essays on some determinants of food-security and
TRANSCRIPT
Essays on Some Determinants of Food-Security and
Consumption of Nutrients in India
by
Amlan Das Gupta
A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF
THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
in
The Faculty of Graduate and Postdoctoral Studies
(Economics)
THE UNIVERSITY OF BRITISH COLUMBIA
(Vancouver)
October 2014
© Amlan Das Gupta, 2014
ii
Abstract
This thesis is motivated by the unique experience of India regarding economic growth and
the corresponding impact on nutritional status. Despite sustained periods of growth,
malnutrition levels in India have shown modest improvement. Moreover, average calorie
consumption in the country is going down even as consumption expenditure and incomes go
up. The objective of this thesis is to shed light on some possible causes of this puzzling
phenomenon. The second chapter is a theoretical exploration of the possibility that
preference for conspicuous consumption could be a factor contributing to this decline in
calorie consumption. This chapter starts by incorporating status seeking preferences in a
dual-economy general equilibrium model and then demonstrates that, in such a setting,
economic growth could lead to a fall in calorie consumption across the income distribution
even with incomes rising at the same time. In Chapter 3 the implications of the main
assumption of "keeping up with the Jones" preferences, from the theoretical model of
Chapter 2, is tested in the data. This assumption implies that household calorie consumption
should decline with peer group income. So the effect of peer group income on calorie
consumption is estimated using World Bank data collected from rural India. Using these
estimates it is roughly estimated that Veblen competition can account for more than a third of
the missing calories. A unique source of variation in peer group income, based on caste-wise
domination across villages, is used for identification. The fourth chapter looks at the impact
of the public food distribution system (PDS) in India, on the household per capita
consumption of calories and proteins. This effect is identified using random shocks
introduced into the delivery system of PDS through the impact of rainfall on agricultural
output in the state which is the largest supplier of grains to the system. The results suggest
that a rise in PDS performance has different effect in different regions in the country. Yet, for
those who benefit from this system, the impact on nutrient consumption and malnutrition is
significant and large.
iii
Preface
The identification and design of this research program and the performance of the various
parts of the research, and analysis of the research data was done by the author.
iv
Table of Contents
Abstract ................................................................................................................................................. ii
Preface .................................................................................................................................................. iii
Table of Contents ................................................................................................................................ iv
List of Tables ........................................................................................................................................ v
List of Figures ...................................................................................................................................... vi
Acknowledgement .............................................................................................................................. vii
Dedication .......................................................................................................................................... viii
1. Introduction ...................................................................................................................................... 1
2. Dual-economy Model with Veblen Preferences ............................................................................. 7
3. Empirical Estimation of Veblen Induced Peer Income Effect on Calorie Consumption ....... 37
4. Impact of Public Distribution of Food in India on Consumption of Nutrients. ....................... 68
5. Conclusion ..................................................................................................................................... 108
Bibliography ..................................................................................................................................... 112
Appendix to Chapter 2 ..................................................................................................................... 118
v
List of Tables
Table 2.1: Simulation, Rise in Industrial Productivity ........................................................... 23
Table 3.1: Summary Statistics ...................................................................................................... 41
Table 3.2: Individual Characteristics by Caste ........................................................................ 42
Table 3.3: Village Characteristics by Caste Dominance of Villages .................................... 51
Table 3.4: Individual Characteristics by Village Dominance ............................................... 52
Table 3.5: OLS Estimates of Peer Income on Calorie Consumption .................................. 54
Table 3.6: OLS Estimates on Sample of Low Castes Only .................................................... 55
Table 3.7: IV First Stage, Village Caste Dominance on Peer Income ................................ 56
Table 3.8: IV Second Stage, Peer Income on Calorie Consumption ................................... 58
Table 3.9: Testing for Sanskritization ........................................................................................ 62
Table 3.10: Second Stage IV Using Non-Food Expenditure .................................................. 63
Table 3.11: Comparison of IV Estimates Using Veblen Calories to Total Calories ........ 64
Table 4.1a: Summary Statistics Main Variables ...................................................................... 82
Table 4.1 b: Summary Statistics (control variables and household characteristics) ...... 84
Table 4.2: OLS Estimates, PDS Uptake on Nutrient Consumption .................................... 87
Table 4.3a: Despatches of PDS Grain from Other States by Zone ...................................... 94
Table 4.3b: PDS Grain Despatches from Donor States .......................................................... 95
Table 4.4: Changes in Key Variables Over Time, North-East Zone ................................... 98
Table 4.5: IV Estimates, Quota-Uptake Difference on Nutrient Consumption .............. 103
Table 4.6: Change in Quota-Uptake Gap and Calorie Consumption Over Time .......... 104
Table A.1: Simulation: Effect of Rising Productivity on Food Demand ......................... 128
vi
List of Figures
Figure 1.1: Percentage of Underweight Children Less Than 5 Years................................... 2
Figure 2.1: Path of Food Demand as Productivity Grows ..................................................... 21
Figure 2.2: Calorie Engel Curves ................................................................................................ 33
Figure 4.1: Local Polynomial Estimates of PDS Grain as Function of Income ................ 75
Figure 4.2: Relationship Between Income and PDS Grain (Schematic) ........................... 76
Figure 4.3: Histogram of Income, Sample of PDS Users........................................................ 77
Figure 4.4: Demand and Supply Curves for PDS Grains ...................................................... 78
Figure 4.5a: PDS Rice Procurement from Major Supplier States ....................................... 89
Figure4.5b: PDS Wheat Procurement from Major Supplier States ................................... 90
Figure 4.6: Rice Procurement by PDS Zones ........................................................................... 96
Figure A.1: Path of Food Demand as Productivity Grows .................................................. 129
Figure A.2: Path of Food Demand as Productivity Grows (labourers) ............................ 130
Figure A.3: Path of Food Demand as Productivity Grows (landowners) ........................ 130
vii
Acknowledgement
For his essential guidance and help with this thesis I would like to thank my supervisor Dr.
Mukesh Eswaran and also my supervisory committee members Dr. Patrick Francois and Dr.
Kevin Milligan. I am also grateful for useful comments and suggestions from Dr. Siwan
Anderson, Dr. Kaivan Munshi, and the participants of the empirical workshop at the
University of British Columbia, especially Dr. Thomas Lemieux and Dr. Nisha Malhotra. I
have also benefitted greatly from discussions with my fellow graduate students.
1
Chapter 1: Introduction
In its path towards development, India is perhaps one of the most closely followed countries
in the economics literature. In the past two decades the Indian economy has been growing
rapidly. Now it is often classified along with China as one of the emerging economies with
the potential to rival the economic clout of the western developed nations. However, unlike
China, India is still suffering under the heavy burden of extensive poverty and malnutrition.
Food, primary healthcare and basic sanitation are still out of reach for large sections of the
people. This thesis concentrates on one such basic aspect of development in India, namely,
the nutritional status of the Indian population.
In many ways the Indian experience regarding nutrition has been particularly interesting if
not unique. There are some clear patterns to the trends in nutritional intake in India that
seems to challenge conventional economic logic. One such trend is the very slow decline in
malnutrition in the country despite rapid economic growth. As a result of this, the
malnutrition figures in India are still comparable to the poorest regions of the world (See Fig
1.1) although it is far ahead in terms of both income and other development indicators.1 The
state of nutrition, however, has reached a critical stage where it has ceased to be a
humanitarian problem restricted to the very poor and has started to impinge upon the
economic/productive capacity of the whole country. According to a World Bank report
published in 2005 (Gragnolati et al. 2005) micronutrient deficiencies alone may cost India
US$2.5 billion annually.
1 However Panagariya (2013) has questioned the validity of the reference group used by the WHO on the basis
of which the malnutrition figures are calculated. He argues that the reference group is genetically different from
Indians.
2
The second stylised fact about nutrition in India is that on an average calorie and protein
consumption is declining in the population. This fact becomes even more surprising when
viewed in conjunction with the first. However this is quite established in the literature now
and is sometimes referred to as the calorie puzzle of India (Patnaik 2007, Deaton and Dreze
2009). Studies have also shown that increasing proportions of the population are slipping
below the standard of calorie2 consumption being used by the government to calculate the
poverty line (Mehta et al. 2000).
.
2 These are 2100 Kilocalories for the urban areas and 2400 Kilocalories for the rural areas.
3
Latest estimates using the National Sample Survey data show that in 2009-10 even the
average calorie consumption for the whole population was more than 15 percent lower than
these standards.
Another characteristic of calorie consumption in India was pointed out by Deaton and Dreze
(2009) when they showed that calorie Engel curves for India are upward sloping and yet
shifting downwards over time. This means that the fall in calorie consumption alluded to
earlier is happening not only at the average but at every point in the income distribution,
which suggests that the reason behind this phenomenon is not just impoverishment.
Something has happened which is turning tastes and preferences away from food. For a
country like India with the current levels of child malnutrition such trends are an alarming
sign. In fact this threatens to stymie and even reverse the modest rates of decline in
malnutrition achieved till now.3
This thesis aims to address all of the above stylised facts on Indian nutrition by looking at
some of the factors which may have caused these general trends to appear. The next two
chapters exclusively concentrates on the last stylised fact by looking at a hitherto unexplored
reason for the fall in calorie consumption throughout the income distribution. This is the
increased desire for conspicuous consumption driven by competition for status within ones'
peer group. In an environment where rapid economic growth is pushing up incomes there is a
possibility that agents constantly feel pressured to keep pace with the consumption of their
neighbours. If this pressure is strong enough we might see a reduction in consumption of
essentials like food. In order to lend support to this theory two important questions need to be
3 According to the NFHS 3 report percentage of children under 5 years deemed to be wasted has actually gone
up in 2004-5 compared to 1998-99. However in terms of stunting and underweight figures, modest gains have
been observed over the same period.
4
answered. Firstly, it needs to be ascertained whether such preferences for conspicuous
consumption indeed exists in India, even for the poor. And, secondly, we must examine
whether the negative effect on calorie consumption, which this is liable to generate, is strong
enough to overturn any other factors (for example rising incomes) that may be helping
calorie consumption.
Chapter 2 addresses the second question as described above. It starts by looking at the
implications of "Keeping up with the Jones" (KUJ) preferences on the choice between
conspicuous consumption and non-conspicuous consumption goods (food). These
preferences imply that the marginal utility of conspicuous consumption would now be rising
with the average conspicuous consumption in the agent's peer group. In a partial equilibrium
setting KUJ preferences directly translate into a negative relationship between non-
conspicuous consumption and peer group income of the consumer. As the next step, these
KUJ preferences are incorporated into the demand side of a Dual Economy general
equilibrium model where prices, incomes and peer group incomes are endogenously
determined. In this setting it is shown that productivity driven economic growth could lead to
a fall in calorie consumption even as incomes go up. One caveat here is that the income
effect on food consumption needs to bounded. If this is too high it might overturn the
negative effect of rising peer group income generated by the KUJ preferences. However,
whether the income effect is smaller than the KUJ effect in reality is a question that may only
be answered through empirical investigation.
Chapter 3 is primarily an empirical exercise aimed at answering the two questions that
emerge after the theoretical analysis in Chapter 2. Firstly, do KUJ preferences exist, even
amongst the poor rural population in India? Secondly, even if KUJ exists is it strong enough
5
to explain the falling calorie consumption? In particular, is the negative effect of peer group
income on calorie consumption, as generated by KUJ, larger than the positive own income
effect? As a first step it is argued that in a rural Indian setting, own caste members living in
the same village are a good approximation for the peer group that such an agent would be
exposed to. Next using a World Bank household and village level survey data the effect of
this peer group's income on household calorie consumption is estimated. Unlike most peer
effects estimation self - selection is not an issue here as caste members are born into their
peer groups as per the rules of the Hindu caste system. However, the small sample size of
this dataset introduces significant classical measurement error into the peer group income
measure. To deal with this problem a unique source of variation in peer group income based
on village level land ownership pattern is used for identification. The estimates indicate that
not only is there a negative and significant effect of peer group income on calorie
consumption, it is in fact almost twice as big as the positive own income effect. Using the
estimates it is roughly estimated that about a third of the missing calories may be accounted
for through this channel.
The relevance of status competition in the determination of choice of food amongst
households, as pointed out in Chapters 2 and 3, is not just of theoretical interest. It also
suggests that any policy intervention whereby income supplements are provided to all
households to boosts their nutrient intake would, in all likelihood, not have the desired effect.
This observation is extremely relevant to the current debate regarding the efficacy of the
Public Distribution System (PDS) of food grains in India. The PDS is the biggest and most
extensive food security program of the Indian government. Its function is to collect all grains
offered up for purchase at the government determined minimum price and then store this
6
stock for distribution at subsidised prices through its vast network of fair price shop.
However the system has often come under criticism for being highly inefficient as well as
ineffective. One area of strong criticism has been the leakage of PDS grain into the black-
market while being transported or from storage. This is touted as the main reason why most
eligible households are unable to buy their full quota of grains. Many are of the view that the
government should relinquish the responsibility of transporting the grain and just pass on the
intended amount of subsidy to eligible households as cash transfers.
Chapter 4 attempts to contribute to this debate by looking at the impact of PDS grain
consumption on nutrient consumption. The identification strategy relies on the unique nature
of PDS grain consumption which is often restricted by lack of supplies rather than by the
amount of the quota or the lack of demand. This implies that any exogenous shock to PDS
supply while keeping demand constant would directly affect amount of PDS grain consumed.
So random weather shocks in the states that supply the bulk of PDS grain is used as
identifying variation. Interestingly, the estimation results do not provide clear cut evidence in
favor of PDS. However results from the strongest specification suggests that there is a
positive impact on calorie consumption. Back of the envelope calculations based on these
estimates indicate that scrapping the PDS and replacing it with a cash transfer equivalent to
the subsidy would negatively impact calorie consumption by 1-9 percent depending on
assumptions.
7
Chapter 2: Dual-economy Model with Veblen Preferences
2.1 Introduction
Recently there has been a lot of concern and surprise at the finding that per-capita calorie
consumption in India has been falling in spite of rising incomes over the last couple of
decades or so. This chapter attempts to provide an explanation to this phenomenon on the
basis of Veblen preferences. It starts with a simple dual-economy setting with status seeking
preferences. Later on an income distribution is introduced into the model to highlight the
effects of income inequality on status seeking. The chapter demonstrates the theoretical
possibility of the calorie reduction phenomenon.
As mentioned in Chapter 1 the phenomenon of falling calorie consumption across the income
distribution was demonstrated in the most detailed manner by Deaton and Dreze (2009).
Their paper also explore many possible explanations for this puzzle. They show, for
example, that it is not driven by either food prices or fall in incomes. They also rule out the
possibility that it is driven by a budget squeeze on food due to the introduction of new non-
food essentials to the choice set (like education and sanitation). This theory does not seem
persuasive to them since if it were the case then we would expect people to shift towards
cheaper calorie sources. Instead people are shifting to costlier foods. Two possible
explanations that the authors feel might be having some merit are that either people’s tastes
and preferences have changed, or their calorie requirements have fallen due to the
introduction of machines in all walks of life. The latter explanation was tested in Li and Eli
(2010), where the authors constructed measures of calorie requirements based on activity
8
levels reported by survey respondents. But although this explained a lot of the changes in
food composition it could not explain the fall in calorie consumption over time. The question
that arises is what other explanation could be plausible?
One such explanation is suggested by Banerjee and Duflo in their book The Poor Economics
(2011). Here it is pointed out that the poor may be misinformed about the nutrient values of
different foods. So they may be spending more on fats and less on cereals without the
knowledge that the total nutrition intake is falling due to this switch. It is very plausible that
when people get richer they would switch to tastier foods, but what is strange is that people
often reduce their overall calorie intake while doing this. This would suggest that there may
be some factor that urges people to shift to costlier, tastier calories more often than is healthy
for them.
One interesting observation is that the fall in average calorie consumption is not particular to
India. This phenomenon has been observed previously in England during the industrial
revolution (1770-1850) and also in China more recently (Clark et al. 1995, Du et al. 2002
and Meng et al. 2009). In all these situations, the economy at the time was experiencing high
growth rates and incomes were going up. This leads one to suspect that rising incomes might
have something to do with lowering demand for food over the entire income distribution.
Another paper by Subramanium and Deaton (1996) offers some interesting insights about
this. This paper attempts to verify empirically the notion that poverty leads to low nutrition,
which in turn causes people to have low productivity and receive lower wages. They find,
using National Sample Survey data from the Indian state of Maharashtra that the required
calories for a poor worker to lead a healthy life can be bought with about 5% of their daily
wages. So they conclude that income is not a constraint on nutrition; rather it is the other way
9
round. The main reason why the elasticity of food remains very low is due to the substitution
of cheaper calories to more expensive ones. One possible way this could happen is through
higher conspicuous consumption by agents with a status-seeking motive.
In 1899 in his book titled The Theory of the Leisure Class Thorsten Veblen introduced the
idea of conspicuous consumption into economics. Although similar ideas had been voiced by
Adam Smith (1759) before, Veblen was perhaps the first to recognize the role of visible
consumption as a signal of one’s status. Status seeking behaviour is very common amongst
the human race and one of the most important determinants of status is wealth. While it is
difficult to communicate one’s true wealth level by one’s appearance, visible consumption
may provide hints that are easily picked up by others. In some cases people might even be
able to make themselves appear wealthier than they really are.
This status-seeking behaviour has potentially significant consequences for the economy (as
shown in Coelho 1985). In particular when all members of society are experiencing growth
in income, status competition might get intensified so much so that agents would be willing
to reduce food consumption in order to supplement the consumption of status goods. A
similar phenomenon has been noted in outcomes in a general equilibrium framework in the
literature already. Kelly (2005) and Cooper et al. (2001) have used this phenomenon to
motivate economic growth in their models. In these models growth is driven by endogenous
investment in newer varieties of either a normal or a Veblen good.4 Higher income leads a
person to buy more status goods, which in turn causes more investment in them. They show
that in equilibrium all new investment goes into the Veblen sector and results in lowering of
utility with economic growth. Hopkins and Korienko (2004) in a game theoretic model,
4 A Veblen good may be defined as one which provides no utility except as a status enhancement.
10
investigate the consequences of a change in income distribution in the presence of Veblen
preferences. They show that when the society gets richer almost everybody consumes more
of the Veblen good and utility declines at every income level. While investigating the
relevance of Veblen goods for the well known “Happiness Paradox”,5 Eaton and Eswaran
(2009) conclude that not only is the Veblen hypothesis capable of explaining the paradox,
but it should not be restricted to the developed world only.
