Abstract
Geographic dispersion is routine in new product development, yet it remains unclear whether employee geographic distance (EGD) slows project execution. We develop a coordination cost perspective in which EGD is postulated to increase project delays because it raises attention allocation and information search costs. We test this framework by using a longitudinal employee–project–month panel from a large manufacturing firm that spans 9,729 observations across Europe, North America, and Asia. We find that greater geographic distance is associated with longer project delays, and that the effect is contingent on employee and project characteristics. The EGD–delay link is attenuated by employee familiarity and project similarity, consistent with lower attention allocation costs, and for specialist employees, consistent with differences in information search costs. We also find that the link is amplified for cross-functional geographic dispersion. Additional analyses document that the association between EGD and project delay is concentrated at later project gates and in high-risk or highly innovative projects. A difference-in-differences test around daylight-saving-time transitions indicates that temporal misalignment alone does not account for the EGD–delay relationship. Taken together, the results inform the literature on new product development, project management, and global work by demonstrating when geographic distance is most consequential for project delay.
Introduction
Geographically distributed work in new product development (NPD) projects is increasingly common, with an estimated 39%–53% of employees working at a distance from their collaborators (Gartner, 2023; Tripathy and Eppinger, 2013). Firms embrace employee geographic distance (EGD) to access dispersed talent, increase capacity, and reduce costs while maintaining quality and efficiency of project work. While prior research shows that geographic dispersion can hinder communication, collaboration and technical coordination in distributed product development (e.g., Anderson and Parker, 2013; Gokpinar et al., 2014; O’Leary and Cummings, 2007; Sosa et al., 2002), it does not establish whether EGD translates into downstream project delay, which is a critical yardstick of NPD success. Because collocation is often infeasible, establishing whether and when EGD lengthens project schedules is critical for aligning employee assignments and project characteristics with coordination demands.
To address this question, we develop a theoretical framework linking EGD to project delay, adopting a coordination cost perspective. Specifically, we argue that EGD fragments limited attentional resources and induces higher effort in information search, access, and integration, thereby creating bottlenecks in projects and delaying project delivery. Refining prior research that documents average dispersion penalties for task completion and quality (e.g., Gokpinar et al., 2014), we propose four conditions that alter the magnitude of attention allocation and information search costs: employee familiarity, project similarity, specialization, and cross-functional geographic dispersion.
We test this conceptual framework with a longitudinal employee–project–month panel from geographically distributed NPD projects in a large manufacturing firm, comprising 9,729 observations. The dataset integrates HR records, work registries, monthly project reports, and job catalog information, with employees located across Europe, North America, and Asia. Our empirical strategy consists of employee fixed effects, month and year fixed effects, and employee- and project-level controls, along with three-way clustered standard errors. We discuss and address issues related to employee selection and endogeneity.
Our findings extend the literature on NPD, project management, and global work in several ways. First, we contribute with a theoretical framework that argues how EGD can raise attention allocation and information search costs that, in turn, are reflected in the delay of NPD projects. The framework and findings answer the ongoing calls to provide longitudinal evidence of how geographic dispersion can impact global project work (Froese et al., 2025) and add to the prior literature on the impact of geographic dispersion on employees, teams, and organizations (Anderson and Parker, 2013; Gokpinar et al., 2014; Reiche et al., 2019; Salvador and Madiedo, 2021; Sosa et al., 2002). Additionally, we contribute to the research stream on NPD and project management by highlighting an influential antecedent of the timely delivery of projects (Crama et al., 2019; Kavadias and Loch, 2003; Krishnan and Ulrich, 2001; Staats, 2012).
Our second contribution stems from unpacking the conditions under which the positive association between EGD and project delay can be attenuated or amplified. Employee familiarity and project similarity attenuate the attention allocation costs as they can reduce attention fragmentation and minimize interruptions and switching, hence mitigating the positive EGD–delay relationship. What is striking is that at high levels of familiarity or project similarity, the EGD penalty shrinks substantially. In other words, unlike previous research that argues EGD is almost always detrimental, we demonstrate that this is not always the case. These findings add to the previous research that argues for the benefits of prior collaboration experience for team performance outcomes (Cummings and Haas, 2012; Huckman et al., 2009; KC and Staats, 2012; Wu et al., 2026).
