Jennifer Bo wrote: > Hi, > > I have been reading a lot of articles on empirical studies using > multiple regreession. Many scientists use a scatterplot of the > standardized residuals and standardized PREDICTED values to test for > homoskedasticity. If there > was no pattern they concluded that the variance of the residuals was > costant across all level of the INDEPENDENT variables > (homoskedasticity). > > This confuses me. Wouldn't I have to use scatterplots of EACH of the > INDEPENDENT variables and the residuals to make sure there is no > heteroskedasticity? > > Best regards, > > Jennifer Borck
Yes, but what is done is a shortcut. In theory, what you describe should be extended to include all possible independent variables, not just those accepted into a first-pass unweighted regression model. If there are lots of independent variables and the sample size isn't extremely large then examining lots of such plots is likely to find odd effects in some of them just by chance. In addition I guess practical experience has been that heteorscedactic effects are usually related to the size of the predicted value. David Jones . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
