[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Bootstrap after xtregar

From   Helene Ehrhart <>
Subject   Re: st: Bootstrap after xtregar
Date   Thu, 19 Mar 2009 23:12:10 +0100

Thank you Marteen for this useful precision.
So I won't use this method to do my estimations and I will post another thread to find how I could then estimate in 2SLS with random effects correcting for autocorrelation.

Best regards,

Hélène Ehrhart.

Quoting Maarten buis <>:

--- On Thu, 19/3/09, Helene Ehrhart wrote:
Since I saw that it is possible to use bootstrap after
xtregar, I  tried again with my database changing options.
In fact with the default values for reps and size, it
tells me that I don't have enough observations.

I have 255 observations so the bootstrap run when I specify

bs, reps(250) size(40): xtregar equation, re

Since it then use a sample of only 40 observations, I am
not sure that the results will be very reliable.

You should not use the -size()- option. The idea behind the
bootstrap is that you sample with replacement N out of N
observations. In your case you get an estimated se if you
had a sample of 40 instead of 255, i.e. you are underestimating
the precision of your estimate.

However, I don't think the bootstrap facilities in Stata are
appropriate for your type of problem. You have highly structured
data: there is a clustering in groups and there is a time
dimension, both of which should be respected in the way the
bootstrap samples are drawn, while the standard bootstrap just
assumes a simple random sample. In particular maintaining the
autocorelation structure in your bootstrap samples is going
to be hard (if not impossible).

Sorry for not being more chearful,

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen

*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2021 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index