Get full text report (pdf file; Read by ADOBE Acrobat Reader)
J Occup Health year 1998 volume 40 number 4 page 307 - 312
Classification Original
Title Using Pilot Data to Estimate Sample Size and Compare Question Forms for a Crossover Study*
Author Maxia DONG, Martin R. PETERSEN and Mark J. MENDELL
Organization National Institute for Occupational Safety and Health
Keywords Sample size, Pilot study, Crossover study, Intervention, Office worker, Indoor air, Questionnaire, Visual Analog Scale (VAS)
Correspondence
Abstract Using Pilot Data to Estimate Sample Size and Compare Question Forms for a Crossover Study: Maxia DONG, et al. National Institute for Occupational Safety and Health-A pilot study was conducted on thirty office workers to help determine if a scannable form of symptom severity questions would yield similar results as a nonscannable form. There were three goals of the pilot study: first, to observe if, in a questionnaire using two forms of Visual Analog Scales (VAS), questions using a scannable sequence of "boxes" for responses would elicit different mean responses than questions using unbroken "lines"; second, to observe if questions using a sequence of "boxes" would elicit different within person and week response variability than questions using unbroken "lines"; and third, to estimate the sample size needed for a crossover study, depending on the particular form of the question used, and the number of crossovers. The pilot study, consisting of three sequential weekly questionnaires, provided week, subject, and error variance components for each of three dependent variables from the two different VAS forms. Most of the calculations were performed with a log transformation of the data. For each VAS form, the number of subjects necessary for desired study power for each symptom was calculated. Based on this pilot study, neither the mean nor the within person and week variance component was consistently larger or smaller for the VASbox form than for the VASline form. The linear models analysis showed that the two forms filled out by the same person on the same day had similar mean values and were highly correlated for all symptoms (R2>0.95). Thus we chose the VASbox form because of scanner compatibility and estimated the required number of subjects for our full-scale study based on this chosen form.