# Follow-up on Kruskal-Wallis

#### pleonasm

##### New Member
Hi everybody!

I just found this forum and became so happy because I'm in a bit of trouble and have no book to help me right now.

I am doing an statistical analysis based on the answers from a survey a friend of mine created. The respondents are visitors to a music festival and the survey was divided into 8 sections (examples of these sections include: satisfaction with camping area, satisfaction with festival area, knowledge of sponsors etc.)

The items are most often in Likert-form 1-5 from (Very good - very bad).
The items are not clustered together with each other and thus treated as individual responses. I have therefore treated the Likert-scale as being on an ordinal scale level.

Q1) Is that correct or could I treat it as interval when I present the descriptive statistics (i.e. should I present mean and standard deviation or median and mode?)

Back to the real issue -->

The distributions are skewed anyway so in my analysis I use non-parametric tests.

For every item I do 4 comparisons based on Age, Gender, Previous Experience and Living arrangement (camping site, at home or other).

For my IV 'Age' for instance I have five different groups: 1997-1995, 1994-1992, 1991-1989, 1988-1985, 1985 and earlier.

I have done a Kruskal-Wallis and where the null hypothesis is rejected I want to do a follow up.

The problem is I have tons of data and not much time. Could I compare every group against an "other" group with Mann-Whitney? i might lose some information if you compare it to doing pairwise comparisons but it's okay within the time frame I have.

Example:

1997-1995 against "all others"
1994-1992 against "all others"
...
...
1984 and earlier against "all others"

I will of course make a Bonferroni adjustment so that the alpha level will be 0,05/5 = 0,01 in this case.

My question is: Is this possible? I have a nagging feeling that this is not a correct way to do things. I know that I lose some information but that would be okay, I am only interested in see if any group is different to "the rest". Do I have to do a pairwise comparison either way or is this enough?

Kindest Regards
Linus Lind

Last edited: