Skip to content

random factor variance

8 messages · João R., Andrew Dolman, Ken Beath +1 more

#
On 20/05/2009, at 11:43 AM, Jo?o R. wrote:

            
This means that the variance of the random effect needed to explain  
your data is zero.  The clusters vary by the same amount or less than  
if there was a random effect, that is they can all be explained by  
subject variation.

Ken
#
On Wed, May 20, 2009 at 8:37 AM, Andrew Dolman <andydolman at gmail.com> wrote:
I think this often happens when there is a difference between groups
on the surface, but the model is telling you these differences are
actually nothing more than what would be predicted to occur by chance,
given the other parameters of the model (e.g. individual/subject
variation).
#
On 20/05/2009, at 9:35 PM, Jo?o R. wrote:

            
Because the value can't go below zero, it may end up as zero. Even if  
the true value is small but non-zero then sampling variation may cause  
the estimate to be zero.
Think about a population and then dividing it randomly into a number  
of groups. I will assume normal distributions but the same ideas apply  
to binomial etc. We would expect that the groups will have different  
means, and we know how they will vary based on the population  
variance. When we fit a random effect the question we are asking is do  
these vary more than predicted from the population, in which case our  
random effect variance will be greater than zero.
I suspect the problem is that your model is overfitted, because of the  
number of possible covariates, and the stepwise selection it has  
constructed a model that fits well without need for a random effect.

Ken
#
This makes sense but doesn't it lead to the conclusion that a non-zero
random effect is always "statistically significant"? In fact I think
that is not necessarily so, and you need to run a test to determine
whether a given variance is significant (i.e. better than chance)...

D