WPS6650 Policy Research Working Paper 6650 Up in Smoke? Agricultural Commercialization, Rising Food Prices and Stunting in Malawi Benjamin Wood Carl Nelson Talip Kilic Siobhan Murray The World Bank Development Research Group Poverty and Inequality Team October 2013 Policy Research Working Paper 6650 Abstract Diversification into high-value cash crops among causal effect by comparing impact estimates informed smallholders has been propagated as a strategy to improve by two unique samples of children that differ in their welfare in rural areas. However, the extent to which cash exposure to an exogenous domestic staple food price crop production spurs projected gains remains an under- shock during the early child development window (from researched question, especially in the context of market conception through two years of age). The analysis finds imperfections leading to non-separable production and that household tobacco production in the year of or consumption decisions, and price shocks to staple crops the year after child birth, combined with exposure to that might be displaced on the farm by cash crops. This an exogenous domestic staple food price shock, lowers study is a contribution to the long-standing debate on the child height-for-age z-score by 1.27, implying the links between commercialization and nutrition. It a 70-percent drop in z-score. The negative effect is, uses nationally-representative household survey data however, not statistically significant among children who from Malawi, and estimates the effect of household were not exposed to the same shock. The results put adoption of an export crop, namely tobacco, on child emphasis on the food insecurity and malnutrition risks height-for-age z-scores. Given the endogenous nature materializing at times of high food prices, which might of household tobacco adoption, the analysis relies have disproportionately adverse effects on uninsured cash on instrumental variable regressions, and isolates the crop producers. This paper is a product of the Poverty and Inequality Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at tkilic@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Up in Smoke? Agricultural Commercialization, Rising Food Prices and Stunting in Malawi , Carl Nelson† Benjamin Wood∗ , and Siobhan Murray§ , Talip Kilic‡ JEL Classifications: C26, I15, O13, Q12 Keywords: Malawi, child nutrition, cash crops, tobacco, food prices Sector Boards: Agriculture and Rural Development (ARD), Health, Nutrition, and Population (HE) ∗ Corresponding author: Post-Doctoral Fellow, International Initiative for Impact Evaluation (3ie), Suite 450, 1625 Massachusetts Avenue NW, Washington, DC 20036, USA. bwood(at)3ieimpact.org † Associate Professor, Department of Agricultural and Consumer Economics, University of Illinois at Urbana-Champaign, USA. ‡ Research Economist, Poverty and Inequality Group, Development Research Group, The World Bank, Washington, DC, USA. § Technical Specialist, Computational Tools Group, Development Research Group, The World Bank, Washington, DC, USA. We would like to thank Mary Arends-Kuenning, Kathleen Beegle, Marc Bellemare, Andrew Dillon, Craig Gundersen, Nancy McCarthy, Nick Minot, Lia Nogueira, Raka Banerjee, and the participants at the 2011 Midwest International Economic Development Conference, the 2012 International Conference of Agricultural Economists, and the 2012 International Agricultural Trade Research Consortium for their comments on the earlier drafts of this paper, and the Malawi National Statistical Office and the Malawi Tobacco Control Commission for allowing us access the portions of the data used in the study. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of 3ie, the University of Illinois at Urbana-Champaign, The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. 1 1 Introduction Nearly 75 percent of the extreme poor in sub-Saharan Africa reside in rural areas, and over 90 percent of the sub-Saharan African rural extreme poor participate in agriculture as smallholders. In this context, spurring agricultural sector growth is a key target in national development plans across the region, based largely on the premise that aggregate agricultural growth brings disproportionate gains to the poorest in the developing world.1 In policy circles, one strategy that has surfaced to attain this target is increased commercialization of agriculture and diversification into high-value, labor-intensive cash crop production, particularly among smallholders. Proponents of increased commercialization of agriculture assert that resource-poor smallholders have a comparative advantage in labor-intensive cash crops, originating from cost savings that are realized by employing family laborers at below market remuneration. The reliance on household members on the farm is also argued to result in fewer agency problems, hence lower supervision costs.2 Cost-effective production processes on small farms, combined with relatively higher farm gate prices for cash crops, has fostered the expectation among international donors and policy makers that increased commercialization of agriculture could be a viable rural development strategy to improve welfare. However, the extent to which the economic gains of cash crop production result in household and individual welfare improvements remains an under-researched empirical question (Harrigan, 2008; Carletto, Kilic, and Kirk, 2011). The optimism should be balanced in view of (i) price shocks to staple food crops whose production on small farms might be displaced partially or fully by cash crop production, (ii) land, labor and credit market failures, compounded by limited stocks of household physical, financial and human capital, that might constrain the adoption and sustainability of cash crop production, (iii) intra-household dynamics that surround the control of income from cash crop production and that may not necessarily lead to welfare-enhancing expenditure patterns, and (iv) potential risks tied to the production and marketing of cash crops, partly driven, in the context of export-oriented crops, by increased international regulation and competition in international markets. 1 Ligon and Sadoulet (2007) document that a 1 percent rise in agricultural gross domestic product results in 6 percent income growth for the lowest income decile of the population. 2 Empirical evidence on this claim, however, appears to be limited. 2 In the case of Malawi, where agricultural commercialization has centered on tobacco as the main export, recent governmental policies have emphasized the importance of domestic maize and other food crop production to avoid national food deficits, the effects of volatile food prices, and the risk of long-term aid dependence (Ecker, Breisinger, and Pauw, 2011; Levy, Barahona, and Chinsinga, 2004).3 The emphasis on formulating policies that aim to boost staple crop production often presume, in the context of market failures, a direct link between on-farm production and nutrition outcomes. The presumed link has in turn raised concerns about the nutritional effects of commercialization of staple crops and displacement of staple food production by cash crop production. On one hand, there is evidence that greater food production or economic growth does not necessarily translate into improved nutritional health (Headey, 2011; Pelletier et al., 1995). On the other hand, earlier studies on Malawi had documented the nutritional outcomes of children originating from commercialized vis-à-vis subsistence farming households to be statistically indistinguishable from one another (von Braun, 1995; Kennedy, Bouis, and von Braun, 1992), and more recent empirical work on the country identified a negative correlation between child anthropometric measures and household tobacco adoption (World Bank, 2007). Focusing specifically on tobacco, its production is an inherently risky venture in Malawi given the predominantly rain-fed and unmechanized nature of agriculture, high costs associated with tobacco production, and uncertain returns from tobacco sales at auction floors. Moreover, considering incomplete land markets and the implications of high population density on shrinking farm sizes, tobacco production often comes at the expense of staple crop production on the farm. This in turn might expose uninsured smallholders to unexpected and large food price increases that limit their purchases of additional food required after foregoing staple food production (Sahn and Arulpragasam, 1991). In view of the gaps in the empirical literature on the effects of agricultural commercialization on welfare outcomes, this paper estimates the effect of household tobacco adoption on stunting among 6 to 59 month olds in Malawi when the child’s household is exposed to an exogenous staple food price shock during the early development window (from conception through two years of age).4 We 3 To further support agricultural production, the Malawian government created and supports the Agricultural Devel- opment Marketing and Corporation (ADMARC), which is a parastatal organization mandated to market agricultural produce and inputs. Typically at least one ADMARC outlet could be found in each district center in Malawi. 4 The definition of the early development window that is central to our empirical analysis is driven by research on long-term consequences on maternal and child undernutrition, which is explained in more detail in section 4 (Victora 3 identify the approximate bounds of the price shock of interest based on the monthly time series data on average maize prices, the geo-spatial time series data on rainfall, and the data on annual national maize production dynamics. The nationally-representative, multi-topic household survey data that include child anthropometric measures and detailed agricultural information, among other topics, underlie our distinct contribution to the relevant literature that has been dominated by studies using non-representative samples and exploring the welfare effects of commercialization during times of relative staple food price stability. We find that household tobacco production in the year of or the year after child birth and exposure to a domestic staple price shock lowers child height for age z-scores by 1.27, implying a 70 percent drop in z-score. These results put renewed emphasis on the food insecurity and malnutrition risks that materialize at times of high food prices and that might have disproportionate adverse effects on uninsured cash crop producers who limit their staple crop production. The paper is organized as follows. Section 2 presents a succinct review of the vast empirical literature on the links between agricultural commercialization and nutrition, followed by the history of tobacco commercialization in Malawi. Section 3 describes the data and Section 4 details the identification strategy. The results are presented in Section 5. Finally, the paper concludes with a summary of our results and their implications for policy and future research. 