A Further Clarification

As any reader will know who has followed subject this far, this note is in reply to a reply (see preceding article) to a reply [3] to a criticism [4] of a series of published analyses [1, 2]. In our reply to original criticism, both of which appeared in August, 1965, JMR, we gave our views on each of points raised by Messrs. Fothergill and Ehrenberg. According to them, however, our response does not by itself resolve all difficulties. Fothergill and Ehrenberg seem to have continuing on three points: (1) difference between standard deviations of .57 and .73; (2) extent of built-in correlation in our model; and (3) alleged selectivity of our data. We shall try to clarify these points further; we hope that this will remove all residual difficulties for Fothergill and Ehrenberg, since reader interest may not be sufficient to support a regular, quarterly exchange of views. First, with respect to difference between a standard deviation of .57 market share points and one of .73 points, Fothergill and Ehrenberg state that Numerically, difference itself turns out to be one or two thousands of total 100 point range of variation. We can only reply: what total 100 point range of variation? The dependent variable in analysis from which two standard deviations were derived is change in market share for a brand from one four-month period to next (MSt MSt_-1). As shown in one of our published reports [2, p. 35], standard deviation of MSt MSt_-, for 50 observations available to us, was .81 share points; biggest change in market share among 50 was 2.05 share points, and only 15 of 50 were equal to or greater than one share point in absolute value. We believe that these figures are reasonably typical of variability of short-term changes in market shares for nondurable consumer products, at least in terms of order of magnitude. Consequently, we regard a comparison with the total of 100 point range of variation as irrelevant.