We borrow our current feelings as a read on whatever we are judging, even when the mood came from somewhere else entirely. It is a fast, useful shortcut, but an easy one to mislead.
The core idea is affect-as-information (Schwarz & Clore, 1983). Faced with a judgment that would take real effort to compute, like "how satisfied am I with my life?", people quietly swap in an easier question: "how do I feel right now?" The current feeling becomes a stand-in for the answer. This is fast and often sensible, because our feelings really do summarise a lot of experience. The problem is that a feeling does not come with a label saying where it came from. So a mood produced by one thing gets read as a verdict on another.
That is why the source of the feeling matters more than its mere presence. The signature study makes the point cleanly. People interviewed on sunny days reported being happier with their lives as a whole than people interviewed on rainy days. But when the interviewer opened with a casual "by the way, how's the weather down there?", the weather effect disappeared. Once people noticed the real cause of their mood, they stopped reading it as information about their lives (Schwarz & Clore, 1983). The same study found that bad moods prompted more of this source-checking than good moods, as if feeling low makes people ask "why?" while feeling good just feels right.
Two distinctions sharpen the picture. First, incidental feelings (carried in from something unrelated) get confused with integral feelings (genuinely about the thing you are judging). The whole bias is incidental affect masquerading as integral. Second, the effect is hard to switch off by willpower. In review after review, simply warning people about the carry-over often fails to remove it (Lerner, Li, Valdesolo & Kassam, 2015). The feeling does its work before deliberate thought gets a vote.
The other half of the consensus is that feelings are not a single good-to-bad dial. Specific emotions carry specific tendencies, an idea called the appraisal-tendency framework. Each emotion comes bundled with a way of seeing the world. Fear grows out of a sense of uncertainty and low control, so it tilts judgments toward caution and seeing more risk. Anger grows out of a sense of certainty and control, so it tilts the very same person toward optimism and risk-taking. In the key experiment, fearful people made pessimistic, risk-averse choices while angry people made optimistic, risk-seeking ones, even though both emotions are "negative". Strikingly, angry people's risk estimates looked more like happy people's than like fearful people's (Lerner & Keltner, 2001). Valence, the good-or-bad of a feeling, is only one ingredient.
The liveliest fight is about how much structure feelings really have. The simple view treats mood as one dimension, good versus bad. The appraisal-tendency view says you need the specific emotion, because fear and anger split apart even though they share a valence. The evidence leans toward the richer view, but not unconditionally: a 2025 registered-report replication of the fear-and-anger work reproduced some effects (general optimism about risk) and failed to reproduce others (risk preference, and optimism about clearly negative events), and an attempt to extend the framework to hope did not pan out (Lu, Efendić & Feldman, 2025). The framework is real, then, but narrower and patchier than the original headline.
A second, deeper debate is about what emotions even are. The basic-emotions tradition holds that a handful of emotions are biologically built in, each with its own signature, including a recognisable facial expression. The constructionist challenge argues that emotions are assembled on the spot from raw feeling plus learned concepts, and so vary across people and cultures. A large review came down hard on one popular assumption: across more than a thousand studies, facial configurations did not reliably signal specific emotions across contexts, so you cannot read "anger" or "fear" off a face the way the common view assumes (Barrett, Adolphs, Marsella, Martinez & Pollak, 2019). That finding matters well beyond the lab, because a whole industry of "emotion AI" reads faces as if it could.
There is also a historical caution worth keeping. The famous two-factor theory, that an emotion is bodily arousal plus a cognitive label for it, shaped decades of thinking (Schachter & Singer, 1962). But the original evidence was thin and later attempts to reproduce it were weak, so it is better treated as an influential idea than as a settled finding.
Affect-as-information has a built-in fragility: it depends on the feeling being mistaken for information in the first place. Make the real source obvious and the effect tends to shrink or vanish, which is reassuring but also means the bias is conditional rather than constant. Most of the evidence also comes from brief, mild moods nudged into being in a lab, with hypothetical choices and student samples, so the size of the effect out in messy real life is harder to call. And emotion itself is genuinely hard to measure, which is part of why the facial-reading shortcut became so tempting and so oversold.
When exactly does a feeling get used as information, and when does it get set aside? How far do the specific-emotion effects stretch beyond fear and anger, and beyond the laboratory, given that even the classic results only partly replicate? And the foundational one: are discrete emotions natural categories carved into us, or convenient labels we construct? Both camps still overclaim.
The single most useful habit here is to treat a feeling as a question rather than an answer: where did this come from, and is it actually about the thing I am judging?
Two practical edges. The first is measurement. If you survey customers, their mood at that moment leaks into the scores, so a satisfaction or brand rating can be partly a reading of the weather, the previous question, or the wait on hold. Neutralise it where you can (timing, question order, larger samples) or you will mistake mood for attitude. The second is that specific emotions need specific responses. A frightened customer wants reassurance and certainty; an angry customer wants control and a way to act. Treating both as merely "upset" misreads what will calm them. And be sceptical of facial-coding "emotion analytics", because the science says a face does not reliably reveal a discrete emotion (Barrett et al., 2019); read those outputs as rough good-or-bad signals at best.
Emotional weather shapes political judgment, and the specific emotion decides the direction. Fear pushes people toward caution, threat-avoidance, and sticking with the safe option; anger pushes them toward action, blame, and risk (Lerner & Keltner, 2001). An appeal built on fear and one built on anger will move the same voter differently, so match the emotion to the behaviour you actually want, and know that an incidental national mood can colour judgments of things it has nothing to do with.
Risk communication runs straight into this. People judge a hazard partly by how it feels, not only by its numbers, so dread-heavy risks (contamination, rare side effects) are not defused by statistics alone and need trust and tone as well. Public messaging also has an emotional lever to choose deliberately and ethically: fear tends to make people risk-averse and compliant in the moment but can also paralyse, while anger mobilises but raises appetite for risk. And official wellbeing or satisfaction figures should be read knowing that timing, weather, and survey wording move them.
Before you trust a judgment, yours or someone else's, locate the feeling behind it. If it might have come from somewhere unrelated, name that source out loud, because naming it is what shrinks the bias. Then resist collapsing the feeling into "good" or "bad": ask which emotion it is, since fear and anger send the same person in opposite directions. And when you are trying to measure what people think, control the conditions that set their mood, or you will end up measuring the mood instead.
.png)