Ryan Fairchild and Dr. Joseph Price, Economics
In the United States, two out of three teenagers have sex before graduating from high school. As a result there is a high probability that young women are at risk to become pregnant while still in high school . Studies have shown that teenage pregnancy is associated with a number of adverse outcomes for both mother and child, including less human capital investment and lower future wages . Because of this, factors that could lower teenage pregnancy are important in improving the lives of both young women and future generations. Over the last 50 years, researchers have investigated various mechanisms that have driven teenage pregnancy rates steadily downward. However desegregation, a major social and political change that occurred between the 1960s and 1980s, has been left almost entirely unexplored.
The mechanism through which desegregation could drive a reduction in teenage pregnancy rates is through interracial matching as studies have shown a strong social tendency against interracial relationships. Yancey (1998) discusses issues that lower the occurrence of interracial matching—such as caste theory—while Moe, Nacoste, and Insko (1981) show that a stigma against interracial marriage existed amongst 9th graders in the South during 1966. The expected result of this propensity against interracial matching is that changing the racial composition at desegregated high schools would effectively reduce the number of potential sexual partners and therefore reduce the number of sexual encounters and pregnancies.
In his 2004 paper studying the effects of desegregation on high school dropout rates, Jon Guryan notes that there was no universal, one-time desegregation change in school districts around the nation, but that instead these changes were implemented over time ranging from 1961 to as late as 1982. He exploits these differences to create an estimator that compares the dropout rates between school districts that instituted integration policies 10 years apart from each other and controls for various district characteristics to minimize variation not resulting from segregation policies. He laments that better data was unavailable to complete a year-by-year evaluation and instead resorts to a simplified decade study comparing all 1970s desegregators to those that desegregated before and after that period. However, because pregnancy rates are available for many of the years in which desegregation occurred, we are able to create a more thorough analysis using a difference-in-difference (DID) estimation strategy—an econometric tool that accounts for time trends and geographic factors in order to more effectively isolate the effect of a particular intervention, desegregation in this case.
The biggest challenge is finding, combining, and then cleaning data. We draw on two datasets to construct our measures of teen pregnancy. First, we use data from US birth certificates, the NCHS Natality data, which provides information on the mother’s age and race; state and county of residence; and in many cases the race and age of the father. We combine the count of the number of teen births by age and race with population estimates from the United States Census Bureau to construct teen pregnancy rates for each state. We then condense the data on these districts to create panel data across the years of interest. For the 200 counties that are identified in the natality data we can construct counts of the number of teen births by age and race. Finally, Johnathan Guryan’s data, which includes the timing of desegregation policies at the county level, will be used to complete the data set. We combine it all using a crosswalk that will allow us to connect FIPS and NCHS geographic codes, the two types used in the various datasets. Also, Guryan notes that districts tend to encapsulate one or more counties and matches counties to districts accordingly. Some of these do not match exactly, but are close enough that complications are not anticipated.
Using STATA, a statistical analysis tool, to then run our DID estimator, we compare those who desegregated before 1973 (the average year of desegregation) to those who were yet to implement desegregation policies. Our results are both small and statistically insignificant. The interaction term indicates that desegregation, according to our data, had an effect of increasing teenage pregnancy rates by 0.8% and is not statistically different from 0. Some of the initial results, such as the strong coefficients on the black and white race variables, are somewhat surprising and will be further investigated. These will probably have little effect on our findings of desegregation unless they indicate serious data errors, which is unlikely given other robustness checks. We have contacted Jon Guryan about using his dataset and statistical files directly to act as a control and augment our results. The most likely case is that our results reflect that teenagers are more likely to partner with those in their neighborhoods, and while high schools often reflect that, in this case desegregation policies simply changed the high school racial mix rather than the neighborhood demographics.