SOCIAL SCIENCE RESEARCH
ARTICLE NO.
26, 377–398 (1997)
SO970591
Legitimation, Geographical Scale, and Organizational
Density: Regional Patterns of Foundings of American
Automobile Producers, 1885–1981
Lyda S. Bigelow and Glenn R. Carroll
Haas School of Business, University of California, Berkeley
Marc-David L. Seidel
School of Business, University of Texas at Austin
and
Lucia Tsai
Department of Economics, University of California, Berkeley
Do organizational processes of legitimation and competition operate within different
boundaries corresponding to different geographical levels of analysis? Following Hannan
et al. (1995), this analysis explores the possibility that legitimation operates on a broader
geographical scale (less constrained by political and physical barriers) than does competition. We test the argument by examining founding rates of American automobile producers
from 1885 to 1981, within the framework of density-dependent modeling. Our findings
suggest that within the United States, legitimation operated on a national scale while
competition proceeded primarily on a regional level. Comparison with automobile producer populations in Europe yields differences in application and interpretation of the
theory. r 1997 Academic Press
Organizational theorists have long been fascinated with questions about the
boundaries of formal organizations (Scott, 1992). Recently, similar interest has
developed around questions about the boundaries of processes involving sets of
The research reported here is part of a collaborative research project with Michael T. Hannan,
Stanford University. We appreciate his comments on an earlier draft as well as those of John Torres and
Elizabeth Dundon. The project is supported by grants from the Sloan Foundation and the Institute of
Industrial Relations, University of California, Berkeley.
Address correspondence and reprint requests to Glenn R. Carroll, Haas School of Business,
University of California, Berkeley, S545 Student Services Building, Berkeley, CA 94720-1900.
377
0049-089X/97 $25.00
Copyright r 1997 by Academic Press
All rights of reproduction in any form reserved.
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03 kris
No. of Pages—22 First page no.—377 Last page no.—398
378
BIGELOW ET AL.
many organizations. For instance, institutionalists delineate societal sectors before they concern themselves with issues such as the diffusion of normatively
‘‘appropriate’’ organizational forms within societal sectors (Scott and Meyer,
1983). Similarly, resource dependence theorists demarcate networks so that they
can analyze opportunities and constraints within interdependent webs of firms
(Burt, 1992). And, likewise, organizational ecologists define organizational population boundaries in order to model competition processes among populations of
organizations (Hannan and Carroll, 1992).
Recent work in organizational ecology proposes a theoretically motivated
strategy for dealing with the boundaries of organizational processes. In continued
development of the density-dependent model of organizational evolution, researchers have specified that the two main processes hypothesized to operate do so
within different social and geographical boundaries.1 In particular, Hannan and
Carroll (1992) and Hannan et al. (1995) contend that legitimation of an organizational form (the first process) occurs on a broader geographical scale than does
competition among organizations characterized by the form (the second process).
That is, they argue that ‘‘competitive environments tend to be more local than
institutional environments’’ (Hannan et al., 1995 p. 6).
The argument yields important differences in the specification of mathematical
models of organizational evolution and their empirical testing. So, in their
analysis of the European automobile industry, Hannan et al. (1995) specify a
model that shows legitimation operating on a continent-wide basis, while competition occurs primarily within nation–state boundaries.2 Hannan et al. (1995)
present evidence supporting this specification for Belgium, France, Germany, and
Italy, but it does not hold for Britain. Given Britain’s geographical location, this
pattern of findings suggests that physical geography may play a role in delineating
population processes. The thesis of relatively local competition might also work
itself out at different geographical levels in correspondence to different sociopolitical boundaries. In theoretical terms, the issue concerns the extent to which
legitimation and competition are circumscribed by sociopolitical boundaries
(Barnett and Carroll, 1993) or by physical geography (Carroll and Wade, 1991).
Our efforts here represent an attempt to continue exploring questions of
multiple boundaries for processes pertaining to organizational populations. In
order to facilitate cumulativity, we do so in a way that closely follows the research
of Hannan et al. (1995). Specifically, we continue to examine questions of
geographic level of applicability for density-dependent processes of legitimation
and competition. We also continue to study the automobile industry. However, our
analysis does shift locales: we investigate the American automobile industry
rather than the European industry. The American industry initially sprang up and
1 Barnett and Carroll (1993), Carroll and Wade (1991), and others have used this same strategy
earlier but because Hannan et al. (1995) have a more developed theory about effects at different levels
we focus on their paper.
2 The full implementation of this strategy empirically requires collection and analysis of data on at
least two different levels of analysis.
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
379
LEGITIMATION, SCALE, AND DENSITY
grew with much contact from European producers. However, it quickly surpassed
the European countries in size and scale (see Carroll and Hannan, 1995 for a
capsule review). The United States is also obviously much larger geographically
and more distant from the core countries of Europe than Britain. These facts allow
us to explore whether legitimation and competition are constrained in this context
primarily by nation–state boundaries or by geographical factors. We do so by
contrasting performance of the multilevel density model specified for the entire
country relative to Europe as opposed to a specification of internal geographical
regions relative to the nation–state. The findings show the value of Hannan et al.’s
(1995) theoretical argument but suggest additional issues requiring consideration
in application.
REGIONAL PATTERNS OF INDUSTRY EMERGENCE IN THE U.S.
AUTOMOBILE INDUSTRY
Although the automobile was first invented and developed in France and
Germany in the early 1880s, technical knowledge and societal expectations
regarding it quickly spread to the United States. Three regions of the United States
soon came to harbor pockets of emerging automotive manufacturing enterprises.
The Duryea brothers of Springfield, Massachusetts built the first gasoline automobile (destined for production) in the United States in 1893. Soon after, in 1894,
Elwood Haynes and the Apperson brothers completed their prototype gasoline
automobile in Indiana. That same year, in Philadelphia, the Morris and Salom
electric car was built and soon the firm grew into a large holding company
producing electric cars outside New York City (Flink, 1988). Thus, the three
regions of New England, the Midwest (primarily Indiana, Ohio, and Michigan),
and the New York metropolitan area represented the centers of initial industry
development.
Early automobile technology diffused widely. The proliferation of trade journals and magazines devoted to the ‘‘horseless carriage’’ spurred this diffusion. So
too did widely publicized endurance races such as the Chicago Times–Herald race
in 1895, in which just 2 of 83 entrants were able to complete the 55-mile course
(Rae, 1959; Flink, 1970). Races were held throughout the three regions of initial
automotive activity. They soon spread to other parts of the country as automobiles
and automobile manufacturing firms appeared in other states.
The early history of the automobile industry in the United States hints at the
possibility of varying geographical influences. On the one hand, automotive
historians (e.g., Rae, 1959, 1984; Flink, 1970) note that technological development did not occur in isolation and ideas about what automobiles should look like
and, by extension, the organizations that produced them diffused widely. Such a
process would support Hannan et al.’s (1995) contention that legitimation
operates at a broad level, and that the diffusion mechanisms which foster
legitimation are relatively unconstrained by local factors. On the other hand,
historians also note that in the early period of the industry, the three regions of
initial automotive development each spawned and developed its own technology.
SSR591
@xyserv3/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
rich
380
BIGELOW ET AL.
In particular, the New York metropolitan area cultivated the production of electric
cars; the New England region, propulsion by steam technology; and the Midwest,
gasoline cars.3 The fact that alternative automobile technologies coexisted in all
five regions highlights the scope of the diffusion process. However, the association of regions with specific technologies suggests that local conditions and local
resources may have played an important role in subsequent founding patterns.
The spatially bounded development of alternative automobile technologies is
the main reason we explore regional, rather than state- or city-level density in the
model. Rae (1959, 1984) and Rubenstein (1992) argue that regionally based
differences in proximity to raw materials, natural endowments, terrain, and
density of urban centers led to the evolution of the three distinct automobile
manufacturing regions identified above. Rubenstein (1992), for example, describes the existence of a network of six recharging stations between New York
and Philadelphia at the turn of the century (at a time when such stations were rare)
as evidence of the popularity of electric cars in the Mid-Atlantic region. In another
study analyzing locational patterns of automobile producers from 1895 to 1958,
Boas (1961) also presents empirical evidence of the early establishment of
distinct multistate regions of activity. He finds that four regions emerged by 1900
(roughly consistent with our Midwest, New England, Mid-Atlantic, and West
regions described below) and remained centers of production throughout the
period of his study.4 He argues that these regions emerged in conjunction with
other manufacturing and urban centers. He also emphasizes that automobile
production thrived in areas beyond southern Michigan.
More generally, we note with interest Hounshell’s (1984) argument that many
manufacturing methods were regionally constrained. He observes that among
bicycle manufacturers (an important precursor industry, see Carroll et al., 1996,
for further discussion) differences in production technique were associated with
regions. Bicycle manufacturers in the Midwest were more likely to use metal
stamping techniques than were bicycle manufacturers located elsewhere. Metal
stamping was eventually incorporated in the automobile production process.
In sum, historians and economic geographers alike routinely describe regionally based rather than city- or state-based clusters of activity.5 So we explore its
feasibility as the appropriate level for examining variations in organizational
evolution within the United States.
Figures 1 through 3 depict organizational density (counts of the number of
3 Of course, these technological distinctions only represented initial trends—by 1914, most car
manufacturers were producing gasoline cars.
4 Though by the end of this period a handful of cities emerge as the dominant centers of production,
e.g., New York, Boston, Los Angeles, Cleveland, Detroit, and Chicago, the location of clusters of
smaller scale production continues to correspond, generally, to the four of the five regions used in the
analysis presented here.
5 The justification for regional distinctions is stronger for the New England, Mid-Atlantic,
Midwest, and West regions than for the South. We include all regions in the analysis for completeness.
We believe completeness is important, in part, because of the comparison with national density
variables.
SSR591
@xyserv3/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
rich
381
FIG. 1.
Total U.S. density and foundings.
LEGITIMATION, SCALE, AND DENSITY
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
Comparison of regional densities.
BIGELOW ET AL.
FIG. 2.
382
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
383
FIG. 3.
Comparison of regional foundings.
LEGITIMATION, SCALE, AND DENSITY
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
384
BIGELOW ET AL.
operating organizations) as well as foundings for the period 1885–1981. (We
describe our data sources below.) Note that with the exception of the South,
foundings begin in earnest at roughly the same time, around 1895, across the
United States. This is consistent with the historical record and observers such as
Rae (1959) and Flink (1970, 1988) who emphasize that the diffusion of automobile technology in the United States was not reliant on a single source of
innovation. Rather, technology diffused from discrete pockets of innovators who
all presumably had access to the technology which had been developed in Europe
10 years earlier. The rough coincidence of initial founding patterns facilitates
regional comparisons.
Examining density region-by-region shows some notable differences. Like
national density, the Midwest region reached its peak around 1910. New England
and the Mid-Atlantic experienced maximum density roughly 10 years earlier. The
West and South regions exhibited dual peaks in density, one in the pre-WWI
period and another much later. The second (small) peak may be attributable to the
concentration of alternative-fuel car producers in California and Florida and to the
rise of other specialist producers.
ORGANIZATIONAL EVOLUTION IN MULTILEVEL SYSTEMS
Extensive empirical research demonstrates that diverse organizational populations—labor unions, newspapers, breweries, life insurance companies, savings
and loans, among others—exhibit similar long-term patterns of growth, stabilization, and decline over time. Organizational ecology explains these patterns with
the theory of density-dependent legitimation and competition (Hannan, 1986;
Hannan and Freeman, 1989; Hannan and Carroll, 1992). According to this theory,
in the initial phase of a population’s development, when density is low, increases
in density are associated with legitimation or social ‘‘taken for grantedness.’’ That
is, legitimation increases as more organizations enter the population initially.
Subsequent increases in density yield relatively smaller gains in legitimation such
that legitimation increases with density but at a decreasing rate. Rising legitimation increases founding rates and depresses failure rates. At high levels of density,
competition processes dominate because competition increases with density at an
increasing rate. Competition entails a tightening of the resource base on which
extant organizations in a population depend and which potential entrants may
consider when making their entry decision. So competition depresses founding
rates and increases mortality rates. A third component of the theory—the so-called
density delay effect—predicts that density at the time of founding has an enduring
competitive effect on mortality. Putting the arguments together implies that both
founding and failure rates track contemporaneous density nonmonotonically. The
relationship between contemporaneous density and founding rates shows an
inverted U shape, and the relationship between contemporaneous density and
mortality displays a U shape. Density at the time of founding shows a monotonic
positive relationship with mortality.
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
385
LEGITIMATION, SCALE, AND DENSITY
We focus here on whether different geographic levels of analysis lead to
changes in the effects of density on population vital rates. A variety of available
studies explore this question empirically (see Carroll and Wade, 1991; Swaminathan and Wiedenmayer, 1991; Hannan and Carroll, 1992; Baum and Singh,
1994; Haveman and Romanelli, 1994; Lomi, 1995; Hannan et al., 1995). Results
generally support the prediction that local density has a stronger competitive
effect, at least on founding rates, than global or nonlocal density. The few studies
which estimate both vital rates suggest that foundings may be more sensitive to
geographic specification of density than mortality.
In a study of American brewers, Carroll and Wade (1991) find modest evidence
that founding models are more sensitive to density than mortality when varying
the geographic level of analysis. They report the predicted nonmonotonic regional
density effects in eight of nine regional founding models as opposed to seven of
nine models which use national density terms. The predicted nonmonotonic
density effects are revealed in only four of nine regional mortality models which
specify density at the regional level as compared to seven of nine mortality
models which use national density terms. In an analysis of mortality rates of
Bavarian brewers, Swaminathan and Wiedenmayer (1991) find that both regional
and national density terms show the predicted effects. They, too, suggest that the
case might be different for foundings.
Haveman and Romanelli (1994) and Baum and Singh (1994) also find empirical support for stronger localized competitive effects in populations of savings
and loans and day care centers, respectively. Lomi (1995) in a founding analysis
of Italian cooperative banks also finds that competitive effects are manifested
more strongly through local than nonlocal density.
The theoretical question remains, however, as to why, and under what conditions, different levels of density should have different effects on either vital rate.
Most of the above studies develop convincing ad hoc interpretations but they fail
to advance a general theoretical argument. Hannan et al. (1995) address this
important issue. They argue, generally, that legitimation processes operate more
broadly than competition processes for two primary reasons. First, legitimation of
an organizational form relies on a ‘‘cultural’’ diffusion process that is less
constrained by political boundaries than competition. As Hannan et al. (1995)
explain, even totalitarian regimes find it difficult, if not impossible, to curtail the
flow of cultural images from other, more permissive countries. Second, competition revolves primarily around local resources, although the degree to which local
resources drive competition may vary according to the characteristics of a given
population (i.e., degree of labor or capital intensiveness).
Hannan et al.’s (1995) argument implies that the part of the density model
associated with legitimation—the first-order term of contemporaneous density—
should be specified at the higher geographical level. The competition components
of the model—density at time of founding and the second-order term of contemporaneous density—should remain specified at the lower or local geographical
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
386
BIGELOW ET AL.
level. Thus, the model may be specified as
li(t) 5 exp (b0 1 b1 Nit 1 b2 n 2it),
(2.1)
where the founding rate of a geographic unit, li(t), depends on density of both a
higher-scale geographic unit, Ni, and the squared density of the unit itself, n2i . It is
predicted that b1 . 0, b2 , 0.6
In Hannan et al.’s (1995) analysis of European automobile producers, legitimation effects are specified at the continent level while competition effects result
from the national level. A straightforward application of this specification to the
U.S. population of automobile producers would cast legitimation-related variables at the world, continent or partial two-continent (Europe and the United
States) levels and competition at the national level. However, after estimating and
evaluating numerous such models, we have found no consistent pattern of effects
and have not been able to reach a plausible interpretation. Hannan et al. (1995)
found a similar problem with the automobile industry in Britain (although
Belgium, France, Germany, and Italy all yielded the expected pattern).
Both Britain and the United States are geographically separated by water from
continental Europe. Perhaps, this separation slows or inhibits the cultural diffusion associated with legitimation. Moreover, the sheer size and geographical scale
of the U.S. industry suggest that it might have evolved more independently,
perhaps even requiring its own legitimation process. Anecdotal historical evidence (described briefly above) suggests that much competition in the industry
was circumscribed by region rather than by the nation–state, at least for the
Midwest, New England, and Mid-Atlantic areas. For these reasons, we use here a
respecified version of Hannan et al.’s (1995) argument that competition is a more
local process than legitimation. In this analysis, we explore models with legitimation effects of density cast at the national level and competition effects at the
regional level.
Following the logic of the Hannan et al. (1995) study, this analysis proceeds in
three stages. First, foundings in each region are modeled using only regional
density terms (ni and n2i ). Thus the model takes the form
li(t) 5 exp (b0 1 b1 nit 1 b2 n 2it).
(2.2)
Next, foundings in each region are modeled using only total remaining U.S.
density terms (Ni and N 2i ). Ni is calculated by subtracting focal regional density
from total U.S. density. This avoids using overlapping density counts in the
6 This multilevel specification of density differs from that of previous studies, which varied the
geographical boundaries of density variables but always cast all terms at the same level. A common
specification of this is
l(t) 5 exp (b0 1 b1 Nt 1 b2 N 2t ),
where Nt represents the populations density at time t, and where, again, it is predicted that b1 . 0,
b2 , 0.
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
387
LEGITIMATION, SCALE, AND DENSITY
models. The model takes the form7
li(t) 5 exp (b0 1 b1 Nit 1 b2 N 2it).
(2.3)
Finally, two models are run for each region which include both Ni and ni terms.
These models are designed to test our hypothesis concerning the impact of
regional over national density on founding rates. In the first formulation the model
is specified as
li(t) 5 exp (b0 1 b1 Nit 1 b2 N 2it 1 b3 nit 1 b4n 2it).
(2.4)
Then, in an attempt to follow exactly Hannan et al.’s specification, the following
model is also tested:
li(t) 5 exp (b0 1 b1 Nit 1 b2 nit 1 b3 n 2it).
(2.5)
According to the original formulation of density dependence, the first-order
density terms should have a positive effect and the second-order density terms
should have a negative effect. Using this approach it is possible to ascertain which
level of geographic density has the greatest impact on foundings. Specifically, it is
possible to test the hypothesis that legitimation is driven by both national and
regional density but that competition is driven primarily by regional density.
DATA AND METHODS
To address the question of whether regional density fits the founding model
better than national density, we use data collected by the research team headed by
G. Carroll and M. Hannan (see Hannan et al., 1995; Carroll et al., 1994). The data
base includes information on all firms which engaged in automobile production
activity in the United States since the inception of the industry until 1981.
In constructing the data base, the team primarily relied on three sources: The
New Encyclopedia of Motorcars (Georgano et al., 1982), The World Guide to
Automobile Manufacturers (Baldwin et al., 1987), and the three volumes of The
Standard Catalog of American Cars (Kimes and Clark, 1989; Flammang et al.,
1989; Gunnell et al., 1987). These sources provide comprehensive information on
automobile producers including descriptions of the models produced, start and
end dates of production, and historical accounts of organizational and production
processes. Rules for defining the population were based, in part, on the guidelines
set forth by the automotive historians who compiled these encyclopedias. To be
included in the population of U.S. auto manufacturers, firms had to be engaged in
the production of cars for sale; i.e., individuals who produced one-of-a-kind
7 It should be noted that this model differs from a conventional density-dependent model since N is
i
not total U.S. density, nor is the founding rate that of the entire population’s, but rather only the focal
region’s. For purposes of comparison with previous analyses of founding rates, we ran a conventional
model, i.e., a model in which the national founding rate is dependent on national density terms. We
found that the signs of the coefficients reflect the predicted pattern. (For a confirmation of these results
see Carroll et al., 1994.)
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
388
BIGELOW ET AL.
machines for their own use were not included. Further, the study eliminated firms
which exclusively produced trucks, commercial vehicles, and race cars.
The total number of producers included in the study is 2197, with a peak of 345
in 1910. While this may seem like a surprisingly large number of firms, it should
be emphasized that this data set has been constructed using archival resources and
data-base programs which were not available to earlier researchers (e.g., Epstein,
1928). Also, our definition of a firm includes many of the short-lived individual
producers who failed to qualify for inclusion in similar studies.8
Because the sources list information by model, not firm, we recoded the data
such that models were grouped together under the same firm, as needed. Also, we
retrieved and coded information on specific types of organizational foundings and
deaths so that complicated firm genealogies (which include mergers and acquisitions) could be accounted for and analyzed appropriately. Thus, once model data
were recorded and specific starting and ending events were coded, the data base
contained records on a firm basis, which could then be converted to annual spell
data. Based on Hannan et al.’s (1995) study, the time period of the study presented
here begins in 1885, shortly after the first world producer appears. Observation
stops at the end of 1981, the last year for which there is complete, reliable
information. Regional location information was coded using the state from each
firm’s address.9 States were assigned to one of five geographical regions: New
England, Mid-Atlantic, Midwest, West and South.10 If a firm moved from one
region to another, we adjusted the density count accordingly. A relocation by
itself, however, was not coded as either a founding or death.
We also collected data on relevant industry and socioeconomic covariates for
which a complete time series could be found. These include: total U.S. car
production, compiled from Ward’s Automotive Yearbook, Automotive News, and
World Motor Vehicle Data, U.S. resident population, CPI, and GNP from
Statistical Abstract of the United States, and three time-period dummy variables
based on the research of Altshuler et al. (1984). These periods of development in
the auto industry are the mass production period (Mass) which begins in 1902, the
product differentiation period (Diff) which begins in 1950, and the Japanese
technology period (JT) which begins in 1968.
Although the sourcebooks sometimes give exact dates for the beginnings of
spells of production, at other times only the year in which a founding occurred is
8 See Carroll (1997) for an extended discussion of the theoretical motivation behind these
differences.
9 Prior to WWI, multiplant firms were rare. For purposes of consistency, multiplant firms are
assigned one location based on the firm’s headquarters.
10 Regions are constructed as follows. New England: Maine, New Hampshire, Vermont, Massachusetts, Rhode Island, Connecticut. Mid-Atlantic: New York, New Jersey, Pennsylvania, Maryland,
Delaware, District of Columbia, Virginia. Midwest: Ohio, Michigan, Illinois, Indiana, West Virginia,
Kentucky, Wisconsin, Minnesota, Iowa, Missouri, Kansas, Nebraska. West: Washington, Oregon,
Wyoming, Idaho, Montana, North Dakota, South Dakota, Colorado, New Mexico, Arizona, California, Nevada, Utah, Hawaii, Alaska. South: North Carolina, South Carolina, Georgia, Florida,
Alabama, Mississippi, Louisiana, Texas, Oklahoma, Arkansas, Tennessee.
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
389
LEGITIMATION, SCALE, AND DENSITY
available. This lack of precise dating means that, consistent with most previous
studies of organizational founding, we analyze time-series data of annual event
counts. We estimate models with covariates log linearly related to the rate,
assuming a negative binomial model and using quasi-likelihood estimators11 (see
Barron, 1992). We lag all covariates one year to ensure exogeneity.12
FINDINGS
Table 1 presents the estimates of models restricting density to the focal region
(using the specification in Eq. (2.2)). As expected, the first-order regional density
term (here the natural log of ni) is positive and significant for all regions. In
contrast, the second-order effects are much weaker. Only New England exhibits a
negative and significant coefficient. The covariates, with the exception of the first
two time periods, show few consistent or significant effects. However, the mass
production period dummy is consistently negative and significant in the three
largest regions (New England, Mid-Atlantic, and Midwest).13 The differentiation
period dummy coefficients seem to have the opposite effect on foundings, given
the largely positive effect, although the West is an exception.
Table 2 shows model estimates when only the national density terms are
included in the analysis (as specified in Eq. (2.3)). Again, the first-order density
term is positive and significant across regions, as predicted. And again, the
second-order density term fails to produce the expected negative effect, except in
the Midwest region, which does have a small but significant and negative
coefficient. Thus, there is evidence of a ‘‘spillover’’ legitimation effect from other
regions, but there is no strong evidence of a competitive effect as manifested
through a depressed founding rate. The mass production coefficients remain
consistently negative and significant in three of the five regions and the differentiation coefficient remains largely positive.
Table 3 presents the strongest evidence in support of the hypothesis that
legitimation takes place at a broader level of geographic density while competition occurs at a more local level of analysis. (These models are run using the
specification in Eq. (2.4).) Both national density (ln Ni) and regional density (ni)
show consistently positive effects, and these are statistically significant in 7 of the
10 estimates. Overall, the pattern of estimates suggests that national and regional
legitimation work in concert.
11 We do not use estimation procedures that make assumptions for unobservable heterogeneity, as
does Lomi (1995), because these estimates typically lack robustness (see Trussell and Richards, 1985).
12 The negative binomial is preferred to the Poisson when there is ‘‘overdispersion’’—meaning that
the variance of the event count is larger than the mean. Note that the sigma reported in Tables 1–4
indicates the degree of correction for overdispersion. As sigma approaches zero, the degree of
overdispersion declines.
13 This is consistent with the notion that mass production is associated with an increase in firm size
to assure throughput (Chandler, 1990), which may be achieved at the population level by industry
consolidation. Thus a decrease in the founding rate would be expected. (Although mergers, which are
treated as births, would increase, this increase would be more than offset by the decline in foundings of
new firms and new entrants.)
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
390
BIGELOW ET AL.
