Drivers of serendipity
A quantitative approach to the factors that cause serendipity to happen
Master thesis
Master of Business Administration; Entrepreneurship & Innovation track Faculty of Economics and Business
University of Amsterdam
Name: Rob Brouwers Student number: 10902481
Date of submission: 23-06-2017 Version: final
First supervisor: Ieva Rozentale Second supervisor:
2
Statement of originality
This document is written by Student Rob Brouwers who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3 Table of contents Abstract 1 Introduction 6 1.1 Objective 7 2 Research method 9
2.1 Scale development research methods 9
2.2 C-OAR-SE 10
2.3 Adaptations 10
3 Construct, aspects and attributes 11
3.1 The concept of serendipity 11
3.2 Aspect classification 12
3.3 Attribute classification 13
3.3.1 Attributes related to the individual 14
3.3.2 Attributes related to the environment 15
4 Conceptual framework 17
4.1 Items for scale development and testing 17
4.1.1 Items related to the aspects 17
4.1.2 Items related to the individual and environmental attributes 19
4.2 Conceptual framework 21 5 Data 23 5.1 Group 1 23 5.1.1 Group 1 sampling 23 5.2 Group 2 24 5.2.1 Group 2 sampling 24
4
5.3 Sample size 25
6 Survey 26
7 Results 27
7.1 Descriptive statistics 27
7.1.1 Descriptive statistics of the respondents 27
7.1.2 Descriptive statistics of the Independent Variables 28 7.1.3 Descriptive statistics of the Dependent Variable 29
7.2 Assumptions for multiple linear regression 30
7.2.1 Correlations and multicollinearity 31
7.2.2 Linear relationship 33
7.2.3 Multivariate normality 33
7.2.4 Autocorrelation 34
7.2.5 Homoskedasticity 34
7.3 Multiple linear regression analysis 34
7.3.1 Regressing the DV and five main IV’s 34
7.3.2 Coefficients of the separate variables 35
7.4 Independent samples t-test for additional grouping variables 36
8 Discussion and limitations 37
8.1 Scaling serendipity 37
8.2 Drivers of serendipity 38
8.3 Limitations 40
9 Conclusion and avenues for future research 42
References
5 Abstract
Serendipity is an elusive concept often confused with luck. This paper aims to provide a more applied understanding of serendipity to complement the rich qualitative literature on the subject. This is accomplished by relating theorized drivers of serendipity to a proposed scale for measuring the receptiveness to serendipity of an individual. Testing this model was done through a survey conducted to one group of respondents without prerequisites and another group of respondents working in a business center aimed at startups and small businesses. After performing a multiple regression analysis both high workload and openness to experience appeared to be strong predictors of receptiveness to serendipity. Furthermore entrepreneurs scored substantially higher on this variable than non-entrepreneurs. This paper has taken a first step in quantifying the abstract concept that is serendipity and opens up multiple avenues of further research to extend our practical understanding of this phenomenon.
6 1 Introduction
Serendipity is a term coined by an English novelist, Horace Walpole. The term was inspired by a tale of three princes from Sri Lanka (known as Serendip), who were always making discoveries by accidents and sagacity, of things which they were not in quest of (Merton & Barber, 2006). The most well-known scientific discovery that has been attributed to serendipity is the discovery of penicillin by Sir Alexander Flemming. He valued an anomaly in his test results for its potency instead of discarding it as a failure.
In a time where every firm is actively looking for ways to innovate, serendipity concerns finding novelty where one was not actively looking for. This is an interesting approach and questions a lot of rigid deterministic management practices. Though it has to be stressed that serendipity is not just simple luck: “While good luck may befall the inert or lazy, serendipitous discovery occurs only in the course of an energetic quest, a quest in which lucky discoveries of an unanticipated kind can be recognized through alertness and then flexibly exploited” (Denrell et. al, 2003, p.989). Various groundbreaking discoveries have been attributed to serendipity and the concept is getting more and more attention in recent years, from archival research (Tamboukou, 2015) to entrepreneurs in tech-startups (Silverman, 2013).
Being able to influence serendipity could very well be an ability that would allow one to control a rich source of innovative power. However, if an event becomes predictable, it inherently would not classify as being serendipitous. Still a number of scholars have argued that while this is the case, it is possible to enhance the awareness as well as the odds of serendipitous encounters (Cremonini, 2016). Research on influencing this ‘receptiveness’ to serendipity has begun to emerge only recently and focuses on different aspects such as temporal, locational and social contexts (André et. al, 2009; Khefalidou & Sharples, 2016), the digital environment (Cremonini, 2016), information encounters (Sun et. al, 2011; Foster & Ford, 2003), recommender systems (Ge et. al, 2010) and individual insight (Makri & Blandford, 2012).
7 The difficulty with empirically researching serendipity is that it is a subjective phenomenon; it lies in the eye of the beholder (Friedel, 2001). This is also why the aforementioned empirical research on receptiveness to serendipity employs qualitative methods, primarily diary studies (e.g. Khefalidou & Sharples, 2016) and/or interviews (e.g. Makri & Blandford, 2012). While providing interesting insights, these methods do however not allow for generalization and bridging the gap between theory and practice.
1.1 Objective
Serendipity is an abstract concept that holds the promise of novelty. Interesting for its potential as a source of innovative power, but unsuitable for investing resources, due to its inherent incompatibility with prediction and control. Still, serendipity is not a completely random phenomenon, an individual and/or its environment need to be receptive to serendipity to reap the benefits. This thesis aims to find out which factors are the drivers of serendipity, focusing on factors at the individual level as well as the environment in which the individual operates. Therefore, the research question of this thesis is:
Which individual and environmental factors are drivers of serendipity?
This thesis differs from prior research by being the first study that quantifies the abstract construct that serendipity is. This quantification is only possible because of the rich qualitative literature on serendipity on which this thesis builds. This thesis shows that the available literature allows for taking a first step in bringing serendipity in practice and accessible to management jargon. This will be done by using existing qualitative research to propose quantifiable scales that indicate to what extent an individual is receptive to serendipitous encounters. These scales will then be related to individual attributes and environmental characteristics that are expected to influence this receptiveness. This analysis is done on a survey conducted among those who work in the ‘Hooghiemstra’ building, which is
8 a business center in Utrecht with over 100 small businesses and startups working in different sectors and covering a multitude of disciplines as well as among a sample of individuals. These two groups provide the opportunity to gather and compare rich and useful data to analyze the drivers of individuals’ receptiveness to serendipity.
