Hello Researchers,
Is it possible to check whether factor loading obtained after EFA are significant or not? using just a cutoff (.3 and above) for accepting loadings doesnt seem very good criteria.
I believe confidence intervals for factor loadings can be of help. FACTOR (http://psico.fcep.urv.es/utilitats/factor/) provides these when asked to perform robust analyses.
Thanks, I will surely try this. But is there anyway to calculate significance level or EFA loadings in SPSS through Macro etc?
Although some textbook agree with .3 as cutoff for accepting loading, it is preferred to use .4 as a cutoff. The EFA don’t give significant judgment on the loading. However, once you adopt the final structure of the scale, you can run the reliability test.
@Pablo The software is crashing on my Windows 10, even compatibility mode doesn't work. I am trying to open an excel file with 300 cases. It doesn't import .txt at all. Any suggestion for running it successfully?
The data must be in a tab delimited text file with a .dat extension. No other information but the data must be included (i.e. a line with variable names is not allowed). This can be easily done by saving the data from SPSS using the .dat format. You can also do it by saving the data from Excel in tab delimited format and then changing the extension to .dat. (I'd suggest you use the latter method because saving from SPSS seems to create a problem in the first element of the file). I'm attaching a paper where you can find a step-by-step guide to FACTOR. Note, however, that it refers to a previous version of the program, where neither missing data nor robust analysis were still available. It's quite intuitive, though, and you'll find it easy to figure it out. (Be warned: your analysis may take a while to be done, but the result is worth the time spent.)
Sanjay,
A factor loading is a correlation between scores on the variable of interest and scores on the factor. As such, determining statistical significance (from zero would be the customary threshold) is pretty straightforward. With sample sizes of 100, any loading of at least |.197| is statistically different from zero (alpha of .05). Since that level is well below most folks' threshold for a loading to be considered salient, you can see that statistical significance is usually a weak criterion. That is, with a large sample size, it's trivially easy to say that a loading is different from zero. (See Richard Gorsuch's 1983 text, Factor analysis, 2nd ed.)
If you have a lot of loadings to evaluate this way, you'd be wise to use a Bonferroni type adjustment. For example, if you had 30 variables and wanted to keep the aggregate risk of a false positive declaration at .05 or less, then each individual decision would be conducted against a (.05 / 30 = .0016666) significance level. With 100 cases, the minimum correlation for statistical significance at the .0016666 level is |.312|.
Good luck with your work.
@Pablo: Thanks for the suggestions, I will follow it. But I am unable to download the article. Please share a link or pdf of the same. Regards
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS