This post is based on the paper: “The role of cognitive abilities in decisions from experience: Age differences emerge as a function of choice set size,” by Renato Frey, Rui Mata, and Ralph Hertwig that appeared in Cognition 142 (2015) 60–80.
People seldom enjoy access to summarized information about risky options before making
a decision except for things like weather forecasts that explicitly state a probability. Instead, they may search for information and learn from the environment—thus making decisions from experience. Many consequential decisions—including health care choices, finances, and everyday risks (e.g., driving in bad weather; crossing a busy street)—are made without full knowledge of the possible outcomes and their probabilities so we must make decisions from experience. According to the authors, the mind’s most notable transformation across the life span is a substantial decline in processing speed, working memory and short-term memory capacity —all components potentially involved in search and learning processes.
Frey et al suggest that a fundamental building block of decisions from experience is information search. Ceteris paribus, the more a person searches (explores), the better her estimate of an option’s value. Reliance on small samples can prompt systematically higher or lower levels of risk taking relative to those observed in decisions from description (in which outcomes and probabilities are explicitly presented, as in the traditional gambling tasks ubiquitously investigated by psychologists and economists). One reason is that small samples may not include rare but highly consequential events and, if they do, they may under represent them. Depending on whether these rare events are desirable or undesirable, experienced-based choice will involve either more or less risk aversion relative to description-based choice.
The researchers used models to set up their experiments. They not only used what they called a Neo-Bernouillian model like cumulative prospect theory, but also learning models. A learning model assumes that choices are based on expectations that are updated from trial to trial; that is, from draw to draw in decisions from experience. Specifically, an expectation of an option’s payoff is updated by adding a prediction error to the previous expectation, weighted by an individual learning rate. The prediction error (post the Prediction Machine) is the difference between the old expectation of an option’s payoff and the most recently experienced outcome.
One of these learning models, the instance-based learning model (IBLM), deserves particular attention according to Frey et al because it makes detailed assumptions about specific cognitive processes, such as forgetting and memory retrieval. It builds on the ACT–R cognitive (post ACT-R) architecture and its assumptions regarding the role of frequency and recency in human memory. The IBLM implies that a person faced with outcome sequences A and B (i) updates the activation of experienced outcomes (i.e., the instances) during the sampling process. In order to make a choice at the end of the sampling process, she (ii) retrieves the instances associated with an option from memory. The probability of retrieving instances from memory is, in turn, a function of their activation, and activation is a function of the instances’ recency and frequency of observation. The person then (iii) determines the blended value of each option based on the probability of retrieving the associated instances and their corresponding outcomes, and finally (iv) chooses the option with the higher blended value.
The researchers analyzed younger and older adults’ decisions from experience on two levels, namely, search and choice. They reported three studies that used a sampling paradigm to investigate younger (M = 24 years) and older (M = 71 years) adults’ decisions from experience. In Study 1 (N = 121) participants made 12 decisions between pairs of payoff distributions in the lab. Study 2 (N = 70) implemented the same paradigm using portable devices, collecting 84 decisions per individual over a week. Study 3 (N = 84) extended the sampling paradigm by asking participants to make 12 decisions between two, four, and eight payoff distributions (in
the lab). Overall, the behavioral results suggest that younger and older adults are relatively
similar in how they search and what they choose when facing two payoff distributions
(Studies 1 and 2). The results from Study 3, in contrast, suggest that age differences become evident as information load and, by extension, the demands on cognitive resources increase.
Although younger and older adults potentially still rely on the same updating strategies (e.g., the simple delta-learning rule) with more than two options, they have to keep track of their current evaluations of multiple (up to eight) payoff distributions. Arguably, this task is
cognitively much more taxing than the tracking of two options. Consequently, older adults may aim to reduce complexity and cognitive burden by decreasing search effort per option. There is also an order effect for older adults. In other words, they seem to suffer less from choice overload if they have the chance to start with few options and work toward the more complex decision problems, as compared to when they have to start with the complex decision problems right away.