Document View

We synthesize organization learning theory and organizational ecology to predict systematic patterns in the founding and growth of organizations over time. Our central argument is that competition triggers organizational learning, which in turn intensifies competition that again triggers an adaptive response. We model this self-exciting dynamic-sometimes referred to as the 'Red Queen' in general evolutionary theory-to explain organizational founding and growth rates among the thousands of retail banks that have operated in Illinois at any time from 1900-1993. We find strong evidence that Red Queen evolution led some organizations to grow quickly and to place strong competitive pressure on rivals. Red Queen evolution also helped establish barriers to entry. However, this same evolutionary process appears to make organizations more susceptible to 'competency traps', ultimately slowing their growth rates and inviting new market entry. Organizations confronted by a widely varying distribution of competitors grow more slowly and are more likely to face new entrants. Overall, the results suggest that processes of organizational creation and growth emerge from ecologies of learning organizations. More generally, we discuss the use of ecological theory and models to study the empirical consequences of organizational learning. [PUBLICATION ABSTRACT]

Copyright Oxford University Press(England) Apr 2002

1. Introduction

How do organizations develop over time? Some of the earliest work in organizational sociology addresses this question (Blau, 1955; Stinchcombe, 1965),and it remains the central focus of two vital research areas. The literature on organizational learning focuses on processes of organizational development, and on the adaptive and maladaptive consequences of these dynamics (March, 1988). Meanwhile, organizational ecology seeks to understand the dynamics of organizational founding, growth, change, and failure in a variety of contexts over long periods of time (Hannan and Freeman, 1989; Carroll and Hannan, 2000). Comparing these schools of thought, it is striking how distinct each has remained from the other. If one listed the most significant theoretical developments in organizational learning theory, rarely, if ever, would one find mention of ideas or findings from organizational ecology. Similarly, nearly every major empirical finding in organizational ecology can be explained without recourse to organizational learning theory-the exception being duration-dependent models, arguably the least 'ecological' findings in that literature. Due to this theoretical divide, our understanding of how organizations develop depends greatly on which of the two leading dynamic perspectives we choose: organizational learning theorists point to processes of internal organizational development, while ecologists point to interdependencies among organizations.

Although each of these theoretical approaches has made considerable progress on its own, we believe the combination of these two perspectives can provide new insights into organizational dynamics. We take a step in this direction here, beginning with two propositions that create a complementary link between organizational learning theory and organizational ecology: (i) competition among organizations triggers internal organizational learning processes; and (ii) learning increases the strength of competition generated by an organization. Taken together, these propositions suggest that competition and learning reinforce one another as organizations develop-a process known in general evolutionary theory as the 'Red Queen' (Van Valen, 1973).1 This paper seeks to derive and test predictions in an empirical model of Red Queen evolution.

In particular, our theory analyzes organizational evolution as a socially constrained process. Most current theories of organizational evolution depict, at least implicitly, a process of development in which systems tend to become more efficacious. At one extreme, this assumption lies at the heart of economic renditions of industrial evolution that presume the survival of efficient organizations (see Nelson, 1994).More generally, as Carroll and Harrison (1994) observe, most theories of organizations assume a process that brings about 'historically efficient' forms of organizations (March and Olsen, 1989). By contrast, our theory emphasizes the importance of social constraints, where each organization's fitness depends on its coevolutionary relationships with other organizations. This social context constrains organizational evolution both over time and across relationships. We argue that these social constraints, under certain conditions, lead organizations to make maladaptive changes in a process that often creates disequilibria.

Although we emphasize maladaptation and disequilibria, our theory nonetheless implies predictable patterns in organizational evolution. Specifically, our model can adjudicate the fundamental trade-off between the size and the number of organized social units (Hawley, 1950). This trade-off lies central to at least two streams of literature: researchers seeking to explain the rise of large, powerful organizations; and studies that note that while some organizations grow to be very large, others proliferate in numbers and then work together in organizational communities (Barnett and Carroll, 198 7; Lazerson, 1988). Our theory explicitly illuminates this trade-off, identifying the conditions under which formal organization expands through the growth of existing organizations versus through the proliferation of new organizations. Red Queen evolution, we predict, leads to periods where existing organizations become particularly well adapted, so that they grow rapidly and effectively prevent the founding of new organizations. This dynamic gives way, also predictably, to periods when organizations coevolve into maladaptive dead ends-suffering lower growth rates and inviting new challenges from entrepreneurial organizations.

Empirically, we test our model in an analysis of organizational founding and growth among the thousands of retail banks that have existed in Illinois at any time since 1900. Over this period, hundreds of local banking markets expanded, sometimes through the proliferation of new banks and in other cases through the growth of incumbent banks. In our view, these competing processes hinged predictably on Red Queen evolution. To motivate our empirical study, we first state our theory in more general terms.

2. A social ecology of organizational learning

We begin with the theory of organizational learning developed by March and his colleagues (March and Simon, 1958; Cyert and March, 1963; March, 1988, 1994). Organizational learning theory has been elaborated in various subtle and interesting ways, but we limit our attention to a few central ideas. We assume that members of organizations 'satisfice' in day-to-day decision situations, searching for alternatives when performance falls below some aspiration level and stopping this search once performance again meets this threshold. We also assume that such 'problemistic search' takes place sequentially and locally. Members of organizations begin by looking at close alternatives to improve performance. If these fail, they may consider moving to a more distant option. When participants fail to find a solution that restores performance to a satisfactory level, the learning model allows organizational members to adjust aspirations downward. Aspirations can also adjust upward when performance exceeds aspirations. In any case, the model describes an organization that adjusts either performance or aspirations until it reaches equilibrium (March, 1994).

What triggers this problemistic search to begin with? For several reasons, we think it worthwhile to analyze one particular trigger to organizational learning, competition from other organizations. First, organizations rarely exist in isolation. Typically they belong to a population of competing organizations. Second, competition is a major cause of poor performance among organizations-both direct competition from known rivals and more indirect, diffuse competition that generates scarcity even when rivals do not perceive the competition (Hannan and Freeman, 1989). Finally, by examining the link between competition and organizational learning theory, we open the door to the synthesis of learning theory and organizational ecology.

To illustrate this link, we examine the dyadic case, where one organization faces performance problems due to competition from another. According to the learning model, when these performance problems become sufficiently large, organizational members begin to search for alternatives that might restore performance to an acceptable level. Once performance reaches satisfactory levels again, searching stops. Seen within a single organization, the process ends with this improved performance and the learning model makes no additional predictions.

Consider, however, what takes place at the rival organization. The first organization's performance improvement probably comes, at least in part, at the expense of its rival. Thus, as performance improves at the first organization and its search ends, performance falls at the rival, triggering search within that organization. The rival organization now enters its own search process similar to what we saw in the first organization. This process also ultimately ends when performance returns to satisfactory levels. Once again, this improvement at the rival organization most likely comes at the expense of the first organization, again triggering the first organization's search processes and so forth. In this way, the full sequence of causal links in organizational learning takes place both within and among organizations. The complete process of organizational development involves an ecology of learning organizations, with each organizational solution sowing the seeds of a rival's next challenge (Mezias and Lant, 1994).

Although we present the case of learning among a pair of rivals for simplicity, the idea applies to larger numbers of coexisting rivals triggering each other's development along a coevolutionary path. In general, we expect two contrary effects to emerge from this process. On the one hand, organizations that have been exposed to more competition historically are likely to develop more and better capabilities. Under certain assumptions (which we make explicit and relax below),these capabilities improve the organization's fit to the environment.

On the other hand, such organizations also tend to face more capable rivals, since these rivals too have emerged from a history of competition. Consequently, the organization will find satisfactory performance more difficult to attain given that it must now compete against more capable rivals. These two opposing outcomes-better-adapted organizations and more potent rivals-emerge together from Red Queen evolution.2 On the other hand, organizations may respond to performance problems by adjusting their aspiration levels, rather than searching for better routines. How might adaptive aspiration levels affect Red Queen evolution? Consider two extreme cases, where in each case organizational members set their aspiration levels through social comparison to other organizations (March, 1994). At one extreme, competing organizations respond to rivalry only by lowering their aspiration levels. Under such extreme downward adjustment, adaptive aspirations prevent Red Queen evolution from developing. Instead, increased competition simply implies lower aspirations and the predictions of our empirical model will not hold. At the other extreme, all organizations set their aspiration levels according to the highest level of performance observed among rivals. In this case, most organizations will continuously search for performance enhancing changes. In this extreme upward-adjustment case, adaptive aspirations intensify Red Queen evolution and our predictions should be strongly supported.

We expect that reality falls somewhere between these two extremes. Some organizations may stall out of the Red Queen, adjusting aspirations lower. Other organizations, however, will maintain their aspiration levels or adjust them up to those of their more successful rivals. These organizations should experience Red Queen evolution, as we predict. Given this plausible intermediate case, we expect to see evidence of Red Queen evolution even with adaptive aspirations.

Although our theory depicts organizations as having strategic interactions, it does not require strong assumptions about organizational rationality. We emphasize this point because, for many, the idea of Red Queen evolution conjures an anthropomorphic image of organizations as individual actors involved in a race. This imagery might lead one to focus on the behavior of top managers responding to each other's strategic decisions. This framing has become commonplace in the field of evolutionary game theory, for instance, where hypothetical strategists play out ecologies of games (Axelrod, 1985).While these situations often occur, we think that Red Queen evolution applies to a much broader range of organizational behaviors. Organizational responses in this model include the many, individually small actions taken by organizational members responding to their own immediate concerns, who may well be ignorant of the larger strategic consequences of their actions. The clerk who alters a procedure in an effort to complete a job in time, the banker who seeks out new business by developing her social network, and the engineer who alters a product specification in response to performance problems-each of these individuals intends to improve performance within his or her limited domain. Our model allows such actions to have strategic importance, whether or not these individuals realize the larger strategic posture of their organization.

2.1 Social constraints on the Red Queen

Thus far we have described Red Queen evolution under the assumption that an organization can search for alternatives without constraint. This approach provides a baseline set of predictions that hold if evolutionary processes tend to create and sustain adaptive outcomes. But if social processes constrain evolution, then such a model will not likely provide an accurate picture of reality. In particular, we think that a theory of Red Queen evolution will have greater predictive value if it allows for two general types of constraints on learning: those that emerge with the passage of time, and those that operate across organizational relationships.

Constraints in time. Should we assume that lessons learned in the past continue to apply into the future without end? Ignoring the timing of experience produces an implausible model in which coevolutionary development that occurred just last year is interchangeable with coevolution that occurred a century ago. Moreover, the nature of coevolution may change over the history of competitive interaction. We relax this assumption in our model by allowing the consequences of coevolution to depend on when it occurs in time. Specifically, our model can detect both weak- and strong-form time constraints.

Weak-form constraints cause coevolutionary development to diminish, possibly to zero, with the passage of time. Weak-form time constraints appear, for example, in discussions of organizational 'memory'. As March (1994: 91) observes, organizational memory becomes less accurate and less complete with the passage of time. Organizations 'remember' rules, procedures, and practices, but not the experiences that generated these institutional characteristics. If knowledge of underlying rationales helps make organizational procedures more adaptive, then an organization may benefit less from experiences as time passes and the rationales underlying routines fade away.

Under a strong-form time constraint, distant-past experience turns maladaptive. In the 'competency trap' (Levitt and March, 1988),organizations fare especially badly by applying yesterday's well-learned solutions to today's very different problems. Seen at the organization level, the competency trap can account for individual cases where an especially well-developed organization fares particularly poorly as its environment changes. In light of the Red Queen, we expect to see competency traps develop for entire cohorts of competing organizations as their environment changes over time. In this situation, coevolving organizations have mutually reinforced each other's development in a direction that no longer fits the demands of the current environment.

We believe that a coevolved competency trap could be especially maladaptive. When an individual organization finds itself in a competency trap, it may look to other referent organizations to find more satisfactory solutions to its problems (Haveman, 1993; Greve, 1996). Furthermore, if selection processes eliminate relatively maladapted organizations, then ultimately organizations suffering from competency traps will fail. Through either process, organizational populations can 'correct' an individual organization that 'goes astray' due to a competency trap (Miner and Haunschild, 1995). By contrast, when entire cohorts of competing organizations move together into a competency trap, social comparisons reinforce maladaptive behavior. Moreover, selection processes are unlikely to eliminate these poorly adapted organizations because selection distinguishes among organizations according to their relative fitness-which remains constant in this case since Red Queen evolution affects entire cohorts of organizations. Consequently, precisely where social comparison processes reinforce the maladaptive outcomes of the Red Queen, selection processes have the least leverage to correct the problem. Notorious cases where entire organizational communities become unfit, such as the American automobile industry of the 1980s,ma y illustrate such a mutually reinforcing industry-level competency trap.

Constraints across competitive relationships. Our discussion of the Red Queen has considered coevolution among only a single cohort of rivals. Within a cohort of rivals, organizations share a common history of competitive moves and counter-moves occurring with the same sequence and timing. But most organizational populations comprise multiple cohorts that have entered at very different times in history. Important differences in organizational forms typically correspond to these cohorts, reflecting the variegated social conditions present at the time each began (Stinchcombe, 1965).How does Red Queen evolution take place in organizational populations that include different cohorts, each experiencing a unique sequence and timing of competitive events?

To answer this question, it helps to think of each cohort as moving along its own Red Queen, so that multiple Red Queens develop when multiple cohorts coexist. For instance, if all organizations out of a population of 100 began at the same time, then all 100 will have shared a similar coevolutionary history. By contrast, if these 100 organizations arrived in ten different cohorts, then a more complex coevolution would take place. Early arrivals would coevolve for a few years, and then would face an entirely new set of competitors. Again and again new cohorts of different rivals would appear. These new organizations will not likely enter using the same practices and strategies to which the industry incumbents have coevolved. Rather, each new cohort will bring very different competitive threats into the population-the 'waves of creative destruction' described by Schumpeter (1934).

If we disregard this multiple cohort problem in our model, we implicitly assume: (i) that search costs do not increase when an organization engages in multiple coevolutionary processes; and (ii) that each coevolutionary process is independent of all others. Under these assumptions, an organization can freely develop along with every new cohort of rivals that it confronts-unaffected by the fact that it may be facing many other cohorts of rivals as well. In this scenario, our organization facing 100 organizations in a single cohort differs little from one encountering its 100 rivals in many separate historical waves.

To relax these assumptions, first consider the costs that an organization must bear to engage in coevolution. When an organization faces all of its rivals within the same cohort, it effectively plays a single coevolutionary game and so spreads the costs of playing this game across all rivals (Slatkin and Maynard Smith,1979). At the other extreme, an organization that faces each of its rivals in a separate cohort effectively plays as many games as it has rivals, incurring considerably greater costs to adaptation than it would if all rivals came from the same cohort. Consequently, an organization that coevolves with many different cohorts of rivals incurs substantially greater search costs than an organization that coevolves with all its rivals concentrated into one cohort. We expect, then, organizations facing more varied cohorts of rivals to be less fit and weaker competitors as a result of the Red Queen.

Second, we relax the assumption that each coevolutionary game is independent. Relaxing this assumption, adaptations made to deal with one cohort of rivals may constrain what an organization can do in response to another cohort of rivals (Sorenson, 2001). For instance, a retail bank might respond to a customer-service improvement by a rival, but this may in turn constrain what it can do to respond to a different rival pursuing a low-cost, low-service policy. More generally, as Padgett and Ansell (1993) observed, a 'robust' strategy that can deal with multiple, conflicting demands in an ecology of games precludes many courses of action that would be well-suited to any one game. In the context of our model, this idea suggests that the more varied the cohorts faced by an organization, the more constraint it faces in its various coevolutionary struggles. Again, this leads us to predict that organizations facing more varied cohorts of rivals will be less fit and generate weaker competition.

At first glance, this prediction seems to stand in sharp contrast to the notion that variety gives organizations an evolutionary advantage (Miner and Haunschild,19 95). The advantage of variety, however, stems from having a wide range of routines on which the organization can draw to meet environmental challenges (Sorenson, 2000a). Variety among an organization's rivals, by contrast, increases the complexity of the environment and makes the organization's task of fitting the environment more difficult. Although in the long run such an experience may confer some advantage, this possibility does not make the task of taking on a wide variety of rivals any easier. Thus we predict that the difficulties involved in competing simultaneously with a wide variety of competitors should reduce an organization's fitness and competitiveness.

