1. Market access and nutrition smoothing: Access to towns and cities protects farm children against poor health conditions at birth in the DRC William A. Masters ab http://sites.tufts.edu/willmasters Amelia F. Darrouzet-Nardi a http://sites.tufts.edu/ameliadarrouzetnardi a Friedman School of Nutrition Science and Policy b Department of Economics (by courtesy) Tufts University Seminar at the Delhi School of Economics 3 December 2014

2. How important is access to towns and cities for child nutrition and health on farms? 1.Access to markets could improve or worsen child nutrition : – could raise purchasing power, but also raise cost of caregivers’ time – and alter relative prices of nutritious foods 2.Market access also changes the ag-nutrition relationship – can separate decision-making between farm and household, – and creates opportunities for consumption smoothing 3.This paper focuses on resilience to seasonal health shocks – loosely inspired by Burgess & Donaldson (2010), "Can Openness Mitigate the Effects of Weather Shocks? Evidence from India's Famine Era." American Economic Review, 100(2): 449-53. – farmers’ vulnerability to shocks may be increasingly important over time – towns and cities offer diverse channels for consumption smoothing labor markets, migration and remittances product and asset markets public services and insurance networks Market access and farm household nutrition motivation | method | results | robustness

3. This paper is about separability Qty. of nutritious foods (kg/yr) Qty. of farm household’s labor time (hrs/yr) Qty. of farm household’s other goods (kg/yr) Other employment (allows sale of labor to buy food) Can towns and cities help rural children overcome shocks to nutrition and health production? Qty. of nutritious foods (kg/yr) Once farmers are actively trading, production decisions are “separable” from consumption choices, linked only through purchasing power Rural food markets (allows sale of other goods to buy food) In self-sufficiency, production =consumption Consumption Production Consumption Production That same separability applies whether households are buying or selling, and allows consumption smoothing over time Market access and farm household nutrition motivation | method | results | robustness

4. Farmers’ access to towns and cities is a major focus for public investment and is increasing rapidly around the world, but causal impact is unclear – Markets arise and grow where people have something to sell – And people who have things to sell move towards markets How might one identify (some of) urbanization’s effects on rural people? – Randomized trials isolate specific interventions, and cannot reveal combined effects of transport, communications, investment and trade – Many surveys occur around natural experiments, but access to towns and cities varies only slowly and predictably Here, we focus on spatial and temporal variation in seasonal risks – Our natural experiment is the timing of conception and birth Relative to spatial and temporal variation in weather shocks In a country that offers “placebo” regions with little seasonality Market access and farm household nutrition motivation | method | results | robustness …and about identification strategy What can cross-sectional survey data reveal about nutrition and health behavior?

5. Birth timing relative to seasonal variation creates a recurring natural experiment The “treatment” is having the worse season (if there is one) occur during the period of greatest vulnerability – late pregnancy and early infancy are highly sensitive for child growth – wet seasons often bring both hunger and disease exposure Market access may be protective – Households can trade to smooth consumption – Households can access health and other services We expect less effect of birth timing, and less protection from market access, in regions with less seasonal fluctuation in rural conditions Market access and farm household nutrition motivation | method | results | robustness

6. The D.R. Congo is the size of India, but much poorer Market access and farm household nutrition motivation | method | results | robustness Source: http://www.ifitweremyhome.com/compare/IN/CD

7. The D.R. Congo has low density and straddles the equator Market access and farm household nutrition motivation | method | results | robustness equator Towns and cities depend on mining etc.; seasonality depends on latitude

8. Within each cross-sectional survey, we have a triple difference-in-difference design Household location and child birth timing Region has a distinct wet season? (= farther from the equator) Yes No (placebo region) Child was born in or after wet season? (=Jan.-Jun. if lat.<0, Jul.-Dec. otherwise) Yes (at risk) No (control) YesNo Household is closer to town? (=closer to major town) Yes (protected?) NoYesNoYesNoYesNo Hypothesized effect of birth timing:Neg.None Note: To test our hypothesis that market access protects against seasonality, the identifying assumptions are that birth timing occurs randomly between seasons (tested), and that seasonal risk factors would have been similar in the absence of towns (untestable). Market access and farm household nutrition motivation | method | results | robustness

9. Seasons depend on rainfall and temperature equator At the equator, average monthly rainfall fluctuates from 100 to 200 mm, and average monthly temperature fluctuates from 24 to 26 degrees Celsius. Market access and farm household nutrition motivation | method | results | robustness

10. “Winter” is a drier period, farther from the equator equator Away from the equator, there is a drier, colder winter, here May through August. Latitude -6 Market access and farm household nutrition motivation | method | results | robustness

11. In the other hemisphere, winter is 6 months later equator Here in the Northern Hemisphere, the drier season occurs from November through February. Latitude +4 Market access and farm household nutrition motivation | method | results | robustness

