Friday, July 16, 2021

Chapter review: Judgments and Models



Context
The 2021 book, Noise: A Flaw in Human Judgment, deals with the normal human illusion that the world is a usually causally understandable and predictable place. The book was written by Nobel laureate Daniel Kahneman and two other authors. The main point of the book is to argue that noise in the real world is an undeniable and important but shockingly ignored factor in human judgment error. Judgment errors have two components, bias and noise. In the context of this book, noise is observed and measured as an innate human tendency to make errors that are not due to a known bias. With bias, one can generally predict the kinds of error people will tend to make. This is aspect of the human condition is very well-known and and generally well accepted. A few people with the moral courage to face it, even try to account for their personal biases to qualitatively and/or quantitatively reduce their errors.

On the other hand, noise in human judgment does not lead to predictable errors. That is not so well known and hard for many people to accept. It manifests as random scatter in erroneous judgments with no apparent reason other than the uniqueness of the person making the mistakes.

Even various experts in social science, medicine and commerce, who all should know better, simply blow this aspect of reality off because it is psychologically uncomfortable and/or threatening. Human egos tend to be fragile and reflexively self-defensive in the face of threat. That helps mask unpredictability and sources of causation in real life. There is an awful lot of overconfidence going on among an awful lot of people, but it may be inevitable. Most humans just don’t easily learn and practice things outside the comfort zones of their cognitive capacity or their ego.

Most of us believe most events that have happened were causally understandable, predictable and easily explained. That is sometimes true in the physical sciences, but it usually isn’t true in the social sciences and human life generally, including politics. The reasons for overconfidence to the point of illusion are now reasonably understood. They are grounded in (i) how the human mind evolved to deal with reality with as little effort as possible, and (ii) an innate human inability to see and think about the people and the world according too sets of rules, even very simple rules, that significantly (but modestly) reduce human error. 

Based on the human error rates and the reasons therefore discussed in Parts III, Noise in Predictive Judgments, and V, Improving Judgments, it seems reasonable to believe (my estimate) that if that content was mandatory subject matter in public schools and in post high school education, society maybe could avoid about 10-15% of the error and waste inherent in how we do things. In a $20 trillion/year economy, that might translate to ~$2 trillion less economic and human waste per year, maybe more. 

Chapter 9, like most of the rest of the book, relies heavily on statistics and basic statistical concepts. Because statistics is not how the human mind usually thinks, this review downplays statistical concepts and jargon as much as possible without rendering it incoherent or incorrect. 


Chapter 9: Judgments and Models
Chapter 9 focuses on why even models of reality that are so simple as to seem ridiculous are almost always better than most experts and non-experts in making real world predictive judgments. That means judgments about all kinds of things, from medical diagnoses and corporate hiring decisions to predictions of war and decisions whether to grant bail before trial to someone in jail. 

What models of reality do is take noise out of human judgments. That’s all they do. It is the only thing they do. Simple models can be flawed, but even then, they usually beat most everyone most of the time. Few people can beat models, even simple flawed ones. Why is this the case?

It is the case because humans usually apply a form of thinking that Kahneman calls “clinical judgment” to real world problems and decisions. Clinical judgment has noise in it because it is unique to the person applying it. Models do not have this variability. Applying the rules of a model to solving a problem or making a decision uniformly applies various rules relevant to the problem or decision. Kahneman calls that kind of a process “mechanical judgment.” It lacks noise because the rules are applied uniformly.

And that is where the difficulty in seeing and accepting the power of mechanical judgment compared to clinical judgment hits a brick wall. Imagine a successful, experienced doctor, or a researcher with a PhD and many peer-reviewed publications, or a high level business executive with years of training and experience being told their judgment is not as good as a simple model based on just two or three rules. It does not matter how solid the evidence is that the model is better most of the time. The model threatens ego and professional esteem.[1] Most people making judgments and predictions suffer from the illusion of validity, which is a false belief that their judgment is significantly better than it really is. One problem is that the future is usually quite uncertain and people just do not accept that reality. People believe they have the information they need to make the right judgment, but fail to understand how likely their decision is to be undone by fate. Another is that the mind tends to morph bad decisions into good ones over time by memory distorting processes. Kahneman summed it up like this:
“If you are as confident in your predictions as you are in your evaluation of cases, however, you are a victim of the illusion of validity. .... The reaction is easy to understand: Mheel’s[1] pattern contradicts the subjective experience of judgment, and most of us will trust our [often illusory] experience over a scholar’s claim.”
And, it gets even weirder. Researchers found that when they build a model of individual professionals, the model beats the professional most of the time. In other words, the model of you beats you. Models of judges beat judges, and that was based on 50 years worth of data. Kahneman comments:
“The model-of-the-judge studies reinforce Meehl's conclusion that the subtlety is largely wasted. Complexity and richness do not generally lead to more accurate predictions. .... In short, replacing you with a model of you does two things: it eliminates your subtlety and it eliminates your pattern (personal) noise.”
And, it gets even weirder than that. When random models of individuals were built and tested, not the best simple models, the random ones generally did better than the experts:
“Their striking finding was that any linear model, when applied consistently to all cases, was likely to outdo human judges predicting an outcome from the same information. .... Or, to put it bluntly, it proved almost impossible in that study to generate a simple model that did worse than the experts did.”
Poor humans. What a hot mess. Professionals tend to dislike this line of research. No one likes their judgments being called noisy. Egos get banged up and shorts are in a twist. There is much consternation, huffing, puffing and criticisms of the research. So far, all the criticisms (over 20 by now) have been fully rebutted. The research and data are sound.


Footnote: 
1. In 1954 psychology professor Paul Meehl published a book that reviewed twenty studies where there was clinical judgment and then mechanical judgment predicting how the clinical judgments turned out.  His analysis clearly indicated that mechanical beat clinical, but professionals really did not like that outcome. “Meehl discovered that clinicians and other professionals are distressingly weak in what they often see as their unique strength: the ability to integrate information.” Turns out that their perceived unique strength was their actual unique weakness.

No comments:

Post a Comment