A majority of corporate mergers fail. Failure occurs, on average, in every sense: acquiring ? rm stock prices tend to slightly fall when mergers are announced; many acquired companies are later sold off; and profitability of the acquired ? rm is lower after the merger (relative to comparable nonmerged ? rms). 1 Participants report a lot of con? ict during the merger, resulting in high turnover (Buono et al. 1985, Walsh The most conclusive evidence of lower postmerger pro? tability comes from studies by Ravenscraft and Scherer (1987, 1989).
They use Federal Trade Commission line-of-business data to compare companies’ lines of business after they were acquired with a proxy for what their performance would have been without the merger (using comparable control businesses). Operating income as a percentage of assets is lower by 0. 03 for the merged target businesses. This is a substantial (and statistically signi? cant) drop because their pretakeover operating income/asset ratio averaged 0. 115. Also, McGuckin et al.
(1995) provide support for the hypothesis that mergers and acquisitions fail on average, even though their overall interpretation is the opposite (but not clearly supported by their analysis). Speci? cally, they ? nd that acquisitions decrease productivity and employment at the ? rm level (even though acquiring ? rms were highly productive before the acquisition) and this is similarly supported in their initial plant-level analysis. They manage to overturn the productivity result at the plant level only for a subset of plants (those belonging to larger ?
rms). 1 1988). 2 Participants express disappointment in the mergers’ results, and surprise at how disappointed they are. Curiously, widespread merger failure is at odds with the public and media perceptions that mergers are grand things that are almost sure to create enormous business synergies that are good for employees, stockholders, and consumers. Two examples may help illustrate our ideas about cultural con? ict in mergers. In the period leading up to the Daimler-Chrysler merger, both ? rms were performing quite well (Chrysler was the most pro?
table American automaker), and there was widespread expectation that the merger would be successful (Cook 1998). People in both organizations expected that their “merger of equals” would allow each unit to bene? t from the other’s strengths and capabilities. Stockholders in both companies overwhelmingly approved the merger and the stock prices and analyst predictions re? ected this optimism. Performance after the merger, however, was entirely different, particularly at the Chrysler division. In the months 2 Walsh and Ellwood (1991) ? nd that the high rate of turnover among management at acquired ?
rms is not related to poor prior performance, indicating that the turnover is not due to the pruning of underperforming management at the acquired ? rm. Management Science © 2003 INFORMS Vol. 49, No. 4, April 2003, pp. 400–415 0025-1909/03/4904/0400$05. 00 1526-5501 electronic ISSN WEBER AND CAMERER Cultural Con? ict and Merger Failure following the merger, the stock price fell by roughly one half since the immediate postmerger high. The Chrysler division, which had been pro? table prior to the merger, began losing money shortly afterwards and was expected to continue to do so for several years (CNNMoney, February 26, 2001).
In addition, there were signi? cant layoffs at Chrysler following the merger (that had not been anticipated prior to the merger) (CNNMoney, February 29, 2001). Differences in culture between the two organizations were largely responsible for this failure (Vlasic and Stertz 2000). Operations and management were not successfully integrated as “equals” because of the entirely different ways in which the Germans and Americans operated: while Daimler-Benz’s culture stressed a more formal and structured management style, Chrysler favored a more relaxed, freewheeling style (to which it owed a large part of its premerger ?
nancial success). In addition, the two units traditionally held entirely different views on important things like pay scales and travel expenses. As a result of these differences and the German unit’s increasing dominance, performance and employee satisfaction at Chrysler took a steep downturn. There were large numbers of departures among key Chrysler executives and engineers, while the German unit became increasingly dissatis? ed with the performance of the Chrysler division. Chrysler employees, meanwhile, became extremely dissatis?
ed with what they perceived as the source of their division’s problems: Daimler’s attempts to take over the entire organization and impose their culture on the whole ? rm. 3 While cultural con? ict often plays a large role in producing merger failure, it is often neglected when the bene? ts of a potential merger are examined. For instance, following the announcement of the AOLTimeWarner deal, a front-page Wall Street Journal article (Murray et al. 2000) discussed possible determinants of success or failure for the merger (such as synergies, costs, competitor reaction, and so forth). The only clear discussion of possible cultural con?
ict is a single paragraph (out of a 60-column-inch article) revealing how the “different personalities” of AOL’s For instance, one joke told at Chrysler is: “How do you pronounce DaimlerChrysler? ‘Daimler’ the ‘Chrysler’ is silent. ” 3 Steve Case and TimeWarner’s Gerald Levin re? ect cultural differences between the two ? rms. A similar article (Jubak 2000) included a single paragraph entitled “What could go wrong with the synergy strategy. ” Moreover, in these sorts of short, cursory, obligatory discussions of possible cultural con? ict, there is rarely discussion of what steps might be taken if there is dramatic con? ict.
While culture may seem like a “small thing” when evaluating mergers, compared to product-market and resource synergies, we think the opposite is true because culture is pervasive. It affects how the everyday business of the ? rm gets done—whether there is shared understanding during meetings and in promotion policy, how priorities are set and whether they are uniformly recognized, whether promises that get made are carried out, whether the merger partners agree on how time should be spent, and so forth. This paper introduces a simple experimental paradigm to explore cultural con? ict as a possible cause of merger failure.
The guiding hypothesis is that an important component of failure is con? ict between the merging ? rms’ cultural conventions for taking action, and an underestimation by merger partners of how severe, important, and persistent con? icts are. Cultural conventions emerge to make individual ? rms more ef? cient by creating a shared understanding that aids communication and action. However, when two joined ? rms differ in their conventions, this can create a source of con? ict and misunderstanding that prevents the merged ? rm from realizing economic ef? ciency. We hypothesize that the extent of these con?
icts are unexpected because observers focus on tangible aspects of ? rms’ practices (such as technology, capital, and labor costs) and ignore aspects that are more dif? cult to measure such as culture. This leads to overestimation of the value of a merged ? rm at the time of the merger. Our emphasis on cultural con? ict is not meant to suggest, of course, that other potential causes of merger failure are not important. Certainly, agency problems, optimism, and hubris may lead top managers to undertake mergers that are bad for shareholders. Also, holding cultural incompatibility aside, con? icts of interest between employees in two merged ?
rms may also harm the merger. For instance, 401 Management Science/Vol. 49, No. 4, April 2003 WEBER AND CAMERER Cultural Con? ict and Merger Failure employees in each of the two ? rms may have reasons to prefer maintaining the “old way of doing things”—possibly because of learning costs, inertia, and so on—and may, therefore, intentionally resist adopting the other ? rms’ practices. While we recognize these other potential sources of merger failure, our focus is on one speci? c cause: differences in culture may simply make it dif? cult for members of the merged organization to see things in the same way.
Our paradigm also allows us to explore the development of a speci? c form of tacit knowledge in groups, which we use as a metaphor for culture. The experiments we report in this paper speci? cally explore what happens when two groups that have independently developed this tacit shared knowledge—which allows them to operate ef? ciently—need to combine their knowledge and anticipate how dif? cult it will be to do so. Organizational Culture Organizational culture has received considerable attention from organizational researchers, and substantially less attention from economists. While agreement on a precise de?
nition of the concept has proven dif? cult, there are a few important elements shared by most de? nitions. Culture is usually thought of as a general shared social understanding, resulting in commonly held assumptions and views of the world among organizational members (Wilkins and Ouchi 1983, Schall 1983, Rousseau 1990, Schein 1983). Culture is developed in an organization through joint experience, usually over long periods of time. It is useful because it allows an organization’s members to coordinate activity tacitly without having to reach agreement explicitly in every instance.
Language— in the form of codes, symbols, anecdotes, and rules about appropriate statements—plays an important role in organizational culture, constituting a large part of the shared understanding held by organizational members (Schall 1983, Schein 1983, Cremer 1993). However, despite agreement that culture is important, it is dif? cult to precisely measure and study (Schein 1996, Marcoulides and Heck 1993, Rousseau 1990). Researchers have relied on a few different approaches to study culture in organizations (Schein 402 1990, Rousseau 1990). Much of this research is ethnographic observation of
interactions in small numbers of organizations (e. g. , Schein 1983, 1990; Barley 1983). While informative and helpful for inspiring theory, the small samples involved in this type of analysis usually make it dif? cult to draw ? rm conclusions. Another approach consists of questionnaires administered to large numbers of members of a few organizations (e. g. , Schall 1983, Hofstede et al. 1990, O’Reilly et al. 1991, Chatterjee et al. 1992). The questionnaires are usually designed to measure important elements of culture that can then be compared across ?
rms to draw conclusions about how they differ in culture and how culture affects organizational performance. These studies are useful in that they provide concrete empirical measures of differences between ? rms on several dimensions related to culture. However, there is often little agreement from one investigation to the next concerning the key elements of culture. Moreover, these studies often have small numbers of independent observations (? rms) and the usual concerns in survey research like response bias due to the sample of selected ?
rms not being determined randomly or due to nonresponses being correlated with dependent variables, or the fact that respondents retrospectively recall and evaluate cultural variables. Culture has received considerably less attention from economists. Kreps (1990) argues that culture presents organizations with a solution to problems resulting from multiple equilibria in which there may be uncertainty about the appropriate behavior— cultural rules are “focal principles” that point to a socially understood solution, limiting the need for explicit communication.
Hermalin (2001) summarizes the economic approach to culture and presents a formal model in which culture is an ef? ciency-improving asset in which ? rms can invest. Arrow (1974) discusses culture as “codes” developed by organizations to help coordinate activity and points out that these codes are path dependent and may, therefore, differ greatly between ? rms, even though each is ef? cient. Lazear (1999) notes the connection between culture and language, particularly in the extent to which it facilitates ef? cient economic exchange. He de?
nes culture as the shared expectations and patterns of Management Science/Vol. 49, No. 4, April 2003 WEBER AND CAMERER Cultural Con? ict and Merger Failure behavior among individuals, and uses language as a proxy for measuring culture. Cremer (1993) builds on Arrow’s (1974) concept of codes to de? ne culture in a way similar to what we use in our experiments. He de? nes culture as “the part of the stock of knowledge that is shared by a substantial portion of the employees of the ? rm, but not by the general population from which they are drawn” (p. 354).
4 In Cremer’s (1993) model, the organization must respond to outside messages in a coordinated manner, and this is less costly to accomplish when the stock of shared knowledge is greater, because of less time needed for communication. Our experiments can be seen as ways of creating these focal principles and codes in the laboratory, and then measuring their empirical consequences. Taken together, previous work on organizational culture points to a couple of key elements. One is shared understanding among organizational members, which usually comes about through shared experience (or a process of socialization and handing down of traditions).
Culture arises endogenously through this shared experience and is, therefore, path dependent and idiosyncratic. Organizational researchers and economists also agree that this shared understanding is helpful because it allows members of a ? rm to successfully coordinate activity. Our implementation of culture in the laboratory—similar in many ways to Arrow’s (1974) and Cremer’s (1993) de? nition—includes these elements and, therefore, allows us to study culture empirically in a controlled and novel way.
Organizational Culture and Merger Failure Many previous studies have touched on aspects of merger failure, though there are none that conclusively document the causal effect of cultural con? ict. Most studies simply document success or failure of mergers, without directly addressing differences in culture (e. g. , Ravenscraft and Scherer 1987, 1989). Some studies examine the effects on postmerger profitability of product and resource relatedness (Singh This de? nition is similar to the conceptualization underlying the empirical approach of Hofstede et al. (1990) for identifying culture; see also Barley (1983).
4 and Montgomery 1987, Shelton 1988, Harrison et al. 1991). Most evidence suggests that mergers are more successful when merging ? rms make related products. Other studies also examine similarities or differences in some areas related to corporate culture, without directly addressing culture (Shanley and Correa 1992). The studies best able to establish causal effects of con? ict and merger failure simply are not conclusive, due to reasons that often arise when working with ? eld surveys and real-world data. One kind of such study typically surveys members of two ?
rms that were merged, asking them to recall aspects of their premerger ? rm, and then tests the relationship between similarity of premerger ? rms and subsequent performance (e. g. , Chatterjee et al. 1992). Another approach is to rely on existing literature (such as published case studies) that describes previously occurring mergers in detail, coding the content of each article for things like premerger similarity and postmerger synergy realization, and then using these variables to explore the determinants of merger success (e. g. , Larsson and Finkelstein 1999).
While both of these kinds of studies indicate that postmerger integration and performance are related to the cultural similarity and ability to capitalize on synergies of premerger ? rms, the results are not quite conclusive because of problems associated with obtaining the ideal data for such a study. While existing studies provide support for the hypothesis that cultural integration plays an important role in mergers, the causal effects are not clearly determined. This is not because of a ? aw in the way the research was conducted, but due to problems that frequently arise when working with realworld data.
For instance, in the studies relying on survey responses from employees in the merged ? rm, these problems include possible response bias, survivorship bias (only people who stuck with a merged ? rm may return questionnaires), and responses based on possibly biased respondent memory (employees who are a part of a successful merger may recall more similarities between two ? rms before the merger). Similarly, the studies that code from existing literature may also suffer from selection bias (extremely successful or unsuccessful mergers may be written 403 Management Science/Vol. 49, No. 4, April 2003 WEBER AND CAMERER Cultural Con?
ict and Merger Failure about more frequently and “halo effects” (coding bias that leads synergy realization and organizational integration to be retrospectively coded as closer together than they really are). Of course, both of these kinds of studies also suffer from the impossibility of determining how well ? rms would have performed if they had not merged and from possibly weak measures of actual postmerger performance. In pointing out these problems, our goal is not to pick on these studies. In fact, we believe they did the best they could, given the limitations of working with real-world data, and that they provide valuable evidence.
Our goal is to highlight how dif? cult it is to have an ideal ? eld study and, hence, how helpful (at the margin) some exploratory experiments might be as a complement to these kinds of studies. Experiments allow us to control for all these concerns and, therefore, serve as a useful complement to—but not a replacement for—the above studies. This control comes, of course, at a price—reduced generalizability—that springs from the fact that the experiments lack all the features of real-world mergers. However, while experiments often seem to be overly simpli?
ed recreations of complicated realworld phenomena, using them can help get a handle on the basic processes underlying these phenomena. 5 While experiments often give up realism and generalizability, the added control and the ability to precisely measure variables of interest mean that they can be a useful complement to ? eld studies. Culture in Our Experiments In our experiments, culture is a specialized homemade language a pair develops to solve a task. In the task, two subjects with the same set of pictures have to learn to jointly identify a subset of the entire set of pictures.
To do this, they must develop tacit shared knowledge, creating a common way to quickly describe the pictures so that a “manager” subject can guide the “employee” to pick the prespeci? ed subset. Two pairs of subjects, or “? rms,” See, for instance, Knez and Camerer (1994, 2000), Weber (2000), and Weber et al. (2001). All of these papers use simple experiments to help understand complicated real-world phenomena and processes such as organizational growth, leadership, and transfer of practice. 5 separately develop cultures. Then, the two pairs are “merged.
” One manager must then describe pictures simultaneously to two employees. Because the manager previously participated only with one of the two employees, we anticipate that the con? ict in homemade languages will make it dif? cult for the manager to get the new employee to pick the correct pictures, and will also slow down the old employee. We, therefore, expect that performance postmerger will be signi? cantly affected by the difference in languages. The dif? culty of cultural integration is well known. Therefore, the expected result that differences in the languages leads to lower performance should not be surprising.
The more interesting question—given our hypothesis that participants in a merger typically focus on the gains to be obtained (because of synergies, technology, and so forth) and neglect potential cultural integration dif? culties—is whether subjects are aware of the extent of the dif? culty associated with merging. If not, they will underestimate the degree of dif? culty in resolving cultural con? ict. This is measured by having subjects make guesses (for money) about how quickly pictures can be matched in the postmerger phase. Treating these values as subjects’ estimates of the value of the merged ?
rm, we can investigate whether the subjects’ estimates tend to be more optimistic than the actual performance of the merged ? rm, implying that underestimation of cultural con? ict is one important source of merger failure. 6 Finally, we also explore subjects’ attributions about the sources of possible integration dif? culties. Our hypothesis here is that if subjects underestimate the effect of cultural con? ict on postmerger integration and performance, then they will also attribute failure to incompetence on the part of others, rather than to differences in perspectives due to cultural differences.
The picture-matching task in our experiments is analogous to situations in which workers can perform any of several possible related activities, but the This overvaluing of mergers is likely to be aggravated in realworld situations by the “winner’s curse” (the fact that in auctions for objects of common unknown value, bidders fail to adjust for the fact that high bids re? ect the most optimistic guesses about the object value and, statistically, will tend to be higher than the object’s true value). See Roll (1986). 6 404 Management Science/Vol. 49, No. 4, April 2003
WEBER AND CAMERER Cultural Con? ict and Merger Failure correct one depends on some information or knowledge held by managers. Managers have to communicate the information they have to employees quickly and accurately. If a concise natural language to do this does not previousy exist, the organization must develop one to ef? ciently perform. 7 The building blocks of our paradigm are minimal analogies to key features of corporate culture. Each pair must learn to jointly react to an external event (the subset of pictures selected by the experimenter) to accomplish a speci?
c task in which the employee has to perform the appropriate actions for that particular event (selecting the correct subset of pictures). Subjects must begin by using elaborate descriptions of the pictures that focus on several features. After several rounds, their descriptions can be honed to a pithy phrase or word that instantly conveys a shared understanding of what is most distinctive about the picture (this is like the development of jargon, distinctive anecdotes, grammars, and so forth—e. g. , Schall 1983, Schein 1983).
We should note that we are capturing only one facet of the many complex elements of organizational culture. “Culture” in our experiments is a simpli? ed version of the real-world phenomenon: The groups in our experiments are really developing a form of shared knowledge (what are the identifying aspects of each picture) that allows them to ef? ciently perform a task. We use this shared knowledge, or language, as a metaphor for broader ideas of culture. 8 Culture Some similar experiments of this sort have been recently done in game theory (e. g. , Blume et al.
1998). They study whether players in sender-receiver games with common interests can develop a homemade language that maps privately observed types into choices of symbols (such as *, &, and #). Language formation works well, although it is substantially undermined when there is con? ict of interest between the sender’s desire to hide their type and the receiver’s interest in ? guring it out. 8 The kinds of languages we have in mind are the precise codes that arise in police work, air traf? c control, ? lm direction, the military, and so forth.
Examples of development and learning of specialized jargon are policemen (? guring out what crime code to use in reporting an incident), lawyers (categorizing cases and applying precedents), ? lmmaking (“key grip,” “best boy”), and doctors (medical jargon). Note in each case how long it takes newcomers to learn the language and how important it is for them to do so. Also interesting are businesses in which words are used 7 is usually de? ned as a system of values and ideals (what’s good), norms (what’s expected), and conventions of behavior (how things are done).
Agreeing on code names for objects is a huge simpli? cation of all that culture is. But as we stress throughout, it is meant to be just a starting point. Moreover, several researchers have noted the connection between language and culture, speci? cally noting that a group’s language is, perhaps, the most important and most easily directly observable aspect of its culture, because it re? ects group members’ shared understanding and way of representing the world (Schall 1983, Barley 1983, Hofstede 1984, Cremer 1993, Lazear 1999).
What we explore with the merger process is how this tacit shared knowledge is transferred between two groups that need to reconcile disjunctions in their knowledge, and how subjects may be biased in their perceptions of the dif? culty of this process. While it is possible to create something more like a rich organizational culture experimentally, this minimal culture serves as a conservative test of our hypothesis. If even this minimal aspect of culture results in postmerger dif? culties in simple ?
rms consisting of only 2 or 3 employees, then we would expect that differences in more elaborate forms of real organizational culture would result in potentially even greater con? ict. We predict that merger failure will occur if the culture each pair develops is path dependent and idiosyncratic. Even if the two ? rms ef? ciently perform in the premerger period because of their concise languages, if their languages are different, then postmerger communication among all parties will be dif? cult. (Indeed, the more deeply ingrained ? rm-speci? c language is—and the more ef? cient the ?
rm—the harder the integration may be. ) After a merger, it will take some time to either “train” the new employee to understand the acquiring ? rm’s language or to to describe visual or sensory images that are not ordinarily translated into language—such as wine tasting, art, and music. (Musicians in recording studios create a rich language to describe, in words, the kinds of sounds they are striving for—calling a drum sound “too crunchy,” “too tinny,” “cheesy,” and so forth. They also use analogy if it is likely to be commonly understood, as in “How about that ’In the Air Tonight’ sound?
” or “More ’Born in the USA,’ please. ” Management Science/Vol. 49, No. 4, April 2003 405 WEBER AND CAMERER Cultural Con? ict and Merger Failure develop a new language shared by all members of the organization. Note that the “? rms” and “mergers” in our experiment are simpli? ed versions of their real-world analogies. Our ? rms consist of only two or three employees performing one simple task, and the merger consists of taking two companies that consist of one manager and employee each and creating one ? rm with one manager and two employees.
What we are trying to capture is a much more complex process in which many more employees all perform varied and multiple tasks, and in which mergers mean much more complicated combinations of the two ? rms. As with much experimental research, our goal is simply to use a few key elements to recreate a simpli? ed version of the real world in the laboratory, and use this to help understand the phenomenon in a more controlled environment than the real world. Methods The task in our experiments is based on studies by Clark and Wilkes-Gibbs (1986) and Schober and Clark (1989) to address how shared meaning arises in language.
9 In the experiments, every subject is presented with the same set of 16 pictures depicting of? ce environments (see Figure 1 for examples). While most of the pictures share some common elements—people, furniture, room characteristics, and so forth—each picture is unique. Among the aspects that vary are the number of people and their characteristics (gender, clothing, ethnicity), physical aspects of the room (high ceilings, objects on walls, furniture), and the actions of the people (conversing with others in the picture, talking on the telephone, working at a computer).
In every round of the experiment, the experimenter indicated 8 of the 16 pictures in a speci? c order to one subject, in the role of “manager. ” The manager then described the pictures any way he or she liked to the other subject (the “employee”). The employee had to select the correct 8 pictures, in the same order as indicated by the experimenter to the manager. The manager and employee each earned the same amount of money based on how quickly the task was completed.