This paper examined whether the sample rotation scheme of the Current Population Survey (CPS) results in an underestimation of current smoking prevalence in the Tobacco Use Supplement to the Current Population Survey (TUS-CPS). The TUS-CPS has been administered as part of the CPS, which has eight rotation groups of households in each month that are repeatedly interviewed based on a sample rotation scheme. Previous research has found that even though all eight rotation groups in the CPS are independent random samples of the population, some estimates, such as unemployment rates, tend to be significantly higher in the first rotation group than among other rotation groups. The probit regression results of this paper showed that although current smokers are more likely to attrite than nonsmokers in all years of the TUS-CPS, for the six waves of TUS-CPS before 2003 there is no evidence that current smoking prevalence estimates were significantly affected by the rotation scheme of the CPS. For the three waves of TUS-CPS since 2003, however, the results showed that current smoking prevalence has been underestimated likely due to differential panel attrition. It appears that rotation group bias in these waves was caused by the substantially increased number of additional questions current smokers had to answer.
Read full abstract