DOES TRADE AFFECT CHILD HEALTH?

Frankel and Romer (1999) documented positive effects of geographically-determined trade openness on economic growth. At the same time, critics fear that openness can lead to a “race to the bottom” that increases pollution and reduces government resources for investments in health and education. We use Frankel and Romer’s gravity model of trade to examine how openness to trade affects children. Overall we find little harm from trade and potential benefits largely through slightly faster GDP growth.

The consensus among mainstream economists on the desirability of free trade remains almost universal.

-- Mayda and Rodrik (2001)

Economists generally agree that economic openness is a good thing. The basis for this support ranges from the theory of comparative advantage and political economic theories of rent seeking (e.g., Krueger, 1974) to empirical evidence such as longitudinal studies correlating trade opening with rising living standards and declining poverty (Dollar and Kraay, 2004) and cross-sectional studies with sophisticated identification strategies (e.g., Lee, Ricci and Rigobon 2004). One of the most convincing studies is Frankel and Romer (1999), which showed that even the portion of trade determined by plausibly exogenous geographic factors predicts higher GDP per capita.

The hypothesis that trade improves living standards is more controversial outside of economics (e.g., Weissman 2003; Tabb 2001; and Danaher and Burbach 2000; Mayda and Rodrik [2001] provide the case against the claim that trade improves living standards from within economics). Some critics are dubious that trade promotes economic growth. Others argue that even if trade grows the economy, trade’s benefits either do not trickle down to most citizens or are outweighed by its costs. These potential costs may include environmental degradation, increased exposure to disease, decreased public spending due to lower ability to tax capital, increased exposure to international financial crises, and increased demand for low-skill labor and subsequent reduced returns to human capital acquisition.

In this paper we ask whether openness to the international economy affects children’s health. This question goes to the center of the debate about globalization. Children’s health is an important end in its own right (Sen 1999). Health is also an important means to achieving long-run economic growth. Even if trade promotes short-run economic growth, it is unlikely to drive long-run growth if it harms health substantially.

A simple correlation between trade and a measure of children’s health does not necessarily reveal the causal effect of trade (Rodrik and Rodriguez 2001; Helpman 1988; and Harrison 1996). For one, countries that trade more might be different from countries that trade less in ways related to children. For example, high prevalence of tropical disease might reduce both trade and health (McArthur and Sachs, 2001). Intuitively, traders avoid malarial regions. Also, the causality might be reversed. Healthy children tend to become more productive adults. If higher incomes lead to greater demand for variety, countries that have healthier children might choose to trade more. Similarly, Alsan, Bloom and Canning (2004) discuss channels connecting improved health with higher inflows of direct foreign investment. They document a positive correlation in a panel dataset (although the causality remains unclear).[1]

Building on the work of Frankel and Romer (1999), we address these concerns about causality by predicting how much countries trade using exogenous geographical characteristics. To determine the exogenous portion of trade we first estimate a “gravity” model of trade as a share of GDP. In the gravity model, trade between two nations is assumed to depend on such factors as whether they have a coastline, share a border, are near each other, and are large in terms of population and area. Cumulating predicted bilateral trade flows across all potential trading partners provides an estimate of predicted trade for each nation. This estimate of predicted trade is highly correlated with actual trade and is also plausibly exogenous; we do not believe child welfare has an effect on the location of nations.[2] Intuitively, Chad is a land-locked nation far from the population centers of the globe, and so the gravity model predicts low trade for Chad. We then use the predicted geographic component of trade estimated from the gravity model to obtain a cross-sectional estimate of the effect of trade on children’s welfare. As noted above, Frankel and Romer (1999) found openness (as predicted by geography) causes higher economic growth. Others have used this method to show openness reduces pollution (at least at higher income levels, Frankel and Rose 2002), reduces child labor (Edmonds and Pavcnik 2004) and increases democracy (Lopez-Cordova and Meissner 2005).

We use two pairs of measures of children’s health: (1) mortality rates for infants and for children up to age 5; and (2) anthropometric measures of child stunting (low height for age) and wasting (low weight for height). We find that trade predicts lower infant and child mortality and lower stunting (height for age) but has no statistically discernible relationship with wasting (weight for height). We also find that trade predicts higher life expectancy. Our results imply that for the average country a 15 percentage point increase in predicted trade as a share of GDP (an increase of about one standard deviation) corresponds to approximately 4 fewer infant deaths per 1000 births and 4 fewer deaths before age 5 per 1000 births. Each of the mortality declines is about 10% of both the mean and the standard deviation; thus, the effects of trade are economically important. We also find that trade predicts higher income, higher immunization rates for measles, and larger expenditures on public health as a share of GDP, and that some of the benefits of trade on child health appear to operate through these causal channels. The remaining effects of trade may operate through unmeasured channels such as flows of information or may be due to mismeasurement of incomes and the other observed channels.

Potential Links Between Openness and Children’s Outcomes

Trade might affect children’s health through a number of pathways. These include: income, public and private incentives, public policy, environmental quality, urbanization, and Westernization. We briefly discuss each potential pathway.[3]

Trade might affect children’s welfare by increasing economic growth. Since Adam Smith and David Ricardo, economic theories have predicted that trade increases income. In turn, rising incomes appears to improve health (Pritchett and Summers 1996). Channels may include: improved nutrition for mothers and for children (Fogel, 1994); improved access to clean water (from bottled water, treated drinking water and/or better community sanitation); and improved access to health care (especially if many parents are liquidity constrained [Becker and Tomes, 1985] or if governments invest tax revenue in public health).

Trade might also affect children’s health by influencing the degree to which governments are willing and able to fund public health. On the one hand, openness to the international economy can lead to financial crises and debt build-up, which can increase the influence of the IMF and World Bank.[4] Critics of these organizations emphasize cases where such international institutions frequently pushed governments to reduce spending on social services (Weissman, 2003). Moreover, in open economies, governments have a hard time taxing capital; in the extreme, bidding wars for factories can reduce resources for investments in children. At the same time, the increased risk associated with greater openness to the international economy can increase public demand for government safety nets, and children may benefit from this if these government safety nets are disproportionately targeted toward health (Rodrik, 1999). As noted above, any increase in economic growth might also increase the government’s tax revenue, which can increase spending on children’s health.

Trade might also affect children’s health by affecting environmental quality. Outward-oriented industrialization initiatives are closely related to trade, and factories emit air and water pollution. Air pollution is a major source of acute respiratory infections such as pneumonia, which are a common cause of infant mortality. Encouragingly, Frankel and Rose (2002) test whether openness (as predicted by the gravity model) predicts higher pollution. They find that trade does not increase pollution and might even reduce it.

Factories also tend to be located in or around cities. Whether this is good or bad for children is not possible to predict. On the one hand, person-to-person and water-borne pathogens are more easily transmitted in urban settings, and air pollution tends to be worse. On the other hand, children in cities are usually closer to health care than children in rural areas. In addition, indoor coal and wood fires – major sources of indoor air pollution – are less prevalent in urban areas.

Openness can also help spread infectious diseases, as seen in the European arrival in the Western Hemisphere 500 years before our sample and in the epidemic spread of HIV/AIDS along many highways (although HIV/AIDS only affected national mortality rates after our study period).

Finally, the gravity model’s measures of openness to trade presumably capture openness to cultural influences. For example, centuries ago, Islam spread along trade routes to long expanses of coastline Africa and East Asia, while Western science and culture have spread around the globe along with traded goods. To the extent that openness as measured by the gravity model increases understanding of the germ theory of disease, the value of immunizations, and the value of literacy and science, openness to trade can improve children’s outcomes. At the same time, if openness brings greater recreational drug use, more consumption of Coca-Cola, and less breastfeeding (in the pre-HIV epidemic era we study), then openness might reduce children’s health.

Empirical Analysis

The net result of the forces described above is unclear from theory, and so we proceed to our empirical analysis. We present the gravity model, describe our data, and present results. We then look at several of the channels noted above: income, immunization rates, urbanization, and public expenditures on health to see if they mediate the links between trade and children’s outcomes.

Geographical Gravity Model

We use the cross-sectional gravity model from Frankel and Romer (1999). The amount of trade between any two countries i and j is modeled as a function of the distance between them (Dij), their populations (Ni), their areas (Ai), whether or not they are landlocked (Li), whether or not they share a common border (Bij), and several interactions:

The fitted values from equation (1) are the predicted geographic components of each country’s trade with each other country in the world. For each country, these fitted values are summed to obtain the total predicted geographic component of trade:

(2) ,

where b is a vector of coefficients in equation (1), (b0, b1, ..., b13), and Xij is the vector of right-hand side variables (1, lnDij, ... , Bij[Li +Lj]).

Specification

Our cross-sectional specification uses either actual trade as a share of GDP or the predicted level of trade as a share of GDP, , and a vector of control variables (Z), to predict child welfare (W):

(3) ln(Wi) = b0 + b1Ŝi + Zib + ei .

Data

Table 1 contains summary statistics of our cross-sectional sample. Depending on the variables we analyze, the sample consists of data for 100-130 countries. Our primary selection criterion was data availability. For 134 countries we had a complete set of observations for infant mortality, GDP per capita, geographical trade share and actual trade share. Given this base sample, the only countries we omitted from our primary analysis were Bulgaria, Hungary, Poland and Romania. We left these countries out because they were all Soviet bloc countries and our data come from a time of significant political and social upheaval there. Including them in the analyses presented below had no substantive effect on the results. The complete set of countries included in our sample is listed in the Appendix.

We use infant and child mortality rates and stunting and wasting rates to measure children’s welfare. The infant and child mortality rates are from the World Bank’s World Development Indicators (1990). The stunting and wasting data are largely from the World Health Organization (WHO).[5] Stunting largely captures persistent shortfalls in nutrition and/or a high disease burden, while wasting captures more recent shocks to nutrition and illness (WHO 1995, ch. 5). While genes are important determinants of height and weight variation within a well-fed population, they have little to do with variation across populations with high and low levels of nutrition (ibid.).

Our measures of actual and predicted trade shares are the actual and predicted (geographical) trade shares reported by Frankel and Romer (1999). They use 1985 trade, population and income data from the Penn World Table. We also include the percentage of land in a tropical area (from Gallup and Sachs 1999). We sometimes, but not always, include GDP per capita, immunization rates for measles, and government expenditures on public health as a share of GDP. We are interested in comparing how the relationship between trade and children’s health changes when we condition on these variables. These measures come from the World Development Indicators (1990).[6] When used, each measure is from 1990.

Finally, measurement error is a serious concern for all of our data. Accurately measuring social indicators such as infant mortality across countries is notoriously difficult (Pritchett, 1997; Krueger and Lindahl, 2001). We discuss how measurement error may affect our results.

Identification

The validity of our identification strategy depends on our instrument - geographical trade share - satisfying three conditions: (1) It must be correlated with actual trade share; (2) It must not be affected itself by children’s welfare; and (3) It must be uncorrelated with other factors that affect children’s welfare.

Condition (1) is easily satisfied in that our predicted trade share strongly correlates with actual trade. Table 2 contains the results of a regression of actual trade share on geographical trade share (our instrument), the percentage of land in the geographical tropics, and region indicator variables. The F-statistic for geographical trade share is 86, which by conventional standards is a very strong first stage (Staiger and Stock, 1994).

We can rule out condition (2) by virtue of the construction of geographical trade share. Children’s health cannot affect a country’s geographical characteristics.