Psychologists often conclude from research subjects’ behavior in psychological experiments that humans are irrational.
New research indicates that humans are in fact quite rational; they just do not trust what people in lab coats tell them. The research suggests that by taking doubt into account, psychologists have the opportunity to strengthen the predictive power of many commonly used models and potentially better understand human behavior.
The research by Vanderbilt University computer scientist and psychologist David Noelle and University of California-San Diego psychologists Craig McKenzie and John Wixted is published in the September issue of the Journal of Experimental Psychology: Learning, Memory, and Cognition.
Noelle argues that accounting for doubt from the start of an experiment is more effective than trying to convince subjects that what they are being told is true.
“Part of the message of this paper is rather than banging your heard against the wall repeatedly trying to come up with a mechanism that pushes their level of trust in your instructions up as high as you can, why not just incorporate a parameter capturing their level of belief in those instructions into your account of their performance?” he said.
To account for doubt, Noelle and his colleagues developed a new model, dubbed the “trust model,” to interpret subjects’ performance in a simple task. The trust model represents the first time an attempt has been made to quantify and account for the impact of doubt on a subject’s performance in an experiment.
“The trust model describes exactly what a perfectly rational person would do if they harbored some amount of doubt in the task instructions,” Noelle said. “The good fit of the trust model to actual performance suggests that humans might be behaving perfectly rationally in these experimental settings.”
Noelle suspects that many subjects, who are often psychology students, experience doubt because they have been taught about classic psychological experiments where researchers did in fact attempt to mislead the subjects.
“It may be that subjects are responding in a perfectly adaptive way if you see it through their eyes: this strange experience of walking into a laboratory and meeting someone who tells them to do something they’ve never done before,” he said. “Researchers assume when subjects don’t behave as we expect them to that there’s something wrong with their rational reasoning process. Instead, we need to consider that maybe they don’t believe the rules of the game and are hedging their bets in some way. We believe their performance can be explained in terms of that hedging, that skepticism and lack of trust.”
Noelle and his colleagues used a relatively simple experiment to test their theory. Undergraduate psychology students were shown pairs of letters, “B,” “E” and “R,” and numbers, “5,” “6” and “8.” During the first phase of the experiment, they were asked to separately rate their confidence in the appearance of a “B” and the appearance of a “6.” This phase gave the researchers baseline information about how well the students could detect the “B” and the “6” individually.
For the second phase, the students were first instructed that a “B” and a “6” would never appear together. Despite this instruction and the fact that “B” and “6” never appeared together, many still reported seeing the pair.
When questioned after the experiment about their belief in the instructions, 40 percent of the students said they did not fully trust that the researcher was being honest when he said the two characters would not appear together.
The researchers then designed the trust model, which factored in a level of doubt, to interpret the students’ behavior in this task. The level of doubt was a free parameter adjusted to fit the actual performance of each subject during the second phase. The researchers then looked at the relationship between the level of doubt as expressed by this parameter and the actual doubt reported by the student.
“We expected that the students who expressed doubt at the end of the experiment would have exhibited this doubt in their performance, as reflected in the value of the doubt parameter, and that is what we found,” Noelle said.
Funding from the National Science Foundation supported the research. Noelle is an assistant professor of computer science and psychology and a member of the Vanderbilt Center for Integrative and Cognitive Neuroscience.