“Sharing is Caring”
An experimental study on the impact of stress on knowledge sharing
Master thesis
Behavioural Economics and Game Theory
15 ECTS By Dafne Schröer 9854223 ***The University of Amsterdam Amsterdam School of Economics
Supervisor: dr. T. Buser
Second corrector: dr. J.J. van der Weele
July 2016
This document is written by Student Dafne Schröer 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.
Alchemists turned into chemists when they stopped keeping secrets.
Eric Raymond
(*1957, American computer programmer, and author)
Abstract
Knowledge sharing within an organization has become increasingly more important in the fast moving world we live in. Most organizations use shared databases to store and reuse knowledge. Experience and previous research have shown that without intervention, people will not automatically share the knowledge they have via a shared database. This makes it important to get a better understanding of all factors that have an impact on the decision to share. This paper examines an experiment to find the impact of stress on knowledge
sharing. Stress is seen as one of the most significant work-‐related health risks and has an impact on behaviour, decision making and mental functioning. Because of this I hypothesise that stress could be expected to have an impact on the decision to share knowledge as well. This is why in this paper the knowledge sharing and stress are connected. A negative
relationship between being stressed and sharing was found and some of the findings
provided weak statistical evidence in favour of the hypothesis, but future research is needed to draw any definite conclusions.
Contents
1 Introduction 1
2 Literature review 4
2.1 Previous literature on knowledge sharing 4
2.1.1 The impact of knowledge sharing 4
2.1.2 Possible issues with knowledge sharing 4
2.1.3 Possible solutions for knowledge sharing issues 6
2.2 Previous literature on stress 4
2.2.1 Stress and some of it’s effects 8
2.2.2 Stress and decision making 9
2.2.3 Stress and human interaction 10
2.3 Connecting knowledge sharing and stress 12
3 Hypothesis 13
4 Method 13
4.1 Treatment group 14
4.1.1 Execution of the TSST-‐G 15
4.1.1.1 TSST-‐G phase 1 15
4.1.1.2 TSST-‐G phase 2 16
4.1.2 Knowledge sharing part of the experiment 17
4.1.2.1 Knowledge sharing phase 1 17
4.1.2.2 Knowledge sharing phase 2 18
4.2 Control group 19
4 Results 20
6 Discussion 26
6.1 Discussion of the results 26
6.2 Limitations of the study 26
6.3 Suggestions for future research 29
7 Summary and concluding remarks 31
Appendix 33
1. Introduction
Humans have always shared their knowledge and knowhow by telling each other stories about their thoughts and experiences (Smith, 2001). Despite this innate capacity for sharing and storytelling, capital, labour and raw materials were historically seen as much more valuable than creating and applying knowledge (Smith, 2001). Indeed, within the field of commerce, organizations used to focus mainly on processing information to solve problems (Nonaka, 1994). But this is changing, and today it is widely recognised as more important to manage human intellect and convert it into new and innovative products and services (Smith, 2001).
It has become a main focus for organizations to prevent valuable human knowledge
resources from being wasted. In particular, knowledge in the form of knowhow is often lost by outsourcing, mergers and downsizing (Smith, 2001). Suppiah and Manjit (2011)
estimated that 90% of all knowledge within an organization resides in the minds of its staff. This makes it all the more important for organizations to find ways to store, sort and use the knowledge they have at their disposal (Smith, 2001).
Organizations often use electronic databases to share and (re)use corporate knowledge (Cress and Hesse, 2004). To be able to create knowledge, it has to be understood that this process begins its life as data. Data are initially transformed into information and only then can this information be transformed to knowledge. Information alone has little value until it has been given meaning or is being used on the job (Smith, 2001). This shows the
importance of well functioning knowledge sharing systems.
Economics is interested in human decision making. The question “what decisions do people make?” is important, but equally important is “why do people make the decisions they make?” What drives people to do what they do? And how can we influence that? In the case of knowledge sharing one could ask why would people share knowledge, and why do some people share their knowledge with others, while others do not.
In the context of knowledge sharing within institutions, for knowledge to be stored in databases, employees often need to make an active decision to contribute their knowledge. The decision to share has been studied in depth and the problems that arise with sharing via databases have been described in multiple studies. Understanding how to address these problems to create well-‐functioning knowledge management systems is also a focus of existing literature.
At the outset, human inertia is thought to be one of the biggest obstacles to the creation of well-‐functioning knowledge management systems (Smith, 2001). This is exacerbated by the “free rider” problem that is inherent in common pool resources such as knowledge
databases. A solution to the free-‐rider problem that ensures most knowledge will be shared has yet to be found, however prior research on knowledge sharing and possible
interventions and their results are discussed in the first part of section 2 of this paper.
One possible reason for the lack of knowledge being shared in organizations that has not yet been studied is stress. Stress has been found to change behaviour and cognition by
changing neuronal activity (Joëls and Barams 2009; Sandy and Haller 2015; Dawans et al. 2012). It creates social withdrawal (Sandy and Haller 2015), impaired physical and mental functioning (Kalia 2002) and the majority of the brain regions affected by stress response are also involved in the decision making process.
Apart from knowledge sharing problems, organizations also have to deal with stress related issues on a daily basis. In 2001 the World Health Organization noted that stress is one of the most significant health risks in the twenty-‐first century (Kudielka et al. 2007). In 2014 the European Agency for Safety and Health at Work described a project carried out by Matrix (2012) on the estimated cost of work-‐related stress and mental health issues in the
European Union in 2012. Matrix (2012) found that the cost of work-‐related depression was estimated at €620 billion. Although the authors are not clear on what part of the work-‐ related depression is actually caused by stress, the findings make it clear that work stress is not just a health problem, but also an economic problem. Further, TNO (2014) found in their study on labour market stress that one in eight workers in the Netherlands is suffering from work-‐related stress. In the United States these numbers are even higher, with 40% of
workers reporting their job as very or extremely stressful and 25% of workers viewing their job as the number one stressor in their lives (Sauter et al. 2014). These complementary findings indicate that most organizations will have to deal with stress in one way or another.
Taken together, stress and knowledge sharing are two extremely important issues for most organizations, and both have been researched extensively. Given all these findings, one could expect stress to have an impact on knowledge sharing. However, until now, this impact has not been studied. This paper seeks to fill the gap in the existing literature by examining the impact of stress on knowledge sharing. Could (part of) the lack of knowledge sharing possibly be explained by stress? Based on the theoretical view of stress and
knowledge sharing described in section 2, this paper describes an experimental setting that investigates people’s sharing behaviour when influenced by stress (and benchmarked against a non-‐stressed treatment) as discussed in section 3.
The experimental design used to study the impact of stress on knowledge sharing follows the well proven Trier Social Stress Test for Groups (Dawans et al. 2010), followed by a knowledge sharing task. The complete design of the experiment used is discussed in more detail in section 4. Results are presented in section 5 and discussed in section 6. Although the data from this experiment came from a very small and noisy sample (meaning that in most cases, the results were either not statistically significant, or provided only weak evidence in favour of the tested hypothesis), this paper does find a negative relationship between stress and sharing.
Section 6 will finish with a discussion on some limitations of this study that should be taken into account when interpreting the results, followed by some implications for future
research. This paper will end with a summary and some concluding remarks in section 7.
2. Literature review
2.1 Previous literature on knowledge sharing
2.1.1 The importance of knowledge sharing
Cress and Kimmerle (2009) describe a vision of a world that allows people to share and combine knowledge all over the world, regardless of the place and time. This vision was spurred by the emergence of Web 2.0 websites that emphasize user-‐generated content, interoperability and usability, and social software (DiNucci, 1999). The new breed of software makes it possible for people to start conversations with each other and create a comprehensive pool of knowledge which is kept up to date by users from all over the world. The fact that everyone has access to this knowledge and can change it means that the quality of the common knowledge pool will be very high. The first to make this vision real is the online encyclopaedia Wikipedia. Because millions of people use this encyclopaedia and can change and add knowledge as they wish (as long as it is objective and supported by references) (Cress and Kimmerle, 2009), Wikipedia has successfully become a high quality source of information, as shown by Giles‘ (2005) research which compares Wikipedia and Encyclopaedia Britannica.
Corporate knowledge is one of the most critical assets in modern organizations. With the rise of internet-‐based disruptive technologies, it has become increasingly difficult for organizations to sustain their competitive advantage even with more advanced technology, innovative products, and better services. Perhaps the only thing that is sustainable is a knowledge advantage, because it is usually difficult to imitate and is embedded in multiple aspects of an organization, including its history (Cress and Kimmerle, 2009; Alavi and Leitner 2001; Armistead and Meakins, 2002), culture and identity, routines, policies, systems and documents, as well as individual employees. This is why organizations are constantly trying to find ways to maintain their own corporate knowledge (Wang, 2005).
2.1.2 Possible issues with knowledge sharing
Knowledge sharing is about maintaining the knowledge flows in an organization while knowledge pools (such as databases) promote preservation, sharing and reuse of
knowledge. Knowledge pools allow many people to search for knowledge without having to contact the person who originally developed the knowledge (Wang and Ahmed, 2003). This gives more people access to information. It also helps to create a better use of employees’ steadily growing wealth of knowledge to solve problems and to achieve goals (Smith, 2001). Although very important, technology is only a facilitator in knowledge management. The human aspect is still vitally important; and human behaviours, like sharing, do not always occur naturally, however they are of huge importance for knowledge management systems to work properly (Armistead and Meakins, 2002).
For knowledge pools to function as an effective way to store organizational knowledge, it is not only important that stored knowledge is used by others, but maybe even more essential that people share the knowledge they have (Wang and Ahmed, 2003). The problem – that people will very gladly use the information from a pool of jointly compiled knowledge, but in many instances those same people will not help to grow the pool by adding knowledge of their own – is seen in many situations where shared knowledge pools are created. This is also the case for Wikipedia – millions of people use Wikipedia, but only a small percentage actually contributes to the knowledge pool (Cress and Kimmerle, 2009).
This same problem arises in organizations where people need to share knowledge through databases, such as file servers or document management systems. When people have to make a decision to share knowledge with others by entering it in a shared database, a social dilemma arises. A social dilemma is a situation in which the decision that gives the best result for the individual is not necessary also the best decision for the group as a whole. One reason for not sharing information could be that knowledge is often seen as a kind of power that people do not want to give away (Cress and Hesse, 2004). This gives rise to a conflict between the interests of individual group members and the interests of the group as a whole (Cress and Kimmerle, 2009). The knowledge sharer has private costs associated with sharing knowledge (she must expend time and effort to put knowledge into the database) but receives no private benefit from this effort. She already has access to her own
knowledge, whether or not she contributes any of the knowledge to the shared database (Cress and Kimmerle, 2009). Hence, while everyone else can benefit if the knowledge sharer puts her knowledge into the database, the knowledge sharer herself receives no benefit at
all, and indeed bears a personal cost. Without intervention, this situation also means that the knowledge sharer has no reason to expect any reciprocity from the people that use her information (Cress and Hesse, 2004). Indeed, the knowledge sharer only has a slight hope to benefit from the knowledge contributed by others (Fulk et al. 1996). The cost-‐benefit
analysis therefore means that the most efficient strategy for every individual, independent of what everyone else is doing, is to withhold information. But the group as a whole will have a lower benefit if members withhold information from the shared database. This makes the most efficient strategy of withholding information an Inferior-‐Nash Equilibrium (Cress and Hesse, 2004).
The explanation above shows that we can look at the exchange of knowledge via a database as a public goods dilemma. Characteristics of a public good are non-‐rivalry and non-‐
excludability. In the case of knowledge sharing via a database, the database can be seen as the public good. The value of the database will not diminish if people use it, so it is non-‐rival, and it does not matter if a group member contributed to the database or not, everyone in the group can use it, making it non-‐excludable (Cress and Hesse, 2004; Cress and Kimmerle, 2009).
2.1.3 Possible solutions for knowledge sharing issues
The free-‐riding problem described here – that people will use the available information shared by others without contributing themselves – raises the question of how to motivate individuals to behave in the group’s interest and share their knowledge. This means that, given the described dilemma, people need to behave contrary to the individually rational option (Cress and Hesse, 2004).
There are some interesting aspects to this dilemma. In 1994 Constant et al. found that people react differently depending on the knowledge to be shared. They suggested that to be able to better understand the process of knowledge sharing in an organization, the attitude and perceptions of employees need to be measured. Cress and Kimmerle (2009) described this phenomenon from a psychological perspective, including a focus on the motivational situation of a user of knowledge management systems. They show how this motivational situation leads to free-‐riding and explain why the readiness to cooperate will
always be a fundamental part of a successful knowledge process.
Many researchers note that expected rewards are one of the most important factors for people to share their knowledge with others (Wang, 2005). Cress and Hesse (2004) did a number of experiments to test the impact of different motivational situations on the amount and quality of knowledge shared. They found that providing metaknowledge about relative importance, implementing a use-‐related bonus (a bonus received if the shared information is used by others) and reducing the cost of adding information to the database all resulted in a rise of the quality of information shared. Cress and Hesse (2004) also tested prescriptive rules (that is, rules that prescribed subjects should share a given number of contributions via the shared database) and provided subjects with feedback about the mean contributions of their teammates. Although there were some positive effects, none of these motivational situations was able to stop the free-‐riding problem completely. 1
It is not only motivation that stimulates people to contribute their knowledge – culture can also play a vital role. A good knowledge sharing culture in an organization has been
prescribed as one of the most important issues in a well functioning knowledge
management system (Wang, 2005; Chow et al. 2000). Tong et al. (2015) use a field study in the ICT sector to show that organizational culture is positively related to knowledge sharing and impacts job satisfaction. Further, Cropanzano et al. (2003) executed two field studies in which they found that emotional exhaustion has negative consequences for both individual employees and their employers. Emotional exhaustion negatively affected job performance, organizational commitment and turnover intentions.
Further, according to Wang (2005), individual differences are also an important explanatory factor for knowledge sharing. To be willing to share one’s knowledge with others, one must be willing to interact with others. If and how people share knowledge could even go beyond individual differences as described by Wang (2005). Chow et al. (2000) looked into the
1 As an aside, it is worth pointing out from an experimental economics perspective that Cress and Hesse (2004)
use a form of deception in this experiment – telling subjects that they would be sharing via a database with their team members, but these team members were in fact made up. This may have some impact on the validity of the findings.
importance of cultural differences in knowledge sharing. They argue that because of the globalization of economic activity it is getting more important to take cultural differences into account, because people from different countries and cultures often differ in the way they react to work related conditions (Triandis et al. 1988; Chow et al. 2000). Members of different cultural groups are often distinguished by the relative emphasis they place on their self-‐interest. Hence we can distinguish cultural groups from one another using two basic values: individualism and collectivism. When there is a conflict of needs, people from collectivist cultures are more inclined to give up their needs to benefit the group. These cultures tend to cooperate more. In contrast, people from individualistic cultures tend to be more competitive and place their own needs above those of the group. In a situation of knowledge sharing a difference could evolve between people from a collective culture and those from an individualistic culture, especially when there is a conflict between collective interest and self interest in the knowledge sharing situation. Given all these findings the study of Chow et al. (2000) shows that the extent and motivation in knowledge sharing may differ across countries and should be taken into account in the search for possible solutions.
2.2 Previous literature on Stress
2.2.1 Stress and some of it’s effects
In the past several decades there has been a great deal of interest in the sociological study of stress (Perlin and Bierman 2013). In this paper, stress will be defined as the activation of the neurophysiological stress response (Sandy and Haller 2015). Stress changes behaviour and cognition by changing neuronal activity. This happens in situations that are threatening, and it is crucial for survival that this happens rapidly and enduringly (Joëls and Baram 2009; Sandy and Haller 2015; Dawans et al. 2012).
Sandy and Haller (2015) discuss a range of research on coping behaviours in response to stress. Social withdrawal is one of the coping behaviours found in response to stress. Stress and related stress disorders have a big impact on the way people function (Kalia et al. 2002). Apart from social withdrawal, studies have also shown an increase in irritability and displays of anger as other patterns of short-‐term social stress responses (Repetti and Wang 2016). Kalia et al. (2002) discuss the fact that it has consistently been shown that the negative
impacts caused by stress are as severe as those caused by other medical conditions such as arthritis, hypertension and diabetes.
Employee stress has been shown to have significant consequences for organizations. Stressed people experience impaired physical and mental functioning and more work days are lost. Some of the consequences of stress found are absenteeism, accidents and human errors, with the increase in human error in particular having a further impact on the occurrence of accidents and on the quality of production. But there are also consequences of worker stress found in the interpersonal relationships within an organization. These include disputes between workers, conflicts and interpersonal problems; but also violence, the resistance to change and loss of intellectual capital (Kalia, 2002). It is becoming clear that stress is a phenomenon that negatively affects a growing number of people in the workplace (Brun and Lamarch, 2006).
2.2.2 Stress and decision making
Decision making and stress are intricately connected. Many (economic) decisions have to be made under stress. When a stress reaction occurs, it originates in the hypothalamus which offsets a physiological and endocrine response. This stress response affects different brain regions including the orbitofrontal cortex, the amygdala, basal ganglia, limbic, hippocampus and the prefrontal cortex. These regions are very sensitive to stress hormones, due to the fact that they have many stress hormone receptors (Starke and Brand, 2012).
Complex neural networks are also involved in decision making, with a majority of the brain regions affected by stress response also involved in the decision making process. For example, decisions that involve moral dilemmas, reward processing, uncertainty and adjustments from automatic response are all decisions that take place in one or more of these brain regions with many stress hormone receptors. These findings show that the brain regions that are associated with decision making are sensitive to stress induced changes (Starke and Brand, 2012).
Different studies on decision making under acute laboratory stress have shown that the underlying mechanisms of decision making may be altered by stress (Bos et al. 2009; Leder
et al. 2012; Gathman et al. 2014). Particularly in people with substantial cortisol responses, the brain regions involved in decision making also show changed neural activation (Starke and Brand, 2012). This shows that stress has cognitive consequences (Tomova et al. 2014). Some of the underlying mechanisms affected are the strategy used and the processing of feedback, because the working memory is sensitive to stress induced changes (Starke and Brand 2012; Leder et al. 2012). In addition, the adjustment from automated response, the sensitivity to reward and punishment, the reduction in the action outcome contingencies and the explicit knowledge of participants in these studies are underlying mechanisms altered by stress (Starke and Brand, 2012).
Delaney et al. (2014) look at the impact of stress on economic decision making. They find that subjects that are exposed to stress have increased subjective discounting rates, exhibit more present-‐focused preferences and are less likely to explore available options. Starke and Brand (2012) also mention the way stress affects fine-‐tuned decision making. Stress likely triggers cardiovascular, neural and hormonal reactions and this all can lead to a performance shift in decision making.
2.2.3 Stress and human interaction
More recently, there has been continuing research on the impact of stress on the ability of people to tune into others (Tomova et al. 2014), which may have significant consequences for social interactions in a workplace environment. This recent research has led to some interesting findings, especially with regard to the differences between men and women in social stress responses. The generally regarded social stress response for people is the “fight-‐or-‐flight” response (Dawans et al. 2012). Fight-‐or-‐flight responses are less recourse demanding and more automatic strategies. This means that people in a fight-‐or-‐flight response tend to fall back to more egocentric processes when having to judge emotions or perspectives of others, since paying attention to feelings, needs and intentions of others is resource demanding (Tomova et al. 2014).
Research has shown that the fight-‐or-‐flight response is mainly found in men (Dawans et al. 2012), with research by Taylor et al. (2000; 2006) finding a different strategy in women. The researchers found that women do not become more egocentric under stress, but become
more social. They use the metaphor “tend-‐and-‐be-‐friend” to describe this female stress response. The evolutionary explanation Taylor et al. (2000; 2006) provide is that women tend to create and maintain their social network to protect themselves and their offspring from threats. However, it should be noted that at the time of the study there was still little known about the immediate prosocial and antisocial responses following stress, particularly in men (Dawans et al. 2012), and the research of Taylor et al. (2000; 2006) was mainly based on research done on rodents (for the endocrine response to stress) and primates (for the behavioural response to stress). It could be questioned if the same responses would apply to human females. Tomova et al. (2014) did not find any differences in psychological and physiological stress responses between men and women. This means that the differences in self-‐other distinctions between men and women may not be due to gender differences (Tomova et al. 2014).
Research by Dawans et al. (2012) aimed to look at the effect of a psychosocial laboratory stressor on both social and anti-‐social behaviour. In this study, the researchers focussed specifically on the stress responses of males, because the idea is that the tend-‐and-‐be-‐friend response specifically characterizes women’s stress response. Dawans et al. (2012) found that stress exposure increased trust, trustworthiness and even sharing behaviour in social interactions in men. It should be noted that because this research only used male
participants, one cannot compare the male tend-‐and-‐be-‐friend response to the female tend-‐and-‐be-‐friend response or draw any conclusions on male and female behaviour in a mixed sex group. To do that, new research is necessary that compares both. Dawans et al. (2012) also found that non-‐social risk taking was not impacted, but stress did impact the willingness to accept risks arising through social interactions. This finding indicates that the pro-‐social behaviour following stress is not due to a general increase in the readiness to bear risks. These findings show that the social approach behaviour found for females in response to stress by Taylor et al. (2000; 2006) also holds for males. Overall, this shows that humans seem to have the tendency to provide and receive protection within groups in threatening times (Dawans et al. 2012).
2.3 Connecting knowledge sharing and stress
One question left to ask is: what causes the stress to occur? As with knowledge sharing, research has shown that organizational structure and culture are of great importance in stress and burnout problems. For example, Finney et al. (2013) find that organizational structure and culture are the most important predictors of stress and burnout in
correctional officers. Further, in terms of stress and burnout in healthcare workers, multiple studies have found that the organizational structure and culture have a large impact
(Fiabane et al. 2013; Tennant, 2001; Cooper and Cartwrite, 1994).
Bringing these seemingly disparate areas of research together, we know that knowledge sharing is very important in organizations, and that stress is seen as one of the most significant work-‐related health risks in organizations. Organizations deal with knowledge sharing issues and stress on a daily basis. If we look at the research done on both, there seems to be a number of similar creators and consequences in both stress and knowledge sharing problems. Organizational structure has a significant impact on both stress and the willingness to share knowledge. Interpersonal relationships are also an important influence on both, and since stress has an impact on decision making and creates more present-‐ focused preferences, the decision to share one’s knowledge with others might be impacted by stress as well.
Given all the results described above, there is evidence to suggest stress will have a negative impact on the willingness of people to share their knowledge with others. While the tend-‐ and-‐be-‐friend response described by Taylor et al. (2000; 2006) indicates a potential positive impact of stress, most of the evidence came from studies of rodents and primates, not from human subjects. Further, the study of Dawans et al. (2012), which did find sharing behaviour in response to stress, used experiments in which human participants interacted face to face with others. Especially in a situation where participants share information in an anonymous way via a database (as is the case in the experiment studied in this paper) the participants should not expect any reciprocity. This means there would be no reason to “tend-‐and-‐be-‐ friend” others. This is why in this study a negative impact of stress is expected.
As far as I am aware, there has not yet been any research on the impact of stress on knowledge sharing via databases. Therefore, this study sets out to investigate the effect of psychosocial stress on the willingness to share in a laboratory experiment.
