LISTSERV at the University of Georgia
Menubar Imagemap
Home Browse Manage Request Manuals Register
Previous messageNext messagePrevious in topicNext in topicPrevious by same authorNext by same authorPrevious page (October 1999)Back to main SPSSX-L pageJoin or leave SPSSX-L (or change settings)ReplyPost a new messageSearchProportional fontNon-proportional font
Date:         Wed, 13 Oct 1999 16:23:15 +0100
Reply-To:     "Daniel P. Mears" <dpmears@MAIL.LA.UTEXAS.EDU>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         "Daniel P. Mears" <dpmears@MAIL.LA.UTEXAS.EDU>
Subject:      Deviation contrast in GLM
Content-Type: text/plain; charset="us-ascii"

List-servers:

I'm curious about what exactly the deviation contrast in GLM-Univariate (and in the logit module) does. The SPSS online help indicates that the deviation contrast provides information on whether EACH category of an independent variable differs statistically from the grand/overall mean of ALL of the categories of the independent variable. Thus, with a 5-category independent variable, it provides information on whether the mean for category 1 differs statistically from the grand/overall mean for categories 1-5, on whether the mean for category 2 differs statistically from the grand/overall mean for categories 1-5, and so on.

There are two issues about which I am particularly confused. First, to obtain the full set of contrasts (i.e., for each category of the independent variable), you have to run the GLM-Univariate option twice, once with the first category of the independent variable "omitted" and a second time with the last category "omitted." The very nature of the deviation contrast option would seem to suggest that "omitting" ANY category makes no sense.

Second, how does this deviation contrast option "work"? For example, it seems like those categories that contribute disproportionally more cases to the grand/overall mean would consistently be more likely to be closer to the grand/overall mean, and thus it seems likely that any derived statistic would be suspect. (I speak here as a humble user of statistics, not as one who is particularly knowledgeable in this area.)

Any comments or clarification would be appreciated.

Dan Mears UT-Austin


Back to: Top of message | Previous page | Main SPSSX-L page