Effect of Cross-Departmental Collaboration on Performance: Evidence from the Federal Highway Administration

Abstract

Cross-departmental collaboration, one of the most salient administrative reforms, has been promoted to resolve cross-jurisdictional administration issues over the previous three decades. Nearly all previous empirical studies have examined the direct impact of cross-departmental collaboration on organizational performance without accounting for the indirect effects of managerial practices. Using data from the Federal Highway Administration, this study develops an integrated structural equation modeling and Bayesian network model used to examine both direct and indirect impacts of cross-departmental collaboration on organizational performance. The structural model indicates that cross-departmental collaboration has a direct effect on organizational performance and indirect effects through its influence on resource acquisition and knowledge creation. The scenario-based simulation suggests the optimal integration of managerial actions to improve agency performance, which is achieved by encouraging cross-departmental collaboration and supporting the knowledge creation process. Finally, implications are provided to present practical managerial actions from the Federal Highway Administration as an exemplar for other highway agencies.

1. Introduction

Highway agencies confront long-term governance issues and challenges under the federalist system due to cross-jurisdictional networks and organizational silos [1]. Since 1990, federal transport law has changed focus from highway-centered planning governed by state transportation authorities to a more collaborative approach and intermodal network. Federal regulations also mandate state agencies to collaboratively develop transport improvement plans identifying approved projects in each fiscal year [2]. Highway planning includes networks of transportation departments participating in collaborative decision-making to resolve cross-jurisdictional administration issues. The interdepartmental networks are usually a formal structure with designated members operating within the current governance structure [2,3]. Multijurisdictional coordination can solve multi-sectoral issues in highway management, especially megaregional planning [4], disaster recovery [5], and operational resiliency [6].

Cross-departmental collaboration, one of the most salient administrative reforms, is being adopted to enhance the federal departments’ organizational performance [7]. Because highway systems transverse jurisdictional boundaries and modal networks, highway agencies’ cross-departmental collaboration is multi-directional coordination through geographic and political boundaries, consisting of horizontal and vertical collaboration [8]. Horizontal collaboration is coordination between agencies on the same scale concerning geographical adjacency and proximity. In contrast, vertical collaboration entails multi-level connections between federal, state, regional, and local jurisdictions. Cross-departmental collaborations in both horizontal and vertical approaches develop collaborative networks for highway management and operation. These networks could respond to unexpected accidents, manage highway assets, and integrate highway policies, ultimately resulting in improving organizational performance.

Nearly all previous empirical studies have examined the direct impact of cross-departmental collaboration on organizational performance [3,9,10,11]. However, there has been limited research that accounts for the indirect effects of cross-departmental collaboration through the role of managerial practices. This study enables the investigation of the causal pathways between resource acquisition and knowledge creation through which cross-departmental collaboration leads to organizational performance improvements. Resource acquisition is grounded on the resource dependence theory (RDT) and the resource-based view (RBV), which are well-established organizational theories. The RDT was adopted to empirically investigate mutual dependence between organizations, resulting in valuable insights and leveraging more understanding of the organization [12]. The RDT states that organizations opt to collaborate in exchange relationships due to the requirement for critical resources [13]. The RBV was used as a theoretical lens to understand resource acquisition mechanisms to sustain performance. The RBV highlights that an organization’s sustainable advantage is derived from the strategic resources that the organization possesses [14]. Nonaka’s theory was adopted to frame knowledge creation. From a process viewpoint, knowledge creation is defined as an organization’s ability to create new knowledge, transfer it within the organization, and incorporate it into products, services, and processes [15].

Furthermore, no research to date has empirically established the causal relationships between these constructs using a sample from a highway agency. It is vital to highlight that a lack of theoretical and empirical study on causality among these organizational constructs provides the scientific problem investigated in this research. Therefore, the purpose of this paper is to address this research gap by exploring the direct impacts of cross-departmental collaboration on organizational performance and unraveling the organizational mechanism of indirect effects through resource acquisition and knowledge creation in the Federal Highway Administration (FHWA).

2. Conceptual Model

2.1. Theory

2.1.1. Cross-Departmental Collaboration

Cross-departmental collaboration is defined as any joint action across diverse departments that is intended to increase public value and address difficult public challenges through coordination, partnering, conflict resolution, and cooperation [3,11]. In rapidly changing public services, cross-departmental collaboration has become an essential activity for each agency to implement. Cross-departmental collaboration is a process that encourages other related departments to work together in joint activities to increase public value and improve governance. In this study, the authors use the terms “cross-departmental collaboration” and “interdepartmental collaboration” interchangeably throughout the manuscript.

The characteristics of interdepartmental collaboration can be either dyads or two-party interactions between department levels within government agencies or at the inter-organizational level [9]. To solve wicked problems that cannot be solved through the traditional approach, collaboration is crucial to deal with these complicated social problems. In the past, cross-departmental collaboration was considered an informal or coincidental activity, and at best, it was an administrative experiment [10]. Currently, cross-departmental collaboration has been developed and accepted as a common concept. For example, in the United States, interdepartmental collaboration has become the gold standard in public management. Furthermore, federal, state, and local departments need to coordinate under collaborative governance to receive financial support for the public service program [9].

2.1.2. Resource Acquisition

Resource acquisition refers to how an organization acquires tangible and intangible resources from internal and external environments [16]. This study adopts two concepts of resource acquisition: resource dependence theory (RDT) and resource-based view (RBV). The RDT proposes that, to succeed, organizations depend on critical resources from the external environment to maintain their existence and development [13]. The solution to reducing dependency is to collaborate with other organizations to secure the needed critical resource. From another perspective, the RBV theory states that a subset of the resources enables the organization to achieve effectiveness and contributes to long-term efficiency. The RBV theory highlights the significance of the ability to exchange the information and knowledge systems with collaborative partners [17]. Organizations should sustain the critical resources of tangible assets (e.g., capital, machinery, and workforce) and intangible assets (e.g., intellectual capital, social resources, technology, capabilities, and competencies). Several methods can be applied to acquire resources, including resource attraction, resource purchase, and external resource sharing [16].

2.1.3. Knowledge Creation

Knowledge creation is defined as a dialectical process, where various contradictions are integrated through dynamic and interactive synergy among employees, the organizations, and the working environment [15]. Knowledge creation is a generative process and involves accumulating organizational knowledge stock through collecting knowledge from internal and external sources through individual learning. Each member interacts with other members by synthesizing existing knowledge and sharing knowledge and lessons learned through collaborative networks. Knowledge creation can be principally categorized into two perspectives. The first perspective, called a stock view, means the activities and initiatives employees engage in accumulate their organizational knowledge stock [18]. Another perspective is defined as a process in which an organization can collectively generate novel knowledge; distribute it within the organization; and integrate it with products, systems, and services. In the knowledge-creating theory, Nonaka and Toyama [15] divide knowledge into two categories: explicit and tacit. Explicit knowledge is the documented knowledge in textbooks, procedural manuals, and contract documents (e.g., drawings, specifications, and conditions of contract). This type of knowledge is easy to codify, capture, and disseminate [19]. In contrast, tacit knowledge is the skill, expertise, intuition, and judgment of an individual accumulated from work experience, which is not simple to exchange among the organization members. Knowledge is created through a synthesized process, occurring from repeated interactions between explicit and tacit knowledge.

2.1.4. Organizational Performance

Delivering efficient and effective public service is a critical organizational outcome of government agencies. The organizational outcomes can be measured by perceived organizational performance in terms of an organization’s ability to effectively achieve its objectives and missions through work quality [7]. Measuring perceived performance is appropriate for this study because the perception of an employee, acting as an internal constituency, can further the understanding of actual organizational performance as they have direct experience with their organization’s mission accomplishments. Since the administration’s goal is to enhance organizational performance, operating results can provide guidance for executives to improve public service. Therefore, many government agencies have deployed organizational performance as the focus of the agencies’ agenda.

2.2. Hypothesis Development

We ask the following research question: what is the interplay between cross-departmental collaboration, resource acquisition, and knowledge creation in improving organizational performance in the FHWA? Hypotheses have been formulated to answer this research question and to confirm the conjectured relationships.

2.2.1. Cross-Collaboration and Resource Acquisition

Highway agencies strive to receive funding through the mechanism of annual budgets and the authorization of permanent legislation [20]. Contracts are a common and formal tool for highway agencies to obtain necessary resources. They are intended to minimize future risks that may exist in the process of an exchange agreement. Nevertheless, highway agencies’ success is also directly or indirectly related to their capability to acquire essential resources from collaborative partners. The informal approach of applying more collaborative strategies, like alliance, co-optation, and integration, can help agencies seek more profound commitment from partners [21]. From the RDT perspective, resource dependence between highway agencies and their partners has formed a self-reinforcing structure of mutual power and increased pragmatic benefits, creating network ties to improve resource acquisition.

Because the highway agencies’ public services are policy, planning, operation, and maintenance of highway networks, both tangible and intangible resources are primary factors for organizations’ efficiency and effectiveness [22]. Organizations possess a different set of tangible and intangible resources that uniquely present their distinctive identities. These organizations can acquire intangible assets through network ties based on joint activities between agencies, so they do not usually need a contribution of tangible resources (e.g., finance, machines, and materials). Practice-based collaboration through social and informal networks, supported by the leaders, results in an accumulation of intangible resources such as technology transfer between agencies [16]. The outside-in capabilities of agencies also include exchanging diverse competencies and skill sets between informal networks, so these networks are important channels for agencies to acquire resources [23]. Thus, based on previous findings, cross-departmental collaboration supports resource acquisition from mutual relationships between departments.

Hypothesis 1 (H1).

Promoting cross-departmental collaboration will increase resource acquisition.

2.2.2. Cross-Departmental Collaboration and Knowledge Creation

Interdepartmental collaboration is a primary factor in supporting employees’ access to diverse ideas, various information sources, and broader knowledge [24]. Such collaborations equip the employee with in-depth knowledge and experience-based skill. These activities help create knowledge repositories and improve job-related skills, leading to the accumulation of innovative knowledge that spans across inter-organizational boundaries [25,26]. Knowledge sharing also occurs through collaboration between departments, enabling improved organizational learning capability.

Because highway planning involves multi-jurisdictional cooperation, highway agencies are required to expand their duties beyond a geographic proximity, making their boundaries increasingly blurred [7,27]. Therefore, the collaboration between organizational boundaries is an effective learning channel for knowledge creation among the various parties in the provision of highway services [19]. This interaction encourages engineers to acquire knowledge and skills related to the duties for which they are responsible. Previous literature has led to the assumption that a high level of cross-departmental collaboration supports knowledge creation in an organization.

Hypothesis 2 (H2).

Promoting cross-departmental collaboration will increase knowledge creation.

2.2.3. Resource Acquisition and Knowledge Creation

RDT posits that organizations need to garner necessary resources from their collaborative networks to maintain operation and development [13]. Ding and Huang [28] revealed that sufficient resource sharing positively impacts collaborative knowledge creation. The success of collaborative knowledge creation depends on adequate and timely allocation and sharing of resources. Knowledge creation is recognized as a resource-sharing process through interdepartmental collaboration between individuals and divisions and as extra-organizational entities formed as dyads or networks [14].

In the built environment, Goh and Loosemore [29] asserted that the critical intangible asset is social resources, which are agencies’ connections to their collaborative networks. Thus, agencies’ connections with other federal authorities, state agencies, and municipalities are crucial channels for engineers to engage in the knowledge creation process [17,20]. Engineers can create technical knowledge of highway planning based on legislation relevant to federal, state, and local standards [30]. Because engineers can acquire knowledge from collaborative networks through highway agencies’ connections, they can create new knowledge based on highway agencies’ existing knowledge [19]. The newly created knowledge can be tacit (e.g., experience, judgment) or codified (e.g., drawings, specifications, and manuals). This mechanism can help them share a lesson learned from projects, which leads to the development of engineers’ technical abilities [14]. According to these studies, resource acquisition is hypothesized to be directly related to knowledge creation.

Hypothesis 3 (H3).

Resource acquisition has a positive impact on knowledge creation.

2.2.4. Cross-Departmental Collaboration and Organizational Performance

Many benefits arising from cross-departmental collaboration support the achievement of public policy goals. The collaborative network has access to resources among various organizations; thus, they can collaborate to overcome budget constraints. Another advantage of collaboration is risk sharing, especially in highway construction projects, which inherently have high-impact risk factors [27]. The benefits of departmental collaboration could significantly impact the performance of public organizations, as the transition would be more radical than in less bureaucratic and hierarchical private companies [7].

Cross-departmental collaboration is considered a means to enhance an organization’s effectiveness and efficiency, leading to improved organizational performance [10]. To present transportation evidence, Margerum and Parker [2] examined a multilevel network of Area Commissions on Transportation, established for transportation planning and management in Oregon. These collaborative networks encouraged cooperation and communication to solve interjurisdictional issues in transportation investment, which in turn improved agencies’ performance.

Hypothesis 4 (H4).

Cross-departmental collaboration has a positive effect on organizational performance.

2.2.5. Knowledge Creation and Organizational Performance

There are several empirical studies on the positive impact of knowledge creation on organizational performance in the construction industry. Based on a survey of small- and medium-sized construction enterprises in Malaysia, knowledge creation, contributing to inter-organizational learning through the established external links of inter-organizational relationships, had a significant impact on these enterprises’ performance [31]. In a study of Singaporean construction companies, Pheng [32] also examined the relationships between organizational learning, construction productivity, and performance. They found that organizational learning significantly impacts construction productivity, which increases competitive advantages and improves company performance.

In highway agencies, significant construction delays are caused by external issues, such as the unsolved acquisition of land, increased prices of construction materials, limited coordination and knowledge sharing among parties, and public utility relocation [19,33]. Engineers who participate in the knowledge creation process tend to possess more developed skills and an increased depth and range of knowledge, equipping them with the critical thinking necessary to effectively solve such delays [19,34]. Alashwal et al. [19] conducted a case study of highway projects in Malaysia to develop the knowledge utilization process that can be applied to manage project delay. The empirical research presented how engineers create and utilize new knowledge to develop solutions to resolve delay issues, consecutively improving project performance. In the public sector context, Ngah, Tai, and Bontis [35], in a study of the Roads and Transport Authority of Dubai, indicated that knowledge creation, as the component of knowledge management capability, is the critical antecedent variable of organizational learning and improves organizational performance. Thus, knowledge creation is expected to positively enhance organizational performance.

Hypothesis 5 (H5).

Knowledge creation has a positive impact on organizational performance.

2.2.6. Resource Acquisition and Organizational Performance

The RBV argues that resources are acknowledged as a primary driver for competitive advantage if they are unique, valuable, inimitable, and company-specific [36]. Dzeng and Wen [37] revealed construction firms should highlight the significance of the acquisition of tangible resources assets (e.g., plant or construction equipment) and intangible resources (e.g., patented construction methods, in-house geographic data). Both assets interact in specific ways to develop construction companies’ core competencies and distinct identities [29]. Barrett and Sexton [38] argued that construction innovation occurred because of how firms utilized their unique resources to generate innovation. In the proactive resource-push view, which applies the RBV as a theoretical lens, construction firms develop more innovation capability if they innovate because they have sufficient abilities. This approach provides a more solid cornerstone for construction innovation than market-pull orientation, where clients demand innovation implementation [39]. Therefore, acquiring unique and sufficient resources is essential to survive the competition, innovate, and generate high profit, which is crucial for improving firm performance.

Hypothesis 6 (H6).

Resource acquisition has a positive impact on organizational performance.

The theorizing process is based on a deductive approach, which involves a development of hypotheses based upon literature and theories, then followed by hypothesis testing [40]. Prior empirical studies and theoretical literature were reviewed to develop these six hypotheses. The conceptual model was proposed based on the hypothesized relationships, as displayed in Figure 1.

3. Research Methodology

3.1. Study Design and Participants

The survey approach is used because the study aims to assess the causal relationships between organizational constructs by using a questionnaire tool to address a scientific problem. This study used the collected data from the Office of Personnel Management, which conducted an annual employee census, known as the Federal Employee Viewpoint Survey (FEVS). The survey was organized to collect data from officials of diverse backgrounds in the US federal departments. The FEVS employed a stratified sampling method to provide survey results, representing the entire workforce of federal departments and staff within individual departments [41]. The samples are full-time, part-time, permanent, non-seasonal employees of large departments and small or independent agencies. The online questionnaire was administered to measure aspects of attitudes, opinions, and perceptions toward the organization’s life. The questions related to the employees themselves, as well as their managers and their organizations, covering diverse topics regarding work collaboration, resource allocation, knowledge management, and organizational performance. The FEVS covers various aspects of public administration and provides generalizability and representativeness for the US federal agencies. The FEVS dataset has also been used for analysis to publish more than 40 articles related to organizational research studies [42].

The study used the data drawn from the Federal Highway Administration (FHWA), which is an agency under the United States Department of Transportation. The agency manages federal budgets for the National Highway System and supports state and local governments in the highway’s design, construction, and maintenance. The FHWA is headquartered in Washington, DC, and has one division office in each state. There are approximately 3000 employees nationwide, among whom civil engineer is by far the most typical profession, followed by transportation experts [43]. A sample of the 2019 FEVS dataset was selected for data analysis because this dataset was the most recent published for public use. For analysis, responses with missing values in any one of the research items were excluded from the sample, leaving a final sample size of 1535. To briefly summarize respondents’ characteristics, about 66% of the sample were male, while 34% were female. Regarding education level, employees were well educated: 94.4% held a tertiary degree (bachelor’s degree and beyond). As for employment length, 35.7% had job tenure fewer than ten years, 33.1% were between 10 and 20 years, and 31.2% were longer than 20 years. In terms of position, 19.8% were in supervisory positions and 80.2% reported a non-supervisory role. According to the ethnicity, 75.9% were white.

