In 1974, Amos Tversky and Daniel Kahneman described an effect that later became one of the most reproducible in behavioral economics. In their experiment, participants spun a wheel of numbers that stopped at either 10 or 65, and then answered a question that had nothing to do with the number: what percentage of UN countries are African nations. Those who got 10 gave, on average, a significantly lower estimate than those who got 65. A random number, generated right before the participants' eyes, shifted their estimate.
The effect was called anchoring. The mechanism is that when estimating an unknown quantity, a person does not calculate the answer from scratch, but starts from the first reference point they come across and adjusts it—usually insufficiently. An anchor influences even when it is obviously irrelevant and when the person is aware of its existence.
Salary negotiations are nearly ideal conditions for this effect. The exact value of the role is unknown to both sides, the range is wide, and the decision is made quickly under social pressure. The first number named becomes the point around which the entire subsequent conversation revolves—and even strong arguments usually move the outcome less than the choice of this point itself.
From this follows something that was not obvious for a long time. Asking about a candidate's current or past salary is not a harmless way to get oriented. It sets an anchor. If a person names a figure below market value, all further bargaining centers around it, rather than around the weight of the role. And this perpetuates: whoever once earned less continues to earn less, regardless of how well they perform.
This is precisely why asking about past salary has been banned in many jurisdictions. Studies of such regulation showed a noticeable increase in salaries for those who changed jobs, as well as a narrowing of gaps between groups. The mechanism is exactly the same: when the anchor is unavailable, negotiations begin to be based on the role.
What a company that wants a fair process does: Determines the salary band before the conversation—based on the weight of the role, not the candidate's expectations. Names it first: since an anchor will emerge anyway, let it be a justified range rather than a random number from the past. Does not ask about the current salary—even if it is legal, doing so degrades the quality of its own decision. And checks whether the variance within a single grade is explained by criteria or by bargaining power: if by the latter, you do not have a compensation system, but the result of a series of conversations.
An anchor will appear in any case. The only question is whether you will set it—or it will come from someone else's past.