12. The data are... Demographic and Health Surveys (DHS), in 2007 and 2013, for – Height and weight of the index child (N=8,435 children) – Mortality of children ever born to the respondent (N=69,641 births)m which permits us to control for mother fixed effects – Demographic controls (age, sex, whether firstborn or a short birth interval) – Wealth quintile (relative to other DHS respondents) The Armed Conflict Location and Event Dataset (ACLED) for – Exposure to armed conflict near the child’s home during their birth year The FAO’s Multipurpose Africover Database on Environmental Resources, for – Proximity to the nearest of 160 towns and cities – Latitude (and hence exposure to seasonality) Market access and farm household nutrition motivation | method | results | robustness

13. Birth timing: Presence of seasons: Jan.-June No N=18,009 Jan.-June Yes N=18,973 July-Dec. No N=16,724 July-Dec. Yes N=15,935 All Births N=69,641 Child status Children Alive (%)84.6%84.5%83.7%85.2%84.5% HAZ-1.51 (1.68)-1.51 (1.62)-1.61 (1.92)-1.26 (1.80)-1.47 (1.86) WHZ-0.31 (1.25)-0.47 (1.12)-0.24 (1.41)-0.45 (1.31)-0.38 (1.33) Age (months)28.24 (17.57)28.00 (17.29)29.70 (17.10)29.88 (16.69)29.16 (16.53) Firstborn (%)23.8%24.9%23.8%23.5%24.5% Short interval (%)28.2%27.9%26.1%19.74%25.6% Boys (%)50.5%51.2%50.4%50.2%50.6% Household Wealth (quintile)2.61 (1.27)3.20 (1.46)2.60 (1.26)3.25 (1.45)2.92 (1.40) Proximity (km -1 )0.11 (0.23)0.16 (0.27)0.10 (0.23)0.15 (0.27)0.13 (0.26) Environment Conflicts108.72 (716.5)15.03 (65.7)93.52 (596.8)15.95 (69.7)31.28 (66.9) Latitude (abs val)1.91 (1.36)6.14 (2.01)1.98 (1.17)5.99 (2.02)4.31 (2.64) We split the population into groups by risk exposure Note: Data shown are means and standard deviations (in parentheses). Births labeled as Jan.-June occurred in calendar months July-December for children born in the Northern hemisphere (N=418). Conflicts are number of fatalities during the child’s birth year in the respondent’s 1-degree square grid-cell of residence.

14. Market access and farm household nutrition motivation | method | results | robustness (1)(2)(3) VariablesUnit/typeChild is aliveHAZWHZ Age spline 1Linear spline-0.017***-0.074**-0.107*** Age spline 2Linear spline-0.002**-0.072***0.011*** Age spline 3Linear spline -0.006 Child is maleBinary-0.115*-0.133**-0.108** Child is firstbornBinary-0.288***0.021-0.026 Short preceding birth intervalBinary-0.594***-0.148*-0.020 Ln(fatalities during birth year)Continuous-0.062***-0.114***0.031** Household Wealth indexCategorical0.145***0.250***0.053*** Absolute value (latitude)Continuous-0.046***-0.015-0.017 Proximity to townkm -1 0.281**-0.0220.162 Born Jan.-JuneBinary0.134**-0.1070.075 Constant 2.940***-0.2560.407*** ObservationsN18,8453,4053,473 R2R2 R2R2 0.1790.073 An exploratory regression with continuous variables describes the relationships between them

15. Market access and farm household nutrition motivation | method | results | robustness (1)(2)(3) VariableUnit/typeChild is aliveHAZWHZ Age spline 1Linear spline-0.016***-0.080***-0.100*** Age spline 2Linear spline-0.002***-0.067***0.010*** Age spline 3Linear spline -0.009*** Short preceding birth intervalBinary-0.510***-0.187***-0.039 Child is maleBinary-0.149***-0.164***-0.116*** Ln(fatalities during birth year)Continuous-0.057***-0.087***0.018 Proximity to townkm -1 0.744***0.3690.144 Born Jan.-JuneBinary0.080-0.097-0.022 Absolute value(latitude)Continuous-0.0040.045***-0.019 Born Jan.-June*ProximityInteraction0.1040.877**0.232 Born Jan.-June*Abs(lat)Interaction-0.0020.0180.007 Abs(lat)*ProximityInteraction-0.0530.038-0.014 Born Jan.-June*Proximity*Abs(lat)Interaction-0.021-0.201***-0.000 Constant 3.081***0.2000.627*** ObservationsN18,8453,4053,473 R2R2 R2R2 0.1440.056 Splitting each variable into categories, we can run a triple difference-in-difference regression Notes: The linear age splines are actually ‘time elapsed in months since birth’ for the mortality regressions. Age splines control for child’s age at observation. Born Jan.-June is actually born July-Dec. in Northern hemisphere to account for inversion of seasons at the equator. Age splines control for child’s age at observation. Conflicts are the cumulative count in the child’s cluster of residence during the child’s birth year. Errors clustered by DHS survey cluster (v001), * p<.10, ** p<.05, *** p<.01.

