Transformation
Transformations apply a mathematical function to all the response data.
By default, the transformation option is set to “None” meaning that the response data will be analyzed as is. Transformations may be needed to help meet the residual assumptions and make the ANOVA valid. Residuals are assumed to be normally distributed with a constant variance. Check the diagnostic plots to validate these assumptions. If the plots don’t look right, go back and try some transformations.
Here is a list of the transformations and some data types that may benefit from using that transformation:
Square Root |
Count or Frequency data |
Natural Log |
Variance (std dev) or Growth data |
Base 10 Log |
Variance (std dev) or Growth data |
Inverse Square Root |
|
Inverse |
Rate data, Decay data |
Power |
|
Logit |
Bounded data, Yield data |
Arcsin Sqr Root |
Probability, Fraction defective |
The Base 10 Log is the most commonly used data transformation.
Look at the Box-Cox plot on the Diagnostics button for guidance.
If a transformation is applied the analysis will need to be redone.
The model often becomes simpler and has fewer significant terms once a transformation is applied.