Académique Documents
Professionnel Documents
Culture Documents
Manufacturing
Author(s): Frank M. Gollop and James L. Monahan
Source: The Review of Economics and Statistics, Vol. 73, No. 2 (May, 1991), pp. 318-330
Published by: The MIT Press
Stable URL: http://www.jstor.org/stable/2109523 .
Accessed: 12/08/2013 13:20
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to The Review of
Economics and Statistics.
http://www.jstor.org
ing that applyingthe index at the aggregatetwo- Jacqueminand Berry(1979) propose the entropy
or three-digitlevel meaningfullymeasures prod- index:
uct dissimilarity.This conclusionfollows, the ar-
gumentcontinues,because this form of d1 would d4- Esi ln(llsi). (5)
register only production activity outside the Like d2, both d3 and d4 are sensitiveto changes
broadly defined industry aggregate housing the in the number and distributionof products. In
primaryfour-digitproduct.Though it lacks some fact, each index can be decomposedinto number
desirablefeatures,but because of the readyavail- and distribution components. Unlike their
ability of PPSR data in Census publications,d1 Herfindahlcounterpart,however,neither d3 nor
remains a popularlyused measure of diversifica- d4 is bounded from above by a constant; 1 <
tion.10 d3 < n and 0 < d4 < ln(n). Another distinctionis
Berry (1971) and McVey (1972) apparentlyin- notable as well. As the simple numberof prod-
dependently propose and apply the Herfindahl ucts increasesor the distributionof productsbe-
index as a measure of corporate diversification. comes more equal, the Herfindahl-basedindex d2
The indexhas the form ESY2, where s1 is the share increases at a decreasingrate, d3 increases at a
of the ith product in total sales. McVey (1972) constantrate, and the entropyindex d4 increases
uses the index directly;Berry(1971) suggeststhe at a decreasingrate in productnumberbut at an
form increasingrate in productdistribution.The char-
d221-Es7.
acterizationsof diversificationby the three in-
(2) dexes are
quite different. In the end, however,
The index does not addressproduct heterogene- none of the three indexes satisfies the hetero-
ity, but it does satisfyboth the numberand distri- geneity property.
bution properties.This can be demonstratedby Pomfret and Shapiro(1980) introducea diver-
addingand subtracting1/n to (2), where n is the sification measure that does attempt to address
numberof distinctfour-digitproducts: the heterogeneityrequirement:
d5 - ESjVijx (6)
d + 1: - - S? ~~~~~(3)
( n) (n2
where i and j represent primaryand secondary
The first term in (3) varies directly with the
industries,respectively,and
numberof products,thereby satisfyingthe num-
ber property.The second term can be interpreted vii = 0 if the producingagent operates in one
as the difference between two Herfindahl in- 4-digit industryonly,
dexes, one based on actual product shares (si) = 1 if j is in the same 3-digit industryas i,
and a second calculatedon assumedequal shares = 2 if j is in the same 2-digit industryas i,
(1/n). The term satisfies the distributionprop- and
erty. It takes a zero value when the n products = 3 if i and j are in different2-digit indus-
are distributedequally.It takes increasinglynega- tries.
tive values as the distributionof productionbe- Except for the case where n = 1, however, the
comes more unequal. index d5 is not necessarilysensitive to variations
Two other indexes have similar properties. in either the number or share distributionof
Utton (1977) constructs a "cumulativediversifi- products.The index is calculatedby simplycom-
cation curve" and introducesan index based on biningproductshares aroundcommondummy
sj
the area above that curve: variablesvi identifyingSIC boundaries.1" Chang-
d32Erisi - 1, (4) ing the number or distributionof products as-
signed to any vij categoryleaves d5 unchanged.
where ri is the shipmentsrankof the ith product None of the existing measures of D compre-
and products are ranked in descending order. hensivelysatisfies the five properties introduced
10The measure appears 11The "concentricindex"introducedby Caves,Porter,and
d, frequentlyin the literature.See,
for example,Rhoades(1973),Gorecki(1974, 1975),Greening Spence (1980)addressesinterproductdistancein nearlyiden-
(1980),and Caves(1981). tical fashion.
