# Require help in ordinal data analysis.

#### Rita08

##### New Member
Hello Everyone,
I have 4 groups and the responses are of ordinal data- 7 rating scales.

My objective is to find out whether which group is better and preferred.

I have found out that when we deal with ordinal data, we could analyse it using ordered logistic regression model (polr package in R) and then plot the predictions using least square means- (lsmeans package in R).

My question is:
1. Is the above method provided a correct way to analyse the ordinal data especially if I have an objective to find out which group is better?
2. Can we come to know whether which particular group is better just based on coefficient values of the logistic model??

Thank you! Wishing you’ll a very happy new year!

#### Karabiner

##### TS Contributor
A common analysis would be using the Kruskal-Wallis H test, for determining whether
the grouping variable is associated with the ordinal scaled dependent variable. If this
turns out statistically significant, you can perform Mann-Whitney U tests for pairwise
comparisons between groups.

With kind regards

Karabiner

#### Rita08

##### New Member
Thank you!
Just a small doubt, wouldn’t I be requiring to check the assumptions of Mann-Whitney test??
I guess the data needs to be symmetric in shape. My data is clearly not symmetrical.

#### Rita08

##### New Member
Hello,
I didn’t understand.
My null hypothesis is there is no significant difference between Group 1 and Group 2.

#### Karabiner

##### TS Contributor
Some examples for null hypotheses (the hypotheses to be nullified):

The proportion of people with overweight in group A is the same as in group B.
The correlation between amount of food intake and depression is = 0.

The mean difference in intelligence between population A and population B is 0.000.
The median difference in intelligence between population A and population B is 0.000.
The probability that a member of population A ranks higher with regard to intelligence than a member of population B is = 0.50.

With kind regards

Karabiner

#### fed2

##### Active Member
Provided that significant differences were detected by this Kruskal-Wallis test, one may be interested in applying post-hoc tests according to Dunn for pairwise multiple comparisons of the ranked data.

#### Karabiner

##### TS Contributor
Provided that significant differences were detected by this Kruskal-Wallis test, one may be interested in applying post-hoc tests according to Dunn for pairwise multiple comparisons of the ranked data.
Seems more appropriate than M-W (or Wilcoxon, respectively),
according to e.g. explanations accompanying the STATA and R
procedures https://rpkgs.datanovia.com/rstatix/reference/dunn_test.html