The CAUSALMED Procedure

You use the EVALUATE statement to specify variable levels or values for evaluating various effects. In the assignments, you can specify one or more of the following variable levels:

Each assignment is of the form

var-key=value-key

where var-key specifies a variable and value-key is either its numerical value, its character value, or a keyword (such as MEAN) that generates a value from the variable.

Because of interaction effects and nonlinear models, computation of causal mediation effects usually depend on the values or levels of the treatment, control, mediator, or covariate levels. PROC CAUSALMED assigns values or levels to these variables automatically when it performs a default mediation effect analysis. By setting these default variable levels, you obtain "overall" measures of causal mediation and related effects. For more information about how PROC CAUSALMED sets the default values of levels, see the section Evaluating Causal Mediation Effects.

However, to address your particular research questions more directly, you can provide EVALUATE statements with specific variable levels to evaluate mediation and related effects. Specifying covariate levels, the treatment level (of the treatment variable), or the control level (of the treatment variable) changes the estimates of all mediation effects and decompositions. Specifying the controlled level of the mediator variable does not change the estimates of the total effect (TE), the natural direct effect (NDE), or the natural indirect effect (NIE). But it does change the estimates of the controlled direct effect (CDE) and the reference interaction (IRF).

You can provide as many EVALUATE statements as you want. Each statement specifies an assignment scheme that defines the mediation effects and produces a summary of effects, decompositions of effects (if requested), and percentage decompositions of effects (if requested).

To distinguish the results that are produced by different EVALUATE statements, you can specify a distinct label in each EVALUATE statement. A maximum of 256 characters is allowed for each label. This label is displayed in the output tables.

For example, suppose that C1 and C2 are continuous covariates in the mediation model and you want to evaluate the mediation effects at C1=5 and C2=10. You can request that by providing the following statement:

evaluate 'Set C1=5 C2=10' C1=5 C2=10;

In this statement, the quoted string, 'Set C1=5 C2=10', labels the set of assignments for evaluating the mediation effects and is followed by the assignments, C1=5 and C2=10.

If you want to evaluate the mediation effects conditioned on a different set of covariate values, you can add another EVALUATE statement. For example,

evaluate 'Scheme 1 -- C1=5  C2=10' C1=5  C2=10;
evaluate 'Scheme 2 -- C1=10 C2=5'  C1=10 C2=5;

Meaningful labels for the EVALUATE statements are highly recommended in practice.

If you use '_Default' as the label, PROC CAUSALMED overrides the default variable levels for evaluating mediation effects. For example, the following statement generates only one set of mediation effect output tables, which replace the default tables:

evaluate '_Default'  C1=5 C2=10;   /* Overrides the default assignment scheme */

In addition to the use of fixed value assignments (such as C1=5 in the preceding examples), PROC CAUSALMED provides several ways to specify the var-key and the value-key in an assignment.

You can use the following var-keys in an assignment:

In all the preceding examples, actual variable names have served as var-keys and numerical values have served as value-keys. The following statements show examples of assignments that specify keywords for var-keys:

evaluate 'Scheme 3' _treatment=max _control=mean _mediator=last C1=mode;
evaluate 'Scheme 4' _A1=0 _A0=1 _mstar=10 C1='Boys';
evaluate 'Scheme 5' _A1=.5 _A0=-.5 _mediator=0 C1=2;

You can use the following value-keys in an assignment:

The following statements show examples of assignments that use different types of keywords for value-keys:

evaluate 'Evaluation 6' _treatment=.5(SD) _control=-0.5(SD) _mediator=min
                        C1=mode;
evaluate 'Evaluation 7' _A1=first _A0=last _mediator=mean C1='Boys';

After specifying the assignments in an EVALUATE statement, you can use one or more of the following options to control the displays of mediation effects and decompositions that are generated by the EVALUATE statement:

For more information about how variable levels are related to the interpretation of causal mediation effects, see the section Evaluating Causal Mediation Effects. For illustrations of the use of the EVALUATE statement, see Example 38.2 and Example 38.3.