Re: Propensity Scores

*To*: mathgroup at smc.vnet.net*Subject*: [mg125590] Re: Propensity Scores*From*: Barrie Stokes <Barrie.Stokes at newcastle.edu.au>*Date*: Tue, 20 Mar 2012 02:21:35 -0500 (EST)*Delivered-to*: l-mathgroup@mail-archive0.wolfram.com*References*: <201203170750.CAA05061@smc.vnet.net>

Hi Sylvia As long-time Mathematican and a long-time Bayesian, a few thoughts. Mathematica is (at least) a massive CAS (Computer Algebra System), with lots of mathematical knowledge built-in, and is also a marvellous programming system for *very* high level coding, algorithm development, etc. Having looked at the paper you are reading, I note in the synopsis the author says, in part: "The estimates of mean effect obtained with *five* (my emphasis) different techniques were compared and nonparametric bootstrap was recommended as superior tool for propensity score analyses." You say: It's enormously frustrating (given the degree of depth in Mathematica documentation) to realize the absence in Mathematica of a commonly used statistical function in SAS and STATA, pointing to the real difference in need between mathematicians and statisticians to impute missing data from large datasets using bootstrapping." I think you*ve put your finger on it. There *are* myriad "real difference in need between mathematicians and statisticians", and that is reflected in the difference between SAS and Mathematica. Mathematica is not in the first instance a statistics package, but you can do advanced mathematical (i.e., symbolical) statistics in it that you can*t do in a non-CAS system like SAS. To some extent (Mathematica has fantastic numerical functionality) it*s horses for courses. I don*t think you should expect that Mathematica should come with scores or even hundreds of different and very specific statistical model fitting procedures built-in. If I want to use the procedures built in to SAS, Stata, etc., on data to hand, I use them as is. If I want to try a variation on a theme, or produce a particular graphic, I turn to Mathematica and code it myself. (Mathematica has full matrix algebra capability for multivariate calculation.) Logistic Regression is essentially built-in to Mathematica . You also say: "For whatever reason, I assumed otherwise since Mathematica so exquisitely lends itself to predicting the pathways of cellular automata." Hmm. There is a, er, well known connection between cellular automata and Mathematica that possibly explains that observation. But you can*t expect that Mathematica has ready-made versions of most of the algorithms in most scientific fields. The tools are there, though. And some Google code searching sometimes turns up something useful. Cheers Barrie PS Jerry-rigged? Jerry-rigged? Never. Carefully coded, you mean. PPS I found http://www.jhsph.edu/bnrc/Research_Libr/Harding2003.pdf and http://icvts.oxfordjournals.org/content/early/2011/11/25/icvts.ivr005.full after a spot of Googling. >>> On 17/03/2012 at 6:50 pm, in message <201203170750.CAA05061 at smc.vnet.net>, <sylviahobbs at aol.com> wrote: > Newbie Here! Finally, I have my very own Mathematica license, to have and to > hold until death do us part. We are already arguing over the best way to > construct a propensity score given the absence of a dedicated function in > Mathematica. It's enormously frustrating (given the degree of depth in > Mathematica documentation) to realize the absence in Mathematica of a > commonly used statistical function in SAS and STATA, pointing to the real > difference in need between mathematicians and statisticians to impute missing > data from large datasets using bootstrapping. For whatever reason, I assumed > otherwise since Mathematica so exquisitely lends itself to predicting the > pathways of cellular automata. Mathematica license and I are on honeymoon > here in Boston for St. Paddy's weekend reading aloud to each other from a > very nice 2009 paper on the impact of exclusion decisions in random sampling > by Maciej G=C3=B3rkiewicz published in the Biocybernetics and Biomedical > Engineering entitled "Usi! > ng Propensity Score with Receiver Operating Characteristics and Bootstrap to > Evaluate Effect Size in Observational Studies." > [http://ibib.waw.pl/bbe/bbefulltext/BBE_29_4_041_FT.pdf] > My question is to the Bayesians out there, what Mathematica functions have > you jerry-rigged for propensity scoring? Holla Back! > > Sylvia Hobbs

**References**:**Propensity Scores***From:*sylviahobbs@aol.com