Understanding the Factors Influencing the Value of Person-to-Person ...

Proceedings of the 39th Hawaii International Conference on System Sciences - 2006

Understanding the Factors Influencing the Value of Person-to-Person Knowledge Sharing Susan A. Brown University of Arizona [email protected]

Alan R. Dennis Indiana University [email protected]

Abstract It is generally accepted that knowledge sharing is a difficult task for organizations. Many reasons for this difficulty have been proposed. In this paper, we offer another. Specifically, we build on Zander and Kogut’s work [29] and examine the relationship between knowledge dimensions and knowledge sharing. Departing from their study, we focus on person-to-person, rather than organization-to-organization, knowledge sharing. We surveyed 68 employees of a Workman’s Compensation Board in Canada. To analyze the data, we employed Hierarchical Linear Modeling. The results demonstrate that complexity and teachability increased person-to-person knowledge sharing but observability did not. Contrary to expectations, the availability of codified knowledge in a knowledge management system (KMS) had no impact on person-to-person knowledge transfer; individuals were as likely to share knowledge person-toperson regardless of whether there was a KMS available that contained appropriate knowledge.

1. Introduction Knowledge and intellectual capital have been recognized as key organizational assets [e.g., 9, 18, 28]. As with other organizational assets, it is important to manage knowledge to ensure that its use achieves organizational goals. Unlike most other assets, the greatest value of knowledge within an organization is derived when it is shared, as this improves work and facilitates the development of new knowledge [4]. Knowledge sharing is not an easy task. A number of organizations have implemented knowledge management systems (KMS), only to find that employees do not use them [9]. Issues such as motivating employees to share knowledge [27], creating positive attitudes around knowledge sharing [2], and trust [17] continue to be addressed in research and in practice. Yet, even with technology and incentives in place to enable it, knowledge sharing remains a difficult task for

Diana B. Gant National Science Foundation [email protected]

organizations to accomplish. Research suggests that personal and institutional factors [2, 13, 27] and aspects of the KMS [7] can increase or inhibit knowledge sharing. The nature of knowledge itself is also a key aspect in effective knowledge sharing [20, 21, 28, 29]. In this paper, we examine how the nature of knowledge impacts the value of person-to-person knowledge sharing. We examine the impact of three dimensions of knowledge (teachability, observability, complexity) and the presence of codified knowledge in a KMS on person-to-person knowledge sharing. Specifically, we address the following question: Do characteristics of the knowledge and the amount of relevant codified knowledge in an organizational KMS affect the value of knowledge received from colleagues via person-to-person communication?

2. Theoretical Background Hansen, Nohria, and Tierney [8] suggest that there are two primary approaches to managing knowledge sharing within organizations: codification and personalization. With the codification approach, organizations rely heavily on computers, carefully codifying the knowledge and storing it in documents in a KMS in order to make it accessible to a large number of people in the organization. This codification approach is useful for organizations whose strategic focus is on the standardization of knowledge, and focuses on sharing knowledge through documents. Organizations employing codification typically hope to reap the benefits of knowledge reuse. With the personalization approach, organizations attempt to connect knowledge seekers to the people who have the needed knowledge. Similar to ‘T-shaped managers’ [9], this knowledge management approach is more one of managing connections to who has the knowledge as opposed to managing the knowledge per se. The personalization approach is useful for organizations in which the strategic orientation is expertise. From the user perspective, codification and personalization are very different approaches. With a codification approach, the knowledge seeker finds

0-7695-2507-5/06/$20.00 (C) 2006 IEEE

1


relevant knowledge documents in the KMS and acquires knowledge by reading the documents. Knowledge is inherently tied to a specific context when it is first created because it is closely tied to the originating user, task, organizational unit, etc [8, 12, 26]. If the knowledge in the knowledge documents was not developed in the same context as the user’s context, then the user must understand the context in which the knowledge was created, and assess how close his or her target context is to the original context. The knowledge must be deconstructed from its original context and put in a general form tobe transferred and then reconstructed in the user’s target context making whatever adjustments are needed for the new context. This deconstruction, sharing and reconstruction can be challenging. With a personalization approach, knowledge is received from a person, not a document. The exchange process is therefore more interactive than with a codification-based KMS [16]. Thus the potential knowledge user can work together with the knowledge source to jointly and interactively contextualize the knowledge into the user’s context. Regardless of which strategy an organization adopts, individuals are also capable of bypassing the formal KMS and using their own social networks to acquire knowledge [9]. If the codification or personalization KMS does not provide needed knowledge or does not provide knowledge in an easy-to-consume form, users can simply seek knowledge from individuals they know. Very little is known about how the nature of knowledge influences the use of such informal person-to-person knowledge sharing outside of a formally sanctioned codification-based KMS.

2.1 The value of knowledge sharing In order to determine whether and how much knowledge sharing has occurred, we examined the perceived value of the knowledge sharing process. Perceptions have been used in other research as a proxy for the value of knowledge sharing [e.g., 15] and provide a useful mechanism for applying the research model in a wide variety of settings. The alternative is to employ tests associated with the actual knowledge content; this approach would have lower utility across research settings, due to the context-specific nature of the tests. The research model depicted in Figure 1 includes four components to assess the value of knowledge sharing: efficiency (the sole metric used by Zander and Kogut [29]), quality, learning, and understanding. While these outcomes can be achieved using a KMS, as well as person-to-person knowledge sharing, our focus here is on the person-to-person approach (as reflected in our measures in Table 1). The knowledge sharing process should assist employees in saving time in the performance of their jobs. For some jobs, there are key elements that are fairly standard, and having knowledge about those elements readily available will save time [29]. Further, as Hansen et al. [8] point out, knowledge reuse is a critical component of successful operations in some organizations. The ability to capitalize on knowledge that has already been created within the organization results in significant time savings, thus improving the efficiency of individual workers, as well as the organization as a whole. It is equally important that knowledge sharing improve the quality of work performed. Knowledge sharing should not be such an onerous task that it detracts from job performance. In fact, the quality of work should improve as more knowledge is available to more individuals in the organization [8].

Knowledge Dimensions Codifiability Value of Person-to-Person Knowledge Sharing

Procedural Complexity

Efficiency Quality Learning Understanding

Teachability

Observability

Control Variables

System Dependence

Gender Time in Job

Figure 1: Research Model

2


For knowledge sharing to be truly valuable, it should enhance learning [1, 14]. Having a solid base of existing knowledge enables individuals to recognize the value of new knowledge [4]. Thus, assessing the degree to which an individual has acquired new knowledge as a result of the sharing is a key aspect of evaluating it. Ko et al. [14] propose that learning and the ability to apply that learning (i.e., understanding) are separate and distinct aspects of knowledge sharing. An organization truly benefits when its members not only pass along knowledge, but also share the deep structures necessary to apply that knowledge. Taken together, these four aspects comprise a broader view of the value of knowledge sharing than was considered by Zander and Kogut [29]. We believe that all four are important in assessing the outcomes and ultimate success of knowledge sharing.

2.2 Knowledge dimensions Zander and Kogut [29] argue that five dimensions of knowledge affect the ability to efficiently share knowledge outside the bounds of a formal KMS: codifiability, procedural complexity, teachability, system dependence and observability. Codifiability refers to the degree to which knowledge can be encoded and stored. Procedural complexity refers to the variety of procedural resources that must be combined to create knowledge. Teachability refers to the degree to which the knowledge can be shared via training, either in school or on the job. Observability refers to the degree to which knowledge can be imitated or copied by observing the performance of a task or its products. System dependence refers to the degree to which knowledge is dependent on and derived from many different people . In theory, knowledge that is less complex, more codifiable, more teachable, and more observable should be easier to share [29]. Likewise, the greater the extent to which a task is system dependent the more likely knowledge will be shared [29]. Zander and Kogut [29] found that for inter-organizational knowledge transfer, only codifiability and teachability had significant impacts. Our focus is somewhat different than Zander and Kogut’s [29]. They examined the impact that these five knowledge dimensions had on the speed at which innovations spread among firms. Rather than examining inter-organizational knowledge sharing between manufacturing firms, we choose to examine the person-toperson sharing of knowledge within the same organization. There are, however, some significant differences as we move from inter-organizational knowledge transfer to intra-organizational knowledge sharing. In inter-organizational transfer, one organization is trying to imitate, and thus take advantage of another organization’s knowledge. The knowledge must be

extracted from products or announcements in the public domain. In this environment, knowledge transfer diminishes the value of knowledge in the first organization. In intra-organizational knowledge sharing, individuals are exchanging knowledge in order to advance their organization. This knowledge may be exchanged personto-person or via a KMS. In this environment, knowledge sharing enhances the value of knowledge in the organization. Thus, while there may be differences in the settings, we believe that these same five dimensions will be important in person-to-person knowledge sharing. The first dimension is not only whether knowledge for a given task can be codified, but whether it has been codified. If there is little codified knowledge available in a KMS to support a task, then we would expect the value of person-to-person knowledge sharing to be high. Conversely, as the amount of task-relevant codified knowledge available in a KMS increases, we would expect the value of person-to-person knowledge sharing to gradually decrease because knowledge is now available in the KMS and some individuals will choose to use the KMS rather than seek knowledge from another person. Thus, we offer the following hypothesis: H1: The existence of codified knowledge is negatively associated with the value of person-toperson knowledge sharing. The complexity of the knowledge task is a second important dimension. As the complexity of knowledge required to perform a task increases, people are more likely to seek knowledge from people rather than from a codification-based KMS [3]. For simple tasks, the needed knowledge is often available in a small number of formal sources (e.g., documents in a KMS), that are relatively quick and simple to locate. As complexity increases, the need for interactivity [16] and help in contextualizing the knowledge increases, so the value of person-to-person knowledge sharing should increase. As complexity increases, the number of different sources from which knowledge is needed increases, so that the knowledge seeker is more likely to rely on people rather than documents. Departing slightly from Zander and Kogut [29], we refer to this as procedural complexity, to capture the complexity associated with procedural tasks. Thus, we offer the following hypothesis: H2: Complexity is positively associated with the value of person-to-person knowledge sharing. The third dimension, teachability, is directly related to person-to-person knowledge sharing. If knowledge is easy to teach, and thus easy to learn, then it is more likely to be shared from person-to-person [29] thus increasing the value of person-to-person knowledge sharing. Conversely, if knowledge is difficult to teach, it is less likely that the knowledge will be shared from person-toperson and its value should decrease. Thus, we offer the

3


following hypothesis: H3: Teachability is positively associated with the value of person-to-person knowledge sharing. The fourth dimension, observability, presupposes that the knowledge needed to do the task is readily apparent through observation [28], not interaction. Thus, when knowledge is observable, there is less need to for personto-person interaction to share knowledge; individuals can simply watch and learn. Conversely, if through the process of watching someone perform their job, the observer can not extract the underlying knowledge, then person-to-person interaction is essential to sharing knowledge and the value of that interaction increases. Thus, we offer the following hypothesis: H4: Observability is negatively associated with the value of person-to-person knowledge sharing. The final dimension of knowledge is system dependence. If task knowledge is derived from and dependent on a larger number of people, processes, or systems, then those seeking that knowledge are more likely to seek knowledge from many different sources, more of which are likely to be people, rather than KMS documents [3]. Thus the value of person-to-person knowledge sharing should be high. Conversely, if knowledge is not highly system dependent, then a few sources, mostly documents, will likely suffice and the value of person-to-person sharing should be lower. Thus, we offer the following hypothesis: H5: System dependence is positively associated with the value of person-to-person knowledge sharing.

3. Method This research was conducted at a Workman’s Compensation Board (WCB) in one of the ten Canadian provinces. The WCB assesses the safety of working conditions and trains organizations how to create a safer working environment. At the time of the study, the WCB had implemented a simple KMS for storing documents, regulations, procedures, and basic knowledge, and was in the process of assessing the viability of implementing a new KMS that provided more features. The subjects in this study were the 180 prevention officers and managers whose job it was to assess and respond to safety concerns in organizations throughout the province. Each eligible employee received an email requesting their participation in an online survey. We received complete responses from 68 employees, resulting in a 38% response rate. Approximately 53% of the respondents were occupational health officers, 34% were occupational safety officers, and 13% were managers. On average, the respondents had 10.7 years of job experience, and were overwhelmingly male (87%). These demographics are consistent with the

