I'm developing a model of academic achievement using student responses to approximately 20 established scales shown to be related to academic achievement. There are over 200 items. I am wondering how best to identify factors using factor analysis. Would it be appropriate to put all of the items in a single analysis? Or would sub-grouped analyses be better? I have measures at different levels of specificity from broad personality traits (e.g., conscientiousness) to more specific social cognitions (e.g., grade goals, academic self efficacy). Any guidance/advice gratefully received.
1
-
$\begingroup$ It seems you already have an unregistered account on this site, michelle. Can you confirm this? We will merge them but you will need to register to take full advantage of SE facilities. $\endgroup$– chlCommented Oct 13, 2012 at 19:29
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
It sounds, from your question, like the scales are well established; in this case, they may have already been factor analyzed. Certainly the "big 5" personality traits have been factor analyzed many times. This might point to a confirmatory factor analysis.
But you say you want to "develop a model of academic achievement", which seems like you want to use these factors to predict achievement, perhaps using some form of regression. In this case, something like partial least squares might be best.
However, if your model is just for general use - a theoretical construct - then an exploratory factor analysis of all 200 items is probably what you want.
One problem that is often overlooked in EFA is that you can't find things that aren't there. That is, if some aspect of academic achievement is not included in your 200 items, you won't be able to find it.
answered Oct 13, 2012 at 12:57
-
$\begingroup$ Thanks Peter. Lots of the measures have already been validated. However, the problem is that many of these measures overlap. I'm wanting to identify this overlap and reduce the number of measures. However, some of the items are more similar than others and I am wondering if it is appropriate to lump them all together e.g., whether I will miss important factors? $\endgroup$– michelleCommented Oct 13, 2012 at 13:16
-
$\begingroup$ Having multiple measures that are similar is OK in a factor analysis; but you could miss things for factors that are under-represented. There's no real way around that, unfortunately. $\endgroup$ Commented Oct 13, 2012 at 13:20