0

I have a questionnaire consisting of 7 multi item scales. The first four multi item scale each represents if a respondent performs a giving role and the remaining 3 multi item scales each represent a respondents identification with e.g. organization, colleagues, top-level managers. All the multi-item scales are based on likert scales summated into a multi-item scale (e.g. 6 questions added up to a total score and then divided with number of items within the scale (6) which gives the mean for the scale - if the mean is above a certain point, I then have created a dummy variabel (e.g. if respondent A has a mean for multi item scale 1 of more than 3, he will get a 1 (and a 0 if not). That is done for each of the scales meaning they only consist of 1's and 0's. Is that the way to do it, if I want to knoq the following?

I want to know if the role identification can explain why a certain role is performed more than other roles?

It should be noted that each multi-item scale is based on reliable theory and I have used cronbachs alpha to confirm it, which stated values for each of 7 scales of more than 0,70.

What should I do next?

Nick Cox
  • 48,377
  • 8
  • 110
  • 156
user48829
  • 1
  • 2
  • Are you saying you want to know if there is a statistical relationship between any of the first 4 scales and any of the last 3 scales? Or is your question more sophisticated than that? In any case, keeping the scale scores (rather than dichotomizing) will give you more information and should result in a more powerful statistical analysis. – Joel W. Jun 23 '14 at 17:46
  • Hi Joel, I have made a summative scale by adding up five variables into 1 based on theory. I have checked for correlation among these variables within the summated scale and cronbachs alpha reports above 0,7. All the 7 multi item scales have cronbach alpha above 0,7. – user48829 Jun 24 '14 at 08:20
  • I am trying to figure if there is a relationship between the variables. The first four could be named: Role A, Role B, Role C and Role D. The remaining three variables could be named identification with A, identification with B and identification with C. I want to find out, if Identification with A leads to Role A, If identification with A leads to Role B etc and also if identification with B leads to Role A, if it leads to role B etc. And the same with identification with C. – user48829 Jun 24 '14 at 08:23
  • As all multi item scales are summated based on ordinal scales (Likert), I assume to be using spearman's rho when checking for linearity. Correct? I found some advice about this here: http://stats.stackexchange.com/questions/8071/how-to-choose-between-pearson-and-spearman-correlation – user48829 Jun 24 '14 at 08:24
  • If you are interested in just predicting roles with single identifications, you could calculate correlations between identifications and roles. That would be 3 identifications and 4 roles, so 12 correlations. If you think the 3 identifications together might predict roles, you might try 4 regression analyses, predicting one role at a time with the three identifications. Which are you interested in. – Joel W. Jun 24 '14 at 13:16
  • Hi Joel, Thanks that is also what I have been doing so far. It's a bit time consuming that's all. – user48829 Jun 24 '14 at 19:54
  • I was wondering if I could do a MANOVA? So that I could calculate if a respondent within identification A would report role levels of X on role A, of Y on B, of Z on C etc. And also done on the same respondents (and the other respondents) on identification B and identification C. That should let me know whether Identification with A will lead to a higher level of role A than e.g. role B, C and D. That is my what I am researching. And if you - or others - have an idea what else I could be doing, it would be much appreciated! Thanks! – user48829 Jun 24 '14 at 20:00
  • Oh Joel, I saw you asked a question in the last sentence. I am interested in both. The main idea is to know what identification can lead to. For this research project - my thesis - I want to find out if identification in any way can predict roles (the 4 roles represent 4 strategic roles). Identification A is made up by 6 variables summed up and divided by 6 to get the mean score for each respondent. The same is done for identification with B and for identification with C. – user48829 Jun 24 '14 at 20:11
  • Some universities have statistics departments that offer statistical consulting to their grad students at no or little cost. – Joel W. Jun 25 '14 at 01:10
  • Hi Joel, That's right. I just thought that I might ask here as well. Partly because here is the place to go in the future. – user48829 Jun 25 '14 at 08:45

0 Answers0