011 min
Where the halo effect shows up
Picture a design lead hiring a product designer. Five people apply, and the design lead scores each one from 1 to 5 on research, systems thinking and craft.
The first candidate arrives with a polished portfolio site and talks with ease. The case studies turn out to be thin. Even so, the design lead gives this candidate a 5 on all three skills, because the first ten minutes felt so good.
The second candidate is quiet and slow to answer. Their case studies show careful research and a clear system of reusable parts. The design lead gives them a 3 on all three skills, including the two that the work plainly proves.
Nobody in the room planned to favour the first candidate. Each score felt like a separate judgment about a separate skill. But each candidate got the same score on every skill. That pattern is the halo effect, and it is easy to see once the scores sit side by side.
021 min
Where the halo effect comes from
The design lead's problem is older than design. In 1920 the psychologist Edward Thorndike published a short paper on how military officers rate the men under them. Each officer scored men on separate qualities, among them physique, intelligence, leadership and character.
One flight commander rated 137 aviation cadets, and his intelligence ratings tracked his ratings for physique, leadership and character, with correlations of .51, .58 and .64. A correlation measures how closely two sets of scores rise and fall together, and 1.0 is a perfect match. Thorndike judged the figures too high and too even to be real differences between men.
His explanation was that a rater forms a general feeling that a man is rather good or rather poor, and that feeling changes every specific rating. He used the word halo for the way a general good opinion spreads to specific abilities.
Telling raters to judge each quality alone did not help
Thorndike's instructions asked the officers, very firmly, to judge each quality on its own. The ratings above came from officers who had been told exactly that. A rule against mixing the judgments is therefore not enough to prevent the mix.
031 min
Why the halo effect happens
Thorndike's explanation leaves one question open. Why would a general feeling answer questions about specific traits? Judging a person's intelligence or craft takes evidence, and evidence takes time. A general feeling about the person takes none. Researchers usually explain the halo effect as the mind using that quick feeling to answer questions it has not had time to check. This is an interpretation that fits the results below. Nobody has watched it happen inside a head.
Richard Nisbett and Timothy Wilson tested the idea in 1977. Students watched a lecturer and rated his accent, his mannerisms and his looks. Different groups saw the same man behave in a friendly way or in a cold way. Their ratings of his accent, mannerisms and appearance followed how likeable he seemed.
People believed the cause ran the other way
Afterwards the students were asked what had shaped their view of the lecturer. Many students believed that his accent, looks and mannerisms had made him likeable, which is the reverse of the true order. They could not see how strongly the feeling had influenced their own ratings.
041 min
The halo effect and physical appearance
A halo can start from many things, and the best-known one is how a person looks. In 1972 Karen Dion, Ellen Berscheid and Elaine Walster gave people a photograph of someone near their own age. Researchers had already sorted the photographs into attractive, average and unattractive. People who judged the attractive photographs expected those people to be more socially desirable and to have better job and marital prospects.
The effect reaches decisions that cost people something. In 1974 a study by Efran found that people gave milder sentences to attractive defendants than to unattractive ones for exactly the same crime.
Most of this evidence comes from experiments in which people judge photographs or written cases. Real hiring and sentencing give people far more information, so the size of the effect outside the laboratory is harder to know.
052 min
The halo effect around company results
A halo does not have to come from a person. It can come from a result. The business professor Phil Rosenzweig argued in his 2007 book The Halo Effect that a company's performance creates a halo. That halo changes how people see its leaders, strategy and culture. People then explain a good result with a good culture, and a bad result with a bad one. Judging a decision by how it turned out is a close relative, called outcome bias.
Cisco, 2000 to 2001
Rosenzweig's main example is Cisco Systems. While Cisco grew during the internet boom, the business press praised its management, its selling and its run of company purchases. Then technology shares fell. Between April 2000 and April 2001 the price of a Cisco share dropped from $80 to $14. Fortune wrote that acquisitions, forecasting, technology and senior management had all failed the company. Rosenzweig asks whether Cisco's leaders, culture and strategy could really have changed that much in so short a time. His point is that mostly the opinion changed.
Rankings of admired companies
Rosenzweig adds that popular rankings, such as Fortune's yearly list of the most admired companies, are heavily shaped by the same effect. A firm with good results tends to score well on qualities that nobody measured separately.
Business books built on tainted data
Rosenzweig also doubts the research methods behind some best-selling business books. One of them is Good to Great, which tried to explain why some good companies went on to greatness and others did not. If the success is known before the traits are rated, the ratings may reflect the success. The traits then look like causes when they are only descriptions.
062 min
How the halo effect shows up in product, design, and AI
Products get the same treatment as people. Say the design lead's team tests a new prototype that looks clean and attractive. Participants say they like it and call it easy to use. The session notes show that many of them could not finish the tasks.
Research on websites shows the same split. In 2002 Lindgaard and Dudek asked users to rate the visual appeal of a group of websites. The sites with high appeal ratings were then tested for usability. On average, participants failed more than half of the tasks on the high-appeal sites, but their satisfaction ratings stayed high.
A beautiful interface can start a halo, because people see it before they use it. Test usability by what people do and count the failures. Do not rely on what people say they feel.
Brands that pass their reputation to new products
The same spread works across a company's products. The popularity of Apple's iPod raised enthusiasm for the company's other products. Marketers use this on purpose when they launch a new item under a trusted brand.
AI judges and the halo effect
The same worry applies to language models that score essays, answers or designs against a list of criteria. If a model gave every criterion the same score, its rubric would add nothing. A 2026 preprint audited 12 model judges on 2,377 essays in two languages. One halo comparison in that study was overturned by a stricter check. When instrument and calibration were matched, the authors found no credible evidence that the models' halo goes beyond the range of trained human raters. This is one preprint, so treat it as a result to watch and not a settled finding.
072 min
Where the evidence on the halo effect is contested
That raises a fair question: is the halo effect always a mistake? For decades it was treated as one. In 1993 Kevin Murphy, Robert Jako and Rebecca Anhalt questioned that view. Murphy, Jako and Anhalt reviewed the research behind that view and concluded that each main part of it was either wrong or open to question.
A match between ratings is not always an error
Some people really are strong in several related areas. When ratings of those areas match, part of the match is true. The mistake is to let a general feeling replace evidence about each trait, and ratings alone cannot show which of the two happened.
Fewer halo errors does not mean better ratings
Researchers often score a rater's halo and treat a low score as a sign of accuracy. In 1989 Murphy and Balzer pooled data from 10 studies and found an average correlation of .05 between measures of rater error, halo among them, and rating accuracy. A low halo score therefore says almost nothing about whether the ratings are right.
Halo is not a fixed trait of the rater
In a laboratory study, Murphy and Anhalt found that halo errors were highly unstable when the people rated or their behaviour changed. The stability coefficients were .20 and .18 in those two cases. Halo errors were only moderately stable when the people and the behaviour stayed the same. The result suggests that halo depends partly on the rating situation. Labelling a person as halo-prone is a poor description.
Where this leaves a practitioner is clear enough. Treat the halo effect as a risk of the setup. Change what people see and when they score, and do not spend effort looking for biased people.
082 min
How to guard against the halo effect
Back to the design lead. A feeling of being fair is poor evidence, because the students in the lecturer study could not see what had shaped their own ratings. The fix has to be a change to the process.
Marking exams one question at a time
Daniel Kahneman met the problem as a young professor who marked essay exams. He marked one student at a time and read all of that student's essays in a row. A high mark on a student's first essay made Kahneman more generous with vague or ambiguous statements in that student's later essays. He then switched to marking every student's answer to one question before moving on to the next question. He became convinced that the new method was better, because a mistake on one answer no longer carried over into the next.
The design lead can copy the idea. Score every candidate on research before anyone is scored on craft. The scores for one skill are then not coloured by what the candidate showed on another.
Hide what the question does not need.
Remove names, photos and school names from work samples before they are scored.
Write the criteria first.
A rubric fixed before meeting anyone says what a 5 and a 3 look like in the work itself.
Score alone, then talk.
Each panel member submits scores before the discussion, so one strong first impression does not set the others' scores.
Check the pattern afterwards.
A candidate with the same score on every skill is a signal to re-score the skills one at a time.
The last step before any hire is to compare the scores with how people later perform. That check is the only one that shows whether the ratings were right.
091 min
The halo effect compared with nearby concepts
Several ideas sound close to the halo effect. One question separates each of them from it: what spreads, and to what.
| Concept | What it does | How it differs from the halo effect |
|---|---|---|
| Not in the library yetHorn effect | ||
| PsychologyFirst impression bias | ||
| PsychologyConfirmation bias | ||
| PsychologyAnchoring | ||
| PsychologyPygmalion effect |
?4 questions
Questions people ask
Does the halo effect only come from good looks?
Is a halo always a mistake?
Does knowing about the halo effect protect you from it?
Who named the halo effect?
§13 sources
Sources
Show all 13 sourcesHide sources
Edward L. Thorndike (1920), A Constant Error in Psychological Ratings, Journal of Applied Psychology. (Thorndike).pdf
Richard E. Nisbett and Timothy D. Wilson (1977), The Halo Effect: Evidence for Unconscious Alteration of Judgments, Journal of Personality and Social Psychology.
A summary of the Nisbett and Wilson study, PsyBlog (2007).
Karen Dion, Ellen Berscheid and Elaine Walster (1972), What Is Beautiful Is Good, Journal of Personality and Social Psychology.
A summary of the physical attractiveness stereotype, Wikipedia.
A summary of halo effect research including Efran (1974), Wikipedia.
An article on Phil Rosenzweig and his book The Halo Effect, Knowledge at Wharton.
Halo Effect: Definition and Impact on Web User Experience, Nielsen Norman Group.
Kevin R. Murphy, Robert A. Jako and Rebecca L. Anhalt (1993), Nature and Consequences of Halo Error: A Critical Analysis.
Kevin R. Murphy and William K. Balzer (1989), Rater Errors and Rating Accuracy, Journal of Applied Psychology.
Kevin R. Murphy and Rebecca L. Anhalt (1992), Is Halo Error a Property of the Rater, Ratees, or the Specific Behaviors Observed?, Journal of Applied Psychology.
A blog post that quotes Daniel Kahneman, Thinking, Fast and Slow, on marking exams, Crooked Timber (2012).
LLM Judges as Raters: A Pre-Registered Audit of Severity, Halo, Reliability, and Version Instability in LLM Essay Scoring, arXiv preprint (2026).





