ÂÜÀòÉç¹ÙÍø

 : : : Posting

: : content

The conclusions in the article which Armstrong cites by Evanschitsky et al. (2007) which found that only 1.2% of empirical results in marketing are replicated (actually 0.53%, since only 44% of the replications replicated) are naïve and are unlikely to be heeded. The purpose of a “straight” replication is to find out whether a specific finding that supports or rejects a hypothesis is repeatable. However, this information is given by the /p/-value of the hypothesis test in the original study and unless the /p/-value is marginal there is no need to replicate (see my 2003 JBR article titled “Qualifying the Importance of Findings”). The purpose of replication “with extension” is to find out how generalizable the findings are. But Evanschitsky et al. ignore the generalizability of the original study’s findings. Generalizability is increased by replication of respondents (respondent heterogeneity) and replication of products or ads (stimulus heterogeneity). Editors’ widespread insistence on replication with nonstudents improves the former and reviewers are increasingly rejecting studies based on a single product or a single ad, which improves the latter. Single-stimulus studies where the single stimulus differs in the replication has to be the main reason for “failure” to replicate. This, plus the likelihood that only the poorer studies are chosen for replication in the first place. Or are the authors suggesting that “labs differ” or “experimenter effects” are responsible for discrepant findings? I don’t think so. Scott Armstrong and his colleagues don’t really understand what “replication” means and they continue to tell their horror story about the lack of replications and make totally misleading implications.