[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Quantile Regression With Groupwise Fixed Effects

From   "Austin Nichols" <>
Subject   Re: st: Quantile Regression With Groupwise Fixed Effects
Date   Thu, 5 Jun 2008 11:27:36 -0400

Thomas Mayock <>:
There are a number of relevant threads in the Statalist archives.  You
may also want to read
or at least p.4, which states "In least squares applications the usual
strategy would be to transform y and X to deviations from individual
means, and then compute \hat{\beta} from the transformed data. For
quantile regression this decomposition of projections isn't available
and we are required to deal directly with the full problem."

You can't specify clustering for -qreg but -bootstrap- supports the
-cluster- option, if you are mainly worried about getting standard
errors slightly more robust to clustering.  But I don't know how
-bootstrap- with the -cluster- option will perform in your data--you
may want to run simulations to see its small-sample properties in data
like yours.

On Thu, Jun 5, 2008 at 10:40 AM, Thomas Mayock <> wrote:
> Dear Statalisters
> I am attempting to estimate a quantile regression model with many (over 100) geographic fixed effects. My initial attempt at estimation was to just include group indicator dummies. The drawback to this approach is computational cost: it takes several hours for the model to run.
>    As the group fixed effects are of no interest in the study, I was thinking of performing the group-wise within transformation on the data, then implementing qreg. As I am new to quantile regression, however, I was unsure as to whether or not this is valid. Is anyone aware of other studies that have faced this issue or methodological papers that address a clustering structure in cross-sectional quantile regression?
>     Any information is greatly appreciated.
> Tom
> *
> *   For searches and help try:
> *
> *
> *
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index