Online Program

Teen Sex and Propensity Score Matching: Will the Media Report Both? a Case Study
*Janet Rosenbaum, Johns Hopkins STD Center 

Keywords: propensity scores, matched sampling, media, adolescent health

Health studies vary in appropriateness of their methodology to address a target policy question. Statistics is popularly viewed as obscure and impenetrable, but without dissemination of statistical issues studies with more appropriate statistical methods may not prevail in the policy arena. The author published a matched sampling study on the sexual behavior of virginity pledgers in the journal Pediatrics in December 2008. Matched sampling is not widely practiced within social sciences. One might expect that the media would not report on methods at all. This presentation examines the tactics that the author used to encourage the media to report on methodological issues and the extent to which methodological issues became part of the public discourse in discussing the evidence about the sexual behavior of virginity pledgers.

The earlier work on virginity pledgers used logistic regression and found that on average pledgers delay sex by 18 months compared with non-pledgers. The author's Pediatrics paper used 3:1 exact and nearest-neighbor matching preferably within propensity score calipers with replacement using the R Matchit package and verified balance on 120 variables. The data used was a subset (n=3400) of the National Longitudinal Study of Adolescent Health, a federally-sponsored nationally representative survey of adolescent sexuality. The control variables were measured at wave 1 in 1995, the pledge was measured at wave 2 in 1996, and outcomes were measured at wave 3 in 2001. There were 289 pledgers and 3100 non-pledgers; matching yielded 645 similar non-pledgers.

Media interviews were conducted between December 22, 2008 and January 6, 2009. In interviews, the author stressed the similarity of pledgers and matched non-pledgers on over 100 variables in order to ``compare apples to apples' and analogized regression to turning an orange into an apple using arithmetic. Media sources including the Washington Post, the Today Show (NBC), the Wall Street Journal op-ed, and the evangelical magazine Christianity Today mentioned the study's methodology prominently within their reports about the study, including explicit mention of the 100 variables on which pledgers and matched non-pledgers were balanced.

Rigorous quantitative research can successfully sell its statistical strengths within the public arena.