Download PDF

Maize Hybrids, Diversity of Diets and Sources of Vitamin A among
Smallholder Farmers in Zambia
Melinda Smale1, Mourad Moursi2, Ekin Birol2
Invited paper presented at the 4th International Conference of the African Association of
Agricultural Economists, September 22-25, 2013, Hammamet, Tunisia
1
2
Michigan State University, corresponding author. [email protected]. +703 231 8492.
HarvestPlus, Washington, DC.
1
Maize Hybrids, Diversity of Diets and Sources of Vitamin A among
Smallholder Farmers in Zambia
Abstract
Since a 1994 study by Kumar, we are not aware of analyses that have related the adoption of
hybrid seed to dietary diversity and smallholder farmers in Zambia, despite the policy
importance of nutrition and food security. We apply various econometric approaches to test
the relationship of hybrid seed use to dietary diversity and diversity in sources of Vitamin A
among smallholder maize growers in Zambia, based on a survey of 1,128 households. Four
dietary diversity indicators are defined according to recent research advances: food group
diversity, vitamin A diversity, food frequency, and frequency of consuming vitamin Afortified food. Growing hybrid seed, whether measured as a binary variable or in terms of
quantities planted, is strongly and positively related to all four indicators. Findings are robust
to econometric models. Although it is often hypothesized that growing hybrids promotes crop
specialization and reduces the range of on-farm food sources, these results suggest that
instead, hybrid growers enjoy a greater range of food sources from farm production, market
purchase, or both. In Zambia, smallholder maize farmers who do not grow hybrid seed are
therefore a disadvantaged group, not only with respect to maize productivity, but other key,
diet-related welfare indicators.
2
Maize Hybrid Impacts on the Diversity of Diets and Sources of
Vitamin A among Smallholder Farmers in Zambia
I.
Introduction
Research implemented by the International Food Policy Research Institute during the late 1980s
in the Eastern Province of Zambia (Kumar 1994) concluded that adoption of maize hybrids was
substantial even among smallholder farmers and that the impact of adoption on food intake was
greater on farms under 4 ha than on larger farms. While staple food consumption rose in areas
where adoption rates were higher, Kumar (1994) found that, among smallholders, dietary
diversity may have declined due to greater reliance on own maize production and fewer
purchased food types. Historically, Eastern Province has had lower rates of hybrid seed adoption
and farmers have insisted on continuing to grow local maize varieties because of strong
consumption preferences for flint-type grain. Thus, the impact of adoption on food intake where
hybrid maize was a cash crop, particularly among smallholders, was a striking finding.
Since the study by Kumar (1994), we are not aware of analyses that have related hybrid
maize use to dietary diversity per se, although malnutrition and food security have occupied the
center stage of agricultural policy in Zambia. For example, despite progress in supplementing the
consumption of vitamin A through health programs and fortified sugar, vitamin A deficiency,
and other micronutrient deficiencies, continue to jeopardize maternal and child health. Vitamin A
deficiency, which is widespread in the region, is a cause of blindness, night blindness, impaired
growth, weakened immune systems and increased risk of death due to infection among children;
among women in pregnancy, vitamin A deficiency may also contribute to increased risk of
morbidity, increased risk of neonatal mortality, and nightblindness (West 2002). In Zambia, over
half of children under five years of age are considered to be vitamin A deficient, as indicated by
low plasma retinol concentrations (NFNC/CDC, 2004). These rates could be higher in the
presence of infection (Hotz et al. 2011). Based on data drawn from Zambia’s 2005/06 Living
Conditions Monitoring Survey, Fiedler et al. (2012) found that while sugar fortification has
reduced inadequate vitamin A intake (IVAI) from 87% to 79%, sugar, alone, will not
substantially improve the nutrient intake status if the Zambian people.
Growing hybrid seed could contribute in potentially contradictory ways to the diets of
smallholder, maize-growing households in Zambia. On one hand, growing higher-yielding maize
may enable smallholders who rely for their staple food requirements on their own production to
reduce land needed to satisfy consumption needs and re-allocate it to production of other food or
cash crops. Reliance on own production can persist in remote areas with uncertain markets,
where farm sizes are small, and where production is particularly risky. Thus, growing more
diverse crops can contribute to dietary diversity through on-farm production and consumption.
On the other hand, hybrid maize often plays the role of a cash crop when there are no more
remunerative alternatives or where there are strong consumption preferences for local maize, as
3
has been documented for decades in Eastern Province. In that case, smallholder farmers would
sell their harvest and have the opportunity to diversify the foods consumed by their households.
However, under subsidized input programs, households are often bound by farmer organizations
to sell their harvest in order to repay their contributions to the cost of inputs provided through
loans. Diversification of diets through consumption requires not only steady cash flows but also
reliable markets for a range of food products.
Our null hypotheses are that maize hybrid use affects dietary diversity and the range of
foods consumed that are sources of vitamin A in the diet. However, we have no apriori reason
for predicting the direction of effect. In this paper, we document dietary diversity among
smallholder maize growers and test the effect of hybrid seed use on dietary diversity and
diversity in sources of Vitamin A. Dietary diversity indicators are defined according to recent
research advances (Arimond et al., 2010). Dietary diversity is modeled as a reduced-form
outcome of optimal decisions by the agricultural household to grow or purchase food. We relate
hybrid seed use directly to dietary outcomes based on a cross-sectional survey of 1,128 maizegrowing farm households in the major maize-producing regions of Zambia, applying various
econometric methods to test our hypotheses.
This paper is intended to provide information of use in designing strategies to introduce
vitamin-A enriched maize among Zambian smallholders, and also on its potential impact.
Dietary diversity scores estimated here may also serve as a baseline for Harvest Plus.
Since Shubh Kumar’s in-depth research in Eastern Province twenty years ago, we are not aware
of published analyses in Zambia that have specifically tested the relationship of maize hybrid use
to household dietary diversity or dietary sources of Vitamin A. HarvestPlus proposes to reduce
the risk of vitamin A deficiency in at-risk populations through the introduction of improved crop
varieties biofortified with provitamin A (Bouis et al., 2011). Zambia was selected as a country
that might benefit greatly from the introduction of provitamin A rich maize varieties. This is
especially true given that although maize is the most commonly consumed food, only 23% of
Zambian households purchase pre-milled roller and breakfast maize—the only maize meal
products which are considered to be fortifiable (Fiedler et al. 2012).
II.
Measuring dietary diversity
Recognition of the prevalence and irreversible consequences of micronutrient malnutrition
among both urban and rural populations in South Asia and Sub-Saharan Africa, combined with
measurement challenges, led to the development of indices of household dietary diversity that
could be constructed via cost-effective survey instruments based on recall. Research
implemented by the International Food Policy Research Institute (e.g., Hoddinott and Yohannes,
2002) confirmed that a more diversified diet is associated with improvement in birth weight,
child anthropometric status, and improved hemoglobin concentrations, as well as caloric and
protein adequacy, percentage of protein from animal sources (high quality protein), and per
capita consumption (a proxy for household income). Studies that validated dietary diversity
4
against nutrient adequacy in developing countries confirmed a positive relationship and a
consistently positive association between dietary diversity and child growth (Ruel 2002). Even in
very poor households, increased food expenditure resulting from additional income is associated
with increased quantity and quality of the diet.
Ruel’s in-depth review (2002) concluded that although dietary diversity was universally
recognized as a key component of healthy diets, there was still a lack of consensus on how to
measure and operationalize it. Reference periods ranged from 1 to 15 days, and questions
remained regarding the classification of foods by group, portion size and frequency of intake,
scoring systems, cutoff points and references periods.
In a widely-used approach documented by Swindale and Bilinsky (2006), the household
dietary diversity score was operationalized as a count over 12 food groups consumed in either a
seven day or twenty-four hour reference period. To consider micronutrients, food groups were
expanded and/or regrouped by micronutrient and counted. Researchers had learned that even
where micronutrient-rich foods are available, inequitable intra-household distribution may
prevent access of women to micronutrient-rich food. Typically, women’s scores were found to
be representative of their children’s. Thus, in cases where the organization of the household and
decision-making processes suggested that distributional considerations were important, analysts
proposed scores constructed on the basis of individual interviews. For example, these were
recommended in the case of polygamous or extended family groups, or when focusing
specifically on the nutrition of women and children. Accordingly, Smale et al. (2012) applied
these in the Douentza region in rural Mali, where agricultural production (and consumption)
units are headed by a patriarch and composed of multiple households of the patriarch and his
married sons.
More recently, Arimond et al. (2010) completed an analysis of different food group
diversity indicators used to predict micronutrient adequacy of women’s diet among poor
populations. They constructed 8 candidate diversity indicators and assessed their performance
against the mean probability of adequacy for 11 micronutrients in five countries. Findings
confirmed the predictive strength for the diversity indicators, although the best performing
indicator depended on the country context. Based on their findings, they recommended an
indicator constructed on nine food groups: 1) all starchy staples; 2) all legumes and nuts; 3) all
dairy and dairy products; 4) organ meat; 5) eggs; 6) flesh foods and other small animal protein;
7) Vitamin A-rich dark green leafy vegetables; 8) other Vitamin-A rich vegetables and fruits; 9)
Other fruits and vegetables. Fats and oils are an optional group which was excluded from their
analysis, which focused specifically on the prediction of micronutrient adequacy among women.
In the context of micronutrient analysis, fats and oils are considered as calories only.
The food group diversity indicator Arimond et al. (2010) recommend is based on 24-hour
recall. This reference period has been shown to perform better when there was a minimum
quantity of consumption required for a food group to “count” in the diversity score. In their
analysis, researchers used data collected from individuals, and set the minimum consumption cut
5
off requirement at 15 grams. In other others, an individual had to consume at least 15 grams of
meat for instance in order for her to receive a count of +1 in the flesh food category.
In this analysis, we use four diversity indicators, each of which illuminates a different
aspect of consumption. The first, which we refer to as “food group diversity,” was constructed
over ten groups. An inventory of food consumed was elicited from the primary female decisionmaker in the household with reference to the preceding 24-hour period. Groups are defined as: 1)
starchy staples (maize, other cereals, sweet potato, or other roots and tubers);2) legumes and nuts
(beans and groundnuts or other pulses); 3) dairy (milk or cheese); 4) organ meats (kidney or
liver); 5) eggs; 6) flesh foods (fish, red meat, or poultry); 7) Vitamin A-rich fruits (mango,
papaya, guava); 8) Vitamin A-rich vegetables (green leafy vegetables); 9) other fruits and
vegetables; 10) Vitamin A-fortified foods (sugar or Blue Band margarine). Thus, to the 9 groups
recommended by Arimond et al. (2010), we added a tenth group for fortified foods to reflect
conditions specific to Zambia. The ten groups were constituted from the 19 categories included
in the survey instrument (See Annex A, Table 1). Given the scale of the baseline survey and its
multiple objectives, we were unable to control for quantities consumed in each category. We
define Vitamin A-rich foods as a “source” based on the Codex Alimentarius definition (60
Retinol Activity Equivalents (RAE) per 100g of Vitamin A)1. The food group diversity indicator
ranges in value from 1 to 10.
Our second indicator is a count over the groups containing foods that are sources of
Vitamin A or beta-carotene, which we call “vitamin A diversity.” Again, we based our
construction on the original 19 groups in the survey instrument. Both 24-hour and 7-day
indicators were constructed, but the 7-day indicator was preferred for econometric analysis
because it has a more normal distribution (Annex A, Figure 1). The Vitamin A diversity
indicator ranges from 0 to 9.
The third indicator we use is a “food frequency” index, constructed according to Arimond
and Ruel (2002). One of the main limitations of the first two indicators is that neither takes into
account consumption differences within a given food group. For instance, one household or
individual may have consumed meat three times and other only once during the reference period
but both score +1 for meat consumption. The authors proposed a conversion of data obtained for
a seven-day recall period to account for frequency of consumption. For each food group, a
household or individual receives a score of zero for frequencies fewer than four days per week, a
score of unity for frequencies from 4 to 6 (inclusive) times per week, and a score of two for
frequencies of seven or more. The diversity count is then summed across food groups, but with
ten groups, the hypothetical range of this indicator is considerably greater (1 to 20), and in the
data, the maximum is 17.
Our fourth indicator is the frequency of consuming Vitamin A fortified foods, which
include sugar and blue band margarine. Since these are costly consumption items, we
hypothesize that the relationship of growing hybrid maize to this outcome variable could differ
1
RAE differs from Retinol Equivalents (RE), which were common in older food composition
tables.
6
from that of the other indicators. We refer to this indicator as “Vitamin A fortified food
frequency.” The values of this indicator are (0,1, and 2).
III.
Conceptual framework
According to the framework of the agricultural household (Singh, Squire, and Strauss 1986; de
Janvry et al. 1991), the household combines farm resources and family labor to maximize utility
over consumption goods produced on the farm or purchased on the market, and leisure.
Decisions are constrained by a production technology, conditioned on the farm physical
environment and land area; family labor time, allocated to labor and leisure; and a full income
constraint. The full income constraint stipulates that a season’s expenditures of time and cash
cannot exceed the sum of net farm earnings and income that is ‘exogenous’ to crop and variety
choices. In a single-period model, ‘exogenous’ income includes stocks, remittances, pensions
and other transfers from the previous season.
The agricultural household approach is suitable for analyzing the decisions of farmers
who are not fully commercialized, and/or who operate with missing markets or imperfectly
functioning markets. When markets are perfect and farmers are neutral to risk, consumption and
production decisions are separable and the model of the agricultural household simplifies to
profit-maximization. Crop and variety choices are then based on relative prices and farm
physical conditions. Although there is variability in the orientation of Zambian maize growers,
most do not operate in a context of perfectly functioning markets, and we cannot assume they are
neutral to risk. In this setting, consumption and production decisions are non-separable.
Household characteristics that affect preferences and access to markets influence crop and
variety choices. In addition, the prices actually faced by farmers are not market prices but
shadow prices that reflect their household characteristics as well as market characteristics.
We follow Van Dusen’s (2006) adaptation of the household model to the analysis of crop
diversity, applying it instead to dietary diversity. Household utility is defined over the
consumption of goods produced on the farm (X) and purchased goods (Z), given a vector of
exogenous socioeconomic and household characteristics (Φ hh). Households maximize utility
subject to a full income constraint, a time constraint for household labor valued at the local
market wage, w, a non-tradability constraint, a constraint on production technology, and a
diversity constraint defining the optimal bundle of food attributes or combination of foods
consumed at the household level. Households choose the level of production of j crops, j = 1, 2,
…J, denoted by Qj. The cost function C(Q; Φ farm) incorporates the technological constraints
for the household, where Φfarm is a vector of exogenous farm characteristics. Following the
standard agricultural household model presented in Singh, Squire and Strauss (1986) and Van
Dusen (2006), the model can be expressed as follows:
Max U(X, Z; Φ hh)
(1)
7
Z = p(Q-X) – C(Q; Φ farm) + Y + wT)
(2)
Hi(Qj, X; Φ market) = 0
(3)
D = D(Qj, X, Z; Φ market)
(4)
The household chooses a vector of consumption levels (X,Z), and output levels, Q, such that the
general solution to the maximization of household utility under binding constraints is a set of
constrained optimal production levels, Qc, consumption levels Xc, and purchase levels Z:
Q = QCj(p, Φ hh, Φ farm , Φ market)
(5)
X = XCj(p, YC, Φ hh, Φ farm, Φ market)
(6)
Z = Z(p, YC, Φ hh, Φ farm, Φ market)
(7)
where YC represents the full income for the constrained optimal production levels QC.
The household’s constrained dietary diversity outcome can be expressed as follows:
DC = D(XCj,Z ( p, YC, Φ hh, Φ farm, Φ market))
(8)
Prices (p) are endogenous to the household, and are in turn, functions of household and market
characteristics that determine transactions costs and effective prices.
IV.
Empirical strategy
A detailed description of the sample design is found in De Groote et al. (2011). The population
domain incudes five provinces (Central, Copperbelt, Eastern, Lusaka, Northern and Southern
Province), located in three agroecological zones (I, IIA and III) of Zambia. A stratified, twostage sample was designed. The three agroecological zones (AEZs) served as strata. The total
number of households in the sample was allocated proportionate to population and maize
production (20% for zone I, 40% each for the other two zones). First stage sampling units were
standard enumeration areas (SEA). Numbering 113, these were selected with probability
proportionate to size, by AEZ, from lists maintained by the census bureau. The second stage
units were all households living in each SEA. Ten households were selected in each SEA by
simple random sample drawn from a list. By design, data are self-weighted. Data were collected
by three survey teams, each including a supervisor and five enumerators, in June and August of
8
2011. The full sample consists of 1128 households, of which only 19 cultivated more than 20 ha.
In Zambia, farmers cultivating less than 20 ha are defined as “smallholders.”
IV.
