The predictive ability paradox in behavioral accounting research

The purpose of this brief paper is to illustrate a problem with certain “types” of behavioral experiments predicated on a test of predictive ability. Such studies possess the following common characteristics: (1) They employ the implicit prediction models of individual human subjects; (2) They utilize as criteria for assessing predictive usefulness events whose outcomes cannot be independent of expectations about them (e.g. bankruptcy); (3) They use historical values for both criterion events and treatment factors. The difficulty with such experimental designs lies in the necessity to assume unalterable prior behavior in order to utilize a known value for the criterion event. Such an assumption makes it difficult to demonstrate the predictive usefulness of any information treatment factors.