Date post: | 22-Nov-2023 |
Category: |
Documents |
Upload: | independent |
View: | 0 times |
Download: | 0 times |
1
DIFFERENTIAL NUTRITIONAL RESPONSES ACROSS
VARIOUS INCOME SOURCES AMONG EAST AFRICAN PASTORALISTS: INTRAHOUSEHOLD EFFECTS, MISSING
MARKETS AND MENTAL ACCOUNTING
Kira M. Villa, Cornell University Christopher B. Barrett, Cornell University
David R. Just, Cornell University
August 2010
We thank seminar audiences at Cornell University and an anonymous referee for their helpful
comments. These data were collected by the Pastoral Risk Management Project of the Global
Livestock Collaborative Research Support Program which is funded by the Office of Agriculture
and Food Security, Global Bureau, USAID, under grants DAN-1328-G-00-0046-00 and PCE-G-
98-00036-00. Partial funding for write-up was provided by the USAID Assets and Market
Access CRSP. The opinions expressed do not necessarily reflect the views of the U.S. Agency
for International Development.
2
DIFFERENTIAL NUTRITIONAL RESPONSES ACROSS VARIOUS INCOME
SOURCES AMONG EAST AFRICAN PASTORALISTS: INTRAHOUSEHOLD
EFFECTS, MISSING MARKETS AND MENTAL ACCOUNTING
Abstract
In this paper we explore the relationship between dietary diversity and income in pastoralist
households in East Africa. Previous estimates of income elasticities of nutrient demand have
ranged from zero to unity. However, these estimates are always based on total income. One
possible reason for this wide range is that dietary behavior may respond differently to different
sources of income if, for example, agents engage in “mental accounting”, the practice of treating
distinct income sources as not fully fungible. Estimating income elasticities with total income
may mask these differential responses and result in very different income elasticity estimates
depending on which income source changes. Using dietary diversity as a measure of dietary
quality, we find that differential dietary responses do exist across income sources among the
pastoralist households studied. Possible explanations for this result include market failures for
certain commodities, intrahousehold bargaining and mental accounting. These differential
effects persist after accounting for intrahousehold bargaining, market failures and after using
exogenous variations of the different income sources. While we cannot test it explicitly as an
explanation, mental accounting does appear to play some part in explaining the dietary patterns
evident in this sample.
3
I. Introduction
The field of development economics is largely devoted to exploring ways of combating
poverty, along with its many adverse effects. In countries at all levels of development, the food
insecurity and low nutrient intake of the poor have occupied a central place in the study of
poverty. Until recent decades, it was generally thought that the most effective way to combat
hunger and malnutrition was through economic growth and, more specifically, raising the
income of the poor. Conventional wisdom has held that while nutrient intake may not rise one
for one with income, the income elasticity of nutrient demand is still substantially greater than
zero.
In the last few decades however, some studies have challenged this idea arguing that
increases in income will not produce substantial improvements in nutrition (Behrman and
Deolalikar 1990, Behrman and Deolalikar 1989, Behrman and Deolalikar 1987, Behrman et al.
1988, Bouis 1994, Bouis and Haddad 1992). If this claim holds true, then it has significant
implications for how economists and policymakers think about the effects of economic growth
and development on hunger, malnutrition and household food security.
Traditionally it has been thought that the low nutrient intake of the poor is largely due to
low income. Substantial resources have therefore been devoted to income growth programs
aimed at improving nutrition in poor communities. However, despite many studies on the
subject, there is still little agreement over the extent to which nutrient consumption in poor
households responds to changes in their income. Studies examining this matter often look at the
intake changes of specific nutrients, particularly calories, in relation to changes in some measure
of income. The scope of the debate on this relationship ranges from studies arguing that the
calorie-income curve is essentially flat (Behrman and Deolalikar 1990, Behrman and Deolalikar
4
1987, Bouis 1994, Bouis and Haddad 1992, Wolfe and Behrman 1983) to the other extreme
where studies have estimated income elasticities of caloric demand close to one (Pitt 1983,
Strauss 1984). Other studies find a concave or elbow-shaped calorie-income curve (Ravallion
1990, Strauss and Thomas 1995, Strauss and Thomas 1990, Subramanian and Deaton 1996).
These latter findings indicate that among the very poor, nutrient intake would increase with
income up to a certain level, after which, the nutrient-income elasticity would decline possibly to
zero.
A number of reasons have been proposed for the wide range of estimates of the nutrient-
income elasticity for poor households.1
The rest of this paper explores the possibility of differential dietary responses to changes
in various income sources. Section II briefly reviews evidence thus far for the existence of
differential responses and reasons why they might occur, such as intrahousehold dynamics,
However the vast majority of studies on this matter have
not considered the possibility that nutrition may respond differently to different sources of
income. Most of these studies have thus far only explored the nutrition-income relationship
using total income. Yet, there are reasons to believe that where income comes from may change
how it is used in the household. These reasons include intrahousehold dynamics, market
imperfections or missing markets for certain goods, and mental accounting. Thus dietary
patterns may have differential responses to changes in different income sources. Therefore the
impact of income on nutrition might be more appropriately evaluated with income disaggregated
into different sources.
1 Behrman and Deolalikar (1987) argue that an aggregation bias resulting from common methods of inferring nutrient intake cause an upward bias in nutrient-income elasticity estimates. Bouis and Haddad (1992) claim that estimates are often overestimated due to ‘wastages and leakages’ often unobserved by common data collection methods as well as because of correlated measurement errors between explanatory and dependent variables. A number of other studies also point to non-linearities in the relationship between nutrient intake and income that are often unaccounted for in functional forms modeling this relationship (Ravallion 1990, Strauss and Thomas 1995, Strauss and Thomas 1990, Subramanian and Deaton 1996).
5
market imperfections and mental accounting. Section III describes our empirical model and
strategy for examining the relationship between dietary behavior and various income sources.
Section IV describes the data set used for this paper and the setting from which it comes.
Section V describes the econometric specifications used for estimation. Sections VI and VII
discuss estimation results and test for differential responses of dietary diversity to different
income sources. Finally, Section VIII concludes.
II. Differential Dietary Response across Income Sources
There is evidence that changes in certain income sources may impact food intake
differently than changes in other sources. There is a large literature that shows households in the
United States tend to have a larger marginal propensity to consume food out of food stamps than
out of cash income, even when households are unconstrained, 2 implying that food stamp income
has a different impact on household consumption than cash income (Devaney and Fraker 1989,
Fraker et al. 1995, Senauer and Young 1986). This phenomenon is often referred to as the food
stamp cash out puzzle. A number of studies have also found intrahousehold dynamics in resource
control and allocation cause different sources of income to have differential effects on household
consumption and expenditure activities3. Other studies have found differential expenditure and
consumption responses to changes in various income sources that were seemingly due to mental
accounting (Duflo and Udry 2004, Hoffmann 2007, Kooreman 2000, O’Curry 1997). 4
2 A household is unconstrained if it receives food stamps and but also spends a positive amount of cash income on food.
Differential dietary responses across income sources may also result due to the failure or absence
of markets for certain home-produced goods (de Janvry et al. 1991).
3 Some of this literature is reviewed in Behrman (1997), Haddad et al. (1997), Haddad et al. (1996) and Quisumbing (2003). 4 For a good discussion of the mental accounting literature see Thaler (1999).
6
Intrahousehold Bargaining Effects
Intrahousehold dynamics might cause household consumption to respond differently to
changes in different source of income due to differences in resource control and preferences
across various household members. Income sources typically controlled by household members
more concerned with diet and nutrition may have a very different impact on household diet and
food intake than other income sources controlled by members less interested in nutrition.
Indeed, there is a large literature discussing findings that show increases in women’s status,
income and control of resources have a positive effect on child health and expenditures on health
and human capital goods (Behrman 1997, Haddad et al. 1997, Haddad et al. 1996, Quisumbing
2003, Thomas 1990).
Breunig and Dasgupta (2005) examine the food stamp cash out puzzle and conjecture that
this discrepancy is driven primarily by the effects of intrahousehold bargaining. If so, then one
would expect multiple-adult households to exhibit this behavior but not single-adult households.
Studying households in San Diego, Breunig and Dasgupta (2005) find that single-adult
households show no difference in their marginal propensity to consume food out of food stamp
or cash income, while multiple-adult households have an approximately six to eight times higher
marginal propensity to consume food out of food stamp income than cash income. The authors
interpret this economically and statistically significant difference as supporting their
intrahousehold hypothesis.
7
Market Failures and Missing Markets
Household specific market failures, or missing markets in an extreme case, for particular
commodities may also cause varying expenditure responses to different income sources.
Selective market failures for certain home-produced goods may result when household
transactions costs associated with market participation for those goods increase to the point
where those goods are rendered non-tradable for the household in question. This then induces
such households to be autarkic producers and consumers of those particular goods (de Janvry et
al. 1991). Thus increases in household production of those goods would increase consumption
of just those goods, but have little to no impact on other household consumption goods. For
example, if market failures cause a household to be an autarkic producer and consumer of maize,
then marginal increases in maize production would increase household maize consumption but
have no substantial impact on household education expenditures. Thus, household-specific
missing markets or market failures may cause expenditure activities to respond differently to
changes in different sources of income. This explanation may be particularly relevant in some of
the more remote areas of developing countries that have limited access to formal markets.
Mental Accounting
Finally, households may spend various income sources differently due to what behavioral
economists refer to as mental accounting. One component of mental accounting is that income is
not fungible across different sources as standard economic theory would suggest. Instead,
people may assign certain expenditure activities, implicitly or explicitly, to specific ‘mental’
accounts funded by different sources of income. Thus changes in income and wealth in one
8
mental account, such as a windfall, are not perfect substitutes for income changes in another
account, such as wages for labor.
Instances of mental accounting are fairly well documented in consumer behavior in
developed countries and in experimental economics. However, studies explicitly testing for
mental accounting in developing countries are scarce. In a study conducted in Uganda,
Hoffmann (2007) found that households who received insecticide-treated mosquito nets, and
given the opportunity to sell them on site, were more likely to use them for household members
vulnerable to the effects of malaria. Alternatively, if households instead received cash to
purchase the nets on site, the nets were much more likely to be used by main income-earners in
the household.
Duflo and Udry (2004) found evidence of mental accounting in the expenditure patterns
of households in Cote d’Ivoire. They found that yam cultivation, which is typically male-
controlled, had a strong positive association with spending on household public goods and basic
necessities, such as education, food staples and overall food consumption. Meanwhile, they
found that changes in income from male non-yam crops and female cultivated crops were
strongly associated with consumption of adult and prestige goods (tobacco, alcohol, jewelry,
adult clothing and non-staple foods). Furthermore, increases in yam income were associated
with decreases in spending on adult and prestige goods, whereas increases in income from male
non-yam crops resulted in decreases in spending on food.
Factors associated with intrahousehold dynamics, missing markets and mental accounting
could all cause differential dietary responses across various income sources. With this in mind,
it would be prudent to test the appropriateness of nutrition demand models using aggregated
income as an explanatory variable rather than disaggregated income. Instead of looking at
9
dietary responses to changes in total income, it might be more appropriate to explore which
income sources appear to be more important to food consumption and what are the dietary
responses to changes in different income components.
III. Model and Empirical Strategy
In order to investigate the possibility of differential dietary response to different income
sources we employ two approaches. In the first approach we use simple multivariate regression
to compare nutrition-income elasticities estimated when using a model with total income as an
explanatory variable versus one with income disaggregated into different components. We then
test for whether income elasticities estimated in the disaggregated model are equal across income
sources. In a manner similar to Breunig and Dasgupta (2005), once differential dietary responses
are established, we then explore whether market failures or intrahousehold bargaining can solely
account for our results by estimating our model on specific sub-samples of our data.
In our second approach, we implement a reduced form method similar to that used by
Duflo and Udry (2004). In this approach we attempt to control for the potential endogeneity of
income sources by using a set of exogenous shock variables, which affect different income
sources differently, as regressors in a model estimating dietary behavior. We then test whether
dietary responses are equal across the exogenous variations of different income sources.
Multivariate Approach
Most prior studies the development literature examining the nutrition-income relationship
have assumed equivalent income elasticities across various income sources. A common
functional form in this literature follows the log-linear equation:
10
(1) ivtiv
C
c
cvt
cF
f
fivt
fJ
j
jvt
jivtivt VHPYN εµθγδβα ++++++= ∑∑∑
=== 111lnlnln
where
i is an index for the individual,
v indexes the village or location,
t indexes the time period,
N is some indicator of nutrition or dietary quality,
Y is income,
P is the price of food commodity j,
H is household specific characteristic f,
V is village or location specific characteristic c,
µ is unobserved individual-specific effect and
ε is the disturbance term.
However, if household diet responds differently to different income sources, the estimated
nutrition elasticities with respect to total income may be misleading.
To illustrate, say a household earns income from two different sources, Y1 and Y2.
Suppose the household uses income Y2 primarily for food purchases and income Y1 mostly for
other expenditure activities and rarely for food expenditures. Suppose further that the nutrition
elasticity with respect to Y2 is positive while that with respect to Y1 is zero. Say this household
experiences a large increase in Y1 and a very small increase in Y2. Consequently the household
experiences a large increase in total income but a very small increase in food expenditures,
resulting in an estimated nutrition elasticity with respect to total income that is close to zero.
This result masks the positive nutrition elasticity with respect to Y2. On the other hand, if the
large increase in total income was primarily due to an increase in Y2, then a larger positive total
11
income elasticity of nutrition would be estimated masking the zero nutrition elasticity with
respect to Y1. Either case could lead to mistargeted income growth programs concerned with
nutrition.
The nutrition literature has estimated nutrition-income elasticities ranging from near zero
(Behrman and Deolalikar 1990, Behrman and Deolalikar 1987, Bouis 1994, Bouis and Haddad
1992, Wolfe and Behrman 1983) to almost one (Pitt 1983, Strauss 1984). One possible
explanation for this wide range of estimates is that many of these studies are not accounting for
the possibility of differential dietary responses to changes in various income sources.
To explicitly test the assumption of equivalent nutrition elasticities across income sources
we disaggregate income by source in (1) to get the following:
(2) itiv
C
c
cvt
cF
f
fivt
fJ
j
jvt
jK
k
kivt
kivt VHPYN εµθγδβα ++++++= ∑∑∑∑
==== 1111lnlnln
where k indexes income sources and
(3) kivtivtivtivtivt YYYYY ++++= ...321 .
Using (2) we can test the null hypothesis
(4) H0: β1 = β2 = β3 = ... = βκ
ΗΑ: nm ββ ≠ for some { }Knm ,...,2,1, ∈ and nm ≠
A rejection of the null hypothesis would indicate that there exist differential nutritional responses
to different income sources and thus (1) is not an appropriate model for estimating nutrition
elasticities with respect to income.
However, this model is only adequate if the composition of income is similar among
households at different levels of wealth. There is substantial evidence that nutrition-income
elasticities vary at different levels of wealth (Sahn (1988), Strauss and Thomas 1990,
12
Subramanian and Deaton 1996). Therefore if there are systematic differences in income
composition between poor and rich households, which are not controlled for explicitly, then a
rejection of the above null hypothesis may just be picking up differences in nutrition-income
elasticities at various levels of wealth as opposed to differences due to income source. Therefore
(2) must be further modified to allow for non-linearities in the relationship between nutrition-
income elasticities and income level.
To allow income elasticities to differ over different levels of income, dummy variables
indicating the income quantile to which the household belongs are included in the model as both
intercept shifters as well as interacted with income and price variables to allow for income and
price elasticities to change with the level of income.5
(5)
This gives us the following equation:
itiv
C
c
cvt
cF
f
fivt
fJ
j
jvtl
jl
K
k
kivtl
kll
L
llivt VHPYQN εµθγδβα ++++++= ∑∑∑∑∑
===== 11111]lnln[ln
where Q is an indicator variable equal to one if the individual i belongs to income quantile l and l
= 1, 2, 3,…L. This gives the following testable hypothesis:
(6) H0: Kllll ββββ ==== ...321 for all { }Ll ,...,2,1∈
ΗΑ: nl
ml ββ ≠ for some { }Knm ,...,2,1, ∈ and nm ≠ and for some { }Ll ,...,2,1∈
Equation (5) allows for income and price elasticities to vary across different levels of
income and also controls for the possibility of differential dietary responses due to wealth
differentials as opposed to income source differentials. A rejection of the null hypotheses in
hypothesis (6) indicates that dietary responses are not equivalent across changes in different
income sources within a particular income quantile. This would indicate that a model using
5 As another way of capturing these effects, Strauss and Thomas (1990) proposed a few functional forms of log-inverse log models that seemed to work well. However these models are problematic if there are a number of observations where income is between 0 and 1. Since in this paper, income is disaggregated there are many zero observations for each income source variable. This proved to be problematic when working with the log-inverse functional forms proposed by Strauss and Thomas (1990).
