non r-class program may not set r()

[Zuhumnan Dapel]

Dear All,
I have been trying to estimate a Non Linear Least Square (NLS) IV model using the

bys id: gen runsum_ly = sum(ly)
gen yb_bar = runsum_ly/m
gen x_demean = x - xm
reg ly yb_bar
predict res, r
reg ly res x_demean
predict lyhat
reg x res x_demean
predict xhat
cap nl (y = lyhat*({g}^gap) + {b}*xhat) if m > 1, iter(500) nolog
if e(converge) == 0 | _rc != 0 {
ret sca enlsiv_g = .
ret sca enlsiv_b = .
}
else {
ret sca enlsiv_g = _b[/g]
ret sca enlsiv_b = _b[/b]

but been getting this error message: "non r-class program may not set r()." What could be the reason please?. How do I handle with it?.

Thanks,
Dapel

[daniel klein]

If you want to return something in r(), you need a set up like

program myprogram ,rclass
        ...
        return scalar myscalar = value
end

alternatively, you can use Mata to put contents in r(), as in

mata : st_numscalar(r(myscalar), value)

Best
Daniel

[Zuhumnan Dapel]

Thanks Dan. How do I recode these by on any of your suggestions?

ret sca enlsiv_g = _b[/g]
            ret sca enlsiv_b = _b[/b]
        }

and

ret sca enlscreiv_g = .
            ret sca enlscreiv_b = .

because I got "unexpected end of file" as an error message after using

program myprogram ,rclass
              ret sca enlsiv_g = .
            ret sca enlsiv_b = .
        }
        else {
        program myprogram ,rclass
            ret sca enlsiv_g = _b[/g]
            ret sca enlsiv_b = _b[/b]
        }

[Nick Cox]

The second instance of

program myprogram ,rclass

is illegal (and indeed makes no sense). Indeed, the program makes no sense as there is no if block which you need to give meaning to the else block.

You should back up and tell us what you want to do.

This looks like "premature programming" to me, that is,

1. You are trying to write a slightly complicated program before you understand how to write simple ones.

2. You are writing a program with wired-in variable names when you would be better off writing a do-file.

[Zuhumnan Dapel]

Dear Prof, Thank you for your firm and helpful comments. Here is what I have been wanting to do: to estimate a Non Linear Least Square (NLS) IV model using the

bys id: gen runsum_ly = sum(ly)
gen yb_bar = runsum_ly/m
gen x_demean = x - xm
reg ly yb_bar
predict res, r
reg ly res x_demean
predict lyhat
reg x res x_demean
predict xhat
cap nl (y = lyhat*({g}^gap) + {b}*xhat) if m > 1, iter(500) nolog
if e(converge) == 0 | _rc != 0 {
ret sca enlsiv_g = .
ret sca enlsiv_b = .
}
else {
ret sca enlsiv_g = _b[/g]
ret sca enlsiv_b = _b[/b]

but been getting this error message: "non r-class program may not set r()." What could be the reason please? How do I deal with it?

Thanks,
Dapel

[Nick Cox]

Sorry, but I can't see that you are adding anything to your previous posts. You haven't (yet) shown a need to make this an r-class program. I'd get this working before you try to make it fancier.

[Zuhumnan Dapel]

Sorry Prof. I want to estimate a DPD model, balanced but with gaps. With N=108 and T=24 (actual) and Ti=5 (observed). Is this a bit clearer? First, I used POLS. Here are the

But we are able to account for change only between two periods. How do we address the nonlinearity across the entire 5 periods within 24 years? I don't know if this is plausible.

[Zuhumnan Dapel]

 preparing the data for the regression analysis
keep  id, year, pce, cpi
generate rpce = pce/cpi
generate ln_rpce=log(rpce)
tsset id year

the regression for 1980 & 1985
regress ln_rpce l5.ln_rpce if inrange(year, 1980, 1985), robust
nlcom  _b[l5.ln_rpce]^(1/5)
scalar ar_85_80=_b[l5.ln_rpce]^(1/5)

the regression for 1985 & 1992
regress ln_rpce l7.ln_rpce if inrange(year, 1985, 1992), robust
nlcom  _b[l7.ln_rpce]^(1/7)
scalar ar_85_92=_b[l7.ln_rpce]^(1/7)

the regression for 1992 & 1996
regress ln_rpce l4.ln_rpce if inrange(year, 1992, 1996), robust
nlcom  _b[l4.ln_rpce]^(1/4)
scalar ar_92_96=_b[l4.ln_rpce]^(1/4)


the regression for 1996 & 2004
regress ln_rpce l8.ln_rpce if inrange(year, 1996, 2004), robust
nlcom  _b[l8.ln_rpce]^(1/8)
scalar ar_96_04=_b[l8.ln_rpce]^(1/8)

view the AR coefficients from the four estimations
scalar dir

[Nick Cox]

You have spelled out the substantive background, and someone may be able to help. But I am lost on the connection between your latest post and your earlier questions.

[Zuhumnan Dapel]

OK Prof. Here is the difference: I just switched discussion from one estimator, NLS-IV in the initial post to another, POLS in the most recent post. Thanks

[daniel klein]

I do not fully get why you think the scalars you are trying to obtain should be put into r(). The code you show in #8 has statements like

scalar name = value

I cannot see any obvious reason why, in the code in your initial post, you do not simply replace

ret sca name = value

with

scalar name = value

just the way you do with your pooled regression models. Why do you want to put them in r() so desperately?

Best
Daniel

[Zuhumnan Dapel]

After exempting some lines in the initially used codes, I got a some results. I'm reasonable OK now. Thank you all very much!