This thesis contributes to the existing literature on serendipity that has produced multiple qualitative results, but does not provide clear grounds for comparisons. The thesis will provide grounds to distinguish which types of individuals have, and which types of environments provide a higher probability for serendipity to occur. While not providing control over serendipity, the results of this thesis help us to better understand the particular features to foster for serendipitous encounters to occur.
This thesis is further structured as follows: Chapter 2 explains the research method. Chapter 3 provides an overview of the concept of Serendipity, the aspect classification and the attributes, based on the current literature. In chapter 4 the conceptual framework is explained. Chapter 5 discusses the rater entity and sample size. Chapter 6 explains the survey. Chapter 7 describes the results of the survey. Chapter 8 will present discussion and limitations. Finally chapter 9 provides conclusions and avenues for future research for this thesis.
9 2 Research method
As explained, the aim of this thesis is to identify the drivers of serendipity by proposing scales that indicate the extent to which an individual is receptive to serendipitous encounters and test these against the expected drivers. And while there is a lot a lot of qualitative research available on serendipity, there is no prior research that approaches serendipity in a comparable, quantitative way to draw upon. Therefore a lot of new ground has to be broken and the research method is crucial to safeguard validity and reliability.
2.1 Scale development research methods
This thesis will be based on an existing scale development research method, to be able to provide the structure it misses due to the absence of quantitative research in the field of serendipity. Scale development is what most clearly distinguishes this thesis from the extant literature and therefore this line of research method is chosen.
An interesting method for scale development is proposed by MacKenzie et. al (2011) which is aimed at MIS and behavioral sciences and is seemingly a good fit for this thesis. The problem however lies in the focus on assessing validity multiple times during the process. As mentioned before, there is no existing comparable literature, therefore construct validity is hard to warrant.
C-OAR-SE on the other hand is a procedure for developing scales to measure marketing constructs, developed by Rossiter (2002). While created for marketing research purposes, this procedure will be used in this thesis because of its emphasis on Object and Attribute Classification. This emphasis provides the means to use the rich qualitative literature to develop the scales. Nevertheless, this procedure cannot be used one-to-one and requires adaptations to be suitable for this thesis, which are discussed further on.
10 2.2 C-OAR-SE
The C-OAR-SE procedure stands for: Construct definition, Object classification, Attribute classification, Rater identification, Scale formation, and Enumeration and reporting. This procedure insists on a clear up-front identification of the concept and classification of objects, attributes and rater entities. For this thesis these aspects are also crucial, because of the abstract concept that serendipity is and the absence of clearly identifiable antecedents.
The C-OAR-SE procedure also addresses validity and reliability in a different way compared to traditional procedures. In C-OAR-SE only content validity is essential, for it warrants that the items properly represent the construct. Face validity is deemed inadequate, because it only takes the items into account that were included; Items that have been omitted and the reason why they have been, are not part of the validation. Construct validity, according to this procedure, is content validity, properly established. The prevailing method of determining construct validity is through a multitrait – multimethod matrix (MTMM), however Rossiter notes: “Validity should be established for a scale independently, and as far as possible, absolutely, not relatively via its correlation or lack of correlation with other measures” (2002, p. 326).
Regarding reliability, again content validity is paramount, because differences when performing test-retests are not a sign of a lack of reliability, but rather of the item not properly representing the construct. Reliability of a scale is instead obtained by composing a sound group of raters, in which case reliability (or precision-of-score) is determined based on the type of attributes and the rater entity. This will be addressed further on in this thesis.
2.3 Adaptations
While following the structure as described in the C-OAR-SE procedure, it is decided to not fully stick to the procedure. First of all, the construct is defined in terms of objects and attributes, but serendipity cannot be classified into different focal objects to be rated. Instead this thesis will define the construct in terms of aspects and attributes, this makes more sense when rating an abstract concept.
11 Secondly, the C-OAR-SE procedure draws the objects and attributes of the construct from open-ended interviews with a sample group of the target raters. In this thesis these objects and attributes will be drawn from the existing literature, because serendipity is a little-known phenomenon and without knowledge of the term, interviews are expected to raise more confusion than useful items.
Finally, while the C-OAR-SE procedure uses input from open-ended interviews for the items, this thesis draws the items from the literature. Items that will be used to scale the aspects are based on qualitative research on serendipity and tested using Q-sorting. Items used for scaling the attributes are all taken from existing tests and therefore do not require pre-tests.
12 3 Construct, aspects and attributes
In this chapter the construct ‘drivers of serendipity’ will be defined based on the aspects and attributes that make up for it. Firstly the concept of serendipity is further explained to give an overview of what is associated with it. Next the aspects derived from this overview are classified and placed in perspective. Finally the attributes will be related to items used in existing literature.
3.1 The concept of serendipity
Merton & Barber (2006) identify serendipity as an ‘‘unanticipated, anomalous and strategic datum that becomes an occasion for developing a new theory or for extending an existing theory’’. Unanticipated, because a serendipitous discovery is a fortuitous by-product, an unexpected observation; anomalous, because a serendipitous discovery is surprising as it appears inconsistent with previous theories or expected facts, and strategic when serendipitous discovery permits implications which bear upon generalized theory (Cunha et. al, 2007). Therefore serendipity illustrates the factors that cannot be captured under prediction and control. Though it is not pure chance, rather some kind of luck that tends to favor prepared minds, i.e. those ready to benefit from it.
Denrell et. al (2003) explain serendipity as “a consequence of effort and luck joined by alertness and flexibility, where the effort was not initially directed to the specific end realized, alertness is required to recognize the lucky appearance of a new possibility and flexibility is displayed in redirecting the effort.” This explanation allows for a clearer understanding of serendipity and is closer to current management jargon. Most managers can relate to the terms effort, alertness and flexibility. Though for research purposes these terms are too vague and hard to control for.
A new perspective is introduced by Dew (p.739, 2009), who conceptualizes serendipity based on three interacting building blocks: “a resource (sagacity), an event (contingencies), and an activity (the
13 individual is already on a journey)”. Whereas Ge et. al (2010) explain serendipity as a function of unexpectedness and usefulness. Another perspective on serendipity is by André et. al (2009), stating serendipity is discovery met by sagacity. Both views combined explain serendipity as being a triangulation of unexpectedness, insight and value (Makri & Blandford, 2012). Yet Sun et. al (2011) define serendipity as an unexpected finding of information followed by an unexpected connection. Finally von Hippel & von Krugh (2015) employ a completely different wording of serendipity: as a discovery of a viable need-solution pair without ex ante identification of a problem.
All in all, serendipity appears hard to pin down, therefore it is suggested by different scholars to not treat serendipity as an orthogonal concept, but rather as a space (Makri & Blandford, 2012) or a continuum (André et. al, 2009). In this way it is possible to indicate whether one discovery is more serendipitous then another, instead of excluding one and including the other.
3.2 Aspect classification
It becomes clear there is no agreed understanding of what Serendipity precisely is. This makes it difficult to distinguish aspects that represent the construct. Still, when making an overview of the literature certain recurring terms can be identified which are most likely to allow for establishing these aspects. After identifying these aspects, they have to be translated into items that can be rated (as suggested by the C-OAR-SE procedure), to make them usable for a survey.
The most common term associated with serendipity, found in practically every article published on the subject, is ‘unexpectedness’ which is safe to associate with serendipity. An “unexpected encounter” can be classified as a concrete aspect, nearly everyone will describe the aspect identically.
Next is the process at the individual level, which has been defined more diversely, from sagacity, alertness, insight, bi-socation, connection-making to ‘an intellectual leap of understanding’ (André et. al,
14 2009). Because both the personal aptitude and the process are important, this thesis will use “insightful connection-making” as the aspect to represent the process at the individual level. This aspect can be classified as an abstract formed aspect. According to the C-OAR-SE procedure, this classification requires the aspect to be broken down into concrete singular components. But for this thesis it is decided to keep this aspect abstract formed for the rater entity to rate subjectively, this in accordance with prior research on rating the extent to which an encounter is serendipitous (Makri & Blandford 2012).
Interestingly other aspects of serendipity have seen a significant shift in relevance. Earlier work stated the importance of being on a quest/involved in an activity, for serendipity to occur (Denrell et. al, 2003). Recent work neglected this aspect and instead places emphasis on the value/usefulness of a discovery for it to be deemed serendipitous (Ge et. al, 2010). This thesis chooses to follow this shift. Firstly, because ‘involvement in an activity’ is problematical to rate and classify. Secondly, because the meaning of the word ‘serendipity’ simply has evolved during the years and value has been increasingly associated with it. Therefore a “valuable encounter” is the third aspect, which can be classified as an abstract aspect that suggests somewhat different things to a group of raters.
3.3 Attribute classification
Now that the presumed aspects of the construct have been identified, the next step is to look at the attributes that shape the construct. Because of the fact that serendipity is not simple luck, the relative differences in quantity and quality of serendipitous encounters between individuals depend on their personal attributes and those of their environments. Qualitative research on serendipity has reported on numerous properties that appear to be related to making serendipitous discoveries. These properties will function as the attributes, even though only reported as being correlated, because of the absence of clear causal attributes.
15 3.3.1 Attributes related to the individual
As portrayed in the entomology of serendipity, the princes who were making all these serendipitous discoveries were doing so by paying attention to seemingly unimportant information. In other words they expressed openness to new information. Kefalidou and Sharples (2016) relate open-mindedness to experiences of serendipity. Another study looking into entrepreneurial opportunism, found that entrepreneurs who were always open for new opportunities and did not plan far in advance, were entering new businesses they had never expected (Rosa, 1998). The attribute that will be used to reflect these findings is “openness to experiences”, which has been researched extensively as part of ‘the big five personality traits’ (e.g. Goldberg, 1990).
Next to openness to experiences, another trait from ‘the big five’ is also found to positively influence serendipitous discoveries. This trait is “extraversion”, Pickering & Gray (2001) showed that extraversion and openness to experiences increased receptiveness to the unexpected. Their reasoning for extraversion being influential is that extraverts have a higher need for external incentives. Furthermore, Heinström (2007) also reported “the strongest connection between incidental information acquisition and personality seemed to be a positive link to extraversion” (p.588). McCay-Peet et. al (2015) can also see why extraversion may play a role in serendipity, because interactions between people is important throughout the whole serendipity process.
Another attribute that appears to be related to enabling serendipity and is also intuitively associated with it, is discussed by Makri & Blanford (2012). They identified a collection of emotional states, which their interviewees mentioned enabled serendipity for them. A few were closely related, which were: being relaxed, alert and in a good mood. To include these in this thesis they will be represented by the attribute “optimism”. Furthermore, optimism is also mentioned by Williams et. al (1998) as a key factor for the ability to recognize opportunities presented by a chance event and act upon it. Optimism is part of the widely used Life Orientation Test, which allows for testing it reliably.
16 With regards to the C-OAR-SE procedure, all three of these attributes can be classified as eliciting attributes (Rossiter, 2002). These are attributes that represent “an internal trait or state that has outward manifestations, which are mental or physical activities, as components”. In the case of this thesis, the components that are referred to are the items used in in the corresponding tests (The big five personality traits and the Life Orientation Test). Furthermore Rossiter also states that to use an eliciting attribute, it has to be internally consistent by a coefficient alpha of approximately 0.8. The specific test used in this thesis for openness to experiences and extraversion is “the short five (S5)” (Konstabel et. al, 2012) which shows an α score of .78 for openness to experiences and .89 for extraversion. For the optimism the revised version of the LOT is used, the LOT-R (Scheier et. al, 1994) which has an α score of .78. Therefore these attributes and the corresponding items can be used for testing the scales for this thesis.
3.3.2 Attributes related to the environment
To start with it is well worth looking at the work environment as a whole, because as André et. al (2009) mentioned: the social and physical environment are important when it comes to serendipity. George & Zhou (2001) illustrate this further by stating that a negative work environment means that individuals are not receiving the support they need. And as one might have thought, it is important for an environment to enable connection-making in a social as well as a physical context, to increase the opportunity for serendipitous discoveries to occur (Cremonini, 2016; McCay-Peet et. al, 2015). The attribute that reflects this mechanism is “work environment”, more specifically, how positively/negatively the social and physical work environment is experienced. To measure this attribute part of the Work Environment Impact Scale – Self Rating (WEIS-SR) (Wästberg et. al, 2016) will be used.
17 Another environmental aspect that is shown to be related to serendipity is workload. Csikszentmihalyi & Sawyer (1995) already suggested that insight in general occurs during idle times. And later research on serendipity mentioned even more specifically that less workload appears to be related to increased serendipity (Makri & Blandford, 2012; Kefalidou & Sharples, 2016). The corresponding attribute is then of course “workload”, for which a test is developed: the NASA Task Load Index (NASA-TLX) (Hart & Staveland, 1988).
Like the attributes related to the individual, these aspects related to the environment are both also classified as eliciting. Therefore they both also require a coefficient alpha of approximately 0.8. In the case of the work environment, the collection of items from the WEIS-SR used for rating the social and physical work environment has an α score of .79 (Dorsey et. al, 2016). For the NASA-TLX a modified version will be used, namely the Raw TLX (RTLX), which leaves the pairwise weighing of items out of the test (Hart, 2006), while maintaining a high correlation with the weighted score (Grier, 2015). The NASA-TLX has an α score of over 0.8. (Xiao et. al, 2005).
18 4 Conceptual framework
4.1 Items for scale development and testing
As discussed, several aspects have been identified that are suggested to make up the receptiveness to serendipity. Furthermore, several attributes have been mentioned that are expected to influence this receptiveness on the level of the individual as well as his or her environment. According to the C-OAR-SE procedure items will have to be generated based on these attributes. Because of the absence of literature on quantifying serendipity and no access to experts, the items will be generated the way that will be discussed in this section.
4.1.1 Items related to the aspects
The three aspects that have been defined are: (1) ‘an unexpected encounter’, (2) ‘insightful connection-making’ and (3) ‘a valuable encounter’. These aspects are the building blocks of the construct: receptiveness to serendipity. Based on these aspects items have to be generated to test an individual’s receptiveness. Important to mention again, is that receptiveness to serendipity will not be treated as nominal data, because as André et. al (2009) stated, serendipity is a continuum. Therefore, receptiveness to serendipity will be conceptualized as ‘the extent to which an individual has serendipitous encounters’. This receptiveness is then broken down into two parts: (1) ‘How often does an individual have encounters that are potentially serendipitous’ and (2) ‘To what extent are these encounters serendipitous’. This combination of both the quantity and quality are expected to determine the receptiveness to serendipitous of an individual.
The quality of potentially serendipitous encounters is in effect the combined score of the three aspects: unexpectedness, insight and value. The higher an individual scores on these, the more serendipitous his or her encounters are. This is based on a system Makri & Blandford (2012) used for judging potential serendipitous encounters their interviewees described. They asked themselves three questions related to
19 - The unexpectedness of the circumstances
- How insightful the connection-making was - The (expected) value of the encounter
The difference for this thesis will be that participants in the survey will rate themselves on these variables, instead of the researcher. This will be done using items with a 6 point Likert scale. Furthermore, participants will not be asked to rate specific encounters on these three items, but instead rate their encounters with novelty in general.
But as mentioned before, the quality alone does not completely cover the construct, because someone who has a handful of very serendipitous encounters may be regarded less receptive to serendipity than someone who has a lot of relatively serendipitous encounters. The relevance of quantity can even be traced back to the origin of the word serendipity by Horace Walpole. He related the word to the princes from Sri Lanka, who were always making discoveries, by accidents and sagacity, of things which they were not in quest of (Merton & Barber, 2006).
To incorporate the quantity, items will be created that determine how often an individual has encounters with novelty in general. These items will be ordinal using a 6 point Likert scale, instead of interval, because there is no information on average numbers of encounters within a given period.
In the end it is desirable to have one score that represents all these items combined. Therefore to assess ‘the extent to which an individual has serendipitous encounters’ the outcomes of both the quantity and quality of potential serendipitous encounters have to be weighted in such a way that it provides a meaningful score. In this case “meaningful” means that the score should reflect two things.
• The first is that quality weighs heavier than quantity, this is in line with the general misconception that luck plays an important role in serendipity (Denrell et. al, 2003). People tend to think everyone can have a stroke of genius by chance, when the right contingency comes along (Dew, 2009).
20 • Secondly the weighing method should inflate the weighted score, the higher an individual scores on the items. This reflects the notion that, even though serendipity can be considered a continuum, all three aspects should be present substantially, otherwise we only find diluted serendipity (Makri & Blandford 2012). Therefore, in other words, lower scores on the items should deflate the weighted score.
The proposed formula for weighing the score based on these points is as follows:
ܴ = ඥݍ
× ݍത
௨௩R = Receptiveness to serendipity (the extent to which an individual has serendipitous encounters) qf = Quantity of potentially serendipitous encounters (frequency of encounters with novelty)
quiv = Quality of potentially serendipitous encounters (consists of unexpectedness, insightfulness and
value)
By multiplying the score for quantity by the average score of the three components that make up for the quality, higher scores on both will inflate the weighted score relatively more. Furthermore by taking the root of the score for quantity, the score for quality weighs heavier. Therefore this formula meets the two criteria for a meaningful weighted score.
Finally, using 6 point Likert scales leads to a weighted score for R between 1 and 14.7. This score can then be standardized to a weighted score will be standardized to a scale ranging from 1-10.. Although this has no added benefit for the purpose of analysis, the standardized scale allows for an easy intuitive interpretation.
4.1.2 Items related to the individual and environmental attributes
Regarding the items representing the attributes, these are drawn from proven tests in other studies. They have already been briefly mentioned together with their corresponding attributes and can be found in the following table.
21
Attribute Test Items Scale
Individual Openness to experiences ‘Short Five’ (S5) 6 5 point Likert
Extraversion ‘Short Five’ (S5) 6 5 point Likert
Optimism LOT-R 10 4 point Likert
Environment Work environment WEIS-SR 7 6 point Likert
Workload NASA-RTLX 6 21 point Likert
Table 1. Items generated based on the attributes
the various attributes are measured using different Likert scales. For purpose of uniformity, ease of analysis and forced choice, all these tests will be adapted to 6 point Likert scales. The argumentation for doing so is as follows:
- Again ease of intuitive interpretation plays a role. Testing all items on a 6 point Likert scale offers a clear overview of the scores.
- Analysis of the data is more reliable because the different scores do not have to be transformed using for example a Chi-square test (which does not account for ordinality). In addition, using different scales for the same items does not significantly change the variance, kurtosis or skewness, (Dawes, 2007). Moreover, when comparing 4, 5, 6 and 11 point Likert scales for the same items, Leung (2011) found that only the 6 and 11 point scales follow the normal distribution.
- By using a 6 point scale respondents are forced to choose, which is in line with the aim of the survey to inflate high scores and deflate low scores. Furthermore an even-point Likert scale helps eliminating social desirability bias (Garland, 1991).
22 Understandably the tests that will be used are designed with their corresponding Likert scale for a reason and adapting it will question their validity. Still given the reasoning above it is expected that the benefits of adaptation outweigh any potential distortion of the scores.
4.2 Conceptual framework
The conceptual framework of this thesis is based on the following hypotheses:
H1: The scale for measuring ‘receptiveness to serendipity’ correlates with the scores on openness to experience, extraversion, optimism, work environment and workload.
. The first step in identify the drivers of serendipity is to find a way to measure different levels of “aptitude for serendipity” between individuals, or as it is called in this thesis: receptiveness to serendipity. Therefore a scale has been proposed based on qualitative literature, but to validate this scale it has to be tested. Various attributes at the level of the individual and his or her environment are said to influence serendipitous encounters, if scores on these attributes are found to correlate with scores on the proposed scale, it speaks in favor of the validity of the scale.
To further look into the effects of the different attributes on receptiveness to serendipity it is expected that each attribute is correlated with the DV.
H2.1: Openness to experience is positively related to receptiveness to serendipity. H2.2: Extraversion is positively related to receptiveness to serendipity.
H2.3: Optimism is positively related to receptiveness to serendipity.
H2.4: Positive experience of work environment is positively related to receptiveness to serendipity. H2.5: High workload is negatively related to receptiveness to serendipity.
23 Finally, there are variables related to demographics that can provide additional insights into the drivers of serendipity and where they are manifested, which is also why two different groups of respondents have been used.
H3: Individuals working in a business center for startups and creative small businesses are more receptive to serendipity than individuals working in other places.
H4: Entrepreneurs are more receptive to serendipity than individuals with a different type of employment.
This final hypothesis focuses more specifically on entrepreneurs building on the literature relating entrepreneurs to higher levels of serendipity (Dew, 2009; Rosa, 1998)
These hypotheses are visualized by the following conceptual framework:
Figure 1: the conceptual framework
Individual Openness to Experience Extraversion Optimism Environment Work environment Workload Receptiveness to serendipity Quantity of Quality of potentially potentially serendipitous serendipitous encounters
X
encounters - Unexpectedness - Insight - ValueWorking in ‘Hooghiemstra’ Entrepreneur Demographics
24
5 Data
As mentioned before, the survey has been sent to two groups of respondents: group 1 is without predefined demographics and group 2 consists of people working in the same business center. Both groups and sampling techniques will be further discussed in this chapter.
5.1 Group 1
This group of respondents has no prerequisites, anyone is eligible for being included in this group to take the survey. This approach is taken to gain access to a large sample with relative ease. The research question of this thesis allows for setting this broad category, because every individual can be regarded as being receptive to serendipity to a certain extent. The attributes of the construct, which will be tested in the survey, do expect the respondent to be working currently or to have worked in the past. This means that the data of respondents who have never worked will not be used in the analysis because for these respondents the survey will be incomplete Therefore, in the analysis this group only includes those who work or have worked in the past.
5.1.1 Group 1 sampling
The sampling technique for group 1 is a combination of convenience and snowball sampling, through multiple channels everyone in the network of the researcher will be asked to take the survey. These channels are e-mail, LinkedIn and Facebook, which provide access to a broad range of individuals. Furthermore, everyone is asked to spread access to the survey within their own network. To provide an incentive for filling in the survey, a private boat trip has been raffled among the respondents who provide their email address. Although it is clear that a representative sample of the population will not be achieved, having a larger sample by not setting prerequisites and using snowball sampling does provide a reliable base to test the proposed scales.
25 5.2 Group 2
The shared characteristic of this group of respondents is that all of them are working for an organization that has an office in the Hooghiemstra building. The business center Hooghiemstra is a large monumental industrial building, which has been renovated and currently houses 103 small firms ranging from animation studios to translation agencies and to legal advisors. The mission of Hooghiemstra is not just to house entrepreneurs and small businesses, but also to connect and guide them. Several of these firms have won prestigious awards (amongst which an award at the film festival of Cannes) which contributes to Hooghiemstra being regarded as a highly creative place.
This is also why this group of firms is interesting for this thesis, because one might expect to find individuals who are more receptive to serendipity in such an environment. This is not expected to make a difference for validating the scales compared to the other group as it is expected that the scales measure the extent to which someone is receptive to serendipity equally for those more or less receptive. However, it enables us to analyze if there is a difference in the receptiveness to serendipity between both groups.
5.2.1 Group 2 sampling
The sampling technique for group 2 is convenience sampling, the raters having the choice to participate or not. A factor that speaks in favor of the reliability for the sampling of group 2 is that all those employed in the Hooghiemstra building have the option to participate. Through the front office of Hooghiemstra all organizations in the building have been approached by email to answer the questionnaire. To provide an incentive for filling in the survey, again a private boat trip has been raffled among the respondents who provided their company name.
26 5.3 Sample size
Determining the required sample size is a much debated topic that has produced different formulas and rules of thumb. For this thesis correlations are tested between 1 DV and 5 IV, for which a useful formula is argued by Green (1991). He found out that for medium effect-size studies for testing multiple correlations with a small number of predictors (<7), the minimum number of respondents should be 50+8m (where m is the number of predictors).
Van Voorhis & Morgan (2007) take a different approach, stating that when using more than 5 predictors the absolute minimum of participants should be 10 per predictor. Though they add it is better to opt for 30 participants per IV.
If we take the number from Green as the minimum number of respondents and the rule of thumb from van Voorhis & Morgan as the desired amount, the aim for this thesis will be to gather 150 responses, with a minimum of 90.
27
6 Survey
The questions of the survey can be found in appendix A. The survey for both group 1 and 2 were exactly the same save for the introduction in which the respondents were addressed differently and the final question of the survey asked group 1 for an email address to enter the raffle and group 2 for their company name.
The first five questions asked for demographics of the respondents. These were all multiple-choice to increase the ease of filling in the survey. These ordinal items are cumbersome for analysis but since these demographics were not part of the IV’s this limitation is accepted.
The following twelve items made up the measurement tool for scaling receptiveness to serendipity. As explained in section 4.1.1 these are based on extant qualitative literature and for further refinement the questions were determined using Q-sorting (Nahm et. al, 2002). For this method 5 questions for each aspect (unexpectedness, insightfulness, value and quantity) were devised in agreement with a professor in labor and learning of which two questions were reversed. Afterwards these questions were presented to a group of Business Administration master students (specialization: Entrepreneurship & Innovation) in random order. They were asked to group the questions in the categories unexpectedness, insightfulness, value and quantity. The three questions that scored the best on each of the three groups were used in the survey. See appendix B for the results of the Q-sorting.
The remaining questions were all copied directly from the original tests. Though as discussed in 4.1.2 the Likert scales for these tests were adjusted to a 6-point scale for all tests.
