Hi James

you could provide both pattern and structure coefficients from both pattern matrix and structure matrix. In both tables, emergence of factors is likely to be same. If the factor loadings are greater than 0.40 under a particular factor, you can consider those particular items to be belonging to that factor.

Hi Kailash Jandu thanks for your reply...

So using both matrix's for oblique will provide a good indication for factor loading and it's interpretation and so forth? Is there ever a reason that I would use only one of the matrix's in contrast to using both?

And for an orthogonal rotation, I would use the rotated factor matrix, as the only other matrix appearing (at-least in my own testing) is the factor matrix?

Kind regards,

James

Hi James Phillip White

In oblique rotation (oblimin or promax), you get pattern matrix, structure matrix, and factor correlation matrix. Pattern matrix talks about the factor loading of each item onto the factors (that is actually regression coefficient). Structure matrix represents correlation between each item and factor. Factor correlation matrix gives correlation between factors themselves. I would suggest to use both pattern and structure matrices since factor loadings and correlations between items and factors are shown in two different matrices in oblique rotation contrary to orthogonal rotation in which both loadings and correlations are presented in same matrix (Rotated component matrix).

so yes, for orthogonal, only rotated component matrix.

Thanks Kailash Jandu that is very helpful.

Also, because I only have one eigenvalue =>1, when doing the initial testing without the number of factors to extract option (based off eigenvalue) do you think that I only assess that factor? Or should I do some exploring of a variety of the different number factor solutions (e.g., one-factor, two-factor, three-factor and so on) although there is only the one factor that is =>1?

If that question makes sense...

James Phillip White you could definitely try for some other options. There is no need to be dependent on eigen value. Based on the theoretical framework, you could determine the number of factors to be extracted. What matters is the variance explained by the factor for a particular item.

Hi Kailash Jandu when referring to other options, are you referring to other types of rotations, the number of factors extracted, a mixture of both (or something else)?

And in relation to your comment "what matters is the variance explained by the factor for a particular item", are you referring to the extraction sums of squared loading's % of variance or if not, then total? Could you please elaborate a little (if possible). Apologies, I am trying to get my head around all of the potential possibilities...

Hi James Phillip White

you could try other extraction method, rotations, and number of factors extracted.

What I mean by variance explained is that, in EFA, our focus is on getting meaningful loadings in pattern matrix (for oblique) or rotated component matrix (for orthogonal). These matrices give us idea about how much of variance in a particular item is being explained by a particular factor.

If the loadings are not so meaningful, you could consider the possibility of either making a common factor based on cross-loadings or deleting a particular item if that seems appropriate.

It must also take into account the communities and the item-test relationship, in addition to the reliability by dimensions to complement the exploratory analysis, several autopres indicate a total variance of more than 50%.

Hi Cristian Ramos-Vera, are you referring to the inter-item correlation for internal consistency? E.g., Cronbach's Alpha? My analysis reveals that there is one factor that emerges from 8 variables, however, the total variance explained is only around 35%.

Kailash Jandu in relation to the communalities table in the SPSS output (the one with the initial and the extraction loading's), do they also need to be considered before moving onto the matrix tables, and if so, what is the advised cut off range? I have some values that are lower, e.g., as low as .240....

Hi James Phillip White

It is suggested that communilities above 0.40 only should be considered for interpretation or further analysis purpose. If it is below 0.40, you could delete that particular item.