why always (invalid name

[Johnny YANG]

Here is my code as shown below, when I run it, it always report error "( invalid name". Somebody please help me with this.

tempname myresults
postfile `myresults' threshold intercept gradient se using myresults.dta

forval x=1(1)65{
capture noisily xi:regress lnSale lnYearlyIndOut fyear if mi==`x',robust
post `myresults' (`x') (`=_b[_cons]') (`=_b[lnYearlyIndOut]') (`=_se[lnYearlyIndOut]')
}

[Joseph Coveney]

Clean up your code.

tempname myresults
postfile `myresults' byte threshold double intercept double gradient double se using myresults

forvalues x = 1/65 {
    quietly count if mi == `x'
     if r(N) > 1 { // Ought to be >2 if you want to assure both regression coefficients, and ought to be >3 if you want to assure their standard errors aren't missing-valued
        regress lnSale c.(lnYearlyIndOut fyear) if mi == `x'
        post `myresults' (`x') (`=_b[_cons]') (`=_b[lnYearlyIndOut]') (`=_se[lnYearlyIndOut]')
     }
}

[Johnny YANG]

Clean up your code.

tempname myresults
postfile `myresults' byte threshold double intercept double gradient double se using myresults

forvalues x = 1/65 {
quietly count if mi == `x'
if r(N) > 1 { // Ought to be >2 if you want to assure both regression coefficients, and ought to be >3 if you want to assure their standard errors aren't missing-valued
regress lnSale c.(lnYearlyIndOut fyear) if mi == `x'
post `myresults' (`x') (`=_b[_cons]') (`=_b[lnYearlyIndOut]') (`=_se[lnYearlyIndOut]')
}
}

Thank you Joseph! I tried your code, but it still doesn't work. Here is the error reported :
( invalid name
program error: code follows on the same line as close brace
r(198);

It doesn't make sense to me, the braces are obviously closed.

[Joseph Coveney]

First, you don't need to macro-evaluate the regression output; you can just put them in as-is. See below.

.ÿ
.ÿversionÿ15.1

.ÿ
.ÿclearÿ*

.ÿ
.ÿquietlyÿsysuseÿauto

.ÿ
.ÿtempnameÿfile_handle

.ÿtempfileÿtmpfil0

.ÿ
.ÿpostfileÿ`file_handle'ÿbyteÿregionÿdoubleÿslopeÿdoubleÿslope_seÿusingÿ`tmpfil0'

.ÿ
.ÿforvaluesÿregionÿ=ÿ0/1ÿ{
ÿÿ2.ÿÿÿÿÿÿÿÿÿquietlyÿregressÿgear_ratioÿc.mpgÿifÿforeignÿ==ÿ`region'
ÿÿ3.ÿÿÿÿÿÿÿÿÿpostÿ`file_handle'ÿ(`region')ÿ(_b[mpg])ÿ(_se[mpg])
ÿÿ4.ÿ}

.ÿpostcloseÿ`file_handle'

.ÿ
.ÿexit

endÿofÿdo-file


.

Second, for the close-brace problem, try

set trace on

to see where the error arises.

Last, I forgot to mention this, but there's another thing that you might need to trap. Insert

drop if mi(lnSale, lnYearlyIndOut, fyear)

before the loop.

[Johnny YANG]

First, you don't need to macro-evaluate the regression output; you can just put them in as-is. See below.

.ÿ
.ÿversionÿ15.1

.ÿ
.ÿclearÿ*

.ÿ
.ÿquietlyÿsysuseÿauto

.ÿ
.ÿtempnameÿfile_handle

.ÿtempfileÿtmpfil0

.ÿ
.ÿpostfileÿ`file_handle'ÿbyteÿregionÿdoubleÿslopeÿdoubleÿslope_seÿusingÿ`tmpfil0'

.ÿ
.ÿforvaluesÿregionÿ=ÿ0/1ÿ{
ÿÿ2.ÿÿÿÿÿÿÿÿÿquietlyÿregressÿgear_ratioÿc.mpgÿifÿforeignÿ==ÿ`region'
ÿÿ3.ÿÿÿÿÿÿÿÿÿpostÿ`file_handle'ÿ(`region')ÿ(_b[mpg])ÿ(_se[mpg])
ÿÿ4.ÿ}

.ÿpostcloseÿ`file_handle'

.ÿ
.ÿexit

endÿofÿdo-file


.

Second, for the close-brace problem, try

set trace on

to see where the error arises.

Last, I forgot to mention this, but there's another thing that you might need to trap. Insert

drop if mi(lnSale, lnYearlyIndOut, fyear)

before the loop.

That works! Thanks Joseph!
But another question, in the loop there are some regressions where the coefficient of lnYearlyIndOut is 0(omitted), so there is no variable to be stored. Thus Stata reported no observations
no variables defined
no variables defined
invalid syntax

I want the loop to continue and just skip the coefficients that does not exist, and save all those that exist. Could you help me with that?

[Joseph Coveney]

in the loop there are some regressions where the coefficient of lnYearlyIndOut is 0(omitted)

As I illustrated earlier: trap the conditions when your operation cannot be performed, and whenever those conditions are detected, instruct the loop to continue (skip to the next iteration).

I recommend avoiding the casual use of capture for such a task in production code.