13
aggregated income as an explanatory variable is less appropriate than one in which income is
disaggregated.
Reduced Form Approach
In our second approach we implement a reduced form method similar to that employed
by Duflo and Udry (2004). More precisely, we use a set of exogenous shock regressors, which
affect different income sources differently, to predict each income source separately. We then
use these sources of exogenous variation to test for differential responses of dietary diversity to
different income sources.
We assume the following function for the dietary behavior of person i in location v at
time period t:
(7) ∑∑==
+++++=B
bivti
bt
bV
v
vvivtivt SVYN
13
1210 lnln µµαααα
where N is still some indicator of nutrition or dietary quality, V is a regional dummy variable, Y
is household income and S is seasonal dummy variable.
We predict the logarithm of income from source k as a function of R exogenous shock
regressors using the following relationship:
(8) ∑ ∑∑∑∑= ====
++++++=V
vivti
B
b
bbvrrvR
r
V
v
vvR
r
rrkivt SVZVZy
1 143
112
110ln ννβββββ
Using the predicted level of each income source ( kivtyln ) estimated from (8), we use the
following relationship to predict the total income of person i in location v at time period t:
(9)
ivti
L
l
li
l
B
b
bbV
v
vvK
k
kivt
li
lkL
l
K
k
kivt
kivt
Q
SVyQyY
ωωθ
θθθθθ
+++
++++=
∑
∑∑∑∑∑
=
=====
15
14
13
12
1110 ˆlnˆlnln
14
where liQ is a dummy variable indicating whether person i is in income tercile l and
{ }Ll ,...,2,1∈ . Combining (7) and (9) we get the following relationship:
(10)
ivti
L
l
li
l
B
b
bbV
v
vvL
l
kivt
li
klK
k
K
k
kivt
kivt
Q
SVyQyN
µµδ
δδδδδ
+++
++++=
∑
∑∑∑∑∑
=
=====
15
14
13
12
1110 ˆlnˆlnln
In order to use (10) to test for differential effects of various income sources on dietary behavior
the following overidentifying restriction must hold:
(11) nivt
ivt
nivt
ivt
mivt
ivt
mivt
ivt
yY
yN
yY
yN
ˆlnln
ˆlnln
ˆlnln
ˆlnln
∂∂
∂∂
=
∂∂
∂∂
{ }Knm ,...,2,1, ∈∀ .
This restriction says that the exogenous regressors used to predict kivtyln only effect
ivtNln through their effect on each income source but and do not affect ivtNln directly. In other
words, (11) says that none of the exogenous shock regressors on the right-hand side of (8) also
belong on the right-hand side of (7).
If (11) is satisfied, we can then test for differential dietary responses across income
sources by testing the following hypothesis:
(12) Ho: lKKll21
22
21
12
11 ... δδδδδδ +==+=+ { }Ll ,...,2,1∈∀
HA: ln2121 δδδδ +≠+ nlmm for some { }Ll ,...,2,1∈ and for some { }Knm ,...,2,1, ∈ nm ≠
Using this method we can look at how dietary behavior responds to particular exogenous shock
variables that are related to different sources of household income. We can then test whether
dietary responses are equivalent across exogenous variations of different income sources.
15
IV. Data and Setting
The data for this paper come from a comprehensive set of panel data collected by the
USAID Global Livestock Collaborative Research Support Program (GL CRSP) project
“Improving Pastoral Risk Management on East African Rangelands” (PARIMA). Households
were surveyed in five locations in southern Ethiopia and six in northern Kenya6, all in one
livestock production and marketing region (Barrett et al. 2008). In total, 337 households are
included in the data.7 In each household the household head was surveyed along with up to two
adult, non-head household members. Because investigating intrahousehold dietary patterns is
beyond the scope of this paper, only household heads are included in this study. Surveys were
conducted in March 2000 for baseline information and then quarterly from June 2000 to June
2002, resulting in 10 quarterly observations for each household.8
The baseline survey gives information on individual and household characteristics such
as household size, sex, age and education. The repeated surveys provide information on income
earned from various sources such as trade, wages and salary, crop value and remittances. They
also report information on households’ livestock holdings, trade and production over the last
quarter.
Survey intervals were chosen
to correspond to the bimodal rainfall patterns of the study region. Further details on these data
are provided in Barrett et al. (2008).
Households in the area are typically male-headed. Female household heads are generally
poorer and are often widowed or divorced. 31% of the household heads in the sample are
6 The six study locations in Kenya were Dirib Gumbo, Kargi, Logologo, Ng’ambo, North Horr, and Sugata Marmar. In Ethiopia the study sites were Dida Hara, Dillo, Finchawa, Qorate, and Wachille. 7 Due to some issues of attrition, interruption and missing observations of particular variables for certain individuals or communities the number of observations per survey period ranges from 186 to 303. 8 The baseline survey in March 2000 did not provide dietary information or information on income over the quarter. Therefore this study included only 9 quarterly observations from June 2000 to June 2002 and information from the March 2000 survey was only used for baseline information on the household.
16
female. Of those, only eight percent of the female heads are married and 49% of them are in the
bottom income tercile. Table 1 provides some descriptive statistics on the study population.
While a number of households are involved in activities such as trade, wage labor, or, to
a very limited extend, crop cultivation, primary economic activities for most households in the
area are centered on livestock. Pastoralism allows households to be opportunistic in the arid and
semi-arid lands of the study region where uncertain rainfall makes primary production risky
(Coppock, 1994). Only five households in the data do not own livestock over the study period.
Mean annual rainfall in the study area is just around 400mm, making crop cultivation difficult.
Therefore, pastoralist households rely chiefly on livestock for income. Average household herd
size in the data is 13.31 tropical livestock units (TLU).9
In order to estimate equations (5), (9) and (10) the sample is broken up into three income
terciles based on households’ mean intertemporal income. Table 2 describes income and its
composition for the lower, middle, and upper terciles. On the whole, there are not substantial
differences in the composition of household income across income terciles. The most drastic
difference across income terciles is in the share of remittances in total household income. Net
remittances make, on average, 40% of total household income in the lower tercile but only 24%
and 18% in the middle and upper terciles, respectively.
Production of livestock products makes
up on average roughly 43% of total household income earned in the study population over the
study period. Livestock trade is 13%, wages and salary is 8%, net remittances is 28%, non-farm
non-livestock trade and business make up 6%, and crop value comprises only 3% of total
household income earned in the study area.
9 TLU is a standard measure for aggregating herd size across various species. 1 cattle = 1 TLU, 1 camel = 0.7 TLU, 1 sheep=0.1 and 1 goat = 0.11 TLU.
17
The repeated surveys also ask individuals to recall their own food and beverage
consumption of the past 24 hours. This includes food and beverages consumed both in and away
from the home. This information was used to calculate the dietary diversity of each individual in
each survey period. Dietary diversity is defined here as the number of unique food and drink
items consumed over the 24-hour recall period. Mean (median) dietary diversity in the sample
population is 3.24 (3.00). Maize, tea and especially milk are by far the most consumed items in
the study area.
V. Econometric Specification
In order to test for differential dietary responses to changes in different income sources,
we use dietary diversity as a measure of individual-level dietary quality. Dietary diversity has
been proposed and widely used as an alternative indicator of dietary quality and food security
(Arimond and Ruel, 2004, Hatloy, et al., 1998, Hoddinott and Yohannes, 2002, Ogle, et al.,
2001, Onyango, et al., 1998, Ruel, 2002, Ruel, 2003, Torheim, et al., 2004). A diverse diet has
been long associated with enhanced nutritional well-being. Indeed, a number of studies have
shown dietary diversity to be highly correlated with dietary quality and nutrient adequacy
(Arimond and Ruel, 2004, Hatloy, et al., 1998, Hoddinott and Yohannes, 2002, Ogle, et al.,
2001, Onyango, et al., 1998, Torheim, et al., 2004). Studies have also shown a consistent and
positive association between child growth and dietary diversity (Arimond and Ruel, 2004,
Onyango, et al., 1998). Dietary diversity is also unlikely to suffer from some of the same
measurement errors and bias problems that prove problematic in many more conventional
nutritional indicators, such as food expenditures or estimated nutrient intake volumes (Hoddinott
18
and Yohannes, 2002). 10
Dietary diversity is usually defined as either the number of unique food and drink items
consumed or the number of unique food groups consumed over a recall period. We define
dietary diversity here as the number of unique food and drink items consumed over the 24 hour
recall period. For example, if an individual consumes three helpings of maize, one helping of
beans and two servings of milk, his dietary diversity count would be three.
Consequently dietary diversity is used as the dependent variable in
equations (5) and (10).
While many previous studies have used nutrient intake or food expenditures to indicate
nutrition or dietary quality, such information is unfortunately unavailable in these data. Detailed
nutrient intake data were not taken in these surveys. While the PARIMA surveys do not have
good individual nutrient intake or expenditure information, they do provide good data on
individual-level dietary diversity.
To estimate equations (5), (8), (9) and (10), we disaggregate income into six different
sources: income earned from non-farm and non-livestock trade and business such as from crafts,
firewood and water; income earned from wages and salary; income earned from livestock trade;
the value of livestock products produced; the value of crops harvested; and net remittances,
which include the value of cash and in-kind gifts as well as food aid. 11
Control variables for equation (5) are as follows. village level food prices included in the
model are those for maize, tea and milk, which are by far the most important food staples in the
Although there may be
some exceptions, generally, income earned from the production of livestock products and non-
farm, non-livestock trade and business is typically controlled by women. Conversely, men
generally control income earned from livestock trade and wages and salary.
10 See Strauss and Thomas (1995) and Hoddinott and Yohannes (2002) for a discussion of these problems. 11 For further details on how each income source was constructed see Barrett et al. (2008).
19
study region. Age, education, gender and household size are included as controls for individual-
and household-level characteristics. We include four seasonal dummy variables for the long-
rain, long-dry, short-rain and short-dry seasons. As location-specific controls we also included
village-level average quarterly rainfall and regional dummy variables for 10 of the 11 study
locations.
To estimate equation (8), we use six exogenous variables: average rainfall in the current
quarter; average rainfall over the previous quarter; whether or not a household member
experienced an illness in the last quarter that prevented him or her from working; and
standardized Normalized Differential Vegetation Index (NDVI) cumulative over the current
quarter, over the last two quarters, and over the last 3 quarters.12
NDVI indicates the amount of vegetative cover and thus forage availability based on
photosynthetic activity observed from satellites (Chantarat, 2009). NDVI has been shown to be a
key driver in herd dynamics and can be used to compare current vegetative conditions with
previous periods to detect abnormal conditions (Chantarat, 2009).
We standardize dekadal (taken every ten days) NDVI for each study location with NDVI
data taken every ten days from 1981-2008. Thus we get dekadal standardized NDVI (zndvivdn)
for location v in dekad d in year n such that:
( )( )vdnd
vdndvdnvdn NDVI
NDVIENDVIzndvi
σ−
=
where ( )vdnd NDVIE and ( )vdnd NDVIσ are the mean and standard deviation of NDVI values for
location v in dekad d taken over the period 1981-2008. We then generate three cumulative zndvi
variables for each quarterly observation such that dekadal zndvi measures are cumulative over
12 Since all the exogenous variables except illness were measured at the village-level, illness is the only exogenous variable that was interacted with location dummies in our specification of equation (8).
20
the previous quarter, the previous two quarters, and the previous three quarters. A positive
cumulative zndvi measure indicates a better than normal period while a negative cumulative
zndvi measure indicates a worse than normal period. These three cumulative zndvi measures are
likely to affect income sources differently. For example herd mortality will likely respond
slower to bad vegetative conditions than say, milk production. Since in this area men tend to
earn the income received from livestock trade while women earn the proceeds from milk sales,
these exogenous shocks can effectively identify different income sources, including those that
reflect gender differences in livelihood patterns.
VI. Results from Multivariate Approach
For the purposes of this paper we will focus just on estimated income elasticities. Full
regression results are presented in the Appendix as Tables A1 and A2. Table 3 reports income
elasticities estimated from equations (1) and (5) which use aggregated and disaggregated income,
respectively, as explanatory variables. All estimated income elasticities are rather small in
magnitude. Income elasticity estimates using the aggregated income model are 0.04 for the
lower and middle tercile and 0.03 for the upper tercile. All are statistically significant at the one
percent level.
Turning now to estimates from the disaggregated income model, although small in
magnitude, the estimated source-specific income elasticities appear statistically different from
each other and from those estimated for aggregate income. Income from wages and salary and
net remittances appear to have an impact on dietary diversity for all income terciles with
estimated income elasticities all significant at the one or five percent levels. The income
elasticities specific to wages and salary estimated for the lower tercile is 0.04 and is 0.02 for the
21
middle and upper tercile. Estimated income elasticities for remittances are 0.03, 0.04 and 0.02
for the lower, middle and upper tercile, respectively.
Income from trade and business has a statistically significant impact on dietary diversity
for only the lower and upper tercile with estimated income elasticities of 0.06 and 0.02,
respectively, both significant at the one percent level. Income due to crops has a statistically
significant impact on dietary diversity only for the middle tercile with an estimated income
elasticity of 0.02. Income specific to livestock trade only has a statistically significant impact on
the dietary diversity of the lower tercile with an estimated income elasticity of 0.02. Income
specific to livestock products does not have a statistically significant impact on dietary diversity
in any of the income terciles.
Thus there do seem to be differences in the relative magnitudes and significance of
estimated income elasticities specific to the various income sources. However, we must
statistically test for differences in dietary diversity response across income sources, particularly
since the estimated income elasticities are very low. A Wald test on hypothesis (6) confirms this
impression and rejects the equality of income elasticities across sources and income terciles at
the one percent level with a test statistic of ( ) 45.44152 =χ .
Since income from livestock trade and wages and salary is typically associated with men
and income specific to livestock products and non-livestock trade and business is typically
associated with women, we can roughly examine whether the differences in elasticity estimates
is due to differences in the income sources themselves or due to the gender-specific preferences
of those controlling the different sources of income. A Wald test rejects the equality across
terciles of estimated income elasticities specific to the generally female controlled income
sources of livestock products and trade and business at the one percent level with a test statistic
22
of ( ) 39.1332 =χ . The test also rejects at the one percent level across all income terciles the
equality of estimated income elasticities specific to the typically male controlled income sources
of livestock trade and wages and salary with a test statistic of ( ) 87.1132 =χ . This is not a
perfect test since the gender associations of each income source are not perfect. However, it
does provide evidence that cultural differences in men’s and women’s control over different
income sources is probably not the sole driver behind the differential responses of dietary
diversity across income sources. We will test this more formally in the next sub-section.
The next step then is to investigate more fully some possible explanations behind the
differential impacts of different income sources on dietary diversity. As noted earlier, three
possible explanations for this result are discussed in the literature: intrahousehold dynamics of
resource control and allocation; missing markets for certain home produced commodities; and
mental accounting. There is no way to explicitly test for mental accounting as the cause of
differential responses. However, missing markets and intrahousehold effects can be directly
tested as explanations. If neither can fully account for the differential responses of dietary
diversity to various income sources, then this is evidence in favor of mental accounting playing
some role in our result.
Intrahousehold Bargaining as Possible Explanation for Differential Responses
Many pastoralist households in the study region practice polygamy. In addition,
households often include extended family members such as parents, siblings of the household
head or his spouse(s), and adult children. Therefore intrahousehold processes of bargaining and
resource control could provide a reasonable explanation for the differential responses of dietary
diversity across income sources. In such households, preferences surely vary among household
23
members. So if different income sources were typically associated with different household
members, it would result in the differential dietary diversity responses we find.
In order to test for intrahousehold effects as the cause for differential responses, we adopt
a method similar to that implemented by Breunig and Dasgupta (2005). In their study of the
food stamp cash-out puzzle Breunig and Dasgupta (2005) restrict their analysis to single-adult
households in order to test whether the effects of intrahousehold bargaining cause cash transfers
to impact household consumption differently than food stamp transfers. Since the differential
impact of the two income sources on consumption was established only for multiple-adult
households and not single-adult households, they conclude that intrahousehold bargaining is
behind this result.
