Overview of Variable Transformations

Transforming data is an important technique in exploratory data analysis. Centering and scaling are simple examples of transforming data.

More complex transformations are useful for a variety of purposes. A variable that violates the assumptions of a statistical technique can sometimes be transformed to fit the assumptions better. For example, a variable that is not normally distributed can be transformed in an attempt to improve normality; a variable with nonhomogeneous variance can be transformed in an attempt to improve homogeneity of variance.

You can create new variables in a data set by transforming existing variables. SAS/IML Studio provides a Variable Transformation wizard that enables you to quickly apply standard transformations to your data. These include normalizing transformations (such as logarithmic and power transformations), logit and probit transformations, affine transformations (including centering and standardizing), and rank transformations.

You can create your own transformations within the Variable Transformation wizard by using SAS DATA step syntax and functions. These enable you to recode variables, to create variables with simulated values from known distributions, and to use arbitrarily complex formulas and logical statements to define new variables.

Most SAS/IML Studio transformations create a new numerical variable from an existing numerical variable. You can define custom DATA step transformations that use and create variables of any type.

You can apply transformations to all observations, or you can apply the transformation only to observations that are included in analyses.