Saving option within a program

[Dario Maimone Ansaldo Patti]

Dear All,

I have the following rough code (it needs further amendements):

capture program drop marginteff
program define marginteff, rclass
version 10

syntax varlist(min=3 max=3 numeric fv ts), grp(varlist min=1 max=1 numeric) [MIXed] [savedata(string) savegraph(string)]
marksample touse

tokenize `varlist'
local var1="`1'" 
local var2="`2'"
local var12="`3'"

tokenize `grp'
local g="`1'"

tempname B
tempname xb
tempvar rslope
tempvar rint
tempvar cross
tempvar cross_se
tempvar m_cross
tempvar m_cross_se
tempvar m_m_cross
tempvar m_m_cross_se
tempvar t_m_cross
tempvar t_m_m_cross
tempvar rslope
tempvar rint
tempvar prob

mat `B' = e(b)
mat score double `xb' = `B' if `touse'

if "`mixed'"==""{
    qui predict `prob', pr
    qui predictnl `cross'=(_b[`var1']+_b[`var12'])*(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+_b[`var2']+_b[`var12']*`var1'))*(1-(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+_b[`var2']+_b[`var12']*`var1')))-_b[`var1']*(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'))*(1-(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'))) if `touse', se(`cross_se')
}

else if "`mixed'"!=""{
    qui predict `prob', mu
    qui predict `rslope' `rint' if `touse', reffect
    qui su `rslope' `rint'
    qui predictnl `cross'=(_b[`var1']+_b[`var12']+`rslope')*(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+_b[`var2']+(_b[`var12']+`rslope')*`var1'+`rint'))*(1-(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+_b[`var2']+(_b[`var12']+`rslope')*`var1'+`rint')))-(_b[`var1']+`rslope')*(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+`rslope'*`var1'+`rint'))*(1-(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+`rslope'*`var1'+`rint'))) if `touse', se(`cross_se')
}

bysort `g': egen `m_cross'=mean(`cross') if `touse'
qui su `m_cross'
label variable `m_cross' "(Avg) Interaction term"
scalar `m_m_cross'=r(mean)
bysort `g': egen `m_cross_se'=mean(`cross_se') if `touse'
qui su `m_cross_se'
label variable `m_cross_se' "(Avg) S.E."
scalar `m_m_cross_se'=r(mean)
gen `t_m_cross'=`m_cross'/`m_cross_se' if `touse'
quietly su `t_m_cross'
label variable `t_m_cross' "(Avg) T-statistic"
scalar `t_m_m_cross'=r(mean)

if "`savedata'"!=""{
    preserve 
    keep `m_cross' `m_cross_se' `t_m_cross'
    restore
}  

if "`savegraph'"!=""{
    preserve
    twoway(scatter `m_cross' `prob')
    "`savegraph'"
    restore
}

label variable `m_cross' "Cross effect"
label variable `m_cross_se' "Standard Errors"
label variable `t_m_cross' "t-test"

su `m_cross' `m_cross_se' `t_m_cross'


di
di
di "Significance test for cross term"
di as text _dup(50) "-"
  di as text "     t-statistic  = " in  result %9.3f `m_m_cross'
  di as text "     s.e.         = " in  result %9.3f            `m_m_cross_se'
  di as text "     t-test       = " in  result %9.3f `t_m_m_cross'
  di as text "     P>|t|        = "   in  result %9.3f ttail(10000,`t_m_m_cross')
di as text _dup(50) "-"
end

I included two options *savedata* and *savegraph*. After I run my analysis, I would like to save some data in a separate file and some graphs too. I would like to be able to include within savadata option the location for where I want to save the data (specifically, New_Analysis\Results). But I am not able to do so. Specifically, I would like to have something like:

marginteff y x1 x2 x12, mixed savedata(New_Analysis\Results\dataint.dta) savegraph(New_Analysis\Results\my_graph)

How can I do that? Specifically, I cannot find the file dataint.dta in the directory I selected (there is no file dataint.dta at all on my pc). What's wrong with my code?

A second issue I have is that before generating the summariy statistics of `m_cross' `m_cross_se' `t_m_cross' I try to label them. But it seems Stata cannot do that.

 Variable |        Obs        Mean    Std. dev.       Min        Max
-------------+---------------------------------------------------------
    __000007 |    186,443    .0121636    .0029053   .0055672   .0186525
    __000008 |    186,443    .0059492    .0016279   .0025022   .0092233
    __00000B |    186,443    2.106395    .3716783   .7090537   2.815147

Thanks for your kind help.

Dario

[Dario Maimone Ansaldo Patti]

If it can help in understanding my second problem above, I used set trace on. I noted the following:

- label variable `t_m_cross' "t-test"
= label variable __00000B "t-test"
  --------------------------------------------------------------------------------------------- begin label ---
  - version 10.0
  - local vv : display "version " string(_caller()) ", missing:"
  - gettoken val : 0
  - if (strpos("`val'", "val") > 0 ) {
  = if (strpos("variable", "val") > 0 ) {
    gettoken val 0 : 0
    syntax anything [, nofix]
    if "`fix'" != "" {
    local fix ", nofix"
    }
    gettoken var rest : anything
    while `"`rest'"' != "" {
    gettoken lab rest : rest
    local label "`lab'"
    }
    local vlist : list anything - lab
    if "`lab'" == "." {
    local lab ""
    }
    foreach var of varlist `vlist' {
    `vv' _label `val' `var' `lab' `fix'
    }
    }
  - else {
  - `vv' _label `macval(0)'
  = version 14, missing: _label variable __00000B "t-test"
  - }

I do not know why Stata cannot label the variable correctly.

Dario

[William Lisowski]

The following example run in Stata 17.0 suggests that variable labels are ignored by the summarize command.

. set obs 100
Number of observations (_N) was 0, now 100.

. generate x = runiform()

. label variable x "whatever"

. describe x

Variable      Storage   Display    Value
    name         type    format    label      Variable label
------------------------------------------------------------------------------------------------
x               float   %9.0g                 whatever

. summarize x

    Variable |        Obs        Mean    Std. dev.       Min        Max
-------------+---------------------------------------------------------
           x |        100    .4973389      .30869   .0030522   .9874847

.

and this continuation shows successful labeling of a temporary variable and use of the label by the tab command, suggesting the failure to use the labeling lies in summarize.

. tempvar t

. generate `t' = runiform()

. label variable `t' "whichever"

. describe `t'

Variable      Storage   Display    Value
    name         type    format    label      Variable label
------------------------------------------------------------------------------------------------
__000000        float   %9.0g                 whichever

. summarize `t'

    Variable |        Obs        Mean    Std. dev.       Min        Max
-------------+---------------------------------------------------------
    __000000 |        100     .510208    .2838849   .0013631   .9944572

. tab `t' in 1/5

  whichever |      Freq.     Percent        Cum.
------------+-----------------------------------
   .5948808 |          1       20.00       20.00
   .6701937 |          1       20.00       40.00
   .6880603 |          1       20.00       60.00
   .7970893 |          1       20.00       80.00
   .9794578 |          1       20.00      100.00
------------+-----------------------------------
      Total |          5      100.00

With regard to savedata options, you never actually issue the command

save `"`savedata'"'

to save the dataset, and note that in your example, the

New_Analysis\Results\

directory will already need to exist as a subdirectory of your working directory.

[Dario Maimone Ansaldo Patti]

William Lisowski thanks a lot for the clarifications. Regarding the first issue, you are totally right. I did not pass the command to save the file. Regarding the labelling issue, it is a bit annoying that summarize cannot accept a label of the variables. Maybe I can work the problem out in some way, maybe copying the variables into new temporary ones with the name I like and then use the sum command. Thanks a lot for your help.

[Dario Maimone Ansaldo Patti]

William Lisowski Thanks again for your help. I have another small issue. I changed the code to pass the save command. Specifically, I wrote:

**# Program to calculate marginal effects for the cross term

capture program drop marginteff
program define marginteff, rclass
version 14

syntax varlist(min=3 max=3 numeric fv ts), grp(varlist min=1 max=1 numeric) [MIXed] [savedata(string) savegraph(string)]
marksample touse

tokenize `varlist'
local var1="`1'" 
local var2="`2'"
local var12="`3'"

tokenize `grp'
local g="`1'"

tempname B 
tempname xb
tempvar rslope
tempvar rint
tempvar cross
tempvar cross_se
tempvar m_cross
tempvar m_cross_se
tempvar m_m_cross
tempvar m_m_cross_se
tempvar t_m_cross
tempvar t_m_m_cross
tempvar rslope
tempvar rint
tempvar prob
tempvar N

mat `B' = e(b)
local j: colsof(`B')
mat score double `xb' = `B' if `touse'
local N=e(N)
local df=`N'-`j'-1

if "`mixed'"==""{
    qui predict `prob', pr
    qui predictnl `cross'=(_b[`var1']+_b[`var12'])*(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+_b[`var2']+_b[`var12']*`var1'))*(1-(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+_b[`var2']+_b[`var12']*`var1')))-_b[`var1']*(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'))*(1-(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'))) if `touse', se(`cross_se')
}

else if "`mixed'"!=""{
    qui predict `prob', mu
    qui predict `rslope' `rint' if `touse', reffect
    qui su `rslope' `rint'
    qui predictnl `cross'=(_b[`var1']+_b[`var12']+`rslope')*(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+_b[`var2']+(_b[`var12']+`rslope')*`var1'+`rint'))*(1-(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+_b[`var2']+(_b[`var12']+`rslope')*`var1'+`rint')))-(_b[`var1']+`rslope')*(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+`rslope'*`var1'+`rint'))*(1-(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+`rslope'*`var1'+`rint'))) if `touse', se(`cross_se')
}

bysort `g': egen `m_cross'=mean(`cross') if `touse'
qui su `m_cross'
label variable `m_cross' "(Avg) Interaction term"
scalar `m_m_cross'=r(mean)
bysort `g': egen `m_cross_se'=mean(`cross_se') if `touse'
qui su `m_cross_se'
label variable `m_cross_se' "(Avg) S.E."
scalar `m_m_cross_se'=r(mean)
gen `t_m_cross'=`m_cross'/`m_cross_se' if `touse'
quietly su `t_m_cross'
label variable `t_m_cross' "(Avg) T-statistic"
scalar `t_m_m_cross'=r(mean)

if "`savedata'"!=""{
    preserve 
    keep `m_cross' `m_cross_se' `t_m_cross'
    save `"`savedata'"'
    restore
}  

if "`savegraph'"!=""{
    preserve
    twoway(scatter `m_cross' `prob')
    save `"`savegraph'"'
    restore
}

tempname cross_effect
tempname se
tempname t_test

gen `cross_effect'=`m_cross'
gen `se'=`m_cross_se'
gen `t_test'=`t_m_cross'

su `cross_effect' `se' `t_test'

di
di
di "Significance test for cross term"
di as text _dup(50) "-"
  di as text "     t-statistic  = " in  result %9.3f `m_m_cross'
  di as text "     s.e.         = " in  result %9.3f            `m_m_cross_se'
  di as text "     t-test       = " in  result %9.3f `t_m_m_cross'
  di as text "     P>|t|        = "   in  result %9.3f ttail(`df',`t_m_m_cross')
di as text _dup(50) "-"
end

I tried to add a replace option like the following:

save `"`savedata'"', replace

or

save `"`savedata', replace"'

or in the command window I typed

marginteff centeraf global centerafglob, grp(id) mixed savedata(New_Analysis\Results\mydata.dta, replace)

However the file is saved with the extension .dta, replace and in any case, it does not me to replace the existing file. What is wrong with that way I am using?
Regarding the second problem, I tried to generate new variables, but still sum does not display the correct name. I guess there is no any other trick I can use..

Best

Dario

[daniel klein]

I did not read the details here but I wanted to point out that there is no such thing as the

...marginal effects for the cross term

in the sense that the product of two variables, say, x12=x1*x2, cannot change independently from those two variables, i. e., x1 and x2, that make up the product.

[Dario Maimone Ansaldo Patti]

Thanks for pointing this out daniel klein. I am aware of this. I am adapting to my needs the procedure suggested by Norton and Ai (2004). Here there is another paper, which shows an example using margins and the command inteff which produces basically the same results.

