Freeman NicheWidthDynamics 1983

Niche Width and the Dynamics of Organizational Populations
Author(s): John Freeman and Michael T. Hannan

Source: American Journal of Sociology , May, 1983, Vol. 88, No. 6 (May, 1983), pp. 1116-
1145
Published by: The University of Chicago Press
Stable URL: https://www.jstor.org/stable/2778966
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 [email protected].
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend
access to American Journal of Sociology
This content downloaded from

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
All use subject to https://about.jstor.org/terms
Niche Width and the Dynamics of
Organizational Populations'
John Freeman
University of California, Berkeley
Michael T. Hannan
Stanford University
This paper explores the effects of environmental variability and

grain on the niche width of organizational populations. It develops a
model of the manner in which environmental variations affect the
life chances of specialist and generalist organizations. This model
predicts that death rates of generalists exceed those of specialists in
fine-grained environments, regardless of the level of variability, but
that generalists have lower death rates when environmental variation
is both coarse grained and large. The model is applied to a sample of
restaurant organizations in 18 California cities. Maximum likelihood
estimates and tests confirm the major predictions of the model.
I. INTRODUCTION
Organizational ecology seeks to explain the dynamics of the relation of

organizations to the environment. In the broadest terms, population
ecologists ask, Why are there so many kinds of organization and why
are there no more? In seeking to answer this question, ecologists assume
that structural inertia limits the capacity of organizations to make radical
changes in strategy and structure. If inertial forces are strong, adaptation
by individual organizations cannot be the main force for change in organiza-
tional populations. In addition to facing strong inertial forces, most organiza-
tions face changing environments. Each time an environment changes, the
survival chances characterizing each population also change. The pattern
of structural variation at any time reflects the previous history of selection
1 This research was supported by National Science Foundation grants SES79-12315 and
SES81-09382. It was begun with funding from the School of Business Administration,
University of California, Berkeley. We wish to acknowledge the assistance of Michael
G. Fisher and Jack Brittain in the data-acquisition phase of the research and of Charles
E. Denk in the data analysis. Glenn Carroll and the participants in the Conference on
Formal Theories of Organizations at Werner-Reimers-Stiftung, Bad Homburg, made
helpful comments on an early draft. Nancy Brandon Tuma has made a major contribu-
tion to the methods used in this research and has also made helpful suggestions. Requests
for reprints should be sent to John Freeman, School of Business Administration, Uni-
versity of California, Berkeley, California 94720.
? 1983 by The University of Chicago. All rights reserved.

0002-9602/83/8806-0002$01.50
1116 AJS Volume 88 Number 6

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
(differential rates of birth and death). That is, the distributions of organiza-
tional characteristics reflect the underlying selection pressures. Therefore,
processes of change in biotic and organizational populations have important
similarities. Earlier (Hannan and Freeman 1977), we urged organizational
analysts to exploit these parallels and use the powerful and suggestive
models of population ecology.
Modern bioecology emphasizes evolutionary dynamics that underlie
patterns in nature. Organizational theory also focuses on structural arrange-
ments but lacks a satisfactory theory of change. Therefore, we emphasize
patterns of change in developing ecological theories of organizations. The
research reported here is part of that general effort.
In particular, our earlier work proposed that modern theories of niche
width, like Levins's (1968) fitness-set theory, be used to explain the distribu-
tion of specialist and generalist organizations across social environments.
We adapted Levins's model to suit organizational analysis and applied it
to populations of restaurants in 18 California cities. Our goal was to learn
whether our version of fitness-set theory describes selection processes in
these organizational populations.
In applying ecological theories to sociological ends, one important
strategic choice concerns the question whether to analyze equilibrium
distributions or the underlying dynamics. Modern ecological theory empha-
sizes the fact that environmental changes affect observed distributions
largely by altering birthrates and death rates of populations using different
adaptive strategies. Competition is invoked as a mechanism for linking
environmental change and changes in these vital rates. Despite this empha-
sis, much ecological analysis focuses on equilibrium distributions rather
than on rates. Such an emphasis may be appropriate when time scales are
long, competitive pressures are strong, and environments do not fluctuate.
Most interesting organizational populations face continually changing,
uncertain environments. Therefore, we doubt that equilibrium analysis is
appropriate for organizational ecology. Even if competitive pressures on
organizational populations are strong, equilibrium analysis of organization-
environment relations seems unlikely to be helpful. For this reason, we do
not follow Levins and other niche theorists exactly.
Theories that account for the full range of ecological phenomena, and
provide a comprehensive answer to the question of observed organizational
variability, presume rather detailed knowledge of the constituent processes
leading to evolution of organizational forms. Rates of birth and rates of
death must be studied separately if net mortality under various circum-
stances is to be explained. In this first effort we focus attention on mortality
rates.
A key element in our research strategy is an attempt to infer differences
in specialism/generalism from observed (or reported) organizational
1117

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
American Journal of Sociology
strategy and structure. Using data from interviews with key staff members,
we place each establishment in one of a series of types on a continuum
from specialism to generalism.
The important implication of Levins's model is that the advantages of
generalism depend in a complex way on the distribution of environmental
variations. We estimate models in which the relative death rates of special-
ists and generalists depend in a nonlinear way on overall variability and
the pattern or "grain" of variability. Estimates from this model permit
inferences about whether life expectancies of specialists and generalists
vary across environments in a manner that agrees with niche theory.
It is important to point out that our tests of hypotheses actually concern
three broad questions. Does our theory describe, at least approximately, the
processes governing organizational niche width? Do organizational niche-
width processes operate at least partly as a force of mortality? Are our
procedures for inferring differences in specialism/generalism of organiza-
tions from strategy and structure sound?
The paper is organized as follows. Section II reviews Levins's theory of
niche width, and Section III develops an extension appropriate for organiza-
tional analysis. Section IV explains why we test the model with data on
restaurant organizations, and Section V describes our research design.
Section VI tells how we measured organizational and environmental proper-
ties. Section VII describes how these measures are distributed in our sample.
Our objective is to estimate the effects of combinations of organizational
and environmental characteristics on death rates. Section VIII describes
how we used Tuma's (1980) RATE program to estimate such effects, and
Section IX reports our findings. The concluding section discusses implica-
tions of the findings.
II. NICHE-WIDTH THEORY
The concept of niche plays a central role in modern ecology, expressing the
ways in which the growth rates of populations depend on resources and on
the actions of other populations. Although many niche dimensions can be
studied, most empirical work concentrates on one dimension-niche width.
Niche width refers to a population's tolerance for changing levels of re-
sources, its ability to resist competitors, and its response to other factors
that inhibit growth. A population which has wide tolerance, meaning that
it can reproduce in diverse circumstances, is said to have a broad niche.
Populations with broad niches are commonly called generalists. Populations
with more limited ranges of tolerance are called specialists. Niche-width
theories are formulated to explain how environmental variations affect the
life chances of specialists and generalists.
Niche-width theories address a Jack-of-all-trades problem. They consider
1118

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
obvious trade-offs between tolerance of widely varying conditions and

capacity for high performance in any particular situation. Levins (1968),
whose lead we follow, proposed a principle of allocation: each population
has a constant sum of fitness that may be allocated across strategies for
playing the evolutionary game. Specialist populations follow the strategy
of betting all of their fitness chips on specific outcomes; generalists hedge
their bets.
We argued in earlier papers that the distinction between specialist and
generalist organizations involves issues of slack or excess capacity. General-
ists must hold some productive capacity in reserve. But specialists commit
most of their resources to a few tactics for dealing with the environment.
Thus specialist organizations will appear to be leaner than generalists, to
have less organizational fat. So in a sense we ask whether organizational
slack provides any evolutionary advantage.
Niche-width theories address both temporal and spatial variations in
environments. They typically focus on two features of temporal and spatial
distributions. The first is the level of environmental variability, the variance
of the series about its mean. The second is one characteristic of the pattern
of variation, grain. Grain refers to the degree of mixing of different types of
outcomes in the temporal or spatial distribution. Think of a spatial distribu-
tion composed of small squares of several types. When the squares are well
mixed, one will seldom encounter large patches of a single kind of square.
Such a distribution is said to be fine grained. If, instead, there are large
clumps of homogeneous squares, the distribution is said to be coarse grained.
When one considers change over time, grain refers to the length of typical
periodicities. When fluctuations occur frequently, so that tenure in any
one environmental condition is short, the environment is said to be fine
grained. Fluctuating environments in which tenure is long are coarse
grained. Except for the extreme case of complete stability, variability and
grain can vary independently.
In Levins's model, variability and grain predict optimal niche width,
given certain assumptions about tolerances. We have discussed sociological
interpretations of Levins's theory elsewhere (Hannan and Freeman 1977).
Here we review briefly its main elements.
Levins defines a fitness function that tells how well phenotypes bearing
each level of some trait fare in a particular environment. The fitness function
tells which level of y is optimal in that environment and how departures
from the optimal level lower fitness. Letting the trait under study be
denoted by y, the fitness function in environment 1 can be written as
w(y, E1). Levins assumes that this function is nonnegative, symmetric
around the optimal level, y*, and decreases to zero as y* - y increases.
To give an organizational example, y might represent degree of formalization
of procedures. In any given environment defined in terms of legal-political
1119

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
structures, social conditions, market characteristics, and technology, the

fitness function would tell which level of formalization was optimal and how
higher or lower levels would penalize an organization in its life chances or
chance of reproducing itself.
The degree of generalism manifested in a population is the rate at which
w[y, E(t)] declines. Consider figure 1. In it, a specialist with a peaked fitness
function is contrasted with a generalist, whose function is flatter.
Environments vary in ways that make alternative solutions to the
problem of survival more or less substitutable. Figure 2 plots illustrative
fitness functions for populations in two local environments, called "patches,"
in which it has fitness functions w(y, E1) and w(y, E2). The curves overlap
substantially in figure 2a, which means that the two patches pose similar
adaptive challenges for the population. Figure 2b illustrates a case where
the patches are dissimilar; the average curve has two peaks. High fitness in
E1 precludes high fitness in E2.
Levins represented fitness sets with curves like those in figure 3, whose
axes are fitnesses in the two kinds of patches. Each population with a distinct
strategy is a point in the fitness set. Selection chooses points on the boundary
of the fitness set since they are points of maximum fitness. Figure 3a repre-
Fitness
(w)
Level of Trait Y
FIG. 1.-Specialists and generalists
1120

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
11 1
>1o >
4-J
, ~~~~~~~~o
.H
> \ .~~~~0
*~~~~~~~~~ >
_ H 0
U)~~~~~~~
cn ~ ~ ~ ~ H L
Cd
4-J~~~~~~~~~~~~~~~~
c~ ~ ~ ~ ~ ~ ~ ~ n
C>> ? 5-~~~~~~~~~~~~~~-
X \ v~~~
H

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
* r~~~~~~~~~ I-4
V)
3
U)
40
\ Ics~~~ En
V \ 1 3~~~~~~~~~~~
. /~~~~~~~~~~~~~~~
_~~~~~~~~~~~~~~~~
W X~~~~~~~~~~~~~~~~

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
sents the fitness sets for the curves in figure 2a. This fitness set is convex
along the upper-right boundary. That is, all straight lines connecting
points in the set fall within the set. The case of dissimilar patches produces
a concave fitness set, as in figure 3b.
Levins introduced a graphical method for finding how optimal strategies
depend on patterns of environmental change. This method involves using
a so-called adaptive function, which tells how selection weights the fitnesses
in the various patches. Levins considered cases in which there are only two
kinds of patches which can be distributed with either fine grain or coarse
grain. In the case of fine grain, actors encounter the environment in small
patches. Since a given patch will be large for some forms and small for
others,2 patch size must be defined relatively. A patch is considered to be
small for some unit if the typical duration in the patch is short relative to
its life expectancy. Fine-grained environments are experienced as an average
of the various types of patches. For example, an environment fluctuating
rapidly between hot and cold may seem warm. Levins reasoned (see also
Roughgarden 1979, p. 269) that selection in fine-grained environments
weights fitnesses in the two patches additively. This reasoning implies that
overall fitness is a linear combination of the two fitnesses weighted by the
relative frequency of the patches: A(f) = Ow(Ei) + (1 - O)w(E2), where 0
is the probability that the environment is in patch type E1. When the
typical duration in a patch is long relative to life expectancy, the environ-
ment is not experienced as an average. Survival requires enduring long
spells of either kind of patch. Forms that are poorly suited to either kind
have a low probability of survival. This reasoning implies that selection
processes in coarse-grained environments respond to some multiplicative
function of fitnesses in the two patches. Levins chose the log-linear form,
ln A(c) = Ow(El) + (1 - O)w(E2). Roughgarden (1979, p. 269) gives a
population-genetic justification of this specification. In figures 4 and 5, the
two adaptive functions, A (f) and A (c), are represented by dashed lines.
Optimal populations in an environment are those points in the fitness set
that are tangent to the highest-valued adaptive function that intersects
the fitness set. A graphical analysis of the case of high variability (6- .5)
is shown in figure 4a. Variability is high in the sense that E1 and E2 are
equally likely at any time. This figure shows that generalism is favored in
all kinds of fluctuating environments when fitness sets are convex. Both the
linear and log-linear adaptive functions select populations with moderate
fitnesses in both kinds of patches, that is, generalists.
Suppose that typical fluctuations are large relative to tolerances, which
2 Judgments about the sizes of typical patches must consider the adaptive capacities and
life expectancies of forms. A short-term business cycle might be coarse grained for a small
business and fine grained for a large one.
1123

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
b0
*~~~~~~~~~~
~~~~~~~~~~~~- 4
rO ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~bi)
n W Q~~~~~~~~~~~~~~~~~~~~~~~~~~~~~E
40
/ cd
1124

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
. d
A .W~~~~~~~~~~~
/00
/d CNI
0.
\C) 12)
bD
1125~~~~~~~~~~~~~~~~-
1125~~~~~~~~~~~~~~~~~~~~~~~~~~1

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
means that fitness sets are concave. Fine-grained variations favor specialists,
but coarse-grained variations favor generalists, as is shown in figure 5.3
Finally, stable environments favor specialist organizations regardless of
grain and the shape of the fitness functions; that is, specialists will tend to
dominate all stable environments.
III. A MODEL FOR ENVIRONMENTAL EFFECTS ON ORGANIZATIONAL

DEATH RATES
Earlier (Hannan and Freeman 1977) we proposed that the theory reviewed
in Section II be adapted and applied to the problem of explaining niche
width in organizational populations. We pointed out that not all of the
predictions of niche-width theory agree with the conventional wisdom
about organizations. Organizations supposedly manage their dependencies
in turbulent environments by expanding into new geographical areas,
integrating vertically, or diversifying products. Each of these adaptations
makes an organization more generalized. Thus received theory implies
that variable environments favor generalists.4 In contrast, we predict that
only coarse-grained variations favor generalism.
We do not apply fitness-set theory directly. Instead, as we pointed out
in the Introduction, we propose a model for environmental effects on
organizational death rates which agrees qualitatively with fitness-set
theory. Our model is less general since it pertains only to death rates of
organizations. At the same time, it adds some substantive meat to the bare
bones of fitness-set theory, telling more specifically how environmental
variations affect organizational niche width.
This study concentrates on the effects of variations in demand for services.
Restaurant sales fluctuate with seasons, business cycles, and consumer
income cycles. They also respond to purely local events and to fashions and
fads. Are typical demand variations large relative to the adaptive capacities
3 Actually, Levins concluded that coarse-grained variations favor polymorphic popula-
tions-single populations that include subforms specialized to alternative patches. Some
kinds of organizations are close analogues to polymorphs. Conglomerate multinational
firms are a good example. The possibility of organizational polymorphism must be con-
sidered in any completely general analysis. However, we are dealing with simple organi-
zations for which polymorphism is unlikely. Either restaurant establishments are spe-
cialized to a particular environmental dimension or they are not. Therefore, for the
application at hand, we conclude that generalists are favored in all coarse-grained, vari-
able environments. If we studied restaurant companies rather than establishments, the
problem of polymorphism would be more serious. Some corporations operate fast-food
establishments, dinner houses, and a variety of specialized restaurants in addition to
their other businesses. Furthermore, many corporations in the restaurant business are
diversified in lines of business having nothing to do with restaurants. Selection processes
working on the restaurants are not easily observed at the corporate level.
4Representative citations for arguments of this sort are to be found in Ashby's (1952)
Law of Requisite Variety, Katz and Kahn (1978, p. 131), Lawrence and Lorsch (1967,
p. 8), Chandler (1962, pp. 42-49), Thompson (1967, pp. 34-37), and Pfeffer and Salancik
(1978).
1126

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
of restaurants? Most restaurants seem to operate close to the margin,

suggestirg that they are sensitive to even small fluctuations in demand.
This means that the fitness sets of restaurants are concave with respect to
typical variations in demand. The rest of our argument depends on this
assumption.5
Although Levins's model predicts only equilibrium distributions, we
want to introduce dynamic considerations. Instead of assuming that
equilibrium has been reached, we study the impact of various factors on
death rates of organizational forms with specialist and generalist charac-
teristics.
The rest of this paper develops implications of niche theory for the
effects of environments on death rates of specialist and generalist organiza-
tions. For simplicity, as well as to impose nonnegativity on rates, we
specify models in which environmental characteristics have linear effects on
the natural logarithm of the death rate.
The most important step in operationalizing niche theory is specifying
the precise forms of environmental effects. As we mentioned above, Levins
considered situations in which the environment fluctuates between only
two kinds of patches. We extend his treatment to the case in which environ-
ments fluctuate over continua. Let r indicate the death rate of an organiza-
tion at time t. We propose that
r = exp (a + OG + yGV + 6GS + VGVS + A), (1)

where G is a dummy variable that tells whether the organization is a general-
ist, V records the variability of its environment, and S is a dummy variable
that equals one if the environment has coarse grain. The remaining term,
A, summarizes the effects of other variables6 that affect death rates. This
model can be reduced to each case in figures 4 and 5 by setting some combi-
nation of parameters to zero, as we now show.
The simplest way to interpret the model in (1) is in terms of relative
death rates of generalists and specialists, A = r,/r,, where rg and r, are the
death rates for generalists and specialists, respectively. The baseline for all
comparisons is the death rate for specialists, obtained by setting G = 0 in
equation (1):
r8 = exp (a + A) . (2)
The death rate for generalists, obtained by setting G = 1 in (1), is
rg = exp (a + , + yV + 6S + DVS + A). (3)
'When fitness sets are convex, variable environments favor generalists regardless of
grain. Thus if restaurant organizations have convex fitness sets, grain will not play any
role in affecting death rates. Our hypothesis will fail.
6This model does not make any predictions about the main effects of variability and
grain. We do, however, include such effects in the empirical analysis reported below in
order to ensure that we do not confound main effects and interaction effects.
1127

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
The ratio (3) to (2) is the relative death rate,
A = exp (G3+yV+6S+. VS). (4)

Niche theory makes one global prediction and several specific ones. The
global prediction is that selection processes for organizations differ in fine-
grained and coarse-grained environments. Since we measure coarse grain by
seasonality, the prediction is,
hypothesis 1: 6 5z 0 and v - 0.
Our specific hypotheses are predictions about how selection processes

differ with grain. In the case of fine grain, niche theory implies that special-
ism is optimal at all levels of variability (see figs. 4a and 5a). Setting S = 0
in expression (4) gives A(f), the relative death rate in fine-grained environ-
ments, as
(f) = exp (@ + 'yV). (5)
Niche theory implies that A(f) exceeds unity over the range of variation in
V. This requires
hypothesis 2a: , > 0,

and
hypothesis 2b: ,B + -yV > 0 over the range of V.
Hypothesis 2b requires that either -y > 0, or, if -y < 0, I yV(max) I < 11,
since V is a variance and cannot be negative.
Hypothesis 2a is common to all well-known theories of organization-
environment relations. It is widely assumed that stable environments favor
specialist organizations. Hypothesis 2b is distinctive to the model that we
have proposed. The more conventional view holds that increasing vari-
ability increases the relative advantage of generalists, that -y is negative.
The case of coarse grain is more complex. Figure 4 implies that specialists
are favored when seasonality (coarse grain) combines with low variability,
but that generalists are favored when seasonality combines with high
variability. When S equals unity, the relative death rates of generalists and
specialists in coarse-grained environments, A(c), have the form
A(c) = exp [, + 6 + (-y + .)V] . (6)
As we stated above, niche theory implies that specialists are favored

when variability is low, but that generalists are favored when variability
is high. In terms of the model, this means that,
hypothesis 3: A(c) > 1 for V(min) but < 1 for V(max),

where V(min) and V(max) denote the minimum and maximum levels of
environmental variability. This hypothesis has two implications for the
parameters of the model:
hypothesis 3a: /3 + 6 > 0,
1128

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
and
hypothesis 3b: y + r < 0.
Most important, the values of the four parameters must imply that the
expression in brackets in (6) changes from positive to negative within the
observed range of V.
IV. WHY STUDY RESTAURANTS?
Although the restaurant industry is huge in terms of employment and sales,

individual restaurants are neither large nor powerful. Why study restaurants
rather than the large firms and government bureaus which have pre-
occupied organizational research? The choice of restaurants offers six
advantages for testing niche-width theories:
