|
- Question:
- What is best transformation of Y so that T(Y) vs. X is
most linear?
- Definition:
- Vertical Axis : Correlation coefficient of T(Y) and X
-   where T(Y) is the Box-Cox transformation
of Y
- Horizontal Axis: Box-Cox power transformation parameter
(lambda)
- Interpretation/Conclusion:
- Best transformation is lambda = 0.6. Note that the
square root transformation (lambda = 0.5) is a reasonable
choice in this case.
- NIST/SEMATECH Handbook Chapter:
- 1.
Exploratory Data Analysis
- 4.
Process Modelling
- References:
- NIST/SEMATECH e-Handbook of Statistical Methods,
Box-Cox Linearity Plot, 6/2003.
- Box and Cox (1964), "An Analysis of of transformations,"
Journal of the Royal Statistical Society,
Series B 26 (2): 211–252.
|