Why Do People Prefer Randomisation? An Experimental Investigation

Abstract

The standard assumption in economics is that people have well-defined preferences and are able to specify their preferred choice. This is usually adopted by experimenters, when eliciting preferences, by asking subjects to choose between two options: "I choose option A" or "I choose option B". Recently there has been a tendency for experimenters to add a third option to these standard two options, namely a middle column, usually labelled "I am not sure what to choose". The implications for subjects of choosing this middle column vary across experiments. Some experiments provide no direct financial implications: what is ’played out’ at the end of the experiment is not influenced by subjects choosing this middle column. In other experiments, if the middle column has been checked, then the payoff is determined by a randomisation of A and B. I report on an experiment, which adopts this latter incentive mechanism, and ask the question as to why people might choose this option, that is "why do they prefer randomisation?" I explore four distinct stories and compare their goodness-of-fit in explaining the data. The first story is that the decision-maker (DM) has strictly convex preferences and actually prefers a mixture of A and B, and the DM has random preferences in every problem; the second is that the DM prefers a mixture of A and B only if it gives the highest utility, but the DM may tremble in expressing his or her preference; the third is that the DM cannot distinguish between A and B unless their difference exceeds some threshold; the fourth is that the DM actually prefers to delegate the choice (to the coin), shifting the ‘responsibility’ to the coin, though the DM may tremble in expressing his or her preference. My results show that the first and the third have the most empirical support. I include a discussion of whether my results have anything to say about preference imprecision.

Keywords: individual preferences, randomisation, risk, laboratory experiment

JEL Classification: C91, D81, D01