"I don't believe in coincidences"

Indeed, gambling is one of the practical examples of the oft-studied phenomenon of just how badly humans typically estimate probability by intuitive means. Pseudo-statistical notions of a "run" of bad luck, or that snake-eyes are "due" because they haven't been rolled in a while, are the psychological basis behind the existence of Las Vegas in all its glitzy glory.

This is why I keep a statistician within communication distance. Lots of engineering decisions are based on probability and prediction of failure (i.e., reasoning through a sea of consequence and coincidence), and my staff is becoming very attuned to the notion that their armchair estimates of probability are not good measurements for what we should actually do. Statistical probability consistently defies even well-informed intuition. Hence "suspicious" coincidence is often predicated on wrong estimates of probability and "normal" outcomes, coupled with a misunderstanding of the law of Large Numbers.

Consider the classic fair coin toss. The probability of heads in any one toss is always p = 0.5, and always p = 0.5 for tails. Always. But intuitively we know that the law of large numbers means that for some long string of recorded coin tosses, the proportion of heads to tails outcome will tend toward 1:1. But if our running count is 453 heads to 472 tails, it is not true that the system will somehow "correct" itself by making heads a more likely outcome on the next toss. Yet this is how many gamblers think, and the basis upon which billion-dollar casino/hotels are funded. But in fact there is no "history" in this sort of statistical process. A coin toss always has evenly split probability -- always.

Considered another way, it's just as likely you'll be dealt a royal flush as any other combination of cards. The difference is solely in how those combinations are scored for poker. It is that imposed significance, not any sort of behavior of the underlying cards, that creates the sense of coincidence.

Your need for a Statistician to help correct intuitive judgement sounds like it reflects the need to account for failures in thinking as Kahneman outlines in "Thinking Fast and Slow".
 
Last edited:
Your need for a Statistician to help correct intuitive judgement sounds like it reflects the need to account for failures in thinking as Kahneman outlines in "Thinking Fast and Slow".

Definitely the case, however in the engineering world intuition often gets expressed quantitatively. That is, engineers know that some particular design factor is quantitative, but rather than compute it appropriately based on statistical models, they settle for estimates based on first-order computational methods or straw-poll surveys of past performance.

Estimating the reliability of a punch press in an assembly line, for example, helps determine the efficiency and capacity of the factory it belongs to. You can just pluck a PDF out of your orifices, or you can ask the foreman for his recollection of how often punch presses break down, or you can play around with the punch press yourself and see how reliable it seems to be, or you can go to the library and read rigorously controlled academic studies on the matter.

Kahneman gets into more qualitative territory, and it's all very valid. Illusion of validity goes into all aspects of estimation and prediction, not just that for which we retain statisticians. Indeed we might be thinking about problems all wrong and mistaking what factors we should consider. But even within correct models we have parameters that can either be estimated or computed. The statisticians help us compute them properly, and in so doing show us how previously bad our estimates were.
 

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom