27 Apr
2011
27 Apr
'11
2:03 p.m.
Jeremy,
(sorry this has to be really quick as I am running out the door)
Simulation is always appropriate given the model. The only new
assumption simulation is introducing is that the # of simulations is
high enough to asymptotically approach the correct value and you can
just set the #. Once you have a vector of means and variance
covariance matrix you have already assumed any conditional
exchangeability of observations.
Brandon
On Wed, Apr 27, 2011 at 1:58 PM, Jeremy Ferwerda <ferwerda(a)mit.edu> wrote:
As an addendum: would simulation be appropriate if it
was a panel dataset? My inclination is that it would not be, since the rows aren't
interchangeable
Jeremy
________________________________________
From: gov2001-l-bounces(a)lists.fas.harvard.edu [gov2001-l-bounces(a)lists.fas.harvard.edu]
On Behalf Of Brandon Stewart [brandonmstewart(a)gmail.com]
Sent: Wednesday, April 27, 2011 1:46 PM
To: Class List for Gov 2001/E-2001
Subject: Re: [gov2001] calculating standard errors for multiple imputation
You can definitely do it using the covariance but its waaay easier to
just simulate. Simulate 1000 times from each dataset, combine the
simulations and then take the SE as normal.
Brandon
On Wed, Apr 27, 2011 at 1:43 PM, James Conran <jkconran(a)mit.edu> wrote:
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
--
gov2001 mailing list served by Harvard-MIT Data Center
List Address: gov2001(a)lists.gking.harvard.edu
Subscribe/Unsubscribe: http://lists.gking.harvard.edu/?info=gov2001
Hi all,
Can anyone tell me how to calculate appropriate multiple imputation standard errors for
the marginal effect of an independent variable that is part of an interaction term?
That is, the basic model is:
Y = beta1*x1 + beta2*x2 + beta3*x1*x2.
The standard errors for each of the three coefficients can be calculated with the usual
MI adjustment but to calculate the standard error of the marginal effect you need to deal
somehow with the covariance term between x1 and x2.
Does anyone know whether the covariances can be adjusted in a similar way to what happens
with the standard errors or do we have to calculate the standard error of the marginal
effect for each dataset/regression and then perform the SE adjustment?
Cheers,
James
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l