[R] comparing attitudes of 2 groups / likert scales?

[Dieter Menne]

Mona_m wrote:
> 
> For my dissertation, which is due way too soon, I am doing a survey,
> comparing attitudes of 2 independent groups, with 5 scale likert
> questions.
> Basically I want to show if they have similar or different attitudes. I am
> testing 4 hypotheses, and have in total about 20 questions. 
> 
> 

Using an unpaired (in you case) t-test on Likert scale is a bit risky,
because the Gaussian distribution might be severely violated. It might be Ok
if your data are reasonable centered around "moderate", but frequently we
have responses where all but one subject replied with "very good". If you
can create a sum of scores, these are frequently more suitable for being
analyzed by some quasi-continuous method, and using a non-parametric
Wilcoxon test might avoid reviewer comments in some areas of research.

If you only have few levels, something like polr (in MASS) and the plots
created from it by Fox/Anderson might be an alternative (google for
Fox/polytomous effects). These results are more difficult to interpret and I
have seen cases where papers using this where rejected in medical journals
(why don't you use Wilcoxon?).

Overall, when you have more complex cross-over designs with additional
crossed variables, violating Gaussian assumptions for me seems to be the
lesser evil compared to violating independence. "Assess independence, equal
variance and normality -in that order" (van Bell, Statistical rules of
thumb). I remember Douglas Bates mumbling something along the same lines,
but he mentioned a 10 level scale.

Dieter

-- 
View this message in context: http://n4.nabble.com/comparing-attitudes-of-2-groups-likert-scales-tp2015738p2015812.html
Sent from the R help mailing list archive at Nabble.com.