Hi,
This is a new question following on my previous question on before- and after-matching propensity score graphs: http://www.statalist.org/forums/foru...score-matching
As I mentioned in that post, it is useful to show plot the propensity scores of treatment and control observations before and after matching.
When a nearest-neighbor matching is used it is easy to identify which control observation(s) is matched with each treated observation, and -psmatch2- produces a variable (_n1) to indicate that.
However, when a kernel-based matching is used the process becomes complicated because in a kernel-based matching every treated observation is matched with all the control observations where the control observations with the closest propensity score to that specific treated observation is assigned the biggest weight; the farther the propensity score of the control observations from the specific treatment observation the smaller their weight.
Now, I wonder how this is dealt with when one wants to plot the propensity scores of treatment and observation after matching when kernel-based method is used.
Thanks,
Navid
This is a new question following on my previous question on before- and after-matching propensity score graphs: http://www.statalist.org/forums/foru...score-matching
As I mentioned in that post, it is useful to show plot the propensity scores of treatment and control observations before and after matching.
When a nearest-neighbor matching is used it is easy to identify which control observation(s) is matched with each treated observation, and -psmatch2- produces a variable (_n1) to indicate that.
However, when a kernel-based matching is used the process becomes complicated because in a kernel-based matching every treated observation is matched with all the control observations where the control observations with the closest propensity score to that specific treated observation is assigned the biggest weight; the farther the propensity score of the control observations from the specific treatment observation the smaller their weight.
Now, I wonder how this is dealt with when one wants to plot the propensity scores of treatment and observation after matching when kernel-based method is used.
Thanks,
Navid
Comment