Learning in Context: Extensively Computerized Work Groups as
Communities-of-Practice

Joey F. George
Florida State University

Suzanne Iacono
Boston University

Rob Kling
Indiana University, Bloomington

Accounting, Management and Information Technology. 5(3/4)(1995), 185-202.

---

Acknowledgment: The authors would like to thank all of the many people who have read and commented on earlier versions of this paper and the many people who participated in the Desktop Computing Project as respondents and as research colleagues. Partial funding for the Desktop Computing Project came from National Science Foundation grant #IRI-87-09613. The Government has certain rights in this material. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. We would also like to thank the editor and reviewers. Special thanks to Karen George. Copies of the research instrument are available from the authors.

---

Abstract

As computing becomes increasingly integral to organizational life, how work groups learn to successfully use computing becomes a critical issue. The current focus in the management information systems literature is on individual training and teaching methods. The context in which people and groups learn is overlooked in these studies. But work groups provide different types of learning environments, some which encourage learning while others discourage it. Three characteristics of work group environments help explain why learning varies: differential valuation of work roles in organizations (clerical versus professional work groups); differential participation in legitimate peripheral learning (through the presence of local expertise and time to interact and learn); and differential levels of participation in noncanonical communities-of-practice (especially through grass roots computing implementations). Two contrasting case studies illustrate how these concepts result in different learning environments. Professional work groups are more highly valued in organizations and members are given more autonomy to participate in legitimate peripheral learning and emerging communities-of-practice, while clerical groups are less valued, isolated from other practitioners and more rigorously held to canonical work practices. Participation in computing implementations also provides opportunities for learning that are missing in groups that have computing forced on them by management.

Key Words: organizational learning, work groups, desktop computing, clerical work, social practices

---

Introduction

Computing is a ubiquitous and in many cases essential part of modern organizational life in North America. The latest U.S. Census Bureau numbers indicate that one-third of American workers used a computer at work in 1989, up from one-quarter in 1984 (Kominski, 1991). Although Canada lags behind the U.S., Japan, and Western Europe in the diffusion of computing in organizations, increasingly Canadian organizations are implementing a variety of computer-based information systems (Grayson, 1993). As computing becomes a salient part of a large number of North American work settings, individuals, work groups, and organizational subunits must learn how to leverage computing to successfully perform their work. To date, most of the literature which focuses on learning about computing is highly individualistic and context-free. The primary objective of most of these studies is to better understand training methods for specific information systems and application packages (e.g., Davis and Bostrom, 1993; Davis and Davis, 1990). But formal training, typically individual instruction distant from workers' desks and every day worklife, is only one mechanism by which computing skills and knowledge are acquired and distributed in organizations. We argue, instead, that learning is a group-level phenomenon which is an essential part of daily work practices. People learn about computing during the course of adopting, altering, and using it in their work.

In our studies of extensively computerized work groups, defined as groups of people who provide a bundle of products or services, report to a common supervisor, and have at least one local microcomputer or terminal for every two group members, we found that how groups learn to use computing differed across work groups. These differences were neither random nor unique. Instead, we found systematic differences related to the occupational status of the work group and the manner in which their computer systems were implemented. In an earlier paper (Lepore et al., 1989), we linked these factors to work group productivity and satisfaction with computing. Here we investigate the question: Why should learning how to use and leverage computing in work group settings be influenced by occupational status and computing implementation practices?

The rest of the paper is organized as follows: First, we examine context-free learning, as derived from the end user training literature, and its limitations. This is followed by a discussion of learning-in-context as derived from the organizational learning literature. We then consider three characteristics of work group environments that influence learning-in-context. Next, we discuss our study of computing in white collar work groups, the Desktop Computing Project. Then, we contrast brief case histories of two of our work groups. The first case history focuses on a group of professionals who engaged in noncanonical work practices and implemented computing themselves, a process we refer to as a grass roots implementation strategy. Our second case history focuses on a group of clerical workers who had a new information system thrust on them by management when they wanted to cut costs. We refer to the latter type of implementation strategy as top down. While the case histories are derived from actual work groups, they are distinct archetypes illustrating polar differences in how in situ groups come to fit computing into their work lives. Finally, we conclude that due to the valuation of different work roles in organizations, the degree of legitimate peripheral participation in learning, and the degree of innovation permitted during the implementation process, communities-of-practice and learning-in-context may or may not emerge.

Context-free versus in-context learning

In the end user computing literature, learning about computing is characterized as a highly individualized, context-free activity. The self-described goal of this literature is to find ways to improve end user training, both in terms of content and delivery, to ``...ensure that users acquire the necessary EUC [end user computing] skills...in the most efficient and effective ways possible (Sein and Bostrom, 1990)." Based on a series of experiments that investigated end user computing training and individual learning style, Sein, Bostrom and colleagues recommend that training be tailored to match differences in individual learning modes (Bostrom, Olfman, and Sein, 1990; Sein and Bostrom, 1989; Sein and Bostrom, 1990). In general, laboratory studies of computer training seek to find ways to improve the effectiveness of teaching computer skills. One experimental study showed lectures to be better than self-study and level of education and individual learning style to be critical (Davis and Davis, 1990); another study found that experts in both task domain and specific software packages solved problems differently than people who were experts in either one alone or who were novices in both (Mackay and Lamb, 1991); still another found that subjects using Macintosh-like interfaces for training outperformed subjects using command-line-based interfaces (Davis and Bostrom, 1993). One of the few studies that has investigated the effects of an end user training program in situ found evidence of time-savings and increased productivity resulting from systems developed by end users after they had participated in a training program (Cronan and Douglas, 1990). Different training methods were not compared in this particular study, however; the point was to demonstrate that users could be trained to develop effective systems.

The benefit of doing the types of experimental studies described above is better formal training programs in organizations. Ostensibly, better training programs will lead to more skilled, knowledgeable users and more productive organizations. However, results of on-site surveys about end user training indicate that users depend less on formal training programs than they do on informal sources found in the users' immediate work environments. For example, Lee (1986), in a study of 311 end users in thirteen units of twelve organizations, found that 89.4% of the respondents reported consulting colleagues regarding microcomputer usage, compared to the 48% who consulted IS staff. Nelson and Cheney (1987), in a study of 100 management end users in twenty organizations, report that "self-training" is the predominant mode of computer-related training among their respondents. Eason (1988) reports that users are more likely to turn to local experts for help than to the organization's technical support staff because the local expert understands both the users' primary work and the computer systems they use. Based on the results of these field studies, one may conclude that, in general, formal training programs are less well used than expected; instead, end users are more likely to train themselves or seek help in their local environments when necessary.