2 The Nexus between Agricultural Commercialization and Nutrition Exploration of the comparative advantages of developing countries in labor-intensive cash crops is rooted in the seminal development article by Johnston and Mellor (1961). Although an extensive literature exists surrounding agricultural commercialization, much of it was undertaken during times of relative food price stability and has focused on the documentation of short-term effects. The majority of these articles document large, positive and statistically significant effects of agricultural commercialization on household and child welfare outcomes.5 et al., 2008). 5 For more information on agricultural commercialization in the literature see (Côte d’Ivoire: Sahn (1990); The Gambia: von Braun, Puetz, and Webb (1989); Guatemala: von Braun, Hotchkiss, and Immink (1989); Katz (1994); Barham, Carter, and Sigelko (1995); Kenya: Kennedy and Cogill (1987); Philippines: Bouis and Haddad (1990); India: Birthal, Joshi, and Gulati (2005); Senegal: Maertens and Swinnen (2009); Madagascar: Minten, Randrianarison, and Swinnen (2007); Maertens and Swinnen (2009) provides a review of evidence from various sub-Saharan African settings). 4 Despite the presence of encouraging findings, the literature also features microeconometric studies that recover either negative or positive but statistically insignificant short-term effects of household agricultural commercialization on welfare outcomes (von Braun, de Haen, and Blanken, 1991; Kennedy, 1989; Immink and Alarcon, 1993; Immink et al., 1995). Complementing these results, Carletto et al. (2011) present evidence on the long-term effects of agricultural commercializa- tion on household welfare by using a panel data set spanning the period of 1985-2005, and focusing on non-traditional agricultural export crop adoption in the Central Highlands of Guatemala. The authors document that while consumption status improved for all household groups in the surveyed communities between 1985 and 2005, the extent of improvement among long-term adopters was lower in comparison to the changes experienced by non-adopters. Other factors influencing the relationship between household agricultural commercialization and welfare outcomes include unfavorable rainfall outcomes and staple food price shocks. Carter and Maluccio (2002) use South African panel data to show that the inability of communities to insure against covariate shocks results in increased levels of stunting after economic disasters. Similarly, Akresh, Verwimp, and Bundervoet (2007) demonstrate that crop failures in Rwanda lead to disproportionate reductions in the height of low-income girls, and Hoddinott (2006) finds that a Zimbabwean drought has long-term stunting effects on children less than two years old. 2.1 The Malawian Context The links between household agricultural commercialization and welfare are particularly important to understand in the case of Malawi where the headcount poverty rate stands at 51 percent and 85 percent of the population reside in rural areas, primarily engaged in agriculture (NSO, 2012). The sector is not only an essential part of the social fabric, but also the backbone of the economy. 84 percent of Malawian households own and/or cultivate land, and the contribution of agriculture towards the Gross Domestic Product (GDP) stands at 30 percent.6 Maize is the main staple crop, accounting for close to 60 percent of household caloric intake and being grown by nearly 100 percent of all farm households.7 Maize availability typically defines the food security status of the 6 The estimate of the contribution of agriculture towards the GDP is for 2011, and is obtained from http://data.worldbank.org/indicator/NV.AGR.TOTL.ZS. The estimate of the percentage of Malawian households owning and/or cultivating land is based on the Third Integrated Household Survey (IHS3) 2010/11. 7 The estimates are based on the IHS3 2010/11 data. 5 country. While household and national food self-sufficiency goals have been at the core of rural devel- opment strategies in the country, agricultural commercialization, particularly tobacco production, has been deemed essential for Malawian growth (Peters, 1996; World Bank, 1989, 2009; United Nations Development Programme, 2009; Republic of Malawi, 2000; Tobin and Knausenberger, 1998). Tobacco has played a prominent role in Malawi’s historical economic development. The production of the crop initially centered on large scale estates but, with international encouragement, Malawi began liberalizing their tobacco industry in the early 1990s.8 Smallholders first sold tobacco on the Malawian auction floor, under a quota, in 1991 (Tobin and Knausenberger, 1998). In 1994, Malawi repealed the Special Crops Acts that favored estates and smallholders quickly rushed into tobacco production (Tsonga, 2004). Eventually smallholder tobacco restrictions, in the forms of quota systems, control boards and grower’s clubs, were abolished. By 1998, almost 20 percent of Malawian households, including over 400,000 smallholders, produced tobacco (Kadzandira, Phiri, and Zakeyo, 2004). Almost all of the new households who entered the market grew burley tobacco, with smallholders currently accounting for 70 percent of Malawi’s total production (Lea and Hanmer, 2009). Although production levels have fluctuated over time, tobacco reigns not only as Malawi’s most important cash crop but also accounts for the majority of all Malawian exports (Orr, 2000; Jaffee, 2003). Malawi is well-suited for growing burley tobacco, much of which is used as a low-cost filler supplied to international cigarette producers. Tobacco production, with its delicate cultivation and particular harvesting requirements, exhibits few economies of scale (Jaffee, 2003). Additionally, burley tobacco’s air-curing process is not capital intensive, further lending itself to small-scale Malawian production (Takane, 2008). All legally sold tobacco in Malawi goes through regional auction floors depicted in figure 1.9 While tobacco production originally concentrated itself in the South, growth quickly spread throughout the country. Most smallholders continue to sell their crop through tobacco clubs, typically including 10 to 20 farmers, but recent legal changes allow them to sell directly to the 8 See Chirwa et al. (2008) for additional information on the history of tobacco in Malawi. Zeller, Diagne, and Mataya (1998) provide insights into smallholder tobacco adoption decisions, but subsequent lifting of the governmental entry restrictions may have altered production choices. 9 The original Limbe floor is in Southern Malawi, but a floor opened in the Central region in 1979 and in the Northern region in 1993. 6 floors if so desired. The Central region now houses the busiest tobacco auction floor with the most registered tobacco clubs, although thousands of clubs exist in each of Malawi’s three geographic regions (Jaffee, 2003; Tsonga, 2004). All producers must sell at least one bale of tobacco to be eligible for the auctions, thus farmers on small plots may resort to selling to intermediary buyers at below market prices (Takane, 2008). Malawi’s auction sales have plateaued somewhat over the last decade, with reports of illegal exporting to neighboring countries, lower auction prices due to plastic contamination, and possible collusion amongst tobacco purchasers (Kadzandira et al., 2004). The available empirical studies that explore the welfare effects of household tobacco adoption in Malawi mostly rely on limited data from the 1990s, when the Malawian government restricted smallholder production (Kees van Donge, 2002). Kennedy includes Malawi in cross-country reports depicting the effects of commercialization on child nutrition, and finds insignificant differences in nutritional outcomes between tobacco- and non-tobacco-producing households during times of food price stability (Kennedy, 1994; Kennedy et al., 1992). Masanjala (2006) examines the effect of smallholder Malawian tobacco market liberalization on poverty alleviation. He concludes that tobacco adoption in the mid-1990s increased total household income, with a negative effect on non-farm household income. His study determines that tobacco farming significantly decreased caloric-intake food security, with the majority of the children in his sample being stunted. This paper constitutes a new contribution to the longstanding international and Malawi-specific debate on the links between agricultural commercialization and nutrition. We bring together the cash crop adoption literature with the work on covariate shocks in developing economies for the purpose of examining the effects of smallholder tobacco adoption on child stunting during an exogenous staple food price shock in Malawi. Our research aims to confirm whether households producing tobacco during the food price spike have experienced nutrition distress due to inadequate food stocks and limited market availability of staple goods, resulting in worse child height-for-age outcomes. 3 Data Our primary source of data is the Malawi Second Integrated Household Survey (IHS2) which was implemented from March 2004 to February 2005 by the National Statistical Office (NSO). The 7 IHS2 used a two-stage stratified sample selection process. The primary sampling units (PSU) were the Enumeration areas (EAs), selected from each stratum on the basis of probability proportional to size (PPS). The household population figures used to select the EAs at the first stage originated from the 1998 Population and Housing Census. The 30 IHS2 strata are distributed over 3 regions of the country, namely North, Center, and South, and include all 4 urban centers, namely the cities of Mzuzu, Lilongwe, Blantyre, and Zomba, and 26 administrative districts, which excluded Likoma Island and were all considered rural. In total, 564 EAs were selected for the survey. At the second stage, each selected EA was relisted, and 20 households were randomly selected. The final IHS2 sample included 11,280 households. The IHS2 featured a multi-topic household questionnaire that solicited (i) detailed information on household income, consumption, and agricultural production, and (ii) data on demographics, ed- ucation, health and labor at the individual-level, among other topics. Weight and height information were recorded for children aged between 6-59 months in kilograms and centimeters respectively. In terms of agriculture, the household reported detailed information on cultivation and production for the reference rainy and dry agricultural seasons, and on tobacco production specifically, including a history of tobacco cultivation covering the period of 1999/2000-2003/2004 rainy agricultural seasons.10 On the whole, we limit our analysis to smallholder households, defined to hold arable land of seven acres or less.11 In exploring the effect of household tobacco adoption on child nutrition, we identify for each child whether his/her household was a tobacco grower in the year of or the year after his/her birth. This dummy variable is the main explanatory variable of interest.12 We limit our cash crop production definition to tobacco producers, as Malawian production of other inedible cash crops like cotton or tea pale in comparison to tobacco (World Bank, 2007). Few Malawian farmers monocrop tobacco. Due to food necessities and risk reduction through 10 The retrospective questions on previous tobacco cultivation activities only captured incidence. Hence, with the exception of the reference rainy season, we do not have data on the extent of tobacco cultivation on the farm and the associated production outcomes for the past seasons. 11 The seven acre cutoff covers 95 percent of the farming Malawian household population. Our results are robust to alternative cut-off points. Box-plots underlying our decision for the seven acre threshold and empirical results under different thresholds are available upon request. 12 Our results robust to alternative definitions of household tobacco production, including tobacco production (i) in the year of and the year after child birth, (ii) only in the year of child birth, (iii) only in the year after child birth, or (iv) only during the staple food price shock of interest. 8 diversification, tobacco-farming households often grow additional crops. Almost all of the tobacco producing households also reported growing non-tobacco crops in the current planting season. Regardless, smallholders who planted and tended to labor-intensive tobacco made a significant production decision. Malawian farm households typically experience inadequate food production, with a well-defined hungry season before early green maize is edible for consumption. By directing a portion of their land away from food production, tobacco growers reduced the amount of on-farm food available for their families. In fact, the IHS2 data indicate that with respect to non-tobacco counterparts, tobacco-producing farm households, on average, record lower values for (i) the share of all agricultural plots that is cultivated with maize, and (ii) the share of total farm area that is devoted to maize, and that the differences are statistically significant at least at the 5 percent level. The main outcome of interest in our analysis exploring the nutritional impact of household tobacco production is stunting, a robust long-term measurement of childhood undernutrition that compares a child’s height and age to a globally representative World Health Organization (WHO) reference population (Habicht et al., 1974).13 Abnormally short children are defined as having a z-score of two standard deviations or more below the mean of the WHO population (Waterlow et al., 1977; Dibley et al., 1987). A wide range of literature has demonstrated that malnutrition in general, and height-for-age in particular, affects long-term wage, health and educational attainment opportunities (Strauss and Thomas, 1998; Chang et al., 2010; Gandhi et al., 2011).14 Stunting has been shown to accurately measure long-term childhood nutrition levels (Beaton et al., 1990). In comparison to standard reference populations, stunted children experience greater probabilities of early mortality, decreased physical capabilities and diminished mental capacity (Grantham-McGregor et al., 2007; Fawzi et al., 1997). Over 30 percent of the children under the age of 5 around the world were estimated to be stunted in 2005 (Black et al., 2008). Researchers attribute a large number of childhood deaths in Africa to undernutrition and have called for future Malawian health interventions to focus on stunting (Espo et al., 2007). Recent updates to the WHO’s Child Growth Standards have further emphasized the importance of nutritional deficiencies in newborn rural Malawian children (Prost et al., 2008). 13 Stunting is identifiable within the child’s first year. We use anthropometric measurements for children from six to sixty months of age, which only eliminates one observation from the sample. 14 See Hoddinott et al. (2008) and Maluccio et al. (2009) for discussion of a positive nutritional shock in Guatemala, which is shown to increase long term economic productivity and educational attainment. 9 Following the restriction on smallholder households, the sample includes 5,714 children with height measurements.15 Table 1 presents the sample means and the results from the tests of mean differences by household tobacco adoption status in the year of or the year after child birth.16 The rate of stunting among children born into tobacco-producing households stands at 51 percent, significantly higher than the 43 percent for children in non-tobacco-producing households. The aggregate household income statistics, however, show tobacco households to have more overall income than non-tobacco-producing farm households. On average, tobacco producing households also control more land, have larger families, have younger households heads and are more likely to live in communities without a health clinic. 4 Identification If we assume that systematic differences between children from tobacco-farming households and their comparators from non-tobacco farming households are captured conditional on observables, the coefficient θ from the following linear regression would yield an unbiased estimate of the impact of household tobacco adoption on child stunting: yih = α + θDih + χCih + φHih + ωVih + εih (1) where i denotes individual; h denotes household; y is the height-for-age z-score for children 6 to 59 months old, with lower z-scores implying worse nutrition outcomes; α is the constant; D captures household tobacco adoption as defined above; C , H and V are vectors of child-, household-, and community-specific variables, respectively; and ε is the error term. Previous work on health outcomes provides a wealth of information on significant contributing factors to reducing nutritional deficiencies.17 The range of controls introduced in our specification account for variables that enter the health production function described by Strauss and Thomas 15 Mei and Grummer-Strawn (2007) suggest excluding from analysis observations with height for age measurements that are either below six deviations or above five deviations with respect to the mean. Applying this criterion, we omitted 97 observations, accounting for less than 2 percent of the whole sample. 16 All of the variables originate from the IHS2 data. The income variables are obtained from the Food and Agricultural Organization’s Rural Income Generating Activities (RIGA) database. We elected to use the version of the total income aggregate that calculates the value of own consumption from the food consumption module of the IHS2 household questionnaire. 17 Recent literature questions past results in clarifying the links between health and development (Deaton, 2003). 10 (2007). The model expresses child stunting as a function of prices of consumption and health inputs, along with observed and unobserved variables that influence utility and health, including demographic variables, human capital and background, non-labor income, and disease environment. We do not observe the prices of consumption and health inputs, so we include a rich set of variables that determine health outcomes and that are observed in our survey data. Following the introduction of child-, mother-, household-, and community-level controls, there might remain unobserved child- /household-level attributes that jointly determine the outcome of interest and household tobacco adoption (i.e. ε is correlated with D in equation 1). Consequently, the coefficient θ estimated with an Ordinary Least Squares (OLS) regression would be subject to omitted variable bias that is likely to be positive. One solution to this particular problem is the use of a Two-Stage Least Squares (TSLS) re- gression, where the idea is to isolate the movements in D that are uncorrelated with ε by finding an instrumental variable (IV) that predicts D but exerts no direct impact on the outcome variable. This in turn permits a consistent estimation of the coefficient θ. In the traditional TSLS set up, the regression estimated at each stage is linear. The predicted values for the endogenous variable are obtained in the first stage by regressing D on C , H and V from equation 1, collectively denoted as W , and the instrumental variables, denoted as Z . In the second stage, the outcome variable is regressed on W , and the predicted values of D. The coefficient associated with the predicted value of D is consequently taken as the consistently-estimated version of the coefficient θ from equation 1. In our case, an additional complexity is introduced by the fact that the endogenous