difference between pscore and psmatch and choosing matching algorithm

[Naika Sangroo]

Dear all,

I am planning to use propensity score matching for my paper. I am a little bit confused as to what is the difference between pscore and psmatch.

I gathered some previous codes and information online. created global list (local would not work with my stata for some reason. it would work the fist time then when i run it again the second time it would not work)

***treatment

global TREAT treatment

*** x-list

******  $TREAT $HH $GEO $HH_RESP $LAND $EDU $WATER

probit $TREAT $HH $GEO $HH_RESP $LAND $EDU $WATER
predict pre_probit, p
    sum pre_probit
    sum pre_probit if migrant == 1
    sum pre_probit if migrant == 0

psmatch2 $TREAT $HH $GEO $HH_RESP $LAND $EDU $WATER
    sum _pscore
    sum _pscore if migrant == 1
    sum _pscore if migrant == 0

I am not very certain what exactly am i doing here. is pscore after balance? and pre_probit what it was like before balance? Why do we perform pstest? and to do the next step nearest neighbor, radius, kernal and stratification which codes do i use? if you can guide me to the right direction on successfully using PSM i would be very grateful. Thank you.

[Lorien Nair]

Refer this for help on pscore Vs psmatch2: https://www.bgsu.edu/content/dam/BGS...ta-example.pdf

[Olive Bat]

Hi.. i still have the same problem which is better pscore or psmatch2.. and i also wanted to check for robustness using sensitivity analysis.. anyways i would recommend you this link on you tube which is quite helpful on pscore and various forms of matching which has been explained really well...https://sites.google.com/site/econom...score-matching

[Naika Sangroo]

Thank you Lorien and Olive for your advice! The sources are very helpful!

[Naika Sangroo]

I am still a little bit confused as i should only used the observations that have matched for my analysis should i then use minimum and maximum comparison and delete the observations that do not match? How do i actually do this step in order to move to choosing matching algorithm

count if _pscore < .2249672 //There are 2 comparison units whose estimated p-score is less than the minimum estimated p-score for the treated units (.2249672).  


count if _pscore > .8357755 // There are 10 treated units  whose estimated p-score is greater than the maximum estimated p-score for the comparison units (.8357755).


* delete all observations whose propensity score is smaller than the minimum and larger than the maximum in the opposite group

    drop if _pscore < .2249672
    drop if _pscore > .8357755

I am stuck in this very beginning part. thank you so much for your help.

[Lorien Nair]

You should see if you can use the comsup option in pscore to ensure that the common support assumption holds and that you are working with observations that satisfy this condition.
Also, it is bad practice to delete observations from the dataset, instead enforce common support in the algorithm for better matches.