Hi,

I am using a questionnaire comprising 70 questions, each one being a 5 item Likert scale ranging from 1=strongly disagree to 5=strongly agree. The 70 questions can be grouped into 7 sub-groups of varying quantity. The sample consists of 80 participants randomised to two groups each receiving one of two methods of administering the same medical treatment. I want to compare the attitudes of the two groups; the primary hypothesis being that one group will perceive more adverse effects than the other and will therefore express less positive attitudes than the other.

My assumption is that the attitudes expressed are ordinal and the Likert scale numbers used to code the responses are merely labels. As such they cannot be summed and I cannot compute means and SD's to compare the two groups. I have considered summing the ranks and comparing the medians but the results miss a lot of real information. I am aware of the debate regarding the ordinal or interval nature of data but I would prefer to consider this data as ordinal as it is the attitudes of participants which is the primary concern.

My question is; how do I create a score or a ranking system to compare the individual sub-sections and the overall score/ rank between the two groups?

I'm using this data as part of a PhD and I'm really getting nowhere trying to solve this problem. I would greatly appreciate some advice.

Many thanks,
Adam

Look into an independence test split by group

Statman,

Thanks or the reply. I've tried to do an independence test split by group by;
In menu toolbar - Data, then split file, compare groups, moving my variable which defines my two groups in the 'groups based on' box. This split the file layered by the variable. A Mann-Whitney U Test would not perform then!

Am I going about this wrong? I'm using SPSS 16.0

Thanks again,
Adam
P.S. Have you come across any type of a worked example which analyses Likert data this way?

Sorry, not the SPLIT file under Data but set the grouping variable as Group under the independence test

Statman,

Appreciate the reply and I've followed that. It works fine except this procedure produces a Mann Whitney test with related results for each individual question. Bearing in mind that (1) I will have 70 questions which is a lot of data to analyse and report the results of as individual questions and (2) i will be selecting the most 'important' question from each sub-section to compare using chi-squares between the two groups, I was hoping to find a way of comparing the sub-sections as sub-sections, essentially comparing some form of score or ranking of the sub-sections between the two groups. This may not actually be possible as my data is ordinal and not interval. A colleague has suggested that I need to conduct a factor analysis which will allow comparision of subscales between groups. I don't see an alternative! Would you agree?

Again, I appreciate your advice.

Adam