[Meng Zhang]

Reinstall Rif from ssc and use the command bsrifhdreg

Thank you very much, it works.

[Meng Zhang]

Reinstall Rif from ssc and use the command bsrifhdreg

Thank you very much, it works.

[Bagus Suyanto]

Dear FernandoRios,

Thank you for this very useful command. I tried to use qregplot after xtqreg as follows:
global ylist asinh_real_total_assets
global xlist i.post i.rdetl_sts timetreat
xtqreg $ylist $xlist, q (0.05 0.1 0.25 0.75 0.9 0.95 0.99) i(group_id)
qregplot, ols olsopt(abs(group_id) robust) label mtitles("Post Treatment" "Treatment" "Total Assets")

However, I always got these lines:

WARNING: .29666916% of the fitted values of the scale function are not positive
WARNING: .29666916% of the fitted values of the scale function are not positive
WARNING: .29666916% of the fitted values of the scale function are not positive
WARNING: .29666916% of the fitted values of the scale function are not positive
WARNING: .29666916% of the fitted values of the scale function are not positive
WARNING: .29666916% of the fitted values of the scale function are not positive
WARNING: .29666916% of the fitted values of the scale function are not positive
WARNING: .29666916% of the fitted values of the scale function are not positive
WARNING: .29666916% of the fitted values of the scale function are not positive
estimates post: matrix has missing values
If the above message is not useful, please contact me at [email protected]

Could you please provide me with the solution to resolve this issue? I can send my sample dataset to your email. Many thanks.

[FernandoRios]

Hi Bagus
This is not a problem of qregplot, but, as the message says, a "warning" from xtqreg
If you read on The original paper , you will see that the qreg identification is based on on a linear model with linear heteroskedasticity. Because of that, some times, they model will predict a variance that is negative, violating the estimation assumption.
This, however, is not always a problem, if the share of cases is small. WHile there is no rule...a 0.3% is very small.

HTH
Fernando

[Bagus Suyanto]

Hi Bagus
This is not a problem of qregplot, but, as the message says, a "warning" from xtqreg
If you read on The original paper , you will see that the qreg identification is based on on a linear model with linear heteroskedasticity. Because of that, some times, they model will predict a variance that is negative, violating the estimation assumption.
This, however, is not always a problem, if the share of cases is small. WHile there is no rule...a 0.3% is very small.

HTH
Fernando

Dear FernandoRios,

Thanks a lot for your prompt response. I highly appreciate that. It is indeed not a problem in xtqreg as you mentioned, considering the small percentage. When I run xtqreg command separately, it provided me with the result without a problem. However, it became a problem when I wanted to plot it into graph using qregplot. Any advice on how to get it resolved?

Kind regards,
Bagus

[FernandoRios]

Ohh I just realized the other problem
perhaps you can’t estimate the qreg for 99th percentile

[Bagus Suyanto]

Ohh I just realized the other problem
perhaps you can’t estimate the qreg for 99th percentile

Dear FernandoRios,

Thank you for your suggestions. I've tried to plot the quantile regressions using qregplot for a couple of days. However, I could not get the graph that fully corresponds with the xtqreg command. I have sent you an email and attached three different xtqreg specifications and results. The first one has actually satisfied my expectation since it plotted the xtqreg quite accurately using qreqplot. Nonetheless, I could not get the expected result for the rest. For the second one, I could produce a graph but it did not seem to correspond with the coefficient values, whilst for the last one I could never get the graph. I always get the similar notification as I mentioned earlier which was "estimates post: matrix has missing values".
I hope you are willing to respond my email. Thank you very much.

Kind regards,
Bagus