0

I am computing multiple regression coefficients for these variables (for several cases):

         y,x1,x2,x3
     log(y)= a(log(x1))+b(log(x2))+c(x3)+d
     res=lm.fit(log(y)~log(x1)+log(x2)+(x3))

All values of y,x1,x2,x3 are between 0 and 1.

The values of the coefficients range between -10 to 10 for a

The values of the coefficients range between -5 to 5 for b

The values of the coefficients range between -5 to 5 for c

The values of the coefficients range between -3 to 3 for d

I wonder if we should expect a certain range for each coefficient.or in other words,how to know that my regression analyses did a good job.

amith
  • 103
  • 2
  • This question puzzles me. Why do you believe your coefficients should be in sprecific ranges (and these ranges symmetric around zero)? And why do you believe that to be relevant for model performance? – Roland Feb 12 '14 at 14:45
  • I put approximate ranges but not necessarily symmetric. what confuses me is how to know if my model is correct and so the coefficients are? all explanatory variables have positive linear relationship with the response`y` but some values of the coefficients are negative – amith Feb 12 '14 at 14:50
  • It seems like you don't have a good understanding of linear regression. You should study some textbook. To assess model quality you could look at `summary(lm(log(y)~log(x1)+log(x2)+(x3)))` and in particular the R-squared value and the result of the F-test. You should also look at diagnostic plots: `plot(lm(log(y)~log(x1)+log(x2)+(x3)))` – Roland Feb 12 '14 at 15:01
  • Ways to check that a regression analysis is working well are discussed, *inter alia*, at http://stats.stackexchange.com/questions/32600. For more material on this, please search our site for [regression diagnostic](http://stats.stackexchange.com/search?q=regression+diagnostic). – whuber Feb 12 '14 at 17:36

0 Answers0