Researchers frequently use constructed week samples to approximate content for larger populations of textual data in content analysis projects. To date, this sampling method has not been validated in longitudinal contexts necessary for the conduct of large-scale health communication research. This study uses Monte Carlo bootstrap sampling to determine the number of constructed weeks necessary to accurately estimate one- and five-year population values for different types of variables in a quantitative content analysis. Five years (1999-2004) of four different daily newspapers were coded for four variables that varied on type (count vs. rating), amount of missing data, and distribution (normal vs. non normal). Results suggest that sampling a minimum of six constructed weeks was most efficient for both time frames. Missing data lowers sampling precision, although a correction can be calculated if the amount of missing data can be estimated. Using an efficient method of sampling newspapers such as constructed week sampling can help communication researchers to more easily study health coverage in the media.
Bibliographical noteFunding Information:
This project was funded through the U.S. Centers for Disease Control and Prevention contract U48/CCU710806. The authors thank Matthew Kreuter, Robert Logan, Brooke Asbury, Caren Bacon, Darigg Brown, Vicki Collie-Akers, Kristy Davidson, Mark Graves, Heather Jacobsen, Keri Jupka, Susan Lukwago, Vinay Reddy, Jyothi Varanasi, and Anitha Vempaty for their assistance in planning and carrying out the everyday activities of the study. Some statistical analyses and other support were provided by the Saint Louis University Center of Excellence in Cancer Communication Research (1P50CA095815).