Empirical model
Our regression model is a reduced form equation that relates hybrid seed use and other
explanatory factors to dietary diversity among maize-growing farm households in Zambia
according to equation (8), rewritten as:
δi = β0 + β1Xi + γAi + εi
i= 1,..,N
(9)
Where δ expresses dietary diversity, X is a vector of exogenous explanatory variables, ε is the
random error term, and i indexes households. A is use of hybrid seed in maize production,
introduced explicitly to test the hypothesis of interest. Following the conceptual framework of
the agricultural household, we control for household, farm and market characteristics among
exogenous variables.
It is possible that our variable of interest, A (hybrid seed use), is endogenous due to
measurement error, simultaneity, or selection bias. That is, the unobserved factors that predict
the dietary diversity indicators might be correlated with the household’s decision to grow hybrid
seed. In that case, estimating Model (9) would result in biased estimates, overstating the impacts
of hybrid seed use on dietary diversity.
The model is estimated via two-stage least squares equation, with a binary variable
measuring hybrid seed use in the first stage even though the assumed model is linear. Angrist
and Krueger (2001) state that even in the case of a dichotomous variable in the first of the two
equations, two-stage least squares produces consistent estimators that are less sensitive to
assumptions about functional form. They advocate this approach over use of nonlinear models
such as probit or logit (2001: 80). Two-stage least squares, which relies on the central limit
theorem, is considered to be robust; even with a dummy endogenous variable, second stage
estimates are consistent (Kelejian, 1971).
To validate our regression results based on the binary variable, we also test the
endogeneity of hybrid seed quantities (kg) in each outcome equation. The amount of hybrid seed
planted is a variable that is censored at zero, for which a Tobit regression, which is a nonlinear
model, is appropriate. Instrumental variables regression with two-stage least squares, as
estimated above, is no longer suitable. The Control Function Approach which enables us to test
for the endogeneity or self-selection bias of a linear model with a potentially endogenous
censored variable. As in a two-stage instrumental variables model, the control function approach
requires an instrumental variable to be used in the first stage. In the second stage, however, the
structural model is estimated with the observed endogenous variable and the residual from the
first stage as explanatory variables. The test of endogeneity is the statistical significance of the
coefficient of the residual, estimated with bootstrapped standard errors. The control function
9
approach is fully developed in Wooldridge (2010), building on early work by Smith and Blundell
(1986) and others.
Finally, we compare the regressions estimated by instrumental variables and ordinary
least squares methods for the first three indicators (food group diversity, Vitamin A diversity,
and food frequency) to count regressions estimated with the assumption that the underlying data
are generated by a Poisson process. We apply ordered logit regression to test hypotheses
concerning the fourth indicator, which has three outcomes that are ordinal and ordered (from less
frequent to more frequent, where more frequent is hypothesized to be “better”).
Separate regressions were estimated for each outcome variable of interest. Outcome
variables include four dietary diversity scores: 1) the 24-hour household dietary diversity score;
2) the 7-day Vitamin-A dietary diversity score; 3) the food frequency score; and 4) the frequency
score for food fortified with Vitamin A. These and explanatory variables are defined and
summarized in Table 1.
As discussed in the introduction, we hypothesize that the impact of hybrid seed use could
be either positive or negative, depending on the orientation of the households’ production toward
consumption or sale and consumption preferences related to either grain texture or food
purchases on the market.
Human capital indicators, included the quality of labor (literacy) and household supply of
labor (number of active adults) are expected to affect the use of hybrid seed positively. The
number of dependents, on the other hand, may dampen these effects. A vast literature on
adoption of new seed varieties suggests strongly that households with more assets and more
likely to adopt. Adoption is also expected to be higher in more favorable production
environments, toward the northern areas of the country, and with less extreme temperatures. We
include both dummy variables for agroecological zones, as well as average temperatures and the
range of temperature keyed to georeferenced coordinates recorded for each household.
Temperatures are based on high-resolution monthly climate data from 1950–2000 (Hijmans et al.
2005), provided by Kai Sonder at CIMMYT (pers.comm., March 1, 2012). Subsidies clearly
affect hybrid seed use positively.
We also expect capital endowments to be positively related to dietary diversity through
better access to resources for farm production and markets for sales of products and food
purchases. Farming systems, and hence food produced on farms, varies by agro-ecology and
rainfall conditions. We have no hypothesized direction of effect for agro-ecological zone.
Subsidies influence dietary outcomes via the use of hybrid seed.
10
Table 1: Variable Definition and Summary Statistics
Variable
Dietary diversity outcomes
Construction
Food group diversity
Count over 10 food groups (see text)
5.27
1.66
Vitamin A diversity
Count over 19 food groups, of which 9 contain
sources of Vitamin A or beta-carotene (see text)
Count over 10 food groups by frequency category
(see text)
Frequency category for foods fortified with
Vitamin A (sugar, blue band margarine)
3.94
1.54
8.01
2.94
1.09
.921
Household grows named F1 hybrid=1, 0
otherwise
Total kg planted, named F1 hybrid
19.3
41.0
Literacy
Number of literate household members
3.66
2.35
Dependents
Number of household members <15 and >64
years of age
Number of households >15 and <64 years of age
3.58
1.89
3.28
2.09
63.6
320.9
20.9
1.36
13.4
1.20
Agro-ecological zone I
Total value of assets owned by household (mill
ZMK)
Average mean monthly climate data “1 km2”
resolution from 1950–2000
Average maximum less average minimum
temperature at 1 km2 resolution from 1950–2000
AEZI=1 for zone 1, 0 otherwise (see Figure 1)
.207
.405
Agro-ecological zone III
AEZIII=1 for zone 3, 0 otherwise
.381
.486
0.654
0.476
Food frequency
Vitamin A fortified food
frequency
Mean
St. Dev.
Potentially endogenous variables
Grow hybrid
Hybrid seed planted
Explanatory variables
Active adults
Assets (mill)
Average temperature
Temperature range
Instrumental variable
Received seed subsidy
1=received maize seed subsidy; 0=otherwise
Source: Authors.
V.
Results
A. Descriptive statistics
Table 3 shows the percentage of respondents who reported consuming foods classified among
the ten groups used to construct the individual dietary diversity score, considering all households
and comparing those who grow maize hybrids with those who do not.
As would be expected, the most frequently consumed items were starchy staples,
consisting predominantly of maize but including other cereals. Other Vit A rich fruits and
11
vegetable were next in order of importance. These are items that are more likely to be consumed
in one season than another, and in relatively small quantities as ingredients in stews or as snacks,
such as pumpkin, tomatoes, mango, and papaya. Other fruits and vegetables were also consumed
often by respondents interviewed. Nuts and legumes were next in order of overall frequency,
mostly reflecting the consumption of groundnuts. Vit A fortified foods (primarily sugar) were
consumed by over two-thirds (67.8 percent) of women interviewed in the previous 24 hours.
Dark green leafy vegetables were consumed by nearly one quarter (24 percent). Again,
consumption of this last group is seasonal. Flesh foods, including red meat, poultry and fish
(fresh or dried) were consumed by over half (53.8 percent). Less frequently consumed food
groups included dairy products (22.4 percent), eggs (21.9 percent), and especially, organ meats
(5 percent).
At significance values of 5% or less, households growing hybrid maize are more likely to
have consumed foods classified in any of the food groups, with the exception of other fruits and
vegetables and nuts and legumes. Bivariate relationships are strongest for foods containing large
amounts of protein, such as dairy products, flesh foods, eggs or organ meats, but there are also
highly significant for the group containing food that is fortified in Vitamin A (sugar, Blue Band
margarine).
Table 3. Comparison of food groups consumed in 24 hours preceding survey, by use of
maize hybrids
Grow maize hybrids
No
Yes
All
farmers
P-value*
Starchy staples
86.9
90.8
89.5
0.0550
Nuts and legumes
71.2
73.3
72.6
0.4810
Dairy
16.3
25.3
22.4
0.0010
Organ meats
Eggs
Flesh foods and other
small animal protein
Vit A rich dark green
leafy vegetables
Other Vit A rich fruits
and vegetables
Other fruits and
vegetables
Vit A fortified foods
2.6
6.1
5.0
0.0150
17.2
24.1
21.9
0.0100
45.6
57.7
53.8
0.0000
20.3
25.8
24.0
0.0530
85.2
89.4
88.0
0.0470
82.6
83.2
83.0
0.7950
58.7
72.0
67.8
0.0000
*Pearson chi-squared test comparing distributions.
Mean scores for all four diversity indicators are shown in Table 4, comparing users of F1
hybrid maize seed and non-users across the entire sample of maize-growing households. Out of
a total of 10 possible categories of food, households growing maize hybrids consumed a mean of
12
5.5 in a 24-hour period, compared with a mean of 4.9 among non-users. Vitamin A diversity was
also higher among hybrid maize growers, averaging 4.1 out of a total of 9 groups, compared to
3.6 among farmers who did not grow hybrids. Mean food frequency was 8.3 among hybrid maize
growers and only 7.5among non-growers. However, both of these scores are less than half the
hypothetical maximum of 20 for this indicator, illustrating that considering the frequency of
consumption has a dampening effect on relative dietary diversity. As expected given the findings
reported in Table 3, the frequency of consumption of foods fortified with Vitamin A is also
greater among maize-growing households how plant hybrid seed.
Table 4. Comparison of mean values of dietary diversity indicators, by use of maize hybrids
Household Dietary
Diversity
Vitamin A
Diversity
Food
Frequency
Vitamin A
Fortified
Frequency
Grow maize hybrids
No
4.87
3.56
7.45
0.901
Yes
5.48
4.13
8.28
1.17
5.28
3.95
8.01
1.09
0.0000
0.0000
0.0000
0.0000
All maize growers
pvalue, ttestdiff
Source: Authors.
Thus, descriptive statistics support the hypothesis that the net direction of various
possible effects of hybrid seed use on dietary diversity appear to be positive. Next, we test this
hypothesis econometrically using multivariate analysis and testing for selection bias with
instrumental variables.
B. Econometric results
Results of diagnostic tests performed with instrumental variables regression do not support that
the decision to use hybrid seed is endogenous in household dietary diversity, Vitamin A
diversity, or food frequency. The null hypothesis of homoskedasticity could not be rejected in
any of the regressions.
As a consequence of these findings, we estimated all regressions with ordinary least
squares (OLS) and the observed, binary variable for hybrid seed use as an exogenous variable
(equation 1 above). Recognizing potential clustering effects of households due to the sample
design, robust standard errors were estimated using clustering by village. Regression results are
shown in Table 5. Poisson regressions were also tested, but the similarity of the regression
coefficients and standard errors led us to conclude that no explanatory power was sacrificed with
OLS methods.
The impact of growing hybrid seed on household dietary diversity is significant at 1%,
raising the score by 0.35 points on average. Location in AEZIII, which is the maize-growing
13
Table 5. Ordinary least squares (OLS) regression, impact of growing hybrid maize seed (binary choice) on food group
diversity, Vitamin A diversity, and food frequency
Food group diversity
Vitamin A diversity
Robust
Coef.
Std. Err.
Food frequency
Robust
t
P>t
Coef.
Std. Err.
Robust
t
P>t
Coef.
Std. Err.
t
P>t
Grow F1 hybrid
0.3556
0.1415
2.5100
0.0120
0.4089
0.1307
3.1300
0.0020
0.3442
0.2159
1.5900
0.1120
Literacy
0.1421
0.0399
3.5600
0.0000
0.1037
0.0396
2.6200
0.0090
0.3141
0.0600
5.2300
0.0000
Dependents
-0.0254
0.0278
-0.9100
0.3620
-0.0035
0.0279
-0.1300
0.9000
-0.0177
0.0509
-0.3500
0.7280
Active adults
-0.0412
0.0458
-0.9000
0.3690
-0.0309
0.0444
-0.7000
0.4870
-0.1486
0.0705
-2.1100
0.0360
Assets
0.003789
0.001223
3.1000
0.0020
0.004639
0.000984
4.7200
0.0000
0.006611
0.002904
2.2800
0.0230
Average temperature
-0.0168
0.0539
-0.3100
0.7560
-0.0358
0.0464
-0.7700
0.4410
-0.0109
0.0860
-0.1300
0.9000
Temperature range
-0.1876
0.0526
-3.5700
0.0000
-0.0965
0.0515
-1.8700
0.0620
-0.2586
0.0831
-3.1100
0.0020
AEZI
-0.3124
0.3327
-0.9400
0.3480
-0.1692
0.2049
-0.8300
0.4090
0.0373
0.1908
0.2000
0.8450
AEZIII
0.7434
0.1633
4.5500
0.0000
0.2594
0.1529
1.7000
0.0900
2.0059
0.2926
6.8500
0.0000
Constant
7.3169
1.4489
5.0500
0.0000
5.1615
1.3123
3.9300
0.0000
10.0273
2.3217
4.3200
0.0000
F statistic = 13.86
Prob > F = 0.0000
R-squared = 0.1605
Source: Authors. Robust errors are clustered by village.
F statistic = 7.34
Prob > F
= 0.0000
R-squared = 0.0733
14
F statistic = 25.85
Prob > F = 0.0000
R-squared = 0.2495
region farthest to the north of the three, is associated with a very large increase in household
dietary diversity (0.74). Higher asset values, which are strongly correlated with land size, and
adult literacy rates, are strongly associated with more diverse diets. Dependents and labor
supply have notdiscernible effect. A wider long-term temperature range is a negative factor,
perhaps because it is associated with a less diverse cropping system found in the southern region
of the country.
Findings with respect to richness in sources of Vitamin A are similar, although the impact
of growing hybrid seed is even larger in magnitude (0.41) and significance (<1%). Temperature
range, and location in AEZIII have weaker effects on Vitamin A diversity. Considering the
frequency of food group consumption weakens the impact of grow hybrid maize, canceling out
its statistical significance. In this regression, the number of active adults in the household is
negatively associated with food frequency, perhaps because it reduces availability per capita,
other factors held constant. This result points to the importance of considering various,
comparative definitions for diversity measurement before drawing policy conclusions.
Similar results were also obtained when examining the impact of the scale of hybrid seed
use on dietary diversity. The statistical insignificance in the second-stage regressions of the
residual from the first-stage, reduced form regression explaining the kgs of hybrid seed planted
attests again to the exogeneity of growing hybrid maize in diversity outcomes. Ordinary least
squares regressions that treat the scale of hybrid seed use as exogenous are presented in Table 6.
With respect to either dietary diversity of vitamin A richness, regression results are statistically
stronger overall when the explanatory variable is measured in terms of the scale of hybrid use as
compared to the decision to grow hybrid seed. The impact of an additional kg of hybrid seed on
dietary diversity or vitamin A richness scores is small in magnitude compared to the effect of the
decision to grow hybrids. The magnitudes of the effects of literacy and farming in AEZIII remain
large. In general, regression results are similar for Vitamin A diversity and food frequency. Most
importantly for our comparative analysis, greater scale of hybrid maize production is associated
with higher dietary diversity when frequency of consumption is taken into consideration.
Results of the ordered logit regression testing the impacts of growing hybrid seed on the
frequency of consuming food that is fortified with Vitamin A (sugar, Blue Band margarine) are
shown in Table 6. The values of the dependent variable are (0, 1, 2). Planting hybrid maize seed
is associated with an average frequency increase of an average of about 2 on the ordinal scale, at
the 6% level of significance. We cannot reject the endogeneity of hybrid seed use in this dietary
outcome. As with the other dietary diversity indicators, asset values, adult literacy, and location
in AEZIII play positive roles in consumption of these items, which are luxury items. The number
of dependents in the household is weakly and negatively associated with consumption, given
more dependents implies, literally “more mouths to feed” by income-earners.
15
Table 6. Ordinary least squares (OLS) regression explaining the impact of scale of hybrid maize seed (continuous choice) on
food group diversity, Vitamin A diversity, and food frequency
Food group diversity
Vitamin A diversity
Robust
Coef.
Kgs F1 hybrid seed
Literacy
Std. Err.
0.0053
0.0012
Food frequency
Robust
t
P>t
4.5300
0.0000
Coef.
Std. Err.
0.0071
0.0012
Robust
t
P>t
5.8800
0.0000
Coef.
Std. Err.
0.0079
0.0022
t
P>t
3.6600
0.0000
0.1372
0.0394
3.4800
0.0010
0.0953
0.0368
2.5900
0.0100
0.3017
0.0579
5.2100
0.0000
Dependents
-0.0312
0.0276
-1.1300
0.2590
-0.0118
0.0272
-0.4300
0.6660
-0.0278
0.0498
-0.5600
0.5770
Active adults
-0.0547
0.0457
-1.2000
0.2320
-0.0477
0.0429
-1.1100
0.2670
-0.1653
0.0678
-2.4400
0.0150
0.003784
0.001297
2.9200
0.0040
0.004666
0.001064
4.3900
0.0000
0.006696
0.002981
2.2500
0.0250
-0.0078
0.0538
-0.1500
0.8840
-0.0194
0.0485
-0.4000
0.6900
0.0148
0.0858
0.1700
0.8630
-0.1884
0.0523
-3.6000
0.0000
-0.0974
0.0513
-1.9000
0.0580
-0.2594
0.0819
-3.1700
0.0020
Assets
Average
temperature
Temperature range
AEZI
-0.1772
0.2018
-0.8800
0.3800
0.0256
0.1960
0.1300
0.8960
-0.3271
0.3350
-0.9800
0.3290
AEZIII
0.8010
0.1651
4.8500
0.0000
0.3401
0.1522
2.2300
0.0260
2.1019
0.2929
7.1800
0.0000
Constant
7.3382
1.4117
5.2000
0.0000
5.0462
1.3266
3.8000
0.0000
9.6608
2.2585
4.2800
0.0000
F statistic = 15.74
F statistic = 10.18
F statistic = 27.73
Prob > F
Prob > F
Prob > F
= 0.0000
R-squared = 0.1706
Source: Authors. Robust errors are clustered by village.
