On 3 Jul 2003 05:07:48 -0700, [EMAIL PROTECTED] (Gartland, Myles) wrote: > How robust are multinomial logistic regression estimates when the > continuous variables are non-normal? > > I could not find if the estimates are sensitive to this. >
All your typical regression procedures are intended to yield a prediction equation. How robust is a prediction equation if it has to subsume values that are extreme outliers? - Like the ordinary logistic, the 'prediction' allows for prediction into infinity (essentially) so a single extreme value or two *might* not hurt. But a fortuitous fit for outliers is not a rescue that you ought to count on. And, as Herman posted, it is the residuals that matter, where they matter. (P.S., logistic residuals are slightly different from 'normal' but that is a trivial distinction.) -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html "Taxes are the price we pay for civilization." Justice Holmes. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