3.2. Measurement

The authors selected the questionnaire items used in this study after an extensive review of all the questions in the 2019 FEVS. Eight questionnaire items were selected and grouped according to the literature of each latent construct. The questions were measured with a 5-point Likert scale (i.e., 1 = “strongly disagree”, 3 = “neutral”, and 5 = “strongly agree”). The exogenous construct is the practice of cross-departmental collaboration (CC). The first question was “Managers promote communication among different work units”. The second item was “Managers support collaboration across work units to accomplish work objectives”. These two items were used to measure interdepartmental collaboration by Lee [7]. Cronbach’s alpha (α) was 0.934.

The first endogenous construct, resource acquisition (RA), was measured using the two questions: “I have sufficient resources (for example, people, materials, and budgets) to get my job done”, and “I have enough information to do my job well”. The reliability alpha for this construct was 0.719.

The second endogenous construct, knowledge creation (KC), was measured using the following two questions: “The workforce has the job-relevant knowledge and skills necessary to accomplish organizational goals”, and “The skill level in my work unit has improved in the past year” (Cronbach alpha = 0.702).

The dependent construct is organizational performance (OP), measured by two questionnaire items: “How would you rate the overall quality of work done by your work units? ” and “My agency is successful at accomplishing its mission”. In prior studies, these questionnaire items were administered to measure organizational performance [7,44]. Cronbach’s alpha was 0.721. All the constructs have Cronbach’s alpha values greater than 0.70, indicating that the item questions are reliable and present uni-dimensionality within their measurement scale [45]. Harman’s single-factor test was conducted to detect the possibility of common source bias in the self-reported survey method [46]. The result indicates that the first factor accounted for less than 50% of the variance. Thus, the common source bias did not suggest any serious concern.

3.3. Research Design

This research adopted a hybrid approach by integrating structural equation modeling (SEM) with the Bayesian network (BN). SEM is a confirmatory technique that examines casual relationships in a conceptual model, and it is appropriate to explain established theoretical relationships from pre-existing knowledge [47]. In contrast, BN is an exploratory technique to provide theoretical explanations by learning the quantitative probabilities from the data [47,48]. This novel approach combines a theoretical construction based on an empirically validated structural model with a graphical interaction’s BN. A conceptual model was formulated to test the hypothesized relationships between the constructs using SEM. Subsequently, the BN, developed based on the structural model’s causal relationships, was utilized as a decision support model.

3.3.1. Structural Equation Modeling

Structural equation modeling is a multivariate statistical technique combining factor analysis and path analysis. This technique allows researchers to develop, test, and confirm causal relationships among constructs, represented by multiple indicators that help address the problem of measure-specific errors [49]. This study adopted a two-step approach that consisted of measurement model validation and structural model assessment, analyzed using AMOS 22.0.

A measurement model is formed by two linear equations that identify the relationship between the latent construct and the observed variable, with terms denoted as follows:

$$x={\Lambda }_x\xi +\delta$$

(1)

$$y={\Lambda }_y\eta +\epsilon$$

(2)

where

$x$ is a column vector of exogenous, or independent, variables

${\Lambda }_x$ is the coefficient matrix of exogenous factor loadings of $x$ on $\xi$

$\xi$ is a vector of the independent latent variables, exogenous variables

$\delta$ is a column of measurement errors in $x$

$y$ is a column vector of endogenous variables

${\Lambda }_y$ is the coefficient matrix of endogenous factor loadings of $y$ on $\eta$

$\eta$ is a vector of latent dependent, or endogenous, variables

$\epsilon$ is a column vector of measurement errors in $y$.

The form of the structural model is expressed by the following equation:

$$\eta =\beta \eta +\Gamma \xi +\zeta$$

(3)

where

$\beta$ is a coefficient matrix of direct effects between endogenous variables

$\Gamma$ is a coefficient matrix of regression effects of the exogenous variables

$\zeta$ is a column vector of the residual error.

3.3.2. Bayesian Network Modeling

Bayesian networks, also called belief networks, are probabilistic graphical models representing joint probability distribution over a set of random variables. The network topology of a BN is represented by a directed acyclic graph (DAG), which is a pair G = (V, E). A set of nodes that represents variables or attributes is denoted by V, and a set of directed edges, represented by E, connects the nodes representing causal relations [50]. The DAG is a graphical representation of a set of nodes and directed edges. In the DAG, when there is a directed edge from node Vi to node Vj, node Vi is called the parent node of node Vj, and node Vj is the descendent of node Vi, also known as a child node [51]. A conditional probability table (CPT) describes each child node’s joint probability distribution, expressing the relationships’ strengths conditioned by combining the parent nodes’ values. If a node has no parent node, it is expressed by a prior probability. The following formula expresses the joint probability distribution of the BN [50]:

$$P(V_1, V_2,\text{…}, V_n)= {{\displaystyle \prod }}_{i=1}^{N}P(V_i|Parent(V_i))$$

(4)

where nodes are random variables denoted as $V_1$, $V_2$,…, $V_n$. A DAG consists of $n$ number of nodes. The set of all random variables $V_i$ with an arc connect node $i$ and $j$ is represented by $Parent(V_i)$.

4. Results

4.1. Structural Model

The first step is to examine the measurement model using confirmatory factor analysis (CFA). The purpose of this step is to consider whether the observable variables (i.e., indicators) combine to represent the latent variables (i.e., construct) and confirm that indicators are hypothesized to measure each construct [52]. The study applied the maximum likelihood approach to conducting the CFA. Furthermore, the study adopted several descriptive goodness-of-fit indices to establish the validity of the model: goodness-of-fit index (GFI), comparative fit index (CFI), Tucker-Lewis index (TLI), incremental-fit index (IFI), standardized root mean square residual (SRMR), and root mean square error of approximation (RMSEA). The fit indices are categorized as absolute fit indices (i.e., Chi-square, GFI, SRMR, and RMSEA) and incremental fit indices (i.e., CFI, TLI, and IFI). Absolute fit indices are a direct measure of the degree to which the hypothesized model fits the observed data [49]. In contrast, comparative fit indices assess how well a hypothesized model fits, comparing with a null model [53]. The Chi-square statistic calculates the difference between a hypothesized model and observed data. The GFI evaluates the proportion of the variance in the sample variance-covariance matrix. The SRMR is calculated from covariance residuals, describing the difference between observed data and the hypothesized model, while the RMSEA assesses the lack of fit to the saturated model. The CFI compares the proportionate improvement in the model fit of the hypothesized model over a null model. The TLI determines a correlation for model complexity. The IFI calculates the chi-square difference between the hypothesized model and a baseline model with uncorrelated variables [54].