We likewise restrict our data sample to households where the household head is
unmarried. Since many households also house extended family members, the sample was further
restricted to households in which the household head is single and the oldest non-head household
member was no older than 10 years less than the age of the head. In order to account for gender
differences in preferences and resource control we also include only female-headed households
in the single adult sub-sample.13 By restricting the sub-sample to only households with one
adult, this necessarily excludes any households that are affected by processes of intrahousehold
bargaining, since only one member in the household has any significant bargaining power.14
Summary statistics on this sub-sample are provided in Table 4. Restricting the sample
resulted in a much smaller data set. This single-adult sub-sample has only 567 of the original
2495 observations in the full sample. Therefore to conserve on degrees of freedom we only
13 By excluding male-headed households in the single-adult sub-sample we only exclude two households. Including these households has no effect on our results but we nonetheless leave them out for clarity. 14 This of course assumes that children have no substantial or systematic bargaining power over resource allocation within the household. While we cannot test this assumption, we feel safe in making it.
24
include controls for region, season and income tercile.15
Table 5 reports income elasticities estimated from equations (1) and (5) using the single-
adult sub-sample. Using the aggregated income model, the estimated total income elasticity of
dietary diversity is only statistically significant for the lower tercile with a point estimate of 0.04.
Looking at estimated income elasticities from the disaggregated income model, source-specific
income elasticities again appear to differ, statistically, across income sources.
The female household heads in the
study region are often widowed or divorced and are typically poor. Approximately half of the
households in this sub-sample are in the lower income tercile.
The pattern of results observed from these estimates is largely similar to those observed
using the full sample with a few minor differences. Again there appears to be a difference in the
relative magnitudes and significance of estimated income elasticities across income sources. In
fact the estimated income elasticity specific to income from livestock trade in the upper tercile is
statistically significant and negative. If mental accounting does play a role in the behavior of
these households, then we might see a negative estimated income elasticity such as this if
generating income from livestock trade is not only unimportant to dietary diversity but also
substitutes resources away from earning income from sources more important to dietary
diversity.
A Wald test again rejects that estimated income elasticities are equivalent across income
sources at the one percent level with a test statistic of ( ) 99.92152 =χ . Thus even after
controlling for the possibility of intrahousehold bargaining, we still observe differential dietary
diversity responses to different income sources. Since all household heads, and thus all adult
15 Including these controls does not significantly change our results or conclusions.
25
household members, in this sample are female, we can conclude that neither gender differences
nor intrahousehold bargaining is solely driving these results.
Missing Markets as Possible Explanation for Differential Responses
Almost all households in the study area own livestock and thus produce milk. However,
income earned from livestock products, a large part of which is milk production, appears to be
less important to dietary diversity than other income sources. The estimated income elasticities
specific to livestock products are not statistically significant in any of the income terciles using
the full sample or the single-adult sub-sample. Given the prominence of livestock products as an
income source in this population (livestock products make up on average 36%, 49% and 45% of
household income in the lower, middle and upper terciles, respectively) one might reasonably
expect it to play a larger role in influencing dietary diversity.
A possible reason behind this result is that many pastoralist households might not
participate in formal markets for livestock products, rendering those goods non-tradable. If the
transaction costs associated with market-based exchange induce the household to be an autarkic
milk producer and consumer, any increased income from production of that non-tradable good
necessarily expands only the quantity of milk that household consumes, not the variety of food it
consumes. Therefore production of milk and other livestock products may do little to enhance
dietary diversity.
In order to test the possibility that household-specific non-tradable home-produced goods
causes the differential dietary diversity responses observed in the full sample, we use a similar
approach as we did in the previous section to control for intrahousehold effects. We restrict the
data sample to observations where households recorded positive milk sales, as opposed to just
26
positive milk production. Those who sell milk necessarily treat milk as tradable. By focusing
only on these observations, we exclude any household-period observations for which milk might
have been non-tradable. Further, positive milk sales indicate that households participated in
more formal market settings in settlements where they were not only able to participate in the
market for milk but would also have the opportunity to participate in markets for other goods as
well.
As with the single adult sub-sample, restricting our sample resulted in a much smaller
data subset. The milk market sub-sample has only 370 of the full sample’s 2495 observations.
Therefore, again, to conserve degrees of freedom we include extra controls only for region,
season and income tercile when we estimate equations (1) and (5) using this sub-sample.
Descriptive statistics on the milk market sub-sample can be found in Table 6. Income
elasticities estimated from the aggregated income and disaggregated income models using this
sub-sample can be found in Table 7. Here we do see more of a difference in the general pattern
of results from that when using the full sample. This does lead one to believe that missing
markets play a role in some of the differential effects we observe. While we do see a difference
in the general pattern of results, estimated source-specific income elasticities still differ
statistically from each other and from those estimated with respect to aggregate income.
Additionally, even though livestock products is a very important source of income in the sub-
sample, the estimated income elasticities specific to livestock products are still not statistically
significant in any of the income terciles. Thus even though the sample has been restricted such
that milk is necessarily treated as tradable, livestock product income still has little to no effect on
dietary diversity.
27
A Wald test again rejects equivalent income elasticities across income sources and
terciles at the one percent level with a test statistic of ( ) 77.74152 =χ . Again, in order to examine
if these differential effects are driven by differences in male and female control of different
income sources, we test whether estimated income elasticities specific to trade and business and
livestock products are statistically equal and whether those specific to wages and salary and
livestock trade are statistically equal. A Wald test fails to reject that income elasticities
estimated for income from trade and business and livestock products are equal. Thus, in this
sub-sample, the two income sources mostly associated with women do not appear to be treated
differently with regards to dietary diversity (both of these estimated income elasticities are not
statistically different from zero in any tercile). A Wald test, however, rejects at the ten percent
level the equality of estimated income elasticities specific to income from wages and salary and
livestock trade. Thus these two typically male-associated income sources still appear to have
different impacts on dietary diversity.
While missing markets may play some part in explaining the differential dietary diversity
responses we observe in the full sample, they do not appear to be able to solely explain these
results. Further, when looking at the effects of more gender-specific income sources while
controlling for the effects of missing markets, there still appears to be a difference in how dietary
diversity responds to the more male-controlled income sources of livestock trade and wages and
salary. While this test is imperfect, it can lead one to believe that there is a reason beyond (and
possibly in addition to) gender differences and missing markets that contributes the existence of
differential responses of dietary diversity across income sources. While we cannot test for it, we
believe that mental accounting is a sensible candidate explanation.
28
VII. Results from Reduced Form Approach
While we attempt to control for particular household characteristics (intrahousehold
bargaining and household-specific missing markets) that might be driving the differential dietary
diversity responses we observe, one still might be concerned about the possibility that income
sources are endogenous to certain household characteristics that are not accounted for but could
nonetheless be driving our results. We therefore also estimated dietary diversity demand as a
reduced form using exogenous variables affecting various income sources as regressors.
Full regression results from (8) are reported in the Appendix in Table A3. In each of the
seven regressions the exogenous regressors are jointly statistically significant. The estimated
coefficients between regressions are also significantly different from each other in each pairwise
combination. Thus the various income sources are each affected differently by different
realizations of the exogenous variables. We can thus test whether income sources are treated
similarly with regards to dietary diversity by examining whether the exogenous variables
affecting the different income sources have differing effects on dietary diversity above and
beyond their effect on total income.
Table 8 reports coefficient estimates on predicted income sources from equations (9) and
(10). Equations (9) and (10) estimate the predicted source-specific income elasticities of total
income and dietary diversity, respectively, where each income source is predicted by the
exogenous variables. A non-linear Wald test fails to reject the overidentification restriction
(hypothesis (11)) with a test statistic ( ) 46.10152 =χ . Thus the coefficients in the dietary
diversity regression are all of statistically equal proportion to the corresponding coefficients in
the total income regression, meaning that the overidentification restriction is satisfied.
29
Looking at the dietary diversity regression, the estimated source-specific income
elasticities again appear to differ from each other. Predicted income from livestock trade and
crop value appear to have a relatively stronger positive effect on dietary diversity while income
from wages and salary appears to have a negative effect on dietary diversity. Predicted income
from livestock products and remittances has no statistically significant effect on dietary diversity.
Predicted income from trade and business has a statistically significant effect only on the dietary
diversity of households in the lower tercile. A Wald test rejects the equality of estimated income
elasticities across all income sources and terciles at the one percent level.
Examining whether there still exist differential effects within income sources specific to
each gender, we again test the equivalence of estimated income elasticities specific to just
female- and just-male associated income sources. A Wald test fails to reject that the female
associated income elasticities specific to trade and business and livestock trade are equal. Thus
there appears to be no difference in the impact of these two typically female associated income
sources on dietary diversity. A Wald test however rejects the equivalence of the two typically
male-associated income elasticities specific to livestock trade and wages and salary at the one
percent level. Therefore the two generally male-associated income sources appear to affect
dietary diversity differently and thus the differential responses we observe cannot be entirely
explained by gender differences in resource control.