above, but the cumulativehistoryprovides some lently. As for the Utton index, both derivatives
useful insights. First, most empiricalstudies ap- equal zero. While reasonable people may dis-
plying any of the above indexes treat the output agree as to the most appropriatefunctionalrep-
of each establishment as a homogeneous resentationof diversification,the naturalinclina-
product.12 Regardlessof its productmix, all ship- tion of economists is to posit that the marginal
ments are assigned the plant's primaryproduct contributionto diversificationof a change from
code. The changinginternal product diversifica- one to two or two to three distinct products
tion of establishmentsis masked. The resulting exceeds the marginal contributionof a change
indexes are measuresof industrialdiversification, from producing80 to 81 products.
not product diversification.The latter is pre- The Herfindahl-basedindex d2 exhibits the
ferred. Second, while a number of studies ac- desired negativesecond derivativesin both prod-
knowledge the importance of accounting for uct number and distribution. Furthermore, as
productheterogeneity,those that formallyincor- demonstratedby Davies (1979), for n < 100 the
porate some notion of product dissimilarityuni- Herfindahlindexplaces equalweighton the num-
versallydefine distance between productswholly ber and distributioncomponentsidentifiedin (3),
in terms of SIC codes. The Utton and Goto a desirablepropertygiven the lack of prior infor-
applicationsof d, and the Pomfret and Shapiro mation regardingwhich characteristic(if either)
index d5 are representative.Such indexes, how- should dominate.
ever, registerheterogeneityonly as productscross
SIC borders.As a result, the measured distance III. An Index of Diversification
between chemical and petroleumproductsis as-
sumed equal to the distance between chemical The critical issue is how to generalize the
productsand electricalmachinery.A more mean- Herfindahl-based index d2 to make it sensitiveto
ingful index of diversificationrequiresa distance product heterogeneity. Fortunately,the Herfin-
metric more sensitive to dissimilaritiesamong dahl index already has embedded in it an elemen-
products.Third,and most important,of the basic tary treatment of heterogeneity. Two products
measures described above, the Herfindahl-based are either identical or different.If identical,out-
index appears the most promisingcandidate for put units are simplyaggregatedand given a com-
furthergeneralization.Index d2 fails the hetero- mon product subscript.If different,output bun-
geneity requirementbut otherwise satisfies the dles are given distinctsubscripts.
remainingfour propertiesof a well-designedin- Makingthis bivariatenotion of producthetero-
dex. geneityexplicitis the key to properlygeneralizing
The Utton (d3) and entropy(d4) indexes also d2. An exampleis instructive.Consideran enter-
satisfythe numberand distributionpropertiesbut prise that producesthree differentfive-digitprod-
(1) neither is bounded in the appealingzero-one ucts yi with shares si. The resultingvalue of d2 is
interval and (2) their second derivativesdo not
exhibit wholly desirable properties. The second =
d d=1 - S2
_525 - 2 -2
S_32
derivativesof the entropy index with respect to
the numberof productsand the extent of equality =(1-3)- (si2- (7)
in productdistributionare negative and positive,
respectively.Unless strongargumentsexist to the Contrastthis with the scenariowhere Y2 and y3,
contrary,the derivativesshouldbe signed equiva- perhaps because they fall within the same four-
digit productgroup, are treated as identical:
12
Examples include Berry (1971, 1974), McVey (1972),
Rhoades (1973), Gorecki (1974, 1975), Utton (1977), 2= 1-5S2- (52++ S)
Jacqueminand Berry (1979), Greening(1980), and Pomfret
and Shapiro (1980). Goto (1981) and Gorecki (1980) are
exceptions.Goto considersthe commoditycompositionof the = (1- )-E si2-9 )2S2S3 * (8)
124 largest Japanese firms in the 1963-75 period. Gorecki
tests three mechanicalapproximationsto the distributionof
plant output against actual output vectors in 1970 for 890 The number and distributioncomponentsin (8)
plantsin 155 Canadianfood processingenterprises. still are computedon the basis of the true num-
to n productsyields
The input cost shares {Wkj} and {wi} serve as
d2= 1- ES- - E SiSkZik estimatesof the parameters{I,kj} and {/,ij} defin-
i i koi
ing the Cobb-Douglastechnologies Yk = ki
and y1= HlXfii for the kth and i h products,
( n) E(i n2 respectively.Assumingcompetitivemarkets,
E ESiSkZik, (9)
i koi
dln y1 X_plj _
where =
E
wlj,
The heterogeneityterm in (10) focuses solely wheneverZik = 1. For Zik = 0, however, ik takes
on productdistancebased on differinginput sets some positivevalue in the zero-one interval,ap-
and attempts to incorporatethis dimension for- proachingunity as the productsbecome increas-
mally into a measure of diversification.This pro- ingly dissimilar.
duction orientationfor otik is consistentwith the It follows that d'"is bounded in the zero-two
model'sultimateapplicationto Censusdatawhich interval. Dividing (10) by 2 bounds d'2 in the
focus on manufacturingactivityat the production desirablezero-one interval:
site.
