Taking Serial Correlation Seriously: An Illustration from an Assessment of Temporal Trend of Racial Disparity in Chlamydia Prevalence Among Black and White Women

• Activity Number: 530
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Lin H. Tian*+ and Catherine L. Satterwhite and Jim R. Braxton and Samuel L. Groseclose
• Companies: CDC and CDC and CDC and CDC
• Keywords: Time series data ; serial correlation ; regression with ARMA errors ; temporal trend ; racial disparity

Abstract

We assessed the temporal trend of absolute disparity in chlamydia prevalence from 2006-2008 between black women and white women using two methods - standard linear regression (ignores serial correlation) and regression with ARMA error model (accounts for serial correlation). Racial disparity significantly decreased (-0.122% per two-month interval; standard error (SE) =0.052, P-value=0.019) using regression with ARMA error analysis, but not in standard regression analysis (-0.107% per two-month interval, SE=0.095, P-value=0.260). This suggests ignoring serial correlation in observed data can introduce substantial bias in regression coefficient standard error estimation and mislead statistical inference. In the example, ignoring correlation would have incorrectly led us to conclude that the female black-white racial disparity did not significantly improve over time.