty with the condition of the poor, thus necessitating the definition and identification of the poor9. There is an alternative view where one can consider an individual as a member of a community and it is the situation of the individual within the
8 For a fuller discussion of some of these issues one may see Section II of Kale Memorial lecture of Vaidyanathan (Vaidyanathan (2001)).
9 It is this aspect that gave the poverty line a central role in measurement of poverty. Once the poverty line is so defined the attention was unwittingly shifted to income and income distribution rather than to consumption and consumption distribution and consumption deprivation.
7
community that determines what the necessities are for that community. This is the relative nature of poverty that the expert group noted above. It is this interaction of an individual with other members of his community that determines his priorities in consumption, and what his requirements are. It is the distribution of the consumption expenditure on essential commodities relative to the norm determined by the community that provides us information on consumption deprivation. It is the consumption deprivation of an entire community with respect to all essential commodities that constitutes poverty of that community. This is the approach Kumar et. al. (1996), and Sitaramam et.al. (1996) took. In the latter work a hierarchy of needs was established, while in the former the most essential of those needs, cereals, was taken and cereal consumption deprivation was used as a poverty measure.
3.2 Engel curve provides a wealth of information on the community’s consumption behavior at various levels of total expenditure and for different family compositions. Dandekar and Rath (1971), Rao (1981), and Deaton and Tarrozi (2000) came very close to taking a full view of the Engel curve but confined to look only at the poverty line portion of the Engel curve. In determining the total expenditure at which the minimum calorie requirements are met Dandekar and Rath used the observed empirical relation between food expenditure and the total expenditure. Dandekar and Rath looked at the Engel curve based on the National Sample Survey data to determine the total expenditure that would permit an expenditure on food which would meet the calorie requirements. They used the Engel curve only for that purpose. Bhanoji Rao used the typical properties of an Engel curve of a necessity to suggest that the proportion of food expenditure increases, reaches a maximum and then declines. He suggested that the point where this proportion reaches a maximum could be taken as the threshold for acute poverty and suggested that one and half times that level can be taken as the poverty line. Again he left Engel curve behind after deriving a poverty line from it. The Engel curve for a commodity is actually the demand function for that commodity, keeping the prices constant. As such it depicts the consumption behavior of persons or households with different levels of total expenditure. Deaton, who has done considerable work on consumer behavior and consumer demand functions, also examined poverty issue only from the traditional approach and ignored looking at the entire Engel curve for essential commodities. The Engel curve not only suggests how to fix a poverty threshold it also provides a distribution of mean consumption expenditure on food according to income. Thus the Engel curve summarizes the economic equilibrium consumption expenditure on food.
3.3 Sitaramam suggested using commodity-specific consumption deprivation, such as cereal or food consumption deprivation, as a measure of poverty. Kumar, Gore, and Sitaramam (1996) demonstrated that this new measure satisfies various properties that a poverty measure should satisfy, except that it has no poverty line for identifying who the poor are. They even suggested that there is no need to have a poverty line to measure poverty. Ideally an Engel curve for an essential commodity must saturate at a finite level of expenditure on that commodity. They
suggested using that saturation level of commodity-specific consumption of an essential commodity as a norm set by the community. Any shortfall from that level may be taken as the commodity-specific consumption deprivation. They showed that a standardized cumulative commodity-specific consumption deprivation index falls within the class of deprivation functions Atkinson suggested for a class of poverty indices.
3.4 Figure 3.1 below depicts the typical forms Engel curves could take
Total Expenditure
Expe
nditu
re o
n a
spec
ific
com
mod
ity
Figure 3.1Different Types of Engel Curves
Luxuries
Necessities
Taking an essential commodity for which the Engel curve is concave they postulated that the Engel curve would admit a saturation level of expenditure. As the normally used Engel curves in economics do not admit a finite asymptote they used a functional form that does admit such a finite asymptote, a form suggested by the saturation curves used in catalysis10. The deprivation function derived from the saturating Engel curve and the associated deprivation or poverty index are depicted in Figure 3.2 below:
10 They compared the new specification with some of the commonly used Engel curve specifications and found that the new specification was to be preferred on the basis of least error sum of squares and randomness and Normality of the estimated errors (Sitaramam et. al. (1996).
