Dear Dr. Geiser Thank you very much for this informative video (and all of your other videos - they have helped me a lot in my research thus far, I even managed to (hopefully) fit my first LST model not too long ago). In regards to the topic of this video I would like to pose a question. I have been doing a lot of EFAs and fairly recently moved on to CFAs, as I have been working for a pair of researchers who prefer the exploratory approach to factor analysis. As such, I cannot help but wonder, if there really is no additional benefit in conducting a CFA after EFA. One of their main arguments for relying on the EFA is to remain inductive/abductive to new findings based on the data. Within this framework, the CFA may seem like too much of a deductive approach, in which one "merely" tests whether one's preconceptions about the dimensionality of the factor can hold statistically. In other words, is it not possible, that one may be blinded to alternative coneptualizations of dimensionality if solely relying on the CFA? (Of course, dimensionality should always be judged according to the theoretical conceptions, thus the abductive approach). It should be mentioned for context, that their work relies mostly on self-constructed scales (to some degree inspired by prevalidated scales, but always contextualized to the particular field) and measuring constructions that I (based on my recent LST) now know to be only moderately trait-like, with varying degrees of occassion-specificity (if I managed to correctly compute that coefficient, of course). Once again, thank you very much for your work. Kind regards, Casper
Hi Casper, I think it is fine to use only EFA is this seems to be more conducive to the research questions and goals and appears to deliver more detailed insights than CFA. My main point in this video is that there is no benefit in running a CFA after an EFA on the same sample/set of data (or a very similar set of data drawn from the same population). The reason is that there would not be any new or interesting conclusions or insights. You cannot confirm something that you detected based on the same set of information. A random split of a sample should not yield anything different beyond random sampling error. I hope this helps. Good luck with your LST analyses! Best, Christian Geiser
Thank you for sharing all this amazing content Dr. Geiser, if someone is interested in learning about applying the sem framework is Quant psych the best curriculum, generally speaking?
Yes, quantitative psychology and/or educational measurement PhD programs typically offer good coverage of basic and advanced structural equation modeling techniques. Best, Christian Geiser
Is it acceptable to perform EFA and CFA using the same samples with measurements taken in different years? For instance, EFA using 2018 data then CFA using 2019 data….. Thank you!
Sure, why not. Or just use longitudinal CFA for both time points combined. That way, you could also test for measurement invariance across time. See: ruclips.net/video/ZI7E3hL9f2c/видео.html ruclips.net/video/WRrZYFBPxDY/видео.html Best, Christian Geiser
Insightful video, Dr. Geiser. Please, in the event where a validated scale (in a different context) is be validated in another context ( with cultural difference), should CFA still be done? Or EFA should come first?
In cases where you already have a theory about the factor structure from previous studies, I would prefer CFA in this context unless there is reason to believe that the factor structure may be totally different across cultures. Best, Christian Geiser
Dr Geiser, thanks for making excellent informative videos. May I extend the question posed by @ericasare8864 a little further. I have been working to validate the English version of one of the scales, developed and standardized in US, in Indian adolescents. Please also note that I am using the English version only and not the translated (local language) version of this scale. Although the original scale as developed by authors has 36 items and has a popular six factor structure, but is still marred by inconsistencies. Some studies do provide evidence for the original six factor structure but many studies also report the five-factor model as exhibiting better model fit. Further studies also report cross loading problem with certain items, some items loading onto different factors than in the original scale, some studies retaining less than 36 items and of Couse studies also differ wrt the magnitude of factor loadings. Do you think the above facts provide reasonable ground to try EFA first and then go for the CFA? Or should one go for the CFA directly?
@@naseerbhat7366 I would try CFA first since specific competing factor structures have already been proposed/examined. If none of the previous factor structures fits, then I would try EFA. Best, Christian Geiser
I would start with CFA unless you have no idea what the factor structure may look like. EFA is for exploring factor structures. In the case of a previously validated scale, you are not exploring. However, this could change if the previously determined factor structure does not hold in the other context/culture. In that case, it may be useful to follow up with an EFA to examine why the hypothesized CFA model failed. Best, Christian Geiser
I'd say it depends on whether you have clear hypotheses about the factor structure. If yes, then I'd start with CFA. If not, EFA may be a better starting point. Best, Christian Geiser
Your videos are very informative. What is your take on using Lorenzo-Seva’s Solomon method, which a sample equally in terms of variance using KMO. Sometimes random sampling can artificially inflate variance in one sample and deflate it in the other. Cheers, 🇦🇺🦘
Thank you for making this excellent video on this topic!