28
7 Results
7.1 Descriptive statistics
In total 159 respondents filled in the survey of which 99 belonged to group 1 (no-prerequisites) and 60 to group 2 (working in the Hooghiemstra business center). Of these 159 respondents, 113 filled in the whole survey (71 for group 1 and 42 for group 2) and thus were used for the analysis. As mentioned before a valid sample size would ideally have been 150 with a minimum of 90, with 113 valid responses this research had an acceptable sample size to perform the analysis.
7.1.1 Descriptive statistics of the respondents
The distribution on the demographics of the 113 respondents can be found in table 2. To get absolute numbers for age and work hours those variables have been recoded to the central number of their respective categories.
Demographic descriptives of the respondents
Total Group 1 Group 2
N Mean St.dev. N Mean St.dev. N Mean St.dev.
Age 113 41.16 13.30 71 43.63 14.23 42 37.00 10.43
Gender (0=M 1=F) .46 .50 .42 .50 .52 .51
Highest level education (1-4) 3.33 .78 3.27 .86 3.43 .63
Work hours per week 35.96 9.08 34.18 10.00 38.96 6.30
Entrepreneur (0=yes 1=no) .28 .45 .15 .36 .50 .51
29 The average age of the respondents is 41.16 years which is close to the Dutch national average of 42.41.
The ratio of male to female is not too far from 50%. Regarding the highest level of education it is known that 45.7% of the Dutch population has attained tertiary education 1, which would correspond with a
mean of approximately 2.30 on this category scale. The average score of the respondents in this dataset is 3.33 which shows that the average level of education of the respondents is relatively high. Also the average number of weekly working hours of the respondents (36) is high compared to the average number of working hours in the Netherlands of 30.31. Finally, the number of self-employed individuals
in the Netherlands is 1.28 million2, which translates to 7.5% of the population. The percentage of
entrepreneurs of the respondents is 28.3%, which was to be expected because of the respondents who work in the Hooghiemstra business center.
When comparing the two groups of respondents it is clear that the first group comes closer to the average of the population in terms of highest level of education, work hours per week and percentage of entrepreneurs than the second group. As was expected, group 2 is a more homogeneous group as is shown by the smaller standard deviations. It can be described as a group of relatively young highly educated individuals who work a lot of hours and of whom half is self-employed.
7.1.2 Descriptive statistics of the Independent Variables
The 5 main IV’s are the attributes that were discussed in chapter 3.3. The items for these IV’s were taken from existing, highly robust tests. Some of these tests contained dummy questions and/or reversed questions which were removed or recoded, respectively. All items were tested on 6 point Likert scales which allows for easy intuitive interpretation of the results. Table 3 lists the descriptives of the independent variables. When looking at the descriptives two things stand out. The first being the difference in standard deviation between the individual attributes (Openness to Experience, Extraverion and Optimism) and the environmental attributes (Work Environment and Workload). The second being
1http://ec.europa.eu/eurostat/web/database 2
30 the difference in Openness to Experience and optimism between groups 1 and 2, where the other IV’s are more or less the same.
Descriptives of the Independent Variables
Total Group 1 Group 2
N Mean St.dev. N Mean St.dev. N Mean St.dev.
Openness to Experience 113 4.55 .96 71 4.41 1.05 42 4.80 .72
Extraverion 4.36 .86 4.39 .85 4.31 .88
Optimism 4.45 .81 4.54 .76 4.29 .89
Work Environment 4.91 .59 4.92 .55 4.90 .65
Workload 3.41 .54 3.40 .58 3.41 .49
Table 3: Descriptives of the Independent Variables
7.1.3 Descriptive statistics of the Dependent Variable
The dependent variable ‘receptiveness to serendipity’ consists of different aspects which are combined and weighted as discussed in section 4.1.1. This score is rescaled to fit a 1-10 scale. The descriptives of the different aspects that make up the DV can be found in appendix C. A brief overview of the means and standard deviations are shown in table 4. The table shows that, in line with hypothesis 4, entrepreneurs have a higher receptiveness to serendipity than non-enterpreneurs .
Descriptives of the Dependent Variable
Total Group 1 Group 2
N Mean St.dev. N Mean St.dev. N Mean St.dev.
Receptiveness to Serendipity 113 5.21 .97 71 5.21 .97 42 5.21 .99 (scaled 1-10)
31
Total Non-entrepreneurs Entrepreneurs
N Mean St.dev. N Mean St.dev. N Mean St.dev.
Receptiveness to Serendipity 113 5.21 .97 81 5.02 .93 32 5.68 .94 (scaled 1-10)
Table 4: Descriptives of the Dependent Variable
What stands out right away is that the means for group 1 and group 2 are exactly the same which is not in line with hypothesis 3. However, when looking at the difference between entrepreneurs and non-entrepreneurs there is a clear distinction.
Finally, although not part of the formulated hypotheses, there was one more demographic that showed a clear difference in score on the DV: male respondents score a higher receptiveness to serendipity than females (See table 5).
Descriptives of the Dependent Variable
Male Female
N Mean St.dev. N Mean St.dev.
Receptiveness to Serendipity (scaled 1-10) 61 5.40 .91 52 4.99 1.01
Table 5: Descriptives of the Dependent Variable by gender
7.2 Assumptions for multiple linear regression
As the hypotheses look at the correlation between the DV and IV’s, a multiple linear regression is the model of choice for the analysis. To perform this analysis the data should meet certain assumptions as discussed by Cohen et. al (2013) and others.
32 7.2.1 Correlations and multicollinearity
The correlations between the independent variables can be seen in table 6. A score of 0.1 indicates a small effect, 0.3 a medium and 0.5 a large effect (Cohen et. al, 2013). For the variables to be suitable for regression it is important that they are not too highly correlated. The threshold above which it is advisable not to include two correlated variables in the regression is 0.7 (Tabachnick et. al, 2001). Correlations Age Gender Educati on level Work hours Openne ss to Exp. Extrav ersion Optmis m Work Enviro nment Work Load Group 1/2 Entrep eneur Age 1 Gender -.150 1 Education level .035 .045 1 Work hours -.048 -.285** .139 1 Openness to Experience -.002 .180 .205* -.141 1 Extraversion .112 -.003 .219* -.039 .466** 1 Optmism .213* .055 .129 -.035 .119 .239* 1 Work Environment .087 .218* .162 -.029 .097 .207* .429** 1 Work Load .078 -.093 -.082 .130 .161 -.064 -.203* -.284** 1 Group 1/2 -.243** .098 .100 .263** .199* -.045 -.155 -.014 .008 1 Entrepeneur .131 -.107 .038 .178 .115 .068 .032 .107 .108 .370** 1
*. Correlation is significant at the 0.05 level (2-tailed). **. Correlation is significant at the 0.01 level (2-tailed).
33 The table shows that work hours are negatively correlated to gender, indicating that men work significantly more hours than women. Furthermore, respondents’ level of education is significantly positively correlated with their openness to experience and extraversion. It appears that older respondents are more optimistic and less represented in group 2 of which the latter already became clear in section 7.1.1. Moreover, the respondents of group 2 work significantly more hours than those in group 1, as mentioned before. Finally women rate their work environment relatively higher than men.
Correlations among the independent variables require a closer look, because strong correlations might indicate multicollinearity, which can harm the regression model. The table shows that openness to experience and extraversion correlate highly significant with each other, which is a known phenomenon (Aluja, 2002). Extraversion and optimism also correlate significantly, but the effect is not too strong; the same holds for the correlation between extraversion and work environment.
Work load correlates negatively with the work environment, which makes sense, just as work load having a negative correlation with optimism. A strong positive correlation is indicated between optimism and work environment, which can almost be classified as a large effect (.429 against .5 ), still although the effect is large, the correlation is not too strong to exclude the variable. Finally, there is a highly significant correlation between the being an entrepreneur and group 2, which already became clear from the demographics of the sample.
Another way of testing for multicollinearity is by looking at the tolerance and VIF values of the independent variables. In table 7 these values are listed. For a variable not to be collineated the tolerance should be over 0.2 and the VIF value should be under 10 for strong models and under 2.5 for weak models (Hair et. al, 1995). For this dataset there is clearly no case of multicollinearity. Collinearity Statistics Tolerance VIF Openness to Exp. .690 1.449 Extraversion .716 1.398 Optmism .758 1.318 Work Environment .751 1.332 Work Load .836 1.197 Group .774 1.292 Entrepeneur .822 1.217
34
Figure 3: Histogram residual values
7.2.2 Linear relationship
For a multiple linear regression the relationship between the DV and IV’s needs to be linear. This linearity assumption can best be tested using a scatterplot. The standardized residuals are plotted against the predicted values to show if there is a linear relationship. See figure 2 for the scatterplot of the model used for analysis. The plot shows a clear linear relationship which meets the linear relationship assumption. Furthermore, the plot shows that there are not many outliers that might distort the analysis and that the variances appear equally
for different values of X. This will be discussed more extensively further on.
7.2.3 Multivariate normality
The error between observed and predicted values should be normally distributed. This assumption can be checked by reviewing Q-Q plots of the different IV’s against the DV (See appendix D). The plots all show a near normal distribution, but to make sure that there is a normal distribution the residual values of the model can also be plotted on a histogram to review the distribution. See figure 3 for the histogram. This plot shows a clear bell-shaped distribution which is also confirmed by the Shapiro-Wilk test for normality (F: .989; p: .479).
35 7.2.4 Autocorrelation
To test the data for autocorrelation there is one method that is proven to give the best results, which is the test developed by Durbin & Watson (1951). This test is part of a multiple linear regression and gives a value between 0 and 4, where a value around 2 indicates no autocorrelation. The values for the model of regression that is used is 2.025, indicating no autocorrelation.
7.2.5 Homoskedasticity
As mentioned with the scatterplot of figure 1 there does not appear to be inequality in variance for different values of X. Furthermore, the Breausch-Pagan test for heteroskedasticity also proofs to be insignificant (LM: 3.475; p: .627). Therefore the variables meet the homoscedasticity assumption.
7.3 Multiple linear regression analysis
Section 7.2 showed that all assumptions for being able to do the regression are met and the model is estimated in SPSS through a linear regression.
7.3.1 Regressing the DV and five main IV’s
The main hypothesis of this paper seeks to correlate the DV with the five main IV’s and is as follows:
H1: The proposed scale for measuring ‘receptiveness to serendipity’ correlates with the scores on openness to experience, extraversion, optimism, work environment and workload.
The result for the model fit is an adjusted R squared of .143 (table 8) which means that the model accounts for 14.3% of the variance in receptiveness to serendipity with a significant F-test (F=3.564, p=0.005) which shows that this difference in variance is significant. Therefore, hypothesis 1 can be accepted, though it has to be noted that the variance accounted by the model is small.
36
Table 9: Multivariat estimation results of receptiveness to serendipity
Model Summary
Model R R Square Adjusted R Square
Std. Error of the
Estimate Durbin-Watson
61 5.40 .91 52 4.99
Table 8: Multiple regression model fit
7.3.2 Coefficients of the separate variables
The five sub-hypotheses that are part of hypothesis 2 refer to the beta values of each separate IV, which are shown in table 9.
The coefficients indicate that the only IV that appears to have a significant relation with the receptiveness to serendipity is Workload. Openness to experience is almost significant related to the DV. Now that the coefficients of the IVs are known hypothesis 2 can be evaluated as follows:
- Openness to experience is positively related to the receptiveness to serendipity. The correlation is strong, but only weakly significant, therefore hypothesis 2.1 is rejected.
- Extraversion is significantly related to the receptiveness to serendipity’. Therefore hypothesis 2.2 is rejected.
- Optimism is not significantly related to the receptiveness to serendipity. Therefore hypothesis 2.3 is rejected. Beta t Sig. Openness to Experience .187 1.796 .075 Extraversion .114 1.092 .277 Optmism .087 .857 .393 Work Environment .104 1.019 .311 Workload .204 2.121 .036
37 - Positive experience of work environment is not significantly related to the receptiveness to
serendipity’. Therefore hypothesis 2.4 is rejected.
- Workload is strongly and significantly correlated with ‘receptiveness to serendipity’ but the relation is positive instead of negative as hypothesized. Therefore hypothesis 2.5 is rejected
7.4 Independent samples t-test for additional grouping variables
Finally there are two hypotheses that introduce additional predictor variables, the first being:
H3: Individuals working in a business center for startups and creative small businesses are more receptive to serendipity than individuals working in other places.
For this hypothesis an independent samples t-test is performed with ‘receptiveness to serendipity’ as the test variable and ‘group’ as the grouping variable. This results in a t-value of -.015 with a significance score of .988, this means that the variable group has no significant influence on the DV and therefore the hypothesis is rejected. This conclusion was already clear because of the exact same score on the DV of both group 1 and 2 in section 7.1.3.
H4: Entrepreneurs are more receptive to serendipity than individuals with a different type of employment.
For this hypothesis again an independent samples t-test is performed with the DV as test variable and ‘entrepreneur’ as the grouping variable. This results in a t-value of -3.377 with a significance score of .001, this shows a highly significant effect of being anentrepreneur on the DV.Therefore the hypothesis is accepted. Again this already hinted at in paragraph 7.1.3 where a clear distinction in the score on the DV was visible.
