I am doing a little research on comparison of Quality of life of diabetic and non diabetic patients.
For this purpose I've two questionnaires, one for the diabetic people, including 39 questions and the other for non diabetic people, including 29 questions. The data is likert. (Attaching the 1st page of the questionnaires)
Now some of the questions are similar to each other(in each questionnaire separately) That's why I want to use factor analysis so that only important factors could be extracted. I am thinking of using factor analysis separately on the two questionnaires and when the factors are extracted I will use logistic regression for the comparison of the quality of life of both groups.
Now my question is whether applying factor analysis is right or not? plus if it is right and I do so and the factors are extracted, how would I enter the data for logistic regression analysis? How it would be run and interpreted?
A ratio of at least five, better ten data sets per item of the utilised QoL scales would be good for reliable factor results.
You might decide to perform two seperate EFAs for the two scales or - if filled in by the same participants - one comprehensive EFA (if i got it right, you have two samples, so two EFAs are required).
Regarding your regressions, you might integrate those items found to be associated (loading on the same factor in the rotated solution) to a scale. However, you should test consistency of each defined scale before running subsequent regression analyses.
Best wishes!
First of all I am not clear why you want to use separate scale for assessing quality of life of diabetic and non-diabetic people if your intention is to compare these two groups. My suggestion would be to develop a scale, generate items that can be used for measure QoL of both groups. Then perform EFA and assess the dimensionality of your scale (Principal component analysis with varimax rotation). You can decide on which items should be retained based on the factor loading of individual item. Also check the internal consistency of your scale by computing Cronbach's alpha.
Once you finalize your scale, you can collect data from both the groups and run independent sample t test to find out whether there exist any difference between the two groups or not. You can further assess the validity of your scale by conducting CFA..
I agree that you can do the comparison and even fit regression analysis. However, my concern is the point raised above by Arjun. If you are planning comparing two groups it is a must that you use the same scale to measure QoL in both groups. Your current plan of comparing the two groups is not fair and will not lead to a valid result.
As for the use of regression models, you have to first extract the scale (using PCA) and generate the score for the scale (factor scores---you may name it as QoL score in your case). Then there are two possibilities. One is to fit linear regression and interpret the beta coefficient with its 95%CI. The other is to rank the respondents based on the factor scores. For example, you may rank the as poor, midium, high quality. This forms the dependent variable to fit a multinomial logistic regression model. But not that this is to identify predictors of QoL not to compare the groups.
To compare the groups please do what is suggested by Arjun.
factor analysis helps to determine questions in this case that measure the same property in order to decrease the amount of variables. If those questionaries are new and You have alot of patients (at the very least 100) then it would make sense. Howevere if those scales are already beign used in practise I would useprviously desribed methodology of using the questionaire.
You have 29 questions in one questionnaire and 39 in another. Some of them surely overlap in terms of the qualities they are measuring. Not only you can use factor analysis but you should. Reduce the number of variables you have. Not only that will facilitate the analysis but it will also reduce the likelihood of multicollinearity in your model.
Independent t sample test can be used to see the significant difference between two population on one variable.
Quality of life or health-related quality of life? There are some disease-specific measures that are useful. QoL measures vary because the definition of this construct varies. Yes, it sounds like you are comparing group means: t-tests or a one-way ANOVA.
Nnote: The QoL score of the first diabetic subject has nothing to do with that of the
first non-diabetic subject. When two samples have the same population variance, the independent samples t-test uses the pooled variance when computing standard error.
If unequal variance, individual variances need to be used instead in computation, and degrees of freedom should be approximated. The folded F test is used to evaluate the equality of two variances.
- Badges
- Science topic
- Similar topics
- Medicine
- Disease
- Metabolic Diseases
- Glucose Metabolism Disorders
- Diabetes