M > 1

Abstract

Increasingly, communication expm'ments are incorporating replication factors for the purpose of controlling confounds and increasing generalizability. If replications are considered to be samples of possible treatment implementations, treating the replication factor as random is more uppropriute than treating it asfixed. Study 1 shows that treating sampled replications as afixed effect leads to potentially swious alpha inflation in the test of the treatment effect while treating sampled replications as random controls alpha at its nominal level. Study 2 addresses a common objection to treating replications as random: the argument that to do so will lead to unacceptably low power in statistical testing. Although experiments with very few replications are likely to be deficient in power, the results c f Study 2 establish that power can be improved to an unexpected d e p e bya relatively modest increase in the numberofreplicatwns. Sally lackson is Associate Professor of Communication at the University of Arizona. Dale E.Brushers is now an instructor in the Department of Communication at Ohio State University.The authors thank Bert Meuffels for useful comments and suggestions on an earlier version of this article. Special thanks go to Daniel J. OKeefe for supplying effect s u e data and other meta-analysis statistics and for helpful comments on the manuscript. The support of the Netherlands Institute for Advanced Study in the Humanities and Social Sciences and of the University of Oklahoma Provost's Office is gratefully acknowledged.

Gain-Framed Messages Do Not Motivate Sun Protection: A Meta-Analytic Review of Randomized Trials Comparing Gain-Framed and Loss-Framed Appeals for Promoting Skin Cancer Prevention

“…That is, rather than conducting an experiment examining just one pair of messages (as is the current practice), a researcher could instead compare multiple message pairs. Such designs require statistical analyses adapted to the presence of message replications, but obviously have substantial benefits with respect to establishing dependable generalizations about message effects [58,59,60]. Thus in future research aimed at examining the relative effectiveness of two message kinds for encouraging skin cancer prevention (or any other behavior), the research community need not wait for replications to be published; researchers can-and arguably should-be asked to include message replications in their designs, so as to provide evidence that any claimed effects generalize beyond that particular pair of messages.…”

Methylnadimide End-Capped Polyimides for Use in High Temperature Composites—A New Approach to Their Synthesis

“…However, the position of the endotherm attributed to the retro Diels-Alder reaction is found to be somewhat variable in different models. In the case of the DDS model ms showed the evolution of methylcyclopentadiene in this temperature range [14].…”

Wprowadzenie do analizy wariancji z czynnikami losowymi

“…Badacze uznają próbę badanych osób jako źródło losowej zmienności nawet wtedy, gdy próba uczestników nie jest losowa, co w badaniach eksperymentalnych zdarza się bardzo często. Na podobnej zasadzie należy potraktować próbę replikacji (Jackson, Brashers, 1994b). Kwalifikacja replikacji jako czynnika losowego umożliwi przeprowadzenie dla replikacji "oddzielnego" wnioskowania statystycznego -analogicznego do wnioskowania, jakie przeprowadza się wobec badanych jednostek z myślą o tym, by wnioski objęły populację jednostek.…”

“…A zatem pod względem sprawowania kontroli nad błędem I rodzaju, pożądanymi schematami eksperymentalnymi będą te, które replikacje traktują jako czynnik losowy, jako że pozwalają one utrzymać ten błąd na założonym poziomie alfa (Jackson, Brashers 1994b). Taki sam wniosek sformułowali Kenny i Judd (1986: 428), pisząc, że remedium na brak niezależności pomiarów jest uznanie czynnika grupującego obserwacje za losowy i włączenie go do analiz.…”