VIII. Conclusion
We find evidence of differential dietary diversity responses to various income sources.
These differential impacts across income sources persist after controlling separately for the
effects of intrahousehold bargaining and household-specific missing markets. While each of
30
these explanations may play a role in causing the differential effects we observe, neither can
solely account for our results. We also observe different dietary diversity responses across
income sources when looking at exogenous variations of each income source in a reduced form.
Further, in all of our specifications, save that using a single-gender, single-adult sub-sample, we
reject that two typically male-associated income sources, livestock trade and wages and salary,
equally impact dietary diversity. Thus neither can gender differences fully account for our
results. We therefore conclude that mental accounting may also play a significant role in the
behavior of the households we study.
For the most part, research on the nutrition-income relationship in developing countries
has investigated the dietary impacts of changes in total household income. However, where
income comes from may change how it is used in the household. Consequently, it may be more
accurate to examine the impact of different sources of income on dietary behavior rather than
merely aggregate income. If income source matters to how households respond to changes in
income, then this has important implications for how the relationship between income and diet is
assessed. Income generating programs concerned with the food intake of poor households may
be ineffective, or even counter productive, if they are not targeting appropriate income sources.
Recognizing and better understanding the consumption behavior of poor households as it relates
to various income sources could substantially improve policy targeting and development efforts
on the whole. Treating all income equally may lead to inadequate assessments of income-
consumption relationships.
31
REFERENCES
Arimond, M., and M. T. Ruel. "Dietary Diversity Is Associated with Child Nutritional Status:
Evidence from 11 Demographic and Health Surveys." Journal of Nutrition. 134, no.
10(2004): 2579-2585.
Barrett, C. B., Getachew, G., McPeak, J., Mude, A., Vanderpuye-Orgle, J., and Yirbecho, A.
(2008) Codebook for Data Collected Under the Improving Pastoral Risk Management on
East African Rangelands (PARIMA) Project.
Behrman, J. R. (1997) Intrahousehold Distribution and the Family, in Mark R. Rosenzweig and
Oded Stark, eds. Handbook of Population and Family Economics, Amsterdam Elsevier,
pp125-187
Behrman, J. R., and A. B. Deolalikar. "The Intrahousehold Demand for Nutrients in Rural South
India." Journal of Human Resources 25, no. 4(1990): 665-696.
Behrman, J. R., and A. B. Deolalikar. "Is Variety the Spice of Life? Implications for Calorie
Intake." Review of Economics & Statistics 71, no. 4(1989): 666.
Behrman, J. R., and A. B. Deolalikar. "Will Developing Country Nutrition Improve with
Income? A Case Study for Rural South India." Journal of Political Economy 95, no.
3(1987): 492.
Behrman, J. R., A. B. Deolalikar, and B. L. Wolfe. "Nutrients: Impacts and Determinants." The
World Bank Economic Review 2, no. 3(1988): 299.
Bouis, H. E. "The effect of income on demand for food in poor countries: Are our food
consumption databases giving us reliable estimates?" Journal of Development Economics
44, no. 1(1994): 199-226.
Bouis, H. E., and L. J. Haddad. "Are estimates of calorie-income elasticities too high?: A
recalibration of the plausible range." Journal of Development Economics 39, no. 2(1992):
333-364.
32
Breunig, R., and I. Dasgupta. "Do Intra-Household Effects Generate the Food Stamp Cash-Out
Puzzle?" American Journal of Agricultural Economics 87, no. 3(2005): 552-568.
Chantarat, S. "Pro-Poor Risk Management: Essays on the Economics of Index-Based Risk
Transfer Products." Dissertation, Cornell University, 2009.
Coppock, D. L. The Borana Plateau of Southern Ethiopia: Synthesis of Pastoral Research,
Development, and Change, 1980-91: International Livestock Centre for Africa, 1994.
de Janvry, A., M. Fafchamps, and E. Sadoulet. "Peasant Household Behavior with Missing
Markets: Some Paradoxes Explained." The Economic Journal 101, no. 409(1991): 1400-
1417.
Devaney, B., and T. Fraker. "The Effect of Food Stamps on Food Expenditures: An Assessment
of Findings from the Nationwide Food Consumption Survey." American Journal of
Agricultural Economics 71, no. 1(1989): 99-104.
Duflo, E., and C. Udry (2004) Intrahousehold Resource Allocation in Cote d'Ivoire: Social
Norms, Separate Accounts and Consumption Choices, NBER.
Fraker, T. M., A. P. Martini, and J. C. Ohls. "The Effect of Food Stamp Cashout on Food
Expenditures: An Assessment of the Findings from Four Demonstrations." The Journal
of Human Resources 30, no. 4(1995): 633-649.
Haddad, L., J. Hoddinott, and H. Alderman. Intrahousehold Resource Allocation in Developing
Countries: Models, Methods, and Policy. Baltimore, MD: Johns Hopkins University
Press for the International Food Policy Research Institute, 1997.
Haddad, L., C. Pena, C. Nishida, A. Quisumbing, and A. Slack. "Food Security and Nutrition
Implications of Intrahousehold Bias: A Review of Literature." International Food Policy
Research Institute FCND Discussion Paper, no. 19(1996).
Hatloy, A., L. E. Torheim, and A. Oshaug. "Food variety--a good indicator of nutritional
adequacy of the diet? A case study from an urban area in Mali, West Africa." European
Journal of Clinical Nutrition 52, no. 12(1998): 891.
33
Hoddinott, J., and Y. Yohannes. "Dietary diversity as a food security indicator." Food
Consumption and Nutrition Division Discussion Paper 136(2002).
Hoffmann, V. (2007) Mental accounts, gender, or both? The intra-household allocation of free
and purchased mosquito nets. Ithaca, NY, Cornell University.
Kooreman, P. "The Labeling Effect of a Child Benefit System." American Economic Review 90,
no. 3(2000): 571-583.
O'Curry, S. "Income Source Effects." Unpublished working paper, DePaul University (1997).
Ogle, B. M., P. H. Hung, and H. T. Tuyet. "Significance of wild vegetables in micronutrient
intakes of women in Vietnam: an analysis of food variety." Asia Pacific Journal of
Clinical Nutrition 10, no. 1(2001): 21-30.
Onyango, A., K. G. Koski, and K. L. Tucker. "Food diversity versus breastfeeding choice in
determining anthropometric status in rural Kenyan toddlers." International Journal of
Epidemiology 27(1998): 484-489.
Pitt, M. M. "Food Preferences and Nutrition in Rural Bangladesh." Review of Economics &
Statistics 65, no. 1(1983): 105.
Quisumbing, A. R. Household Decisions, Gender, and Development: A Synthesis of Research.
Washington, D.C.: International Food Policy Research Institute, 2003.
Ravallion, M. "Income Effects on Undernutrition." Economic Development and Cultural Change
38, no. 3(1990): 489-515.
Ruel, M. T. "Is dietary diversity an indicator of food security or dietary quality? A review of
measurement issues and research needs." Discussion Paper-Food Consumption and
Nutrition Division, International Food Policy Research Institute (2002): 58pp
Ruel, M. T. "Operationalizing Dietary Diversity: A Review of Measurement Issues and Research
Priorities." Journal of Nutrition. 133, no. 11(2003): 3911S-3926.
Sahn, D. E. "The Effect of Price and Income Changes on Food-Energy intake in Sri Lanka."
Economic Development & Cultural Change 36, no. 2(1988): 315.
34
Senauer, B., and N. Young. "The Impact of Food Stamps on Food Expenditures: Rejection of the
Traditional Model." American Journal of Agricultural Economics 68, no. 1(1986): 37-43.
Strauss, J. "Joint determination of food consumption and production in rural Sierra Leone :
Estimates of a household-firm model." Journal of Development Economics 14, no.
1(1984): 77-103.
Strauss, J., and D. Thomas (1995) Human Resources: Empirical Modeling of Household and
Family Decisions, in J. Behrman, and T. N. Srinivasan, eds. Handbook of Development
Economics vol. 3A, Amsterdam Elsevier, pp. 1883-2023.
Strauss, J., and D. Thomas. The shape of the calorie-expenditure curve. Center Discussion Paper
No. 595: Economic Growth Center Discussion Paper No. 595, Yale University, 1990.
Subramanian, S., and A. Deaton. "The demand for food and calories." Journal of Political
Economy 104, no. 1(1996): 133.
Thaler, R. H. "Mental accounting matters." Journal of Behavioral Decision Making 12, no.
3(1999): 183-206.
Thomas, D. "Intra-Household Resource Allocation: An Inferential Approach." The Journal of
Human Resources 25, no. 4(1990): 635-664.
Torheim, L. E., F. Ouattara, M. M. Diarra, F.D. Thiam, I. Barikmo, A. Hatloy, and A. Oshaug.
"Nutrient adequacy and dietary diversity in rural Mali: association and determinants."
European Journal of Clinical Nutrition 58(2004): 594-604.
Wolfe, B. L., and J. R. Behrman. "Is Income Overrated in Determining Adequate Nutrition?"
Economic Development & Cultural Change 31, no. 3(1983): 525.
35
Table 1: Full Sample Summary Statistics N=2495
Number of Households=319
Proportion of Total Income
Mean Standard Deviation
Min Max
Total Income 1.00 131.12 255.26 0.00 7402.22
Trade and Business 0.06 8.16 56.80 0.00 1652.22
Wages and Salary 0.08 18.46 73.59 0.00 763.16
Livestock Trade 0.13 19.94 101.54 0.00 4033.66
Livestock Products 0.43 60.43 151.64 0.00 5551.66
Crop Value 0.03 5.70 45.87 0.00 1512.62
Remittances 0.28 18.43 59.54 0.00 1896.96
Herd Size (TLU) 13.31 24.44 0.00 484.89
Dietary Diversity 3.24 1.55 0.00 8.00
Head is Male 0.69 0.46 0.00 1.00
Head is Married 0.68 0.47 0.00 1.00
Proportion which attended
Adult/Literacy School 0.06 0.24 0.00 1.00
Primary School 0.05 0.22 0.00 1.00
Secondary School 0.07 0.25 0.00 1.00
36
Table 2: Full Sample Summary Statistics by Income Tercile Number of
households % of
Household Income
Mean Standard Deviation
Min Max
Lower Tercile 107 Total Income 40.41 41.21 0.00 379.49 Trade and Business
0.06 2.07 8.14 0.00 113.45
Wages and Salary
0.04 1.79 11.16 0.00 228.31
Livestock Trade 0.11 7.14 20.85 0.00 233.22 Livestock Products
0.36 14.82 24.40 0.00 248.45
Crop Value 0.03 1.61 9.96 0.00 139.99 Remittances 0.40 12.98 22.57 0.00 379.49 Middle Tercile 106 Total Income 1.00 96.34 91.71 0.00 614.80 Trade and Business
0.06 5.14 17.70 0.00 322.40
Wages and Salary
0.04 4.14 24.41 0.00 479.96
Livestock Trade 0.14 17.60 38.67 0.00 381.58 Livestock Products
0.49 50.78 68.99 0.00 487.09
Crop Value 0.03 5.12 31.06 0.00 342.77 Remittances 0.24 13.56 18.54 0.00 191.30 Upper Tercile 106 Total Income 1.00 281.03 417.37 0.00 7402.22 Trade and Business
0.06 19.04 101.33 0.00 1652.22
Wages and Salary
0.15 55.25 124.50 0.00 763.16
Livestock Trade 0.13 38.12 178.87 0.00 4033.66 Livestock Products
0.45 126.64 253.63 0.00 5551.66
Crop Value 0.02 11.30 76.19 0.00 1512.62 Remittances 0.18 30.68 103.60 0.00 1896.96
37
Table 3: Full Sample Elasticity Estimates N=2495
Groups=319 Lower Middle Upper
Elasticity Standard
Error Elasticity Standard
Error Elasticity Standard
Error
Disaggregated Income Model
Trade & Business 0.06*** 0.01 0.01 0.01 0.02*** 0.01
Wages & Salary 0.04*** 0.01 0.02** 0.01 0.02*** 0.01
Livestock Trade 0.02*** 0.01 0.001 0.01 -0.001 0.01
Livestock Products 0.01 0.01 0.01 0.01 0.004 0.01
Crop Value -0.01 0.02 0.02** 0.01 -0.001 0.01
Remittances 0.03*** 0.01 0.04*** 0.01 0.02*** 0.01
Aggregated Income Model
Aggregated Income 0.04*** 0.01 0.04*** 0.01 0.03*** 0.01 ***, **, * significant at the one percent, five percent and ten percent levels respectively
38
Table 4: Single Adult Sub-Sample Summary Statistics Number of
Households % of
Household Income
Mean Standard Deviation
Min Max
Lower Tercile 34 Total Income 1.00 40.35 43.27 0.00 268.28 Trade and Business
0.06 2.47 8.37 0.00 73.04
Wages and Salary 0.05 2.88 18.45 0.00 228.31 Livestock Trade 0.09 7.35 24.78 0.00 233.22 Livestock Products
0.34 13.21 21.98 0.00 142.73
Crop Value 0.02 1.23 7.80 0.00 76.89 Remittances 0.44 13.22 18.26 0.00 167.54 Middle Tercile 23 Total Income 1.00 87.78 95.76 0.00 498.39 Trade and Business
0.05 2.79 6.56 0.00 50.84
Wages and Salary 0.03 4.62 36.06 0.00 479.96 Livestock Trade 0.13 16.39 42.80 0.00 381.58 Livestock Products
0.48 46.69 69.52 0.00 397.69
Crop Value 0.05 7.19 37.83 0.00 318.11 Remittances 0.26 10.10 13.23 0.00 126.46 Upper Tercile 15 Total Income 1.00 278.18 194.57 3.81 941.22 Trade and Business
0.12 27.69 85.34 0.00 497.57
Wages and Salary 0.38 133.32 163.10 0.00 576.63 Livestock Trade 0.06 14.81 38.11 0.00 247.50 Livestock Products
0.23 58.83 83.45 0.00 417.50
Crop Value 0.01 2.68 26.04 0.00 261.68 Remittances 0.21 40.86 75.83 0.00 612.37
39
Table 5: Single Adult Sub-Sample Elasticity Estimates N= 567
Groups= 72 Lower Middle Upper
Elasticity Standard
Error Elasticity Standard
Error Elasticity Standard
Error
Disaggregated Income Model
Trade & Business 0.067** 0.028 0.023* 0.013 0.032* 0.017
Wages & Salary 0.065*** 0.017 0.000 0.020 0.030*** 0.012
Livestock Trade -0.004 0.011 0.004 0.012 -0.039*** 0.016
Livestock Products 0.014 0.010 -0.005 0.014 -0.006 0.019
Crop Value 0.026 0.038 0.019 0.013 0.041*** 0.017
Remittances 0.025** 0.013 0.030* 0.018 0.018 0.012
Aggregated Income Model
Aggregated Income 0.040*** 0.015 0.024 0.024 0.034 0.028 ***, **, * significant at the one percent, five percent and ten percent levels respectively
40
Table 6: Milk Market Sub-Sample Summary Statistics Number of
Households % of
Household Income
Mean Standard Deviation
Min Max
Lower Tercile 40 Total Income 1.00 47.55 34.67 0.00 181.97 Trade and Business
0.02 0.73 3.06 0.00 17.57
Wages and Salary 0.00 0.07 0.54 0.00 4.81 Livestock Trade 0.10 7.07 15.78 0.00 85.09 Livestock Products
0.63 29.43 23.86 0.00 117.21
Crop Value 0.00 0.13 1.25 0.00 11.64 Remittances 0.25 10.10 12.46 0.00 64.76 Middle Tercile 45 Total Income 1.00 127.39 91.60 10.35 614.80 Trade and Business
0.06 7.72 13.19 0.00 68.07
Wages and Salary 0.01 2.17 9.42 0.00 72.55 Livestock Trade 0.14 20.14 38.26 0.00 220.94 Livestock Products
0.59 75.44 71.57 0.00 440.70
Crop Value 0.04 5.65 23.88 0.00 225.12 Remittances 0.15 16.28 22.26 0.00 121.58 Upper Tercile 53 Total Income 1.00 282.51 218.71 5.85 1673.54 Trade and Business
0.11 39.97 159.39 0.00 1652.22
Wages and Salary 0.04 12.30 53.75 0.00 381.28 Livestock Trade 0.11 29.58 51.99 0.00 251.16 Livestock Products
0.58 149.87 120.01 0.00 745.35
Crop Value 0.05 24.63 84.74 0.00 500.95 Remittances 0.11 26.15 88.11 0.00 775.98
41