2.2 Selection processes and Red Queen evolution

Though we have focused on the interplay of organizational learning and competition, some aspects of Red Queen evolution could develop, instead, through the interactions of natural selection and competition-even without organizational learning. For example, consider the extreme case where no organizational learning takes place. A simple formulation of this case requires several assumptions. (i) Holding environmental conditions constant, assume that fitness varies across organizations, but not within any individual organization over its lifetime; (ii) Assume that competition increases failure rates, selecting organizations on the basis of their fitness. Under these conditions, those organizations that survive exposure to competition will be more fit than those organizations that have not faced much competition (Sorenson, 2000b). This result emerges not from learning, but rather from greater selectivity in survivorship among organizations that have faced more competition. Thus, selection, even without learning, could produce a pattern consistent with our theory.

Selection processes also could generate a competence trap in Red Queen evolution. Relaxing the assumption that environmental conditions remain constant, organizations that survived competition during an earlier era may fit particularly poorly with the new conditions. Under this scenario, organizations find themselves in a competence trap, but they did not develop their competencies in response to competition. Rather, survivors from the earlier era outlasted rivals on the basis of (now outdated) competencies that they had from the start; they then found themselves out-of-step once the environment changed. Thus the competence-trap aspect of the Red Queen could also result from a purely selection-driven process.

A Red Queen theory stated purely in terms of selection processes is especially plausible; such a theory requires only that organizations differ initially with respect to fitness, and that competition enhances selection on the basis of that fitness. Nonetheless, we see merit in a learning-based theory of Red Queen evolution as well. The learning-based theory features a self-exciting causal process. If organizations learn from exposure to competition, this learning intensifies competition-rather than simply sorting firms on fixed, initial differences as the pure-selection theory predicts. Furthermore, this intensification accumulates in a learning-based model: increased fitness levels and competition further stimulate learning, there by strengthening competition, and so on. Because of this auto-catalyzing process, even small initial differences can importantly influence the evolutionary trajectories of industries. By contrast, pure selection processes can only operate on the variety available in the population; thus, the entrepreneurial process and its ability to generate better firms constrain the potential fitness and competitiveness of the population. Consequently, though selection processes probably contribute to the process, we think that organizational learning likely plays a critical role in driving Red Queen evolution.

Because both selection and learning probably play a role in Red Queen evolution, our approach seeks to empirically distinguish between the results of selection-based and learning-based evolution. The empirical tests in this study feature models of organizational growth that allow us to isolate within-organization variation over time, as opposed to the between-organization variation on which a pure-selection mechanism operates. This empirical distinction could not be made in an earlier investigation that modeled the implications of the Red Queen for organizational failure rates (Barnett and Hansen, 1996).

2.3 Density delay versus Red Queen evolution

The logic of Red Queen evolution also appears to contradict the logic of 'density delay'-a theory proposed by Carroll and Hannan (1989) and supported in various empirical studies (Carroll and Hannan, 2000). The theory of density delay focuses on the number of organizations in a population at the time of an organization's founding. Organizations founded at times of high density are predicted to suffer an enduring frailty as a result of resource scarcity at the time of founding. By contrast, organizations founded during more munificent, low-density conditions are predicted to enjoy an ongoing survival advantage. Thus the logic of density delay suggests that competition imposes lasting negative effects on organizational life chances, while the logic of Red Queen evolution implies the opposite: that competition enhances organizational survivability.

Can we resolve these apparently contradictory arguments? Swaminathan (1996) offered one potential resolution by proposing that the life-threatening consequences of initial scarcity fade in time (rather than being permanent as argued in the original theory). Furthermore, Swaminathan argued that as time passes, organizations that survived the 'trial by fire' of initial scarcity would prove more likely to survive, consistent with the idea that competitive conditions select for fit organizations. We agree with the 'trial by fire' argument, but see no reason to limit it to being triggered by competition during the year of an organization's birth. Our elaboration of Red Queen evolution, in essence, extends Swaminathan's reasoning, all owing competition throughout an organization's lifetime to enhance fitness (and allowing for learning as well as selection).

In the spirit of Swaminathan's approach, we reconcile our theory with density delay by allowing for two distinct processes, each specified separately in our empirical models. In one set of terms, we allow exposure to competition in general (at any time) to have initially adaptive but eventually maladaptive consequences, as predicted by our theory of Red Queen evolution. In another term, we then treat competition at founding as an exceptional circumstance with its own distinct, on going effects consistent with the theory of density delay. Taking this approach, we model the ongoing learning and selection of Red Queen evolution, while granting that competition at birth may present a special case. This seems reasonable, given that the quality of an organization's initial structures, processes, strategy, and people might vary as a function of initial resource scarcity-but that continued exposure to competition likely triggers natural selection and organizational learning. In sum, we allow for both density delay and Red Queen evolution, with the relative strength of each depending on when competition has the most severe effect on an organization's life.

2.4 Modeling the experience distribution

To test our ideas empirically, we first build a model of organizational viability, depicting how organizational life chances develop as organizations coevolve. We then model organizational viability in terms of observed rates of organizational growth and founding. Most existing models of organizational development depict a process that takes place as organizations become older, larger, or more structured (e.g. Barron et al., 1994; Ingram and Baum,1 997). By contrast, R ed Queen evolution features not individual organizational properties, such as age or size, but rather relationships among organizations. Consequently, we need a model that includes measures of each organization's competitive relationships to study Red Queen evolution empirically.

We find it useful to describe each organization's competitive relationships according to its experience distribution. We measure for each organization j a set of clocks, [tau]jk, each recording the duration of j's competitive relationship with one of its N rivals k. Each organization j's experience distribution can then be described by the moments of the distribution-such as the mean and variance-of its [tau]^sub jk^s. For example, we measure [mu]^sub j^, the mean duration of j's competitive relationships, and [sigma]^sup 2^^sub j^ = [[summation operator]^sub k^ ([tau]^sub jk^ - [mu]^sub j^)^sup 2^]/N, the variance of the distribution (as defined, N does not include the focal organization j). To allow the effects of coevolution to vary depending on historical recency, we calculate for each organization separate means [mu]^sub Rj^ and [mu]^sub Dj^, representing organization j's mean competitive experience in the recent and distant past, respectively.3

These measures of the experience distribution distinguish among organizations according to their competitive histories. A higher mean indicates that an organization has had long competitive relationships on average. Meanwhile, a high variance in the experience distribution implies that an organization faces competition from a wide variety of different cohorts of rivals. These measures differ from organization to organization, depending on where and when it competes, but they also differ over the life of an organization. For instance, an organization might have considerable experience concentrated into a single cohort of rivals, in which case it would have a high mean and zero variance in its experience distribution. If a new cohort of rivals appears, the variance of the organization's experience distribution would then increase considerably while the mean of the experience distribution would fall.

To model the effects of competitive experience on organizational life chances, we depict the relative fitness v of each organization j as a proportionate function of the mean and variance of its competitive experience: v^sub j^ = (a^sub R^[mu]^sub Rj^ + a^sub D^[mu]^sub Dj^ + b[sigma]^sup 2^ ^sub j^), where a^sub R^, a^sub D^, and b represent coefficients to be estimated. These terms demonstrate the effects of organization j's competitive experience on its fitness. As implied by our theory, we also allow the strength of competition from j's rivals, k, to hinge on the rival's competitive experience. Specifically, we assume that the competition generated against j by a given rival, k, follows as a linear function of its competitive experience: [alpha]+ c^sub R^[mu]^sub Rk^ + c^sub D^[mu]^sub Dk^, where [mu]^sub Rk^ and [mu]^sub Dk^ denote k's recent and distant competitive experience, respectively. Summing over all N of j's rivals, k, the total competition faced by organization j is [alpha]N + c^sub R^[summation operator]^sub k^[mu]^sub Rk^ + c^sub D^[summation operator]^sub k^[mu]^sub Dk^. Using the approach of Barnett (1997),we then allow the competition faced by organization j to affect its viability, resulting in the complete model:

v^sub j^ = a^sub R^[mu]^sub Rj^ + a^sub D^[mu]^sub Dj^ + b[sigma]^sup 2^ ^sub j^ + [alpha]N + c^sub R^[summation operator]^sub k^[mu]^sub Rk^ + c^sub D^[summation operator]^sub k^[mu]^sub Dk^

The consequences of Red Queen evolution appear distinctly in this model, enabling us to restate our predictions operationally. If a^sub R^ > 0 and a^sub D^ < 0,then recent experience improves j's viability while distant-past experience makes j less viable, consistent with Red Queen development constrained by a competency trap. On the flip side, we expect j's rivals, k, to be more potent due to recent competition and less potent due to distant past competition, which suggests c^sub R^ < 0 and c^sub D^ > 0. Finally, if competing with various cohorts of rivals harms j's viability, as we expect, then b < 0.

Our model of Red Queen evolution generalizes the density-dependent model typically used in organizational ecology (Hannan and Carroll, 1992). Density measures one aspect-the simplest aspect-of the experience distribution: It represents the number of competitive relationships involving a specific organization. With density controlled, our model also allows an organization's viability to depend on the durations of these competitive relationships-both in their mean and variance. Furthermore, we allow a rival's potency to depend on its competitive experience. In this way, the density-dependent model provides a useful baseline, depicting the null hypothesis that competition does not develop over time as in Red Queen evolution.

Note that this model overcomes the problem of Red Queen evolution being self-concealing. It separates the two countervailing effects of the Red Queen: organizations can become more viable, but their rivals can become more potent. If one looked for evidence of a Red Queen simply by examining descriptive data, such as the relationship between growth rates and exposure to competition, these contradictory effects might suppress one another-concealed by the fact that the most developed organizations also face the most developed competitors.

3. Organizational founding, growth, and the Red Queen

We use our model to predict rates of founding and growth among retail banks that operated in Illinois at any time from 1900 to 1993. An earlier analysis of failure rates among these organizations (Barnett and Hansen, 1996) found support for the Red Queen model. Here, we extend the analysis to organizational founding and growth rates-studying these two processes together because the sizes and numbers of organizations result from two competing processes. Often, establish ed organizations grow, sometimes becoming extremely large. Nevertheless, the founding of new organizations is also common. When the growth of established organizations outpaces the creation of new ones, organized activity concentrates among a small number of incumbents. By contrast, when new organizations arrive faster than existing organizations grow, many individually small organizations transform industries and markets (Schumpeter, 1934).

How can we predict whether organizational expansion takes place through the growth of incumbents or the proliferation of new ventures? The social processes of internal organizational development offer one possible solution to this problem. Organizational theorists argue that well-developed organizations possess capabilities that allow them to grow (Penrose, 1954; Selznick,195 7). By this thinking, growth slows when organizations become poorly adapted over time-due either to organizational inertia or dysfunctional developments within organizations such as extreme bureaucratization or excessive control (March and Simon, 1958). When existing organizations stagnate, new activities more likely arise through the founding of new organizations than by the expansion of existing ones (Hannan and Freeman, 1984).

Another answer points to forces in the organizational environment that constrain individual organizations as they attempt to grow. Social conditions shape the forms that organizations take when founded (Stinchcombe, 1965),which in turn constrains the organization's growth trajectory. Socially constructed norms limit the range of legitimate organizational action-again impinging on growth prospects (Meyer and Rowan, 1977). Dependence relations with other organizations that control access to scarce resources also constrain organizations (Pfeffer and Salancik, 1978), especially rivals that compete over growth opportunities. Similarly, networks of relations among organizations structure flows of information and resources, working to the disadvantage of organizations that hold weak positions in these networks (Burt, 1992). Generally, various environmental forces limit organizational growth rates, so that only organizations positioned to overcome these constraints enjoy sustained growth (Podolny et al., 1996).

To summarize, existing theories predict that three general types of causes control the relationship between the competing processes of organizational growth and organizational founding: (i) characteristics of individual organizations that encourage or retard their growth,(ii) environmental characteristics that either favor the growth of incumbents or the founding of new organizations, and (iii) competition from existing organizations, which slows the growth rates of rivals and inhibits the founding of new organizations. Our baseline models include variables from each of these categories. With these factors controlled, we then add our specification of Red Queen evolution, which we expect to affect both rates of organizational founding and growth.

3.1 Organizational founding

We model organizational founding rates using the approach developed in organizational ecology (Hannan and Freeman, 1989). In these models, the 'risk set' of potential entrepreneurs is unknown. Consequently, researchers treat the organizational environment itself as the unit of analysis. Nearly all studies of organizational founding take as the unit of analysis some aggregate geographic area, such as a nation state. Then they typically model the arrivals of new organizations as a stochastic process, usually calculating the founding rate by observing counts of the numbers of foundings in each year aggregated over the geographic area being studied (Barron, 1992).

Our study differs from this approach in one important way. Our data include detailed information on the 1182 separate population centers in Illinois and on each of the 2940 banks that ever operate in any of them (not including the Chicago area). Each of these locations represents a distinct banking market for two reasons. First, the communities of Illinois tend to be spread out geographically and separated by rural areas, such that each forms a distinct banking market. Second, throughout the study period, Illinois banking authorities prohibited branching. Consequently, each bank operated only in a single location and competed only with other banks in that local area.

These factors allow us to model the rate of bank founding in each of the 1182 separate locations in Illinois that ever supported at least one bank. We gain several major advantages by studying disaggregated geographic units. One advantage is that we can identify precise competitive arenas. We know that banks operating together in a small community almost certainly compete with each other. By contrast, organizations in large geographic areas may well lie so far apart that they do not compete directly. Another advantage is that we can observe considerable variance among the competitive histories of otherwise similar locations. This heterogeneity helps us identify the parameters of our Red Queen model. After all, if we measured just one competitive location over time, then only one competitive history would inform the estimates. That approach might not provide sufficient independent variance among the competitive history measures to allow for the estimation of their effects. Our data avoid this problem. Finally, our data structure allows us to estimate the hazard of organizational founding using conventional waiting-time models (rather than event-count models). In contrast to most event-count models, these models permit the estimation of duration-dependent founding rates.

Our model takes the form:

r^sub i^ (t) = r^sub i^ (t)[low *]exp[ c^sub fR^[summation operator]^sub k^[mu]^sub Rk^ + c^sub fD^[summation operator]^sub k^[mu]^sub Dk^ + b^sub f^[summation operator]^sub k^[sigma]^sup 2^ ^sub k^ ]

where r^sub i^ (t) is the instantaneous rate of organizational founding in location i and r^sub i^ (t)[low *] is the baseline founding rate specified as a function of variables. We model duration dependence using a piecewiseexponential model, which only mildly restricts the functional form of duration dependence (Barron et al., 1994). By reducing the possibility of misspecifying the functional form of duration dependence, we reduce the likelihood of spuriously finding evidence of Red Queen evolution in our competitive history effects. The baseline rate also incorporates the number, or 'density', of banks. Because we track the precise geographic location of all banks in the data, we can separate the effects of banks within the same locale from those in other locales (Barnett and Carroll, 1987; Carroll and Wade,199 1). In particular, we model local density as a quadratic to allow the founding rate to increase initially with density due to legitimation and then to decline as competition sets in at higher density levels (Hannan and Carroll, 1992). To allow for competitive differences in markets with only one bank, we include a dummy variable to indicate whether a market is monopolized. Also, we include a measure of per-capita local density to allow competition to intensify as density increases relative to the size of the local human population. Non-local density (in the natural logarithm) takes a monotonic specification to allow for legitimation by the proliferation banks in other areas (see Barron, 1999).

The other terms in the model test the Red Queen hypothesis (with the subscript f denoting that the parameters are for the founding model). If our theory holds, then we should find c^sub fR^ < 0 and c^sub fD^ > 0, evidence of lower founding rates when incumbent banks have recent competitive experience and higher founding rates when they have distant past experience. Also, we expect to find b^sub f^ > 0, indicating higher founding rates when incumbent banks have faced amore varied set of cohorts in their competitive histories.

3.2 Organizational growth

Organizational size distributions tend to be highly skewed, with a handful of very large organizations in any given population. Building on this fact, Herbert Simon and his colleagues developed a useful baseline model to study organizational growth. They showed that skewed size distributions result when proportionate growth occurs at random and independent of size-a regularity known as 'Gibrat's Law of proportional effect' (Ijiri and Simon, 1977). In particular, this model depicts organizational growth as a discrete, stochastic process:

S^sub t2^/S^sub t1^= S^sub t1^ ^sup [beta]^[epsilon]

where S denotes organizational size at a given time and [epsilon] is an error term with mean equal to 1 that allows for random growth and decline. The coefficient [beta] equals 0 when size does not affect the growth rate. More recently, this model has been generalized to allow for multivariate estimates of the growth rate, making it extremely valuable for our purposes.

Empirically, tests typically reveal that organizations violate Gibrat's law, with [beta] < 0, suggesting that smaller organizations grow disproportionately. This means that a given organizational unit tends to grow more if it 'stands alone' than it does if it belongs to a larger organization. Some researchers have argued that this finding shows the importance of small organizations in driving economic progress and job creation, a possibility raised some time ago by analysts of job-generation patterns (Birch and McCraken, 1982; Granovetter, 1984). Skeptics have tended to look for various problems of specification error and interpretation in such findings (Brown et al., 1990). Despite these disagreements, the research in this area offers several clear conclusions: that estimates of organizational growth models depend on whether data cover the full range of observed organizational sizes, on how size is measured, and on whether the growth process is well specified (Evans, 1986; Hall, 1987).

Meanwhile, growth also appears to depend on organizational age. Several studies report that younger organizations grow at a higher rate (Barron et al., 1994).With this in mind, and inspired by Hannan and his collaborators' (1998) recent work in which age and size have non-proportionate effects, we adopt the normal Gibrat's law baseline model of organizational growth with one modification: we allow size-dependent growth to differ over an organization's age, using a piecewise age-dependence specification.

With age- and size-dependence well specified, we expand the model to allow for the trade-off between the sizes and numbers of organizations. Recent research accomplishes this by allowing organizational growth rates to vary as a function of competition from other organizations, so that growth rates slow down as the numbers of rival organizations increases. Carroll (1981) pioneered this approach to modeling growth; it has since appeared in a variety of empirical and simulation studies of organizational growth rates (e.g. Barnett and Carroll, 1987; Hannan and Ranger-Moore, 1990; Baum and Mezias, 1993; Barnett, 1994; Barron et al., 1994). Here, we generalize the competitive model of organizational growth to allow for the consequences of Red Queen evolution:

S^sub t2^/S^sub t1^= S^sub t1^ ^sup [beta]^f[N] exp[a^sub gR^[mu]^sub Rj^ + a^sub gD^[mu]^sub Dj^ + b^sub g^[sigma]^sup 2^ ^sub j^ + c^sub gR^[summation operator]^sub k^[mu]^sub Rk^ + c^sub gD^[summation operator]^sub k^[mu]^sub Dk^][epsilon]

where the parameters subscripted with g denote effects on the rate of growth. The function of N includes the various density effects as specified in the founding model. Also, the growth models include the density of banks at an organization's birth to allow for density delay, as described above. f[N] also includes the Euclidean distance with respect to size of each bank from all other banks in its locale to allow for 'size-localized' competition (Carroll and Hannan, 2000).

The other model parameters test our predictions. If our theory holds, then organizations with substantial recent competitive experience will grow faster, while those with a lot of distant-past experience will grow at a slower rate: a^sub gR^ > 0 and a^sub gD^ < 0. We also expect j's rivals, k, to be stronger competitors if they have experienced intense recent competition and weaker rivals if they have had more competitive experience in the distant past, predictions supported if c^sub gR^ < 0 and c^sub gD^ > 0. The spatial-constraints idea predicts growth to be slower for those organizations that compete with varied cohorts of rivals, so that b^sub g^ < 0.

One advantage of studying the Red Queen in terms of organizational growth (as opposed to organizational failure as in Barnett and Hansen,1996) is we can examine size observed multiple times over each organization's lifetime. Consequently, we can estimate restricted versions of the growth model that include a fixed effect for each organization (considered more restricted models because they cannot include time-invariant organization-specific variables such as density at founding). This specification confines attention to intra-organizational variation (over time) in growth, and so selection across organizations cannot explain its estimates. If evidence of Red Queen evolution appears even in this specification, then we can rule out the possibility that natural selection processes alone drive the results.

Note that effects in the growth model should be interpreted in terms of their compounding effects on size. Each year, each variable has its effect on the incremental growth of the organization over that year. This change in size then becomes the baseline size for the following year's proportionate change in size. Consequently, effects that appear to have small marginal effects on size in any one year may compound over time to generate considerable differences across firms.

4. Data and method

Our data include every retail bank that operated in Illinois (excluding Chicago and its surrounding county) at any time from 1900 through 1993. Historically, retail banks in the United States take deposits into personal and business accounts, and lend money to consumers and organizations; this definition excludes various other financial services organizations: investment banks, brokerages, credit unions, and savings and loan associations. Illinois offers an ideal setting for this study because it restricted branching over most of this century, enabling us to identify clearly each bank's geographic market and allowing us to treat the many separate locations in Illinois as independentmarkets.4 The study period begins in 1900 because data quality declines as one goes back into the 19th century. For banks founded prior to 1900,the data source records their year of founding so that the ages of these banks can be used in the analysis. The data, including the sizes, locations, and times of founding and ending for every bank, came from industry directories (Rand McNally, various years; Thomson, various years).

Size, in assets, was recorded for each bank at its birth, death, and at five-year intervals.5 We adjusted asset values for inflation using the Consumer Price Index (CPI) published by the United States Bureau of Labor Statistics. For the construction of independent variables in the founding model, we linearly interpolated size (total assets) across missing years. However, for the growth models, we only used cases with observed size values. Change in size from panel to panel is the dependent variable in these models, annualized to correct for minor differences in panel size and due to the shorter panels at the start and end of an organization's life. This approach made it possible for us to include virtually every organization in the growth models, even if they only survived a short period.

The industry directories list many different types of bank-ending events. Our coders recorded every event as listed in the data sources. We then classified these events as foundings, mergers, and failures. Failures included cases where banking authorities brought in other banks to acquire the assets and/or liabilities of a failing bank. Figures 1-4 show the various events and the number of these banks over time. Note that every directory listed events for several prior years, so that the occasional missing directory did not create a blank window where a bank could start and fail without being noticed. For this reason, we are confident that we have included every bank that ever operated in the sample area.

We calculated several variables to test the Red Queen model. For each bank j we calculated in each year [mu]^sub j^, the mean duration of j's competitive relationships in that year-computed by summing the number of years that j had competed against each of its rivals as of a given year, divided by the number of rivals. To measure the variance of this distribution, we calculated [sigma]^sup 2^ j for each organization in each year. We also split the mean competitive experience measure into j's recent and distant competitive experience using a moving window. To test the sensitivity of our results, we estimated the model with two different recency windows, five years and ten years. As of a given year, competitive experience over the previous five (or ten) years contributed to recent competitive experience ([mu]^sub Rj^),while competitive experience dating back to before the past five (or ten) years added to distant competitive experience ([mu]^sub Dj^).

Summing these terms over j's rivals generated measures of the competitive experience of the rivals faced by j in a given year. To allow for competitive intensity to vary also with the ages of rivals, we summed these variables over all of j's rivals in a given year. This allows us to determine whether the effects of competitive experience arose simply from the aging of rivals (Barnett,199 7)-rather than due to the fact that rivals had competed in the past, as we claim. To sort out this possibility, competition was allowed to vary both with the age and experience of rivals. We kept these two measures distinct by subtracting the mean duration of rivals' experience from the sum of their ages. So computed, the aggregate age term is larger when an organization's rivals have aged with little competition, while aging with competition results in a larger measure for competitive experience.

To help prevent failures from generating selectivity bias in our growth models, we employed Lee's (1983) generalization of Heckman's (1979) sample-selection correction. This approach corrects for bias that might otherwise result as failing organizations leave the data, resulting in a sample of predominately successful, growing organizations. We estimated the selectivity term [lambda] using the failure rate model estimated on these data by Barnett and Hansen (1996).

We included several variables to control for exogenous forces that might affect either growth or founding rates. The size and recent growth of the human population in each location were recorded decennially from the bank directories (and linearly interpolated for intervening years). From US Census sources, we also recorded several control variables for each county: the proportion of the population in urban areas, the number of manufacturing establishments, the number of farms, the number of retail establishments, the number of wholesale establishments, the average farm value, and the average wage per manufacturing worker (US Department of Commerce,19 88, 1992).We adjusted all dollar value measures using the Consumer Price Index.

We also included, in all models, several variables known to affect the development of organizational populations. Lagged foundings, failures, and mergers of banks controled for population dynamics (e.g. Delacroix and Carroll,19 83). The ratios of cumulative failures to cumulative foundings, and of cumulative mergers or acquisitions to cumulative foundings helped to control for differences in the composition of the organizational population due to selection (Sorenson, 2000b). In the founding models, we also included recent change in assets to distinguish between growing and declining markets.

We estimated the growth models on the pooled panels, with independent variables at the start of each panel predicting change over the intervening period. This format yields 21 439 observations of the 2940 organizations. We estimated the growth models using a generalized least squares (GLS) specification that allowed for organization specific random effects. Standard error estimates used White's (1980) method, which is robust to heteroscedasticity. We also estimated a growth model with organization-specific fixed effects using ordinary least squares (OLS),t o determine whether our results held under this more conservative specification.

For the founding models, we segmented the history of each of the 1182 local banking markets into one-year segments so that independent variables could be updated annually. We could not know the first year that each location was at risk of receiving a new bank. Consequently, we allowed the first founding event in each location to indicate that it was now at risk of experiencing bank foundings. Taking this approach, we could not model the first founding in any location. After that first founding, each place entered the risk set every year in which it had a bank. In a few cases, the number of banks in a location fell to zero and the place then left the risk set for future foundings. If a new bank appeared in those cases, the place would once again enter the analysis but that 'new' first founding event would not. The resulting founding data include 1360 foundings in the 1182 local banking markets, split into 72 586 location-year segments. Table 1a,b describes the data. We estimated the founding models using the piecewiseexponential specification available in the statistical program TDA (Blossfeld and Rohwer, 1995). This model suits our data well because many of the places had banks at the start of the study period. These 'left truncated' locations could bias the results if we estimated amore restrictive parametric model (Guo, 1993).

4. Results

Tables 2 and 3 report the estimates of the growth models. Models 1 and 2 estimate random-effects GLS specifications that do not consider the historical timing of competitive experience. Models 3-6 present full specifications that allow for varying effects depending on the timing of experience. Models 4 and 6 re-estimate models 3 and 6, respectively, with fixed organization-specific effects. Models 3 and 4 use a five-year recency window, while models 5 and 6 use a ten-year window. We estimate the effects of organizational age, and the age-specific effects of organizational size, for all models but only report them for model 5 in Table 3. Table 4 presents estimates of various founding model specifications. Each founding model includes historical period and piecewise duration effects. Table 5 reports estimates of these effects for model 10,the most complete specification. In both the growth and founding models, we begin with a baseline model in which competition depends only on the density and ages of rivals. Then, we estimate more elaborate models that include the various measures of competitive experience to test our theory.

Across both Tables 2 and 4,the results strongly support the Red Queen hypotheses. Banks with recent competitive experience grew at a higher rate, decreased their rivals' growth rates, and lowered the founding rate of new banks. By contrast, banks with competitive experience in the distant past grew slower, increased the growth of their rivals, and experienced more new rival bank foundings. Banks facing various cohorts of rivals also grew slower and invited new bank founding. Overall, these results held across estimation methods and for different specifications of historical recency.

Note that support for the Red Queen model does not appear until we estimate the fully specified model. For instance, mode l 2 shows no evidence that competitive experience enhances an organization's growth rate, nor that competitive experience increases the strength of competition from rivals. Distinguishing between recent and distant past experience, models 3-6 shows higher growth rates for organizations with recent competitive experience, and lower growth rates for organizations with competitive experience in the more distant past. We predicted this pattern, but these opposing effects suppress one another in model 2-a specification that does not take the importance of temporal constraints into account. Similarly, models 3-6 show that rivals competed more strongly if they had recent competitive experience, but this effect reversed if rivals had competitive experience in the more distant past. Again this result matches our prediction but was suppressed in model 2. (This result does not hold, however, when estimated with fixed effects and a ten-year recency window.)

As predicted, variance in an organization's competitive experience significantly reduces its growth rate. This effect strongly affected banks with especially high variances-depressing growth rates at the high end by about 11% based on the estimates in model 5. Note that when plotting this effect against the beneficial effect of recent competition, however, the vast majority of observations show organizations benefiting from the Red Queen in the form of higher growth rates despite this variance effect.

The founding models also support the Red Queen hypotheses. Locations with recent competition showed considerably lower founding rates-37% lower per year when evaluated at the mean level of recent competition than for a location that has not had recent competition using a ten-year window, and 53% lower using a five-year window. At the maximum observed levels of recent competition, the chance of a founding falls nearly to zero. Thus, the Red Queen appears to produce a powerful barrier to entry against new banks. These strong effects completely reverse over time, however. As predicted, incumbent organizations with distant-past competitive experience invite new foundings. On average, this effect increased founding rates by about4.4% using a five-year recency window, and about 5% using a ten-year window.

Founding rates increased, as predicted, when incumbent banks had greater variance in their competitive experiences, although this result is not robust when specified using a ten-year window. As with the growth models, the substantive effect o f varied experience differed considerably over the distribution of that variable. At the mean observed value, variance in competitive experience increased founding rates by only about 1% (using the estimate in model 9),while at its maximum the variable nearly tripled the founding rate. When organizations compete with multiple cohorts of rivals simultaneously, organizational founding rates increase considerably.

Both the founding and growth models reveal density-dependent competition. In the growth models, the most robust local density effect is of the monopoly dummy variable. These estimates suggest that a bank's first competitor decreases its growth rate 3-4%. Local density does not have robust effects beyond this dummy variable. The mix of positive effects of local density and negative effects of per-capita local density suggest that competition beyond the first rival depends on whether banks are crowded relative to the size of the human population. Meanwhile, non-local density consistently shows a negative effect in the growth models. Also, note that neither the Euclidean distance measures, nor density-delay, have significant effects in the growth models. [Note, however, that Barnett and Hansen (1996) found that density-delay did significantly increase failure rates among these organizations.] In the founding models, markets with two firms had lower founding rates than monopolized markets. Beyond this level, more banks in a locale increased the founding rate, consistent with the idea that numbers of banks helped to legitimate the form. Overall, however, competition within a locale appears to depend more on the experience of rivals than on the number of rivals.

The effects of age on competition deserve attention. Simply aging (without competing) led to a much smaller increase in competitive intensity in the growth models than aging while competing. Meanwhile, aging without competing actually invites new foundings. If learning enhances competition, it appears to do so not as an automatic consequence of aging, but only when stimulated by the existence of competitors.

Evidence of 'mass dependence' is mixed. The sum of rivals' assets does not have robust effects across the growth models, and positively affects the founding rate. With organizational density controlled, this suggests that entry increases in markets with larger banks on average. However, the prior year's change in bank assets strongly and negatively influences the founding rate. This captures the trade-off between organizational growth and founding; when growth rates of incumbent banks increase, founding rates of new banks decline on average.

4.1 Control Variables

We find evidence that banks grow more slowly if they acquire their rivals over time, and that this strategy increased founding rates considerably. We see this as evidence that organizations attempt to forgo competition and thereby 'kill' the Red Queen, but that this loss of competition leads them to grow more slowly which, in turn, encourages new entrants. Selection by failure also increases the founding rate, but such selection does not appear to decrease the growth rates of incumbents.

Lagged failures increased the founding rate, evidence of a renewal process (Delacroix and Carroll, 1983), though lagged foundings were negatively related to the founding rate. (Lagged mergers, by contrast, had no apparent effect on the founding rate.) In the growth models, lagged foundings lowered the growth rate in three of the four complete models-again evidence that growth and founding act as substitutes. Lagged mergers increased the growth rate, but only in the fixed-effects specifications of the complete models.

Also interesting are the effects of organizational age and size in Table 3. Except for the very young and very old, age and size dependence appear relatively constant. Organizations grew considerably faster both when very young and very small. This raises the possibility that prior tests of Gibrat's law misspecified size and age dependence as proportional. It may be, in fact, that the only violations of Gibrat's law occur for brand-new organizations. Finally, except for the first year, we do not find evidence of duration dependence in the founding rate (see Table 5).

For the exogenous control variables, we report only effects robust across all specifications. Founding rates fell over time as measured by market age. The size and growth of the human population in a locale increased both the growth and founding rates. Urbanization increased the growth rates of existing banks, as did the density of retail establishments. Manufacturing establishment density decreased the founding rate. Finally, counties with higher average farm values saw lower growth rates of existing banks but higher founding rates of new banks.

5. Discussion and conclusion

We began with a call to combine ideas and models from organization learning theory and organizational ecology. Our empirical results, we believe, show the value of such theoretical integration. As predicted, we found that the experience distributions of organizations strongly affect rates of organizational founding and growth. Organizations with recent experience grow considerably faster, and at the same time generate strong competition that slows the growth of other organizations and decreases the founding rate of new organizations. However, as time passes and this competitive experience recedes into the more distant past, these effects reverse-evidence of a competence trap. Furthermore, organizations facing various cohorts of rivals suffer lower growth rates, and their niches show greater vulnerability to the entry of new organizations-evidence of interdependence in coevolution. Overall, the se results provide strong support for our theory, and for a model of Red Queen evolution that depicts organizational learning as a socially constrained process.

Consider the dynamics implied by our findings. Although Red Queen evolution can erect barriers to entry, these barriers appear to come at a cost. When a new firm does enter a market, this new firm can seriously retard the growth of the incumbents. In fact, incumbents experience a type of double jeopardy. First, the new entrant decreases the mean duration of experience for incumbents, thereby lowering growth-rates and entry barriers. Second, the new entrant increases the variance in competitive experience, which further lowers growth-rates and invites even more entries. This effect becomes increasingly disruptive as the time since the last new entrant grows. For example, if a new bank entered a market that had been occupied by a stable duopoly for five years, the entry of the new bank would depress the annual growth rate only incrementally (less than a percentage point by our estimates). If the new bank entered a market that had been a duopoly for twenty years, by contrast, the entry of the new bank would depress the growth rates of the incumbents by several percentage points. (A large effect given that these effects on growth rates cumulate.) Although Red Queen evolution can benefit its participants in some ways, it leaves them ill prepared to deal with new competitive threats.

Managers of organizations often are acutely aware of this problem. In fact, the function of 'strategic management' in modern commercial organizations exists in large part to overcome the Red Queen. Typically, strategic management involves taking actions that isolate an organization from the forces of competition, attaining the 'positional advantage' enjoyed by large monopolists or uniquely high-status organizations. Ironically, our results paint a grim picture for these isolated strategic players. First, these organizations do not enjoy the developmental benefits associated with the Red Queen, a fact that will haunt them should new competitors ever break their monopoly. Second, organizations with positional advantage often attain this stature by strategically acquiring rivals. We show, however, that this strategy dramatically slowed organizational growth rates, suggesting that firms that 'kill' the Red Queen through acquisitions suffer negative consequences in the form of lower growth rates as a result. Because competition drives the development of capabilities, domination and monopoly today imply competitive weakness in the long run.

One interesting implication of our results concerns the corrigibility of organizations. The growth-enhancing consequences of recent experience exceed considerably (more than an order of magnitude in some specifications) the growth inhibiting consequences of distant-past experience. This pattern suggests that, conditional on survival, these organizations could rapidly mend their ways when confronted with new competitive threats. The fact that this conclusion depends on survival, however, should lead us to interpret this pattern cautiously. It may be, as some have speculated, that organizations stay with current practice despite being in a competency trap (Levitt and March, 1988). Declaring that firms can rapidly correct the competency trap in an analysis of survivors misses the fact that the least correctable organizations most likely fail.

Some scholars react to our model by questioning why greater variance among competitive relationships decreases fitness, given that variation often receives billing as an adaptive characteristic of evolving systems (Campbell, 1969). Keep in mind, however, that whether variation proves adaptive depends on environmental conditions (Hannan and Freeman, 1977; Sorenson,20 00a). Specialists fit a particular context well, but have difficulty under other environmental conditions. Generalists, by contrast, 'spread' their fitness so that they fit somewhat in many situations but match none optimally, an organizational 'jack of all trades'. In this light, our argument concerns the benefit of specializing to a few competitive relationships. Low variance in the experience distribution means that an organization continues to coevolve with the same handful of rivals. As long as these conditions remain, we expect (and find) such an organization to be particularly well adapted. Whether this specialized organization will have more trouble adapting to new competitors than will an organization that has experience across a wide variety of competitive relationships raises a different question.

One limitation to our study is that we do not directly observe learning within these organizations. This limitation raises the possibility that our findings might result from processes that do not involve learning. The primary alternative explanation is that our results arise from selection processes culling among organizations, rather than to learning among individual organizations (Sorenson, 2000b). Indeed, as we explained, selection processes likely play an important role in Red Queen evolution: those organizations that survive periods of intense competition will be those that fit better to begin with, and they will grow faster and generate stronger competition even without learning. When times change, these organizations will fall victim to competency traps just as do the learning organizations in our story. We think it likely that some of our explanatory power comes from selection processes rather than through learning-driven organizational change. For several reasons, however, we do not think that most of our results can be attributed to selection in the absence of learning. First, it is difficult to construct a pure selection story that predicts that fitness declines due to high variance in the experience distribution. Second, our models attempt in several ways to control for selection effects (namely, the sample-selection correction in the growth models, and the cumulative failure/foundings measures in all models), although such controls can never completely eliminate the possibility that selection plays a role. Most importantly, for the growth models, the fixed-effects specifications look only at intra-organizational variation over time, and so cannot be driven by heterogeneity across organizations as in the pure-selection argument. The fact that our results hold in the fixed-effects specifications makes us especially confident that selection processes alone cannot explain our findings.

A second alternative explanation is that our historical experience effects, in fact, capture complicated patterns of organizational age dependence. If the age controls do not fully capture the effect of maturation, then those processes may spuriously affect other variables that change with age. We believe that this alternative explanation lacks merit in this case for two reasons. First, we use an extremely flexible piecewise specification of age dependence in the growth models (and we allow for age-size interactions as well). The flexibility of this specification makes it unlikely that we have misspecified the functional form of age dependence. Second, our historical experience terms measure competitive relationships, rather than simply the age of the organization itself or of its constituent parts (cf. Ingram and Baum,199 7). Indeed, just looking at the effects of age itself, our results suggest that older organizations invite higher entry rates-consistent with Baron et al.'s (1994) senescence/obsolescence arguments. The Red Queen requires competitive experience over and above ageing.

A third alternative explanation concerns our variance effect-where organizations facing more varied cohorts of rivals grow slower and invite new entries. It may seem that locations with higher founding rates will, due to frequent entry, have incumbent organizations with higher variance among cohorts of rivals. This might lead to a statistical association between variance and future entries, but not for the reasons that we state. Keep in mind, how ever, that the variance term captures not how much founding activity has taken place, but rather how widely a given number of foundings have been spread over time. (Other variables capture the level of founding activity.) Given a certain level of founding activity in a given location, incumbents will have a larger variance term if these foundings arrive in multiple cohorts instead of concentrating into one or a few cohorts. In this way, our measure of variance in competitive experience should vary independently of the level of founding activity.

In conclusion, we repeat our call to integrate current theories of organizational development. We hope that this synthesis will broaden the approach taken in most developmental theories, where attention remains on characteristics of organizations themselves such as age or size. Rather, we shift attention to the competitive relationships of organizations. These relationships are the engines of development in our theory. Sometimes they make organizations more viable, and in other cases we show that this development slows growth and invites new rivals. But in either case, the development of these organizations depends in large part on their competitive relationships and the social constraints implied by them.

Acknowledgements

We thank Jon Bendor, Ron Burt, Glenn Carroll, Mike Hannan, Dan Levinthal, Jim March, Joel Podolny, Laszla Palos, Arthur Stinchcombe, Toby Stuart and two anonymous reviewers for their useful comments, and the Stanford Graduate School of Business for its financial support. The usual disclaimer applies.

[Reference]  »  View reference page with links

References

Axelrod, R. (1985), The Evolution of Cooperation. Basic Books: New York.

Barnett, W. P. (1994), 'The liability of collective action: growth and change among early American telephone companies,' in J. A. C. Baum and J. V. Singh (eds), Evolutionary Dynamics of Organizations. Oxford University Press: New York, pp. 337-354.

Barnett, W. P. (1997), 'The dynamics of competitive intensity,' Administrative Science Quarterly, 42,128-160.

Barnett, W. P. and G. R. Carroll (1987), 'Competition and mutualism among early telephone companies, 'Administrative Science Quarterly, 32,400-421.

Barnett, W. P. and M. T. Hansen (1996), 'The red queen in organizational evolution,' Strategic Management Journal, 17,139-157.

Barron, D. N. (1992), 'The analysis of count data: overdisperson and autocorrelation,' Sociological Methodology, 22,179-220.

Barron, D. N. (1999), 'The structuring of organizational populations,' American Sociological Review, 64,421-445.

Barron, D. N., E. West and M. T. Hannan (1994), 'A time to grow and a time to die: growth and mortality of credit unions in New York City,1914-1990, ' American Journal of Sociology, 100, 381-421.

Baum, J. A. and S. J. Mezias (1993), 'Competition, institutional linkages, and organizational growth,' Social Science Research, 22,131-164.

Birch, D. L. and S. MacCracken (1982), The Small Business Share of Job Creation-Lessons Learned from the Use of a Longitudinal File. Small Business Administration: Washington, DC.

