Number of Observation in Quantile Regression

[Attaullah Shah]

Hi All
I have a data set that has 147000 observation. When I use qunatile regression using qreg2 from SSC, I get the same number of observation ( i.e. 147000) in each regression of quantile of .2, .4, .6, .8. I am not able to understand this as in my opinion the number of observation in each quantile should be 147000*.2=29400. Your comments will be appreciated. The regression code is

qreg2 LDTA CR TIDINEW lDCPSF lmkt IND1- IND60 y1- y16, quantile(.2 )
qreg2 LDTA CR TIDINEW lDCPSF lmkt PROF SZ1 COLAT3 RD LIQ TAX CAPEX IND1- IND60 y1- y16, quantile(.4 )
qreg2 LDTA CR TIDINEW lDCPSF lmkt PROF SZ1 COLAT3 RD LIQ TAX CAPEX IND1- IND60 y1- y16, quantile(.6 )
qreg2 LDTA CR TIDINEW lDCPSF lmkt PROF SZ1 COLAT3 RD LIQ TAX CAPEX IND1- IND60 y1- y16, quantile(.8 )

[Nick Cox]

Not so. The need is to find a quantile by finding a given fraction below and the complementary fraction above, given predictors. (It often needs a lot of information to do that well, especially when, as in your problem, you are using a large number of predictors.)

Consider this similar question: how many observations are used in finding a median in a univariate distribution? Answer: all of them.