Can You Train Away Bias? What the Studies Tried and What They Found

Start hereIn a disguised business case, graduate students who had one training session were 19% less likely to pick the answer that only confirmed their hunch. Whether gains like that reach everyday decisions is still argued.

Training can reduce some thinking biases on tasks like the ones practised, and in several studies the effect lasted for weeks. Whether it carries over to new kinds of problems, and to real decisions, is less certain.

Below is what “debiasing” means and the main methods researchers tried. Then comes what changed, for how long, and the “bias blind spot” that makes all of this harder.

A 2025 review found 54 randomised trials in education; pooling 41 of them showed a small but significant improvement in avoiding biases, compared with control groups. Its authors wrote that questions remain about the depth of the learning and whether it carries beyond the classroom.

011. What “debiasing” means

Debiasing means trying to reduce a known thinking bias, by changing either the person or the situation. A cognitive bias is a steady tilt in how we judge, the same way, again and again.

One well-known example is confirmation bias: looking for evidence that fits what we already expect. For example, if you think a coworker is careless, you may remember their one typo and forget their many clean reports.

In a 1982 chapter, Baruch Fischhoff compared debiasing to “destructive testing” in engineering: studying the conditions that make a bias disappear. A 2004 handbook chapter by Larrick grouped the strategies into motivational, cognitive and technological ones. A 2015 guide by Soll, Milkman and Payne sorted them by whether they change the person or the environment.

In 2015, Carey Morewedge and colleagues wrote that early attempts to reduce decision biases with training “met with little success”. That led researchers to focus on incentives and on changing how decisions are presented. The sections below take the main methods one at a time.

022. The methods researchers tried

Some methods helped in the lab, some barely did, and one backfired when overdone. Each method below has what was tried, what was found, and how sure the evidence is.

Warnings and “try to be fair”

Knowing about a bias, or being told to avoid it, did little in the studies we read. For example, telling yourself “I’ll be objective about this job candidate” names the goal but not a way to reach it.

A 2021 review noted that merely teaching knowledge about biases has appeared insufficient to reduce them. Anchoring, leaning too hard on the first number you see, is one example.

In a 1996 set of five studies, anchoring appeared difficult to avoid even when people were warned in advance. In 1984, telling people to be “as fair and unbiased as possible” corrected their judgments less than asking them to consider the opposite.

How sure: Moderate several lab studies point the same way, though each tested only one or two biases.

Consider the opposite

Asking yourself why you might be wrong reduced several biases in lab studies; in one set, a long list backfired. The idea is to think of reasons the opposite could be true before you decide. For example, before you conclude a friend is ignoring you, write two other reasons they might not have replied.

In two 1984 experiments, considering the opposite reduced bias when people weighed new evidence on social issues and tested first impressions of a person. In a 2000 study with experts in a real-world setting, listing arguments against a given number reduced anchoring.

A 2018 review of health-related debiasing found 60 of 87 interventions at least partly successful, and methods such as consider-the-opposite were among those mainly tested.

There is a catch. Hindsight bias is the feeling, after something happens, that it could have been seen coming. In two 2002 studies, people who listed ten ways an event could have turned out differently showed more hindsight bias, not less.

Those who listed only two showed no significant hindsight effect. The authors’ explanation was that finding ten reasons felt hard, and that difficulty suggested there were few other ways things could have gone. The idea behind that explanation, that how easily things come to mind shapes judgement, has had mixed results in replications.

How sure: Moderate it worked across several biases in lab studies and one study of experts; overdoing it backfired in one set of studies.

Incentives

Paying people to get it right helped on some tasks, but did not make biases disappear. For example, a bonus may make you double-check a sum; it may not stop the first number you heard from colouring your estimate.

A 1999 review of 74 experiments found that the most common result of paying for performance was no change in average performance. Higher pay often did help on judgement tasks that respond to more effort. The authors noted that no replicated study had made rationality violations disappear purely by raising incentives.

How sure: Moderate one large review of 74 experiments, and a 2021 study with stakes above a month’s income found the same.

Training games and videos

One session with a game or video reduced several biases on later tests, and in one study on a disguised real task. Most of the studies that tested lasting effects used this method. Think of it as a short course with practice rounds and feedback on your own answers.

