proc CALIS for mediation analysis - type of variables

[Emma_at_SAS]

I have a question about using proc CALIS for mediation analysis. 

independent variables: Ordinal Likert type (10 variables with 5-7 Likert levels)

Mediator: binary (1 variable)

response variable: binary (3 variables)

1- Is proc CALIS a good choice for mediation analysis or do you have a suggestion?

2- Can proc CALIS model different types of variables, in my case ordinal and binary?

3- My response variables are measured as continuous variables, but transformed to binary to match other research in the field. However, I think the type of variable does not matter for mediation analysis and I can use the continuous variable (a score between 1 and 38). Do I need to transform these scores/check for normality?

Thank you,

Mary

[SteveDenham]

Easy part first - yes CALIS is a good choice.  Other options are GLIMMIX, CAUSALMED and CAUSALGRAPH.  Of these, CAUSALMED is designed specifically for mediation analysis.

Hard part - you say you have 3 binary response variables.  Are these categorizations of the continuous score variable?  if you wish to stay with this, then CALIS is perhaps the only choice.  If you move to the continuous score as the response, then CAUSALMED would be the best choice.  See Example 37.4 Mediation Analysis by Linear Structural Equation Modeling for reasons to use CAUSALMED.  Check the other examples for code to get started.

SteveDenham