organizational distribution of employees, thus suggesting that non-response issues did not unduly bias the sample. The survey asked respondents to assess the, codifiability, procedural complexity, teachability, observability, and system dependence of knowledge for four job tasks performed by all occupational health and safety officers (accident reports, inspections, education, and consulting). They were also asked to report the extent to which codified knowledge existed for each of those tasks and the outcomes of knowledge sharing from other prevention officers (as opposed to the KMS) for each task. We also asked each respondent to report his or her gender and the number of years experience he or she had as a prevention officer, which we used as controls. All independent measures were drawn from Zander and Kogut [29] and slightly adapted to the WCB environment. For example, instead of stating “new manufacturing personnel” the items were changed to state “new prevention officers.” The items for the value of knowledge sharing were drawn from Ko et al. [14], and elaborated on and expanded for use in this context. The constructs and the specific items used to measure them are presented in Table 1. All constructs are measured from the perspective of the knowledge receiver, because the receiver is the best person to judge whether he or she has benefited from knowledge sharing [14].

4. Analysis We began by conducting a factor analysis on the survey items to ensure they loaded on the constructs as intended (see Table 2). All items loaded as expected on the six major constructs with minimal cross-loading. All items had adequate reliability, except for system dependence (see Table 1), which had a Cronbach’s alpha of .41, compared to an alpha of .64 in Zander and Kogut’s [29] original study. Because of its low alpha, we removed system dependence from the study. We therefore had four matched sets of independent and dependent variables from each survey respondent (one set for each of the four tasks performed), so we could not use standard regression techniques to analyze the data [11, 23, 25]. With traditional regression, there is a problem with the unit of analysis. If the data are analyzed at the lowest level (the four knowledge tasks), then the impact of the individual must be omitted. Because there is likely to be significant correlation among the four knowledge sharing scores for a specific individual, this approach can erroneously inflate the significance and cause type 1 errors. If the data are analyzed at the second level (individual in our case), then we cannot include the four first level variables (codifiability, complexity teachability, and observability) in the model, except in aggregate, which removes precision.

4


Table 1: Measures

Value of Person-toPerson Knowledge Sharing (alpha=.92)

KS1. The advice I received from other Prevention Officers helped me to learn how to gather the necessary information to complete the task. KS2. The advice I received from other Prevention Officers allowed me to learn how employers in their workplaces will use this work product. KS3. The advice I received from other Prevention Officers has increased my ability to ask penetrating questions about conducting this task. KS4. The advice I received from other Prevention Officers has improved my knowledge of this work task. KS5. The advice I received from other Prevention Officers will allow me to complete this task more efficiently in the future because I will know where to go for advice. KS6. The advice I received from other Prevention Officers will allow me to complete this task more efficiently in the future because I will be able to conduct the task with greater independence. KS7. The advice I received from other Prevention Officers will allow me to complete this task more efficiently in the future because I am more knowledgeable about the task. KS8. The advice I received from other Prevention Officers will allow me to improve the quality of future work products because I will know where to go for advice. KS9. The advice I received from other Prevention Officers will allow me to improve the quality of future work products because I will be able to conduct the task with greater independence. KS10. The advice I received from other Prevention Officers will allow me to improve the quality of future work products because I am more knowledgeable about the task. T1.

Teachability (alpha=.72)

Existence of Codified Knowledge (alpha=.82)

T2. T3. T4. T5. CK1. CK2. CK3.

Procedural Complexity (alpha=.72)

PC1. PC2. PC3.

Observability (alpha=.73)

O1. O2. O3.

System Dependence (alpha=.41)

SD1. SD2. SD3.

New Prevention Officers can easily learn how to do this task by talking to skilled Prevention Officers. New Prevention Officers can easily learn how to do this task by studying relevant documentation. Educating and training of new Prevention Officers to complete this task is a quick and easy job. New Prevention Officers know enough after WCB new employee training to do this task. New Prevention Officers know enough after participating in the Prevention Division mentoring program to do this task. Large parts of the documentation for this task are embodied in the shared drive. Extensive documentation describing critical parts of the process for completing this task exists within the Prevention Division. Extensive documentation describing critical parts of the process for completing this task exists in the WCB. Processes for using reference materials are important to do this task. Processes for collecting information are important to doing this task. Processes for assembling reports are important to doing this task. A Prevention Officer can easily learn how to do this task by analyzing existing reports. A Prevention Officer can easily learn how to do this task by using an existing report as a template. A Prevention Officer can easily learn how to do this task by observing other Prevention Officers doing this task. It is impossible for any one Prevention Officer to know everything about this task. To obtain high work performance, it is very important that the Prevention Officers have long experience in the specific offices where they are working. One Prevention Officer can do this job in isolation from other Prevention Officers without product quality suffering. (Reverse coded)

Therefore, we used Hierarchical Linear Modeling (HLM) which is designed to analyze this type of multilevel research [11, 23, 25; see also 6 for an application in knowledge sharing]. In our case, we have a two level model: the lowest level (level 1) is the knowledge task (which has the matched set of knowledge sharing and the four independent variables); the second level (level 2) is the individual knowledge worker and his or her job experience and gender. Because we have two sets of models, one for level 1 and one for level 2, we now can calculate R2 at both level 1 and level 2 [25]. We have only

four knowledge tasks per person, so we can treat a maximum of three variables as random effects factors (which means that they are different for each individual employee). We chose to treat the intercept, teachability, and observability as random effects factors and leave the existence of codified knowledge and procedural complexity as fixed effects factors, which means that they are treated exactly like standard regression coefficients – there is one value for each coefficient that is calculated for everyone in the sample and that remains constant across all data. We believed that the assessments of the

5


existence of codified knowledge and a task’s procedural complexity were less likely to differ in meaningful ways among the individuals in our sample; they were more “objective” than assessments of observability and teachability. An analysis of the data revealed that, as expected, procedural complexity had a low coefficient of variation, but that the coefficient of variations for the other three variables were similarly high. Table 2: Factor analysis 1 .685 .663 .819 .829 .787 .861 .865 .736 .759 .734

2

Components 3 4

5 6 KS1 KS2 KS3 KS4 KS5 KS6 KS7 KS8 KS9 KS10 T1 .768 T2 .549 .351 T3 .687 T4 .544 T5 .384 .624 CK1 .781 CK2 .883 CK3 .835 PC1 .721 PC2 .757 PC3 .305 .698 O1 .792 O2 .842 .513 O3 SD1 .626 SD2 .825 SD3 .490 Note: Loadings lower than .300 are not displayed

5. Results We followed the HLM analysis process recommended by Hoffman [11] and Snijders and Bosker [25]. Table 3 summarizes the results of each step in this process. Step 1 shows the results of an unconditional model (the ICC was 0.266). Step 2 builds a random coefficient model by adding the four level 1 independent, resulting in a Level 1 R2 of 32%, a Level 2 R2 of 30%, and a significant decrease in deviation, indicating that this model better fits the data than the step 1 unconditional model. Step 3 builds a level 2 intercept-only model that adds job experience and gender as independent variables for the level 1 intercept; that is, a model to explain mean individual performance. Once again, there is an increase

in Level 1 and Level 2 R2, but the decrease in deviation is not significant. Step 4 builds a level 2 slope and intercept model that adds job experience and gender as independent variables for the impact of teachability and observability as well as level 1 intercept. Once again, the decrease in deviance is not significant compared to the model in Step 2 and there are only slight improvements in Level 1 and Level 2 R2. The random coefficient model (Step 2 in Table 3) shows that teachability and procedural complexity have significant positive impacts on the value of person-toperson knowledge sharing, while observability and the existence of codified knowledge have no effect. The addition of the individual-specific factors (Step 3, Table 3) shows that job experience is significantly negatively related to the value of knowledge sharing (presumably because the individual has less to learn as experience increases) but that gender is not significant. However, the deviance statistic suggests that the addition of job experience does not significantly improve overall model fit. Therefore, we conclude that the model in Step 2 is the most efficient model. As an aside, we note that the Level 1 unexplained variance at Step 4 is significantly different from zero (Ȥ2 = 189.58, df = 44, p