st: Propensity scoring using -teffects- in Stata 13; puzzling features, documentation issues,

["Lacy,Michael" <Michael.Lacy@colostate.edu>]

Dear Colleagues,

Per the subject, I've found things that puzzle me me while trying out -teffects-, the new Stata 13 collection of modules
that implement propensity scoring analysis and related procedures. I'd like some reactions to either correct or support my
perceptions.  There may be more issues. My impression is that -teffects- is very complete and sophisticated, but 
perhaps not entirely polished yet.


1) When doing nearest neighbor matching propensity score matching: e.g. ,

teffects psmatch (Y ) (Tx X* ),  atet nneighbor(2) gen(nay) caliper(0.06)

a) I can't find a way to request matching *without* replacement or even any mention of the replacement/no
replacement issue in the documentation. To my understanding, there is no received wisdom saying "always pick
matched subjects with replacement," and many situations in which w/o replacement would be desirable.

b) As documente, nearest neighbor matching always selects *at least* the specified number of
matches for each, not *exactly* the specified number.   So, some of my treated subjects had 2 matches,
others 150 matches.  I can understand that one could argue for the virtues of always using  >=, i.e.,  "take all the 
matches you can get," but requiring that one's analysis find all possible matches within the caliper seems strange,
if for no other reason than that obtaining 150  matches for some Tx subjects when the user only wanted 2 can 
unecessarily be very slow.

Is there a reason to require a) and b) be enforced, apparently without option?  


2) Errors(?) in documentation regarding the -caliper- option.

In *several* places, the caliper distance for matches is described incorrectly as a *minimum* (!) distance,
elsewhere correctly as a *maximum.*

e.g. -help teffects psmatch- ... caliper(#)          " specify the minimum [sic] distance for which two
observations are potential neighbors"

This apparent confusion of min and max also appears in the detailed (pdf) documentation, and in the dialog box for 
-teffects psmatch-.

This would seem to me as something to be fixed in an update, unless I'm somehow confused.


3) Imprecision in documentation about obtaining ids of matching cases

The documentation says about the generate option

generate(stub)  generate variables containing the row indices of the nearest neighbors

As it turns out,  "row index" means "observation number in the dataset prior to execution of the -teffects-
command, prior to any application of [if] [in] that the -teffects- command contains."   To my knowledge, "row
index" has not previously been standard Stata terminology (-search row index- yields nothing).  


Regards, 

Mike Lacy
Assoc. Prof./Dir. Grad. Studies
Dept. of Sociology
Colorado State University
Fort Collins CO 80523-1784
970.491.6721 (voice)


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/