To be an acceptable model, all six indices need to pass the following criteria: GFI, CFI, TLI, and IFI > 0.90; SRMR < 0.05; and RMSEA < 0.08 [49,55]. Since chi-square is sensitive to large sample size, and chi-square and degree of freedom (df) were used for descriptive information [56]. The CFA results revealed that the measurement model presented acceptable fit indices (${\chi }^2$ = 189.2, df = 14, GFI = 0.97, CFI = 0.98, TLI = 0.95, IFI = 0.98, SRMR = 0.03, and RMSEA = 0.08). As presented in

Table 1. Indicators of measurement model.

Construct	α	R2	CR	AVE
Cross-Departmental Collaboration	0.934	0.94	0.94	0.88
Resource Acquisition	0.682	0.83	0.71	0.55
Knowledge Creation	0.719	0.70	0.74	0.59
Organizational Performance	0.721	0.78	0.72	0.57

, all the constructs of the measurement model had R² values greater than 0.50, confirming convergent validity. The composite reliability values (CR in Table 1), ranging from 0.71 to 0.94, exceed the suggested values of 0.70, demonstrating that these indicators have sufficient internal consistency and represent the respective constructs. The average variance extracted values (AVE in Table 1) of all constructs also passed the suggested 0.50 cut-off, indicating that questionnaire items for each construct captured more variance in the underlying construct than the amount of variance caused by measurement error [49,55]. After validation, the measurement model was subsequently adopted for structural model assessment.

In the second step, SEM was conducted to test the theoretical relationship and estimate the paths’ strength between constructs simultaneously. Because indicators measure underlying latent constructs, SEM can be used to correct the measurement error [55,57]. In model comparison, SEM reports fit indices, which can be used as a criterion to select the best-fitted model. Thus, SEM is methodologically appropriate to test the conceptual model’s hypotheses.

The fit indices for the structural model in this study are presented in

Table 2. Fit indices of structural model.

Model	Fit Indices
	${\mathit{\chi }}^2$	df	GFI	CFI	TLI	IFI	SRMR	RMSEA	BIC
Conceptual model	197.82	16	0.968	0.975	0.956	0.975	0.027	0.086	237.81
Revised model	189.14	15	0.970	0.976	0.955	0.976	0.027	0.079	231.14

. After considering the fit indices, the RMSEA value was 0.09, greater than the acceptable level of 0.08; therefore, the conceptual model might not have been the best-fitted model. The post hoc modification was applied using model trimming to delete the path, which results in a best-fitted model. Model trimming was conducted by deleting the path in which the standardization residual was higher than $\left|4\right|$ [54] and then determining improvements in the fit indices. Complying with the criteria of model trimming, the revised model was developed by deleting the hypothesized path from the resource acquisition to the organizational performance.

The revised model presented an acceptable level of model fit (${\chi }^2$ = 189.14, df = 15, GFI = 0.97, CFI = 0.98, TLI = 0.96, IFI = 0.98, SRMR = 0.03, and RMSEA = 0.08). Additionally, the Bayesian information criterion (BIC), appropriate to compare models with large sample sizes, was used to assess parsimony for model comparison [58]. The model with lower BIC is more parsimonious than the compared model. The BIC of the revised model was less than the value of the conceptual model. Consequently, the revised model was considered a parsimonious model and thus was accepted as the final structural model, as displayed in Figure 2.

Using a deductive approach, structural equation modeling has been employed to empirically test the hypotheses because deductive reasoning establishes firm criteria for verifying hypotheses [40]. The standardized regression coefficients are presented in

Table 3. Standardized path coefficients and structural equations.

Paths	Structural Equations	Coefficient	S.E.	C.R.
CC → RA	ZRA = 0.692(ZCC)	γ = 0.692	0.026	19.542 ***
CC → KC	ZKC = 0.274(ZCC) + 0.576(ZRA)	γ = 0.274	0.032	6.540 ***
RA → KC		β = 0.576	0.050	11.770 ***
CC → OP	ZOP = 0.111(ZCC) + 0.942(ZKC)	γ = 0.111	0.035	20.165 ***
KC → OP		β = 0.942	0.020	3.164 ***

Note: *** p < 0.001; S.E., standard error; C.R., critical ratio.

. Cross-departmental collaboration was considered an exogenous factor (γ), while the other remaining constructs were considered endogenous factors (β). The critical ratio (C.R.) is calculated by dividing a path parameter’s regression weight by its standard error. The C.R. is interpreted similarly to the Z-test, whereby the path with a C.R. greater than 3.29 is significant at the 0.001 level [54]. Cross-departmental collaboration exerted a strong and positive influence on resource acquisition (0.692, p < 0.001), supporting H1. Cross-departmental collaboration had a moderate and positive impact on knowledge creation (0.274, p < 0.001), thus H2 is supported. The association between resource acquisition and knowledge creation was strong and positive (0.576, p < 0.001), providing support for H3. Cross-departmental collaboration is positively related to organizational performance (0.111, p < 0.001), supporting H4. Knowledge creation exerted a highly positive impact on organizational performance (0.942, p < 0.001), accepting H5. Finally, the proposed path from resource acquisition to organizational performance was removed, which implies the rejection of H6.

4.2. Bayesian Networks

4.2.1. Bayesian Network Construction

The first step to develop a BN model is structure learning, which is qualitative. Typically, the structure of a BN can be learned by expert knowledge and data learning. However, the expert judgments cannot confirm the objectivity and accuracy of the results, which may lead to spurious relationships, while it is difficult for the simple data-driven technique to learn the order among the nodes and critical information concealed in the investigation reports [50]. This study applies the integrated approach to address these weaknesses in structure learning. The integrated approach connects the structural model to Bayesian networks by constructing the DAG based on an empirically validated structural model. The DAG was developed by deriving the causal relationships between latent constructs in the structural model, as shown in Figure 3.

The second step is parameter learning to specify joint distributions, which is quantitative. The parameter learning calculates the CPT of each node in the BN. The parameter learning of the BN $G=\left(A,ᴪ,P\right)$ calculates the parameter values $\Theta$ relating to the $DAG’ᴪ’$ and distribution from the data set $’D$′. Generally, if $G=\left(A,ᴪ,P\right)$ is a Bayesian network with parameter values $\Theta =\left\{{\Theta }_i\right\}$ where ${\Theta }_i=\left\{{\Theta }_{ij}\right\}$ and ${\Theta }_{ij}=\left\{{\Theta }_{ijk}\right\}$ such that ${\Theta }_{ijk}=P\left(A_i=k\left|parent\left(A_i\right)\right.=j\right);{\forall }_{i,j,k}$. Accordingly, the parameter learning is to calculate the parameter of ${\Theta }_{ijk}$ from the data set $\prime D$′ [59].

The CPT can be calculated through expert knowledge or learning from sample data. The expert knowledge could be incomplete and subjective, which might affect the network’s accuracy [50]. Therefore, this study calculated the conditional probability distribution for each node by learning from sample data. The most straightforward method to learn the parameter is the counting algorithm. In addition, the counting algorithm should be applied in all possible circumstances because it is acknowledged as a true Bayesian learning algorithm [48]. Thus, this study adopted the counting algorithm to calculate the CPT from sample data. The total aggregation method was applied by averaging responses to questionnaire items in each construct to form a single variable as a node. By determining the occurrence frequency, the numerical value of each node was discretized to transform the scale into three states: [1–2.5] as low, [2.5–4] as medium, and [4,5] as high [48]. This study used the commercially available software package Netica to develop the BN. The DAG was drawn as a cognitive map, and then the CPTs were automatically learned from the training dataset containing 1535 cases.

Table 4. Conditional probability table for organizational performance node.

Cross-Departmental Collaboration	Knowledge Creation	Organizational Performance
		Low	Medium	High
Low	Low	41.0	46.2	12.8
Medium	Low	16.7	44.4	38.9
High	Low	9.1	54.5	36.4
Low	Medium	2.9	64.7	32.4
Medium	Medium	0.6	38.9	60.5
High	Medium	0.5	15.0	84.5
Low	High	2.6	23.7	73.7
Medium	High	0.6	11.3	88.1
High	High	0.1	1.3	98.6

shows an example of CPT presenting the probability distribution of the organizational performance node. In this example, when the cross-departmental collaboration and knowledge creation nodes are both in the low state, the probabilities of low, medium, and high states of organizational performance node are 41.0%, 46.2%, and 12.8%, respectively.

After learning the CPTs, the probabilistic inference was conducted using the ‘Compile the net’ function in Netica. Then, the overall graphical representation of the proposed BN was automatically developed in the form of a belief bar, as depicted in Figure 4. The BN presents the existing condition of the organizational performance, impacted by the antecedent variables as evidence from the sample data. The BN is used to explain the effect of different predictor variables on the outcome variable in the workplace environment. In the current scenario, high cross-departmental collaboration (70%), high knowledge creation (68%), and high resource acquisition (62%) are likely to occur. These three predictor variables result in a high organizational performance node (85%), corresponding to the mean value of 4.28. Although most employees perceived organizational performance to be high, there is still an opportunity for improvement in this outcome variable.

4.2.2. Sensitivity Analysis

Sensitivity analysis is a diagnostic method used to assess input variables that have significant impacts on the output variable. Sensitivity analysis is applied to identify the most influential input variables that can minimize uncertainty in predicting the output variable. The diagnostic results present the influence of related nodes on the mean value of the target node.

This study applied the variance reduction method to analyze the sensitivity analysis. The aim is to present which varying nodes can reduce the uncertainty of the query node, as expressed by the degree of reduction in variance. This method calculates the variance reduction of a query node G caused by varying node H. The variance of the query node G given the input node H, denoted as V(G/h), can be calculated by the following equation [51]:

$$V\left(G/h\right)={\displaystyle \sum }_{g}p\left(\frac{g}{h}\right){\left[X_g-E\left(\frac{G}{h}\right)\right]}^2$$

(5)

where g is the state of query node G, h is the state of input node H, p(g/h) is the conditional probability of g given h, E(G/h) is the expected real value of G due to finding h for node H, and Xg is the numeric value relating to state g.

Table 5. Sensitivity analysis of organizational performance node.

Factor	Variance Reduction	Percent Variance Reduction	Normalized Variance Reduction
Cross-Departmental Collaboration	0.1039	23.8	1.00
Knowledge Creation	0.0819	18.7	0.79
Resource Acquisition	0.0365	8.34	0.35

presents the sensitivity analysis of the query node, which is organizational performance. The indicators consist of a variance reduction, a percentage of variance reduction, and a normalized variance reduction for each input node. The normalized variance reduction presents the relative sensitivity resulting from the percent variance reduction. The normalized variance reduction is the relative sensitivity resulting from the percent variance reduction of each node divided by the highest percent variance reduction [48].

Sensitivity analysis ranked the critical factors as per their explanatory power on a target node. If the input node’s variance reduction is comparatively high, this input node exerts relatively high explanatory power on the target node. The critical factors of organizational performance are cross-departmental collaboration and knowledge creation, with variance reductions of 23.8% and 18.7%, respectively. Cross-departmental collaboration and knowledge creation have the greatest explanatory power over organizational performance node and thus are significant factors in improving agency’s performance.

4.2.3. Scenario Analysis and Discussion

This study applied a what-if analysis for scenario-based simulation to explore possible consequences based on changing factor(s). In the first scenario, the impact of cross-departmental collaboration on other nodes can be considered when the chance of 100% occurrence was conditioned at a high state, as presented in Figure 5. The chance of high knowledge creation increased from 68.4% to 80.0%, reflecting an increase of 17.0%. Consequently, the chance of high organizational performance increased from 85.3% to 95.3%, indicating an improvement of 11.7%.

Encouraging cross-departmental collaboration among highway agencies is necessary because highway networks extend throughout geographical and political boundaries. The transportation agencies are also confronted with budget constraints and conflicting demands due to overlapping jurisdictions and ambiguous boundaries [60]. Transportation agencies’ decision-making is characterized by dynamic multi-jurisdictional issues that are meant to be resolved by collaborative networks. To solve these issues, transportation agencies develop a mutual relationship and policy coalition through collaborative networks [27].

For example, the Oregon Transportation Commission has chartered Area Commissions on Transportation (ACTs), consisting of government and non-government organizations [2]. The ACTs are characterized as multi-level collaboration networks, in which representatives are nominated from the state, regional, and local agencies [8]. The purpose of the ACTs is to improve collaboration at the regional level. Furthermore, the ACTs also act as advisory bodies in developing the Statewide Transportation Improvement Program, allocating and prioritizing projects for funding. Collaboration within the ACTs structure is a prime driver in facilitating joint decision-making among the participating parties. Therefore, agencies coordinate in these collaborative networks to help solve complicated inter-jurisdictional issues due to the limited budget and increased travel demand, thus contributing to successful statewide decision-making in transportation planning and policy.

The most optimistic scenario in improving organizational performance can be achieved by enhancing both critical factors that have the most significant explanatory factors. This scenario is simulated by entering the opportunity of 100% occurrence of high state for both cross-departmental collaboration and knowledge creation, as depicted in Figure 6. As a result, the probability of high organizational performance rises from 85.3% to 98.6%, representing an increase of 15.6%.

The scenario analysis emphasizes the significance of knowledge creation to improve organizational performance. Knowledge creation begins with socialization, which is an effective process of exchanging and creating tacit knowledge within a construction project. Because tacit knowledge is difficult to codify and is workplace-specific, it can be accumulated by directly sharing team members’ experiences in social interactions. Junior engineers can develop tacit knowledge through observation and informal discussion, as well as working together with senior engineers to acquire hands-on experience in day-to-day construction activities [31]. Converting tacit knowledge to explicit knowledge occurs in the externalization stage. Storytelling and dialogue are efficient methods to articulate a hidden concept of tacit knowledge and then codify tacit knowledge into formalized documents [15]. The project team uses brainstorming sessions to create new explicit knowledge by developing a written and understandable standard operating procedure for the new construction method. The combination is the process that systemizes and applies explicit knowledge from inside and outside the organization. During the combination process, the explicit knowledge is gathered, edited, and integrated to formulate more practical and systematic knowledge. The new explicit knowledge is transferred among project members for future usage [35]. Team members convert explicit knowledge into tacit knowledge to develop a shared mental model through the internalization stage. For instance, junior engineers can embody explicit knowledge, such as construction methods and specifications, by reading and understanding these written documents during action and practice [19]. Subsequently, they can enrich their tacit knowledge through learning-by-doing in their professional practices. The knowledge creation process is a spiral movement that continuously stimulates the conversion between explicit and tacit knowledge. The four modes of knowledge creation amplify a new spiral of knowledge conversion; therefore, more knowledge can be created in an organization.

5. Discussion

5.1. Theoretical Contributions

These empirical results offer insights into how cross-departmental collaboration influences organizational performance through the role of resource acquisition and knowledge creation. Cross-departmental collaboration indirectly impacts organizational performance through resource acquisition and knowledge creation. The results reveal that cross-departmental collaboration has a positive and sizable effect on resource acquisition, which in turn positively impacts knowledge creation. As hypothesized, the results indicate that the effect of cross-departmental collaboration on knowledge creation is positive. Knowledge creation, in turn, has a positive and substantial effect on organizational performance. This finding is consistent with previous research from transportation agencies, presenting that collaborative networks among metropolitan planning organizations (MPOs) and other related partners are formulated to develop and implement active transportation policies in the United States [61]. The research revealed that cross-departmental collaboration was significantly related to resource sharing and knowledge creation around active transportation policies. Organizing MPOs as collaborative intermediaries played a crucial role in creating explicit and tacit knowledge and sharing tangible and intangible resources among actors and organizations. Therefore, MPOs could solve complicated transportation planning problems that arise from metropolitan areas’ silo effect and bureaucratic culture, thus improving the performance of transportation agencies.

The empirical results also reveal that cross-departmental collaboration directly impacts organizational performance, but the effect size is relatively small. Although cross-departmental collaboration through multijurisdictional schemes has been promoted over the previous three decades, functional fragmentation between transportation agencies is still a longstanding and complex issue across jurisdictional boundaries and modal networks [62,63]. The complexity of organizational collaboration may be another reason cross-departmental collaboration has a relatively small direct impact on performance. Sanders [64] studied the inherent complexity of collaboration and revealed an organizational mechanism that interdepartmental collaboration impacts intradepartmental collaboration, ultimately improving organizational performance. Collaboration is usually effective after the working relationship has been developed. Therefore, it may take substantial time, particularly in the early stage of developing a collaborative network, to facilitate interdepartmental collaboration and engage a high level of intradepartmental collaboration to be synergistic for reaping performance benefits.

This empirical result does have a significant implication concerning the influence of resource acquisition. Samaddar and Kadiyala [18] argue that sufficient resource acquisition had a statistically significant impact on the success of collaborative knowledge creation. This finding is consistent with the resource-based view (RBV) that highlights the significance of the critical resource of outside-in capabilities when engaging with strategic alliances [17]. As an example, sharing lessons learned from projects with external partners leads to a sustainable advantage for organizations by creating their technical and managerial knowledge [14]. Besides, this finding is in line with the main argument of the resource dependence theory (RDT) that organizations need to enter into collaborative networks to maintain critical intangible resources [12]. Consequently, they can coordinate to develop professional expertise to manage the technical complexity of highway mega-projects [13].

This study also provides theoretical contributions to the knowledge management literature by investigating how knowledge creation can improve organizational performance. Knowledge creation has a positive and substantial impact on organizational performance. This finding is in line with the knowledge-based theory that knowledge is a strategic resource used by the organization to develop organizational ability leading to improved performance [65]. Through this theoretical lens, organizations are regarded as integrated repositories of tacit and explicit knowledge, in which diverse knowledge bases serve as main antecedents of long-term sustainable organizational performance [66].

Further highlighting the policy reliance of this research, cross-departmental collaboration has emerged as a prime mover for organizational performance. Notably, it improved performance by promoting resource acquisition and knowledge creation required for task accomplishment. The two variables capture most of the positive effects on performance, which means that these variables are very significant for organizational success. Remarkably, the degree of knowledge-intensive activities is substantial for the FHWA as a research and advisory organization for state and local highway departments, so the improved performance gain through knowledge creation is very critical.

5.2. Practical Implications

Cross-departmental collaboration, such as collaborative networks of transportation agencies, is considered a critical approach to solving increasingly complex inter-jurisdictional and wicked issues with no solutions, such as budget constraints and increased demands and climate change [2,10]. The highway agencies should promote cross-departmental collaboration across multi-jurisdictional agencies. For instance, the FHWA recognizes the necessity of highway planning and management from a megaregional perspective. The US megaregions are a network of metropolitan areas combined through similar characteristics and mutual interests of social, economic, topographic, political, climatic, and infrastructural issues [67]. Because the US highway systems transcend many jurisdictional boundaries, highway planning and management is an inherently megaregional issue. The FHWA has continuously funded the collaborative design, construction studies, and project works at the megaregional scale for a decade. Since 2016, the FHWA has been actively promoting collaborative research, convening seminars, and sponsoring peer-to-peer work in large-scale highway issues of megaregions [43]. These initiatives reveal that the FHWA embraces cross-departmental collaboration as a crucial priority.

Additionally, to improve organizational performance, highway agencies should acknowledge the benefit of knowledge creation, presuming that the most significant assets are skills and knowledge. For example, the FHWA organizes knowledge disciplines around technical specialty, occupation, and profession. Examples of disciplines are design, finance, planning, and safety. A discipline champion is appointed to lead each discipline and execute mandatory activities that each discipline needs to accomplish [68]. The FHWA adopted the Discipline Support System, an integrated knowledge management system created to transfer knowledge across the agency, share best practices, and network within the discipline [69]. With geographically dispersed offices across the country, a virtual environment for knowledge creation provided by the Discipline Support System is essential to sustain and grow the knowledge discipline. The discipline champion also facilitates mentoring programs to capture and transfer explicit and tacit knowledge from senior engineers to the younger workforce, which helps solve the issue of knowledge loss because of retirement. The evidence illustrates the effort that the FHWA has exerted to inculcate knowledge creation into the organizational culture.

5.3. Limitation and Future Directions

Using secondary data from the FEVS has some limitations that researchers should acknowledge and may address in future research. First, the survey data is from federal departments in the United States, exposed to Anglo-Saxon culture. As a result, some of the findings may not be generalizable to eastern countries in which the patronage system dominates public sector organizations. Future studies should be conducted to test the benefits and costs of cross-departmental collaboration in these eastern countries, in which collaboration activities are impeded by significant barriers from bureaucratic culture, the silo effect, jurisdictional fragmentation, and an autocratic leadership style. Second, the study did not investigate the impact of demographic variables that could also affect organizational performance. Analyzing innovation activities in the Australian Public Service, Demircioglu [70] revealed that demographic variables, including education, experience, and managerial position, are positively related to innovation implementation, thus improving federal departments’ performance. Therefore, future research should include demographic variables in a causal model to explore their effects on organizational performance.

Finally, because the FEVS data come from a cross-sectional design, we cannot dismiss the possibility of reverse causation, which excludes inference for any causal claims. Furthermore, cross-departmental collaboration takes considerable time to develop a trusted working relationship, so it will not lead to immediate benefits in improving performance. Additionally, collaboration continuously facilitates resource acquisition and knowledge creation in the long term; therefore, the ultimate results on organizational performance should be measured in the long run. Thus, future work can expand our study by applying a longitudinal research design, which collects data at several time points. Latent growth modeling can be used to analyze the time-lagged panel data [71]. The longitudinal model can examine whether dynamic organizational attributes increase the level and growth rate of organizational performance. This longitudinal model can also explore the dynamic nature of organizational attributes and the relationships among their dynamics (i.e., rates of change) [72]. This research would theoretically contribute to the literature of time-phased organizational studies and investigate the role of cross-departmental collaboration as a significant determinant of the dynamics and longitudinal organizational performance trajectories.

6. Conclusions

A hybrid SEM-BN approach is proposed based on an integrated method that connects an empirically validated structural model to Bayesian networks. As a confirmatory tool, the SEM is performed to identify causal relationships between organizational constructs and empirically validate the conceptual model. The BN is used as an exploratory technique to identify the critical organizational variables and analyze the influence of improvement in organizational attributes on organizational performance.

Using the data from the FHWA, this SEM was conducted to validate an organizational-centric structural model. This study examined the direct effect of cross-departmental collaboration on organizational performance and explored indirect effects through managerial practices of resource acquisition and knowledge creation. The results indicate that cross-departmental collaboration positively impacts resource acquisition (H1) and knowledge creation (H2). Resource acquisition is also positively related to knowledge creation (H3). Cross-departmental collaboration has a positive impact on organizational performance (H4). Meanwhile, knowledge creation is substantially associated with organizational performance (H5). Finally, we removed the hypothesized path between resource acquisition and organizational performance, implying the rejection of H6. The post hoc modification was conducted using the model trimming approach, which removed the direct path from resource acquisition to organizational performance, implying that resource acquisition by itself did not directly improve organizational performance. Instead, agencies need to acquire resources, which are indispensable to prepare the foundations for knowledge creation.

The BN models are applied for scenario-based simulation to highlight critical pathways to enhance organizational performance. The simulation was conducted using sensitivity analysis to identify critical factors that significantly impact organizational performance. The most critical factors are cross-departmental collaboration and knowledge creation. Two scenarios were analyzed using what-if analysis to demonstrate how the cross-departmental collaboration practice and knowledge creation process directly and indirectly impact organizational performance. The scenario-based simulation provides recommendations for cross-departmental collaboration practice and knowledge creation processes in improving highway agencies’ performances.