My understanding of jackknifing is that it is virtually _defined_ as
leaving each value
out in turn and combining all the results to see what that indicates
about
variability. It may be that you want some other cross-validation
procedure,
and well and good, but Stata's jackknifing procedures will not I think
provide it.
Nick
[email protected]
Shahrul Mt-Isa
Many thanks for the reply JV. I feel this could work. My understanding
of
jackknifing is it works from dependent random groups technique. Creating
a
dummy cluster would be a good way to start. I will double check the
literature to make sure this does yield something meaningful.
From: jverkuilen <[email protected]>
I am not entirely sure what you want to do but you might want to
consider
jackknifing by a cluster variable. This is usually done in the context
of
jackknifing by subjects in a repeated measures context, where you throw
out
all observations within a given subject. But you could make an
artificial
cluster variable by generating a new variable that has fewer values than
the
number of subjects and is unrelated to any variable in the study. Then
cluster according to that.
I would be wary doing this actually converges to something meaningful,
though---check the literature. Also recall that jackknifing is only
appropriate when the statistic in question is continuously dependent on
the
data.
"Shahrul Mt-Isa" <[email protected]>
I am trying to do a jackknife on a large dataset. Stata's -jackknife-
command and -,vce(jackknife)- deal with this I understand. However, it
is
very time consuming as this assumes straightaway that I want number of
groups A=n number of patients. Is there a way for me to choose A, for
example A=2 instead to make computation less intensive?
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
