Very simple example of foreach over one string variable with several values

[Anne Dam]

Dear forum,

Im struggling to learn how to use the foreach loop using a string variable with several values as indicator var for regression. The examples I have found so far have been very intricate and not exactly what I was looking for.

In the example below, imagine that I just want to use each value in "drug" as an indicator variable and display each of them.

How do I do that?

sysuse cancer,clear

tab drug,miss

*two different aproaches I have tried

*method 1
foreach x of varlist drug {
    display "`x'"
}

*method 2
foreach y in drug {
    display "`y'"
}

Thank you so much for advise and tips!
---
Im using stata 17 for windows

[Joro Kolev]

Loops are not needed to use "a string variable with several values as indicator var for regression". The variable drug is not a string variable:

. sysuse cancer
(Patient survival in drug trial)

. des drug

Variable      Storage   Display    Value
    name         type    format    label      Variable label
--------------------------------------------------------------------------------------------------
drug            byte    %8.0g      type       Drug type

. summ drug

    Variable |        Obs        Mean    Std. dev.       Min        Max
-------------+---------------------------------------------------------
        drug |         48       1.875    .8410986          1          3

. tab drug

  Drug type |      Freq.     Percent        Cum.
------------+-----------------------------------
    Placebo |         20       41.67       41.67
      Other |         14       29.17       70.83
         NA |         14       29.17      100.00
------------+-----------------------------------
      Total |         48      100.00

it is a byte variable that has 3 levels.

How you use such a variable in regression expanded as dummies is

. reg died i.drug

      Source |       SS           df       MS      Number of obs   =        48
-------------+----------------------------------   F(2, 45)        =      9.14
       Model |  3.17202381         2   1.5860119   Prob > F        =    0.0005
    Residual |  7.80714286        45  .173492063   R-squared       =    0.2889
-------------+----------------------------------   Adj R-squared   =    0.2573
       Total |  10.9791667        47  .233599291   Root MSE        =    .41652

------------------------------------------------------------------------------
        died | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
        drug |
      Other  |  -.5214286   .1451444    -3.59   0.001    -.8137644   -.2290928
         NA  |  -.5214286   .1451444    -3.59   0.001    -.8137644   -.2290928
             |
       _cons |        .95   .0931375    10.20   0.000     .7624113    1.137589
------------------------------------------------------------------------------

[Anne Dam]

Dear Joro,
you are absolutely right, I am sorry my example to match my own data was bad - drug is of course a numerical var !

But you pointing that out to me, also helped me realize how to solve my own dataset problem where "drug" is a string variable.
(Instead of trying foreach on my string var, I convert my own string to numeric using encode and apply forvalues ! )


*So instead of trying something like 'if drug was a string':

foreach x of varlist drug {
regress died studytime if drug=="`x'"

I shoud instead use:

*encode own_string_var, gen(drug)
forval i=1/3 {
regress died studytime if drug==`i' 
}

[Clyde Schechter]

Assuming the string variable has a relatively limited number of values, the approach in #3 is practical. However, sometimes a string variable is an identifier (e.g. medical record number with 10 digits) and the number of values is too large of use with -encode-. Or, there is also the issue of looping over the values of a numeric variable whose values cannot be neatly summarized in a short list like 1/3. So the general approach is:

levelsof x, local(xlist)
foreach x of local xlist {
    display `"`x'"'
}

This works for both numeric and string variables. (If x is a numeric variable, the outermost `" "' in the -display- command are not needed, though they do no harm if there. But the innermost ` ' around x must still be there.)

[Anne Dam]

That is very neat - thank you so much to both Joro and Clyde for your ideas and solutions!