variable is binary, such that a linear first stage estimation could produce predicted values that are below 0 or above 1. Newey and McFadden (1984) show that in general estimation models that include generalized method of moments, given endogenous variables, G, and exogenous variables, W , the asymptotically optimal instruments are: Z = Var(W )−1 E(G|W ) (2) where E(G|W ) is the predicted probability from a binary response model such as Probit. In this context Var(W )−1 can be ignored because it is just a scaling factor. Hence, the excluded IVs, 11 namely Z, are used in a first stage Probit model, together with W . The predicted probability of household tobacco production, which is generated from the first stage Probit, is used as the instrument for D in a just-identified linear generalized methods of moments (GMM) second stage estimation. The first stage Probit is expressed as: ∗ Di = Φ(Z β ) + Vi (3) ∗ where Di is a latent variable underlying household tobacco cultivation in the year of or the year after ˆ . In this set up, we do not observe D∗ child birth, and the recovered predicted probabilities are Φ i ∗ ∗ and instead observe Di to be equal to 1 or 0 for Di ≥ 0 or Di < 0, respectively. The second stage estimation accounts for heterogeneity of the variance across the sample, by using the inverse of the estimated covariance matrix clustered at the EA level as the GMM weight matrix. The estimates solve the exactly identified GMM criterion function: −1 ˆ min( Zi Ui ) ( Zi Ui ) (4) β i i 4.1 Instrumental Variables For TSLS regressions to work, each IV must satisfy two conditions, namely instrumental relevance and instrumental exogeneity. If an instrument is relevant, then the variation in the instrument is related to the variation in the instrumented variable; as such the IV should assume in the first stage regression a highly statistically significant coefficient with a sign that is in line with the theoretical relationship between the instrument and the instrumented variable. In addition, the IV must satisfy the exclusion restriction, i.e. it must be uncorrelated with the outcome variable. Although the claims regarding instrumental relevance and exogeneity should be assessed on theoretical merits, test results regarding the strength and the validity of the IVs will be reported in section 5 to support the theoretical arguments made here. The selection of the IVs is in part underlined by the following reasoning. Maize is the traditional crop in Malawi, and tobacco is a new crop that smallholders consider for adoption. Most crop adoption models, like Eckstein (1984), make the decision forward-looking because the benefit of a new crop is the revenue streams that it will generate. Smallholder expectations of future revenue 12 streams are based on past values of observed variables that are likely to be correlated with crop revenues, but not necessarily in line with their production capabilities. The IVs that we rely on originated from outside the IHS2 database, and are namely (i) the tobacco-maize relative land suitability index for the 9x9 km grid that each respective IHS2 EA is associated with and that stems from the Global Agro-Ecological Zones (GAEZ) database18 , and (ii) the 1997/1998 district-level estimate of the number of tobacco farmers originating from the First Integrated Household Survey (IHS1) data. Both IVs are mapped in figure 2. Our first IV, the GAEZ tobacco-maize relative land suitability index, is constructed as the difference between tobacco and maize land suitability indices. The suitability index is modeled using data on average climate conditions, soil and terrain characteristics and an assumption of farm management type. Input climate parameters derived from rainfall and temperature time-series include the timing, length and quality of the growing period as well as water availability. Soil characteristics relevant to plant production, including nutrient availability and retention capacity, rooting conditions, oxygen availability, salinity and toxicity, and management constraints are derived from the Harmonized World Soil Database. Terrain variables, elevation and degree of slope or slope class, are incorporated as limiting factors on production of different crops, as well as expected yield reduction factors due to erosion. All input layers are combined with an assessment of crop- specific requirements to produce estimates of land suitability for tobacco and maize production, on a continuous scale from 0 to 100. For the purposes of this paper, intermediate input level, rain-fed tobacco and maize production is assumed.19 While we use an average suitability index value derived for each IHS2 EA, the GAEZ model outputs are produced at a spatial resolution of 5 arc minutes (approximately 9x9 km), corresponding to an area larger than most EAs. As a result, although the GAEZ suitability index clearly has sub- district specificity, it may not capture all variation at or within EA-level. It is also important to note that the suitability index is based on average climate conditions calculated over a baseline period of 1960-1990. Inter-annual climate variability is incorporated to some extent, as a yield-reduction factor, but recent climate trends and prior year conditions are not reflected in this IV. Changes over 18 For additional information on the GAEZ dataset, see gaez.fao.org. Following the GAEZ instruction, we applied a 0.01 scalar to the suitability index values. 19 The intermediate input level assumes an improved management system. Production is for subsistence as well as commercial sale through manual labor with some mechanization. Improved seeds and some fertilizer and pesticides are also used. Our results are robust, however, to the use of suitability index values derived under low-input scenarios. 13 time, in weather, soil fertility, or technology, would likely also have an effect on tobacco adoption. Exploring these relationships in depth would require data of a higher spatial resolution and greater frequency than is currently widely available. However, despite the constraints associated with spatial and temporal resolution, the tobacco- maize relative land suitability index is a composite variable that is envisioned to capture the local knowledge acquired by farmers through years of experience working their land. We expect this understanding of the potential and limitations of the physical environment to be an important factor in tobacco adoption. The variation in the tobacco-maize relative land suitability index is, therefore, expected to positively predict household tobacco adoption, fulfilling in principle the requirement of instrumental relevance. Moreover, the tobacco-maize relative land suitability index is defined based on average climate conditions for the period of 1960-1990, avoiding direct links to children’s nutritional outcomes obtained during the IHS2. Since this IV is also uniformly dependent on exogenous factors, and does not necessarily reflect differences in genetic factors, skills or other unobserved child-/household-level attributes, we claim that it does not influence the outcome variable beyond its impact through tobacco adoption, and thus satisfies the exclusion restriction. Our second IV, the IHS1 district-level estimate of tobacco farmers, is based on the premise that farmers are likely to grow commercialized crops after seeing others in their neighborhood adopt. The estimate is predetermined so it will not be correlated with time-variant unobserved variables. In this respect, the variable is expected to predict household tobacco adoption in a statistically significantly fashion, fulfilling the requirement of instrument relevance. Since the IHS1 information precedes the births of the IHS2 sample children that underlie our analysis, the IV avoids any direct links to children’s current nutritional health and also satisfies the instrumental exogeneity requirement. 4.2 Identification through Staple Food Price Shocks As argued above, our instruments are defined at a higher level than the household as well as the community, and should not be related to the outcome of interest beyond their association with the household tobacco adoption. We address concerns regarding any remaining omitted variable bias linked to within-district unobserved heterogeneity by identifying the causal impact of interest as the difference across the tobacco adoption effects informed by two unique samples of children that 14 differ in terms of exposure to a recent, exogenous domestic staple food price shock. More specifically, the agricultural season in Malawi spans two calendar years, lasting anytime from November to July, depending on the spatial differences in the timing of the rains as well as the type of crops cultivated. The harvest season typically covers the period of March to July, again depending on the same factors underlying the definition of the overall agricultural season. August 2001 marks the approximate beginning of the post-harvest period following the 2000/2001 agricultural season, during which unfavorable rainfall outcomes underlined a 35 percent drop in maize production compared to the 1999/2000 season (Dorward et al., 2008). The effects of poor harvest were compounded in the post-harvest period by mismanagement/premature exports of strategic grain reserves, import delays, and slow donor response (Devereux, 2002; Minot, 2010). Poor rains, combined with a 44 percent reduction in the amount of fertilizer disbursed under the government’s subsidy program with respect to the prior season, led to even lower levels of maize harvest tied to the subsequent 2001/2002 agricultural season (Dorward et al., 2008). Historical rainfall patterns, depicted in figure 3, demonstrate the back-to-back agricultural seasons with unfavorable rainfall outcomes with respect to the long-term average.20 These trends are also supported by depressed annual maize production and maize yield estimates associated with the 2000/2001 and 2001/2002 agricultural seasons, as shown in figure 4. From the available monthly data on maize prices portrayed in figure 4, we observe that the heightened prices were sustained from August 2001 to March 2003, which would mark the start of the 2002/2003 agricultural season harvest period. The period of August 2001-March 2003 is in turn assumed to underlie the exogenous food price shock of interest.21 20 Rainfall summaries for figures 3 and 2 were produced from unbiased RFE v 2.0, using GeoWRSI v 3. GeoWRSI v 3 is stand-alone software implemented by the USGS for the FEWS NET Activity, and distributed by the Climate Hazards Group at University of California, Santa Barbara. During the 2000/2001 and the 2001/2002 rainy seasons, 28 and 70 percent of the pixels across the country had a rain shortfall of at least 5 percent with respect to the pixel-specific long term seasonal average, respectively. The comparable statistics at the national-level for the severe rain shortfall of at least 25 percent with respect to the pixel-specific long term seasonal average stand at 6 and 39 percent for the 2000/2001 and the 2001/2002 rainy seasons, respectively. The rainfall-related statistics for the 2000/2001 season, however, masks other forms of abnormal rainfall outcomes that had materialized, including late, heavy rains in February-March 2001 leading to localized flooding and waterlogging of fields (Dorward et al., 2008; Devereux, 2002). During that season, the official maize production estimates were revised downwards three times, disguising the severity of the production shortfall to both consumers and assistance organizations (Devereux, 2002). 21 Figure 5 reports reports nominal average maize prices computed across the data from markets in Chitipa, Karonga, Lilongwe, Nkhata Bay, Rumphi, Mzuzu, Mitundu, and Lunzu. The dynamics are consistent with the nominal and real price trends reported in World Bank (2007). While maize prices varied throughout Malawi, Hartwig and Grimm (2011) characterize this specific food crisis to be one of the worst in recent Malawian history. Dorward and Kydd (2004) provide a comprehensive discussion regarding the necessary changes to the market intervention and liberalization 15 Following the definition of the shock period, we move on to explain what is meant by “exposure.” While widely accepted that children under the age of 2 are greatly influenced by undernutrition, less attention has been focused on nutritional shocks to intrauterine fetuses. Barker (1997) clarifies the link between intrauterine undernutrition and negative health outcomes. He specifically focuses on undernutrition during pregnancy, suggesting that maternal health during the fetal stage strongly influences the growth and development potential for children later in life. Shrimpton et al. (2001) mainly emphasized the influence of negative intrauterine nutritional statuses on weight outcomes. More recent research has also demonstrated the strong negative effects of fetal undernutrition on children’s height measurements (Victora et al., 2010). Nutritional studies have identified an early development window, from the point of conception to around the child’s second birthday, which is the primary determinant of children’s nutritional status (Victora et al., 2008). Our intrauterine period is defined as 8 months before the child was born, and the early development window as the child’s intrauterine period plus his/her first two years of life. We use the early development window concept to differentiate children exposed to the shock versus those who were not exposed to the shock during this critical stage of growth. This means that at the start of the shock period in August 2001, children born in September 1999 had one month of exposure to the shock within their early development window. Since the shock lasted until March 2003, children had up to 19 potential months of exposure to the shock within their early development window. As the maize prices receded to more typical levels after March 2003, children born between March and October 2003 only experienced the shock intrauterinally, while those born after October 2003 were not exposed to it.22 Given what is meant by the shock period and the child exposure to the shock, our identification strategy identifies the causal effect of household tobacco adoption on child stunting by estimating a TSLS regression using only the sample children that were not exposed to the shock during their early development window, and differencing out the resulting coefficient of interest from the coefficient that is obtained from a TSLS regression using only the sample children that were exposed to the shock during their early development window.23 The common unobserved factors policies to avert similar crises in the future. 22 The estimation results are robust to longer and shorter definitions of the price shock group. For a more detailed look at the price trends in Malawian staples food in general, and maize in particular see Minot (2010). 23 We later test the robustness of our results by separating the shock exposed group into (i) children with a year or less of exposure and (ii) children with more than 1 year of exposure. 16 that influence the effect of household tobacco adoption during the shock period should also be present in the period of price stability. The difference in coefficients produces a “difference in difference” type identification. Hence, the fortified approach of comparing TSLS estimates obtained from by two unique samples of children that differ in terms of exposure to the shock of interest should help remove any remaining within-district unobserved heterogeneity that might otherwise plague our estimates. Further examining household summary statistics from an exposure to shock perspective, table 2 hints at the influence of cash crop production during times of food price shocks. Tobacco-producing households are more dependent on the price of nutrition than maize-producing households. Periods of short supply are likely to increase the price of nutrition significantly. 5 Results Table 3 presents the overall OLS and TSLS estimates of the effect of household tobacco adoption in the year of or the year after child birth.24 The OLS estimate confirms the descriptive relationship previously documented by World Bank (2007). The TSLS estimate of the overall effect indicates the existence of positive correlation between D and ε in equation 1 (i.e. positive omitted variable bias) that would have otherwise plagued the main coefficient of interest. The negative, TSLS-informed overall relationship between the endogenous variable and the outcome is seen in a different light in table 4 when the sample is split in accordance with the shock within the early development window and the regression is estimated separately for each sub-sample. The results in table 4 are central to our fortified identification strategy that relies on TSLS in the context of child exposure to an exogenous staple food price shock. We find that the tobacco coefficient of interest, while negative, is no longer statistically significant among children that did not experience the shock. Conversely, children who experienced the shock within their early development window and originated from households producing tobacco in the year of or the year after child birth were significantly more likely to be stunted. Given our theoretical arguments in section 4.2, and the differences in the tobacco coefficients reported in table 4, the average causal effect of household tobacco adoption on child height-for-age z-score during staple food price spikes is estimated at -1.27, which is statistically significant at the 1 percent level and implies a 70 percent 24 In tables 3 and 4, the standard errors are clustered at the EA-level and the regressions are weighted using the IHS2 household sampling weights. The results are consistent with or without the use of the sampling weights. 17 increase in stunting. Furthermore, the results from the statistical tests that are reported in tables 3 and 4, and that are used in gauging the strength and the exogeneity of our IVs lend support to the theoretical arguments presented earlier in section 4.1.25 The coefficients from the first stage regressions and the Cragg-Donald Wald F statistics from the TSLS estimations discussed above indicate that the direction of the relationship between the IVs and the endogenous variable is as expected and that the IVs are sufficiently strong to avoid weak IV bias. Since an over-identification test could not be run in our preferred, just-identified model that uses Probit-based predicted probability as the IV for the dummy endogenous variable in a TSLS framework, we provide the over-identification test results from the traditional TSLS estimations in which our two IVs are directly included as excluded instruments. We fail to reject the null hypothesis that the instruments are orthogonal to the error term, and this result, which holds true regardless of the child sample of interest, is in favor of the theoretical arguments concerning the exogeneity of the IVs. Conceptually, our findings highlight some of the conditions that may constrain children from reaching their full mental and physical capabilities, and support the view that commercialized farmers may face a distinct set of disadvantages during food price spikes and in the context of non-separable production and consumption decisions at the household level. Children that originated from tobacco-producing households that were exposed to the shock within their early development window could have suffered from undernutrition with enduring long-term effects on health and educational outcomes (Field, Robles, and Torero, 2009; Almond and Currie, 2011). A few results associated with the control variables in table 4 are also worthwhile to mention. Male children, on average, fared significantly worse than females in terms of stunting. Children born into higher-income households showed a general positive trend in height-for-age z-scores, although some of the insignificant differences lend weight to the idea that household bargaining might be influencing purchasing decisions.26 While limited to the sample of children that were exposed to the shock, child access to bed nets, improved housing conditions (proxied by the housing index)27 , and larger arable land holdings are associated with higher height-for-age z-scores as well. 25 See Stock and Yogo (2005) for additional information on these tests. 