Noise: A Flaw in Human Judgment
Daniel Kahneman, Olivier Sibony, and Cass R. Sunstein
Geoffrey W. Sutton
We are constantly exposed to opinions. Some of those opinions are judgments. And some of those judgments affect our opportunities to work, obtain healthcare, receive fair treatment by government entities, and earn fair evaluations in school. Some people are paid to make informed judgments. Unfortunately, some judgments are noisy—they vary. Noise is about the differences in judgments that affect our lives.
When the authors provide examples of variation in judgments, they are writing about variability in a statistical sense. As a retired professor who taught research and statistics to undergraduate and graduate students, I’m not sure the authors were entirely clear—at least not clear enough for readers who are either new to the concept or haven’t drawn on their statistics knowledge for some years. In any event, I think the book deserves a look because it draws attention to a real problem—a problem with which I’m familiar.
The authors often provide examples of judges in the criminal justice system. For example, they note that judges can vary in the length of a sentence for the same offence. They also provide examples of different diagnoses provided by physicians for the same set of symptoms—this is especially true in the diagnosis of mental disorders.
The authors introduce the problem of variation in judgments by referring to shots at target. Given a group of people firing at a target, there will likely be some variation. If they are experienced, we would expect them to be close to the bullseye. The degree to which the holes are scattered is variance. The variation is like noise in judgments, which deviate from accuracy. Those deviations represent error.
Let me suggest dropping the term variance in judgments in favor of differences. We can expect people to have differences of opinions about one thing or another but when it comes to a medical diagnosis, we want an accurate decision. When different experts arrive at different diagnoses, that’s noise. And that can be scary when the diagnosis leads to very different types of treatment.
Bias is a related concept. Bias is a systematic type of error. Using the target analogy, bias reveals a tendency for all the deviations from the bullseye to be located in the same area. In psychology, it might be a tendency for some clinicians to diagnose anxiety rather than ADHD or ADHD rather than anxiety when observing fidgety children. Bias can be found in numerical scores too such as when some psychologists tend to obtain lower scores than others on intelligence tests.
The authors also cover the problem of transient differences or occasion noise. That’s the kind of inconsistency that can happen when the same person looks at the same data but comes up with a different judgment on two different occasions. The authors mention some well known influences like time of day and hunger affecting judgments.
I’ll skip ahead to their recommendations. After providing us with additional terms and many examples, the authors offer suggestions for controlling unwanted noise. One major suggestion is to rely on algorithms based on the evidence that computerized assessment of all relevant data can often beat human decision-makers in accuracy. The authors recognize this won’t sell well to a lot of readers but they do offer a defense against common objections.
A second, and in my mind more palatable approach is to create a structured approach to decision-making. This can be as simple as guidelines, checklists, and preset questions to use in various fields. In applied psychology and counseling, students learn to use checklists and decision trees when making a diagnosis. Others learn to use scales and questionnaires and ways to aggregate available information relevant to both diagnoses and treatment plans.
There’s a lot more in this book both in terms of examples of noise as well as suggestions for reducing noise in different areas of life. They supplement their work with useful appendixes: How to conduct a noise audit, A checklist for a decision observer, correcting predictions.
Kahneman, D., Sibony, O., & Sunstein, C.R. (2021). Noise: A flaw in human judgment. New York: Hatchette.
Also, consider connecting with me on FACEBOOK Geoff W. Sutton
You can read many published articles at no charge: