Agriculture-Nutrition Linkages, Cooking-Time, Intrahousehold Equality Among Women and Children: Evidence from Tajikistan

Abstract

Household-level agriculture-nutrition linkage (ANL) tends to be strong where food markets are imperfect. In such an environment, markets for cooking services may also be imperfect. Nutrition outcomes of household members, including intrahousehold inequality among women and children can depend on a household’s self-production of food and cooking services, provided both by household labors, particularly of women. Using the primary data in Tajikistan, we show that longer cooking-time by women in the household often strengthens household-level ANL and also reduces intrahousehold inequality in nutritional outcomes among women and children. These effects are stronger in areas with lower nighttime light intensity and for households with lower values of cooking assets. In a context where household-level ANL is strong, ANL may also depend on households’ self-production of complementary inputs, including cooking services. This dependence reveals both unique opportunities for and vulnerabilities of ANL for the rural poor.

Résumé

Les liens entre l’agriculture et la nutrition (en anglais : Agriculture-Nutrition Linkage, ANL) au niveau des foyers ont tendance à être forts là où les marchés sont imparfaits. Dans ces environnements, les marchés pour les services d’élaboration et de cuisson des repas peuvent aussi être imparfaits. La nutrition des membres d’un foyer, ainsi que l’inégalité nutritionnelle entre les femmes et les enfants du ménage, peuvent dépendre de la production d’aliments et des service de cuisson fournis au sein du foyer, souvent par le travail des femmes. Utilisant des données primaires au Tadjikistan, nous démontrons que des temps de cuisson plus longs par les femmes renforce l’ANL au niveau du foyer, et d’ailleurs réduit les inégalités nutritionnelles parmi les femmes et les enfants. Ces résultats sont plus marqués dans les régions ou il y a moins d’intensité lumineuse la nuit, et pour les foyers avec moins de ressources pour la cuisson. Dans les contextes ou l’ANL au niveau des foyers et élevée, l’ANL peut dépendre aussi d’autres contributions complémentaires, y inclus les services de cuisson. Cette dépendance révèle soit des opportunités uniques, soit des vulnérabilités ANL chez les pauvres rurales.

Appendices

Appendix A: List of Abbreviations

ADS: Agricultural diversity score

ANL: Agriculture-nutrition linkage

APP: Agricultural production practices

APV: Agricultural production values

BMI: Body mass index

CMAD: Children’s minimum acceptable diet

CMDD: Children’s minimum acceptable dietary diversity

DEA: Data envelopment analyses

GDP: Gross domestic product

GPS: Generalized propensity score

GPS-IPW: Generalized-propensity-score-based inverse probability weighting method

IFPRI: International Food Policy Research Institute

IPW: Inverse probability weighted

IV: Instrumental variables

IVGMM: Instrumental-variable generalized method of moments

NGPS-IPW: Nested GPS-IPW

PR: Primary respondent

SD: Standard deviation

TAWA: Tajikistan Agriculture and Water Activity

USAID: United States Agency for International Development

VCE: Value of cooking equipment

WAZ: Children’s weight-for-age Z-score

WDS: Women’s dietary diversity score

WDSM: Women’s minimum acceptable dietary diversity score

WHZ: Children’s weight-for-height Z-score

WLS: Weighted least square

WRA: Women of reproductive age

ZOI: Zone of influence

Appendix B: Theoretical Framework

Role of Household Cooking-Time on ANL

The demand for nutritional outcomes of individual $i$ ($N_{i}$) in household $h$ is selected by maximizing utility,

$$ U = U\left( {N_{i} ;Z_{h,i} } \right) $$

(6)

which depends on the inherent preference that depends on individual and household characteristics $Z_{h,i}$. Here, for simplicity, we drop the subscript $h$ and $i$ (we will reinstate them later when discussing intrahousehold equality).

$N$ is supplied by available food F (either home produced $A$ or purchased $F_{P}$), household members’ cooking-time $T$, and functions of cooking technologies C, level of food-market access $M$, and again Z (which can affect non-food sources of nutrition).

$$ N = f\left( {F, T;C,M, Z} \right). $$

(7)

Note $T$ does not include cooking provided by non-household members, such as purchased cooking services. The equilibrium nutritional outcome $N^{*}$ is jointly determined by these demand and supply factors.⁸

In low-income countries with food poverty (so that nutrition is still constrained by food availability F), we are likely to have

$$ \frac{{\partial N^{*} }}{\partial F} > 0. $$

(8)

The ANL literature generally suggests that, in rural areas with low $M$, we tend to have

$$ \frac{\partial F}{\partial A} > 0 $$

(9)

because of relatively higher costs for acquiring $F_{P}$ given $Z$.

Because low M would indicate relative scarcity of pre-cooked food, or most food items must be cooked before becoming digestible, or retaining bioavailability of nutrients in food items may require more careful cooking, $\frac{{\partial N^{*} }}{\partial F}$ responds more positively to cooking inputs. Furthermore, because low M is also associated with high cooking-service costs, cooking inputs are supplied mostly by households’ own cooking inputs, $T$. Thus,

$$ \frac{{\partial^{2} N^{*} }}{\partial F\partial T} > 0. $$

(10)

It is also likely that cooking-time T and cooking technologies are substitutes, so that $\frac{{\partial^{2} N^{*} }}{\partial T\partial C} < 0$.⁹ Therefore, combined with (10), we are likely to have

$$ \frac{{\partial \left( {\frac{{\partial^{2} N^{*} }}{\partial F\partial T}} \right)}}{\partial C} < 0. $$

(11)

Because we have $\frac{\partial F}{\partial A} > 0$, condition (11) suggests

$$ \frac{{\partial \left( {\frac{{\partial^{2} N^{*} }}{\partial A\partial T}} \right)}}{\partial C} < 0. $$

(12)

Household cooking-time $T$ may also be substituted with $M$, because as $M$ rises, the access to either pre-cooked food or affordable cooking services may improve. Similarly, higher M may be associated with more urban income-earning opportunities which can provide alternative means to improve nutrition, and thus higher M raises the opportunity cost of T for most individuals. Therefore, $\frac{{\partial^{2} N^{*} }}{\partial T\partial M} < 0$. Using the same argument as cooking technologies, we then have

$$ \frac{{\partial \left( {\frac{{\partial^{2} N^{*} }}{\partial A\partial T}} \right)}}{\partial M} < 0. $$

(13)

Conditions (12) and (13) lead to hypothesis 1:

Hypothesis 1

Conditional on household characteristics Z, household agricultural production (greater diversity or production scale) and cooking-time is generally complementary in improving equilibrium nutritional outcomes of women and children, and this effect is stronger for households with lower cooking technologies or in rural areas.

The hypothesis 1 may not hold if some of the aforementioned conditions fail. Therefore, whether hypothesis 1 holds or not is an important empirical question.

Intrahousehold Inequality Among Women and Children

Equilibrium nutritional outcomes of each household member maximize a certain utility function. Based on the realization of $N_{i}^{*}$, some indicators of intrahousehold equality $\sigma^{*}$ are obtained:

$$ N_{i}^{*} = \arg { \hbox{max} }_{{N_{i} }} U\left( {N_{i} ;Z_{h,i} } \right) \sigma_{h}^{*} = \sigma \left( {N_{i}^{*} ;Z_{h,i} } \right).$$

(14)

Note $\sigma_{h}^{*}$ is conditional on Z (and its intrahousehold variation) because observable Z is assumed to explain most of the variations in intrahousehold allocations (including activity-level, pregnancy status, etc.).

It is beyond the scope of this study to assess how the utility function is maximized. For the sake of presenting a theoretical framework, we consider the case in which allocation is made by a decision-maker in the household, with the goal of maximizing the aggregate consumer surplus (or utility) of all members while meeting the resource constraint. In this particular case, $N_{i}^{*}$ is determined such that

$$ \left. {\frac{{\partial U\left( {N_{i} ;Z} \right)}}{{\partial N_{i} }}} \right|_{{N_{i} = \hat{N}_{i} }} = \left. {\frac{{\partial U\left( {N_{j} ;Z} \right)}}{{\partial N_{j} }}} \right|_{{N_{j} = \hat{N}_{j} }} \forall j \ne i $$

(15)

so that every member’s marginal utility from an additional unit of nutrition is equated. When this holds, and if the utility function is the same across members conditional on $Z$ (so that all variations in inherent preference are solely explained by observable characteristics), we should have $\sigma_{h}^{*} = \sigma \left( {N_{i}^{*} ;Z} \right) = 0$, meaning perfect intrahousehold equality.

Even in such a case, $N_{i}^{*}$ may deviate from $\hat{N}_{i}$ if transaction costs in achieving the aforementioned allocations are greater than the returns from reallocations. These transaction costs will thus affect $\sigma_{h}^{*}$, conditional on $Z$.

$A$ and $T$ may jointly reduce $\sigma_{h}^{*}$ if they affect such transactions costs in such ways. For example, the greater overall availability of food and longer cooking-time may reduce such transactions costs. However, they can also operate in opposite ways.¹⁰ The discussion here is made for illustrative purposes, and how $A$ and $T$ affect $\sigma_{h}^{*}$ conditional on $Z$ is therefore an empirical question which we aim to investigate.

