In order to do EFA and CFA for assessing psychometric characteristics of an instrument, is it obligatory to randomly select split file or a different data set for undertaking CFA ?
Many thanks.
Hi Vahideh,
While there are writes (Worthington ve Whittaker, 2006) who suggest conduct EFA and CFA same data set because of providing ampirical evidence about data set.
But most writers (Fabrigar, Wegener, MacCallum ve Strahan, 1999) who suggest conduct EFA and CFA different data sets. If your data is big you can split your file as random %50 (for EFA) and %50 (for CFA).
hope it helps.
Article Evaluating the Use of Exploratory Factor Analysis in Psychol...
Article Scale Development Research
Thank you so much.
the total sample size is 343. I do not think it is enough big for splitting. right?
Yes You are true. It is small for splitting. I think use EFA if you develop a scale. If you know the construct already, you can use CFA.
Good luck.
It is not a new instrument. we just assessed the its validation in our population.
Dear Vahideh,
conducting you CFA with the same sample as your EFA is a bit like a self-fulfilling prophesy. Once you have factors that fit your data, there is no point in running a CFA with the same sample. The only thing you would get is a really good model-fit, which is not surprising. The split-file method comes with some perequisites (that I do not know much about) but I have continously been told by statisticians that this is no longer a very good way to confirm your factors.
You sample size is sufficient for both EFA and CFA.
Kind regards
Marcel
Thank you so much Dr. Marcel.
So, do mean a total of around 165 sample is enough for both EFA and CFA for an instrument with originally 33 items?
Dear Vahideh
have not got my PhD yet, but thank you anyway. In an earlier post you wrote that your sample size is 343. That I would consider big enough for both CFA and EFA. 165 is sufficient for EFA, however, I do not have a source at hand giving a ballpark figure for the necesaary sample size for CFA. I have only been told many times that one should aim for no less than 300.
Kind regards
Marcel
One addition: people used to calculate the number of participants for EFA by multiplying the number of items with 5 or sometimes even 10. That is no longer done. Today, it is all down to average communalities, items per factor and, of course, factor loadings.
Later!
Marcel
One addition: people used to calculate the number of participants for EFA by multiplying the number of items with 5 or sometimes even 10. That is no longer done. Today, it is all down to average communalities, items per factor and, of course, factor loadings.
Later!
Marcel
The sample is quite marginal for running CFA. However, there is no strict rule on the adequacy of the sample, but given that some CFA models require a lot of extra parameters to include, your sample size seems a bit low. I agree that if you can run a de novo CFA using a theory for building your model would be a good idea.