TABLE 1
GQL Estimates of Regional Density on Regional Foundings of U.S. Automobile
Manufacturers 1885–1981
Independent
variables
Constant
ln (ni )
(ni )2
Car production
Total population
GNP
Mass
Diff
JT
CPI
Number of observations
(years)
x2
s
New
England
Mid-Atlantic
Midwest
West
South
23.088*
(1.466)
1.633*
(0.191)
20.0003*
(0.0001)
5.07e-09
(1.97e-07)
2.52e-05
(2.46e-05)
20.021*
(0.009)
20.699*
(0.222)
5.278*
(1.667)
4.627
(2.742)
0.005
(0.004)
20.819
(0.761)
1.096*
(0.094)
4.39e-06
(1.68e-05)
23.96e-09
(7.49e-08)
23.38e-06
(1.01e-05)
0.002
(0.002)
20.683*
(0.138)
0.475
(0.464)
20.589
(0.608)
20.001
(0.001)
1.784*
(0.765)
1.166*
(0.103)
7.35e-06
(5.59e-06)
24.13e-08
(7.54e-08)
24.74e-05*
(1.02e-05)
0.005*
(0.002)
20.500*
(0.248)
2.045*
(0.432)
20.546
(0.629)
0.0003
(0.001)
23.502*
(1.133)
1.603*
(0.229)
7.97e-05
(0.0009)
28.34e-09
(9.15e-08)
2.44e-05*
(1.29e-05)
20.002
(0.002)
20.483
(0.299)
21.261*
(0.467)
0.205
(0.572)
20.002
(0.002)
20.368
(1.207)
1.445*
(0.214)
20.001
(0.0008)
27.92e-08
(1.24e-07)
21.63e-05
(1.65e-05)
0.003
(0.003)
20.448
(0.434)
1.168
(0.827)
21.211
(0.713)
20.0005
(0.001)
97
46.21
0
97
87.60
0.001
97
117.74
0.049
97
65.14
0
97
73.18
0
Note. ni denotes regional density. Ni denotes total U.S. density minus ni , or total remaining density.
Figures in parentheses are standard errors.
* p , .05.
Also striking in Table 3, is the strong competitive effect of regional density,
reflected in the negative second-order regional density coefficients (n2i ). While the
national density (N 2i ) terms all have a small but depressing effect on foundings,
none of these is statistically significant. In contrast, the regionally based n2i terms
exhibit strong and statistically significant negative effects on foundings. The
exception to this pattern is, interestingly, the Midwest region which is where the
greatest density evolves and which, of course, emerges as the geographic center of
the auto industry.
Table 4 provides corroboration of the impact of regional density on competition. Here, following the approach of Hannan et al. (1995), the higher level
second-order density term is omitted from the analysis (as in Eq. (2.5)).
These estimates further support the hypothesis of legitimation operating on a
broader geographic scale than competition. The first-order effect of national
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
391
LEGITIMATION, SCALE, AND DENSITY
TABLE 2
GQL Estimates of Total Remaining U.S. Density on Regional Foundings of U.S. Automobile
Manufacturers 1885–1981
Independent
variables
Constant
ln (Ni )
(Ni )2
Car production
Total population
GNP
Mass
Diff
JT
CPI
Number of observations
(years)
x2
s
New
England
Mid-Atlantic
Midwest
West
South
0.933
(1.564)
1.139*
(0.126)
27.63e-06
(5.01e-06)
21.28e-07
(2.36e-07)
23.45e-05
(2.69e-05)
20.013
(0.009)
20.742*
(0.258)
5.470*
(1.629)
4.449
(2.681)
0.002
(0.003)
20.006
(0.848)
1.219*
(0.108)
26.02e-06
(5.00e-06)
2.01e-08
(8.62e-08)
22.90e-05
(1.11e-05)
0.004
(0.003)
20.955*
(0.256)
0.878
(0.491)
20.336
(0.654)
20.001
(0.001)
20.893
(0.906)
1.248*
(0.111)
22.41e-05*
(9.94e-06)
22.37e-07*
(7.71e-08)
21.63e-05
(1.12e-05)
0.005*
(0.002)
0.487*
(0.241)
0.537
(0.458)
20.734
(0.612)
20.003*
(0.001)
23.751*
(1.689)
0.688*
(0.233)
9.82e-06
(8.34e-06)
9.61e-08
(1.26e-07)
1.56e-05
(2.02e-05)
20.001
(0.004)
21.222
(0.685)
0.548
(0.663)
20.986
(0.713)
0.0009
(0.002)
27.510*
(2.515)
1.180*
(0.334)
6.86e-06
(7.70e-06)
28.92e-08
(1.36e-07)
2.81e-05
(2.13e-05)
0.002
(0.004)
20.961
(0.537)
20.628
(0.775)
21.102
(0.811)
20.0006
(0.002)
97
111.16
0
97
153.08
0.049
97
105.98
0.082
97
82.68
0.485
97
105.02
0.131
Note. Ni denotes total U.S. density minus ni , or total remaining U.S. density. Figures in parentheses
are standard errors.
* p , .05.
density remains positive and the second-order effects of regional density retain
the same signs as in Table 3.
Attempts to replicate exactly the Hannan et al. (1995) model by incorporating
only a first-order higher level density term were unsuccessful. That is, when the
models in Table 4 were run with only ln (Ni) and n2i , evidence of a strong
competitive effect disappeared and model fit diminished. These findings suggest
that regional levels of density played an important role in legitimating the
organizational form.
Given the late stage resurgence of density in four of the five regions, we also
ran models which control for industry age and age–density interaction effects.
These models are derived from Hannan and Carroll (1995) and Hannan (1997)
who extend the conventional density model to allow density to have time-varying
effects on legitimation and competition based on the age of the population. The
extended models add variables for population age and the interactions of popula-
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
392
BIGELOW ET AL.
TABLE 3
GQL Estimates of Regional and Total Remaining U.S. Density on Regional Foundings of U.S.
Automobile Manufacturers 1885–1981
Independent
variables
Constant
ln (Ni )
(Ni )2
(ni )
(ni )2
Car production
Total population
GNP
Mass
Diff
JT
CPI
Number of observations
(years)
x2
s
New
England
Mid-Atlantic
Midwest
West
South
20.727
(1.539)
0.566*
(0.207)
21.72e-06
(5.18e-06)
0.094*
(0.029)
20.001*
(0.0003)
29.17e-09
(2.28e-07)
25.65e-09
(2.75e-05)
20.017*
(0.008)
20.895*
(0.269)
4.121*
(1.523)
3.721
(2.506)
0.005
(0.004)
20.185
(0.721)
0.446*
(0.182)
26.05e-06
(4.03e-06)
0.050*
(0.013)
20.0002*
(7.88e-05)
7.08e-08
(8.19e-08)
27.18e-06
(1.06e-05)
25.58e-05
(0.002)
20.761*
(0.178)
0.497
(0.473)
20.187
(0.608)
20.0001
(0.001)
20.148
(0.782)
0.939*
(0.153)
26.70e-06
(8.37e-06)
0.005
(0.005)
5.51e-06
(1.76e-05)
26.83e-08
(7.78e-08)
21.69e-05
(9.75e-06)
0.002
(0.002)
20.067
(0.267)
0.731
(0.453)
20.602
(0.604)
20.0005
(0.001)
23.276*
(1.260)
0.050
(0.195)
26.34e-06
(6.34e-06)
0.405*
(0.057)
20.009*
(0.002)
21.16e-07
(9.64e-08)
2.40e-05
(1.29e-05)
20.003
(0.003)
20.048
(0.391)
21.156*
(0.476)
0.028
(0.661)
2.45e-05
(0.002)
22.432
(1.714)
0.441
(0.246)
1.05e-06
(6.18e-06)
0.245*
(0.068)
20.006*
(0.002)
24.56e-08
(1.30e-07)
22.80e-06
(1.86e-05)
0.003
(0.003)
20.731
(0.508)
0.629
(0.827)
21.109
(0.754)
20.001
(0.002)
97
115.95
0.009
97
110.08
0.028
97
69.45
0
97
66.28
0
97
92.46
0.004
Note. ni denotes regional density. Ni , denotes total U.S. density minus ni , or total remaining U.S.
density. Figures in parentheses are standard errors.
* p , .05.
tion age with density. These facilitate testing basic density dependence theory in
contexts where legitimation and competition express diminishing strength over
time.
Although estimates of these models provide some limited support for the
proposed extensions, they do not consistently improve on those models reported
in Table 4.14 Further, the lack of a consistent pattern across regions and across
14 We estimated a set of nested models designed to allow a hierarchical comparison. Beginning
with a baseline model (the model reported in Table 4), we added variables incrementally. For each
region we first estimated the baseline model (Model 4) then added the following variables one at a
time: region-specific industry age (year minus start year of industry in region), region-specific
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
393
LEGITIMATION, SCALE, AND DENSITY
TABLE 4
GQL Estimates of Constrained Effects of Total Remaining U.S. Density and Regional Density
on Regional Founding Rates of U.S. Automobile Manufacturers 1885–1981
Independent
variables
Constant
ln (Ni )
(ni )
(ni )2
Car production
Total population
GNP
Mass
Diff
JT
CPI
Number of observations
(years)
x2
s
New
England
Mid-Atlantic
Midwest
West
South
20.727
(1.547)
0.531*
(0.178)
0.097*
(0.027)
20.001*
(0.0003)
2.19e-08
(2.09e-07)
4.46e-07
(2.77e-05)
20.017*
(0.008)
20.944*
(0.225)
4.113*
(1.524)
3.751
(2.513)
0.005
(0.003)
20.056
(0.712)
0.386*
(0.179)
0.049*
(0.014)
20.0002*
(8.35e-05)
1.16e-07
(7.79e-08)
26.42e-06
(1.09e-05)
20.0007
(0.002)
20.919*
(0.178)
0.421
(0.481)
20.219
(0.616)
0.0006
(0.001)
20.087
(0.759)
0.859*
(0.113)
0.006
(0.005)
2.70e-06
(1.70e-05)
25.62e-08
(7.61e-08)
21.55e-05
(9.43e-06)
0.002
(0.002)
20.123
(0.254)
0.734
(0.450)
20.565
(0.599)
20.0003
(0.001)
23.225*
(1.196)
20.046
(0.161)
0.397*
(0.056)
20.009*
(0.002)
28.03e-08
(9.09e-08)
2.67e-05*
(1.25e-05)
20.004
(0.003)
20.235
(0.345)
21.171*
(0.474)
20.015
(0.660)
0.001
(0.002)
22.429
(1.733)
0.462*
(0.218)
0.245*
(0.068)
20.006*
(0.002)
25.42e-08
(1.21e-07)
23.68e-06
(1.79e-05)
0.003
(0.003)
20.705
(0.484)
0.663
(0.804)
21.113
(0.753)
20.001
(0.001)
97
112.35
0.017
97
110.32
0.027
97
68.98
0
97
66.82
0
97
92.36
0.005
Note. ni denotes regional density. Ni denotes total U.S. density minus ni , or total remaining U.S.
density. Figures in parentheses are standard errors.
* p , .05.
models makes it impossible to draw sound conclusions from these results.
Age-squared interactions with density are never significant and of the age
interactions, only age–density proved significant in two of the five regions—the
Mid-Atlantic and the Midwest. In these regions, the age-squared interaction with
density has no effect while the age–density interaction is negative and significant,
potentially indicating that density has a reduced effect over time. Further, in all of
age-squared, interactions of region-specific industry age with regional density, interactions of
region-specific industry age with regional density-squared, interactions of region-specific industry
age-squared with regional density, and finally, interactions of region-specific industry age-squared
with regional density-squared. We then compared the chi-squared values as well as the value and
significance of coefficients across models within regions. Given the great number of coefficients
estimated and the inconsistent findings (described in the text), we do not report the estimates in detail.
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
394
BIGELOW ET AL.
the models with age-squared–density interactions and in most of the models with
age–density-squared interactions, the main regional and national density effects
disappear.
We do find modest evidence that population age matters. Region-specific
industry age has a strong, positive, and significant effect on entry in New England
and the chi-squared value indicates an improvement in model fit. (Region-specific
industry age-squared, however, has no effect.) Region-specific industry agesquared has a small, positive and significant effect in the South and Mid-Atlantic.
(Region-specific industry age, however, has no effect.)
DISCUSSION
A wide variety of organizational scholars maintain interest in questions of
entry, boundaries, and industry emergence. Such questions include, for example,
to what extent new entrants respond to local conditions over national or international market conditions. A promising direction for investigating such questions
within the organizational ecology framework involves examining effects of
density at different geographical levels on population vital rates. Indeed, the
theory of density-dependent organizational evolution has recently been extended
in a geographically informed way that addresses questions of this kind (Hannan et
al., 1995).
The research presented here investigates and develops these issues. Following
Hannan et al. (1995), our main purpose here was to test the hypothesis that
density-dependent legitimation operates on a global scale, while densitydependent competition operates on a local level. The empirical findings support a
general interpretation of this argument. For the United States, ‘‘global’’ turns out
to mean national and local means regional; whereas in Europe the boundaries
were found to be continental and national, respectively. The difference between
these findings and those of Hannan et al. (1995) suggests to us that geography and
physical distance account for the different scale of effects of legitimation and
competition rather than nation–state political boundaries. That is, nation–state
boundaries do not seem to constrain competition and legitimation consistently in
the two studies—physical space does.15
We also find some limited evidence that these processes interact with industry
age, such that the effects of density diminish with age. However, only two of five
regions in the United States show evidence of this effect. We conclude that
regional boundaries, which carry no real political standing in the United States,
are less capable of capturing population-level inertial effects inherent in this
specification, than nation–state boundaries, which attach to meaningful institutions (See Hannan and Carroll, 1995).
The findings presented here also suggest an interaction between geography,
physical space, and technology worthy of further study. The association of three
15 It is possible that different constraints operate in the two settings; also note that the conclusion
about physical space does not imply that nation–state boundaries are irrelevant.
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
395
LEGITIMATION, SCALE, AND DENSITY
distinct technologies with three regions in the early period of the automobile
industry might be accidental or might arise from factors associated with these
regions. These factors could include differences in terrain, weather conditions,
extant transportation infrastructure, natural resource endowments (such as the
availability of water for steam or gasoline for gas-powered engines), and local
technical expertise.
Other researchers have noted similar positive effects of geographic proximity
on technological development and the innovation process (Dosi, 1988; Jaffe et al.,
1993; Feldman, 1994). Carlson and Jacobsson (1991) argue that the boundary of
new technology development is primarily regional. They suggest that during the
early period of technology development, i.e., before standardization, close collaboration between firms is required. Geographic proximity reduces the hazards of
technological uncertainty through collaboration and facilitates the transmission of
tacit knowledge. Once standardization occurs, the benefits of geographic proximity are unclear. Agglomeration benefits may be subsumed by increased competition or proximity may continue to impart a net technological advantage.
Our results indicate that competition processes operate more strongly through
regional than national density except in the case of the Midwest. This is an
unexpected finding. One possible explanation for it may rest on the interaction
effects of geographic proximity and technology on vital rates. It may be that in
populations where different standards or competing technologies exist, the region
which spawns the eventual dominant technology may be more sensitive to
multilevel density, due to its function as a technological source for the entire
population. Thus, competition effects are difficult to detect at the regional level for
this region. However, regions which are initially associated with alternative
technologies may remain more sensitive to locally driven competition due to their
late adoption of the ultimate ‘‘winning’’ technology. Local capital, labor, supplier,
and customer networks, whose limits are defined by physical space, would have to
be reconfigured to conform with the new technology, impeding foundings. Could
it be that slow adoption of a standard technology reduces agglomeration benefits
and prolongs local competition? How does the composition of technological
alternatives within a region effect these processes? Clearly more work is needed
in this area (see Bigelow and Seidel (1995) for further discussion).
More abstractly, multilevel models of density dependence show similarity to
recently developed models of path dependence in economic geography (Arthur,
1988, 1989; David, 1988).16 Assumptions common to both types of models
include (a) early conditions have enduring and pervasive effects, (b) initial growth
displays positive-feedback effects, and (c) nearly chance events can have potentially far-reaching effects on population vital rates (Hannan and Carroll, 1992;
Carroll and Harrison, 1994; Krugman, 1991; Arthur, 1990). Where the two
theories part company, not surprisingly, is on the issue of how to incorporate
16 Similarities between path dependence and density dependence have been noted elsewhere (e.g.,
Carroll and Harrison, 1994; Baron and Hannan, 1994).
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
396
BIGELOW ET AL.
sociological and economic forces in the respective models. Although densitydependent models downplay the role of firm characteristics, they do incorporate
firm-specific and industry-specific control variables when possible. In contrast,
economic models of path dependence offer little or no consideration of how
sociological forces might determine population boundaries or how legitimacy
effects population vital rates.
In density-dependent models, changes in density over time reflect the waxing
and waning of the opposing processes of legitimation and competition. In the
early stages of a population’s evolution, according to the theory, the observed
increase in density is a manifestation of the combined effects of strong legitimation processes and weak competitive processes. In the path-dependent economic
location models, the initial increase in density in a given region is purely a
function of weak or latent competition. The problem with ignoring legitimation,
and the underlying social, political, and geographical factors associated with it, is
that it makes it difficult to explain the advantages of density. Agglomeration
benefits, knowledge spillovers, externalities, and the like are difficult, if not
impossible, to use as justification for density benefits across populations. This is
precisely the issue which a density-dependent model can account for once it is
respecified at a lower or regional level of analysis.
REFERENCES
Altshuler, A., Anderson, M., Jones, D., Roos, D., and Womack, J. (1984). The Future of the
Automobile, MIT Press, Cambridge.
Arthur, W. B. (1988). ‘‘Urban systems and historical path dependence,’’ in Cities and Their Vital
Systems (J. H. Ausubel and R. Herman, Eds.), National Academy Press, Washington, DC.
Arthur, W. B. (1989). ‘‘Competing technologies, increasing returns, and lock-in by historical events,’’
Economic Journal 99, 116–131.
Baldwin, N., Georgano, G. N., Sedgwick, M., and Laban, B. (1987). The World Guide to Automobile
Manufacturers, Facts on File, New York.
Barnett, W. P., and Carroll, G. R. (1987). ‘‘Competition and mutualism among early telephone
companies,’’ Administrative Science Quarterly 32(3), 400–421.
Barnett, W. P., and Carroll, G. R. (1993). ‘‘How institutional constraints affected the organization of
early United States telephony,’’ Journal of Law, Economics and Organization 9(1), 98–126.
Baron, J. N., and Hannan, M. T. (1994). ‘‘The impact of economics on contemporary sociology,’’
Journal of Economic Literature 32(3), 1111–1146.
Barron, D. N. (1992). ‘‘The analysis of count data: Overdispersion and autocorrelation,’’ in Sociological Methodology (P. V. Marsden, Ed.), Basil Blackwell, Oxford.
Baum, J. A., and Singh, J. V. (1994). ‘‘Organizational niches and the dynamics of organizational
founding,’’ Organization Science 5(4), 483–501.
Bigelow, L. S., and Seidel, M.-D. L. (1995). ‘‘It’s not what you do, but where you do it: The link
between technology and geography in the U.S. automobile industry,’’ Paper presented at the
Strategic Management Society Annual Conference, Mexico City, Mexico.
Boas, C. W. (1961). ‘‘Locational patterns of American automobile assembly plants, 1895–1958,’’
Economic Geography 37, 218–230.
Burt, R. S. (1992). Structural Holes, Harvard Univ. Press, Cambridge, MA.
Carlson, B., and Jacobsson, S. (1991). ‘‘What makes the automation industry strategic,’’ Economics of
Innovation and New Technology 1, 93–118.
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
397
LEGITIMATION, SCALE, AND DENSITY
Carroll, G. R. (1997). ‘‘Long-term evolutionary change in organizational populations: Theory, models
and empirical findings from industrial demography,’’ Industrial and Corporate Change, in press.
Carroll, G. R., Bigelow, L., Seidel, M., and Tsai, L. (1996). ‘‘The fates of De Novo and De Alio
producers in the American automobile industry, 1885–1981,’’ Strategic Management Journal 17,
117–137.
Carroll, G. R., and Hannan, M. T. (1995). ‘‘Automobile manufacturers,’’ in Organizations in Industry:
Strategy, Structure and Selection (G. R. Carroll and Michael T. Hannan, Eds.), pp. 195–214,
Oxford Univ. Press, New York.
Carroll, G. R., Hannan, M. T., Bigelow, L., Seidel, M., Teo, A., and Tsai, L. (1994). Before Production:
Organizing Activities and Founding Events in the American Automobile Industry from 1886–
1982, Working Paper No. 94-14, Center for Research in Management, University of California,
Berkeley.
Carroll, G. R., and Harrison, J. R. (1994). ‘‘On the historical efficiency of competition between
organizational populations,’’ American Journal of Sociology 100, 720–49.
Carroll, G. R., and Wade, J. B. (1991). ‘‘Density dependence in the organizational evolution of the
American brewing industry across different levels of analysis,’’ Social Science Research, 20,
271–302.
Chandler, A. (1990). Scale and Scope, Harvard Univ. Press, Cambridge, MA.
David, P. A. (1988). Path-Dependence: Putting the Past into the Future of Economics, Institute for
Mathematical Studies in the Social Sciences Technical Report 533, Stanford University.
Dosi, G. (1988). ‘‘The nature of the innovation process,’’ in Technical Change and Economic Theory
(G. Dosi, R. Nelson, G. Silverberg, C. Freeman, and L. Soete, Eds.), Pinter, London.
Epstein, R. C. (1928). The Automobile Industry: Its Economic and Commercial Development, A. W.
Shaw Co., New York.
Feldman, M. P. (1994). The Geography of Innovation, Kluwer Academic, Dordrecht.
Flammang, J. M. (1989). Standard Catalog of American Cars 1976–1986 (Second ed.), Krause,
Iola, WI.
Flink, J. J. (1970). America Adopts the Automobile, 1895–1910, MIT Press, Cambridge, MA.
Flink, J. J. (1988). The Automobile Age, MIT Press, Cambridge, MA.
Georgano, G. N. (Ed.) (1982). The New Encyclopedia of Motorcars: 1885 to Present (Third ed.), E. P.
Dutton, New York.
Gunnell, J. A., Schrimpf, D., and Buttolph, K. (1987). Standard Catalog of American Cars 1946–1975
(Second ed.), Krause, Iola, WI.
Hannan, M. T. (1997). ‘‘Inertia, density, and the structure of organizational populations,’’ Organization
Studies, in press.
Hannan, M. T. (1986). A Model of Competitive and Institutional Processes in Organizational Ecology,
Technical Report 86-13, Department of Sociology, Cornell University.
Hannan, M. T., and Carroll, G. R. (1995). ‘‘Theory building and cheap talk about legitimation,’’
American Sociological Review 60, 539–544.
Hannan, M. T., and Carroll, G. R. (1992). Dynamics of Organizational Populations, Oxford Univ.
Press, New York.
Hannan, M. T., Carroll, G. R., Dundon, E. A., and Torres, J. C. (1995). ‘‘Organizational evolution in
multinational context: Entries of automobile manufacturers in Belgium, Britain, France, Germany, and Italy,’’ American Sociological Review 60, 509–528.
Hannan, M. T., and Freeman, J. (1989). Organizational Ecology, Harvard Univ. Press, Cambridge, MA.
Haveman, H. A., and Romanelli, E. (1994). Geographic Diffusion of Foundings in the U.S. Savings
and Loan Industry, unpublished working paper, Cornell University, Ithaca, NY.
Hounshell, D. A. (1984). From the American System to Mass Production, 1800–1932: The Development of Manufacturing Technology in the United States, Johns Hopkins Press, Baltimore.
Jaffe, A. B., Trajtenberg, M., and Henderson, R. (1993). ‘‘Geographic localization of knowledge
spillovers as evidenced by patent citations,’’ Quarterly Journal of Economics 108, 577–598.
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris
398
BIGELOW ET AL.
Kimes, B. R., and Clark, H. A. (1989). Standard Catalog of American Cars 1805–1942 (Second ed.),
Krause, Iola, WI.
Krugman, P. (1991). ‘‘Increasing returns and economic geography,’’ Journal of Political Economy 99,
483–499.
Lomi, A. (1995). ‘‘The population ecology of organizational founding: Location dependence and
unobserved heterogeneity,’’ Administrative Science Quarterly 40, 111–144.
Motor Vehicle Manufacturing Association. Various Years. World Motor Vehicle Data, Motor Vehicle
Manufacturing Assoc., Detroit.
Rae, J. B. (1959). American Automobile Manufacturers: The First Forty Years, Chilton Co.,
Philadelphia.
Rae, J. B. (1984). The American Automobile Industry, Twayne Publishers, Boston.
Rubenstein, J. M. (1992). The changing U.S. Auto Industry: A Geographical Analysis, Routledge,
Chapman & Hall, New York.
Scott, W. R. (1992). Organizations: Rational, Natural, and Open Systems (3rd ed.), Prentice-Hall,
Englewood Cliffs, NJ.
Scott, W. R., and Meyer, J. W. (1983). ‘‘The organization of societal sectors,’’ in Organizational
Environments (J. W. Meyer and W. R. Scott, Eds.), pp. 129–153, Sage, Beverley Hills, CA.
Swaminathan, A., and Wiedenmayer, G. (1991). ‘‘Does the pattern of density dependence in
organizational mortality rates vary across levels of analysis? Evidence from the German brewing
industry,’’ Social Science Research 20, 45–73.
Trussell, J., and Richards, T. (1985). ‘‘Correcting for unmeasured heterogeneity in hazard models
using the Heckman–Singer procedure,’’ Sociological Methodology, 242–276.
SSR591
@xyserv1/disk4/CLS_jrnlkz/GRP_ssrj/JOB_ssrj97ps/DIV_230z03
kris