Managerial Pay and Corporate Performance

ing, for the moment, from potential statistical and measurement problems, and in the absence of theoretical reasons to specify an alternative form of functional relationship, we may begin by postulating that a top executive's compensation is related in linear fashion to both the profits and sales of the firm he manages. The structural form of the relationship can be written: (1) Cit = ao + a,-rxi + a2Sij + ui, Where, C, r, and S represent executive compensation, corporate profits, and corporate sales, respectively, and u is a random disturbance term. Subscript i denotes the firm and subscript t the period to which the measure applies. By supplying a basis for observing the magnitude of the coefficients a, and a2 and the levels of statistical significance attaching thereto, the above specification provides a natural vehicle for inferring the relative influence of the two independent variables upon compensation, and thereby testing the alternative hypotheses. The emergence of a positive value for the constant term ao would imply, in effect, that executive rewards rise less than in proportion to company sales and/or profits. Thus, it seems probable that a $50 thousand difference in annual profits between two firms in the $100 million profit range would result in a smaller difference in the pay of their respective chief executives than would the same dollar profit difference in the case of two firms whose yearly earnings were in the $100 thousand range. Represented graphically, the compensation vs. profits or compensation vs. sales relationship would therefore be expected to be concave downward for a sample of enterprises differing widely in size, and the linear approximation to any segment of such an underlying relationship would necessarily include a positive intercept value. It follows, then, that a, and a2 must be interpreted in marginal-although constant for the sample range-terms throughout. Statistical Problems Unfortunately, direct application of equation (1) to any generalized sample of crosssectional data can be expected to encounter several possible sources of statistical bias. For one thing, the efficiency of least square estimates depends upon the variances of the disturbance terms being constant. Examination of scatter diagrams of pilot regression runs using equation (1) revealed, as anticipated, that the error terms were not constant but were approximately in proportion to the dependent variable. Moreover, and as one might also suspect, those firms relatively large by virtually any scale criterion were also characterized by relatively high sales and profits levels. This scale-associated linkage between the independent variables poses the threat of serious collinearity,4 with re4The high degrees of correlation between the independent variables in their natural form were indicated by the presence of simple correlation coefficients which in most cases, exceeded .9. This high degree of observed This content downloaded from 157.55.39.104 on Mon, 20 Jun 2016 05:40:33 UTC All use subject to http://about.jstor.org/terms