Is it acceptable to merge data from 2 different studies for the same scale? both studies are almost conducted on similar population. or, to do analysis for the psychometrics of the 2 sets of data seperatly and compare the similarities/differences. If same factors structure arise, then this validate the new outcome among the new popluation.
This depends on several factors, the most important being what hypothesis you are testing. If you are, for example, developing and validating a novel questionnaire, I would opt for two separate analysis.
Dear Prof Muayyad, this was what I did with my co worker. I sent my SPSS data file to her, and she added on to the file '4 Romanian'. This was because I had put my Malaysian students ethnic groups as 1, 2, 3. All the other data was added by her. We used the same 2 questionnaires, and I already had defined the variables. We helped each other, that is what co workers should do. And we merged our data into a larger sav file. Thanks.
This depends on several factors, the most important being what hypothesis you are testing. If you are, for example, developing and validating a novel questionnaire, I would opt for two separate analysis.
In my opinion it depends on the type of study that you want to conduct.
Hello Muayyad,
I would go for option 2. Differences in terms of the sample characteristics will always be used as serious limitations.
Regards, Witold
I would not merge the two datasets, but do a comparison - e.g., Cronbach's alpha coefficients, ICCs, factor loadings, etc. - to see whether there are elements to support the validity of constructs assessed between the two samples.
I suggest CFA on separate samples first and if you get satisfactory fit for both then proceed with measurement invariance across these samples. If scalar invariance is achieved then you will be able to compare latent means.
As stated, we can not be very sure of what "similar populations" means. Has the data been collected at the same time and in identical geographical population? Random samples allows that you can draw conclusions of the sample to the whole population, and this is not the case, I understand here that in merging (in this case overlapping cases of both studies in the same data set, assuming you have the same questions in both studies) you will need a variable to identify both studies. To do some kind of tests this merging may be useful, always considering that they are different samples and not the same populations. In analysis (not referring here to the practical problem of merging files), you should use option 2.
It depend on the analysis you want to do what you need. I would always merge the data. I could run the analysis for each group, but I would like to test if parameters for different groups are equal. I do not want to compare the differences, I would like to test if both (all) groups have the same parameters, In SPSS this type of analysis is not easily done, You could test it for IRT or CFA in R, LISREL or MPlus,
I would like to see that all measurement instruments have their own variable-identification to ensure ease of merging. It would help the advancement of science to be able to research for differential item functioning (DIF) in IRT models.
Combining data sets can be consuming a lot of time if items or variables are different and have to make to fit to each other.
I would prefer splitting the 2 sets of data in separate analysis. then find similarities in the result.
Thank you colleagues and scholars for adding your valuable opinions. thats help alot
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS