Definition
An outlier is an observation that differs conspicuously from the rest — a measurement or entry error, a different population, or a genuine rare event. It affects the mean, standard deviation, correlation and regression coefficients, so identify it before analysis (box plot, 1.5 × IQR rule, z-score, Cook's distance) and decide how to treat it.
Data & visualisation
Do not delete outliers automatically: verify the cause first. Correct or remove errors; keep genuine values and use robust methods (median, nonparametric tests, robust regression) or run the analysis with and without them and describe the difference.
In multivariate data a point may look ordinary in each variable separately yet be an outlier in their combination — Mahalanobis distance or PCA helps. In regression distinguish outlying response (large residual) from outlying predictors (high leverage).
In Statistica
Statistica flags outliers in box plots, in regression diagnostics (standardized residuals, Cook's distance, leverage) and in control charts; points can be selected in an interactive graph, traced to the spreadsheet and temporarily excluded with case selection conditions without altering the data. Data-cleaning tools are in the Data menu.
Related terms
- Box plotA box plot shows a distribution with five numbers: the box spans from the lower to the upper quartile (the middle 50 %…
- Mean, median and modeThe mean, median and mode are measures of central tendency — a single number summarizing where the data lie.
- Nonparametric testsNonparametric tests do not assume a normal distribution — they work with the ranks of values rather than the values themselves.
- Linear regressionLinear regression describes how a dependent variable (yield, price, consumption) depends on one or more explanatory…
- Control charts (SPC)A control chart is the basic tool of statistical process control (SPC): it plots measurements over time with a centre…
Knowledgebase guides
FAQ
- Should I remove an outlier?
- Only when you can demonstrate an error. Otherwise keep it and use robust methods, or report results with and without it.
- Which rule defines an outlier?
- Most often 1.5 times the interquartile range beyond the quartiles (box plot) or |z| > 3; in regression a Cook's distance above 1 or 4/n.
Try it on your own data
Statistica free for 30 days
Full version, no credit card. Or get a pricing estimate in a minute.