8
9
Total Expenditure
Expe
nditu
re o
n a
spec
ific
com
mod
ityFigure 3.2
Consumption Deprivation
Engel Curve for Necessities
Consumption Deprivationcurve
Deprivation or Poverty
Taking an essential commodity such as cereals we get the following commodity-specific consumption deprivation index (poverty index):
1.3)()(
0*
*
−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−⎟⎟⎠
⎞⎜⎜⎝
⎛ −= ∫
∞
dyyfC
yCCD
Where C* is the maximum value of consumption expenditure, and C(y) denotes the actual mean consumption at a given level of total expenditure y, and f(y) is the probability density function of the total expenditure. As the deprivation function is a monotonic decreasing function in y the above infinite integral converges and has a finite value.
3.5 We may list here some important aspects of this new measure of deprivation. While the poverty line is subjectively chosen by researchers the norm used in our deprivation index is determined by the socioeconomic setting in which a household is situated. It is objectively derived empirically from the observed Engel curve. To the extent that this is estimated from an estimated Engel curve it is subject to probabilistic errors. Kumar, Gore, and Sitaramam (1996) suggest several interesting statistical problems that arise as a result. This index is based on economic considerations and not on nutritional and other normative considerations. The index shares one feature that was inherent in Sukhatme’s approach to measuring hunger. This measure is crucially dependent on the distribution of “actual consumption deprivation” of an essential commodity in the community relative to the community-specific saturation norm (assumed to be
10
the same for all individuals in the community).11 The traditional measures of poverty are related to the consumption deprivation only through an indirect and weak link between income and or total expenditure and consumption deprivation. The index is such that if there are more persons or households with greater consumption deprivation greater is the contribution of that group to poverty. Thus this deprivation index measures severity of deprivation. We thus notice that the Engel curve based measure of deprivation takes care of all the problems of the traditional poverty indexes cited by the Expert Group and the criticism Lipton had leveled against them.
IV EMPIRICAL COMPUTATION OF POVERTY INDEX BASED ON ENGEL CURVE
4.1 We use the data on consumer expenditure at the household level provided by the National Sample Survey Organization (NSSO) for the 43rd, 5oth and 55th rounds covering 1987-88, 1993-94, and 1999-2000.12 We illustrated the proposed methodology in detail in a recent work (Kumar, Mallick, and Holla (2008)). The present work is an extension of that work. We examine the household level data in 15 major states of India, and also at All-India level, both in rural and urban settings. For completeness, and to make the presentation self-contained, we repeat some of the salient features of our empirical approach to the measurement of consumption deprivation. We also notice that the NSSO’s sample design is such that relatively richer households were oversampled. As the sample design is a multi-stage stratified sampling the data do not represent the population unless we take due note of the underlying sample design and use the corresponding multipliers.
4.2 The earlier study was undertaken with per capita monthly expenditure on cereals and per capita total monthly expenditure. As different households have different household compositions such per capita figures do not reflect comparable expenditures between households. This aspect again was not considered in our earlier study, although the Engel curve specification used allowed for shifts in the Engel curve due to changes in household size. It was mentioned in the earlier study that the food consumption pattern is different for different states and that there seems to be a substitution of cereals by some other food items. That study also did not include cereal substitutes which constitute the staple food for some poorer households in some states. In this study we take cereal substitutes also into account and include all food items in calculating food deprivation index as a poverty measure. We also convert the expenditure into adult equivalent scales.
11 The traditional measures of poverty use instead the distribution of income or total expenditure below the norm of a poverty line.
12 There are other data sources that one can use for estimating the extent of poverty. The data collected by National Nutrition Monitoring Bureau is one. The panel data collected for a few villages by ICRISAT is another. One may see Vaidyanathan (2001) for a discussion of relative merits of such different data sources.
11
4.3 Engel curves depict the consumption expenditure of a household against the total expenditure of the household. Different households have different household composition. There seems to be a priori reasons to suspect that the actual and normative levels of consumption vary by age and sex. Combining them by a simple addition does not reflect the underlying distribution of expenditure by age and sex. In most household surveys it is the total household consumption expenditure that is obtained and not individual expenditure by each member of the household. The issue of adult equivalent scale arises for two reasons. First, an adult male’s consumption living alone is different from his consumption when he is a member of a two-person household after marriage. This issue is referred to as economies of scale within a household. The other issue is simply that one cannot compare the per capita consumption of a four member household with two children with the per capita consumption of a four member households with all four being adults. Sabatese et.al. (2001) find statistical evidence for significant differences in consumption per person by age, by sex, and by country.
4.4 To address this issue we provide adult equivalent scales that are specific to India where the consumption expenditure of an adult male is taken as the reference and the food expenditures of adult females, male and female children are expressed as a multiple (fraction) of the food expenditure of an adult male. Usually some sophisticated theoretical justification is often given and theory is called upon to determine a methodology of deriving the adult equivalent scales. This justification is based on welfare comparisons between individuals and between households. What we compare is not the utility of consumption of food of one person with that of another person. Such interpersonal comparison of utilities cannot be done in general. We compare, instead, the consumption deprivation in terms of a shortfall in expenditure on food at constant prices. While we admit that such shortfalls have different disutilities at different income or total expenditure levels and hence cannot be added, we also view that in the absence of any known law of variable marginal utility of income we assume it to be constant justifying the addition of such nominal values of food consumption deprivation.
4.5 Two major considerations for choosing the adult equivalent scale are: (i) they must be stable so that comparisons can be made according to those scales across time and space, (ii) they must be specific to the country. There are several studies that examined the effect of using different adult equivalent scales on the poverty estimates and many poverty studies use adult equivalent scales. Rarely do we find any justification for the scales used. As there is ambiguity of what scales one should use, quite often researchers resort to only performing a sensitivity analysis of poverty measures to different adult equivalent scales. The most popular among these is the OECD scale that assigns the first adult a weight of 1, the second and subsequent adults a weight of 0.7, a child a weight of 0.5. There is no reason to assume that these widely used scales are applicable to India.
4.6 Sen and Sengupta (1983), in their study of malnutrition of rural children in West Bengal, observe possible sex bias as reflected in greater malnutrition of girl child compared to a male
12
child. In order to find scales that are suitable to Indian conditions we estimated a regression equation of household expenditure on food with the independent variables being total expenditure, the number of male and female adults, and male and female children, and household size square. Using the estimated regression line additional food consumption is calculated for adult male, adult female, male child, and female child. The adult equivalent scales are derived from these estimates. These are presented below:
The Adult Equivalent Scales for Food Expenditure in India (based on NSSO data)
1987-88-Rural 1 0.7591317 0.5215729 0.47118906
Urban 1 0.87164084 0.38990162 0.44167062
1993-94
Rural 1 0.80906785 0.55370166 0.47533897
Urban 1 0.88298132 0.52116718 0.37658138
1999-00 Rural 1 1.0373348 0.97217663 0.74000138
Urban 1 0.72746812 0.53903435 0.49041613
These AESs are based on All-India samples. The AES s of 1999-2000 seems to differ from those of the previous years. Using these estimates and the observations made by Sen and Sengupta to guide us we altered slightly the commonly used OECD scales. The AES s we used are: Adult male 1, adult female 0.8, male child 0.5 and female child 0.4. We used these adult equivalent weights in our analysis for determining food consumption expenditure per adult male equivalent and for replacing the household size by the number of adult male units in the household.
4.6 There is an issue of contamination of reported data in the 55th round of NSSO. This was
because two reference periods were used and the normally used 30 day reference period data
may have been adjusted by the respondent to correspond with the 7 day period response. The
effect of that on the estimates, and comparability of the estimates between the rounds was
examined by Deaton and Dreze (2002), Sundaram and Tendulkar (2003), Sen and Himanshu
(2004), and Deaton and Kozel (2005). In order to address this issue and make the results
comparable between the rounds for the 55th round data we used the correction factor for food
given in table 5 of Sen and Himanshu (2004).
