Dear all, We are modeling in AMOS with multiple input variables (3) and multiple mediating variables (2) and two output variables (2). NOw statistically speaking, 2 versions of the model show a satisfactory fit (one is almost ideal). Now I am wondering, can one have two output variables? Most models I have seen so far have one output variable so we want to make sure that it is theoretically and statistically ok to have two output variables. Also, what about the association between the output variables. I guess that should be assumed as no covariance. Please note that I am using AMOS and all variables are Likert scale (all standardized before inputting into AMOS). I would be very grateful for any clarification, help, or discussion. Best wishes, Gulnaz
If you can have two or more output variables that are not covariating, unless you add that parameter according to your theoretical model
Gulnaz Anjum, yes, you can easily include two outcome variables in your SEM model. Further, you can correlate the residuals of the two outcome variables if you think that they are conceptually related.
Dear Gulnaz,
beyond what my predecessors correctly noted, including a) several mediators and b) outcomes add further necessities to wrap your head around the reasons for correlations among mediators and among outcomes. Including vs. non-including error correlations in contrast to inter-effects (e.g., mediator 1 --> mediator 2) is not a question of fashion but has some implications.
1) In most cases, you will have to include anything that addresses a residual correlation. Failures to do so (i.e., fixing inter-effects and error covariances) expresses the implicit assumptions that all sources of covariance are already included in the model. For instance, if you have two predictors of two outcomes and you fix error covs, you implicitly say that the reasons for the correlation of both outcomes are only these two shared predictors. This is non-reasonable in almost all imaginable cases. Errournously fixing error covariances is the no 1 reason for unnecessary misfit of the model
2) If you include instead an error covariance among mediators and among outcomes, you are on safe ground. The interpretation of this error covariance, however, expresses again an implicit assumption, namely that there are further--outside the model--common causes of both variables. However, this can be wrong when the reason (or ONE reason) for the covariance among mediators/outcomes is the effect of one mediator [outcome] on the other mediator [outcome]. The question is: what happens if you fail to include this effect and instead estimate the error covariance. Imagine you have a predictor x1 and two mediators, m1 and m2. In reality, x1 affects m1 and m2 and, m1 affects m2. Hence there is a partial mediation model with x1 affecting m2 directly and via m1. If you now fail to estimate the m1-m2 effect and instead estimate an error cov, you will get no bias, as the x1-->m2 effect will be the total effect of x1 and m2 that would result by the sum of the direct and indirect effect. The only downside is that you fail to learn that m1 is a mediator in this system (what is actually bad).
Hence, think about all relations in the model. Complex models imply complex efforts to thing hard :)
Best,
Holger
Dear Cristian Ramos-Vera, Mehmet Mehmetoglu , and Holger Steinmetz Thank you so much for your answers, they are very helpful. Holger Steinmetz I am incredibly thankful to you for your detailed answer. I will certainly make sure to include and think about all relationships, variances (both co- and error), and all fitness indicators. Thank you so much!
Best wishes,
Gulnaz
Dear Holger
Can you help me with this
I have 6 IVs, 4 Mediators and I DV in my model. All Medtrs. mediate the path between all 6 IVs and Dv. Now, to calculate the specific indirect effects I created user-defined estimand by multiplying the reg weight of the path of IV1 to M1 and M1 to DV. But I am getting the syntax error "declaration expected"
I am not able to figure out what's the mistake that I am making in creating the estimand? How do I create this estimand to find the specific indirect effect? or is there some other way too to calculate that?
Dear Gunjan,
this might be a software-related error. I never saw this error? What's the software you work with?
Best,
--Holger
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers