How General is the Strength-Weight Bias in Probability Updating?
59 Pages Posted: 13 Jul 2020
Date Written: June 19, 2020
Experimental research by Griffin and Tversky (1992) has identified a strength-weight bias in probability updating: when assessing sets of binary signals, individuals excessively focus on the strength or extremeness of the given evidence while insufficiently accounting for the weight or reliability of the information. We challenge the generality of this prominent experimental finding and argue that the simple two-state setting explored by Griffin and Tversky (1992) is an extreme and peculiar case with respect to the normative importance of information weight and strength. We explore the range of reasonable information settings and demonstrate that information weight (strength) is much less (more) relevant in other environments. The findings of Griffin and Tversky (1992) could thus be driven by a general strength-weight bias, but also by an insensitivity to the specific information environment. To explore these competing hypotheses, we conduct a balls-and-urns experiment in which we systematically vary the information environment. We find that there is no consistent strength-weight bias but rather insensitivity to the specific information environment. Given that many real world information environments do not resemble the simplified two-state setting considered by Griffin and Tversky (1992), our results caution against an indiscriminate transfer of their findings to a broad set of real world applications.
Keywords: Information Weight, Over- and Underinference, Judgment Biases
JEL Classification: C91, D91
Suggested Citation: Suggested Citation