Date:         Wed, 7 Jul 1999 13:59:57 -0500
Reply-To:     "Nichols, David" <nichols@SPSS.COM>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         "Nichols, David" <nichols@SPSS.COM>
Subject:      Re: Interactions

Randomly assigning those cases at a given value to one of two groups is
easy. After recoding the other cases into 0 or 1 on that variable, just use
command syntax of the form:

IF FIRSTVAR=medianvalue NEWVAR=RND(RV.UNIFORM(0,1)).
EXECUTE.

You should be aware of the fact that doing this may substantially alter the
relationship you're looking at, especially if you do a bivariate median
split. There was an article in the quantitative methods section of
Psychological Bulletin about five years ago that demonstrated some of the
dangers of bivariate median splits.

David Nichols
Principal Support Statistician and
Manager of Statistical Support
SPSS Inc.

> -----Original Message-----
> From: Rita Patel [mailto:RXPATEL@OPM.GOV]
> Sent: Wednesday, June 30, 1999 1:48 PM
> To: SPSSX-L@LISTSERV.UGA.EDU
> Subject: Interactions
>
>
> Hi,
>
> I would like to do a moderator median split to display my interaction
> effects. Both my independent variables are continuous. One of these
> independent variables is a single item measure (5 point
> scale). I would
> like to do a median split on this variable (lo=values below median;
> hi=values above median; values at median randomly assigned to either
> the lo or hi group).
>
> I can use the compute statement to assign hi or lo groupings. How do I
> randomly assign the values at the median for this variable?
>
> Any help would be greatly appreciated.
>
> Also, two quick stats. questions: (1) can anyone recommend a book with
> a good discussion on Bonferroni corrections? (2) In doing a test on
> mediation, is there a formula for comparing coefficients if see if the
> coefficients are significantly different--can you suggest a reference?
>
> Thanks again,
>
> Rita
>
> *******************************
>   E-Mail scanned for viruses.
> *******************************
>