Compute a linear equation between two or more variables

Are the two variables correlated (no equation)?
Parametric Approach: Pearson's r. Requires a linear relationship and that residuals follow a normal distribution.
Nonparametric Approach. Spearman's rho or Kendall's tau. Relationship may be linear or nonlinear. No normality requirement.

See Helsel & others (2017) Chapter 8.


Is my one explanatory (X) variable a significant predictor of Y?
Parametric Approach: Linear Regression. Requires a linear relationship and that residuals follow a normal distribution.
Bootstrap Approach: Computes significance test with a bootstrap. Requires a linear relationship, but not normality of residuals.
Nonparametric Approach. Theil-Sen nonparametric line. Requires a linear relationship, but not normality of residuals.

See Helsel & others (2017) Chapters 9 and 10.


Which of several possible explanatory variables are significant predictors of Y?
Parametric Approach: Multiple Regression. Requires a linear relationship on partial plots and that residuals follow a normal distribution.
Nonparametric Approach. A two-step process. Compute the residuals of Y from a LOESS surface smoothed versus one or more X variables. In step two, model and test the relationship between the residuals and the remaining primary X variable of interest using the Theil-Sen line. Assumes a linear relationship between residuals and the primary X variable, but does not require normal residuals.

See Helsel & others (2017) Chapters 11 and 12.

Online at:


Chapter references are to the 2017 Second Edition of Statistical Methods in Water Resources.