vlook

* Stata code, custom command
* vlook is a powerful command for rapidly assessing the capability of current data to explain one variable

* This code creates a stata command that does a diagnostic of a single variable in a data set.


* Drops the program vlook if it already is in memory
cap program drop vlook


* Defines the program vlook
program define vlook, rclass


  * Summarizes the variable of interest
  sum `1', detail  
    
  * Creates a histogram of the variable
  hist `1'
  
  *** This sections checks the correlation between the variable and all other variables in the dataset
  * Declare empty variable holders
  local corr_matrix25
  local corr_matrix5
  local corr_matrix75
  
  * Loops through all possible variables
  foreach v of varlist * {
    * Excude the variable being examined
    if ("`v'"!="`1'") {
        * Correlates each of the variables in the dataset
      cap corr `v' `1'
        * If there is no error _rc and the correlation is greater than .25 and less then .5 add `v' to the 
        * global called corr_matrix25. 
      if  (_rc==0&abs(r(rho))>=.25&abs(r(rho))<.5) local corr_matrix25 `corr_matrix25' `v'
        * If there is no error _rc and the correlation is greater than .5 and less then .75 add `v' to the 
        * global called corr_matrix5.
      if  (_rc==0&abs(r(rho))>=.5&abs(r(rho))<.75) local `corr_matrix5'  `corr_matrix5' `v'
        * If there is no error _rc and the correlation is greater than or equal to .5 and less then .75 add 
        * `v' to the global called corr_matrix75.
      if  (_rc==0&abs(r(rho))>=.75) local `corr_matrix' `corr_matrix75' `v'
    }
  }
  
  di "`1' is correlated with $corr_matrix25 at (.25<=corr<.5)"
    corr `1' $corr_matrix25
  di "`1' is correlated with $corr_matrix5 at (.5<=corr<.75)"
    corr `1' $corr_matrix5
  di "`1' is correlated with $corr_matrix75 at (.75<=corr)"
    corr `1' $corr_matrix75


  qui reg `1' $corr_matrix75 $corr_matrix5 $corr_matrix25
  * This saves the r2 from the previous regression.  It is the upper limmit of explainable variance by use
  * of the available variables which are at lease .25% correlated with the variable of interest
  local upperlimit = string(e(r2),"%9.2f")


  *** This sections makes a list of variables with explanatory power to analysze with respect to the variable
  *** of interest.
  * Creates an empty global to hold variables.
  local reg_vars
  * Loops through only variables that are at least 50% correlated with the variable of interest.
  foreach v in $corr_matrix75 $corr_matrix5 {
    * This creates a flag to tell stata to exclude a variable from the regression.
    local flag=0
* Loops through all of the variables that have already been added to reg_vars.
    foreach vv in $reg_vars {
    qui corr `v' `vv'
* If any variable is too correlated with any other variable it is excluded from the regression.
    if (("`v'"!="`vv'")&(abs(r(rho))>=.90)) {
      local flag=1
      di "Excluding `v' from regression because of high correlation with `vv'"
      * Ends the most nested loop
 break
    }
    }
  * If the flag has not been sprung add the variable `v' to the regression
  if (`flag'==0)   local reg_vars `reg_vars' `v'
  }
  
  return local regvars `regvars'
  
  *** This does a detailed quantile analysis of the explanatory power of the variables
  *** around different quantiles.
  
  local quantile_list 125 250 375 500 625 750 875
  * First declare a matrix to hold the results of the quantile regression
  
  local nrows = wordcount("`reg_vars'") 
  local ncols = wordcount("`quantile_list'") 
  
  mata: A=J(strtoreal(st_local("nrows"))+2, strtoreal(st_local("ncols")), 0)
  
  * Assign the quantile levels to the first row of the matrix
  mata: i=0
  foreach v in `quantile_list' {
    mata: i++
    mata:  A[1,i]=strtoreal(st_local("v"))/1000
  }


  
  mata: i=0
  
  foreach q in `quantile_list'  {
    mata: i++
qui di "qreg `1' `reg_vars', quantile(`=`q'/1000')"
qui qreg `1' `reg_vars', quantile(`=`q'/1000')
    mata: ii=1  


    foreach v in `reg_vars' _cons {
      mata: ii++
 cap local temp=string(_b[`v'],"%9.2f")
 if _rc!=0 local temp=.
 mata: A[ii,i]=strtoreal(st_local("temp"))
    }
  }
  
  * This displays the variables corresponding to what rows in the quantile regression.
  di "Rows of the coefficient matrix on the quantile regression:"
  local i = 0
  foreach v in quantile `reg_vars' _cons{
   local i = `i'+1
   di "`i' - `v'"
  }
  mata:A


  di _newline "reg `1' `reg_vars'"
  reg `1' `reg_vars'
  
  di "The upper limit on explainable variance is `upperlimit'"


end




*******************************************************
*                  Command Example                    *


use http://www.ats.ucla.edu/stat/stata/dae/crime, clear
vlook crime

Program to recode all missing in SPSS to Stata missing

* Stata code
* Converts all values that equal missing such as -9999 to Stata missing values .

* Drops the program if it is already in the memory.
cap program drop recode_missing


* Defines a new program called recode_missing.
program define recode_missing

  local variable_changed = 0
  local variable_skipped_count = 0
  local variable_skipped


  * Loops through all variables identified by the first argument in the command.
  foreach v of varlist `1' {
   * Replaces the variable observations which match the second argument.
    capture replace `v'=. if `v'==`2'
   if _rc==0 local variable_changed = `variable_changed' + 1
   if _rc!=0 { 
      local variable_skipped_count = `variable_skipped_count' + 1
      local variable_skipped `variable_skipped' "`v'" 
   }
  }


  di "Recoded observations from `1' which equalled `2' into stata missing"
  di "# of variables changed: `variable_changed' "
  if `variable_skipped_count'!=0 {
 di "# of variables skipped: `variable_skipped_count' "
 di "skipped:  `variable_skipped'"
  }
end


* Example syntax
* . recode_missing * -9999
* This command will recode all variables that contain a -9999 into stata missing values ".".






***********************************************************
*          example of the command  in action              * 
use http://www.ats.ucla.edu/stat/stata/dae/logit.dta, clear
gen test="adsf"


* Does not make any logical sense to recode 0s as missing
recode_missing * 0


* Recoded observations from * which equaled 0 into stata missing
* # of variables changed: 4 
* # of variables skipped: 1 
* skipped:  test