Lindley's Paradox — When Logic Wears a Disguise

Lindley's Paradox occurs when frequentist and Bayesian statistical methods produce contradictory conclusions from the same data. Specifically, a result can be statistically significant (low p-value) in a frequentist test while the Bayesian posterior probability strongly favors the null hypothesis. This disagreement becomes more pronounced with large sample sizes.

Also known as: Lindley-Jeffreys paradox, Jeffreys-Lindley paradox

How It Works

With large samples, frequentist tests can detect arbitrarily small effects and produce significant p-values for practically meaningless differences. Bayesian analysis penalizes vague alternative hypotheses because the likelihood is spread thinly across the parameter space, so the precise null hypothesis receives comparatively more support.

A Classic Example

A clinical trial with 100,000 participants finds a treatment effect of 0.01 units with p = 0.03. The frequentist rejects the null hypothesis. However, a Bayesian analysis with a reasonable prior concludes there is a 95% probability that the null hypothesis is true, because the observed effect is so small that it is more consistent with noise than a real effect at the prior's scale.

More Examples

A large government survey of 500,000 households finds that people in one region earn on average $200 more per year than the national average, with p = 0.04. The frequentist analyst declares a statistically significant regional wage gap. A Bayesian economist, incorporating prior knowledge that regional wage differences of that magnitude are extremely rare, concludes the posterior probability of a true gap is less than 15%.

A genomics study scanning 1 million genetic variants finds one SNP associated with a disease at p = 0.04 after correction. The frequentist flags it as significant. A Bayesian analysis incorporating the prior that most of the million tested variants have no true effect concludes the probability that this specific variant is a true positive is below 20%, suggesting the result is likely a false discovery.

Where You See This in the Wild

This paradox frequently arises in large-scale epidemiological studies, genomics (genome-wide association studies with millions of data points), and social science research with big data, where tiny effects routinely reach statistical significance.

How to Spot and Counter It

Report effect sizes alongside p-values. Consider Bayesian approaches or Bayes factors when sample sizes are large. Evaluate whether a statistically significant result is also practically meaningful. Be explicit about prior assumptions and the distinction between statistical and substantive significance.

The Takeaway

The Lindley's Paradox is one of those reasoning errors that sounds perfectly logical at first glance. That's what makes it dangerous — it wears the costume of valid reasoning while smuggling in a broken conclusion. The best defense? Slow down and ask: does this conclusion actually follow from these premises, or am I just connecting dots that happen to be near each other?

Next time someone presents you with an argument that "just makes sense," check the structure. The feeling of logic is not the same as logic itself.