Estimation

Estimating a quantity

The "Lady Tasting Tea" story captures the core idea: we want to estimate a quantity (here, the lady's true ability to discriminate cups), not merely declare "she's right." Two main approaches:

  • Maximum likelihood estimation (MLE) — choose the parameter value that makes the observed data most probable.
  • Bayesian estimation — combine the data with a prior belief to produce a posterior distribution over the parameter.

Both yield an estimate plus uncertainty, which is the point of estimation — quantify it as a range with the Confidence Interval Calculator.

Estimating and comparing effects

  • The treatment effect — the difference in outcome between manipulated conditions; this is what a randomized experiment is built to estimate (see Experiments).
  • Measures of effect size — cohen's d, odds ratios, correlations, etc., describe how big an effect is, independent of sample size; the Effect Size Calculator converts between them.
  • Pros and cons of standardizing. Standardized effect sizes (e.g., Cohen's d) are comparable across studies but depend on the sample's variability; raw units are interpretable but not comparable. Report both when possible.

Estimating the relationship between variables

Beyond a single contrast, regression-style estimation summarizes how an outcome changes with predictors. This bridges directly into Models.

Key terms

  • Effect size — the magnitude of an effect, in standardized or raw units.
  • Point estimate — a single best guess of a parameter, usually with an interval expressing uncertainty.
  • Posterior — in Bayesian estimation, the updated belief after seeing data.

Further reading

Last updated: 2026-10-20

Menu
Loading