Stata学习：如何进行PSM-DID？
momo
一、详细解读
0. 数据导入
导入自带数据集：
webuse nlswork.dta,clear1. 前期处理
设置面板：
xtset idcode year, delta(1)
panel variable: idcode (unbalanced)
time variable: year, 68 to 88, but with gaps
delta: 1 unit
变量加工与描述性统计：
xtdescribe
g age2= age^2
g ttl_exp2=ttl_exp^2
g tenure2=tenure^2
global xlist "grade age age2 ttl_exp ttl_exp2 tenure tenure2 not_smsa south race"
sum ln_w $xlist 2. 传统DID
在这个数据集中，政策执行时间为1977年，因此设置：
g time = (year >= 77) & !missing(year) 而政策执行地方为idcode大于2000的地方，因此设置：
g treated = (idcode >2000)&!missing(idcode) 生成政策变量：
g did = time*treated 采用OLS估计：
reg ln_w did time treated $xlist
Source | SS df MS Number of obs = 28,091
-------------+---------------------------------- F(13, 28077) = 1295.76
Model | 2405.08413 13 185.006471 Prob > F = 0.0000
Residual | 4008.77975 28,077 .142778066 R-squared = 0.3750
-------------+---------------------------------- Adj R-squared = 0.3747
Total | 6413.86388 28,090 .228332641 Root MSE = .37786
------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
did | .0124603 .0094099 1.32 0.185 -.0059836 .0309042
time | -.0694754 .0097324 -7.14 0.000 -.0885513 -.0503996
treated | -.0142815 .0078374 -1.82 0.068 -.0296431 .0010801
grade | .0633445 .0010293 61.54 0.000 .061327 .065362
age | .0484968 .0036834 13.17 0.000 .0412771 .0557165
age2 | -.0008259 .0000583 -14.18 0.000 -.0009401 -.0007117
ttl_exp | .0250849 .002395 10.47 0.000 .0203906 .0297793
ttl_exp2 | .0002872 .0001266 2.27 0.023 .000039 .0005353
tenure | .0460783 .001968 23.41 0.000 .0422209 .0499356
tenure2 | -.0019604 .0001341 -14.62 0.000 -.0022233 -.0016976
not_smsa | -.1686808 .0051713 -32.62 0.000 -.1788169 -.1585447
south | -.1017625 .0055672 -18.28 0.000 -.1126745 -.0908504
race | -.0528909 .0049329 -10.72 0.000 -.0625597 -.0432221
_cons | .1400083 .0535917 2.61 0.009 .0349661 .2450506
------------------------------------------------------------------------------
固定效应模型：
xtreg ln_w did time treated $xlist i.year, fe
Fixed-effects (within) regression Number of obs = 28,091
Group variable: idcode Number of groups = 4,697
R-sq: Obs per group:
within = 0.1781 min = 1
between = 0.3576 avg = 6.0
overall = 0.2666 max = 15
F(23,23371) = 220.25
corr(u_i, Xb) = 0.1840 Prob > F = 0.0000
------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
did | .019761 .008746 2.26 0.024 .0026182 .0369037
time | -.3296955 .2012367 -1.64 0.101 -.7241326 .0647416
treated | 0 (omitted)
grade | 0 (omitted)
age | .0665613 .0105138 6.33 0.000 .0459536 .0871689
age2 | -.0009369 .0000616 -15.20 0.000 -.0010577 -.000816
ttl_exp | .0392161 .003072 12.77 0.000 .0331947 .0452375
ttl_exp2 | -.0001047 .0001352 -0.77 0.439 -.0003697 .0001603
tenure | .0338566 .0018579 18.22 0.000 .0302151 .0374981
tenure2 | -.0018164 .000126 -14.42 0.000 -.0020633 -.0015695
not_smsa | -.086641 .0095117 -9.11 0.000 -.1052846 -.0679974
south | -.0597302 .0109246 -5.47 0.000 -.0811431 -.0383173
race | 0 (omitted)
year |
69 | .0423343 .015528 2.73 0.006 .0118984 .0727702
70 | -.0339863 .022938 -1.48 0.138 -.0789464 .0109737
71 | -.0299658 .0318704 -0.94 0.347 -.092434 .0325023
72 | -.0584678 .0412716 -1.42 0.157 -.1393627 .0224272
73 | -.0961513 .0508379 -1.89 0.059 -.195797 .0034943
75 | -.1513149 .0698104 -2.17 0.030 -.2881479 -.0144818
77 | .1558488 .1129486 1.38 0.168 -.0655379 .3772354
78 | .1417381 .1027217 1.38 0.168 -.0596031 .3430793
80 | .0826885 .0834407 0.99 0.322 -.0808607 .2462377
82 | .0268096 .0638067 0.42 0.674 -.0982557 .151875
83 | .0103961 .054166 0.19 0.848 -.0957727 .116565
85 | .007777 .0347411 0.22 0.823 -.0603178 .0758717
87 | -.0223798 .0162763 -1.37 0.169 -.0542825 .0095229
88 | 0 (omitted)
_cons | .5038214 .1945871 2.59 0.010 .1224179 .8852249
-------------+----------------------------------------------------------------
sigma_u | .3532234
sigma_e | .2898202
rho | .59764928 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(4696, 23371) = 6.58 Prob > F = 0.00003. PSM-DID
一般匹配
定义种子：
set seed 0001生成随机数并随机整理：
g tmp = runiform()
sort tmp 近邻匹配：
psmatch2 treated $xlist, out(ln_w) logit ate neighbor(1) common caliper(.05) ties
Logistic regression Number of obs = 28,091
LR chi2(10) = 10093.80
Prob > chi2 = 0.0000
Log likelihood = -13636.63 Pseudo R2 = 0.2701
------------------------------------------------------------------------------
treated | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
grade | -.0342869 .0070082 -4.89 0.000 -.0480227 -.0205511
age | .1726049 .0227237 7.60 0.000 .1280673 .2171424
age2 | -.0026816 .0003693 -7.26 0.000 -.0034054 -.0019577
ttl_exp | -.0681925 .0155242 -4.39 0.000 -.0986193 -.0377657
ttl_exp2 | .0040423 .0008363 4.83 0.000 .0024031 .0056815
tenure | -.0138884 .0128354 -1.08 0.279 -.0390453 .0112686
tenure2 | .0002263 .0008671 0.26 0.794 -.0014733 .0019259
not_smsa | .4024444 .0355904 11.31 0.000 .3326885 .4722002
south | 2.865226 .0409688 69.94 0.000 2.784928 2.945523
race | .7655284 .034012 22.51 0.000 .6988662 .8321907
_cons | -3.400584 .3254663 -10.45 0.000 -4.038487 -2.762682
------------------------------------------------------------------------------
----------------------------------------------------------------------------------------
Variable Sample | Treated Controls Difference S.E. T-stat
----------------------------+-----------------------------------------------------------
ln_wage Unmatched | 1.62754443 1.75722818 -.129683748 .005816206 -22.30
ATT | 1.62798811 1.70873922 -.080751107 .021132148 -3.82
ATU | 1.75707925 1.75230645 -.004772802 . .
ATE | -.051698588 . .
----------------------------+-----------------------------------------------------------
Note: S.E. does not take into account that the propensity score is estimated.
psmatch2: | psmatch2: Common
Treatment | support
assignment | Off suppo On suppor | Total
-----------+----------------------+----------
Untreated | 2 10,733 | 10,735
Treated | 20 17,336 | 17,356
-----------+----------------------+----------
Total | 22 28,069 | 28,091 检验协变量在处理组与控制组之间是否平衡：
pstest $xlist, both graph
----------------------------------------------------------------------------------------
Unmatched | Mean %reduct | t-test | V(T)/
Variable Matched | Treated Control %bias |bias| | t p>|t| | V(C)
--------------------------+----------------------------------+---------------+----------
grade U | 12.341 12.853 -22.3 | -18.01 0.000 | 1.18*
M | 12.348 12.645 -12.9 42.1 | -12.03 0.000 | 1.15*
| | |
age U | 29.223 28.848 5.6 | 4.56 0.000 | 0.94*
M | 29.216 30.875 -24.7 -342.5 | -23.65 0.000 | 1.05*
| | |
age2 U | 897.8 878.68 4.7 | 3.86 0.000 | 0.95*
M | 897.33 994.81 -24.1 -409.7 | -22.80 0.000 | 1.01
| | |
ttl_exp U | 6.2403 6.2447 -0.1 | -0.08 0.938 | 1.05*
M | 6.2334 6.9942 -16.4-17031.2 | -14.80 0.000 | 0.92*
| | |
ttl_exp2 U | 61.008 60.008 1.2 | 1.01 0.312 | 1.03
M | 60.849 72.756 -14.8 -1090.6 | -12.91 0.000 | 0.79*
| | |
tenure U | 3.0891 3.1782 -2.4 | -1.93 0.053 | 0.96*
M | 3.0899 2.9777 3.0 -25.9 | 2.87 0.004 | 1.10*
| | |
tenure2 U | 23.41 24.482 -1.9 | -1.60 0.110 | 0.90*
M | 23.423 21.518 3.5 -77.7 | 3.41 0.001 | 1.14*
| | |
not_smsa U | .32945 .20689 27.9 | 22.36 0.000 | .
M | .32897 .32066 1.9 93.2 | 1.65 0.099 | .
| | |
south U | .618 .07219 140.2 | 107.35 0.000 | .
M | .61756 .61802 -0.1 99.9 | -0.09 0.930 | .
| | |
race U | 1.3912 1.1631 50.3 | 39.53 0.000 | 1.91*
M | 1.3896 1.3763 2.9 94.2 | 2.43 0.015 | 1.05*
| | |
----------------------------------------------------------------------------------------
* if variance ratio outside [0.97; 1.03] for U and [0.97; 1.03] for M
-----------------------------------------------------------------------------------
Sample | Ps R2 LR chi2 p>chi2 MeanBias MedBias B R %Var
-----------+-----------------------------------------------------------------------
Unmatched | 0.272 10149.10 0.000 25.7 5.2 145.8* 3.70* 88
Matched | 0.019 932.83 0.000 10.4 8.2 33.1* 1.16 88
-----------------------------------------------------------------------------------
* if B>25%, R outside [0.5; 2]
去掉不满足共同区域假定的观测值：
gen common=_support
drop if common == 0 根据上面匹配好的数据，进行DID。
首先是OLS：
reg ln_w did time treated $xlist
Source | SS df MS Number of obs = 28,069
-------------+---------------------------------- F(13, 28055) = 1294.85
Model | 2402.32966 13 184.794589 Prob > F = 0.0000
Residual | 4003.8761 28,055 .142715241 R-squared = 0.3750
-------------+---------------------------------- Adj R-squared = 0.3747
Total | 6406.20576 28,068 .228238769 Root MSE = .37778
------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
did | .0130378 .0094101 1.39 0.166 -.0054064 .0314821
time | -.069613 .0097317 -7.15 0.000 -.0886876 -.0505384
treated | -.0145157 .007837 -1.85 0.064 -.0298765 .0008452
grade | .0635306 .0010326 61.53 0.000 .0615067 .0655545
age | .0487295 .0036856 13.22 0.000 .0415054 .0559535
age2 | -.0008306 .0000583 -14.24 0.000 -.0009449 -.0007163
ttl_exp | .0246821 .0023991 10.29 0.000 .0199798 .0293843
ttl_exp2 | .0003217 .0001271 2.53 0.011 .0000726 .0005708
tenure | .0461767 .0019693 23.45 0.000 .0423168 .0500365
tenure2 | -.0019786 .0001343 -14.73 0.000 -.0022419 -.0017154
not_smsa | -.1681968 .0051738 -32.51 0.000 -.1783376 -.158056