What you have said is very insightful. Thank you as always
Dear Dr. Geiser
Thank you very much for this informative video (and all of your other videos - they have helped me a lot in my research thus far, I even managed to (hopefully) fit my first LST model not too long ago).
In regards to the topic of this video I would like to pose a question. I have been doing a lot of EFAs and fairly recently moved on to CFAs, as I have been working for a pair of researchers who prefer the exploratory approach to factor analysis. As such, I cannot help but wonder, if there really is no additional benefit in conducting a CFA after EFA. One of their main arguments for relying on the EFA is to remain inductive/abductive to new findings based on the data. Within this framework, the CFA may seem like too much of a deductive approach, in which one "merely" tests whether one's preconceptions about the dimensionality of the factor can hold statistically.
In other words, is it not possible, that one may be blinded to alternative coneptualizations of dimensionality if solely relying on the CFA? (Of course, dimensionality should always be judged according to the theoretical conceptions, thus the abductive approach).
It should be mentioned for context, that their work relies mostly on self-constructed scales (to some degree inspired by prevalidated scales, but always contextualized to the particular field) and measuring constructions that I (based on my recent LST) now know to be only moderately trait-like, with varying degrees of occassion-specificity (if I managed to correctly compute that coefficient, of course).
Once again, thank you very much for your work.
Kind regards,
Casper
Hi Casper,
I think it is fine to use only EFA is this seems to be more conducive to the research questions and goals and appears to deliver more detailed insights than CFA. My main point in this video is that there is no benefit in running a CFA after an EFA on the same sample/set of data (or a very similar set of data drawn from the same population). The reason is that there would not be any new or interesting conclusions or insights. You cannot confirm something that you detected based on the same set of information. A random split of a sample should not yield anything different beyond random sampling error.
I hope this helps. Good luck with your LST analyses!
Best,
Christian Geiser
Thank you for sharing all this amazing content Dr. Geiser, if someone is interested in learning about applying the sem framework is Quant psych the best curriculum, generally speaking?
Yes, quantitative psychology and/or educational measurement PhD programs typically offer good coverage of basic and advanced structural equation modeling techniques.
Best, Christian Geiser
thanks
Is it acceptable to perform EFA and CFA using the same samples with measurements taken in different years? For instance, EFA using 2018 data then CFA using 2019 data….. Thank you!
Sure, why not. Or just use longitudinal CFA for both time points combined. That way, you could also test for measurement invariance across time. See:
ruclips.net/video/ZI7E3hL9f2c/видео.html
ruclips.net/video/WRrZYFBPxDY/видео.html
Best, Christian Geiser
Insightful video, Dr. Geiser.
Please, in the event where a validated scale (in a different context) is be validated in another context ( with cultural difference), should CFA still be done? Or EFA should come first?
In cases where you already have a theory about the factor structure from previous studies, I would prefer CFA in this context unless there is reason to believe that the factor structure may be totally different across cultures.
Best, Christian Geiser
Thank you.
Dr Geiser, thanks for making excellent informative videos. May I extend the question posed by @ericasare8864 a little further. I have been working to validate the English version of one of the scales, developed and standardized in US, in Indian adolescents. Please also note that I am using the English version only and not the translated (local language) version of this scale. Although the original scale as developed by authors has 36 items and has a popular six factor structure, but is still marred by inconsistencies. Some studies do provide evidence for the original six factor structure but many studies also report the five-factor model as exhibiting better model fit. Further studies also report cross loading problem with certain items, some items loading onto different factors than in the original scale, some studies retaining less than 36 items and of Couse studies also differ wrt the magnitude of factor loadings. Do you think the above facts provide reasonable ground to try EFA first and then go for the CFA? Or should one go for the CFA directly?
@@naseerbhat7366 I would try CFA first since specific competing factor structures have already been proposed/examined. If none of the previous factor structures fits, then I would try EFA.
Best, Christian Geiser
I would start with CFA unless you have no idea what the factor structure may look like. EFA is for exploring factor structures. In the case of a previously validated scale, you are not exploring. However, this could change if the previously determined factor structure does not hold in the other context/culture. In that case, it may be useful to follow up with an EFA to examine why the hypothesized CFA model failed.
Best, Christian Geiser
Sir , please answer ..if items of questionnaire are taken from different papers for the "same construct" . What should we run efa or cfa
I'd say it depends on whether you have clear hypotheses about the factor structure. If yes, then I'd start with CFA. If not, EFA may be a better starting point.
Best, Christian Geiser
Thankyou sir @@QuantFish
Your videos are very informative. What is your take on using Lorenzo-Seva’s Solomon method, which a sample equally in terms of variance using KMO. Sometimes random sampling can artificially inflate variance in one sample and deflate it in the other. Cheers, 🇦🇺🦘