Hi Israt,

one way would be to split the data set and derive a model from the first set with exploratory factor analysis and do a confirmatory factor analysis with the derived model on the second half. Then you can see with some common fit indices like RMSEA how well the model fits to the independent data set. There are some crucial points in doing an exploratory factor analysis like Number of factors to extract, Rotation method and some more. We wrote a paper on exactly that topic, maybe you want to have a look.

Best,

Martin

Martin, Thank you so much. It will be a great help. 

To ensure the validity and reliability of the factor structureAlpha, split-half coefficients and test–retest reliability.  See the attachment for your kind perusal.

Hi Senthil, thanks for the file.

Martin and Senthil, we did not have CFA in our course outline. Therefore, we need to find out the reliability and validity of the extracted factors without using Confirmatory Factor Analysis. I can ensure the face validity with consulting other PhD students. However, I am not sure how to prove other validity or reliability without using CFA.

Dear Israt,  Good Morning.

Confirmatory factor analysis is a hypothesis testing approach to factor analysis where one defines the factor structure apriori and using the structural equation modeling approach how far the data fits with this predefined factor structure. If this is your objective then as suggested by others you will have to use special software like AMOS, LISREL, EQS etc.

However, if your objective is to do a cross validation of earlier reported factor structure then EFA is advised. Use any one method of EFA (preferably the one used in earlier research) and then use some tests to determine the number of factors to retain for subsequent rotation. It would be better if one adopts several criteria to decide the number of factors such as scree plot, parallel analysis, Velicer's MAP. C-Hull method etc.

If you get same number of factors as obtained in previous research even then you can not say that you have arrived at a similar factor structure unless you demonstrate that all the factors have pattern of factor loadings similar to the earlier one. This can be done by computing factor structure congruence that has been nicely described in the Book on factor analysis by Gorsuch.

Hi Senthil,

Thank you so much. I will read the book written by Gorsuch. I am not supposed to use CFA. I am only supposed to use EFA to find the factor structure and then report how I did it and if it is valid and reliable. Thank you so much for your help. I will post what I come up with once I finish my report.

Cheers!

Give the restriction to EFA, you can perhaps execute EFA using whatever method you covered in the course. One way to infer validity is to look at the factor loadings. First of all, they need to be significant, but even more important are the sizes. Factor loadings that are greater than or equal to 0.7 (absolute value) are generally regarded as ideal. Such loadings will indicate high item validity. Even as I say this generally, you should be aware that different methods of estimation can produce different loadings. You should therefore use the 0.7 only as a guided. In addition, the acceptable loading size can also vary depending on the size of the sample.

Overall construct validity can be inferred from the average variance extracted in the accepted factor model. This can be calculated as the average of the squared standardised factor loadings in CFA. Average variance extracted greater than 0.5 is indicative of adequate factor level convergent validity. With respect to EFA with SPSS, I think that the program provides a % of the variance extracted in the output. This may be used to allow an evaluation of the factor level convergent validity. Larger extracted variances are better. However, this should not be your criterion for determining the number of factors to accept. If is is used at all to inform model selection, it is usually in combination with other criteria and in my experience, it is mostly employed to reject less parsimonious models that improve overall fit but only improve the extracted variance marginally. I should note also that some researchers will not recommend the use of the % of the variance extracted. They might prefer to focus on the size of the loadings (and AVE) for validity and on the explained variance of the items themselves for reliability. I am not sure if SPSS gives the latter.

Once the factor structure is determined you can work backwards to determine reliability. Here you can evaluate the set of items forming the various dimensions. I am tempted to mention Cronbach Alpha here because is it perhaps the most traditionally recognised method. However, there are many problems with this measure so you will need to determine reliability depending on what you have focused on in your course.

If you can get your hands on them, some good references for these issues are Hair et al (2006). Multivariate Data Analysis (http://www.abebooks.com/Multivariate-Data-Analysis-6th-Edition-Hair/3039467044/bd)

Tacq (1997) Multivariate Analysis Techniques in Social Science Research (http://books.google.be/books/about/Multivariate_Analysis_Techniques_in_Soci.html?id=hS-f5kc_03UC&redir_esc=y)

I hope that this helps.