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

Re: st: strata in bootstrap

From   "Timothy Dang" <>
Subject   Re: st: strata in bootstrap
Date   Thu, 20 Mar 2008 17:15:28 -0700

Hello Austin et al-

Thanks for the two good bits of advice. I am using -cluster- now (it's
where I started, but then talked myself out of it, silly me). To deal
with the fact that I'm looking for proportions, I'm using -bca- for
the confidence intervals:

bootstrap m=r(mean), bca reps(1000) level(90) cluster(session
Subject): ci MyFee if repeated==1
estat bootstrap, all

Not sure why I'm still using -ci- within the bootstrap, instead of
-sum-. O well.

On the face of it, this looks good to me. If you know something about
-bca- which makes it inappropriate for proportions (I'm looking at
proportions in the 3% to 20% range), I'd be grateful to hear about it.

Otherwise, I think I'm good to go. Thanks lots.


On Wed, Mar 19, 2008 at 12:02 PM, Austin Nichols
<> wrote:
> Timothy Dang <>:
>  Computing the mean across individuals is also a form of ignoring
>  heterogeneity, in some sense.  But to answer your main question, the
>  -strata- option you specified constrains -bootstrap- to draw (for each
>  replication) 12 observations with replacement within each subject
>  record. This does not seem to address your main concern that
>  observations within subject are likely highly correlated, but the
>  -cluster- option does.  The -svy:tab- command I showed is just a trick
>  to get good confidence intervals for proportions; there are others.
>  See [R] bootstrap or SJ4:312-328 for more detail on -bootstrap-:
>  On Wed, Mar 19, 2008 at 1:56 PM, Timothy Dang <> wrote:
>  > Hello Austin-
>  >
>  > To answer your question, I'm ignoring learning as being much less
>  > important than heterogeneity.
>  >
>  > I obviously need to think about this more. I haven't used any of the
>  > -svy- commands in Stata before, so I'll try to figure them out.
>  >
>  > Meanwhile, I'm still wondering about the effect of -strata-. Does it
>  > cause bootstrap to draw one (independent) sample from each stratum for
>  > a replication? Or is it something else?
> *
>  *   For searches and help try:
>  *
>  *
>  *

Timothy O'Neill Dang / Cretog8
One monkey don't stop no show.
*   For searches and help try:

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