Expertise Redundancy, Transactive Memory, and Team Performance in Interdisciplinary Care Teams

Abstract

Objective

To examine how expertise redundancy and transactive memory (TM) in interdisciplinary care teams (ICTs) are related to team performance.

Data Sources/Study Setting

Survey and administrative data were collected from 26 interdisciplinary mental health teams.

Study Design

The study used a longitudinal, observational design. Independent variables were measured at baseline, 6, and 12 months: expertise redundancy (the extent to which team members possess highly overlapping knowledge), TM accuracy (the extent to which team members accurately recognize experts in relevant knowledge domains), and TM consensus (the extent to which team members agree on who is expert in which knowledge domain). Team performance was measured as risk‐adjusted average number of client hospitalization for the 6 months following each survey.

Data Collection Methods

Survey data were collected by the authors. Administrative data were collected by the state's administrative agency.

Principal Findings

Expertise redundancy had a negative effect on performance. TM accuracy had a positive effect on performance, and such effect was stronger when expertise redundancy was higher. No significant effect was found on TM consensus.

Conclusions

Transactive memory could serve as a cognitive coordination mechanism for mitigating the negative effect of complex knowledge structure in ICTs.

The use of interdisciplinary care teams (ICTs) in health care continues to grow. The Institute of Medicine identified ICTs as an integral strategy for improving care quality and transforming health care systems (Donaldson and Mohr 2000; Institute of Medicine 2001). Many recent innovations and reform initiatives were built upon the idea of team‐based care (Crabtree et al. 2010; Schuetz, Mann, and Everett 2010; Boult et al. 2011). ICTs are particularly advantageous for bringing a broad range of expertise to address complex health care needs (Heinemann 2002). However, using diverse disciplinary knowledge in ICTs can be challenging because it requires health care professionals to effectively identify and exploit different information and perspectives (van Knippenberg, De Dreu, and Homan 2004). It is imperative to understand what makes ICTs effective in identifying and utilizing diverse knowledge to achieve high performance.

Research in both health care and non–health care settings shows that diversity in team composition is not necessarily associated with effective teamwork or better performance (Opie 1997; Shortell et al. 2004; Mannix and Neale 2005; van Knippenberg and Schippers 2007). Such evidence underscores the need for knowledge coordination in diverse groups. As van Knippenberg, De Dreu, and Homan (2004, p. 1011) stated, “it is not the availability of information per se but the use of this information in group task performance that lies at the basis of diverse groups' potentially superior performance.”

The literature suggests several mechanisms for knowledge coordination (Okhuysen and Bechky 2009). One foundational mechanism uses formal roles to indicate expertise and coordinate task performance (Hackman 1990). A role is a bundle of expertise and tasks that are expected of those who occupy a position (Biddle and Thomas 1966). Role‐based coordination can be very effective in settings where the role structure is rigorous and enduring such that both expertise and tasks can be clearly distinguished and tied to specific persons (Bechky 2006). For example, doctors, nurses, and registration clerks in a primary care clinic may coordinate well based on their distinctive roles. In other settings where the role structure is imprecise and dynamic, formal roles can only serve as proxies for expertise. Informal expert roles may be enacted by team members as they negotiate and construct their work arrangements (Turner 1986). Examples of such settings include tumor boards and Assertive Community Treatment (ACT) teams in which group members possess both unique and overlapping knowledge and are expected to utilize such knowledge to make joint decisions or co‐perform tasks. These groups need to supplement formal roles with more adaptive coordination mechanisms such as team cognition and scaffolding (Brandon and Hollingshead 2004; Okhuysen and Bechky 2009).

This study focuses on the latter settings and examines how two aspects of a team's knowledge structure, expertise redundancy and transactive memory, interactively affect team performance. Expertise redundancy is the extent to which team members possess highly overlapping knowledge, a complex structure that may heighten the need for knowledge coordination. Transactive memory is team members' shared knowledge of who knows what. We draw on data collected from 26 ACT teams to test our hypotheses. Our results show that expertise redundancy is negatively and transactive memory accuracy is positively associated with team performance. Further, transactive memory accuracy has a stronger positive relationship with team performance when expertise redundancy is higher.

Theory

In ICTs with imprecise and dynamic role structures, knowledge overlap may arise from a mix of the following scenarios: (1) teams needing multiple members with the same expertise to serve large clienteles; (2) teams members developing extra‐role knowledge over time through cross‐training and interdisciplinary learning; and (3) team members sharing the responsibility for developing knowledge and performing tasks that are not tied or loosely tied to specific roles (Dyer 2003; Smith and Hou 2014). Through these activities, team members may develop expertise and informal expert roles that are not defined by their formal roles.

Some knowledge overlap may be useful to facilitate mutual understanding of diverse perspectives. However, excessive knowledge overlap or expertise redundancy may negatively affect team performance for several reasons. First, from a design perspective, expertise redundancy creates a more complex knowledge structure that is difficult for members to navigate. Expertise redundancy may induce ambiguity in members' roles and responsibilities, leading to inefficiency (e.g., more team members assuming the same role and responsibility than needed) or ineffectiveness (e.g., failure to fulfill certain role and responsibility due to lack of coordination) in task performance.

Second, from a process perspective, expertise redundancy may hinder effectiveness in information processing and communication. Information sampling research shows that shared information is more likely to be recalled and discussed in group discussion than unshared information (Stasser and Titus 1985; Stasser, Taylor, and Hanna 1989; Larson, Foster‐Fishman, and Keys 1994). This literature suggests that two mechanisms may contribute to information sharing biases: (1) Shared and redundant information has a higher probability of being recalled because more members possess such information (recall bias); and (2) shared information may contribute more to the group discussion and decision because it tends to conform to more members' initial preferences (discussion bias). Another study suggests that, in a high‐overlap situation, group members tend to make false assumptions about what other members already know and give inadequate clues when communicating unshared information, which results in confusion and errors in communication (Wu and Keysar 2007). Therefore, we expect that there is a negative relationship between expertise redundancy and team performance.

Hypothesis 1: Expertise redundancy is negatively associated with team performance in ICTs.

Team research suggests that well‐functioning teams can overcome challenges associated with excessive knowledge overlap in ICTs through assigning expert roles and developing shared team cognition. Research indicates that assigning group members clear expert roles in ad hoc teams facilitates group information processing (Stasser, Stewart, and Wittenbaum 1995; Stewart and Stasser 1995). In longstanding ICTs, expert roles evolve over time as members obtain and master new knowledge. In such teams, transactive memory systems theory provides a general framework for examining the relationship between teams' knowledge structure and performance.

Transactive memory (TM) is team members' shared knowledge of who knows what or a team's meta‐knowledge of the location of team expertise (Wegner 1987, 1995; Lewis and Herndon 2011; Ren and Argote 2011). Several factors contribute to the development of TM in teams, including members' recognition of individual traits, roles, and other status cues, task interdependence, group training and shared experience, and communication (Liang, Moreland, and Argote 1995; Bunderson 2003; Ren and Argote 2011). The literature suggests that TM is an important cognitive coordination mechanism that differentiates well‐functioning teams from their counterparts (Moreland 1999). Two properties of TM are particularly important for knowledge coordination: the extent to which team members can accurately recognize experts in relevant knowledge domains (TM accuracy), and the extent to which team members agree on who is expert in which knowledge domain (TM consensus; Austin 2003).

Transactive memory systems theory suggests that TM accuracy and consensus affect performance by fostering cross‐understanding and helping team members use available knowledge effectively (Hollingshead et al. 2011). TM accuracy facilitates matching tasks with people and expertise, leading to better decision quality (Brandon and Hollingshead 2004). It also improves information search and retrieval (Hollingshead 1998) and tacit coordination in collective task performance (Wittenbaum, Stasser, and Merry 1996). TM consensus contributes to reducing misleading cues in communication and decision making, which in turn improves performance (Bunderson 2003). Consensus may evoke a process of specialization through which team members become accountable for encoding, storing, and sharing information in their areas of expertise, making the entire team's knowledge system more efficient (Wegner 1987). These mechanisms suggest that TM accuracy and consensus are positively related to team performance.

