summary description of a group of bivariate observations. One may of course ask, "How does one know whether a figure of 0.55 for correlation coefficient is significantly high or not? In the absence of any test of significance, one's judgment would be subjective'*. It is in order to form an objective judgment that the econometrician conducts a test of significance and makes a statement of the following type: "The probability that the true value of the correlation coefficient is non-zero is more than 0.99". But the objectivity thus gained is of quite an illusory character. Firstly, in many instances, the scatter of points is unique and no true value exists for the correlation coefficient: the value of 0.55 is all that there is to the scatter by way of correlation coefficient and there is no means of going beyond it. Secondly, as we have seen before, the figure 0.99 would be based on certain unverifiable subjective assumptions. Therefore, one's subjective judgment regarding the significance of 0.55 does not become objectivised by putting next to it the subjective figure of 0.99. Nevertheless, the figure 0.55 for the correlation coefficient is far from being meaningless. The same can be said about many other statistical tools. Different types of averages: the arithmetic mean, the geometric mean, weighted averages, etc, have been extensively used with benefit in many different fields of quantitative work without any probability assumptions being made. The standard deviation is a commonsense measure of the degree of dispersion in a set of one dimensional observations. The method of least squares had proved its usefulness over centuries before the invention of the term Maximum Likelihood Method. Fitting of curves to empirical frequency distributions or to time series is a useful statistical device for condensing data: a large number of observatons may be replaced by two or three numbers representing fitted parametric values and this is useful even without any probability interpretations. Even such test statistics as Student's t and Fisher's F have got commonsense meaning and may be used with benefit, provided they are made to refer strictly to the data under analysis and not invested with any spurious probability significance.
