THE PROCESS OF SCHEMA EMERGENCE: ASSIMILATION, DECONSTRUCTION, UNITIZATION AND THE PLURALITY OF ANALOGIES

Transcription

1 THE PROCESS OF SCHEMA EMERGENCE: ASSIMILATION, DECONSTRUCTION, UNITIZATION AND THE PLURALITY OF ANALOGIES STEVEN KAHL University of Chicago Booth School of Business 5807 South Woodlawn Avenue Chicago, IL Tel: (773) CHRISTOPHER B. BINGHAM Kenan-Flagler Business School The University of North Carolina at Chapel Hill 4200 Suite, McColl Building Chapel Hill, NC Tel: (919)

2 THE PROCESS OF SCHEMA EMERGENCE: ASSIMILATION, DECONSTRUCTION, UNITIZATION AND THE PLURALITY OF ANALOGIES Abstract Schemas are a key concept in strategy and organization theory. While much is known about the value of schemas and their use and structure, direct examination of how new schemas emerge are lacking. Our study of the life insurance industry s development of the business computer schema from addresses this gap. We find that schema emergence involves three key processes: (1) assimilation into an existing schema, (2) deconstruction of the existing schema, and (3) unitization of the new schema into a single cognitive unit. Intriguingly, our study also shows that these three processes are shaped by the interplay of multiple analogies over time. More broadly, beyond contributing to the psychological foundations of strategy and organization by setting forth the processes of emergence, these findings have important implications for the process of change and managerial cognition. ii

3 A longstanding finding in strategy and organization theory is that managing environmental change is difficult. Research suggests that it may not be the environmental changes per se that create difficulties, instead it may be executives cognitive assessments of the change (Kaplan, 2008). Environmental changes like technological innovations make it challenging for organizations to effectively engage in coordinated action since the meaning of the innovations may be unclear. Consequently, the management literature increasingly has focused on the cognitive aspects of and implications for market behavior (Kaplan & Tripsas, 2008; Tripas & Gavetti, 2000;). Research suggests that schemas are central to the ways organizational members deal with these cognitive challenges. Schemas are defined as knowledge structures that contain categories of information and relationships among them (Dane, 2010; DiMaggio, 1997; Elsbach, Barr, & Hargadon, 2005; Fiske & Dyer, 1985; Gick & Holyoak, 1983). Schemas are important because they help give meaning to environmental changes and so help stimulate and shape action (Elsbach et. al., 2005; Nadkarni & Naryayan, 2007; Ruef & Patterson, 2009). A primary agenda for firm research on schema is examining their impact (Schminke et al., 1997), change (Elsbach et. al., 20007) or structural attributes such as size, complexity or focus (Dane, 2010; Nadkarni & Narayanan, 2007). Surprisingly, however, there is very little understanding in strategy and organization theory about how schemas emerge. Work in psychology does provide some understanding of schema emergence. It points to the role of analogies where individuals think back to some situation they experienced or heard about and then apply lessons from that situation to the current one. This research typically relies on lab studies where individuals are carefully provided an unambiguous analog that exactly fits the focal situation (Gick & Holyoak, 1983). Yet, firms rarely operate in such controlled settings with single analogies. So while it appears likely that analogies play a role in the formation of a collective schema, there are no studies that explicitly address how this may occur. Overall, although managerial cognition scholars have continued 1

4 to call for research that moves beyond understanding the content of important cognitive concept like schema, there is very little understanding about schema emergence. As Gavetti and colleagues (2005: 709) stated, Especially intriguing in the question of where cognitive representations come from. Our study addresses this research gap by asking how does a new collective schema emerge over time? Through content analysis, we empirically study how life insurance firms developed a new schema for the business computer from The commercial introduction of the business computer in 1954 was a significant new technology and created the opportunity for potential consumers to develop a new schema to interpret it. Our central contribution is a theoretical framework that helps open the black box of schema emergence. We identify three novel processes. First, assimilation appears important in schema emergence. We find that insurance firms faced multiple analogies for the computer, but largely adopted the one most related to prior technological solutions. While the related analogy helped the new technology gain traction, it had the effect of making the emerging new schema get assimilated into an existing schema. Deconstruction is another key process since we find that through ongoing use with the computer, the existing schema gets deconstructed. The categories and relations of the central analogy become more general and less valid whereas the categories and relations of a marginal analogy become more specific and more valid. Finally, we find support for unitization as a process related to schema emergence. Data reveal that categories and relations related to the marginal analogy become increasingly connected over time so that they become a conceptually distinct stand-alone cognitive unit. Overall, like existing research in psychology, our study reveals the relevance of analogies in schema emergence. However, unlike existing research in psychology, our study uncovers a much more nuanced role of analogies that underscores their number, nature, and continued interplay. Besides contributing to 2

