Random Matching with Observable Past Actions: Experimental Tests of the Reputation Effect

Abstract

我們對實驗對象提供不同層面的過去行為的歷史，以及採用不同長度的實驗，來比較合夥人與陌生人在公共財實驗中對公共財捐獻的差異。我們的實驗數據顯示，在二十回合的實驗中，當群體捐獻的歷史不對陌生人揭露時，在大多數回合中合夥人的捐獻都比陌生人多。若群體捐獻的歷史對陌生人揭露，則此一差異縮小，若我們再進一步對合夥人及陌生人都揭露個人捐獻的歷史，則此一差異終將消失。然而，在兩個序列的十回合實驗中，即使個人捐獻的歷史為已知，合夥人的捐獻仍較陌生人多。這些觀察建議，若過去行為的資訊在合夥人與陌生人間較對稱也較詳細，則合夥人與陌生人間捐獻的差異將縮小，但實驗回合數必須夠長才能使此一差異明顯下降。 We compare the difference between giving by Partners and giving by Strangers in experiments with different dimensions of histories of past actions and different lengths of trials. In experiments with twenty trials, our experimental evidence shows that Partners contribute more than Strangers in most rounds when the histories of group contributions are not revealed to Strangers. The difference becomes smaller if this information is made available to Strangers, and eventually disappears if both Partners and Strangers are further informed about the histories of individual contributions. However, in experiments with two sequences of ten trials, Partners still contribute more than Strangers even if the histories of individual contributions are known. These observations suggest that more symmetric and detailed information regarding past actions reduces the difference between giving by Partners and giving by Strangers, but sufficient repetitions are necessary for this difference to be markedly reduced.