Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# st: goodness-of-fit McElroy's R-squared for NLSUR

 From "Pedro Tiago Carvalho" <[email protected]> To <[email protected]> Subject st: goodness-of-fit McElroy's R-squared for NLSUR Date Tue, 13 Mar 2012 19:56:03 -0000

```Dear Miss or Mrs,

I would like to obtain the overall goodness-of-fit McElroy's R-squared for
the whole NLSUR model (with several equations) on STATA  11.
Can someone tell me how I can obtain it? How can I calculate it?

I'm trying to compute McElroy's R-squared from the program below

"
clear all
use "C:\Users\Pedro\Desktop\r2reg3.dta", clear
nlsur (y1 = {a} + {b1}*x1+ {b2}*x3) (y2 = {a} + {b1}^x2+ {b3}*x4)
r2reg3
"

but gives the following error:

"
. r2reg3
equation() misspecified
r(303);
"

Can someone help me?

Thank you so much.

Pedro Carvalho

*********************************************************************
*! r2reg3 V3.0 28oct2011
*! Assistant Professor
*! Agricultural Research Center - Agricultural Economics Research Institute
- Egypt
*! Email: [email protected]

program define r2reg3 , rclass
version 10
tempvar E`var' Yb_Y`var' Sig2 YMAT R4S SYY
tempname Ybv Yb Yv E`var' Yb_Y`var' Sig2 Omega YM RS1 RS2 RS3 RS4 RS5
tempname Y Ev E W Omega IMn Dt YMAT Mat1 Mat2 Mat3 R2Mat Trm RSQ
tempname SSE1 SSE2 SSE3 SSE4 MSS1 MSS2 MSS3 MSS4 SST1 SST2 SST3 SST4
marksample touse
{
preserve
scalar N=e(N)
scalar K=e(k)
scalar Q=e(k_eq)
scalar DFF=(Q*N-K)/(K-Q)
scalar DFChi=(K-Q)
local N=e(N)
local K=e(k)
local Q=e(k_eq)
local DF1=K-Q
local DF2=Q*N
local DFChi=(K-Q)
mat `Omega'= e(Sigma)
local varlist `e(depvar)'
foreach var of local varlist {
qui predict `E'`var' if `touse' , equation(`var') res
qui summarize `var' if `touse'
qui gen `Yb_Y'`var' = `var' - `r(mean)' if `touse'
}
mkmat `E'`var'* if `touse' , matrix(`E')
mkmat `e(depvar)' if `touse' , matrix(`Y')
mkmat `Yb_Y'`var'*   if `touse'  , matrix(`Yb')
svmat `Y' , name(`YMAT')
matrix `Ybv'=vec(`Yb')
matrix `Yv'=vec(`Y')
matrix `Ev'=vec(`E')
matrix `W'=inv(`Omega')#I(N)
matrix `Sig2'=det(`Omega')
scalar Sig2=`Sig2'[1,1]
matrix `SSE1'=det(`E''*`E')
matrix `SSE2'=`Ev''*`W'*`Ev'
matrix `SSE3'=`Ev''*`Ev'
matrix `SST1'=det(`Yb''*`Yb')
matrix `SST2'=`Ybv''*`W'*`Ybv'
matrix `SST3'=`Ybv''*`Ybv'
scalar v=1/N
matrix `IMn'=J(N,N,v)
matrix `Dt'=I(N)-`IMn'
forvalues i =1/`Q' {
scalar R2_`i'=e(r2_`i')
}
qui egen `SYY' = varprod(`Yb_Y'`var'*) if `touse'
qui sum `SYY' if `touse' , meanonly
scalar SY=r(mean)
matrix `Trm'=trace(inv(`Omega'))*SY
scalar R5=1-(Q/`Trm'[1,1])
qui gen double `R4S' = . if `touse'
scalar R4 = 0
forvalues i =1/`Q' {
mkmat `YMAT'`i' if `touse' , matrix(`YM')
matrix `Mat1'=`YM''*`Dt'*`YM'
matrix `Mat2'=`Yv''*I(`Q')#`Dt'*`Yv'
scalar `Mat3'=`Mat1'[1,1]/`Mat2'[1,1]
matrix `R2Mat'=R2_`i'*`Mat3'
scalar R4`i'=`R2Mat'[1,1]
qui replace `R4S' = R4`i' if `touse'
qui sum `R4S' if `touse' , meanonly
qui replace  `R4S' = r(mean) if `touse'
scalar R4 = R4 + r(mean)
}
forvalues i = 1/3 {
matrix `MSS`i''=`SST`i''-`SSE`i''
matrix R`i'=1-(`SSE`i''*inv(`SST`i''))
scalar R`i'=R`i'[1,1]
}
forvalues i = 1/5 {
scalar F`i'=R`i'/(1-R`i')*DFF
scalar Chi`i'= -N*(log(1-R`i'))
scalar PChi`i'= chiprob(DFChi, Chi`i')
scalar  PF`i'= fprob(`DF1',`DF2', F`i')
}
scalar LSig2=log(Sig2)
scalar LLF=-(N*Q/2)*(1+log(2*_pi))-(N/2*abs(LSig2))
restore
}
di as txt
"{bf:{err:========================================================}}"
di as txt "{bf:{err:* Overall System R2 - Adjusted R2 - F Test - Chi2 Test
*}}"
di as txt
"{bf:{err:========================================================}}"
mat `RSQ'=`RS1' \ `RS2' \ `RS3' \ `RS4' \ `RS5'
mat rownames `RSQ' = Berndt McElroy Judge Dhrymes Greene
mat colnames `RSQ' = R2 Adj_R2 F "P-Value" Chi2 "P-Value"
matlist `RSQ', twidth(8) border(all) lines(columns) rowtitle(Name)
format(%8.4f)
di as txt "  Number of Parameters         =" as res _col(35) %10.0f K
di as txt "  Number of Equations          =" as res _col(35) %10.0f Q
di as txt "  Degrees of Freedom F-Test    =" as res _col(39) "(" K-Q ", "
Q*N ")"
di as txt "  Degrees of Freedom Chi2-Test =" as res _col(35) %10.0f DFChi
di as txt "  Log Determinant of Sigma     =" as res _col(35) %10.4f LSig2
di as txt "  Log Likelihood Function      =" as res _col(35) %10.4f LLF
return scalar f_df1 = `DF1'
return scalar f_df2 = `DF2'
return scalar chi_df = DFChi
return scalar k=`K'
return scalar k_eq=`Q'
return scalar N=`N'
return scalar lsig2=LSig2
return scalar llf=LLF
return scalar chi_g = Chi5
return scalar chi_d = Chi4
return scalar chi_b = Chi3
return scalar chi_j = Chi2
return scalar chi_m = Chi1
return scalar f_g = F5
return scalar f_d = F4
return scalar f_b = F3
return scalar f_j = F2
return scalar f_m = F1
return scalar r2_g = R5
return scalar r2_d = R4
return scalar r2_b = R3
return scalar r2_j = R2
return scalar r2_m = R1
end
*********************************************************************

```

Attachment: r2reg3.dta
Description: Binary data