teffects psmatch r(459) error no match with caliper

[Emile Tenezakis]

Good morning everyone,

I found only a few posts concerning my problem and unfortunately none of them provided clear solution steps... Here is the issue :
I have a household survey dataset. I'm interested in the effect of the treatment elec on child labor, i.e. variable "wwork" for subsample with kid==1. I want the AT effect on the treated. (elec==1 if treated).
The code:


frequency weights computation
gen tp=1000*weight
gen newweight=round(tp)

subsample with kids
gen mysample=1 if kid==1

propensity scores computation to define the common support interval
logit elec nb_kids06 floor_area nb_rooms cons nb_asscap cement HH_age HH_age2 HH_lit water if mysample==1 [fw=newweight]
predict score if e(sample)
gen mysupport=1 if score>=0.005 & score<=0.99

command teffects with caliper
teffects psmatch (wwork) (elec nb_kids06 floor_area nb_rooms cons nb_asscap cement HH_age HH_age2 HH_lit water, logit) if mysupport==1 & mysample==1 [fw=newweight], caliper(0.04) atet

I get the following error message:
no propensity-score matches for observation 3 within caliper 0.04; this is not allowed
r(459) ;

using (osample) to remove non-matched observations doesn't work (as previous users pointed it out). What is the solution when one wants to use teffects psmatch with a caliper but some observations don't find match with that condition ?
Note that I suspect more problems considering that even with a caliper of 0.5, it doesn't work....

Thank you very much for your attention

[Emile Tenezakis]

Update: I found a solution which is the option vce(iid) to use default Abadie-Imbens standard errors. The code works fine. Apparently there was an issue with standard errors computations this is why I tried this option, but I have no clue why it actually works now. If anyone has an explanation it would be more than welcome...