In two 2015 experiments, one session of a computer game or an instructional video produced medium to large reductions in bias right away. The effects persisted at least two months later.

Games cut bias by at least 31.94% at once and at least 23.57% two months later; videos by at least 18.60% and 19.20%. Games that gave personal feedback and practice did better than videos. The reductions also appeared on problems in contexts and formats the training had not covered.

The next step was to test it outside the lab. In a 2019 study, 290 graduate students had one training session, either before or after an unannounced business case; the training schedule, not a random draw, decided which. The case was modelled on the decision to launch the Space Shuttle Challenger.

Trained students were 19% less likely to choose the inferior solution that only confirmed their hypothesis.

A 2021 review reports that the effect was retained for up to 52 days.

In a 2025 study, a single session of a different kind reduced confirmation bias in national risk analysts in a European country and in master’s students. The session was an explanation and a group case exercise.

Bias was measured before and after the session, at the same event. The analysts taking part were more than half of that country’s national risk analysts.

Reviewers of this work are more cautious. A 2021 systematic review found only 12 studies that adequately tested whether effects lasted at least 14 days. Only one of them also tested transfer to other tasks and contexts. It noted that the positive results on lasting effects came largely from one research programme.

Its authors concluded there is currently insufficient evidence that these methods will substantially help people decide better in real life. They also argued that the 2019 field study measured one narrow form of confirmation bias, so the training may have taught a specific, limited trick.

How sure: Moderate effects on similar tests lasted weeks in several studies; real-life evidence is limited, and most positive results come from one research programme.

Statistical training

Learning statistical rules, like the law of large numbers, carried over to new everyday problems, though less after two weeks. The law of large numbers says that as samples get larger, their average tends to get closer to the true average. For example, a café with three five-star reviews tells you less than one with three hundred.

In 1987, Richard Nisbett and colleagues reported that even brief formal training in reasoning rules may improve how people reason about everyday events. In a 1991 study, training on the law of large numbers in one area carried over strongly to another, when tested right away.

Two weeks later, performance in the untrained area had dropped significantly but was still better than in the control group. A 1988 study found that graduate training in psychology and medicine had large effects on statistical and methodological reasoning, while chemistry training had none.

A separate 2015 team tested 154 people four weeks after training on ten biases. The analogy-based training appeared to improve scores only on tasks that measured statistical principles.

For skills in general, the weight of the evidence we read leans towards far transfer being rare, with action games and music training the main exceptions claimed. Our Range summary covers that debate.

How sure: Moderate several studies agree, though most of them come from one research group.

Taking the outside view

Checking how similar past cases turned out removed over-optimistic forecasts in a 1994 planning study. The “inside view” builds a forecast from your own plan; the outside view asks how things like it usually go. For example, before promising a report by Friday, check how long your last three reports took.

In 1993, Daniel Kahneman and Dan Lovallo wrote that over-optimistic forecasts come from an “inside view” that “anchors predictions on plans and scenarios”. The planning fallacy is the name for underestimating how long your own tasks will take. In the 1994 planning studies, the optimistic bias “was eliminated” for people told to connect past experiences with their predictions.

In a 2014 report on a large forecasting tournament, training was one of three things linked to accuracy. That training “corrected cognitive biases, encouraged forecasters to use reference classes”. Our decision tools page has more on the inside view and that tournament.

How sure: Moderate a clear result in one planning study, and a training package that included it helped in one large forecasting tournament.

033. Does it carry over, and does it last?

Effects usually held on problems like the trained ones for weeks; whether they reach real decisions is still argued. Researchers call the spread to unlike problems “transfer”. For example, spotting confirmation bias in a quiz is one thing; noticing it while choosing a flat is another.

The 2025 meta-analysis said the evidence suggests training can reduce bias on targeted tasks, and that more research is needed on whether this translates into real-world decisions.

All the studies it reviewed had an unclear or high risk of bias, and there was some risk of publication bias. In plain words: the studies inside the meta-analysis had design weaknesses, and some with no effect may never have been published.

The researchers read the same evidence differently. The 2019 field study’s authors called their result promising evidence that debiasing training effects transfer to field settings. The 2021 reviewers called for longer studies, lasting at least one year. Our reading: the lab effects look real, and the long-term, real-life evidence is still thin.

