Using weights in Quantile regression

[Karim Syed]

Dear all

I have unbalaced firm-year data from 15 countries. Observations from some countries are too many. In one case, almost 30% of the sample belongs to single country. I like to run quantile regression with weights to overcome this problem.But unfortunately qregcommands does not accept aweight option. PLease help how can I apply weights to quantile regression model.

Regards

[Clyde Schechter]

Actually, aweights are not appropriate for this situation anyway, and are inapplicable, even in theory, to quantile estimation. What you are trying to do is best accomplished using -iweights-, which are allowed in -qreg-.

Many people don't understand what aweights are for. They are used in a situation where each observation in the dataset represents a composite result for multiple units. For example, one might have a data set where each observation represents a person who has been measured multiple times, and the data in the observation gives the average of the measurements for that person. If different people have been measured different numbers of times to generate this data, then aweights would be used to account for the fact that the different people's averages have different error variance because they are based on a different number of measurements. Another situation would be where the observations correspond to organizations, each with different numbers of members (employees, whatever). If the variables being used in the analysis are the averages of observations of the members of the organization, then, again aweights would be used.

Your situation is not like that. You have different numbers of observations per country, not a single observation per country with those observations representing averages over different numbers of people. So aweights wouldn't be appropriate even if they were available and conceptually applicable to quantile estimation (which they are not).

[Karim Syed]

Many thanks, Dear Clyde Schechter for your detailed respose.