We want to give everyone the opportunity to influence the discussion that will take place during the workshop. Feel free to post comments and questions on the topic of the workshop here.

Peter Murmann & Jan Rivkin

DATE: July 31, 2004

TO: Peter Murmann and Jan Rivkin

FROM: Sid Winter

RE: Your AOM PDW on Adaptation and Selection

I really wish I could participate in your PDW. This memo sets forth much of what I would surely say if I could be there.

The framing that you offer in the description of the PDW contrasts the view that "individual organizations are largely inert" with the perspective that organizations are "fairly flexible and open to redesign." You identify the former with organizational ecology and the latter with the positioning school.. Evolutionary economics, you say, occupies an intermediate position, but one that emphasizes the difficulties of change - hence "selection is crucial." These perspectives at the individual organizational level then imply corresponding answers to the question of how industries, or populations of firms, change. On the one view, it is mostly by way of the selection process, which is interpreted exclusively in survival terms, i.e., organizational births and deaths. On the other, change at the individual organization level (adaptation) plays a much more important role in industry change.

The framing you offer is, I concede, widely accepted. As such, it is reasonable enough as a starting point for your PDW discussion. Nevertheless, in my view this framing is far from satisfactory. Thus, I hope the PDW discussion will treat the framing as problematic.

Over the years, some evolutionary economists, organization theorists and others have sought to emphasize three closely related points that tend to be obscured by the accepted framing.

First, we cannot slide by the very basic question of what we mean by change. This is a problem with deep and ancient philosophical roots. Some of us, in fact, have talked about this in terms of the clash between two familiar and ancient quotations. Heraclitus said that "You cannot step into the same river twice, for new waters are ever flowing onto you." In the Book of Ecclesiastes, however, it is written that "there is no new thing under the sun." These are two very strong claims, each with an element of truth, and it is challenging to sort out how they can be related and reconciled.

My own efforts to grapple with this got a big boost when Michael Cohen pointed out to me the key role of the resolution, or "granularity" of the observations of the thing that is alleged to be changing / not changing. For example, at a sufficiently coarse level of resolution, the Heraclitus claim defies common sense: certainly there are numerous identifiable respects in which a river stays the same long enough for you to step into it twice. However, the explanatory comment about "new waters" indicates that Heraclitus proposes that we should look closely. If we look closely enough at the river, we will see that it is always changing noticeably, even over very short periods of time. (Over longer periods, we see change in terms of water levels, erosion patterns and even gross features of the river's course.) On the other hand, to side with Ecclesiastes in the clash of quotations is to rely on observations of coarser grain, and perhaps also to propose some particular choices regarding the attributes that are "significant" enough to be worth attending to.

These basic considerations transfer directly from rivers to business firms and other organizations . Firms change all the time-we might say that "you can never observe the same firm twice." Clearly, there is a flow of customers, employees, inputs and outputs, technologies, etc. that changes from day to day and implies change in the larger entity, much like the waters of the river. To make the claim that the firm is inert is either to adopt a level of resolution at which change is invisible, or it is to discount visible change on the ground that it is, from some viewpoint, not important.

Essentially the same puzzle appears in a variety of forms, and at the conceptual level the solution is much the same. Granted that they change in some ways, do firms "adapt" or perhaps undergo "strategically significant" change? If not all change "counts," when we try to answer such questions, we first need some basis for determining what sort of change does count. Having settled that, we then need to ask whether an apparent absence of change in that sense might be just a matter of resolution/ granuality - i.e., with more powerful instrumentation, we might see the change. There is also the spatial version of the puzzle; if we want to assess whether McDonalds or Intel operates two different establishments with the same routines, we need to understand "sameness" in a way that addresses the problems of granularity and significance.

The second of my three points brings routines and capabilities explicitly into the story. When the importance of routinized behavior is acknowledged, the distinction between "adaptation" and routinized response presents significant conceptual difficulties, and even more significant ones involving operational measurement. The issues here are similar to, and in fact overlap, those involved in the change/ no change distinction.

One part of the theoretical picture that Nelson and I tried to paint involves the image of a firm that is unchanging or "inert" specifically in the sense that its routines do not change. Firms that operate with unchanging routines can do an enormous amount of changing by Heraclitus' standards, but it is reactive change, pre-programmed response to a changing environment. Hence, some would prefer to view the same history in the Ecclesiastes perspective, as no change at all - the dismissive remark is that this is not the kind of change that matters, (but of course there are many respects in which it does matter). From a strategic perspective, one important question is what happens when the environment departs from its accustomed patterns of fluctuation and presents more substantial novelty - how well will the established routines do then? As we know, it often appears that an organization capable of very complex and flexible routine responses can be highly inert in the face of such a challenge

To me, at least, it is a common-sense, obvious proposition, that routine response governs a large fraction of day-to-day change in a business firm. Indeed, it is as obvious as (and similar to) the proposition that you can step into the same river twice. And it seems to me that this sort of change is not what people ordinarily mean, or should mean, by "adaptation." Presumably nobody thinks that "adaptation" is involved when the bookstore sells more Harry Potter books when a new Harry Potter book comes out, or more Bill Clinton autobiographies when that comes out. Since I am writing this on a delayed flight to Chicago, it also occurs to me to remark that I do not think that the typical adjustments in an airline flight occasioned by bad weather and traffic delays at O'Hare count as "adaptation" by the airline-even though this is not the fully "nominal" routine and the occurrence of the deviation may be hard to predict. (I would agree, however, that sometimes weather conditions impact airline operations sufficiently so that "adaptation" does come into play.) In these, and a great many other examples, our sparse knowledge of the processes is nevertheless enough to make us quite confident that the appropriate responses are largely obvious and familiar to the relevant organization members. And that means, at least to me, that it is not adaptation.

If routinized response should not be counted as "adaptation", there are troublesome implications for observing the latter. Visible, constructive and frequent change in firm behavior cannot be accepted as a reliable indication of "adaptation"—because a lot of that occurs in the form of routinized response. But then, what is a reliable indication of adaptation? Where do you draw the line? How do you establish that an observed change was not latent in the existing routines all along, and the change, when it happened, more a manifestation of inertia than of adaptation? Heraclitus comes back to trouble us here, because the less demanding our criteria for identifying a "constant routine," the more the routines will look constant, and the less adaptation we will see.

This puzzle appears in the most perplexing form when we consider dynamic capabilities - patterned and practiced modes of firm behavior that allow the firm to change its operational routines in a systematic way. My favorite way of dramatizing this point is to quote a statement by Ralph Gomory, made in a meeting at Yale during the time when he was VP for R&D of IBM Somebody made reference to innovation in semiconductors, whereupon Ralph said: "Innovation in semiconductors? I'm not sure there is innovation in semiconductors. They just keep doing the same thing, over and over." Exactly. The semiconductors themselves are changing dramatically and producing dramatic effects in the world as a result, but the processes by which that is happening are so highly crafted and repetitive that words like "innovation" or "adaptation" seem out of place. As Intel goes on enacting Moore's Law time after time, how much innovation/ adaptation is really involved?

It is pretty clear that the only path to resolving the adaptation vs. routinized response difficulty is by way of improved observation and understanding of routinized processes. Only if you know what the routine is can you be confident about the judgment that what is happening at a particular time is not the routine in operation - or parse in an intelligent way the various "matters of degree" that reality generally presents to us.

(continued on next post)

(continued from previous post)

This emphasis on the question ìin what respectsî ñ dimensionalizing and quantifying, making the case for or against particular traits as QGTs ñ is I think complementary to the ìwhenî question you pose for the PDW ñ the proposal for a contingency approach. Needless to say, I am all for a contingency approach ñ but I think the ìrespectsî part of the problem offers a lot of interest and promise, and has been sorely neglected.

Best of luck with the PDW, and extend my regards to my friends and colleagues there!

EXCERPT FROM DRAFT PAPER FOLLOWS
"on recognizable character and the regeneration of organizations"
Excerpt from a paper in progress
authored by Jeremy Birnholtz and Michael D. Cohen,
School of Information, University of Michigan
version of 4 August 02004

Many have argued [including Pentland and Feldman, and Cohen] that when studying recurring action patterns of organizations, such as routines, there is substantial truth in both of these proverbs:

* one does not step into the same river twice (Heraclitus)
* there is no new thing under the sun (Ecclesiastes)

From one perspective (what might be called extremely close up), no situation of action repeats itself identically. From an alternate perspective (what might be called standing far back), every action situation appears as but a variant on, or recombination of, some predecessors or known general types of actions. [ftnt: It's interesting that, in their extreme versions, each of these perspectives is tied to a gloomy assessment of our chances of controlling our world, which is either too variable (chaotic), or not variable enough (overdetermined).]

This "paradox" of the (n)ever-changing world may be more or less apparent, but it never really vanishes. Always in its shadow, organizations must synthesize the diverse, typically inconsistent, capabilities and preferences of their members into a coherent ecology of recurring actions that affects the world in a recognizable way. If the organization is to continue
receiving the resources it requires, it must become able to affect the world at a level exceeding some minimal competence. Indeed, if that loop is not closed, we should hesitate to call the system organized.

All organizations must address the problem of turning perpetual novelty into actionable similarity. [see Weick, sense-making work] (They must also avoid mistaking significant novelty as mere repetition.) The former issue is particularly acute in the conditions of our chosen case. In seasonal organizations such as the summer camp we discuss, activity is
interrupted for long periods and the assembled actors for the next cycle are mainly inexperienced in their roles or are complete novices to the organization. As they gather to (re)create a summer camp together, very many of the "things under the sun" seem new. Nonetheless, only a week later, returning campers and alumni do feel that they have stepped again in the same river. They recognize the resulting organization as being "the same", as having maintained its character.

END EXCERPT

Peter Murmann: Gustavo's study in my mind is interesting example of researching the relative role of adaption versus selection. Here is a overview of his disseration.

The aim of this dissertation is to contribute to the knowledge about the determinants of both aggregate and plant level productivity growth in Developing Countries, with special interest in Latin America. In order to do this I focus on three specific research questions:

(A) What are the forces behind the process of aggregate total factor productivity growth (TFPG)? Aggregate TFPG can be decomposed into plant level improvements (usually as a consequence of so-called learning or retooling processes) and selection (which can be additionally separated into market share reallocations by the incumbents, entrants and exitors). The degree to which these non-mutually exclusive explanations are important matters for at least three reasons. First, from a theoretical point of view, to determine to what extent the growth process can be better captured by the ìrepresentative agent modelî typically used in the neo-classical tradition or by a ìheterogeneous agents modelî largely used in evolutionary theory. Second, from the policy point of view, it helps to characterize the costs of the growth process. And third, because the long-run perspectives of growth in each case are different.

(B) What are the determinants of productivity heterogeneity? The necessary condition for selection is the presence of heterogeneous agents. An examination of the literature suggests that the following factors may account for plant-level productivity heterogeneity: uncertainty, managerial skills, capital vintages, location, localized learning, diffusion lags and shocks. However, there has been much less investigation about the relative contributions of some of these hypotheses, particularly in the context of developing countries. In order to deal with this research question I augmented the standard plant vintage model suggested in the literature by introducing learning and plant-specific effects.

(C) What are the determinants of plant level productivity growth? In this section we relax the assumption that only new plants embody new technology by introducing the possibility that existing plants can adopt new technology. It is at this point that the sector level model of (B) can be expanded to deal with issues about technology adoption costs, learning-by-doing and externalities.

The empirical research is being carried out on a dataset of 105,750 observations corresponding to 12,500 Chilean manufacturing plants classified by 3-digit ISIC sectors and 360 municipalities for the period 1979-2000. Among others the main findings (mainly regarding topics A and B above) are as follows:

First, the long-run TFPG for total manufacturing has been about 2% per year, this result being relatively robust to the methodology used to compute productivity index numbers.

Second, TFPG has been a major component of long-run output growth. We observe that between 40% and 50% of total growth is explained by TFPG. However, analyzing sub-periods there is a negative correlation between TFPG and capital formation, as also reported in studies for industrialized economies.

Third, the major source of aggregate TFPG is found within micro-sectors rather than in the structural change that reallocates resources between sectors. Actually, the contribution of structural change to aggregate TFPG is never larger than 10%. Additionally, when I distribute this figure between static and dynamic shifts, I found that static shifts dominate the structural change. This means that resources are being re-allocated towards sectors with high initial productivities, rather than sectors that have fast growing productivity. These are also sectors with low productivity growth rates, suggesting a case of wrong specialisation.

Fourth, net-entry plus market share re-allocation effects are the main driving forces underlying TFPG growth. Indeed, the combined effect of these two sources represent between 60% and 70% of aggregate TFPG. Of these two, the more important source is the replacement effect given by new highly productive plants substituting for low efficient ones.

Fifth, the previous point does not mean that all the new plants are always better than the incumbents as the standard vintage model would predict. Some of these new plants also have low productivity levels and in some cases this fact leads them to exit not long after entry. This issue is captured both by the shrinking in the productivity spread of the new cohorts as the cohorts age and by the negative correlation obtained between exit and productivity in case of the new plants.

Sixth, the productivity advantage of entrants is due to the presence of high productivity levels at the entry date. There is no clear evidence that the contribution of the net-entry is due to misallocation of post-entry learning or growth effects.

Seventh, the contribution of net entry is time dependent. Net-entry always makes a positive contribution to total factor productivity growth in all the sub-periods except during the last one. This is because the quality of the entrants in the last period deeply deteriorates. This fact coincides with the phase of the largest capital formation and with the increase in the initial scale by the entrants (and potentially their technological complexity). The deterioration of the performance by the new plants is also shared by the evolution of the incumbents.

Ninth, using relatively ìcomparableî methodologies both for TFP computation and decomposition analysis, we found that Chile would be located right in the middle between a developed country like the UK and a fast-growing Asian tiger like Taiwan. The within-plant improvements by the survivors are more important than in the UK, suggesting the existence of an important process of local assimilation and adaptation of technology that, in the Chilean case, is mainly imported. On the other hand the replacement effects and market share re-allocations are less important. In the comparison with Taiwan instead, the within-plant improvements are less relevant, the net-entry is relatively similar, but what makes the difference is the important role of the between survivor market share re-allocations.

Tenth, the (unconditional) productivity of a cohort declines over time as a consequence of the vintage effects. New plants are on average definitively better than the old ones. However, the importance of these vintage effects is remarkably higher in plants born after the main economic reforms of the mid 1970s were already implemented. This suggests that pro-market transformation has been able to create or generate a stream of better plants. In addition to this, while these vintage effects were completely disconnected from the evolution of the technological frontier before the reforms, this is not true in the period after the reforms: changes in the technological frontier feed back on the quality of the new plants born after the economic reforms.

Eleventh, once I control for vintage effects, the conditional productivity of a given cohort of plants grows over time This is due to learning when the surviving members of the cohort are young but more importantly by the selection and exit of the less productive members of the cohorts. We also found negative correlation between learning effects and vintage effects, pointing towards differences in the knowledge creation process of the different sectors: that is, while there are sectors where growth is mainly due to learning, other sectors show growth mainly due to vintage effects (or exogenous technological change).
The combination of all the previous results allows us to describe the Chilean manufacturing sector as one of average TFP growth in comparison to international standards, with a dynamic due mainly to plant turnover within micro-sectors rather than structural change, and with a major role played by market share reallocations and net entry/exit effects. If we were to try to understand the fundamentals of TFP growth in an economy such as Chile, it seems clear that, together with trying to explain the TFPG at sector level, we have to incorporate the issue of heterogeneity. When we have differences in productivity among plants within the same sector of the magnitude that we have found here, it is very hard not to question the assumption of the representative plant as valid, and also to ignore the importance of the competition process and selection as a key source of aggregate productivity growth.

Food for thought provided by Tanya Sammut-Bonnici

Evolutionary theory recognizes three mainstream theories of evolution: Natural Selection, Probability and Complexity. (Depew, Weber, 1995)

The answer of whether Selection or Adaptation prevails may be in this accepted triad of evolutionary thinking. Adaptation Dynamics would be explained through the Complexity paradigm.

At Warwick Business School, UK we're looking at these dynamics in the telecommunication industry and have found that players in the mobile communications market moderate their strategic behavior and opt for strategies that imply a form of complex adaptive behavior.

The common goal becomes the collective survival of the firms in order to eliminate the probability that one firm will prevail and the rest will fail. A complex set of isomorphic strategies emerges at the level of technical standards, network platforms and the consumer platform.

As a result market shares of competing firms are dramatically similar and network externalities are reconfigured to act for the benefit of the whole industry.

Network economics theory predicts that in this industry a dominant player would emerge.

Instead we find that firms perform to very similar levels as a result of a form of complex tacit complicity and co-operation at the level of network standards, platforms and consumer strategies.

Message from Colin Jones

Dear All,

As I will be unable to attend the forthcoming forum, I have a question that I would love addressed it time permits. In recent times, both geoff Hodgson and Thorbjorn Knudsen have advocated that a 'pure' Darwinian approach enables us to nest a Lamarckian (adaptive) perspective within an evolutionary theory that is ultimately governed by a neo-Darwinian (selectionist) force. Thus a dialogic relationship is present due to the continual interaction of both forces.

My question is, does the works of Hodgson and Knudsen, a Darwinian perspective that entertains both Lamarckian and neo-Darwinian processes, already represent a useful contingency approach?

Regards

Colin

wow. lots of great read here. thanks!