In all of these studies, especially in the laboratory studies that focus on training techniques, the emphasis is on programmatic and individual skill acquisition. But work is rarely carried out by individuals in isolation and learning is more effective when it occurs just-in-time. New theories emphasize the interdependent and collaborative nature of work, requiring an understanding of practical learning by groups and organizations. This shift in emphasis incorporates people's preferences for local help with computing problems on an as-needed basis rather than scheduled formal training or dependence on outside experts. The current focus of the EUC literature on determining the best ways to teach individuals abstract computing skills in special-purpose but context-isolated teaching facilities, as formal training programs do, is really missing the point. A better understanding of learning environments, including the working conditions and non-canonical practices of co-workers, is what is important, not the abstract skills themselves. Brown and Duguid (1991) argue that the emphasis on teaching should shift to an emphasis on learning in context, that:

...learners can in one way or another be seen to construct their understanding out of a wide range of materials that include ambient social and physical circumstances and the histories and social relations of the people involved....What is learned is profoundly connected to the conditions in which it is learned (pp. 47-48).

Perhaps the mismatch between the fiction of teaching abstract knowledge to individuals and the social reality of learning in the work place can explain why the popular business press is full of articles about low levels of computer literacy and the need for better training for workers (Davidson, 1994; Riendeau, 1994; Sullivan, 1994; "Why Johnny Can't Compute," CIO, 1994). Despite the billions spent by industry and government on desktop computers and formal training programs, many users still lack the know-how to successfully leverage computing in their work. The traditional literature on teaching methods and individualized learning seems inadequate to help us understand how workers are able to learn about computing, not as a series of isolated abstract skills, but as an integral part of their actual work practices.

Why Does Learning Differ Across Work Groups?

Organizational learning theorists have long recognized that organizations are not monolithic and that learning varies from group to group (e.g., Attewell, 1991; Brown & Duguid, 1991; Dodgson, 1993; Levitt & March, 1988). Within organizations, multiple learning processes can take place at the same time (Dodgson, 1993). Learning may not occur at the same depth or even at all within some organizational subunits. Levitt and March (1988) refer to this aspect of organizational learning as the "ecologies of learning" that exist in organizations:

Organizations are collections of subunits learning in an environment that consists largely of other collections of learning subunits....a routine may produce different outcomes at different times, or different routines may produce the same outcome at different times (p. 331).

But what causes ecologies of learning to co-exist in organizations? We have noted that occupational status and implementation practices matter (Lepore et al., 1989), although we lack a comprehensive theoretical explanation for understanding why. By integrating these findings with Brown & Duguid's (1991) unified view of working, learning and innovating, we gain an understanding of how communities-of-practice emerge in extensively computerized work groups. Three characteristics of the work group environments influence learning in context: 1) the differential valuation and misrecognition (Bourdieu, 1977) by management of work roles in clerical and professional work groups; 2.) the degree of participation in legitimate peripheral learning permitted under different working conditions (Lave and Wenger, 1991); and 3) opportunities for participation in innovative, noncanonical communities-of-practice such as grass roots computing implementations.

The first characteristic focuses on the degree to which certain work roles are valued or misrecognized. In organizations, clerical jobs have traditionally been undervalued (Iacono & Kling, 1987). It is almost axiomatic in the scholarly literature on the sociology of work that clerical workers have less autonomy and control in the work place than do professional workers due to their position in the organizational hierarchy and the lower valuation of their work (Kanter, 1977; Ritzer, 1977). On the other hand, professional workers have more freedom to act, higher status, and make more highly valued contributions to the organization.

Related to the different valuation of work roles is the misrecognition of work practices. Many work practices are "misrecognized" (Bourdieu, 1977) by outsiders who do not understand the complex, improvisational skills needed to handle changing working conditions. Due to the misreading of work practices, training programs are often overly simplistic and superficial, or do not exist at all. Brown & Duguid (1991:49) argue that management observers in organizations tend to assume that most of their workers engage in the canonical practices of their formal job descriptions and that:

...complex tasks can be successfully mapped onto a set of simple, Tayloristic, canonical steps that can be followed without need of significant understanding or insight (and thus without need of significant investment in training or skilled technicians) (Brown & Duguid, 1991:42.)

Misrecognition is particularly problematic for low valued work which is assumed to be uncomplicated and straightforward. Evidence of misrecognition is available in the myriad studies of the computerization of clerical groups. Fearfull (1992) writes about British organizations where the "junior system" of training clerical workers was eliminated after computerization, resulting in a new class of clerks who knew how to operate computers but who had little understanding of the clerical jobs they were performing. To outside observers, clerical jobs in these organizations appear to be deskilled (Braverman, 1974) due to the introduction of computers. Although the jobs of the new clerks were somewhat impoverished, the jobs of the clerks who had worked at these organizations before and after computerization were actually expanded. This case illustrates how easy it is for outside observers to misread work practices (Fearfull's insights came from having worked as a clerk for over nine years) and provides one explanation for the common perception that computerization deskills clerical jobs, even though other observers have found both enskilling and deskilling (Attewell, 1987; Gallie, 1991).

Other studies of the computerization of clerical work have found that negative outcomes are often not the direct result of computerization. Instead they result from changes in seating patterns, inadequate training, excessive expectations about productivity, or a bad fit between actual work practices and the new computing system (Gregory & Nussbaum, 1982; Iacono & Kling, 1987; Kraut et al., 1989; Zuboff, 1988), all dimensions of clerical work that are misrecognized by those who manage the implementation.

The second characteristic is based on Lave and Wenger's (1991) "legitimate peripheral participation" theory of learning. According to this theory, learners do not learn abstract knowledge. Instead, they learn how to become a member of a community, whether that community is an office, a shop floor, a high school or a playground. Learners participate at the periphery of community activity and thereby begin to assimilate and adopt the community's viewpoint and language. Learning occurs through practice. As Brown and Duguid point out, "The central issue in learning is becoming a practitioner not learning about practice (p. 48, original emphasis)."

According to Lave and Wenger (1991), learners must have legitimate access to the peripheries of communication, e.g., war stories shared at the water cooler or in the hallways, tacit knowledge passed from more experienced people, actual observation of work practices or participation in work-in-progress. At the periphery, they learn what constitutes acceptable practices and how to communicate appropriately. Eventually, they become practitioners and insiders, able to pass on tacit knowledge to other peripheral learners. Nowhere is the difference between practitioner and non-practitioner more obvious than in talk about computing. In our own research, we have found that the language surrounding certain computer-based information systems, such as MRP systems (Kling & Iacono, 1984), clearly differentiates insiders from outsiders. These cleavages also act as barriers. Only those who have the legitimate right to become insiders can participate at the boundary of the community and become a practitioner. While professionals can engage in water cooler discussions or in electronic bulletin boards, clerical groups are often isolated from these experiences, either because they are physically located in separate rooms, basements or back offices, or because time spent at the water cooler is not considered part of their jobs.

The third characteristic that differentiates work groups is participation in innovative communities-of-practice (Brown & Duguid, 1991). Actual work practices comprise many unrecognized, noncanonical activities. For example, work group boundaries may be fluid, emergent, and interpenetrative as people work, learn, and innovate together. "Maverick" communities may provide alternative models of organizational action and engage in naturally occurring experiments. In the course of innovating, group members may break rules and ignore traditional expectations. While people at the core of the organization may view these practices as counterproductive if they learn about them, many professional groups have substantial power to act independently and to influence the trajectory of change in their organizations.

We have observed such noncanonical activities during the course of the Desktop Project. Many computing implementations were not controlled by IS personnel (Kling and Iacono, 1989). Like those who have studied the phenomenon of end-user computing (Rockart & Flannery, 1983), we observed the use of applications that had been completely developed by the work group members themselves, with little or no involvement from anyone outside the group. However, we also observed cases where some, or all, of the equipment and software had been purchased, configured and deployed entirely by work group members themselves. Often, these actions were hidden from the central information systems subunit or other organizational authorities to the point where entire sub-cultures existed uncontrolled by central IS. Despite the prevalence of these emergent, noncanonical practices, the literature on user involvement, even that of Mumford (1981) and other proponents of socio-technical design (Bjorn-Andersen & Hedberg, 1977; Pava, 1983), assumes that computer implementations are planned and controlled by the organization's information systems subunit. The end-user computing literature only recognizes control over application development. Nowhere is there recognition of the conflicts, hidden work practices and other non-canonical activities that many work groups engage in to purchase their preferred equipment and configure, implement and effectively use it on their own. We recognized these activities as critical social practices related to the computerization of work groups and have distinguished between grass roots and top down computing implementations.

The Desktop Computing Project

The Desktop Computing Project is a longitudinal study of the technical and social dimensions of desktop computing in work groups in several organizations in Southern California and Arizona. Desktop computing is defined as multi-functional computer services that are located on or near most work group members' desks (Kling and Iacono, 1989; Lepore, et al., 1989; George, Kling and Iacono, 1990; Kling, Iacono, and George, 1990). The analytic focus of desktop computing is much broader than the end user computing concept in that it includes all the different forms of computing services that have emerged over the years and which are now available at the desktop. For example, desktop computing includes traditional centralized computing services available through terminals or terminal emulation, end user computing or application development, standalone personal computing packages, and new communication packages for computer supported collaborative work (e.g., electronic mail, bulletin boards and computer conferencing). Many aspects of desktop computing are made possible through the increased availability and functionality of computer networks and distributed applications, as in client/server architectures.

The Desktop Project was begun in 1985 by Kling and Iacono with a pilot study, and ended in 1992 with final follow-up data collection. The primary objective of the study was to investigate changes in the social dimensions of work in extensively computerized groups. Important dimensions included work group participation in decisions about computing, extent of computing use, variety of tasks performed with computing, changes in the quality of work life, and how work groups learned about the computing that they used each day. Consequently, the main unit of analysis was the work group.

The Sample and Data Collection

For work groups to be included in the study, they had to fit our criterion of extensive computerization -- one PC or terminal for every two workers. Our rational for this criterion was simple -- if computerization causes work group transformations, we would be most likely to observe those changes in the groups that were most extensively computerized (Kling and Iacono, 1989). We identified potential organizations and the extensively computerized work groups within them through existing contacts, personal solicitation at local computing professional meetings, newspaper accounts of computerization, and systematic phone calls to the MIS departments of the largest companies in Orange County, California, and the surrounding four counties (available on public lists) and small professional organizations, such as law and architecture firms (available from the yellow pages).

Only private sector, for-profit organizations were solicited to participate, with groups in California being the first to be identified. Groups in Arizona were chosen later with an eye toward supplementing and complementing the California groups. For example, two of the Arizona groups were part of one of the organizations in California, AIRCRAFT. Thirty-eight work groups in seven southern California organizations and three Arizona organizations were selected. Twenty-five of the work groups are in three large organizations (more than 2500 employees) -- INSURE, AIRCRAFT, and COAST PHARMACEUTICALS. The remaining 13 work groups are in five medium-sized (250 to 2500 employees) and two small organizations (fewer than 250 employees). Data were collected in 1988, 1989 and 1990 through a combination of intensive case studies and longitudinal survey methods.

Since little was known about extensive desktop computerization in the mid-1980s, several preliminary case studies were conducted as part of a pilot project to understand the variety of social and technical arrangements which support extensively computerized work settings and to learn what the critical issues were for users. Based on findings from the pilot study, a questionnaire was developed over the course of two years of intensive pre-testing and revision. It includes approximately 200 closed-response questions covering topics such as patterns of individual computer use and job characteristics, patterns of computer use and computer practices in the work group, and changes in worklife that the respondent attributed to desktop computerization. All members of the thirty-eight selected work groups were administered the questionnaire once each year (1988, 1989, and 1990). The specific questionnaire data used in this paper are from 357 respondents (representing an 86% response rate) who completed the instrument in 1988.

In addition to the survey methods, intensive case studies were conducted in selected work groups. While surveys offer broad comparisons, intensive and comparative case analysis allows for close examination of social choices and their associated practices in several selected work groups. Seventy intensive interviews with representative members of the selected work groups were conducted. Interviews were semi-structured, based on a specific outline of questions to be asked (for comparability across work groups) but which allowed for open-ended responses from the interviewees and targeted probes from the researchers. Interviews lasted from one to three hours.

Our sample is not random; instead, we employed an alternative sampling scheme called theoretical sampling whereby one samples along theoretically interesting dimensions. Theoretical sampling is closer to the ANOVA designs of experimental psychology and the grounded theory approach advocated by Glaser and Strauss (1967). It is especially attractive when sample sizes are modest. However, for it to work, the researcher must have some kind of theoretical argument about which characteristics of work groups or information technology are likely to explain major differences in outcomes (Kling, 1991). Consequently our research was designed in stages -- each stage leveraging and informing the next. The preliminary case studies informed both questionnaire design and the sampling schema for the work groups to be included in the next stage of the research.

The Sampling Schema

Based on our pilot study, we developed a sampling schema along dimensions where salient differences in the practices of computing and outcomes were likely to be found. Our schema included: different occupational mixes (predominantly professional, predominantly clerical, or mixed), and implementation strategy (either top down or grass roots).⁽¹⁾

In building our sample, we sought to have an even distribution of clerical and professional work groups. We employed a two-stage categorization process to determine these groups. In the first stage, an initial interview with the work group manager was used to determine whether a work group was clerical or professional. In the second stage, questionnaire responses to a job classification question were checked and categorizations changed when necessary. Our clerical groups included several order entry groups, accounting groups, and word processing groups. Most members of such groups were clerks with one or two supervisors. Our professional groups consisted of actuaries, financial analysts, engineers, and systems analysts. A typical professional group consisted mainly of professionals with a single secretary or several clerks. Several of our work groups, selected because they fit other schema requirements, were difficult to categorize in that they had equal numbers of clerical and professionals workers. We labeled them mixed work groups. For example, a guaranteed annuity contract administration group had equal numbers of underwriters, computer application analysts, and clerical processors. Work groups were coded as "1" for mostly clerical,"2" for mixed clerical and professional, and "3" for mostly professional.

We also sought an even distribution of work groups with top down and grass roots implementation strategies. A similar two-stage process was employed. Based on the initial interview with the work group manager, groups were selected for the study. During the second stage, responses to an implementation strategy question on the survey were checked and categorizations adjusted where necessary. The survey question asked level of agreement with the following statement: "Some people from our work group had to fight to get the desktop computing equipment we now have." Since a seven point scale was used, responses above 4.0 indicated "grass roots' implementations while scores below 4.0 indicated top down implementations. Implementation strategy was coded as either a "1" for grass roots or a "2" for top down.

Study Procedures

The typical procedure for conducting the study with each work group was as follows. Once the work group had agreed to participate, one of the authors visited the site for an initial tour and orientation. The researcher then set up some initial interviews with representative group members and established procedures for the distribution of the questionnaire. The researcher manually distributed questionnaires to each member of each work group and returned to the sites to personally retrieve the questionnaires, which helped produce relatively high response rates. After results were compiled, the researchers either returned to the sites for feedback meetings with management and group members, or they provided written feedback. The researchers also scheduled follow-up interviews with some group members at this point. For the second and third years of the project, the last three steps were repeated: the questionnaire was distributed and collected, feedback was provided, and follow-up interviews were conducted. In parallel with the second and third year surveys, groups were selected for intensive case studies based on our theoretical sampling scheme and by pinpointing for further study groups that were particularly interesting, e.g., those showing evidence of significant transformations of work from one year to the next.

Survey Results

In our questionnaire, we asked several questions about the work group learning environment. Individual responses to the four items in Table 1 were averaged to obtain group means for each of the 38 work groups in our sample. The first and fourth items in the scale focus on the existence of computing expertise in the group. While the first item asks for an evaluation of self expertise, the fourth item asks for an evaluation of the degree of expertise (help) in the environment (from others) when needed. The second and third items focus on whether the work group environment engenders learning about computing. Item two focuses on the availability of formal training while item three focuses on the amount of time tolerated for engaging in self-training. All four items used a 7-point scale that varied from "NO!NO!NO!" for "1" to "YES!YES!YES!" for "7," with "4" being neutral. The four mean scores were combined into a single scale, ranging from 4 to 28, that measures how work group members evaluate their own learning environments. The higher the number, the better the group perception of their learning environment. To form the scale, items 2 and 4 in Table 1 were reverse coded at the individual level before group means were computed. Group responses for all four items were then added for the overall scale, which had a reliability score (Cronbach's alpha) of .71. According to Nunnally (1978), an alpha of over .7 is adequate for social science sales in the early stages of development.

Table 1: Questionnaire items used in scale of skill acquisition

The means and standard deviations for implementation strategy and work group learning environment, by occupational composition of the work groups, is shown in Table 2. The table also contains descriptive statistics for occupational composition and work group learning environment by implementation strategy. Table 3 contains a Pearson correlation matrix for these three variables.

Table 2: Descriptive statistics for key desktop variables.

* means first, standard deviations in parentheses

Table 3: Pearson correlation matrix for key desktop variables.

Two of the correlations are statistically significant at the p < .05 level, with the third nearly so. One implication of these results is that occupational mix for a work group is strongly associated with how information systems are implemented in the work group: professional work groups are associated with grass roots implementations, while clerical groups are associated with top down implementations, and mixed occupation work groups are squarely in the middle. Also, the work group learning environment is positively associated with occupational mix and negatively associated with implementation strategy: the ability of a work group to engender learning is associated with professional groups and grass roots implementations.

Why should the work group learning environment be associated with professional work groups and with grass roots implementation? If learning takes place in situ, with understanding emerging from the working conditions and the interactions of co-workers, why should communities of practice emerge in some groups and not in others? Our intensive case studies provide some insights, as detailed in the next section. Our purpose is to understand the social context in which computing practices arise, focusing specifically on the interplay among the three characteristics of learning environments mentioned previously.

Two Case Studies of Learning in Extensively Computerized Work Groups

The research literature rarely reports detailed case data about social choices and their associated practices over extended periods of time. Some case studies have focused on end user participation during the systems development period (Newman & Noble, 1990). Others have focused on participative design episodes and their long-term impacts (Clement & Van den Besselaar, 1993). Our approach is to illustrate important theoretical relationships by focusing on two contrasting and archetypical cases. We first examine a case that is typical of the professional work groups we studied. We then examine a case that is typical for clerical groups.

Learning in a Professional Work Group
In 1988-1989, this work group consisted of eight people in a financial planning group in the central finance department of a large manufacturing organization. This work group had three primary areas of responsibility: manpower planning; capital analysis; and overhead estimation for use in contract negotiation. As part of central finance, the group provided these services for the entire plant. To do those parts of their work that can be done using computing, they predominantly used stand-alone personal computers, which they reported using almost all of the time. (At the time data were collected, local area networks were starting to become more widely available in organizations such as this one, but diffusion was still in its early stages.)

At that time, the group consisted of five college educated financial analysts, two clerical workers, and the group's manager. The manager, both clerical workers, and four of the five financial analysts were women. On the average, in 1988, work group members spent 22.5 hours per week at their workstations (the number of hours per week was reported at 21.2 in 1990). However, the amount of time a given individual worked at the computer depended on the type of work being done. For example, the analyst in charge of manpower planning estimated she spent about 15% of her work time at the computer. Most of this time was spent working with the mainframe-based manpower planning system, although some manpower planning tasks, such as manpower tracking, were done using a standalone PC software package. The analyst in charge of capital analysis, however, estimated that she spent 90% of her time using software packages. In fact, so much of her work involved a personal computer, one of their machines was essentially dedicated to her, despite the fact that analysts had to share computing resources. The group's performance was evaluated on their ability to complete assignments on time and on the quality of their analyses, both in terms of content and appearance. As the group was required to provide information for senior management, sometimes hurriedly, their ability to perform well under time pressure was highly valued. As their manager said, they "...value[d] time, sometimes over accuracy."

Were it not for the work group's manager, there would not have been any microcomputers for the work group to use. The manager fought with the organization's central finance department to purchase microcomputers for the group's use. When she first became manager of the work group, in 1986, she was amazed to find the group doing most of their work by hand with calculators. The work demands on the group exceeded their ability to get the work done on time -- "They weren't able to turn things around." "What if" analyses were very difficult to do with written spreadsheets, and the quality of such analyses depended greatly on the expertise of the previous manager. The current manager brought to the group a knowledge of personal computers and spreadsheet software, and she asked financial management to provide desktop computerization for the group. Management refused, not seeing the necessity of sponsoring local computer access. Refusal was a typical response from the controller of the manufacturing facility at that time. He often referred to people who used computers frequently as "green faces" (color monitors were not widespread in the company at then). The group's manager was determined to bring in personal computers anyway, despite management refusal and despite her own limited capital purchasing authority. She clandestinely obtained computers and spreadsheet software, purchasing the hardware and software without exceeding her authority. She was not able to procure a desktop computer with a single purchase order, but she was able to purchase a discounted portable computer. She bought two portable computers and spreadsheet software in three separate transactions. She gave the portables to her financial analysts to use.

The analysts, working together as a group, learned very quickly how to model their manual work with spreadsheet software. At first, it was just a matter of transferring paper-based spreadsheets to a computer screen, but with use and constant group interaction, the analysts soon learned about other useful features in the spreadsheet software, such as creating simple graphics and being able to perform "what if" analyses with their spreadsheet models. The analysts became dependent on the portables, unwilling to go back to paper and calculators, and the manager was able to use their current work practices as leverage to get financial management to procure two desktop computers during the 1986 fiscal year. Other personal computers were obtained in 1987 and 1988. According to the manager, computerization resulted in improved accuracy, time and labor savings ("Overhead analyses still take time but now can be turned around in a couple of days whereas it took a couple of months before."), and the ability of fewer people to do more things with the same resource levels.

As group members used the spreadsheet software more and more, they began to build more complex models. The analyst in charge of capital analysis built spreadsheet models for all of her work. She and other PC users in the group essentially developed and maintained all of their own computing applications. Eventually, one of the analysts used the spreadsheet software to develop templates, which were distributed to other work groups for uniform entry of data needed by the financial planning group. As group members mastered the spreadsheet software, they asked their manager for other packages to support their work, such as database management and more sophisticated graphics software. As with the spreadsheets, group members enhanced their graphics and database computing skills by learning these new packages on their own.

Learning within the group was continuing even at the time interviews were conducted. For example, one of the analysts was working hard to learn as much as she could about advanced spreadsheet macros and how to incorporate them into her work. Meanwhile, she and the other group members were all attempting to deal with another more serious problem, how to find the data they needed and how to put them in a format the group could use. As was indicated earlier, the financial planning group worked with many different computer systems, including mainframe and minicomputer-based systems. Their analyses required many different types of data from all over the organization. No one in the group, and few people in the organization, knew where to find all of the data the group needed for their work. As one analyst put it, "You have to get it [data] yourself and figure out where it is." In addition, once needed data were found (if they were found), it was not possible to download them into the personal computers. Most of the time, analysts had to input data by hand from computer reports generated somewhere else in the organization. At the end of 1988, the analysts and their manager were beginning to learn how to use Culprit to find some of the data they needed, and they were working with financial management to acquire the data communications capabilities necessary for downloading data to their personal computers.

Financial planning group members considered themselves to be experts at the computing tasks they performed, relying very little on outside experts, if at all, even though the organization had a formal training program, staffed by professional information systems personnel. Information systems staff were seldom asked for support because they were judged to be unresponsive, or they were perceived to not understand the systems and software the group used. In fact, the financial planning group became so expert at using personal computer software to support their work that members of other work groups in the organization often sought them for assistance with their own work as they began using spreadsheet software not yet supported by central IS. Financial planning group members suggested that the organization's formal training programs were probably very good for other groups, but not for them. Even after IS began offering spreadsheet classes, they reported that they were simply too busy to give up the one to five days required to complete the formal training away from their own office area, particularly when they weren't sure how useful the classes would be.

Learning in a Clerical Work Group
In 1988-89, this group, an accounting department in a regional mortgage banking firm, Western Mortgage, consisted of nine clerical bookkeepers who handled traditional accounting functions such as payroll and accounts payable (Jewett and Kling, 1990). In 1988, group members used computing about nine hours per week, on average. By 1990, their computing use had doubled to almost 19 hours per week. While none of the bookkeepers had a college degree, they were known for their accuracy and most had been with the company for a long time. Their performance was evaluated on a daily basis based primarily on the accuracy of the information they entered in their systems.

The banking firm where the group was located underwent some structural changes during the time of the study. Some loan offices were consolidated in 1988 and layoffs occurred in 1991-1992. By 1992, due to a decrease in interest rates, they were anticipating an increase in volume. The clerical staff in the work group had been reduced to five and the company wanted to be able to handle the increases without hiring anyone new. A new information system was implemented by management in 1988 to cut costs. They also wanted to generate more accurate information in a more timely manner. Although the clerks had used personal computing for spreadsheets, word processing, and scheduling, members of the accounting work group had previously prepared most accounting transactions on a manual form, which was then routed to a service bureau for processing. The new system, developed by an outside software vendor, automated such basic accounting functions as general ledger and accounts payable. It ran on a small mainframe computer and was implemented by in-house staff.

The new system changed how members of the accounting work group performed their jobs. Previously, transactions were recorded on manual forms; after the conversion, clerks entered numbers directly into the computer system themselves. Previously, group members moved around the office from their desks to file cabinets and back; after the changeover, they stayed at their desks, consulting computerized databases. Accounting clerks developed some new computing skills but also lost many opportunities for social interaction. Many felt at the time that their jobs had lost status as they had shifted from the role of accounting clerk to data entry clerk.

The transition from the old system to the new was distressing for the group. Members of the work group were not involved in analysis, design or implementation planning. They attended a single vendor training session just before the new system was implemented. Even as this training session was occurring, changes were still being made to the new system. After the system was installed, changes in procedures occurred almost daily. In reaction to the resulting confusion and uncertainty, group members began to consult each other and system manuals, and to phone the vendor, in order to deal with the system. The accounts payable supervisor reported "it takes twice as long to do our work" with the new system.

After a year, one supervisor reported that they were "out of crunch mode" and work with the new system was "getting better." There was, however, no formal training program. When asked whether time for informal, on-the-job training was allowed, the manager reported that he "would hope that the type of people that I hire[d] would be motivated to [learn to] make their jobs easier in that learning more would speed things up and make their work more accurate and more timely." The group developed its own outline of the basics of system operation, a set of canonical procedures that users were expected to follow. Company standards for the amount of time required to enter new transactions were not changed in light of the more time-consuming procedures required by the new system. These standards were based on long experience both within the firm and in standard accounting practices and were therefore not a negotiable issue for this work group.

While some learning took place in this clerical work group, it never went beyond the basic minimum required to perform the canonical procedures required to use the new system. Group members did not participate in the development of their own computing environments or develop any of their own applications as part of system use. Because daily throughput and speed of transaction inputs were tantamount, they never had time to modify any of the existing applications that were part of their new systems.

Differences in Work Group Learning

In the introduction, we asked why the emergence of communities of practice should be related to a group's occupational status and to the manner in which computer systems were implemented. To shed some light on these issues, we turned to the literature on organizational learning and uncovered three characteristics of learning environments: 1) differential valuation and misrecognition of work roles; 2) differential access to legitimate peripheral learning; and 3) opportunities for participation in innovative noncanonical communities-of-practice.

In the professional financial planning group, the manager was a maverick who ignored precedent and broke through conventional boundaries despite the scorn of the financial controller. She had considerable autonomy in implementing new systems in her work group, where highly innovative and experimental computing practices took place. Consequently, the group, comprised of college-educated professionals and a manager with some PC and spreadsheet expertise, soon learned how to model their work on the new systems and to use these systems to provide new capabilities. Group members even developed a spreadsheet-based template for other work groups to use. They also distributed what they had learned to other group members who came to them for help, thus expanding the boundaries of their noncanonical work practices to include other spreadsheet and PC users throughout the organization. What they had learned was highly valued by other professionals in the organization who wanted to become participants in this emergent community-of-practice. There were no outside or a priori expectations about how quickly they could perform their work with the new system. Indeed, as much of what they did was hidden from management, they could take time to learn and innovate as required for their own work.

For the clerical accounting group, learning was minimal and a priori expectations about group productivity were high. It took them a year to learn their systems well enough to satisfy the expectations of their supervisor. The implementation of the new system was motivated by a desire to cut costs and increase information accuracy. The work group supervisor was a novice like other members of the group; they could not learn from her and initial learning from the vendor was off-site. While they shared what limited information they had with each other, informal observation and picking up "know-how" from war stories or through collaboration was limited. Instead, after the new system was installed, clerical workers became increasingly isolated from each other physically (see Kraut, Dumais, and Koch, 1989 for another example of increased physical isolation of clerical workers after implementation). Rather than engage in noncanonical work practices, perhaps leading to innovation and experimentation, they were expected to sit at their computers and focus on their work. The canonical practices associated with clerical computer use were enforced, effectively disallowing them from walking around or interacting with each other. Due to the low valuation of their work and its misrecognition, training was inadequate and performing their work with the new system was difficult. Company standards were not changed to allow for any slack in performance expectations at the time of the implementation, creating additional stress. There was some hope that individuals would take the initiative to learn more to increase their speed, but in the four years of the study there was rarely time. The accounting clerks were passive participants during a top down implementation arranged by management and carried out by outside vendors and MIS. They had no say in the development of their own applications, and as a result, did not experiment or innovate. They learned just enough about computing to use their new system.

In other Desktop Project work groups, we observed similar patterns of work practices and learning about computing as those described above in the financial analyst and accounting groups. For example, a clerical payroll supervisor told us that she had changed the menu screen on her group's computers, and the information systems analysts had become furious. They told her she couldn't change the menu herself. The payroll supervisor thought this was silly because it was easier to change the menu herself than call for outside help. In another case, clerical workers in a university work group used personal computers and shared applications on a local area network. Yet these clerical workers were completely unaware of the underlying nature of the information systems they were using. In fact, many responded to our questionnaire that they used stand alone personal computers, despite the fact that every morning they were greeted with a sign-on screen that told them they were using the department's local area network. In a third case, information systems analysts provided clerical workers with a host of functions for performing their jobs, but the analysts only told the clerical workers about some of the functions that were available. In interviews, the analysts explained how they expected the clerical workers to eventually come to them and ask if it would be possible for the analysts to design and implement additional functions. At that point, the analysts laughed, they would tell the clerks, "You already have that function implemented. In fact, you've had it for some time now."

These episodes from several of our clerical work groups are examples of how the low valuation of clerical jobs and IS insistence on clerks' following canonical work practices around their computing use prevent clerical workers from becoming insiders in communities-of-practice. The question arises, however, as to why clerical group members do not insist on more training or inclusion. One plausible explanation is learned helplessness (Seligman & Garber, 1980). Through constant reminders to remain in their clerical roles, i.e., to follow strict procedures and not innovate, eventually clerical workers become helpless about any computer use that falls outside the confines of its specific purpose in their canonical roles (White, 1992). On the other hand, there is evidence that when conditions allow for clerical groups to make demands, they actually do. Clement (1990) describes a case study where clerical workers initiated an organizational learning process after they had a new system imposed on them by management. Clement's case is based on clerical groups that were part of Canadian national unions.

It is also possible for professionals to have computing systems imposed on them from the outside in a top down manner, as illustrated by Orlikowski's (1992) case study of a consulting firm where Lotus Notes was implemented with little prior explanation. The professionals had little concept of the paradigm of collaboration underlying Notes or of the package's actual functionality and, consequently, it was not used. These examples allow us to separate the impacts of occupational status from implementation strategy. Due to the higher valuation of their work roles, professional groups may retain the right to refuse to use a system they don't understand even when the system is implemented by top management fiat. In our clerical accounting group example, we observed that clerical workers also lacked knowledge about how to use the new system, but they were forced to use it under stressful conditions. Seen through this analytical lens, it is not surprising that many clerical groups report dissatisfaction with their computing systems. Their low valuation in the firm results in misrecognition of their jobs, oversimplifying and deskilling them, and poor mapping of their work practices onto new computer-based information systems.

Conclusions

We began this paper by noting that our study of white collar work groups in several organizations revealed that in situ learning about computing was influenced by the strategy used to implement information systems and by the occupational status of the group. The end user training literature, with its emphasis on teaching individuals abstract skills away from their work environments, provided little insight into why these factors should be associated with how groups learn about computing. An examination of the organizational learning literature focused on groups learning-in-context and uncovered three ways in which the work practices of groups influence learning: 1) differential valuation and misrecognition of work roles; 2) the degree of participation in legitimate peripheral learning; and 3) opportunities for participation in innovative noncanonical communities-of-practice.

An important issue for future research is understanding the ways in which learning about computing are related to work performance. We did not collect explicit performance data in our study, so the best we can do is to speculate about its relationship to learning. If the work groups we studied are representative of work groups in general, then professional work groups appear to be better situated than clerical groups to effectively leverage computing in their work through the emergence of communities of practice. Their participation in grass roots implementations led to increased work group expertise, local control, and self-reliance. For example, the financial planning group portrayed here was able to improve the quality of the work they did for top management, and increase the timeliness and accuracy of their work after group members had implemented their own systems. Learning in the group continued after implementation, with group members working to solve the problems they faced. On the other hand, in the clerical group we profiled, performance worsened after system implementation and then only gradually improved, due in part to the misrecognition of the complexity of the clerical workers' jobs, which resulted in very limited training on the new system, and the ways in which the workers were discouraged from learning from each other. Based on these cases alone, we cannot articulate perfectly the links between learning and performance, but we can emphasize the importance of the relationship between how work groups learn about the computing they use in their jobs and how they perform in those jobs.

Our findings should be regarded with some skepticism, however, since our sampling procedure was not random. However, theoretical sampling and comparative case studies are appropriate for many purposes (Yin, 1989), such as the multi-method exploratory research conducted in the Desktop Computing Project. We have not tried to explain every factor involved in group learning about computing. We instead focused on two factors, occupational status and implementation strategy, both derived from a grounded theoretical approach, which proved critical in explaining why some groups excel in their computer use, innovating, and experimenting with new uses, and others barely manage to fit the system to their work practices. We strongly encourage more research which would explore further the critical relationship between group learning-in-context and the emergence of communities of practice around certain computing applications and systems.

References

Attewell, P. (1987). The deskilling controversy. Work and Occupations, 14(3), 323-346.

Attewell, P. (1992). Technology diffusion and organizational learning: the case of business computing. Organization Science, 3(1), 1-19.

Bjorn-Andersen, N. & Hedberg, B. (1977). Designing information systems in organizational perspective. In P. Nystrom & W. Starbuck (Eds.), Prescriptive Models of Organizations. Amsterdam: North-Holland.

Bourdieu, P. (1977). Outline of a Theory of Practice. Cambridge: Cambridge University Press.

Bostrom, R.P., Olfman, L., & Sein, M.K. (1990). The importance of learning style in end-user training. MIS Quarterly, 14(1), 101-119.

Braverman, H. (1974). Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century. New York: Monthly Review Press.

Brown, J.S. & Duguid, P. (1991). Organizational learning and communities-of-practice: Toward a unified view of working, learning, and innovation. Organization Science, 2(1), 40-57.

Clement, A. (1990). Cooperative support for computer work: a social perspective on the empowering of end users. In CSCW 90 Proceedings, (pp. 223-236).

Clement, A. & Van den Besselaar, P. (1993). A retrospective look at PD Projects. Communications of the ACM, 36(4), 29-37.

Cronan, T.P. & Douglas, D.E. (1990). End-user training and computing effectiveness in public agencies: An empirical study. Journal of Management Information Systems, 6(4), 21-39.

Davidson, H.M. (1994). Training in computer use. The Tax Adviser, 25(4), 244-245.

Davis, S.A. & Bostrom, R.P. (1993). Training end users: An experimental investigation of the roles of the computer interface and training methods. MIS Quarterly, 17(1), 61-85.

Davis, D.L. & Davis, D.F. (1990). The effect of training techniques and personal characteristics on training end users of information systems. Journal of Management Information Systems, 7(2), 93-110.

Dodgson, M. (1993). Organizational learning: a review of some literatures. Organization Studies, 14(3), 375-394.

Eason, K. (1988). Information Technology and Organisational Change. London: Taylor & Francis.

Fearfull, A. (1992). The introduction of information and office technologies: The great divide? Work, Employment & Society, 6(3), 423-442.

Gallie, D. (1991). Patterns of skill change: Up-skilling, de-skilling or the polarization of skills? Work, Employment & Society, 5(3).

George, J. F., Kling, R., & Iacono, S. (1990). The role of training and support in desktop computing. In K. Kaiser & H. Opelland (Eds.), Desktop Information Technology (pp. 237-252). Amsterdam, the Netherlands: Elsevier Science Publishers B.V.

Glaser, B. and Strauss, A. (1967). The Discovery of Grounded Theory. Chicago: Aldine Publishing Co.

Grayson, J.P. (1993). Skill, autonomy, and technological change in Canada. Work and Occupations, 20(1), 23-45.

Gregory, J. & Nussbaum, K. (1982). Race against time: Automation of the office. Office: Technology and People, 1:197-236.

Iacono, S. & Kling, R. (1987). Changing office technologies and transformations of clerical jobs: A historical perspective. In R. Kraut (Ed.), Technology and the Transformation of White Collar Work. Hillsdale, N.J.: Lawrence Erlbaum Associates.

Ives, B. & Olson, M.H. (1984). User involvement and MIS success: A review of research. Management Science, 30(5), 586-603.

Jewett, T., & Kling, R. (1990). The work group manager's role in developing computing infrastructure. In ACM Conference on Office Information Systems, (pp. 69-78).

Kanter, R.M. (1977). Men and Women of the Corporation. New York: Basic Books.

Kling, R. (1978). The impacts of computing on job characteristics of managers, data analysts, and clerks. (Working Paper). Irvine: University of California, Public Policy Research Organization.

Kling, R. (1991). Adapting survey methods to study the social consequences of computerization. In K. Kraemer (Ed.), The Information Systems Research Challenge: Survey Research Methods. Cambridge, MA: Harvard Business School Press.

Kling, R. & Iacono, S. (1984). The control of information systems after implementation. Communications of the ACM, 27(12), 1218-1226.

Kling, R. & Iacono, S. (1989). Desktop computerization and the organization of work. In T. Forester (Eds.), Computers in the Human Context (pp. 335-356). Cambridge, MA: The MIT Press.

Kling, R., Iacono, S., & George, J. F. (1990). Occupational power, patterns of desktop computer use, and quality of worklife. In K. Kaiser & H. Opelland (Eds.), Desktop Information Technology (pp. 63-107). Amsterdam, the Netherlands: Elsevier Science Publishers B.V.

Kominski, R. (1991). Computer Use in the United States: 1989. U.S. Department of Commerce, Bureau of the Census.

Kraut, R., Dumais, S., & Koch, S. (1989). Computerization, productivity, and quality of work-life. Communications of the ACM, 32(2), 220-238.

Lave, J. & Wenger, E. (1991). Situated Learning: Legitimate Peripheral Participation. Cambridge: Cambridge University Press.

Lee, D. M. S. (1986). Usage patterns and sources of assistance for personal computer users. MIS Quarterly, 10(4), 313-325.

LePore, S. J., Kling, R., Iacono, S., & George, J. F. (1989). Desktop computerization and quality of worklife: the role of implementation process and infrastructure. In Tenth International Conference on Information Systems, (pp. 223-235). Boston, MA:

Levitt, B. & March, J.G. (1988). Organizational learning. In W.R. Scott & J. Blake (Eds.), Annual Review of Sociology,14 (pp.319-340).

Mackay, J.M. & Lamb, Jr., C.W. (1991). Training needs of novices and experts with referent experience and task domain knowledge. Information & Management, 20(3), 183-189.

Mumford, E. (1981). Participative systems design: Structure and method. Systems, Objectives, Solutions, 1(1), 5-19.

Nelson, R. R., & Cheney, P. H. (1987). Training end users: an exploratory study. MIS Quarterly, 11(4), 547-559.

Newman, M. & Noble, F. (1990). User involvement as an interaction process: A case study. Information Systems Research, 1(1), 89-113.

Nunnally, J.C. (1978). Psychometric Theory (2nd ed.). New York: McGraw-Hill.

Orlikowski, W. (1992). Learning from Notes: Organizational issues in groupware implementation. Proceedings of the Conference on Computer Supported Collaborative Work, Toronto, Canada.

Pava, C.H.P. (1983). Managing New Office Technology: An Organizational Strategy. New York: The Free Press.

Riendeau, E. (1994). Lack of basic skills can stall system project. National Underwriter Property & Casualty-Risk & Benefits Management, 29(5), 5-6.

Ritzer, G. (1977). Working: Conflict and Change. (2nd ed.) Englewood Cliffs: Prentice Hall.

Rockart, J. F., & Flannery, L. S. (1983). The management of end-user computing. Communications of the ACM, 26(10), 776-784.

Sein, M. & Bostrom, R. (1990). An experimental investigation of the role and nature of mental models in the learning of desktop systems." In K. Kaiser & H. Opelland (Eds.), Desktop Information Technology (pp.253-276. Amsterdam, the Netherlands: Elsevier Science Publishers B.V.

Sein, M. & Bostrom, R. (1989). Individual differences and the training of novice users. Human-Computer Interaction, 4(3), 197-229.

Seligman, M.E.P. & Garber, J. (Eds.) (1980). Human Helplessness: Theory and Application. New York: Academic Press.

Sullivan, D. (1994). How do you train employees to become more computer-literate? American Banker, 159(127), 10A.

White, H.C. (1992). Identity and Control: A Structural Theory of Social Action. Princeton, N.J.: Princeton University Press.

Why Johnny can't compute. (1994). CIO, 8(2), 14-15.

Yin, R. (1989). Case Study Research: Design and Methods. Newbury Park, CA: Sage Publications.

Zuboff, S. (1988). In the Age of the Smart Machine: The Future of Work and Power. New York: Basic Books.

1. ¹ Work groups were also selected according to the primary tasks for which they used computing (text processing, analysis/design, and recordkeeping.) However, these distinctions proved less powerful than the other two and are not included in this analysis.