13
4.7 We need price deflators to convert the expenditures between the rural and urban sectors,
between states, and between different rounds. Deaton and Tarozzi (2000), Deaton and Dreze
(2002), and Deaton (2003) discuss in detail regarding this issue. We employ the price indices
derived from NSSO data given by Deaton (2003) to deflate the estimates of cereal consumption
for comparison purposes between years, between states, and between rural and urban sectors.
The price deflators derived from Deaton (2003) are presented in Table 1 with All India Rural for
1993-94 as the base (100).
14
Table 1: The Price Deflators Used in the study (Derived from Deaton (2003)
4.8 Data do not come the way the theory conceptualizes it. While we expected food to be an
essential commodity that would saturate at a finite level of total expenditure, it is not confirmed
by the data. Adult equivalent monthly food expenditure plotted against adult equivalent monthly
total expenditure shows that at higher levels of total expenditure the proportion spent on food
increases. This is because the food expenditure includes food products purchased, and the rich do
spend a lot on purchased sweets and food preparations in restaurants. In addition, this could be
due to such affluent households spending more on food for guests and for family and religious
festivals.
4.9 In order to take care of this situation one must determine a cut-off point in total
expenditure where the Engel curve would turn from concave to convex. Using the notion that
any curve can be approximated by a polynomial of sufficient degree and that a cubic allows for
the curve to be both concave and convex in different segments, we estimated a third degree
polynomial between adult male equivalent food and total expenditure. We then determined a cut-
off point such that below that cut-off point the cubic fit would be concave with an asymptote13.
Subsequently we used households whose monthly per capita total expenditure is less than the
cut-off in further analysis14. The Total expenditure cut-off points so obtained empirically are
presented in Table 2.
13 The procedure was iterative. When we first determined the cut-off and estimated the cubic Engel curve again for the truncated sample it remained convex at higher ranges of total expenditure. Hence we repeated this process until we obtained an Engel curve that is entirely concave up to the cut-off level of total expenditure.
14 This procedure is equivalent to our treating this cut-off point as a poverty line. This is analogous to, but different from the suggestion of Rao (1981). Thus while Kumar et al. (1996) dispense with the focus axiom we bring it back here, as the data do not conform to the theoretical specification.
Table 2: Total Expenditure Cut-off Points Obtained from a Cubic Engel Curve Fitted for the Entire Sample
In Rs Per Adult Male Unit Per Month in 1993-94 Prices
4.11 We estimated all these functional forms. We found that in a majority of cases Banks-
Blundell-Lewbel form is the best fitting in terms of highest RP
2P. However, the saturating
functional form used by Kumar, Gore, and Sitaramam (KGS) (1996) has, in a majority of cases,
the second best RP
2P. Furthermore it is found that these RP
2P s associated with KGS specification are
quite large and the percent shortfall from the RP
2P of best-fitting (Banks-Blundell-Lewbel form is
negligibly small, the maximum percent shortfall being only 1.45%. The RP
2Ps associated with these
Engel curves, along with the maximum RP
2P and the shortfall of KGS specification are presented in
Tables 1a, 1b, and 1c of Appendix. We also found, using Normality tests, that the errors are
distributed as a Normal distribution. Hence we chose KGS specification of the Engel curve.TP
15PT A
TP
15PT Although we used adult equivalent scales the adjusted household size (zBi B) may still appear in the Engel curve as
the adult equivalent scale only corrects for scale in food expenditure and the relation between these two variables (Engel curve) may still be subject to economies of scale. In fact we did observe, in a majority of cases, a statistically significant negative coefficient for the adjusted household size.
typical picture of the type of fits obtained is presented below in Figure 4.1 for the fits obtained
for All India Rural 1993-94. The estimates of the parameters of the saturating KGS Engel curves
are presented in Tables A2a, A2b, and A2c of the Appendix.
Figure 4.1: The Relative Performance of Alternate Engel Curves
The type of fits obtained by the cubic Engel curve (BBL) and the saturating Engel curve of KGS
is depicted for 1993-94 All India Rural in Figure 4.2 below:
Figure 4.2: The Closeness of KGS and BBL Engel curves and the actual data
0
100
200
300
400
500
600
700
Axi
s Ti
tle
Graph of Saturating Engel Curve for 93-94-all-India Rural: Actual Vs Predicted
Actual
KGS
BBL
18
19
V ESTIMATES OF FOOD CONSUMPTION DEPRIVATION BASED ON ENGEL CURVES
5.1 We define poverty index as the food deprivation index of a chosen community (rural or
urban sector of a particular state). We term this food deprivation index a modified Sitaramam
6.1 Given the recent debate on the traditional poverty estimates in India we find it useful to
revive a decade and half old suggestion of Kumar, Gore, and Sitaramam to measure poverty
using an Engel curve for essential commodities. We used household level data from three recent
large consumer expenditure surveys of NSSO and estimated food consumption deprivation. We
showed that this index has very little spatial correlation with the traditional measures of poverty.
6.2 NSSO data has some limitations but it seems to be the best available data. The imputed
values for home produce and meals eaten outside, the meals served for guests and domestic help
etc. leave sufficient room for subjective judgments, and quite often the field investigator may be
prompting the respondent on what prices to use for such imputations. There is a need to do
alternate pilot surveys and studies to focus on individual questions of this sort.
6.3 One may estimate the Engel curves for some other necessities such as health services,
primary and secondary education, water, fuel and light, etc. and estimate the associated
consumption deprivations. The question then arises as to how to combine them into a single
index of poverty. The study by Nathan, Mishra and Reddy (2008) provides a measure based on
Euclidean distance between the target or norm and actual multidimensional deprivation. Human
development indices and poverty indices can be brought into the manifold of microeconomics of
consumer demand through Engel curve analysis of essential goods and services.
REFERENCES
Atkinson, A.B. (1987), ‘On the measurement of poverty’, Econometrica, 55: 749–764.
Banks, J., R. Blundell, and A.Lewbell (2006),”Quadratic Engel Curves and Consumer Demand”, Review of Economics and Statistics, Vol. 79, pp.527-539.
Behreman, J. and Anil Deolalikar (1987), “Will Developing Country Nutrition Improve with
Income? A Case Study for Rural South India”, Journal of Political Economy, Vol.95, pp.492-
507.
Dandekar, V.M, and Nilkanth Rath (1971), Poverty in India, Indian School of Political
Economy.
28
Deaton, A. (2003), “Prices and Poverty in India: 1987-2000”, Economic and Political Weekly, 38 (4): 362-368.
Deaton, A., and Jean Dreze (2002), “Poverty and Inequality in India”, Economic and Political
Weekly, 37 (36): 3729-3748.
Deaton, A., Kozel, V. (2005), ‘Data and dogma: The great Indian poverty debate’, World Bank Research Observer, 20 (2): 177-199.
Deaton, A. and C. Paxon (1998), “Economies of Scale, Household Size, and the Demand for Food”, Journal of Political Economy, 106 (5): 897-930.
Deaton, A. and A. Tarozzi (2000), “Prices and Poverty in India”, Princeton Research Program in Development Studies, Available from http:www.wws.princeton.edu/~rpds.
Foster, J., J.Greer, and E. Thorbecke (1984), “A Class of Decomposable Poverty Indices”,
Econometrica, Vol. 42.No. pp.
Government of India (1993), Perspective Planning Division, Planning Commission, Report of the
Expert Group on Estimation of Proportion and Number of Poor.
Kumar, T.K., A.P. Gore and V. Sitaramam (1996), ‘Some conceptual and statistical issues on measurement of poverty’, Journal of Statistical Planning and Inference, 49 (1): 53-71.
Kumar, T.K, S.K. Mallick, and Jayarama Holla (2007), “Estimating Consumption Deprivation in
India using Survey Data: A State-Level Rural-Urban Analysis before and during Reform
Period”, Working Paper No.7 Centre for Globalization Research, Queen Mary’s, University of
London. (Forthcoming in Journal of Development Studies 2008)
Lewbel, A. (2006), “Engel Curves”, in New Palgrave Dictionary of Economics, Second Edition Lipton, M. (1997), “Editorial-Poverty: Are There Holes in the Consensus”, World Development, Vol. 25, No.7, pp.1003-1007.