1. Most comparative research on organization-environment relations can-
not specify and measure well the environments of organizations. The
environments of most large firms have unclear spatial boundaries. They
include capital markets, product markets, and consumer markets which
span national economies and often cross national borders. Moreover,
national states have become an increasingly important feature of the
environment of many firms, particularly large ones. Such firms, interesting
though they may be, are not good candidates for simple ecological analysis.
It is just too difficult to specify variations in relevant environments. Restau-
rants, however, typically depend on local environments for capital, labor,
factors of production, and especially consumers. Such localized dependence
simplifies the task of measuring relevant environmental variations.
2. Since organizational death rates vary widely, it is necessary to study
large populations facing similar environments. Restaurants are small
enough that most cities have many of them. Hence it is feasible to estimate
the distributions of effects of environmental variations on restaurants.
3. All kinds of restaurants exist in diverse urban environments. Conse-
quently, there is a good chance that the effects of grain and variability can
be distinguished empirically.
4. Restaurants have diverse forms or strategies. For example, fast-food
forms, many types of ethnic restaurants, coffee shops, luxury dinner houses,
and natural food restaurants can be readily identified. Moreover, an
obvious spectrum of specialism/generalism underlies the variety of forms.
Some restaurants, such as sandwich shops and nightclubs, specialize in a
narrow price range; others span much of the range. Similarly, some restau-
rants, such as doughnut shops and luncheonettes, specialize in one part of
the day; others operate 24 hours a day. Finally, some restaurants specialize
in a narrowly defined cuisine; others provide a broad array of menu items.
Overall there appears to be great variability in terms of specialism-general-
ism.
1129

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
5. Restaurants supposedly have much shorter life expectancies than

most other kinds of firms. If death rates are sufficiently high, observation
over short periods can yield enough failures to permit meaningful survival
analysis. We chose to study restaurants partly for the reason that geneticists
study Drosophila. Both forms turn over rapidly enough that one is not
restricted to retrospective analysis.
6. Finally, it is easy to find reasonably complete listings of restaurants
in urban areas. Since restaurants depend on local consumer environments,
they usually purchase listings in the Yellow Pages. This publication provides
a convenient universe from which to sample restaurants.
All these reasons suggest that restaurants are a nearly ideal kind of
organization on which to conduct a prospective test of our niche-width
hypotheses.
V. RESEARCH DESIGN
We conducted prospective analysis because following establishments

forward in time increases the likelihood of recording events for short-lived
organizations. We wanted to avoid sampling only the most successful
members of a population, as is common in organizational research. In the
present context, picking only successful organizations is tantamount to
sampling on the dependent variable.-'Such sampling plans make it extremely
difficult to obtain unbiased estimates of structural relationships.
The unit of analysis here is the individual establishment. The goal of
our design was to maximize variation in grain and in variability among
establishments. However, we knew nothing about the environments of
particular establishments before sampling. Therefore, we used information
about variation in sales among cities in designing a plan for sampling
establishments. We analyzed time series of aggregate restaurant sales by
quarter for the 200 largest cities in California for 1974-77 and chose 18
cities that fell toward the extremes in the joint distribution of seasonality
and variability of aggregate sales. Sampling establishments within these
cities should maximize variation in grain and variability and minimize the
correlation between them.
In 1977 we sampled up to 50 establishments in each city. In cities with
more than 50 restaurants we took a random sample; in smaller cities we
selected the whole population. Establishments that had already closed or
that refused7 to participate were replaced with a randomly chosen alterna-
tive when the local population had not been exhausted.
We conducted telephone interviews with owners or managers to obtain
information about each restaurant's history and current operations. The
interviews were repeated in 1978 and 1979. In later interviews, we added
7 The refusal rate for the three years of study was 6%.
1130

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
establishments to replace those that had failed. In an effort to obtain more

variation in life-cycle characteristics, we oversampled newly listed restau-
rants in the second and third waves. In all, we sampled 985 establishments.8
The period of observation ranges from one to three years because restau-
rants were added in the second and third years of the study. This feature
of the design requires use of a method of analysis that adjusts properly for
differential length of exposure to the risk of an observed death. We discuss
this issue in Section VIII.
Though most establishments in the sample are independently owned
local establishments, 128 are branches of large chains. We suspect that the
contingencies of selection differ for local establishments and members of
chains. For one thing, chains operate in many local markets. Since we have
information only on local environments, we restrict attention to organiza-
tions that operate completely within the city. This restriction leaves us
with 810 observations. In 22 of these cases, data on one or more key vari-
ables were missing. Excluding these cases, the effective sample size is 788.
Of these, 104 (13%) died during the observation period and 99 (13%)
were sold.
VI. MEASUREMENT
Specialism and Generalism
An organization can specialize on many dimensions. For example, a restau-

rant can specialize in type of cuisine, style of service, hours of operation,
price range, diversity of menu items, and range of services (e.g., cocktail
service, entertainment, banquet facilities). Niche theory applies to special-
ism on any dimension that affects survival chances. Only detailed study of
a population of organizations can tell which dimensions are relevant.
Probably several dimensions need to be considered for most organizational
populations.
In the first test of the model, we use a single measure of specialism. This
measure reflects our 'understanding of the main strategic choices involved
in attempting to establish a restaurant. Some entrepreneurs attempt to
appeal to a narrow range of the population of potential customers.9 Others
8 This excludes establishments that appeared following the sale of an original sample
member at the same address. In an earlier report (Hannan and Freeman 1981), we ana-
lyzed the full set of 1,097 establishments. However, this number is somewhat deceptive
because we lack complete data on almost all of these new entrants. Therefore, we have
decided to concentrate here on the original sample members.
I Any resident, commuter, or visitor to a city is likely to purchase meals at several res-
taurants. Therefore, each person should be considered a multiple customer. Persons
presumably distribute their purchases across an array of establishments. Restaurants
should be considered specialized or not with respect to these patterns. Even if every
person eats once a year at a particular type of ethnic restaurant, the ethnic form may still
be quite specialized.
1131

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
try to appeal to the "average" consumer who occupies the middle of the
market range reflecting quality and price. After studying complete inter-
views and advertisements in the Yellow Pages, we classified establishments
into 33 forms. This coding relied heavily on menu specialization.10 It also
took account of hours of operation, number of entrees, and staff composi-
tion. We distinguished cafes, steakhouses, coffee shops, health food restau-
rants, taco stands, and Basque restaurants, to name only a few.
The measure used in this analysis collapses the 33 forms into three:
generalist, fast-food establishment, and other specialist. To be coded a
generalist, a restaurant had to meet three criteria: (1) offer a fairly general
menu, not limited to such items as pizza or hamburgers and not dominated by
a specific ethnic" cuisine; (2) employ at least one person in the differentiated
role of chef 12 or cook; and (3) offer seating in the restaurant. Of the restau-
rants in the sample, 31% are generalists. The fast-food form contains
establishments that specialize in one of the following: pizza, hamburgers,
hot dogs, fried chicken, tacos, doughnuts, or ice cream. This category
includes 34% of the establishments. The remaining 35%, the specialists,
tend to feature ethnic cuisine or have a very limited menu.
We assume that these broad forms respond differently to changing
demand, that they have different fitness functions. Demand for restaurant
services has many distinguishable components, such as the business lunch
trade, family dining, tourism, meetings, and banquets. We assume that
the various components do not change proportionately each time aggregate
sales rise or fall. Generalists are presumably less sensitive to fluctuations in
aggregate demand because they offer a greater range of services to more
diversified populations of consumers.
Environmental Characteristics
Since many respondents were reluctant to reveal detailed financial infor-

mation, we could not measure variability in sales for individual establish-
ments. It is not clear that it would have been appropriate to do so. Extreme
fluctuations in sales for a particular establishment may be confounded
10 Whenever restaurants advertised their specializations in the Yellow Pages, we used

this information in coding strategies.
11 One might argue that all of the restaurants in our sample specialize in terms of cuisine,
that the generalists specialize in American food. But it is important to remember that
specialism must be defined relative to the environment. A perfectly ordinary generalist
in Bangkok or Marrakesh is a specialist in San Francisco. Likewise, American restaurants
are considered quite exotic in most of the world.
12 Many restaurants do not have a staff member with the formal role of chef or cook.
This is true particularly of those that offer a very limited menu that can be mastered by
most employees, e.g., sandwiches. Investment of resources in employing chefs and cooks
is one of the fixed costs of generalism in the restaurant business. The value of having
such staff is that they can prepare a wide variety of menu items and can shift production
as tastes change.
1132

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
with death. Restaurants may manipulate their prices, hours of operation,

quality of food or service, and other features of their operations in ways
which influence sales temporarily as business failure threatens. So, for both
practical and theoretical reasons, we do not measure variability at the
establishment level. Instead, we give each establishment the variability
score for the city in which it is located. We used the time series of quarterly
gross sales by city mentioned above. Figure 6 displays some of the extreme
cases among the 18 cities.
The ecological literature discusses effects of both variability and unpre-
dictability of variations. Although both uncertainty and pure variability
conceivably affect niche width, the existing literature is not very clear on
this subject. For example, Levins did not draw a distinction between
these two components of environmental fluctuations. Both may be relevant
for many classes of organizations."3 In this paper we concentrate on only
8 - 1 - I - - t T i - I I l i ! t i i
Riverside
0 < ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~1
Palo Alto
4-4 -
.~Rodwood
(U city
2
- South Lake Tahoe
I I I I, I I I I , I l I l 1 ,
Time
FIG. 6.-Quarterly aggregate restaurant sales in thousands of constant dollars for se-
lected cities from fourth quarter of 1976 to first quarter of 1980.
13 We decided not to specify the model in terms of uncertainty because the certainty of
variations depends on an organization's information-processing capacity as well as on
the pattern of fluctuation. Introducing considerations of certainty will tend to shift the
focus toward decision making. We want to emphasize the role of environmental forces
rather than adaptation by individual organizations. Therefore, we focus on variability
of environments rather than the certainty of variations. Of course, we recognize that
obviously predictable patterns, such as the large increase in restaurant sales during the
1133

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
one of them-variability. We argue that variability constrains niche width,

whether or not it is predictable. Our measure of variability is the coefficient
of variation of each city's gross restaurant sales by quarter over the period
extending from the fourth quarter of 1976 to the first quarter of 1980:
Vk = ok/lXk, where Xk is the mean for the kth city of gross sales over th
period and 0ak is the standard deviation of the time series over the period
for the kth city.
The second environmental property is grain. It is not obvious a priori
whether a particular patch size is large or small for a class of organizations.
Should we analyze weekly variations in demand, quarterly variations,
yearly variations? Our understanding of the economics of the industry is
that restaurants can average over variations from one week to the next
but that several months of low sales create serious problems. This view
suggests that seasonal variations in sales ought to be considered coarse
grained. Therefore, for our purposes a restaurant has a coarse-grained en-
vironment when its sales variations have a strong seasonal component.
We have tried two approaches to measuring seasonality. The first is the
method used in selecting cities, the strength of the seasonal component in
the time series of aggregate sales for the city in which the establishment is
located. The second method uses a self-report from each establishment
about the seasonality of its sales. We obtain sharper results with the self-
reports.'4 This is not surprising, since there are likely to be important dif-
ferences among subenvironments in each city. For example, restaurants
located near highway interchanges probably face different seasonal pat-
terns than those in central business districts. Moreover, the calendar
quarters used in official statistics do not square well with the seasonal
patterns actually faced by restaurants in some cities.15 Therefore, we use
self-reported seasonality in the analysis reported below:
S 1 if sales are seasonal,

Sik l if sales are not seasonal.
Covariates
We also collected data on dates of founding, seating capacities, and owner-

ship-whether the establishment is independently owned, franchised, or
part of a chain. We recorded distributions of staff across job categories
Christmas season, may be anticipated by almost all organizations. Thus a realistic model
should include both variability and predictability. In the interest of simplicity, we do
not present such a model here.
14The patterns of estimates are similar with the two measures, but the standard errors
of estimates are considerably smaller with the self-report.
15 For example, the ski season in Lake Tahoe falls in two quarters, as does the spring
season in Palm Springs.
1134

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
and learned whether a restaurant offers a variety of special services, such

as bar service, takeout meals, and counter service. We also obtained the
average number of meals sold per week and price ranges. By multiplying
the midprice by number of meals per week, we calculated an estimate of
total sales volume for each establishment.
Each of these covariates lacks data for some observations. Sometimes
respondents refused to reveal some information or claimed that they did
not know-this was particularly problematic for sales data. Other times,
language problems interfered (even though some of our interviewers spoke
Spanish or Cantonese), or respondents ceased to cooperate part way through
the interview. Since the number of missing observations differs from vari-
able to variable, the sample size changes as we vary the specification of
the model.
As we indicate in Section IX, preliminary analysis revealed that only
age and the logarithm of sales (average meals times midprice) need to be
included in our models. When we use these two variables, the sample size
drops from 788 to 466.
VII. DESCRIPTIVE STATISTICS
Our analysis concentrates on the effects of three variables: generalism,

seasonality, and variability. As we mentioned above, 31%o of the establish-
ments in the sample are generalists. Seasonal sales patterns were reported
by 23%o. Variability ranges from 4.3 to 26.8 and has a mean of 15.2.
It turns out that generalism and seasonality are not correlated-the x2
statistic for independence in this fourfold classification for the 788 cases is
only .02. Generalism has a weak correlation with variability (r = .02), and
the correlation between variability and seasonality is .33.
Some of the analyses reported in Section IX control for the effects of age
and sales volume. Table 1 presents the means and standard deviations of
these variables by combinations of generalism and seasonality. Most dif-
ferences among the means in this table are small. Seasonal restaurants tend
to be slightly older than nonseasonal ones and to have slightly higher sales.
TABLE 1
MEANS AND STANDARD DEVIATIONS OF AGE AND LOG SALES FOR THE
SAMPLE OF 466 ESTABLISHMENTS WITH COMPLETE DATA
SPECIALISTS GENERALISTS
Nonseasonal Seasonal Nonseasonal Seasonal
Age. .... 10.7 (11.7) 14.0 (14.6) 10.7 (11.7) 10.8 (10.3)
Log sales .... 7.9 (1.2) 8.3 (1.7) 7.9 (1.2) 8.8 (1.5)
NOTE.-SDs in parentheses.
1135

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Specialists are older than generalists but have slightly lower average sales.
The fact that the differences are small is substantively important. It suggests
that we are not comparing generally successful organizations with generally
unsuccessful ones when we compare specialists with generalists (or seasonal
with nonseasonal establishments).
VIII. METHOD FOR ESTIMATING CAUSAL EFFECTS ON DEATH RATES
As we mentioned in Section V, our design provides observation periods of

varying lengths. We must use a methodology that adjusts for the length of
time that a restaurant is exposed to the risk of failing.'6 More important,
we must adjust for the fact that most observations are censored on the
right, that is, the study ends before most establishments go out of business.
Censoring Problems
We have only partial records-the data are censored on both left and right.
Figure 7 illustrates the three types of cases we must consider. The time
marked tb is the date on which three illustrative establishments were
founded. The times, Ti, T2, and T3 are the dates of interviews for the three
cases.-" The td's are times at which failures occur.
tb d (1) 1 d (2) 2 3
FIG. 7.-Illustration of three types of data records: tb denotes the date of birth, the
td's are dates of deaths, and the T, are dates of interviews.
16 It might seem more direct to calculate crude death rates for specialists and generalists
across combinations of environmental conditions. Unfortunately, crude death rates can
be very misleading when the lengths of observation periods vary and more than one kind
of event (such as a sale or a refusal to continue participation) can end an observation.
17 We use t to denote the timing of an event and T to denote the timing of observations.
1136

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
The case indicated by a solid line in figure 7 dies before the first interview.
Not only is the date of failure unknown, but we have no information on this
establishment. The case indicated by the dashed line dies between waves
of observations. We have recorded structural information on this case from
the first interview, but it has closed before the second interview. We cannot
determine the death date exactly. We know only that the date falls between
the two interview dates. The case indicated by the dotted line is right
censored. This establishment is still alive at the last interview, so we do not
observe its death date. We merely record that it had not yet failed by T3.
We do not know any straightforward procedure for dealing with left
censoring in this type of study. Since we have no information about the
establishments that died prior to observation, other than names and ad-
dresses, we cannot include them in analysis.
To minimize the damage caused by left censoring, we use age of the
establishment at ri as a covariate. This procedure allows us to examine the
pattern of age dependence in the selected portion of the sample. These
results may guide future work on the problem of left censoring.
In the case of right censoring, we note the date at which the establishment
is censored and use the information that it survived to that date in esti-
mating effects on death rates (see Tuma and Hannan [1978] for illustrations
of alternative approaches to dealing with right censoring).
The fact that we do not learn the exact timing of failures during the
study does not hamper attempts to estimate effects in a continuous-time
model, as we demonstrate below.
Maximum Likelihood Estimation
The death rate is not directly observable. It must be estimated from infor-
mation on lengths of lifetimes or distributions of times of deaths. The condi-
tional survivor function G(t I z, to) tells the probability that an organization
with characteristics summarized in the vector z and founded at time to
will not die before t. That is, if T denotes the time of failure,
G(tf z) = Pr [T > t - to z] .
This probability can be estimated directly from data on lifetimes. The

death rate is defined as
r(tf z) = -d lnG(tI z)/dt. (7)
It tells the probability of dying at (very near) time t, conditional on the

covariates, z. The differential equation in (7) can be solved, subject to the
initial condition that G(O) = 1, giving
G(t| z) = exp [-fO r(s)ds] .
1137

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
If the covariates are fixed, as we assume here,'8 the parameters of the death
rate function are closed-form functions of the survivor function.
In order to estimate effects on rates, one must specify how the rates
depend on exogenous variables. We assume that the death rate is constant
over time and depends log linearly on the exogenous variables: ln r = z'a.
Then
G(tl z) =-exp (z'a)t (8)
is the probability of not dying by time t. In other words, equation (8) gives
the probability under the model that an observation will be censored on
the right at time t. Suppose the second wave of interviews took place at r2.
The probability of dying between the first and second waves of interviews
equals 1 - G(r2J z), the probability of dying before T2, given that the
restaurant was still alive at the first interview. If the establishment closes
between the second and third waves, its likelihood is just G(r2 z)[1 -
G(r3 - T21 z)]. These simple rules and obvious extensions yield likelihoods
for each type of observation.
Because restaurants in a city compete for customers and staff, survival
of restaurants attempting to occupy the same niche is probably not com-
pletely independent. At the extreme, there is no independent variation in
rates within cities and the effective sample size is only 18. But this is not
realistic. Knowing the timing of the death of one establishment probably
does not help much in predicting the dates of events for others. We assume
as a first approximation that competition is diffuse, that all establishments
compete for common resources but do not face strong head-on competition.
This assumption implies that cases within a city are approximately pair-
wise independent. With this assumption, the likelihood for the entire sample
is the product of the likelihoods of all observations. We use Tuma's (1980)
RATE program to choose parameter estimates that maximize this likelihood
function.
We actually observe two kinds of transitions. Establishments may die (go
out of business) or be sold. Since we do not want to treat sales and deaths
alike, we analyze transitions among three states: alive, dead, sold. However,
we specify that the states dead and sold are absorbing states. We use
information on the timing of transitions to death to estimate death rates.
A sale is a censoring event with respect to this transition. Restaurants that
are sold during the study contribute information about death rates during
the time they are alive-the information that they have not yet died.
Tuma, Hannan, and Groeneveld (1979) give full details on estimation of
three-state and more general models using event-history data. These prin-
ciples carry over directly to the case considered here, for which exact
timing of events is unknown.
18 Since the observation times are short, it is not unreasonable to assume that the covari-
ates are fixed. In fact, there is some observable change in some of them. We ignore changes
in covariates in this paper but will address the issue in subsequent reports.
1138

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
IX. RESULTS
Death Rates
Our analysis of death rates used the model in equation (1). We began by
exploring the effects of various control variables, including measures of the
size of the establishment (number of employees, number of seats, number
of meals served per week), its age, the logarithm of its gross sales, and
characteristics of its local environment such as total restaurant sales, sales
per establishment, and measures of growth and decline in aggregate sales.
Only age at time of first observation and the log of gross sales had system-
atic effects on death rates. Therefore, we include age and sales volume in
some of the following analyses. However, because the maximum likelihood
estimator did not converge when both age and log size were included in the
model, we report estimates of models containing the two variables sepa-
rately. Since we lose many cases when these variables are included, we also
report estimates of models that exclude them.
We also studied differences between fast-food and other specialist forms
by including four terms in the model in equation (1): a main effect for the
fast-food form and the three interactions involving grain, variability, and
the dummy variable for the fast-food form. None of the four effects is
either large or significant at the 10% level. The likelihood ratio test of the
null hypothesis that all four fast-food effects are zero has a likelihood ratio
x2 statistic of only 2.61 with 4 df. As we had suspected initially, restaurants
specializing in fast food do not differ from other specialists regarding selec-
tion processes. Therefore, we do not include any terms for the distinction
between fast food and specialism in subsequent steps. In effect, we collapse
the three-category measure into a two-category distinction, specialist versus
generalist.
With these preliminaries, we arrive at our final specification:
rjk = exp (a + 3Gjk + YGjkkVk + 3GjkSjk + tGjkVkSJk + Ajk), (9)
where A contains the main'9 V and S and, in some models, the effects of
age or of the log of sales.20 The subscripts in equation (9) are a reminder
that seasonality, generalism, age, and sales vary among establishments
but that variability varies only among cities.
19 We have also introduced an interaction between V and S. We do not report the results
of models with this term because we have not always obtained convergent solutions.
However, in those cases in which the iterative maximum likelihood routine does converge,
we find that the VS interaction is small and not significantly different from zero. More
important, presence of a VS interaction does not change the qualitative patterns of
effects reported below (including levels of significance).
20 A referee wondered why we do not report estimates for a model estimated on the 466
cases but without either age or log size. The reason is that we are not intrinsically inter-
ested in the smaller sample. However, the pattern of estimates for such a model is very
similar to that reported for eqq. (3) and (4) in table 2.
1139

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Table 2 reports estimates of four models. Models 1 and 2 contain only

terms involving G, V, and S; therefore they can be estimated using 788
cases. Model 1 differs from model 2 in excluding the interactions involving
seasonality. Models 3 and 4 are similar to model 2 except that they contain
the effects of age and sales, respectively.
The first hypothesis we presented above is that selection processes differ
by grain. The likelihood ratio test of model 1 against the less constrained
alternative model 2 has a x2 of 10.9 with 2 df. The difference in fit between
models 1 and 2 is significant at the 1% level. Tests comparing models 3 and
4 with comparably constrained alternatives give similar results. Thus the
first hypothesis is supported-the effects of environmental instability on
organizational death rates differ according to the grain of typical variations.
We offered two specific hypotheses about fine-grained environments.
Hypothesis 2a is that ( > 0. Although /3 is positive in models 3 and 4, it is
essentially zero in model 2 (model 1 is not relevant given the test reported
in the previous paragraph). In none of the three cases does 3 differ signifi-
cantly from zero.2' Thus support for the hypothesis is somewhat mixed.
Point estimates tend to bear out the hypothesis, but standard errors are
too large to give much confidence in the findings.
Hypothesis 2b also receives mixed support-the point estimates from
models 2, 3, and 4 conform to the hypothesis but do not differ significantly
TABLE 2
MAXIMUM LIKELIHOOD ESTIMATES OF EFFECTS ON THE

LOG OF THE DEATH RATE
MODELS
INDEPENDENT PARAMETER IN
VARTABLE EQUATION (1) (2) (3) (4)
Constant ............. ... -8.65*** -8.63*** -9.13*** -7.788***

Variability (V) ........ ... .021 .017 .069* .091**
Seasonality (S) ........ . 135 .220 -.335 -.529
Generalism (G) ........ 1.29t -.001 .164 .546
G*V ................. y - .083t .012 .015 .004
G*S ................. a... 6.50tt 8.16tt 8.85tt
G*V*S ............... ... -.467tt -. 580tt - .647tt
Age (at first observation
in years) ......... ... ... -.031** ...
Log sales. ....... ... ... -.243***
N .... 788 788 466 466
x2 (VS. constant rate
model) .... 3.5 14.4 21.9 24.7
(4 df) (6 df) (7 df) (6 df)
*P < . two-tailed test. t P < .05 one-tailed test.

* P < .05 two-tailed test. tt P < .01 one-tailed test.
*** P < .01 two-tailed test.
21 We used one-tailed tests for the directional hypothesis stated in Sec. III. We report
two-tailed tests for other parameters since we did not make directional predictions.
1140

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
from zero. According to these estimates, specialists are favored somewhat

in fine-grained environments that fluctuate greatly, contrary to the con-
ventional view. As it is, we cannot reject the null hypothesis that y is zero.
Still, it is encouraging that the effect has the predicted sign.
There is no ambiguity about the success of the theory in the case of
coarse-grained (seasonal) environments. The relative death rates implied
by the estimates agree with the predictions of the theory. Consider, for
example, the results for model 4. The estimated relative death rate function
is A(c) = exp [.546 + 8.85 + (.004 - .647)V]. The expression in brackets
is positive for small values of V and negative for large values of V, as pre-
dicted. It changes sign when V equals 14.6, well within the observed range.
The estimates of models 2 and 3 imply similar patterns. Because seasonality
and variability are correlated, estimates of effects at the extremes are un-
likely to be precise. Nonetheless it is clear that the qualitative pattern
implied by the model holds strongly. Not only are the implied differences
in death rates of specialists and generalists large; in addition, both a and
v are significantly different from zero 5% level. Recall that these parameters
are associated with first-order and second-order interaction effects. Given
the high collinearity between V and S, it is surprising that we can reject
the null hypothesis that either of these parameters is zero.
Since previous organizational research has not reported effects on transi-
tion rates, it may be helpful to re-express the findings in another metric.
One way to do so uses the concept of half-life. The half-life implied by a
rate is the expected time by which half of the units at risk of the event will
have experienced it. The half-life of a firm is the expected age by which
half of a large population of homogeneous organizations exposed to similar
environments will fail. When rates are constant, the half-life is the value of
t that solves exp [-rit] = .5. That is, the half-life equals -ln (.5)/r.
Restaurant death rates appear to decline with age. However, the esti-
mated half-lives (assuming constant rates) are short relative to the strength
of the age dependence in the rate. In the interest of simplicity, we ignore
age dependence in calculating half-lives. Table 3 reports ratios of estimated
half-lives at birth for generalists and specialists, using the estimates of
model 4. The first row of table 3 shows the pattern in nonseasonal environ-
ments. The half-life of generalists is only half as long as that of specialists
TABLE 3
RATIO OF ESTIMATED HALF-LIVES

OF GENERALISTS TO SPECIALISTS:
IMPLIED BY MODEL 4 IN TABLE 2
V = lo V= 20
Nonseasonal ..55 .53

Seasonal .. . . .03 13.1
1141

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
at both levels of V. Increasing environmental instability does not appear

to favor generalists, as is claimed so often.
The second row of table 3 shows what happens in coarse-grained (sea-
sonal) environments. When seasonal environments have slight variability
(V = 10), the half-life of generalists is only a small fraction of the half-life
of specialists-there is strong selection pressure in favor of specialism in
this case. When seasonality combines with high variability (V = 20), the
situation is reversed. The half-life of generalists is an order of magnitude
longer than that of specialists. In this case there is strong selection pressure
in favor of generalists. The ratios in the second row of table 3 show the
reversal predicted by the theory.
Sale Rates
It is instructive to learn that the sale process differs from the death process.
There was no reason to suspect that grain and variability would have
systematic effects on the rate at which establishments are sold. After all,
establishments may be sold as often because of success as because of failure.
It turns out that none of the variables that affect death rates has any
effect on the sale rate. This is true even for age.22 This difference increases
our confidence that we have identified an important process. It is not
simply the case that restaurants turn over more rapidly in some cities and
that those cities happen to have certain combinations of grain and vari-
ability.
X. DISCUSSION
Our model makes several predictions about the effects of environmental

variables on the relative death rates of specialist and generalist organiza-
tions. The global prediction is that the patchiness of environmental variation
affects the selection process. Analysis of this problem in terms of seasonality
of demand confirms the prediction. Establishments that report seasonal
sales patterns are affected differently by variability in aggregate sales.
The model also predicts how the relative death rates will vary by grain.
The predictions are confirmed weakly for the case of fine grain. Signs of
estimated parameters agree with the predictions but do not differ signifi-
cantly from zero. However, as we noted in the previous section, received
theory fares even worse than fitness-set theory. The usual argument is that
any kind of environmental instability favors generalist strategies. Our
estimates imply that this is not so in fine-grained environments. In particu-
lar, our estimates imply that specialists are actually favored in high-
22 This difference illustrates the danger of combining the outcome states. For examp
if we combine sales and deaths into one analysis with end of establishment as the out-
come state, we find no apparent pattern of effects.
1142

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
variation, fine-grained environments. This finding agrees with the prediction

of niche-width theory.
The second case, involving coarse-grained environments, gives strong
support to the model. The implied relative death rates show the predicted
reversal over the range of V. The parameters associated with this pattern
differ significantly from zero in the predicted direction.
Differences in death rates determine variations in prevalences of forms
