Fitting a regression discontinuity model in R with covariates using rdrobust package

Question

How do I specify covariates for regression discontinuity model using 'rdrobust' package in R?

The problem is with 'covs' parameter in the code below. It is supposed to allow specifying additional covariates for the model. I have tried providing a vector of vectors from the dataset (as in the code below) and a vector of variable names but neither seem to work.

fm = rdrobust(datasets$CONTINUANCE,
              dataset$T,
              p=3,
              covs=c(dataset$CONTENT_LENGTH, dataset$CONTENT_SNIPPET),
              kernel='uniform',
              weights=dataset$USERS,
              bwselect='msetwo',
              vce='nn',
              nnmatch=7,
              level=95)

As a result, I get the following error:

Error in if (c <= x_min | c >= x_max) { : 
  missing value where TRUE/FALSE needed
Calls: rdrobust
Execution halted

score 4 · Answer 1 · answered Aug 21 '19 at 21:11

The documentation for this package is unhelpful, unfortunately. You specify the covariates like a formula (var1 + var2) rather than as a vector (c(var1, var2)), so something like this should work:

fm = rdrobust(datasets$CONTINUANCE,
              dataset$T,
              p=3,
              covs=dataset$CONTENT_LENGTH + dataset$CONTENT_SNIPPET,
              kernel='uniform',
              weights=dataset$USERS,
              bwselect='msetwo',
              vce='nn',
              nnmatch=7,
              level=95)

Thank you! It works now. Maybe I drop an email to the authors of the package to update the documentation. — Dr. Kaufman, Aug 22 '19 at 13:33

Sebastian Calonico · Accepted Answer · 2019-08-23T18:05:55.250

1

In the current version of rdrobust (0.99.4) the covariates should be incorporated as a combined object via cbind. In the example above:

fm = rdrobust(datasets$CONTINUANCE,
              dataset$T,
              p=3,
              covs=cbind(dataset$CONTENT_LENGTH, dataset$CONTENT_SNIPPET),
              kernel='uniform',
              weights=dataset$USERS,
              bwselect='msetwo',
              vce='nn',
              nnmatch=7,
              level=95)

We will keep this in mind for the next version coming soon. Thanks, Sebastian.

edited Aug 23 '19 at 18:05

answered Aug 22 '19 at 17:26

Sebastian Calonico

101
4

Thank you Sebastian - very helpful! – Dr. Kaufman Aug 22 '19 at 17:34

Fitting a regression discontinuity model in R with covariates using rdrobust package

2 Answers2

Linked