{smcl}
{* 17feb2010}{...}
{cmd:help cgmwildboot}{right:version:  2.0.0}
{hline}

{title:Title}

{p 4 8}{cmd:cgmwildboot}  -  Linear regressions with multi-way clustered standard errors, bootstrapped
for one cluster dimension.{p_end}


{title:Syntax}

{p 4 6 2}
{cmd:cgmwildboot}
{depvar}
[{indepvars}]
{ifin}
{weight}
, cluster({varlist}) bootcluster({varname}) [null({it:numlist}) reps({it:integer}) seed({it:integer})]

{title:Options}
{p 4 6 2}
- If specified, the {opt null} option requires a list of numbers that indicate the null hypothesis for the regressors other than the constant. For example, if the command regresses {it:y} on {it:x1} and {it:x2}, the user could 
specify {opt null(0 0)} for a zero null. The user could also specify, say, {opt null(. 1)} if there is no null for {it:x1} and there is a null of 1 for {it:x2}.
{p_end}
{p 4 6 2}
- The bootstrap treats regressors differently depending on whether there is a null specified. See Cameron, Gelbach and Miller (2008) for details.
{p_end}

{title:Notes}

{p 4 6 2}
- You must specify at least one clustering variable ({opt cluster}) and one cluster to bootstrap ({opt bootcluster}).{p_end}
{p 4 6 2}
- The p-value is based on the empirical distribution of {it: t-statistics}. It is twice the fraction of t-statistics
above (below) the initial t-statistic for positive (negative) t-statistics.
{p_end}
{p 4 6 2}
- The confidence interval is based on the empirical distribution of {it: coefficients}. It reports the 5th and 95th
percentiles of the coefficients from the bootstrap iterations.
{p_end}
{p 4 6 2}
- The covariance matrix included in {cmd:e(V)} is 'fake'. It is a diagonal matrix with the elements that will
allow the display of coefficient p-values in regression tables (e.g. {cmd:estout}). It should not be used in 
any post-estimation inference.{p_end}
{p 4 6 2}
- Updates on 2013-09-13: P-values for negative (positive) coefficients forced to use lower (upper) tail of
bootstrap distribution of t-statistics; altered the 'fake' covariance matrix to handle extreme p-values of
zero or one to allow reporting in post-regression commands (e.g., estout). 
{p_end}
{p 4 6 2}
- Updates on 2013-10-18: Corrected P-values for the number of non-missing t-statistics; corrected error with
the confidence intervals. 
{p_end}
{p 4 6 2}
- Updates on 2014-02-04: Updated table so R2 and adjusted R2 are clearly labeled. Relevant update to the
cgmreg command called by this program. 
{p_end}
{p 4 6 2}
- Updates on 2015-03-10: Updated p-values so to avoid p-values greater than one. The code looks at
the upper (lower) tail of the bootstrap distribution for positive (negative) coefficients, which was
sometimes causing p-values greater than one (e.g., positive coefficient with t-stat in the bottom half
of the bootstrap distribution) 
{p_end}


{p 4 6 2}
- {bf: WARNING:} The program uses a macro string variable containing the list of regressors. If you are
using a version of Stata that limits the length of string variables (e.g., Intercooled), you may have
to shorten variable names so that the entire listing of variables can fit in the string macro.
{p_end}

{title:Description}

{p 4 4 2}
{opt cgmwildboot} resembles the {opt cgmreg} regression command except that it bootstraps along one of the clustering
dimension using the wild bootstrap procedure decribed in Cameron, Gelbach and Miller (2008). The program is modified
from the cgmreg command provided by Miller.{p_end}


{title:Example}
{phang}{stata "webuse nlsw88" : . webuse nlsw88}{p_end}
{phang}{stata "reg wage tenure ttl_exp collgrad if ~missing(age), cluster(industry)" : . reg wage tenure ttl_exp collgrad if ~missing(age), cluster(industry)}{p_end}
{phang}{stata "cgmreg wage tenure ttl_exp collgrad, cluster(industry age)" : . cgmreg wage tenure ttl_exp collgrad, cluster(industry age)}{p_end}
{phang}{stata "cgmwildboot wage tenure ttl_exp collgrad, cluster(industry age) bootcluster(industry) reps(50) seed(999)" : . cgmwildboot wage tenure ttl_exp collgrad, cluster(industry age) bootcluster(industry) seed(999)}{p_end}
{phang}{stata "cgmwildboot wage tenure ttl_exp collgrad, cluster(age) bootcluster(industry) reps(50) seed(999)" : . cgmwildboot wage tenure ttl_exp collgrad, cluster(age) bootcluster(industry) reps(50) seed(999)}{p_end}


{title:Return values}

{col 4}Scalars
{col 8}{cmd:e(N)}{col 27}Number of observations
{col 8}{cmd:e(df_m)}{col 27}Model degrees of freedom
{col 8}{cmd:e(df_r)}{col 27}Residual degrees of freedom
{col 8}{cmd:e(r2)}{col 27}R-squared
{col 8}{cmd:e(r2_a)}{col 27}Adjusted R-squared
{col 8}{cmd:e(nreps)}{col 27} Number of successfull bootstrap resamples
{col 8}{cmd:e(NC)}{col 27}Number of cluster variables
{col 8}{cmd:e(S)}{col 27}Number of cluster combinations
{col 8}{cmd:e(N_i)}{col 27}Number of clusters for cluster variable i

{col 4}Macros
{col 8}{cmd:e(bootclust)}{col 27}Name of bootstrap cluster variable
{col 8}{cmd:e(clusvar)}{col 27}Names of cluster variables
{col 8}{cmd:e(clustvar)}{col 27}Names of cluster variables
{col 8}{cmd:e(vcetype)}{col 27}FAKE: The covariance matrix is 'rigged' (See above)
{col 8}{cmd:e(predict)}{col 27}Program used to implement {cmd:predict}
{col 8}{cmd:e(properties)}{col 27}b V
{col 8}{cmd:e(cmd)}{col 27}{cmd:cgmwildboot}
{col 8}{cmd:e(depvar)}{col 27}Name of dependent variable

{col 4}Matrices
{col 8}{cmd:e(b)}{col 27} Coefficient vector
{col 8}{cmd:e(V)}{col 27} Variance-covariance matrix of the estimators (See above)

{col 4}Functions
{col 8}{cmd:e(sample)}{col 27} Marks estimation sample


{title:Reference}


{p 4 6 2}
- Cameron, A., J. Gelbach and D. Miller. 2006. Robust inference with multi-way clustering. NBER Technical Working Paper 327.{p_end}

{p 4 6 2}
- Cameron, A., J. Gelbach and D. Miller. 2008. Bootstrap-based improvements for inference with clustered errors. {it:Review of Economics and Statistics} 90(3): 414-427.{p_end}

{p 4 6 2}
 - Petersen,M. 2009. Estimating standard errors in finance panel data sets: Comparing approaches. {it:Review of Financial Studies} 22(1): 435-480.{p_end}

{title:Author}

{p 4 4 2}Mitchell Petersen at Northwestern University (mpetersen@northwestern.edu) wrote logit2, which is available at http://www.kellogg.northwestern.edu/faculty/petersen/htm/papers/se/se_programming.htm.{p_end}

{p 4 4 2}Douglas Miller at UC, Davis (dmiller@ucdavis.edu) wrote cgmreg, which is available at http://www.econ.ucdavis.edu/faculty/dmiller/statafiles/index.htm.{p_end}

{p 4 4 2}Judson Caskey, University of Texas, judson.caskey@mccombs.utexas.edu{p_end}



{title:Also see}

{psee}
Manual:  {bf:[R] regress}

{psee}
Online:
{helpb regress}, {helpb _robust}
{p_end}