across environments unless birthrates compensate. If birthrates of special-
ists and generalists are roughly constant over environments, the pattern
implied by the death rates holds. But it seems likely that entrepreneurs
are aware of some of the effects we have uncovered. If so, birthrates will
tend to track the previous history of deaths. Such a birth process only
increases the speed at which the distribution converges to the pattern
implied by the death rates. It is difficult to imagine that entrepreneurs
would be so perverse as to found specialist firms at disproportionately
high rates in those environments in which their half-lives are much shorter
than those of generalists. But this is the only sort of birth process that
would undo the long-term consequences of the differences in death rates
that we observe.
We studied restaurants only for convenience. Our arguments about
environmental determinants of organizational niche width are general. In
particular, we think that they hold for all kinds of organizations. Only the
scope of the environment and the time scale change when one shifts from
small to large organizations (see Hannan and Freeman 1977, 1982).
This is the first test of the organizational niche-width hypotheses. We
plan to conduct two more shortly. We have collected event-history data on
populations of national labor unions and firms in the semiconductor in-
dustry. Both studies collect measures of aspects of niche width: the range
of occupations and industries that a union tries to organize, and the diver-
sity of products that a firm attempts to market. These studies also measure
temporal variations in key environments. Labor unions and high-technology
firms differ sufficiently from restaurants in structure to give a good reading
on the generality of the findings reported here.
It is important to extend the scope of niche-width models to include
other dimensions of the strategies used by organizational forms. One
promising direction for such work involves study of variations in what
ecologists call life-history strategies. Some organizational forms specialize
in the ability to respond to new opportunities and to the appearance of
ephemeral resources. Other forms depend on competitive ability, the
ability to enter densely occupied markets and compete well. In a biotic
context such strategies are called r-strategies and K-strategies, respectively.
Brittain and Freeman (1980) show important differences between such
strategies among firms in the semiconductor industry. They also trace the
1143

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
implications of such strategies for the internal structure of the firms and
the careers of their employees. Such arguments clarify the processes that
govern organizational niche width.
Our treatment does not elaborate on the systemic properties of environ-
ments. In this paper, we have focused on such sources of environmental
variation as seasonality. It is obvious that for many purposes, such patterns
are generated by the social, political, and economic system in which organi-
zations operate. Ecologically dominant organizations in an area create
patterned environmental variation for subdominant organizations. Uni-
versity towns show just such characteristics, and patterns of university
operations can be expected to have effects on the restaurant population
much like those attributed to seasonality in general here. More broadly,
though, such large-scale changes as modifications in government procure-
ment and accounting policies, changes in legislation and operations of the
courts all have an obvious impact on the selection processes of organiza-
tional populations. We do not preclude such views of the environment. In
fact, we hope this work leads in that direction. It is useful, however, to
start with simpler conceptions.
Clearly the research reported here is just a first step in developing an
organizational ecology theory with predictive power. But, since organiza-
tional ecology has contained such a high ratio of promises to evidence, it
is an important first step. It shows that at least some of the fundamental
processes identified in ecological theories can be tested in a straightforward
manner.
REFERENCES
Ashby, W. R. 1952. Design for a Brain. New York: Wiley.

Brittain, J., and J. Freeman. 1980. "Organizational Proliferation and Density-dependent
Selection." Pp. 291-338 in Organizational Life Cycles, edited by J. Kimberly and
R. Miles. San Francisco: Jossey-Bass.
Chandler, A. 1962. Strategy and Structure. Cambridge, Mass.: MIT Press.
Hannan, M. T., and J. Freeman. 1977. "The Population Ecology of Organizations."
American Journal of Sociology 82:929-64.
. 1981. "Niche Width and the Dynamics of Organizational Populations." Tech-
nical report no. 2, Organizational Studies Section, Institute for Mathematical Studies
in the Social Sciences. Stanford, Calif.: Stanford University.
. 1982. "Structural Inertia and Organizational Change." Technical report no. 4,
Organizational Studies Section, Institute for Mathematical Studies in the Social
Sciences. Stanford, Calif.: Stanford University.
Katz, D., and R. L. Kahn. 1978. The Social Psychology of Organizations. 2d ed. New
York: Wiley.
Lawrence, P. R., and J. W. Lorsch. 1967. Organization and Environment: Managing Dif-
ferentiation and Integration. Cambridge, Mass.: Harvard University Press.
Levins, R. 1968. Evolution in Changing Environments. Princeton, N.J.: Princeton Uni-
versity Press.
Pfeffer, J., and G. Salancik. 1978. The External Control of Organizations: A Resource
Dependence Perspective. New York: Harper & Row.
1144

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00
Niche Width
Roughgarden, J. 1979. The Theory of Population Genetics and Evolutionary Ecology: An

Introduction. New York: Macmillan.
Thompson, J. D. 1967. Organizations in Action. New York: McGraw-Hill.
Tuma, N. B. 1980. "Invoking RATE." Mimeographed. Menlo Park, Calif.: SRI Inter-
national.
Tuma, N. B., and M. T. Hannan. 1978. "The Censoring Problem in Event History
Analysis." Pp. 202-40 in Sociological Methodology 1979, edited by K. Schuessler. San
Francisco: Jossey-Bass.
Tuma, N. B., M. T. Hannan, and L. P. Groeneveld. 1979. "Dynamic Analysis of Event
Histories." American Journal of Sociology 84: 820-54.
1145

176.77.155.192 on Wed, 17 Apr 2024 22:20:56 +00:00

Freeman NicheWidthDynamics 1983

Uploaded by

Copyright:

Available Formats

Freeman NicheWidthDynamics 1983

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Freeman NicheWidthDynamics 1983

Uploaded by

Copyright:

Available Formats

Niche Width and the Dynamics of Organizational Populations

Author(s): John Freeman and Michael T. Hannan

Stable URL: https://www.jstor.org/stable/2778966

This content downloaded from

This paper explores the effects of environmental variability and

Organizational ecology seeks to explain the dynamics of the relation of

? 1983 by The University of Chicago. All rights reserved.

1116 AJS Volume 88 Number 6

This content downloaded from

This content downloaded from

II. NICHE-WIDTH THEORY

This content downloaded from

obvious trade-offs between tolerance of widely varying conditions and

This content downloaded from

structures, social conditions, market characteristics, and technology, the

FIG. 1.-Specialists and generalists

This content downloaded from

This content downloaded from

This content downloaded from

This content downloaded from

This content downloaded from

This content downloaded from

III. A MODEL FOR ENVIRONMENTAL EFFECTS ON ORGANIZATIONAL

This content downloaded from

of restaurants? Most restaurants seem to operate close to the margin,

r = exp (a + OG + yGV + 6GS + VGVS + A), (1)

The death rate for generalists, obtained by setting G = 1 in (1), is

rg = exp (a + , + yV + 6S + DVS + A). (3)

This content downloaded from

The ratio (3) to (2) is the relative death rate,

A = exp (G3+yV+6S+. VS). (4)

Our specific hypotheses are predictions about how selection processes

hypothesis 2a: , > 0,

A(c) = exp [, + 6 + (-y + .)V] . (6)

As we stated above, niche theory implies that specialists are favored

hypothesis 3: A(c) > 1 for V(min) but < 1 for V(max),

hypothesis 3a: /3 + 6 > 0,

This content downloaded from

IV. WHY STUDY RESTAURANTS?

Although the restaurant industry is huge in terms of employment and sales,

This content downloaded from

5. Restaurants supposedly have much shorter life expectancies than

We conducted prospective analysis because following establishments

This content downloaded from

establishments to replace those that had failed. In an effort to obtain more

Specialism and Generalism

An organization can specialize on many dimensions. For example, a restau-

This content downloaded from

Since many respondents were reluctant to reveal detailed financial infor-

10 Whenever restaurants advertised their specializations in the Yellow Pages, we used

This content downloaded from

with death. Restaurants may manipulate their prices, hours of operation,

This content downloaded from

one of them-variability. We argue that variability constrains niche width,

S 1 if sales are seasonal,

We also collected data on dates of founding, seating capacities, and owner-

This content downloaded from

and learned whether a restaurant offers a variety of special services, such

VII. DESCRIPTIVE STATISTICS

Constant ............. ... -8.65* -8.63* -9.13* -7.788*