Displaying 1 result from an estimated 1 matches for "iter76".
Did you mean:
item76
2012 Oct 22
1
glm.nb - theta, dispersion, and errors
...issing value where TRUE/FALSE needed
What happens is that, looking at theta iterations, values become too large,
and thereby infinite, and thereby NaN.
This is an example for model 2 above, with a theta of 4125 and a
significantly smaller dispersion value than 1 of .903 (p<.000):
>theta.ml: iter76 theta =1.34118e+16
>theta.ml: iter77 theta =3.0232e+16
>theta.ml: iter78 theta =-Inf
>theta.ml: iter79 theta =NaN
One symptom causes this problem:
1: In sqrt(1/i) : NaNs produced
How would you recommend to deal with these problems? Are results reliable
although these errors occur?
Again...