Three characteristicsof the heterogeneityterm
D 1/241 - -i E ESi Sk(Zik -
Oik)].
EESiSk(Zik - oTik) in (10) are worth noting. First,
ii k0i
the notion of productdistancerepresentedby the (13)
difference [wkj - wi,1 is consistentwith the basic
distance function underlyingthe Herfindahldis-
The index takes a zero value for any entity that
tributionterm E(s2 - 1/n2). The latter is a sim-
producesonly a single product;D = 0 when si =
plified form of (1/2n)EE(si - sk)2. The distribu-
1 and Zik and otik are undefined.In contrast, D
tion term therefore defines product distance in
approachesunity as a producingunit approaches
terms of output shares (si - Sk); the heterogene-
a state of "perfect diversification;"lim D = 1 as
ity term defines product distance in terms of -> 0 and ik
s7 1 for all i, k.
-*
input shares (Wk1 - wij). Second, EESiSk(Zik -
Finally, a simplifying restriction can be im-
aOik) measures the heterogeneity of the firm's
posed. Since the index D is to be applied to the
whole productmix and can be rewrittenin a form
five-digit product data reported to the Census
revealing the heterogeneity contributionarising
Bureau by each manufacturingestablishment,it
througheach product:
is reasonableto assume that productsidentified
E E SiSk(Zik - ik) = EaSi7o (12) by a reportingplant as distinctfive-digitproducts
i k* ii satisfy some minimum,howeverminimal,notion
4 Sk(Zi/ - ik) represents the across of product dissimilarity.It follows that all Zik
where o five-digitproduct pairs take zero values.
product heterogeneityspecific to product i and
Symmetrically,it is reasonableto assume that all
si(oi identifies the contributionof that product's
output units grouped by a respondingestablish-
heterogeneity to overall firm diversification.
ment within a common five-digitcode are suffi-
Third, note that the product shares Sk play an
ciently similarso as to be aggregateddirectlyas
importantrole in the definitionof oi and there-
in (10), therebymakingunnecessaryall corrective
fore Esiaoi. The extent of product heterogeneity
adjustments- 2SiSk Zik where Zik would take a
arising through oa1 in, say, a three-productfirm
unit value. In short, the length of the product
where s1 = 0.5, s2 or 53= 0.49, o-12= 1.0, and
vector reported by establishmentsis accepted,
a13= 0.1 depends on whether 52 or S3 has the
therebyreducing(13) to its final form:
0.49 share. Intuition suggests that the measured
heterogeneityof the firm'sproductmix shouldbe
higher (lower) if s2 equals 0.49 (0.01). Hetero- D = 1/2 1 - ESi E SikoSikY (14)
geneity should be higher (lower) when it is the ii koi
two dominant products that are very dissimilar
(similar).The variablesSk ensure this result. One nice propertyof the index is that it can be
Heterogeneity thus depends importantly on decomposed into unique number, distribution,
both Zik and aik. The discrete variable Zik con- and heterogeneitycomponents:
tinues to play its correctiverole, ensuringthat the
index of diversificationis not upwardbiased by
any disaggregationof a homogeneous output D = 1/2[(1 --) + E - s 2)
bundle into artificiallydistinct product groups.
The continuousvariableaik, in contrast,captures
the extent of dissimilaritybetween any two prod- + E ESiSkTikj (15)
ucts for which Zik = 0. By definition, aoik = 0 i koi
The definition of product heterogeneity re- The establishment-basedindexes not only are
quiresa vector of input cost shares (wij) for each small but also have decreased over the 1963-82
five-digit product. Ten input categories in the period. Seventeen of twenty industries exhibit
Census establishmentfiles are distinguished:pro- decliningdiversification.The three exceptionsare
duction workers, other labor, fuel, electricity, tobacco (an increase in the number of outputs,
purchased services, agriculturalmaterials, min- though each new product has a very small share
eral inputs, nondurablematerials,durablemate- in production),apparel (a slight increase in the
rials, and capital. The representativeinput mix number of dissimilarproducts),and leather and
for each five-digitproduct is defined as the un- leather products (an increase in the number of
weighted mean of input cost shares in those es- products and a move to a more heterogeneous
tablishmentsmanufacturingonly that product.16 productmix).20
This procedure guarantees product-specific, The ranking of industries, however, has re-
Cobb-Douglasinput requirementvectors. mained relatively unchanged with only two no-
Diversificationindexesare constructedfor each table exceptions:rubberand plasticproducts,and
establishmentand each enterprise and then are leather. The formerexperiencedthe most severe
assigned to their appropriate two-digit indus- decline. Its rankfell from fourth to eighteenth as
tries.'7 Unweighted and shipments-weighted its index declined steadily from 0.280 in 1963 to
mean diversificationindexes are calculated for 0.106 in 1982, the result of the industry'smove to
each industryaggregate.18 a more limitednumberof productsand to a more
Table 1 reportsweightedand unweightedaver- homogeneous product mix. The latter, leather
age diversificationindexes for establishmentsin and leather products, experienced precisely the
each two-digit industry. Consider first the opposite trend, as its rank climbed from nine-
weightedindexes,the preferredmeasureof diver- teenth in 1963 to thirteenthin 1982.
sification within an industry.The 1982 indexes The unweightedindexes, also reportedin table
range from a low of 0.049 in tobacco to 0.428 in 1, generally are small and declining over time.
petroleum and coal products. Sixteen of twenty The importantobservation,however, is that the
industries have 1982 indexes below 0.206, the unweighted index is smaller than its weighted
manufacturingaverage.Decompositionof this av- counterpart in every instance, suggesting that
erage into its number, distribution,and hetero- large establishments are more diversified than
geneity components yields 0.244, - 0.101, and small ones. The 1982 all-manufacturing weighted
0.063, respectively.Given the definitionsin (15), average index is 0.206; the unweighted mean
the componentsrevealthat establishmentsin 1982 is only 0.051. The fact that the rate of decline in
on averageproduced 1.95 fairly similarproducts the unweighted index over the 1963-82 period
in very unequal (86:14) proportions.On the in- (-37%) far exceeds the correspondingrate for
dex scale of zero to one, individualmanufacturing the weighted index (- 8%) indicatesthat the de-
plants, with the exception of those producing cline in diversificationis especiallysevere among
petroleum and coal products and chemicals,are the smallestestablishments.
not very diversified.19 Table 2 presents mean enterprise-basedin-
dexes for each industry.It has the same formatas
table 1, but there the similaritystops. The 1982
16
weighted indexes range from 0.252 in lumber to
Increasingthe numberof input categoriesto twenty in-
creases the establishmentand enterprise diversificationin-
0.589 in petroleum and coal products.Six of the
dexes, on average,by approximately 0.01 points. twenty industries have average enterprise-based
17
The 1963 Census of Manufactures data file contains diversificationindexes above 0.50. Four have in-
296,033 establishmentsand 265,778 enterprises.There are dexes greater then 0.55. These include paper,
348,385 establishmentsand 294,397 enterprisesin the 1982
Censusfile. chemicals, petroleum and coal products, and
18 Unweightedand shipments-weighted indexesfor both es- transportationequipment.The 0.464 all-industry
tablishmentsand enterprisesalso are constructedfor each
4-digitindustryand are availableon requestfromthe authors
or the CensusBureau.
19 The 1982number,distribution,and heterogeneitycompo-
20
nents take 0.413, -0.122, and 0.137 values, respectively,in The parentheticalexplanationsare derivedfromobserved
petroleumand coal productsand 0.297, -0.111, and 0.090 movements in the number, distribution,and heterogeneity
values,respectively,in chemicals. componentsfor each industry'sindex.
Shipments-Weighted Unweighted
1963 1967 1972 1977 1982 1963 1967 1972 1977 1982
Food and Kindred Products .242 .240 .230 .217 .192 .106 .090 .088 .079 .080
Tobacco Manufactures .040 .052 .055 .052 .049 .024 .027 .030 .025 .025
Textile Mill Products .180 .169 .167 .156 .163 .071 .064 .062 .053 .060
Apparel .118 .119 .121 .125 .137 .038 .033 .034 .037 .041
Lumber and Wood Products .147 .130 .120 .132 .141 .054 .029 .030 .040 .044
Furniture and Fixtures .202 .196 .180 .172 .185 .097 .066 .070 .060 .069
Paper and Allied Products .184 .194 .177 .162 .155 .067 .070 .061 .049 .053
Printing and Publishing .250 .237 .226 .230 .201 .155 .077 .072 .074 .064
Chemicals and Allied Products .286 .297 .289 .282 .276 .133 .118 .112 .092 .093
Petroleum and Coal Products .434 .486 .469 .464 .428 .108 .096 .090 .082 .082
Rubber and Plastics Products .280 .244 .195 .176 .106 .080 .072 .056 .058 .034
Leather and Leather Products .079 .072 .142 .128 .142 .041 .035 .054 .051 .053
Stone, Clay and Glass Products .097 .086 .081 .079 .072 .050 .032 .029 .026 .029
Primary Metal Industries .289 .286 .283 .272 .227 .100 .090 .084 .075 .069
Fabricated Metal Products .155 .155 .136 .131 .116 .069 .059 .057 .049 .046
Machinery, except Electrical .223 .229 .228 .217 .191 .060 .049 .045 .039 .040
Electrical Equipment and Supplies .233 .224 .200 .165 .143 .074 .069 .060 .043 .043
Transportation Equipment .200 .206 .161 .163 .188 .073 .064 .052 .044 .045
Instruments and Related Products .227 .212 .217 .186 .169 .052 .047 .048 .038 .038
Miscellaneous Manufacturing .114 .124 .120 .113 .112 .045 .034 .028 .029 .030
All Manufacturing .223 .224 .208 .211 .206 .081 .059 .056 .052 .051
Shipments-Weighted Unweighted
1963 1967 1972 1977 1982 1963 1967 1972 1977 1982
Food and Kindred Products .400 .436 .479 .472 .428 .103 .081 .074 .062 .063
Tobacco Manufactures .355 .410 .464 .477 .532 .041 .049 .063 .060 .074
Textile Mill Products .428 .434 .468 .456 .448 .070 .062 .061 .050 .059
Apparel .182 .224 .287 .266 .290 .040 .033 .035 .037 .040
Lumber and Wood Products .259 .248 .352 .359 .252 .053 .028 .028 .038 .041
Furniture and Fixtures .245 .262 .318 .263 .255 .095 .062 .066 .058 .065
Paper and Allied Products .498 .524 .539 .541 .561 .070 .069 .067 .056 .064
Printing and Publishing .316 .313 .328 .334 .332 .156 .074 .068 .069 .059
Chemicals and Allied Products .491 .598 .590 .584 .563 .128 .102 .095 .070 .076
Petroleum and Coal Products .545 .610 .631 .606 .589 .101 .090 .088 .091 .093
Rubber and Plastics Products .480 .497 .487 .444 .373 .078 .070 .055 .057 .037
Leather and Leather Products .216 .225 .331 .308 .266 .040 .034 .044 .041 .046
Stone, Clay and Glass Products .314 .333 .412 .412 .294 .053 .034 .032 .030 .032
Primary Metal Industries .553 .567 .576 .592 .542 .099 .091 .085 .075 .073
Fabricated Metal Products .296 .346 .347 .355 .276 .069 .058 .056 .050 .048
Machinery, except Electrical .350 .398 .476 .462 .409 .056 .044 .040 .035 .037
Electrical Equipment and Supplies .529 .509 .494 .465 .454 .071 .065 .056 .039 .042
Transportation Equipment .529 .563 .571 .567 .561 .073 .064 .052 .045 .046
Instruments and Related Products .420 .440 .420 .424 .426 .052 .044 .045 .036 .038
Miscellaneous Manufacturing .232 .237 .319 .247 .294 .046 .033 .028 .029 .030
All Manufacturing .430 .462 .486 .488 .464 .080 .055 .051 .048 .047
average contrasts with the establishment-based unweighted indexes increased in only two in-
averageof 0.206. stances." The opposite trends in the all-manu-
The number, distribution, and heterogeneity facturing indexes in table 2 are striking. The
componentsshare the limelightin table 2. These underlyingdistributionof enterprise diversifica-
enterprise-basedcontributionsto the all-manu- tion has become bimodal. Enterprise numbers
facturing index 0.464 are 0.390, - 0.083, and and shipmentshave clusteredin opposite tails.
