looping regressions with different versions

[Anne-Claire Jo]

Hello statausers,

I am working on generating regression tables with nested loop - for different versions - and it is also one of the follow-up questions: https://www.statalist.org/forums/for...t-observations.
In particular, I would like to export regression tables in excel file in which it contains: Columns (1) to (4) are so-called "Aggregated" version and columns from (5) until the end is either version 0 or 1.
Aggregated version is when every country is included in the regression, 0 is when one country is excluded from the dataset at the time (for e.g. if countrylist is USA, UK, FRA, CAN then first set is excluding only USA, second set is UK, and so on), and 1 is when we are regressing one-by-one country (for e.g. only USA, then UK, then FRA so on).
To implement this, i used the following command:

gl folder folderdirectory
cap encode country, gen(cty)

loc job var1 var2 var3 var4

loc analysislevel 0        // 1 if one-by-one (country); 0 if excluding one country

loc iteration 1

levelsof country, local(countries)

foreach c in "" `countries' {

    * Excel file replace/append option
    if `iteration' == 1 local replace_op replace
    if `iteration' > 1  local replace_op append

    * Regression versions: Aggregate + 0/1
    if missing("`c'") {        // Aggregate
        loc regif
        loc ex_lab
        loc cn Aggregate
    }
    if `analysislevel' {     // 1; one-by-one
        loc regif "if country =="`c'""
        loc ex_lab
        loc cn `c'
    }
    else {                    // 0; excluding
        loc regif "if country !="`c'""
        loc ex_lab "Excluding, "`c'","
        loc cn `c'
    }
    
    noi di _n "=================       Country: `cn'     ==================="


    foreach var of local job {
        
        * Basic setup for regression
        loc model_name "5-year `var'"
        loc FElbl13     i.cty#i.year i.cty#i.ind_a38
        loc FElbl24        i.cty#i.ind_a38#i.year
        loc FE13        "C-I C-Y"
        loc FE24         "C-I-Y"
        loc cl_CI        i.cty#i.ind_a38

        * Regression (Columns 1 to 4)
        forval col = 1/4 {
            
            * Setting regressors for columns 1&2 // 3&4
            if inlist (`col', 1, 2)    local xvars ib3.p_LogLP_VA LogL_av     // Columns 1&2
                else local xvars F1LogLP_VA_av ib3.p_LogLP_VA `var'_1L LogL_av // Columns 3&4
        
            * Setting FE/labels for columns 1&3 // 2&4
            if inlist(`col', 1, 3)    local fe `FElbl13' local fe_table `FE13'
                else local fe `FElbl124' local fe_table `FE24'
            
            
            * REGRESSION
            capture reghdfe `var'_5L `xvars' `regif' [aw=weightvar], a(`fe') vce(cluster `cl_CI')
            
            if c(rc) == 0 {
                
                levelsof country if e(sample), clean local(countrylist)
                local n_countries = `r(r)'
                levelsof ind_a38 if e(sample)
                local n_inds = `r(r)'

                * Export to excel file
                outreg2 using "${folder}/Baseline/`analysislevel'_`var'.xlsx", `replace_op' ctitle(`model_name') label ///    
                        addtext(Fixed effects, `fe_table', Country List, `countrylist', `ex_lab' Countries, `n_countries', Industries, `n_inds')
            }
            else if !inlist(c(rc), 2000, 2001) {
                display as error `"Unexpected regression error: var = `var', country = `c'"'
            }
        
        }
    
    } // end of job loop
    
    local ++iteration
    
} // end of countries loop

However, i get the following error:

=================    Country:    ===================
inlist not found
r(111);

I am unsure where to find the issues.. and to fix them eventually.
I understand that my code is a bit complicated but hope the question is clear!

Thanks a lot in advance.

[Nick Cox]

 
 inlist (`

should be

inlist(

as otherwise Stata is looking for a variable or scalar called inlist -- and reporting correctly that none such can be found.

Debugging is often difficult. I spent about two hours at the weekend looking for a bug before I found it. In this case, the problem was presumably with code using inlist() so I looked for that.

[Anne-Claire Jo]

thanks Nick for your reply.
I changed to inlist() but now it shows error:

 
 Unexpected regression error: var = `var', country = `c'

for every country and every variable. it seems the code is not working as wanted..

[Nick Cox]

I can't help with that one. You could at least tell us which command produced that error.

[Clyde Schechter]

That error message almost certainly arises from

 else if !inlist(c(rc), 2000, 2001) {
     display as error `"Unexpected regression error: var = `var', country = `c'"'
 }

although it is very strange that it actually prints out `var' and `c' instead of the values of var and c when the error was triggered.

The way that part of the code is reached is when the -capture reghdfe- command encounters an error other than an insufficient estimation sample size. So you need to figure out what is going wrong with -reghdfe-. To do that, put

set tracedepth 1
set trace on

immediately before the -capture reghdfe- command. (You probably want to put -set trace off- immediately after the -capture reghdfe- to avoid printing out a lot of other trace information that probably will not be helpful in this context.) That way Stata will show you what that command looks like when all the local macros in it have been interpreted. Also, change -capture- to -capture noisily- so that Stata will show the error message that -reghdfe- produces. You will probably find some non-existent variables mentioned, or something of that nature. In any event, the answer will almost surely be found in the results of that trace.

