Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: Normally distributed error term & testing normality of residuals
From
"JVerkuilen (Gmail)" <[email protected]>
To
[email protected]
Subject
Re: st: Normally distributed error term & testing normality of residuals
Date
Sat, 13 Oct 2012 11:40:13 -0400
On Sat, Oct 13, 2012 at 10:26 AM, Ebru Ozturk <[email protected]> wrote:
> Thank you. So, it leaves me with only one choice to follow Cameron and Trivedi. But what would be if unobserved data censored at zero?
I wouldn't say that it's your only choice, just that model assessment
in the context where you have a partially observed variable is tricky
and there isn't a definitively supported diagnostic.
One refinement over simply looking at predicted values for the fully
observed data would be, if your dataset is large enough, to generate
some random subsets and see how well the model fit on the other cases
predicts, i.e., do cross-validation or some kind of bootstrapping.
Nick Longford has a nice paper in JRSS-A on using parametric
bootstrapping in assessing multilevel models and the basic ideas could
probably be adapted here. Essentially you simulate coefficients from
the asymptotic distribution of the estimates and then simulate linear
predictors. Finally you'd simulate observations according to your
censoring rule. These are more variable than the predicted values. You
might be able to trick -mi- into doing that for you without a lot of
programming (but not in Stata 10).
Longford, N. T. Simulation-based diagnostics in random coefficient
models. Journal of the Royal Statistical Society Ser. A 164, 259–273,
2001.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/