There are some intractable social problems which are extremely difficult to ameliorate but low-birth weight shouldn’t be one of them for a country as rich as ours. It would appear a prudent approach for mediating the adverse effects of low-birth weight outcomes is to know where to look for such outcomes in order to mobilize and deploy sufficient resources to address the problem. In this study I exploit two datasets I merged to understand the association of several county-wide variables with low-birth rates. Before we move to the study’s findings let’s ask why we should be concerned about low-birth weight.
Why is Low-Birth Weight a Major Problem?
The adverse consequences of low-birth weight have been known for a long-time. Low-birth weight outcomes have been correlated with deviant behavior, difficulties in language development, poor academic achievement, lower intelligence, high school dropouts, hyperkinesis, autism, involvement in childhood accidents, poor cognitive function, slow physical growth, respiratory disease, high blood pressure, diabetes, and heart disease. In addition, women with low income, have low levels of educational attainment, are Black and younger than 17 or older than 35 are more likely to give birth to low-birth weight babies (Bhutta, et al., 2002; Caputo andWallace, 1970; Hack, et al., 1991; Hack, et al., 1995; Paneth, 1995; March of Dimes, 2014). The first line of defense associated with low-birth weight outcomes should be prevention (Reichman, 2005).
Data, Variables and Hypotheses
The focus of this study is county variations in low-birth weight outcomes in the US. Two datasets were merged: 1) Robert Wood Johnson Foundation and the University of Wisconsin Population Health Institute, County Health Ranking and, 2) American Community Survey, Demographics – Poverty; Social Determinants of Health – Employment.
With the merged dataset the following variables were available:
The specific research hypotheses are:
- The percent low-birth weight outcomes is directly related to the percent of county-wide adults who report no leisure time physical activity.
- The percent of low-birth weight outcomes is directly related to the percent of county-wide children who live in single-parent households.
- The percent of low-birth weight outcomes is directly related to the percent of county-wide adults who report not getting social/emotional support.
- The percent of low-birth weight outcomes is inversely related to county-wide high school graduation rate.
- The percent of low-birth weight outcomes is directly related to the county-wide inequality measure, the Gini coefficient. (The Gini coefficent varies between 0, representing complete equality, and 100, representing complete inequality.)
- The percent of low-birth weight outcomes is related to regions of the country: South, Northeast, Midwest and West.
As the dependent variable, low-birth rate, is a continuous variable and the covariates are a mixture of continuous and one categorical variable, the statistical tool of choice is multiple regression.
The multiple regression produced the following output.
The full model is statistically significant, Prob > F = 0.0000, as well as all covariates. The model accounts for a large amount, 64%, of the variance in low-birth weights among counties in the United States.
I could spend a few paragraphs interpreting each coefficient with a phrase something like this: With all other variables controlled, a one unit increase in variable x is associated with a y (coefficient) increase/decrease (depending on the sign of the coefficient) in the percent low-birth weight outcomes.
One problem with this approach to interpretation is some of the covariates are measured in different scales. A scale-free method is provided by using elasticity coefficients (Wooldridge, 2002). Below is a table of elasticity coefficients associated with this regression analysis. Elasticity coefficients indicate a proportional change in the dependent variable is associated with a given proportional change in the independent variable.
Controlling for the influence of all other covariates, we can interpret each elasticity coefficient in the following manner:
- A 1% increase in the ‘percent of adults whot report no leisure time physical activity’ is associated with a 0.231% increase in babies categorized as low-birth weight. The research hypothesis is supported.
- A 1% increase in the ‘percent of children who live in single-parent households’ is associated with a 0.287% increase in low-birth weight babies. The research hypothesis is supported.
- A 1% increase in the ‘percent of adults who report not getting social/emotional support’ is associated with a 0.125% increase in low-birth weight outcomes. The research hypothesis is supported.
- A 1% increase in ‘Gini coefficient, measuring inequality,’ is associated with a 0.239% increase in low-birth weight babies. The research hypothesis is supported.
- A 1% increase in ‘high school graduation rate’ is associated with a 0.145% decrease in low birth weight outcomes. The research hypothesis is supported.
With the regression formula available we can utilize the regression coefficients to predict each county’s expected low-birth weight mean. By sorting the predictions we can identify the top ten counties with the most favorable low-birth weight outcomes and the worst ten counties with the least favorable low-birth weight outcomes.
Ten counties with the most favorable low-birth weight outcomes are:
Colorado, Utah and Minnesota dominate US counties in producing the best low-birth weight outcomes.
Ten counties with the worst low-birth weight outcomes are:
Counties in Southern states exhibit some of the worst low-birth weight outcomes. Mississippi has the unenviable distinction of having 6 of the 10 counties with the least favorable low-birth outcomes.
The impact of the Southern region on low-birth outcomes is highlighted in the following graphs.
As the Gini coefficient increases, capturing greater levels of economic inequality, the percent of low-birth outcomes increases in all regions but Southern counties exhibit higher levels of low-birth weight outcomes throughout the Gini coefficient distribution.
As the percent of children who live in single-parent households increase the percent of low-birth weight babies increase but the regions of the Northeast, Midwest and West exhibit lower predicted low-birth weight means throughout the ‘percent of single household’ distribution.
An increase in the ‘percent of adults who report not getting social/emotional support’ is associated with an increase in low-birth weight outcomes but again Southern counties exhibit higher means throughout the ‘no social/emotional support’ distribution.
Low-birth weight outcomes are directly related to the ‘percent of county-wide adults who report no leisure time physical activity’ in all regions of the US but Southern counties have higher predicted low-birth means than the other US regions.
As hypothesized there is an inverse relationship between low-birth weight outcomes and high school graduation rates. While this relationship holds for all US regions we nevertheless observe Southern counties with higher, unenviable low-birth predicted means throughout the graduation rate distribution.
The observant reader will have noticed that the predicted low-birth weight means appear to be very close in the above graphs for the West, Midwest and Northeast regions. There are differences in the predicted means but are they significantly different?
Let’s begin by using Stata’s ‘contrast’ command to test the overall effect of the factor variable, region_cat.
The above table demonstrates the overall effect of region_cat is significant (P>F = 0.000). We can say that region_cat deserves to be in the model.
What about the contrasts between the other regions? To answer this question we use Stata’s ‘pwcompare’ command, producing the following table.
The first three contrasts duplicate the results we just interpreted. However, the last three contrasts reveal non-significant differences in low-birth weight means between West, Midwest and Northeast regions. The only significant mean difference contrasts are those which involve the South. The other contrasts are not significantly different at an conventionally used level of significance.
Conclusion and Policy Implications
This analysis has demonstrated low-birth weight outcomes can successfully be predicted by county-wide variables. If we can successfully predict where low-birth weight outcomes are more likely to occur it then becomes prudent to mobilize and deploy resources in those areas where known medical and social interventions can mediate the adverse effects of low-birth weight.
Bhutta, Adnan T., et al. (2002). “Cognitive and behavioral outcomes of school-aged children who were born preterm: A meta-analysis.” Journal of the American Medical Association. Volume 288, Number 6: 728–37.
Caputo, Daniel V.; Mandell, Wallace. (1970). “Consequence of low birth weight.” Developmental Psychology, Vol 3(3, Pt.1), 363-383. doi: 10.1037/h0030030
Hack,Maureen; et al. (1995). “Long-term developmental outcomes of low birth weight infants.” The Future of Children. Vol. 5, No. 1, pp. 176-196.
Hack, Maureen; et al. (1991). “Effect of very low birth weight and subnormal head size on cognitive abilities at school age.” New England Journal of Medicine. 325:231-237/ doi: 10.1056/NEJM199107253250403
March of Dimes. (2014). Working Together for Stronger, Healthier Babies.
Paneth, Nigel. (1995). “The problem of low birth weight.” The Future of Children. Vol 5, No. 1: 19–34.
Reichman, Nancy E. (2005). “Low birth weight and school readiness.” The Future of Children. Volume 15 Number 1.
Wooldridge J. M. (2002). Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press.