Document View

We discuss the creation of organizations and their survival as distinct selection processes, and consider the significance of their divergence. In particular, to understand the implications of entrepreneurial booms, we propose the possibility of asymmetric selection, where entry selection and exit selection differ from each other in strength. An observed increase in founding rates hence may reveal a decline in the selection threshold for entry—implying lower average fitness among boom-time entrants. When such an expansion occurs, organizations born during these periods of heightened entry should suffer higher failure rates if the fitness threshold required for survival remains stable or becomes more stringent. We also discuss other processes that might educe founding waves, and explain the different implications of these accounts for our empirical model. Estimates of the model support our theory of asymmetric selection in two out of three markets using a comprehensive dataset describing organizations in the US computer industry. [PUBLICATION ABSTRACT]

Copyright Oxford University Press(England) Aug 2003

1. Introduction

What can we infer when we observe large numbers of new organizations being created? Does a wave of new foundings indicate that entrepreneurs and capital have flocked to an industry to take advantage of its favorable prospects? If so, increasing rates of entry would signal the likely future fortune of these firms. Or, alternatively, might a surge of new organizations imply lenient times, in which even organizations with slim chances of surviving can marshal the resources necessary for entry? This paper develops an evolutionary model that can tease apart these alternative interpretations of founding waves, thereby advancing our understanding of how organizational founding, like organizational failure, operates as a selection process.

To develop our model, we draw on the two leading approaches to the evolutionary analysis of organizations: organizational ecology (Hannan and Freeman, 1989), and organization learning theory (March, 1988). Our modeling approach comes directly from organizational ecology (Carroll and Hannan, 2000), giving us the advantage of a well-understood framework for the analysis of organizational founding and failure. Our theoretical focus, meanwhile, closely follows one of James March's primary insights: that selection processes may have asymmetric implications for adaptability, where characteristics favored under one circumstance become liabilities in other contexts. Many examples of March's focus on asymmetry come to mind: the 'competency trap', where actors persist in using routines appropriate at one point in time even after the environment has shifted in a way they makes them inappropriate (Levitt and March, 1988); over-investment in 'exploitation' at the expense of 'exploration' due to an asymmetry between short-run and long-run performance expectations (March, 1991); and errors that arise when actors extrapolate from organizational experiences too limited (March et al., 1991) or too exceptional (Denrell and March, 2001) to reflect future circumstances accurately. In each of these instances, March and his colleagues reveal self-defeating dynamics where apparently adaptive organizational processes form and select routines that, in turn, prove to be maladaptive.

In the spirit of March's insight, we consider here the significance of asymmetry between the processes of organizational founding and failure. To begin, we think about the creation of new organizations as a selection process. If the requirements for such 'entry selection' map directly onto the antecedents of survival, then waves of new foundings reflect a wealth of viable opportunities and portend heightened odds of organizational survival. By contrast, what if waves of new foundings indicate only that entry has become particularly easy? In this case, large cohorts of new organizations would experience especially high failure rates if the exit-selection process remained stringent. We refer to this possibility as asymmetric selection, where the threshold for entry selection differs from the threshold for exit selection. In the next section, we attempt to tease apart, theoretically and empirically, the implications of symmetric and asymmetric selection for the dynamics of organizational populations.We then estimate our model using a new data set covering the life histories of all US general-purpose digital computer manufacturers over the period 1951-1994.

2. Asymmetric selection

Social scientists have long considered why and under what conditions new organizations arise. One of the central concerns in these works is how the entrepreneur takes into consideration his chances of success. Most economic renderings, following Schumpeter (1934), start fromthe premise of intended rationality, where entrepreneurs build new organizations when they judge the risk worthwhile (Nelson and Winter, 1982).Weber's sociological approach of course pays considerably more attention to the cultural and social preconditions of entrepreneurship, yet still emphasizes the importance of the entrepreneur's chances of success. Indeed, entrepreneurial success takes on ethical force in Weber's theory, where he argues that, in the wake of the Reformation, moral justification for living required fulfillment of one's 'calling' (in Luther's sense): 'the fulfillment of the obligations imposed upon the individual by his position in the world' (Weber, 1958: 80). Thus, one need not assume a strict rendering of homo economicus to assert that entrepreneurs act in accordance with their perceived chances of success.

In these ways, various received theories suggest that organizational foundings more likely occur when nascent entrepreneurs assess their potential for success favorably. Of course, the intentions of rational action need not materialize, especially in an uncertain process like entrepreneurship. Nonetheless, observers commonly infer from founding activity that opportunities exist, and surges in the founding rate typically lead them to believe that worthwhile opportunities abound. Such a view, however, presumes symmetry between the organizational founding and failure processes: that when organizational viability increases, so too does the arrival rate of new firms. A corollary of this assumption asserts that organizations founded during entrepreneurial booms enjoy particularly high odds of success.

As Stinchcombe (1965) noted, however, the building of new organizations requires not only the motivation to do so, but also the resources and supporting institutions that buttress a society's organizing capacity, such as financial capital, an available and trained workforce, supportive legal institutions, and existing organizations that facilitate trade and provide models for organizing. Stinchcombe's distinction, between the motivation to found organizations and society's organizing capacity, intrigues us. Perhaps organizational founding waves appear not because of an improvement in the prospects for new organizations (the motivational argument), but rather due to an increase in society's organizing capacity. Under these circumstances, founding a firm may become particularly easy as a result of an influx of financial capital seeking to fund new ventures, an oversupply of skilled labor willing to join fledgling firms, or a rapid increase in the willingness of existing institutions and organizations to lend support to start-ups. In such lenient times, even nascent organizations with low chances of success might garner sufficient support to begin operations, leading to 'over-shooting' in organizational entry (see Carroll and Delacroix, 1982). Consequently, founding rates would increase, but the survival rates of new organizations appearing during these periods would fall below those entering in 'stricter' times.

To think more rigorously about asymmetric selection, consider the diagrams in Figure 1. Each illustrates a hypothetical frequency distribution of nascent organizations vying for entry, arrayed from left to right from least to most fit. Case 1, in the upper chart, illustrates the baseline founding process under the assumption that fitness determines which nascent start-ups begin operations successfully. Potential entrants above the selection threshold enter the population of organizations; those to the left of the threshold never materialize. Case 2, the lower diagram, shows how the founding rate will surge if entry becomes easier at times-that is, when the threshold for entry selection relaxes. Under this circumstance, a greater number of nascent organizations successfully enters the population-but a decline in the average fitness of organizations being founded accompanies this surge in the founding rate. As Figure 1 illustrates, if fitness governs the success of nascent start-ups in entering the population, downward fluctuations of the entry selection threshold will lead to increases in the founding rate but decreases in the average fitness of newly founded firms.

The asymmetry comes into play when we look at the exit-selection process. Although fitness drives both entry and exit, we allow the threshold for selection to differ across these two processes. In particular, moving from case 1 to case 2 we assume that the threshold for exit selection remains unchanged, even as entry becomes easier. These circumstances demonstrate the importance of asymmetric selection: as we move from case 1 to case 2, founding rates increase but survival rates fall because the threshold for survival has not moved downwards. Allowing the selection threshold to vary asymmetrically between founding and survival reverses the correlation of these processes; instead of varying together, such that increases in survival mirror increases in founding, decreases in survival rates accompany surges in founding. Alternatively, asymmetric selection can also result because of differences over time between the entry and exit-selection thresholds. At a lenient time for entry, survival criteria might also ease. Over time, however, if the exit-selection threshold again becomes strict, then the asymmetry appears and organizations from large entry cohorts will experience higher failure rates.

We investigate these ideas empirically by estimating the model:

rj(t) = rj(t)*exp[aBj]

where rj(t) represents the exit rate of organization j, allowed to vary as a function of j's market tenure t, rj(t)* denotes j's baseline exit rate estimated in terms of measured variables (specified in an exponential function) and Bj is the size of organization j's birth cohort. Organizations born in boom times (as illustrated in case 2 above) belong to relatively large birth cohorts, while organizations born in small cohorts likely entered the industry during a period of more stringent entry selection. If higher than typical founding rates imply lower fitness, on average, for the organizations being founded, then we should observe a > 0. Or, in the case where exit selection is temporarily eased along with entry selection, then the effects of Bj materialize only after the initial period of relaxed entry and exit thresholds. In our specifications, we allow for both of these scenarios.

2.1 Alternative possibilities

Two alternative accounts suggest situations in which founding rates would increase but fitness levels (and consequently survival) would not decline. First, overall levels of entrepreneurial activity might increase, which would correspond to an upward shift in the distribution in case 1 of Figure 1. In this scenario, the number of nascent organizations above the entry-selection threshold would increase, leading to an increase in foundings. The average fitness of these new foundings, however, would remain unchanged, because the selection threshold has not shifted left, nor has the frequency distribution shifted to the right. An upward shift of the distribution simply implies a greater number of attempted foundings at any given fitness level. If this scenario describes reality better than our proposition, then we would expect to observe a = 0 because the size of an organization's birth cohort would not correlate with its level of fitness.

Founding rates could also increase because, for whatever reasons, the distribution of nascent organizations shifts to the right-reflecting a general increase in the fitness of potential start-ups. Under this scenario, founding rates rise because the frequency of high-quality nascent firms-those likely to fit the environment well-rises. In this case, we would expect to find a < 0 empirically, where members of larger birth cohorts survive longer due to their on-average higher fitness levels. Overall, then, our model allows a test of competing hypotheses about the underlying mechanisms at work during entrepreneurial booms.

2.2 Asymmetric selection or density delay?

The theory of density delay offers empirical implications related to, but clearly distinct from, the expectations of our argument. According to density delay theory, organizations founded during times of crowding have difficulty obtaining the resources necessary to thrive (Carroll and Hannan, 1989). Because these initial difficulties stunt the organization during its crucial formative years, they may have lasting effects on firm performance. Carroll and Hannan (1989), therefore, predict persistently higher rates of failure for organizations founded under crowding. Empirically, this theory implies a model that includes Nfj, the density of organizations at the time of organization j's entry, instead of the size of j's birth cohort Bj, as in our model. Considerable differences in the logic of the two theories underlie these operational differences. Density-delay theory depends on a logic of scarcity during the founding process-with greater scarcity presumed in more crowded environments. The theory of asymmetric selection, by contrast, features variable fitness levels, with lenient selection implying larger-but less fit-birth cohorts. [See Swaminathan (1996) for another application of the logic of variable fitness levels applied to founding conditions.]

We can test the empirical implications of these alternative logics in a single model. Nfj-the density that organization j faces at founding-aggregates the size of j's birth cohort Bj together with the population density prior to these entrants. Specifically, Nfj = Npj + Bj, where Npj represents those organizations in the population that entered prior to organization j's founding (and that still exist in the year j enters). Therefore, we can conduct an empirical test of density delay vs. asymmetric selection by estimating:

rj(t) = rj(t)*exp[aBj+ bNpj].

Density-delay theory predicts that we would improve this model by imposing the constraint a = b, and that a = b > 0. By contrast, the theory of asymmetric selection predicts that a > 0 and a > b. In this way, we can test the competing predictions of these two theories in the same model.

3. Data and method

We collected a comprehensive data set that includes every manufacturer of a digital, general-purpose computer system in the USA from the birth of the industry in 1951 through 1994. Unlike most studies of this sort, which rely on a single published source or on an already existing electronic data file, we combined information from every known source that systematically documents the computer industry. As described in the Appendix, this required that we reconcile many documents for every year in the dataset. The resulting dataset includes 2602 computer manufacturers observed over 10 655 organization-years.

3.1 Measures

Organization characteristics. We calculated an annual measure of each organization's market presence. Determining each firm's markets presented a challenge because market definitions varied not only by source, but also over time (for details see Swanson, 2002). Although some sources used fine-grained market definitions, others provided only coarser classifications. Due to these differences, we settled on three broad market definitions that fit all sources: 'mainframe', or large computers, including supercomputers; 'midrange' computers, including minicomputers, small business computers, servers, workstations and other medium-sized systems; and 'microcomputers'. Any organization could compete in one, two, or all three markets in any given year. Over the course of the industry, roughly 88% of the firms operated in a single market, 11% competed in two markets at any point in time, and only 1% operated in all three markets simultaneously. Figures 2-7 depict the numbers of organizations entering, exiting, and inhabiting each of these three markets over time.

Tenure in the computer industry (in years) records the duration of time elapsed since entry into the computer industry, until the firm exits the industry or is right-censored. We coded the last observation-year to the midpoint for exiting organizations to deal with time aggregation bias in estimation of the hazard models (Peterson, 1991).

Organization size, updated annually, came from three measures available over different time ranges from different sources: the number of employees (from Computers and Automation1), the number of product shipments for an organization (from IDC) or the number of products on the market (from Data Sources) for a given year. Due to the nature of the three markets, we determined each organization's size-small, medium, or large-based on its size relative to other organizations in the market at that time.We based the categories for the size groups on the characteristics of the populations and contemporary renderings of the largest firms in each market, with the number of employees being the primary indicator of organization size, followed by annual shipments. We interpolated values for gaps in reporting within each of the measures. For cases where we only had information on the number of products on the market in a year, we regressed the number of shipments on the number of products to help determine the proper size categories. This procedure yielded size measures for roughly 90% of organization observation-years, with 3510 observations being small, 5327 being medium and 1019 being large.

If an organization ceased to report a measure indicating its size or did not report a size measure until some future date, we projected its organizational size measure for four years into the future or in the past. After the four years, we assigned the organization to the small category unless we found information indicating otherwise. Similarly, we designated organizations with no measures indicating size (n = 115) as small, unless we uncovered information to the contrary.

We determined the dichotomous de alio/de novo distinction for each organization based on our knowledge of the organization's activity prior to entering the computer industry, its founding date, and its name. We define de alio firms as those known to operate in another venue prior to entering the computer industry; conversely, de novo ventures indicate computer start-ups. We coded organizations with contemporaneous founding and entry dates as de novo organizations (n = 185).

Firms founded prior to their first appearance in the computer industry could have been de alio organizations, de novo organizations with pre-production, or de novo organizations with a lagged inclusion in a directory. To account for the latter, we considered firms with a one-year lag between founding and entry to the computer industry de novo (n = 323).At the other end of the spectrum, organizations that entered the industry nine years or more after their establishment received a de alio coding, after a careful screening of organization names. If the name suggested a computer firm (e.g. Advanced Computer Products), then we searched for the organization in Lexis/Nexis and Google (www.google.com) to determine whether its initial activities included computer manufacturing. If no information could be found, we coded such organizations as de alio entrants (n = 474). For organizations with a lag of 2-8 years between reported dates of founding and entry (~51%), we turned first to the organization's name. If it indicated that the firm manufactured computers, we considered it a de novo organization. If the name suggested otherwise, we assigned a de alio code. For the remaining 'gray area' organizations, we looked for evidence of activity prior to entering the computer industry. If the firm operated outside of the computer industry in the interim, we considered the organization a de alio entrant. Likewise, if the organization appeared to have always been involved in computer manufacturing, then we called it a de novo entrant. In the absence of evidence, we defaulted to the de novo coding.

For the set of organizations without founding dates (n = 530), we looked for activity in another setting prior to our observation of them in the computer industry in Lexis/Nexis, Who's Who in Electronics, US Electronics Industry Directory (Harris Publishing Co., various years), and Electronic Buyer's Guide (McGraw-Hill, various years). If we found evidence of activity outside computer manufacturing prior to entering the industry, we coded the entity as de alio. Otherwise, we defaulted again to de novo.Of the 2602 organizations, 1906 entered de novo and 696 entered de alio.

We coded exit from the industry as two different events: merger or ceased to exist as a computer manufacturer. Modeling exits at the level of the industry rather than the sub-market seems most appropriate for these data given that 93% of exits from a segment resulted in industry exit. Only rarely (n = 11) did two organizations of equal stature merge to create a new entity (e.g. Burroughs and Sperry in 1987 creating Unisys).When this happened, we coded the organizations as ceasing to exist under their original identities, and assigned a new ID number, when the collective assets of the two organizations merged to create one, 'new' organization.2 For the remaining exits, we again relied on Lexis/Nexis and contemporary sources or historical accounts for evidence of the nature of the exit. Because this industry has consisted of many players, the press only covers the more prominent organization exits. Most industry exits, in fact, eluded tracing. Therefore, we broadly defined 'ceased to exist', including events such as exiting to another industry, ending computer manufacturing and dissolving the organization. Aside from the mergers among similarly sized organizations, we treat the two ending events of acquisition and ceased to exist both as 'industry exits'.

Industry conditions. We computed two types of density counts from the data: industryand market-level.We measure industry density as the total number of organizations in the population (over all markets) in a given year. For the market densities, we allow manufacturers of a particular type to affect only manufacturers in their own market. In other words, the midrange density count affects only midrange manufacturers and so on. Similarly, we calculate Nfj and Npj at both the industry and market level.

Entry cohort size, Bj, counts the number of organizations that enter the industry in the same year that the organization j enters the industry. We also computed this measure for entry into each of the markets to capture the market-level dynamics of the size of the entering cohort. Market-level entries include organizations that are new to a specific market, even if they have already been operating in other markets within the industry.

Number of current-year entries varies from year to year for each organization, counting the number of organizations entering a given market in a given (current) year. This variable is measured for each of the three markets to capture whether the given market is experiencing a 'boom' in entries in any given year. Note that in the first year that an organization enters a given market, that organization's 'entry cohort size' for that market will be exactly the same as the 'number of current-year entries' for that market. For this reason, when these two variables are included together in the same model, the organization's 'entry cohort size' is set to 0 in its first year so that the number of entries in that year is recorded only in the 'number of current-year entries' variable. We found that this approach reduced collinearity between these variables and so improved the efficiency of our estimates.

We also include measures of exogenous environmental characteristics that may affect the industry exit rate. Three developments in the electronics industry critically influenced the evolution of the computer industry: the silicon transistor in 1954, the integrated circuit in 1961, and the microprocessor in 1971. We represented these advances by creating four period indicators: 1951-1953, 1954-1960, 1961-1970 and 1971-1994. Real gross domestic product is measured annually in billions of 1987 US dollars (US Department of Commerce, c. 2001). Finally, the prime interest rate provides a lagged measure (taken as of the last day of the respective year) of the interest rate banks charge their 'best' customers, indicating the availability of capital (Federal Reserve, c. 2002).

3.2 Method

We specified duration dependence in our model as a piecewise constant hazard rate, an extremely flexible functional form that allows the hazard rate to vary freely from period to period and assumes a constant rate only within each period (Blossfeld and Rohwer, 1997). To allow the independent variables to update, we segmented each organization's life into one-year segments. Descriptive statistics on the segmented life-history data appear in Table 1.We estimated our model using the stpiece STATA routine written by Jesper Sorensen.

4. Results

Table 2 reports pertinent estimates from various failure-rate model specifications. (The effects of the other variables in these models appear in Table 3.) Model 1 includes a quadratic specification of the density of computer manufacturers measured at the industry level, as well as industry-level density measured for each organization in its year of founding.Model 2 disaggregates the density-at-founding variable, allowing the size of an organization's entry cohort to have a distinct effect. Models 3 and 4 mirror models 2 and 3, but with densities and entry cohorts measured at the level of the three more specific computer markets. Model 5 then includes a market-level measure of current year entries.

In model 1, organizational density has a positive first-order effect, and a negative second-order effect; however, the net effects remain positive over the observed range of industry density, evidence of competition among computer manufacturers over the entire range of industry-level density. Also in model 1, density at the time of an organization's entry relates positively to its exit rate as predicted by the theory of density delay. Model 2 then disaggregates the density-at-entry effect into the effect of entry cohort size and density of organizations that existed prior to an organization's founding. Comparing the ?2 statistics of models 1 and 2, model 2 does not improve significantly over model 1. In model 2, we cannot reject the possibility of equality in the coefficient estimates of entry cohort size and density-at-entry. At the industry level, then, the density-delay theory finds support, while the theory of asymmetric selection does not.

Models 3 and 4 repeat these tests, but with densities and entry cohorts measured at the level of each market. When interpreting these findings, keep in mind that an organization in more than one market experiences more than one of these effects. For instance, an organization operating as both a mainframe and midrange computer manufacturer experiences entry-cohort size effects in both of those markets. Also keep in mind that a multiple market organization contributes to more than one of the density or entry cohort terms, so these terms do not sum to the industry level terms in models 1 and 2. Both models 3 and 4 improve over estimates of these models (not shown) constraining all market-specific effects to be equal. In the density effects, competition seems strong, but only reaches levels of statistical significance among midrange computer manufacturers. In model 3, consistent with the theory of density delay, density-at-entry increases the likelihood of exit, but it only reaches statistical significance in the microcomputer market.

Most interesting in light of our theory are the estimates in model 4. Results from two of the three markets support the theory of asymmetric selection-mainframe and midrange computers. In these markets, the larger an organization's entry cohort, the higher its exit rate. Furthermore, the strength of these effects exceeds the effects of the density at entry (less the entry cohort), with model 4 significantly improving on model 3. Hence, these two models strongly support the theory of asymmetric selection, but not density-delay theory. By contrast, the opposite appears true in the microcomputer market. There, we find no statistically significant effect for the size of an organization's entry cohort, but density at entry (less the entry cohort) positively and significantly increases the exit rate. Though consistent with density-delay theory, that theory predicts a positive effect for the entry cohort as well, so it receives only partial support.

The specification in model 5 includes the number of current-year entries by market. This variable is included to allow for the possibility, discussed in our theory section, that failure rates might fall during 'boom times' for entry. If so, then this variable should be negatively related to the failure rate, and we do find evidence of such an effect for the midrange and microcomputer markets. These results are consistent with the idea that asymmetric selection develops over time, with failure rates temporarily falling during entry waves, and then picking up again especially for those organizations founded during the boom.

Table 3 contains several noteworthy results among the other variables. Looking first at the market-specific indicator variables, note the baseline exit rates for organizations in each of the three markets. These baseline rates cumulate for organizations in multiple markets. Given that each is negative, multiple product firms enjoy lower exit rates than single market firms, ceteris paribus, consistent with Sorenson's (2000) findings from a study of workstation manufacturers. In models 1 and 2, the baseline exit rates appear roughly equal in the midrange and mainframe markets, and just slightly higher for microcomputer manufacturers. By comparison, our more complete models show a dramatically lower exit rate for midrange manufacturers. Apparently, models that failed to control for the effects of founding conditions on the exit rate yielded biased estimates of market-specific differences in the exit rate.

Also in Table 3, industry tenure has a relatively constant effect on the exit rate, although there appears to be a slight increase in the rate over the mid-tenure levels.We advanced no priors for the period and macroeconomic effects, although the macroeconomic effects seem to go against what would seem intuitive-with exit rates higher under low interest rates. Before questioning this result, however, one must recognize the endogeneity of interest rate changes. The Federal Reserve typically lowers interest rates in response to weak economic conditions, so this measure may pick up the general economic climate as well as the availability of capital. Organizational size has its typical effects, showing a liability of smallness (middle-sized organizations provide the baseline). De alio organizations have lower exit rates, consistent with the notion that organizations benefit from having established identities. [For a complete study of this issue, see Swanson (2002).]

Finally, Table 4 reports a re-estimate of the full specification (model 5), but with all effects allowed to vary by market. One possible problem with model 5 is that most of the covariates are constrained to affect all organizations similarly, regardless of their markets of operation. Only the variables of theoretical interest are allowed to vary in their effects depending on the markets that an organization inhabits. Consequently, we were concerned that our theoretically pertinent results might be spuriously reflecting differences across the three markets that should be attributed to market-specific effects of the control variables. As it turns out, our results in support of the theory of asymmetric selection are robust even in this ambitious specification. Despite the large number of parameters allowing for the fullest possible variety of market-specific effects among the covariates, entry cohort size continues to predict higher failure rates in two of the three markets.

5. Discussion and conclusion

Evolutionary reasoning has made inroads throughout organization theory; in reading the journals, one frequently encounters theory development that hinges on the operation of selection on organizations. However, by and large researchers conceive of selection occurring in terms of differential rates of exit among organizations that already exist. Entry selection, in contrast, receives very little explicit attention in the theoretical apparatus of evolutionary approaches to organizations. One reason for the dearth of explicit theory about entry selection no doubt resides in the difficulty of obtaining data about nascent firms-the risk set from which entry selection processes must select. As this paper illustrates, however, the problem of data availability does not preclude the study of entry selection.While one cannot argue with a call for more and better data, at issue here is the need to theorize about organizational creation as, itself, a selection process. This shift in theoretical emphasis, we argue, can help us account for patterns in data that already exist-that only an evolutionary perspective can explain.

In particular, we consider the significance of waves of organizational founding. Both casual observers and many scholars might interpret such waves as a sign of munificent opportunities and, consequently, especially viable new firms. Certainly, many entrepreneurs and their financial and institutional backers make such claims during founding booms. These expectations must be tempered, however, by an acknowledgement that founding booms might occur instead as the result of a relaxation in the threshold for entry selection-an increase in the availability of, or an easing in the access to, the resources necessary to found new enterprises. Recognizing this possibility, we investigate empirically whether larger cohorts of new organizations have higher or lower failure rates. In the case of the US computer industry, it appears that precisely those conditions that increase organizational founding end up decreasing the life spans of these new organizations-at least in two out of three markets. Thinking of the founding of organizations as a distinct selection process that can diverge in severity from exit pressure thus leads us to a very different interpretation of a much-observed phenomenon.

Though our results suggest that an easing in the availability of resources for organizational founding rather than a relaxation in performance levels necessary for survival drives these waves of entrepreneurship, our study cannot isolate which particular changes in the environment account for the easing of entry selection. A variety of factors might contribute to this effect, including an expansion in the availability of financial capital available for new ventures, a change in legal regimes or institutions easing founding, or an increase in the pool of skilled labor necessary for a particular type of venture. Future research might usefully attempt to determine precisely which factors ease the resource mobilization problem faced by entrepreneurs.

Regardless of the precise mechanism(s) driving these effects, our results have important implications for research in organizational ecology. A substantial existing literature on organizational failure rates reports that exit rates rise with population density at the time of founding, interpreted as evidence in favor of density-delay theory (for a review, see Carroll and Hannan, 2000). These results, however, might stem either from the current crowding pushing new entrants into unfavorable positions near the edge of the niche (as predicted by density delay), or from the fact that large numbers of firms enter in a short time window when entry barriers fall (asymmetric selection). Would these results remain robust in a model specified to differentiate between these two accounts? Researchers can easily test this, as we have, by decomposing the size of the population at entry into two terms: the number of incumbents and the size of the entry cohort. At least in some cases, it seems likely that asymmetric selection, rather than density delay, might account for these observed effects.

Our results may also have important practical applications for those involved with the creation of new firms. Though one would always expect these actors to evaluate carefully the likely success of a new venture, the potential for asymmetric selection suggests that they should raise their level of skepticism in the face of rising entry rates. That advice might prove quite valuable as existing studies suggest that, on average, actors react in the opposite direction when in the midst of these founding waves. For example, Sorensen and Sorenson (2003) find that the number of entrepreneurs attempting to start a new television station rises in response to successful founding in the previous period. Venture capitalists and other resource holders thus might wisely withhold resources from 'hot' markets.

One cannot claim, however, that entrepreneurs necessarily act irrationally, when they participate in these founding booms. Though the risk of failure increases, the cost of entry (in terms of the effort involved with mobilizing resources) declines.Hence, the risk-adjusted return available to entrepreneurs might actually rise during these manias. A more sinister account of the micro-processes underlying our results might even assert that savvy entrepreneurs recognize the ease of mobilizing resources and hence float inferior ideas into the entrepreneurial marketplace, taking investors (both financial and otherwise) for a ride (essentially, a type of agency problem).

Note that this paper advances only a weak form of the theory of asymmetric selection. Our model allows for asymmetry only in the strength of entry and exit selection-relaxing the threshold for entry selection without relaxing the threshold for survival. One might also consider that entirely different dimensions operate in the founding and survival selection processes. Such differences in the criteria for entry and exit would introduce even stronger asymmetry. Under these circumstances, the least viable organizations (with respect to survival) might actually have the best chances of entry-an evolutionary form of adverse selection.

Research in other contexts hints at the importance of strong form asymmetry in entry and exit. For instance, Barnett and Miner (1992) model selection processes through an internal labor market under conditions where those favored for entering the organization face disadvantages with respect to promotion. This asymmetry, in turn, leads to strong indirect mobility advantages for precisely those least favored in the entry selection process. Another, more entertaining, example of asymmetric selection applied to internal labor markets comes in the form of the Peter Principle (Peter, 1969). The logic of this argument runs as follows: individuals receive promotions on the basis of outstanding performance in their current positions. The skills required for success in their current positions, however,may have little to do with those necessary to succeed at the next level up in the hierarchy. Thus, these individuals do not perform well in their new position. Those who do succeed, however, will receive promotions to the next level of the organization. When one runs this process through to its equilibrium, it implies that all individuals will eventually occupy jobs at which they are incompetent (because they then will not receive further promotion). The deleterious effects, of course, arise from the asymmetric fitness criteria; promotion (selection into the next job level) may have little to do with success in the new job.

Asymmetric selection likely operates in a wide range of contexts.At least with respect to firms, however, the presumption has generally been that entry and exit operate in parallel. We have shown that this need not be the case-and in fact appears not to be true empirically in at least one industry, computer manufacturing. Hence, we encourage researchers to consider where else this symmetry might break down. By thinking about asymmetric selection processes among organizations, we take to the population level a form of reasoning often applied at the organization level by Professor March and his colleagues. We suspect that asymmetric selection will turn out to be as prevalent and important at the population level as it has turned out to be in other aspects of organizational life.

