Dear Joel, Thanks for the great video! 20:10 When you say a sperate test, would it mean a separate CFA for each construct individually? I have come across issues regarding fit indices for my measurement model (7 constructs, 40 items, n=105, relatively non-normal data). When conducting a CFA on each construct individually (there is a term called individual CFA), all constructs show a good model fit, while putting all variables and constructs simultaneously (called pooled CFA) turns into a significant poor fit. Would it be a good idea to conclude that a poor fit through overall measurement model (pooled CFA) can be associated with a small sample size?, As far as I understand, CFA normally requires 5-20 samples per variable. In my case, 40* (5-20)= 200 to 800 sample size is required to get a stable estimate. May I ask your opinion regarding individual CFA? Do you think I can use the results from individual CFA to claim a good fit as long as I present the overall measurement model fit indices and justify it with a small sample size? Also, I see some papers showing restates of CFA for each construct. Does it mean they run an individual CFA? Thank you so much for your time in advance.
Typically you are not running a CFA for a single construct. A CFA is usually performed for all constructs in the proposed model. Run them all at once. If you are running them all at once and have model fit issues, this is probably because of multicollinearity. You need to assess if you have discriminant and convergent validity. If not, then you need to consider dropping items until you have achieved internal validity or worse case recollect with better measures. Hope this helps.
No need to include dichotomous variables in a CFA. A CFA is trying to assess how well your measures capture the unobservable variable. With dichotomous variables there is no need to be included in a CFA
Dear Joel,
Thanks for the great video!
20:10
When you say a sperate test, would it mean a separate CFA for each construct individually?
I have come across issues regarding fit indices for my measurement model (7 constructs, 40 items, n=105, relatively non-normal data).
When conducting a CFA on each construct individually (there is a term called individual CFA), all constructs show a good model fit, while putting all variables and constructs simultaneously (called pooled CFA) turns into a significant poor fit.
Would it be a good idea to conclude that a poor fit through overall measurement model (pooled CFA) can be associated with a small sample size?,
As far as I understand, CFA normally requires 5-20 samples per variable. In my case, 40* (5-20)= 200 to 800 sample size is required to get a stable estimate.
May I ask your opinion regarding individual CFA?
Do you think I can use the results from individual CFA to claim a good fit as long as I present the overall measurement model fit indices and justify it with a small sample size?
Also, I see some papers showing restates of CFA for each construct. Does it mean they run an individual CFA?
Thank you so much for your time in advance.
Typically you are not running a CFA for a single construct. A CFA is usually performed for all constructs in the proposed model. Run them all at once. If you are running them all at once and have model fit issues, this is probably because of multicollinearity. You need to assess if you have discriminant and convergent validity. If not, then you need to consider dropping items until you have achieved internal validity or worse case recollect with better measures. Hope this helps.
can I perform CFA for dichotomous varibles?
No need to include dichotomous variables in a CFA. A CFA is trying to assess how well your measures capture the unobservable variable. With dichotomous variables there is no need to be included in a CFA
Great new video!