0.157, respectively.The numbercomponentis the An equally important insight arises from a
dominantforce (enterprises,on average,produce comparisonof tables 1 and 2. While 17 of 20
4.5 differentfive-digitproducts),but the hetero- industries had declining establishment-basedin-
geneity contributionhas a magnitude equal on dexes between 1963 and 1982, enterprise-based
averageto more than 40%of the numbercompo- diversificationincreased in 14 of 20 instances.
nent. Its contribution(0.157) to enterprisediver- There appears to be a shift over time in enter-
sificationis two and one-half times its contribu- prise activity.Risingenterprisediversificationand
tion (0.063) to establishmentdiversification.The falling establishment diversificationimply that
contributionof product distributionis significant firmsappearto be movingtowardmore homoge-
as well. Comparedto its establishmentcounter- neous production within establishments while
part (-0.101), the enterprise-baseddistribution holding a more diverse portfolio of establish-
term (-0.083) is smaller in both absolute and ments.
relative values.21Given that the bounds on the Table 3 investigatesfurther the link between
distributionterm in (15) are an increasingfunc- establishmentand enterprisediversification.Since
tion of product number, it follows that product an enterpriseis defined as the sum of its estab-
distributioncontributesrelativelymore to enter- lishments,an enterprise'sdiversificationmust be
prise than to establishmentdiversification.The a functionof diversificationwithin and amongits
presence of an unequal product distributionoff- plants. Consider adding and subtractinga ship-
sets 40%of the numbercontributionin the estab- ments-weightedaverageof diversificationindexes
lishment model but only 22% in the enterprise for an enterprise's establishmentsto the right-
model. Overall, enterprises produce a larger hand side of an identityequatingthe enterprise's
numberof more dissimilarproducts and a more diversificationindex with itself:
equal distributionof productsthan do establish-
ments.
Di I = EaijD est + (Dent EaijDest),i
The enterprisediversificationrankingof indus- I I
tries is remarkablystable over the twenty-year (16)
period. The only outliers are rubber and plastic
products and tobacco. The former's ranking
and
dropped from seventh to twelfth between 1963 where ent and est represent "enterprise"
the
and 1982. The latter's ranking moved from "establishment,"respectively,and aij equals
eleventh to fifth and today, most assuredly, is shipments'share of the jth plant in the ith firm.
highergiven the 1985 acquisitionsof Nabisco and The first term on the right-handside of (16)
General Foods by R.J. Reynolds and Philip reflects the contributionof within-establishment
Morris,respectively. diversificationto overall Dient. Each Dfest is calcu-
mix and
One of the more interesting insights surfaces lated wholly in terms of the product
of the establishment. The
from a comparisonof weighted and unweighted input requirements jth
that differences in prod-
enterprise indexes. Quite clearly, large enter- second term recognizes
are
prises are more diversifiedthan small ones. The uct mix and input combinationsacrossplants
enterpriseindex captured in the enterprise measure Dient but not
1982weightedall-manufacturing
is ten times its unweightedcounterpart.But con- in the individual Dest. The difference between
sum
siderthe trends.Weightedindexesincreasedover Dient and the weighted establishment
the 1963-82 period in 14 of 20 industries,while 22
Three out of four industriesexhibit a decline in enter-
prise diversificationbetween 1977 and 1982. This may be
21
Recall, an increasinglyequal productdistributionreduces more a statisticalartifactthan an indicationof a trend given
the negativedistributionterm in absolutevalue. the poor state of the economyin 1982,the censusyear.
