Causal Induction from Continuous Event Streams: Evidence for Delay-Induced Attribution Shifts

Contemporary theories of Human Causal Induction assume that causal knowledge is inferred from observable contingencies. While this assumption is well supported by empirical results, it fails to consider an important problem-solving aspect of causal induction in real time: In the absence of well structured learning trials, it is not clear whether the effect of interest occurred because of the cause under investigation, or on its own accord. Attributing the effect to either the cause of interest or alternative background causes is an important precursor to induction. We present a new paradigm based on the presentation of continuous event streams, and use it to test the Attribution-Shift Hypothesis (Shanks & Dickinson, 1987), according to which temporal delays sever the attributional link between cause and effect. Delays generally impaired attribution to the candidate, and increased attribution to the constant background of alternative causes. In line with earlier research (Buehner & May, 2002, 2003, 2004) prior knowledge and experience mediated this effect. Pre-exposure to a causally ineffective background context was found to facilitate the discovery of delayed causal relationships by reducing the tendency for attributional shifts to occur. However, longer exposure to a delayed causal relationship did not improve discovery. This complex pattern of results is problematic for associative learning theories, but supports the Attribution-Shift Hypothesis