Date: Thu, 13 Jul 2006 07:29:33 -0400
Reply-To: Jim Groeneveld <jim2stat@YAHOO.CO.UK>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Jim Groeneveld <jim2stat@YAHOO.CO.UK>
Subject: Re: running ANOVA to test for differences
Hi Duklaa,
If you want to check whether respondents didn't answer unseriously, you
have to define some (arbitrary and disputable) criterion for checking
that. And you have to consider that a respons spectrum of one respondent
may be fake, while the same one of another respondent may be quite
serious. So this is very difficult, if not impossible, to see from the
obtained scores.
Any normal or not normal distribution of the responses does not apply. You
may have survey statements with which everyone completely agrees (if your
Likert scale is on agreement).
And honestly, ANOVA and GLM are not at all intended for anything like this.
Only if you have different groups according to one or more criteria (other
than the Likert scale variables) you may perform analysis of variance to
show differences in scoring between groups on one or more dependent
(survey) variables. And as that concerns a parametric test you may
question whether your limited and discrete 5 values can be applied to that
(if your scales have 5 points). You might consider a nonparametric test,
but I think a parametric one might do. But that is not what you are
telling us about.
Sorry, forget about your PROC GLM coding. Perhaps syntactically correct,
but conceptually nonsense. Are you a statistician? Is there a statistician
nearby?
Can you be more specific with regard to what you want to show?
How many points does you Likert scale(s) have?
Do you have 10 such scales?
Even if you could "show" (or at least suspect) that certain respondents
did not answer the scales seriously (which is very hard) then what would
you do with them? Remove them? Would that be justified? Did you foresee
this before you began? What is written in the protocol about it?
I am sorry to have to warn you that, what you apparently have in mind, is
wrong anyway. Do not run it and do not try to "interprete" the results.
And I am sorry too for not being able to give some good advice, as I don't
exactly know what you want and what you have. Please give us more
information.
I wish you a lot of success.
Regards - Jim.
--
Jim Groeneveld, Netherlands
Statistician, SAS consultant
home.hccnet.nl/jim.groeneveld
On Thu, 13 Jul 2006 03:42:34 -0700, duklaa <duklaa@GMAIL.COM> wrote:
>Hello,
>
>I have responses to a survey which asked a bunch of questions with
>Likert scale responses. I want to make sure that the responses are all
>different (i.e. the respondents didn't just click uniformly for all
>questions to answer quickly).
>
>How do I run ANOVA to test for this? I don't believe that the responses
>are normally distributed across the scale. Should I use GLM instead?
>
>I used the following syntax:
>
>proc glm data=Work.IMData outset=Tempout;
> class CMF1 CMF2 CMF3 CMF4 CMF5 CMF6 CMF7 CMF8 CMF9 CMF10;
> model CMF1 = CMF2 CMF3 CMF4 CMF5 CMF6 CMF7 CMF8 CMF9 CMF10;
>run;
>
>Is this correct? I am not entirely clear on how "class" and "model"
>work.
>
>How do I interpret the output?
>
>Additionally, if the answers are all different, I want to know which
>pairs are similar, if any. Is there any way to test that apart from
>running pairwise T-tests?
>
>Thanks in advance for the help.
>
>- D