16. Market access and farm household nutrition motivation | method | results | robustness Our preferred specification is to split the sample, taking advantage of relatively large sample size (1)(2)(3)(4)(5)(6) VariableUnit/type Alive Seasons Alive No Seasons HAZ Seasons HAZ No Seasons WHZ Seasons WHZ No Seasons Age spline 1Spline-0.021***-0.022***-0.051-0.135***-0.098***-0.101*** Age spline 2Spline-0.003***-0.002***-0.086***-0.090***0.010***0.012*** Age spline 3Spline -0.005-0.003 Short intervalBinary-0.284***-0.302***-0.385***-0.449***-0.172***-0.062 MaleBinary-0.117***-0.126***-0.029-0.293***-0.104*-0.038 Conflict exposedBinary-0.0430.0360.1390.249**-0.074-0.062 Jan.-JuneBinary-0.127**0.079-0.0970.0630.051-0.093 Jan.-June*RemoteInteraction0.128*-0.025-0.329**-0.188-0.0340.132 Constant 0.1580.537**0.524***0.624*** ObservationsN17217172974224421143124319 R2R2 R2R2 0.2900.2990.0830.077 Note: The mortality tests (col. 1 and 2) include mother fixed effects, and the linear age splines are actually time elapsed since birth, in months. Born Jan.-June is actually born July-Dec. in Northern hemisphere to account for inversion of seasons at the equator. Age splines control for child’s age at observation. Conflict exposure is a binary indicator of whether there was civil conflict in a 1-degree square of the child’s residence during the child’s year of birth. Errors clustered by DHS-cluster (v001), * p<.10, ** p<.05, *** p<.01

17. Market access and farm household nutrition motivation | method | results | robustness Note: Data shown are the number of children ever born in each month, as recorded across each DHS survey in the DRC. The solid line refers to calendar months, and the dashed line uses a seasonal adjustment by hemisphere, where dates north of the equator are recorded as “January” for births in June, “February” for July, etc. In our regressions, these “rain months” are aggregated into six-month periods, since as children in higher latitudes who are born in the January-June period are more exposed to heavy rains and subsequently poor health outcomes than those born in the rest of the year. Could the correlations we see be driven by selection into healthier birth timing? This turns out to be the less healthy season in which to be born, suggesting no attempt at selection into healthier timing of conception and birth Factors other than the health of the child must be driving seasonality in conception and birth

18. Market access and farm household nutrition motivation | method | results | robustness Note: Dependent variable is a binary indicator of birth during the Jan.-June wet season. Regression estimated using fixed-effects logit. All results include fixed effects for survey clusters (N=840), with notation and variable definitions as in Table 6. p-values in parentheses ; * p<.10, ** p<.05, *** p<.01. Could the correlations we see be driven by selection into healthier birth timing? (1)(2)(3) VariableUnits/typeBorn Jan.-June Seasons Born Jan.- June No seasons Child is MaleBinary0.0090.0230.005 (0.762)(0.632)(0.895) Wealth indexCategorical-0.015-0.0570.002 (0.384)(0.106)(0.919) Ln(fatalities)Continuous0.0140.0030.018 (0.125)(0.830)(0.152) Proximity to townkm -1 0.319*0.538-0.047 (0.069)(0.227)(0.875) Abs val (latitude)Continuous0.021 (0.138) Observations 18804706011728 The only correlation we see is with proximity to town, e.g. from a seasonal migration effect

19. Market access and farm household nutrition motivation | method | results | robustness Note: Data shown are coefficient estimates and 95% confidence intervals for “average treatment effects” in our preferred specification (Table 5), for our three dependent variables of interest followed by five ‘placebo’ variables for which no effect is expected of our ‘treatment’, due to the absence of any plausible mechanism of action. Among our robustness checks, we do “placebo” tests for desirable outcomes that could not be caused by birth timing Hypothesized effects on survival, heights, weights No significance where no effect is expected

20. Conclusions and implications In the DRC, farm households that are closer to towns use it to protect themselves from seasonal shocks to nutritional status Possible mechanisms underlying this effect include: – Specialization and trade, to overcome diminishing returns on the farm – Consumption smoothing, via separability of production & consumption – Access to public services Future work may be able to distinguish among mechanisms – But all of them provide opportunity for farm households to exploit or respond to their own idiosyncratic, diverse circumstances – Policies and programs based on markets cannot prescribe what households will do, only that they can do it more easily! Market access and farm household nutrition motivation | method | results | robustness