Insight from these papers suggests that status seeking preferences have the potential to
generate ex-post utility reducing decisions from agents. But, in order to investigate the
possibility of calorie reduction in India we need to look at the consequences of adding
Veblen preferences to a general equilibrium model of a developing economy. For this
purpose I look at the dual economy model. The concept of dual economy, first introduced by
Arthur Lewis in 1954, is frequently used to model developing economies. This classic model
tells the story of development economies where better technologies in the industrial sector
draws out labour from the agricultural sector, eventually raising incomes in both sectors.
This model naturally introduces heterogeneity in the economy which is a setting where
Veblen competition becomes more potent. Here (section 2.2 below), I present a model based
on Eswaran and Kotwal (1993) dual economy setup. I demonstrate that under certain
conditions Veblen preferences do generate falling food consumption with economic growth
in equilibrium. Then in section 2.3 the same model is presented in a small open economy
setting. Section 2.4 has a slightly more complicated version of the model with the addition of
capital as a factor of production. This model serves to highlight the situation when rising
5 This refers to a finding, mostly in developed countries, that happiness measures do not go up as fast as wealth.
Easterlin (1974, 1995)
11
income inequality becomes essential for the calorie puzzle. The last section concludes and
discusses the implications of the findings.
2.2 The Basic Model
We start with a description of the preference structure. Suppose that the agent has a choice
between two goods X and Z. Here X is a non-Veblen good which we take as food, and Z is
the Veblen good. Next we make two assumptions which will define the preference structure
for the agents.
Firstly it is assumed that the utility function of the agents are quasilinear in food (X) and the
Veblen good (Z). This kind of preference may be written as follows:
Here lower case denotes quantities of these goods. The function may be any standard
increasing and strictly concave function with
; the next term is a linear
function of the Veblen good Z. This form, although not crucial for the results, helps to
highlight the workings of the Veblen preferences.6 A feature of this utility function is that at
lower levels of consumption of X, marginal utility per dollar spent will be higher than that
from the Veblen good (which does not change). So until an agent’s marginal utility per dollar
from X has fallen to a constant (which is equal to divided by the price of the industrial
good) there will be no consumption of Z. After the marginal utilities per dollar have been
equalised, all extra income will go into consumption of Z. This feature is a very apt
description of preferences in poorer areas where initially agents devote most of their income
to subsistence requirements. Luxury goods and other non-necessary goods are consumed
6 See the appendix for an analysis of the model with the quasilinearity assumption relaxed.
12
only after the basic needs have been fulfilled. This assumption also implies that Engel curves
for food will not be upward sloping after a certain threshold of income.7
The next assumption is that preferences for the Veblen good Z exhibits "Keeping up with the
Jones" (KUJ) a term coined by Gali (1994) and later formalized further by Dupor and Liu
(2003). Unlike the previous assumption this one is crucial and is the main driver of the
results. According to the formalization of this concept presented in Dupor and Liu (2003),
preferences for Z exhibits KUJ8 if the marginal utility of Z consumption is rising in the
average Z consumption of the agent's peer group .
So we have:
In order to maintain quasilinearity the second derivative of the Veblen function with respect
to the Veblen good Z is assumed to be 0.
Partial Equilibrium
Next I present a simple partial equilibrium result that follows from my assumptions on the
preferences. Assume that food X is the numeraire and the price of the Veblen good Z is p.
7 Flat Engel curves can be achieved, as in Eswaran Kotwal (1990), using hierarchical preferences that are
lexicographic. But quasilinear specification is more tractable as it implies that food demand would respond to
price and in the presence of Veblen preferences also to own income. 8 There is another aspect to preferences involving the Veblen consumption or relative consumption. This is the
differentiation between jealousy and admiration (Dupor and Liu (2003)). For example, if the agent feels jealous
that people in the neighbourhood have fancy cars her utility will be falling with a rise in average consumption
of fancy cars. This would imply that . On the other hand if she admires the higher consumption of fancy
cars in the neighbourhood the preference should have . The results in the present model will go through
with either of these assumptions as long as I have quasi-linearity and KUWJ.
13
Proposition 1: Consumption of the non-Veblen good (X) falls with the average income of
the group once the agent has started consuming Veblen good (Z), ceteris paribus.
Proof: See Appendix.
The proposition implies that if there are two similar individuals with the same incomes and
facing the same prices, the one living in a richer surrounding will consume less food. This is
a direct implication of “Keeping up with the Jones” aspect of the preferences. When an agent
who has already started consuming both goods moves to a richer peer group, she essentially
move to a group with higher average Z consumption. This drives up her own marginal utility
of Z consumption, prompting her to substitute away from food. This is the feature which
drives the results in this model. This is also the point which is taken to data in subsequent
empirical Chapter 3 and used as a test for the existence of KUJ for the agents in the sample.
General Equilibrium
The partial equilibrium result derived in Proposition 1 above holds own income and prices
constant. But during the period when average calorie consumption in India was falling,
average consumption expenditure and incomes were increasing rapidly (Deaton and Dreze
2009). Since the objective here is to explain the actual occurrences of that period, it is
important to allow for the simultaneous movement of own income and peer group income
(relative food prices were roughly constant) and examine whether calorie consumption can
still fall. This, of course, can be theoretically modeled only in a general equilibrium
framework, which is introduced below.
For the general equilibrium analysis a specific quasi-linear utility function with KUJ is
assumed. It is customary in the literature related to Veblen consumption to model the utility
14
function as increasing functions of or (see Eaton and Eswaran 2009). In general
the Veblen function is assumed to be strictly concave. But, as discussed above, the particular
context of rural/poor households here makes a quasi-linear specification ideal. With the
removal of the quasi-linear assumption the general equilibrium results presented below are
valid under a different set of conditions. More details are provided in the Appendix.
The specific functional form assumed for an individual i for the sake of tractability is as
given below:
Using these preferences9 it is possible to calculate the demand as a function of price, own
income and group income assuming food X to be the numeraire:
Demand for Z is given by:
9 Note that this specification displays jealousy as long as . Beyond this level of Z consumption we
switch to admiration. But this feature is not relevant for the results.
15
Supply Side
This section outlines the production system and resources. The population, which is also the
labour force, is normalized to 1, proportion of this population is landless and the rest own
an equal share of the total land. All agents supply labour inelastically. The two goods are
produced according to the following production functions:
Food production requires both land and labour inputs with decreasing returns to labour. Z is
assumed to be an industrial good which only uses labour as input and exhibits constant
returns. Here represents land which is a fixed factor, owned equally by the landlords and
fully used in the production of X. and are the labour allocations to and A and B are the
productivity parameters for the agricultural and industrial sector respectively. This completes
the description of the supply side.
Demand functions have already been calculated, so the last thing is to define the peer group
of the agents. To keep things simple I assume that the peer group g is the same for all agents
and comprises of the entire economy. So now is the average peer
group income used in Veblen comparisons where and are the incomes of the landless
labour group and the landowner group respectively.
The equilibrium of this model is defined by the prices and labour allocation where all
markets clear. But due to the unique nature of preferences assumed, demand for grain is
kinked in income (see expression for food demand in equation 2.1 above). This gives us
16
three different types of equilibria, and which one obtains depends on the magnitudes of the
productivity parameters for the two sectors. In the first equilibrium both landlords and
landless agents are too poor and cannot afford the Veblen good Z, in the second only the
landlords are rich enough to consume Z, and in the third both types of agents have become
satiated with food and have started consuming Z. In our subsequent analysis we will focus on
the third sort of equilibrium since our objective is to characterize the effect of conspicuous
consumption on food consumption it is essential that agents do consume some Z to bring
them into the analysis.
Effect of Productivity Improvements
Next we conduct some comparative statics exercises in the model by raising the agricultural
and industrial productivity parameters (A and B respectively).
Proposition 2: In an equilibrium in which all agents are satiated with food (X), equilibrium
allocation of labour to the agricultural sector ( will fall with any rise in either agricultural
or industrial productivity (A or B respectively).
Proof: See Appendix.
Intuitively, when agricultural productivity (A) receives a positive shock, incomes rise. So
demand for the industrial Veblen good Z goes up (since agents are assumed to be satiated
with food and devote all further income to the consumption of luxury goods). This draws
labour out of the agricultural sector in order to cater to the increased demand for Z. Similarly,
when industrial productivity (B) goes up it makes the value of the marginal product of labour
in the industrial sector higher than that in the agricultural sector (creating an imbalance). This
17
also leads to movement of labour out of the agricultural sector into the industrial sector. In
either case, labour is drawn out of the agricultural sector raising the land to labour ratio and
increasing wage rate. This result is similar in spirit to Eswaran-Kotwal’s (1993) results but
there is an extra dimension introduced by the presence of status seeking behaviour. The rise
in income raises average consumption of the Veblen good in the economy. This further
pushes up demand for Z (due to KUJ), leading to magnified effects on the wage rate as
compared to a model without Veblen competition. So Veblen preferences can facilitate a
faster reduction in poverty, however it can have a deleterious effect on food/calorie
consumption as we see below.
A Possible Resolution to the Calorie Puzzle:
The next proposition investigates the possibility of a calorie reduction as documented by
Deaton and Dreze (2009), among others, in Indian data. As mentioned earlier, they have
shown that the Engel curves for calorie consumption have been shifting down over time even
when the relative price of the said good is constant; so calorie reduction has occurred across
nearly all income groups. In order to investigate such a possibility in the present model, two
things need to be ensured. Firstly, the relative price of the industrial good has to be kept
constant, and secondly incomes of both groups need to be rising. According to Deaton and
Dreze (2009) these were the circumstances prevailing in the Indian economy at the time the
calorie reduction was observed (1985 – 2005). In this model we can derive the relative price
of food as a function of parameters and the labor allocation to agricultural sector from the
labour market equilibrium condition (equating marginal products in the two production
sectors) and is given by:
18
In the equilibrium where everybody is consuming the Veblen good Z, a rise in the
productivity parameters B or A will lower labour allocation to agriculture (Proposition 2).
So it is possible to change A and B in such a way that price remains constant. We have no
way to endogenise the changes in productivity so we will assume here that productivity has
changed keeping the relative price constant. The second point about income is also important
because, as mentioned before, the Deaton and Dreze (2009) paper actually show that calorie
Engel curves are shifting down over time. Although it is not guaranteed in this model that
any productivity shock will raise incomes for all, we can however, show the consequences on
food consumption in the event that all incomes rise.
Proposition 3: In an equilibrium where all are satiated with food, if agricultural and
industrial productivity go up simultaneously keeping relative price of the industrial good
constant, and if landlord’s incomes do not fall, food consumption for both groups fall.
In order to see this, observe the food threshold for both groups after they are satiated with
grain, from the demand for food given in equation 1:
Substituting in the value of using the production function of Z (equation 3) we have:
19
The shock applied to the system is such that price is constant and B (industrial productivity)
is raised. It has already been shown in Proposition 2 that any rise in productivity will
cause (labor allocation to agriculture) to go down. Thus we see that in the expression for
food demand (above) the numerator remains constant and denominator increases, resulting in
the fall of the overall quantity.
If the relative price of the Veblen good is held constant, the only reason for change in the
food demand is the rise in average consumption of Z. Due to the rise in A and B, total
income in the economy goes up. This means that total Z consumption (in effect ) will go up
as well (since now all extra income goes into Z consumption). This will lead to a fall in the
food demand by way of raising the marginal utility of the consumption of Z. So, while the
income of both groups go up, food consumption falls. In effect, we will observe the falling
Engel curves. (see Figure 2.1)
Discussion
A fall in calorie consumption will not be observed always. Assuming the conditions outlined
in Deaton and Dreze (2009) (constant relative price and rising income) we do see a fall in
food consumption. In the appendix it is shown that the reduction in calories will hold even if
the landlords' income falls in this model. The stipulation that landlord’s income does not fall
is a (strong) sufficient condition, not a necessary condition, for the result in Proposition 3 to
hold.
Amongst the two main assumptions imposed on the preference structure the role of quasi-
linearity is not crucial (the other being KUJ). This assumption implies that income effect is
totally absent in the demand for food, and is essentially an extreme representation of the fact
20
that food elasticity is found to be quite low for people with sufficient income (see Behrman
and Deolalikar 1987). If on the other hand, a positive (but modest) income elasticity of food
is introduced in the model it would work against the Veblen effect and push up food
consumption when incomes rise. In this case, whether calorie reduction obtains would come
down to which of the two effects are stronger. In the appendix a model with positive income
effect and KUJ has been used to demonstrate that calorie reduction can still obtain.
Whether KUJ preferences actually exist in the Indian populace and whether or not it can
dominate the income effect is an empirical question. This will be the focus of the next
chapter.
22
2.3 Small Open Economy
In this section I assume that the economy is free to import goods from the world market. I
also assume that the domestic agents take the world price of food as given, in other words it
behaves like a small open economy. A small open economy takes the relative price of
food/industrial good as a constant ( say). Here profit maximisation by producers in the
industrial sector means that they still equate their marginal cost of production i.e. the wage to
the value of the marginal output from 1 unit of labour ( ). This means that wages are now
pegged at . In this setting we have the following result.
Proposition 4: In the small open economy setting, a rise in industrial/agricultural
productivity leads to calorie reduction for both groups when both groups are consuming the
Veblen good.
Proof: Increase in A leads to a rise in to re-equate agricultural wage to . This change
keeps wage constant but raises land rental income for the landlords. As a result landlords
demand more Z. Since domestic production of Z cannot be increased (as workers are drawn
into agriculture) the increased demand for Z is met by importing Z. As a result, average Z
consumption of the economy goes up and food consumption for everybody falls. This creates
some excess of grain which is then exported to pay for the import of Z.
Since determines the wage an increase in B leads to a rise in income of both groups along
with a fall in labour allocation adjusts to equate agricultural wages to the industrial
wages. This should lead to a fall in land income, but for the sake of comparison we assume
that total income rises for the landowners as well. This will in turn lead to a rise in Z
consumption of both groups since they have already been satiated with food. Then, from the
23
expressions of the food thresholds we know that we will have a fall in the food thresholds.
Below is the expression for the food threshold/demand at constant relative price of food.
One other way to see this is that since productivity has gone up total income of all agents in
the economy will have to go up, this means that average level of Z consumption will also go
up (since all agents are assumed to be satiated with grain already). So now we can read from
the expression for the food threshold above that food consumption will fall. Below is a
simulation result using parameter values . Here
and are food consumption and incomes for the two income groups respectively.
Productivities Incomes Food Consumption Agri. Lab
A B
1 1 1 2.578 0.789 0.789 0.622
1 1.5 1.5 2.861 0.680 0.680 0.463
We can see the calorie reduction result here. A rise in B leads to rise in income of both
groups (although landlords enjoy less proportionate increase due to fall in ).
Table 2.1: Simulation Showing Fall in Food Consumption with Rise in Industrial Productivity.
Notes: Table shows a simulation result of raising Industrial Productivity B from 1 to 1.5 keeping other
parameters constant. Parameter values used are . Here and
are food consumption and incomes for the two income groups respectively.
24
The small open economy setting is a good approximation for a less-developed economy like
India. India is typically self sufficient in food grain production but imports a lot of industrial
goods. Although the volume of trade is large, India is definitely unable to affect world prices.
So this assumption is quite close to reality.
One distinct advantage here is that we don’t need to worry about keeping the relative price
constant while we change the productivities. The calorie reduction result is strengthened in
this setting. It is particularly interesting that the result holds for growth in agricultural
productivity, which is quite analogous to what we observe from India.
2.4 Capital as a Third Factor of Production
Model with Capital
In this section I am going to present a version of the above model which is more general in
various ways. To begin with I will not be assuming any specific functional form for the
Veblen preference term in the utility function. This form will be analogous to the way
Veblen preferences are modeled in general in the literature (see Eaton and Eswaran 2009).
There are, however, certain restrictions which are discussed below. The other difference is
that I introduce a third factor of production (capital) which is unequally distributed
throughout the population (unlike the other factors) and is used exclusively in industrial good
production. As we shall see the rate of return for capital will be growing more than all the
other factors. As a result of this, rising income inequality with economic growth gets
introduced into the model.
25
This features of rising inequality with growth in this model is interesting for two reasons.
Firstly, it is generally accepted that income inequality has been on the rise in India (see
Deaton and Dreze 2002). So the prediction of the previous model that income inequality is
going down was not representing reality. Also, since Veblen competition is driven by the
desire to catch-up with the rich, one would expect rising income differences to have
significant effects. Here the new model makes a significant contribution by pointing out that
even if marginal utility from consumption is falling for both goods, in the presence of rising
income inequality and status competition, an income driven rise in own consumption of the
status good could lead to a rise in marginal utility of the status good. This is due to a higher
average consumption of the status good in the peer group where average income has gone up
by more than the individual's. This is the key insight of the capital model.
Also worth pointing out is the fact that in the previous model we could never have observed
rising income inequality with economic growth with certainty. The reason is that land rental
rates may be falling even if we make our regular assumption that all agents have higher
incomes after the productivity growth. So we may still have a lowering of income inequality,
leading to a rise in food consumption. This problem persists even if we assume unequal
distribution of land. Although an unequal distribution of land also introduces a continuous
distribution of income, it still does not ensure rising income inequality.
Lastly, in this section I work with the assumption that an agent’s comparison group consists
of all the agent’s who can be observed to be consuming more of the Veblen good than her.
This is a much more realistic assumption than what I had earlier. In the previous model we
had just two income groups and it was impossible to make even the simplest assumption that
26
the agent’s compare themselves with their richer peers. The comparison group assumption
turns out to be quite important as it determines the movement of the average Z consumption.
An even better assumption would be that people only compare with a subset of agents
consuming more of the Veblen good (not with agents who are much richer), but in the
context of this model it will only add complications rather than extra insight.
Formally the three new assumptions are as follows:
Assumption 1: An agent’s peer/comparison group consists of all agents earning a higher
income than her.
A disadvantage with this, and in fact with any peer group assumption I make, is that it
presumes that people’s peer groups are essentially determined by their incomes. Although in
reality income is definitely an important determinant of one’s social circle it is never the only
criterion. Occupation, religion, race, caste etc are also important determinants. But in this
model I will abstract from this. Perhaps this abstraction is the reason why I observe certain
aberrations from reality when I look at the Calorie Engel curves predicted by this model.
This feature is discussed in more detail below.
For the next assumption I introduce an index for each agent as i which is continuous between
0 and 1.
Assumption 2: All agents are endowed with a single unit of labour; total land is equally
distributed, but total capital is distributed according to a function
27
I will not be imposing any further restriction on the distribution of capital. I believe an
unequal distribution of capital assets is quite a reasonable assumption. Typically capital
would be accumulated through savings but here I will abstract from that and assume the total
stock of capital is fixed at . The shape of the distribution function of capital, although
irrelevant for the main results, has important implications for the predicted food Engel
curves. I will elaborate further on this after I characterize the equilibrium. Also the
assumption about equal distribution of land is essentially a simplifying assumption. Having
the landowner class as before introduces many other issue like how much capital land-
owners should have etc. This unduly complicates the model without adding to the basic
insight.
Assumption 3: The utility function is now given as follows:
v
Note that with the new Assumption 1 average Z consumption relevant to an agent is
dependent on the agent. It is now given as follows for person i with i not equal to 1:
With the assumed utility function the demand for food is given as follows:
28
Observe that the income at which agent starts consuming Z is not the same for all. More is
said about this in the section below about food Engel curves.
This specific preference structure is assumed for the purpose of highlighting the scenario
where rising income inequality becomes essential for the calorie reduction phenomenon. The
main model presented above used quasilinear preferences which assumed away income
effect on food. Hence, income did not play any significant part in those results. Here it is
assumed that the Veblen component of the preferences is concave. Moreover, the influence
of own consumption of Z and peer group's consumption of Z is assumed to be opposite but
equal in magnitude. This feature makes it essential that if demand for X has to fall in any
equilibrium with constant prices, the peer group's consumption of Z has to increase more
than own consumption. In the general equilibrium this condition translates into rising income
inequality. An even more general model is presented in the appendix where the magnitude of
influence of own and peer group's consumption of Z is allowed to be different. In this model
the results simply depends on the relative strengths of own and peer income effects.
Supply Side
There are certain modifications to the supply side as well, but these are standard. The
production functions are now given as:
Next let the factor prices for labour, land and capital be w, v and r respectively. Then we can
write the incomes of each agent as follows:
29
These expressions can be expressed in terms of the labour allocations alone once we have
worked out the factor prices. So we look at the conditions for profit maximisation in the two
sectors. From the industrial goods sector we have the following:
..... ( 5)
..... ( 6)
Also from the grain producer’s profit maximisation condition we have:
..... ( 7)
..... ( 8)
Since there is perfect mobility of labour between the two sectors only a single wage may
exist. So using the two first order conditions with respect to labour we can obtain an
expression for the price of the industrial good in terms of resource allocations only.
... ( 9)
Demand for grain and the industrial good are as before. Using the expressions for the
incomes of the different classes from above and equations 5) - ( 9) we can express the
demands for grain and industrial goods only in terms of factor allocations. Next we impose
the market clearing conditions for land, labour and capital.
30
The market clearing conditions that are yet to be taken care of are the grain market and the
market for industrial goods. We can drop one of these by Walras’ Law. The other equation
can be expressed fully in terms of either the labour allocation to grain or industrial goods
sector. Once we solve this equation for the labour allocation we can go back to our equations
5) - ( 9) to calculate all the relevant variables.
Now we can write down the final grain market clearing condition in terms of labour
allocation to the grain sector using the demand for food calculated earlier:
.... ( 10)
Here the food demand is a function of incomes and price of Z which are given by equations
( 5), ( 6), ( 8), and ( 9). Equation 10) can be expressed only in terms of the labour
allocation to agriculture and other parameters. Assuming that equilibrium exists, we can
solve for this key variable which allows us then to compute all other parameters for this
general equilibrium.
Now various equilibria may exist for this model depending on the proportion of population
that has enough income to consume Z. For our purposes the most interesting equilibrium is
the one where all have graduated to become Z consumers. Henceforth I will restrict my
attention to this equilibrium only. The next proposition looks at the comparative statics when
I vary the productivity parameters for this equilibrium.
31
Proposition 5: In the equilibrium where all are satiated with food, any increase in industrial
or agricultural productivity (represented by the parameters B and A respectively) leads to a
fall in the equilibrium labour allocation to agricultural sector.
Intuition: The intuition for this proposition is completely analogous to that of the
corresponding proposition from the model without capital. Any increase in productivity
raises income. A rise in the incomes of the Z consumers raises the demand for Z and this
draws labour out of the agricultural sector. The effect is enhanced due to the presence of the
Veblen parameter. What is noteworthy is that this result holds exactly as it did in the
previous model even with positive income effect on food. The reason why this follows so
simply is because of the assumptions on the peer group and income distribution. The
construction of the model dictates that any productivity driven economic growth leads to an
increase in the gap between own and peer group's consumption of the Veblen good. This
feature increases marginal utility of Veblen consumption and helps to counter the income
effect on food which might have turned the result by increasing food demand. Proof is in the
appendix.
Construction of the Food Engel Curves
Since the objective of this chapter was to provide an alternative explanation of the findings in
Deaton and Dreze (2009), I will try to compare the slope of the Engel curves that this model
predicts to the one that we find in their paper. They found, quite reasonably, that the Engel
curves for food are positively sloped. This of course includes the effect of Veblen
preferences. It turns out that the slope of the Engel curve in the equilibrium where all agents
have started consuming Z depends on the distribution of capital.
32
In this variant of the model the food consumption is given by:
When we look at the Engel curves we keep the relative price of industrial good constant. So
the only thing determining the food consumption is the term . Individual agent i's
Engel curve for food rises with income until income hits the mark, given by the above
expression at zero consumption of Z.
This is a positive quantity since we assume that first derivative of the Veblen function is
always positive and finite. After this, food consumption goes up but slower than income
(assuming that is constant). So as shown in figure 2.4 below, an individual’s food Engel
curve begins with a straight stretch coinciding with the 450
line, shown by the black line.
After reaching the curve rises but with slope less than one (shown by the red lines). The
exact shape of this section depends on the function V in Assumption 3.
Also note that since in this model varies by the individual, the food threshold is also
different for different people as shown in the figure by the other red lines. The richest agent
in this model is the agent indexed by i=1. When she graduates to become a Z consumer her
income is given by:
33
When the agent who is just behind agent 1 graduates, then agent 1 has already done so. So
for this next agent income at which she graduates is lesser than x1d:
This is evident from the fact that V' is falling in its argument. So each agent along the index
line 0 – 1, graduates to a different income level.
34
If we look at the Engel curve for the population, of course we are going to see a different
picture compared to an individual.10
Different individuals are on different points on their
individualized Engel curves indexed by i. The population Engel curve will be an aggregation
of these in the general equilibrium. This is essentially what the empirical Engel curves in
Deaton and Dreze (2009) represent.
Calorie Puzzle Revisited
Also, in this setting we are again interested in investigating the possibility of a calorie puzzle
occurrence. Now the expression for relative price of food is a bit more complicated (given by
equation (2.9)), yet it is still possible to raise A and B simultaneously in such a way that
relative price remains constant. Although the following proposition mirrors an earlier one,
the demonstration here is under more general conditions.
Proposition 6: In an equilibrium where all are satiated with food, if productivity parameters
both rise keeping the relative price constant, is large enough to ensure that the poorest
person's income is going up, then food consumption falls for all agents
Proof: See Appendix.
10
To obtain this relationship we take the average food consumption of all agents with a particular income and
then plot this against income. Now due to our assumption on the capital distribution income is unique for all
agents with the same index. Here we again consider the situation where everybody has started consuming Z. So
food consumption for agent i is given by the expression for food demand given by:
Clearly, the shape of the Engel curve will be totally determined by how varies with income. Now
because the distribution of capital is assumed to have a positive slope, income is rising with the index of the
agent. So in order to have rising Engel curves, we need to rise with i (because v is concave).
Note that at i = 1 this quantity is equal to 0. But at i = 0 it is:
The above inequality holds as the integral is always higher than . Hence, if the functions are smooth, will be rising over some part of the income distribution. This implies that the Engel curves will be rising for
those ranges.
35
Intuition: Two of the assumptions we make are important for this proposition. The fact that
agents only compare with those richer than themselves, and the rise of income with the index
of the agents, (induced by the distribution of capital) ensure that the difference between own
and peer group average consumption of Z goes higher (becomes more negative) for all
agents. This is the implication of rising income inequality. Next as the prices are constant,
this raises marginal utility per dollar of consuming Z. As a result food consumption falls.
The predictions of this model with capital are in some ways fundamentally different from
that of the former model. The most significant difference is the importance of rising income
inequality in the capital model. The food demand functions for the two models are given by:
Whereas in the previous model food consumption fell with the average Z consumption, in the
latter model it falls with the difference between own consumption and the average. As a
result of this, we observe reduction in food consumption in the capital model only if
productivity growth leads to a rise in income inequality within the comparison group. So
according to this model if we are observing a downward shift of the calorie Engel curves, we
must be associated with rising income inequality all through the income distribution. We can
achieve this condition theoretically by making a natural assumption that capital assets are
unequally distributed.
2.5 Conclusion
The basic model has a quite few interesting insights to offer on the role of status seeking
behaviour in a simple dual economy. The first is that the effects of rise in productivity are
36
enhanced in the presence of conspicuous consumption. With everybody having higher
income, Veblen competition is intensified according to our preferences. This leads to higher
incomes and faster growth of the industrial sector. It is this trait of status seeking that others
like Cooper et al. (2001) have used to generate growth. This is also reflected in the finding
that labour allocation to agricultural sector would be much higher in the absence of Veblen
preferences.
The second interesting contribution is to demonstrate the possibility that the calorie–
reduction puzzle documented by Deaton and Dreze (2009) and others may be attributed to
status seeking behaviour. This reduction in calories has been shown to possibly obtain both
in the closed as well as the small open economy version of a model incorporating Veblen
preferences. In fact as we change the assumptions to either include capital or extend it to the
small open economy case the results arise more naturally with fewer and more realistic
assumptions.
37
Chapter 3: Empirical Estimation of Veblen Induced Peer Income
Effect on Calorie Consumption
In this chapter the partial equilibrium result from Chapter 1 (Proposition 1), which predicts a
negative relationship between peer group income and food consumption in the presence of
KUJ is taken to the data using the Living Standard Measurement Survey collected by the
World Bank. The goal here is to provide evidence for the existence of preferences for
keeping up with the Joneses among rural agricultural communities in India. This exercise is
similar to those in other papers which attempt to measure the effect of peer group income on
economic outcomes e.g. Luttmer (2005) on happiness measures and Charles et al. (2009) on
conspicuous consumption. Of these, the approach followed in this paper is closest to that of
Charles et al. (2009).11
They use the fact that in USA, being non-white is a negative signal
for wealth and the so-called poorer races have to spend more on conspicuous consumption to
make up for this handicap. The authors find that the differences in spending on visible
consumption by race can be explained by the average incomes of the particular race. But,
while the Charles et al. (2009) paper addresses interracial differences in income driving
Veblen consumption, here within caste income distributions and its effect on conspicuous
consumption are examined. Also, rather than highlight the signalling aspect of the Veblen
goods, the setting here is closer to the “keeping up with the Joneses” motivation as regards
the consumption of status goods (see Galli 1994).
11
Khamis et al. (2010) use the same methodology as Charles et al.(2009) with Indian data and find mixed
results.
38
One paper which measures the impact of peer group income on consumption expenditure for
rural Indian households is Munshi and Rosenzweig (2009). Although in a different context
which pertains to a test of the hypothesis that households enjoy full insurance from caste
based insurance networks, the paper finds a positive relationship between own sub-caste
(jati) consumption and own consumption. In this exercise we are going to concentrate on the
peer income effect on calorie consumption. However, we are going to be using the same peer
group for the individual households as used by the Munshi and Rosenzweig paper, which is
castes and village based.
One important contribution in this empirical exercise is the unique instrument used to tackle
measurement issues with the peer group income variable (also encountered in Munshi
Rosenzweig 2009). In this sample, low caste individuals living in low caste dominated
villages earn higher agricultural income than low castes living in high caste dominated
villages (Anderson 2011). This creates an exogenous variation in own caste income among
the low castes across the two types of villages. On using this instrument, estimates for peer
income effect turn out to be negative and significant. Moreover they are twice the magnitude
of the estimated own income effect (which are positive as expected), which is exactly in line
with the theoretical model.
At this juncture I should point out that due to the limitations of the theoretical model
presented in the previous chapter it is not possible to calculate the exact magnitude of peer
income effect and own income effect that would lead to the calorie reduction phenomenon.
In fact different assumptions on the nature of the Veblen/KUJ part of the preferences could
lead to different conditions being required for calorie reduction. For example, in the model
with capital presented in the previous chapter it is implicitly assumed that own and peer
39
income effects on food consumption are equal in magnitude but opposite in sign. This
assumption leads to the requirement of rising income inequality for calorie reduction to go
through.12
However in this chapter the attempt will be to estimate the elasticities of own and
peer group income without much reference to income inequality. The results suggest that
own income effect is about half the size of peer income effect. This implies that rising
income inequality may not be required for the possibility of calorie reduction with income
growth. However if income inequality is going up within the peer group it will only magnify
the negative impact on calorie consumption.
3.1 The Data
The data used here is from the World Bank series of Living Standards Measurement Survey
(LSMS) conducted in several countries. This particular data is from the Survey of Living
Conditions conducted for the two Indian states of Uttar Pradesh (U.P) and Bihar. Here
between December 1997 and March 1998, 1035 households were surveyed in Bihar from 57
villages in 13 different districts, and another 1215 households were surveyed in U.P from 63
villages in 12 districts. The survey is restricted to rural households and most households
derive livelihood from agriculture. It provides fairly detailed information about
demographics, member’s characteristics, consumption, access to facilities, means of
livelihood and also information about the environment in which the household is situated
(village level characteristics). Although similar datasets like Indian Human Development
Survey or the Rural Economic Development Survey also contain village level information,
12
In fact rising income inequality would be a requirement for calorie reduction whenever own income effect is
assumed to be equal or greater than peer income effect.
40
LSMS is the only dataset where the variation in village level land-ownership patterns may be
used to identify my main coefficient of interest.
The main variables of interest for this exercise are income, caste, price of food and total
calorie consumption. The caste of each household is known. The income of the household is
generated (following Anderson 2011) by adding total wage income of all household
members, income from enterprise, total proceeds from crop sales, transfers and the total
value of home production of in-kind receipts of crops and foods. The calorie consumption
variable is constructed by using the household consumption of different food items and
multiplying them with estimates of calorie content from Gopalan et al. (1974).13
Food items
used are rice, wheat, barley, maize, bajra (course millet), pulses, sugar, gur (jaggery), eggs,
meat/fish, milk, milk-products, potatoes, edible oil and vanaspati (a form of oil). It should be
mentioned that certain assumptions had to be made while constructing variables. Firstly, the
quantities of rice, pulses etc are multiplied by average amount of calorie content for all types
of rice and pulses. This is because the data does not provide information about the exact type
of these foods. Also, these calorie estimates are constructed from data about food
purchased/produced by the households and it does not imply that it coincides with calorie
consumption. In particular, household members could have taken meals outside their house
or other guests could have had meals in the household, but there is no information about this
13
There have been updated versions of Gopalan (1974). I preferred to use the version closest to Deaton and
Dreze (2009) so as to make my variables as consistent with theirs as possible. However there does not seem to
be much difference in the two versions particularly in the context of this dataset which only lists broad
categories of food items like rice or pulses. In contrast NSS datasets list much more detailed classification of
food items consumed.
41
in the data.14
The price of food is calculated by averaging the prices of all food items used for
the calorie consumption, with weights proportional to their share in the food expenditure.
Table 3.1 below provides summary statistics for some of the primary variables of interest.
14
Although I can see the monetary value of meals consumed outside from the data, there is no reliable method
of calculating the nutritive value of these meals. In the main specification I control for expenditure on outside
meals, but they do not appear to have any significant effect on calorie consumption.
42
The large standard deviation in the average per-capita calorie consumption of the household
perhaps reflects the difference in household member’s characteristics.
Patterns of occupations, degree of mechanisation in one’s job, availability of public goods
are other factors that may contribute to differences in calorie consumption (using the
arguments of Deaton and Dreze (2009) and Li and Eli (2010) about calorie requirements).
43
So controls for occupation, crop type choices, factors determining strenuousness of everyday
life (like distance to water source) etc will be used in the main regressions.
Table 3.2 above, is the breakup of the sample according to occupation (proportions). Note
that this is just the primary means of livelihood as reported by agents. Most of the agents
have their own farms. The next biggest chunk are the casual labourers who work for daily
wages in other people's farms. Some are also permanent (having long term contracts with
employers) agricultural workers. Salaried workers and traders (shop owners) are other
sizeable proportions. This information along with that on hours worked will be important as
we attempt to control for calorie requirement of the agents. The break down in Table 3.2
suggests that the sample is essentially rural and agricultural.
Apart from income generating activity, people may also be engaged in other, strenuous day
to day activities. For example, fetching drinking water from far away is one such activity.
Also the type of crop being cultivated, fertility of the land owned, amount of irrigation
available and degree of mechanization of the different farm processes may also be important
determinants of calorie requirement. Table 3.2 gives an idea of the information available
about these factors. From the statistics it seems that we should not be too concerned about
these factors. Most people fetch water from relatively close quarters. Also, the average
quality of land is the dominant reported soil type. Nevertheless, I will be controlling for them
in my regressions.
44
3.2 Empirical Strategy
The primary objective of this study is to investigate the relationship between calorie
consumption and the average income of the agent’s social group after controlling for own
income, price of food and other factors affecting calorie requirement. The first step of this
analysis is the identification of an appropriate peer group for the agents. Here, the relevant
reference group for an agent is assumed to consists of her own caste people living in her own
village. This assumption also implies that economic outcomes of other castes are either
unobservable or do not matter to agents while they make their own consumption decisions.
In terms of the model presented before we are essentially assuming that the entire population
for an agent is her own caste members living in her own village. While choosing the
reference group we need to ensure that the agents interact sufficiently within this group and
have the opportunity to observe each other’s consumption patterns. For our purposes the
assumption of own caste people seems to be the best way to proceed. Firstly, it is a
documented fact that in rural India caste based social interaction is very important. For
example, the information passed through these social interactions is good enough to sustain
informal credit networks within castes (Munshi and Rosenzweig 2005, 2006). The
arrangement provides added insurance to rural agents against unforeseen shocks like illness
or marriages. As such they are highly valued by the members. Information networks within
castes and sub-castes (jatis) are strengthened by strict rules forbidding members to marry
outside the caste. These strong caste networks have also been cited as the main reason why
spatial migration is extremely low in India (Munshi Rosenzwieg 2005). Anderson (2011)
conjectures that inter-village migration is rarely seen in India probably because the caste
based networks are not known to extend beyond the village boundaries.
45
Another piece of evidence on the importance of caste networks is manifested in politics.
Especially in the states of UP and Bihar (on which this analysis is based) caste based political
parties are very strong. There is evidence that people will tolerate a lower quality politician
in office as long as she belongs to the party that represents one’s caste (Banerjee and Pande
2007).
Given the strong interaction within the community for people born into the same caste, it
seems natural to assume that own caste members will be a good representation of an agent’s
reference group.15
Here I will restrict attention to own caste members from the same village
because presumably these people will be most visible to the agent and the people they meet
and socialize with. Also, it is often seen that when people compare themselves they tend to
do so with people they think are similar to themselves. For example a poor farmer would be
less affected with the information that a millionaire has bought a private jet than if told that
her fellow farmer in the village has just bought a new television set.