R-squared
16
= 0.0000
= 0.1004
R-squared
= 0.0000
= 0.2600
Table 6. Ordered logit regression, impact of growing hybrid seed (binary choice) on the
frequency of consumption of Vitamin A-fortified food
Robust
Coef.
Grow F1 hybrid
Residual
Literacy
Std. Err.
z
P>z
2.2917
1.2243
1.8700
0.0610
-2.3399
1.2452
-1.8800
0.0600
0.1001
0.0588
1.7000
0.0890
Dependents
-0.0636
0.0394
-1.6100
0.1070
Active adults
-0.0197
0.0583
-0.3400
0.7360
Assets
0.023064
0.007874
2.9300
0.0030
Average temperature
0.2076
0.1017
2.0400
0.0410
Temperature range
0.0079
0.0802
0.1000
0.9220
-0.1499
0.2160
-0.6900
0.4880
1.0207
0.2312
4.4100
0.0000
AEZI
AEZIII
Wald chi2(10) =
Prob > chi2
=
62.22
0.0000
Log pseudolikelihood = -669.64947
Source: Authors. Robust errors are clustered by village.
VI.
Conclusions
Although much is known about maize production and also about household nutrition in Zambia,
we are unaware of studies that relate use of maize hybrids to dietary diversity other than the
analysis conducted by Kumar in 1994. In this paper, we sought to formally test the impact of
growing hybrids on the dietary diversity of maize-growing households based on baseline data
collected from over 1000 households in the major maize-producing areas of Zambia in 2011 by
Harvest Plus.
We have tested the association using a econometric approach based on the theoretic
frameworks of the agricultural household and the operational methods of quasi-experimental
impact evaluation. We tested the endogeneity of the decision to grow hybrid seed, due primarily
to the selection bias among adopters, using instrumental variables methods. Our instrumental
variable was subsidy receipt. Diagnostic statistical tests led us to fail to reject the exogeneity of
the decision to grow hybrids, suggesting the hybrid seed use and dietary diversity are related in a
recursive way. We then estimated a sequence of dietary diversity equations (food group
diversity, Vitamin A diversity, and food frequency) with hybrid seed use treated as an exogenous
variable. For robustness, we also tested the same models as Poisson regressions, given that
dietary diversity scores are measured as counts across food groups. Results were similar, but not
as strong statistically. Our fourth dietary diversity equation, explaining the frequency of
consumption of food fortified with Vitamin A, was estimated with ordered logit regression.
17
Finally, we tested for the potential endogeneity nature of the scale of hybrid seed use (kgs) using
an instrumented control function approach. As in the case of the binary adoption variable, we
failed to reject the exogeneity of hybrid demand (scale) in dietary diversity outcomes. Positive
scale effects of hybrid seed use on dietary diversity indicators were even stronger than adoption
effects.
Findings are also robust across econometric models, although the overall statistical
strength of the regressions and individual coefficients varies among them. Growing hybrid seed,
whether measured as a binary variable or in terms of kgs, has a powerful effect on the numbers
of food categories consumed by household members in either a 7-day or 24-hour reference
period, as well as the number of food groups that are sources of vitamin A. Other than this factor,
agro-ecological zone, which is related to the diversity of crops grown as well as market
infrastructure, plays an important role. Human capital, expressed in terms of adult literacy, as
well as household assets, which proxies for the capacity to generate income over time, are key
factors.
One major implication of this result is that dietary diversity, vitamin A supplementation,
and hybrid seed adoption appear to go hand-in-hand among smallholder farmers in Zambia. That
is, hybrid seed use does not counteract dietary diversity, as has often been hypothesized based on
the argument that growing hybrids tends to promote crop specialization and drive out crops that
serve as alternative food sources. Furthermore, it is well known that higher farm incomes from
crop sales do not necessarily translate to better nutrition or greater dietary diversity. While the
negative relationship between areas planted to maize and other crops have been observed at a
regional scale in Zambia, especially before and after structural adjustment when policies heavily
favored hybrid maize (Smale and Jayne 2003), we are not able to observe these in these crosssectional data collected from agricultural households in 2011. Unfortunately, the baseline data do
not permit us to explore this relationship with a more complete, multi-period, farmer decisionmaking model and other information on crops grown. This might be a suitable topic for further
research.
18
References
Arimond M. and Ruel M.T. 2002. Progress in developing an infant and child feeding index: an
example using the Ethiopia demographic and health survey 2000. IFPRI FCND
Discussion Paper 143. IFPRI: Washington DC.
Arimond, M. and M.T. Ruel. 2010. Dietary diversity is associated with child nutritional status:
evidence from 11 demographic and health surveys. Developing simple measures of
women’s diet quality in developing countries: An overview. J. of Nutrition : 2579-2585.
Bouis H.E., C. Hotz, B. McClafferty, J.V. Meenakshi, W. Pfeiffer. 2011. Biofortification: A new
tool to reduce micronutrient malnutrition. Food and Nutrition Bulletin. March 32(1
Suppl):S31-40.
de Janvry, A., Fafchamps, & Sadoulet, E. (1991). Peasant household behavior with missing
markets: Some paradoxes explained. The Economic Journal, 101, 1400-141
Fiedler, J.L., R. Zulu, G. Kabaghe, K. Lividini, J. Tehinse, O. Bermudez. 2012.Towards
Overcoming the Nutrition Information Gap-Induced Impasse: Identifying Zambia’s
Fortification Policy Options. Manuscript. HarvestPlus, Washington, DC.
Hoddinott, J. and Y. Yohannes. 2002. Dietary Diversity as a Household Food Security Indicator.
Food and Nutrition Technical Assistance Project, Academy for Educational
Development, Washington D.C., 2002.
Hijmans, R. J., S. E. Cameron, J. L. Parra, P. G. Jones and A. Jarvis. 2005. Very High Resolution
Interpolated Climate Surfaces for Global Land Areas. International Journal of
Climatology 25: 1965–1978.
Hotz, C., U. Palaniappan, J. Chileshe, E. Kafwembe and W. Siamusantu. 2011. Nutrition
Survey in Central and Eastern Provinces, Zambia 2009: Focus on Vitamin A and Maize
Intakes, and Vitamin A Status among Women and Children. National Food and Nutrition
Commission of Zambia, Tropical Diseases Research Centre, Zambia, and Harvest Plus:
Lusaka and Washington, D.C.
Kennedy, G., N. Fanou-Fogny, C. Seghieri, M. Arimond,Y. Koreissi,R. Dossa, F.J. Kok, and I.
D. Brouwer. 2010. Food groups associated with a compositive measure of probability of
adequate intake of 11 micronutrients in the diets of women in urban Mali. J. of Nutrition
140: 2070S–2078S.
Kumar, S. 1994. Adoption of hybrid maize in Zambia: Effects on gender roles, food consumption
and nutrition. International Food Policy Research Institute. Research Report 100.
Washington DC. IFPRI.
National Food and Nutrition Commission (NFNC) of Zambia, University of Zambia, MOST,
CDC. 2004. Report of the national survey to evaluate the impact of vitamin a
interventions in Zambia in July and November 2003.
Ruel, M.T. 2002. Is Dietary Diversity an Indicator of Food Security or Dietary Quality? A
Review of Measurement Issues and Research Needs. FCND Discussion Paper No. 140.
International Food Policy Research Institute (IFPRI), Washington, DC.
19
Ruel, M., M. Dietchler, and M. Arimond. 2010. Developing simple measures of women’s dietary
quality in developing countries: an overview. J. of Nutrition 140: 2048S-2050S.
Singh, I. L. Squire, and J. Strauss (eds.) 1986. Agricultural Household Models: Extensions,
Applications and Policy. The World Bank and Johns Hopkins University Press,
Washington, DC and Baltimore, MD.
Smale, M. L. Diakite and N. Keita. 2012. Millet transactions in market fairs, millet diversity and
farmer welfare. Environment and Development Economics 17, Special Issue 05, October:
523-546 DOI: http://dx.doi.org/10.1017/S1355770X12000216 (About DOI), Published
online: 16 May 2012.
Swindale, A. and P. Bilinsky. 2006. Household Dietary Diversity Score (HDDS) for
Measurement of Household Food Access: Indicator Guide (v.2). Washington, D.C.: Food
and Nutrition Technical Assistance Project, Academy for Educational Development.
UNICEF 2009a. Human Development Report.
UNICEF 2009b. Tracking Progress on Child and Maternal Malnutrition.
Van Dusen, M.E. 2006. Missing markets, migration, and crop biodiversity in the milpa system of
Mexico: A household-farm model. In M. Smale (ed.), Valuing Crop Biodiversity: Onfarm Genetic Resources and Economic Change. Wallingford, UK: CABI.
West, K.P., Jr. 2002. Extent of vitamin A deficiency among preschool children and women of
reproductive age. J. of Nutrition 132: 2857S-2866S.
WHO. 2009. Global Prevalence of Vitamin A Deficiency in Populations at Risk 1995-2005.
20
Annex. Table 1. Food groups as elicited in the HarvestPlus baseline survey
IA1
Food Item
1
Maize
2
Other cereals
3
Beans and other pulses (such as cowpeas, etc.)
4
Nuts and seeds
5
Sugar
6
7
8
9
10
11
12
13
Sweet potato, orange fleshed
Other roots and tubers (irish potatoes, cassava, white fleshed sweet potato
etc.)
Fruits (mango, pineapple, guava, pawpaw etc.)
Wild fruits
Eggs
Milk, cheese
Dark-green leafy vegetables
Other vegetables (pumpkin, tomatoes etc.)
14
15
16
Fish and other seafood (shrimp, crab etc.)
Red meat (cow/goat/pig/sheep/pork etc.)
Animal liver, kidney, and other offals (abats)
17
Poultry (chicken, duck, etc.)
18
Blue band margarine (vitamin A fortified)
19
Fats and oils (butter, other margarine, soybean, mustard, ghee, etc)
21
IA2
IA3
IA4
Did you consume [food
item] in the last 24 hours?
Did you consume [food
item] in the last 7 days?
How many times in the
last seven days did you
consume [food item]?
1 = Yes >> IA4
2 = No>> IA3
-99 = Don't know/remember
1 = Yes >> IA4
2 = No >> next item
-99 = Don't
know/remember
Number of times
.4
0
.2
Density
.6
.8
Annex. Figure 1. Distributions of Vitamin A diversity scores, by reference period
2
4
6
Vitamin A diversity, 7-day reference period
0
2
4
6
8
Vitamin A diversity, 24-hour reference period
8
.4
.2
0
Density
.6
.8
0
22
10
23