3. Hypothesis
As discussed in section 2.2 above, stress changes people’s behaviour and cognition (Joëls and Baram, 2009; Sandy and Haller, 2015; Dawans et al. 2012) and can create a response of social withdrawal (Sandy and Haller, 2015). In a situation where one is asked to share knowledge with others via a shared database, social withdrawal may make subjects less empathetic and therefore potentially less willing to share.
Further, the impact of stress on interpersonal relationships (Kalia, 2002) will plausibly influence subjects’ willingness to share knowledge with others. When a person has negative feelings towards people she has to share knowledge with, she might be less willing to put in the time and effort required to share. To share knowledge via a database, subjects must make the decision to put in the time and effort to share their knowledge. When people are stressed, their fine-‐tuned decision making is affected. Therefore, with induced stress, it is expected that subjects will become less sensitive to feedback, and reward and punishment measures taken to try to stimulate people to share their knowledge (Starke and Brand, 2012). In addition, the resistance to change as an effect of stress might have a negative impact on subjects’ willingness to share (Kalia, 2002).
The arguments described above are the reason why the hypothesis of this paper is that stressed people will share less of their knowledge with others via a shared database.
4. Method
The goal of this research is to look at the impact of stress on knowledge sharing. The hypothesis was studied in a laboratory study, which took place at the CREED laboratory at the University of Amsterdam. Both the control groups and the treatment groups were
tested in the same evening. There were 24 participants of which 18 were master students in Economics at the University of Amsterdam and six participants were from outside the university. Sixteen participants were male and eight participants were female. The Ethics Committee Economics and Business (EBEC) at the University of Amsterdam approved the experimental design.
Participants were randomly assigned to either the control group or the treatment group. A control group was included to ensure the observed effects in the treatment group were a direct result of the induced stress. There were two control groups and two treatment groups, each comprising six participants. The experiment consisted of two main parts: the stress test part and the knowledge sharing part. The control group only participated in the knowledge sharing part of the experiment.
4.1 Treatment group
To induce stress, the treatment group started with a stress test. The method used to induce stress was the Trier Social Stress Test for Groups (TSST-‐G) developed by Dawans et al. in 2010. The TSST-‐G is a standardized motivated performance task protocol that combines high levels of socio-‐evaluative threat and uncontrollability in a group format. In their research, Dawans et al. (2010) use heart rate devices to test autonomic stress responses. Apart from the autonomic stress response, they also measure the endocrine stress responses, because studies have found the measurement of cortisol as an indicator for adrenocortical activity to be of high predictive value for psychosocial stress (Foley and Kirschbaum, 2010). They collect saliva samples from participants before, during and after the stress test to measure cortisol levels. Apart from the biological parameters, the authors also measure different psychological parameters such as discomfort and anxiety using questionnaires such as the trait anxiety scale of the State-‐Trait Anxiety Inventory (STAI) (Spielberger et al. 1970). The research of Dawans et al. (2010) and others has shown reliable psychological and biological (cortisol, heart rate) stress responses (further examples in Dawans et al. 2012; Pabst et al. 2013; Bos et al. 2009; Gathman et al. 2014). Primarily due to financial and logistical constraints, the TSST-‐G designed by Dawans et al. 2010 was used in this experiment, without actually measuring biological and psychological stress responses.
4.1.1 Execution of the TSST-‐G
The TSST-‐G consists of two phases: a preparation phase and an interview phase (see figure 1). At the start of the experiment, participants entered the first lab and were randomly allocated a seat number. Each participant had a pen, some paper and the instructions on their desk. The instructions were for both phases of the experiment. This was done to keep the time between the stress test and the knowledge sharing test as short as possible. After reading the instructions, participants answered some control questions to verify they understood the task.
4.1.1.1. TSST-‐G phase 1
After the control questions, the preparation phase started. Participants had 10 minutes to prepare a 2-‐minute speech about themselves and their suitability for a fictitious sales job. After 10 minutes, the experimenter asked participants to take their second seat number and quietly form a line at the door. Participants were not permitted to bring notes into the next phase of the experiment.