On the information search side, a potential counterintuitive pattern emerges: specialists seem to be less penalized by distance than generalists. Interestingly, recent research suggests that generalists are better able to transfer and integrate knowledge across contexts and help bridge communication and handoff gaps, whereas specialists’ deeper organization-specific knowledge raises adjustment and coordination costs (De Stefano and Bidwell, 2026; Fahrenkopf et al., 2020; Sabel and Sasson, 2023). We complement this evidence by showing that specialists are likely to be less impacted by EGD, possibly because they sustain lower information search costs. Specifically, specialization likely narrows the information search space and speeds up the identification and integration of relevant knowledge.
Furthermore, cross-functional geographic dispersion amplifies the information-search costs of EGD. Importantly, this is not a blanket indictment of splitting R&D and production. Firms may still realize labor cost or capacity gains from cross-country allocation; however, our results indicate that such arrangements are associated with delays in project execution. This pattern is consistent with the theoretical challenges of coordinating work across dispersed functional interfaces, where the absence of standardized processes and routines can increase coordination frictions (Salvador and Madiedo, 2021; Sosa et al., 2004). Finally, our additional analyses reveal that the effects of EGD are more pronounced at later project stages and in high-risk or highly innovative projects, and are absent when employees are highly co-located.
The remainder of the paper proceeds as follows: Section 2 develops the framework and hypotheses; Section 3 describes the setting and data; Section 4 outlines the empirical models; Section 5 presents the results and mechanism tests; and Section 6 discusses the implications for theory and practice.
Theoretical framework
Figure 1 shows our conceptual framework.
Conceptual framework.
We define EGD as the average geographic distance between a focal employee and their collaborators within the same project. Building on classic and contemporary work on organizational attention and information processing, we focus on two employee-level components of coordination costs: attention allocation costs and information search costs 1 (March and Simon, 1958; van Knippenberg et al., 2015). Attention allocation costs concern the extent to which employees can focus their attention. Information search costs reflect the effort required to find, access, and integrate relevant information across geographic distance.
Attention allocation costs: Effective project work depends on the disciplined allocation of a limited employee attentional budget. EGD fragments this budget by forcing employees to juggle multiple collaboration contexts over geographic distance. EGD implies that employees need to adapt to different platforms, reporting templates, and technical processes, which increases attention demands. EGD also induces stop–start work rhythms, which further drain attention (KC and Tushe, 2021). Employees need to spend more time coordinating over geographic distance, which entails waiting for responses and arranging ad hoc virtual meetings. These interruptions and the associated switching create attention residue and “time famine,” thereby lowering productivity (Perlow, 1999). As a result, employees struggle to sustain focus on project milestones and tend to default to more salient local tasks, even when priorities lie elsewhere (O’Leary et al., 2011). In summary, by fragmenting attention over geographic distance, EGD slows progress on interdependent tasks and increases the likelihood of delays.
Information search costs: EGD increases information search costs 2 as employees need to expend effort to find, access, and integrate relevant information over geographic distance (Carlile, 2004). Because locations often govern and store information in distinct repositories and under different protocols, employees must first locate authoritative, up-to-date sources and secure access permissions, which can be hindered by tacitness and asset specificity (Teece, 1977). Even after access, cross-location differences in taxonomies, formats, and documentation standards necessitate integration work, including translation, reformatting, and reconciliation with artifacts and procedures from other locations. Consistent with this logic, geographic dispersion is associated with weaker inter-unit knowledge flows and more difficult information elaboration in teams (Tzabbar and Vestal, 2015; van Knippenberg et al., 2015). In aggregate, this locate–access–integrate work lengthens the time required to retrieve and apply knowledge, thereby delaying milestone completion. Overall, we expect the following relationship:
The moderators of the coordination costs of EGD
Building on the coordination cost perspective, we propose four moderators that alter the severity of the attention allocation and information search costs. Employee familiarity and project similarity are theorized to shape attention allocation costs by reducing attention fragmentation and minimizing interruptions and switching. Specialization lowers information search costs by narrowing the domain of knowledge, shrinking the information search space, and easing information transfer over distance. Cross-functional geographic dispersion raises information search costs by increasing the translation and alignment work required to reconcile specifications, standards, and test procedures over geographic distance.
Employee familiarity
Research shows that familiar teams face fewer challenges in communication and coordination because repeated collaboration facilitates day-to-day teamwork and lowers the cognitive demands of individual employees (Skilton and Dooley, 2010; Staats, 2012). Familiar employees can anticipate one another's work habits and communication styles, so they require fewer reminders, follow-ups, and scheduling loops to secure responses, thereby reducing interruptions and switching, and limiting attention residue (Cummings and Haas, 2012; Leroy and Glomb, 2018). Familiarity also streamlines task handoffs and role assignments. Familiar members can more quickly align on “who does what,” which helps sustain focus on project milestones rather than renegotiating roles and responsibilities (Huckman et al., 2009). By contrast, low degrees of familiarity can lead to attention fragmentation as employees need to repeatedly redirect their focus, clarify responsibilities, and wait for responses. Accordingly, we expect that:
Project similarity
Project similarity reduces attention allocation costs by lowering the technical heterogeneity of tasks that employees must manage across their project portfolio. When projects are similar, employees can reuse schemas, routines, and technical language, which eases switching and limits fragmentation of attention (KC and Staats, 2012). Prior research shows that moving between similar tasks reduces the cognitive load and productivity losses associated with context switching (O’Leary et al., 2011). In many cases, employees can build directly on work they have done before. For example, a software developer can reuse code from a previous project, or an engineer can apply the same tools across projects. In other words, project similarity reduces technical heterogeneity and frees up attentional capacity, allowing employees to better manage the coordination demands imposed by geographic distance. Accordingly, we expect that:
Employee specialization
Employee specialization lowers information search costs by narrowing and structuring the information domain. For a specialist, a smaller, well-defined problem space narrows the search set and enhances the ability to identify authoritative sources and filter out noise over geographic distance. Specialists also face less uncertainty about task-specific abilities among dispersed collaborators and develop stable maps of where expertise resides and which artifacts (e.g., documents, tools, routines) are diagnostically useful, thereby shortening routing and verification cycles (De Stefano and Bidwell, 2026; Reagans and McEvily, 2003). Furthermore, specialization enables faster interpretation and integration of incoming information, even when local specifications vary. By contrast, generalist roles expand the search space and require reconciling more heterogeneous inputs without in-depth knowledge, thereby lengthening the time needed to locate and integrate information. Even when procedures or product specifications change over time in a given location, specialists are more likely to spend less time searching and acting on the necessary information, given the narrower scope of their tasks (Goradia and Byron, 2024; Vessal and Sommer, 2025). Accordingly, we expect that:
Project cross-functional geographic dispersion
Cross-functional geographic dispersion is conceptualized as instances when a project's R&D and production activities are located in different countries. While such cross-functional geographic dispersion can generate efficiency gains by enabling cost advantages and capacity utilization across locations, it also introduces coordination costs. Interfaces between R&D and production often require repeated exchanges to clarify tolerances, documentation, and test procedures. When these interfaces span countries, employees must navigate heterogeneous repositories, access rules, and vocabularies, which increases the time required to locate, access, and make usable the information needed to proceed (Carlile, 2004). These cross-functional frictions can cascade into delays, particularly when geographic distance is high (Sosa et al., 2004). For example, misunderstandings in the design–production process generate rework and change orders, asynchronous handoffs, and additional rounds of clarification, which impact project timelines. We expect that:
Empirical setting and data
We test our theoretical framework with data from a multinational company operating in an NPD context. The focal company is a global leader in hydraulic pumps, headquartered in Denmark, with roughly 20,000 employees across more than 50 countries and a net turnover exceeding USD 4 billion. Data were collected through a multi-year collaboration with the focal company. The company granted access to a longitudinal dataset that links employees to projects over time, allowing for ongoing discussions with managers to ensure the research questions, measures, and results remained managerially relevant. The research team conducted several on-site and virtual meetings with managers, human resources (HR) representatives, and IT specialists to extract and merge monthly project reports, HR records, and the job catalog. Multiple feedback rounds with senior managers ensured accuracy and contextual validity. Table A-1 in the E-Companion outlines the data collection timeline and activities.
NPD process in the organization
The NPD process is managed through projects that follow a standardized seven-gate model. In the first four stages, the projects follow a path from idea generation and validation to concept development, production, and manufacturing. The final three gates involve scaling up production and then deploying a marketing and sales plan for the product release (Project Gates 5, 6, and 7). Project managers file monthly reports with a uniform structure. Each report lists the project identifier, production sites, product type and characteristics, target markets, team composition and roles, project hierarchy, and planned versus realized timelines at each gate. These elements enable the construction of a project delay analysis that compares expected versus realized gate transitions and collects a rich set of project-level measures.
