I wonder the extent to which it is acceptable to use validation data for further (correlational and descriptive) analysis in situations like the fictional one described below:
Suppose I am planning to validate a 30 item scale A into Turkish. I collect data (n=300) for the final A translation, alongside with previously validated scales B and C addressing different constructs. I collect data and drop 2 items for low psychometric performance and adapt the factorial structure for the Turkish populations. I use the scores from the latest 28 item version A with adapted factorial structure, together with the ones from B and C to check for correlations or regressions. I draw conclusions from these analyses and write my report transparently reporting the procedure.
As i understood your case, i feel that the sample size is ok, and you can run your test , and conclude.
how will u adapt the factorial structure? u need to test its fit, so it will be good to keep part of the dataset aside for testing the fit (cross validation)
Dear Ramachandra,
Lets say I would split the sample into two and make the cross validation. If the validation is successful; Would it be okay to use the original 300 subject dataset for further analysis?
I agree with the researchers, yes acceptable under the criteria that researchers already mentioned.
I agree, if anything you are providing further evidence of validity (i.e., your scale is significantly related to constructs that it should theoretically be related to).
Thanks for all the replies. I particularly asked this question to understand whether this kind of a practice would be regarded as a malpractice (i.e. duplication of data) or contrary to the logic of scientific research. As far as I understand, the repliers points no problems with respect to this area.
Thanks Everyone for the informative responses. As far as I understand, although the fictionary design can not be regarded as malpractice, it would end up in bad science.