Hi there,
I am working on my Ph.D dissertation. I have run a factor analysis of 56 items with maximum likelihood and direct oblimin method. It turns out that only one factor can be extracted (scree plot and Egenvalue>1) from a pool of 56 items, and one item has excessively large loadings (about 70%) on the single factor.
Examining respondents' responses to the items, I found many cases of identical answers to consectutive questions (at least 5). Does this suggest for the revision of the survey?
Thank you very much.
Manzu He, there could be different reasons for your results. One is that the items are essentially measuring the same kind of thing - at least from the perspective of your respondents. I could be wrong, but I suspect you are conducting your research in China. In some research that my colleagues and I have been involved in, we found Chinese participants' responses loaded on a single factor in comparison with western participants who, on the same scale, tended a lot more to compartmentalise. We suggested that our Chinese participants might integrate their ideas (maybe a Confucian tendency) more than do westerners - but that might depend on what is being investigated.
Another reason for a single factor, resulting from strings of similar responses, is that your participants have done what is called satisficing, i.e., rushed quickly through your survey to get the task over and done with as quickly as possible. If they don't really care about the research (or, perhaps they believe there are too many items to respond to), they might choose the same response consistently. They might even take pleasure in sabotaging your research. It happens!
I suspect you should consider reducing the number of items on your survey so that only the most important things are asked, and also trying to do things that encourage participants to put honest effort into responding. Sometimes the whole design of a survey is important, e.g., ensuring the instructions make sense, that items are worded well, and that all "necessary" response options are provided.
Maximum likelihood is only done according to your data that meets the normality distribution of the data, you should use generalized least squares or parallel analysis. You should confirm the results by CFA using SEM indicators (IFC, TLI, etc) to demonstrate the unidimensionality of the instrument that should be in accordance with your theoretical model.
Hello Manzu,
In addition to Robert Trevethan's excellent suggestions, I would add that:
1. scale construction and validation is not an easy or foolproof process (as you may have discovered);
2. any item having no variation (e.g., if everyone chose the "5" value as their response) will not correlate with any other item and will not contribute to the scale; and
3. the fact that one item has a much stronger loading than others does not suggest a flaw with the item. However, if there is only one item that satisfies your criterion for salient affiliation with the factor, then most folks would argue that you don't have a (common) factor.
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts