Very interesting. For the chapter non identification: "There appears to be one more fitted parameters than the data can support". Does this happen because your example you have too many parameters to estimate vs the number of manifest variables in the model (underidentification?) Do error parameters set to 1 count in the parameter equation?
@@mronkko the model+output from minute 10.23 (sorry I'm on my phone I can't link the time). I have seen this "not full rank and more parameters than the data can support" in several of your videos on data convergence and wonder if it's always linked to underidentification, it has appeared for me too already and using the ((k+1)*2)/2 rule didn't suggest it unless I made a calculating mistake (the model is quite large)
@@queenofcopas Yes, that always means nonidentification. But identification can be either a feature of the model or it can be empirical underidentificatiin. The error will be the same because these conditions lead to the same outcome. The parameter count rule that you are referring to is a necessary but not sufficient condition for identification. A model with positive degrees of freedom (as defined in the SEM literature, which differs from how Stata reports df for non identified models) can still be non identified.
Thank you so much for your answer. I will watch your other videos on SEM and model identification + workflow and take a better look on my past data again, I didn't remember having checked the SE in detail. Thank you so much for your work, it's very precious for us !
Very interesting. For the chapter non identification: "There appears to be one more fitted parameters than the data can support". Does this happen because your example you have too many parameters to estimate vs the number of manifest variables in the model (underidentification?) Do error parameters set to 1 count in the parameter equation?
What time in the video?
@@mronkko the model+output from minute 10.23 (sorry I'm on my phone I can't link the time). I have seen this "not full rank and more parameters than the data can support" in several of your videos on data convergence and wonder if it's always linked to underidentification, it has appeared for me too already and using the ((k+1)*2)/2 rule didn't suggest it unless I made a calculating mistake (the model is quite large)
@@queenofcopas Yes, that always means nonidentification. But identification can be either a feature of the model or it can be empirical underidentificatiin. The error will be the same because these conditions lead to the same outcome.
The parameter count rule that you are referring to is a necessary but not sufficient condition for identification. A model with positive degrees of freedom (as defined in the SEM literature, which differs from how Stata reports df for non identified models) can still be non identified.
Thank you so much for your answer. I will watch your other videos on SEM and model identification + workflow and take a better look on my past data again, I didn't remember having checked the SE in detail. Thank you so much for your work, it's very precious for us !