Date: Tue, 14 Aug 2007 13:15:52 -0700
Reply-To: David L Cassell <davidlcassell@MSN.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: David L Cassell <davidlcassell@MSN.COM>
Subject: Re: GLM procedure
Content-Type: text/plain; format=flowed
Paige sagely replied:
>On Jul 23, 5:38 am, Mike P <michael.pearm...@tangozebra.com> wrote:
> > Hi All,
> > More of a stats general question here, I'm doing lots of surveys where
> > people have been exposed to some advertising or not, (control and
> > exposed groups) and sometimes i have the same type of question that
> > appears on each questionnaire, for example brand awareness.
> > I've now collected 100 surveys and wish to make some kind of market
> > norm for this brand awareness question, i thought i could take the
> > variable from each dataset and work out a % figure for this say
> > Exposed % - Control % to give me a difference figure.
> > Is it statistically sound to then run an ANOVA on this data to find if
> > there are differences in different vertical markets?
> > My initial thoughts are no, as the samples each survey are different,
> > Also does anyone have a better idea of how i could have some kind of
> > statistically sound market norm?
>If the response is continuous and the "usual" assumptions are met
>(errors are identically and independently normally distributed, etc.),
>then yes you could certainly do ANOVA. But with only two groups, that
>would be equivalent to a t-test.
>But then you start talking about percents, and if your responses are
>percents, then you can't do ANOVA. You would probably just do a
>comparison of the two proportions.
>paige\dot\miller \at\ kodak\dot\com
I agree, for the most part. But if the independent values of the 'percent'
yield a model that has roughly normal residuals, then we *ought* to be able
perform a typical ANOVA type of analysis. If that is what our poster is
I also agree with the poster that there are methodological problems with
slapping numbers from different surveys together, since they may not all
reflect the same instrumentation. I mean, if one survey asks a slightly
question, or phrases it somewhat differently, then it may reflect different
responses, due to the way the question is put forth. There's a lot of
on this type of problem in the sample survey literature, and there's an
field of work on questionnaire design.
So it may not be meaningful to compare across questionnaires. I would
that the poster try to aggregate by questionnaire structure, so that the
would be comparable. At that point, the poster might look at comparing the
David L. Cassell
3115 NW Norwood Pl.
Corvallis OR 97330
Puzzles, trivia teasers, word scrambles and more. Play for your chance to