The authors of Noise: A Flaw in Human Judgment* examine the random variations that occur in judgmental decisions and recommend ways to make more consistent judgments. Variability is observed when two or more qualified decision makers review the same data or face the same situation and come to different judgments or conclusions. (Variability can also occur when the same decision maker revisits a previous decision situation and arrives at a different judgment.) The decision makers may be doctors making diagnoses, engineers designing structures, judges sentencing convicted criminals, or any other situation involving professional judgment.** Judgments can vary because of two factors: bias and noise.
Bias is systematic, a consistent source of error in judgments. It creates an observable average difference between actual judgments and theoretical judgments that would reflect a system’s actual or espoused goals and values. Bias may be exhibited by an individual or a group, e.g., when the criminal justice system treats members of a certain race or class differently from others.
Noise is random scatter, a separate, independent cause of variability in decisions involving judgment. It is similar to the residual error in a statistical equation, i.e., noise may have a zero average (because higher judgments are balanced by lower ones) but noise can create large variability in individual judgments. Such inconsistency damages the credibility of the system. Noise has three components: level, pattern, and occasion.
Level refers to the difference in the average judgment made by different individuals, e.g., a magistrate may be tough or lenient.
Pattern refers to the idiosyncrasies of individual judges, e.g., one magistrate may be severe with drunk drivers but easy on minor traffic offenses. These idiosyncrasies include the internal values, principles, memories, and rules a judge brings to every case, consciously or not.
Occasion refers to a random instability, e.g., where a fingerprint examiner looking at the same prints finds a match one day and no match on another day. Occasion noise can be influenced by many factors including a judge’s mood, fatigue, and recent experience with other cases.
Based on a review of the available literature and their own research, the authors suggest that noise can be a larger contributor to judgment variability than bias, with stable pattern noise larger than level noise or occasion noise.
Ways to reduce noise
Noise can be reduced through interventions at the individual or group level.
For the individual, interventions include training to help people who make judgments realize how different psychological biases can influence decision making. The long list of psychological biases in Noise builds on Kahneman’s work in Thinking, Fast and Slow which we reviewed on Dec. 18, 2013. Such biases include overconfidence; denial of ignorance, which means not acknowledging that important relevant data isn’t known; base rate neglect, where outcomes in other similar cases are ignored; availability, which means the first solutions that come to mind are favored, with no further analysis; and anchoring of subsequent values to an initial offer. Noise reduction techniques include active open-mindedness, which is the search for information that contradicts one’s initial hypothesis, or positing alternative interpretations of the available evidence; and the use of rankings and anchored scales rather than individual ratings based on vague, open-ended criteria. Shared professional norms can also contribute to more consistent judgments.
At the group level, noise can be reduced through techniques the authors call decision hygiene. The underlying belief is that obtaining multiple, independent judgments can increase accuracy, i.e., lead to an answer that is closer to the true or best answer. For example, a complicated decision can be broken down into multiple dimensions, and each dimension assessed individually and independently. Group members share their judgments for each dimension, then discus them, and only then combine their findings (and their intuition) into a final decision. Trained decision observers can be used to watch for signs that familiar biases are affecting someone’s decisions or group dynamics involving position, power, politics, ambition and the like are contaminating the decision process and negating actual independence.
Noise can also be reduced or eliminated by the use of rules, guidelines, or standards.
Rules are inflexible, thus noiseless. However, rules (or algorithms) may also have biases coded into them or only apply to their original data set. They may also drive discretion underground, e.g., where decision makers game the process to obtain the results they prefer.
Guidelines, such as sentencing guidelines for convicted criminals or templates for diagnosing common health problems, are less rigid but still reduce noise. Guidelines decompose complex decisions into easier sub-judgments on predefined dimensions. However, judges and doctors push back against mandatory guidelines that reduce their ability to deal with the unique factors of individual cases before them.
Standards are the least rigid noise reduction technique; they delegate power to professionals and are inherently qualitative. Standards generally require that professionals make decisions that are “reasonable” or “prudent” or “feasible.” They are related to the shared professional norms previously mentioned. Judgments based on standards can invite controversy, disagreement, confrontation, and lawsuits.
The authors recognize that in some areas, it is infeasible, too costly, or even undesirable to eliminate noise. One particular fear is a noise-free system might freeze existing values. Rules and guidelines need to be flexible to adapt to changing social values or new data.
Our Perspective
We have long promoted the view that decision making (the process) and decisions (the artifacts) are crucial components of a socio-technical system, and have a significant two-way influence relationship with the organization’s culture. Decision making should be guided by an organization’s policies and priorities, and the process should be robust, i.e., different decision makers should arrive at acceptably similar decisions.
Many organizations examine (and excoriate) bad decisions and the “bad apples” who made them. Organizations also need to look at “good” decisions to appreciate how much their professionals disagree when making generally acceptable judgments. Does the process for making judgments develop the answer best supported by the facts, and then adjust it for preferences (e.g., cost) and values (e.g., safety), or do the fingers of the judges go on the scale at earlier steps?
You may be surprised at the amount of noise in your organization’s professional judgments. On the other hand, is your organization’s decision making too rigid in some areas? Decisions made using rules can be quicker and cheaper than prolonged analysis, but may lead to costly errors. which approach has a higher cost for errors? Operators (or nurses or whoever) may follow the rules punctiliously but sometimes the train may go off the tracks.
Bottom line: This is an important book that provides a powerful mental model for considering the many factors that influence individual professional judgments.
* D. Kahneman, O. Sibony, and C.R. Sunstein, Noise: A Flaw in Human Judgment (New York: Little, Brown Spark) 2021.
** “Professional judgment” implies some uncertainty about the answer, and judges may disagree, but there is a limit on how much disagreement is tolerable.