Hypothesis 2

Household agricultural production (greater diversity or production scale) and cooking-time are generally complementary in reducing intrahousehold nutritional inequality among women and children, and this effect is stronger for households with lower cooking technologies or in rural areas.

Technical Efficiency

The theoretical framework on the effects of technical efficiency in agriculture is similar to hypothesis 1. With higher efficiency, either greater $A$ or greater resources allocated for other purposes can improve nutritional outcomes. On the one hand, if greater $A$ is achieved, and if cooking-time $T$ is complementary, then higher technical efficiency and $T$ can be complementary to improving nutrition. On the other hand, the opportunity costs of resources used to raise technical efficiency in the area may be high (for example, they can be used for other means to more directly raise nutrition) to produce the same amount of food, or higher technical efficiency may not lead to sufficient food availability if households simply reduce inputs use. In such a case, raising technical efficiency may not improve nutritional outcomes.

Hypothesis 3

Higher technical efficiency in agriculture and cooking-time are complementary to each other in improving nutritional outcomes, and this effect is stronger for households with lower cooking technologies or in rural areas; these conditions may largely hold in the current setting of the studied area in Tajikistan.

Appendix C: Generalized-Propensity-Score-Based Inverse Probability Weighting Method

GPS is estimated in the following way (Hirano and Imbens 2004; Bia and Mattei 2008).¹¹ Each household $h = 1, \ldots , H$ is associated with a set of potential outcomes $\varpi_{h} \left( {T_{h} } \right)$, which is conditional on treatment $T_{h}$ (cooking-time in our case). Each $h$ is associated with observed covariates $Z_{h,i}$, cooking-time $T_{h} \in \left[ {t_{0} ,t_{1} } \right]$ where $t_{0}$ and $t_{1}$ are the lower and upper bounds of treatment level.

The conditional density of $T$ given the covariates $Z$ can be denoted as $r\left( {T,Z} \right) = f_{T|Z} (T|Z)$. Specific GPS, based on the observed $T_{h} ,Z_{h},$ can be denoted as $R_{h} = r\left( {T_{h} ,Z_{h} } \right)$. It is assumed that, within strata with the same $r\left( {T,Z} \right)$, the probability that $T = T_{h}$ is independent of $Z$, which is another way of saying that $T_{h}$ is independent of $Z_{h}$ once conditional on $R_{h}$, so that the changes in outcomes can be attributed solely to $T_{h}$ once conditional on $R_{h}$.

GPS is estimated in the following way. First, $T_{h}$ or its particular transformation (such as Diewert transformation) $g\left( {T_{h} } \right)$ are regressed on $X_{h}$, through the maximum likelihood method with normally distributed disturbance term. This regression estimates $r\left( {T,Z} \right) = \frac{1}{{\sqrt {2\pi \hat{\sigma }^{2} } }}\exp \left[ { - \frac{1}{{2\hat{\sigma }^{2} }}\left\{ {g\left( T \right) - \zeta \left( {\hat{\gamma },Z} \right)} \right\}} \right]$ in which $\zeta \left( {\hat{\gamma },Z} \right)$ is a function of $Z$ and parameters $\hat{\gamma }$. Given $r\left( {T,Z} \right)$, we compute $R_{h} = r\left( {T_{h} ,X_{h} } \right)$ for each observation $h$.

Balancing-Tests

Balancing tests for GPS methods are typically conducted by comparing GPS-adjusted means of $Z_{h}$ across subgroups that are defined based on the range of $T_{h}$. A standard approach (Hirano and Imbens 2004; Kluve et al. 2007) is to split the sample into three subsamples $G_{j} \left( {j = 1, 2, 3} \right)$ based on the terciles of $T_{h}$, divide each subsample into five blocks $G_{jk} \left( {k = 1, 2, 3, 4, 5} \right)$ based on the quintiles of the $R\left( {T_{h,j}^{m} ,Z_{h} } \right)$ evaluated at the median of $T_{h}$ within the tercile $j$, and then calculate the t-statistics for the equality of means of $Z_{h}$’s between blocks $G_{jk}$ and $G_{\xi k} \left( {\xi \ne j} \right)$.

Appendix D: More Results

See Tables 14, 15, and 16.

Table 14 Factors associated with APP (probit).

Table 15 Generalized-propensity-score estimation of cooking-time.

Table 16 Full results for some of the main equations on the effects of crop diversification score on various nutritional outcomes estimated from IPW (statistically significant signs)