4.1.1.2 TSST-‐G phase 2
The second phase of the stress test consisted of two parts. For the first part of the second phase, participants entered a second lab and quietly stood in front of two interviewers, a man and a woman dressed in formal business clothes. Participants stood next to each other, but were separated by mobile dividing walls that restricted any eye contact and social interaction with the other participants (see figure 2). The interviewers gave a brief verbal summary of the forthcoming task and explained to the participants that they would call them one by one in random order to do their two-‐minute speech. The speeches were also recorded on a camera that stood very visible on the interviewers’ desk. This was done to further increase subjects’ stress.
The interviewers withheld any form of verbal or non verbal feedback while participants were speaking. If a participant was quiet for three or more seconds, the interviewers were instructed to respond in a standardized way, saying: “Go on, there’s still some time left” after which they would look at the participant in silence. If a participant finished their speech before the two minutes were over, the interviewers were instructed to also respond in the standardized way saying: “Go on, there’s still some time left” after which they would look at the participant in silence.
After all participants had given their speech (12 minutes in total), the second part of the second phase of the stress test started. In this phase, the interviewers asked the
participants to serially subtract the number 13 from a given number (e.g. 1632, 1619, 1606, etc.) as quickly and accurately as possible. All participants received an individual starting number to avoid learning effects. If a participant made a mistake the interviewers were instructed to interrupt them saying: “Sorry, that’s wrong. Start over please”. The participant had to go back to his or her individual starting number and start again. Each participant was called upon multiple times, resulting in an average of 80 seconds of calculating for each participant (a total of 8 minutes). After the calculations finished, the interviewers asked the participants to quietly form a line in front of the door to the third lab, where the second part of the experiment was held.
4.1.2 Knowledge sharing part of the experiment
The second part of the experiment was the knowledge sharing part (see figure 1). The design of the knowledge sharing part is a modification of the design used by Cress and Hesse (2004). It is set up as a public goods dilemma in which the database is the public good everyone can contribute to. This second part of the experiment also consisted of two
phases. A brief summary of the instructions read at the beginning of the first part of the experiment were read aloud.
4.1.2.1 Knowledge sharing phase 1
In phase 1, participants were asked to calculate monthly salaries of sales personnel. Each participant was responsible for two months of the year, meaning that in each group of six participants, a full year of information was available. Participants had nine minutes to
calculate as many salaries as possible. After each calculation participants had to make a choice regarding whether they would share their calculation by putting it in a shared database. To contribute a salary calculation to the database would take approximately 15 seconds of time per calculation. After filling in all fields of the database form, the participant could move on to the next calculation and so on. Participants could also choose not to share their calculation, in which case they could move straight on to the next calculation. The decision to share or not had to be made after each calculation. It is important to note that it was made clear in the instructions that there was no obligation to contribute to the
database, and that the decision to contribute calculations to the database had no direct personal benefit, because the participant would have access to her own phase 1 calculations in phase 2. Putting information in the database took time which could not be spent on calculating more salaries -‐ meaning there was a personal cost -‐ but it would help the other participants in the group, because those participants would then not have to calculate that monthly salary in phase 2. Participants’ potential payoff in both phases would only depend on how many calculations they got right, not on how many the other members of their group got right. Their earnings were 10 cents for every monthly salary calculated correctly in phase 1 and in phase 2 they could earn 30 cents for every salary calculated correctly.
4.1.2.2 Knowledge sharing phase 2
At the beginning of phase 2, all participants received a database from the experimenter which showed all shared phase 1 calculations. In phase 2, participants had 12 minutes to calculate as many annual salaries as possible. To do so, they could use their own calculations from the previous phase and the calculations contributed to the shared database by their group members. If not everything was contributed to the shared database, participants had to calculate the missing monthly salaries themselves.
It is clear to see the dilemma that arises through the amount of money people can earn during the experiment: each participant maximises their own potential earnings by withholding information from the shared database. