I'm pretty sure that means that there are literally no matches (that is, the data is so unbalanced that no treated units are in the same strata as control units). This could mean that you just need to set wider bins. It could also be something nonsensical in your particular data- so if you are matching on country and year fixed effects in a TSCS dataset for example, you could by construction, never have any matches. Brandon On Wed, Apr 13, 2011 at 2:14 PM, Slawa Rokicki <slawa.rokicki(a)gmail.com>wrote;wrote: _______________________________________________ gov2001-l mailing list gov2001-l(a)lists.fas.harvard.edu http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l

So matching missing values is one way to go about it, but that presumes a certain homogeneity to the missingness. Plus when you reach the analysis stage, the missingness will still be a problem. The better strategy is to impute the missing values and then use CEM. The CEM manual elaborates on how to do this (I think its also described in one of the CEM papers that Gary wrote). Brandon On Wed, Apr 13, 2011 at 2:25 PM, Slawa Rokicki <slawa.rokicki(a)gmail.com>wrote;wrote: _______________________________________________ gov2001-l mailing list gov2001-l(a)lists.fas.harvard.edu http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l