Two of our other pages meet the same question. Our logical fallacies page covers short lessons on manipulation techniques. Our SIFT guide covers how long lessons on checking news online have been found to last.

How sure: Unsettled reviews agree training helps on targeted tasks; researchers disagree on how far that reaches real life.

044. The bias blind spot

People see bias more easily in others than in themselves, and in two studies being clever did not shrink that gap. Psychologists call this the bias blind spot. For example, you may spot a friend’s wishful thinking about a new car while feeling sure your own view of your job is clear-eyed.

The term was introduced in 2002 by Emily Pronin, Daniel Lin and Lee Ross. In three surveys, people rated themselves as less subject to biases than the average American, their classmates and fellow airport travellers. In a follow-up study, people who rated themselves better than average insisted their ratings were accurate and objective.

They held to this even after reading how the bias could have affected them. In 2021, three preregistered experiments with 969 people repeated the finding, with a large gap between ratings of self and others. One explanation: we judge our own bias by looking inward, and other people’s by what they do.

In two 2012 studies, none of these blind spots shrank with higher cognitive ability or bias-related thinking habits. If anything, higher ability went with a larger blind spot. Being free of the blind spot did not help people avoid the classic biases themselves. In seven studies, many classic thinking biases were unrelated to cognitive ability, though some others were related.

A 2015 scale for the blind spot was distinct from measures of intelligence. People who scored high on it were more likely to ignore the advice of others. Our Think Again summary looks at whether clever people are better at changing their minds.

Can the blind spot itself be trained? In the 2025 study of risk analysts, the training session reduced the blind spot, but both analysts and students still showed it afterwards.

In a 2024 study, people saw more of their own biases in an algorithm trained on their decisions than in the decisions themselves. They were also more likely to correct decisions they thought came from the algorithm. Our reading: a second pair of eyes can show you what looking inward misses.

How sure: Strong the 2002 finding was repeated in three preregistered experiments; evidence that training shrinks it is thinner.

055. What this means in practice

The studies point towards building checks into how you decide, rather than relying on willpower to stay unbiased. That is our reading of the whole picture. Warnings did little, incentives did not remove biases, and training effects were real on targeted tasks, with less evidence beyond them.

A check built into a routine is there every time, whether or not you remember the lesson.

A 2018 review of health-related studies saw technological strategies, such as visual aids, as a route to large-scale debiasing outside the lab. The 2015 guide by Soll, Milkman and Payne looks at how organisations might alter the work environment through “cognitive repairs”.

Benjamin Franklin tried something like this on himself. He kept a little book that gave each of his thirteen virtues a page and each day a column. Of humility, he says he had little success with “the reality of this virtue, but I had a good deal with regard to the appearance of it.”

Our Franklin summary tells the rest. Changing a habit of mind is slow, and partial.

Try it (our suggestion): pick one decision this week and build in one check:

• Before you decide, write the strongest reason you could be wrong. Stop at two reasons.
• Before you estimate time or cost, look at how long similar past tasks took.
• Before you see anyone else’s figure, write down your own.
• Ask someone else to look at your reasoning. People tend to see bias more readily in others than in themselves.

For a team, a pre-mortem is one way to make the first check a fixed step; our decision tools page explains it. Our 7 thinking skills page has exercises for intellectual humility. Our Thinking, Fast and Slow summary and cognitive biases page cover the biases themselves.

06Quick answers

07Use this

08Quick quiz

How we checked this · Last checked October 2026

• We read the abstract of every study through PubMed, OpenAlex, Crossref, Europe PMC or the publisher's own page, and the full text of the 2021 review and the 2025 risk-analyst study.
• The 2025 meta-analysis is behind a paywall, so we read its abstract only.
• A 2020 correction notice for the 2019 field study could not be opened; the 2021 review, written after it, reports the same figures.
• Where researchers read the same evidence differently, we show both readings: the field study's authors and the 2021 reviewers.
• Our wording on far transfer matches our Range summary, and the 1994 planning-study sentence matches our Thinking, Fast and Slow summary.

How we check every article →

09Sources