5 the literature on managerial cognition, our study contributes fresh insights to change theory and to the increasingly influential psychological foundations of strategy and organization. THEORETICAL BACKGROUND As scholars become increasingly interested in the cognitive underpinnings of markets, technological change, and strategic decision-making, they have begun to characterize the cognitive concepts that influence how market participants interpret their environment. This research often highlights schema. Schemas are defined as the representations of the categories associated with a concept as well as the relations among those categories (Dane, 2010; Hargadon & Fanelli, 2002). As such, schemas act as cognitive frameworks that simplify information processing (DiMaggio, 1997). They enable firm leaders to order and interpret their environment and so promote efficiency as ongoing activities are cast into relatively stable patterns (Misangyi, Weaver, & Elms, 2008). The schema concept is related to a variety of cognitive concepts, including frames, industry recipes, categories, knowledge structures, dominant logics, or interpretive schemes. While these concepts all vary slightly, they share the assumption that any given experience may be understood by a group of individuals or firms in different ways (Bartunek, 1984). Importantly, many of these concepts also share definitional underpinnings with the schema concept. For example, several scholars studying use the term frame (Benford & Snow, 2000; Kaplan, 2008; Kaplan & Tripsas, 2008) to describe how organizational members understand the world around them. Yet, much empirical work on frames (e.g., Kaplan, 2008; Kaplan & Tripsas, 2008) relies on and explicitly references Goffman s (1974/1986: 21) original conceptualization of frames, as schemata of interpretation. Other organizational scholars focus on interpretive schemes, which are defined as cognitive schemata that map experience of the world (Bartunek, 1984: 355), or the term knowledge structures, which is used interchangeably with the term schema (Walsh, 1995: 285). Thus, we focus on the schema concept not only because it is 3

6 central in the organizations literature but also since other frequently invoked cognitive concepts share conceptual underpinnings with the schema concept. Focusing on schemas allows us to more concretely define and measure what we mean by emergence. Because schemas include categories and their relations, we can analyze emergence in terms of the formation and development of both of these elements. Looking at categories alone may not provide a complete understanding because this approach does not address the relations among elements that also appear to impact interpretation and action. For example, Siggelkow (2001) found that the apparel giant Liz Claiborne had difficulty adapting to changes in its environment largely because of the many independent relationships among its business activities. Likewise, Henderson and Cockburn (1996) argued that many pharmaceutical firms are slow to develop science driven R&D partly because of complementarities among existing practices. Further, thinking of schemas as categories and their relations (and not just categories) is important since it enables scholars to address several key questions related to schema such as their structure, use or change. For example, some studies focus on the structural characteristics of schema, a central characteristic of which is complexity (Eden et. al, 1992; Nadkarni & Narayanan, 2005; 2007). Complexity refers to both the number of distinct categories in a schema as well as the degree of connectedness (i.e., number of relations) among those categories (Walsh, 1995). Theorists argue that complex schemas allow organizational members to accommodate a greater variety of distinct strategy solutions in decision making (Nadkarni & Narayanan, 2007). Other studies focus on schema change. They suggest that with the accumulation of experience and understanding, a given schema will become more stable. That is, the categories comprising schemas, in addition to the relations linking categories to other categories, may become harder to change (Dane, 2010; Fiske & Taylor, 1984; Goldstein & Chance, 1980). An important insight in this work is that within market settings, schemas may exist at a 4

7 collective level, and not just an individual level. As support, firm and population studies emphasizing the role of knowledge argue that schemas exist at collective levels that come to represent the varied understanding between the participants (DiMaggio & Powell, 1983; Grant, 1996; Johnson & Hoopes, 2003; Nadkarni &Narayanan, 2007; Weick, 1979). Johnson and Hoopes, for example, investigate whether intraindustry competitors hold a common pattern of beliefs (or schema). Similarly, Hargadon and Fanelli (2002: 294) note that because schemas are influenced by the surrounding social context, firm members often come to share the same schema. Yet, while extant organizations research has been quite explicit about the definitions of schema as well as the structure, use, change and collective-level relevance of schema, it has been far less explicit about schema emergence. Hence, empirical studies use a variety of assessment procedures to identify the presence, evolution, and nature of a schema, but provide little in-depth understanding about how that schema comes to exist. Fortunately, psychology research provides some insight into schema emergence. It highlights the role of analogies and analogical transfer. Analogical transfer is defined as the transfer of knowledge from one situation or concept to another by a process of mapping- the construction of correspondences (often incomplete) between elements of a source analog and those of a target (Gick & Holyoak, 1983; Holyoak & Thagard, 1989). Experimental evidence (Gick & Holyoak, 1983; Novick & Holyoak, 1991; Schustack & Anderson, 1979) and computational analysis (Winston,1980) suggest a strong relationship between the processing of concrete analogs and the formation of a general schema (e.g., learning the general concept of pump by comparing hearts and water pumps) 1. 1 For example, in one of the earliest studies investigating the induction of a problem schema from concrete analogs, Gick and Holyoak (1983: 3) asked subjects to solve Dunker s (1945) radiation problem : You are a doctor faced with a patient who has a malignant tumor in his stomach. It is impossible to operate on the patient, but unless the tumor is destroyed the patient will die. There is a kind of ray that can be used to destroy the tumor. If the rays reach the tumor all at once at a sufficiently high intensity, the tumor will be destroyed. Unfortunately, at this intensity the healthy tissue that the rays pass through on the way to the tumor will also be destroyed. At lower intensities the rays are harm- less to healthy tissue, but they will not affect the tumor either. What type of procedure might be used to destroy the tumor with 5

