[FernandoRios]

Ok so that is a good point
if you use predict you can get model predictions with mmqreg based on any equation you estimate:

ssc install frause
frause oaxaca, clear
mmqreg lnwage educ , q(50)
predict xb50, xb  eq(#3)
mmqreg lnwage educ , q(75)
predict xb75, xb  eq(#3)

However, if you use "abs" option, it will not consider that part of the estimation.

xtqreg (the original command) does allow you to estimate the quantile with fixed effects.
Check the latest update of that command, and the option "predict()"

[david yu id]

.

[david yu id]

Dear Dr. @FernandoRios:

I have a question about testing equality of coefficients across quantiles. I noticed that sometimes the estimated coefficients in both quantiles are insignificant and their 95% confidence intervals largely overlap with each other, but the "test" command will yield results showing the two estimated coefficients are significantly different. Do you know what might have caused this issue?

I run the below code for illustrative purposes (also including "sqreg" for comparison):

 
. use http://fmwww.bc.edu/RePEc/bocode/o/oaxaca.dta, clear
(Excerpt from the Swiss Labor Market Survey 1998)

. mmqreg lnwage female educ exper tenure , q(85 90)
MM-qreg Estimator
Number of obs = 1434
Quantile:  .
------------------------------------------------------------------------------
      lnwage | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
location     |
      female |  -.0841137   .0255318    -3.29   0.001    -.1341552   -.0340722
        educ |   .0847507   .0049257    17.21   0.000     .0750964     .094405
       exper |   .0110983   .0014531     7.64   0.000     .0082504    .0139463
      tenure |   .0077084   .0018578     4.15   0.000     .0040672    .0113496
       _cons |   2.213327   .0682166    32.45   0.000     2.079625    2.347029
-------------+----------------------------------------------------------------
scale        |
      female |   .0579893   .0192994     3.00   0.003      .020163    .0958155
        educ |  -.0125849   .0037233    -3.38   0.001    -.0198826   -.0052873
       exper |  -.0054269   .0010984    -4.94   0.000    -.0075796   -.0032741
      tenure |   .0031301   .0014043     2.23   0.026     .0003778    .0058825
       _cons |   .4823967   .0515647     9.36   0.000     .3813318    .5834617
-------------+----------------------------------------------------------------
qtile_85     |
      female |  -.0176419   .0253919    -0.69   0.487    -.0674091    .0321253
        educ |   .0703249   .0048994    14.35   0.000     .0607222    .0799275
       exper |   .0048776   .0014463     3.37   0.001     .0020429    .0077124
      tenure |   .0112964   .0018473     6.12   0.000     .0076759     .014917
       _cons |   2.766287   .0683739    40.46   0.000     2.632277    2.900298
-------------+----------------------------------------------------------------
qtile_90     |
      female |  -.0014663   .0283704    -0.05   0.959    -.0570713    .0541388
        educ |   .0668144    .005486    12.18   0.000      .056062    .0775668
       exper |   .0033638   .0016384     2.05   0.040     .0001525    .0065751
      tenure |   .0121696   .0020559     5.92   0.000     .0081401     .016199
       _cons |   2.900848   .0805478    36.01   0.000     2.742977    3.058719
------------------------------------------------------------------------------

. test [qtile_85]female  = [qtile_90]female

 ( 1)  [qtile_85]female - [qtile_90]female = 0

           chi2(  1) =    7.29
         Prob > chi2 =    0.0069

. lincom [qtile_85]female  - [qtile_90]female

 ( 1)  [qtile_85]female - [qtile_90]female = 0

------------------------------------------------------------------------------
      lnwage | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         (1) |  -.0161756   .0059899    -2.70   0.007    -.0279157   -.0044356
------------------------------------------------------------------------------

. 
. sqreg lnwage female educ exper tenure , q(85 90)
(fitting base model)

Bootstrap replications (20): .........10.........20 done

Simultaneous quantile regression                    Number of obs =      1,434
  bootstrap(20) SEs                                 .85 Pseudo R2 =     0.1235
                                                    .90 Pseudo R2 =     0.1002

------------------------------------------------------------------------------
             |              Bootstrap
      lnwage | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
q85          |
      female |   -.048664   .0302942    -1.61   0.108    -.1080899    .0107618
        educ |   .0689975   .0040432    17.07   0.000     .0610663    .0769287
       exper |   .0050717   .0014436     3.51   0.000     .0022398    .0079035
      tenure |   .0099555   .0021046     4.73   0.000     .0058271    .0140838
       _cons |   2.807041   .0780006    35.99   0.000     2.654033    2.960049
-------------+----------------------------------------------------------------
q90          |
      female |  -.0542809    .038456    -1.41   0.158    -.1297171    .0211553
        educ |   .0639304   .0072212     8.85   0.000     .0497652    .0780957
       exper |   .0040191    .002412     1.67   0.096    -.0007124    .0087505
      tenure |   .0077403   .0026316     2.94   0.003      .002578    .0129027
       _cons |   2.983513   .1189851    25.07   0.000     2.750109    3.216917
------------------------------------------------------------------------------

. test [q85]female  = [q90]female

 ( 1)  [q85]female - [q90]female = 0

       F(  1,  1429) =    0.06
            Prob > F =    0.8084

. lincom [q85]female  - [q90]female

 ( 1)  [q85]female - [q90]female = 0

------------------------------------------------------------------------------
      lnwage | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         (1) |   .0056169   .0231544     0.24   0.808    -.0398034    .0510371
------------------------------------------------------------------------------

.

Thank you in advance!

[FernandoRios]

One reason is that in mmqreg all qregressions are connected via the scale function
if those coefficients are significant then the difference between two qreg s will be statistically different

[david yu id]

Dear Dr. FernandoRios:

Thank you so much for the reply!

Then in this case is there a way I can "reconcile" the mmqreg equality test results to those of sqreg?

Thanks again!

[FernandoRios]

They are different models
keep in mind sqreg estimates each quantile regression model independently
mmqreg imposes functional form restrictions across all quantiles. More restrictive but efficient too

[david yu id]

Got it. Thanks!

[Emmanuel Owusu Oppong]

Thank you, prof FernandoRios and prof Joao Santos Silva, for the insightful explanations.

1. Using survey data on 1400 respondents and the dependent variable, which is the log of their monthly income and assets, an index of 20 assets. Is it ethically right to use mmqreq to estimate the quantile regression, considering the fact that there is no time variable?

mmqreg asset lnincome saving educationinyears age marriage, quantile(0.1 0.25 0.5 0.75 0.9)

mmqreg lnincome asset  saving educationinyears age marriage , quantile(0.1 0.25 0.5 0.75 0.9)

2. The output revealed that two of the variables under the scale estimation are significant. Do I need to impose a robust standard error for heteroskedasticity?

Thank you so much

[Joao Santos Silva]

Dear Emmanuel Owusu Oppong,

This looks fine to me, but if you do not have fixed effects you can simple use qreg2, right?

Best wishes,

Joao

[Emmanuel Owusu Oppong]

Thank you professor Joao Santos Silva . I do not have fixed effects so I will try qreq2 also. Once again, thank you.

[merve kaya]

[QUOTE=FernandoRios;n1667677]There is no command "yet"
but here a simple example:

[CODE]
use c:\data\oaxaca, clear
ssc install qregplot
ssc install qreg2

keep if lnwage!=.
foreach i in educ exper tenure female {
bysort isco:egen m_`i'=mean(`i')
}

Dear Fernando,
I try to estimate a model with quantile regression. My dataset consist of N:36, T:8. It is recommended to use correlated random effect with small T size.
I get an error when running the above code

. keep if lnreading!=.foreach i in lnhdi lngini lngsfr lnstudentteacherratio {bysort isco:egen m_`i'=mean(`i')}

".foreach invalid name"

Where am i doing wrong?

[FernandoRios]

Please type the commands in
not just copy paste

[merve kaya]

Dear FernandoRios,
I want to use correlated random effect with T:8, N:36 panel data.

When I adapt the code you wrote above to my own data set, I get an error:

keep if lnreading!=.foreach i in lnhdi lngsfr lngini {bysort isco:egen m_`i'=mean(`i')}

Where am I wrong with this code?

[FernandoRios]

hard to say. I cannot see your data, so i cannot run and verify your code,
however, you could use my other command for CRE

1. Install cre: net install cre, from("https://raw.githubusercontent.com/friosavila/stpackages/main")
2. Use CRE for model estimation

Example Qreg with CRE

cre, abs(gfe): qreg y x1 x2 x3, q(10)
done

HTH

[merve kaya]

Thank you Dear FernandoRios