Data sources
We integrated three data sources for January 2015 to August 2016: Monthly Project Reports. Standardized records of project composition by month and exact timestamps for project advancement. These records allow us to map each employee-project combination longitudinally and calculate a precise longitudinal measure of project delay. From the project technical categorization, we obtain the degree of similarity across projects, while from the location information, we trace the countries of the R&D and production sites. The data also contain rich information on a set of project characteristics such as project size, hierarchy, innovation level, program, and concerns, which we use as control variables. HR Records. Longitudinal employee data, including precise latitude and longitude coordinates of the location for each employee, as well as the country, city, site, and building. We use these to calculate the EGD measure. We also obtain employees’ age, gender, tenure, department affiliation, and managerial responsibility. Job Catalog. Descriptions of 92 roles with fields for the role name (e.g., “R&D Global Product Manager”), job track (e.g., “Specialist”), and the purpose of the job role (e.g., “Work within the functional area”), key activities (e.g., “Develop the product line”), knowledge and experience for the role (e.g., “Background in innovative projects”), job family (e.g., “R&D”), and hierarchy as a numerical band (e.g., “Band 6 for R&D Chief Engineer”).
Merging these sources yields an employee–project–month panel comprising 526 employees and 42 projects over 20 months (January 2015 to August 2016), resulting in 9,729 observations.
Variables and operationalization
We describe all the variables in Table 1 (please see the E-Companion for additional details).
Variable description and summary statistics.
Variable description and summary statistics.
We calculate EGD at the employee–project–month level as the average of the natural logs of pairwise distances between the focal employee and all other members of the same project in the same month:
The measure relies on the exact geographic coordinates (longitude and latitude) from HR records. Geographic dispersion in the focal organization is substantial. Employees work across Europe (Denmark, the United Kingdom, Hungary, Serbia, Germany, France, and Finland), North America (the United States, Mexico), and Asia (China, India, Singapore, Taiwan, Russia, and the United Arab Emirates). Distances are computed as described below: Within-city distances: For employees located in the same city, we compute the shortest walking distance between their buildings using Google Maps. Between-city distances: For employees in different cities within the same country, we calculate the shortest travel distance (by car or plane) between them. Cross-country distances: For employees located in different countries, we compute the straight-line Euclidean distance between coordinates.
These definitions follow common practice in prior literature and reflect the most relevant mode of interaction at each geographic scale. Distances are calculated from the perspective of each focal employee e to all other project members j,
An illustrative example of EGD.
We use project delay as the main outcome variable, which is a critical performance indicator in NPD project success (Crama et al., 2019; Krishnan and Ulrich, 2001; Nobeoka and Cusumano, 1997). In NPD, projects are rarely evaluated on immediate financial performance, as revenues and profits materialize years later. Instead, timeliness and adherence to schedule are the principal indicators of project execution effectiveness (Kavadias and Loch, 2003).
We operationalize project delay as the difference (in days) between the latest expected completion date of the current stage gate (defined by management at project inception and revised at the midpoint) and the actual completion date recorded in subsequent project reports. By using the most recent expectation as the baseline, the measure captures incremental, gate-specific deviations rather than cumulative project overruns. This approach ensures that delays recorded at later stages reflect new, localized slippage in execution, rather than mechanically propagating earlier schedule deviations across subsequent observations.
Specifically, each monthly report contains the expected and realized timestamps for all seven-stage gates in the standardized stage-gate process. To capture both planned and revised expectations, we use three distinct timestamps: Initial expectation from the first project report (set at project start). Updated expectation from the midway report (typically Gate 3). Actual realization from the report marking gate completion.
By combining these, we can trace how completion estimates evolve and quantify deviations from both original and revised expectations. The resulting measure reflects the number of days a project gate was completed earlier or later than expected (positive values indicate delay). We compiled these data across 20 months, allowing consistent tracking of completion times. Examples of project reports and delay calculations are provided in Figures A-2 and A-3 in the E-Companion, and the distribution of project delay is shown in Figure A-4 in the E-Companion.
Moderating variable: Employee familiarity
We measure employee familiarity as the number of “familiar” teammates on the focal project. Two employees are considered “familiar” if they have collaborated on other projects, either concurrently in the same period or in prior periods. Thus, familiarity reflects shared project experience beyond the focal project. Illustrations of how familiarity is constructed and distributed across employees are provided in Figures A-5 and A-6 in the E-Companion.
Moderating variable: Project similarity
We measure project similarity using detailed project descriptions provided by the company. Each project belongs to one of four technical categories: Innovation, Line Extensions, Platform Generation, or Product Integration. These categories define the project's technological and process characteristics. Projects in the same category tend to rely on comparable resources, routines, and deliverables.
For each employee in a given month, we calculate the ratio of projects that share the same technical category as the focal project to the total number of projects in which the employee is involved during that month. The metric measures the extent to which an employee's concurrent projects share similar technical content and development processes. The metric ranges theoretically from
Moderating variable: Employee specialization
We measure employee specialization using the firm's official job catalog, which categorizes each role as either a specialist or a non-specialist. Managers from the focal company emphasized that specialists are “the operational people who do the work” and “the engine of the project,” highlighting their central role in project execution. The job catalog lists 92 distinct roles across the organization. Each role is pre-classified by the company as a specialist or not, and our cross-check confirmed the consistency of this classification. Typical specialist roles include Product Configuration Engineer, Product Data Designer, and Product Specialist. Non-specialist roles include R&D Team Leader, Product Manager, and Technology Development Technician. We construct a binary indicator equal to 1 for employees holding specialist roles and 0 otherwise. In our sample, 55 of 92 roles (298 of 526 employees) are classified as specialists.
Moderating variable: Project cross-functional geographic dispersion
We measure project cross-functional geographic dispersion as a binary indicator that equals 1 if the project's R&D country differs from all of its production countries, and 0 otherwise. Each monthly project report specifies a single R&D location and up to five production sites across different countries. When the R&D country is included among the production countries (e.g., R&D in Denmark and production in Denmark, Finland, and Hungary), the indicator takes the value 0. When the R&D country is not represented among the production sites (e.g., R&D in Denmark and production in Finland and Hungary only), the indicator takes the value 1. An illustration of this variable's construction is provided in Figure A-9 in the E-Companion.
Control variables
We include a comprehensive set of controls at both the employee and project levels to isolate the effect of EGD on project delay (further details are available in the E-Companion).
For employees, we control for key demographic and work-related characteristics that may influence coordination costs and access to information. Specifically, we control for age, work experience, seniority, and managerial responsibilities. More experienced and senior employees, as well as managers, typically possess higher authority, broader internal networks, and greater knowledge of where to obtain information, which can lower information-search costs and confound the EGD–delay relationship. We also include the number of parallel projects in which each employee participates, as it can increase coordination complexity (Colicev et al., 2023). Finally, we control for the average absolute time-zone difference between the focal employee and all teammates in the same project–month, capturing temporal misalignment that may reduce communication efficiency (Chauvin et al., 2024).
For projects, we control for project size and the management team's load, which captures the availability of resources for coordination and problem-solving. We include the project's innovation level (1–5), recognizing that more innovative projects face greater uncertainty and coordination challenges (Nobeoka and Cusumano, 1997). We also add project program complexity, measured as the inverse of the natural logarithm of total project-months within the program. This transformation implies that higher values correspond to smaller and less frequently occurring project programs. Such programs are typically less standardized and rely on fewer established routines, which increases coordination uncertainty and interdependence challenges (Krishnan and Ulrich, 2001). We control for whether the project spans multiple countries, accounting for cross-border process heterogeneity. Project load distribution controls for fairness in work effort and removes confounding from overloaded projects or bottlenecks that delay timelines irrespective of distance (O’Leary et al., 2011). Finally, the number of project concerns controls for project problems and issues that require ongoing attention.
Model and results
Our identification assumption is that, conditional on fixed effects and controls, EGD is not systematically correlated with unobserved determinants of project delay. We address this with employee fixed effects to account for all time-invariant employee heterogeneity, and calendar month and year fixed effects to eliminate seasonality and time trends. We also implement a comprehensive set of employee- and project-level controls to address observed heterogeneity. Standard errors are three-way clustered at the project, employee, and month levels, providing conservative inference (Abadie et al., 2023). Additionally, we conduct a range of supplementary analyses that support our primary specification. Our main model is estimated as:
Pairwise correlations are reported in Table A-2 (E-Companion). The main estimates appear in Table 2.
Main results.
Main results.
Margins plots for the moderators.
We find support for H1. The coefficient on EGD is 47.463 (p = .030), indicating that greater geographic distance is associated with longer project delays. The interaction terms for employee familiarity and project similarity are −1.033 (p = .015) and −20.934 (p = .010), respectively, supporting H2a and H2b. The interaction terms for employee specialization and project cross-functional geographic dispersion are −6.536 (p = .041) and 14.711 (p = .046), respectively, supporting H3a and H3b.
