How to reduce bias in performance evaluations

Bias in performance evaluation is difficult to address because nobody experiences themselves as biased. It operates below the level anyone notices, and it produces the same result every year: some people are consistently seen and others are consistently not.

The shapes it takes

The halo effect: one strong trait colours everything else, and a person who presents well ends up rated well across categories nobody actually observed. Its opposite, the horns effect, does the same damage from one bad incident.

Similarity bias, which is the most uncomfortable one. Managers rate people who resemble them more highly than equivalent colleagues who do not, and this holds across gender and race in study after study.

Recency bias: the last six weeks weigh more than the first ten months, so someone who finishes strong beats someone who was steady.

And the pattern that shows up in the language itself. In one analysis of technology company reviews, 58% of the reviews written for women contained criticism of their personality, against 2% of those written for men. Women tend to receive vague and subjective feedback, men specific and actionable feedback, and that difference compounds over a career.

What actually reduces it

Structure first, because bias fills whatever space a form leaves empty. Replace good communication skills with something observable: presents complex ideas clearly in meetings, replies within a day, checks understanding when handing over work. A specific criterion is harder to bend.

Then calibration. Bringing managers together to compare ratings before they are final surfaces the manager who rates everyone highly and the one who rates nobody above average, and it makes demographic patterns visible while they can still be discussed.

Then multiple perspectives, because a single rater’s bias goes straight into the score, while several raters at least disagree with each other.

Then continuous documentation, which is the least glamorous item and possibly the most effective. A record written through the year beats a memory reconstructed in December, and memory is where recency bias lives.

Then check whether it worked

Look at the distribution of ratings across groups. Look at promotion rates for people with equivalent ratings. If a difference appears, you have found something worth investigating rather than something to explain away.

None of this eliminates bias. It makes it visible and expensive to act on, which is the achievable version of the goal.

Sign up for a free TalentRewards trial.