While my interests in politics and public policy are mostly domestic (i.e. within the US), I also have a secondary interest in international politics and economics, something I previously had an increased interest in prior to the 2016 (United States) Presidential Election. The following post was an assignment in the same course where I also completed the two domestic consumption regression model posts. In this post, I create econometric models to predict and explain child mortality in the developing and emerging world, including inserting fixed and random effects to control for country-specific results.
Introduction: In this paper, I will be creating a series of econometric models to predict child mortality (defined as children under 5 years old) in developing and emerging economies using five socio-economic and health variables from the World Development Indicators (WDI) 2014 extended 5-year estimates. These variables are GDP per capita (in 2005 real US dollars), access to sanitation facilities, immunization rates of children aged 1-2 against measles and DPT (two separate variables), female literacy rates, and average fertility per woman. The unit of analysis are individual countries in the developing and emerging world, with data encompassing 5-year intervals from 1990 to 2010.
Theory: Generally, as a economy GDP per capita increases, child mortality should be expected to decline because increased economic wealth allows for better health services, thus reducing child mortality (Bar-Zeev et al., 413). An increase in access to sanitation generally should be associated with a decrease in child mortality because having proper sanitation reduces the risk of diseases and other health problems brought on by poor sanitation, such as cholera (Banerjee and Duflo, 46). Meanwhile, an increase in immunizations among children should be associated with a decline in child mortality, as immunizations will make children less vulnerable to diseases they are being vaccinated against (Banerjee and Duflo, 60). One of the most important explanatory variables is female literacy, expressed as a percent per country. Many studies have found that increasing the proportion of women who are educated would have significant health benefits, including decreased child mortality, because women would be able to have increased health knowledge, are more often to marry later, and are made aware of civil rights that they can use to help themselves (Fischetti). Finally, an increase in the average fertility rate should be associated with an increase in child mortality because mothers experiencing higher child mortality are more likely to have a higher fertility rate to ensure that at least one or two children survive (Roser).
Research Design: From the WDI 2014 extended 5-year estimates, ChildMortality will be the response variable, coded as a discrete variable representing the number of child deaths per 1,000 births. In regards to the explanatory variables, GDP Per Capita and FertilityPerWoman will be expressed as discrete variables. The other explanatory variables, ImprovedSanitation, ImmunizeMeasles, ImmunizeDPT, and FemaleLiteracy will be expressed as a percent representing the proportion of women in each country meeting each variable's underlying indicator.
Three regression models will be used for this brief, with robust standard errors to account for possible heteroscedasticity. The first will be an OLS model testing the explanatory variables against the response variable. The second will utilize two sub-models, one utilizing a fixed effects F-Test, and the other a random effects chi-squared test to test for random effects, to test if country-specific results are statistically significant as binary or random variables, respectively. The F-Test’s purpose in this paper will be to account for between-country differences within the data by treating each country as a binary variable and determining if between-country differences are statistically-significant.
The three models are constructed as follows:
- Model 1: Yi ChildMortality = β0 + βn [All Explanatory Variables] Xn + ε
- Model 2a: Yi ChildMortality = β0i + βn fe [All Explanatory Variables] Xn fe + ε, fe = Country
- Model 2b: Yi ChildMortality = β0 + βn re [All Explanatory Variables] Xn re + e, re = Country
Results:
Table 1: Regression Results
*p < .05, ** p < 0.01, *** p < 0.001
Note: Values in parentheses indicate robust standard errors, which account for possible heteroscedasticity in the data.
Table 1 displays the regression results. Under Model 1, the normal OLS, all explanatory variables except sanitation are statistically-significant at the .05 level or higher, and GDP per capita, female literacy, and female fertility are statistically significant at the 0.001 level or higher. However, the direction for rates of child immunization against measles is the opposite of what was predicted, having a positive rather than a negative association.
Under Model 2a, the country F-Test fixed effects model, the F-value was 0.000, making country-specific data results statistically-significant when individual countries are treated as binary variables. Under this model, only GDP per capita and female literacy are statistically-significant at the 5% level or higher, and nether are statistically-significant above the 1% level. However, while measles immunization flips direction to have a negative association with child mortality (as predicted), GDP per capita remains statistically-significant at the 1% level, but has a positive relationship with child mortality.
Similarly, for Model 2b, country random effects, the chi-squared value is 0.000, making these results statistically-significant if country-specific results are treated as random variables. In the regression output, changing the model from a fixed effects F-Test to a random effects Chi-squared test had some notable results. Particularly, female fertility became significant at the 0.05 level again, and the statistical significance of female literacy increased. Meanwhile, measles immunization flipped to Model 1’s direction, as did GDP per capita. However, the sanitation and two immunization variables remained statistically insignificant at the 5% level or higher in Models 2a and 2b, whereas they were in Model 1. Finally, the statistical significance of GDP per capita declined from .01% to 1% from Model 1 to Models 2a and 2b.
Conclusion and Implications: The results above seem to indicate that at first glance, not all health and socio-economic variables are programs with declines in child mortality, and if anything, have no statistical relationship at all. One reason why this may be is that the data are macro-level, which may hide “within-group” differences within countries. In other words, because country-level data simplifies the whole country to a single number, the data may hide outliers or deviations that do follow predicted trends. A recommended way to get around this problem is to focus more on individual-level data collection and studies. Another possible reason why the F-Test results seem to go against predicted trends is that when treating each country as a binary variable, there is less data per “binary variable,” (there were three or four observations per country), making the results more erratic and less significant. Meanwhile, the general swing towards the OLS results during the Chi-Squared test could be indicative of the diversity of the data among countries being studied, in that some may be extremely poor and have very bad health outcomes, while others are increasingly developing and are vastly improving their health outcomes. In turn, the results reflect the “randomness” of the data, which despite the aforementioned problem of the small sample sizes within “country categories,” may make it a better predictor of child mortality.
In addition to the above analysis, there are two significant implications for global health and economic policy from this study. First, female literacy is one of the most powerful statistical indicators of decreased child mortality, which falls in line with the theory and research presented earlier. As such, increasing female literacy around the world would likely be one of the most effective tools to reduce child mortality, and elevate the social and economic standing of women. Second, improved economic prosperity generally is another development tool that is statistically associated with a decrease in child mortality, although the association is not as powerful as female literacy. However, it can be argued that economic development can be a means to help achieve increased female literacy through fostering educational institutions using the newfound wealth to help women gain knowledge to improve the socio-economic and health incomes of themselves and their communities.
Works Cited:
Banerjee, Abhijit V., and Esther Duflo. Poor Economics: A Radical Rethinking of the way to fight Global Poverty. New York City, PublicAffairs, Perseus Books Group, 2011.
Bar-Zeev, Naor, Levison Chiwaula, Innocent Kauta, and Bernadette O’Hare. “Income and child mortality in developing countries: a systemic review and meta-analysis.” Journal of the Royal Society of Medicine, vol. 106, no. 10, 2013, www.ncbi.nlm.nih.gov/pmc/articles/PMC3791093/pdf/10.1177_0141076813489680.pdf.
Fischetti, Mark. “Female Education Reduces Infant and Childhood Deaths.” Scientific American, 7 Jul. 2011, www.scientificamerican.com/article/graphic-science-female-education-reduces-infant-childhood-deaths/.
Roser, Max. “Fertility Rate.” Our World in Data, last revised 2 Dec. 2017, ourworldindata.org/fertility-rate. Accessed 12 Dec. 2019.
Data Source:
Bhargava, Alok. In the author's possession.
Nathan Parmeter
Author and Host, The Parmeter Politics and Policy Record
No comments:
Post a Comment