The Bizarre Effects of Anchoring

Research with trial judges shows that arbitrary numeric reference points have a powerful effect on their decisions.


What is the average price of a textbook in the nearest college campus bookstore? Before you answer, is it greater or less than $7,128.53?

Most people quickly agree that the average price of a textbook is less than that absurdly high amount. But contemplating that outrageous number affects judgment. When first given an absurdly high numeric reference point people make higher estimates of the average price of a textbook then when they are not first provided with this reference point.

Psychologists call this phenomenon "anchoring." Amos Tversky and Daniel Kahneman first demonstrated that people tend to rely on numeric reference points when making numeric judgments. Doing so is a helpful and useful mental shortcut. If you are going to buy a new car with a sticker price of $36,345, you know that you likely will not pay that amount. You might negotiate that number down and there might be additions not listed. The sticker price, however, provides a good estimate of the final price. Anchoring is a thus useful way to estimate how much you will pay for the car. Problems arise when anchors influence judgment even when they are meaningless or arbitrary.

Psychologists have shown that all manner of numeric reference points influence numeric judgments, no matter how bizarre or fanciful. In one study researchers asked undergraduates to identify the year that Attila the Hun was defeated. In this study researchers first asked their subjects whether Attila the Hun was defeated before or after the year corresponding to the last three digits of their phone number plus 400 (A.D.). Then they asked subjects to identify when Attila was defeated. (For the record, Attila suffered his only defeat at the Battle of the Catalaunian Plains in 451 A.D.). The students certainly knew that their numbers had nothing to do with Attila the Hun, but their guesses correlated remarkably with their phone numbers.

Psychologists have proposed three main theories to explain the influence of anchoring. Tversky and Kahneman believed that people adjust inadequately from the anchor. They proposed that cognitive effort is required to adjust mentally the anchor down (or up) to the correct estimate and that people commonly fail to put in enough cognitive effort. If so, the starting point would thus affect the final estimate. Other psychologists suggested that once people entertain the possibility that the anchor is correct, their evaluation of that anchor guides how they arrive at the actual estimate. For example, when people contemplate whether the price of the average textbook exceeds $7,000, they call to mind examples of inordinately expensive textbooks. Although even the most expensive books are not that pricey (not yet!) considering whether that inordinately high number is accurate calls to mind expensive examples of textbooks, which in turn influence their estimate of the average price.

Finally, others suggest that anchors temporarily alter perception, creating an illusion of judgment. Just as an adult can seem small when standing next to an elephant and large when standing next to a cat, anchors induce people to perceive reasonable estimates as either too large or too small in comparison. Having just considered whether the average price of a textbook exceeds $7,000, an estimate of $150 seems too small. All three of these influences might be at play because anchoring is a potent and reliable phenomenon.

Judges commonly must make numeric estimates when determining damage awards, criminal sentences, bail amounts, or civil fines. Suggestions from lawyers, statutory damage caps, or insurance policy limits can anchor judges' decisions. Using experimental methods with sitting judges as research subjects, we have found that anchors have profound effects on judges.

In one of our studies, federal judges who first considered a frivolous motion to dismiss for failure to meet the jurisdictional amount of $75,000 in a diversity action in federal court awarded 29% less in a serious personal injury case than judges who did not have to rule on the motion. We found that offers made during settlement discussions in a civil case anchored judges' ultimate awards—even though we reminded these judges that discussions during settlement talks are inadmissible.

We also showed that damage caps influenced civil damage awards of American, Canadian, and Dutch judges. In a criminal context, judges anchored on sentences in unrelated cases; that is, after sentencing a defendant in a minor case to several months in prison, judges then sentenced a more serious offender to far less time than when the order of the cases had been reversed.

Every anchor we tested influenced judges. We thus began experimenting with an outlandish anchor. We created a hypothetical case for judges in which we described a roadhouse (or a "club") that had violated a local noise ordinance. Our materials stated that the roadhouse had put a band out on an outside deck late into the evening on a couple of occasions in its opening week, disturbing nearby residents. We asked 526 judges from seven different jurisdictions to identify an appropriate fine that would be sufficient to deter future misconduct and appropriately reflect the "degree of disruption" that the roadhouse had caused to its neighbors. For half of the judges, we identified the facility as "Roadhouse 58", named after its street address; for the other half of the judges, we called it "Roadhouse 11,866," also named after its address.

Roadhouse 11,866 would be well advised to either remain quiet at night or change its name. The median fine judges levied on Roadhouse 11,866 was $1,683, as compared to $1,000 against Roadhouse 58. The difference between the groups was significant. (Mann-Whitney test, z = 2.03, p =.04.) Although the address was irrelevant, it clearly drew the judges' attention. Seven of the judges evaluating Roadhouse 11,866 levied a fine of $11,866!

Making numeric evaluations is difficult. No natural metric exists to convert a qualitative degree of pain and suffering into a quantitative dollar value, nor a sense of desert into a number of years in prison. Judges (and juries) must nevertheless perform this task constantly. Research on anchoring suggests that the task will be influenced by misleading influences like irrelevant anchors.

What can be done? The only sure-fire cure for the influence of misleading anchors is to avoid learning about them, but that is impossible in many cases. Counter-anchors are not effective either, as research suggests that decision makers will focus on one of the available anchors. Well-crafted constraints like sentencing guidelines can contain the influence of irrelevant anchors, although they might introduce other structural problems. Reasonable anchors hold some promise. Recent research by Valerie Hans and Valerie Reyna suggests that people are more apt to be influenced by meaningful, reasonable anchors than misleading, unreasonable ones. To the extent judges can either avoid exposure to anchors, rely on sensible guidelines, or identify reasonable anchors, they can avoid some of the erratic influence of anchors.

Our full article on anchoring in judges: Jeffrey J. Rachlinski, Andrew J. Wistrich & Chris Guthrie, Can Judges Make Reliable Numeric Judgments? Distorted Damages and Skewed Sentences, 90 Indiana L.J. 695 (2015).

Tomorrow: Gains and Losses:  The Arbitrary Effect of Decision Frame.

NEXT: Today in Supreme Court History: June 9, 1970

Editor's Note: We invite comments and request that they be civil and on-topic. We do not moderate or assume any responsibility for comments, which are owned by the readers who post them. Comments do not represent the views of or Reason Foundation. We reserve the right to delete any comment for any reason at any time. Report abuses.

  1. As a civil practice defense lawyer I deal with this question regularly. In the damages trial summation plaintiff’s counsel will suggest an amount to the jury. You don’t know what it will be and as the defendant your summation goes first. Do you suggest a number?

    Some defense attorneys go by the dogma that you never suggest a number because that establishes a “floor” that the jury will go above. But what if the floor is an acceptable number? You’ve already lost the case. What if plaintiff suggests a ridiculous number for the injury involved (say, $2 million for a broken ankle) and the jury has no idea whether in the world of jury verdicts this is what that injury’s worth? If you suggest $5,000 that might bring them down to earth. It’s a case by case decision, at least for me, invoking elements of poker and roulette.

    1. I think suggesting a number works to your advantage.

      If you are going to buy a new car with a sticker price of $36,345, you know that you likely will not pay that amount. You might negotiate that number down and there might be additions not listed. The sticker price, however, provides a good estimate of the final price.

      There’s your answer. There’s a reason cars have MSRP’s.

      1. I almost always suggest a number. Most of my trials are assigned to me by liability insurers, they have a bureaucratic mentality and are wedded to the play-it-safe, never-suggest-a-number dogma. So in reporting on the verdict I don’t tell them what I’ve done. Fortunately they never ask to look at the transcript.

    2. You should be candid. The value of the injury is nothing. Or people are better off with that person dead. The plaintiffs owe money to the defendant for ridding them of this toxic person. The value of life is excessive in order to unjustly enrich the plaintiff bar. You, as a defense lawyer, owe your job to them, not to the client. Deter them, you lose your job. You have some thinking to do.

      It should be the standard of professional practice to utterly destroy the lives of the plaintiff lawyers, the plaintiffs, the plaintiff experts, and especially that horrible judge. They should regret living after suing your client. That is not zealous representation. That is adequate representation. They are the enemy of our nation. The law is totally biased, and all out lawfare must be declared on them. (In the thin skull doctrine, the defendant is expected to pay for a pre-existing conditon, for example.)

    3. In Massachusetts, defendants sometimes avoid pleading statutory limits on liability because the case is obviously worth less than the limit and a jury would be tempted to award the maximum once they heard a number. I don’t have details on how common this is — it’s something I read about, not something I experienced. The risk is, you might be stuck with a verdict in excess of the legal limit.

      1. That’s a factor not present where I practice (New York), where there are no statutory limits.

        But in Mass., if the jury awards more than the limit, it doesn’t count, right? So why be afraid of an over the limit award?

        1. If you don’t plead the limit it is waived. There is a $20,000 cap on tort liability for a charitable organization. If you don’t plead it and the jury awards $20,000,000 you are stuck with the verdict. There is sometimes a strategic reason not to use this particular limit. Employees are not protected by the liability limit but they won’t be sued as long as the employer has deep pockets available.

