First Order vs Second Order Factor Structures in SEM
HTML-код
- Опубликовано: 16 сен 2024
- This lecture explains the difference between a first order (one and two factor) factorial model and a second order factorial model.
Here are some example Syntaxes for the three models mentioned in the Video:
Single First Order Latent Factor:
MODEL:
GRIT by Item1-item12;
Two First Order Latent Factor Model:
MODEL:
POE by Item1 Item4 Item6 Item9 Item10 Item12;
COI by Item2 Item3 Item5 Item7 Item8 Item11;
Second Order Factorial Model
POE by Item1 Item4 Item6 Item9 Item10 Item12;
COI by Item2 Item3 Item5 Item7 Item8 Item11;
GRIT by POE COI;
If you dont get Convergence in your model, you can do the following:
POE by Item1 Item4 Item6 Item9 Item10 Item12;
COI by Item2 Item3 Item5 Item7 Item8 Item11;
GRIT by POE* COI; ! Free constraint on POE - Let it estimate freely
GRIT@1; ! Constrain Variance to 1
The resources for this series of lectures (Slides, syntaxes, data) can be downloaded here: bit.ly/31iGzfc
Here are some example Syntaxes for the three models mentioned in the Video:
Single First Order Latent Factor:
MODEL:
GRIT by Item1-item12;
Two First Order Latent Factor Model:
MODEL:
POE by Item1 Item4 Item6 Item9 Item10 Item12;
COI by Item2 Item3 Item5 Item7 Item8 Item11;
Second Order Factorial Model
POE by Item1 Item4 Item6 Item9 Item10 Item12;
COI by Item2 Item3 Item5 Item7 Item8 Item11;
GRIT by POE COI;
If you dont get Convergence in your model, you can do the following:
POE by Item1 Item4 Item6 Item9 Item10 Item12;
COI by Item2 Item3 Item5 Item7 Item8 Item11;
GRIT by POE* COI; ! Free constraint on POE - Let it estimate freely
GRIT@1; ! Constrain Variance to 1
I just read so many papers when I should have just watched your video! Thank you!
Glad it was helpful! :) And thank you for the feedback. I just posted a video on Bifactor models if you are interested
would request you to make some more videos for budding researchers like me.
Some suggested topics are given below:
1. Introduction to MPlus software and its interface.
2. Testing and Reporting EFA, Single Level CFA, and SEM Models
3. Theoretical Understanding of Multilevel analysis of data
4. Testing and Reporting of Multilevel CFA, and SEM Models
5. Testing and Reporting of Multilevel Mediation Models (1-2-2, 2-1-2, 2-2-1, etc)
6. Testing and Reporting of Multilevel Moderation Models (2x2-2 etc)
Note: I have seen many videos on youtube but most of them do not explain how to report the results in research papers and thesis, which is the main issue of your researhers.
Thank you for explanation. It was very clear
What if first order model and second order model have the same model fit information such as CFI, TLI, RMSEA, SRMR, AIC, BIC, and Adjusted BIC. How can we decide to choose the first order or second order?
This is usually the case (especially if there's 3 first order factors). I would go with the model thats more inline with my theoretical assumptions.
Which one is a higher-order factor analysis?
Thank you so much for explaining the second-order model, but my friends told me that the second-order model can not consist of two first-order factors, it will consist of more than three? Can you give me some tips about that or give me some reference to read?
Thanks so much for a great video! I have a question regarding counting the overall sum score from a questionnaire. I've seen in some articles that an overall score was used with model 2 as well as with model 3. Is this correct to use a sum composite overall score in the case of model 2, or in this case only separate scores for each dimension should be calculated? For example, if the theory says that grit is composed of POE and COI (they both together represent the theoretical concept of grit), and those dimensions are correlated as in model 2, is it appropriate to count an overall sum score?
Hi Maja, you are asking two questions here. Ill adress the one you arent directly asking first. 1) I would be super careful when using "sum" scores for SEM models. The mean of these scores will always be zero. So Id rather use the factor scores and work with that. 2) In Model 2, you cant use the "sum" score for both factors together as you are not estimating a unidimentional or higher order factor. You'd use the scores of each factor independently (but again, refer to point 1 as the mean would be zero when estimating the latent mean). For Model 3 (in like SPSS), I would then use the mean for POI and the Mean for COI /2 as the mean for the second order factor. In Mplus, again, it will be difficult given Point 1. Hope that helps
@@LlewellynVanZyl Thank you for your reply. I don't mean counting sum scores for SEM models for latent variables. I mean interpreting scores in SPSS. Does model 2 justify using the sum score for the questionnaire if the theory informs us it's ok? Or do we first need to check if we can estimate model 3 to be able to use a total score for a scale? In other words, what are the practical implications and differences between model 2 and 3?
Thanks a lot for the video, worked perfectly here. Can you show the code to get standardized estimates?
Just add STDXY in the output command
@@LlewellynVanZyl Tks!!
I am getting convergence and discriminant validity issue in second order cfa ……I couldn’t understand the solution of it you gave, towards the end of the video….kindly guide me on how i can fix this?
What is the specific error you are getting? And see the pinned comment for the solution to get convergence for an under identified model.
Sir, which model is to be retained if the model fit for first order and second order are exactly the same?
The one thats closely alligned to your theory. So for example, the Mental Health Continuum has three components (emotional, psychological and social wellbeing). We can use all three components in isolation, but our theory tells us that they build up to a larger mental health factor. So then Id use that. But again, it has to do with what type of question you want to answer and if the second-order factor would make the explanation easier. Hope that helps!
Thanks
@@LlewellynVanZyl Can we say that both model are equally plausible?
Then what is a zero order factor?
A zero order model is a unidimensional model where all items load onto a single First order factor. Hope that helps
Hi sir how we can decide the scale is first order or 2nd order
Your theory should inform you. Theory always leads. If your theory says that Engagement is a higher order factor comprised out of vigour, dedication and absorption, and you have similar fit between a first order factor model and a second order factor model.... then you chose the second order model as this is what theory tells us.
Hello, prof.
Can i request ESEM analysis in MPLUS
Hi Soeharto. Thanks for the comment. I made a video, created a tool, and wrote a how to paper on ESEM. You can view the video tutorial here: ruclips.net/video/yK_myOYNi-o/видео.html Hope that helps!!
@@MplusforDummies finally, thank you
@@soehartosoeharto8471 hope it helps. Let me know what you think. Our tool makes ESEM estimation really easy. Hope my explanation is ok and meaningful! Looking forward to your feedback 😊
@@LlewellynVanZyl I will make an essential review after checking the software and watching the video fully.
would request you to make some more videos for budding researchers like me.
Some suggested topics are given below:
1. Introduction to MPlus software and its interface.
2. Testing and Reporting EFA, Single Level CFA, and SEM Models
3. Theoretical Understanding of Multilevel analysis of data
4. Testing and Reporting of Multilevel CFA, and SEM Models
5. Testing and Reporting of Multilevel Mediation Models (1-2-2, 2-1-2, 2-2-1, etc)
6. Testing and Reporting of Multilevel Moderation Models (2x2-2 etc)
Note: I have seen many videos on youtube but most of them do not explain how to report the results in research papers and thesis, which is the main issue of your researchers.