R&D and Israel's Industrial Export Performance: An Application of Multidimensional Scaling

Abstract

Two recent articles [10, 11] present econome tric applications of the technique of non metric multidimensional scaling (henceforth MS). Although quite common in the psychometric lit erature, this method of multivariate analysis is new in the economics literature. Syrkin [11] illus trates the applicability of the technique in the con text of development economics, comparing his results with those of the well-known 1967 factor analysis study of Adelman and Morris. Maital [10] demonstrates several econometric applications of this technique, among them an interesting analy sis on national input-output tables in a manner which may establish a new useful approach for meaningful international comparisons of macro economic structures. It is argued here that MS should prove to have a wide and general applicability as a prescreen ing device when the common problem of a too large and too complex a body of data is faced by the empirical investigator. The accepted approach in such cases is to adopt a specific theory or mod el as applicable. This immediately allows the re searcher to focus on a manageable number of potentially interesting variables and defines the nature and direction of the relevant relationships. Typically, several variables are taken as exogen ously determined, and one or more as endogenous. Then some form of regression analysis is applied. The results are said to indicate that a certain per centage of the variance is in fact explained by the included variables. Remaining unsaid, generally, is that if the particular model explicitly tested is not the one, the R2's are meaningless. Fur thermore, even if the model is correct, the speci fication may not be, in which case the included variables may in fact represent more correct variables which have been excluded. Clearly some device which could condense and organize the ob servations, allowing for an efficient use of all of the available data should generally be useful as a screening. Such a device should generally avoid a