Blau, P. M. (1955), The Dynamics of Bureaucracy. University of Chicago Press: Chicago.

Blossfeld, H.-P. and G. Rohwer (1995), Techniques of Event History Modeling: New Approaches to Causal Analysis. Lawrence Erlbaum: Mahwah, NJ.

Brown, C., J. Hamilton and J. Medoff (1990), Employers Large and Small. Harvard University Press: Cambridge, MA.

Burt, R. S. (1992), Structural Holes: The Social Structure of Competition. Harvard University Press: Cambridge, MA.

Campbell, D. T. (1969),' Variation and selective retention in socio-cultural evolution,' General Systems, 16,69-85.

Carroll, G. R. (1981), 'Dynamics of organizational expansion in national systems of education,' American Sociological Review, 46,585-599.

Carroll, G. R. and M. T. Hannan (1989), 'Density delay in the evolution of organizational populations: a model and five empirical tests,' Administrative Science Quarterly, 34,411-430.

Carroll, G. R. and M. T. Hannan (2000), The Demography of Corporations and Industries. Princeton University Press: Princeton, NJ.

Carroll, G. R. and J. R. Harrison (1994),' On the historical efficiency of competition between organizational populations,' American Journal of Sociology, 100,720-749.

Carroll, G. R. and J. Wade (1991),'Density dependence in the organizational evolution of the American brewing industry across different levels of analysis,' Social Science Research, 20, 217-302.

Cyert, R. M. and J. G. March (1963), A Behavioral Theory of the Firm. Prentice Hall: Engelwood Cliffs, NJ.

Delacroix, J. and G. R. Carroll (1983), 'Organizational foundings: an ecological study of the newspaper industries of Argentina and Ireland,' Administrative Science Quarterly, 28,274-291.

Evans, D. S. (1986), 'Gibrat's Law, firm growth, and the Fortune 500,' Fordham University working paper.

Granovetter, M. (1984),'S mall is bountiful: labor markets and establishment size,' American Sociological Review, 49,323-334.

Greve, H. R (1996), 'Patterns of competition: the diffusion of a market position in radio broadcasting,' Administrative Science Quarterly, 41,29-60.

Guo, G. (1993),'E vent-history analysis for left-truncated data,' in P. Marsden, (ed.) Sociological Methodology. Basil Blackwell: Cambridge, MA, pp. 217-243.

Hall, B. H. (1987),'The relationship between firm size and firm growth in the US manufacturing sector,' Journal of Industrial Economics, 35,583-606.

Hannan, M. T. and G. R. Carroll (1992), Dynamics of Organizational Populations: Density, Legitimation, and Competition. Oxford University Press: New York.

Hannan, M. T. and J. Freeman (1977), 'The population ecology of organizations,' American Journal of Sociology, 82,929-964.

Hannan, M. T. and J. Freeman (1984), 'Structural inertia and organizational change,' American Sociological Review, 49,149-164.

Hannan, M. T. and J. Freeman (1989), Organizational Ecology. Harvard University Press: Cambridge, MA.

Hannan, M. T. and J. Ranger-Moore (1990), 'The ecology of organizational size distributions: a microsimulation approach,' Journal of Mathematical Sociology, 15,67-89.

Hannan, M. T., G. R. Carroll, S. Dobrev and J. Han (1998), 'Age dependence in organizational mortality revisited, p art I: European and American automobile industries,188 5-1981,' European Sociological Review, 14.

Haveman, H. A. (1993), 'Follow the leader: mimetic isomorphism and entry into new markets,' Administrative Science Quarterly, 38,593-627.

Hawley, A. H. (1950),Human Ecology: A Theory of Community Structure. Ronald Press: New York.

Heckman, J. J. (1979), 'Sample selection bias as specification error,' Econometrica, 47,153-161.

Ijiri, Y. and H. A. Simon (1977), Skew Distributions and the Sizes of Business Firms. North Holland: New York.

Ingram, P. and J. A. C. Baum (1997), 'Chain affiliation and the failure of Manhattan hotels, 1898-1980,' Administrative Science Quarterly, 42,68-102.

Lazerson, M. H. (1988),' Organizational growth of small firms: an outcome of markets and hierarchies?' American Sociological Review, 53,330-342.

Lee, L. F (1983), 'Generalized econometric models with selectivity,' Econometrica, 51,507-512.

Levitt, B. and J. G. March (1988),' Organizational learning,' Annual Review of Sociology, 14, 319-340.

March, J. G. (1988), Decisions and Organizations. Basil Blackwell: Cambridge, MA.

March, J. G. (1994), A Primer on Decision Making: How Decisions Happen. Free Press: New York.

March, J. G. and J. P. Olsen (1989),Rediscovering Institutions: The Organizational Basis of Politics. Free Press: New York.

March, J. G. and H. A. Simon (1958), Organizations. John Wiley: New York.

Mezias, S. J. and T. K. Lant (1994),'Mimetic learning and the evolution of organizational populations,' in J. A. C. Baum and J. V. Singh (eds), Evolutionary Dynamics of Organizations. New York: Oxford.

Meyer, J.W. and B. Rowan (1977),'Institutionalized organizations: formal structure as myth and ceremony,' American Journal of Sociology, 83,340-363.

Miner, A. S. and P. R. Haunschild (1995), 'Population level learning,' Research in Organizational Behavior, 17.

Nelson, R. R (1994), 'Evolutionary theorizing about economic change,' in N. Smelser and R. Swedberg (eds), The Handbook of Economic Sociology. Princeton University Press: Princeton, NJ, pp. 108-136.

Padgett, J. F. and C. K. Ansell (1993),'Robust action and the rise of the Medici,1400-1434,' American Journal of Sociology, 98,1259-1319.

Penrose, E. (1954), The Theory of the Growth of the Firm. Oxford University Press: New York.

Pfeffer, J. and G. R. Salancik (1978), The External Control of Organizations. Harper and Row: New York.

Podolny, J. M.,T. E. Stuart and M. T. Hannan (1996), 'Networks, knowledge, and niches: competition in the worldwide semiconductor industry,19 84-1991,' American Journal of Sociology, 102,659-689.

Rand McNally International Bankers Directory (Annual directories from 1900-1990),Rand McNally, Chicago, IL.

Schumpeter, J. A. (1 934), The Theory of Economic Development. Transaction Books: New Brunswick.

Selznick, P. (1957), Leadership in Administration. Harper and Row: New York.

Slatkin, M. and J. Maynard Smith (1979),'Models of coevolution,' Quarterly Review of Biology, 54,233-263.

Sorenson, O. (2000a), 'Letting the market work for you: an evolutionary perspective on product strategy,' Strategic Management Journal, 21,277-292.

Sorenson, O. (2000b), 'The effect of population level learning on market entry: the American automobile industry,' Social Science Research, 29,307-326.

Sorenson, O. (2001),'Interdependence and adaptability: organizational learning and the long-term effect of integration,' UCLA working paper.

Stenseth, M. C. and J. Maynard Smith (1984),'Coevolution in ecosystems: Red Queen evolution or stasis?,' Evolution, 38,870-880.

Stinchcombe, A. L. (1965), 'Social structure and organizations,' in J. G. March (ed.), Handbook of Organizations. Rand-McNally: Chicago.

Swaminathan, A. (1996), 'Environmental conditions at founding and organizational mortality: a trial-by-fire model,' Academy of Management Journal, 39,1350-1377.

Thomson Bank Directory (Annual directories from 1991-1993), Thomson Financial Publishing: Skokie, IL.

US Department of Commerce (1988), Countyand City Data Book. US Department of Commerce, Social and Economic Statistics Administration, Bureau of the Census: Washington, DC.

US Department of Commerce (1992), USA County Statistics [CD-ROM database]. US Department of Commerce, Social and Economic Statistics Administration, Bureau of the Census: Washington, DC.

Van Valen, L. (1973), 'A new evolutionary law,' Evolutionary Theory, 1,1-30.

White, H. (1980), 'A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity,' Econometrica, 48,817-830.

[Author Affiliation]

Address for correspondence

William P. Barnett: Graduate School of Business, Stanford University, Stanford, CA 94305,USA; barnett_william@gsb.stanford.edu.

Olav Sorenson: Anderson School of Management, University of California, Los Angeles, 110 Westwood Plaza, Box 951481, Los Angeles, CA 90095, USA; olav.sorenson@ anderson.ucla.edu.