s-news
[Top] [All Lists]

Re: Linear model question: independence?

To: <naomi@stat.psu.edu>, <gwgilc@wm.edu>, <s-news@wubios.wustl.edu>
Subject: Re: Linear model question: independence?
From: "Hugh Jones" <hajones@dlwc.nsw.gov.au>
Date: Fri, 28 Mar 2003 12:13:15 +1100
I would agree with Naomi but would add that the continent effect should also be included as a blocking factor, an additive effect. I would probably treat continent as fixed in this case. Flies are nested within population so why not analyse the means of the 20 flies? After all, interest is primarily in the latitudinal effect.
 
Hugh Jones
Senior Biometrician
Centre for Natural Resources
NSW Depart. Land & Water Conservation
PO Box 3720, Parramatta NSW 2124
 
Tel:    61-2- 9895 7628
Fax:   61-2- 9895 7867

>>> Naomi Altman <naomi@stat.psu.edu> 03/28/03 09:26am >>>
As far as I can see, you have a mixed effects model with random effects
population (and fly) and fixed effect latitude.  You could test the
interaction (which is a mixed effect, usually modeled as random) by a LRT
or, since the design is balanced, by an F-test using the MSE from the model.

--Naomi Altman

At 05:09 PM 3/27/2003 -0500, George W. Gilchrist wrote:
>I have a statistical question that hours of discussion with various folks
>(and editors...) have not settled in my mind, so I ask your scholarly
>opinions.
>
>I have measured wing lengths of 20 flies from 10 different populations along
>a latitudinal gradient on two different continents. I want to do a
>regression of wing length on latitude. The question is, should I regress the
>mean wing length of each population on latitude (giving 16 degrees of
>freedom) or should I regress each individual's wing length on latitude (396
>df)? The residuals are normally distributed and homoscedastic across
>latitudes.
>
>The null hypothesis is that there is no clear gradient in size across
>latitudes. Both of these models reject the null. The question comes when we
>compare the slopes between continents. While it looks visually like there is
>a large difference in the slopes, the means model cannot reject the null
>hypothesis that they are parallel. Of course, the model with the 396 df does
>reject the null.
>
>Thank you for any advice you can offer!
>
>Cheers, George
>
>==================================================================
>George W. Gilchrist                        Email #1: gwgilc@wm.edu
>Department of Biology, Box 8795          Email #2: kitesci@cox.net
>College of William & Mary                    Phone: (757) 221-7751
>Williamsburg, VA 23187-8795                    Fax: (757) 221-6483
>http://gwgilc.people.wm.edu/
>
>
>--------------------------------------------------------------------
>This message was distributed by s-news@lists.biostat.wustl.edu.  To
>unsubscribe send e-mail to s-news-request@lists.biostat.wustl.edu with
>the BODY of the message:  unsubscribe s-news

Naomi S. Altman                                814-865-3791 (voice)
Associate Professor
Dept. of Statistics                              814-863-7114 (fax)
Penn State University                         814-865-1348 (Statistics)
University Park, PA 16802-2111


--------------------------------------------------------------------
This message was distributed by s-news@lists.biostat.wustl.edu.  To
unsubscribe send e-mail to s-news-request@lists.biostat.wustl.edu with
the BODY of the message:  unsubscribe s-news
<Prev in Thread] Current Thread [Next in Thread>