Message/Author 


Dear Linda or Bengt, I conducted a syntax with independend measures x1, x2, and x3, mediator m, and dependend measure y; x3 servs also as moderator of regresions of x1 and x2 on m. The syntax works and I have two questions. (1) Does syntax look plausible? (2) As I have a same missings in x3, some students are exluded from the analysis completely. Can I add covariates (e.g. using "Modell x3 on x5x7;") for estimation of missings in x3 only? I mean, whether the fit values for the regression model can be estimated without this covariates. variable: x1 x2 x3 y m; missing = all(99); usevar = x1 x2 x3 y m X1X3 X2X3; cluster = cluster; define: CENTER x3(GRANDMEAN); X1X3 = x1*x3; X2X3 = x2*x3; analysis: type = complex; model: y ON x1 m x1 x2; m ON x1 x2 x3 X1X3 X2X3; model indirect: y IND x1; y IND x2; y IND x3; output: sampstat standardized std stdy; Thanky you! 


(1) Model Indirect is not used correctly since you have moderation. See the UG for the MOD option and also our new book (see homepage). Also, y IND x3 should not be there. (2) You can bring x3 into the model by mentioning it mean or variance. The theory for this is also described in our book, Chapter 10. 


Dear Bengt, thank you! I have your new book and I looked also into manual. Concerning your first comment I changed the syntax following the following sample in the manual y MOD m z (1 1 0.1) mz x; First: I get warning and the warning was: "Variables on the righthand side of a MOD statement are not allowed to have missing data. Problem in observation 10 for variable m." Is it not allowed to have missings in any of the idependent or mediation variables and the cases with missings are not excluded bevore the analysis? Second: As I centred (but not zstandartized) the moderator, are (1, 1 and 0.1) meaningfull options in this part of the syntax? Thank you! 


I think you want to use y MOD m z (1 1 0.1) xz x; because you wanted to moderate the influence of x on m. If you have an mz interaction you can't include subjects with missing on m. 


Thank you Bengt! Unfortunately, also after excluding missings on m, the syntax did not work. Either in its simple form y MOD m xz x; no in the form y MOD m z (1 1 0.1) xz x; Warning is: "In a MOD statement in MODEL INDIRECT, the value for xz must equal the product of the values for x and m: Problem found in observation 91 where: xz(0.514103) does not equal x(1.000000) * m(4.333333) The next questions is: Might it be a problem to estimate two moderation effects one at a time? As an x variable has three stages (CG1, CG2 and EG), I separated it in the analysis in x1 and x2 (both are dummies). Further, I included two interaction term into the model z*x1 and z*x2 (see the syntax in the first post). Thus, an indirect effect of x1 on y via m (and moderated by z) and an indirect effect of x2 on y via m (and moderated by z) needed to be estimated, in order to test indirect effects of EG on y. But is it possible to estimate two indirect, moderated effects in the same model? 


With xz interaction you don't have to eliminate cases with missing on m. First check why this error occurs in your data for observation 91: xz(0.514103) does not equal x(1.000000) * m(4.333333) 


Thank you Bengt! Now the analysis of inderect effects works. I got the indirect effects of x1 (OR of x2) on y modereated by z. Is 1 and 1 a meaningfull option for range of z OR is it better to use a standard deviation? If e.g. SD of z is 0.50 it might be y MOD m z (1 1 0.5) xz x; The results show, how the inderect effects change, if the moderator decrease or increase in one or two SDs. I looked for syntax to bring x3x5 into the model by mentioning it mean or variance and estimate missings on z (moderator). The theory for this is indeed described in the book, Chapter 10. However, I did not finde a sample of the (simple) syntax. I will be very greatfull, if you can help me! 


1 SD is a reasonable choice. But the SD value should not be given in the 3rd spot. So if you have a variable with mean zero and SD=0.5, you would say z(0.5 0.5 0.1) All the runs in the book are posted on our website at http://www.statmodel.com/Mplus_Book_Tables.shtml 


Thank you Bengt. I looked at the runs and read chapter 10 in your book. If I understood well, one need to pose the xvariables in the model section only, in order to estimate missings on xvariables. However after doing this, I still received the same warning and FIT index became significantly worth. "Data set contains cases with missing on xvariables. These cases were not included in the analysis. ..." The syntax was: model: y ON x3 m x2 x1; m ON x1 x2 x3 X1X3 X2X3; x1 x2 x3; The last line was the new one that I used for icluding missings on xvariables into analysis. (I thought, it might work similar to approach 7 in Table 10.25.) 


If you create the X1X3 and X2X3 interactions in Define those are new x variables with missing if the 3 x's have missing. You can mention those 2 interaction variables on the last line as well, although it isn't clear that this is an optimal way of doing things given that this does not acknowledge that the 2 interaction variables are functions of other x variables. See discussion on page 44 of the book referring to a multiplegroup alternative. 


Thanks! The input runs and regressions are very similar to the ones with excluding missings on x. The only problem is that the "model indirect" does not work in this case. The warning is *** ERROR in MODEL INDIRECT command Variables on the righthand side of a MOD statement are not allowed to have missing data. Problem in observation 1 for variable SEFV. If I delete "MODEL INDIRECT" in the input, it runs. The input was model indirect: y MOD m x3 (1 1 0.1) X1X3 x1; 


Then just delete missings on X1 and X3. 


Thank you! 


Dear Bengt, is it possible to estimate indirect moderated effects (e.g. by using using "y MOD m z (1 1 0.1) xz x;") with missings on x? 


Sorry, I mean ... with ESTIMATION of missings on x? 


Not using MOD but if you bring the x into the model you can do what MOD does yourself in Model Constraint. 


Thank you. It works well and unstandardized estimators for indirect, direct and total effects are presented now! However, I do not see standardized estimators. Is it possible to calculate standardized ones for indirect and total effects? 


Add STANDARDIZED to the OUTPUT command. If they are available, you will get them. 

Back to top 