I am interested in the Self Determination Theory and particularly in intrinsic motivation in the workplace. I have the opportunity to analyse the data of an existing questionnaire on the quality of life at work. The 81 questions are designed as followed: "I find my work interesting" The choices are "Completely , Rather yes , Rather not , Not at all" and different for each question. I am a bit lost on what method to use and what I could extract.
I think I could extract the important data on wellbeing and see if it is predicted/linked by others data on intrinsic motivation. But which method shoud I use?
Would anybody have an advice? Thanks
Dear Sarah, I'm afraid I don't fully understand your question. Are you inquiring about the way in which you should add up items in order to form composite measures of the constructs you're interested in? If so, such information should be found in the manual and/or relevant references about the questionnaire, since you say it's an already existing one. If it's not, then the factor structure of the test must be explored with exploratory factor analysis.
If your question is about a different thing and I completely missed the point, please help me understand what you're really asking. Regards.
It sounds like you have items that were part of pre-existing scales. For example, are there a separate set of questions that are specifically about wellbeing?
What I have understood that you have used a questionnaire made by someone else and you want to analyze the data. If that is the case, then the analysis depends upon how the items were measured.
Your question suggests that data was collected using 81 items and measurement was on ordinal scale but the each item was presented on different ordinal scale. Well in that case, I am afraid, you will have to analyze each statement separately. Mean may not give you much of information in this regard, but percentages, mode and median may show something if you want to make a comparative analysis (among genders, age groups, educational levels etc.). This approach may not provide anything fruitful. A rather better approach would be to do cluster analysis. The respondents can be divided into groups, based upon intrinsic motivation and further the clusters (groups) can be used for further analysis.
Had all 81 items been measured on similar scale "completely to not at all" I would have suggested to use Exploratory Factor Analysis to identify factors, which then can treated as constructs. Once factors are identified, the composite variables can be create and further analysis can be carried out.
N.B. Thanks for asking this question. I was rather stuck with a set of data as to how to analyze it and while answering to your query, I found a way to analyze it. I have 74 statements on 7 point ordinal scale but I cannot do factor analysis (doesn't give anything that I could explain); will do clustering now :)
Ali Farooq, since EFA only requires correlations as input, I don't see any problem with using Likert-scored items that had different "end-points." Having similar end points throughout would probably lead to higher correlations, but it doesn't prohibit examining the existing correlations.
David, thank you so much for the correction. I have not come across any study where items measured on different scales were use in EFA, most of the studies showed EFA on the items having similar scales and of course that would lead to higher correlations.
So what I take from here is that if I have items measured on different scales for example, 1-5 scale, 1-3 scale and 1-7 scale, there is no problem in conducting EFA on such items.
Kindly guide on this as well: If I want to compute variable based upon the factors determined during the EFA. Wouldn't it effect the composite score of the variable when items measured on two different scales are combined using Mean(I1, I2, I3,I4), where I1 and I2 were 5 point scales and I3 and I4 were 3- and 7-point scales respectively?
Thanks,
I haven't ever seen anyone mix together 3-point, and 5-point etc. items in the same scale. I do think that would be unwise, but from a sheer computational point of view, you could still calculate correlations, and that would be sufficient for EFA.
Tools/Techniques to be used will depend on two things:
Define them clearly to get specific answer.
Thank you for your answer Ali Farooq, It is the first time I analyse data.
And yes, you got it right. I am using a questionnaire made by someone else who only looks at the quality of life at work. As explained a question of the 81 is for example As part of my employment, “I have the opportunity to manage my own pace of work (time, speed, breaks)”. The possible answer can be chosen from Strongly Agree, Agree, Disagree, Strongly or Disagree. The choices for the answers are different from time to time. You made me realise that I came pack the questions in groups that have the same answers and they fall into categories with similar themes that I can analyses easily with scales. I did not choose this questionnaire unfortunately and still I want to analyse the data for it to be interesting. I wish to link each question to competence, relatedness or autonomy or eudominic wellbeing. I then wish to analyse combine data to link the intrinsic motivation answers (competence + autonomy + relatedness) to the eudominic wellbeing ones but I don't know what is the most relevant method for that.
thanks a million for your time! this is precious!
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude