Core topic 02 of 13
Cognitive biases, sorted by how well the evidence holds up
A cognitive bias is a systematic error in judgement, one that leans the same way every time instead of scattering randomly. Roughly a dozen have survived large replication tests. The rest of the famous list is thinner than it looks.
Daniel Kahneman and Amos Tversky started this line of work in the early 1970s with a simple observation: people do not reason badly at random. Ask a room full of people to estimate a number and they land in the same wrong place, in the same direction, for the same reason. That regularity is what makes bias research useful. You can predict the error before it happens.
What follows is grouped by evidence rather than by mechanism, because the practical question is which of these you can rely on.
Biases with strong replication evidence
Confirmation bias Replicated
You search for evidence that fits what you already believe and treat contradicting evidence more harshly. Peter Wason demonstrated it in 1960 with a number sequence task where people tested only the rule they had guessed. Sixty years of work since has found it in medical diagnosis, jury decisions and online news reading.
The uncomfortable part is that intelligence does not protect you. Dan Kahan's work on politically charged numeracy problems found that people with stronger maths skills were better at reaching the answer their side preferred.
Anchoring Replicated
The first number you see drags your estimate toward it, even when the number is obviously irrelevant. Anchoring was one of the findings tested in Many Labs 1, which repeated it across 36 samples and found it consistently. Estate agents shown different asking prices for the same house valued it differently, and said afterwards that the asking price had not influenced them.
Availability heuristic Replicated
You judge how likely something is by how easily an example comes to mind. Vivid, recent and emotionally loaded events come to mind fastest, which is why people overestimate deaths from plane crashes and underestimate deaths from heart disease. Media coverage feeds directly into this, and the effect shows up in insurance purchases after floods and earthquakes.
Hindsight bias Replicated
Once you know the outcome, you remember having expected it. Baruch Fischhoff showed this in 1975 and a 2012 meta-analysis by Roese and Vohs confirmed it across hundreds of studies. It quietly corrupts learning from experience, because it deletes the memory of your actual uncertainty at the time.
Real, but the size or the story is argued over
Loss aversion Contested
Losing 100 is supposed to hurt about twice as much as gaining 100 feels good. The asymmetry is easy to demonstrate, but the 2 to 1 ratio does not hold across contexts, and David Gal and Derek Rucker argued in 2018 that many demonstrations are better explained by inertia and by how the question was framed. The direction survives. The number does not.
Dunning-Kruger effect Contested
In the 1999 studies, people in the bottom quartile of a test overestimated their score by a wide margin while top performers underestimated theirs. That pattern replicates. What does not follow is the version you see quoted, where confidence peaks among the least skilled. Nuhfer and colleagues in 2017 and several statisticians since have shown that part of the curve appears in random data because of regression to the mean.
Fundamental attribution error Contested
You explain your own lateness by traffic and someone else's by their character. The effect is robust in Western samples and much weaker in East Asian ones, which is why many researchers now call it a correspondence bias and treat the strong universal version as overstated.
Where biases actually cost you
Bias research earns its keep in decisions you make once, under time pressure, with no feedback loop. Hiring, medical triage, sentencing and investment all fit that shape. Decisions you repeat daily tend to correct themselves because reality answers back.
The practical countermeasures are unglamorous. Write down your prediction before the outcome so hindsight cannot rewrite it. Get the estimate from two people before either hears the other's number, which kills anchoring inside a group. Ask what would have to be true for you to be wrong, which is the only reliable move against confirmation bias.
Watch this on the channelThe channel covers one bias per video with the study, the sample size and what it does not prove.
Open Psychology Cue on YouTubeQuestions people ask about this topic
What is a cognitive bias in simple terms?
A systematic error in how you judge or decide, one that pushes in a predictable direction rather than randomly. Anchoring pulls almost everyone toward whatever number they saw first.
How many cognitive biases are there?
Popular charts list around 180, but most are variations on a few mechanisms: what comes to mind easily, protecting an existing belief, and reacting to framing. The long lists overstate how many distinct effects have been demonstrated.
Can you train yourself out of cognitive bias?
Partly. Carey Morewedge and colleagues found that training transferred to real decisions, but changing the decision environment works better than changing the person. Checklists, independent second opinions and forcing yourself to argue the opposite all beat simply knowing the bias exists.
Is the Dunning-Kruger effect real?
The pattern replicates. The popular reading, that incompetent people think they are experts, goes past what the studies show, and part of the curve appears in randomly generated data because of regression to the mean.