# 11-153/1 (2011-11-03)

Adrian de Groot Ruiz, Radboud University Nijmegen; Theo Offerman, University of Amsterdam; Sander Onderstal, University of Amsterdam
Cheap Talk, Refinement, Selection, Experiment, Neologism Proofness, Announcement Proofness, ACDC
JEL codes:
C72, C92, D82, D83

We test the Average Credible Deviation Criterion (ACDC), a stability measure and refinement for cheap talk equilibria introduced in De Groot Ruiz, Offerman & Onderstal (2011b). ACDC has been shown to be predictive under general conditions and to organize data well in previous experiments meant to test other concepts. This experiment provides the first systematic test of whether and to which degree credible deviations matter for the stability of cheap talk equilibria. Furthermore, it tests ACDC in a new setting. We also introduce a neologism dynamic to explain the main dynamic characteristics of our data. Our main result is that credible deviations matter and matter gradually, as predicted by ACDC. In addition, our data support the predictions of ACDC in settings where existing concepts are silent. Finally, we test the prediction derived in De Groot Ruiz, Offerman & Onderstal (2011a) about bargaining power and information transmission. We find that, as predicted, less information is transmitted as the Sender's relative power decreases.