MerQuR+Statistics Glossary

Statistics Glossary → Percentiles (quantiles) and percentile definitions

Percentiles (quantiles) and percentile definitionsconcept

The value below which a given share of the ordered data falls; the 90th percentile, for example, is the point below which about 90% of the observations lie.

A quantile is expressed as a proportion (between 0 and 1) and a percentile as a percentage: the 0.25 quantile is the 25th percentile (first quartile, Q1), and the 0.50 quantile is the median. Percentiles summarise location and spread regardless of the shape of the distribution; the interquartile range (IQR = Q3 − Q1) and the box plot are built on them. They are much less affected by outliers than the mean and standard deviation.

In a finite sample the requested proportion usually falls between two observations, and there is more than one rule for filling that gap. Hyndman and Fan (1996) classify nine definitions: type 7 is the default in R and Excel, type 6 in SPSS and type 2 in SAS. The differences are small in large samples, but in small samples different programs can report different Q1 and Q3 values from the same data, so the definition used should be reported.

An everyday example

A paediatrician says that a 2-year-old's height is at the 85th percentile. This means that about 85% of children of the same age are shorter than this child. The child's position among peers is clear without knowing the average height in centimetres.

In MerQur

Descriptive statistics has a 'Percentiles' list and a 'Percentile definition' option (Type 7, Type 6 SPSS HAVERAGE, Type 2 SAS PCTLDEF=5, R type 1/3/4/5/8/9). In the bootstrap CI the 'Statistic' list includes the percentile and the 'CI method' can be Percentile or BCa; 'Quantiles' in the weighted mean gives weighted quantiles, 'Quantiles (CSV)' in quantile regression sets the τ values to model, and parametric survival has 'Additional output: Predicted percentiles and S(t)'. Many analyses, such as the permutation test, Cohen's kappa and mediation analysis, also give percentile bootstrap intervals.

Analyses where this concept appears

Descriptive StatisticsOne-Sample t-TestIndependent-Samples t-TestPaired t-TestBootstrap CIPermutation TestFriedmanCohen's Kappa (including Fleiss κ and Krippendorff α)Correlation AnalysisBland-Altman Agreement AnalysisPropensity Score AnalysisROC CurveConfusion Matrix MetricsDBSCAN / HDBSCANPrincipal Component Analysis (PCA)Cronbach's Alpha and Reliability AnalysisStructural Equation Modeling (SEM)Weighted Mean (Survey Means)Nonlinear RegressionQuantile RegressionProbit RegressionMediation AnalysisPath AnalysisParametric Survival (AFT)Interval-Censored SurvivalAnomaly DetectionVAR (Vector Autoregression)PLS-SEM (Partial Least Squares Path Modeling)Hexbin Map

Related concepts

Interquartile range (IQR) and Tukey fencesBox plot