The trouble with estimates of priming.

This paper argues that the existing evidence for priming stands on a weak foundation. Although many experiments have found evidence of priming, their designs only randomly assign part of the treatment necessary to test for priming, and the part they do assign seems likely to influence the part they only observe. Consequently, the findings of these studies are potentially biased, probably upwards. The most troublesome of these biases results from the treatments in the studies causing people to adopt attitudes that are consistent with the dependent variable, usually presidential approval or vote choice. Although the amount of bias this causes is unknown, I illustrate its potential dangers by assuming a simple model of attitudes, calculating the potential biases, and showing that they can be very large. I also review attempts by researchers to avoid this bias, and find that they most likely fail. [ABSTRACT FROM AUTHOR]