Thanks to Kit Baum, a new version of the user-written command ovbd has been put up on SSC. ovbd creates a sample of random binomial data with a user-specified correlation structure and pattern of proportions. Its primary utility is in simulation exercises.
This is a new version of the command, which was originally posted on SSC in 2007 and was written for Stata Release 9, when Stata employed the older pseudorandom number generator. The new version is written for the current Release 18 of Stata.
There are several differences from the former version in its use.
First, ovbd no longer has a dependency on the user-written command ridder. Neither does the new version call drawnorm. Both of these functionalities are now internalized in the command’s code.
The clear option is no longer mandatory.
Likewise, the verbose option behaves differently from that of the first version in that details of failures to find suitable roots are always reported. ovbd now reports the involved proportions and their sought correlation coefficient for each pair where root finding fails along with the reason for the algorithm’s failure (essentially exclusively bracketing failures). The purpose of the verbose option now is to display an informational message whenever the transformed correlation matrix is not positive definite so that the user may take the indicated precaution.
Allowed syntax (values and combinations) for varlist, n() and clear follows that allowed by drawnorm.
Last, the two component functionalities—transformation of the correlation matrix and means vector, and generation of the correlated random binomial variables from them—are now contained in separate subcommands, ovbdc and ovbdr respectively, which may be called by the user directly. This serves to increase efficiency in the typical simulation use case.
The ancillary do-file that illustrates usage of the command has been updated as well. There is also a new ancillary do-file that illustrates the use of the two component commands in the context of a simulation exercise.
This is a new version of the command, which was originally posted on SSC in 2007 and was written for Stata Release 9, when Stata employed the older pseudorandom number generator. The new version is written for the current Release 18 of Stata.
There are several differences from the former version in its use.
First, ovbd no longer has a dependency on the user-written command ridder. Neither does the new version call drawnorm. Both of these functionalities are now internalized in the command’s code.
The clear option is no longer mandatory.
Likewise, the verbose option behaves differently from that of the first version in that details of failures to find suitable roots are always reported. ovbd now reports the involved proportions and their sought correlation coefficient for each pair where root finding fails along with the reason for the algorithm’s failure (essentially exclusively bracketing failures). The purpose of the verbose option now is to display an informational message whenever the transformed correlation matrix is not positive definite so that the user may take the indicated precaution.
Allowed syntax (values and combinations) for varlist, n() and clear follows that allowed by drawnorm.
Last, the two component functionalities—transformation of the correlation matrix and means vector, and generation of the correlated random binomial variables from them—are now contained in separate subcommands, ovbdc and ovbdr respectively, which may be called by the user directly. This serves to increase efficiency in the typical simulation use case.
The ancillary do-file that illustrates usage of the command has been updated as well. There is also a new ancillary do-file that illustrates the use of the two component commands in the context of a simulation exercise.
Comment