Replication: qreg gives very different results to authors' program

Thu Hoang

Join Date: Oct 2023
Posts: 2

Replication: qreg gives very different results to authors' program

20 Oct 2023, 01:08

Hi everyone, I'm new to this forum, so if there is any issue with my post, please feel free to let me know!

I am doing a replication exercise for this paper, "Estimating the Impact of Microcredit on Those Who Take It Up: Evidence from a Randomized Experiment in Morocco" (https://www.aeaweb.org/articles?id=10.1257/app.20130535) - data and the replication package is available here: https://www.openicpsr.org/openicpsr/...ersion/V1/view.

To replicate figure 1 in the paper, I ran qreg2 as follows:

Code:

set maxiter 16000
xi: qreg2 profit_total treatment $householdcontrols $respondentcontrols i.paire if samplemodel == 1, quantile(.9) cluster(demi_paire)

But the results I got are quite different from the results in the paper - the authors wrote their own program for the quantile treatment effect regression.

Click image for larger version

Name: qreg.PNG
Views: 1
Size: 9.6 KB
ID: 1730964

The authors' program for quantile treatment effect regression:

Code:

tab1 paire,gen(p)

capture program drop NewregqM
prog def NewregqM, rclass
args var
clear matrix
sca drop _all
local sca var Varb Varbs n
count if treatment==1
sca ntot1=r(N)
count if treatment==0
sca ntot0=r(N)
matrix T=J(81,1,1)#(0\1)
matrix P=I(81)#J(2,1,1)
matrix X=(T,P)
matrix Isig=J(162,162,0)
foreach i of numlist 1 2 3 4 5 {
        matrix Sig`i'=J(162,162,0)
        }
matrix  q_0=J(5,1,0)
matrix  q_1=J(5,1,0)
foreach nump of numlist 1/81 {
        foreach val of numlist 0 1 {
                quietly centile `var' if p`nump'==1 & treatment==`val',centile(10 25 50 75 90) level(95)
                sca n`val'=r(N)
                matrix q_`val'=q_`val'+n`val'*(r(c_1)\r(c_2)\r(c_3)\r(c_4)\r(c_5))/ntot`val'
                foreach i of numlist 1 2 3 4 5 {
                        matrix Isig[(`nump'-1)*2+`val'+1,(`nump'-1)*2+`val'+1]=n`val'
                        matrix theta`i'=nullmat(theta`i')\r(c_`i')
                        sca var`i'=((r(ub_`i')-r(lb_`i'))/(invnormal((1+.95)/2)-invnormal((1-.95)/2) ))
                        matrix Vecvar`i'=nullmat(Vecvar`i')\var`i'
                        matrix Sig`i'[(`nump'-1)*2+`val'+1,(`nump'-1)*2+`val'+1]=var`i'
                        }
                }
        }

foreach i of numlist 1 2 3 4 5 {
        matrix b=invsym(X'*Isig*X)*X'*Isig*theta`i'
        matrix b1=b[1,1]
        disp `i'
        matrix list b1
        matrix uhat=theta`i'-X*b

        matrix uhat2=hadamard(uhat,uhat)
        matrix uhatres=uhat2-Vecvar`i'
        matrix RegT=(J(162,1,1),T)
        matrix Sig01=invsym(RegT'*RegT)*RegT'*uhatres
        matrix Dsig01=J(2,2,0)
        matrix Dsig01[1,1]=Sig01[1,1]
        matrix Dsig01[2,2]=Sig01[1,1]+Sig01[2,1]
        matrix Sig2=Sig`i'+I(81)#Dsig01
        matrix Isig2=invsym(Sig2)

        matrix Sig3=J(162,162,0)
        foreach nump of numlist 1/81 {
        
                foreach val of numlist 0 1 {
                        matrix Sig3[(`nump'-1)*2+`val'+1,(`nump'-1)*2+`val'+1]=uhat2[(`nump'-1)*2+`val'+1,1]
                        }
                }


        matrix b2=invsym(X'*Isig2*X)*X'*Isig2*theta`i'
        matrix Varb2R=invsym(X'*Isig2*X)*X'*Isig2*Sig3*Isig2*X*invsym(X'*Isig2*X)
        sca b2=b2[1,1]
        sca Varb2R=Varb2R[1,1]
        sca sb2R=sqrt(Varb2R)
        matrix res=nullmat(res)\(b2,sb2R,b2/sb2R,2*(1-normal(abs(b2/sb2R))))
        }
matrix res=((10\25\50\75\90),(res,q_0,((res[1,1]/q_0[1,1])\(res[2,1]/q_0[2,1])\(res[3,1]/q_0[3,1])\(res[4,1]/q_0[4,1])\(res[5,1]/q_0[5,1]))))


end
  
gen id=.
gen vars=""
gen N=.
gen n_surveyed=.
gen mean_surveyed=.


gen coef10=.
gen coef25=.
gen coef50=.
gen coef75=.
gen coef90=.

gen se10=.
gen se25=.
gen se50=.
gen se75=.
gen se90=.

global variables_quantile profit_total assets_total consumption income income_dep

Below is a data extract of the variables used in the regression:

Code:

* Example generated by -dataex-. For more info, type help dataex
clear
input float(profit_total treatment paire)
   6200 1 10
      0 0 66
      0 0 66
      0 0 66
   -151 1 19
      0 0 66
      0 0 66
      0 0 66
   3580 0 66
      0 0 66
  19200 0 66
      0 1 32
      0 0 10
  -1200 1 32
  21940 0 66
      0 0 66
      0 0 66
      0 1 13
  -2240 0 66
      0 0 66
      0 0 66
   4800 1 32
      0 0 66
      0 0 66
  -2100 0 76
      0 0 66
   2482 1  9
  21478 0 66
      0 0 66
      0 0 66
      0 0 66
      0 0 66
      0 0 10
    680 1 19
      0 0 32
      0 1 80
  79920 0 66
      0 0 66
   8560 1 80
   1590 0 66
      0 0 66
 -30040 0 32
      0 0 10
      0 0 10
 -17007 1 80
      0 1 66
      0 1 66
      0 1 66
      0 1 66
14135.5 1 80
   3880 1 66
 -12600 0 32
      0 1 66
   1550 1 66
      0 0 32
    300 1 66
   3480 1 66
      0 1 66
   2400 0 32
      0 1 66
      0 1 66
      0 1 31
      0 1 66
 1227.5 0 13
      0 1 31
      0 1 31
      0 1 31
      0 1 66
      0 1 31
      0 1 31
      0 1 66
      0 1 31
      0 1 31
      0 1 31
      0 1 31
  -4900 1 31
      0 1 66
      0 1 31
    250 1 31
  -2490 1 31
      0 1 31
      0 1 31
   -428 1 31
    955 1 31
   2300 1 31
  84000 1 66
      0 1 66
  48024 0 76
      0 1 66
  21400 0 31
      0 1 66
  -1880 0 31
      0 1 66
    -80 0 31
      0 1 66
      0 1 66
   -100 0 31
  -2550 0 31
   1000 0 31
   2000 0 31
end

Household control variables:

Code:

