离散被解释变量的DID

计量经济学

1.关于Probit DID的讨论
https://www.statalist.org/forums/forum/general-stata-discussion/general/1373003-probit-difference-in-difference-model

2.关于Logit DID的讨论
https://www.statalist.org/forums/forum/general-stata-discussion/general/1409741-difference-in-difference-logit-model
3.胡日东，林明裕，2018，双重差分方法的研究动态及其在公共政策评估中的应用，财经智库，03（03）：84-111 页
此文中有论述到二元离散被解释变量情况下的双重差分模型，但其论述很不清楚，前文说不可使用Probit模型估计的系数直接当做处理效应，后文又称此法被广泛使用，不明个中道理何在。

4.《养老保险制度统一与劳动要素市场化配置》一文使用了Probit-DID模型，并提供了相应代码：

*基本模型
vcemway probit turnout3 rj jibenyanglao i.year , cluster(pid year)
margins,dydx(_all)post
vcemway probit turnout3 rj jibenyanglao gender age age2 education wage2 marriage tenure2 hour1 housetype old1 child1 i.year , cluster(pid year)
margins,dydx(_all)post
vcemway probit turnout3 rj jibenyanglao gender age age2 marriage education wage2 tenure2 hour1 housetype old1 child1 i.year i.provcd , cluster(pid year)
margins,dydx(_all)post
*平行趋势检验
   generate t5 = invttail(9233,0.001)
   vcemway probit turnout3 tj2 tj3 tj4 jibenyanglao  i.year , cluster(pid year)
   margins,dydx(_all)post 
   generate b_1 = _b[tj2]  
   generate se_b_1 = _se[tj2] 
   generate b_1LB = b_1 - t5*se_b_1 
   generate b_1UB = b_1 + t5*se_b_1
   generate b_2 = _b[tj3]  
   generate se_b_2 = _se[tj3] 
   generate b_2LB = b_2 - t5*se_b_2 
   generate b_2UB = b_2 + t5*se_b_2
   generate b_3 = _b[tj4]  
   generate se_b_3 = _se[tj4] 
   generate b_3LB = b_3 - t5*se_b_3 
   generate b_3UB = b_3 + t5*se_b_3
   gen wave=_n*2+2012
   generate b = . 
   generate LB = . 
   generate UB = . 
   replace b = b_1  if wave==2014
   replace b = b_2  if wave==2016
   replace b = b_3  if wave==2018
   replace LB = b_1LB if wave==2014
   replace LB = b_2LB if wave==2016
   replace LB = b_3LB if wave==2018
   replace UB = b_1UB if wave==2014
   replace UB = b_2UB if wave==2016
   replace UB = b_3UB if wave==2018
   keep wave b LB UB
   drop if wave>=2020
   twoway (connected b wave, sort lcolor(navy) mcolor(navy) msymbol(circle_hollow) cmissing(n)) ///
       (rcap LB UB wave, lcolor(navy) lpattern(dash) msize(medium)), ///
        ///
       yline(0, lwidth(vthin) lpattern(dash) lcolor(teal)) ylabel(, labsize(small) angle(horizontal) nogrid)  ///
        ///
       xline(2015, lwidth(vthin) lpattern(dash) lcolor(teal)) xlabel(2014(2)2018, labsize(small)) xmtick(2013(2)2019, nolabels ticks) /// 
       legend(off) ///
       graphregion(fcolor(white) lcolor(white) ifcolor(white) ilcolor(white))