So now we can write out the basic regression as follows:
..... (3.1)
Here is log per-capita calorie consumption of the household i in a typical month, is log
of household income per-capita, is the log average income per-capita of the people of the
same caste as household i living in the same village excluding household i, is log of food
price facing the agent, X is a set of individual or village level controls and e is a random error
term which will be clustered at the village level.
15
The assumption is contrary in spirit to the widely documented phenomenon of Sanskritization (Srinivasan
1952) where the lower castes emulate behaviour of the upper castes. But this is an effect which should be
particular to castes and hence should be taken into account by including caste fixed effects. Also I show later
that most other caste incomes do not affect an agent’s consumption decision.
46
3.3 Measurement Error and Instrument
A possible problem with the specification presented above is that average caste income and
calorie consumption may be endogenously determined.16
We have very little idea about what
determines the incomes of the agents in our dataset. There could be many unobserved
characteristics of either the household or the village that systematically affects both calorie
consumption and caste income. These things will have to be varying by caste and village. For
example, if people belonging to a certain caste enjoy both higher incomes as well as a less
labour intensive life due to their being physically better suited to live in a particular area. Or
certain caste members who are rich might own facilities (for example water or cattle) in their
villages that help them survive on lesser calorie intake. Although I try to control for as many
village characteristics as possible, one can never be sure that all possible sources of
endogeneity have been accounted for.
Another more important problem is the presence of classical measurement error in the main
variable of interest; average income of own caste in the village. This variable is not measured
precisely since sometimes there are very few observations of each caste from each village (2-
3 or even less). Amongst the three lower caste groups in the sample, the Backward
Agricultural castes have 3 or less households sampled in 18 villages, Backward Other Castes
has 3 or less sampled in 31 villages and the SC/ST caste group has 3 or less sampled in 22
villages out of a total sample of 72.17
Also in reality caste networks are organised at the sub-
caste level (also known as jatis see Munshi Rosenzwieg 2005), however there is no
information about sub-castes in this data. So the average caste income that is calculated may
16
A Durban Wu Hausmann test of exogeneity of average caste income with the basic specification yields a p
value of 0.0002 indicating that OLS is not consistent. 17
The villages where just 1 household of a particular caste is sample, are dropped since the caste income cannot
be calculated for such cases.
47
be deviating from its true value quite significantly. However, there is no reason to expect any
correlation between the measurement error and the observed value.18
As such we may take
this to be a case of classical measurement error. This error will likely produce attenuation
bias in the coefficients making them lower in magnitude. The best strategy here would be to
instrument the average caste income variable. It turns out that there exists a candidate
instrumental variable which serves my purpose.
In a recent paper Anderson (2011) uses the same dataset to show that low caste people living
in high caste dominated villages have significantly less income than similar low caste people
living in low caste dominated villages, where dominance is determined by which caste owns
the majority of the land in the village. She goes on to show that this difference is largely
attributable to the fact that, in high caste dominated villages, high caste villagers own the
water sources and do not sell water to the low caste people. This breakdown in water trading
networks reduces the agricultural income of the low castes.
One implication of the finding of Anderson (2011) is that for a low caste agent living in a
low caste dominated village her entire peer/social group will have higher income than a
similar counterpart living in a high caste dominated village. The strategy in this chapter will
be to use this variation in the peer group’s income to identify the Veblen effect on calorie
consumption. As an instrument for average caste income in the village a dummy for village
dominated by the low castes is used. This instrument solves the measurement error problem
in two ways. First, it introduces an exogenous variation to the true value of peer group
income which is uncorrelated with the measurement error and secondly this variation affects
all low caste agents so sub-caste level average incomes are also affected. Using this IV a two
18
The measurement error is not due to any systematic sampling anomaly, rather it is a result of lower sample
size which simply increases the noise in the observed variable.
48
stage least squares estimation procedure is carried out, where in the first stage the following
equation is run on a sample restricted to low caste people:
The equation contains which takes the value one if the responder belongs to a low caste
dominated village. In the second stage the estimated values of from this equation is used
in the original equation (equation 3.1) to estimate the parameter of interest. Here also the
sample is restricted to low castes only.
In order for to be a good instrument it is important that this dummy indicating the
agent’s village dominance does not affect calorie consumption on its own except through its
effect on peer group income.19
The first concern is that the breakdown in water trade in the high caste dominated villages
(the main reason of the variation according to Anderson 2011), by itself may be affecting
calorie consumption. The results could be biased if people with lower access to water (as
should be the case in high caste dominated villages) have to expend a lot of energy trying to
irrigate their crops by digging or carrying water. To control for this, variables indicating
nearness to water sources like rivers and canals, availability of groundwater and public
groundwater projects are included in all the specifications.
A second issue is the presence of unobservables that may vary by caste dominance across
villages (for example public goods, resources etc.). As a response to this concern, this paper
follows Anderson (2011) in arguing that the variation in caste dominance across villages is
19
Of course the dummy will also have an effect on own income but that will be controlled for.
49
largely exogenous to any present economic outcomes. The first step in this is to argue that
caste dominance is exogenously determined. The pattern of settlement in this area is
determined by migration patterns from over a thousand years ago. Later, just before
independence in 1947 land ownership tended to be concentrated in the hands of the upper
caste. So there was prevalence of absentee landlordism, that is, the landlord might be living
in some other village altogether (Metcalf 1979). After independence the government
undertook a land reform drive to redistribute land to the landless tenants. This resulted in
ownership of land passing on to some low caste tenants (Neale 1962). From then on land
ownership rights have been passed down by older generations by way of inheritance. Formal
selling of land is very rare. It has been observed that only 1% of land is sold each year (Dreze
et al. 1999). So land ownership in a particular village is exogenously determined by the
settlement patterns and the land reform drive.
The second step in the argument is to show that the two kinds of villages are very similar on
most observable characteristics. This is an extremely important observation in this context
since village fixed effects are not being used; it is thus quite comforting to know that the
villages are similar at least on the caste dominance line (see Table 3.3 below).20
Table 3.3
compares some village level variables and Table 3.4 looks at household characteristics across
the two types of villages. As far as village characteristics are concerned the equivalence of
means cannot be rejected for any of the variables. This suggests that the villages of the two
types are similar in most ways. Regarding the household level characteristics there are some
that are significantly different like literacy, but these variables can be controlled for in my
20 Table 3.3 and 3.4 are replications of Tables 1 and 2 in Anderson (2011), but they are equally relevant here. I
have of course added and subtracted some variables according to the requirements in the present context.
50
regressions. Also the comparison of calorie consumption and mean caste income gives us
some indication about the result we might expect later.
The last concern regarding the IV is the possibility of caste based migration across villages
or inter-caste mobility. But the latter concern can be dismissed altogether because of the
strict rules of the Hindu caste system in India. Though sociologists have documented a
phenomenon “Sanskritization” where lower castes try to graduate into a higher caste by
trying to emulate the ways and practices of the upper caste, this process is not the same thing
as caste mobility. See Srinivas (1952). Also to be noted is that if Sanskritization has been
going on it will only make my results weaker. Since there is a complete absence of high caste
people from the low caste dominated villages the residents of low caste dominated villages
have less opportunity to emulate high castes and so should be consuming more calories.
Permanent inter-village migration is also very minimal in India (Munshi and Rosenzweig
2005, 2006). The main reason for this is believed to be the caste based consumption
smoothing/insurance networks that are not known to extend beyond the village.
Also to test the validity of the instrument intuitively I follow Murray's (2006) suggestion and
include the low caste dominated dummy in a reduced form regression of the second stage.
The coefficient of the dummy is negative and significant in all the specifications (not
reported here).
53
3.4 Results
This section begins with the basic OLS results of equation 5 as presented in Table 3.5 and
Table 3.6 (below). The coefficient on the average income of the household's caste from that
village is very small, mostly insignificant and always negative (except for one specification).
The basic regression includes controls for state, caste fixed effects, total number of hours
worked by the members of the household, number of women and children in the household
and average age in the household. Technically it is also possible to include village fixed
effects in this regression but since there are about 120 villages and about 2000 households
there is not enough data to precisely estimate all these parameters. Subsequent specifications
introduce dummies for each of the 25 districts21
(in column 2), occupation controls (column
3), and dummies to indicate whether the household is engaged in the cultivation of cereals,
cash crops, oilseeds or bulbous roots (column 4). The rationale for inclusion of these controls
is to either take care of any spatial characteristics of the location of the agent (for example
people in hilly areas may have less calories than others living in plains) and remove the
effects of some special occupation or crop cultivation that requires more effort than others
(for example paddy cultivation is supposed to be a back breaking work). The fifth and sixth
specifications (column 5 and 6) includes controls for distance to nearest water source,
quality of land owned, percentage of land owned irrigated, and also distance to basic
facilities like public distribution shops, hospitals/medical centers and schools. Also, some
parts of calorie consumption may also be motivated by Veblen competition. For e.g. more
expensive foods like meat and milk products may carry a status value in itself. Introducing
21
Each state in India is divided into a number of districts. 13 districts from the state of Bihar and 12 from Uttar
Pradesh are included in this survey. Public good distribution is largely administered by the district level
bureaucracy in India.
54
controls for demographics and occupations of households helps control for different
perceptions among households about what constitutes Veblen goods and Veblen bads.
Amongst the controls added own income has a positive and significant coefficient, as
expected. Some other controls not reported are also interesting. For example, average age in
55
the household and average age squared are both significant. Coefficient of average age is
positive and age-squared is negative indicating a concave relationship between calorie
consumption and average age in the household. Caste dummies for the two lowest castes are
also significant and both have a positive sign. Amongst the controls for occupation and crop
cultivation the only significant ones are interest income (negative effect) and pulses
cultivation (negative effect).
56
In Table 3.6 the same regressions using log calorie consumption and log incomes are
reported using a sample of low castes only (castes included are backward agricultural,
schedule castes and schedule tribes). This is done in order that we may be able to compare
the OLS and the IV results which is also run on a sample of just the low-castes. Here the
coefficient of interest (average own caste income) is negative and the base specification (one
without any controls except age and hours worked) is also significant. But all coefficients are
small in magnitude and much lower than those of own income which are always significant.
Next, the main IV results are presented, the first stage (Table 3.7) followed by the second
stage (Table 3.8). The basic specification (column 1) has own income, literacy, state and
57
caste controls, and also controls for availability of water and average age of household
member. In the next 5 regressions I add the controls analogous to those added in the OLS
specifications earlier. The last one is a regression on a restricted sample of people working
on their own farms and uses controls for degrees of mechanization like use of tractors etc.
The purpose of this is again to address the calorie requirement issue. This is a two stage least
square estimation. Recall from equation (3.2) that the first stage equation is as follows:
is again log average caste income of the village where agent i lives.
Significance of the coefficient of the dominance dummy ( ) indicates the validity of the
instrument in terms of being a determinant of the variable average caste income. As we can
see that the F statistic for most of the columns are very high. The F values are way higher
than the rule of thumb value of 10 indicating the joint significance of the first stage equation.
Next in the second stage the estimated values of log average caste income according to
the above specification are used to instrument for it in our main equation. Here the dependent
variable is per-capita calorie consumption. Equation (3.1) is given below for reminder.
The calorie measurements are per-capita calorie intake per month measured in kilocalories.
As can be seen from Table 3.8 below, the main coefficient of interest (that of own caste
average income) is negative in all specification and significant for all except the basic one
(without any controls). The result without controls is somewhat expected as the uncontrolled
regression cannot account for idiosyncratic variations in calorie requirement, tastes and
preferences or availability of helpful public resources.
58
The results (refer to Table 3.8 Column 7, specification with all controls) indicate that a 1 %
change in average income of the agent’s caste results in a 0.22 % fall in the per-capita
monthly calorie consumption. In order to get some idea about meaning/significance of this
magnitude we first note that mean low caste income per-capita in high caste dominated
villages is Rs. 665.53 on average. The same statistic for low caste dominated villages is Rs.
1141.17. So we can say that living in a low caste dominated village implies that the person
has a comparison group which on an average is richer by Rs. 475 per-capita annually. This
sort of a differential in income of the comparison group leads to a fall in daily calorie
consumption per-capita of about 15.7%. These magnitudes are much higher than the
59
magnitudes we obtain through the OLS estimation. This disparity may be due to the
attenuation bias from the measurement error in peer group income.
The next question is how effective are these numbers in explaining the calorie reduction
documented in Deaton and Dreze (2009). In order to get an idea about this a couple of rough
calculations will be useful. Now according to Deaton et al. (2009) rural per-capita household
expenditure went up from 251.3 in 1983 (38th
round of NSS) to 318.3 in 2004-5 (61st round
of NSS) in real terms. Assume that the average income of the comparison group goes up by
exactly the same amount as per-capita household expenditure. Then, using the estimates in
this chapter, the fall in mean per-capita calorie consumption due to rise in peer income alone
is about 5.8 %. Actual findings from the Deaton et al. (2009) paper puts the fall in calorie
consumption during this period to about 9%. Next if we take into account the effect of own
income (which increases calorie consumption) the figure above has to be revised down to
3.2%. But, even this is more than a third of the actual fall as calculated by Deaton and Dreze
(2009). So the estimates explain a significant portion of the calorie puzzle.
More importantly, in the context of the theoretical model presented in the previous chapter,
these results demonstrate that own income has about half the influence that peer group
income has on calorie consumption. Although we cannot conclude from this evidence that
the observed calorie reduction is necessarily through the Veblen channel, it seems very
probable.
60
3.5 Robustness
1.Influence of High Castes: Sanskritization?
The first robustness check is designed to test the assumption made earlier about the
irrelevance of other group incomes to the calorie consumption decision. The most relevant
test for this would be to include the other castes’ incomes in the basic regression. The main
argument against the assumption made earlier is the concept of "Sankritization" as first
pointed out in Srinivas (1952). This is the notion that lower castes are continuously trying to
emulate the behaviour of the upper castes with the view to climbing up the social status
ladder. Since higher castes are richer and they consume fewer calories on an average, the
existence of Sanskritzation would imply that agents living in high caste dominated areas will
be consuming lower calories, thus reducing their calorie consumption differential with their
companions in low caste dominated regions. (Note that there are no high castes in the low
caste dominated villages, so no such mechanism could be working there.) In other words, if
this mechanism is working the previous analysis underestimates the Veblen effect.
In order to test this, high caste average income is included as a regressor into the basic
regressions presented in Table 3.6. Of course the specification suffers from the whole gamut
of problems that were initially present (mainly the measurement error issue). Another
problem with this is the very few numbers of observations due to the complete absence of the
high caste agents from low caste dominated villages. Also, for the very same reason, it is not
possible to include high caste average income in the IV regressions since all the low caste
dominated agents will have a missing value totally eliminating the variation.
61
The results are reported in Table 3.9. Inclusion of the high caste’s average income does not
change the result much. In fact the coefficient on the high caste income variable is
insignificant lending further credence to the assumption made earlier.
2. Non-food Expenditure
Although we have seen that food consumption falls with rise in peer group income we still
have no idea where the saving from this lower consumption goes. The next set of regressions
is designed to investigate where the money generated from lowering calorie consumption is
being used. Data is available for some durables and some non-food items (although many of
the non-food items may be classified as essentials). The IV estimation done in tables 3.7 and
3.8 is repeated here with a series of expenditures on non-food items as the dependent
variable. In each regression all the controls used in table 3.8 are also included.
The results are reported in Table 3.10. All coefficients except the one on children’s clothing
turn out to be insignificant. Thus not much can be deduced about where the money is going
from the lower food consumption. Yet the coefficient on children’s clothing is perfectly
consistent with our story.
64
3. Calories from Veblen Foods
Agents derive calories from all kinds of food items. Some of these might be Veblen goods
themselves. For example for many societies meat or dairy products are considered to be elite
foods associated with a higher status or wealth. This sub-section focuses on certain food
items which may be viewed by agents as high status items. Here, the IV regression is run
with the same specification as before but calories derived from more elite/fashionable foods
such as milk, meat, eggs, sugar etc as dependent variable.
65
The coefficient for own caste income (reported in Table 3.11 column 2) are significant and
positive except in column 7 where all controls are included. This suggests that for foods
which may be perceived as status goods by the agents the effect goes the other way round.
One of course, would have expected all the coefficients here to be positive and significant,
but there is some confounding in this data since our choice of food items as Veblen or non-
Veblen is likely to be quite arbitrary. A proper analysis of what is considered richer food can
only be done with more fine data like for example what Charles et al. (2009) have done with
a survey specially designed for the purpose of identifying Veblen goods from non-Veblen
ones. Another interesting feature here is that the coefficient for Veblen food prices is positive
and significant in all specifications whereas the coefficient for own income is negative
(although insignificant in some specifications). Since positive price effect is usually
associated with conspicuous consumption goods, this evidence suggests that the choice of
food items is perhaps not grossly incorrect.
3.6 Conclusion and Discussion of Alternate Explanations
The objective of the paper was to demonstrate that the decrease in calorie consumption in
India during the period 1985-2005 may have been due to (to some extent at least) the
manifestations of status competition. The paper contributes towards this in two parts. First in
the theoretical part (Chapter 2) it was shown that in the general equilibrium of a standard
dual-economy model with Veblen preferences (as represented by "Keeping up with the
Jones" specification), it is possible for food consumption to decrease with economic growth
under certain conditions. It turns out that the elasticity of food demand with respect to own
66
income is required to be modest in comparison to the Veblen effect in order for us to see
calorie reduction with income growth.
In the empirical part, the main contribution of this paper is to demonstrate a robust negative
relationship between peer group’s income and own calorie consumption. What is more, the
magnitude of the coefficients of peer group income are much larger than those of own
income. The take away message is that peer group’s behaviour does have some effect on the
behaviour of an individual. This is consistent with the conditions generated in the Chapter 2,
although it is not totally conclusive.
An alternate interpretation of this result is that people derive information about how to spend
their money by observing the behaviour of their peer group. Although this explanation
cannot be ruled out, it is possible to claim that this is also consistent with the Veblen theory
as long as agents only look at richer neighbours while deciding what to consume. The strong
relationship with higher income may perhaps be indicative of this.
It is also worth mentioning that people can be getting status signals from various overlapping
peer groups. For example people may be moving in two different groups and they may be
rising in the income hierarchy in one and falling in the other at the same time. This makes
measurement and quantification of the Veblen effect very complicated. Here the attempt has
been to isolate and identify the Veblen effect from one particular peer group while
controlling as best as we can the influence of other groups. The objective was to establish the
existence of the effect and also to have some idea about its magnitude.
As far as the calorie puzzle in India is concerned, findings of this paper suggests that status
competition can be a significant factor contributing to it. Incomes have been rising across
67
almost all levels of the income distribution in the last couple of decades. This fact, when
coupled with my finding that a rise in income of the peer group leads to a fall in the calorie
consumption of agents, leads to predictions very similar to the falling Engel curves as
described in Deaton and Dreze (2009). Using the estimated rise in consumption over 1984 to
2004 by Deaton et al. (2009) we see that these estimates can explain a significant part of the
missing calories (roughly one third). However, evidence presented here is from a small
sample of rural Hindu low caste households from just two states in India. The chapter does
not claim to have identified the peer income effect for the whole of India. Also, the relevance
of this evidence for the calorie puzzle is largely suggestive in nature. However, in the limited
scope of this study I find no evidence that might suggest anything that runs contrary to the
theory presented in Chapter 2.
If the evidence presented here is to be believed, these findings also have some policy
implications. Firstly, it suggests that policy makers should account for the fact that just
aiming to increase incomes (like paying out cash transfers) for all may not help to achieve
goals like improving nutritional intake. In my estimates, the coefficient on own income is
positive but almost half of that of the peer group income. So calorie intake would not rise as
long as the agent does not perceive own income to be going up much more than her peer
group’s income. On the other hand, information dissemination drives that try to sell food
consumption as an important household goal (and may be fashionable goal) might help
temper the perception that nutrition comes second to taste or status.
68
Chapter 4: Impact of Public Distribution of Food in India on
Consumption of Nutrients.
4.1.Introduction
In the last two decades the Indian economy has experienced a sustained period of high
growth.22
Yet, this economic success has had very limited impact on the incidence of
malnutrition in the country. India's child malnutrition levels have been and are still
comparable to those in the poorest regions of the world (Gragnolati et al. 2005). Even more
alarming is the fact that increasing fractions of the population are slipping below the
recommended calorie consumption standards23
of the government even as their incomes and
consumption expenditures are going up (Deaton and Dreze 2009). In the light of this
situation, food security has to be one of the biggest concerns for policy-makers.
One recent step taken by the government in this direction is the Food Security Bill 2013.
This bill enhances the coverage of the public distribution of subsidised food (PDS) from
about 30% of the population to 70%. The move is the latest in a series of government efforts
designed to either streamline, re-focus or improve the efficiency of this gigantic public
program. From its humble origin as a largely urban based food rationing scheme under
British rule, the PDS has now grown into the biggest food security program of the Indian
Government, both in terms of its scale,24
and in terms of the investments involved.25
The
22
Indian GDP has been growing on an average at more than 5% per annum since the late 1980s. 23
These standards are 2100 Kilocalories for the urban areas and 2400 Kilocalories for rural areas per person per
day. 24
With more than four hundred and fifty thousand fair price grain outlets spread all over the country.
69
important question that is yet to be answered is whether the impact of this program on food
security and malnutrition in the country is significant and in proportion to its size and
expense. Also, is it possible to achieve the same impact using some other smaller, less
complicated mechanism? Some of these questions relevant to the PDS debate will be the
focus of the present paper.
By most accounts the PDS program is quite popular amongst its beneficiaries.26
Yet it has
been severely criticised for its inefficiency and ineffectiveness by both internal evaluations as
well as external agencies like the World Bank and the FAO. For example, it has been pointed
out that for every rupee of subsidy passing on to the targeted beneficiary the central
government has to spend Rs 4.27. Also it was estimated that in the absence of PDS poverty
in the whole country would have gone up by just 2 percent points, highlighting the
ineffectiveness of the PDS in impacting either poverty or food security in proportion to its
cost. (Radhakrishna et al. 1997).
Besides ineffectiveness the PDS also suffers from gross inefficiencies mainly due to
corruption leading to the leakage of grains into the black market (Jha and Ramaswami 2010).
Khera (2011a) estimates that in 2007-08 37.2 percent of the rice and 57.7 percent of the
wheat issued to the states for distribution from the central pool, were lost due to diversion.
These diversions are generally believed to be the reason why eligible households only
consume about 50 percent of their allotted quota (Svedberg 2012, Khera 2011 a, b). Due to
such alarming losses many are of the view that the distribution mechanism should be
replaced by cash transfers by the amount of the subsidy (Kapur et al. 2008, Kotwal et al.
25
The subsidy cost to the Central Government was 5.2% of its total expenditure in the 10th Plan period 2002-
07. (Source Planning Commission of India) 26
People usually travel the long distances to the PDS shops repeatedly in the hope of getting their share of
allocated grain although they are frustrated very often. Khera (2011a)
70
2011, Svedberg 2012). Others argue that the elasticity of income transfers and the effect
of PDS quotas are not equivalent due to the possibility of transferred cash being spent
on non-food items (Dreze 2010, Khera 2011b, Himanshu 2011, Cherian 2013). Besides, the
whole system of procurement and distribution also achieves other state goals like price
stability and famine prevention. Nevertheless, everyone would agree that the PDS needs to
improve its delivery mechanism27
and improve its targeting to better serve the needs of the
poor.
This chapter will attempt to contribute to this debate by answering the fundamental question
regarding the relevance of the PDS in terms of raising calorie and protein consumption in the
population. In simple terms: What would be the impact of an improvement in the supply of
PDS grain on nutritional intake? This efficiency of supply of the PDS could be a measure of
the ability of the system to make the allotted amount of food available for the people to buy.
The exercise is challenging mainly because of the peculiarities of the PDS system. The
government subsidises the grain but also imposes a quota which varies by state of residence,
income and over time. Besides, on many occasions, agents are unable to buy their full quotas
because of supply-side deficiencies on the part of the government suppliers. This implies that
an agent's observed PDS grain consumption is not always her quantity demanded or her full
quota (as simple demand theory would tell us), but it may be something rationed by supply
failures. This feature may be used to identify the relationship between PDS efficiency and
the consumption of nutrients. Factors that may affect the grain delivery mechanism of the
PDS would directly influence the amount of grain available for dispensation in the PDS
shops. One such factor is rainfall, especially in those areas that are large contributors to
27
According to Himanshu and Sen (2011) improving delivery is a higher priority than expanding entitlement
regarding improving the impact of the PDS.
71
the PDS stock. Also stock availability would impact different regions differently. In grain
surplus areas locally procured grain may be used in PDS, but in grain scarce areas
supplies have to be brought in from outside. A combination of these two is used in
this chapter to identify the effect, after controlling for preferences and demand
determinants as best as possible.
There are advantages gained by looking at nutrient (calorie/protein) consumption rather
than amount of grains delivered or bought. Firstly, nutrient consumption of the people has to
be the ultimate aim of any food-security measure. Secondly, there remains a concern that
agents may be buying from PDS shops and then selling them again at higher prices
(Bhagwati and Panagariya 2013). Looking directly at calorie consumption should help
circumvent this problem. Like all other subsidies, the PDS is also a market intervention and
is bound to affect nutrient consumption through various channels including local socio-
economic institutions. The first step here would be to properly identify econometrically the
net effect of PDS on nutrient intake and then to unravel the mechanisms if possible.
4..1 Institutional Background: Evolution of the Public Distribution System in India
The origins of the PDS go back to the early 1940s during the Second World War. The British
regime had introduced rationing of food items due to war time necessities. After
independence in 1947 the Government of India decided to continue with the system.
According to Dantwala (1993) about 54 million people in the urban areas were covered at
this time. The rationale behind urban rationing was that unrestricted markets would draw out
food grains from the rural hinterland in times of scarcity. At this stage the main focus was on
achieving price stability and preventing famines. Although this approach would soon change,
nevertheless it introduced an urban bias in the PDS that would persist for some time.
72
By 1956, the end of the first five year plan period, it was becoming clear that PDS had to be
transformed into a food security program and this transformation took place through the
second five year plan period (1957-1961). The total number of PDS outlets or fair price
shops went up from 18,000 to 51,000 (Nawani 1994). Soon the PDS organisation took its
present shape with the formation of the Food Corporation of India (FCI) in 1965. Every year
the government would announce a minimum support price for food-grains. Any unsold
stocks at the declared minimum price would be bought by the FCI and stored. These stocks
would then be issued out to state governments for distribution through the PDS and also for
maintaining a buffer stock for times of scarcity. The whole country was divided into 5 zones
with most of the grain surplus areas located in the North. Today the FCI maintains a huge
stock of over 54 million tonnes of food grain and distributes throughout the country via
492,000 fair price shops (11th plan, Planning Commission of India).
After the liberalisation of the Indian economy in 1991, the PDS came back into policy
debates mainly because of the huge subsidy burden it was imposing on the government. The
total subsidy for running the program had gone up from 0.04% of GDP in 1970-71 to 0.5%
of GDP in 1991-92. At this time there were a number of reviews of the PDS and its working.
These studies highlighted the large expense being incurred for very small gains. Various
recommendations were made for targeting subsidies to the poor in order to curtail the subsidy
burden (Ahluwalia 1993, Parikh 1994, Radhakrishna et al. 1997, Dutta and Ramaswami
2000, 2001, Indrakant 2000). Responding to the criticism, the government decided to go in
for targeting by income. In 1997 the Targeted Public Distribution System (TPDS) was
introduced. This scheme offered 10 Kgs of grains per month to households below poverty
line (BPL) at half the cost of procurement to the FCI. At the same time subsidies for
73
households above poverty line (APL) were totally eliminated. The quota for BPL households
was eventually raised to 35 kgs per month in 2002. Also many state governments have
tinkered with PDS entitlements on their own. For example Tamil Nadu persisted with the
universal PDS system while states like Andra Pradesh and Kerala have reduced their quotas.
A good summary of these changes can be found in Khera (2011b), Planning Commission
(2005).28
In spite of these changes the problems with the PDS still persist. According to the Planning
Commission's report for the 11th plan, subsidy cost for maintaining the PDS has gone up
from Rs. 51.7 billion in 1996-97 to Rs. 238.3 billion in 2006-07 which is more than an
increase of 3.5 times. Also identification of the poor has not been done well by the
state governments. According to the Planning Commission's survey report, although uptake
by poor households were much higher than previously, only about 57% of the BPL families
were being covered by the targeted PDS. Errors in identification exist both in inclusion as
well as exclusion with many "ghost" BPL cards going around.
As a remedy to the targeting problems the Government came up with the latest legislation in
2013, increasing the coverage of TPDS to 70% of the income distribution in the rural areas
and 50% in the urban areas. Each household will be entitled to 5 Kgs of grains per month at
very subsidised prices for the next three years. These changes might bring about a more
significant improvement in food security in the future but it is unlikely to be simply due to a
rise in entitlement. One major issue is the delivery mechanism and the leakage of grains to
28
See also the article "Simplifying the food security bill" published in the Hindu Newspaper on 12 March 2012.
http://www.thehindu.com/multimedia/archive/00948/Simplifying_the_NFS_948744a.pdf
74
the black market. This problem needs to be addressed and some suggested methods are cash
transfers or cash cards that take the distribution mechanism out of the equation entirely.
The main focus of this chapter, however, is not to suggest wholesale changes or
improvements in supply mechanism but to exploit exogenous changes in supply to estimate
the effect of current PDS consumption on nutrition.
4.2. Motivating the Empirical Setup
As mentioned before the objective of this chapter is to estimate the effect of PDS grain on
the consumption of nutrients by utilising exogenous shocks to PDS supply while controlling
for demand related factors as best as possible. This section attempts to explain the economic
logic behind this approach and the assumptions required for the purpose. Also, the
composition of the sample to be used for analysis is discussed.
The PDS grain consumption that we observe is the outcome of optimization by the
households. One of the key determinants of this optimum quantity would be the household's
income. Figure 4.1 plots a kernel estimation of PDS grain consumption as a function of
annual mean per capita consumption expenditure (MPCE) separately for 2004 and 2009.
Both curves are consistently downward sloping apart from a small section at the beginning.
This shows that for most of the sample PDS grain is an inferior good. PDS consumption is
increasing in MPCE only for the poorest.29
However when we consider an impact of supply
29
It might come as a surprise that most of this sample of PDS users are above the poverty line as indicated by
the vertical lines in Figure 4.1. However, it should be borne in mind that below poverty line status in India is
determined not just by income but by a host of different criterion via a BPL census conducted by the
government every 10 years. Scores are awarded to each surveyed household for each criterion and the final
decision is made according to a cut-off on these scores. For example, owning a motorised vehicle or a
refrigerator would debar you from BPL status, but being shelter-less or belonging to a primitive tribal group
would lead to automatic qualification into the group. Also, the different states have some independence in
determining the cut-off criteria used in their jurisdiction for BPL status determination. This diversity in
75
shocks to the optimum PDS grain consumption we need to take into account the response of
both these groups of households.
Figure 4.2 is a stylized representation of Figure 4.1 for purposes of illustration. Suppose the
household with income is the income for which PDS grain consumption peaks.
Any household with income less than for example , would be treating PDS grain as a
determining criterion for BPL status is the likely cause of the surprisingly low number of PDS users lying
below the Planning Commission's poverty line in this sample.
Notes: These lines depict the relationship between PDS grain consumption (wheat and rice)
and Monthly Per Capita Consumption Expenditure×12. The relationship is estimated by
local polynomial smoothing. The kernel function used is Epanechnikov and the degree of
polynomial used is 1. This estimation was done only on households with MPCE less than or
equal to a hundred thousand rupees. Two vertical lines are at the ten thousand and thirty
thousand rupees mark indicate the official poverty line for India according to the Planning
Commission and the poverty line used by the World Bank.
Source: NSS rounds 61st and 66th
76
normal good. Similarly households to the right of like will consider PDS grain as an
inferior good. However it should be pointed out that households like those with income like
are very few in the sample (refer to Figure 4.3 to see the distribution of MPCE in the
sample).
Next let us analyse how these different households might react to an exogenous supply shock
to PDS grain. While we expect the demand for PDS grain to be some sort of downward
sloping curve in price, the supply curve is dictated by the amount of the quota and the price
at which this quota may be bought. The supply curve facing any particular household is
contingent upon its income status (APL/BPL) and the state in which it is situated. However,
in general the supply function would look something like the curve labelled SS in
Figure 4.4. This figure represents demand and supply of PDS grain for a representative
77
consumer who is allotted a quota of Q kilograms of grain at price P. The quantity supplied is
.
anything between 0 and Q at price P, however at all other prices quantity supplied is 0. This
is the supply curve in an ideal situation. But, as described earlier, PDS is seldom able to
deliver the promised amount of grains to its consumers due to large scale illegal leakage into
black markets. Also buying from the PDS is also associated with travelling long distances to
reach the Fair Price Shop (FPS) and standing in long queues and then often returning empty
handed. In other words, the effective price paid by the consumer is higher than the
official subsidised price. Also, the quantity available for collection is below the quota
46
283
901
1963
3377
4251
4745
50944985
4643
4117
3673
3179
2720
2447
2133
17181587
13201197
0
10
00
20
00
30
00
40
00
50
00
Fre
qu
en
cy
0 20000 40000 60000 80000 100000income
Figure 4.3: Histogram of Income (MPCE), Sample of PDS users with MPCE less than
from NSS 61st and 66th rounds.
Notes: This figure shows the distribution of income/ monthly percapita consumption
expenditure (MPCE) for the sample of PDS users with MPCE less than Rs 1,00,000. Sample
includes all agents from 2004 and 2009 NSS surveys. Amounts are deflated to 2004 prices.
78
amount Q. In Figure 4.4 this is represented as a shift of the supply curve from SS to SE. Now
is the maximum amount available to the consumer and this at an effectively higher price
of .
Next we would like to investigate the implications of a positive supply shock to the PDS. In
such a case we may imagine that the effective price falls to and quantity available for
collection goes up to say . This means that the supply curve now shifts down (as
indicated by the arrow) to SE1. Now consider two different households represented by their
demand curves DD and D1.
79
If these consumers are from the region to the left of from Figure 4.2 then DD represents
the household with higher income amongst the two. If on the other hand the they are from the
range of income to the right of then D1 is the household with higher income. In either
case, however, it is clear that the supply shock leads to higher consumption of PDS grain.
Such a supply shock therefore, will lead to an upward shift in PDS grain consumption, as
shown in Fig. 4.2. This observation motivates the empirical specification used in this paper.
The supply shock introduced into the PDS system by random rainfall shock in PDS grain
supplier states is used to identify the effect of a rise in PDS grain consumption on nutrient
consumption in households. Of course, this approach relies on being able to control for the
factors affecting the demand curve. Also, if we look at Figure 4.1 the estimated relationship
between MPCE and PDS grain in 2004 and 2009 seem to be parallel shifts of each other. It is
unlikely that this change is due to people substituting into PDS grains (due to shift in
preferences). Such a response would have ensued if perhaps there had been a large negative
shock to incomes throughout the distribution (since PDS grain is considered inferior).
However the period between 2004 and 2009 was a period of rapid economic growth in
India. A more plausible explanation lies in the improvement of PDS delivery of food
grains which reduced the gap between the quota and the actual uptake. This is exactly what I
plan to use in the identification strategy.
In this exercise the sample restricts attention to households who have reported a positive
quantity purchased from PDS. This excludes agents who do not own a ration-card and are
ineligible to buy from the PDS, as well as those who chose not to buy from it. The latter
category can again be sub-divided, based on the possible reason for their choice. The richer
households may choose not to consume PDS grain considering it to be of inferior quality. On
80
the other hand, the very poor might be unable to buy from the PDS due to income
constraints. However given the drastically subsidised prices of PDS grain (ranging from Re
1/KG to Rs 5.5/KG for households below poverty line (BPL)), it is unlikely that a household
will be income constrained to buy at least a portion of its allotted quota. To put this in
perspective, consider that the highest possible expenditure on PDS grain for a BPL
household is Rs 210 per month (if the highest quota is purchased at the highest price). The
official poverty line income per day according to the Planning Commission is Rs. 26-33
per capita per-day, which is a monthly household income of about Rs 2400 assuming a
family of three. Therefore, the proportion of households dropping out because of income
constraints is likely to be small.
4.3. Data
National Sample Survey Data
The first exercise is to estimate the impact of PDS grain consumption on calorie/protein
intake. For this I use the NSS consumption data from rounds 66th and 61st (2009, 2004
respectively). These are the large rounds making the combined sample of more than
two hundred thousand households. This is a very detailed consumption survey at the
household level whose sampling method makes it representative at the district, state and
national levels. Both quantity and value of consumption are reported in most cases making it
possible to calculate nutrient consumption on the one hand and also the price paid by the
agent by dividing the value by the quantity. The nutrient consumption tables used in this
exercise are the same that are used by the NSS and the original source is Gopalan et al.
(1974). Data set also provides demographic and household characteristics of the surveyed
households.
81
In the entire sample, about 60,000 households bought rice from the PDS and about 30,000
bought wheat. The total number of households who bought anything at all is about 80,000.
