teffects psmatch with Mahal-metric matching

[Clotilde Mahe]

Dear all,

I would like to build a control-matched sample on which to run an IV (outcome variable is a continuous variable; endogenous dependent variable, a dummy), and would like to do so using a Mahal-metric matching. However, I could not find a way to use teffects psmatch with this matching, in order to obtain correct standard errors... I first built the sample with a Mahal-metric matching using psmatch2, and ran my IV; then, with nearest neighbour matching, using teffects to obtain the correct sample weights to then use in the IV. IV coefficient estimates do not differ much, but still, a little (e.g. around 0.4). Does this matter much? Is there a way to use Mahal-metric matching with teffects?

Thank you for your help.

Clotilde

[David Radwin]

By "Mahal-metric matching," do you mean nearest neighbor matching using the Mahalanobis distance metric? If so, try teffects nnmatch. Mahalanobis is the default method. It will report Abadie-Imbens standard errors, but those standard errors apply to the results of the matching estimates, not to any subsequent IV estimates.

David

[Clotilde Mahe]

Thank you, David, for your reply. Yes, this is exactly what I mean (nearest neighbor matching using the Mahalanobis distance metric. Would using teffects nnmatch be similar to using psmatch2 with Mahalanobis distance metric and the option ai(M), where M is the number of neighbors that are used to calculate the conditional variance? Would you recommend bootstrapping?

Clotilde

[Clotilde Mahe]

Sorry, David, I might have made a mistake. I am using a Mahalanobis matching as is:

psmatch2 depvar [if exp] [in range] , mahalanobis(varlist) outcome(varlist) ai(integer) [population altvariance kernel(kernel_type) llr
bwidth(real) caliper(real) w(matrix) ate]

Sorry for the confusion...

Clotilde

[David Radwin]

I don't know for sure, but I think the answer to your first question is yes. See http://www.statalist.org/forums/foru...814#post752814 .

I'm sorry that I don't know the answer to your second question.

[Clotilde Mahe]

Thanks, David, for your reply!