Message/Author 

Anonymous posted on Wednesday, October 20, 2004  9:57 am



I am interested in running a vanishing tetrad test to determine if the items on a scale are effect or causal indicators. To run this test, I need to obtain the implied covariance matrix for a model with causal indicators. However, I am unsure of how to do this in MPlus. I would definitely appreciate any guidance. Thank you! 


The RESIDUAL option of the OUTPUT command is used to request model estimated means, variances, and covariances. 

Anonymous posted on Wednesday, October 20, 2004  10:09 am



Thanks for getting back to me so quickly! I'm actually wonder how to set up the model. I have no problem specifying an effect indicators model where the paths go from the latent variable to the individual items. (It's just a CFA model.) However, I'm not clear on how to switch the direction so that the paths go from the individual items to the latent variable. 


I copied this from a former post by Bengt. "In some applications it may be more natural to let a factor be defined as being influenced by indicators, rather than influencing the indicators. An example is SES. Mplus handles this modeling by the following model statements. f by y*; f@0; f on x1@1 x2 x3; Note that f by y* (that is, freeing the factor loading for the single indicator y) is the same as y on f. f has to be defined by some y  if instead you use the dummy definition f by y@0; y on f; the estimate shows up in f by y." 

Anonymous posted on Wednesday, October 20, 2004  10:40 am



Wondeful! Thanks so much! 

ds posted on Monday, November 21, 2005  8:10 am



I am trying to run a cfa with 3 latent variables and a categorical outcome variable. As part of this model, I have one "latent" variable with 2 categorical indicators (hs grad, minority) and 1 continuous indicator (age). I want to use these 3 indicators as effect indicators of background risk. However, I am not clear how to accomplish this using Mplus (again, switching the arrows). Any help would be appreciated! 


f BY y*; ! defining the factor; same as ! regressing y on f f@0; f ON age@1 hsgrad minority; 

ds posted on Saturday, November 26, 2005  11:01 am



Thank you! In terms of the first line f BY y*; For the CFA, should I also be including all latent variables included where the above says "y*"? DS 


f BY is used only to define F as a factor name. The factor itself is specified using the f ON statment. 

Al Farrell posted on Tuesday, January 03, 2006  1:59 pm



I'm trying to run a causal indicator model using the f by Y* syntax, but keep getting an error message: *** ERROR in Model command Unknown variable: Y Does Y have to be an actual variable in the data set? Thank you! 


Yes, the variables have to be in the data set and part of the NAMES or USEVARIABLES options. 


I want to do a MIMIC MODEL (30 items as causal indicators, and 2 items as effect indicators) , but in the end, I want to calculate the latent score variable, how can I obtain this? thanks 


You can obtain factor scores by using the FSCORES option of the SAVEDATA command. Example 5.8 shows a MIMIC model. 


I am trying to run the following causal indicator model with four continuous indicators (api_kop, api_tijd, api_mee and api_and) for the latent variable action_i, and with smerenbn as the only dependent variable related to action_i. action_i BY smerenbn@0; smerenbn ON action_i; action_i@0; action_i ON api_kop@1 api_tijd api_mee api_and; Is this correct? Although this model above runs without problems, I run into trouble when adding any other paths with smerenbn as dependent variable (which is necessary for my model). I keep getting error messages: THE DEGREES OF FREEDOM FOR THIS MODEL ARE NEGATIVE. THE MODEL IS NOT IDENTIFIED. NO CHISQUARE TEST IS AVAILABLE. CHECK YOUR MODEL. Is it not possible to use the y variable (smerenbn) in any other instance than to define the factor? 


You do not need to put a latent variable behind an observed variable in Mplus. Just say: smerenbn ON api_kop@1 api_tijd api_mee api_and; If you want to put a latent variable behind an observed variable, the correct format is: action_i BY smerenbn@1; smerenbn@0; 


Looks like you are doing formative indicator analysis which is fine as you have specified it, although all you have to say is action_i BY; to define the action_i factor (nothing needs to follow BY). There is no problem in letting your y (smerenbn) be a dependent variable for other predictors as well. If you can't see the problem, send your input, output, data, and license number to support@statmodel.com. 


I have a latent variable model with one latent variable being determined by 3 causal (or formative) indicators. There are no effect indicators in this model. I also have this latent variable determining 3 endogenous variables. For the purpose of scaling, can I set any of these 6 paths to 1? Also, in addition to this, do I have to constrain the error variance of this latent variable in any way? 


Yes, you can fix any path to one. The error variance must be fixed at zero. 

Jon H posted on Friday, May 01, 2015  12:41 pm



Not sure where to post thisI was wondering if there were any plans to include the vanishing tetrad test with future versions of Mplus (rather than having to grab the output and stick it in Hipp's SAS macro). If not, it's something I'd love to have. 


In which contexts do you find that useful? 

Jon H posted on Friday, May 01, 2015  2:26 pm



I should be able to distinguish between casual and effect indicators using theory (Blalock's 'mental experiments'). But I've found that I'm increasingly getting into situations where it's hard to know whether some variables are causing something or the effect of it. I think I know (based on background knowledge and trying out different model structures), but it would be nice to test for it in an easy way. The best example from my own work is using variables related to job quality; I have several jobrelated variables (e.g., having to with work autonomy, or pay) and I think I have a pretty good sense of which ones are causing a jobrelated outcome (e.g., work autonomy) and which ones are reflections of the jobrelated outcome. But it would be nice to be able to test it to make sure. To be fair, I can test it, using the suggestions in this thread, it just requires using multiple programs and takes a fair amount of time. It would be great if I could do it all within Mplus. 


Ok; we'll add it to our list. 

Back to top 