south | -.101765 .0055661 -18.28 0.000 -.1126748 -.0908551
race | -.0534454 .0049452 -10.81 0.000 -.0631381 -.0437526
_cons | .1361316 .0536086 2.54 0.011 .0310562 .2412069
------------------------------------------------------------------------------然后是固定效应模型：
xtreg ln_w did time treated $xlist i.year, fe
Fixed-effects (within) regression Number of obs = 28,069
Group variable: idcode Number of groups = 4,696
R-sq: Obs per group:
within = 0.1782 min = 1
between = 0.3585 avg = 6.0
overall = 0.2678 max = 15
F(23,23350) = 220.19
corr(u_i, Xb) = 0.1861 Prob > F = 0.0000
------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
did | .0199922 .0087501 2.28 0.022 .0028414 .037143
time | -.3259785 .2014472 -1.62 0.106 -.7208281 .0688712
treated | 0 (omitted)
grade | 0 (omitted)
age | .0664258 .0105214 6.31 0.000 .0458031 .0870485
age2 | -.0009367 .0000617 -15.18 0.000 -.0010576 -.0008158
ttl_exp | .0388635 .0030813 12.61 0.000 .0328238 .0449031
ttl_exp2 | -.0000845 .0001359 -0.62 0.534 -.0003508 .0001819
tenure | .0339768 .0018608 18.26 0.000 .0303295 .0376241
tenure2 | -.0018292 .0001263 -14.48 0.000 -.0020768 -.0015817
not_smsa | -.0865921 .0095179 -9.10 0.000 -.1052478 -.0679365
south | -.0600416 .0109379 -5.49 0.000 -.0814807 -.0386026
race | 0 (omitted)
year |
69 | .0425357 .0155411 2.74 0.006 .0120741 .0729974
70 | -.0332855 .0229601 -1.45 0.147 -.0782887 .0117178
71 | -.0290703 .031901 -0.91 0.362 -.0915985 .0334578
72 | -.0573452 .0413132 -1.39 0.165 -.1383217 .0236314
73 | -.0947357 .0508901 -1.86 0.063 -.1944837 .0050123
75 | -.1496357 .0698826 -2.14 0.032 -.2866102 -.0126612
77 | .1540973 .1130656 1.36 0.173 -.0675187 .3757134
78 | .140131 .1028285 1.36 0.173 -.0614196 .3416817
80 | .081445 .0835287 0.98 0.330 -.0822767 .2451668
82 | .0260429 .0638735 0.41 0.683 -.0991533 .151239
83 | .0097373 .0542219 0.18 0.857 -.0965412 .1160159
85 | .007336 .0347776 0.21 0.833 -.0608304 .0755023
87 | -.0225526 .0162929 -1.38 0.166 -.0544878 .0093826
88 | 0 (omitted)
_cons | .5068104 .1947252 2.60 0.009 .1251361 .8884847
-------------+----------------------------------------------------------------
sigma_u | .3531224
sigma_e | .28988636
rho | .59740195 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(4695, 23350) = 6.56 Prob > F = 0.0000逐年匹配
这里采用Xu等（2022）的思路：
- Xu, L., et al. (2022). How has China's low-carbon city pilot policy influenced its CO2 abatement costs? Analysis from the perspective of the shadow price
- Appendix B. Supplementary data 【数据+Stata+Matlab】
*（1）matching data
use psmdata.dta, clear
forvalue i = 2003/2018{
preserve
capture {
keep if year == `i'
set seed 10101
gen ranorder = runiform()
sort ranorder
psmatch2 treated $xlist , outcome(lnMAC) logit neighbor(2) ///
ties common ate caliper(0.05)
save `i'.dta, replace
restore
clear all
use 2003.dta, clear
forvalue k =2004/2018 {
capture {
append using `k'.dta
save ybydata.dta, replace
**(2) regression
use ybydata.dta, clear
*- PSM-DID3（samples that meet the Common Support Assumption ）
reghdfe lnMAC LCC $xlist if _support == 1, $regopt //Column (3) of Table 5
outreg2 using psm.rtf,bdec(4) sdec(4) append
*- PSM-DID4（frequency-weighted regression）
gen weight = _weight * 2
replace weight = 1 if treated == 1 & _weight != .
reghdfe lnMAC LCC $xlist [fweight = weight], $regopt //Column (4) of Table 5
outreg2 using psm.rtf,bdec(4) sdec(4) append4. DID的条件检验
4.1 共同趋势假设
生成年度虚拟变量
tab year, g(yrdum)
interview |
year | Freq. Percent Cum.
------------+-----------------------------------
68 | 1,375 4.82 4.82
69 | 1,232 4.32 9.14
70 | 1,686 5.91 15.05
71 | 1,851 6.49 21.53
72 | 1,693 5.93 27.47
73 | 1,981 6.94 34.41
75 | 2,141 7.50 41.91
77 | 2,171 7.61 49.52
78 | 1,964 6.88 56.40
80 | 1,847 6.47 62.88
82 | 2,085 7.31 70.18
83 | 1,987 6.96 77.15
85 | 2,085 7.31 84.45
87 | 2,164 7.58 92.04
88 | 2,272 7.96 100.00
------------+-----------------------------------
Total | 28,534 100.00生成政策实行前的那些年份与处理虚拟变量的交互项
forval v=1/7{
g treated`v'=yrdum`v'*treated
}做三个回归：
- 第一列：没有控制变量
- 第二列：如果did依然显著，且treated*这些政策施行前年份交互项并不显著，那就好
- 第三列：工会会影响这个处理组和控制组的共同趋势，因此看看union=0的情形
xtreg ln_w did treated* i.year ,fe
outreg2 using result_1.doc,replace tstat bdec(3) tdec(2) ctitle(ln_w,Full-sample)addtext(Controls, No,Year Effect, Yes) keep(did treated*)
xtreg ln_w did treated* $xlist i.year ,fe
outreg2 using result_1.doc,append tstat bdec(3) tdec(2) ctitle(ln_w,Full-sample)addtext(Controls, Yes,Year Effect, Yes) keep(did treated*)
xtreg ln_w did treated* $xlist i.year if union!=1 ,fe
outreg2 using result_1.doc,append tstat bdec(3) tdec(2) ctitle(ln_w,union=0)addtext(Controls, Yes,Year Effect, Yes) keep(did treated*)
4.2 政策干预时间的随机性
将政策执行时间分别提前到1975年和1976年，重新构造DID变量：
g time1 = (year >= 75) & !missing(year)
g treated_1= (idcode >2000)&!missing(idcode)
g did1 = time1*treated_1
g time2 = (year >= 76) & !missing(year)
g treated_2= (idcode >2000)&!missing(idcode)
g did2 = time2*treated_2 回归结果：
xtreg ln_w did1 $xlist i.year,fe
outreg2 using result_2.doc,replace tstat bdec(3) tdec(2) ctitle(ln_w,Full-sample)addtext(Controls, Yes,Year Effect, Yes) keep(did*)
xtreg ln_w did2 $xlist i.year,fe
outreg2 using result_2.doc,append tstat bdec(3) tdec(2) ctitle(ln_w,Full-sample)addtext(Controls, Yes,Year Effect, Yes) keep(did*)
4.3 控制组将不受政策影响
考虑一个并没有受政策影响地方假设其受到政策影响的情况：
g time = (year >= 77) & !missing(year)
g treated_3= (idcode<1600 & idcode>1000)&!missing(idcode)
g did3 = time*treated_3
xtreg ln_w did3 $xlist i.year,fe
------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
did3 | .0047208 .0137331 0.34 0.731 -.022197 .0316387
...did3不显著，证明控制组不受政策影响。符合预期。
4.4 政策实施的唯一性
至少证明这个政策才是主要影响因素。需要寻找某些受到其他政策影响的地方：
g time4 = (year >= 77) & !missing(year)
g treated_4= (idcode<3000 & idcode>2300)&!missing(idcode)
g did4 = time4*treated_4
xtreg ln_w did4 $xlist i.year,fe
------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
did4 | -.009966 .0129562 -0.77 0.442 -.035361 .0154289
...这里did4不显著，可以证明政策实施的唯一性。但有时候did4可能依然显著，但是系数变小，证明还受到其他政策影响。
4.5 控制组和政策影响组的分组是随机的
用工具变量来替代政策变量，解决因为分组非随机导致的内生性问题：
xi:xtivreg2 ln_w (did=hours tenure) $xlist i.year,fe first 二、期刊示例
示例1：psmatch2
文献来源
- Zhang, X., et al. (2023). The impact of carbon markets on the financial performance of power producers: Evidence based on China
- Appendix A. Supplementary data 【数据+Stata】
示例代码
cd "C:\Download\1-s2.0-S0140988323006175-mmc1\data and command"
use "data.dta",clear
gen top10=ln(Top10)
gen Soe=0
replace Soe=1 if 公司属性=="中央国有企业"|公司属性=="地方国有企业"
destring stkcd,replace
xtset stkcd time
gen miss1=missing(ROA ,did ,Size ,Leverage ,Growth ,top10 ,Soe)
drop if miss1==1
set seed 10101
gen ranorder = runiform()
sort ranorder
*==========进行一对一匹配========================================
psmatch2 did i.time Size Leverage Growth top10 Soe, outcome(ROA) n(1) ate ties logit common
*==========平衡性检验============================================
qui psmatch2 did i.time Size Leverage Growth top10 Soe, outcome(ROA) n(1) ate ties logit common
pstest Size Leverage Growth Top10 Soe, both graph
*====================未匹配结果==================================
psmatch2 did i.time Size Leverage Growth top10 Soe, outcome(ROA) n(1) ate ties logit common
*========(1)一对一匹配===========================================
bootstrap r(att) r(atu) r(ate), reps(100): psmatch2 did i.time Size Leverage Growth top10 Soe, outcome(ROA) n(1) ate ties logit common
gen psm_sample1=1 if _weight!=.
drop _pscore _treated _support _weight _id _n* _pdif
*========(2)邻近匹配，k=4========================================
bootstrap r(att) r(atu) r(ate), reps(100): psmatch2 did i.time Size Leverage Growth top10 Soe, outcome(ROA) n(4) ate ties logit common
sum _pscore
dis 0.25 * r(sd)
gen psm_sample2=1 if _weight!=.
drop _pscore _treated _support _weight _id _n* _pdif
*========(3)卡尺匹配=============================================
bootstrap r(att) r(atu) r(ate), reps(100): psmatch2 did i.time Size Leverage Growth top10 Soe, outcome(ROA) n(4) cal(0.01) ate ties logit common quietly
gen psm_sample3
=1 if _weight!=.
drop _pscore _treated _support _weight _id _n* _pdif
*========(4)半径匹配=============================================
bootstrap r(att) r(atu) r(ate), reps(100): psmatch2 did i.time Size Leverage Growth top10 Soe, outcome(ROA) radius cal(0.01) ate ties logit common quietly
gen psm_sample4=1 if _weight!=.
drop _pscore _treated _support _weight
*========(5)核匹配===============================================
bootstrap r(att) r(atu) r(ate), reps(100): psmatch2 did i.time Size Leverage Growth top10 Soe, outcome(ROA) kernel ate ties logit common quietly
gen psm_sample5=1 if _weight!=.
drop _pscore _treated _support _weight
*========(6)局部线性回归匹配=====================================
bootstrap r(att) r(atu) r(ate), reps(100): psmatch2 did i.time Size Leverage Growth top10 Soe, outcome(ROA) llr ate ties logit common quietly
gen psm_sample6=1 if _weight!=.
drop _pscore _treated _support _weight _id _n* _pdif
*========(7)样条匹配=============================================
set seed 10101
bootstrap r(att) r(atu) r(ate), reps(100): psmatch2 did i.time Size Leverage Growth top10 Soe, outcome(ROA) spline ate ties logit common quietly
gen psm_sample7=1 if _weight!=.
drop _treated _support _weight _id _n*
*========(8)马氏匹配=============================================
psmatch2 did, outcome(ROA) mahal(i.time Size Leverage Growth top10 Soe) n(4) ai(4) ate
gen psm_sample8=1 if _weight!=.
drop _treated _support _weight _id _n*
*****************************************************************
*========生成结果=============================================
sum psm_sample1 psm_sample2 psm_sample3 psm_sample4 psm_sample5 psm_sample6 psm_sample8
xtset stkcd time
cap drop _est*
qui{
reg ROA did Size Leverage Growth top10 Soe i.time if psm_sample1==1,cluster(stkcd) r
est store m1
reg ROA did Size Leverage Growth top10 Soe i.time if psm_sample2==1,cluster(stkcd) r
est store m2
reg ROA did Size Leverage Growth top10 Soe i.time if psm_sample3==1,cluster(stkcd) r
est store m3
reg ROA did Size Leverage Growth top10 Soe i.time if psm_sample4==1,cluster(stkcd) r
est store m4
reg ROA did Size Leverage Growth top10 Soe i.time if psm_sample5==1,cluster(stkcd) r
est store m5
reg ROA did Size Leverage Growth top10 Soe i.time if psm_sample6==1,cluster(stkcd) r
est store m6
reg ROA did Size Leverage Growth top10 Soe i.time if psm_sample8==1,cluster(stkcd) r
est store m8
outreg2 [m1 m2 m3 m4 m5 m6 m8] using "倾向得分匹配PSM(内生性2).doc",word replace tstat tdec(2) bdec(3) rdec(3) nonote drop(_Itime_*) sortvar(ROA)addtext(Firm FE,YES,Year FE,YES)得到结果
示例2：kmatch
文献来源
- Shangguan, Y., et al. (2023). Environmental bonuses of employment protection: Evidence from labor contract law in China
- Appendix B. Supplementary data 【数据+Stata】
示例代码
cd "C:\Download\1-s2.0-S0140988323007363-mmc1\DATA & CODE Environmental Bonuses of Employment Protection"
use "Environmental Bonuses of Employment Protection.dta",clear
sum LC2_dum
kmatch ps LC2_dum (lny1), ///
ebalance ( lnFixasset lnLabor acf_sale age sa hhi c_d roa ) ///
gen(var) wgen(wvar) nn(1)
drop if _KM_ps == .
reghdfe lny1 did2 lnFixasset lnLabor acf_sale age sa hhi c_d roa [pweight=_KM_mw ] , ///
a( id year ind4a_year ind4a_cnamecode cnamecode_year ) clu(id) 得到结果
Propensity-score nearest-neighbor matching Number of obs = 93,066
Neighbors: min = 46512
Treatment : LC2_dum = 1 max = 46554
Covariates : (none)
PS model : logit (pr)
Entropy bal.: targets = 1
balance tolerance = .00001
covariates = lnFixasset lnLabor acf_sale age sa hhi c_d roa
basis = matching weights
Matching statistics
------------------------------------------------------------------------------------------
| Matched | Controls | Balance
| Yes No Total | Used Unused Total | loss
-----------+---------------------------------+---------------------------------+----------
Treated | 46512 0 46512 | 46554 0 46554 | 4.14e-13
Untreated | 46554 0 46554 | 46512 0 46512 | 1.52e-16
Combined | 93066 0 93066 | 93066 0 93066 |
------------------------------------------------------------------------------------------
Treatment-effects estimation
------------------------------------------------------------------------------
lny1 | Coefficient Std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
ATE | .2756808 .0174657 15.78 0.000 .2414481 .3099134
------------------------------------------------------------------------------
Stored variables
Variable Storage Display Value
name type format label Variable label
---------------------------------------------------------------------------------------------
var byte %8.0g Treatment indicator
_KM_nc long %10.0g Number of matched controls
_KM_nm long %10.0g Number of times used as a match
_KM_mw double %10.0g Matching weight
_KM_ps double %10.0g Propensity score
_KM_strata byte %8.0g Matching stratum
wvar double %10.0g Matching weights for ATE
HDFE Linear regression Number of obs = 88,762
Absorbing 5 HDFE groups F( 9, 17649) = 62.55
Statistics robust to heteroskedasticity Prob > F = 0.0000
R-squared = 0.8876
Adj R-squared = 0.8219
Within R-sq. = 0.0917
Number of clusters (id) = 17,650 Root MSE = 0.4057
(Std. err. adjusted for 17,650 clusters in id)
------------------------------------------------------------------------------
| Robust
lny1 | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
did2 | -.118895 .0215247 -5.52 0.000 -.1610854 -.0767046
lnFixasset | -.1014892 .0089206 -11.38 0.000 -.1189745 -.0840039
lnLabor | -.1498149 .0124582 -12.03 0.000 -.1742341 -.1253956
acf_sale | -.3305972 .0283343 -11.67 0.000 -.3861353 -.2750591
age | .0001202 .0006171 0.19 0.846 -.0010895 .0013299
sa | .0235574 .0060927 3.87 0.000 .0116151 .0354997
hhi | -3.597629 1.952235 -1.84 0.065 -7.424202 .228945
c_d | -.0109467 .0124806 -0.88 0.380 -.0354099 .0135164
roa | -.0889202 .044236 -2.01 0.044 -.1756272 -.0022132
_cons | 4.526707 .25066 18.06 0.000 4.035389 5.018026
------------------------------------------------------------------------------
Absorbed degrees of freedom:
---------------------------------------------------------+
Absorbed FE | Categories - Redundant = Num. Coefs |
-----------------+---------------------------------------|
id | 17650 17650 0 *|
year | 10 0 10 |
ind4a_year | 3194 10 3184 |
ind4a_cnamecode | 9357 279 9078 ?|
cnamecode_year | 3072 274 2798 ?|
---------------------------------------------------------+
? = number of redundant parameters may be higher
* = FE nested within cluster; treated as redundant for DoF computation期刊排版
示例3：psmatch2
文献来源
- Wang, W., et al. (2024). The impact of energy-consuming rights trading on green total factor productivity in the context of digital economy: Evidence from listed firms in China
- Appendix B. Supplementary data 【数据+Stata】
示例代码
/**Near-neighbor matching**/
clear
use data.dta
set seed 103
gen u=runiform()
sort u
psmatch2 du SIZE AGE LEV TOBINQ ROA CAPITAL GROWTH TOP1 INDEP EPI FDI GDP IND,outcome(FGTFP) n(2) ate ties logit common
pstest SIZE AGE LEV TOBINQ ROA CAPITAL GROWTH TOP1 INDEP EPI FDI GDP IND,both graph
psgraph
reghdfe FGTFP c._treated#c.dt SIZE AGE LEV TOBINQ ROA CAPITAL GROWTH TOP1 INDEP EPI FDI GDP IND if _weight!=.,absorb(id year)
reghdfe FGTFP c._treated#c.dt SIZE AGE LEV TOBINQ ROA CAPITAL GROWTH TOP1 INDEP EPI FDI GDP IND if _support ==1,absorb(id year)
/**Radius matching**/
clear
use data.dta
set seed 101
gen u=runiform()
sort u
psmatch2 du SIZE AGE LEV TOBINQ ROA CAPITAL GROWTH TOP1 INDEP EPI FDI GDP IND,outcome(FGTFP) radius ate logit cal(0.01) common
pstest SIZE AGE LEV TOBINQ ROA CAPITAL GROWTH TOP1 INDEP EPI FDI GDP IND,both graph
psgraph
reghdfe FGTFP c._treated#c.dt SIZE AGE LEV TOBINQ ROA CAPITAL GROWTH TOP1 INDEP EPI FDI GDP IND if _weight!=.,absorb(id year)
reghdfe FGTFP c._treated#c.dt SIZE AGE LEV TOBINQ ROA CAPITAL GROWTH TOP1 INDEP EPI FDI GDP IND if _support ==1,absorb(id year)
/**Nuclear matching**/
clear
use data.dta
set seed 104
gen u=runiform()