The CAUSALGRAPH procedure examines the structure of graphical causal models and suggests statistical strategies or steps that enable researchers to estimate causal effects that have valid causal interpretations.
Causal models are encoded in the form of directed acyclic graphs (DAGs) (Pearl 2009a, 2009b), which are the primary input for the procedure. In the following example, you input a causal model in the MODEL statement and inquire about the identification conditions for the causal effect of PFAS on Duration in the IDENTIFY statement:
proc causalgraph;
model "MyModel"
Age ==> Parity PFAS Education,
Parity ==> PrevBF Duration PFAS,
PrevBF ==> PFAS Duration,
PFAS ==> Duration,
Education ==> Duration Employment PFAS BMI Alcohol Smoking,
Employment ==> Duration PFAS BMI Alcohol Smoking,
BMI Alcohol Smoking ==> Duration;
identify PFAS ==> Duration;
run;
PROC CAUSALGRAPH interprets and encodes the (causal) directional relationships among the variables that you specify in the MODEL statement. It then applies the causal DAG theory to determine appropriate statistical strategies that you can use to estimate the causal effect of PFAS on Duration. For example, PROC CAUSALGRAPH outputs the valid sets of covariates that you can use for adjustments in outcome regression analysis or for matching in propensity score analysis. By using regression adjustment or propensity score matching techniques to estimate the target causal treatment effects, you can minimize the biasing effects due to the confounding covariates.
PROC CAUSALGRAPH provides a number of criteria for identifying causal treatment effects:
constructive backdoor criterion
backdoor criterion
parents-of-treatment criterion
parents-of-outcome criterion
joint ancestor criterion
instrumental variables
To identify the sets of adjustment covariates or instrumental variables, you can either request that the procedure list these sets (as in the preceding example) or test the validity of the sets that you specify. For more information about different criteria of identification, see the METHOD= option in Chapter 37, The CAUSALGRAPH Procedure.
PROC CAUSALGRAPH supports the specification of multiple models, multiple treatments, and multiple outcomes. Other main features of PROC CAUSALGRAPH include the following:
identification of joint causal effects from multiple treatments on multiple outcomes
enumeration of all the observationally testable assumptions that are encoded by a causal model
listing of causal and noncausal treatment-to-outcome proper paths
ability to import and export saved models
ODS graphics to visualize causal models
For more information about PROC CAUSALGRAPH, see Chapter 37, The CAUSALGRAPH Procedure, and Thompson (2019).