Anchoring Bias: Why the First Number Changes the Decision

Why an initial number can pull later estimates toward itself, what a new 2026 meta-analysis says about the size of the effect, and why not every random number becomes an anchor.

A number can enter the room before the evidence does

Imagine being asked two questions.

Is the population of a city higher or lower than 800,000?

What is the population?

Now imagine another person receives:

Is it higher or lower than 3 million?

What is the population?

Even if both people know the first number is only a reference point, their final estimates can move in different directions.

The first number has become an anchor.

Anchoring is one of the most robust effects in judgment and decision-making research.

It is also commonly exaggerated.

The careful claim is not:

Any number can secretly control your mind.

It is:

An initial numeric reference can systematically shift later judgments, especially when it is relevant enough to enter the estimation process.

The effect is old, but the newest synthesis is very current

Anchoring entered the heuristics-and-biases literature through work by Amos Tversky and Daniel Kahneman in the 1970s.

For decades, studies showed that initial values could influence estimates of:

• prices;
• quantities;
• probabilities;
• legal awards;
• negotiations;
• valuations;
• factual estimates.

In 2026, Dan Schley and Evan Weingarten published a major meta-analysis covering fifty years of anchoring research.

The dataset contained 2,601 effect sizes, including 1,280 comparisons between high and low anchors.

The overall high-versus-low anchoring effect was large, with a reported Hedges’ g of about 0.83.

That result strengthens the basic conclusion:

numeric anchoring is not a fragile laboratory curiosity.

But the same meta-analysis also shows why the popular version is too simple.

Not all anchors are equal

The 2026 synthesis found smaller or null effects in several conditions.

Anchoring tended to weaken when:

• the anchor was merely incidental;
• the anchor came from a different dimension;
• the number was obviously random;
• incentives encouraged more careful judgment;
• debiasing interventions were used;
• the anchor provided little directional information.

This is crucial.

A relevant asking price and a random phone number are not psychologically equivalent.

The fact that a number appears before a judgment does not guarantee a strong anchor effect.

Why anchors work

There is no single universally accepted mechanism.

Several explanations have been proposed.

Insufficient adjustment

The classic idea says people begin from an anchor and adjust away from it.

The adjustment often stops too early.

This model is intuitive, especially when people deliberately use a starting point.

Selective accessibility

Another account suggests the comparison question changes which information becomes mentally accessible.

If asked whether a value is above a high anchor, the mind temporarily searches for reasons or facts compatible with a high value.

That evidence then influences the later estimate.

Scale distortion

A numerical reference may alter the way the judgment scale itself is represented.

The anchor can change what feels “high,” “low,” “large” or “small.”

Information interpretation

Sometimes an anchor is not arbitrary at all.

If a seller asks for €20,000, you may reasonably infer:

They probably know something about the value.

Now part of the effect may reflect information use rather than pure bias.

The modern literature increasingly treats anchoring as a family of processes rather than one mechanical reflex.

A negotiation illustrates the problem

Suppose a used car is listed at €18,900.

You inspect it and estimate that a fair price might be somewhere around €16,000.

Would you have generated exactly the same estimate if the listing had been €14,900?

Maybe not.

The listed price has several roles.

It is:

• a number;
• a reference point;
• a signal about the seller’s expectations;
• a possible cue about market value;
• a starting point for bargaining.

Calling the entire effect “irrational anchoring” can therefore oversimplify the situation.

Some information carried by the anchor may be legitimate.

The challenge is separating useful signal from reference-point pull.

Why expert knowledge does not automatically remove anchoring

People often assume expertise makes a person immune.

Research does not support such a clean divide.

Experts can show anchoring effects.

But domain knowledge, incentives and relevant information can reduce dependence on arbitrary starting points.

This is more realistic than saying either:

“Experts are just as biased as everyone else.”

or:

“Experts are immune.”

Expertise changes the information environment.

An experienced valuer has more independent reference points available.

That can dilute the influence of one external anchor.

It does not guarantee zero effect.

The most dangerous anchors often look reasonable

An absurd anchor is easy to reject.

A plausible anchor is harder.

A salary offer that is slightly low.

A house asking price within the local range.

A “normal” project deadline.

A suggested donation amount.

A doctor’s initial estimate.

A forecast from a respected analyst.

These values do not look arbitrary.

That is precisely why they can influence judgment.

The mind does not treat them as random noise.

It treats them as candidates for useful information.

Anchoring is not only about being fooled by someone else

We also create our own anchors.

Yesterday’s price.

Last year’s salary.

The previous relationship.

The first estimate in a project.

The original deadline.

The amount already invested.

Once a number becomes the reference point, later information may be evaluated relative to it.

This can be useful.

Reference points reduce cognitive work.

The risk appears when the reference remains influential after its informational value has disappeared.

The first estimate can become a hidden commitment

Imagine estimating a renovation at €8,000 before detailed quotes arrive.

Later, evidence suggests €12,000 is more realistic.

The original estimate can still shape reaction.

€12,000 now feels “too high” because it is being compared with €8,000.

The question quietly changes from:

What does the evidence now support?

to:

How far away are we from the first number?

That is anchoring at the level of project thinking.

The first estimate becomes psychologically privileged.

Why “consider the opposite” can help

Debiasing research has tested methods that force attention away from the anchor.

One approach is to generate reasons why the anchor might be wrong or why a different estimate could be justified.

The 2026 meta-analysis found evidence that debiasing interventions reduce anchoring effects.

This does not mean a checklist eliminates the bias.

The principle is stronger than the trick:

Build an estimate from independent evidence before negotiating with the reference value.

If you can generate your estimate first, the anchor has less opportunity to define the scale.

A practical example: salary negotiation

Suppose a recruiter asks:

“What salary are you expecting?”

If you answer immediately, you create an anchor.

If they name a range first, they create one.

A more evidence-based process would begin with:

• market rates;
• role scope;
• location;
• experience;
• benefits;
• competing offers;
• company size;
• scarcity of skills.

Only then should the external number be integrated.

The goal is not to ignore the first number.

It may contain information.

The goal is to ensure it is not the only coordinate system available.

Anchoring can interact with confidence

A person can know about anchoring and still experience it.

Awareness is not the same as immunity.

This is true of many cognitive effects.

The problem is that people often feel they adjusted enough.

The final estimate feels internally generated.

That subjective feeling can hide the anchor’s contribution.

Anchoring is therefore a good example of a broader DarkBrain theme:

The source of a judgment is not always visible from inside the judgment.

What the evidence does not support

The evidence does not support these universal claims:

• every number creates an anchor;
• random numbers are as powerful as meaningful ones;
• anchoring proves people are irrational;
• experts are always equally vulnerable;
• knowing about anchoring eliminates it;
• anchors only matter in negotiation;
• the first number always wins;
• any difference after a reference point must be caused by anchoring.

The 2026 meta-analysis shows strong average effects and substantial heterogeneity.

Both facts matter.

The DarkBrain model: anchor, relevance, independent estimate, revision

A practical four-step model:

Anchor

What number entered first?

Relevance

Why might that number contain legitimate information?

Independent estimate

What would you estimate if the anchor had never been presented?

Use outside data where possible.

Revision

After integrating the anchor’s legitimate information, how much should the estimate actually move?

This sequence prevents two opposite errors.

Blindly accepting the anchor.

Blindly rejecting useful information because you are afraid of bias.

Why this matters

Anchoring is powerful because it turns a reference point into part of the judgment environment.

The strongest defense is not motivational.

It is architectural.

Build more than one reference point.

Estimate independently.

Separate evidence from suggestion.

Then revise deliberately.

The first number may still matter.

It no longer gets to define the entire problem.

Continue exploring

Next: Availability Heuristic: Why Vivid Examples Feel More Common

The next article examines why easy-to-recall events can distort estimates of frequency and risk, and why modern research is more nuanced than the simple claim that “the media makes dramatic risks feel common.”