This number is expected to be far less than the number who are eligible to buy. There is no
data about possession of ration cards (cards that show eligibility for PDS) in the 66th round
about 76% have ration cards amongst the households sampled in the 61st round. As such it is
difficult to identify the households below poverty line (BPL). The best that can be done is to
infer the type of household by checking the price paid since it is known that BPL household
receive a subsidised price. This information is used to build the main variable of interest
which is the gap between allotted quota and uptake of grains. Quotas vary by state, income
and type of grain. The source for information about PDS quotas used for this exercise is
Khera (2011b), Planning Commission (2005) and the article "Simplifying the food security
bill" published in the Hindu Newspaper on 12 March 2012. Although the best efforts were
made to ensure accuracy, it has to be admitted that the available information leaves some
room for confusion in some cases.
Another issue is the problem of identifying all households who have access to PDS. As
described earlier there is no information about ration cards holding in the 66th round data. So
most of the analysis here is conducted on a restricted sample of households who have
reported some quantity purchased from the PDS. The potential selection problem here arises
from people who are eligible yet choose not to consume. But the identification strategy (as
hinted at in the introduction) would rely on supply shocks to the PDS and these people would
be unlikely to be affected by these shocks anyway. Besides, by restricting the sample it is
made certain that the household has access to a fair price shop and has a ration card.
82
The summary statistics for some of the key variables for this restricted sample is reported in
Table 4.1a and Table 4.1b. The change from 2004 sample and 2009 sample is also reported.
83
There are two interesting aspects in Table 4.1a. Firstly, it shows that calorie consumption per
capita has fallen from 2004 to 2009. Although this is the general trend for the whole
population as reported by Deaton and Dreze (2009), it is a bit surprising to find the same
trend in a restricted sample of PDS users who would be in general poorer than the
population. Protein consumption however seems to have gone up in the sample which is
contrary to what is going on in the whole population. The second interesting aspect is that
over time PDS does seem to have expanded both in terms of coverage and grain
delivery. Both number of PDS users (number of observations) and quantity consumed of
PDS rice and wheat have gone up, this in spite of the 2004 NSS full sample being bigger
than the 2009 sample. Yet, the gap between quota and uptake has gone up, perhaps reflecting
the expansion in quotas and entitlements that have taken place in between these two years. In
all other respects the 2009 sample seems to be doing better.
85
4.4. Empirical Strategy and Results
4.4.1 OLS
We begin the analysis with a pooled OLS estimate where the main variables of interest are
household intake of rice/wheat from the PDS and the difference between quota allotted to the
household and total uptake.
The equation to be estimated is as follows:
Here represents the Monthly Per Capita household consumption of calories or
proteins in logarithm for household i in district d in time t in state s. is the main
variable of interest. It is either PDS uptake of rice or wheat or the difference between the
quota of food grains and the uptake of grains for household i in district d in time t in state s.
and are district and year controls, are state-time controls, are household level
economic and demographic characteristics and is the error.
The results are reported in Table 4.2. The OLS results are exactly as we would expect. Both
PDS consumption of rice and wheat have positive significant coefficients indicating that it
does contribute to nutrient consumption positively. The signs of the variable of interest also
change when we switch to quota-uptake differences. Negative significant coefficients here
indicate that PDS inefficiency impacts both protein and calorie consumption negatively and
significantly. Controls used are of three types. Firstly we use a set of demographic
controls like caste and religion of the household, next are household characteristics like size,
number of females and adults, method of cooking, land-ownership, whether they consume
out of their own produce and electrification and lastly there are a set of incomes (proxied by
86
Monthly Per Capita consumption expenditure MPCE30
) and prices of PDS and open market
purchases of wheat, rice and sugar (the three main items obtained via the PDS).
Occupation dummies following National Classification of Occupation codes 2004, and
seasonal dummies (using NSS sub-rounds31
) have also been included.
The size of the coefficients imply that a one kilogram increase in the quota - uptake gap leads
to a fall of 0.8 percent in daily calorie intake and 0.69 percent fall in daily protein intake.
Using the mean figures for the sample these numbers may be translated to a 15.2 kilocalorie
fall in average calorie consumption per person every day.
30
Following Hnatkovska et al. (2012), MPCE figures have been converted to 2004 prices using state-level
deflators calculated using state level poverty lines as reported by the Planning Commission. Planning
Commission (2004-05, 2009-10) 31
NSS collects data in four sub-rounds which are July-September, October December, January - March and
April - June. Controlling for these sub-rounds allows to take out seasonal variation in prices and grain demand.
88
4.4.2 Omitted Variable Bias and the Instrument
One major concern regarding identification is the presence of omitted variables that might
have prompted agents to self-select into the sample. Firstly, there are many government
sponsored welfare programs running at the same time (for example NREGA and IRDP32
).
Suppose a politically well connected household has better access to all these programs. Then
the political connection variable may explain both higher access to PDS and higher calorie
consumption giving us a spurious correlation. The other side of the story is that a politically
marginalised households may have very low access to resources and be forced to consume
from the PDS which may be looked upon as an inferior good. In either case calorie
consumption and PDS efficiency will be found to be spuriously correlated. These factors
essentially cause a selection bias where a particular kind of households (e.g. politically aware
or well connected) get self selected into the sample. In order to circumvent this potential
problem I need an exogenous variation in the quota-uptake difference which will not be
correlated with these unobservable factors. Now I will describe the instrument I use for this
purpose.
The PDS in India works by procuring grains directly from the farmers and then redistributing
them at subsidized prices through fair price shops. As mentioned, for procurement the
government declares a minimum price each year and buys up all the grains offered up at that
price. This operation of buying, storing, and supplying the grains is done by the
Food Corporation of India (FCI). Among the states, Punjab and Andhra Pradesh have been
the biggest
32
National Rural Employment Guarantee Act (NREGA) promises 100 days of employment at minimum wages.
Integrated Rural Development Programme (IRDP) provides loans to poor households at subsidised rates.
89
suppliers of rice to the FCI, whereas the supply of wheat is heavily dominated by Punjab (see
Fig. 4.5). The salience of Punjab as a supplier of the FCI leads one to expect that any
variation in the agricultural production of Punjab would affect the stocks of food grains with
the FCI and in turn the supply of grains at the fair price shops.
The variation in agricultural production is determined by various factors. Of these some are
totally random like rainfall. Although Punjab is a state with large tracts of irrigated land (one
of the reasons why it produces large surpluses to be sold to the FCI), its agriculture is still
rain dependent. One of the reasons for this is that irrigation itself is rain dependent. Drought
years may adversely affect agricultural output. Hence we should observe a link between the
rainfall in Punjab and the availability of PDS grain in all states in that year. This variation in
rainfall in Punjab is what I will be using to instrument for PDS leakage.
0
10
020
030
040
0
Ric
e p
rocure
men
t in
hu
nd
red
tho
usa
nd
ton
nes
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Total Punjab
Andhra Pradesh
Figure 4.5a: Rice Procurement from Major Supplier States.
Source: Food Corporation of India website
90
For this purpose I construct a variable using rainfall data from the state of Punjab over the
time period 1990-2009. The data is available at the website of Indian Institute of Tropical
Meteorology.33
For each of the years I construct a metric for above average rainfall by
subtracting the average rainfall over two decades i.e. 1990-2009, from the current year’s
rainfall.
Two assumptions are made in this argument. First, I am assuming that rain has some effect
on the agricultural output in Punjab. This may be safely assumed for most areas in India
since Indian agriculture is notoriously susceptible to the vagaries of the monsoon. Yet,
33
The rainfall data may be downloaded free from the website of Indian Institute of Tropical Meteorology at:
http://www.tropmet.res.in/
050
10
015
020
025
0
Wh
ea
t pro
cure
men
t in
hu
nd
red
tho
usa
nd
tonn
es
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Total Punjab
Haryana
Figure 4.5b: Wheat Procurement from Major Supplier States.
Source: Food Corporation of India website
91
Punjab is one of the states where irrigation through canals and pumps is used a lot, hence
there might be some suspicion that the rain dependence of Punjabi agriculture may have
lessened somewhat in recent years. To test this I looked at the total value of rice and wheat
production in Punjab from 2004 to 2010. The correlation coefficient between these figures
and my rainfall metric is 0.16 and 0.17 respectively. So although it is true that agriculture is
not as rain-dependent as it used to be earlier, there is still positive relation between the two.34
Another link to be examined is whether an increase in agricultural production in Punjab also
causes larger uptake of grains from Punjab by the FCI. Here as well there may be some
subtle reasons to question this very innocuous looking assumption. For example: in a year
when rains are good in the entire country Punjab would have high output, but there would be
extra grains to be brought up from other states as well. In such a scenario it is possible that in
good years less grain may be picked up from Punjab and more from other states which are
more dependent on rain. In years of low rainfall the bulk of FCI purchases are brought from
Punjab which is less rain dependent and more mechanized in its farming techniques. But
once again I find that the correlation between the uptake from Punjab by the FCI and value
of total production is 0.43 and 0.28 for rice and wheat respectively for the years between
2004 and 2010 (Source Ministry of Agriculture and Cooperation website, Government of
India). So the assumptions involved in the argument for the instrument seem to be valid.
The Punjab rainfall instrument has a major disadvantage, however, which is that it varies
only by year. This means that, with this instrument, year fixed effects and state-year fixed
effects cannot be used. Also, in India rainfall is often determined by the quality of the
34
Due to the rampant use of pumps for irrigation, some of the areas in Punjab are facing salinity and water table
depletion.
92
monsoon,35
which is liable to affect rainfall all over the country. Thus rain in Punjab may be
correlated with rain in other states, which in turn may have a bearing on consumption of
nutrients in those other areas. So it is essential that state-year fixed effects be used.
One solution to this problem would be to interact the basic rainfall metric with some other
variable which varies at the region, district, or household level which could enhance the
effect of the original variable. A similar strategy was used by Nunn and Qian (2012), where
they try to estimate the impact of U.S. food aid on conflict in Africa. Here they instrumented
for US food aid by an interaction between US agricultural production and the propensity for
a country to partake of US aid as represented by their average receipts of aid. In the current
setting, a similar strategy would be to use the differential impact of Punjab's rainfall on the
different zones that the PDS apparatus is divided into.
Zonal variation in PDS grain supply
One of the main functions of the Food Corporation of India is to move food-grains from the
surplus areas to the deficit areas. It is perhaps due to the good work of the FCI that the threat
of famine has largely disappeared from the country. For this purpose, FCI divides the country
into five zones which are North, South, East, West and the North-East. The North zone
includes the states of Punjab, Haryana, Uttar Pradesh which happen to be the major grain
surplus states both for wheat and rice. The South has states like Andhra Pradesh which is a
big surplus state for rice, but the others like Tamil Nadu, Karnataka and Kerala are more or
less neutral. In the East, West Bengal and Orissa do have surpluses from time to time but
would be best described as self sufficient. Most of the grain deficit areas fall in the West and
35
The seasonal winds from the sea which bring rain after the summer are known as the monsoons.
93
North-East zones. As such there has to be a lot of movement of grains out of the North zone
into the West and the North-East.36
This structure and the distribution of the surplus and deficit areas leads us to expect that any
rain shock to the Punjab would lead to differential impact on PDS supplies in different zones.
We would expect the biggest improvements to happen in the grain deficit areas and only
marginal or no improvements for the East or South.
The patterns in inter-zone grain transfers by the FCI become more apparent when we look at
the figures representing planned and actual despatches of grain into and from various zones.
These figures for the year 2012-13 are presented in Tables 4.3a and 4.3b. Table 4.3a shows
amounts of inter-state grain transfers by the FCI into states belonging to the five zones. In
terms of total volume of transfers the North-East definitely receives less than the other
zones, but this is not surprising considering that only about 3 percent of the total population
of the country lives in this zone. The size of the population in the other zones is clearly
reflected in the amounts of the grain transfers they get. However it should be noted that for
all zones except the North-East, these figures are liable to include intra-zone transfers as
well, that is transfers from other states belonging to the same zone. Also the fact that most of
the states belonging to the North are surplus grain producers is reflected in the relatively low
inter-state transfers into these states.
33
North East zone has Assam, Manipur, Meghalaya, Arunachal Pradesh, Mizoram and Nagaland.
94
Table 4.3b on the other hand, shows where the grain dispatches are coming from. While
there exists at least one state in each of the four big zones which contributes a non-zero
amount to the PDS stock in other states, there is no contribution of this kind from any state in
the North-East. This is hardly surprising if one takes into account the very low amounts that
are procured by the FCI from the North-East. Procurement of rice by zones is shown in
Figure 4.6 for the years 2000 to 2010.37
The procurement of rice from North-East is almost
negligible when compared to the other zones. So it is perhaps safe to assume that the North-
East would be totally dependent on other states for its PDS supplies. In particular it should be
dependent on the North which is not only the biggest supplier of grains but also sends out a
lot of grains to other states (Table 4.3b). At the same time the North receives very few inter-
state grain transfers (as evident from Table 4.3a) indicating that most of the grains sent out
from the northern states (like Punjab) end up in states outside the North zone.
37
The corresponding figure for wheat is not shown since the North and the states of Punjab and Haryana always
make up more than the 90% of the procurement while the South and North-East do not contribute at all.
95
The North-East zone is of special interest to this exercise for a number of reasons. First, it
happens to be a low producer of food grains and hence is largely dependent on the surplus
zones for its supplies. On top of that the North-Eastern states are quite remotely located and
as such provide ample opportunities for illegal diversion of food grains into the black market
while grains are being transported. This conjecture is consistent with the findings of Khera
(2011a) who reports that over 80% of the grains allocated to the state of Assam never reach
the people. Table 4.4 reports some descriptive statistics of key variables in the NE zone
along with the change from 2004 to 2009. The interesting thing in this table is that the PDS
seems to have made good progress during this time. Both the consumption of PDS rice and
96
wheat have improved and with no change in quotas during this time the gap between
quota and uptake has also reduced significantly. Along with this there has also been a rise
in the number of PDS users. But calorie consumption has still gone down on average and
protein consumption has shown no change.
4.4.3 Exclusion Restriction
This interaction between zones and the Punjab rain metric is to be used in a two stage least
square estimation as an instrument for quota-uptake difference. The main identifying
assumption here is that the differential impact of Punjab rainfall on the PDS supply in the
different zones is the only way it can affect calorie consumption in households. In order that
050
10
015
0
Zo
na
l R
ice P
rocu
rem
en
t, in 1
000
00
Ton
ne
s
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
North South
East West
North-East
Figure 4.6: Zonal Rice Procurement by Zones
Source: Food Corporation of India website
97
this assumption is not violated a number of precautions have been taken. Here the main
concerns about the validity of the instrument are discussed.
The first concern is that rainfall in Punjab could affect calorie consumption independent of
its effect on PDS efficiency. It is possible that higher rainfall in Punjab causes a nationwide
shock to the markets which affects agricultural output, agricultural prices and economic
activity all over the country. This might be another channel affecting calorie/protein
consumption in other states apart from its effect on the efficiency of the PDS. However these
concerns are taken care of through the state-year fixed effects which control for state specific
shocks for each year. Also prices of PDS and open market food grains (rice and wheat) and
sugar are controlled for to ameliorate the effect of unobservables on the grain market.
Also, rain in Punjab may be correlated with rain in other states which may cause a boom in
these states as well. From Figure 4.1 we know that a negative relationship exists between
income and PDS consumption, hence we may expect a boom year to drive down PDS uptake
in that year. This may lead to confounded results from the 2SLS estimation. In order to
address this concern the first step is to exclude Punjab and some other states which are close
to it (like Haryana and Himachal Pradesh etc.) from the sample since in these states we
would most likely be picking up the effects of rainfall on local conditions rather than PDS
efficiency. For the rest of the states it is unlikely that rain in Punjab would directly affect
local conditions. However, rainfall in Punjab may be correlated with rain in these other states
through the quality of the monsoons in that year. This year specific effect would be taken
care of using the year fixed effect and the state-year fixed effects.
98
There is still another possible scenario where the identifying assumption may be violated,
which is when rainfall in Punjab causes supply spillover effects which are zone specific
(apart from the PDS generated effects). However, this is unlikely since the zonal partitioning
is, to the best of my knowledge, particular to the PDS only. No other government programme
99
uses this partitioning system to either distribute resources or to fix target groups. As such it is
difficult to imagine reasons why any spillover effect would be zone-specific. If there are
spillovers dependent on geographical features or nearness to Punjab then the year state
effects should absorb that as well. A zone specific spillover effect, if any, would most likely
be caused by the resultant surge in PDS supply itself. Although with the current approach
this effect cannot be disentangled from the main effect, it may be argued that any attempt to
measure the effect of rise in PDS supply on calorie consumption should include any spillover
effects of a rise in PDS supply to that region anyway.
4.4.4 Two Stage Least Squares
So now, the first stage of the 2SLS is given as follows:
Here is Punjab's rainfall, measured as described above, in the current year. Variables
are now also indexed with the PDS zones which is represented by j. is a set of
dummies that indicate which zone the household belongs to (j takes five values indicating the
five PDS zones). Two separate specifications are run, differentiated by the set of dummies
used for the instrument. In the first specification only the NE dummy is used, which makes it
a comparison between households in the NE zone to the rest of India. In the second
specification all dummies except North are used, which implies a comparison of the different
zones with North as the reference group. Also, the dummy for the respective PDS zones
100
is also controlled for directly in each of the regressions. Standard errors are clustered at
the year-zone level.
The first stage results are reported in the first two columns of Table 4.5. While interpreting
these coefficients one needs to keep track of the comparison group. For example, the
negative significant coefficient for NE zone in column one indicates that rain in Punjab has a
negative impact on quota-uptake gap in the NE zone when compared to the rest of India, as
expected. On the other hand the coefficients in column two uses North zone as the
comparison group. Since Punjab itself is situated in the North zone it is expected that
negative impact on quota-uptake gap would be quite high for this region. Hence it is not
surprising that the South and East have statistically significant coefficients that are positive.
This indicates that the impact of Punjab's rain on the gap/inefficiency is smaller in these
areas when compared to the North zone. However, the West has an insignificant coefficient
indicating that the impact of rain is statistically equivalent to that in the North. The only
negative significant negative coefficient is for North-East which is exactly what one would
expect given this region's overdependence on outside suppliers38
.
The second stage results are reported in the last three columns of Table 4.5. The coefficients
of quota-uptake are negative and significant in the first specification (column 3, with only the
North-East interaction as IV) providing evidence to support the claim that PDS supplies or
efficiency is positively related to calorie consumption at the household. These results imply
that a fall of one kilogram in the quota-uptake gap leads to a 0.55 percent point rise in calorie
consumption (about 8.5 to 9 Kcal at the average). The magnitude of the coefficient is smaller
38
Although both specifications are jointly significant, the F-statistic for the specification with all four zonal
interactions is just higher than 8, slightly below the accepted norm of 10.
