2

I got these results from a GLM from R, but I am not sure what it means that the intercept is significant but none of my variables actually are. So is nothing really significant? Here are my results:

Groups Name        Variance   Std.Dev.  Corr   
 Plot   (Intercept) 1.4842e+07 3852.4696        
        Year        3.6783e+00    1.9179 -1.000 
Number of obs: 121, groups: Plot, 50

Fixed effects:
                   Estimate Std. Error z value Pr(>|z|)   
(Intercept)        -3.90239    1.36721  -2.854  0.00431 **
Wald_ManagBUBAdi  -16.50851 9563.66009  -0.002  0.99862   
Wald_ManagBUBAj     0.45015    2.31227   0.195  0.84564   
Wald_ManagBUBAun    1.63915    1.46892   1.116  0.26447   
Wald_ManagBUdi     -0.03527    2.25725  -0.016  0.98753   
Wald_ManagBUplen    1.14402    1.52959   0.748  0.45450   
Wald_ManagFIBA      2.92878    1.53228   1.911  0.05595 .
Nick Cox
  • 48,377
  • 8
  • 110
  • 156
RaB
  • 23
  • 1
  • 3
  • I just posted a question which might have similar problem as this post: http://stats.stackexchange.com/questions/72824/a-categorical-variable-in-glm-shows-signifcance-from-analysis-of-deviance-but-e; maybe you could check your data to see whether all response are 0 or 1 when the variable is Wald_ManagBUBAdi – tiantianchen Oct 15 '13 at 11:25
  • Could well be separation as @tiantianchen suggests: see [here](http://stats.stackexchange.com/questions/11109/how-to-deal-with-perfect-separation-in-logistic-regression/68917) for a discussion of ways to deal with it. – Scortchi - Reinstate Monica Oct 23 '13 at 09:39
  • I kind of have the same thing: a significant intercept effect, but no significant effect of my factor. However, in my case it is interesting to know whether the data differs from 0. Am I allowed to do a post-hoc analysis testing the levels of my factor separately to check for an intercept effect? –  Oct 23 '13 at 09:27
  • Welcome to the site. This is not an answer to the question, so I am converting it to a comment. – Peter Flom Oct 23 '13 at 10:11

2 Answers2

4

The intercept is the predicted value of the dependent variable when all the independent variables are 0. It is rarely of interest, unless the IVs are centered or standardized. Since you haven't told us what any of your variables are, there's no way to say more than that.

Peter Flom
  • 94,055
  • 35
  • 143
  • 276
1

The standard error of this variable 

Wald_ManagBUBAdi  -16.50851 9563.66009  -0.002  0.99862

is kind of alarming. I would suggest trying to see if something went wrong with the data for that variable, or try excluding it from the model and see what happens to the significance levels of the variables left in the model.