{smcl}
{* 16aug2004}{...}
{hline}
help for {hi:bssize}{right:(SJ4-3:st0073)}
{hline}

{title:Bootstrap replication size calculator}

{p 8 17 2}{cmd:bssize initial} {cmd:,} [{cmd:tau(}{it:#}{cmd:)}
{cmd:pdb(}{it:#}{cmd:)} {cmd:pctt(}{it:#}{cmd:)} ]

{p 8 17 2}{cmd:bssize refine using }{it:filename} 

{p 8 17 2}{cmd: bssize analyze} [{cmd:using} {it:filename}] {cmd:,} 
[{cmd:tau(}{it:#}{cmd:)} {cmd:pdb(}{it:#}{cmd:)} 
{cmd:append(}{it:filename}{cmd:)} ]

{p 8 17 2}{cmd: bssize cleanup}

{p 4 4 2}{it:filename} is the name of a file created by the {cmd:bootstrap}
command or the {cmd:postfile} command.


{title:Description}

{p 4 4 2}
The {cmd:bssize} suite of commands implement Andrews and Buchinsky's 
method (2000) of choosing the number of bootstrap replications for 
standard errors and percentile-{it:t} confidence intervals.  For more 
information, see Poi (2004).

{p 4 4 2}
{cmd:bssize} {cmd:initial} provides an initial estimate of the number of 
bootstrap replications needed; it corresponds to step 1 of Andrews and 
Buchinsky's three-step method.

{p 4 4 2}
{cmd:bssize} {cmd:refine} takes as input a file containing the bootstrapped 
values of the statistics and produces a refined estimate of the number 
of bootstrap replications needed; it corresponds to steps 2 and 3 of the 
three-step method.

{p 4 4 2}
{cmd:bssize} {cmd:analyze} takes as input a file containing the bootstrapped 
values of the statistics and performs postestimation analysis of the 
accuracy of the bootstrapped standard errors.  {cmd:bssize} {cmd:analyze} is
not implemented for use with percentile-{it:t} confidence intervals.

{p 4 4 2}
{cmd:bssize} {cmd:cleanup} clears all the global macros set by {cmd:bssize}
{cmd:initial} to store and pass information to {cmd:bssize} {cmd:refine}.
All such macro names begin with {cmd:BSS_}.


{title:Options}

{p 4 8 2}
{cmd:tau(}{it:#}{cmd:)} specifies the probability 1-tau with which the 
bootstrapped statistic differs by no more than {it:pdb} percent from the 
ideal bootstrap.  The default for {cmd:bssize} {cmd:initial} is 5%.  
{cmd:tau()} is required with {cmd:bssize} {cmd:analyze} if {cmd:pdb()} is not 
specified.

{p 4 8 2}
{cmd:pdb(}{it:#}{cmd:)} specifies the maximum percentage deviation between 
the bootstrapped statistic and the ideal bootstrap acceptable with 
probability 1-tau.  The default for {cmd:bssize} {cmd:initial} is 5%.  
{cmd:pdb()} is required with {cmd:bssize} {cmd:analyze} if {cmd:tau()} is not 
specified.

{p 4 8 2}
{cmd:pctt(}{it:#}{cmd:)} specifies the size, as a percentage, of the 
percentile-{it:t} confidence intervals being produced.  If {cmd:pctt()} 
is not specified, {cmd:bssize} assumes standard errors are being 
bootstrapped instead of confidence intervals.

{p 4 8 2}
{cmd:append(}{it:filename}{cmd:)} ({cmd:bssize analyze} only) indicates an
additional file that is to be appended to the {cmd:using} file before doing
the post-estimation analysis.  For example, you might perform an initial
number of bootstrap replications, save the results under one file name,
perform additional replications, and save them under a different name.
{cmd:append()} obviates the need to use Stata's {cmd:append} command to create
one dataset before calling {cmd:bssize} {cmd:analyze}.


{title:Example}

{p 8 12 2}{cmd:. bssize initial, tau(0.01) pdb(5)}

{p 8 12 2}{cmd:. sysuse auto, clear}

{p 8 12 2}{cmd:. set seed 1}

{p 8 12 2}{cmd:. bootstrap "regress mpg foreign displacement" _b, reps(1326) saving(one)}

{p 8 12 2}{cmd:. bssize refine using one}

{p 8 12 2}{cmd:. bootstrap "regress mpg foreign displacement" _b, reps(128) saving(two)}

{p 8 12 2}{cmd:. use one, clear}

{p 8 12 2}{cmd:. append using two}

{p 8 12 2}{cmd:. bstat}

{p 8 12 2}{cmd:. bssize analyze using one, append(two) pdb(2.5)}



{title:References}

{p 4 8 2}
Andrews, D. W. K. and M. Buchinsky. 2000. A three-step method for 
choosing the number of bootstrap repetitions.  {it:Econometrica} 68(1): 
23-51.

{p 4 8 2}
Poi, B. P. 2004. From the help desk: Some bootstrapping techniques.
{it:Stata Journal} 4(3): 312-328.


{title:Author}

    Brian Poi, bpoi@stata.com
    StataCorp