Which test to do when normality test and equal variances are significant?

#1
Hello everyone.
I have been struggeling a lot with my statistics. My research camps with a low N size (31 respondents to my survey). Now I have to work with what I got. I want to find out how background information such as age and gender are linked with how answers were given to a 6-point Likert scale which asked whether participants agreed or disagreed with statements about the joy of certain languages.

I have checked around and came to the conclusion that for most of my background characteristics, I have to use an ANOVA test. However, when I did the ANOVA test for gender I found out that only 6 of my participants were male and 25 female. When checking the normality test and equal variance test, all p values were below 0.05. Now I have read online that the Kruskal-Wallis ANOVA test can be used for this analysis. However, I am not entirely sure because I find different answers every time I search on the web.

What do you think?
Thanks in advance!
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
Yeah, another issue is that Likert-style data can be problematic. Is it and should it be treated as continuous, and what does the distribution of responses look like?

Given this, you don't have to run statistical models on your data. You can just report basic statistics (counts) and let your audience make up their own mind. Running bivariate states may mislead your audience of patterns that may not be there or generalize well.
 
#3
Yeah, another issue is that Likert-style data can be problematic. Is it and should it be treated as continuous, and what does the distribution of responses look like?

Given this, you don't have to run statistical models on your data. You can just report basic statistics (counts) and let your audience make up their own mind. Running bivariate states may mislead your audience of patterns that may not be there or generalize well.
Thanks for the quick reply! Problem is that I kinda have to do statistical analysis here.
 

hlsmith

Less is more. Stay pure. Stay poor.
#4
But do you really? If it is for a class, what do the instructions say? Because if I came across this at a conference or a peer-review I would say the same thing - just create a couple of tables of the counts stratified by the other variables. Doing this will make it apparent that higher level stats would just be a misuse of them.
 

Karabiner

TS Contributor
#5
Thanks for the quick reply! Problem is that I kinda have to do statistical analysis here.
For which research questions? You did not state them explicity.

And by the way, do you have only Likert-type items as dependent variables,
or real Likert scales (which consistes of two or mire items which are summed up)?

With kind regards

Karabiner
 
#6
For which research questions? You did not state them explicity.

And by the way, do you have only Likert-type items as dependent variables,
or real Likert scales (which consistes of two or mire items which are summed up)?

With kind regards

Karabiner
Ok so, one of the research questions is this:
RQ1: What are the language attitudes of Romanian students and alumni in the Netherlands to French, English, and Romanian?

How do I wanna check this out: I have three seperate sections of likert scale statements where each statement is related to language attitude towards one of the languages. I wanna know if I can put the answers to the likert scale question of each category into one so I can see a general consensus of the attitudes towards the three languages. If that makes sense.

The scale itself is 1= strongly disagree and 6= strongly agree
 

Karabiner

TS Contributor
#7
Now I do not know what you mean by checking background characteristics, or what you mean by doing ANOVA in
your initial message.

Seemingly, you want to compare the attitudes of the n=31 participants towards the three languages? If you
want to perform a global test, this can be done using repeated-measures analysis of variance, or using the
Friedman test.

With kind regards

Karabiner
 

hlsmith

Less is more. Stay pure. Stay poor.
#8
What they are doing, is treating Likert as a continous variable and trying to compare subgroups in the n=31. Which sex only has two groups so it wouldn't be ANOVA.


So average likert response for men and women (neglecting other gender groups). Also, that RQ1 example is poorly written. What if they hate French and love English? Are all three languages really grouped together in this one question? Or is this actually broken out into 3 groups. if you are asking a bunch of repeated questions like the latter, than you need to adjust your level of significance to minimize risk of false discovery (e.g., multiplicity).