Endogeneity in Observational Studies: An Example in the Association of Hormonal Contraceptive Use and Cervical Infection

• Activity Number: 215
• Type: Contributed
• Date/Time: Tuesday, August 13, 2002 : 12:00 PM to 1:50 PM
• Sponsor: Section on Statistics in Epidemiology*
• Author(s): Emelita Wong*+
• Affiliation(s): Family Health International
• Keywords: endogeneity ; simultaneous equations approach ; bias in observational studies

Abstract

While a randomized controlled trial (RCT) is the gold standard in assessing unbiased effect of an exposure on outcome, an RCT is not always the most ethically or logistically feasible study design. For example, in determining the association between use of hormonal contraceptive (HC) methods and acquisition of cervical infections, study participants cannot be ethically randomized to different HC methods. Thus, observational studies, either cross-sectional or cohort designs, are the usual study designs. Such designs, however, are likely to yield biased estimates of effects when study participants choose the exposure of interest as well as other covariates that may affect the exposure-outcome relationship. If overlapping unmeasured factors affect choice of exposure, as well as outcomes of interest, endogeneity is said to exist. To adjust for endogeneity, simultaneous equations approach, instead of single equation relating outcome and exposure, is the preferred method of estimating exposure effect. This paper illustrates a test for endogeneity (Bollen, Guilkey, and Mroz) and adjustment for such in the association of HC use and cervical infection in a cohort of over 800 women.