[R] generalized linear regression - function glm - dismissed predictors - more information about simulated data

[Ben Bolker]

Bert Gunter <gunter.berton <at> gene.com> writes:

[snip snip snip]

> Cheers,
> Bert
> 
> P.S. Your results look fishy, even for a simulation. The extremely low
> standard errors and rank deficiencies strongly suggest to me that your
> methodology is faulty and that you are getting a bunch of baloney. I
> would recommend seeking help from your local statistician.
> 
> On Thu, Nov 18, 2010 at 11:23 AM, David Winsemius
> <dwinsemius <at> comcast.net> wrote:

 [snip snip snip]

  I will just note that there is one point in your simulation
that looks like an error.
  You are adding a *single* random normal deviate (the same for
every point) to your simulated expectation.  You probably want
at the very least to put that rnorm() statement inside the loop
so that you get a different epsilon for every individual.
  I don't know if the rest of your approach is correct or not,
but that part looks wrong.

 
> >> // TILL THAT POINT, I RELIED ON THE PROCESS DESCRIBED IN A THESIS.
> >> // THERE WAS NO INDICATION RELATIVE TO THE "ADJUSTMENT" OF EPSILON.
> >> epsilon <- rnorm(1, mean = 0, sd = 0.002)
> >> for (ind in 1:nbInd){
> >>  y[target,ind]<<- c0 + coefficients %*% X[ind,SNPs] + epsilon
> >> }
> >>
> >> Thank you in advance for your kind help.
> >>
> >> C.S.