Dear all,
I am experiencing the following problem and I cannot find the solution anywhere: I am trying to replicate the results from the psmatch2 command with caliper by using the teffects psmatch command with caliper but I do not get the same point estimates. This happens for any dataset I may use. I illustrate the problem below:
use http://ssc.wisc.edu/sscc/pubs/files/psm
psmatch2 WITHOUT caliper
psmatch2 t x1 x2, out(y) neighbor(1) ties
teffects psmatch WITHOUT caliper
teffects psmatch (y) (t x1 x2, probit), first atet nneighbor(1) vce(iid)
As you can see the point estimates are identical when no caliper is used; the SEs are different as expected with "teffects psmatch" reporting the corrected SEs. When I use a caliper of .042793 (i.e. 0.2 * the standard deviation of the propensity score) for each command, I get different results:
psmatch2 WITH caliper
psmatch2 t x1 x2, out(y) neighbor(1) ties radius caliper(.042793)
teffects psmatch WITH caliper
teffects psmatch (y) (t x1 x2, probit), first atet nneighbor(1) caliper(.042793) vce(iid)
Actually, whereas the "psmatch2" estimates a different treatment effect on the treated (.987124793) to when no caliper is used, the "teffects psmatch" results are identical to when I do not use caliper (ATET: 1.019685). It seems that the whole sample (1,000 observations) are used in all four cases above.
Do you have any suggestions as to where this difference may be coming from?
Many thanks,
Chris
I am experiencing the following problem and I cannot find the solution anywhere: I am trying to replicate the results from the psmatch2 command with caliper by using the teffects psmatch command with caliper but I do not get the same point estimates. This happens for any dataset I may use. I illustrate the problem below:
use http://ssc.wisc.edu/sscc/pubs/files/psm
psmatch2 WITHOUT caliper
psmatch2 t x1 x2, out(y) neighbor(1) ties
teffects psmatch WITHOUT caliper
teffects psmatch (y) (t x1 x2, probit), first atet nneighbor(1) vce(iid)
As you can see the point estimates are identical when no caliper is used; the SEs are different as expected with "teffects psmatch" reporting the corrected SEs. When I use a caliper of .042793 (i.e. 0.2 * the standard deviation of the propensity score) for each command, I get different results:
psmatch2 WITH caliper
psmatch2 t x1 x2, out(y) neighbor(1) ties radius caliper(.042793)
teffects psmatch WITH caliper
teffects psmatch (y) (t x1 x2, probit), first atet nneighbor(1) caliper(.042793) vce(iid)
Actually, whereas the "psmatch2" estimates a different treatment effect on the treated (.987124793) to when no caliper is used, the "teffects psmatch" results are identical to when I do not use caliper (ATET: 1.019685). It seems that the whole sample (1,000 observations) are used in all four cases above.
Do you have any suggestions as to where this difference may be coming from?
Many thanks,
Chris
