Date:         Thu, 7 Jan 2010 14:14:02 -0500
Reply-To:     Gene Maguin <emaguin@buffalo.edu>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Gene Maguin <emaguin@buffalo.edu>
Subject:      Re: regression: multiple Y for each X
In-Reply-To:  <8E725808-EDE8-4E4A-A791-5896A0907912@mac.com>
Content-Type: text/plain; charset="us-ascii"

Ian,

I haven't seen Sokal and Rohlf's text so I can't say anything about what
they said. However, your description of what they said seems odd to me. I'd
like to say they are describing an ANCOVA model but I'm not sure. Perhaps
someone else will recognize the description of the computation or be
familiar with that reference.

You said that you want to know whether '... a certain continuous variable
could be a predictor of diet/nutrition in the communities.' So you are
interested in community level relationships. Community is a kind of nuisance
factor. A really crude way to do this would be to average the DV within
community (via Aggregate) and correlate the community level mean with the
community level variable.

I don't think this can be done in Mixed but I could well be wrong (and would
like to be corrected with a syntax example if I am). In an SEM program or
HLM, yes it can.

Gene Maguin


>>Thanks for your suggestions.  For further background, as Keith
requested, I have 9 communities, with between 12 and 93 respondents
per community for a total n of 380.  A scatterplot of the nutrition
variable against the continuous predictor does show an apparent
linear trend, though of course the groups of multiple Ys overlap
broadly.

I don't have much experience with hierarchical or nested designs, and
the approach I've thought about using is one laid out in Sokal &
Rohlf's "Biometry" 1981, example in Box 14.4, page 480.  I guess
maybe that dates me, but it seemed appropriate to the data.  S&R
advocate -- as I understand it -- taking ANOVA sum of squares
(community) minus REGRESSION SS to get deviation from linearity SS.
Then the REGRESSION MS is tested over the deviation from linearity MS.

regards,
Ian

Ian D. Martin, Ph.D.

Tsuji Laboratory
University of Waterloo
Dept. of Environment & Resource Studies


On 07 Jan, 2010, at 11:13 AM, statisticsdoc@cox.net wrote:

> Ian,
> It sounds as though your data has a nested structure.  You might
> want to consider using Hierarchical Linear Modeling.  How many
> communitities are in your study, and how many cases in each community?
> HTH
> Steve Brand
> ------Original Message------
> From: Ian Martin
> Sender: SPSSX(r) Discussion
> To: SPSSX-L@LISTSERV.UGA.EDU
> ReplyTo: Ian Martin
> Subject: regression: multiple Y for each X
> Sent: Jan 7, 2010 10:42 AM
>
> We have a study going on in a number of northern communities,
> measuring diet and nutrition variables.  ANOVA shows differences
> between communities, but we have a hypothesis that a certain
> continuous variable could be a predictor of diet/nutrition in the
> communities.  This predictor is a single value for each community, so
> we have multiple Y (diet or nutrition variable of interest) for each
> value of X.
>
> I seem to remember that it is appropriate to run this first as ANOVA
> (with community as the grouping factor) and then as regression, but
> testing the MS regression over the MS for the community factor from
> the ANOVA model instead of over the MS error from regression.  If so,
> I could do this by hand using the output from the 2 SPSS models, but
> I wondered if there is a way to do this test (or an alternate)
> directly in SPSS.
>
> Comments or suggestions appreciated.
>
> regards,
> Ian
>
> Ian D. Martin, Ph.D.
>
> Tsuji Laboratory
> University of Waterloo
> Dept. of Environment & Resource Studies
>

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@LISTSERV.UGA.EDU (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD