I am using a dichotomous outcome (yes/no DSM-IV diagnosis) as 2 dependent variables in a 2 level model. When I include the command "categorical are XX XX", I get an error message "RECIPROCAL INTERACTION PROBLEM". Without the "categorical" statement, the model runs fine. Is it necessary to label dichotomous variables as categorical in multilevel modeling?
Are you turning one yes/no variable into two variables? It sounds that way but maybe I misunderstand. If so, this is not necessary. Use only one variable. This variable should be placed on the CATEGORICAL list.
Thank you for your response. I will list the 2 variables (they are two separate variables, each with a possibility of 0 for does not meet criteria and 1 for meets diagnostic criteria) as categorical. Now, is there anyway to amend the problem of "RECIPROCAL INTERACTION" with my data - or and I just not able to run this model? Thank you.
Is there a solution to this problem? I have a model with a dichotomous independent variable, one dichotomous mediating variable, three continuous mediating variables and one dichotomous dependent variable. I have listed the dichotomous mediating and dependent variables in the CATEGORICAL list. Is there something that I should do differently so that I don't get the 'RECIPROCAL INTERACTION" message?
Tyler Burch posted on Friday, December 04, 2015 - 12:17 pm
First, thank you so much for what you do to help the discipline. I am trying to run a mediation analysis using a level 1-2-2 configuration. I run the following syntax as suggested by Preacher, although I change the Estimator to WLSMV because my outcome is dichotomous. This does not work as the Estimator defaults to ML. I am wondering if there is a way to run a mediation analysis using a 1-2-2 configuration with a dichotomous outcome. Thanks for your help! ANALYSIS: TYPE IS TWOLEVEL RANDOM; Estimator = WLSMV; MODEL: ! model specification follows %WITHIN% ! Model for Within effects follows aWFCIND; ! estimate Level-1 (residual) variance for x %BETWEEN% ! Model for Between effects follows bFAMSAT SEPCDmx; ! estimate Level-2 (residual) variances for m and y bFAMSAT ON aWFCIND(a); ! regress m on x, call the slope "a" SEPCDmx ON bFAMSAT(b); ! regress y on m, call the slope "b" SEPCDmx ON aWFCIND; ! regress y on x MODEL CONSTRAINT: ! section for computing indirect effect NEW(indb); ! name the indirect effect indb=a*b; ! compute the Between indirect effect OUTPUT: TECH1 TECH8 CINTERVAL; ! request parameter specifications, starting values, ! optimization history, and confidence intervals for all effects
Hello, when I model a level-1 outcome (lifetime drinker, coded 0 or 1, by wave) in a 2-level MSEM (time points within persons), Mplus states that there is no variance for the dependent variable at level 1; it cannot be modeled. Why is this, even though this yes/no variable is time-specific?
OK, thank you. I understand that the mean of a DV is a level-2 parameter, and given the complete dependency of the mean and variance of a dichotomous variable, it would make sense that there is only variance for said DV on level 2.
Is the level-2 variance parameter then interpreted as a "blend" of level-1 and level-2 variance?
Or is there only level-2 variance in a stricter sense? I am running conditional models and want to make sense of the variance explained statistics. I am seeing Mplus only give the variance estimate on level 2, with some interesting changes as I add level-1 and level-2 predictors.
No, that's not how this should be understood. On level 1 there is not a free variance parameter in line with probit/logit regression with a binary DV. Or, it can be seen as fixed (at 1 for probit). On level 2 you consider the random intercept for this binary DV and the random intercept is a continuous variable connected with several free parameters: mean, variance, and regression parameters (Mplus reports the threshold instead of the mean.)
So, no, the level-2 variance is not a blend.
Our short course handouts on this give several good background readings for really understanding this.
Thanks, Dr. Muthen. Could you point me to one of the short course handouts in particular, and I will search for the background readings it provides? I saw several short course documents on the statmodel.com website pertaining to multilevel modeling, and I wonder which one I might view to see the references. Can you help?