New version of weakiv (2.3)

[Mark Schaffer]

With thanks to Kit Baum, a new and improved version of weakiv by Finlay, Magnusson and Schaffer is now available on SSC.

weakiv provides weak-instrument-robust tests and confidence intervals/sets for instrumental variable (IV) estimation of linear, panel, probit and tobit models. The new version of weakiv includes support for dynamic panel data estimation by xtabond2 (Arellano-Bond and relatives), any number of endogenous regressors, projection-based inference and confidence sets, the CLR statistic for all models (iid, robust, cluster-robust, HAC, etc.), simulation-based p-values for the CLR stat, CUE (continuously-updated GMM) -based tests, the subset Anderson-Rubin statistic, and other improvements.

weakiv now requires Stata 11 or better. Users with Stata 10, or users who want to use the previous version of weakiv (1.0), can install weakiv10, also available from SSC.

[Greg Falls]

Mark
I need to perform a tobit analysis with one endogenous regressor using a panel data set and test for the weakness of the instrument. Unfortunately STATA 13 does not have an xtivtobit command. I have two questions. Is there a way perform an instrumental variable tobit analysis using some combination of other STATA commands ? How would I test for the weakness of the resulting instrumental variable ? I would appreciate any help someone can give me.

[Mark Schaffer]

You should probably start a new thread for this - you are asking about testing for weak instruments, whereas -weakiv- supports weak-instrument-robust estimation. That is, -weakiv- will give you coefficient estimates that are "correct" whether the instruments are weak or strong.

That said - do you have many panel units? If you have a small-ish number of panel units, and the number of observations per unit is large-ish (so you are in "fixed-N large-T" land), then you can use dummies for the fixed effects in combination with Stata's -ivtobit- (and with -weakiv-). If you don't, then you probably want to look around for a random effects IV tobit estimator (if one exists).

[Jorge Avila]

Hi Greg,

Did you find a way to use the weak-instrument-robust test with the endogenous Tobit model and random effects? I am using a the Distance function estimation with the Tobit model and re. However, I failed to replicate the weakiv test with my own codes. Please, if you could do something about it, let me know.