38 8 Discussion and limitations
The aim of this paper is to gain a more applied understanding of serendipity, because even though the term is getting more and more attention, research remains observing and descriptive. This is of course inherent to serendipity which purports discoveries that cannot be captured under prediction and control. Still there are recurring observations and theorizations that hint at underlying regularities. Identifying relations between these properties and serendipity could provide insight into some of the factors that cause this phenomenon to happen. Establishing these drivers of serendipity would have a profound impact on how we view innovation, creativity and entrepreneurship.
8.1 Scaling serendipity
To be able to establish possible relations between variables and receptiveness to serendipity a measurement for RtS is required. Literature offers no such tool, therefore this paper proposes a test resulting in a ratio scale. Validating this measurement is done on the one hand through implementing rating criteria from qualitative research. On the other by checking for correlations with those properties the literature posits as factors related to serendipity.
In section 7.3.1 the variance in RtS explained by these factors was significant, which speaks in favor of the proposed scale. Still the variance explained is only 14,3%. Now it is known that the relevance of the R2 score is highly debated (Anderson-Sprecher, 1994; Nathans et. al, 2012) and a lot of adaptations have
been proposed. The partial least squares method (Chin, 1998), generalized r squared (Schumacker & Lomax, 2004) and many more. Still when using different methods the variance explained by the model remains between ,105 and ,198. All in all the exact number is not relevant for determining validity of the scale. What is clear is that either the five IV’s are not enough to explain variance in the DV, the scale does not measure what it should measure adequately, or a combination of both.
39 Concerning the first explanation there is more than enough reason to argue that openness to experience, extraversion, optimism, work environment and workload do not cover the whole concept of serendipity. For example Stoskopf (2005) holds a strong plea for observation and curiosity as prerequisites for serendipity. Van Andel (1994) even accumulated a whole list of what makes a “serendipitist”, he or she is open-minded, perceptive, curious, intuitive, smart, flexible, artistic, humorous and diligent. While being aware of this, including more variables was outside the scope of this research, therefore the five variables that were mentioned more frequently in the literature and could be tested reliably through robust tests were selected.
Regarding the second explanation content validity can rightfully be questioned. There is no quantitative research on serendipity that could be used for establishing the proposed scale. Items in the test for the DV were based on selection criteria for interpreting qualitative research, of which content validity is unknown. Furthermore the resources for this paper were too limited to perform a more extensive q-sorting method.
In conclusion it is safe to assume that both the (number of) variables to test the scale as well as content validity are inadequate for the scale to be deemed robust. Nevertheless it does appear that the proposed scale for the DV relates to some extent to the combination of properties that are theorized to explain it and should therefore not be disregarded. Instead additional testing and adaptations might proof this scale to be more valid than it appears.
8.2 Drivers of serendipity
The title of this paper is ‘drivers of serendipity’ which is what it set out to find. Five ordinal variables and two grouping variables were assumed to be related to differences in receptiveness of serendipity between individuals and/or their environment.
40 Of the five main IV’s only one (workload) is found to relate significantly to RtS and this relation was even conversely of what was hypothesized. Nevertheless the variable ‘workload’ still appears to be related and in such a way that higher workload coincides with a higher receptiveness to serendipity. This goes against research stating that less workload appears to be related to increased serendipity (Makri & Blandford, 2012; Kefalidou & Sharples, 2016). An explanation for this relation could be that a higher workload leads to individuals getting more prompts that function as contingencies for possible serendipitous discoveries (Dew, 2009). Furthermore, though not significant but still strongly related (p=,075) openness to experience shows a positive relation to RtS. This finding is in line with the majority of literature and is also the variable that is most intuitively associated with serendipity. Some additional testing proved openness to experience to be significantly related when an extra variable was added or extraversion was excluded from the model. The correlation between extraversion and openness to experience distorted the analysis. Therefore it can be rightfully argued that openness to experience does have a significant relation to the DV
Regarding the first grouping variable, which separates those working in the Hooghiemstra business center and others, it was expected that those working in a business center housing startups and small creative organizations are more receptive to serendipity. Clearly this was not the case as the scores on RtS for both groups was exactly the same. This outcome is of course not representative for the population because Hooghiemstra was the only business center included in the variable and other centers might very well show differences in the score. But it still warrants a closer look at the supposedly stimulating environment of these ‘office villages’.
The other grouping variable makes the distinction between entrepreneurs and others. It was expected that entrepreneurs are more receptive to serendipity than others. This assumption was confirmed and appeared highly significant (p=,001). Again this relation has been theorized by multiple scholars (a.o. Reynolds, 2005; Martello, 1994; Dew, 2009) but is now also quantitatively reflected in this paper. While
41 not part of hypothesis 1, including this variable in the multiple regression increases the adjusted R2 of
the model by ,055. Therefore it can be argued that being an entrepreneur is a powerful predictor of a relative high score on receptiveness to serendipity.
Finally as mentioned in section 7.3, while not being included in one of the hypotheses gender also showed a significant difference in RtS. Men scored substantially higher on the DV than women. In the literature there is no mention of any kind of relation between gender and serendipity. One explanation could be a difference in self-reporting bias between men and women, which is found in other fields of study (e.g. Sigmon et. al, 2005). But this paper will not take a swing at the gender debate without a sound foundation, therefore this difference will only be remarked as a possible avenue for future research.
8.3 Limitations
It is to be expected being the first to try and quantify the abstract concept that serendipity is, this paper shows a set of limitations. Some have already been mentioned like the validity of the proposed scale that remains to be proven as discussed in 8.1.
Furthermore the aim of this thesis is to identify the drivers of serendipity, but these same possible drivers are used to validate a scale that serves as the DV of these predictors. This method seems paradoxical or hinting at a self-fulfilling prophecy, but these effects are strongly diminished because the measurement tool of the scale is based on thorough qualitative research in the field of serendipity. The IV’s are only used to look at their (relative) correlation with the scale which would also have been the case if the scale was already validated. Admittedly this is not the most pristine way to do research but when breaking new ground one should not pretend to be able to avoid getting dirty.
42 Regarding the respondents the sample does not completely represent the population. As discussed in section 7.1.1 age and gender are representative but work hours, level of education and type of employment deviate substantially from the population average. Though the strength of this limitation is weakened since only ‘highest level of education’ correlates significantly on 2 of the IV’s, all others show no significant correlations on the IV’s. Of course there is also the limitation of self-observing bias which is inherent to these kind of surveys.
