topic Any suggestion about this data-transformation to do regression? in Statistical Procedures

Any suggestion about this data-transformation to do regression?

MikeTurner — Sat, 21 Jan 2012 09:10:49 GMT

Is there is one possibility to do regression on the data through data-transformation? If yes, what kind of transformation I can do?

Thanks.

Any suggestion about this data-transformation to do regression?

SteveDenham — Mon, 23 Jan 2012 12:34:11 GMT

Some more info about the X and Y variables would be a lot of help. Looking at the plot that accompanied your post, I didn't see much but noise. The X variable looks to get very large, so a plot of Y vs. log(X) might be revealing. I don't think there is going to be anything, but you might look at PROC TRANSREG for Box-Cox transformations as a step towards something.

Steve Denham

Any suggestion about this data-transformation to do regression?

Rick_SAS — Mon, 23 Jan 2012 13:29:30 GMT

If the X axis is reversed, this looks like a classic case of heteroskedastic data in which the variance increases with X.

I'd define Z = 1/X and plot Y vs. Z (or use 1/(X+1) if X=0 is a possible value).

You can then use all the standard variance stabilizing transformations (AKA, normalizing transformations) such as Z-->log(Z) or Z-->sqrt(Z).

Depending on the meaning of X, you could also try just a simple reflection such as W = 300000 - x and then work with W.

As Steve mentions, transformations often work best when they are meaningfully related to the data. So ask yourself, is "natural" way to flip around the X axis?