The halo effect

What the halo effect is

The halo effect is a failure of independence between ratings. When you score five attributes of the same thing, those five scores should vary according to five different bodies of evidence. What actually happens is that they collapse toward a single underlying impression, so the correlations between them come out far higher than the underlying facts justify.

That distinguishes it from most of the cognitive bias family. Anchoring distorts one number using another number. Confirmation bias distorts a search. The halo effect distorts the relationship between judgments, which is why it is invisible when you inspect any single one of them. Each rating looks defensible on its own. Only the pattern across all of them gives it away, and that requires you to look at the set, which almost nobody does.

The one sentence version, and the trait that starts it

In one sentence: a strong impression on one dimension becomes the default answer for every other dimension. The starting trait is usually whatever is visible first and takes no effort to assess, such as appearance, confidence, fluency of speech, the polish of a document, or the design of a product's packaging.

Those starting traits share a property worth naming: they are cheap signals. They can be produced by someone who has the underlying quality and equally by someone who has only the signal. A confident presenter and a well informed presenter look the same for the first two minutes, and the rating formed in those two minutes then attaches itself to judgments of accuracy, honesty and expertise made an hour later.

Thorndike in 1920, and Nisbett and Wilson in 1977

The halo effect was named and measured by Edward Thorndike in 1920, in a paper for the Journal of Applied Psychology titled "A constant error in psychological ratings". Thorndike had commanding officers rate soldiers on separate qualities including physique, intelligence, leadership and character. The ratings should have been loosely related at most. They correlated far too tightly: an officer who rated a soldier highly on bearing rated him highly on intellect as well. Thorndike concluded that the raters were not assessing the qualities separately but grading a general impression several times over.

Richard Nisbett and Timothy Wilson added the crucial second finding in 1977, in a study published as "The halo effect: Evidence for unconscious alteration of judgments". Students watched a recorded interview with an instructor who spoke with a foreign accent. One group saw him behave warmly, the other saw the same man behave coldly. Both groups then rated his physical appearance, his mannerisms and his accent, which were identical in the two recordings.

Students who had seen the warm version rated his appearance, mannerisms and accent as appealing. Students who had seen the cold version rated the same features as irritating. Asked afterward whether his manner had influenced their ratings of those attributes, participants said no. Some reported the reverse causal story, that his irritating accent had made him seem cold. That combination, a large effect plus confident denial that it occurred, is what makes the halo effect worth taking seriously rather than merely noting.

Where the halo effect shows up: hiring, marketing and borrowed expertise

The halo effect shows up wherever several ratings of one subject are produced by one person in one sitting. Five settings:

• Hiring. A candidate who interviews fluently is scored higher on technical judgment, reliability and teamwork, none of which the interview measured.
• Performance reviews. One strong quarter raises a manager's ratings of an employee's communication, initiative and planning for the whole year.
• Marketing and brand. A company known for one excellent product is credited with quality in an unrelated product line it has just entered.
• Borrowed expertise. Real authority in one field is treated as authority in another, which is where the halo effect turns into an appeal to authority: the credential is genuine and simply does not cover the question being asked.
• Written work. Clean typography and confident prose raise a reader's estimate of the accuracy of the numbers inside, which no font can affect.

The halo effect and the horn effect, and what else it is confused with

The horn effect is the same mechanism running negatively: one unfavorable impression drags down ratings of unrelated qualities. They are not two effects but one, described from either end, and the research usually treats them together. A candidate marked down for a poor handshake and then marked down for analytical ability has been assessed once and scored twice.

The halo effect also feeds the Dunning Kruger effect from the outside: a person praised across the board, on the strength of one visible skill, receives no signal about the skills nobody actually assessed.

What actually reduces the halo effect

To reduce the halo effect, break the ratings apart so that one impression cannot supply all of them. Five procedures, in order of how much they buy:

1. Score one attribute across all candidates before moving to the next, rather than scoring one candidate on everything at once.
2. Define each attribute with observable anchors, so that a score points to a behavior rather than to an impression.
3. Use different assessors for different attributes, and combine the scores mechanically instead of discussing them into agreement.
4. Blind what can be blinded, assessing a work sample with the name and presentation stripped off.
5. Collect scores before the group meets, because a discussion converges on the most confident voice, which is a halo effect operating on the panel itself.

The principle underneath all five is the one that requires a control group. The placebo effect is the standard illustration of why a trial cannot use its own participants' impressions as the measurement, and a rating panel is in the same position: an assessor who has already formed an impression cannot serve as their own control, so the design has to supply one.

The test to run on a favorable impression

Ask the question that separates evidence from spread: which specific observation supports this particular rating, and would it survive if the person or product had made a poor first impression? If the only answer is a general sense of quality, you have one measurement written down several times.

A second question works on claims rather than people: does this source's credential actually cover the thing being claimed? Genuine expertise in one area is evidence about that area. Extending it is the halo effect wearing academic clothing.

Is the halo effect an unconscious bias?

Yes, the halo effect is an unconscious bias in the specific and demonstrated sense that people cannot detect it operating and deny it when asked, which is exactly what Nisbett and Wilson measured in 1977. Their participants were not concealing anything; they had no access to the process and reported, sincerely, a causal story that ran backwards.

Be careful with the phrase in its looser corporate use, though, where "unconscious bias" often means something broader about attitudes toward groups. The halo effect is narrower, older and better evidenced than that: it is a measurable inflation of correlations between ratings, demonstrated in 1920 and repeatedly since. The honest limit is that the effect is not always an error. A general impression is sometimes genuinely informative, because real qualities do correlate. What the research shows is that the impression is treated as more informative than it is, and that the person doing the treating will tell you, in good faith, that it played no part.

Where to go next

BIASESCognitive biasWhat it is, with examplesBIASESThe Dunning Kruger effectThe Dunning Kruger effect, and what the study actually showedBIASESConfirmation biasDefinition and examplesMETHODThe placebo effectThe placebo effect, explained as a methods problem