Jun Xu <[email protected]> asks why sometimes the observed number of
replications is less than the requested:
> Thanks for Stata Corp people (Jeff) responding to my problem. I would be
> interested in knowing when the update will be realized.
The fix to -bootstrap- is in the wings for the next ado-update.
> One more problem, why the reps(30) I specified is not consistent with the
> number under the Reps column (22) (as well as the matrix e(reps)? Also,
> though the ereturn can be modified to have the e(size) something like that,
> is that possible you could add the size return matrix too for next update?
> Really appreciate.
I'll answer the second question first. Jun Xu can save the estimation sample
size by including it on the list of expressions to bootstrap. For example,
. bootstrap "logit ..." _b size=e(N), ...
^^^^^^^^^
There are two reasons why the observed number of replications for a given
bootstrapped statistic may be different (less than) the number of requested
replications:
1. The expression cannot be calculated after the command is executed
using some of the bootstrap data sets.
2. The command failed for some of the bootstrap data sets.
I believe it is the second that Jun Xu is observing. Some of the logistic
regressions are failing because -bootstrap- is supplying -logit- with a
dependent variable that is either all 0's or all 1's.
The following code will reproduce this behavior:
. sysuse auto, clear
. keep in -25/l
. tabulate for
. set seed 1234
. bootstrap "logit for mpg" _b size=e(N), reps(30) size(20)
Using the auto data, I remove all but the last 25 observations. The
-tabulate- command shows that only 3 out of the 25 cars are Domestic. It is
probable that a bootstrap sample from this data set will result in only
Foreign cars, even more so if we only randomly sample 20 of the 25 cars. The
-noisily- option will display the output from -logit- for each bootstrap
sample. When this option is given, -bootstrap- will also output a message
indicating that is will be posting missing values when either (1) or (2) above
occurs.
A log from the above commands follow:
***** BEGIN: log
. sysuse auto, clear
(1978 Automobile Data)
. keep in -25/l
(49 observations deleted)
. tab for
Car type | Freq. Percent Cum.
------------+-----------------------------------
Domestic | 3 12.00 12.00
Foreign | 22 88.00 100.00
------------+-----------------------------------
Total | 25 100.00
. set seed 1234
. bootstrap "qui logit for mpg" _b size=e(N), reps(30) size(20) noi
bootstrap: First call to (qui logit for mpg) with data as is:
. qui logit for mpg
bootstrap header:
command: qui logit for mpg
statistics: b_mpg = _b[mpg]
b_cons = _b[_cons]
size = e(N)
30 calls to (qui logit for mpg) with bootstrap samples:
. qui logit for mpg
. qui logit for mpg
captured error running (qui logit for mpg), posting missing values
. qui logit for mpg
. qui logit for mpg
captured error running (qui logit for mpg), posting missing values
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
captured error running (qui logit for mpg), posting missing values
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
captured error running (qui logit for mpg), posting missing values
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
captured error running (qui logit for mpg), posting missing values
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
. qui logit for mpg
Bootstrap statistics Number of obs = 25
Replications = 30
------------------------------------------------------------------------------
Variable | Reps Observed Bias Std. Err. [95% Conf. Interval]
-------------+----------------------------------------------------------------
b_mpg | 25 .1438087 .1078638 .198525 -.2659268 .5535442 (N)
| .0174609 .890745 (P)
| .0174609 .5159226 (BC)
b_cons | 25 -1.241594 -2.21652 4.356838 -10.23366 7.750477 (N)
| -18.76268 1.817488 (P)
| -8.075786 1.817488 (BC)
size | 25 25 -5 0 25 25 (N)
| 20 20 (P)
| . . (BC)
------------------------------------------------------------------------------
Note: N = normal
P = percentile
BC = bias-corrected
***** END: log
--Jeff
[email protected]
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
