Dear all,
I am using the Instrumental Variable approach to estimate the causal
effects of TWO endogenous variables in a Mendelian Randomization study.
As long as point estimation is concerned, I have no problem: both "ivreg"
in library "AER" and "tsls" in library "sem" do the job perfectly. The
problems begin
when I try to obtain confidence intervals for these two causal effects.
Of course, I can take the output from ivreg or tsls and compute the Wald-type
confidence intervals using a Normal approximation. But Wald-type confidence
interval are known to have poor coverage properties, and therefore I would
like to switch to more robust confidence intervals, like those provided by
inverting the Anderson-Rubin (AR) or the Conditional Ratio Likelihood
(CLR) tests.
The library "ivpack" has the command "anderson.rubin.ci" which implements AR,
and the library "ivmodel" has the command "confint.ivmodel" which
provides a rich choice of
alternative confidence intervals (OLS, Fuller, LIML, TSLS, AR, CLR).
But both assume that there is only ONE endogenous variable for which
a confidence interval for the causal effect is needed. If there are TWO, or
more, endogenous variables, they stop and give an error.
Any idea of other R libraries which provide commands overtaking this
limitation?
Any suggestion really appreciated. Thanks in advance!!
Gianfranco Lovison
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
