01960cam a22002297 4500001000600000003000500006005001700011008004100028100002100069245006900090260006600159490004100225500001000266520110400276530006101380538007201441538003601513710004201549830007601591856003701667856002601704w0072NBER20140725110656.0140725s1975 mau||||fs|||| 000 0 eng d1 aWolff, Edward N.14aThe Goodness of Matchh[electronic resource] /cEdward N. Wolff. aCambridge, Mass.bNational Bureau of Economic Researchc1975.1 aNBER working paper seriesvno. w0072 a1975.3 aThough the statistical techniques vary, the matching problem is essentially the same in each case and can be stated formally as follows: Given "observations on X,Y from one sample and on X,Z from another sample, when will it be true that by matching observations according to X, an artificial Y,Z sample will result whose distribution is the true joint Y,Z distribution?"(Sims,1972, p. 355). Though the imputed Y,Z distribution will, in general, be different from the true Y,Z distribution, the closeness of the two yields a natural criterion of the goodness of match. By making certain simplifying assumptions, we can make this criterion operational. The goodness of match depends on how much of the relation between Y and Z is transmitted through X - that is, on how X "mediates" between Y and Z. Since the functional form the lower and upper bounds on the true correlation between Y and Z takes depends on the number of X variables, we shall treat the problem in three stages: (a) The case of one mediating variable.(b) The case of two mediating variables. (c) The case of n mediating variables. aHardcopy version available to institutional subscribers. aSystem requirements: Adobe [Acrobat] Reader required for PDF files. aMode of access: World Wide Web.2 aNational Bureau of Economic Research. 0aWorking Paper Series (National Bureau of Economic Research)vno. w0072.4 uhttp://www.nber.org/papers/w0072 uurn:doi:10.3386/w0072