Table 7: Milk Market Sub-Sample Elasticity Estimates N= 370
Groups= 138 Lower Middle Upper
Elasticity Standard
Error Elasticity Standard
Error Elasticity Standard
Error
Disaggregated Income Model
Trade & Business -0.003 0.025 0.02 0.019 0.02 0.016
Wages & Salary -0.10 0.088 -0.05** 0.021 0.02 0.017
Livestock Trade 0.02* 0.014 0.03* 0.015 0.02 0.013
Livestock Products -0.002 0.019 -0.01 0.019 0.02 0.025
Crop Value -0.10* 0.054 -0.01 0.015 -0.0002 0.015
Remittances -0.01 0.016 0.002 0.017 0.04 0.023
Aggregated Income Model
Aggregated Income 0.007 0.024 -0.001 0.026 0.096*** 0.036 ***, **, * significant at the one percent, five percent and ten percent levels respectively
42
Table 8: Reduced Form Regressions Total Income Regression Dietary Diversity Regression
N 2348 2348
Lower Middle Upper Lower Middle Upper Predicted Trade &
Business 0.076 -0.039 0.059 0.129* 0.037 -0.063 Predicted Wages & Salary
-0.141 0.106 0.327* -0.167*** -0.109** -0.022 Predicted Livestock Trade
0.156 0.285* 0.367** 0.133*** 0.063 0.124*** Predicted Livestock
Products 0.255* 0.537*** 0.480*** 0.050 0.022 -0.016 Predicted Crop Value
0.504** 0.449** 0.548*** 0.149*** 0.174*** 0.087* Predicted Remittances
0.271* 0.179 0.005 0.021 -0.017 -0.026 ***, **, * significant at the one percent, five percent and ten percent levels respectively
43
APPENDIX Table A1: Full Regression Results from Disaggregated Income Model
Full Sample Milk Sale Sub-Sample
Single Adult Sub-Sample
N 2495 370 567 Number of Individuals 319 138 72 ln(Trade & Business) 0.060*** -0.003 0.067**
ln(Trade & Business) x Middle Tercile
-0.050*** 0.024 -0.044
ln(Trade & Business) x Upper Tercile
-0.041*** 0.020 -0.035
ln(Wages & Salary) 0.042*** -0.100 0.065*** ln(Wages & Salary) x Middle
Tercile -0.020 0.051 -0.065***
ln(Wages & Salary) x Upper Tercile
-0.025** 0.115 -0.035
ln(Livestock Trade) 0.020*** 0.024* -0.004 ln(Livestock Trade) x Middle
Tercile -0.019** 0.002 0.008
ln(Livestock Trade) x Upper Tercile
-0.021** -0.008 -0.035*
ln(Livestock Products) 0.006 -0.002 0.014 ln(Livestock Products) x Middle
Tercile 0.002 -0.004 -0.019
ln(Livestock Products) x Upper Tercile
-0.002 0.021 -0.020
ln(Crop Value) -0.012 -0.095* 0.026 ln(Crop Value) x Middle Tercile 0.033* 0.088 -0.008 ln(Crop Value) x Upper Tercile 0.011 0.095* 0.015
ln(Remittances) 0.028*** -0.012 0.025* ln(Remittances) x Middle Tercile 0.010 0.014 0.005 ln(Remittances) x Upper Tercile -0.006 0.048 -0.007
ln(Maize Price) -0.021 ln(Tea Price) -0.116***
ln(Milk Price) 0.012 ln(TLU) 0.030***
Adult Education 0.060** Primary School 0.031
Secondary School 0.146*** Male -0.033** Age 0.000
Household Size 0.000 Average Rainfall -0.000
Middle Tercile 0.008 -0.059 0.149** Upper Tercile 0.049 -0.194 0.135
Constant 1.291*** 1.585*** 1.433*** R2 within 0.0411 0.0795 0.0657
between 0.8502 0.6313 0.8840 overall 0.5307 0.5054 0.5747
***, ** and * significant at the one, five and ten percent level, respectively
44
Table A2: Full Regression Results from Aggregated Income Model Full Sample Milk Sales Sub-
Sample Single Adult Sub-
Sample N 2633 407 580
Number of Groups 319 145 72 ln(Total Income) 0.044*** 0.007 0.040***
ln(Total Income) x Middle Tercile
-0.004 -0.008 -0.016
ln(Total Income) x Upper Tercile -0.016 0.090** -0.006 ln(maize price) 0.020
ln(Tea Price) -0.048 ln(Milk Price) -0.005
ln(TLU) 0.018* Adult Education 0.072*** Primary School 0.046*
Secondary School 0.174*** Male -0.036** Age 0.000
Household Size 0.001 Average Rainfall -0.000
Middle Tercile 0.003 0.004 0.103 Upper Tercile 0.038 -0.437** 0.032
Constant 1.388*** 1.498*** 1.455*** R2 within 0.0310 0.0384 0.0429
between 0.8516 0.6432 0.8473 overall 0.5204 0.5030 0.5478
***, ** and * significant at the one, five and ten percent level, respectively
45
Table A3: First Stage Regressions Predicting Source-Specific Income Dependent Variable Total
Income Trade & Business
Wages & Salary
Livestock Trade
Livestock Products
Crop Value
Remittances
Current Rain 0.004*** 0.000 0.002 -0.005** 0.005** 0.006*** 0.002 Lagged Rain -0.000 0.002 -0.002 -0.006*** -0.005* 0.007*** 0.002
1 Quarter Culm ZNDVI
-0.028*** -0.003 -0.002 0.013 -0.045*** -0.035*** 0.015*
2 Quarter Culm ZNDVI
0.008 0.011* 0.014* -0.003 0.007 0.017*** 0.009
3 Quarter Culm ZNDVI
0.014*** -0.002 -0.004 0.002 0.022*** 0.004 -0.004
Illness 0.222 0.090 0.020 0.383 0.259 -0.219 -0.012 Region 1 0.979*** -0.075 -0.178 0.811*** 2.595*** -0.995*** 0.239* Region 2 1.056*** 0.295 1.267** 0.249 0.990*** -1.011*** 0.975*** Region 3 0.179 0.678** 0.931* 0.553** 1.095** -1.240*** -1.049*** Region 4 0.545** 0.793*** -0.195 0.893*** 1.514*** -0.926*** 0.349** Region 5 0.183 1.299*** 0.009 1.079*** 0.114 -1.331*** -0.306 Region 6 -0.295 -0.230 -0.678** 0.405* 0.508 -0.098 -0.676*** Region 7 -0.436* -0.384** -0.700*** 0.257 0.481 -1.048*** -0.086 Region 8 -0.246 -0.259 -0.558* 0.731*** 1.010** -0.634*** -1.821*** Region 9 -0.054 -0.567*** -0.744*** 0.349* 1.733*** -1.032*** -1.473***
Region 10 -0.581*** -0.405** -0.736*** 0.518** 0.727** -0.965*** -1.106*** Illness interacted with
Region 1 -0.292 -0.030 0.410 -0.720 -0.291 0.290 -0.123 Region 2 -0.042 -0.015 0.370 -0.100 -0.316 0.303 -0.292 Region 3 -0.014 -0.333 -0.290 -0.227 -0.642 0.582 0.846** Region 4 -0.009 0.087 -0.042 -0.540 -0.163 0.221 0.248 Region 5 0.233 0.452 -0.308 -0.185 0.580 0.231 0.505 Region 6 -0.433 -0.294 0.052 -0.379 -0.390 0.473 0.191 Region 7 0.346 0.131 0.140 0.019 0.078 0.215 0.831** Region 8 0.265 -0.068 0.008 0.266 -0.050 0.339 0.243 Region 9 0.353 -0.079 -0.003 0.919 -0.036 0.125 1.329**
Region 10 -0.219 0.052 0.095 -0.137 -0.384 0.144 -0.236 Seasonal controls
June 0.006 -0.065 -0.133 -0.095 0.143 0.316*** -0.252** September -0.036 -0.026 -0.013 0.180* -0.690*** 0.208*** 0.631*** December -0.110 -0.108* -0.175** 0.104 -0.333*** -0.061 0.376***
Constant 3.862*** 0.607*** 0.847*** 0.818*** 1.964*** 0.564*** 1.982*** R2-between 0.299 0.311 0.257 0.111 0.291 0.595 0.730
R2-within 0.035 0.022 0.022 0.023 0.079 0.063 0.117 R2-overall 0.166 0.221 0.187 0.051 0.174 0.187 0.374
***, ** and * significant at the one, five and ten percent level, respectively