Causal and diagnostic inferences: A comparison of validity

Abstract

Decision-support technologies are founded on the paradigm that direct judgments are less reliable and less valid than synthetic inferences produced from more “fragmentary” judgments. Moreover, certain types of fragments are normally assumed to be more valid than others. In particular, judgments about the likelihood of a certain state of affairs given a particular set of data (diagnostic inferences) are routinely fabricated from judgments about the likelihood of that data given various states of affairs (causal inferences), and not vice versa. This study was designed to test the benefits of causal synthesis schemes by comparing the validity of causal and diagnostic judgments against “ground-truth” standards. The results demonstrate that the validity of causal and diagnostic inferences are strikingly similar; direct diagnostic estimates of conditional probabilities were found to be as accurate as their synthetic counterparts deduced from causal judgments. The reverse is equally true. Moreover, these accuracies were found to be roughly equal for each causal category tested. Thus, if the validity of judgments produced by a given mode of reasoning is a measure of whether it matches the format of human semantic memory, then neither one of the causal or diagnostic schema is a more universal or more natural format for encoding knowledge about common, everyday experiences. These findings imply that one should approach the “divide and conquer” ritual with caution; not every division leads to a conquest, even when the atoms are cast in causal phrasings. Dogmatic decompositions performed at the expense of conceptual simplicity may lead to inferences of lower quality than those of direct, unaided judgments.