[Anne-Claire Jo]

sorry, in #3, i was mentioning that the error shows continuously like
Unexpected regression error: var = var1, country = country1
Unexpected regression error: var = var2, country = country2
etc

[Hemanshu Kumar]

Anne-Claire Jo Thanks for that clarification. That means your reghdfe command is not working as expected, as mentioned in #5. A good starting point to understand what is going on, would be remove the capture in front of that command, and run your do-file. This will cause the reghdfe command to fail and the error messages from that should be informative. Also add the set tracedepth and set trace commands appropriately, as suggested in #5.

If you still do not understand what is going on, please show us the output of that trace.

[John DSouza]

Hello,
I haven't followed all of your code, but this looks wrong

 
             * Setting FE/labels for columns 1&3 // 2&4             if inlist(`col', 1, 3)    local fe `FElbl13' local fe_table `FE13'                 else local fe `FElbl124' local fe_table `FE24'

Try

if inlist(`col', 1, 3) {
local fe `FElbl13'
local fe_table `FE13'
} 
else {
local fe `FElbl124'
local fe_table `FE24'
}

[Anne-Claire Jo]

thank you all for the reply!

I made a bit of modification as below:

  
 gl folder folderdirectory cap encode country, gen(cty)  loc job var1 var2 var3 var4 
loc analysislevel 0        // 1 if one-by-one (country); 0 if excluding one country

loc iteration 1

levelsof country, local(countries)

foreach c in "" `countries' {

    * Excel file replace/append option
    if `iteration' == 1 local replace_op replace 
    if `iteration' > 1  local replace_op append

    * Regression versions: Aggregate + 0/1
    if `analysislevel' {     // 1; one-by-one
        loc regif "if country =="`c'""
        loc ex_lab 
        loc cn `c'
    }
    else {                    // 0; excluding
        loc regif "if country != "`c'""
        loc ex_lab "Excluding, "`c'","
        loc cn `c'
    }
    
    if missing("`c'")  {        // Aggregate
        loc regif 
        loc ex_lab 
        loc cn Aggregate
    }
    
    noi di _n " =================       Country: `cn'         =================== "


    foreach var of local job {
        
        * Basic setup for regression
        loc model_name "5-year `var'"
        loc FElbl13     i.cty#i.year i.cty#i.ind_a38
        loc FElbl24        i.cty#i.ind_a38#i.year
        loc FE13        "C-I C-Y"
        loc FE24         "C-I-Y"
        loc cl_CI        i.cty#i.ind_a38

        * Regression (Columns 1 to 4)
        forval col = 1/4 {
            
            * Setting RHS for columns 1&2 // 3&4
            if inlist(`col', 1, 2)    ///
               local xvars ib3.p_LogLP_VA LogL_av     // Columns 1&2
            else                     ///
               local xvars F1LogLP_VA_av ib3.p_LogLP_VA `var'_1L LogL_av // Columns 3&4
        
            * Setting FE/labels for columns 1&3 // 2&4
            if inlist(`col', 1, 3) {
               local fe `FElbl13'   
               local fe_table `FE13'
            }
            else {                    
               local fe `FElbl124'  
               local fe_table `FE24'
            }
             
            * REGRESSION
            capture reghdfe `var'_5L `xvars' `regif' [aw=weightvar], a(`fe') vce(cluster `cl_CI')
            
            if c(rc) == 0 {
                
                levelsof country if e(sample), clean local(countrylist)
                local n_countries = `r(r)'
                levelsof ind_a38 if e(sample)
                local n_inds = `r(r)'

                * Export to excel file
                outreg2 using "${folder}/Baseline/`analysislevel'_`var'.xls", `replace_op' ctitle(`model_name') label ///    
                        addtext(Fixed effects, `fe_table', Country List, `countrylist', `ex_lab' Countries, `n_countries', Industries, `n_inds')
            }
            else if !inlist(c(rc), 2000, 2001) {
                display as error `"Unexpected regression error: var = `var', country = `c'"'
            }
        
        }
    
    } // end of job loop
    
    local ++iteration
    
} // end of countries loop

and it is fixed for the issue with unexpected error message, but still I am encountering issue for the output table in the excel file.
Columns (1) to (4) are supposed to be the regressions for the set of aggregated version (which is all the countries / i.e. without if condition in the regression) and from (5) until the end should be the regressions for the set of versions either 0 /1.
However, i see that in the output table, column (1) produces the last part of regression #4 (i.e. written column (4) in the code) and from columns (2) until the end are the set of versions either 0 or 1.
In other words, columns (1)-(3) of the aggregated version is deleted and only column (4) is remained for this version; and the rest are the set of regressions for version 0/1.
Because of the confidentiality issue, I cannot provide the output table but i hope the issue mentioned here is clear for everyone!
Could someone help me finding and solving the issue here?

[Hemanshu Kumar]

If you keep sending outreg2 output to the same sheet of the same Excel workbook, it will replace the prior contents of that sheet. You need to modify your code so you produce a single table with all columns as desired, and then output that full table to the Excel sheet in a single call to outreg2.

[Anne-Claire Jo]

Hemanshu Kumar sorry Im a bit confused.. how should i change?