0

I transformed dependent variable by raising to a power of $0.4$. When the original data is plotted on the back-transformed $y^{1/0.4}$ predicted results, the intercept is substantially larger than zero & gradient lower than $1$. Even by adding $\sigma^{1/0.4}$ to the back-transformed data does not substantially improve it. Anyone have ideas on what more may be done? Regards, Michael.

Nesta
  • 325
Michael P
  • 1

1 Answers1

0

It sounds like you fitted a model of the form

$$y^{0.4} = a x + b$$

Solving for $y$,

$$y = (a x + b)^{1/0.4}$$

Daniel McLaury
  • 24,656
  • Thankyou Daniel - I actually transformed the y, not the x. So it's of the form y^0.4 = ax_1 + bx_2 + c . On back-transforming the predictions as y^(1/0.4) and plotting against observations, I am not getting quite a 0 intercept and gradient of 1. Any ideas? – Michael P Aug 22 '16 at 01:19
  • Updated answer. – Daniel McLaury Aug 25 '16 at 06:19