Noise: A Flaw in Human Judgment

by Daniel Kahneman

“It is easier to bias forensic experts towards the non-committal conclusion of ‘inconclusive’ than to the definitive ‘identification’ conclusion.” Examiners are trained to consider erroneous identification as the deadly sin to be avoided at all costs. To their credit, they act in accordance with this principle.

A task like the analysis of fingerprints seems objective, so much so that many of us would not spontaneously regard it as a form of judgment. Yet it leaves room for inconsistency, disagreement, and, occasionally, error. However low the error rate of fingerprint identification may be, it is not zero, and as PCAST noted, juries should be made aware of that.

The first step to reduce noise must be, of course, to acknowledge its possibility.

Noise can be an invisible problem, even to people whose job is to see the invisible.

a decision hygiene strategy that has applicability in many domains: sequencing information to limit the formation of premature intuitions.

the laboratory keeps them as much in the dark about the case as possible and reveals information only gradually.

linear sequential unmasking.

another recommendation that illustrates the same decision hygiene strategy: examiners should document...

When a different examiner is called on to verify the identification made by the first person, the second person should not be aware of the first judgment.

judgments can be altered by anger, fear, or other emotions, and perhaps you have noted that it is a good practice, if you can, to revisit your judgment at different points in time, when the triggers of occasion noise are likely to be different.

as soon as you know what others think, confirmation bias can lead you to form an overall impression too early and to ignore contradictory information.

“We have more information about this case, but let’s not tell the experts everything we know before they make their judgment, so as not to bias them. In fact, let’s tell them only what they absolutely need to know.”

“The second opinion is not independent if the person giving it knows what the first opinion was. And the third one, even less so: there can be a bias cascade.”

bias and noise (also called inconsistency or unreliability).

ongoing quarterly survey asks the chief financial officers of US companies to estimate the annual return of the S&P 500 index for the next year. The CFOs provide two numbers: a minimum, below which they think there is a one-in-ten chance the actual return will be, and a maximum, which they believe the actual return has a one-in-ten chance of exceeding. Thus the two numbers are the bounds of an 80% confidence interval. Yet the realized returns fall in that interval only 36% of the time. The CFOs are far too confident in the precision of their forecasts.

two noise-reduction strategies that have broad applicability.

selecting better judges produces better judgments.

aggregating multiple independen...

the wisdom of crowds works best when judgments are independent,

prediction markets, in which individuals bet on likely outcomes and are thus incentivized to make the right forecasts. Much of the time, prediction markets have been found to do very well, in the sense that if the prediction market price suggests that events are, say, 70% likely to happen, they happen about 70% of the time. Many companies in various industries have used prediction markets to aggregate diverse views.

Delphi method.

involves multiple rounds during which the participants submit estimates (or votes) to a moderator and remain anonymous to one another. At each new round, the participants provide reasons for their estimates and respond to the reasons given by others, still anonymously. The process encourages estimates to converge (and sometimes forces them to do so by requiring new judgments to fall within a specific range of t...

mini-Delphi, can be deployed within a single meeting. Also called estimate-talk-estimate,

the Good Judgment Project.

given our objective ignorance of future events, it is much better to formulate probabilistic forecasts.

Good calibration is one requirement for good forecasting.

(A well-known response to this criticism, sometimes attributed to John Maynard Keynes, is, “When the facts change, I change my mind. What do you do?”)

Glenn W. Brier in 1950. Brier scores, as they are known, measure the distance between what people forecast and what actually happens.

Margaret, whom we described as a well-calibrated forecaster because she rated 500 events as 60% likely, and 300 of those events did happen. This result may not be as impressive as it seems. If Margaret is a weather forecaster who always predicts a 60% chance of rain and there are 300 rainy days out of 500, Margaret’s forecasts are well calibrated but also practically useless. Margaret, in essence, is telling you that, just in case, you might want to carry an umbrella every day. Compare her with Nicholas, who predicts a 100% chance of rain on the 300 days when it will rain, and a 0% chance of rain on the 200 dry days. Nicholas has the same perfect calibration as Margaret: when either forecaster predicts that X% of the days will be rainy, rain falls precisely X% of the time. But Nicholas’s forecasts are much more valuable: instead of hedging his bets, he is willing to tell you whether you should take an umbrella.

Nicholas is said to have a high resolution in addition to good calibration.

scores reward both good calibration and good resolution. To produce a good score, you have not only to be right on average (i.e., well calibrated) but also to be willing to take a stand and differ...

superforecasters.

real advantage is not their talent at math; it is their ease in thinking analytically and probabilistically.

superforecasters’ willingness and ability to structure and disaggregate problems. Rather than form a holistic judgment about a big geopolitical question

they break it up into its component parts. They ask, “What would it take for the answer to be yes? What would it take for the answer to be no?” Instead of offering a gut feeling or some kind of global hunch, they ask and try to answer an assortment of subsidiary questions. Superforecasters also excel at taking the outside view, and they care a lot about base rates.

In short, what distinguishes the superforecasters isn’t their sheer intelligence; it’s how they apply it.

To characterize the thinking style of superforecasters, Tetlock uses the phrase “perpetual beta,” a term used by computer programmers for a program that is not meant to be released in a final version but that is endlessly used, analyzed, and improved.

“the strongest predictor of rising into the ranks of superforecasters is perpetual beta, the degree to which one is committed to belief updating and self-improvement.”

“What makes them so good is less what they are than what they do—the hard work of research, the careful thought and self-criticism, the gathering and synthesizing of other perspectives, the granular judgments and relentless updating.” They like a par...

people can be trained to be superforecasters or at least to perform more like them.

three major reasons why some forecasters can perform better or worse than others: 1. They can be more skilled at finding and analyzing data in the environment that are relevant to the prediction they have to make. This explanation points to the importance of information. 2. Some forecasters may have a general tendency to err on a particular side of the true value of a forecast. If, out of hundreds of forecasts, you systematically overestimate or underestimate the probability that certain changes from the status quo will occur, you can be said to suffer from a form of bias, in favor of either change or stability. 3. Some forecasters may be less susceptible to noise (or random errors). In forecasting, as in any judgment, noise can have many triggers. Forecasters may overreact to a particular piece of news

BIN (bias, information, and noise) model for forecasting.

all three interventions worked primarily by reducing noise. As the researchers put it, “Whenever an intervention boosted accuracy, it worked mainly by suppressing random errors in judgment. Curiously, the original intent of the training intervention was to reduce bias.”

Tetlock’s training is designed to fight psychological biases. As you now know, the effect of psychological biases is not always a statistical bias.

When working in groups, the superforecasters seem capable of avoiding the dangers of group polarization and information cascades. Instead, they pool their data and insights and, in their actively open-minded way, make the most of the combined information.

Superforecasters are better than others at finding relevant information—possibly because they are smarter, more motivated, and more experienced at making these kinds of forecasts than is the average participant.

The task of assembling a team of professionals to make judgments together resembles the task of assembling a battery of tests to predict the future performance of candidates at school or on the job. The standard tool for that task is multiple regression

It works by selecting variables in succession. The test that best predicts the outcome is selected first.

the next test to be included is not necessarily the second most valid. Instead, it is the one that adds the most predictive power to the first test, by providing predictions that...

if you are assembling a team of judges, you should of course pick the best judge first. But your next choice may be a moderately valid individual who brings some new skill to the table rather than a more valid judge who is highly similar to the first one.