100 is normally OK.
It sounds like you're getting quasi-complete separation problems.
That is, it's not the number of people in the smaller group that's
causing the problem, it's the number of young males of high social
class with low income that's causing the problem.
Jeremy
On 4 July 2011 11:40, Dan <[log in to unmask]> wrote:
> Does anyone have any rules of thumb on this?
>
> It seems to me when I use binary logistic regression and the smaller category in the outcome goes much below 100, I start to get extremely high Wald test statistics and very large coefficients, even though they should as coefficient.
>
> My IVs are standard demographics/SES indicators - gender, age, education (3 dummies), income, social class (4 dummies).
>
> Any help much appreciated.
>
> Dan
>
|