Calculating the Y' variable

[Shachar]

Hi everyone.

I wanted to see if my data could be described better with a non-linear relationship. So I calculated a new Y' variable, based on the coefficients I got from the regression model. Just to make sure I didn't make any mistakes, I rerun the regression, with only the new Y' as the predictor. The results were the same. Then, I used the curve fit option in SPSS to look for a better relationship

However, when I try to do the same thing with an interaction variable, this doesn't seem to work. The regression loses a lot of R^2.
Why is this and what can I do to make it work?

Thanks

 

[obh]

Do you mean you use the same Y' but there is interactions in Y calculation? (first: Y=b+a2x1+a2x2+... via Y=b+a1x1+a12x1x2+a2x2))

 

[Shachar]

No. What I mean is that when I enter the interaction variables I recalculate Y' (Lets call it Y''), according to the new coefficients, and also include the interaction variable. However, while Y' explains all the variance in it's model, Y'' does not.

 

[obh]

Hi Shachar

I think this is what I wrote above...
First: Y=b+a2x1+a2x2. =>new regression based only over estimate Y: ŷ => Y'=b'+a' ŷ high R
second: Y=b+a2x1+a2x2+a12x1x2 =>new regression based only over estimate Y: ŷ=> Y''=b''+a'' ŷ

If not please write an example as I did.

 

[Shachar]

Hi obh.

If I understood correctly then, yes, this is what I'm doing.

 

[obh]

Hi Shachar,

I did a simple test and didn't get the same results as you described.
Can you please attach an excel file with the data you use?
Toda

 

[Shachar]

Hi obh. Sorry for the late reply.

Unfortunately, I can't upload the data. It's not something I can share.
Thank you for your effort!

 

[obh]

No worry Shachar,

I feel that you may do a mistake, or I may misunderstand you.
So when you have some extra time you can try to create a simple example without revealing your real data

 

[Shachar]

Thanks obh!