Shipments-Weighted Unweighted
Contribution Contribution
Enterprise Diversification Diversification Enterprise Diversification Diversification
Diversi- within among Diversi- within among
fication Establishments Establishments fication Establishments Establishments
All Manufacturing:
1963 .430 .038 .392 .080 .069 .011
1967 .462 .037 .425 .055 .045 .010
1972 .486 .035 .451 .051 .039 .012
1977 .488 .023 .465 .048 .036 .012
1982 .464 .014 .450 .047 .033 .014
Two-Digit Industries, 1982:
Food and Kindred Products .428 .018 .410 .063 .035 .028
Tobacco Manufactures .532 .000 .532 .074 .009 .065
Textile Mill Products .448 .013 .436 .059 .027 .032
Apparel .290 .036 .254 .040 .027 .012
Lumber and Wood Products .252 .049 .203 .041 .034 .007
Furniture and Fixtures .255 .051 .204 .065 .050 .015
Paper and Allied Products .562 .006 .556 .064 .029 .036
Printing and Publishing .332 .045 .288 .059 .050 .009
Chemicals and Allied Products .563 .009 .555 .076 .040 .036
Petroleum and Coal Products .590 .004 .586 .093 .028 .065
Rubber and Plastics Products .373 .016 .357 .037 .022 .015
Leather and Leather Products .266 .033 .233 .046 .027 .019
Stone, Clay and Glass Products .294 .011 .283 .033 .020 .013
Primary Metal Industries .542 .007 .535 .073 .038 .035
Fabricated Metal Products .276 .029 .247 .048 .033 .015
Machinery, except Electrical .409 .016 .393 .037 .026 .011
Electrical Equipment and Supplies .454 .007 .447 .042 .022 .019
Transportation Equipment .561 .003 .559 .046 .028 .019
Instruments and Related Products .426 .009 .418 .038 .022 .016
Miscellaneous Manufacturing .294 .020 .273 .030 .023 .007
every census year and in 16 of 20 industriesin lishments has declined markedly, from an all-
1982, in spite of the fact that the weighted diver- manufacturingaverage of 0.038 to 0.014. In
sificationindex exceeds the unweightedindex in contrast,the contributionto enterprisediversifi-
every industry in every year and never by less cation from diversificationamong establishments
than a factor of 4. The measures suggest that has increased from 0.392 to 0.450. By 1982 the
large enterprises,on average,have their individ- latter contribution to enterprise diversification
ual plants producinga more homogeneousprod- was more than 32 times the former.
uct mix than do much smaller firms, those typi- An importantimplicationfollowsdirectly.Con-
cally havingsingle plants. Perhapsthis reflects a trary to the popular argument, technical
lesser degree of flexibility available to those economies of scope appear to play little role in
smallest firms constrained to operate a single explaining the measured increase in enterprise
plant. Whateverthe interpretation,it illustrates diversification.If such technical economies exist
another dimension of importantdifferencesbe- and are activelypursuedby firms,the economies
tween large and small enterprises.23 most likelywouldbe realizedat the establishment
level, but it is precisely here that diversification
V. Conclusion has decreased.Between 1963 and 1982 establish-
ment diversificationhas declined in 17 of 20
A novel feature of this paper is the formal industries.On average,establishmentswere pro-
incorporationof a continuousmeasureof product ducing fewer and less dissimilarfive-digitprod-
heterogeneity into an index of diversification. ucts in 1982 than in 1963. In fact, over this
Census data suggest that the heterogeneitycom- 20-year interval wholly undiversified (D = 0)
ponent is not unimportant.Incorporatinghetero- plants in 13 of 20 industriesincreasedboth their
geneity into the diversificationindex for each proportion of industryestablishmentsand their
manufacturingenterpriseadds, on average,0.157 share of industryshipments.If anything,at the
or more than 50% to the value the indexeswould establishmentlevel firms appear to be seeking
have were product heterogeneityignored (0.464 more conventional scale rather than scope
versus 0.307). The variance in the heterogeneity economies.
contributionover time and across industries is
substantial.The heterogeneitycomponentranges
from 0.054 (apparel, 1963) to 0.256 (petroleum
REFERENCES
and coal products, 1972). Product heterogeneity
should not and need not be ignored. Baumol,WilliamJ., "ScaleEconomies,AverageCost, and the
Profitabilityof MarginalCost Pricing,"in Ronald E.
Applicationof the diversificationindex to U.S. Grieson(ed.), Public and Urban Economics (Lexington,
manufacturing establishments and enterprises MA: LexingtonBooks, 1976),43-57.
leads to a number of interesting interpretative , "On the Proper Cost Tests for Natural Monopoly in
a MultiproductIndustry,"American Economic Review
conclusions but one stands out in importance. 67 (Dec. 1977),809-822.
Enterprisesappear to be restructuringtheir pro- Berndt,Ernst R., and MelvynA. Fuss, "EconomicCapacity
duction activity, moving toward more homoge- Utilization for Multi-ProductFirns with Multiple
Quasi-Fixed Inputs," unpublishedmanuscript(Dec.
neous production within establishments while 1987).
holding a more diversifiedportfolio of establish- Berry, CharlesH., "CorporateGrowthand Diversification,"
ments. While all-manufacturingenterprisediver- Journal of Law and Economics 14 (Oct. 1971),371-383.
, "CorporateDiversificationand Market Structure,"
sificationhas increasedover the 1963-82 period, Bell Journal 5 (Spring1974),196-204.
the contributionof diversificationwithin estab- Caves,RichardE., "Diversificationand SellerConcentration:
Evidencefrom Changes,"this REVIEW 63 (May 1981),
23
289-293.
That the weightedamong-establishment contributionex- Caves,RichardE., MichaelE. Porter,andA. MichaelSpence,
ceeds the unweightedcontributionin table 3 followsautomat- Competition in the Open Economy (Cambridge:Har-
icallyfromthe dominanceof single-plantfirmsin everyindus- vardUniversityPress, 1980).
try. For single-establishment firms,the among-establishment Davies, Stephen, "Choosingbetween ConcentrationIndices:
term (DF"t - Ei aijPD) definitionallyequals zero, a condi- The Iso-ConcentrationCurve," Economica 46 (Feb.
tion applyingto nearly 80% of enterprisesin 1982. These 1979),67-75.
firmscollectivelyhave little effect on the weighted analysis. Federal Trade Commission,Economic Report on Corporate
Each firm, however,whether single- or multi-plant,has an Mergers (Washington,D.C.: U.S. GovernmentPrinting
equalweightin formingthe unweightedmeasures. Office, 1969).
, Statistical Report on Mergers and Acquisitions try Diversification,"this REVIEW 64 (Nov. 1982),
(Washington,D.C.: U.S. GovernmentPrintingOffice, 646-657.
1981). MacDonald,JamesM., "R & D and the Directionsof Diversi-
Gorecki,Paul K., "The Measurementof EnterpriseDiversi- fication,"this REVIEW 67 (Nov. 1985),583-590.
fication,"this REVIEW 56 (Aug. 1974),399-401. McVey,J. S., "The IndustrialDiversificationof Multi-Estab-
, "An Inter-Industry Analysis of Diversification in the lishment ManufacturingFirms: A Developmental
United Kingdom ManufacturingSector," Journalof Study," Canadian Statistical Review 47 (July 1972), 4, 6,
Industrial Economics 24 (Dec. 1975), 131-146. 112-117.
,"A Problemof Measurement,from Plants to Enter- Panzar,John C., and Robert D. Willig,"Free Entryand the
prises in the Analysis of Diversification:A Note," Sustainabilityof Natural Monopoly,"Bell Journal 8
Journal of Industrial Economics 28 (Mar. 1980), (Spring1977),1-22.
327-334. Pomfret,Richard,and Daniel Shapiro,"FirmSize, Diversifi-
Gort, Michael, Diversification and Integration in American cation, and Profitabilityof Large Corporationsin
Industry(Princeton:PrincetonUniversityPress, 1962). Canada," Journal of Economic Studies 7 (Nov. 1980),
Goto, Akita, "StatisticalEvidence on the Diversificationof 140-150.
Japanese Large Firms," Journal of Industrial Eco- Rhoades, Stephen A., "Ihe Effect of Diversificationon In-
nomics29 (Mar. 1981),271-278. dustryProfitPerformancein 241 Manufacturing Indus-
Greening, Timothy, "Diversification,Strategic Groups and tries: 1963,"this REVIEW 55 (May 1973),146-155.
the Structure-Conduct-Performance Relationship:A Sharkey,William W., "Existence of SustainablePrices for
Synthesis,"this REVIEW 62 (Aug. 1980),475-477. Natural MonopolyOutputs,"Bell Journal 12 (Spring
Jacquemin,Alexis, and CharlesBerry,"EntropyMeasureof 1981), 144-154.
Diversificationand CorporateGrowth,"TheJournalof Utton, M. A., "LargeFirmDiversificationin BritishManufac-
Industrial Economics 27 (June 1979), 359-369. turingIndustry,"TheEconomicJournal87 (Mar.1977),
Lemelin,Andre, "Relatednessin the Patternsof Inter Indus- 96-113.