To aid interpretation, Figure 3 reports marginal effects. For binary moderators (employee specialization, project cross-functional geographic dispersion), margins are evaluated at 0 and 1. For continuous moderators, margins are evaluated at substantively meaningful low, mean, and high values. For employee familiarity, we use 0 (since the variable is bounded at zero), the sample mean, and the mean + 1 SD. For project similarity, we use mean − 1 SD, the sample mean, and the upper bound of 1 (since the variable is bounded above at 1). When employee familiarity is low, a one-unit increase in log distance is associated with approximately 29 delay days, whereas at high familiarity the effect declines to about 5 days and is statistically indistinguishable from zero. Similarly, when project similarity is low, the marginal EGD effect is approximately 24 days, compared with about 14 days at high similarity. Specialists experience a smaller distance penalty, about 16 delay days per log unit, compared with about 23 days for non-specialists. In contrast, cross-functional geographic dispersion amplifies the effect of EGD: when R&D and production are located in the same country, the marginal effect is about 16 days, whereas when they are located in different countries, it increases to approximately 31 days.
We conduct multiple additional analyses, which are summarized in Table 3. Several potential selection mechanisms could, in principle, threaten our empirical design. The first concern is self-selection, meaning that employees may choose projects that are easier or better aligned with their personal incentives, which could confound the observed relationship between EGD and project delay. However, discussions with managers at the focal firm confirm that project assignments are made through a centralized, top-down allocation process, whereby managers assign employees to projects based on organizational needs rather than individual choice. This structure minimizes the scope for self-selection that has been documented in more decentralized settings (e.g., lawyers choosing their cases).
Summary of additional analyses.
Summary of additional analyses.
A second potential source of selection bias is residential relocation, whereby employees may move geographically as their work–life preferences evolve, indirectly altering their EGD through reassignment to local projects. 4 For example, an employee who prioritizes family time might request a transfer closer to headquarters, and the resulting decrease in EGD could be misinterpreted as an improvement in coordination efficiency. Our specification with employee fixed effects accounts for each worker's baseline residential choice, thereby reducing cross-sectional bias due to static preferences. Nonetheless, if such moves were common, unobserved lifestyle traits could still confound the estimates. In practice, only 3% of employees change sites during the observation window. When we exclude these “movers” and re-estimate the main model, the results remain virtually unchanged (see Table A-3 in the E-Companion). This suggests that selection effects, both from self-assignment and relocation, are unlikely to explain the observed relationship between EGD and project delay in our setting.
Third, a related selection concern is strategic project staffing, in which managers might reassign employees across projects (and thereby change their EGD) in response to project delays (Hutchison-Krupat and Kavadias, 2015). If such reactive reallocation occurred, our estimates could reflect managerial intervention rather than the impact of EGD itself. Discussions with the focal firm indicate that project staffing is typically established before initiation and rarely altered during the project lifecycle, as mid-course adjustments can spill over to other projects. We test this empirically by regressing EGD on 1-, 2-, and 3-month lagged project delay to detect hidden pre-trends. Across specifications, the lag coefficients are small and statistically insignificant, indicating that prior delay does not predict subsequent changes in EGD (Table A-4 in the E-Companion). We therefore conclude that strategic staffing is unlikely to drive our results.
Finally, we examine whether observable employee characteristics are associated with EGD. We present descriptive plots of EGD by age, gender, seniority, work experience, and managerial responsibility (Figure A-10 in the E-Companion). For visualization, age and experience are grouped into terciles, while gender, seniority, and managerial responsibility are binary. Although the plots suggest that EGD may decline with age and tenure, a regression of EGD on employee characteristics (Table A-5 in the E-Companion) shows no economically meaningful associations, indicating that these traits are broadly evenly distributed with respect to EGD.
We examine several split-sample tests to assess contextual project variables (stage-gates, risk, innovation, collocation, and internet connectivity) that are theoretically expected to affect coordination costs. Across these analyses, the association between EGD and project delay is generally stronger when coordination is more demanding and weaker when it is facilitated, which is consistent with the proposed attention-allocation and information-search cost mechanism.
Variation by project stage-gate: Coordination frictions are expected to be lower in early project gates, where work is more modular and parallelizable, and higher in later gates, where schedules become more rigid, and coordination costs amplify. For instance, late design modification can cascade through procurement and launch schedules, requiring greater coordination effort. We therefore estimate Equation (2) separately for two samples: Gates 1–4 (concept, approval, pre-production) and Gates 5–7 (development, ramp-up, launch). Consistent with the expectations, results (Table A-6 in the E-Companion) show that the association between EGD and project delay is positive and marginally significant in later gates (36.203, p = .08) and non-significant in earlier gates (24.198, p = .233).
Variation by project risk: The association between EGD and project delays should also be stronger under high-risk conditions, where denser coordination, tighter monitoring, and iterative problem-solving should amplify attention and search demands. In contrast, when risk is lower and tasks proceed predictably, EGD should carry fewer coordination frictions, such as less problem-solving, fewer interruptions, and less intensive information exchange. We utilize monthly risk ratings recorded by managers, which reflect the number and complexity of active issues (e.g., technical failures, resource shortages, financial bottlenecks), and categorize projects as low or high risk. We estimate Equation (2) separately for the two project risk samples (Table A-7 in the E-Companion). Consistent with the expectations, we find that the EGD coefficient is positive and significant for high-risk projects (54.221, p = .002) and non-significant for low-risk projects (−18.844, p = .420).
Variation by project innovation level: The association between EGD and project delay should also intensify as projects become more innovative, since novelty increases uncertainty, interdependence, and the need for synchronous problem-solving. These factors heighten attention allocation and information-search costs. Highly innovative projects typically involve more experimentation, design revisions, and coordination across dispersed employees, making it particularly demanding. In contrast, low-innovation projects typically involve more routine tasks and established templates. To test this, we split the sample using the project-level innovation classification recorded by the focal company and estimated the model separately for high- and low-innovation projects (Table A-8 in the E-Companion). Consistent with the expectations, the EGD coefficient is positive and significant for high-innovation projects (80.105, p = .025) and non-significant for low-innovation projects (−9.538, p = .575).
Variation by collocation: In theory, collocation within the same work site should drastically reduce coordination costs and hence the association between EGD and project delay. While we do not observe fully collocated projects (i.e., within the same office), we still observe a high degree of work-site collocation across several projects. Hence, we construct a collocation share variable measuring the proportion of project members located on the same site as the focal employee. We flag projects as highly co-located when at least 95% of members share the same site and as highly dispersed when fewer than 10% do. We then re-estimate the model separately for these two samples (Table A-9 in the E-Companion). Consistent with the expectations, the EGD coefficient is non-significant in the highly co-located sample (51.085, p = .426) but positive and significant in the highly dispersed sample (644.386, p = .040).
Variation by internet connectivity: Internet connectivity can mitigate information search costs of EGD, as higher bandwidth should accelerate information retrieval and usage. We proxy connectivity using Akamai's country-level share of households with average speeds above 4 Mbps (Akamai, 2017). We classify countries as high or low-connectivity (above/below the median) and re-estimate the model for these two subsamples (Table A-10 in the E-Companion). We observe a non-significant EGD coefficient in the low connectivity (222.961, p = .103) and a positive and significant coefficient in the high internet connectivity sample (77.297, p = .025). The results are directionally consistent with the theoretical expectation, but we do not find statistically conclusive evidence that internet connectivity moderates the EGD–delay relationship.
Exploring an alternative source of coordination costs: Temporal misalignment
A plausible alternative is that EGD proxies temporal rather than spatial frictions. Following Chauvin et al. (2024), we examine this channel using a difference-in-differences (DiD) design around daylight saving time (DST) transitions. For each employee–project–month, we set a 
We find that the main EGD coefficient changes only modestly and remains statistically significant (41.399, p = .050). By contrast, the DiD term
Our findings extend the literature on NPD, project management, and global work. First, our framework and findings contribute to the literature on the impact of geographic dispersion on employees, teams, and organizations (e.g., Anderson and Parker, 2013; Gokpinar et al., 2014; Reiche et al., 2019; Salvador and Madiedo, 2021). EGD fragments limited attentional resources and requires increased effort to find, access, and integrate relevant information over geographic distance. Interestingly, we find that effects concentrate where coordination costs are higher, at later-stage gates, and in high-risk and high-innovation projects. Our study responds to the recent calls for longitudinal design in global work (Froese et al., 2025), and we further extend previous research in project management and NPD that studies the antecedents of project performance, showcasing the role of EGD for the timely delivery of projects (Crama et al., 2019; Kavadias and Loch, 2003; Krishnan and Ulrich, 2001; Staats, 2012).