Meenakshi, J.V. and B. Viswanathan (2003), “Calorie Deprivation in Rural India: 1983-1999-200”, Economic and Political Weekly, January 25, pp.369-375.
Nathan, Hippu Salk Kristle., Srijit Mishra, and B. Sudhakar Reddy (2008), “An Alternative Approach to Measure HDI”, January, Working Paper 2008-001, Indira Gandhi Institute for Development Research, Mumbai. http//www.igidr.ac.in/pdf/publication/WP-2008-001.pdf
Pradhan, M. and M. Ravallion (2000), ‘Measuring Poverty Using Qualitative Perceptions of Consumption Adequacy’, The Review of Economics and Statistics, 82 (3): 462-471.
Rao, Bhanoji, V.V. (1981), “Measurement of Deprivation and Poverty Based on the Proportion Spent on Food: An Exploratory Exercise”, World Development, 9 (4): 337-353.
Rath, Nilkanth. (1996), “Poverty in India Revisited”, Indian Journal of Agricultural Economics, Vol. 51(1&2), Jan-June, pp. 76–108.
Rath, Sharadini. (2003), “Poverty by Price Indices”, Economic and Political Weekly, 38 (40): 4260-4268.
Sabatese, Ricardo, Brian W. Gould, and Hector Vellareal (2001), “Household Composition and Food Expenditure”, Food Policy, Vol. 26, pp. 571-586.
Sen, Abhijit, and Himanshu (2004), “Poverty and Inequality in India-I”, Economic and Political Weekly, September 18. Pp.4247-4263.
Sen, Amartya and S. Sengupta (1983), “Malnutrition of Rural Children and the Sex Bias” Economic and Political Weekly, Annual Number, May, pp. 855-864.
Sen, Pronab (2005), “Of Calories and Things: Reflections on Nutritional Norms, Poverty Lines, and Consumer Behaviour in India”, Economic and Political Weekly, October 22, pp. 4611-4618.
Sitaramam, V., S.A. Paranjpe, T.K. Kumar, A.P. Gore and J.G. Sastry (1996), ‘Minimum Needs of Poor and Priorities Attached to them’, Economic and Political Weekly, Special Number, 31 (35-37): 2499-2505.
Subrananiam S. and A.S. Deaton (1996), “Demand for Food and Calories”, The Journal of Political Economy, 104 (1): 133-162.
Sundaram and Tendulkar (2003), “Poverty Has Declined in 1990s: A Resolution of Comparability Problem in NSS Consumer Expenditure Data”, Economic and Political Weekly, 38 (4): 327-337.
Sukhatme, P.V. (1961), “The World’s Hunger and the Future Needs of Food Supplies”, Journal of the Royal Statistical Society, Series A, Vol 124, pp.463-525.
Table A2a: Results of Maximum Likelihood Estimation of Nonlinear Saturating Engel Curve (KGS) All India and 15 Major States :1987-88 RuralFile A SE(A) B B(SE) C C(SE) D D(SE) R2
Table A2.b: Results of Maximum Likelihood Estimation of Nonlinear Engel Curve All India and 15 Major States :1987-88 UrbanFile A SE(A) B B(SE) C C(SE) D D(SE) R2
Table A2.c: Results of Maximum Likelihood Estimation of Nonlinear Engel Curve All India and 15 Major States :1993-94 RuralFile A SE(A) B B(SE) C C(SE) D D(SE) R2
Table A2d: Results of Maximum Likelihood Estimation of Nonlinear Engel Curve All India and 15 Major States :1993-94 UrbanFile A SE(A) B B(SE) C C(SE) D D(SE) R2
Table A2e: Results of Maximum Likelihood Estimation of Nonlinear Engel Curve All India and 15 Major States :1999-2000 RuralFile A SE(A) B B(SE) C C(SE) D D(SE) R2
Table A2f: Results of Maximum Likelihood Estimation of Nonlinear Engel Curve All India and 15 Major States :1999-2000 UrbanFile A SE(A) B B(SE) C C(SE) D D(SE) R2