101
than those obtained by OLS in Table 4.2 which is perhaps because the OLS results were
upward biased due to the self selection of households with better access to PDS and other
public resources. As discussed earlier, this may have been due to their better knowledge
about these availability facilities or their better connectivity in the circles of political power .
However, when we look at the specification with all the zonal interactions as IV we find in
column 4 of Table 4.5 that the coefficients in the second stage are positive and significant.
This is true for both calorie consumption (rise of 0.28 percent for a rise of 1 KG in the gap)
and protein consumption (a rise of 0.7 percent). This result is surprising and goes contrary to
both the first IV specification as well as the OLS results of Table 4.2. Before attempting to
interpret this I would like to point out that the first stage of this particular specification is not
as strong as the first one. F-statistics for the first stage is just over 8 which would lead one to
suspect that the first stage is slightly weak. This could be due to the fact that for regions that
do not depend so much on grain from outside (like south and west) the Punjab rainfall may
not be such a good predictor of the quota-uptake gap. Hence, using all the zonal interactions
somewhat dilutes the effectiveness of the instrument.
If we assume that the coefficients are correct these results (columns 4 and 5) provide
evidence in favor of the critics of PDS. However, the usual argument involving the
inefficiency of the PDS cannot be invoked to explain this. These results imply that
households who are less exposed to PDS inefficiency also consume less protein and calories.
One explanation could be that PDS grains are of much lower quality and switching from
open market grains to PDS grains reduces the quality of nutrition intake. For this to be true
the household in question has to be rich enough to afford the same amount of grain at open
market prices as it could under PDS prices. From the evidence of Figure 4.1 and 4.3 the
102
sample of PDS users do seem to be enjoying quite high consumption expenditures. However,
this explanation is not too persuasive since calorie content of both PDS and open market
grains have been taken to be the same in this exercise.
A second explanation for this result with all zonal interactions may be suggested on the basis
of the trends in calorie consumption and PDS efficiency. As pointed out in detail in the
previous chapters average calorie and protein consumption in India is following a declining
trend. If this trend coincides with a rise in PDS delivery efficiency and coverage (evidence
provided in Table 4.6 below) then we could be spuriously observing the results in the last
two columns of Table 4.5. However time fixed effects were included in the regression
essentially to take care of this eventuality.
The third and most plausible explanation is that the benefits of PDS efficiency is manifested
only on certain sections of the wealth distribution and calorie-consumption spectrum. It
could also differ by regions. When we make a country-wide analysis using the entire income
distribution much of the effect is confounded. In order to unravel the causes of these results it
may be useful to carry out the analysis at various quantiles of the income distribution.
Finally, to provide more insight about the relationship between income (MPCE) and the main
variables of interest (i.e. quota-uptake difference and calorie consumption), Table 4.6
provides a breakup of the sample in terms of MPCE and then displays how different income
groups have evolved from 2004 to 2009 in terms of quota-uptake difference and calorie
consumption. From this we can see that quota-uptake difference has gone down for all
income groups but the fall is much lower for high income groups. Regarding calorie
consumption it seems that low income groups have seen an increase whereas high income
groups show a decline.
105
4.6. Conclusion
The chapter looks at the impact of the PDS on nutrient consumption in some detail.
However, on the basis of the evidence presented here it is still difficult make an
unambiguous statement on the effectiveness of the PDS. The strongest and most robust
specification estimated here is the one with North-east zone interacted with Punjab Rainfall
and this tells us that PDS efficiency does have a positive impact on calorie consumption.
Using these estimates it is possible to make rough evaluations of the impact of the PDS on
the current sample of PDS users. For example, in the event that the PDS is scrapped the grain
consumption derived from the PDS in this sample would go down from 16.85 Kilograms per
month to 0 and the quota-uptake gap would go up by the same amount. This would lead to a
fall in calorie consumption by more than 9 percent or by an absolute amount of 176
kilocalories per day. The average calorie consumption which was just more than 1900
Kcal/day would now go down to just over 1724 Kcal/day.
However this figure is liable to be a gross overestimate because funds that were used to buy
from the PDS would now be used to buy its substitutes from the open market. One crude way
to take this into account would be to calculate the average amount of money freed by the
absence of the PDS and calculate the impact of this amount on calorie consumption using our
coefficient on consumption expenditure. We can also compute the actual subsidy provided
by the government to BPL households by subtracting the central issue prices of the BPL
from the APL. So we can add to the income effect from above the impact of giving the
households a cash transfer equal to the amount of the subsidy. Even after controlling for all
this the loss in per day calorie consumption using the estimates and averages from this
106
sample is still about 150 Kcals, which is 8 percent of the average calorie consumption in the
sample.
However the rough estimates reported above use the sample averages of PDS consumption
and MPCE. We are more interested in the effect that PDS has on households below the
poverty line. BPL households are of course difficult to identify. In fact if we go by the
current poverty line of Rs 28-32 per day of the Planning Commission then the sample of PDS
users contain less than 50 of such households (refer to Figure 4.3). However the
identification of BPL households involve various other criteria that often differ by state. To
be realistic I choose for evaluation the threshold of Rs 40,000 MPCE which is just above the
international poverty line of $ 1.25 PPP per capita per day. At this level of income PDS
consumption is much higher, at about 20 Kgs/month39
and the impact of PDS as calculated
for the full sample above comes to about 170 Kilocalories per person per day. However, it
has to be noted that the coefficients used to generate this figure are estimated at the mean and
are liable to be different if estimated at other points of the income distribution. In the absence
of the relevant coefficients the calculations can be re-done under the assumption that the
money freed by the absence of the PDS grain is spent to buy grains in the open market at the
APL prices (proxy for the market price). In this scenario the negative impact on calorie
consumption is only due to the loss of income of the amount of the subsidy which translates
into just 1 percent drop in calorie consumption. However, given that most studies find a
income elasticity of food to be less than 1 (see Subramanian and Deaton 1996 and Behrman
and Deolalikar 1987), this scenario seems implausible. The real effect would lie somewhere
in between these two figures.
39
Using local polynomial smoothing estimates of the relationship between MPCE and PDS grain consumption
for the sample of PDS users with MPCE less or equal to a Rs 100,000.
107
The other thing that this chapter sheds light on is the nature of inter-regional relationship in
the PDS. While the North zone is the biggest supplier of excess grains, NE and West are the
most significant beneficiaries. While the South and East are quite self-sufficient and also
efficient in terms of their PDS, they are not major users of excess bounty from the North. As
such, outcomes in certain regions respond more than others if PDS is changed or modified.
Since NE and West are more food insecure than the rest, it would be useful for policy makers
to keep this in mind.
From this evidence it does seem that the PDS is having significant impact against
malnutrition . However it is yet to be seen whether the price the government is paying for
this benefit is worth it or not. Counterfactuals where the PDS is replaced by direct cash
transfers may also be estimated in future research by actually estimating demand functions
for PDS grain.
108
5. Conclusion
Amongst all nations of the world, India today suffers from the dubious distinction of housing
the largest number of poor and malnourished people. It is undoubtedly a significant challenge
to the discipline of Development Economics. The present thesis is an attempt to contribute to
the understanding of the factors and mechanisms that help determine food-security and
nutrition in this country. In this attempt the thesis concentrates on two different aspects
where one is related to the demand of food whereas the other pertains to the supply side.
Chapters 2 and 3 are an investigation of the role of status competition and conspicuous
consumption in determining choice of food and in turn calorie consumption. Chapter 2
begins by examining the consequences of having preferences that reflect status competition
(Keeping up with the Jones) in the simplest setting. Then as more factors like prices and
incomes are allowed to be determined endogenously in the model a clearer picture of the
potential of Veblen consumption for influencing calorie consumption emerges. The main
takeaways from this chapter is that status competition generates a negative relationship
between peer group income and food consumption holding everything else constant.
However, in a general equilibrium setting with productivity driven economic growth these
same preferences could generate falling calorie consumption provided that the effect of rising
own income was small, or there was rising income inequality within the agent's peer group.
The findings in this chapter give us a few implications which could be either used to test for
Veblen preferences with the appropriate data or at least allow for comparison with the actual
scenario using stylised facts that we know. For example, it is shown that if own income
effect is modest, any economic growth that leads to a rise in incomes keeping relative price
109
of food constant is liable to reduce food consumption for all groups. This is exactly the
situation in India as described by Deaton and Dreze (2009) during 1985-2004.
Chapter 3 is the empirical investigation that seeks to test the assumptions made in Chapter 1
and also provide some evidence to vindicate the role of status competition in generating the
fall in calorie consumption in India. Using data from rural India and exploiting a unique
source of variation in caste income, estimates of the effect of peer group income on calorie
consumption are obtained. As predicted by the theory, these estimates are negative and
significant. Moreover the magnitude of own income effect is far lower than the negative
effect of peer income. Since this was one of the conditions for calorie reduction outlined in
Chapter 2, this finding reinforces the possibility that status competition has been significantly
contributing to it. Back of the envelope calculations indicate that this factor could be
responsible for a third of the missing calories.
The third chapter also gives us some idea about the mechanism through which status
competition could be influencing calorie consumption. The data does not support the
hypothesis that resources hitherto being used for food purchases are now being used for non-
food items. However, there is some evidence indicating that peer group income increases
calories derived from non-cereal foods, dairy and meat products. This indicates that rather
than considering food in general as a non-conspicuous consumption good, agents only
consider certain food items to be as such. In fact they are shifting away from some of the
traditional foods like cereals towards tastier foods like milk and meat. This finding is
consistent with evidence presented by others (Deaton and Dreze 2009) which shows that
while calorie consumption is falling on an average fat consumption is on the rise. It may also
explain why there is no significant decline in food expenditure.
110
While Chapters 2 and 3 shed light on one of the determinants of food demand in rural India it
may be argued that for the poor food supply is usually the bigger concern. Adequate food
supplies, made available at affordable prices, is the essence of food security. Fortunately,
unlike food demand, this is an area which can be targeted through government policy. In the
fourth chapter the focus is shifted to the supply side. This chapter aims to evaluate the
efficiency and effectiveness of the biggest food security scheme of the Indian government,
namely, the public food distribution system. The effect of a rise in PDS grain supply on
nutrient consumption is estimated using the differential effect of rain induced PDS supply on
the different PDS regions across the country. Using these estimates it is possible to calculate
the impact of the PDS system on calorie consumption under different assumptions.
The fourth chapter also throws some light on the nature of regional distribution of grains by
the PDS. Due to the lack of domestic production in the Northeast and West zones these
regions are heavily dependent on supplies from other areas. These supplies generally come
from the North zone where most of the surplus grain areas like Punjab and Haryana are
located. This feature builds in a natural variation in the PDS mechanism used to deliver
grain to the different regions. The chapter uses this variation for identification.
The results also contribute to the debate about whether subsidised food (as is the PDS) or
cash transfers to the amount of the intended subsidy should be the preferred method to
achieve food security. However in order to answer this question properly, cross price
elasticities and other income elasticities of demand need to be estimated as well. With the
present level of understanding it is only safe to say that there is a possibility that the
replacement of PDS by cash transfer could negatively impact calorie consumption by as
111
much as 9 percent of the present level. In the present scenario, where calorie consumption on
an average is low and falling, this sort of impact could prove to be disastrous.
112
Bibliography
1. Ahluwalia, D. (1993) " Public distribution of food in India : Coverage, targeting and
leakages". Food Policy Volume 18, Issue 1, February 1993, Pages 33–54
2. Alesina, Alberto, Reza Baqir and William Easterly (1999) ‘Public Goods and Ethnic
Divisions’, Quarterly Journal of Economics 114: 1243–84.
3. Anderson, S., (January 2011) “Caste as an Impediment to Trade”, American Economic
Journal: Applied Economics 3: 239–263
4. Banerjee, A.V. and Pande, R. (2007) “Parochial Politics: Ethnic Preferences and Politician
Corruption” BREAD working paper No. 152
5. Banerjee, A., and Duflo, E. (2011) “Poor Economics: A Radical Rethinking of the Way to
Fight Global Poverty”, Public Affairs, New York
6. Banerjee, A., and S. Mullainathan, (May 2010 ) “The Shape of Temptation: Implications for
the Economic Lives of the Poor” NBER Working Paper No. 15973,
7. Behrman, Jere R., and Anil B. Deolalikar. (1987) "Will developing country nutrition improve
with income? A case study for rural South India." The Journal of Political Economy 95.3:
492-507.
8. Bhagwati J. and A. Panagariya, (2014) 'Why growth matters : how economic growth in India
reduced poverty and the lessons for other developing countries'. Chapter 15 "Adult Nutrition
and Food Security". Public Affairs.
9. Card, D. (2001). 'Estimating the return to schooling: Progress on some persistent econometric
problems'. Econometrica, 69(5), 1127-1160.
10. Charles, K. K., Hurst, E., and Roussanov, N., (2009) “Conspicuous Consumption and Race”,
The Quarterly Journal of Economics, May
11. Cherian M. (2013): “Cash transfers are bad for food security”, The Hindu, 11 May
12. Clark, Andrew E. Paul Frijters and Michael A. Shields. “Relative Income, Happiness, and
Utility: An Explanation for the Easterlin Paradox and Other Puzzles (Mar., 2008) "Journal of
Economic Literature , Vol. 46, No. 1, pp. 95-144
13. Clark, Andrew E, and Andrew J. Oswald. (1996) “Satisfaction and comparison income”.
Journal of Public Economics 61 359-381
14. Coelho, P. (1985). “An examination into the causes of economic growth: status as an
economic good”, Research in Law and Economics, vol. 7, pp. 89–116.
15. Cooper, B., Garcia-Penalosa, C. and Funk, P. (2001). “Status effects and negative utility
growth”, Economic Journal, vol. 111, pp. 642–65.
113
16. Dantwala, M.L., (1993), 'Agricultural Policy: Prices and Public Distribution System',
Indian Journal of Agricultural Economics, Vol 48, No.2, April - June
17. Deaton, A., & Dreze, J. (2002). Poverty and inequality in India: a re-examination. Economic
and Political Weekly, 3729-3748.
18. Deaton, A., and Dreze J., (2009) “Food and Nutrition in India: Facts and Interpretations,”
Economic and Political Weekly, 44(7), 42–65.
19. Drèze, J., P. Lanjouw, and N. Sharma. (1999). “Economic Development in Palanpur,
1957–93.” In Economic Development in Palanpur Over Five Decades, ed. Peter Lanjouw and
Nicholas Stern, 114–242. Oxford, UK: Oxford University Press.
20. Drèze, J. (2010): “Unique Facility, or Recipe for Trouble”, The Hindu, 25 November.
21. Duesenberry, J. S. (1949). Income, saving, and the theory of consumer behavior (Vol. 87).
Harvard Univ Pr.
22. Du, S, B Lu, F Zhai and B M Popkin (2002): “A New Stage of the Nutrition Transition in
China”, Public Health Nutrition, 5(1A), pp 169-74. Eaton, B. C., and Eswaran, M., “ Well-
being and Affluence in the Presence of a Veblen Good”, The Economic Journal, 119 (July),
1088–1104.
23. Dupor, B. and Wen-Fang Liu (Mar., 2003) “Jealousy and Equilibrium Overconsumption”,
The American Economic Review, Vol. 93, No. 1, pp. 423-428
24. Dutta, B. and B. Ramaswami (2001) " Targeting and Efficiency in the Public Distribution
System: Case of Andhra Pradesh and Maharashtra". Economic and Political Weekly, vol.36,
No. 18 (May 5-11, 2001), pp. 1524-1532
25. Dutta, B. and B. Ramaswami (2000) 'Reforming Food Subsidy Schemes in India:
Estimating the Gains from Self-Selection', mimeo, Indian Statistical Institute
26. Easterlin, Richard A. (1974). "Does Economic Growth Improve the Human Lot? Some
Empirical Evidence." In Nations and Households in Economic Growth: Essays in Honor of
Moses Abramovitz, ed. R. David and M. Reder. New York: Academic Press, 89-125.
27. Easterlin, Richard A. (1995). "Will Raising the Incomes of All Increase the Happiness of
All?" Journal of Economic Behavior and Organization, 27(1): 35-47.
28. Easterly, W. and R. Levine (1997) ‘Africa’s Growth Tragedy: Policies and Ethnic Divisions’,
Quarterly Journal of Economics, 112(4): 1203–50.
29. Eaton, B. C., and Eswaran, M., (2009) “ Well-being and Affluence in the Presence of a
Veblen Good”, The Economic Journal, 119 (July), 1088–1104.
30. Eswaran, M. and Kotwal. A., (1993) “A Theory of Real Wage Growth in LDCs”, Journal of
Development Economics 42 243-269.
114
31. Fred Arnold, Sulabha Parasuraman, P. Arokiasamy, and Monica Kothari. (2009). Nutrition in
India. National Family Health Survey (NFHS-3), India, 2005-06. Mumbai: International
Institute for Population Sciences; Calverton, Maryland, USA: ICF Macro
32. Galli, Jordi. (1994) "Keeping Up with the Joneses: Consumption Externalities, Portfolio
Choice and Asset Prices", Journal of Money, Credit, and Banking, February, 26(1), pp. 1-8.
33. Gopalan, C, Rama Sastri, B.V & S. C Balasubramanian. (1974) “Nutritive Value of Indian
Foods” National Institute of Nutrition Indian Concil of Medical Research Hyderabad.
34. Gragnolati M., M. Shekar, M. Das Gupta, C. Bredenkamp and Yi-Kyoung Lee. (2005), HNP
Discussion paper: “India’s Undernourished Children: A Call for Reform and Action”
35. Grossman, Gene M., and Elhanan Helpman. (1991) "Quality ladders in the theory of growth."
The Review of Economic Studies 58.1: 43-61.
36. Himanshu and A. Sen (2011): “Why Not a Universal Food Security Legislation?”, Economic
& Political Weekly, 46(12): 38-47.
37. Himanshu (2011): “Food or Cash: The Subsidy Conundrum”, Mint,
http://www.livemint.com/2011/03/01211232/Food-or-Cash-the-subsidy-conu.html?h=B
38. Hnatkovska, V., Lahiri, A., & Paul, S. (2012). Castes and labor mobility. American
Economic Journal: Applied Economics, 4(2), 274-307.
39. Hopkins, E. and Kornienko, T. (2004). “Running to keep in the same place: consumer choice
as a game of Status”, American Economic Review, vol. 94, pp. 1085–107.
40. Indrakant S (2000): "Food Security and Public Support: A Study of Andhra Pradesh" in N
Krishnaji and T N Krishnan (ed), Public Support for Food Security, Sage, New Delhi
41. Jayachandran, S., & Pande, R. (2013). Why Are Indian Children Shorter Than African
Children?. Working paper.
