Johnson s Transformation

[Quanita Shamoon]

My dependent variable doesn't meet the normality assumption. I want to transform it using "jnsn" command. Can someone please guide me on the syntax to be used, I have been trying a few, but I am not sure of it.

[Joao Santos Silva]

Dear Quanita Shamoon,

I cannot answer your question, but I can tell you that there is no requirement that the dependent variable follows a normal distribution, and that it is generally a bad idea to transform the dependent variable. If you tell us more about what you are doing, maybe we can help with that.

Best wishes,

Joao

[Nick Cox]

For once, I only almost agree with Joao Santos Silva For example, if the underlying model is y = ax^b or y = y_0 exp(-bx) generations of researchers have worked with log y and in the first case log x too.

The details that can bite include

1. you need to be clear whether the implications for error distributions are plausible (that depends intimately on the pattern of scatter)

2. back-transformation to get predicted values may need more care than just the obvious exponentiation.

I guess that is much of what Joao has in mind.