KNOWLEDGEBASE - ARTICLE #2071
The Box-Cox transform with GraphPad Prism
What are the Box-Cox family of transforms?
Various forms of the Box-Cox transformation are often used to transform nongaussian data to a distribution that is closer to Gaussian (Normal). Prism does not have any form of the Box-Cox transformation built-in. But Prism lets you add user-defined transforms, so it is easy to do these transforms with Prism.
Equations (functions)
Different statisticians use different forms of the transform.
The usual form of the transform is:
If Y<>0 Y= (Y^lamda -1)/lambda
If Y=0 Y=ln(y)
In Prism, you would enter this as:
Y = IF(Y=0, ln(Y), (Y^lamda - 1)/lamda)
Another form is used when some values are negative, so you need to shift as well as transform.
IF Y= 0, Y=ln(Y + delta)
IF Y<> 0, Y = (Y + delta)^lamda
In Prism, you would enter this as:
Y = IF(Y=0, ln(Y + delta), (Y+delta)^lamda)
You would need to try different values of lamba (and maybe delta) to see what works with your data.
This paper gives even more alternative forms of the Box-Cox transform.
How to enter into Prism
- From a data table, click Analyze and choose Transform,
- Choose a user-defined transform of Y values.
- Click Add to add a new equation to the list.
- Enter the name of the transform ("Box Cox") and enter the equation. OK
- Now you'll see it in the list of user-defined equations.
- You'll need to enter the value of lamda (and delta with the second form) every time you use the transform.