Data Sources and Analyses

This research study used data from national polls conducted by ABC News/Washington Post; CBS News; CNN/USA Today/Gallup; FOX News/Opinion Dynamics; and Newsweek. Some polls included in this study as ABC News/Washington Post were done independently by either organization and not together with the partner. Some polls included as CNN/USA Today/Gallup were actually conducted independently by Gallup, while some identified as CBS News were done jointly with the New York Times, and others were done independently.

Presidential approval ratings from these five organizations were gathered for the period 1997 to 2003. Any data from polls conducted entirely in one day were excluded. All polls were conducted by telephone among adults in the general population ages eighteen and older, except for FOX News polls, which were limited to registered voters. The principal sources of approval data for this study were the PollingReport.com web site and the University of Connecticut's Roper Center iPoll databank. Historical Gallup in-person approval ratings for 1953-88 were taken from Presidential Approval: A Sourcebook, by George C. Edwards III, with Alec M. Gallup.

Spreadsheets were created to analyze data by each organization and included the following information:

• Sample size and description
• Dates of interviewing
• Percent approve, percent disapprove, and percent no opinion (that is, "don't know" or "refused" responses)
• Quarterly averages calculated from individual polls within each quarter
• Yearly averages calculated from quarterly averages within each year

Quarterly and yearly averages were used for most of the analysis in this study. Quarterly averages were computed by averaging approval, disapproval, and no opinion ratings for individual polls within each quarter. Yearly averages were then calculated using the quarterly averages within each year. Tables summarize yearly averages for each president. Graphs using quarterly averages and scatter plots using individual poll results were also created to help with analysis.

An analysis of variance with post-hoc Bonferroni comparisons was run on quarterly averaged results in order to test for differences in approval, disapproval, and DK ratings among the five organizations. To test for associations between DK level and approval or disapproval ratings, linear regressions were run based on individual poll data.