An Experimental Investigation of Updating under Ambiguity

We formulate new hypotheses that take advantage of information updating in order to discriminate between the two major specifications of multi-prior ambiguity models: ``kinked'' and ``smooth''. In particular, across comparable decision settings, we examine the effects of adding or trimming out certain priors, updating the weight on particular beliefs, changing the payoff for a single potential state, and modifying the distribution within certain priors. Our results show that the kinked specification does well in consistently predicting choices from 68% of participants, and the smooth specification predicts well for just 10%. We find evidence that people may use a compound lottery as one of their priors, subjects are insensitive to information that the best prior is more likely, and people place lower values on ambiguous lotteries that are relatively more complex. Our experimental methods are likely to be useful in other contexts, as they allow for simple tests of decision-making under ambiguity without placing restrictions on the weights participants place on priors, or reliance on comparisons to decision-making under risk.