Forecast for 2021

This year will be better or worse than the last one.

Posted Jan 01, 2021

It will get worse before it gets better. —Diverse swamis and wiseacres

When things are really bad, they can only get better. —Some of the same swamis and wiseacres, including E. R. Krüger

Last year, which was a good year for writing manuscripts, I struggled with the question of whether it is harder to make predictions than it is to find explanations, or whether the opposite might be true (Krueger, 2020). I considered arguments of each kind and found them all to have some merit. There seemed to be no way to aggregate over these arguments to arrive at an overall judgment. To put it simply: Predictions suffer from multiple uncertainties, unknown unknowns, and human ignorance; explanations suffer from hindsight biases and overeager pattern recognition; and both suffer from wishful thinking.

When, until yesterday, we were asking ourselves whether 2021 would be a better year than 2020 was, we were engaging in prediction-making under the shadow of uncertainty. For many of us, it seemed – with good reason – that 2020 was far worse a year than any other year in personal memory. Therefore, we were tempted to go with the wisdom of E. R. Krüger (my paternal grandfather) and predict that 2021 would be better. Were we entitled to make this prediction? Maybe not. Recall that wishful thinking is a logical fallacy. The pleasantness of a conclusion (or prediction) has no bearing on its truth value.

Grandpa’s wisdom reflects a recognition of the regression effect (Fiedler & Krueger, 2012). When things are very good (or bad), they are likely to be followed by things that are less good (or less bad). Sir Francis Galton (1869) stumbled into this verity when studying hereditary genius. Perhaps he was hoping that genius would compound over the generations like interest in the Bank of England. Alas, it is not so. Galton, himself a lateral descendant (actually a younger half-cousin) of Charles Darwin’s, was “wicked smaht” (to use a famous phrase from Good Will Hunting), but Uncle Chuck was smarter still. Yet, Galton himself was smarter than most of his relatives that might be brought up for comparison. Regression occurs when we select an extreme event or instance, sample another one from the same population, and find, more likely than not, that the second event or instance is less extreme than the first.

What does the regression effect imply for our prognosis for 2021? Not much, I am afraid. Suppose we had an overall score of badness for all years (such a score is, of course, fiction, but one might take the average judgment made by a large sample of human observers as a proxy). Say we enter the years from 1960 to 2019 into the pool. We draw one year, note its score, and compare it with the score of 2020. Chances are that 2020 was the worse year. Now suppose we have obtained scores for the years 2021 to 2080. We draw one of these years, note its score, and compare it with the score of 2020. We might find that, more likely than not, 2020 was worse than the sampled comparison year. But this would only be so if there was a fairly stable grand mean of good-/badness over all the decades, which is something that we do not know.

Any determination of whether there is a regression effect requires the specification of an underlying statistical model. The properties of long-term stability and random local variation are critical here. In empirical studies, researchers can ensure that these requirements are met, and they can then show that people underestimate regression; that they have a bias to think that extreme events will replicate, or even become more extreme. People over-extrapolate short local trends. 

So then, is optimism about 2021 a case of rational prediction, where people properly account for the regression effect? Not necessarily. Researchers have also documented a bias that is the opposite of regression neglect. This is the so-called gambler’s fallacy (Tune, 1964). This casino-type fallacy encourages risky bets on outcomes not recently seen. It is as though the gamblers think the random process has a memory and a motive to correct itself in the short run (Gold & Hester, 2008). The gambler’s fallacy is not a strong regression effect. People do not predict a return to the mean; they predict an event on the other side of the mean. In our scenario, they’d predict that 2021 would be a particularly good year, perhaps the "best ever," as prognosticated by Mr. Trump at some point.

Again, though, as in the case of regression neglect, the gambler’s fallacy can be demonstrated only if a particular statistical model is specified. Curiously, it’s the same model, featuring a stable mean and a local random variation. Then why do people sometimes neglect regression to the mean, and other times overcompensate for it? The answers to this question are complex. A short, blog-worthy answer is that in the first case, people have a false causal model implying a positive and too-strong an auto-correlation among successive events, whereas in the latter case they have a causal model with a negative auto-correlation. Lest you think I am just tautologically restating the phenomenon, I’ll have you know that research has explored the conditions under which such different types of causal models emerge (e.g., Ayton & Fisher, 2004).

As to our intuition of whether the year 2021 will be better (or worse!) than the year 2020, we ought to recognize that we do not know the statistical model that best describes the variation of annual goodness over historical time. Chances are the model must be more complex than a simple one comprising only a mean and random variation. As students of business, climate, or disease cycles know, the modelling soon becomes complex. And the models, by necessity, fit past data better than future data.

Suspended between complexity and uncertainty, we might just fall back on wishful thinking. Hope trumps despair, one hopes. So let’s make 2021 a good one!   

References

Ayton, P., & Fisher, I. (2004). The hot hand fallacy and the gambler’s fallacy: Two faces of subjective randomness? Memory & Cognition, 32, 1369–1378.

Fiedler, K., & Krueger, J. I. (2012). More than an artifact: Regression as a theoretical construct. In J. I. Krueger (Ed.). Social judgment and decision-making (pp. 171-189). Psychology Press.

Galton, F. (1869). Hereditary genius. Macmillan.

Gold, E., & Hester, G. (2008). The gambler's fallacy and the coin's memory. In J. I. Krueger (Ed.), Rationality and social responsibility: Essays in honor of Robyn Mason Dawes (p. 21–46). Psychology Press.

Krueger, J. I. (2020). Prediction and explanation in a postmodern world. Frontiers in Psychology: Theoretical and philosophical psychology. 11:597706. https://doi.org/10.3389/fpsyg.2020.597706

Tune, G. S. (1964). Response preferences: A review of some relevant literature. Psychological Bulletin, 61, 286–302.