A Meta-Analysis of the 2004 Campaign Polls: An Analogy to Practice and Publication in Psychology
The current practice of relying on single sample null hypothesis tests is being re-evaluated in the behavioral sciences. To highlight the issues raised by both sides of this discussion, a meta-analysis of The Gallup Organization's most recent U.S. Presidential election polling data was conducted. During the 2004 Presidential campaign, most pre-election polling percentages reported Bush ahead of Kerry, although the differences between the voters' preferences were typically within the margin of error. The meta-analyses used in this study showed significant differences between the two candidates' polling percentages, thus yielding a more accurate prediction than the conventional analysis which was based on single samples. These improved predictions provide support for a continued discussion about potential changes in statistical approaches to psychology.