42. Jha, S., & Ramaswami, B. (2010). “How can food subsidies work better? Answers from India
and the Philippines.” (September 1, 2010). Asian Development Bank Economics Working
Paper Series, (221).
43. Jorgenson, Dale W. (1961) "The development of a dual economy." The Economic Journal
71.282: 309-334.
44. Kapur, D., P. Mukhopadhyay and A. Subramanian (2008): “The Case for Direct Cash
Transfers to the Poor”, Economic & Political Weekly, 43(15): 37-43.
45. Keely, Louise C. (2005) “Why isn’t growth making us happier? Utility on the hedonic
treadmill”, Journal of Economic Behavior & Organization Vol. 57 333–355
46. Khera, R. (2011a). "Trends In Diversion Of PDS Grain". Economic and Political Weekly,
Vol - XLVI No. 21, May 21.
115
47. Khera, R. (2011b). 'Revival of the Public Distribution System: Evidence and
Explanations'. Economic and Political Weekly, Vol - XLVI No. 44-45, November 05.
48. Khera, R. (2011c). "India’s Public Distribution System: Utilization and Impact." Journal of
Development Studies, March 07.
49. Kochar, A. (2005). "Can Targeted Programs Improve Nutrition? An Empirical Analysis
Of India’s Public Distribution System." Economic development and cultural change, 54 (1),
20335.
50. Kotwal, A., M. Murugkar and B. Ramaswami (2011): “PDS Forever?”, Economic &
Political Weekly, 46(21): 72-76
51. Kotwal, A. B. Ramaswami, W Wadhwa. (2011) “Economic Liberalization and Indian
Economic Growth: What's the evidence?” Journal of Economic Literature, vol. 49, issue 4,
pages 1152-99
52. Lewis, W.A., (1954), Economic development with unlimited supplies of labour,
Manchester School of Economic and Social Studies 22, 139-191.
53. Li, N., and Eli, S., (2010) “In Search of India’s Missing Calories: Energy Requirements and
Calorie Consumption”, November
54. Linssen. R., L. van Kempen, and G. Kraaykamp. (2010) “Subjective Well-being in Rural
India: The Curse of Conspicuous Consumption”, Published online: 12 May 2010
55. Luttmer, P. (2005). “Neighbours as negatives: relative earnings and well-being”, Quarterly
Journal of Economics, vol. 120, pp. 963–1002.
56. Metcalf, Thomas R. (1979). Land, Landlords, and the British Raj: northern India in the
nineteenth Century. Berkeley: University of California Press.
57. Mehta, J. and Shanta Venkatraman. (Jul. 1-7, 2000) “Poverty Statistics: Bermicide's Feast”
Economic and Political Weekly, Vol. 35, No. 27, pp. 2377-2379+2381-2382
58. Meng, X, X Gong and Y Wang (2009): “Impact of Income Growth and Economic Reform on
Nutrition Availability in Urban China: 1986-2000”, Economic Development and Cultural
Change, 57(2), 261-95.
59. Munshi, K. and M. Rosenzweig (2006), “Traditional Institutions Meet the Modern World:
Caste, Gender, and Schooling Choice in a Globalizing Economy” The American Economic
Review, Vol. 96, No. 4 pp. 1225-1252
60. Munshi, K. and M. Rosenzweig. (2005). “Why is Mobility in India so Low? Social
Insurance, Inequality, and Growth.” Bureau for Research and Economic Analysis of
Development (BREAD) Working Paper 097.
116
61. Munshi, K. and M. Rosenzweig, (2009) Why is mobility in India so low? Social insurance,
inequality, and growth. No. w14850. National Bureau of Economic Research.
62. Nawani, N.P. (1994) " Indian experience on household food and nutrition security" Food and
Agricultural Organisation of United Nations. Chapter 6
63. Neale, Walter C. (1962). Economic Change in Rural India; Land Tenure and Reform in
Uttar Pradesh, 1800–1955. New Haven: Yale University Press.
64. Nunn, N., and N. Qian. (2012). “ Aiding Conflict: The Impact of U.S. Food Aid on Civil
War”. Working Paper 17794. Cambridge, MA: National Bureau of Economic Research.
65. Palmer-jones, R., & Sen, K. (2001). "On Indian Poverty Puzzles and Statistics of Poverty" .
Economic and Political Weekly , 201-17.
66. Panagariya, A. (2013). Does India Really Suffer from Worse Child Malnutrition Than Sub-
Saharan Africa?. Economic & Political Weekly, 48(18), 98-111.
67. Parikh, K (1994): 'Who Gets How Much from PDS - How Effectively Does It Reach the
Poor?', Sarvekshana, January-March
68. Patnaik, U. (2004). "The Republic of Hunger". Social Scientist , 9-35.- (2007) “Neoliberalism
and Rural Poverty in India”. Economic and Political Weekly 3132-50
69. Planning Commission (2005). 'Performance Evaluation of Targeted Public Distribution
System (TPDS).' Programme Evaluation Organisation. Government of India.
70. Planning Commission (2007, 20012), 'Press Notes on Poverty Estimates'
71. Radhakrishna, R. (2005). “Food and Nutrition Security of the Poor: Emerging Perspectives
and Policy Issues”. Economic and Political Weekly 1817-21
72. Radhakrishna, R. K. Subbarao with S. Indrakant and C.Ravi. (1997). " India's public
distribution system: a national and international perspective" World Bank discussion papers ;
no. 380
73. Ramaswami, B. and P. Balakrishnan (2002) "Food prices and the efficiency of public
intervention: the case of the public distribution system in India" Food Policy 27 (2002)
419–436
74. Ray, R & Lancaster, G (2005) “On Setting the Poverty Line Based on Estimated Nutrient
Prices: Condition of Socially Disadvantaged Groups during the Reform Period”,Economic &
Political Weekly, 40(1), 1 January, 46-56.
75. Smith, A. (1759) “The Theory of Moral Sentiments” London, A Milar
76. Srinivas, M.N. (1952). “ Religion and Society Among the Coorgs of South India”
(Oxford, 1952), p. 30:
117
77. Subramanian, S. and Angus Deaton, (1996) “The Demand for Food and Calories”, Journal of
Political Economy, Vol. 104, No. 1, pp. 133-162
78. Svedberg, P. (2012) 'Reforming or Replacing the Public Distribution System with Cash
Transfers?' Economic & Political Weekly Vol XLVII no 7
79. Veblen, T. (1899). “The Theory of the Leisure Class: An Economic Study of Institutions”,
New York: Macmillan Company.
118
Appendix to Chapter 2
Proposition 1: Consumption of the non-Veblen good falls with the average income of the
group once the agent has started consuming Veblen good.
Proof: This can be shown for any general utility function of the form assumed in the text.
The agent’s problem is as follows:
Taking the Lagrange multiplier to be λ we have the first order conditions as follows:
This of course assumes that the poor have enough money to consume some Z. Then totally
differentiating with respect to we have:
The RHS is less than 0 since , , and average Z is sure to go up with
average income as long as any single person has been satiated with food in the group.
119
Solving the General Equilibrium
Using the production functions and resources specified in the main body of the paper we may
calculate incomes of the two types of agents (landlords and landless labourers). Agents
receive wages for their labour which is equal to the marginal product of labour and the
landlords get rent from land which is also the marginal product of land. So if w and v are the
wage and rental rates respectively we can write the incomes of the two groups as follows
Also profit maximization for the industrial good implies that marginal cost of producing one
more unit of industrial good is equal to the wage. So we have:
p = w/B ... (A.3)
Also assume that
... (A.4)
and
... (A.5),
which are the market clearing conditions for land and labour. Next we use Walras’ law to
drop the market clearing condition for the industrial good, and hence we can concentrate on
the food market clearing condition using the demand functions given in equation 2.1 in the
main body of the chapter.
120
The food market clearing equation can be of three kinds depending on which equilibrium we
are in. The first is the case where neither group is rich enough to consume Z. So all income is
spent on food. The supply of grain is given by the production function. So in equilibrium we
must have:
Next if the landlords have been satiated with food but the workers have not we have:
Here I have used equations (A.1) and (A.3) to substitute the values of p and w in the food
threshold expression. Lastly we come to the last equilibrium where both income groups are
satiated with food and have started consuming Z:
Since the relative price of the industrial good and the incomes can be expressed in terms of
we can write all three above equations in terms of the labour allocation to agriculture.
From here we calculate which solves the general equilibrium because the incomes and the
relative price of the Veblen good p, can be easily calculated using equations (A.1), (A.2) and
(A.3).
Proof of Proposition 2
Proposition 2: In an equilibrium where all agents are satiated with the necessary good X,
allocation of labour to the agricultural sector will fall with any rise in either industrial or
agricultural productivity.
121
Proof: The market clearing condition for X determines the allocation of labour to the
agricultural sector . This is the key variable in this model which can be used to then
calculate all the remaining incomes prices and consumptions. This key equation is given by
equation (A.6.3) as:
Here
as given by equation (A.3) and
.
Substituting in the values of we can write the relevant grain market equilibrium
condition (equation (A.6.3)) as follows:
It can be easily verified that increases in both A and B would cause the LHS to increase
leaving the system in disequilibrium. In order to re-equalise the two sides will have to
adjust. So it remains to be seen how the LHS responds to a rise in . If LHS rises with
then a rise in productivity parameters A and B would cause a fall in the equilibrium value
for in this model. Differentiating the LHS with respect to we have the following:
122
If the expression (A.8) above turns out to be positive then we may conclude that allocation of
labour to agriculture falls with a rise in productivity. The expression A1 has two parts. The
first of these, which is always positive, goes to infinity as (as long as ). The
second part is negative and its magnitude is highest at . This is a finite quantity given
by
So since the positive part can be very large and the negative part is finite for any parameter
value we may conclude that there always exists some for which (A.8) is positive and
falls with a rise in productivity. It remains to be seen whether expression (A.8) is positive or
not at the equilibrium value of . In order to check this I substitute into (A.8) the equation
(A.7) which gives us the equilibrium value of .
Equation (A.7) may be rewritten as follows:
Substituting in the value of the LHS from above into expression (A.8.1) we have:
123
This above expression is always positive as lies in between 0 and 1. Thus (A.8.1) is
always positive at the equilibrium . Hence allocation of labour to the agricultural sector
always falls when either of the productivity parameters is raised.
Model with Positive Income Effect
The fact that we observe this falling calorie consumption is due to the two important
assumptions. First is the assumption of “Keeping up with the Joneses” (KUJ) which pushes
up the marginal utility of Z with rising average income. The second important assumption is
that of quasi-linearity which means that there is no income effect to the consumption of X.
For a necessary good like X having no income effect at high levels of income is resonable.
But one may want to see what happens in the model if there is some income effect. In order
to do that I need to remove the quasi-linearity from the preferences. Here I apply the simplest
modification conserving KUWJ and jealousy but replacing the linear term with Z by a
concave one:
124
Here measures the strength of KUJ feature in the utility function. Higher is greater is
the rise in marginal product of Z with peer group income/ Z consumption. With this utility
we can calculate the implicit demand functions as follows:
As expected this demand also goes down with group income, but in this case food demand is
also positively related to own income. So whether food demand falls or not depends on how
own incomes and the peer group income change in the general equilibrium.
Proposition 4: A sufficient condition for the equilibrium labour allocation to agriculture to
fall with a rise in industrial or agricultural productivity is:
Proof: The equilibrium condition for the market for X is as follows:
Substituting in the values in terms of , and with some simplification this can be reduced to
the following:
Clearly, the LHS goes up with any increase of A. But a rise in B causes LHS to rise only
if . Let us suppose for the time being that . Now, in order to prove the
proposition we need to see if the LHS rises with . Differentiating we get:
125
The expression (A.10) above is clearly positive for all . But this is not necessary. Even
for higher values of the expression (A.10) may turn out to be positive for some values
of . To see this I look at the value of expression (A.10) in equilibrium by substituting
equation (A.9) into expression (A.10):
Again this expression cannot be negative if . But even for higher this may be positive
for low values of .
The intuition for this result is slightly involved. First note that if food demand will
increase with a rise in B. To see this, note that a rise in B impacts food demand through two
different channels. Firstly, through “Keeping up with the Jones”, via a rise in
average/aggregate production of Z at the same labour allocation. This effect reduces food
demand. It is in turn countered by the fall in price of Z (which increases demand for food).
126
If the impact of a falling price dominates and demand for food goes up. This would
result in an inflow of labour to the agricultural sector and not out.40
But very high is also a problem in this model. The reason is the dependence of food
demand on the aggregate Z production. After the initial productivity shock the system tries to
restore equilibrium by adjusting (proportion of labour working in agriculture). When
falls aggregate Z production goes up. If the impact of this on food demand is too big it may
never reach equilibrium (food demand falls faster than food supply), or it may reach an
equilibrium by increasing .
When , is not so low as to increase food demand with B and neither so high as to
make food demand fall faster than supply. Hence we would observe falling with rising A
or B. But it should also be pointed out that is a sufficient condition. As above, even
for some food demand may be falling slower than supply with decreasing . The
speed of adjustment of these quantities also depends on the initial equilibrium , before the
productivity shock occurs. So if the initial was low enough we may have falling with
productivity even for some .
Calorie Reduction
Note that no calorie reduction will be observed if does not fall with rise in productivity
parameters in this model. If were to rise, aggregate food production would rise even with
no rise in agricultural productivity, implying that some if not all would have to be consuming
more food. So we expect calorie reduction only when falls. But, this is not enough to
40 Also note, due to the way the industrial sector is set up, any rise in B leaves the expenditure on Z ( unchanged. Hence any change in equilibrium labour allocation will have to be due to changes in the food
market induced by changes in price or average Z consumption
127
ensure calorie reduction as it was in the earlier model. Here we have the income effect to
contend with. While a falling and rising B reduces food demand a rise in income pulls it
in the opposite direction. Of course higher is , greater is the negative impact of the falling
. In the end it boils down to whether the income effect would dominate or the KUJ effect.
Any further conclusions about calorie reduction is difficult to make as I do not have a closed
form solution for . But simulations may give us some further insight into the possibility. In
particular I can display the influence of the KUJ parameter by simulating the food demand
through the model for different values of . In the following table I start with the following
initial parameter values: A=1, B=1, , and . For this combination of
parameters, equilibrium values of the key variables of the model are given below for
comparison:
A B Price ( Z)
1 1 0.44 1.0612 1.06 2.95 0.65 1.81
Here represent food demands and represents incomes of the landless and
landowners respectively
Next I raise A to 2, increase B keeping price constant about 1.0612 and not the changes in
incomes and food demand for the two groups. Then I repeated the process for subsequently
higher values of .
128
A B Price(Z)
1.7 1 1 0.44 1.06 1.06 2.95 0.65 1.81
2 3.18 0.18 1.06 3.37 5.74 1.01 1.72
1.9 1 1 0.47 1.02 1.02 2.97 0.66 1.97
2 3.66 0.14 1.02 3.73 5.87 0.93 1.47
2 1 1 0.49 1 1 2.99 0.66 1.99
2 4 .125 1 4 6 0.88 1.33
2.2 1 1 0.99 0.7 0.71 3.53 0.71 3.53
2 8.3 0.06 0.7 5.87 7.23 0.64 0.77
In each case, (characterized by different values of ), I find that incomes of both groups rise
after the rise in productivity parameters. The relative price of Z has been kept constant at the
original level. Food demand for landowners always falls after the process of productivity
increase. For lower food demand for the landless agent are rising, but note that as we go to
higher levels of (stronger KUWJ) this increase gets smaller and smaller till when
we see a fall in food demand for both groups. The simulations give us an indication that even
without quasilinear specification it is possible to observe calorie reduction if the KUWJ
aspect in the preferences are strong enough. The graphs below simulate the model for an
increase in A from 1 to 3, with B increasing correspondingly to hold relative price constant.
This exercise is carried out for 4 different values of , 1.9, 2.0, 2.1 and 2.15.
Table A.1: Simulation Showing the effect of Rising Productivity on Food Demand for Increasing
Values of
Notes: There are four sets of simulations reported for increasing values of as indicated in Column 1. For each
simulation there are two sets of productivity parameters (second set having higher values) across whom relative
price of the industrial good is held constant (fifth column). The values of other parameters are held constant for
these simulations are given on page 100.
129
The Figure A.1 above, displays how total food consumption evolves as we increase
productivity parameters keeping relative price of Z constant. As expected this is increasing
for lower s but eventually turns negatively sloped for higher . Figures A.2 and A.3 are
similar to the first one, except that they show the food consumption of the landless and the
landowners separately. Food consumption of the landless is a concave function which
becomes more and more concave with higher . Eventually for we observe
falling food consumption at higher values of productivity. On the other hand the
landowner’s consumption of food is always falling. The persistent fall in the landlord’s food
consumption may be attributed to the fact that with falling land rental rates are falling and
hence landlord’s total income does not rise as fast as the landless labourer’s income.
131
Proof of Proposition 5:
Proposition 5: In the equilibrium where all are satiated with food, any increase in industrial
or agricultural productivity (represented by the parameters B and A respectively) leads to a
fall in the equilibrium labour allocation to agricultural sector.
Proof: The grain market equilibrium condition (equation (2.10)) in this case is given as
follows (after substituting in the value of p from equation (2.9). The LHS is the combined
food demand for all agents indexed 0 to 1 and the RHS is the food supply:
First note that whenever A or B increases, ceteris paribus, the LHS goes up by less than the
RHS.
The only way to re –equate the two sides of the equation is to decrease (leading to a fall
in the RHS and a rise in the LHS). The other source of variation in the LHS is the term
. But as is shown in the proof of the next proposition below, this term falls when incomes
go up in this model. As a result this would cause the LHS to fall further leading to a further
lowering of . This is the enhancing effect of the Veblen term that was talked about in the
main text.
Proof of Proposition 6:
Proposition 6: In an equilibrium where all are satiated with food, if productivity parameters
both rise keeping the relative price constant, (capital allocation of the agent with index
132
i=0) is large enough to ensure that the poorest person's income is going up, then food
consumption falls for all agents
Proof: The demand for food in this case is given by the following:
Where is the average Z consumption of all agents with income greater than i.
Since p is held constant the only thing driving the food demand is the difference
between own and peer group’s consumption of Z
In this equilibrium all extra income accrued due to the rise in the productivity
parameters goes into consumption of Z
All changes in income from land and labor are common for all hence any change in
the difference between own and peer group’s consumption of Z has to be driven by
changes in capital income. (Note that incomes will rise for all given our assumption
about k0)
In other words, if the difference in capital income of a particular agent and that of his peer
group rises then food consumption will fall.
Assuming that the change in the rental rate of capital due to the change in productivity
parameters is , we can write that food consumption falls for agent j if:
Or