  2. When I saw the title, being a full time cruiser, finally a look at the hodgepodge of state laws concerning anchoring restrictions.

    Oh well

  3. If the life of career criminal, George Floyd, is worth $27 million, the life of vet, patriot, freedom protester, Ashli Babbit, shot in the face for no reason in the Congress, is worth $100 million. The value of the life of George Floyd was in reality negative. Everyone, including that now wealthy family, boohooing, is far better off with his being dead. Criminals and addicts have a negative value. They destroy value. Once dead, we are all better off.

    Whatever, destructive effect Floyd’s life had, it is small compared to that of a lawyer, especially if a judge. Every year one of them lives, they destroy $5 million. The added value of their deaths to their families, to society, to the economy, should be deducted from any settlement.

    1. This study proves these settlements violate the Fifth Amendment procedural due process rights of defendants. Judges should be replaced by algorithms written by the legislature based on empirical valuations. The EPA places the value of a life at $6 million based on a market risk analysis, not based on the actual value of that life. That is far lower, and could be negative if the person is a tax sucking parasite.

      If the death of the plaintiff resulted in the end of abuse of his family, that value should be deducted from any settlement.

  4. We are sick of you lawyers taking our $trillion and returning nothing of value. This toxic profession must be crushed to save our nation.

  5. My recollection is that a lot of the more extreme anchoring and framing studies haven’t replicated – have you guys checked these out?

  6. “nor a sense of desert into a number of years in prison”

    Huh? Usually I can figure out what was intended in a typo. This one has me flummoxed.

    1. “Desert” as in “just deserts”. It’s related to “deserve.”

  7. There is a really good example of this phenomena at this time with voting fraud. Ask about fraud and the person who’s candidate lost, sees the number in thousands. Generally high enough to reverse the outcome. The person estimate has no basis on how that level of fraud could be carried out, but rather on what it takes to change the outcome. Real investigation of fraud often find the number in small fractions less than 100th of a percent and often less than 100 total cases. In Wisconsin, 27 cases in 3.3 M ballots were found to be suspected of fraud in 2020 election.

    1. Sorry, but no that’s not a good example at all. Dollars are discrete. You may argue about how many there are but there is no doubt about what counts as a dollar. Voter fraud, on the other hand, is a squishy topic where people first have to argue about what counts as fraud. Your Wisconsin example is a case in point. The actual finding was not 27 “suspected of fraud” but 27 “confirmed to a level that satisfied the reporter”. A different person could look at the same data and reach a very different conclusion not because of anchoring but merely because of the subjective definition of “suspected of fraud”.

      1. Your right there is a subjective definition and a number of the suspected cases would actually be lower. A number of cases were people using unacceptable addresses (i.e. UPS addresses). There is no indication these people actually intended to cheat and cast more than a vote they were allowed to cast.

        Fraud also has legal parameters and is not as subjective as you may think. It is necessary to show evidence and this is where the idea of thousand of cases runs in trouble. People either can not explain why they think there are thousand of cases and rely on their “gut feeling”. Or they propose theories that read like a Mission Impossible script.

        1. You are ignoring all the subjective factors that could make the real fraud number far higher than your guess. For example, you have implicitly assumed that absence of evidence means evidence of absence. Your approach misses all the fraud that’s never identified or that (for reasons unrelated to it’s actual existence) can’t be proven to whatever threshold the researcher wants to use. If you, for example, use the legal threshold of fraud, then only actual convictions count – and since it’s better that 10 guilty men go free than that 1 innocent man is punished, that means you’ve potentially undercounted by an order of magnitude on that factor alone. The legal threshold is “beyond reasonable doubt” not only that a crime was committed but that person X was responsible. That omits all the instances where the odds that the crime happened are >50% but not beyond reasonable doubt AND all the instances where we know that a crime happened but can’t necessarily prove who did it. It also omits all the instances of prosecutorial discretion and other factors that stop even well-demonstrated crimes from going to conviction.

          I don’t know what the real voter fraud rate is. My point is that you don’t have a clue either. Your guess that it’s low is as unsubstantiated as the “gut feelings” you deride about the people who think it’s higher. My concern is that people like you have made the conversation about voter fraud so toxic that society can no longer even have an honest conversation about whether voter fraud should be investigated.

  8. I’m sure this a real thing. But our adversarial system makes it the problem of the opposing sides lawyer to deal with.

  9. The CDC used this trick to attempt to deliberately mislead the public about the possibility of COVID transmission outdoors, which the CDC said was “less than 10%.” In reality, 10% was high by around two orders of magnitude. It’s a neat trick. Fortunately, they got called out on it.

Please to post comments