Multiple regression

Again I want to say that I have been using and appreciate this program. Thanks.

Some small additions would make the multiple regression routines more useful.

1. Tolerance and VIF and perhaps the Durbin-Watson statistic for colinearity of predictors

2. Some support for hierarchical regression, howing and testing the significance of additional variables as they are added to a model. These could be F-tests or the Sobel test.

Thanks again for a great program



  • Hmm. That should be "showing" (not "howing") the effects of additional varaibles as they are added.

    Two web sites dealing with this technique for researching interactions ("moderators") and intervening variables ("mediators") are the following:

    Perhaps they may be of interest to others too.

  • Hi Lance

    How does the hierarchical regression you mention differ from a standard forward stepwise regression (as is in statistiXL)?

    Thanks for the input

  • Hi Alan -

    The specifics of the calculations don't differ much at all. But since the investigator trying to use this method must enter variables in a specific sequence and estimate the added contributions they make in order to assess interactions or mediators in field studies, an output table showing that if you enter x after y and z the added contribution is n, and n is significant using either an F-test (the original Baron & Kenny procedure) or the Sobel test (a later addition) would make life a litlle easier. ( I suppose it is a kind of poor man's path analysis though it is quite popular, at least in Psychology. Still ecologists and other researchers probably also sometimes have non-experimental and continuous data to which they wish to fit a model involving interactions between variables or mediational relations between variables. ) I suppose a window allowing the investigator to choose the order of variables entered along with an output table showing the added contributions and their significance would cover it.

    Because interactions and mediational relations usually (always?) involve correlated variables some indicators of colinearity (such as tolerance and VIF) are important in this kind of research. Adding these to StatistiXL would be a little more work.

    I suggested it mostly to make life a little easier for myself but also because I didn't think it would require an enormous programming effort to implement.

    Thanks for the reply. I do appreciate this forum and the program - I am only making suggestions because I like them!!

  • Thanks for the input Lance! We will definately add this to our wish list.


Sign In or Register to comment.