University of Groningen Faculty of Economics and Business
Master Thesis
“The Impact of Project Flexibility on NPD Project Performance in Technologically Uncertain Environments”
MSc. Business Administration Strategic Innovation Management
June 2018
Ingo Ruijs S3531023
Supervisor: Dr. W.G. (Wim) Biemans Co-assessor: Dr. J.D. (Hans) van der Bij
Word Count: 11.565
2 Abstract
New product development (NPD) has been a way for firms to remain competitive in the current dynamic markets. The literature field on NPD management is in disagreement on whether to use a more formal, structured process in managing NPD, or taking a more flexible approach, to allow for adaptation to environmental changes during the project. Biazzo (2009) proposes that NPD processes can be simultaneously structured and flexible to achieve the high- est results. To illustrate this Biazzo (2009) proposed a framework of three flexibility dimen- sions.
Previous research inspected the relation between parts of these three dimensions and one single performance measure (i.e. time to market). No study has empirically tested the full model to ascertain which effect a certain type of flexibility has on NPD project performance as a whole. This research aims to fill this gap in the literature and provide an overview of the flexibility-related antecedents of project performance as a multi-dimensional construct. By do- ing so, this research will assess whether simultaneous structuration and flexibility in NPD pro- jects can benefit the performance of the project on multiple levels. Additionally, the contingent effect of technological turbulence is hypothesized to increase the need for flexibility.
By analyzing NPD projects in 49 Dutch firms through hierarchical multiple regression we find that the efficiency of an NPD project is negatively influenced by the degree of formality of the project’s process structure. And that market share is positively influenced by stage over- lap during the project. Both findings indicate a need for flexibility, no need for simultaneous structuration and flexibility has been found, in contrast with Biazzo (2009). No moderating effects are found for technological turbulence.
Finally, implications for theory and practice are discussed and directions for future re- search will be provided, based on the current study’s limitations.
Keywords: new product development, organizational information processing theory, process
flexibility, innovation performance, flexibility, Stage-Gate
3 Introduction
The current business environment is characterized by turbulence and uncertainty due to tech- nological change and globalization (Fischer, 2017, p. 2). New product development (NPD) has been an important way for firms to cope with this uncertainty and changing environments and has been a way to remain competitive. (Brown & Eisenhardt, 1995; Eisenhardt & Tabrizi, 1995;
Fischer, 2017; Thomas, 2014). But this uncertainty itself can play a large role in the perfor- mance of NPD projects (Ahmad, Mallick, & Schroeder, 2013). The current literature on NPD processes is dichotomized on which approach to take in handling this uncertainty (Biazzo, 2009). The most prevalent approach is a structured and rigid approach of structuring NPD pro- cesses, best illustrated by the influential Stage-Gate model (Cooper, 1990). Large parts of the NPD literature have previously tied this structured approach to high performance (Cooper, 1990; Eisenhardt & Tabrizi, 1995; Griffin, 1997; Kleinschmidt, De Brentani, & Salomo, 2007;
Valeri & Rozenfeld, 2004). Many firms still employ a highly structured and rigid form of NPD project management (Cooper, 2008). On the other hand there is the literature stream that states that this uncertainty calls for a more flexible approach of strategy, as a static approach and rigorous planning are simply not sufficient for managing these levels of uncertainty (Eisenhardt
& Brown, 1998; Eisenhardt, Furr, & Bingham, 2010; Kalyanaram & Krishnan, 1997; Verganti, 1999).
Fischer (2017) states that these two conflicting approaches lead to confusion for both academics and practitioners and suggests that this confusion can be overcome by looking to strategies that are both flexible in some parts of the process and structured in others. These structures are called semi-structures (Eisenhardt et al., 2010), and these are linked to the most successful product portfolio’s (Brown & Eisenhardt, 1997). These are further described this as balancing on the edge of extreme structures of flexibility and structure (Brown & Eisenhardt, 1997). This adoption of simultaneously structured and flexible processes calls for clarification of which dimension of structure should be (less) flexible (Fischer, 2017). To clarify the dichot- omy between structured and flexible processes, Biazzo (2009) proposes a conceptual frame- work consisting of three distinct NPD process dimensions; 1) organizational dimension, 2) informational dimension and 3) temporal dimension.
Current literature in the field of NPD processes has mainly focused on defining flexi-
bility and the three dimensions by Biazzo (2009)(Fischer, 2017, p. 2). So far, meta analyses
4
conducted in this research project (e.g. Fischer, 2017; Borgeld, 2016) have proven to be un- fruitful due to the limited amount of studies relating variables to the flexibility dimensions.
Moreover, no study has empirically tested this model to ascertain which effect a certain type of flexibility has on NPD project performance as a whole. There have been studies that have looked at the relation between these three dimensions and one single performance measure (i.e.
time to market). This research aims to fill this gap in the literature and provide an overview of the flexibility-related antecedents of project performance as a multi-dimensional construct, to better understand the complex relationships in place.
The aforementioned literature gap and goal of this research have led to the development of the following research question as guiding principle for this study;
RQ1 How do the different flexibility dimensions affect NPD project performance?
Additionally, this study will explore the role of turbulence in moderating the relation- ship between the flexibility dimensions and NPD project performance. Biazzo (2009) proposes that the uncertainty regarding markets and technologies an industry faces can decline or in- crease the effectiveness of the used development process. This is something that in the current literature has not been empirically tested with relation to the flexibility dimensions of Biazzo (2009). Therefore, the following secondary research question has been developed to test the contingent factor of turbulence in the model of Biazzo (2009).
RQ2 How does technological turbulence moderate the relationship between the flexibility di- mensions and NPD project performance?
This research contributes to the extensive field of NPD project literature by empirically
grounding Biazzo’s dimensions of NPD project flexibility in the organizational information
processing theory (OIPT). The relevance of the OIPT in assessing the need for flexibility in
today’s business environments will be further solidified. Secondly, the empirical analysis will
further strengthen the theorized relationship between Biazzo’s (2009) dimensions of process
flexibility and project performance and will test for possible detrimental or enhancing effects
of the three separate dimensions on specific performance dimensions. Finally, the contingent
factor of technological turbulence will be used to assess whether the relationship between the
flexibility dimensions and performance are dependent on the level of uncertainty caused by
this turbulence. Practitioners can use the results from this study to better handle uncertainty
and turbulence in their NPD processes, by understanding the complex relationship between the
5
three flexibility dimensions, uncertainty and project satisfaction. Ultimately leading to higher NPD performance.
In order to answer the research questions, data was gathered on 49 innovative Dutch firms. The data was analyzed through hierarchical multiple regression. Findings indicate that the efficiency of an NPD project is negatively influenced by the degree of formality of the project’s process structure and that market share is positively influenced by stage overlap dur- ing the project. Both findings indicate a need for flexibility. No need for simultaneous structu- ration and flexibility has been found, in contrast with Biazzo’s (2009) propositions. No mod- erating effects are found for technological turbulence.
In the following sections, first, a literature review will be performed in which the cur- rent research is positioned in the OIPT and hypotheses will be developed based on this logic.
Secondly, the methodology of the study will be presented, discussing data collection strategies, final sample, measurement scales and the methods of analyses. Thirdly, results will be pre- sented. Here descriptive statistics will be provided, and regression results will be analyzed.
And finally, the study will be concluded and discussed. Theoretical and managerial implica- tions of the study are discussed and suggestions for future research are provided based on the current study’s limitations.
Literature review
The main literature streams used in this research are the Organizational Information Processing Theory by Galbraith (1974) and Tushman & Nadler (1978) and the NPD flexibility literature with a specific focus Biazzo's (2009) paper. The level of analysis for this research will be the project level of NPD. The following sections will discuss these two main literature streams and their position in this research.
Organizational Information Processing Theory
The notion that uncertainty increases the need for information processing during task
execution is the main driver behind the Organizational Information Processing Theory (OIPT)
as proposed by Galbraith (1974). If a task is well-understood before being performed, most of
the activity can be planned beforehand. But when the task is less well understood, and there is
thus a form of uncertainty about the task, there is a higher need of information processing
during execution of the task (Galbraith, 1974). This additional knowledge processing leads to
6
changes in resource allocations, schedules and priorities. All these changes require information processing during task performance. Therefore the main notion of the OIPT is “the greater the task uncertainty, the greater the amount of information that must be processes among decision makers during task execution in order to achieve a given level of performance”(Galbraith, 1974, p. 28).
In case of changing levels of uncertainty, integrating mechanisms should be adopted to handle this change. These mechanisms aim to coordinate action across a large number of inter- dependent functions, as executors of those functions are not able to communicate with all roles they are interdependent with. Galbraith (1974) identified three integrating mechanisms: 1) co- ordination by rules or programs, 2) hierarchy and 3) coordination by targets and goals.
Galbraith (1974) mentions the limiting factor in handling uncertainty is the ability to handle non-routine events through referring them upward through hierarchical channels.
Therefore, in situations of increasing uncertainty the organization must act by changing the organizational design. Galbraith (1974) proposed two ways to reduce the need for information processing and two additional ways of increasing the organization’s capacity to process infor- mation, with the goal of better handling uncertainty. The need for information processing can be reduced though the creation of slack resources and creation of self-contained tasks. The organization’s capacity to process information can be increased through investment in vertical information systems and creation of lateral relations. In the following paragraphs Galbraith’s (1974) explanation of these options will be given.
Creation of slack resources reduces the need for information processing by incorporat- ing several buffers. This reduces the overload on hierarchy since exceptions will be less likely to occur, as there is more capacity to handle uncertainty built into the project. Examples of slack resources are the extension of planning, inventory buffers and larger budgets.
Creation of self-contained tasks is initiating a shift from an input focus to an output focus by creating self-contained groups. This entails supplying the groups with the input they need to reach the desired output. This reduces the need for information processing in two ways.
First, it reduces output diversity. And secondly, information reduction is reached through re- duced division of labor.
Investment in vertical information systems is the investment of making information eas-
ier to transfer upward in the direction of the hierarchy, without overloading this. Formalization
of information makes it possible to transfer an equal amount of information, with a smaller cost
7
in the sense of understanding the information higher in the hierarchy. This leads to easier de- cision on revision of plans in the case of explicit knowledge, this system can be really effective in transferring information needed for decisions. If the information is more tacit or ambiguous, the decision is best resided down in the operational unit.
Creation of lateral relations is moving down the decision making to the place where the information resides, without using self-contained tasks. This destresses the load of infor- mation on the hierarchy through several different lateral processes. Depending on the number of interactions and their complexity the choice for a specific lateral process will be made. In cases of infrequent interactions and low complexity direct contact between team managers suffices. When complexity and frequency of interactions increases the lateral process needed to manage it moves up to; liaison roles, task forces, teams, integrating roles, managerial link- ing roles and eventually a matrix organization can be created
Organizations must employ at least one of the four aforementioned strategies when fac- ing increased uncertainty (Galbraith, 1974). Tushman & Nadler (1978) further developed this theory by adding to this that the choice for a specific information processing strategy must always match with the information processing requirements that are tied to the organizational form and possible control and coordination mechanisms. This need for information processing is defined by the level of the unit’s uncertainty, which is comprised of 1) the unit’s task char- acteristics, 2) the unit’s task environment and 3) inter-unit task interdependence (Tushman &
Nadler, 1978). Additionally, Galbraith (1974) notes that in the case of developing new products or entering new markets, organizations must make some provisions for increased information processing, which is of critical importance for the current study and will be used to construct hypotheses.
NPD Project Flexibility
Biazzo (2009) states that the current literature focusses on a dichotomized approach
towards structuring the NPD process. On the one side there is the stream that emphasizes the
advantages gained by using a structured approach in the NPD process, commonly associated
with the original Stage-Gate model (Cooper, 1990). This approach favors setting an early prod-
uct definition, to avoid costly adaptations later on in the process (Verganti, 1999) and using a
formally structured process consisting of clearly defined tasks and go / no-go decisions. On the
other hand, there is the other literature stream that argues that a highly structured process with
for example an early freeze of product definition might lead to a product that is outdated by the
8
time it is launched, especially in technologically turbulent markets. (Eger, Eckert, & Clarkson, 2005; Kalyanaram & Krishnan, 1997; Verganti, 1999). And therefore, this stream argues that the flexible processes should be employed, in order to reap benefits from newest market and technological knowledge.
The literature is clearly not in agreement on the effects of flexibility on the performance of an NPD project. As performance is a broad construct consisting of multiple dimensions, it is not unthinkable that certain flexibility dimensions only influence certain dimensions of per- formance, rather than the total construct of performance as a whole. Different dimensions of performance have to be reviewed to assess the complex relationship in place. This notion is supported by Fischer (2017) who recommends future research in the field to use a more differ- entiated measure of performance to assess these relationships. Biazzo (2009) proposes a way of organizing NPD projects that is simultaneously structured in some regards, and flexible in others. Biazzo (2009) has identified three dimensions of NPD process flexibility which will be used in this research to assess which forms of NPD process flexibility have positive or detri- mental effects on NPD project performance. These three dimensions are the informational di- mension, organizational dimension and temporal dimension.
In the next subsections these three dimensions of flexibility will be further explained and hypotheses for the relationships between these three dimensions and NPD project perfor- mance will be constructed based on the logic of the OIPT.
Informational dimension. The informational dimension deals with classifying the develop- ment activities and investigating the firm’s product definition approach (Biazzo, 2009, p. 336) The informational dimension of flexibility can be divided into two main categories; 1) problem formulation and 2) problem solving (Fischer, 2017). The problem formulation category relates to the product definition (Bacon, Beckman, Mowery, & Wilson, 1994; Biazzo, 2009), which can be set either early or late depending on the which stream of literature you follow. Problem solving comes after the formulation of a problem and consists of the ‘engineering’ of the solu- tion. The process of reaching the product definition is most often not linear and can take mul- tiple iterations, in which information is processed, before reaching the ‘freeze’ of the definition (Bacon et al., 1994; Biazzo, 2009).
Based on earlier work in this research project by Fischer (2017) we have identified a
late setting of the freeze time as an important predictor of the informational dimension. Bacon
et al. (1994) state that the process of reaching the product definition is iterative, and is influ-
9
enced by changes in customer needs, competitor information, technological risks and opportu- nities and the regulatory environment happening after the initial product definition, also the problem solving influences product definitions in a way that they may need to be redefined.
Bacon et al. (1994) state that setting the product definition early on in the process is infeasible for long-term projects, due to dynamic and uncertain markets and changes in technologies in- volved with the project.
There is a large stream of literature supporting this notion, arguing that a structured process with an early freeze of product definition might lead to a product that is outdated by the time it is launched, especially in turbulent, dynamic markets. Since uncertainty is highest at the early stages of the NPD process, making the decision to freeze here is based on uncertain assumptions and might be detrimental for performance of the project. (Kalyanaram &
Krishnan, 1997; Verganti, 1999). Therefore, it is argued that the product definition should be pushed back, in order to reap benefits from newest market and technological knowledge.
In relation with the OIPT, organizations performing NPD processes have a higher need for information processing. This can be achieved through the creation of slack resources, in this case allowing more time to gather information to set the product definition. By doing this, not all the information needs to be processed during the early uncertain stages, but it can be done during the execution of the project. This information processing is manifested in the feed- back received on different iterations of the product definition (Galbraith, 1974). Setting a late design freeze allows for more information processing, this leads to the product being better adapted to the latest market information and will thus score higher on the performance indica- tors. Thus;
H1: Implementing a late product definition freeze positively influences the NPD project per- formance.
Organizational dimension. The organizational dimension of flexibility refers to the structu-
ration of the process (Biazzo, 2009, p. 336). Classically this dimension can be explained
through the application of a stage-gate-like structure as proposed by (Cooper, 1990). In which
the development process is split up in multiple stages, with go/no go decisions at the end of
each stage, also known as gates. This high level of structuration has previously been linked to
NPD success (Cooper, 1990; Eisenhardt & Tabrizi, 1995; Griffin, 1997; Kleinschmidt, De
Brentani, & Salomo, 2007; Valeri & Rozenfeld, 2004). This can be linked to the ability it cre-
ates to make project members understand the required tasks for the project and the key routines
related to it (Bonner, Ruekert, & Walker, 2002; Kleinschmidt et al., 2007).
10
But on the other hand, there are streams in the NPD literature that find that a highly structured process with a lot of upfront planning can also limit the adaptability to changes (Bonner et al., 2002; Buganza, Gerst, & Verganti, 2010; Eisenhardt & Tabrizi, 1995) and cre- ativity of team members (Bonner et al., 2002), or increased bureaucracy and inefficiency (Kleinschmidt et al., 2007), ultimately leading to decreasing performance. Furthermore, recent adaptions to the stage-gate model (Cooper, 2008, 2017) have adjusted the original, structured, model to allow for more flexibility in stage-gate structure.
It is clear that the literature is not in agreement on the effect of the use of structured processes on NPD performance. But, when looking at this relationship from the perspective of the OIPT, high project process structure limits the ability to create slack resources when facing uncertain situations. Meaning that the organization does not have the ability to process all the information it might need to fully tackle the unforeseen problems created by a changing envi- ronment. This low fit between information processing needs and capabilities will eventually lead to the product being less suited for the environment it was developed for (Tushman &
Nadler, 1978). Thus,
H2: The degree to which the project uses a formal process structure is negatively related with NPD project performance.
Temporal dimension. The temporal dimension of flexibility relates to the execution strategies of development tasks (Biazzo, 2009, p. 336). Different stages in the process can either be per- formed sequentially, or they can overlap, leading to a form of concurrent engineering (Biazzo, 2009; Borgeld, 2016). This overlap has been defined as the most important antecedent of NPD performance in the temporal dimension of flexibility by earlier research in the current research project (Borgeld, 2016).
In engineering management literature concurrent engineering and overlapping stages have been identified to contribute to shorter development times, improved product quality and lower development costs (Yassine & Braha, 2003). This originates in the possibility it brings to share relevant information with development participants working other stages of the pro- cess, for example through direct contact or liaison roles. (Fischer, 2017; Yassine & Braha, 2003)
Linking this to OIPT; the increase of information sharing across ‘teams’ that work on
different development stages increases the information processing capabilities. In the OIPT
this sharing of information laterally between different development stages is called the creation
of lateral relations. This moves down the decision making to the place where the information
11
resides, the NPD project itself, without using self-contained tasks. This is especially preferred over the creation of vertical information systems in the case of tacit and ambiguous information, which is harder to transfer. The creation of lateral relations destresses the load of information on the hierarchy through designing several different lateral processes. These processes are de- pendent on the size and complexity of the issue the lateral relations are being deployed for (Galbraith, 1974).
This increase in information processing capabilities at lower levels in the hierarchy leads to a better fit between information processing needs and requirements in uncertain situa- tions. Thus:
H3: the degree to which the project makes use of overlapping stages is positively related with NPD project performance.
Moderators
Turbulence. Turbulence in terms of technologies associated with a development pro- cess increases uncertainty involved with the project (Bacon et al., 1994; Biazzo, 2009;
Eisenhardt & Tabrizi, 1995; Thomas, 2014; Verganti, 1999). The OIPT states that organiza- tions performing NPD processes have a higher need for information processing in dynamic, uncertain markets and when working with uncertainties regarding the technologies involved (Galbraith, 1974; Tushman & Nadler, 1978). The growing uncertainty caused by technological turbulence will only further increase the need for information processing, and thus flexibility (Buganza et al., 2010). So, following the logic of the OIPT, firms that are able to process more information during the development process in highly uncertain environments, and are thereby more flexible, attain a better fit between information processing needs and capabilities (Tushman & Nadler, 1978). This better fit will ultimately lead to a higher performance (Tushman & Nadler, 1978). Therefore, the moderating effect of technological turbulence on the flexibility-performance relations will always be a strengthening one. Or;
H4a: Turbulence strengthens the positive relationship between the use of a late product defi- nition freeze and NPD project performance, making it more positive
H4b: Turbulence strengthens the negative relationship between the formality of the project’s process structure and NPD project performance, making it more negative
H4c: Turbulence strengthens the positive relationship between the use of overlapping stages
and NPD project performance, making it more positive
12 Conceptual Model
Figure 1 shows the theoretical model of the effects of the three NPD flexibility dimen- sions on NPD project performance, moderated by turbulence.
Methodology
Since the literature field on NPD project flexibility is already a maturing field, and the most important constructs for this research have already been defined, but there are still litera- ture gaps to be filled, a theory testing approach is most appropriate for this study (Aken, van, Berends, & Bij, van der, 2012). The aim of performing theory testing on this rather mature field is to fill the literature gap previously mentioned. Empirical analyses will be performed through multiple regression, based on primary data gathered through a survey. In the following para- graphs the data collection strategy, sample and measurement scales will be discussed, and fi- nally the steps in preparing the data and the actual analysis of performance antecedents will be discussed.
Data Collection
Data was collected through surveys in Dutch firms that actively perform NPD projects.
The projects that were taken into account for the survey are recently launched products. This choice has been made to ensure the project was still relatively fresh in the mind of the respond- ents. In addition to this requirement the product development has to have taken place inside the company, without development in alliances. It is important to ensure this to be able to correlate
H1 +
H4a+
H4b - H4c+
H3 + H2 -
Turbulence
• Technological Turbulence Informational Flexibility
Time until Freezing
NPD Project Performance Efficiency
• Costs
• Time to market Effectiveness
• Quality
• Specifications
• Market Share
• Profitability
• Commercial Success
Control Variables
• Firm size
• Project size
• Tech. radicalness Organizational Flexibility
Formality of Process Structure
Temporal Flexibility Stage Overlap Figure 1 - Conceptual model
13
company specific data to project data. Additionally, initial market performance already has to be available in order to measure all the performance variables. Furthermore, in order to ensure the firms considered for the sample had an organizational structure in place a preferred mini- mum firm size of 25 employees was chosen. Mind that this number was not a ‘hard’ rule as the underlying reason, the presence of an organizational structure, was most important.
Suitable firms were selected from the ORBIS database to ensure complementary data such as the industry type could be accessed digitally. In a team of 7 researchers the suitable firms were contacted and asked to fill out the survey. The survey measures the effect of NPD project flexibility on NPD project performance for one specific project per firm. For this project leaders or other highly involved members of project teams were asked to assess the three types of process flexibility and other project level variables such as the team size and technological radicalness.
Possible common method bias is being addressed by having different respondents score the dependent and independent variables, following Chang, van Witteloostuijn, & Eden (2010) and Podsakoff, MacKenzie, Lee, & Podsakoff (2003). This avoids project managers assessing performance of their NPD projects in a subjective manner. Generally, the second respondent that was asked to assess the performance of the projects was a (senior) manager who was (op- erationally) less involved in the project and could thus supply more objective performance measures.
Constructs in the survey are based on peer-reviewed literature to ensure proper validity and reliability. The survey has initially been constructed in English, based on the available literature. To avoid misunderstandings for respondents filling out the survey we have chosen to translate the survey to Dutch. Two researchers whose mother-tongue is Dutch translated the survey together, discussing possible pain-points in the translation. After this first run the trans- lation was reviewed by the supervisor of the research team and points of improvements have been communicated to the translators. These improvements have then been processed into the translation, after which the total group of 7 researchers and the team’s supervisor did a last check to confirm the quality of the translation.
Sample
The final sample consisted of 49 firms from the Netherlands. The firms are spread over
multiple industries such as electrical engineering, automotive, healthcare etc. The sample con-
sists of small spin-offs to large multinationals of 112.000 employees. This relatively diverse
14
dataset makes for results that will have a good generalizability to the larger population of firms outside of this study. The limited size of the dataset in a large population however, potentially limits this generalizability and it will mean that analyses will have to remain relatively simple with few predictor and control variables in order to preserve statistical power.
Measurements
Scales for measuring the variables in the model have been constructed based on recom- mendations given by literature. This results in measures that are strongly supported by peer- reviewed articles, for optimal validity and reliability of the scales used to measure the con- structs. The constructs are measured using multi-item scales using a 7-point Likert-scale, unless stated otherwise.
Independent Variables.
Time until Freezing. Time until freezing is defined as “The extent to which the project was completed before fixing the product definition”. This construct is measured using a single item based on measurements by Zirger & Hartley (1996), directly asking how far the project was completed (in percentages) when the project definition was fixed.
Project process structure. The formality of the project process structure is in this study defined as “The extent to which the project’s process uses a formal system that provides a template for activities, routines and reviews to be implemented throughout the stages of the project” (adapted from Kleinschmidt et al., 2007). This construct is measured using combining existing scales of Biazzo (2009) and Kleinschmidt et al. (2007). The construct is measured by items like; “During the project a standardized set of stages and go/no go decisions guided the activities from idea to launch”. This construct is measured by a 7-point Likert-scale ranging from 1= “completely disagree” through 7= “completely agree”.
Stage overlap. Stage overlap is defined as “The extent to which separate stages of the development process are handled continuous rather than sequential.” The scale used to meas- ure the overlap between development stages is adapted from Zirger & Hartley (1996). The single item measuring the construct is “The different tasks during the project were carried out;
1=fully sequential, 2=almost sequential, 3=some overlap, 4=overlap, 5=a lot of overlap,
6=almost concurrent engineering (simultaneous), 7=fully concurrent engineering (simultane-
ous)”
15 Moderator Variable.
Technological turbulence. Technological turbulence is defined as “The extent to which the technology in an industry is in a state of flux.” (adopted from Jaworski & Kohli, , 1993, p.
60). This construct is measured using the technological 4-item turbulence-scale by Jaworski &
Kohli (1993). The construct is measured by items like: “The technology in our industry is changing rapidly”. And is measured using a 7-point Likert-scale ranging from = “completely disagree” through 7= “completely agree”.
Dependent Variable.
Performance of the NPD project. Performance is defined as “the subjective level of the project’s success compared to initial expectations”. Previous research in this research pro- ject by Fischer (2017) recommends using a multidimensional way of measuring performance.
This notion is supported by Mallick & Schroeder (2005) who recommend using broader range of measurement for performance, instead of taking a singular perspective. To avoid an unidi- mensional view on project performance, this study adopts the recommendation given by Mal- lick & Schroeder (2005) and uses scales to measure performance based on a three-stage model, including multiple performance aspects; 1) development process performance, 2) development outcome performance and 3) business outcome. To construct the scales for measuring (multi- dimensional)performance, scales by Schleimer & Faems (2016) and Ahmad et al. (2013) have been adapted and combined. This results in a newly constructed scale including all three stages mentioned above, consisting of 7 distinct performance variables. The seven variables are: 1) product development cost, 2) product quality, 3) technical performance, 4) time to market, 5) market share, 6) overall profitability and 7) overall commercial success of the product. An important side note is that, following Mallick & Schroeder's (2005) recommendations, perfor- mance variables are all measured using a subjective scale, comparing outcome to initial expec- tations. This provides us with a common basis to compare projects and controls for the diversity of projects in the dataset (Mallick & Schroeder, 2005). An example of an item measuring per- formance of the NPD project is: “What is the success of the project with respect to the initial expectations in terms of product development costs?”. The scales used are 7-point Likert scales ranging from 1= “significantly worse than expected” through 7= “significantly better than ex- pected”.
Control Variables. Several variables will be added to the regression to control for pos-
sible confounding effects on performance. Following Jansen, Van Den Bosch, & Volberda
16
(2006) firm size will be added. Larger firms might have the advantage of a larger pool of re- sources to draw from, which might influence different performance components. The firm size will be measured using the natural logarithm of the number of employees on SBU level. Fur- thermore, the project group size will be used because of the implications the group size has on costs and performance of the NPD project. And lastly, technological radicalness, which is a source of uncertainty in NPD projects, will be added as a control variable, using a scale that was adapted from de Visser, Faems, Visscher, & de Weerd-Nederhof (2014).
The complete measurement scales used for this research can be found in ‘Appendix I’.
Analysis
As discussed earlier in the methodology section this study will use hierarchical multiple regression to test the hypotheses that have been constructed based on the theory. This subsec- tion will clarify the specific steps that are taken from preparing the dataset to running the re- gression models.
Before analysis could start the dataset had to be cleaned and prepared for use. Several items in our scales were inversely coded, so these were reversed in order to make the scale interpretable. Several variables such as the firm size, and the size of the project group were positively skewed due to the inclusion of very large firms. Therefore, these variables have been altered by taking their natural logarithm to normalize the data (Field, 2013). Since the regres- sion analysis will also contain moderators, variables included in the interaction terms will be mean-centered after the construction of latent variables based on factor analyses, in order to minimize the chance for multicollinearity.
As this study is based on primary data it is important to determine validity and reliability of the gathered data. In determining construct validity, this research performed an exploratory factor analysis. This is used to determine if the items used in the survey actually measure the construct they were theorized to measure. Since the scales have been based on previous re- search, a closed factor analysis with a predefined number of factors was used. Factor loadings smaller than .40 are omitted from the table. Items that would load on multiple factors, the wrong factor, or no factor at all were to be deleted. After each iteration one item was to be deleted until all the items load to their corresponding factor, and none else.
Following this, an analysis of the construct reliability was performed, to assess whether
the remaining items yield reliable results in measuring the constructs they are tied to. For this,
Cronbach’s Alpha’s was calculated for the different constructs. In the case that the Cronbach’s
17
Alpha can be significantly improved by deleting an item from the construct, this will be care- fully considered based on theoretical importance of the item, and properly reported. When re- sults of the factor analysis and construct reliability are sufficient, latent variables will be con- structed for multi-item variables. Next, descriptive statistics will be calculated for the com- pleted constructs to show means, standard deviations and correlations across constructs. These correlations will be analyzed in order to assess whether these give cause to suspect multicol- linearity.
Finally, a multiple regression will be performed to test the hypotheses. As predictors are based on previous literature, hierarchical multiple regression is considered a more appro- priate regression mode than stepwise regression (Field, 2013). The first model of the regression will consist of only the control variables, to see if these have a strong significant effect on the dependent variable. For the second model the independent variables will be added to test hy- potheses about possible positive or negative significant effects of the three forms of flexibility on performance. Following the influential work on moderation of Baron and Kenny (1986), the third model adds the moderator technological turbulence to test if there is a direct effect between the moderator and performance variable. Additionally, interaction terms will be added to test the moderating effect of turbulence. For all models R-squared and F-values will be an- alyzed to assess whether the addition of new variables into the model lead to a better explana- tion of the dependent variable. Additionally, VIF values will be checked for each regression model to check for multicollinearity.
Construct Validity and Reliability
In this section the results of factor analyses and internal scale reliability tests will be
provided and discussed. Firstly, a factor analysis was performed to assess whether multi-item
scales measure the construct they are theoretically tied to. The two multi-item scales used in
the current model are project process structure and technological turbulence, both consisting
of four items. Exploratory factor analysis has been performed on these eight items. Since these
items are based on two constructs that are well described in literature the factor analysis has
been set to extract two factors. A covariance matrix was used upon which Varimax rotation has
been applied. The threshold for factor loadings has been set at a coefficient of .40, meaning
that lower coefficients will not be considered sufficiently factor loadings. The results are pro-
vided in table 1.
18
Firstly, the KMO value was .658, which is sufficiently above the minimum of .50 and Bartlett’s test of Sphericity was significant (p<.001), indicating that the factor analysis will likely yield useful results (Field, 2013). The eight items loaded on their two respective factors without any cross-loadings, indicating the scales used were statistically distinct and had a proper discriminant validity.
Table 1 - Discriminant Validity Factor Analysis IVs Table 2 – Discriminant Validity Factor Analysis DVs
Rotated Component Matrix
aRotated Component Matrix
aRescaled Component Rescaled Component
1 2 1 2
Structure 1 .894 .078 Performance 1 .045 .880
Structure 2 .628 .127 Performance 2 .572 .335
Structure 3 .872 -.025 Performance 3 .555 .234
Structure 4 .711 -.319 Performance 4 .339 .794
Tech. Turbulence 1 -.091 .808 Performance 5 .883 .052
Tech. Turbulence 2 -.032 .620 Performance 6 .699 .201
Tech. Turbulence 3 .069 .614 Performance 7 .855 .092
Tech. Turbulence 4 (inv) .026 .814
Extraction Method: Principal Component Analysis. Extraction Method: Principal Component Analysis.
Rotation Method: Varimax with Kaiser Normalization.a Rotation Method: Varimax with Kaiser Normalization.a a. Rotation converged in 3 iterations. a. Rotation converged in 3 iterations.
Additionally, since the scale for the dependent variable consists of 7 distinct items an exploratory factor analysis was also performed on these items, to test whether the dependent variable has any underlying constructs. For this a similar factor analysis has been performed as on the independent variables, with the only exception that the number of factors has not been specified beforehand but will be based on the eigenvalue scores of the items. This open type of factor analysis allows us to check for underlying construct without any prejudice about the outcome. The results of this second factor analysis are reported in table 2.
Firstly, the KMO value was .748, which is sufficiently above the minimum of .60 and
Bartlett’s test of Sphericity was significant (p<.001), indicating that the factor analysis will
likely yield useful results. The analysis has resulted in two distinct factors without any cross-
loadings. Performance-items 1 and 4 load high on the second factor, indicating that these form
a different underlying construct than the 5 other items. When looking at the corresponding
questions to these items it becomes clear why. The five items that load on factor one belong to
questions regarding the effectiveness to deliver desirable performance, with items such as mar-
ket share and product quality. Whereas, the two items that load on the second factor are more
19
project efficiency-related items, looking at product development costs and time to market.
These findings are comparable to the two performance dimensions used by Weerd-Nederhof, Altena, Visscher, & Fisscher (2005), focusing on process-related performance and outcome- related performance. It might be worthwhile to examine the possible effects of independent variables on these two distinct performance factors, efficiency and effectiveness. Following (Fischer, 2017) who called for analysis of more distinct performance variables instead of one large latent performance variable, the following analyses will use these two dimensions of NPD performance, along with exploring the avenue of analyzing the separate performance variables as dependent variables in the model.
In addition to these factor analyses construct reliability for the identified factors has been analyzed to assess whether the to-be-constructed latent variables are internally consistent and reliable. Results are provided in table 3. No items have been deleted from the scales, as this would not result in significantly better levels of Cronbach’s Alpha for the scales. Results in table 3 show that the Cronbach’s Alpha’s of the four constructs lie between .669 and .816, and are thus reasonable to good based on Nunnally (1978). This indicates that the constructs are sufficiently internally consistent and can be used to construct latent variables to be used in the regression models.
Table 3 - Construct Reliability
Construct # of items Cronbach’s Alpha # of items deleted
Structure 4 .797 0
Technological Turbulence 4 .701 0
Performance Effectiveness 5 .816 0
Performance Efficiency 2 .669 0
Results
In this section the results from analyzing the dataset will be discussed. First, Descriptive
statistics will be reported, discussing correlations and possible multicollinearity. Next, the mul-
tiple regression analysis will be reported and analyzed. Here hypotheses will be tested
20 Descriptive Statistics
Descriptive statistics are reported in table 4. As can be expected, correlations between the different performance variables are relatively high and significant. This can be explained by for example high performing projects, which outperform expectations on nearly all dimen- sions of performance. While these are correlated there is no concern of multicollinearity amongst these variables, especially since they will never be used together in one model. Control variables show expected signs of correlation with dependent variables, this will hopefully lead to regression models with higher significance and explanatory power. When analyzing corre- lations between independent and dependent variables it immediately comes to attention that there are not a lot of correlations, which might prove to make the regression results insignifi- cant. Additionally, the performance dimension measuring efficiency scores considerably lower than the one scoring effectiveness, signifying the reason we found this to be two different di- mensions of performance.
Table 4 - Correlation Matrix and Descriptive Statistics
Regression Results
Hierarchical multiple regression was performed to test the hypotheses. The first model in each regression will be control for the different control variables that have been defined beforehand. The second model adds the independent variables to analyze direct performance effects. In the third model the moderator variable will be entered to assess if the addition of
1 2 3 4 5 6 7 8 9
1. Performance Efficiency
2. Performance Effectiveness .423
**3. Log (Project Group Size) -.322
*.106
4. Log (Employees) -.157 .232 -.022
5. Technological Radicalness -.205 .298
*.174 .082
6. Freezing .065 .196 .218 -.210 -.108
7. Structure -.248† .133 .065 .068 .-.211 .238
8. Stage Overlap -.227 .062 .293
*.112 .447
**.266† -.017
9.Technological Turbulence -.040 .152 .111 .174 -.041 -.099 -.052 .096
Mean 3.63 4.76 1.92 5.00 4.63 .00 .00 .00 .00
Standard Deviation
1.26 .87 0.71 2.34 1.73 31.18 1.12 1.42 1.02
**=significant at p<0.01, *=significant at p<0.05, †=significant at p<0.1
21
turbulence has any direct effect on performance. Lastly, the fourth model adds the three inter- action terms, testing for a moderating relation of technological turbulence on the relationship between the three predictor variables and performance.
Two separate regressions were performed. The first regression will analyze the rela- tionship between the predictor variables and efficiency of the NPD project. The second regres- sion will analyze the relationship between predictor variables and the effectiveness of the NPD project. Since the two performance dimensions are statistically distinct, two different sets of control variables will be used, with the most relevant controls for the specific performance dimension. Both results are reported in table 5. Prior to moving on to the results of the regres- sions it is important to test for possible multicollinearity. The VIF-values for the efficiency regression have a mean of 1.532 and those for the effectiveness regression have a mean of 1.582. The highest individual VIF value that can be found for both regressions is 2.296. This is well below the generally accepted cut-off of VIF ≥10 (Field, 2013; Hair, Anderson, Tatham,
& Black, 1995), and thus there should be no concern for multicollinearity in the analysis. The next two subsections will discuss the results of the regressions, that are reported in table 5.
Table 5- Hierarchical Regression Analysis Result