Hypothesis 2: TM accuracy is positively associated with team performance in ICTs.

Hypothesis 3: TM consensus is positively associated with team performance in ICTs.

Although theories suggest that TM accuracy and consensus both contribute to team performance, evidence supporting such arguments is limited (Ren and Argote 2011). We argue that the effects of TM accuracy and consensus may be contingent on team knowledge structure. TM accuracy and consensus are more likely to affect team performance in situations where teams have complex knowledge structures such as when highly specialized and interdependent areas of expertise coexist. Expertise redundancy complicates a team's knowledge structure because it reduces role clarity and causes members to use non‐task‐specific cues to identify sources of expertise (Bunderson 2003). Expertise redundancy may be more detrimental in decision making teams where information is used to influence decisions and the act of experts is more understated than that of experts in task‐performing teams. Teams in such situations have a heightened need to use TM to facilitate information search and retrieval. Therefore, we expect that TM accuracy and consensus should have an interactive effect with expertise redundancy on team performance.

Hypothesis 4: The positive association between TM accuracy and team performance is stronger when expertise redundancy is higher.

Hypothesis 5: The positive association between TM consensus and team performance is stronger when expertise redundancy is higher.

Methods

The study design is a longitudinal, observational design using survey and administrative data. The unit of analysis is the team.

Setting

Data came from ACT teams in Minnesota. ACT is an evidence‐based practice that uses interdisciplinary teams to provide assertive outreach and constant care to clients with severe and chronic mental illness (Stein and Santos 1998). An ACT team typically consists of one team leader, one psychiatrist, two or more nurses, two or more substance abuse specialists, one or more vocational specialist, and several other service staff (Teague, Bond, and Drake 1998; Witheridge 2010). ACT teams have an imprecise and dynamic role structure because all members, despite their professional backgrounds and titles, are expected to perform a common set of tasks, including client assessment, treatment planning, and in situ interventions, and to have certain knowledge about key areas of community‐based mental health services (Stein and Santos 1998).

Sample and Data Collection

Our sample included all 27 ACT teams in Minnesota, of which 26 teams participated in the study. We collected data using a longitudinal survey design. Respondents completed a survey assessing every team member's expertise at baseline, 6, and 12 months. Team leaders provided additional information on team characteristics and context.

Surveys were distributed to 318, 309, and 304 team members and leaders in the three waves, respectively. We received 287 completed surveys in wave 1, 268 in wave 2, and 275 in wave 3 (response rates = 90 percent, 87 percent, and 91 percent). The gender composition of the sample was 71 percent female and 29 percent male. The racial composition was 91 percent Caucasian, 5 percent Asian, 2 percent African American, and 1 percent others. The role composition was 8 percent team leader, 9 percent psychiatrist, 18 percent nurse, 15 percent substance abuse specialist, 11 percent vocational specialist, and 39 percent others.

Administrative data were used to measure team fidelity to the program standard and team performance. First, the state's Department of Human Services (DHS) assessed team fidelity using a scale modeled after the national ACT standard (Teague, Bond, and Drake 1998; Witheridge 2010). Second, we evaluated team performance for the 6 months after each survey using quarterly client‐outcome data collected by DHS. The research team analyzed team performance by regressing client outcomes on risk factors (diagnosis, age, gender, race, new client) and a team indicator using general linear models. Coefficients of the team indicator (i.e., team effect size) provided a team‐level performance measure. This analysis was conducted at a secure DHS facility with assistance of DHS staff to protect client confidentiality.

Measures

Respondents were asked “Which of your ACT team members have a lot of expertise in the following domains (Include an answer for yourself)?” Eight knowledge domains and a team roster were listed as columns and rows of a table following the question. The eight knowledge domains included psychiatry/medicine, nursing, substance abuse, vocational rehabilitation, courts/civil commitment, housing/subsidies, public assistance, and team coordination. This survey instrument was developed based on the literature and our extensive presurvey fieldwork. During the fieldwork, we observed ACT teams' daily work routines, consulted ACT experts and team leaders, and conducted cognitive testing in two teams to identify what knowledge domains were essential for ACT teams' work. These eight knowledge domains were recognized by ACT experts and practitioners as the essential domains. With this instrument, respondents could identify a team member's expertise in multiple domains. The response captured team member i's perception of team member j's expertise in knowledge domain k on a binary scale (0 = j does not have a lot of expertise and 1 = j has a lot of expertise), which was used to construct expertise redundancy and transactive memory measures (see Appendix SA2 for measurement details).

Expertise Redundancy

Expertise redundancy was measured as the proportion of pairs of team members possessing the same expertise in a knowledge domain, which was then averaged across all knowledge domains. The measure is adjusted for team size by dividing the total number of pairwise overlaps in a given domain by the total number of pairwise comparisons conducted within the team. For a given team size, this measure increases exponentially as the number of team members possessing the same expertise increases, a pattern that captures the idea of excessive knowledge overlap.

Transactive Memory Accuracy

We first calculated an individual accuracy score as the proportion of a team member's perceptions of other members' expertise that matched other members' self‐assessments. After testing the within‐team agreement (ICC = 0.32, 0.22, and 0.26), this measure was averaged across all team members to measure TM accuracy. For sensitivity analysis, an alternative accuracy measure was constructed by comparing a team member's perception with the team's assessment of other members' expertise. We used at least 50 percent team members identifying someone as an expert in a domain as the threshold for team assessment.

Transactive Memory Consensus

Drawing on cultural consensus theory (Romney, Weller, and Batchelder 1986), TM consensus was measured as the average correlation among all pairs of team members on their perceptions of a target member's expertise, which was then averaged across all target members.

Team Performance

The most consistent indicator of well‐performing ACT teams is the reduction of mental health–related hospitalization (Olfson 1990; Burns and Santos 1995). We measured team performance by the risk‐adjusted average number of hospitalization for mental illness among all clients served by an ACT team in the 6 months following a survey wave. We subtracted this measure from one so that a higher value indicates better performance.

Control Variables

We controlled for several factors that may affect both our independent and dependent variables, including team fidelity, team size, number of clients, organizational support, gender diversity, and ethnic diversity. Fidelity is a term describing how well an intervention or program is implemented according to a pre‐established standard (Mowbray et al. 2003). For ACT teams, team fidelity is measured as a composite score assessing how closely a team's structure and procedures adhere to the national ACT standard (1 = low fidelity to 5 = high fidelity; Witheridge 2010). This measure included 18 criteria about ACT teams' structure (e.g., size, composition, and caseload) and procedures (e.g., admission criteria, meeting frequency, diagnostic procedure, service planning, documentation, and reporting). Research shows that fidelity to the program standard is associated with better client outcomes (Olfson 1990; Burns and Santos 1995). Team size was measured by the number of team members. Number of clients was measured as the total number of unique clients a team served during the 6 months following a survey wave. An organizational context that is supportive and provides autonomy can affect team operation and performance (Wageman, Hackman, and Lehman 2005). We measured organizational support using the team leader's response to a seven‐item scale that assessed how much a team could obtain support from its parent organization and was involvement in making budget, financial, and management decisions (1 = Not at all to 4 = A lot; Cronbach's α = 0.70). Gender diversity and ethnic diversity were measured as Blau's indices ($1-\sum p_k^2$, where p is the proportion of team members in kth category) based on team members' gender and ethnicity (Blau 1977; Harrison and Klein 2007).

Statistical Analysis

Because the teams were observed repeatedly over time, we have a panel of 78 observations (26 teams × 3 waves). All variables except team fidelity changed over time. We conducted a variance‐component analysis of our key variables to estimate the proportion of the total variance that was between teams. The intra‐class correlations for team performance, expertise redundancy, TM accuracy, and TM consensus were 0.46, 0.45, 0.21, and 0.24 respectively, which suggested that variations in these variables were due to both between‐team differences and within‐team changes. To account for within‐team dependence, we included a random team intercept in our statistical models and estimated multilevel mixed‐effects linear regression models using the XTMIXED module in Stata 12 (Rabe‐Hesketh and Skrondal 2012).

Results

Table 1 presents descriptive statistics and correlations among all variables. The data suggest the following: (1) teams' mean performance score was 0.8 (i.e., the average number of hospitalization was 0.2), and noticeable variation existed; (2) the average expertise redundancy is 0.12, which corresponds to approximately 38 percent average overlap in the given domains (based on the national ACT standard, we estimate that 25–30 percent average overlap is prescribed in the team model); and (3) 73 percent of team members' perceptions of others' expertise match others' self‐assessments. The low expertise redundancy and high TM accuracy scores might be a result of teams' adherence to the ACT standard (mean fidelity: 4.15) that specifies what types of professionals need to be present on the team and requires team members to perform highly interdependent tasks. TM accuracy is negatively associated with expertise redundancy (r = −0.37, p < .001) and positively associated with TM consensus (r = 0.58, p < .001).

Table 1.

Descriptive Statistics and Correlations

Variable	Mean	SD	Min	Max	Correlation
					1	2	3	4	5	6	7	8	9
1. Team performance	0.80	0.11	0.34	0.98	–
2. Expertise redundancy	0.12	0.06	0.04	0.37	−0.39***	–
3. TM accuracy	0.73	0.04	0.59	0.83	0.24*	−0.37***	–
4. TM consensus	0.46	0.07	0.28	0.60	−0.02	−0.17	0.58***	–
5. Team fidelity	4.15	0.19	3.79	4.50	0.23*	0.01	−0.05	−0.36**	–
6. Team size	11.9	2.49	8	18	0.08	0.17	0.07	−0.01	0.29*	–
7. Number of clients	73.4	19.5	39	109	0.02	0.36 **	−0.07	0.09	0.10	0.72***	–
8. Organizational support	2.79	0.66	1.43	4.00	0.37***	−0.36**	0.34**	0.29*	0.11	0.14	−0.01	–
9. Gender diversity	0.38	0.12	0	0.50	0.19	−0.12	−0.03	−0.06	0.13	0.07	−0.08	0.04	–
10. Ethnic diversity	0.12	0.13	0	0.51	0.03	0.20	−0.35**	−0.20	−0.13	0.11	0.26*	−0.14	−0.18

N = 78.

*p < .05, **p < .01, ***p < .001; two‐tailed.

Table 2 presents results of the regression analyses. Model 1 examines the effects of control variables. Results show that teams with better organizational support (β = 0.04, p < .05) and gender diversity (β = 0.22, p < .05) performed better. Model 2 examines the main effects of knowledge structural variables. After adjustment of other factors, teams with higher expertise redundancy had lower team performance (β = −0.44, p < .05), which supports Hypothesis 1. Teams with higher TM accuracy had higher performance (β = 0.65, p < .05), which supports Hypothesis 2. Hypothesis 3 regarding the main effect of TM consensus is not supported.

Table 2.

Results of Multilevel Mixed‐Effects Model for Team Performance

	Team Performance
	Model 1: Control‐Only		Model 2: Main Effect		Model 3: Interactive Effect
	β	SE	β	SE	β	SE
Intercept	0.13	0.32	−0.05	0.35	1.15**	0.38
Survey wave
Wave 2	0.02	0.02	0.01	0.02	−0.00	0.02
Wave 3	0.03	0.03	0.01	0.03	−0.01	0.02
Control variable
Team fidelity	0.11	0.08	0.09	0.07	0.15*	0.06
Team size	−0.00	0.01	−0.01	0.01	−0.01	0.01
Number of clients	0.00	0.00	0.00	0.00	0.00	0.00
Organizational support	0.04*	0.02	0.03	0.02	0.05**	0.02
Gender diversity	0.22*	0.11	0.21*	0.10	0.25**	0.09
Ethnic diversity	0.09	0.11	0.13	0.10	0.17	0.09
Knowledge structure
Expertise redundancy			−0.44*	0.19	−13.07***	2.26
TM accuracy			0.65*	0.30	−1.64***	0.46
TM consensus			−0.35	0.21	0.05	0.37
Expertise redundancy × TM accuracy					18.65***	3.15
Expertise redundancy × TM consensus					−1.47	2.10
Fit index
Log likelihood	75.21		81.12		95.62
$R_{\mathit{LR}}^2$	0.13		0.25		0.48

N = 78.

*p < .05, **p < .01, ***p < .001; two‐tailed.

Model 3 examines the interactive effects of expertise redundancy and TM accuracy and consensus. Results show that expertise redundancy and TM accuracy have a positive interactive effect (β = 18.65, p < .001), which supports Hypothesis 4. Figure 1 illustrates this interactive effect by showing the marginal effect of TM accuracy on team performance in low‐ and high‐redundancy situations while controlling other covariates at means. When expertise redundancy is low (25th percentile), increasing TM accuracy does not affect team performance. When expertise redundancy is high (75th percentile), increasing TM accuracy significantly improves team performance. We find no support for Hypothesis 5. Figure 2 shows that the effect of TM consensus on team performance remains virtually flat in low‐ and high‐redundancy situations. Among the control variables, team fidelity, organizational support, and gender diversity are positively associated with team performance.

Figure 1.

Expertise Redundancy and Transactive Memory Accuracy Interactive Effect on Team Performance

Figure 2.

Expertise Redundancy and Transactive Memory Consensus Interactive Effect on Team Performance

We performed several postestimation tests. First, we evaluated multi‐collinearity among independent variables using variance inflation factors and found no severe multi‐collinearity. Second, to account for within‐team dependence, we estimated two sets of regression models: one with a fixed team effect and the other with a random team effect. The Hausman test (χ ² = 7.96, p = .72) suggested that there was no correlation between team‐specific errors and the independent variables, which supported the random‐effect models as reported. Third, based on model estimates, team‐specific intercepts varied significantly from the overall intercept. In Model 3, the standard deviation of the random intercept was 0.04 and team‐specific intercepts ranged from 1.07 to 1.22. The Breusch–Pagan Lagrange multiplier test (χ ² = 5.58, p = .01) also supported the inclusion of the random team intercept. Fourth, we computed a goodness‐of‐fit index R ² based on likelihood ratio (Magee 1990; Kramer 2005). Comparing the fit indices across models, the fixed effects in the control‐only model explained approximately 13 percent of variation in team performance that is not explained by the team random effect. The models with knowledge structural variables and their interactions improve the model fit to 25 percent and 48 percent.

We conducted a series of sensitivity analyses to test robustness of our results. First, we examined the reliability of using one's self‐assessment as the benchmark for calculating others' TM accuracy. On average, ACT team members self‐identified as expert in 2.7 knowledge domains. The self‐assessments were relatively stable over time: 69 percent of cases reported no change in expertise over three waves, 22 percent had one change, and 9 percent had two changes. The self‐assessments matched the team assessments 81 percent of time. Using the team assessments to calculate alternative TM accuracy measures produced similar analytical results.

Second, to measure the knowledge structure in ACT teams, our survey instrument included a diverse set of knowledge domains recognized by ACT practitioners as essential for their work. Two knowledge domains (i.e., psychiatry/medicine and nursing) coincided with established professional roles more than other domains did; and one domain (i.e., team coordination) corresponded to teamwork while other domains corresponded to taskwork. The diverse nature of knowledge domains may affect the distribution of expertise redundancy and TM measures and their effects on team performance. Thus, we constructed two sets of alternative expertise redundancy and TM measures: one without psychiatry/medicine and nursing and the other without team coordination as knowledge domains. Removing psychiatry/medicine and nursing from the measurement increased mean expertise redundancy from 0.12 to 0.15, decreased mean TM accuracy from 0.73 to 0.69, and had no effect on mean TM consensus. Removing teamwork decreased mean expertise redundancy from 0.12 to 0.11, increased mean TM accuracy from 0.73 to 0.74, and had no effect on mean TM consensus. We estimated both the main effect and interactive effect models using the two sets of alternative measures and found very similar results. Table 3 presents results from the interactive effect model. Using the alternative measures in regression analyses slightly changed the coefficients of the observed effects, but it did not change the direction or significance of the effects.

Table 3.

Sensitivity Analysis Results

	Team Performance
	Psychiatry/Medicine and Nursing Removed from the Measures		Team Coordination Removed from the Measures
	β	SE	β	SE
Intercept	1.14**	0.34	1.17**	0.37
Survey wave
Wave 2	0.00	0.02	0.00	0.02
Wave 3	−0.00	0.02	0.01	0.02
Control variable
Team fidelity	0.17**	0.06	0.13	0.07
Team size	−0.01	0.01	−0.01	0.01
Number of clients	0.00	0.00	0.00	0.00
Organizational support	0.06***	0.02	0.05**	0.02
Gender diversity	0.22**	0.08	0.24**	0.09
Ethnic diversity	0.14	0.08	0.15	0.09
Knowledge structure
Expertise redundancy	−11.55***	1.95	−12.65***	2.27
TM accuracy	−1.80***	0.40	−1.58***	0.45
TM consensus	−0.01	0.36	0.10	0.36
Expertise redundancy × TM accuracy	15.80***	2.50	18.33***	3.16
Expertise redundancy × TM consensus	0.79	2.11	−2.13	2.06
Fit index
Log likelihood	95.80		95.35
$R_{\mathit{LR}}^2$	0.49		0.48

N = 78.

*p < .05, **p < .01, ***p < .001; two‐tailed.

Lastly, we examined whether our results were influenced by outlier observations. We examined bivariate scatterplots and identified one potential outlier with very low performance, very high expertise redundancy, and below average TM accuracy. We redid the analyses after removing the potential outlier. The results were similar to the reported models except that the main effects of expertise redundancy and TM accuracy were only marginally significant. The sensitivity analyses suggest that our results were robust.

Discussion

In this study, we examined how expertise redundancy and TM interactively affect team performance in ACT teams. We found that expertise redundancy had a negative effect and TM accuracy had a positive effect on team performance. Supporting our argument that TM accuracy is more likely to improve team performance when teams have a complex knowledge structure, our results showed that the positive association between TM accuracy and team performance is stronger when expertise redundancy is higher.

We did not find support for the hypothesized effects of TM consensus. This might be caused by a high correlation between TM accuracy and consensus. We conducted additional analyses to model the effects of TM accuracy and consensus separately. In the consensus‐only models, we did not find support for the main or interactive effect of TM consensus. This suggests that although accuracy and consensus are both properties of knowledge‐related team cognition, accuracy is a stronger predictor of team performance. Consistent with the clinical evidence on which our sample teams were established and the team design literature in general, we found that ACT teams with higher program fidelity (i.e., adherence to team design features), greater organizational support, and greater gender diversity performed better.

While addressing knowledge management challenges in ICTs, the findings also enrich the team literature. Our results suggest that the study of knowledge diversity might benefit from considering team processes or emergent states that are directly related to knowledge processing and utilization to resolve some of the inconsistencies in earlier findings. More specifically, we showed that TM accuracy helps teams with complex knowledge structures to produce superior performance.

Implications for Policy, Delivery, or Practice

Interdisciplinary care teams are increasingly utilized in care delivery. This study offers several policy and practical implications for designing and implementing ICTs. First, from a policy standpoint, explicit team design is crucial for delivering high‐quality care. Our finding suggests that, without effective knowledge coordination, more expertise is not always beneficial. For ICTs, it is important to specify team knowledge structure as clearly as possible. This specification can help ICTs minimize unnecessary knowledge overlap. If some redundancy is required for team performance, team design should specify coordination mechanisms to facilitate experts to communicate and agree on their recommendations and to help other members navigate the complex knowledge structure. For example, the ACT model requires frequent team meetings to develop agreement. Explicit and clinically designed team models are still rare in health care. Policy makers should support and incentivize the clinical field to design, test, and implement such team models.

Second, from a practical standpoint, a team's knowledge structure is a dynamic feature that changes as members practice and learn from each other. Thus, some expertise redundancy may be unavoidable. Our finding identifies one condition under which expertise redundancy in ICTs is likely to produce performance benefits; namely, accurate TM. Practically, managers and team leaders can leverage the benefits and manage the liabilities of expertise redundancy by fostering more accurate recognition of member expertise or implementing interventions when expert roles become blurry or mistakenly perceived. For example, Lewis et al. (2007) found that when teams were instructed to reflect upon their collective knowledge basis before performing tasks, they enacted a more elaborated cognitive representation of member expertise and subsequently performed better. This is particularly practical when team membership changes (e.g., during shift changes or when staff turnover occurs). Team leaders could also explicitly assign roles to team members to signify his or her expert status during task performance, which can rectify the misleading status cues (e.g., age and tenure) that are irrelevant to task performance (Bunderson 2003). Prior research showed that professional roles were important for expertise recognition because team members were more accurate in identifying experts in knowledge domains associated with highly professionalized roles (Zhu and Wholey 2015).

Third, the negative association between expertise redundancy and TM accuracy suggests that having a highly overlapping knowledge structure may increase the difficulty for team members to apprehend who is expert in what area. Managers may need to simplify teams' knowledge structure and differentiate team members' expert roles in order to facilitate the development of accurate TM systems. In fact, TM theory argues that the most efficient knowledge system would be one with minimal redundancy in which highly specialized individual memory systems are accompanied by effective communication. However, an extremely diversified knowledge system (e.g., a system with no expertise redundancy at all) may not be practical given that transdisciplinary learning may occur in many ICT settings. In such settings, managers need to ensure that expertise redundancy and transdisciplinary learning do not interfere with role structures and task performance.

Limitations and Future Research Directions

This research has several limitations that lead to new avenues of research. First, although we observed and collected data from the ACT teams over three waves, our analysis showed that there were significant variations between teams and waves, but no systematic changes over time. In other words, the data contain more cross‐sectional information than longitudinal information. Due to the cross‐sectional nature of the data, we cannot establish causal inferences about the relationships between expertise redundancy, transactive memory, and team performance.

Second, we examined the hypotheses in a context where teams implemented a highly specified team model. The teams were well bounded, and their knowledge domains were clearly defined. Such characteristics are not present in all ICTs. Many ICTs are ad hoc teams that may have less‐specified knowledge structure, leading to different team dynamics and development of different team cognitive structures. Another contextual limitation is that ACT teams are decision–making teams where the effects of expertise redundancy and TM may be more salient. In ICTs that focus on co‐performing tasks (e.g., surgical teams), expertise redundancy may be beneficial for assisting or backing up the primary task performer and TM may be less critical because in such teams, the act of using expert knowledge/skills is more visible to other members. The single research context reduced the generalizability of our findings. Future research is needed to test the relationships between knowledge composition, TM, and performance in other team settings.

Third, team members' professional roles are important status cues for recognizing their expertise. At the team level, formal role composition can affect the development and function of TM systems. However, in the ACT teams that we observed, there was low variation in formal role composition—all teams complied with the role composition criteria specified in the national standard. This limited our ability to test the effects of formal role composition on TM and performance. However, ACT teams exhibited significant variation in the informal expert role structure as captured in our expertise redundancy measure. Future research needs to explore how formal roles influence the development and recognition of informal roles to offer more practical insights on how to manage the dynamic role systems in ICTs.

Conclusions

Our study indicates that, in ACT teams, expertise redundancy and TM accuracy interactively affect team performance. Overall, teams achieved better client outcomes when teams had low expertise redundancy and when team members could accurately recognize who was an expert in specific knowledge domains. Further, TM accuracy had a stronger positive relationship with team performance when expertise redundancy was higher, indicating that it could serve as a cognitive coordination mechanism for mitigating the negative effect of complex knowledge structure. ICTs need to use team design and management strategies to reduce expertise redundancy and foster more accurate transactive memory in order to achieve superior performance.

Supporting information

Appendix SA1: Author Matrix.

Appendix SA2: Expertise Redundancy and Transactive Memory Measures.