Self-reported smoking in online surveys - Prevalence estimate validity and item format effects

Objectives: We assessed validity of self-reported smoking prevalence estimates from an online sample, and explored the impact of different item response formats on estimates. Methods: Self-reported current smoking status was obtained from 110,837 respondents from the Harris Poll Online (HPOL) panel from April 2004 to January 2005. Current smoking prevalence was compared with national estimates from the 2004 Behavioral Risk Factor Surveillance System (BRFSS), 2003 National Health Interview Survey (NHIS), and 2001-2002 National Health and Nutrition Examination Survey (NHANES). All estimates were weighted to reflect the US population. A separate survey section measured smoking prevalence using randomly assigned response formats, including yes/no grid, multiple response, numeric box, category grid, and drop-down box formats. Results: 24.0% (95% confidence interval [CI] = 23.7-24.4) of HPOL respondents reported current smoking. BRFSS, NHIS, and NHANES estimates found 20.9%, 21.5% (95% Cl = 20.9-22.1), and 24.9% (95% Cl = 22.4-27.5), respectively, reporting current smoking. An additional 4.5% of NHANES respondents reporting not smoking had cotinine levels >= 15 ng/mL, indicating current smoking. Estimates of smoking prevalence varied by prevalence period and response format. Conclusions: Prevalence estimates obtained from the HPOL panel are comparable to those from national surveys. Online response format choices result in variation in estimated behavioral prevalence. Online surveys may be useful for public health surveillance of the US population.