[Dario Maimone Ansaldo Patti]

William Lisowski I was able to solve the issue in #5. I changed from r(class) to e(class). Also I worked out to fix the name of the variables using summarize: I changed the code in the following way:

tempname cross_effect
tempname se
tempname t_test

replace cross_effect=`m_cross'
replace se=`m_cross_se'
replace t_test=`t_m_cross'

su cross_effect se t_test

[Matthew Smith Stata]

William Lisowski Thanks again for your help. I have another small issue. I changed the code to pass the save command. Specifically, I wrote:

**# Program to calculate marginal effects for the cross term

capture program drop marginteff
program define marginteff, rclass
version 14

syntax varlist(min=3 max=3 numeric fv ts), grp(varlist min=1 max=1 numeric) [MIXed] [savedata(string) savegraph(string)]
marksample touse

tokenize `varlist'
local var1="`1'"
local var2="`2'"
local var12="`3'"

tokenize `grp'
local g="`1'"

tempname B
tempname xb
tempvar rslope
tempvar rint
tempvar cross
tempvar cross_se
tempvar m_cross
tempvar m_cross_se
tempvar m_m_cross
tempvar m_m_cross_se
tempvar t_m_cross
tempvar t_m_m_cross
tempvar rslope
tempvar rint
tempvar prob
tempvar N

mat `B' = e(b)
local j: colsof(`B')
mat score double `xb' = `B' if `touse'
local N=e(N)
local df=`N'-`j'-1

if "`mixed'"==""{
qui predict `prob', pr
qui predictnl `cross'=(_b[`var1']+_b[`var12'])*(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+_b[`var2']+_b[`var12']*`var1'))*(1-(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+_b[`var2']+_b[`var12']*`var1')))-_b[`var1']*(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'))*(1-(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'))) if `touse', se(`cross_se')
}

else if "`mixed'"!=""{
qui predict `prob', mu
qui predict `rslope' `rint' if `touse', reffect
qui su `rslope' `rint'
qui predictnl `cross'=(_b[`var1']+_b[`var12']+`rslope')*(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+_b[`var2']+(_b[`var12']+`rslope')*`var1'+`rint'))*(1-(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+_b[`var2']+(_b[`var12']+`rslope')*`var1'+`rint')))-(_b[`var1']+`rslope')*(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+`rslope'*`var1'+`rint'))*(1-(invlogit(`xb'-_b[`var2']*`var2'-_b[`var12']*`var12'+`rslope'*`var1'+`rint'))) if `touse', se(`cross_se')
}

bysort `g': egen `m_cross'=mean(`cross') if `touse'
qui su `m_cross'
label variable `m_cross' "(Avg) Interaction term"
scalar `m_m_cross'=r(mean)
bysort `g': egen `m_cross_se'=mean(`cross_se') if `touse'
qui su `m_cross_se'
label variable `m_cross_se' "(Avg) S.E."
scalar `m_m_cross_se'=r(mean)
gen `t_m_cross'=`m_cross'/`m_cross_se' if `touse'
quietly su `t_m_cross'
label variable `t_m_cross' "(Avg) T-statistic"
scalar `t_m_m_cross'=r(mean)

if "`savedata'"!=""{
preserve
keep `m_cross' `m_cross_se' `t_m_cross'
save `"`savedata'"'
restore
}

if "`savegraph'"!=""{
preserve
twoway(scatter `m_cross' `prob')
save `"`savegraph'"'
restore
}

tempname cross_effect
tempname se
tempname t_test

gen `cross_effect'=`m_cross'
gen `se'=`m_cross_se'
gen `t_test'=`t_m_cross'

su `cross_effect' `se' `t_test'

di
di
di "Significance test for cross term"
di as text _dup(50) "-"
di as text " t-statistic = " in result %9.3f `m_m_cross'
di as text " s.e. = " in result %9.3f `m_m_cross_se'
di as text " t-test = " in result %9.3f `t_m_m_cross'
di as text " P>|t| = " in result %9.3f ttail(`df',`t_m_m_cross')
di as text _dup(50) "-"
end

I tried to add a replace option like the following:

save `"`savedata'"', replace

or

save `"`savedata', replace"'

or in the command window I typed

marginteff centeraf global centerafglob, grp(id) mixed savedata(New_Analysis\Results\mydata.dta, replace)

However the file is saved with the extension .dta, replace and in any case, it does not me to replace the existing file. What is wrong with that way I am using?
Regarding the second problem, I tried to generate new variables, but still sum does not display the correct name. I guess there is no any other trick I can use..

Best

Dario

Did you find a solution to the problem regarding the "replace" option?