Dear All,
I am new to factor analysis literature. I have collected some original survey data where I used several batteries of questions measuring different psychological traits. These measurements are validated ones and I have used the exact measures (without dropping some components). Now that I am analysing the data, I have some major doubts. I run a factor analysis first to see if each item load well and some of the items have very low loadings. Since I intend to create an index out of these items, I wonder if I can just drop those items which do not load well in the latent variable and create my index from those which do. I am completely unsure whether I could do it without further collection of new data?. I would appreciate if you could shed some lights on these doubts and/or provide me with some literature on that. Thanks a lot and have a lovely weekend.
Best,
Lala
I guess I have provided you the solution of your problem. Please be kind to check your inbox and email. I have sent you details.
please be kind to share your data in inbox at least. the confidentiality will be maintained. otherwise commenting being difficult for me.
Dear Emran Uddin, I am grateful to you for your potential help. I cannot share the data with you at this stage. Best regards,
Hi Lala!
There are several things (in addition to loadings) that you need to consider before making the decision to drop an item. What about (1) the communalities, (2) rotation, (3) cross-loadings, and -importantly- the (4) substantive meaning and question wording of the problematic items? If, based on all this information, you want to drop items, remember to remove one item at the time and rerun the analysis after each change.
However, if you use validated scales (like NFC), you might consider running a CFA and see whether the scale works (or where there might be problems).
I have some stuff on PCA, EFA and CFA in my office. Stop by when you're back from Chicago ;-) !
- Badges
- Science topic
- Similar topics
- Indexes