Introduction

Organizations that are able to create new knowledge, retain that knowledge, and transfer it throughout the organization are more effective than organizations less adept at managing knowledge (Kogut and Zander, 1992; Teece et al., 1997; Argote and Ingram, 2000). The current study focuses on knowledge retention and argues that organizational structure is a major repository for organizational knowledge. The study analyzes how organizational structure provides a mechanism for retaining knowledge in organizations and transmitting it to new members. More specifically, the study examines how roles and routines can enable knowledge to persist through time in organizations, even in the face of individual member turnover.

In the sections that follow, we provide an overview of research on organizational learning and forgetting. We discuss organizational memory and review studies on the effects of turnover on organizational performance. We develop our theoretical predictions about how turnover and organizational structure interact to affect organizational learning and forgetting and describe the method for testing the predictions. Results are presented and interpreted and theoretical and practical implications are described.

Organizational learning and forgetting

Large improvements in performance are often realized as groups and organizations gain experience in production. The performance gains that accrue as a result of increasing experience are often quite large and have important practical implications for organizations. These 'learning curves' have been found to characterize a wide range of activities (Argote and Epple, 1990), in both manufacturing (Rapping, 1965; Hatch and Mowery, 1998; Epple et al., 1991; Benkard, 2000) and service sectors (Pisano et al., 2001; Reagans et al., 2005).

The learning curve literature is a special case of the more general literature on organizational learning (Levitt and March, 1988; Huber, 1991). The learning curve literature examines changes in performance as a function of organizational experience (see Dutton and Thomas, 1984; Yelle, 1979; Argote, 1999, for reviews). Recent reviews of the organizational learning literature conducted at the intraorganizational (Argote and Ophir, 2002), organizational (Schulz, 2002) and interorganizational (Ingram, 2002) levels of analysis concluded that understanding the conditions under which various types of experience improve performance outcomes in organizations is an important issue.

Although the classic learning curve assumes that learning is cumulative and persists indefinitely though time, recent research suggests that organization learning may decay or depreciate. Argote et al. (1990) provided evidence that organizational learning decayed or depreciated: recent output was a more important predictor of current productivity than cumulative output. Benkard (2000) also found that knowledge depreciated in aircraft production. Depreciation has also been found in hotels (Baum and Ingram, 1998; de Holan and Philips, 2004) and fast food franchises (Darr et al., 1995).

Understanding why organizational knowledge depreciates involves understanding where organizational knowledge is stored or retained in the organization's memory. An organization's memory is distributed across various reservoirs, repositories, or retention bins. Many scholars have emphasized the importance of standard operating procedures and routines as a knowledge repository (Cyert and March, 1963; Nelson and Winter, 1982; Levitt and March, 1988; Gersick and Hackman, 1990; Cohen and Bacdayan, 1994; Feldman and Pentland, 2003). Walsh and Ungson (1991) provided a broader conception of organizational memory and theorized that memory is distributed across five 'retention facilities' in the organization: (i) the individuals who work for the organization, (ii) the organization's culture, which includes shared frameworks, (iii) standard operating procedures and practices, (iv) roles and organizational structures, and (v) the physical structure of the workplace.

The current study examines how knowledge embedded in individuals interacts with knowledge embedded in roles and routines. If knowledge were embedded in individuals, their departure would cause knowledge to depreciate. Consistent with this conjecture, Argote (1999) noted that the fasted depreciation rate was found in the setting with the highest turnover: fast food franchises (Darr et al., 1995). On the other hand, if knowledge were embedded in the organization's roles and routines, the departure of individuals should not have much effect on the organization's stock of knowledge.

Turnover and organizational learning and forgetting

Although research on the causes of turnover has accumulated over many decades, research on the consequences of turnover is a more recent phenomenon (Staw, 1980). Only a few empirical studies have examined the effect of turnover on productivity gains in a learning-curve framework. Argote et al. (1995) found that turnover hurt the performance of laboratory groups performing a production-type task, and further that the effect of turnover was more harmful on simple than on complex tasks. In contrast, Argote et al. (1990) did not find an effect of turnover on the productivity of World War II shipyards. The structures of the shipyards may have buffered the organizations from the effects of turnover. All the yards used a standardized design for the ships (Lane, 1951). Routines were developed for training and transmitting knowledge to new workers. Turnover did not seem to matter in the shipyards, where routines and specific roles existed for doing work.

The current study examines whether the effect of turnover on organizational learning depends on how organizations are structured. Several researchers have suggested that the effect of turnover on performance is likely to depend on how work groups are structured (Price, 1977; Mowday et al., 1982). None of these previous studies, however, varied structure or analyzed the effect of different structures.

Several field studies have found a negative effect of turnover on performance. Studies of sports teams have reported that the number of new players in a particular season had a negative effect on the performance of the team that season (Allen et al., 1979; Brown, 1982; Pfeffer and Davis-Blake, 1986). Although there are rules and regulations about how games should be played, the personal relationships among the players and the interactions among them are very important (Jones, 1974). These interactions are often idiosyncratic and cannot be standardized easily. They depend on individual players and change when new players with different styles join the team. Similarly, Glebbeck and Bax (2004) found that high turnover harmed the performance of offices of a temporary employment agency. Staff of temporary agencies may acquire idiosyncratic knowledge about client preferences that is lost when members depart. Further, the researchers found some support for a curvilinear relationship between turnover and performance: the effect of turnover was negative at high levels and positive or neutral at low levels. Along similar lines, Shaw et al. (2005b) found that the relationship between turnover and performance was nonmonotonic, although the relationship in their study was initially negative and then flattened out. The researchers argued that because machines in the concrete pipe plants they studied were customized, skills and abilities needed are complex and less transferable than in standardized plants. Hence when jobs are not standardized, turnover was found to have a negative effect on team performance.

Other studies have found that turnover had little or no effect on organizational performance. Trow (1960) found that turnover was not disruptive when replacements had experience, were at least as able as their predecessors, and the team had previous experience with the same rate of turnover. Similarly, Grusky (1961) found that although large firms experienced succession more often than small firms, large firms experienced less managerial disruption due to succession than did smaller companies. Grusky suggested that the greater use of written rules, specialization, and hierarchies in the larger companies buffered them from potential negative effects of managerial turnover.

Similarly, Shaw et al. (2005a) found that turnover reduced the amount of sales per employee in locations of a restaurant chain but did not affect other indicators of performance. The authors suggested that due to the high turnover the restaurant industry generally experienced, employees' jobs were designed to be interchangeable. Thus, the effect of turnover was very limited.

Thus, when organizations develop procedures to deal with membership change such that knowledge is not lost from the group and is communicated to new members, turnover usually does not have a negative effect on organizational performance (e.g., see Trow, 1960; Grusky, 1961; Hill and Gruner, 1973; Argote et al., 1990; Shaw et al., 2005a). On the other hand, if work cannot be standardized or routines are not available to handle the loss and addition of members, turnover seems to have a negative effect on organizational performance (e.g., see Allen et al., 1979; Brown, 1982; Pfeffer and Davis-Blake, 1986; Shaw et al., 2005b). Similarly, Carley (1992) found in a simulation study that hierarchies were less affected by turnover than teams.

Research predictions

We hypothesize an interaction between turnover and organizational structure:

If turnover occurs, organizations in which the structuring of activities is high will perform better than groups in which the structuring of activities is low. If there is no turnover, there will be no difference in the performance of groups in which the structuring of activities is low and those in which the structuring of activities is high. Although the main effect of turnover is not the primary focus of the study, we expect turnover to have a negative effect on group performance (e.g., Price, 1977). Groups experiencing turnover are disadvantaged relative to those not experiencing turnover because their members have less experience with the task and with each other. We did not expect structure to have a main effect on group performance. Indeed the manipulation of structure was designed to not have performance effects because we did not want performance to be confounded with the structure variable.

Through a detailed analysis of videotapes of groups performing the task, we also examined the processes through which the hypothesized interaction between turnover and structure may occur. McGrath (1986) has argued compellingly for the importance of examining process. We expect that turnover will have a disruptive effect for groups low in structure because some knowledge may be lost or become inaccessible or irrelevant when a member leaves. Low-structure turnover groups may also have to keep reorganizing and may spend considerable time and effort in transferring knowledge to new members.

Method

Participants

A total of 240 students from two universities participated in the study. Both male and female students participated, but any one group consisted of same-gender members. Some groups consisted of members who were paid for participation while other groups consisted of members who participated to fulfill a course requirement. These variables (gender, university, method of payment) are controlled for in the analyses.

Participants were randomly assigned to the different conditions and to groups within conditions. There were 12 groups in each experimental condition for a total of 48 groups. Each experimental condition contained seven female and five male groups. Each group in the no-turnover condition had three members, whereas each group in the turnover condition had three original members plus four replacements. At any point in time, even in the turnover groups, there were only three members actually working on the task.

Task

Three-person groups performed a task that involved folding two origami products – crowns and ships. Variants of this task have been used in previous studies of group performance (Insko et al., 1980). A production task was chosen to simulate the manufacturing process. Pencils, rulers, 8.5" 11" sheets of paper, and scissors were supplied to all group members. Subjects had to cut the 8.5" 11" sheets of paper into squares in order to fold the crowns and ships. The crown required two basic folds and then it had to be 'popped open'. The ship required three basic folds and then some flaps had to be unfolded. The procedures for making the ship were exactly the same as for the crown up to the second fold.

Procedures

At the beginning of the experiment, participants were told that its purpose was to investigate factors affecting work group performance and that they would be working along with some other people in a group for five work sessions on a task that involved folding two paper products. Each work session lasted 6 minutes. The experimenter informed the participants that there was a reward for the most productive group in the experiment. Participants were encouraged to build as many crowns and ships as they could in the time available, but they were given a restriction: as a group, they had to make at least three crowns and three ships during each work period. Participants were informed that crowns were worth two points to the group and ships were worth three.

The experimenter then trained the participants to fold the crowns and ships. All participants in the experiment received the same training. Participants were told that it was important to make good quality products and that incorrectly or sloppily folded products would not count toward the group's score. They were also told that their group would be penalized if they wasted supplies. Each piece of scrap paper would result in one point being deducted from the group's score. Participants were informed that their work sessions would be videotaped and that they would be asked to fill out a short follow-up questionnaire after the work sessions. At the end of the experiment, participants were debriefed and thanked for their cooperation.

Experimental conditions

Participants were in one of four experimental conditions. Each experimental condition was characterized by high or low structuring of activities and turnover or no turnover.

Turnover

Depending on their experimental condition, groups were told at the end of each work period whether or not they were going to experience turnover in the subsequent period. None of the groups knew about this turnover aspect at the beginning of the study. Groups that did not experience turnover worked together for the five work periods without any change in membership. In the turnover condition, at the end of each work period, one member of each group was asked to leave and was replaced by a new individual. Before joining the group, the replacement received exactly the same training as the original group members.

Structuring of activities

In all conditions, group members were trained to fold both products, crowns and ships, before they began working on the task. In the condition where the structuring of activities in the group was low, once training was complete, group members were allowed to allocate the work and organize themselves in any way that they pleased. They were not assigned any roles and the group was not given any routines to do the task. Participants were allowed to communicate with and help one another while they worked on the task. If turnover occurred in these groups, a random draw determined which participant left the group.

In the condition where the structuring of activities was high, group members were asked to assume specialized roles after the training session was completed. The task involved three distinct jobs: cutting and creasing the paper, making the intermediate folds, and finally either popping open the crown or unfolding the flaps of the ship. Participants were asked to specialize as either the 'cutter,' the 'folder,' or the 'builder,' and were given specific routines to organize themselves for the task. They were told how to set up an assembly line to produce the crowns and ships and how to avoid bottlenecks in that assembly line. Participants were allowed to communicate with and help one another while they worked on the task. Participants were not allowed to switch jobs. If turnover occurred in these groups, either the cutter, the folder, or the builder left the group. The experimenter determined ahead of time who would leave the group in order to balance the replacement process. The newcomer received the same training as the original group members and was asked to perform the job of the person he or she was replacing. The other two members continued with the same jobs.

Measures

Performance data

The main dependent variable in the experiment was group performance. The performance score at the end of each work period was calculated by counting the number of good-quality crowns and ships made by the group, multiplying those numbers by the appropriate point allocations for the crowns and ships, and finally subtracting any point penalties incurred for wasting supplies. The total performance score for any group was calculated as the sum of the performance scores in each of the five work periods.

Post-experimental questionnaire

All participants filled out a questionnaire individually at the end of the experiment. They were assured confidentiality of their responses. The questionnaire consisted of manipulation checks as well as questions about participants' perceptions of the task and their groups' performance.

Videotape data

All the groups in the experiment were videotaped as they worked on the task. The actions and comments of the groups were coded to shed light on the processes underlying their performance. Learning curves reflect productivity gains resulting from increased knowledge. An objective for collecting the videotape data was to obtain measures relating to knowledge that may affect group productivity. The coding scheme used to capture these knowledge measures is described in more detail in the results section. The coding scheme was developed deductively on the basis of our theory and then modified inductively on the basis of what actually occurred in the groups.

Results

Manipulation checks

Participants were asked to rate on a 7-point scale how they divided the work among themselves. Analysis of responses to this question revealed a highly significant effect for structuring of activities, F(1,44)=113.47, P<0.0001, in the predicted direction. There were no other significant effects for this item. In addition, participants were asked whether procedures were given to them by the experimenter or were developed by the group members. Analysis of this question revealed a highly significant main effect for structuring of activities, X²(3)=60.87, P<0.0001, as predicted. There were no other significant effects.

Participants also indicated whether their group had experienced turnover. For this question, there was a highly significant main effect for turnover, X² (3)=36.77, P<0.0001. There were no other significant effects. Thus, the manipulations were effective.

Performance

The total performance data were analyzed¹ in an Analysis of Variance (ANOVA) framework with structuring of activities, turnover, and their interaction as the independent variables. There was a highly significant main effect for turnover, F(1,44)=21.19, P<0.0001. As predicted, groups that experienced turnover did not perform as well as those with stable membership. The structure variable was not significant. The interaction predicted between structuring of activities and turnover was marginally significant, F(1,44)=3.18, P<0.08.

In order to correct for unequal error variance across groups in the various conditions, generalized least squares (GLS) regression was performed. The results from the GLS regression were similar to the ANOVA results. There was no main effect for structuring of activities, F(1,44)=0.0049 (NS), but there was a significant main effect for turnover, F(1,44)=5.51, P<0.02, and a marginally significant interaction effect of turnover and work group structure, F(1,44)=3.13, P<0.08.

Figure 1 illustrates the interaction between structuring of activities and turnover. As can be seen from Figure 1, the interaction is as predicted. Scheffé tests revealed that the mean of the low structuring of activities/turnover condition differed significantly from the mean performance scores in the other three conditions, F(3,44)=7.12, P<0.01. As predicted, when groups did not experience turnover, there was no difference in performance between the high and low structuring of activities groups. When turnover occurred, however, groups in which the structuring of activities was low did not perform as well as groups in which the structuring was high.

Figure 1.

Total performance as a function of turnover and structure.

Full figure and legend (11K)

Additional analyses of the total performance scores were performed to control for gender, whether participants were paid or received class credit for participation, and home university. This analysis revealed a marginally significant effect for turnover, F(1,33)=3.44, P<0.07, a significant interaction between turnover and structuring of activities, F(1,33)=4.32, P<0.04, and a marginally significant effect for gender, F(1,33)=3.05, P<0.07. Women performed somewhat better than men on the task. No other effects were significant. An important point to note about this analysis is that the effect of most interest in this study – the interaction between turnover the structuring of activities – became stronger with the inclusion of appropriate control variables in the model.

A more fine-grained analysis of the performance scores was also performed which disaggregated the total score into scores from each time period. Figure 2 depicts the performance scores as a function of time period, turnover and structure. A repeated-measures analysis of variance was performed on the performance scores with two between-subjects factors, structuring of activities and turnover, and one within-subjects factor, periods.

Figure 2.

Performance as a function of time period, turnover and structure.

Full figure and legend (17K)

As expected, there was a significant main effect for turnover, F(1,44)=21.25, P<0.0001, and a marginally significant structuring of activities by turnover interaction, F(1,44)=3.16, P<0.08. The analysis also revealed a highly significant effect for periods, F(4,176)=85.16, P<0.0001, indicating an increase in performance from one period to the next. There was also a significant periods by turnover interaction, F(4,176)=15.16, P<0.0001: performance in the turnover condition increased at a slower rate than the performance in the no-turnover condition. The main effect for structuring of activities, the structure by time periods interaction and the three-way interaction between periods, structure and turnover were not significant.

The performance data were also analyzed separately for each period. There was a significant main effect for structuring of activities only in the first period, F(1,44)=5.46, P<0.02, with the high structuring of activities groups performing better than the low structuring of activities groups. From the second period onwards (i.e., in periods 2, 3, 4 and 5) there was a significant main effect for turnover. The turnover by structuring of activities interaction was significant in period 3, F(1,44)=4.02, P<0.05, and in period 5, F(1,44)=4.98, P<0.03. Mean performance of the low structuring of activities/turnover groups differed significantly from the mean performance of the groups in the other three conditions in both period 3, F(3,44)=3.95, P<0.05, and period 5, F(3,44)=13.84, P<0.01.

Period by period performance scores were re-analyzed after correcting for heteroskedasticity. The results from the GLS analysis were essentially the same before except for minor differences. There was no main effect for structuring of activities in any of the periods and there was a main effect for turnover only in periods 4 and 5, where none of the original group members were present. The turnover by structuring of activities interaction was significant in periods 3 and 5 as in the previous analysis.

These analyses broken out by time period reinforce the more aggregate results. Turnover has a negative effect in later time periods, after the initial members have departed. The main effect of structure is generally not significant. The interaction of structure and turnover is significant in later time periods, where high structure buffers the groups from the negative effects of turnover.

Videotape data

The videotape data were analyzed to understand why groups in the various conditions performed differently, and to relate performance differences to key knowledge variables. Both actions and comments of group members were coded. All 48 groups were videotaped as they worked on the task. Of these, 46 videotapes were coded because the video recordings of the remaining two groups were of poor quality and could not be transcribed. Two individuals coded the data. First, one of the investigators coded all the tapes. The second individual, who was naive to the research hypotheses, coded one-third of the videotapes – four groups in each of the four experimental conditions. These 16 groups were randomly selected from the available set of videotapes.

The inter-rater reliabilities of the videotape data were computed as Cohen's Kappa (Cohen, 1960). For the high structuring of activities/no-turnover condition, Cohen's Kappa was 0.78; for the high structuring of activities/turnover condition, it was 0.76; for the low structuring of activities/no-turnover condition, it was 0.74; and for the low structuring of activities/turnover condition, it was 0.75. The strength of agreement is substantial (Landis and Koch, 1977).

Table 1 contains the mean frequency of comments for the various knowledge variables. A number of these knowledge variables were highly correlated. Therefore, a multivariate analysis of variance (MANOVA) was performed on the knowledge variables with structuring of activities, turnover, and their interaction as independent variables. The hypothesis of no overall structuring of activities effect was strongly rejected, F(7,36)=7.41, P<0.0001. So was the hypothesis for no overall turnover effect, F(7,36)=9.00, P<0.0001, and no overall structuring of activities by turnover interaction F(7,36)=5.62, P<0.0002.

Table 1 - An analysis of the videotape process data: mean frequency of comments in each knowledge category as a function of experimental condition.

Full table

Because the MANOVA results were significant, univariate analyses were performed. For each knowledge variable in the coding scheme category, an analysis of variance (ANOVA) was performed with structuring of activities, turnover, and their interaction as independent variables. A discussion of these results in each category follows.

Problems in accessing existing knowledge

If group members had trouble accessing knowledge about the task, role specifications, rules and procedures, it was coded as 'Problems in accessing existing knowledge'. Examples include struggling with sheets of paper to build a product, asking other group members for help, or not doing the job assigned. The following segment was taken from the transcript of one of the low structuring of activities/turnover groups in the experiment to illustrate the actions and comments coded in this category.

• No. 4 flips paper around and keeps looking at it (5 s).
• No. 4 unfolds one fold and flips paper again.
• No. 4 to No. 3: What do I do now?
• No. 3 stops what he is doing to look at No. 4's sheet.

For this knowledge variable, there was a significant main effect for structuring of activities, F(1,42)=19.31, P<0.0001; a significant main effect for turnover, F(1,42)=18.42, P<0.0001; and a significant turnover by structure interaction, F(1,42)=4.63, P<0.04. The low structuring of activities groups had more trouble accessing existing knowledge than the high structuring of activities groups. Turnover groups experienced more difficulty than no-turnover groups. Scheffé tests revealed that the most difficulty in accessing existing information was experienced by the low structuring of activities/turnover groups, F(3,42)=13.49, P<0.001. This is consistent with the performance data, where the worst performance was exhibited by the low structuring of activities/turnover groups. The correlation between this variable, problems in accessing knowledge, and performance of the groups was negative and significant, as expected (r=-0.51, P<0.0003).

Gain in knowledge

If group members generated knowledge on their own, it was coded under this category. This knowledge could relate directly to the task, such as finding a short cut to folding the crowns and ships, or it could relate to group strategies. We separated this category into two subcategories: gain in productive knowledge and gain in harmful knowledge. For example, if group members devised a strategy to avoid scrap or to maximize their score it was coded in the subcategory 'gain in productive knowledge'. On the other hand, if the strategy the group came up with resulted in considerable scrap it was coded in the subcategory 'gain in harmful knowledge'. As expected, the 'gain in productive knowledge' variable had a significant positive correlation with group performance (r=0.66, P<0.0001), and the 'gain in harmful knowledge' variable had a significant negative correlation with group performance (r=-0.47, P<0.001). This evidence validates our coding of the knowledge variables.

There was a significant main effect for turnover for both subcategories of this variable, gain in productive knowledge, F(1,42)=23.18, P<0.0001; and gain in harmful knowledge, F(1,42)=12.92, P<0.0008. The no-turnover groups generated more useful knowledge, while the turnover groups generated more knowledge that hampered their performance.

Loss in knowledge

If no one in the group knew or remembered how to make the crowns or ships, or was able to recall other relevant information such as point allocations or penalty for scrap, it was coded under this category. The loss in knowledge had to pertain to the whole group and not just to one or two group members.

There was a marginally significant structuring of activities by turnover interaction for this variable, F(1,42)=3.33, P<0.08. The only groups that experienced total loss of knowledge were those in the low structuring of activities/turnover condition. The correlation between this variable and performance was negative, but did not reach significance (r=-0.17, NS).

Knowledge becomes obsolete

If the group organized itself in a particular way and then had to reorganize itself, the reorganization was coded under this category. For example, in one of the low structure/turnover groups, group members discovered that one person was very good at making crowns and so they organized themselves in such a way that he made crowns while the other two members made ships. At the end of work period turnover occurred and the person making crowns had to leave the group. The new member who joined the group was not able to make the crowns so the group had to reorganize itself and reassign responsibilities.

For this category, the interaction effect between turnover and structuring of activities was highly significant, F(1,42)=30.00, P<0.0001. The low structuring of activities/turnover groups had to keep reorganizing as their knowledge became obsolete. This continual need to reorganize hurt their performance. There was a significant negative correlation between this variable and group performance (r=-0.38, P<0.01).

Transmission of task knowledge

This category included the transmission of knowledge among group members about the task. Comments made about aspects of the task such as the scoring procedure or the ease or difficulty of folding the products were coded in this category. The following quote from a high structuring of activities/no turnover group is an example of comments coded in this category:

Cutter to Builder:

Is folding the hardest?

Builder:

The corners into the center take the longest time. This stuff is easy. The crown looks easier but is harder to open up.

For this subcategory, there was a significant main effect for structuring of activities, F(1,42)=9.49, P<0.004. The low structuring of activities groups transmitted more task knowledge than the high structuring of activities groups. This is consistent with our theory. Because knowledge was embedded in the structure of groups high in structure, those groups had less need to transmit knowledge to new members than groups low in structure.

Transmission of knowledge regarding roles and coordination

This category included the transmission of information among group members about roles or coordination requirements. For the transmission of knowledge regarding roles and coordination variable, the only significant effect was for structuring of activities, F(1,42)=15.01, P<0.0004. The low structuring of activities groups communicated more information regarding roles and coordination to members than did the high structuring of activities groups. An example of comments coded in this category is the following:

No. 2 to No. 1:

Should we do a production line, you think?

No. 1 to No. 2:

Actually no, then two people are idle. I think we should start out doing crowns. Well, you are the best at doing ships. Why don't you make the ships?

No. 1:

We'll start off with one for the time being and then we'll decide what to do.

The videotape data were also corrected for heteroskedasticity and reanalyzed. The results of the Generalized Least Squares (GLS) regression were essentially the same as the ANOVA results.

Discussion and conclusions

The results support our theoretical predictions. If groups experienced turnover, they performed better when roles were specified and routines existed than when roles and work routines were not specified clearly. When there was no turnover, however, there was no difference in the performance of the low structuring of activities groups and the high structuring of activities groups. Further, there was also a significant main effect for turnover. Groups that did not experience turnover performed better than those that did. Consistent with previous studies of organizational learning curves, performance improved with experience across trials. Further, the performance increase was greater in groups that did not experience turnover than in groups that did.

Why did the low structuring of activities/turnover groups not perform as well as the other groups? The videotape data suggest that their poor performance was because group members had trouble accessing existing knowledge, lost knowledge, and had to keep reorganizing when members left the group. The high structuring of activities/turnover groups did not experience the same difficulties because group members had specific jobs to perform and specific methods of coordinating with one another. Thus, they did not have to keep reorganizing whenever new members entered the group, nor did they have to transmit as much information to new group members. Knowledge embedded in the structure enabled the groups to maintain high levels of performance, even in the face of turnover.

An interesting difference emerged in the videotape process data between the turnover and no-turnover groups with respect to the generation of new knowledge on the task. When no-turnover groups generated new knowledge, it seemed to help their performance. They either developed easier ways to do the task or strategies to maximize their score such as not wasting supplies. The turnover groups, on the other hand, generated strategies that hurt more than helped their performance. This could be because a strategy that worked well for a particular set of individuals was not as effective when a new person with different expertise or preferences joined the group. These results are consistent with research on 'transactive memory' (Wegner, 1986) or knowledge of who knows what in groups (Liang et al., 1995; Brandon and Hollingshead, 2004; Lewis et al., 2005). Transactive memory systems developed for one group of individuals may not be effective when the group changes (Moreland et al., 1998). Clear differences also emerged in the videotape process data between the high structuring of activities groups and the low structuring of activities groups. Not only did the low structuring of activities groups have more difficulty accessing existing knowledge, they also spent more time transmitting information about the task and coordination requirements to new members than did groups high in structure.

Results evidenced differences within the 12 low structuring of activities/turnover groups that related to the main predictions of the experiment. The majority of the groups in this condition did not develop specific task assignments or procedures to do the task or change those they developed from one period to the next. There were three groups, however, in the low structuring of activities/turnover condition that developed role specifications and routines that were very similar to those provided to the high structuring of activities groups by the experimenter. These role specifications and routines persisted across all five work periods. The average performance of these three groups – 94 points – was much higher than the average performance of the other nine groups in the low structuring of activities/turnover condition – 78 points. This supplemental analysis provides additional evidence that the structuring of work counteracts the negative effects of turnover.

The purpose of this study was to test whether having knowledge embedded in routines and role specifications would help groups handle the turnover process more effectively, and thus enable them to perform better than groups that did not have knowledge embedded in the structure of work. The performance scores indicated that low structuring of activities groups were hurt by turnover. Groups that had knowledge embedded in the structure of work were able to handle turnover more effectively and hence were not as adversely affected by membership change as groups low in structure.

This study provides empirical evidence about the benefits of structure and standardization for organizational learning. Researchers have debated the effects of structure on organizational learning. Adler and Cole (1993), for example, suggested that the standardization inherent in the Toyota production system used at the Toyota-GM Nummi plant fostered organizational learning by providing a mechanism for capturing and transferring knowledge. By contrast, the researchers argued that the self-managed work groups used at the Volvo plant did not foster organizational learning because knowledge acquired by the various groups was not captured or transferred (see Berggen, 1994, for an opposing view). Our results provide evidence that structure provides a mechanism for retaining knowledge over time in organizations and transferring it to new members.

Negative consequences such as reduced innovation, however, may result from too much structuring of work (Nemeth and Staw, 1989). Thus, the benefits associated with high structure need to be balanced with any costs the high structure may impose. The conditions under which innovation occurs in highly structured groups need to be delineated. For example, new members have been shown to be a source of innovation, even in highly structured groups, when the new member shares a social identity with other group members (Kane et al., 2005).

Our results indicate that structuring work is an effective strategy for retaining knowledge and mitigating the harmful effects of turnover in organizations. Knowledge can persist over time in the face of membership change in organizations when knowledge is embedded in structure. Not only does structure provide a mechanism for retaining knowledge in organizations, it also provides a mechanism for transferring knowledge to new members. Thus, structuring of work is an option organizations should consider for ensuring that their performance continues to improve with experience, even in the face of individual member turnover.

Notes

¹ Because the turnover occurred after the first period, the performance scores were reanalyzed with the data from the first period removed. The results were the same.

References

1. Adler, Paul S. and Robert E. Cole, 1993, "Designed for learning: A tale of two auto plants". Sloan Management Review, 3: 85–94, and Rejoinder, 1994, 2: 45–49.
2. Allen, Michael P., Sharon K. Panian and Roy E. Lotz, 1979, "Managerial succession and organizational performance: A recalcitrant problem revisited". Administrative Science Quarterly, 24: 167–180. | Article |
3. Argote, Linda, 1999, Organizational learning: Creating, retaining and transferring knowledge. Norwell, MA: Kluwer.
4. Argote, Linda, Sarah Beckman and Dennis Epple, 1990, "The persistence and transfer of learning in industrial settings". Management Science, 36: 140–154.
5. Argote, Linda and Dennis Epple, 1990, "Learning curves in manufacturing". Science, 247: 920–924.
6. Argote, Linda and Paul Ingram, 2000, "Knowledge transfer in organizations: A basis for competitive advantage in firms". Organizational Behavior and Human Decision Processes, 82: 150–169. | Article |
7. Argote, Linda, Chester A. Insko, Nancy Yovetich and Anna A. Romero, 1995, "Group learning curves: The effects of turnover and task complexity on group performance". Journal of Applied Social Psychology, 25: 512–529.
8. Argote, Linda and Ron Ophir, 2002, "Intraorganizational learning". In J.A.C. Baum (ed.) Companion to organizations. Oxford, UK: Blackwell, pp 181–207.
9. Baum, Joel A.C. and Paul Ingram, 1998, "Survival-enhancing learning in the Manhattan hotel industry, 1898–1980". Management Science, 44: 996–1016.
10. Benkard, C.Lanier, 2000, "Learning and forgetting: The dynamics of aircraft production". American Economic Review, 90: 1034–1054.
11. Berggen, Christian, 1994, "Nummi vs. Uddewalla". Sloan Management Review, 2: 37–45.
12. Brandon, D.P. and A.B. Hollingshead, 2004, "Transactive memory systems in organizations: Matching tasks, expertise and people". Organization Science, 15: 633–644. | Article |
13. Brown, M.Craig, 1982, "Administrative succession and organizational performance: The succession effect". Administrative Science Quarterly, 27: 1–16. | Article |
14. Carley, Kathleen, 1992, "Organizational learning and personnel turnover". Organization Science, 3: 20–46.
15. Cohen, J., 1960, "A coefficient of agreement for nominal scales". Education and Psychological Measurement, 20: 37–46.
16. Cohen, Michael D. and Paul Bacdayan, 1994, "Organizational routines are stored as procedural memory: Evidence from a laboratory study". Organization Science, 5: 554–568.
17. Cyert, Richard M. and James G. March, 1963, A behavioral theory of the firm. Englewood Cliffs, NJ: Prentice Hall.
18. Darr, Eric, Linda Argote and Dennis Epple, 1995, "The acquisition, transfer, and depreciation of learning in service organizations: Productivity in franchises". Management Science, 44: 1750–1762.
19. De Holan, Pablo Martin and Nelson Philips, 2004, "Remembrance of things past: The dynamics of organizational forgetting". Management Science, 50: 1603–1613.
20. Dutton, John and Annie Thomas, 1984, "Treating progress functions as a managerial opportunity". Academy of Management Review, 9: 235–247. | Article |
21. Epple, Dennis, Linda Argote and Rukmini Devadas, 1991, "Organizational learning curves: A method for investigating intra-plant transfer of knowledge acquired through learning by doing". Organization Science, 2: 58–70.
22. Feldman, Martha S. and Brian T. Pentland, 2003, "Reconceptualizing organizational routines as a source of stability and change". Administrative Science Quarterly, 48: 94–118.
23. Gersick, Connie J.G. and J.Richard Hackman, 1990, "Habitual routines in task-performing groups". Organizational Behavior and Human Decision Processes, 47: 65–97. | Article |
24. Glebbeck, Arie C. and Erik H. Bax, 2004, "Is high employee turnover really harmful? An empirical test using company records". Academy of Management Journal, 47: 277–286.
25. Grusky, Oscar, 1961, "Corporate size, bureaucratization, and managerial succession". American Journal of Sociology, 67: 261–269. | Article |
26. Hatch, Nile W. and David C. Mowery, 1998, "Process innovation and learning by doing in semiconductor manufacturing". Management Science, 44: 1461–1477.
27. Hill, W.F. and L. Gruner, 1973, "A study of development in open and closed groups". Small Group Behavior, 4: 355–381.
28. Huber, George P., 1991, "Organizational learning: The contributing processes and the literatures". Organization Science, 2: 88–115.
29. Ingram, Paul, 2002, "Interorganizational learning". In J.A.C. Baum (ed.) The Blackwell Companion to Organizations. Oxford, England: Blackwell Business.
30. Insko, C.A., John W. Thibaut, Debra Moehle, M. Wetson, W.D. Diamond, Robert Gilmore, M.R. Soloman and Angela Lipsitz, 1980, "Social evolution and the emergence of leadership". Journal of Personality and Social Psychology, 39: 431–448. | Article |
31. Jones, M.B., 1974, "Regressing group on individual effectiveness". Organizational Behavior and Human Performance, 11: 426–451. | Article |
32. Kane, Aimee A., Linda Argote and John M. Levine, 2005, "Knowledge transfer between groups via personnel rotation: Effects of social identity and knowledge quality". Organizational Behavior and Human Decision Processes, 96: 56–71. | Article |
33. Kogut, Bruce and Udo Zander, 1992, "Knowledge of the firm, combinative capabilities and the replication of technology". Organization Science, 3: 383–397.
34. Landis, J.R. and G.G. Koch, 1977, "The measurement of observer agreement for categorical data". Biometrics, 33: 159–174. | Article | PubMed | ISI | ChemPort |
35. Lane, Frederic Chapin, 1951, Ships for victory: A history of shipbuilding under the U.S. Maritime Commission in World War II. Baltimore: The Johns Hopkins Press.
36. Levitt, Barbara and James G. March, 1988, "Organizational learning". Annual Review of Sociology, 14: 319–340. | Article |
37. Lewis, Kyle, Donald Lange and Lynette Gillis, 2005, "Transactive memory systems, learning, and learning transfer". Organization Science, 16: 726–727. | Article |
38. Liang, Diane, Richard L. Moreland and Linda Argote, 1995, "Group versus individual training and group performance: The mediating effects of transactive memory". Personality and Social Psychology Bulletin, 21: 384–393.
39. McGrath, Joseph E., 1986, "Studying groups at work: Ten critical needs for theory and practice". In P.S. Goodman (ed.) Designing Effective Work Groups. San Francisco, CA: Jossey-Bass, pp 362–391.
40. Moreland, Richard L., Linda Argote and Ravi Krishnan, 1998, "Training people to work in groups". In R.S. Tindale, L. Heath, J. Edwards, E.J. Posvac, F.B. Bryant, Y. Suarez-Balcazar, E. Henderson-King and J. Myers (eds) Theory and Research on Small Groups. New York: Plenum.
41. Mowday, R.T., L.W. Porter and R.M. Steers, 1982, Employee-organization linkages: The psychology of commitment, absenteeism, and turnover. New York: Academic Press.
42. Nelson, Richard R. and Sidney Winter, 1982, An evolutionary theory of economic capabilities and behavior. Cambridge, MA: Harvard University Press.
43. Nemeth, Charlan J. and Barry Staw, 1989, "The tradeoffs of social control and innovation in groups and organizations". In L. Berkowitz (ed.) Advances in Experimental Social Psychology, Vol. 22. New York: Academic Press, pp 175–210.
44. Pfeffer, Jeffery. and Alison Davis-Blake, 1986, "Administrative succession and organizational performance: How administrator experience mediates the succession effect". Academy of Management Journal, 29: 72–83. | Article |
45. Pisano, Gary, Richard Bohmer and Amy Edmondson, 2001, "Organizational differences in rates of learning: Evidence from the adoption of minimally invasive cardiac surgery". Management Science, 47: 752–768. | Article |
46. Price, James L., 1977, The study of turnover. Ames, Iowa: The Iowa State University Press.
47. Rapping, Leonard A., 1965, "Learning and World War II production functions". Review of Economics and Statistics, 47: 81–86. | Article |
48. Reagans, R., Linda Argote and Daria Brooks, 2005, "Individual experience and experience working together: Predicting learning rates from knowing what to do and knowing who knows what". Management Science, 51: 869–881. | Article |
49. Schulz, Martin, 2002, "Organizational learning". In J.A.C. Baum (ed.) The Blackwell Companion to Organizations. Oxford, England: Blackwell Business.
50. Shaw, J.D., M.K. Duffy, J.L. Johnson and D.E. Lockhardt, 2005, "Turnover, social capital losses and performance". Academy of Management Journal, 48: 594–606.
51. Shaw, J.D., N. Gupta and J.E. Delery, 2005, "Alternative conceptualizations of the relationship between voluntary turnover and organizational performance". Academy of Management Journal, 48: 50–68.
52. Staw, Barry M., 1980, "The consequences of turnover". Journal of Occupational Behavior, 1: 253–273.
53. Teece, David J., Gary Pisano and A. Shuen, 1997, "Dynamic capabilities and strategic management". Strategic Management Journal, 1: 509–533. | Article |
54. Trow, Donald B., 1960, "Membership succession and team performance". Human Relations, 13: 259–269. | Article |
55. Walsh, James P. and Gerardo Rivera Ungson, 1991, "Organizational memory". Academy of Management Review, 16: 57–91. | Article |
56. Wegner, Daniel M., 1986, "Transactive memory: A contemporary analysis of the group mind". In B. Mullen and G.R. Goethals (eds) Theories of Group Behavior. New York: Springer-Verlag.
57. Yelle, Louis E., 1979, "The learning curve: Historical review and comprehensive survey". Decision Sciences, 10: 302–328.

Acknowledgements

We acknowledge the support of a Dissertation Research Award to Rukmini Devadas Rao from the American Psychological Association. We thank Linda Christie, Becca Fulcer, and Louise Morin for their help collecting and coding the data. Earlier drafts of this paper were presented at meetings of the Midwestern Psychological Association, INFORMS, and seminars at Carnegie Mellon University, the University of Pennsylvania (Wharton), and the University of Southern California. We thank participants in these forums, the Editor, Bruce Kogut, and the reviewer for their very helpful comments.