I am attempting exact matching using Matchit package in R. I get both weights and subclass as outputs of this matching procedure. Based on paper by Ho, Imai, King and Stuart (2007) section 3.3, with subclassification, we need to estimate the average treatment effect within each subclass and then aggregate across subclasses. But, authors also state that when it is not possible to obtain an effect within each subclass, the generated weights can be used.
I appreciate any help or insights for the following questions:
1) When I am aggregating the ATE for subclasses should I eliminate those coefficients that are insignificant? if Not, after aggregating, how can I claim that I have a significant treatment effect?
2) In estimating the ATE for each subclass, should I also use the generated weight?
3) What is the advantage of (in terms of efficiency gain and bias reduction) the method that uses the aggregate of ATE across subclasses vs. the method that uses the generated weight? (I have a very large dataset that makes the reestimation and trial and error very time consuming)