Acknowledgements

We thank the Stanford University Graduate School of Business for research support, and Glenn Carroll and Michael Hannan for useful advice in developing the paper.

[Reference]

References

Barnett, W. P. and A. S. Miner (1992), 'Standing on the shoulders of others: career interdependence in job mobility,' Administrative Science Quarterly, 37, 262-281.

Blossfeld, H. P. and G. Rohwer (1997), Techniques of Event-history Modeling: New Approaches to Causal Analysis. Erlbaum:Mahway,NJ.

Carroll, G. R. and J. Delacroix (1982), 'Organizational mortality in the newspaper industries of Argentina and Ireland: an ecological approach,' Administrative Science Quarterly, 27, 169-198.

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.

Computers and Automation, various years (1951-1973), Berkeley Enterprises: Newtonville.

Data Sources, various years (1982-1996), Ziff-Davis Publishing Co.: New York.

Denrell, J. and J. G. March (2001), 'Adaptation as information restriction: the hot stove effect,' Organization Science, 12, 523-538.

EDP/Industry Report, various years (1967-1983), International Data Corporation: Framingham.

Electronic Buyers' Guide, various years (1950-1986),McGraw-Hill: New York.

Federal Reserve (c. 2002), Statistical Release: Annual Bank Prime Loan. www. federalreserve. gov/release/H15/data. htm.

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

International Data Corporation (1997) Electronic Computer File. International Data Corporation: Framingham.

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

Lexis-Nexis Academic Universe (online) (1999-2001) LEXIS-NEXIS Academic Universe, Congressional Information Service: Bethesda,MD.

March, J. G. (1988), Decisions and Organizations. Blackwell: Oxford.

March, J. G. (1991), 'Exploration and exploitation,' Organization Science, 2, 71-87.

March, J. G., L. S. Sproull and M. Tamuz (1991), 'Learning from samples of one or fewer,' Organization Science, 2, 1-14.

Minicomputer Review, various years (1974-1982),GML Corporation: Lexington, KY.

Nelson, R. R. and S. Winter (1982), An Evolutionary Theory of Economic Change. Belknap Press: Cambridge,MA.

Peter, L. J. (1969), The Peter Principle.W.Morrow: New York

Peterson, T. (1991), 'Time-aggregation bias in continuous-time hazard-rate models,' in P.Marsden (ed.), SociologicalMethodology 1991. Blackwell: Cambridge.

Schumpeter, J. A. (1934), The Theory of Economic Development. Transaction Books: New Brunswick, NJ.

Sorensen, J. B. and O. Sorenson (2003), 'From conception to birth: opportunity perception and resource mobilization in entrepreneurship,' Advances in Strategic Management, 20 (forthcoming).

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

Stinchcombe, A. L. (1965), 'Social structure and organizations,' in J. G.March (ed.), Handbook of Organizations, pp. 142-193. RandMcNally: Chicago, IL.

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

Swanson, A. N. (2002), 'Form coherence and the fates of de alio and de novo organizations in the US digital computer industry: 1951-1994,' unpublished Ph.D. dissertation, Stanford University Department of Sociology.

US Department of Commerce (c. 2001), Real Gross Domestic Product. US Department of Commerce, Bureau of Economic Analysis:Washington, DC.

US Electronics Industry Directory, various years (1990-1992), Harris Publishing Co.: Twinsburg, OH.

Weber, M. (1958), The Protestant Ethic and the Spirit of Capitalism [1904]. Charles Scribner's Sons:New York.

Who's Who in Electronics, various years (1970-1987),Harris Publishing Co.: Twinsburg, OH.

[Author Affiliation]

Address for correspondence

W. P. Barnett, Graduate School of Business, Stanford University, Stanford, CA 94305, USA. Email: Barnett-William@gsb.stanford.edu.

A.-N. Swanson, University of California, Los Angeles, Los Angeles, CA 90095, USA. Email: aswanson@ucla.edu.

O. Sorenson, Anderson Graduate School of Management, University of California, Los Angeles, 110 Westwood Plaza, Box 951481, Los Angeles, CA 90095-1481, USA. Email: Olav.Sorenson@anderson.ucla.edu.