Herman Rubin <[EMAIL PROTECTED]> wrote in message
asiren$[EMAIL PROTECTED]">news:asiren$[EMAIL PROTECTED]...
>
> In many articles, the chi-squared test of goodness of fit
> is used.  Even when used correctly, the distribution is
> NOT a chi-squared distribution; that is an asymptotic
> approximation.  It is quite poor in the tails, which is
> usually where it is used.
>
> Also, it is often used incorrectly.  When parameters are
> estimated using more information than the cell frequencies,
> even the asymptotic distribution is not chi-squared.

Correct, though it is bounded between two chi-squared
distributions (the ones where you remove 0 and p d.f. for
the p estimated parameters), so chi-square tables
may still be somewhat useful in those circumstances.

Glen

.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to