On July 25, 2026, our group of five explored the critical difference between correlation and causation. During our discussion, we examined a claim by the late "President for Life" of Turkmenistan, Saparmurat Niyazov. He advised citizens to strengthen their teeth by gnawing on bones, observing that young dogs given bones had strong teeth, while older dogs whose teeth had fallen out did not. Can the flaw in his logic be exposed by reversing the terms—realizing that strong teeth are required to gnaw on bones in the first place? This is a classic example of confusing correlation with causation (cum hoc ergo propter hoc). But how do we properly establish a true causal connection?
To examine correlation, let’s look at a scatter plot of Cartesian coordinates, placing adult children’s heights on the y-axis and parental heights on the x-axis. The resulting cloud of ordered pairs widens from the southwest corner and tapers off toward the northeast. By drawing a line of best fit through the middle of the mass, we obtain the regression line and calculate r, the correlation coefficient—a measure of the strength and direction of a linear relationship ranging from 1 (perfect positive) to -1 (perfect negative), with values near 0 indicating a weak relationship.
If we draw a line perpendicular to the x-axis where parental height is 6’, the corresponding child’s height on the regression line is 5'9". Conversely, for parents at 5’, the corresponding child’s height is 5'3". Does this imply that tall parents have shorter children and short parents have taller children? No—this is a statistical illusion arising from plotting two Gaussian distributions together, widely known as regression to the mean. We see this elsewhere, such as the "Sophomore Slump," where an exceptional rookie baseball player's second-year performance drops back toward their true average. It isn't laziness; it's simply regression to the mean after an unrepeatable streak. To avoid mistaking regression to the mean (or other confounding variables) for a true causal effect, scientists rely on Randomized Controlled Trials. By randomly assigning subjects to a treatment or control group, any background noise or natural statistical regression occurs equally in both groups, isolating the true impact of the intervention.
The human mind operates like an Aristotelian machine, projecting inherent essences and final causes onto the world around us. Aristotle’s framework relied on the certainty of necessity neatly ordered by circumstances, a chain linking ultimately to a prime mover. David Hume, however, argued that causation is mentally constructed—an expectation that past correlations will persist in the future. He shifted the paradigm to necessity arising from chance and from deduction to induction, reducing causation to statistical probability and relying on key analytical tools to account for confounding variables: 1. Condition: A necessary background state required for a cause to act (e.g., oxygen, fuel, and heat are needed to start a fire). 2. Epiphenomenon: A secondary byproduct with no causal power of its own (e.g., does a rooster's crow make the sun rise?). 3. Overdetermination: Occurs when multiple independent causes are each individually sufficient to produce an outcome (e.g., a firing squad). 4. Preemption: Occurs when two potential causes produce an effect (e.g., two boys throwing rocks at a bottle—whichever strikes first gets the "credit").
Computer scientist Judea Pearl extends this framework to further address the limitation on Hume's correlation through three core structural models. 1. Chain: A direct, sequential pathway. Example: Smoking deposits tar in lungs, increasing the risk of cancer. 2. Fork: A common background variable simultaneously causing two separate outcomes. Example: Hot weather increases both ice cream sales and shark attacks; controlling for heat causes the correlation between ice cream and shark attacks to disappear. 3. Collider: The inverse of a fork, where two independent causes converge on a single outcome. Example: Being conventionally handsome and being a genuinely nice person are independent traits that collide to make someone desirable to date. If your dating pool is restricted to attractive or nice people, you may falsely conclude that handsome people are usually mean—another illusion driven by selection bias and regression to the mean.
Join us at our next meeting, where we will examine why people think the way they do in Steven Pinker’s Rationality: What It Is, Why It Seems Scarce, Why It Matters, BF441.P56 2021, on August 8, 2026, from 2:00 PM to 4:00 PM.
