I have collected my data and I am now analyzing it. Here are the areas I cannot understand:
1. The pre-post tests questions are alike every way (except in how participants respond after exposure to the IV. When I run a factor analysis, do I need to enter both the pretest and posttest questions or just one of them? If only one, would it matter which?
2. "near zero loading" ... what exactly should be "near zero"?
I am new to all this so any assistance would be great appreciated.
Can you explain why you are using a factor analysis?
Personally I would be interested in how each question has changed after the intervention and would therefore be considering a multivariate regression ( I do not mean mean multiple rergression) - have a look at
http://www.bristol.ac.uk/cmm/software/mlwin/mlwin-resources.html#multipleresp
and my answer to this question
https://www.researchgate.net/post/Can_long_format_panel_data_be_used_for_a_range_of_analyses_or_is_it_specific_to_repeated_measures_ie_GLMM_LMM
If you are following a standard protocol then ignore my intervention!
Best of luck.
Hi, Camille.
Kelvyn Jones's answer is a good one. For a tad more "explanation," what you want to know in a pre- to post-test situation is what (if anything) changed. You cannot do that if the ruler itself changes every time you measure. You want there to be one ruler that measures things the same way both times that you measure.
Factor analysis will answer the question of whether items go together to make a single measure. However, when you have FA make that measure, the result will be z-scored to your sample. That means you have lost your base-line ruler from the item scale, and have a new one set to your sample's responses. If you do it again for the post-test, you moved the ruler again, resetting it to the new set of responses.
So, one way to think about this problem is to consider the pre-test responses to be your ruler, against which you are comparing how the post-test responses compare. IRT analysis is also set up to deal with this problem. The original scaling of items into a composite measure (if there is one) will also do it.
Where you need to focus your attention as the researcher is on the meaning of the ruler - what does "change" mean along the ruler you are using? If you can't explain that to the reader, your study won't make sense to us.
The survey was geared at measuring state of 'need' and 'empathy' and after testing my hypothesis and not getting the results I anticipated, I am now investigating whether my measure of these concepts is valid, that is, finding out whether my 'need' and 'empathy' questions are independent. The FA is being done in addition to other tests but I got stuck in the FA bit.
One important thigh to know is if your measure have changed from pre to post test. This could be done in a confirmatory factor analysis framework (I don't know if your sample is big enough for running CFA with ordinal variables, but MPLUS method works well with small sample size).
Also you could do some Rasch analysis and check for differential items functioning between pre and post test. It would we a good think try to figure out why those items show DIF between pre and post-test.
Kindly
Andres
Hi Camille,
I agree with the above comments, and would like to recommend a paper by Hair, Black, Babin, Anderson, and Tatham in their book Multivariate Data Analysis (2006). According to Hair et al., the starting point in factor analysis is the research problem and the underlying variables related to the research problem to determine whether the research problem and variables lend themselves to factor analysis. First, Hair et al. observed “The general purpose of factor analytic techniques is to find a way to summarize the information contained in a number of original detailed variables into a smaller set of new, more generalized composite dimensions” (p. 107). I am attaching another document which I hope you will find useful.
Best,
Elenoa
What if we run FA for both pre and post, and then we choose the items that are in common (if any) between the two tests?
Camille-- For a factor-analysis how-to, there is a great chapter on factor analysis in this book below, "A conceptual guide to statistics using SPSS" (Berkman & Reise, 2012). It's basically a guide (with accessible explanations at the conceptual level) for how to conduct various analyses in SPSS.
http://www.sagepub.com/books/Book233559/reviews#tabview=toc
- Badges
- Science topic