jump to last post 1-2 of 2 discussions (4 posts)

Stat Majors - Help. A Question About Residual Analysis.

  1. My Esoteric profile image91
    My Esotericposted 4 years ago

    Stat Majors - Help.  A Question About Residual Analysis.

    Here is my problem which hours on the Internet has yet to resolve.  I have a log-log linear regression with just an X and Y,   Even with Log-log, the normal prob plot is turned down or up at each end, nice and straight in-between.   HOWEVER, the residual plot has a distinctive negative slope where, as X grows, the residual goes from almost total > 0 until it finally goes negative for the larger Xs; the variance seems more or less constant. 

    What is that residual pattern telling me?  Does anyone have an idea?

  2. Randy M. profile image87
    Randy M.posted 4 years ago

    Way back, I did a lot of residuals analysis on data I generated in my experiments.  If I saw any pattern in the residuals, such as you have, I tried another type of transformation to normalize the data.  What you are shooting for is a random distribution of the residuals.  Whatever transformation that gives you that result is what you are looking for.  I would probably try a natural log transformation e to the x. 

    Hope this helps.  I had the help of a professor of statistics when I was doing this, using SAS for my data analysis in the 80's.

    1. My Esoteric profile image91
      My Esotericposted 4 years agoin reply to this

      Thanks Randy.  While researching for an answer I stumbled across a "Box-Cox Power Transform" method.  It seemed to do the trick in straightening out my residual normal plot once I, by trial and error, got to the right "lambda" value for the transform

    2. Randy M. profile image87
      Randy M.posted 4 years agoin reply to this

      Great! There is usually a solution out there.

 
working