完全在外面当然最好，但稍微覆盖一点也没事，你这个的问题在于被覆盖的太多

好的，有没有办法可以改善这种情况啊

没有看你的代码，不知道你做的是否有问题

mat b = J(500,1,0)
mat se = J(500,1,0)
mat p = J(500,1,0)

forvalues i = 1/500{

use "/Users/eevee/Desktop/77.dta",clear
keep if year==2003
sample 37,count
keep id
save matchcity.dta,replace
  merge 1:m id using "/Users/eevee/Desktop/77.dta"
  gen treat1=(_merge ==1)
  save matchcity`i'.dta,replace
  
  use "/Users/eevee/Desktop/77.dta",clear
  bsample 1, strata(id)
  keep year
  save matchyear.dta, replace
  mkmat year, matrix(sampleyear)

  use matchcity`i'.dta,replace
  xtset id  year
  gen time = 0
  foreach j of numlist 1/57 {  
     replace time = 1 if (id == `j' &  year >= sampleyear[`j',1])
   }
  gen DID=time*treat
  qui reghdfe lnpilu DID lnpeo lnRD lnKTS lnhis , absorb(id year) vce(r)
mat b[`i',1] = _b[DID]
        mat se[`i',1] = _se[DID]
        scalar df_r = e(N) - e(df_m) -1
        mat p[`i',1] = 2*ttail(df_r,abs(_b[DID]/_se[DID]))
}

svmat b, names(coef)
svmat se, names(se)
svmat p, names(pvalue)

drop if pvalue1 == .
label var pvalue1 p值
label var coef1 估计系数

twoway (kdensity coef1) (scatter pvalue1 coef1, msymbol(smcircle_hollow) mcolor(blue)), ///
title("Placebo Test") ///
xlabel(-0.3(0.1)0.6) ylabel(,angle(0)) ///
xline(0.198, lwidth(vthin) lp(shortdash)) xtitle("Coefficients") ///
yline(0.1,lwidth(vthin) lp(dash)) ytitle(p value) ///
legend(label( 1 "kdensity of estimates") label( 2 "p value")) ///
plotregion(style(none)) ///无边框
graphregion(color(white)) //白底

好的，我代码发在这个帖子里了，不知道是不是代码问题