Hi I would appreciate some guidance. I am embarking on a quantitative follow-up to an initial phase of qualitative analysis. I intend to use slightly modified scales from the literature to determine the degree to which the findings generalise in the wider population.
I understand from reading around that it is robust practice to test the validity and reliability of such scales when used in different populations (albeit very similar populations in my case). Can you recommend any alternatives to factor analysis for objectively testing construct validity particularly for smaller samples? I want to demonstrate that the scales representing my variables of interest are indeed measuring what I think they are measuring.
While I understand that there is no 'magic number' of cases required for factor analysis, my population is difficult to access and due to time and resource constraints as well as predicted low response rates I may only get around 100 cases to work with. I understand from reading Field and Pallant and the authorities that is unlikely to be enough to yield a reasonable factor solution.
Hello Nick,
You're back at it! As it happens, I recently made a 2009 article of mine freely available on ResearchGate. It was primarily about measuring people's perceptions about their foot health, but I dealt with a variety of forms of reliability and validity that are more generally applicable. If you can get to my RG site, you should be able to locate it.
If you can't access it, feel free to let me know. I could send you a copy.
Robt.
Hi Nick,
Essentially,you have two issues here. 1) Is the power of your study enough to generalise/infer your findings from your sample to your target population? 2) Is the sample sufficient to compute factor analysis and produce robust construct validity results?
Well, both questions are related to calculating sample sizes but using different methods and purposes.
To answer qn.1, look up sample size calculation for your type of study and estimate how many minimum sample you need to achieve alpha > 0.8 or confidence of 95%. If you have collected your data, you can calculate this retrospectively, to identify whether you have reached power and precision to generalise your finding. If not, you just need to fess up and report in your limitation of study.
To answer qn.2, you are right, it may seem arbitrary but as a rule of thumb, a minimum of 10 samples per item/question/variable is necessary to avoid computational problems later. This depends on how many constructs and how many items per constructs you have. If you have one construct with about four observed variables, then even sample size of 60 is adequate. In any case, give your best to obtain as many sample as you can. Any sample size over 300 is considered adequate for factor analysis. You may also come across those recommending cut-off sample sizes: 50 is very poor, 100 is poor, 200 is fair, 300 is good and 500 is very good, and 1,000 or more is excellent.
You may also try linear discriminant analysis as alternative. Hope this helps!
- Badges
- Science topic
- Similar topics
- Cognitive Science
- Thinking