The last thing they did

The textbook says the opposite. Solomon Asch's 1946 experiments, in which people heard lists of personality traits in different orders, became the foundation of research on how we perceive others.¹ The lesson most people took from that tradition is that first impressions anchor and later information gets bent to fit them. Both are true, and the difference between them explains what the afternoon is doing.

The difference is when you decide. In the model Hogarth and Einhorn proposed in 1992, as a 2025 study summarizes it, a judgment formed once, at the end of a short, simple run of evidence, tends to favour what came first; a judgment revised after every new piece tends to favour what came last.² That study, using legal evidence, found the split: no order effect when people judged at the end, and a recency effect when they updated step by step.²

Meeting a stranger is the first kind of judgment. Knowing someone is the second, taken to its limit: an opinion revised after every encounter, for years, never closed. A first impression and a last-interaction bias are the same machinery run under two conditions. Judge once and the first item anchors. Keep the judgment current and the newest item weighs most.

There is a second effect. Jenifer Siegel, Molly Crockett and colleagues modelled how people learn about others' morality from their choices and found an asymmetry: beliefs about people who behave badly are more uncertain, "and therefore more volatile", than beliefs about people who behave well.³ Bad behaviour does not install a confident new verdict. It unsettles the old one.

An unsettled belief is one that every new observation moves further. So the last interaction looms large not because it rewrote the person but because it made you unsure of them, and unsure judgments are steered by whatever comes next. That is why the week after a falling-out feels decisive in both directions: one more cold reply and the decade really does flip; one decent conversation and much of it comes back.

The authors point to the hopeful side of their own result: the same volatility "could facilitate forgiveness when initial bad impressions turn out to be inaccurate."³ The bias that lets one afternoon erase a decade is also the reason one apology can restore it. A judgment that never updated would protect the decade from the bad afternoon, and it would also make the person unable to change in your eyes.