Risky choice: Selecting between certain and uncertain outcomes.

Abstract

In risky choice situations, individuals are called risk-prone if they select an option with uncertain consequences, and risk-averse if they select an option with certain consequences. When the uncertain option involves either variability in the delay to reinforcement or a probabilistic reinforcer (one which might or might not occur), nonhuman subjects typically exhibit strongly risk-prone choices. This tendency can be explained by the hyperbolic-decay model, which describes how a reinforcer's effectiveness decreases with increasing delay. When the uncertain option involves variability in the amount of reinforcement, some studies have found that nonhumans are risk-averse when food sufficient to meet their needs and risk-prone when it is not. Studies with human subjects have typically found risk-averse choices when they choose between certain and uncertain reinforcers, but the procedures used in these studies differ in several ways from those used with nonhumans. In laboratory and real-world cases where humans experience a series of trials with real reinforcers, their choices tend to be more similar to those of nonhumans. Key Words: Risky Choice Situations; Hyperbolic-Decay Model; Risk-Prone; Risk-Averse; Certain and Uncertain Reinforcers. ********** In everyday life, both humans and nonhumans must frequently choose between alternatives that are almost certain to deliver reinforcement and alternatives with uncertain consequences. A man might consider whether to put his savings in a bank account or in a risky investment that could produce either a large return or a loss of all his money. A student athlete might have to choose between a small college where she is certain to be a starter on the basketball team or a university with a stronger program where she might not even make the team. A predator might have to decide whether to chase a small animal that would be an easy catch or to wait for larger prey that might or might not appear. Because such choices are commonplace, and because they pose interesting dilemmas, many laboratory studies have been conducted to investigate how individuals behave in these situations. For convenience, we can call the alternative that is certain to deliver a reinforcer the certain option, and the alternative with uncertain consequences the risky option. One common classification scheme categorizes choices as risk-prone, risk-neutral, or riskaverse, depending on whether, in a series of repeated choices, the risky alternative would tend to deliver less, equal, or more reinforcement in the long run. To take a simple example, suppose that on each trial a rat must choose between one food pellet delivered with certainty and four food pellets delivered with a probability of 25%. Both of these options would deliver an average of one pellet per trial in the long run. Therefore, if the rat showed a preference for the risky option, its behavior would be categorized as risk-prone. If it chose both options about equally often, its behavior would be called risk-neutral, and if it showed a preference for the certain option, its behavior would be called risk-averse. In different situations and with different subjects, all three types of behavior have been observed. One promising theoretical approach to risky choice, first proposed by Rachlin, Logue, Gibbon, and Frankel (1986), is the hypothesis that probabilistic reinforcers are functionally equivalent to delayed reinforcers. This idea seems very reasonable, because if a reinforcer is delivered on only a certain percentage of the trials, there will often be a delay between a choice response and the eventual delivery of a reinforcer. For instance, if a risky option delivers food to a pigeon on only 25% of the trials, the pigeon will usually have to wait through several trials before it receives food. An advantage of this hypothesis is that there is a substantial body of knowledge about how delay of reinforcement affects choice, and it may be possible to apply this knowledge to choices involving probabilistic reinforcement. …