Wild cluster bootstrap after maximum likelihood estimation

[Irina Simankova]

Hello everybody!

Is it possible to use wild cluster bootstrap after Poisson regression? If not, why? If yes, is there a Stata command for that? Something like "boottest treated, boot(wild)" for OLS.

[FernandoRios]

its possible to obtain an Unconstrained WildBootstrap, but i dont think there is a command for that.
You need to define the Influence functions for the Poisson coefficients, and then apply the noise to determine the Variance Covariance matrix

[Andrew Musau]

See #10 on boottest from SSC and the poisson command: https://www.statalist.org/forums/for...using-ppmlhdfe

[Irina Simankova]

See #10 on boottest from SSC and the poisson command: https://www.statalist.org/forums/for...using-ppmlhdfe

Thank you for your answer.
I am particularly interested in option boot(wild). Unfortunately, I assume #10 answer does not apply to wild cluster bootstrap.

[Andrew Musau]

For ML estimation, the only available option is score bootstrap [option -boot(score)-]. Nonlinear models do not generate well-defined residuals, so one cannot easily implement wild bootstrap. See Kline and Santos' paper: https://eml.berkeley.edu/~pkline/pap...eFinal_web.pdf.

[Irina Simankova]

its possible to obtain an Unconstrained WildBootstrap, but i dont think there is a command for that.
You need to define the Influence functions for the Poisson coefficients, and then apply the noise to determine the Variance Covariance matrix

and

its possible to obtain an Unconstrained WildBootstrap, but i dont think there is a command for that.
You need to define the Influence functions for the Poisson coefficients, and then apply the noise to determine the Variance Covariance matrix

Thank you!