Dear Mike and colleagues,

I am trying to test the association between social effectiveness and consumer's impulsive buying tendency(IBT), which is also hypothesised as being mediated by hedonic shopping values. The social effectiveness is latent variable that is manifested by the bif five personalities. The data set and the R script are attaced. The label names for these variables are A,C,ES,E,O,IBT,SE(hedonic shopping values),g(the latent variable).

According to the output, there are overall 6 constraints in the Smatrix. And the model warned me that: The variances of the dependent variables in 'Smatrix' should be free.

Yet, I only set two constriants in the S matrix: the variance of the mediator and indenpendent are set as 1. May I know what are the other four?

After I sent these constraints free, there is a status1 error 6. I guess this might because the model is not identified as constraints are released. But something went wrong when I check the model by the mxCheckIdentification. May I know, if it is the issuse about model unidentification, shall I keep these constraints in the model? or has to try new models?

In addition, the dependent variable in the SEM is IBT but it seem the metaSEM did not think so..may I know how the package defienes the dependent variables and indenpendent variables?

Sometimes, there are some NA in the lbound or ubound of estimations. For this data, it occurs to the estimation g_o. May I know what the NAs mean. An infinite ubound or a statistic mistake? And given there is no p-values in the random model, may I know how to know whether the estimations are statistically significant or not?

I also failed to draw the sem plot. There are no paths has been drawn among variables by the dataset I attached. I guess it may because the dependent variable and independent variables are not respectively defined by the model. May I know am I right or there are other issues in the plots synax?

Many thanks for your time.

Dear Frank,

Since you are using "diag.constraints = TRUE" in analyzing a correlation matrix, the constraints here refer to the constraints on the diagonals of the correlation matrix. The warning indicates that the variable SE is not an independent variable. The variance of SE should be free. If you free it, there is a total of 7 constraints (A,C,ES,E,O,IBT,SE).

The NA in the lbound or ubound of estimations is likely due to the nonlinear constraints. I usually prefer to use "diag.constraints = FALSE". It works better without any constraints.

When you use intervals.type = "LB", it reports the likelihood-based CI. There is neither z value nor p-value. If the CI includes 0, it is not statistically significant.

The plotting method fails because the names in the data are x1 to x7 whereas they are "A","C","ES","E","O","IBT","SE" in the plotting function.

You may refer to the attached file for the details.

Best,

Mike

Dear Mike,

Many thanks for your quick response！Your feedback is very valueable.

I am still a little confuse about the "dig.constraints" function. As it is shown in Becker09 (page 41, metaSEM, an R packages for meta-analysis... ), "(Stage 2 analysis)...Since there are "mediators" in the model, the arguement diag.constraints=TRUE must be specified". Thus, I supposed that the arguement should be always TRUE as long as there are mediators. May I know is there any guidelines to decide the TRUE or False for these arguement?

And, may I also know is the estimate on g_IBT the value of overall effect or only direct effect?

I am also not sure how the indirect values is calculated as, in page 49 (the same Becker09 example), at the top the indirect effects are reported as "-0.1072, -0.1096 and -0.2168, respectively". However, the estimations on mxAlgebras objects reported in the output are "-0.08770. -0.08955 and -0.17725" respectively. May I know how the indirect effects are calculated?

Besides, is the metaSEM allows me to add a continuous moderator in the model? Is ther any examples?

Many thanks for your time!

Regards!

Frank

Dear Frank,

The recommendation of using "dig.constraints=TRUE" was for an old version of the metaSEM. I have rewritten the tssem2() and wls() functions so that it does not matter now. You may use "dig.constraints=FALSE" regardless of the models. Thanks for pointing out the errors of the indirect effects in the Becker09 example. I will correct the figures in the text in the next version.

If you want to use continuous moderators, you may check the osmasem() function. Here are the examples and the manuscript:

https://psyarxiv.com/ce85j

https://cran.r-project.org/web/packages/metaSEM/vignettes/Examples.html#one-stage-meta-analytic-structural-equation-modeling-osmasem

Best,

Mike

Dear Mike,

Many thanks!

Regards!

Frank

Hi Mike,

Just wonder is the metaSEM allows me to know how much variance that the estimated latent variable (g) and the mediator explain respectively? Why there is no GFI reported in the fitness index?

Kind Regards!

Yours,

Frank

Hi Frank,

Since the variables are standardized, the variance explained is calculated by 1 - the error variance. GFI is not available in the metaSEM because GFI is not now popular in SEM. You may refer to the literature about its limitations.

Best,

Mike