.-
help for ^bootcor^                                                (STB-55: sg138)
.-

Bootstrap inferences about measures of correlation
--------------------------------------------------

   ^bootcor^ var1 var2 var3 [var4] [^if^ exp] [^in^ range]
          [^, r^eps^(^#^) st^at^(pearson^|^icc^|^concord)^
            ^le^vel^(^#^) sav^ing^(^newfile^)^]


Description
-----------

^bootcor^ uses bootstrap resampling to compare correlations between variables.
Bootstrapping, unlike the standard Fisher r-to-z confidence interval method,
does not require assumptions of underlying bivariate normality and therefore
provides more accurate confidence intervals and inferences.  ^bootcor^ also
solves the problem of non-independent samples (i.e., elements in a correlation
matrix) where standard formulas break down.

The user specifies the variables to be compared.  When three variables are
listed, ^bootcor^ will compute confidence intervals for the correlation
between var1 and var2 (r1), and var1 and var3 (r2). When four variables are
listed, ^bootcor^ computes confidence intervals for var1 and var2 (r1) 
and var3 and var4 (r2).

Next, r1 and r2 are compared using a z-test on the difference between r1 and 
r2 for each boostrap sample.  A z-score and p-value are given.  


Options
-------
  
^reps(^#^)^ allows the user to specify how many bootstrap replications
    (B) should be computed.  The default value of B is 50.  It is
    recommended that B be at least 1000 for adequate accuracy when
    estimating correlation coefficients.

^stat(pearson)^ uses the Pearson Correlation Coefficient as the parameter
    to be estimated.  This is the default setting.

^stat(icc)^ uses the Intra-Class Correlation Coefficient as the
    parameter to be estimated.  ^bootcor^ uses a modified version of
    iclassr2.ado to compute the correlation coefficient.

^stat(concord)^ uses the Concordance Coefficient as the parameter to be
    estimated.  ^bootcor^ uses a modified version of concord.ado to compute
    the correlation coefficient.

^level(^#^)^ allows the user to specify the level of confidence for the
    individual correlation coefficients.  Level can range from 1 to 100.  The
    default is 95.

^saving(^newfile^)^ will export the bootstrap replications to a .dct
    file that the user can later analyze in more detail.  Five
    variables are saved, with each resample listed casewise.  r_boot1
    is r1, r_boot2 is r2, z_boot1 is the Fisher transformed value of
    r_boot1, z_boot2 is the Fisher transformed value of r_boot2, and
    z_bootd is the difference of z_boot1 and z_boot2.  The user can
    open this file using the command

     infile using newfile.dct


Examples
--------

 . ^bootcor var1 var2 var3^

 . ^bootcor var1 var2 var3 var4^

 . ^bootcor time1 time2 time3, stat(icc) reps(1000)^

 . ^bootcor var1t1 var1t2 var2t1 var2t2 if(group==1), stat(concord)^

 . ^bootcor var1 var2 var3, saving(boot) level(90)^


Author
------

        Dan J. Neal
        Syracuse, New York
        email: djneal@@syr.edu


Also see
--------
 
   STB: STB-55 sg138