302 QUARTERLY JOURNAL OF ECONOMICS It follows that where within a Census industry there is a roughly balanced diversification of all principal firms among all principal products, the industry concentration measure tends to represent true theoretical concentration for included outputs regardless of whether Census industry or product is a theoretical industry, provided that one or the other is. On the other hand, if there is distinct special- ization of firms among products, the industry concentration measure will be theoretically valid only if all the included products are close substitutes; if they are not, it will misrepresent the true concentration for the component theoretical industries. Assuming indecision as to whether products within a Census industry are close substitutes, we may thus accept as valid for testing our hypothesis only Census industries and related concentration measures where there is either only a single Census product or a roughly balanced diversification of firms among several products. We must reject the others as poten tially invalid. This criterion of selection has been applied along with others previously mentioned. 3 The joint application of the various criteria described above has resulted in a selection of an industry sample by the progressive elimination of Census industries for which(1)geographical market segmentation may destroy the significance of the concentration measure,(2) there is conclusive evidence that the Census industry excludes close substitutes for included outputs, or(3 )there are several Census products, with a significant degree of specialization by princi- pal firms among products. This has left a residual sample of Census component. But if firm I produces 80 per cent of a, firm 2, 80 per cent of b, and firm 3, 80 per cent of c, such an aggregate figure will generally 9. As this criterion has been applied, it has appeared that there is a large number of industri nclude several Census products whie poor substitutes for each other, and within which the various firms are largel specialized in only one or a part of these products. Such industries have been rejected from our sample on the ground that if the products were poor substitutes tion figure would be grossly lower than that for the appropriate theoretical industry groups in the complex. On the other han here is a very considerable number of census industries which include several substitute output groups or Census products, and within which each of the principal firms is relatively unspecialized and has roughly the same standing in each or most of t jor Census products. In these cases, the entration tends to be roughly representative of concentration for each of the concentration ratios, in the absence of geographical difficulties and of a peremp- ory rejection on grounds the Census industry is obviously too narrow, are thu e the industry concentration ssumptions be theoretically component produets are close substitutes
RELATION OF PROFIT RATE TO INDUSTRY lustries which are cleared on the three p ing counts and which nclude a few single-product Census industries and a good many ilti-product industries wherein product and industry concentration measures tend to be very similar. If we assume that for these either the Census product or the Census industry corresponds to a theoretical industry, their concentration measures should be significant for testing our hypothesis The mechanical procedure employed in selecting the sample, and the results of this procedure, may be described as follows. From 340 industries in the 1935 Census of Manufactures I we first selected hose for which some profit data were available from S.E. C. publica- tions for 1936-40, and eliminated the remainder. This left 152 industries, for 149 of which there were also available concentration data. These 149 were then screened to determine which should be rejected on the ground either of geographical market segmentation or of specialization of firms among products within the Census industry. In this procedure we first followed the classification of industries developed in Structure of the American Economy, which segregates Census industries first into those having"national gional, and"local ' markets and second into those which are “ straight", and“ mixed.” Industries with“ national” markets evi- dently represent those which are little affected by geographical market segmentation. "Straight'and"mixed"industries correspond in rough way respectively to industries with"diversified"and"special ized "firms as identified above-a"straight"'industry is one which each manufacturer as a rule engages in the production of all commodities covered by the industry classification, ' and a"mixed industry one "in which manufacturers confine themselves to the production of only part of the commodities included in the industry. 4 Following this classification we then selected only those industries which were classified by the Structure of the American Economy (hereafter referred to as the NRC) as having"national"markets and as being also"straight "A total of 83 industries were th selected, and the remainder rejected The reduced sample of industries was now examined to determine 1. See National Resources Committee, Structure of the American Economy data was decided upon since the S E.C. provided the most adequate ar able uniform source of the requisite 3.Op.cit,pp.264-69