On Wed, 25 Jun 2003 00:15:26 -0500, "praxis" <[EMAIL PROTECTED]> wrote:
> Hi all. > > Assuming I do a multiple regression using ML estimation instead of OLS, do I > still need to meet all the assumptions like normal distribution assumption, > linearity assumption, and/or homoscadesticity assumption? If yes, could > anyone explain why? For the same test, you need the same assumptions. I know that you can set up Discriminant function with other assumptions, and an improved error model, using Logistic regression, and its ML solution. However, in particular, I don't remember this -- Is there a particular ML estimation of "multiple regression"? Is there an ML solution to the Normal equations that isn't the OLS solution? Here is one way to think about tests in general. The OLS test, or any test, is made *efficient* when it makes a useful, true assumption. Whenever we abandon a *useful* assumption -- such as, any of those three named above -- we have to abandon some efficiency at the same time. It is really a useful exercise, now and then, to figure out how to get different results in two analyses. What feature does X test, and what will disrupt X, and what is X robust against? [Can you show what will make the correlated t-test far less powerful than the t-test for groups? ] -- 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/ . =================================================================
