Usage Notes
Other Chapters
Much of what is considered nonparametric statistics is included in other chapters. Topics of possible interest in other chapters are: nonparametric measures of location and scale (Chapter 1, “Basic Statistics”), quantile estimation (Chapter 1, “Basic Statistics”), nonparametric measures in a contingency table (Chapter 5, “Categorical and Discrete Data Analysis”), measures of correlation in a contingency table (Chapter 3, “Correlation”), tests of goodness of fit and randomness (Chapter 7, “Tests of Goodness of Fit and Randomness”), and nonparametric routines for density and hazard estimation (Chapter 15, “Density and Hazard Estimation”).
Other Methods
Many of the tests described in this chapter may be computed using the routines described in other chapters after first substituting ranks (or some other score) for the observed values. (Routine RANKS(see Chapter 1, “Basic Statistics”) may be used to compute ranks.) This method for computing nonparametric test statistics is recommended for cases such as unbalanced one‑way ANOVA designs for which no nonparametric subroutine is provided.
Missing Values
Most routines described in this chapter automatically handle missing values (NaN, “not a number”; see the Reference Material section of this manual). In these routines, observations that are missing are ignored; the variable NMISS is incremented by one for each missing observation. The user should be aware, however, that some routines described in this chapter do not handle missing values. Missing values input to such routines may result in erroneous results.
Tied Observations
Many of the routines described in this chapter contain an argument FUZZ in the input. Observations that are within FUZZ of each other in absolute value are said to be tied. Moreover, in some routines, an observation within FUZZ of some value is said to be equal to that value. In routine SNRNK, for example, such observations are eliminated from the analysis. If FUZZ = 0.0, observations must be identically equal before they are considered to be tied. Other positive values of FUZZ allow for numerical imprecision or roundoff error.