[Johnny YANG]

As I illustrated earlier: trap the conditions when your operation cannot be performed, and whenever those conditions are detected, instruct the loop to continue (skip to the next iteration).

I recommend avoiding the casual use of capture for such a task in production code.

Nice! Thank you!

[Nick Cox]

There might be some commitment to writing your own loops for some other reason, but it may be worth flagging that there are other approaches to repeating regressions over subsets of observations.

For example with rangestat (SSC) by Robert Picard and friends, the example in #4 from Joseph Coveney can be written like this:

. sysuse auto, clear 
(1978 Automobile Data)

. rangestat (reg) gear_ratio mpg, interval(foreign 0 0) 

. tabdisp foreign, c(reg_nobs    reg_r2   b_mpg se_mpg)   

--------------------------------------------------------------------------------------
 Car type |          reg_nobs             reg_r2              b_mpg             se_mpg
----------+---------------------------------------------------------------------------
 Domestic |                52           .3202138          .04007944          .00825853
  Foreign |                22          .21641755          .02089239          .00888933
--------------------------------------------------------------------------------------

rangestat views choice of minimum acceptable sample size differently from some other commands. It won't fall over with too few observations to do what is asked (even zero), but you can always review results later and ignore those based on what you think is too few for comfort or to appease your peers.

[Johnny YANG]

There might be some commitment to writing your own loops for some other reason, but it may be worth flagging that there are other approaches to repeating regressions over subsets of observations.

For example with rangestat (SSC) by Robert Picard and friends, the example in #4 from Joseph Coveney can be written like this:

. sysuse auto, clear
(1978 Automobile Data)

. rangestat (reg) gear_ratio mpg, interval(foreign 0 0)

. tabdisp foreign, c(reg_nobs reg_r2 b_mpg se_mpg)

--------------------------------------------------------------------------------------
Car type | reg_nobs reg_r2 b_mpg se_mpg
----------+---------------------------------------------------------------------------
Domestic | 52 .3202138 .04007944 .00825853
Foreign | 22 .21641755 .02089239 .00888933
--------------------------------------------------------------------------------------

rangestat views choice of minimum acceptable sample size differently from some other commands. It won't fall over with too few observations to do what is asked (even zero), but you can always review results later and ignore those based on what you think is too few for comfort or to appease your peers.

Thank you Nick! This is a very good way! Will note it!

[Johnny YANG]

There might be some commitment to writing your own loops for some other reason, but it may be worth flagging that there are other approaches to repeating regressions over subsets of observations.

For example with rangestat (SSC) by Robert Picard and friends, the example in #4 from Joseph Coveney can be written like this:

. sysuse auto, clear
(1978 Automobile Data)

. rangestat (reg) gear_ratio mpg, interval(foreign 0 0)

. tabdisp foreign, c(reg_nobs reg_r2 b_mpg se_mpg)

--------------------------------------------------------------------------------------
Car type | reg_nobs reg_r2 b_mpg se_mpg
----------+---------------------------------------------------------------------------
Domestic | 52 .3202138 .04007944 .00825853
Foreign | 22 .21641755 .02089239 .00888933
--------------------------------------------------------------------------------------

rangestat views choice of minimum acceptable sample size differently from some other commands. It won't fall over with too few observations to do what is asked (even zero), but you can always review results later and ignore those based on what you think is too few for comfort or to appease your peers.

Hi Nick, I tried the rangestat today, it works really smooth, but the problem is that, I ran the regression by hand, the results are different from the results coming from rangestat

[Nick Cox]

Concrete detail is needed: your data, your code, your results.

[Johnny YANG]

Concrete detail is needed: your data, your code, your results.

egen sic_id=group(sic)
tab sic_id
rangestat (reg) sale nopio, interval(sic_id 1 414)
tabdisp sic_id, c(reg_nobs b_nopio se_nopio)

my data is the fundamental annual data from compustat, 2007-2018.
the results from rangestat is as follows.

group(sic
) reg_nobs b_nopio se_nopio

1 408992 23.583062 .05666472
2 408790 23.582863 .05667833
3 408619 23.582729 .05668994
4 408455 23.582569 .05670099
5 408419 23.582531 .05670342
6 400388 24.219794 .05772926
7 385781 24.304457 .05884326
8 385318 24.304009 .05887762
9 384057 24.303508 .05897171
10 383432 24.303918 .05901932
11 383261 24.303913 .05903223
12 366745 24.310357 .06124795
13 365266 24.364617 .06140111
14 364510 24.364858 .06146397
15 363376 24.382495 .0615424


But if I run the regression when sic_id is 1, the result is as follows,
sale Coef. Std. Err. t P>t [95% Conf. Interval]

nopio -19.72358 4.948205 -3.99 0.000 -29.43763 -10.00954
_cons 764.5555 70.53242 10.84 0.000 626.0901 903.0209

totally different.

[Nick Cox]

Indeed. Your rangestat regressions all pool many identifiers. See how the sample size is always around 400000. You have misread the options for rangestat.

[Johnny YANG]

Indeed. Your rangestat regressions all pool many identifiers. See how the sample size is always around 400000. You have misread the options for rangestat.

so the rangestat is only applicable for small sample, right?

[Nick Cox]

The results for the first identifier are for identifiers 2 up, those for the second 3 up, and so on; not what you want at all. See again #8 and the help. I guess that you want interval(sic_id 0 0).