Hi,
I am trying to use SAS to fit an instrumental variable analysis on a binary outcome. From what I could find online, this can be done using the 2SLS estimator in proc syslin?
The data are as follows: death (outcome, yes/no), gene (exposure, class variable), which regulates "normalized" (mediator, continuous variable). The value of "normalized" is also affected by cohort, female, trg_i, and crp_log. What I want to know is:
1. If "normalized" (as an instrumental variable) is a causal risk factor for death.
2. If there is evidence for bias when using "normalized" as an observational variable vs. as an instrumental variable (using the Durbin-Wu-Hausman test) on risk of death.
Here is my code:
proc syslin data=causalmed 2sls;
endogenous normalized;
instruments gene cohort female trg_i crp_log;
model normalized = gene cohort female trg_i crp_log;
model death = normalized cohort female trg_i crp_log;
run;
proc model data=causalmed;
endogenous normalized;
instruments gene;
death = pnormalized*normalized + interc;
fit death / ols 2sls hausman;
quit;
Is my code correct? Any help would be much appreciated. Thanks!
