Interpreting the "Rotated component matrix" in Varimax Factor Analysis

#1
So I did Varimax factor analysis for 21 questions in order to group them.
As per my understanding the rotated component matrix places each question to where it belongs which is either factor 1, factor 2 etc.

Now two of the questions load under two exact factors: both load under factor 1 and factor 3. what does that mean??

Do I just choose which factor to group them with or is there anything else I have to look at??

THank you in advance!
 

spunky

Smelly poop man with doo doo pants.
#2
Now two of the questions load under two exact factors: both load under factor 1 and factor 3. what does that mean??
that you have cross-loadings and cannot achieve simple structure. it's pretty common to happen, actually. particularly in varimax and other orthogonal rotations because of the very strict assumption of uncorrelated factors.

Do I just choose which factor to group them with or is there anything else I have to look at??
try an oblique rotation method, maybe?
 

noetsi

Fortran must die
#3
What loads onto a factor depends commonly on what you set the minium loading at. For example if you decide the minimum loading is .3 a value of .34 will load on a factor. If you set it at .4 it won't

What did you set as the minimum level to load (and just as important why did you chose that level).

I spent a lot of time recently reviewing the literature on EFA (or more accurately reading reviews of the literature by others). I would guess that few in the social sciences accept loadings lower than .3.

If you find cross loadings there may be a common factor behind the factors you are using. A warning on oblique rotation. Its harder, much harder in my opinion. to interpret the results of an oblique rotation, particularly to someone not familiar with EFA.
 

spunky

Smelly poop man with doo doo pants.
#4
As per my understanding the rotated component matrix places each question to where it belongs which is either factor 1, factor 2 etc.
noetsi brings some very good points and i think i could also add to that... how many factors are you retaining at the end? i can see you mention 'factor 1' and 'factor 3' so i guess there are at least 3 factors. is that all?

i also see you used the word "component". did you use principal component analysis instead of factor analysis?
 

spunky

Smelly poop man with doo doo pants.
#5
A warning on oblique rotation. Its harder, much harder in my opinion. to interpret the results of an oblique rotation, particularly to someone not familiar with EFA.
i think that oblique rotation tends to be a good idea as a data-screening process first. say, for example (i know, sounds crazy) that you end up retaining a solution with 3 or 4 factors... but when you do oblique rotations it turns out that some of them are highly correlated? maybe you could be better off with less factors in such case then but there is no way of knowing that if you only perform orthogonal rotations.