Exploring how counterfactual thinking and memory influences decisions

The research sheds new light on our adaptive ways of inferring value on the unknown.

By Sarah Steimer

We don’t know what we don’t know — but our brains have a way of making inferences about the unknown, based on memory and counterfactual thinking. A new study published in the journal Cognition pushes forward our understanding of how memory interacts with inference and guides adaptive behavior, especially in novel environments. Specifically, the research showed that people generalize inferred values of unchosen options to novel items within the same semantic category. 

Akram Bakkour, assistant professor in the Department of Psychology and director of the Bakkour Memory and Decision Lab, coauthored the study alongside doctoral student Xinyue Li and MA student Zimeng Wang. They began with established findings: Previous work has shown that when people receive a positive outcome for a particular choice, the value of that item increases and the value of the alternative decreases. For example, if a person chooses an apple instead of carrots, and the apple was delicious, their perceived value of the apple increases. At the same time, this person’s perceived value of the carrots decreases, even though they never ate the carrots, which could have been even more delicious than the apple. The inferred value of the unchosen item is acquired through counterfactual thinking. 

Other research has shown that the positive value of the apple — acquired through direct experience — spreads to items connected to it in memory, like other fruit. The research team took the logical next step of testing whether the inferred value of an unchosen option — that was never experienced and acquired through counterfactual thinking — can drive behavior similarly to values acquired through direct experience.

For their study, participants were asked to make a choice between two images, which would later be placed on stamps. Participants were also asked to deliberate on the choice — in one trial, participants were even asked to justify their decision, be it based on aesthetics or other preferences. 

“Ultimately, it doesn't really matter what they end up justifying,” Bakkour explains. “It's a way for us to get them to think about that choice.”

The result, for participants, were chosen and unchosen options from different semantic categories (say, fruits and vegetables). At a later stage, they received feedback that the image they chose will be placed on either a high value or a worthless stamp. They are never told the value of the stamps featuring unchosen images. 

“If they chose the apple, and the apple ended up being placed on a valuable stamp, we know that its value is going to spread to all fruit,” Bakkour says.

Next, participants are asked to choose novel images that they also believe would be valuable, and in this instance they often select ones in the same semantic category as their original, valuable choice (if their first choice was an apple, and the apple was featured on a high-value stamp, they’re apt to choose novel images of other fruit). 

“If the apple was initially chosen over carrots, what happens to carrots?” Bakkour says. “Presumably, its value is going to decrease. But does that decrease in value then also spread to all vegetables in the subsequent task?”

Lastly, participants were given novel options (not encountered in the experiment) to choose from.

“One of them was from an unchosen category that should be inferred to be better and one from an unchosen category that should be inferred to be worse than expected,” Bakkour explains. “And the question is, are they biased? Are they choosing novel exemplars of these unchosen categories that are inferred to be better or not? Are they making this inference on an inference in this task? We see essentially that yes, they do.”

Specifically, the researchers found that participants were more likely to choose or avoid novel items from the same category as an unchosen option that had been paired with a non-valuable or valuable chosen item, respectively.

The results, Bakkour says, suggest that people make inferences to not only guide their behavior based on direct experience, but also to guide their behavior pretty broadly in novel contexts, based on counterfactual thinking.

“It tells us something about the way in which values that are built through direct experience, and values that are inferred through indirect experience — through just thinking about what the alternative could have been — are used kind of similarly,” he says.

Although a person cannot know what the outcome of the alternative was, they inferred it via a sophisticated cognitive process. Now, the research shows that people take it a step further: They treat it as if it were value acquired through direct experience.

“Information is power always,” Bakkour says. “But with a limited cognitive system, you're going to fill in the gaps in information that you don't have. Generally speaking, this is pretty adaptive.”