The 130th Annual Meeting of APHA

Abstract #47486

Jessica R. Schumacher, MS¹, Hao Tang, MD, PhD², and David Cowling, PhD². (1) Tobacco Control Section, California Department of Health Services, 601 North 7th Street, PO Box 942732, MS 555, Sacramento, CA 94234-7320, (916) 324-3719, jschuma1@dhs.ca.gov, (2) California Department of Health Services, Tobacco Control Section, P.O. Box 942732, MS 555, 601 N. 7th Street, MS 555, Sacramento, CA 95814

Context: Weighting and sample design effect are two key elements in tobacco use epidemiology, which measures smoking prevalence, secondhand smoke exposure, smoking cessation rates and other indicators by using large-scale surveillance study. The understanding and interpretation of these two concepts are sometimes beyond methodological concerns because different prevalence rates and variances generated by different methodologies can provide different point of view or evaluation of a given public health program.

Methods: Three waves of California Tobacco Survey (CTS 1990, 1996 and 1999) are used to illustrate the difference of weighting methodologies and the manipulation of sample design effects. CTS were analyzed using weights that accompanied in the data set as well as the new weights that generated from the Census data. Three different computing software packages, SAS (PROC MEANS and PROC SURVEYMEANS), SAS Callable SUDAAN and WesVerPC, are used to compute smoking prevalence rates and variances.

Results: The re-weighted smoking prevalence rates are comparable to the original weighted data. However, if each wave is weighted to different standard population, the trend is different from the results yielded from the data that are weighted to one standard population. The variance estimates are almost identical when sample design methods are specified correctly during the computation using different software.

Conclusion: It is important for public health professionals to understand the function of weighting and sample design effect when key indicator such as prevalence rate is used for decision-making and evaluation in public health settings.

Learning Objectives:

• Discuss the function and importance of weighting and sample design effect in tobacco use epidemiology by using a large sample surveillance data set as example.
• Develop statewide smoking prevalence estimates by performing analyses using SAS, SUDAAN and WesVarPC to identify the difference between these packages and weighting methodologies.

Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.