Nonparametric test for demographics associations with pre and post likert scale

[triunk]

Hi all,

My topic:
Evaluation of a psychology university course.
My problem:
I have a 32 item scale along with demographics (gender, ethnicity, college)
I am using IBM SPSS 20 (so weird from past SPSS...). I am trying to assess diferences (in SPSS) pre and post likert scale (it was actually a 32 item course evaluation scale) based on demographics (gender, ethnicity and college type).

My questions:
Should I use a nonparametric equivalent to an anova form?
If yes this is repeated measures avova equivalent: Friedman anova.

How can I do in IBM SPSS 20 friedman avova per subscale items (if X is the subscale then for X1pre and X1post item and then for X1PRE and X2PRE) based on demographics (e.g. gender groups (female, male) or race (asian, african american, black, white...etc) and provide a graph for it???

Do I have to include as a posthoc any bonferroni corrections?

Many thanks!!!

C

[EliteResearch]

Hi,
Is there a reason you need to do a nonparametric equivalent? If your data are not normal, then you’d want to use Friedman’s because it makes no assumptions about the distribution of the data. If your data are at least somewhat normally distributed, you should be able to do a repeated-measures ANOVA.

SPSS graphs tend to be minimally helpful so I'd recommend using word or excel for your graphs.

If you only have one "treatment" (the psychology course) and your group was only measured at two times (pre/post) then you won't need the bonferroni correction for demographic variables with two levels (e.g. gender). I'm not sure about demographic variables with 3 or more levels.

Good luck!

Amanda

[triunk]

Hi,
Is there a reason you need to do a nonparametric equivalent? If your data are not normal, then you’d want to use Friedman’s because it makes no assumptions about the distribution of the data. If your data are at least somewhat normally distributed, you should be able to do a repeated-measures ANOVA.

SPSS graphs tend to be minimally helpful so I'd recommend using word or excel for your graphs.

If you only have one "treatment" (the psychology course) and your group was only measured at two times (pre/post) then you won't need the bonferroni correction for demographic variables with two levels (e.g. gender). I'm not sure about demographic variables with 3 or more levels.

Good luck!

Amanda

Thank you Amanda! Actually my likert scale is ordinal data that do not follow normal distribution....so I have to use nonparametrics...and based on your advice...friedman anova...So...any ideas how should I use friedman of wilcoxon based on my demographic data are more than welcome!

Many thanks again

X

[Karabiner]

(snip, mis-read this thread)

[triunk]

What exactely do you want to find out? Pre-post differences for different
subgroups, based on gender and/or ethnicity and/or college type? Or to
assess pre-post differences while controlling for gender, ethnicity and
college type?

Kind regards

K.

Thank you very much for your response.
Actually, I want to do both of them if it is possible

Thank you again

X

[Karabiner]

If you want to test for interactions, than you need to do the
usual "parametric" analysis. If your sample size is large, then
you usually do not have to bother about normality too much.
And a Likert scales are often treated like interval scales in
analyses.

Kind regards

K.

[triunk]

"What exactely do you want to find out? Pre-post differences for different
subgroups, based on gender and/or ethnicity and/or college type?

Hi K thank you for this clarification.

Well...for normality I did kolmogorov-smirnov test to a number of items in my likert scales and was shown that my data do not follow normal distribution.
So I for the first question (re-post differences for different subgroups, based on gender and/or ethnicity and/or college type?)
decided to do wilcoxon test for each pre and post item while I have previously split the fales to groups (gender, then ethnicity and then college)
I don't know if I am right...any opinions/corrections most welcome...

With respect to the second question ( Or to assess pre-post differences while controlling for gender, ethnicity and college type?) I really don't know how to handle it...so any insights are most welcome!!!

Thanks again so much!

X

[Karabiner]

Well...for normality I did kolmogorov-smirnov test to a number of items in my likert scales and was shown that my data do not follow normal distribution.

If your sample size is very large, then the K-S test indicates that data
are not from a normal distribution even if the deviation from normal
is practically irrelevant.

So I for the first question (re-post differences for different subgroups, based on gender and/or ethnicity and/or college type?)
decided to do wilcoxon test for each pre and post item while I have previously split the fales to groups (gender, then ethnicity and then college)
I don't know if I am right...any opinions/corrections most welcome...

I very rarely read this kind of analysis. Usually one is interested in
differences of pre-post developments between groups, not in a bunch
of sepereate pre-post results. BTW you usually have to account for
multiple testing, e.g. by using the Bonferroni correction.

With respect to the second question ( Or to assess pre-post differences while controlling for gender, ethnicity and college type?) I really don't know how to handle it...so any insights are most welcome!!!

Mixed ANOVA (mixture of between-subjects (grouping) factors and a within-patient
(time) factor. The interaction between the grouping factor and the time factor
shows whether pre-post-differences differ between groups.

Kind regards

K.