Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Adjustment for comparison of means and %

22 views
Skip to first unread message

Jo Blois

unread,
Nov 24, 2009, 6:16:36 AM11/24/09
to
This is a beginner's obstacle I face. I do not know how I can adjust
for a covariable when comparing groups for means of continuous
variables or % of categorical variable. Can we do this with the
indefendent T-test and Chi-square or do we have to use ANOVA or
regression?

Bruce Weaver

unread,
Nov 24, 2009, 7:01:05 AM11/24/09
to

Have you tried the UNIANOVA procedure (Analyze - General Linear Model
- Univarite in the menus)? In the dialog, your grouping variable is a
Fixed Factor, and variables you want to adjust for are Covariates.

--
Bruce Weaver
bwe...@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/Home
"When all else fails, RTFM."

Ryan

unread,
Nov 24, 2009, 4:49:27 PM11/24/09
to
On Nov 24, 7:01 am, Bruce Weaver <bwea...@lakeheadu.ca> wrote:
> On Nov 24, 6:16 am, Jo Blois <jonathandebloi...@gmail.com> wrote:
>
> > This is a beginner's obstacle I face. I do not know how I can adjust
> > for a covariable when comparing groups for means of continuous
> > variables or % of categorical variable. Can we do this with the
> > indefendent T-test and Chi-square or do we have to use ANOVA or
> > regression?
>
> Have you tried the UNIANOVA procedure (Analyze - General Linear Model
> - Univarite in the menus)?  In the dialog, your grouping variable is a
> Fixed Factor, and variables you want to adjust for are Covariates.
>
> --
> Bruce Weaver
> bwea...@lakeheadu.cahttp://sites.google.com/a/lakeheadu.ca/bweaver/Home

> "When all else fails, RTFM."

It's important to point out that the Covariates box to which Bruce
refers is not intended for variables with three or more unordered
categories (i.e. Race). Also, remember that not allowing there to be
an interaction between the covariate and the grouping variable assumes
homogeneity of slopes. Google "ANCOVA assumptions" for further
details. -Ryan

Bruce Weaver

unread,
Nov 24, 2009, 5:21:14 PM11/24/09
to
On Nov 24, 4:49 pm, Ryan <ryan.andrew.bl...@gmail.com> wrote:

>
> It's important to point out that the Covariates box to which Bruce
> refers is not intended for variables with three or more unordered
> categories (i.e. Race). Also, remember that not allowing there to be
> an interaction between the covariate and the grouping variable assumes
> homogeneity of slopes. Google "ANCOVA assumptions" for further
> details. -Ryan

Good points, Ryan. Regarding the last point, I know that it has
become fairly standard to talk about model "assumptions". But I think
it gets the point across to students more clearly if I say that the
ANCOVA model (with no factor x covariate interaction term) *forces*
(or constrains) the lines for the various groups to be parallel. If a
plot of the data suggests the lines are probably not parallel, I need
to run a model that includes the factor x covariate interaction.
(This is no longer the ANCOVA model, strictly speaking; but it is an
acceptable model.) Inclusion of the factor x covariate interaction
*allows* the fit lines to depart from parallel.

--
Bruce Weaver
bwe...@lakeheadu.ca

0 new messages