[Ainhoa Oses]

Just to add some more detail to the previous post:

In particular, I'm using PISA data, where student data (weights are named w_fstuwt) is nested within schools (weights are named as w_schgrnrabtwt). So I need to do the same as below, but I want this analysis to be done for different quantiles of my dependent variable --math scores-- keeping the student and school weights in the model and including school fixed effects.

  mixed  math_scores X  [pw=w_fstuwt]   || cntschid:, pw(w_schgrnrabwt) pwscale(size)

[Ainhoa Oses]

Dear all,

In relation to posts #90 and #91, I'd highly appreciate it if someone could help me. Please let me know if further clarification is needed. Basically, all I need is to be able to run quantile regressions with fixed effects and with data that contains pweights.

Many thanks.

[valentin schubert]

Hi everyone,

For one of my projects I would like to run a quantile regression with several fixed effects. Basically I have a loan-level data set: each row is a unique observation, so a single bank is observed several time per time period and over twenty years. I would like to run a quantile regression of loan amount on a bunch of loan specific and bank specific independent variables. On top of it, I want to saturate the regression by including: bank, time, and region specific fixed effects.

I have read through the thread here (and others), but could not find a question or answer regarding the use of several fixed effects using the command xtqreg.
Could you help me out? Can this be done? Or is there another command I could use?

Thank you for all your help

Edit: In another thread I read that the code had been adapted in 2019 to allow for more than one fixed effects

[FernandoRios]

Hi Valentin,
I implemented an extension to xtqreg named mmqreg (also available from SSC). It may help you do what you are aiming for, using multiple fixed effects.
Best Regards,
Fernando

[Mingyu Qi]

Hello everyone,

I have been using -xtqreg to run quantile regression on a data set with 5 million records and 15000 clusters. I found the results a bit strange since the p-values are something like 0.965,. I have never seen p-value as large as this before, which makes me wonder if I made any mistake when I ran the regression. Does anyone have run into same trouble before?

This is my code: xtqreg readmission length age sex race drg, i(provider_num) q(0.5)

Thank you in advance!

[Joao Santos Silva]

Dear Mingyu Qi,

Please post your results and the results obtained with xtqreg so that we can comment on them. BTW, is your dependent variable continuous?

Best wishes,

Joao

[OWUSU ANSAH]

Hello, I studying the complementary role of finance on the FDI- growth nexus. I included an interaction between FDI and finance to capture this interaction effects. My quetions are
1. Does isit make sense to include interaction term in the XTQREG comand?
2.Can i include the lag of the dependent variables to the right hand side variables?
3. How different is this estimator from the Galvao one?
This is my command
xtqreg gdppc gdppc_lag FIN FDI FIN_FDI i.year, quantile(.05)
where FIN_FDI IS THE INTERACTION OF FDI AND FINANCE

[Joao Santos Silva]

Dear OWUSU ANSAH,

Interactions ate not a problem. In one of the examples in the paper we include a lagged dependent variable but you have to be careful in that case; please read the part of the paper with where we discuss that example. The estimator is very different from Galvao's. If your interest is on the coefficient of the lagged dependent variable, Galvao's estimator is likely to be preferable; otherwise the MM-QR implemented in xtqreg is probably a better bet.

Best wishes,

Joao

[OWUSU ANSAH]

Thankyou Joao,
Is the Galvao, approach estimated with the same XTQREG?

[Joao Santos Silva]

No, I do not think there is a command for Galvao's estimator.

[Mingyu Qi]

Dear Joao,

Thank you very much for your kind response! Please find below results obtained from xtqreg. The dependent variable is actually binary (0/1), but I assume technically I can treat it as continuous?

[Joao Santos Silva]

Dear Mingyu Qi,

Most quantile regression estimators only work for reasonably continuous dependent variables. The exception is the quantile regression estimator for count data implemented in qcount.

Best wishes,

Joao

[Mingyu Qi]

Dear @Joao Santos Silv,

Thank you for your advice! I am not sure if I can use qcount since I want to include a large number of fixed-effects in the model. Are you suggesting that the binary dependent variable is the reason why my outputs look so weird?

Thanks,
Mingyu

[Joao Santos Silva]

Dear Mingyu Qi,

Do not use qcount; it is not for binary data. Yes, the reason your results are strange is that the estimator is not valid when the dependent variable is binary (or discrete).

Best wishes,

Joao

[Mingyu Qi]

Dear @Joao Santos Silva,

Thank you for your help!

Best,
Mingyu