26 See Zere and McIntyre (2003) for further inquiry into the effects of income distribution on nutritional outcomes 27 The housing index was constructed using principal components analysis. It takes into account: (i) whether the dwelling unit is a permanent structure, (ii) whether the dwelling has a permanent floor, (iii) whether the household has access to a protected water source, (iv) whether the dwelling has a toilet, and (v) the number of rooms per household 18 Finally, to further substantiate our findings, we split the shock-exposed sample to identify those who were exposed for a year or less and those who were exposed 13 to 19 months.28 The story behind the variable of interest remains identical to the previous results. The variable of interest remains negative and statistically significant at the 5 percent level among the children that were exposed to the shock within their early development window for a year or less as well those whose period of exposure was 13 to 19 months.29 As previously shown, the statistically insignificant coefficient of interest for children not exposed to the shock stands in stark contrast with respect to the negative and statistically significant tobacco coefficients discovered within the shock-exposed group.30 6 Conclusion The move from correlation to identification by removing the confounding changes in unobserved variables is an important step in identifying the reasons behind disparities in the effects of food price increases on the Malawian population. Recent work linking economic development and nutritional health has shown the importance of accounting for undernutrition when evaluating the effectiveness of growth plans (Ecker et al., 2011; Headey, 2011). Our results expand on previous literature that explored theoretical reasons for negative health outcomes associated with smallholder agricultural commercialization (von Braun, Kennedy, and Bouis, 1990). Applying these ideas to an actual food price spike, the large negative tobacco coefficients show cash-crop-producing households to be disproportionately affected by the shock, with children significantly more likely to be stunted than their non-tobacco-producing counterparts. Future extensions of this work may examine behavioral or expenditure differences in households by production type. Uncovering possible intra-household allocation constraints hinted at through both this research and anecdotal evidence in the literature would help to explain why adoption decisions influence nutritional outcomes. Applying the work of Lundberg, Pollak, and Wales (1997) member. 28 As the shock lasted for 19 months, it is not possible for children to be exposed to the shock for longer than this amount of time. 29 In the spirit of brevity, we chose to exclude these results, but they are available upon request. The tobacco coefficient was estimated at -1.49 and -1.12 for children with periods of exposure 1-12 and 13-19 months, respectively. 30 The sample restrictions unfortunately prevent us from exploring the heterogeneity of impact of shock exposure beyond the sub-samples informing the table 5 estimates. 19 and Thomas (1990) to the Malawian context may produce fruitful insights into the household expenditure decision-making process. The next round of the Malawian Integrated Household Survey will provide additional information on production and expenditure decisions, allowing for better comprehension of cash cropping choices in the face of food price increases. The cash crop adoption results demonstrate an important vulnerability in developing world agriculture, which is particularly pertinent with the increasing global commercialization of agri- culture. Identifying the individuals most affected by the volatility in food prices is the first step toward remedying their situation. As food markets continue to generally increase, acknowledgment of previous disparities in nutritional outcomes allows for increased effectiveness in targeting the neediest for food aid during times of future food price shocks. Increasing international trade provides overwhelming benefits to many sectors of the global economy. Understanding and accounting for the potential pitfalls associated with agricultural commercialization will only strengthen the movement toward decreasing hunger. This paper helps to pinpoint the determinants behind negative health outcomes for commercialization, and continues the research on the effects of food price shocks on the developing world. Future work, with panel datasets, international food price spikes and more specified price data will allow for an even greater understanding of how food price increases influence the health outcomes of children around the world. 20 References Akresh, R., P. Verwimp, and T. Bundervoet (2007). Civil war, crop failure, and child stunting in Rwanda. Technical Report 4208, World Bank Policy Research Working Paper. 5 Almond, D. and J. Currie (2011). Killing me softly: The fetal origins hypothesis. Journal of Economic Perspectives 25(3), 153–172. 18 Barham, B., M. Carter, and W. Sigelko (1995). Agro-export production and peasant land access: Examining the dynamic between adoption and accumulation. Journal of Development Economics 46(1), 85–107. 4 Barker, D. (1997, Sep). Maternal nutrition, fetal nutrition, and disease in later life. Nutrition 13(9), 807–813. 16 Beaton, G., A. Kelly, J. Kevany, R. Martorell, and J. Mason (1990, December). Appropriate uses of anthropometric indices in children. Nutrition policy discussion paper 7, United Nations Administrative Committee on Coordination- Subcommittee on Nutrition. 9 Birthal, P., P. Joshi, and A. Gulati (2005). Vertical coordination in high-value food commodities: Implications for smallholders. Mtid discussion paper no. 85, International Food Policy Research Institutie. 4 Black, R., L. Allen, Z. Bhutta, L. Caulfield, M. de Onis, M. Ezzati, C. Mathers, and J. Rivera (2008, January). Maternal and child undernutrition: global and regional exposures and health consequences. The Lancet 371(9608), 243 – 260. 9 Bouis, H. and L. Haddad (1990). Effects of agricultural commercialization on land tenure, household resource allocation, and nutrition in the Philippines. Research report no. 79, International Food Policy Research Institutie. 4 Carletto, C., T. Kilic, and A. Kirk (2011). Nontraditional crops, traditional constraints: The long-term welfare impacts of export crop adpotion among guatemalan smallholders. Agricultural Economics 42, supplement 61–75. 2, 5 Carter, M. and J. Maluccio (2002, December). Social capital and coping with economic shocks: An analysis of stunting in South African children. Technical Report FCND discussion paper No. 142, International Food Policy Research Institute. 5 Chang, S., S. Walker, S. Grantham-McGregor, and C. Powell (2010). Early childhood stunting and later fine motor abilities. Developmental Medicine & Child Neurology 52, 831–836. 9 Chirwa, E., I. Kumwenda, C. Jumbe, P. Chilonda, and I. Minde (2008, October). Agricultural growth and poverty reduction in Malawi: Past performance and recent trends. ReSAKSS Working Paper 8, Regional Strategic Analysis and Knowledge Support Group: Sourthern Africa. 6 Deaton, A. (2003, March). Health, inequality, and economic development. Journal of Economic Literature 41(1), 113–158. 10 Devereux, S. (2002). The Malawi famine of 2002. IDS Bulletin 33(4), 70–78. 15 Dibley, M., J. Goldsby, N. Staehling, and F. Trowbridge (1987). Development of normalized curves for the international growth reference: Historical and technical considerations. The American Journal of Clinical Nutrition 46, 736–748. 9 Dorward, A., E. Chirwa, V. Kelly, T. Jayne, R. Slater, and D. Boughton (2008, March). Evaluation of the 2006/7 agricultural input subsidy programme, Malawi. Final report, School of Oriental and African Studies and Wadonda Consult and Michigan State University and Overseas Development Institute. 15 Dorward, A. and J. Kydd (2004, September). The Malawi 2002 food crisis: The rural development challenge. The Journal of Modern African Studies 42(3), 343–361. 15 21 Ecker, O., C. Breisinger, and K. Pauw (2011, February). Growth is good, but is not enough to improve nutrition. 2020 Conference Paper 7, International Food Policy Research Institute. 3, 19 Eckstein, Z. (1984). A rational expectations model of agricultural supply. The Journal of Political Economy 92(1), 1–19. 12 Espo, M., T. Kulmala, K. Maleta, T. Cullinan, M.-L. Salin, and P. Ashorn (2007). Determinants of linear growth and predictors of severe stunting during infancy in rural Malawi. Acta Paediatrica 91(12), 1364–1370. 9 Fawzi, W., M. G. Herrera, D. Spiegelman, A. E. Amin, P. Nestel, and K. Mohamed (1997). A prospective study of malnutrition in relation to child mortality in Sudan. American Journal of Clinical Nutrition 65, 1065–9. 9 Field, E., O. Robles, and M. Torero (2009). Iodine deficiency and schooling attainment in Tanzania. American Economic Journal: Applied Economics 1(4), 140–169. 18 Gandhi, M., P. Ashorn, K. Maleta, T. Teivaanmäki, X. Duan, and Y. B. Cheung (2011). Height gain during early childhood is an important predictor of schooling and mathematics ability outcomes. Acta Paediatrica 100, 1113–1118. 9 Grantham-McGregor, S., Y. B. Cheung, S. Cueto, P. Glewwe, L. Richter, B. Strupp, and The International Child Development Steering Group (2007). Developmental potential in the first 5 years for children in developing countries. Lancet 369(9555), 60–70. 9 Habicht, J.-P., C. Yarbrough, R. Martorell, R. Malina, and R. Klein (1974, April). Height and weight standards for preschool children: How relevant are ethnic differences in growth potential? Lancet 1(7858), 611–614. 9 Harrigan, J. (2008). Food insecurity, poverty and the Malawian starter pack: Fresh start or false start? Food Policy 33, 237–249. 2 Hartwig, R. and M. Grimm (2011). An assessment of the effects of the 2002 food crisis on children’s health in Malawi. forthcoming in Journal of African Economies. 15 Headey, D. (2011, February). Turning economic growth into nutrition-sensitive growth. 2020 Conference Paper 6, International Food Policy Research Institute. 3, 19 Hoddinott, J. (2006, February). Shocks and their consequences across and within households in rural Zimbabwe. Journal of Development Studies 42(2), 301–321. 5 Hoddinott, J., J. Maluccio, J. Behrman, R. Flores, and R. Martorell (2008). Effect of a nutrition intervention during early childhood on economic productivity in Guatemalan adults. The Lancet 371, 411–416. 9 Immink, M. and J. Alarcon (1993, January). Household income, food availability, and commercial crop production by smallholder farmers in the Western highlands of Guatemala. Economic Development and Cultural Change 41(2), 319–42. 5 Immink, M., E. Payongayong, E. Kennedy, and R. Sibrian (1995). Export vegetable crops and poverty alleviation for smallholder farm households: A case study from Guatemala. In Poverty Alleviation through International Trade. UNCTAD. Presented in Santiago, Chile. 5 Jaffee, S. (2003, June). Malawi’s tobacco sector: Standing on one strong leg is better than on none. Africa Region Working Paper 55, 1–58. World Bank Working Paper Series. 6, 7 Johnston, B. and J. Mellor (1961). The role of agriculture in economic development. The American Economic Review 51(4), 566–593. 4 Kadzandira, J., H. Phiri, and B. Zakeyo (2004). Malawi tobacco sector performance audit: The perceptions and views of smallholder tobacco farmers on the state of play in the tobacco sector. Technical report, The World Bank. 6, 7 22 Katz, E. (1994). The impact of non-traditional export agriculture on income and food availability in Guatemala: An intra-household perspective. Food and Nutrition Bulletin 15(4), 295–302. United Nations University Press. 4 Kees van Donge, J. (2002, March). Disordering the market: The liberalisation of burley tobacco in Malawi in the 1990s. Journal of Southern African Studies 28(1), 89–115. Special Issue: Malawi. 7 Kennedy, E. (1989). The effects of sugarcane production on food security, health, and nutrition in Kenya: A longitudinal analysis. Research report no. 78, International Food Policy Research Institute. 5 Kennedy, E. (1994). Agricultural commercialization, economic development and nutrition, Chapter Health and Nutrition Effects of Commercialization of Agriculture, pp. 79–99. International Food Policy Research Institute. 7 Kennedy, E., H. Bouis, and J. von Braun (1992). Health and nutrition effects of cash crop production in developing countries: A comparative analysis. Social Science & Medicine 35(5), 689–697. 3, 7 Kennedy, E. and B. Cogill (1987, November). Income and nutritional effects of the commercialization of agriculture in southwestern Kenya. Research Report 63, International Food Policy Research Institute. 4 Lea, N. and L. Hanmer (2009, October). Constraints to growth in Malawi. World Bank Policy Research Working Paper 5097, World Bank. 6 Levy, S., C. Barahona, and B. Chinsinga (2004, September). Food security, social protection, growth and poverty reduction synergies: The starter pack programme in Malawi. Technical report, The Overseas Development Institute Natural Resource Perspectives. Number 95. 3 Ligon, E. and E. Sadoulet (2007). Estimating the effects of aggregate agricultural growth on the distribution of expenditures. Technical report, World Bank. Background paper for the World Development Report 2008. 2 Lundberg, S., R. Pollak, and T. Wales (1997, Summer). Do husbands and wives pool their resources? Evidence from the United Kingdom child benefit. The Journal of Human Resources 32(3), 463–480. 19 Maertens, M. and J. Swinnen (2009, January). Trade, standards, and poverty: Evidence from Senegal. World Development 37 (1), 161–178. 4 Maluccio, J., J. Hoddinott, J. Behrman, R. Martorell, A. Quisumbing, and A. Stein (2009, April). The impact of improving nutrition during early childhood on education among Guatemalan adults. The Economic Journal 119, 734–763. 9 Masanjala, W. H. (2006). Cash crop liberalization and poverty alleviation in Africa: Evidence from Malawi. Agricultural Economics 35, 231–240. 7 Mei, Z. and L. Grummer-Strawn (2007). Standard deviation of anthropometric z-scores as a data quality assessment tool using the 2006 WHO growth standards: A cross country analysis. Bulletin of the World Health Organization 85(6), 441–448. 10 Minot, N. (2010, January). Staple food prices in Malawi. In Food Security Collaborative Working Papers, AAMP. 15, 16 Minten, B., L. Randrianarison, and J. Swinnen (2007, September-November). Spillovers from high-value agriculture for exports on land use in developing countries: Evidence from Madagascar. Agricultural Economics 37 (2-3), 265–75. 4 Newey, W. and D. McFadden (1984). Handbook of Econometrics, Chapter Large Sample Estimation and Hypothesis Testing, pp. 2112–2245. Elsevier Science. 11 NSO (2012, September). Household socio-economic characteristics report. Technical report, Malawian National Statistics Office. 5 23 Orr, A. (2000). ‘Green gold’?: Burley tobacco, smallholder agriculture, and poverty alleviation in Malawi. World Development 28(2), 347–363. 6 Pelletier, D., K. Deneke, Y. Kidane, B. Haile, and F. Negussie (1995). The food-first nias and nutrition policy: Lessons from Ethiopia. Food Policy 20(4), 279–298. 3 Peters, P. E. (1996). Failed magic or social context? Market liberalisation and the rural poor in Malawi. Technical report, Harvard Institute for International Development. 6 Prost, M.-A., A. Jahn, S. Floyd, H. Mvula, E. Mwaiyeghele, V. Mwinuka, T. Mhango, A. Crampin, N. McGrath, P. Fine, and J. Glynn (2008, July). Implication of new WHO growth standards on identification of risk factors and estimated prevalence of malnutrition in rural Malawian infants. PLoS ONE 3(7), e2684. 9 Republic of Malawi (2000, August). Interim poverty reduction and growth strategy paper - A road map. Technical report, Republic of Malawi. 6 Sahn, D. (1990). The impact of export crop production on nutritional status in Côte d’Ivoire. World Development 18(12), 1635–1653. 4 Sahn, D. and J. Arulpragasam (1991, June). The stagnation of smallholder agriculture in Malawi: A decade of structural adjustment. Food Policy 16(3), 219–234. 3 Shrimpton, R., C. Victora, M. de Onis, R. C. Lima, M. Blössner, and G. Clugston (2001, May). Worldwide timing of growth faltering: Implications for nutritional interventions. Pediatrics 5(5), e75. 16 Stock, J. and M. Yogo (2005). Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg, Chapter Testing for Weak Instruments in Linear IV Regression, pp. 80–108. Cambridge University Press. Editors Donald Andrews and James Stock. 18 Strauss, J. and D. Thomas (1998, June). Health, nutrition, and economic development. Journal of Economic Literature 36, 766–817. 9 Strauss, J. A. and D. Thomas (2007). Handbook of Development Economics, Chapter 54 – Health over the life course, pp. 3375 – 3474. Elsevier. 10 Takane, T. (2008). African rural livelihoods under stress: Economic liberalization and smallholder farmers in Malawi. Institute of Developing Economics, JETRO. 6, 7 Thomas, D. (1990). Intra-household resource allocation: An inferential approach. The Journal of Human Re- sources 25(4), 635–664. 20 Tobin, R. and W. Knausenberger (1998). Dilemmas of development: Burley tobacco, the environment and economic growith in Malawi. Journal of Southern African Studies 24(2), 405–424. 6 Tsonga, E. (2004, November). An analysis of the performance of Malawi’s tobacco production and exports. Technical report, Malawian Ministry of Agriculture, Irrigation and Food Security. 6, 7 United Nations Development Programme (2009). Human development report 2009. Technical report, United Nations. 6 Victora, C. G., L. Adair, C. Fall, P. C. Hallal, R. Martorell, L. Richter, H. S. Sachdev, and Maternal and Child Undernutrition Study Group (2008, Jan). Maternal and child undernutrition: Consequences for adult health and human capital. Lancet 371(9609), 340–357. 3, 16 Victora, C. G., M. de Onis, P. C. Hallal, M. Blössner, and R. Shrimpton (2010, March). Worldwide timing of growth faltering: Revisiting implications for interventions. Pediatrics 125(3), e473–e480. 16 von Braun, J. (1995). Agricultural commercialization: Impacts on income and nutrition and implications for policy. Food Policy 20(3), 187–202. 3 24 von Braun, J., H. de Haen, and J. Blanken (1991). Commercialization of agriculture under population pressure: Effects on production, consumption, and nutrition in Rwanda. Research report no. 85, International Food Policy Research Institute. 5 von Braun, J., D. Hotchkiss, and M. Immink (1989). Nontraditional export cropsin Guatemala: Effects on production, income, and nutrition. Technical Report Research Report No. 73, International Food Policy Research Institute. 4 von Braun, J., E. Kennedy, and H. Bouis (1990, February). Commercialization of smallholder agriculture: Policy requirements for the malnourished poor. Food Policy 15(1), 82–85. 19 von Braun, J., D. Puetz, and P. Webb (1989). Irrigation technology and commercialization of rice in The Gambia: Effects on income and nutrition. Technical Report Research Report No. 75, International Food Policy Research Institute. 4 Waterlow, J. C., R. Buzina, W. Keller, J. M. Land, M. Z. Nichaman, and J. M. Tanner (1977). The presentation and use of height and weight data for comparing the nutritional status of groups of children under the age of 10 years. Bulletin of the World Health Organization 55(4), 489–498. 9 World Bank (1989). Sub-Saharan Africa from crisis to sustainable growth: A long-term perspective study. World Bank. 6 World Bank (2007, December). Malawi poverty and vulnerability assessment: Investing in our future. Full Report 36546-MW, World Bank. 3, 8, 15, 17 World Bank (2009, July). Gross national income per capita 2008, Atlas method and PPP. Technical report, World Bank. 6 Zeller, M., A. Diagne, and C. Mataya (1998). Market access by smallholder farmers in Malawi: implications for technology adoption, agricultural productivity and crop income. Agricultural Economics 19, 219–229. 6 Zere, E. and D. McIntyre (2003, September). Inequities in under-five child malnutrition in South Africa. International Journal for Equity in Health 2(7), online. 18 25 Figure 1: Malawi by Region, Traditional Authorities and Tobacco Floors Mzuzu k Lilongwe k Limbe k k Tobacco auction floor Regional boundary Traditional Authorities 26 Figure 2: 1997/1998 District-Level Estimates of Tobacco Producers and Estimates of Relative Tobacco-Maize Land Suitability Indices Number of Tobacco Tobacco Farmers countries Suitability Index 0 - 1,000 0 - 20 1,001 - 10,000 20 - 40 10,001 - 50,000 40 - 60 50,001 - 100,000 60 - 80 100,001 - 118,376 80 - 100 Maize Tobacco Suitability countries Suitability Index countries Relative to Maize 0 - 20 Lowest 20 - 40 40 - 60 60 - 80 80 - 100 Highest 27 Figure 3: 2000/2001 and 2001/2002 Agricultural Season Rainfall Outcomes in the Context of Long-Term Average Rainfall November November 2000 November 2001 - July - July 2001 - July 2002 Percent of LTAR < 75% 76% - 95% 96% - 105% 106% - 125% > 125% 28 Long-Term Average Rainfall (LTAR, mm) 718 - 900 901 - 1,000 1,001 - 1,200 1,201 - 1,400 1,401 - 1,660 Figure 4: Annual Maize Production Estimates (1996-2005) 29 Figure 5: Monthly Average Maize Grain Prices (1996-2005) 30 Table 1: Sample Means and Results from Tests of Mean Differences by Household Tobacco Production Status Entire Sample All Non-Tobacco Tobacco Difference Instruments: Tobacco-Maize Land Suitability Index −1.30 −2.21 2.74 −4.95∗∗∗ 1997/98 Tobacco Farmers by District (100s) 392.30 339.06 627.40 −288.3417∗∗∗ Child: Height for age zscore −1.84 −1.81 −2.01 0.21∗∗∗ Stunted† 0.45 0.43 0.51 −0.08∗∗∗ Male† 0.49 0.50 0.46 0.04∗∗ Age (Months) 31.6 31.5 32.1 −0.60 Age (Months) Squared 1242.6 1235.5 1273.8 38.30 Bet Nets for Children† 0.36 0.37 0.35 0.02 Mother: Did not Complete Primary‡ 0.79 0.81 0.79 0.02 Completed Primary‡ 0.09 0.08 0.12 −0.04∗∗∗ Above Primary‡ 0.11 0.11 0.10 1.00 Household: Head of Household: Age (Years) 39.2 39.5 37.8 1.77∗∗∗ Household Dependency Ratio 1.5 1.5 1.4 0.00 31 Household Size 5.7 5.7 6.0 −0.27∗∗∗ Housing Index (PCA) 0.0 0.0 −0.2 0.24∗∗∗ Household Asset Income 23388.4 22938.5 25374.9 −2436.4∗ Income Quintile 1‡ 0.20 0.22 0.13 0.09∗∗∗ Income Quintile 2‡ 0.20 0.21 0.18 0.02 Income Quintile 3‡ 0.20 0.20 0.20 0.01 Income Quintile 4‡ 0.20 0.18 0.26 −0.08∗∗∗ Income Quintile 5‡ 0.19 0.19 0.23 0.04∗∗ Land Holding (Acres) 2.4 2.2 3.1 −0.94∗∗∗ Community: Distance to Closest ADMARC (KMs) 7.9 8.0 7.8 0.10 Distance to Closest Auction Floor (KMs) 72.1 71.7 73.8 2.10 Health Clinic Present† 0.30 0.31 0.22 0.10∗∗∗ Southern Region‡ 0.46 0.51 0.21 0.31∗∗∗ Central Region‡ 0.42 0.37 0.65 0.28∗∗∗ Northern Region‡ 0.12 0.11 0.14 −0.02 Observations 5714 4673 1041 1 ***, **, * indicate statistical significance at the 1, 5, and 10 percent level, respectively, with summary statistics weighted in accordance with the complex survey design. 2 Dummy variables are denoted by the † symbol and categorical variables by the ‡ symbol. 3 Annual household income stems from the income aggregate created under the Food and Agriculture Organization of the United Nations (FAO) Rural Income Generating Activities (RIGA) project. It is expressed in terms of United States Dollar, using the average exchange rate from the IHS2 implementation period (U S $1 = M K 108.714). 4 ADMARC is a Malawian parastatal organization mandated to market agricultural produce and inputs. Table 2: Sample Means and Mean Differences Tests by Household Tobacco Production Status with Shock Status Stable Group Sample Shocked Group Sample Non-Tobacco Tobacco Difference Non-Tobacco Tobacco Difference Instruments: Tobacco-Maize Land Suitability Index −1.93 2.65 −4.58∗∗∗ −2.25 2.76 −5.00∗∗∗ 1997/98 Tobacco Farmers by District (100s) 373.00 612.28 −239.28∗∗∗ 334.44 629.67 −295.24∗∗∗ Child: Height for age zscore −1.18 −1.11 −0.07 −1.89 −2.15 0.25∗∗∗ Stunted† 0.26 0.28 −0.02 0.45 0.55 −0.09∗∗∗ Male† 0.51 0.48 0.02 0.50 0.46 0.04∗∗ Age (Months) 18.45 12.55 5.90∗∗∗ 33.22 35.03 −1.80∗∗∗ Age (Months) Squared 701.71 321.34 380.37∗∗∗ 1308.22 1416.67 −108.45∗∗∗ Bet Nets for Children† 0.39 0.33 0.06 0.36 0.35 0.01 Mother: Did not Complete Primary‡ 0.83 0.71 0.11∗∗ 0.81 0.80 0.01 Completed Primary‡ 0.07 0.14 −0.07∗∗ 0.08 0.11 −0.03∗∗ Above Primary‡ 0.10 0.15 −0.04 0.11 0.09 0.018 Household: Head of Household: Age (Years) 38.44 35.22 3.22∗∗ 39.69 38.14 1.54∗∗∗ Household Dependency Ratio 1.48 1.42 0.05 1.49 1.45 0.04 32 Household Size 5.46 5.87 −0.40∗∗ 5.73 5.98 −0.25∗∗ Housing Index (PCA) 0.03 −0.30 0.33∗∗∗ 0.03 −0.19 0.22∗∗∗ Household Asset Income 22196.46 24285.63 −2089.17 23039.61 25538.39 −2498.78∗ Income Quintile 1‡ 0.22 0.14 0.08∗∗∗ 0.22 0.13 0.094∗∗∗ Income Quintile 2‡ 0.22 0.22 −0.01 0.21 0.18 0.028∗ Income Quintile 3‡ 0.22 0.21 0.01 0.20 0.19 0.01 Income Quintile 4‡ 0.16 0.23 −0.06 0.18 0.27 −0.08∗∗∗ Income Quintile 5‡ 0.18 0.21 −0.028 0.19 0.23 −0.05∗∗ Land Holding (Acres) 2.20 3.02 −0.83∗∗∗ 2.19 3.14 −0.95∗∗∗ Community: Distance to Closest ADMARC (KMs) 7.49 7.60 −0.11 8.03 7.85 0.180 Distance to Closest Auction Floor (KMs) 72.39 71.57 0.82 71.60 74.14 −2.55 Health Clinic Present† 0.29 0.21 0.09 0.32 0.22 0.10∗∗∗ Southern Region‡ 0.49 0.24 0.25∗∗∗ 0.52 0.20 0.31∗∗∗ Central Region‡ 0.41 0.67 −0.26∗∗∗ 0.37 0.65 −0.29∗∗∗ Northern Region‡ 0.10 0.09 0.01 0.12 0.14 −0.03 Observations 553 133 4120 908 1 ***, **, * indicate statistical significance at the 1, 5, and 10 percent level, respectively, with summary statistics weighted in accordance with the complex survey design. 2 Dummy variables are denoted by the † symbol and categorical variables by the ‡ symbol. 3 Annual household income stems from the income aggregate created under the Food and Agriculture Organization of the United Nations (FAO) Rural Income Generating Activities (RIGA) project. It is expressed in terms of United States Dollar, using the average exchange rate from the IHS2 implementation period (U S $1 = M K 108.714). 4 ADMARC is a Malawian parastatal organization mandated to market agricultural produce and inputs. Table 3: Overall OLS & IV Results OLS Results IV Results Variable of Interest: Tobacco Producer† −0.147∗∗∗ −1.331∗∗∗ (0.054) (0.482) Child: Male† −0.213∗∗∗ −0.241∗∗∗ (0.039) (0.042) Age (Months) 0.001∗∗∗ −0.099∗∗∗ (0.000) (0.006) Age (Months) Squared 0.001∗∗∗ 0.001∗∗∗ (0.000) (0.000) Bet Nets for Children† 0.175∗∗∗ 0.173∗∗∗ (0.042) (0.051) Mother: Completed Primary‡ 0.043 0.102 (0.068) (0.084) Above Primary‡ 0.178∗∗∗ 0.176∗∗ (0.064) (0.076) Household: Head of Household: Age (Years) 0.000 −0.002 (0.002) (0.002) Household Dependency Ratio 0.046∗ 0.037 (0.024) (0.028) Household Size 0.011 0.013 (0.010) (0.012) Housing Index (PCA) 0.088∗∗∗ 0.065∗∗ (0.018) (0.025) Household Living in Residence Before 1990† −0.079∗ −0.109∗ (0.046) (0.061) Income Quintile 2‡ 0.035 0.084 (0.062) (0.069) Income Quintile 3‡ 0.076 0.123∗ (0.061) (0.072) Income Quintile 4‡ 0.063 0.165∗∗ (0.063) (0.077) Income Quintile 5‡ 0.156∗∗ 0.230∗∗∗ (0.065) (0.077) Land Holding (Acres) 0.004 0.073∗∗ (0.014) (0.033) Community: Distance to Closest ADMARC (KMs) 0.005 0.006 (0.004) (0.006) Distance to Closest Auction Floor (KMs) 0.003∗∗∗ 0.003∗∗∗ (0.000) (0.001) Health Clinic Present† −0.052 −0.116 (0.042) (0.073) Southern Region‡ 0.128∗∗ 0.052 (0.061) (0.112) Central Region‡ −0.042 0.077 (0.061) (0.119) Constant −0.589∗∗∗ −0.498∗∗∗ (0.140) (0.193) Observations 5714 5714 IV Test Results First Stage Probit Coefficients Tobacco-Maize Land Suitability Index − 0.005∗∗∗ 1997/98 Tobacco Farmers by District (100s) − 0.001∗∗∗ Cragg-Donald Wald F statistic 153.92 Endogeneity Test (p-value) 0.02 Hansen Over-identification Test (p-value) 0.37 1 ***, **, * indicate statistical significance at the 1, 5, and 10 percent level. The regressions take into account house- hold sampling weights. The robust standard errors are clustered at the EA-level. 2 Dummy variables are denoted by the † symbol and categorical variables by the ‡ symbol. 3 The endogeneity test and the Cragg-Donald Wald F statistics use the optimal TSLS estimation with the Probit-based predicted probability as an instrument for the endogenous variable. The Hansen Over-Identification Test uses the over-identified TSLS estimation with the directly inclusion of the two IVs as excluded instruments. 33 Table 4: IV Results by Shock Exposure within Early Development Window No Exposure Exposure Variable of interest: Tobacco Producer† −0.994 −1.266∗∗∗ (1.126) (0.476) Child: Male† −0.319∗∗∗ −0.232∗∗∗ (0.111) (0.045) Age (Months) −0.102∗∗∗ −0.092∗∗∗ (0.039) (0.008) Age (Months) Squared 0.001∗∗ 0.001∗∗∗ (0.001) (0.000) Bet Nets for Children† 0.060 0.192∗∗∗ (0.159) (0.052) Mother: Completed Primary‡ 0.110 0.083 (0.280) (0.084) Above Primary‡ 0.414∗∗ 0.128 (0.193) (0.080) Household: Head of Household: Age (Years) −0.011∗ −0.001 (0.007) (0.002) Household Dependency Ratio 0.109 0.029 (0.078) (0.028) Household Size 0.016 0.012 (0.039) (0.012) Housing Index (PCA) −0.003 0.077∗∗∗ (0.066) (0.025) Households living in residence before 1990† −0.094 −0.111∗ (0.152) (0.060) Income Quintile 2‡ 0.170 0.070 (0.172) (0.073) Income Quintile 3‡ 0.217 0.105 (0.166) (0.076) Income Quintile 4‡ 0.099 0.162∗∗ (0.192) (0.082) Income Quintile 5‡ 0.086 0.244∗∗∗ (0.219) (0.080) Land Holding (Acres) 0.021 0.072∗∗ (0.077) (0.033) Community: Distance to Closest ADMARC (KMs) 0.027∗∗ 0.003 (0.013) (0.006) Distance to Closest Auction Floor (KMs) 0.001 0.003∗∗∗ (0.002) (0.001) Health Clinic Present† −0.139 −0.101 (0.149) (0.074) Southern Region‡ 0.254 0.031 (0.247) (0.111) Central Region‡ 0.019 0.068 (0.262) (0.118) Constant −0.498∗∗∗ −0.675∗∗∗ (0.548) (0.201) Observations 686 5028 IV Test Results First Stage Probit Coefficients: Tobacco-Maize Land Suitability Index 0.006 0.005∗∗∗ 1997/98 Tobacco Farmers by District (100s) 0.000∗∗ 0.001∗∗∗ Cragg-Donald Wald F statistic 16.03 150.08 Endogeneity Test (p-value) 0.36 0.03 Hansen Over-identification Test (p-value) 0.64 0.23 1 ***, **, * indicate statistical significance at the 1, 5, and 10 percent level. The regressions take into account household sampling weights. The robust standard errors are clustered at the EA-level. 2 Dummy variables are denoted by the † symbol and categorical variables by the ‡ symbol. 3 Children are placed in a group dependent upon their exposure to the food price shock within their early development window. Children born before September 1999 or after October 2003 are counted as having no exposure. 4 The endogeneity test and the Cragg-Donald Wald F statistics use the optimal TSLS estimation with the Probit-based predicted probability as an instrument for the endogenous variable. The Hansen Over-Identification Test uses the over-identified TSLS estimation with the directly inclusion of the two IVs as excluded instruments. 34