Report
Learning in sender-receiver games
Discussion paper (Center for Economic Research at Amsterdam), no. 9828, Center for Economic Research, Tilburg University
03/1998
Abstract
This paper compares a well-known stimulus-response (SR) model and a belief-based learning (BBL) model using experimental data from sender-receiver games. When the models are fitted to the data by maximum likelihood, the fits are good for both models. In contrast to Camerer and Ho [1997], we compare the models using a formal statistical procedure based on the Davidson and MacKinnon P-test for non-nested hypotheses. The motivation for using this test is that the models are naturally non-nested. Both models involve a certain adjustment parameter, which measures the importance of forgetting. Our results show that the outcome of the test is sensitive to the value chosen for the adjustment parameter. Depending on the value selected, the P-test favors the SR model, the BBL model or neither of the models. A point often overlooked in empirical work is that information from learning can only come from observations where learning occurs. A preliminary examination suggested that our results were affected to some degree by observations taken after behavior has converged. We adjusted the data for this convergence effect and found that the results where not markedly different from our original findings.
Details
- Title: Subtitle
- Learning in sender-receiver games
- Creators
- Andreas BlumeDouglas V DeJongGeorge R NeumannN E Savin - University of Iowa, Economics
- Resource Type
- Report
- Series
- Discussion paper (Center for Economic Research at Amsterdam); no. 9828
- Publisher
- Center for Economic Research, Tilburg University; Tilburg, Netherlands
- Number of pages
- 29 pages, [5] leaves
- Language
- English
- Date published
- 03/1998
- Academic Unit
- Economics; Accounting
- Record Identifier
- 9984963123502771
Metrics
2 Record Views