Most Americans in a Berkeley experiment chose candidates and policies by one factor or a quick tally

Conjoint experiments, the format named in the paper's title, are usually read in aggregate ^[8][10]. Across many paired profiles, researchers estimate how much each attribute moves the choice, such as a candidate's military record or a migrant's English. Banavar and Bansak built a cognitive model aimed at the individual instead, at decision-making midprocess ^[10]. According to the phys.org account, the model works from behaviour alone, inferring a person's values and reasoning from patterns across hundreds or thousands of decisions ^[11].

The individual view matters because a pooled average can hide strategy. Suppose each of the six Senate attributes shows some pull in the pooled data ^[4]. That fits a sample of careful weighers. It fits equally well a sample in which each person picks one favourite attribute and ignores the rest, with different people favouring different ones. The paper's method classifies strategy person by person to tell those apart ^[8].

Bansak said that with complicated choices "we tend to assume they're weighing all the factors," and added: "we find that most of them aren't. Most commonly, people rely on a mental shortcut, effectively deciding based on the single factor they care about most and largely ignoring the rest." ^[3] In the Senate task, a single-factor chooser set aside five of the six attributes on the screen ^[1].

The study's own framing starts from a democratic premise: citizens with access to information will make reasoned decisions ^[14]. In this task, access was complete. Migrant profiles appeared side by side with six factors each, among them education level, age and English proficiency ^[13], and deciding on one factor was still the most common pattern the authors found ^[2]. The result concerns how people use information they already hold. The phys.org account does not report the share of participants in each strategy group, or describe any condition that varied how much information people saw.

Banavar, a Berkeley cognitive scientist and postdoctoral researcher in political science ^[9], said the finding does not mean people are irrational ^[6]. "The kind of rationality that we define in the paper is quite different from what many humans actually do in real life," she said. "We have these amazing brains, but they're not the infinite computational machines that would be necessary in order to do the kind of 'rational' thinking we imagine in our paper." ^[5]

I think she is right to separate the two. Full weighing is a modelling ideal, and missing it on a screen of fictional Senate candidates ^[4] is evidence about how people spend attention. It is weaker evidence about civic competence. The thing this doesn't tell you is whether the strategy mix survives a real candidate met over months of coverage.

Banavar drew the communication lesson herself. If a leader knows that climate opinion is dominated by pocketbook concerns, she said, that single factor might weigh heavily in political advertising, while complex science issues might be downplayed ^[7].