Why Do We Love Statistics That Aren’t Relevant?

We are drawn to statistics because patterns make an uncertain world feel predictable, even when those patterns have little connection to what happens next.

An impressive percentage gives a claim authority. A long streak suggests consistency, while several decimal places create the impression of careful measurement. Yet none of these qualities establishes whether the information is relevant to a decision.

The important distinction is between describing what happened and explaining why it happened. Historical statistics become useful when the relationships behind them help us understand present conditions. When those relationships are missing or no longer apply, even perfectly accurate numbers can mislead.

Why Numbers Make Patterns Convincing

People naturally search for relationships in their surroundings. Recognizing meaningful patterns helps us understand and respond to events, but the same tendency can make coincidence look like evidence.

In a 2008 study published in Science, Jennifer Whitson and Adam Galinsky found that participants experiencing a loss of control were more likely to perceive patterns in random or unrelated information. Across six experiments, participants sometimes identified relationships where none existed.

The study does not establish that every attractive statistic comes from a desire for control. It does illustrate how our desire to make sense of uncertain events can encourage us to find meaning in patterns that do not reliably describe a real relationship.

Adding numbers to those patterns often makes the resulting explanation sound stronger, even before we ask what the numbers actually measure.

The Christmas Day Football Statistic

Imagine an NFL broadcast explaining that a team has won 80% of its Christmas Day games over the last 40 years. The figure appears impressive because it combines a high percentage with a long period of history.

But 80% could represent four wins in five games. The games might involve different players, coaches, opponents, strategies, and rules separated by many years.

The statistic is accurate under the hypothetical example, but it does not explain why the team should be more likely to win its next Christmas game. Unless the holiday introduces a repeatable advantage, the pattern contributes little to an assessment of current performance.

A more relevant measure might describe how the team's offensive line has protected the quarterback in recent games. If the upcoming opponent relies on consistent pressure from four pass rushers, that information identifies a mechanism likely to matter in the matchup.

Both measures summarize historical observations. Only the second directly describes something related to the contest being played.

Small Samples Can Produce Large Percentages

A percentage can disguise how little information supports it. Winning four out of five games produces an 80% record, but it gives us much less evidence about sustained performance than a comparable percentage drawn from hundreds of games.

Precision creates a related problem. A statistic reported to one or two decimal places may sound carefully established, even when the underlying sample is small, selected, or poorly related to the current question.

A percentage is a compact way to represent a proportion. It does not independently establish sample quality, explanatory value, or the persistence of the relationship.

When an unusually impressive percentage appears, one of the first questions should be how many observations produced it and why those observations were grouped together.

Randomness Produces Streaks

Flip a fair coin repeatedly and patterns will emerge. Some sequences will contain several consecutive heads, others several tails, and some will look surprisingly organized.

Those streaks do not establish that the coin has changed. They arise naturally when chance outcomes are observed repeatedly.

The same reasoning becomes more complicated when people evaluate athletes. A basketball player who makes several consecutive shots can appear to have entered a special period of performance, commonly called the hot hand.

Researchers have debated how often that interpretation is justified. An influential 1985 study by Thomas Gilovich, Robert Vallone, and Amos Tversky questioned how reliably observed shooting streaks reflect genuine changes in success probabilities. Later research by Joshua Miller and Adam Sanjurjo identified a statistical bias in the earlier analysis and reported evidence of real hot-hand effects.

The important conclusion is not that every streak is random. It is that observing a streak alone cannot establish its cause. The next step is to determine whether the conditions governing the outcome have actually changed.

Why Extraordinary Statistics Are Easy to Find

Imagine analyzing a football team's record under every possible combination of weather, stadium, opponent, kickoff time, month, holiday, and day of the week.

Thousands of comparisons become possible. Eventually, some combinations will produce extraordinary-looking results simply because so many have been examined.

Perhaps a team has won eight consecutive Thursday games played in November when the temperature was below 40 degrees. The record could be completely true while providing little information about the team's chances in a future game.

The audience sees the exceptional statistic rather than the many combinations that produced nothing remarkable. That selection process makes the discovered pattern appear more important than it is.

This is closely related to the statistical problem of multiple comparisons. As the number of relationships examined increases, so does the opportunity to find results that appear exceptional through chance alone.

The American Statistical Association has warned about related problems in which attractive results are selectively reported after examining multiple analyses. What is interesting to report is not automatically useful evidence.

Historical Success Can Lose Its Relevance

The problem extends beyond sports into business decisions. Consider two hypothetical companies that have each recorded 25 consecutive years of revenue growth.

Company A supplies equipment used to maintain electrical infrastructure. Its customers replace aging components, build new networks, and continue meeting recurring maintenance needs. Historical growth is connected to a set of demand drivers that remain recognizable.

Company B sells equipment based on technology that dominated its market for decades. The company's growth came during a long adoption period with relatively limited competition, but customers have begun moving toward cheaper alternatives.

Both companies can truthfully advertise 25 years of uninterrupted revenue growth. Their histories have not changed, but the environments in which they operate have.

For Company A, the record offers some evidence of performance under demand conditions that continue to exist. For Company B, the record largely reflects success in a market structure that is disappearing.

Neither streak guarantees future results. Company A could lose customers or face unexpected competition, while Company B could adapt successfully. The distinction is whether the conditions that generated the historical record remain relevant to the current decision.

A Statistic Is Not an Explanation

We often treat a historical record as though it explains itself. A company achieved 25 years of growth, a team won several holiday games, or an investor delivered strong returns over a particular period.

Each is an observation. To interpret it, we need to understand the forces that produced it, the alternatives that were possible, and the circumstances under which the relationship might stop working.

A winning record may reflect a persistent competitive advantage, unusual conditions, or chance. Strong business performance may reflect customer demand, operational execution, or an advantage that competitors have not yet challenged.

Without that context, the statistic provides a summary but not a reliable explanation of what to expect next.

Test the Mechanism

Before using a historical statistic to support a decision, begin with a few questions. What exactly does the number measure, and what caused the observed result? Are those causes still operating, and what evidence would show that the relationship has changed?

Then examine how the statistic was selected. Was it identified because someone had a relevant question in mind, or because an extensive search produced an unusual result? Consider sample size and whether more recent information better reflects the conditions involved.

Finally, decide how much weight the statistic deserves. A meaningful pattern can provide evidence without proving what comes next, while a dramatic record can deserve almost no weight when it has little connection to the present situation.

Accuracy Is Only the Beginning

Statistics help us summarize a complicated world. They compress many observations into percentages, averages, streaks, rates, and comparisons that are easier to remember and discuss.

That compression also removes context, including information about the conditions and processes that produced the result. Recovering that context is necessary before using historical records to evaluate current choices.

A statistic can be completely accurate and still be irrelevant. It becomes informative when we establish why the relationship existed and whether that explanation still applies.

The next time a statistic sounds impressive, ask what made it true and whether that still matters.