I agree with Agnieszka, it's better to use both exploratory and confirmatory approaches specially if the instrument will be used in the region for the first time, as I understood that you will translate to Arabic and no one did it before.

Rana

Hi,

whenever you have some idea which latent variables/factors indicators opt to measure, go to CFA. EFA was invented for people randomly creating items with no idea what they would measure. And just as a personal, biased note, I doubt that EFA can compensate for this theoretical failure :)

Hence, if you have assumptions, test them. And: be prepared and open to respond to model misfit. Psychology is full of situations, in which researchers stubbornly keep their beloved scales although the model screams "I am wrong" :)

In addition: If you have a translated measurement instrument, you should/could test for measurement invariance between the Jordanian sample and the sample for which the scale was initially created,

Hope this helps,

Holger

If you intend to use PCA as an extraction method for EFA would actually be not only insufficient but probably wrong for adapting a psychometric instrument's translation. PCA is sensitive to the actual variables you feed it and the original researchers ended up with the particular set of items from a a larger pool. Thus using the refined scale items as a new pool will lead to misleading results. I would say go to CFA without any PCA before. Draw your model structure from the references of the original instrument.

[Furthermore, and as a general note on PCA, Overall's simulation has shown that PCA does not reveal the true structure behind a set of observed variables - it is rather a good summary of the correlations. PCA components are good for prediction purposes, but not for theoretical understanding].

http://psycnet.apa.org/journals/bul/61/4/270/

Interesting comment Nikolaos, thanks. 

First of all, PCA is not a Factor Analysis Method. The main reason of this idea come from the communality of variables. Cause in PCA communality of variables initial values accepted to be 1. But in real FA this value is between 0 and 1. This means in PCA error variance of variables taken account for the analysis and these conditions flatter the factor loadings of variables. But in a work from Snook and Gorsuch (1989) explained the difference between these two approaches in a monte carlo study. According to results, under 30 item the solutions seem similar, but over 30 items results seem different.

For the main question, making FA lead researcher how to conduct CFA..

Snook, S. C., & Gorsuch, R. L. (1989). Component Analysis versus Common Factor-Analysis – A Monte- Carlo Study. Psychological Bulletin, 106(1), 148-154.

As well a using factor analytic methods, is there scope for administering your new instrument alongside another instrument measuring the sane or similar constructs, already translated into Arabic?

I agree with PCA is not a EFA. despite that you will fine PCA under factor analysis in SPSS. EFA could be used to establish a preliminary construct validity but if you have the theory that support the factor structure, it is better to use CFA. if you are intending to use EFA, choose common factor method (in SPSS) not PCA. PCA intended to aggregate the data.