I'm novice with respect to AIC model selection so I'm wondering what are the
hypotheses that model selection addresses and ANCOVA does not?  To provide
an example, imagine we've got y (response variable), x (continuous
predictor), and group (discrete predictor).  I could see fitting three
models:

(1) y as a function of group
(2) y as a function of group + x
(3) y as a function of group + x + group*x

And depending on which model is best supported we conclude different things:
model 1 supported = no support for considering x, model 2 = no support for
considering multiplicative term, model 3 = there is reason to believe all
terms are worthy of inclusion.

Wouldn't an ANCOVA do a similar thing?  We get an ANOVA table that tells us
the sum of squares for each term, and the resultant hypothesis test.  So
depending on the p-values (I know nobody likes to depend on p-values), we
could reach any conclusion roughly equivalent to those I (perhaps wrongly)
supposed for the model selection case.  Equivalent results would be
something like: group significant, OR group & x significant, OR group & x &
group*x significant.

Am I missing something with model selection?  Thanks for your thoughts and
insights,
Andy


On Thu, Apr 1, 2010 at 2:26 PM, David Hewitt <[email protected]> wrote:

> I certainly agree with Gareth that the ANCOVA approach should be
> simple. But it won't address all hypotheses posed by the "slopes
> and/or intercepts" framework. What you need is a model selection
> approach. Fit the various linear models of interest (some of which may
> be ANCOVA-type models), calculate AICc, and see which model(s) are
> best supported by the data. Oh, and check all the assumptions. This is
> a piece of cake in R; see e.g. the packages 'lm' and 'AICcmodavg'.
>
> Dave Hewitt
> Research Fishery Biologist
> USGS Western Fisheries Research Center
> Klamath Falls Field Station, Oregon
> http://profile.usgs.gov/dhewitt
>
>
> From:         Gareth Russell
> Subject:      Re: Comparing slopes and intercepts in linear regressions
>
> I'm afraid ANCOVA is the way to go. It shouldn't be cumbersome though,
> not in any up-to-date software. If you have three columns of data (two
> continuous, one categorical), and specify one of the continuous as the
> dependent and the other two as predictors, then almost all software
> packages will do an ANCOVA.
>
> Gareth Russell
>
> -----
> On Wed, 31 Mar 2010 11:02:07 -0400, Howie Neufeld wrote:
> Dear All -
>
> I have a stats question concerning comparing linear regressions. If
> you have two or more regressions, and want to know if their slopes
> and/or intercepts are significantly different, what procedure would
> you use? I am familiar with SAS mainly. Zar has a two-sample t-test
> equivalent for comparing two slopes, but the procedure for intercepts
> is extremely cumbersome, as is the multiple slope comparison, which
> involves ANOCOVA.
> Thanks!
> Howie Neufeld
>

Reply via email to