The Average Person Is a Statistical Convenience

A mean is useful. It is not a human being.

Averages summarize groups. They can reveal product differences, support estimation and reduce a complex dataset to something decision-makers can use. The problem begins when the summary is mistaken for the people it summarizes.

Two samples can have the same mean and very different distributions. A modest average change can reflect a consistent small shift, two opposing subgroups or a few extreme responders. The same person may also respond differently across occasions as physiology, learning, context and goals change.

Variation can be error or explanation

Some variation is measurement noise. Some reflects uncontrolled conditions. Some is a stable individual difference. Some is a state change. Some is produced by an interaction between product and person.

The job is not to celebrate every difference. It is to determine whether the variation is reliable, interpretable and consequential for the decision.

Group results do not automatically describe individuals

Human processes are often nonergodic: the relationships observed across people do not necessarily describe the relationships operating within one person over time. Group-to-individual generalization therefore has to be tested rather than assumed (Fisher et al., 2018; Molenaar, 2004).

Measurement creates another trap. A task can produce a highly replicable average effect precisely because between-person variability is small; that same task may be unreliable for ranking individuals or correlating differences with other variables (Hedge et al., 2018). A robust group effect is not proof of a useful individual-difference measure.

Segment only when the segmentation earns its keep

Post-hoc slicing can manufacture stories. A useful segment should be based on a defensible construct, measured reliably, replicated where possible and connected to an action the organization can take. It should improve prediction, design, targeting, safety, accessibility or interpretation rather than merely decorate the result.

Practical analysis should make distributions and uncertainty visible. Mixed-effects or hierarchical models can separate person, occasion and product sources of variation. Repeated measures can distinguish stable differences from transient state. Confidence intervals and individual trajectories often reveal more than a single bar chart.

The average consumer is not fake. The average is a valid statistic. It simply is not a person.

When variation changes the decision, the distribution, and sometimes the individual, belongs in the answer.