8 Although these general theoretical arguments linking analogies to the emergence of a new individual-level schema are generally supported in psychology, unresolved issues remain in the organizations literature about how a collective level schema emerges. First, while the organizations literature discusses analogical transfer, it does not directly and explicitly show how it relates to schema emergence. Rather, it links analogies to important organizational outcomes such as innovation (Hargadon & Sutton, 1997), learning (Neustadt & May, 1986), competitive positioning (Gavetti, Levinthal & Rivkin, 2005), problem solving (Schon, 1993), creativity (Hargadon & Fanelli, 2002), and institutional designs (Etzion & Ferraro, 2010). For example, in their empirical study on technology innovation at the design firm IDEO, Hargadon and Sutton (1997: 739), found that designers made analogies between past solutions and current problems. In one case, designers trying to power a door opener on an electric vehicle charger came up with a solution when they remembered an analogous action in the pistons that open the rear window of a station wagon. However, while these empirical studies provide much understanding about the value of analogical transfer in organizations, they do not directly address how it influences schema emergence. A second unresolved issue is that schema emergence in organizations is likely to be more complicated and dynamic than psychology theory suggests. Psychology studies on schema emergence generally focus on lab studies where individuals face clear, concise and concrete analogs that directly map to target problems (Gick & Holyoak, 1983). Yet, decision makers in organizations rarely face such the rays, and at the same time avoid destroying the healthy tissue? Before having subjects try to solve the radiation problem, Gick and Holyoak had them read stories about analogous problems and their solutions. In one story, The General, a general wants to capture a fortress situated in the middle of the land. There are several roads emitting outward from the fortress. But, each contains mines so passage is only safe with a small force of men as a large force will set off the mines. This makes large force direct attack impossible. The general solves the problems by splitting the army into small groups, positioning each at the start of a different road, and having all groups converge at the same time on the fortress. Other stories ( The Red Adair, The Fire Chief, The Commander ) provided by the authors also offer analogous solutions to the radiation problem so that subjects generally induced a convergence schema. This emergent schema helped subjects solve the radiation problem i.e., the doctor could have several low-intensity rays directed the tumor from different directions at the same time, thereby preserving the healthy but having the effects of the lowintensity rays combine to eliminate the tumor. 6

9 clear-cut situations and solutions. Moreover, psychology studies show that in order for a schema to emerge through analogical transfer, many subjects needed an explicit hit to help them see the correspondence between the elements of a source analog and those of a target (Novick & Holyoak, 1991). It is hard to understand how (if at all) the giving of hints might occur in organizations. Finally, the nature of analogies is likely to be different at the organizational level. Studies in psychology find that individuals usually draw on one analogy at time to generate new schema. Serial processing of analogies takes place where each is not in direct competition with another (Gick & Holyoak, 1983). By contrast, in organizations individuals may face multiple competing analogies that must be processed simultaneously, not serially. Further, these analogies may be of varying familiarity some more incremental and some more radical thereby making it less clear which analogies might be adopted and why. Together, these points suggest that existing psychological theory for schema emergence at the individual level may be imprecise for schema emergence at the collective level. Similarly, the nature of an organization complicates understanding for how a collective (vs. individual level) schema might emerge. On the one hand, it seems likely that the number of potential relationships between categories might increase combinatorially with the number of new categories added over time by different users thereby permitting a vast and unpredictable range of different potential schema structures to emerge. Yet, on the other hand, having a large number of relations and categories may be insufficient to guarantee schema emergence; many of the relations or categories may be less relevant or salient. Indeed, it may be the case that certain emergent categories and relations can inhibit schema formation by drawing a disproportionate amount of attention of users and so drown out other important categories and their relations. Existing organizational literature on categories provides some tentative support for these arguments. It highlights the importance of the differentiation between categories in order to establish a distinct category (Hannan, Polos, & Carroll, 2007). Yet because this 7

10 work focuses on only one component of schemas, categories, and overlooks relations among categories, it is not clear if what happens at the category level happens at the schema level. 2 In sum, while the organizations literature is generally clear that collective-level schemas exist and are central to the way firm members come together and have a shared sense of belonging (Bartunek, 1984; Johnson & Hoopes, 2003), this literature it is generally unclear how a collective level schema comes to exist. Overall, these unresolved issues suggest a lack of specific understanding regarding how a collective schema emerges. So, while existing literature suggests that the development of a new schema involves the identification of new categories and relations, there is little in-depth understanding about how this developmental process occurs. In this paper, we address this gap. In what follows, we detail our methods, and then set forth our empirically grounded theoretical framework for schema emergence. DATA AND METHODS Given the general lack of research on schema development, we combined theory elaboration (Lee, 1999) and theory generation (Eisenhardt, 1989) in our analysis. Thus, we were aware of the extant literature on schema and so examined data for the relevance constructs such as categories and relations among categories. But, we also looked for unexpected types of processes by which those categories and relations developed over time and became institutionalized as a new schema. Case Selection and Data Sources 2 Much research on the schema concept in psychology began in parallel to theoretical developments on categorization i.e., how people categorize objects. Categorization research highlights the concept of prototypes, an ideal instance of a category (e.g., a robin a better prototype of the category bird than ostrich). Individuals decide whether a new instance is a member of the category by determining the similarity of its features to those of the prototype. The more features it shares, the more consistently it will be viewed as a typical category member. Although research on categories and schema both share a concern with the way knowledge shapes understanding, they are different in a number of ways. For example, category research helps shed light on how a label is applied. Schema research extends this view by helping explain what effect the label has on a category and how relations link it to other categories. Further, the categorization literature discusses how prototype categories often have typical features (size, color, shape) and most attributes are known, even if actual category members differ so extensively on those features that they become irrelevant for identifying category members (Anderson, 1980). In contrast to the concept of categories, the concept of schema lets some features remain unspecified. Due to this flexibility, research suggests that a schema might be a more efficient cognitive representation than a prototype category since it requires fewer specifics and is more focused on the essence of category membership (Mandler, 1979). 8

11 Our setting is how the life insurance industry developed a new cognitive schema to interpret the business computer from By business computer, we mean those computers used to process and manage transactions as opposed to more computationally focused computers, which were used in the military and as well as in business. In this case, the insurance industry used business computers to process and issue new applications as well as manage the paying of premiums. For simplicity sake, throughout the paper we will use the term computer to refer to these business computers. As a radical new technology, computers did not exist in a previous schema and offers an opportunity to examine how a new schema gets developed. We focus on the insurance industry for several reasons. First, as Yates (2005) notes, insurance was one of the first and largest users of the computer. Thus, understanding how this industry developed its understanding of the computer is important to our general understanding of the commercialization of the computer within industry. Second, the large financial and organizational commitment to purchase a computer left a rich archival record of public discussion about the computer. Last, certain characteristics of the insurance industry s history allow us to address changes within the schema. The collective schema we are interested in is the schema of insurance companies, not the broader market or the shared schema between the various market participants. Thus, this is the schema that consumers used to evaluate the new technology. Yates (2005) notes that three professional and trade associations played the prominent role in shaping how insurance firms interpreted and used the computer: The Society of Actuaries (SOA), Life Office Management Association (LOMA), and Insurance Accountant and Statistical Association (IASA). SOA is the main professional society for actuaries, a key occupation within the insurance industry. LOMA and IASA were more trade associations that focused on general issues related to insurance, and in particular, the use of office technology. At the field level, these associations created committees to investigate the computer, held 9

12 conferences, and as will be discussed in SOA s case, distributed an influential report on the computer before its commercial release. The occupational groups who participated within these associations - accounting, administration, actuary, and the systems group responsible for developing and managing office technology - also represented the main occupational groups who investigated and introduced the computer within the insurance organization. Thus, we take these associations to represent the core collective discourse about the computer for insurance firms. Consequently, our archival efforts focused on the proceedings of these three associations, their electronics committee meetings/reports, as well as an influential book for an insurance professional associated with this group. The proceeding documents included detailed discussions about how members plan to engage or are currently engaging with using the computer in many cases with detailed procedures of how the system works, as well as other important topics such as organizational and career issues of computer-related professionals and how to purchase and evaluate computers. We collected 399 articles, reports, and books from insurance representatives that discussed the preexisting technologies as well as the computer from While the computer was not commercially released until 1954, we collected works dating back to 1945 to get a sense of the pre-existing schema as well as early interpretations of the computer before it was actually used. A unique feature of this case is that insurance firms developed a schema prior to the commercial release of the computer and their use of the computer helps us assess the influence of experience in modifying a schema. While the computer and its schema certainly continue to evolve even today, we stop in the mid 1970s because by then the computer schema had fully emerged as an independent knowledge structure our primarily concern in 3 As noted, the SOA developed important early documents, which we include in our study. But, since we are interested in the business computer and actuaries also worked on computational computers, the bulk of these articles throughout this time period come from IASA and to a lesser extent LOMA. Actuaries were well- represented presenters at the IASA and LOMA meetings. In addition, while computer manufacturers presented at this meetings, we privileged the insurance representative s accounts because we are interested in their schema. Yates (2005) notes how insurance played an important role in shaping the computer manufacturer s perspective. 10

13 this paper. Coding Methodology Following an established research stream, we take written discourse to represent the cognitive schemas (Barr, Stimpert, & Huff, 1992; Tsoukas, 2009). This approach is particularly relevant to the insurance case. As noted in the previous section, written discourse in the form of these reports and proceedings was a primary source of communication and learning between insurance companies. Methodologically we are interested in using these texts to identify and measure the changes in the insurance firms collective schema. By defining schemas in terms of their categories and relations, this process entails identifying the various categories and how they are related within each text. Grammatically, we interpret nouns as categories and verbs as relations, which means capturing all of the nouns within the text as well as the verbs and the nouns which the verbs connect. Some nouns may not be connected through relational verbs, so it is important to identify categories and relations separately. Conventional discourse analysis tools are not well equipped to identify these relational structures, but Carley (1993, 1997) has identified a semi-automatic process, called cognitive mapping that measures cognitive schemas. Cognitive mapping is a form of content analysis that involves processing sentences within a text to isolate the nouns and verbs. Carley and colleagues (1993, 1997) have developed software, called Automap, to assist in this process. We adopted her general procedure for coding, and used Automap to help define the verbs and nouns. More specifically, we scanned each original document and used OCR technology to convert the images to text files. 4 We input each text in the Automap software which identified the instances of all the words, including the parts of speech. Carley notes that an important part of this process involves considering which nouns and verbs to capture within the text. We focused on nouns and verbs that discussed the computer and its operations, 4 11 documents were either unscanable or the OCR results were of too poor quality to use. For these cases, we did not use the Automap software, but coded the category and relational structure by reading through the text. 11

14 choosing to exclude extraneous information introducing the topic or describing the company. We processed the nouns and verbs separately so the identified categories are not dependent upon the relational structure. While this software can also map the relations, we found the texts to be written in a style that made it difficult to accurately capture this in an automated fashion. Therefore, with the generated list of nouns and verbs, we read through each text capturing what categories each verb connects. To give a sense of this process consider the following statement from an article presented at an association meeting: The cards are then mechanically calculated and punched with the unpaid number of weeks and the unpaid portion of the first-year's premium and commission to be withdrawn (Beebe, 1947). Following this procedure, the categories include cards, unpaid number of weeks and portion of the premium, and the unpaid number of weeks and portion of the commission. The relations include calculate and punch. The full relational statements connect card with each of the two data elements. We coded each verbs in its verb form. For example, in this case, calculated would be coded as calculate. However, with nouns, we captured each in its form as represented in the text in order to identify how it changed over time as well as identify new emergent concepts. Carley (1993, 1997) also recognizes that the analyst must decide how to code frequency either as each occurrence or whether it occurs in the text at all. Since many of these texts were procedural in nature, they had many occurrences of the same concepts and relations. Consequently, we only identified whether they were present in the text. We completed this procedure for each text for each year, generating 4,330 unique categories and 451 unique relations 5. Given the inherent subjectivity of this process, we randomly selected ten articles and had two graduate students not associated with the project code each. 92% of the categories, 5 The total number of relations appears lower than the total number of categories. This relates to our coding strategy. The number of categories is high because we coded categories at the lowest level of analysis that is, how the terms were exactly presented in the text. (e.g., IBM_650 or IBM_650_digital_computer). The number of relations is lower since when coding the verbs we focused on relational verbs that connect categories. 12

15 relations, and relational connections of one coder were included in our coding, and 88% of the other. The collective schema for each year represents the aggregate of each text for a given year. Appendices 1 and 2 provides a summary of the most frequently occurring categories (Appendix 1) and relations (Appendix 2) for the collective computer schema from 1947 (corresponding to the first discussion of the computer) through To avoid the idiosyncrasies of year to year fluxuation in the level of discourse, we collapse the initial discussions of the computer prior to commercial introduction ( ) together and group by three year increments after its commercial introduction in We adopt this group for much of our temporal analysis. Each cell represents the total number of articles that used a particular category or relation. The total number of articles for each time period is represented at the top of each table. These frequency counts over time show how certain categories and relations become more or less prevalent over time and so help assess schema emergence. We refer to these Appendices throughout the paper. Finally, it should be noted that throughout our narration of the story we try to use the terms used at the time, but often use the term computer. This is not intended to reify the computer. However, we do note that Appendix 1 shows that the term computer was used frequently early on. Yet, in these texts it is also associated with other terms. We capture this variance and changes throughout the history of the schema emergence. Following Kaplan and Tripsas (2008) call for more historical work within the cognitive processes of technological change, we extend Carley s (1997) method by considering the historical context and the actual use of computers. Our analysis of the associations materials revealed the importance of technological changes in the computer, how the insurance industry used the computer, and broader trends about computer technology such as the development of management information systems (MIS) in the late 1960s and 1970s. Therefore, we supplement our analysis of association proceedings with historical analysis of technological changes, research on the life insurance industry s 13

16 uses of the computer (Yates, 2005), and the general history of these broader movements (Haigh, 2001). HISTORICAL CONTEXT AND SUMMARY OF THE SCHEMA EMERGENCE PROCESS Life insurance firms provide policyholders coverage against potential loss in exchange for premiums. Beyond the actuarial and investment analysis, much of the work in insurance firms are routine and clerical in nature: preparing and processing policy records, determining the premiums paid, dividends, and agent s commissions, as well as notifying and collecting policyholder s premium payments, and the accompanying accounting procedures to record these transactions (Adams, 1946). Historically, insurance firms invested substantially in clerical workers and technology to efficiently manage the policy and accounting processes (Yates, 1989). By the 1940s, virtually all insurance firms used a combination of business machines tabulating machines, sorters, verifiers, calculators, and addressing machines to sort and calculate information represented as punched holes in cards. These punch cards became the record on which policy and accounting information was stored. IBM historians have noted that Insurance companies were among the largest and most sophisticated business users of punched-card machines in the late 1940s and 1950s (Bashe, Johnson, Palmer, & Pugh, 1986). During World War II, industries in general experienced a clerical labor shortage, which was exacerbated in insurance by a post-war life insurance boom (Bureau of Labor Statistics, 1955). As a result, the insurance industry became increasingly interested in new technologies that aided in information processing. One promising, radically new technology that emerged from the war was the computer. The U.S. government in World War II used primitive computers mainly for computational purposes such as calculating missile trajectories. After the war, technology manufacturers in search of a commercial market began developing what became known as business computers, intended not solely for computational work but also for managing business processes such as processing premiums. Part of 14

17 the process of adopting the computer involved developing interpretations of what the computer was i.e., a schema that captured understanding about the computer. Our historical analysis surfaced three distinct temporally phased processes shaping the insurance firms development of the computer schema: assimilation, deconstruction, and unitization. Unexpectedly, we found that each process relates to analogical transfer, but in a more nuanced and dynamic way than what is described in the extant psychology literature. First, our study shows that before the computer was even commercially available in 1954, insurance companies faced two competing but fundamentally distinct analogies for the new technology machine and brain. Insurance companies assimilated the machine analogy into an existing office machine schema, treating it as standard piece of mechanized equipment. We call this stage assimilation. While assimilation helped initiate action to purchase the computer, the novel aspects of the computer, such as programming and decision-making related to the brain analogy, received little attention and so were largely ignored. After gaining some experience with using the computer into the 1960s, insurance companies began to reflect more deeply about how they had assimilated the computer. We call this reflective process, deconstruction, because it led to the weakening of some existing categories and relations in the existing schema and facilitated the incorporation novel categories and relations related to the brain analogy. Thus, the prior analogy of computer as a brain that was largely ignored was made more relevant while the machine analogy that was assimilated was made less relevant. Finally, the last process, unitization, served to strengthen the more novel categories and relations related to brain analogy so that what emerges is a new stand-alone schema that is dissociated with the existing schema. Overall, a key insight is that the schema emergence reflects the ongoing interplay of multiple distinct analogies, rather than the quick selection of one analogy over others as suggested in the literature. In what follows, we discuss the details of each process. 15

18 THE ASSIMILATION PROCESS The initial starting point of the emergence process is that the new object (the computer in this study) must be cognitively recognized. The existing literature argues that making analogies between past technological solutions and new technology is useful because it increases the recognition and ultimately the legitimization of the new technology (Hargadon & Sutton, 1997). Our data, however, reveal that such grounded analogies can initially have negative, not positive, effects since they divert attention away from other relevant analogies that spotlight what is truly unique about the new technology. Specifically, we found that prior to the commercial release of the computer in 1954 insurance companies faced two distinct analogies for the new technology: (1) machine and (2) brain. Because the machine analogy was grounded in past technological solutions (i.e., tabulating machines), insurance companies largely adopted this analogy and not the brain analogy. The result of using the machine analogy was that insurance firms assimilated the new technology (computer) by fitting it directly into their existing schema for office machines. This assimilation however came at the cost of pushing into the background the analogy with the human brain that emphasized the novelty and potentiality of the computer. To show this assimilation process requires identifying the schemas of the proposed analogies as well as the existing schema used to evaluate these alternatives. Yates (2005) historical analysis reveals that insurance firms began to substantially investigate and publicly discuss the computer in the mid-1940s. The initial discourse on the computer included presentations at conferences from insurance and computer manufacturers, commissioned reports to study how the computer could be used, and even books on the subject. 6 Therefore, our initial analysis focused on 1945 through the commercial introduction of the computer in We identified 67 texts, 41 of which addressed how insurance 6 Studies of cognitive practices often focus on the media (Rosa et al, 1999). In this case, the general media did not start actively discussing the computer until closer to its commercial release in

19 firms used the existing office technologies, such as tabulating machines, collators, and printers to process insurance work. The following passage about how Prudential used various tabulating machines to prepare important documents illustrates how insurers thought about the current technology: From approved new business applications, policy writing and beneficiary cards are key punched and verified. The policy writing cards are mechanically sorted and matched with master cards (by collator) to insure the accuracy of the age, kind, premium and amount of insurance in each case. The policy writing and beneficiary cards are next mechanically merged in policy number order and used to write the policies on a bill feed tabulator. After completing the listing of the agents' register sheets, the new business, reinstatement and life transfer cards are reproduced to in-force file cards. (Beebe, 1947: 191-2). This passage highlights the focus on the punch cards as the primary unit of information, the machine doing the work, and the functions through which it gets processed, manipulated, and calculated. The remaining 26 texts developed two dominant analogies for the new technology: one comparing the computer to the brain and another with a machine. Edmund Berkeley (see Yates (1997) for more detailed discussion of Berkeley), an executive from Prudential who had extensive interactions with computer manufacturers, developed an the human brain analogy. At industry meetings, he discussed the computer as a mechanical brain and in 1949 developed the analogy in his 1949 popular book, Giant Brains. He opens the book with the basic construct of the analogy: Recently, there has been a good deal of news about strange giant machines that can handle information with vast speed and skill. The calculate and the reason These machines are similar to what a brain would be if it were made of hardware and wire instead of flesh and nerves (Berkeley, 1949: 1). Where Berkeley s brain analogy focused on the new technology s ability to think and reason much like a human, others within the industry developed a different analogy. Most importantly, the three societies also commissioned committees to generate reports about the computer and its potential use in the insurance industry. In 1952, the Society of Actuaries Committee on New Recording Means and Computing Devices issued a major and influential report, and both IASA and LOMA hosted panels at their conferences about potential use of computers and in 1953 co-sponsored an Electronics 17

20 Symposium focused exclusively on this topic. Unlike Berkeley s work, these reports characterized the computer as an electronic version of existing technology, often using the analogy information processing machine or data processing machine to describe the new technology. The 1952 Society of Actuaries report stated: These new machines have been called computers because they were developed primarily for mathematical work. It is a mistake, however, to think of them today as purely computing machines capable only of a large amount of arithmetic. In recent years, some very important improvements have converted them into machines capable of a wide variety of operations. Nowadays we must think of them as information processing machines with computing representing just a part of their total capabilities. (Davis, Barber, Finelli, & Klem, 1952: 5). While qualitatively different, both analogies tried to make sense of the new technology by invoking aspects of the pre-existing schema. We measure the extent of assimilation by the degree to which the categories and relations of the analogy are the same as those of the existing schema. If many categories and relations are the same, the analogy is more assimilated within the existing schema. To assess the degree of assimilation, we identified the categories and relations for the preexisting schema, the brain, and the machine analogies using the aforementioned coding procedure. This generated 334, 322, and 152 unique categories and 103, 83, 77 unique relations for the pre-existing schema, machine analogy and brain analogy respectively. Figure 1 shows that of the total categories in the pre-existing schema, almost 35% of them were shared with the computer as machine analogy vs. 12% with the computer as brain analogy. Likewise, Figure 1 shows that of the total relations in the pre-existing schema, almost 50% of them were shared with the computer as machine analogy vs. 26% with the computer as brain analogy. ********************** Insert Figure 1 About Here ********************** However, simply sharing categories and relations is not a complete measure of assimilation. It is possible that the shared categories and relations are not part of the central schema. In these cases, the 18

21 shared categories and relations may be with parts of the existing schema that are not very meaningful. Table 1 addresses this issue by comparing which categories and relations associated with the machine or brain analogy are shared with the central categories and relations of the pre-existing schema. We define central in terms of the high frequency categories and relations within the pre-existing schema. Table 1 and Figure 1 show that the machine analogy shares many of the core categories and relations of the existing schema. In contrast, the brain analogy does not (see also Table 1 and Figure 1). Hence, the machine analogy is well assimilated, but the brain analogy is not. ********************** Insert Table 1 About Here ********************** While the texts on the exiting technologies invoked a wide variety of categories and relations, Table 1 shows a concentration around categories such as clerks, punch cards, policy, tabulators and machines. Clerks interacted with the various kinds of machines according to the listed relations to manipulate punch cards to complete a business process. The relations between these categories centered on creating or inputting information on cards through the action of a punch, making computations related to the business process ( tabulate, calculate ), processing the punch cards ( sort, file, merge, matching ) and writing out the output ( print, list, post ). Consistent with Bebe s (1947) passage, generally the main unit of information was the punch card the relations were between the machine and the punch card and not the data found on it. Yet check and verify indicate the need for verification of the information actually punched on the cards. Verification involved both mechanical and manual processes and our analysis revealed that clerks often validated the results from one machine before passing it along to a new machine. Reproduction or the creation of new punch cards with the same information was also frequently required because cards were used for different purposes. Some estimated that life insurance firms had as many as 10 different versions of the same card, making verification and consistency between the cards hard to manage. 19

22 In particular, Table 1 shows the strong similarities between the pre-existing schema and the machine analogy (vs. brain analogy). For the machine analogy, the core categories still focus on machines and punch cards. Relationally, the machine analogy focused on similar kinds of transactionoriented actions: entering in information ( read, punch in ), computing ( compute, add, ), processing information ( sort, ) and writing results ( write ). In contrast, the brain analogy uses categories and relations that generally are not shared with either the pre-existing schema or the computer as a machine analogy. While there are similarities, such as data, computer, and add, the brain analogy concentrates on categories and relations associated with decision-making and thinking. New categories, such as problem and operation, and relations, such as look-up, solve, store, remember, and think, reflect the brain-like processes associated with decision-making. The insurance industry quickly converged on the machine analogy and so largely characterized the computer as such. Of the 26 texts that discussed the computer from , 21 adhered to the computer as a machine analogy. Many rejected the brain analogy outright. E.F. Cooley of Prudential Life, argued: I might use the term "giant brains" to tie in my subject with the more or less popular literature on this subject. But I hate to use that term since there are false implications in it, implications that these machines can think, reason and arrive at logical conclusions, and I don't agree that this is true (Cooley, 1953: 355). Computer manufacturers also emphasized the machine analogy during this time period. They presented 32 texts, ranging from documented question and answer periods in society meetings to exhibits to prepared remarks. Those that addressed the computer used the machine language. IBM even adopted the label electronic data processing machine to describe the computer, which quickly became common after the commercial release of computers in The large-scale adoption of the machine analogy helped develop the initial conceptual schema of the computer. The high percent of shared categories and relations between thinking of the computer as a 20

23 machine and the existing schema indicate that that the machine analogy was assimilated within the existing schema. Interestingly, the categories and relations associated with the brain analogy represented some of the more novel features of the new computer technology and help differentiate it from existing technologies. Berkeley (1949) made this differentiation explicit in his Great Brain book, where he used a table to show how the computer differs in thinking ability from previous technologies. Consequently, these categories and relations related to the brain analogy were not completely ignored, but faded into the background as the insurance industry focused on the more machine-like qualities and functions of the computer. Therefore, much of the novelty of the new technology gets less recognition and development because of the strong preference for the machine analogy. In 1954, the computer was commercially released and the insurance industry shifted from discussing what it could do to developing uses for it. Yet, thinking of the computer in terms of a machine maintained an influence on how insurance firms initially used the computer. A series of surveys conducted by the Controllership Foundation in provides support. It reveals that 68% of firms purchasing computers purchased IBM s 650 a smaller computer that more closely resembled a tabulating machine and could be wheeled in to replace a tabulating machine. The surveys also reveal that rather than create new business processes, life insurance firms converted existing office applications, such as premium billing and accounting, over to the computer (this did not vary by kind of machine). Yates (2005) notes that this incremental machine-like use of the computer persisted throughout the 1950s and into the 1960s. Overall, we find that the initial development of a new schema involves the use of analogies. Why are analogies related to schema development? One reason is that analogies pervade human thought. Studies find that making what is new seem familiar is a fundamental aspect of human intelligence that relies on analogical reasoning (Hargadon & Sutton, 1997). Another reason relates to our context. 21