Second, we show under which conditions EGD is less harmful for project delay. We find that the EGD–project delay link attenuates with higher employee familiarity and project similarity, consistent with lower attention-allocation costs. This finding is not trivial because previous research has almost always advocated for collocation, whereas our research shows that geographic distance is not always detrimental (KC and Staats, 2012; Staats, 2012). A counterintuitive finding is that the effect of EGD on delay is attenuated for specialists. Prior work hints that generalists should be better able to transfer and integrate knowledge over geographic distance, whereas specialists’ deeper, more interdependent, organization-specific work raises adjustment and coordination costs (De Stefano and Bidwell, 2026; Fahrenkopf et al., 2020; Sabel and Sasson, 2023). Our findings are more consistent with the logic that specialists have a narrower search space that speeds up the identification and integration of relevant knowledge (Nagle and Teodoridis, 2020). In turn, the EGD–delay link amplifies under cross-functional geographic dispersion (i.e., when R&D and production are split geographically), where translating specifications and matching interfaces require extra clarification loops and approvals. These findings extend prior work on the role of interfaces and modularity in distributed innovation (Salvador and Madiedo, 2021; Sosa et al., 2004).
Managerially, as firms increasingly rely on a globally distributed workforce to leverage diverse expertise and optimize resource allocation, they face a continuous trade-off between time-to-market and collocation. Our results suggest that firms can mitigate the delay penalty associated with EGD by increasing employee familiarity, assigning employees to similar projects, and leveraging specialist roles. At high levels of employee familiarity, the marginal effect is not statistically distinguishable from zero, but rises to roughly 29 delay days at low familiarity. A similar gradient appears for project similarity, with the effect increasing from about 14 days at high similarity to approximately 24 days at low similarity. When assembling projects that span, for example, Europe and China, managers can prioritize prior collaboration among interdependent team members or assign employees to technically similar projects. Where such allocations are not feasible, managers can recreate these conditions by deploying interface playbooks, shared glossaries, and process templates, thereby reducing the need for repeated clarification cycles. While prior research suggests that generalists may ease coordination by bridging interfaces, our results indicate that specialists experience a smaller distance penalty, 16 delay days per unit of EGD, compared with 23 days for generalists. Cross-functional geographic dispersion amplifies the penalty and therefore requires tighter interface management, clearer handoff ownership, and additional buffer time. When R&D and production are co-located within the same country, the marginal effect is about 16 days; when they are geographically separated, it nearly doubles to about 31 days. In such settings, project managers can move interface freezes earlier, designate a handoff owner with explicit verification authority, and budget additional gate time.
Our study has several limitations. First, we make several attempts to limit endogeneity concerns by testing robustness and proposing alternative explanations. Still, a possible future research direction is to investigate the relationship using a supply-side shock (e.g., a change in employee distance resulting from new government regulations or immigration policy). Second, our data comes from a single large company, and our findings might not directly apply to all settings. Therefore, we call for future firm-level studies to investigate the relationship between EGD and key variables in other settings. Finally, we do not explicitly account for all potential frictions in project work. While we do rule out temporal misalignment as a confounding friction, other coordination mechanisms remain unobserved and warrant further investigation. For example, we do not observe other frictions, such as team-level communication breakdowns, issues with knowledge transfer (e.g., local documentation idiosyncrasies), or country-level constraints (e.g., regulatory and security rules).
Overall, while research often views collocation as a panacea for optimal performance, we demonstrate that geographic distance is not universally detrimental and can be mitigated.
Supplemental Material
sj-pdf-1-pao-10.1177_10591478261443748 - Supplemental material for Far apart, slower together: Employee geographic distance and project delay in global new product development
Supplemental material, sj-pdf-1-pao-10.1177_10591478261443748 for Far apart, slower together: Employee geographic distance and project delay in global new product development by Tuuli Hakkarainen, Anatoli Colicev and Torben Pedersen in Production and Operations Management
Footnotes
Acknowledgments
The authors thank the SMS 2024 conference participants in the Strategic Human Capital track and the EIBA 2024 conference participants in the Human Resource Management track, as well as Matthew Bidwell, Miguel Espinosa, and Deepak Somaya for their valuable feedback. The authors also thank department editor Glen Schmidt, the senior editor, and two anonymous referees for their insightful and constructive comments throughout the review process.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Notes
How to cite this article
Hakkarainen T, Colicev A and Pedersen T (2026) Far apart, slower together: Employee geographic distance and project delay in global new product development. Production and Operations Management xx(x): 1–17.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
