Thanks for your video. I have a question - conceptually, what is the difference between conducting a multilevel path analysis, and the "conventional" multilevel modelling where you call the "lmer" function rather than the "sem" function?
Cool video! Can I ask do I need to specify a level 2 model if everyone gets the same measures (and manipulations)? And also, whether it's possible to do this with fully crossed IVs (2x2)?
Thank you so much! Fantastic explanation! I have few questions which are listed below and your input would be greatly appreciated! a) I have a model where the mediation happens at level 1 instead of level 2. So in that case, would I just adapt the code to specify indirect effects at level 1 instead of level 2?, b) is there way to vary both slopes and interecepts randomly, c) and is there any discussion about mean centring? Which procedures are recommended? E.g. group vs cluster mean centering. Thanks again!
Hi there, thank you for your comments and for visiting! Regarding your questions: a) You should be able to use the same approach by adapting what I provided. HOWEVER, if you are only testing a 1-1-1 type of model [of more complex model at level 1], you will need to still specify some sort of model at level 2. If you go here [lavaan.ugent.be/tutorial/multilevel.html], you will see the authors suggest adding covariances among your level 1 endogenous variables at Level 2. My read on this is that this will result in a fully saturated model at Level 2 which hence should not impact the fit of your model if you are primarily focused on the Level 1 portion of the model. b) Unfortunately, lavaan (at this point) does not have the ability to model random intercepts and slopes. It can only model random intercepts. c) There is a nice discussion posting here [groups.google.com/g/lavaan/c/KLRELn4f5AU] addressing your question. It sounds like it is not necessary to perform centering, as lavaan automatically separates the within- and between- components for level 1 variables. The author suggests the only value of centering with lavaan is how it might assist you in interpretation of certain coefficients. I hope this helps!
@@grungehead12 You might also check out a posting I made this morning. It might also help you think about model specification: mikesquanthub.blogspot.com/2024/04/multilevel-path-model-using-lavaan.html Cheers!
🤯 . Have you been ESPing my thoughts in trying to SEM with R ?Thank you 💕. Also, thanks for the heads up about the “:” . I will revisit this over & over. Damn, you are next level Anaconda.
![[2023] Simple Mediation in R / RStudio using "lavaan" [EASY]](/img/n.gif)
Thanks for your video. I have a question - conceptually, what is the difference between conducting a multilevel path analysis, and the "conventional" multilevel modelling where you call the "lmer" function rather than the "sem" function?
Cool video! Can I ask do I need to specify a level 2 model if everyone gets the same measures (and manipulations)? And also, whether it's possible to do this with fully crossed IVs (2x2)?
Thank you so much! Fantastic explanation! I have few questions which are listed below and your input would be greatly appreciated!
a) I have a model where the mediation happens at level 1 instead of level 2. So in that case, would I just adapt the code to specify indirect effects at level 1 instead of level 2?,
b) is there way to vary both slopes and interecepts randomly,
c) and is there any discussion about mean centring? Which procedures are recommended? E.g. group vs cluster mean centering.
Thanks again!
Hi there, thank you for your comments and for visiting! Regarding your questions:
a) You should be able to use the same approach by adapting what I provided. HOWEVER, if you are only testing a 1-1-1 type of model [of more complex model at level 1], you will need to still specify some sort of model at level 2. If you go here [lavaan.ugent.be/tutorial/multilevel.html], you will see the authors suggest adding covariances among your level 1 endogenous variables at Level 2. My read on this is that this will result in a fully saturated model at Level 2 which hence should not impact the fit of your model if you are primarily focused on the Level 1 portion of the model.
b) Unfortunately, lavaan (at this point) does not have the ability to model random intercepts and slopes. It can only model random intercepts.
c) There is a nice discussion posting here [groups.google.com/g/lavaan/c/KLRELn4f5AU] addressing your question. It sounds like it is not necessary to perform centering, as lavaan automatically separates the within- and between- components for level 1 variables. The author suggests the only value of centering with lavaan is how it might assist you in interpretation of certain coefficients.
I hope this helps!
@@mikecrowson2462 thank you so much! This is very helpful!
@@grungehead12 You might also check out a posting I made this morning. It might also help you think about model specification: mikesquanthub.blogspot.com/2024/04/multilevel-path-model-using-lavaan.html Cheers!
@@mikecrowson2462 Thanks so much!
🤯 . Have you been ESPing my thoughts in trying to SEM with R ?Thank you 💕. Also, thanks for the heads up about the “:” . I will revisit this over & over. Damn, you are next level Anaconda.
Hi Deborah, I'm so glad you found this useful! I'm planning to add more on this topic soon too :) Have a great day!