* Example generated by -dataex-. For more info, type help dataex
clear
input float(members_resid_bl nadults_resid_bl head_age_bl act_livestock_bl act_business_bl borrowed_total_bl members_resid_d_bl nadults_resid_d_bl head_age_d_bl act_livestock_d_bl act_business_d_bl borrowed_total_d_bl)
 7  2 75 1 0 0 0 0 0 0 0 0
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 6  4 52 1 0 1 0 0 0 0 0 0
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 6  4 60 1 0 0 0 0 0 0 0 0
 2  2 27 1 0 0 0 0 0 0 0 0
 6  4 49 1 0 0 0 0 0 0 0 0
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 7  5 80 1 1 0 0 0 0 0 0 0
 0  0  0 0 0 0 1 1 1 1 1 1
 1  1 22 1 0 1 0 0 0 0 0 0
 7  2 40 0 0 1 0 0 0 0 0 0
 3  2 34 0 0 0 0 0 0 0 0 0
 7  6 45 0 0 1 0 0 0 0 0 0
 3  2 59 0 1 0 0 0 0 0 0 0
 6  4 55 0 0 0 0 0 0 0 0 0
 3  2 23 0 0 1 0 0 0 0 0 0
 0  0  0 0 0 0 1 1 1 1 1 1
 3  2 28 0 1 1 0 0 0 0 0 0
 5  2 43 0 0 1 0 0 0 0 0 0
 5  2 25 0 0 0 0 0 0 0 0 0
 6  2 38 0 0 1 0 0 0 0 0 0
 3  2 37 0 0 0 0 0 0 0 0 0
 1  1 40 0 0 0 0 0 0 0 0 0
 6  5 45 1 0 1 0 0 0 0 0 0
 5  2  1 0 1 0 0 0 0 0 0 0
 1  1 70 0 0 1 0 0 0 0 0 0
 4  2 37 0 1 1 0 0 0 0 0 0
 6  2 35 0 0 1 0 0 0 0 0 0
 1  1 75 1 0 1 0 0 0 0 0 0
 9  5 55 0 0 1 0 0 0 0 0 0
 7  4 45 0 0 1 0 0 0 0 0 0
11  9 47 1 1 0 0 0 0 0 0 0
10  7 62 1 0 1 0 0 0 0 0 0
 7  7 47 0 1 0 0 0 0 0 0 0
 1  1 43 1 0 1 0 0 0 0 0 0
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 1  1 45 1 0 1 0 0 0 0 0 0
 0  0  0 0 0 0 1 1 1 1 1 1
 6  2 38 0 1 0 0 0 0 0 0 0
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 2  2 66 1 0 0 0 0 0 0 0 0
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 3  2 30 0 1 0 0 0 0 0 0 0
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 0  0  0 0 0 0 1 1 1 1 1 1
 3  2 24 0 1 0 0 0 0 0 0 0
 0  0  0 0 0 0 1 1 1 1 1 1
 7  6 50 1 0 1 0 0 0 0 0 0
 0  0  0 0 0 0 1 1 1 1 1 1
 7  2 38 0 0 1 0 0 0 0 0 0
 5  5 77 0 0 1 0 0 0 0 0 0
 7  3 40 0 0 1 0 0 0 0 0 0
 5  3 45 0 0 0 0 0 0 0 0 0
 5  2 44 1 0 1 0 0 0 0 0 0
 6  4 60 0 0 1 0 0 0 0 0 0
 8  6 29 1 0 0 0 0 0 0 0 0
11 10 57 0 0 1 0 0 0 0 0 0
 7  4 44 0 0 1 0 0 0 0 0 0
end

Respondent control variables:

Code:

* Example generated by -dataex-. For more info, type help dataex
clear
input float(ccm_resp_activ other_resp_activ ccm_resp_activ_d other_resp_activ_d)
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
1 0 0 0
0 0 0 0
0 0 1 1
0 0 0 0
0 0 1 1
0 0 1 1
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
1 0 0 0
1 0 0 0
0 1 0 0
0 0 1 1
0 0 0 0
0 0 0 0
0 0 1 1
0 0 0 0
1 0 0 0
0 0 1 1
0 0 0 0
1 0 0 0
0 0 1 1
0 0 1 1
0 0 1 1
0 0 1 1
0 0 0 0
1 0 0 0
0 0 0 0
0 0 1 1
0 0 0 0
0 0 1 1
0 0 1 1
0 0 0 0
0 0 1 1
0 0 0 0
0 0 0 0
0 0 0 0
0 0 1 1
1 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 1 1
0 0 1 1
0 0 0 0
0 0 0 0
0 1 0 0
0 1 0 0
0 0 0 0
1 0 0 0
0 0 1 1
0 0 0 0
0 0 1 1
0 0 0 0
0 0 0 0
0 0 1 1
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 1 1
0 1 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 1 1
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 1 1
0 0 0 0
0 0 1 1
0 0 0 0
0 0 0 0
0 0 0 0
0 0 1 1
0 0 0 0
0 0 0 0
0 0 1 1
0 0 0 0
0 1 0 0
0 0 0 0
1 0 0 0
end

Thank you very much for your help.

Tags: None

Announcement

Replication: qreg gives very different results to authors' program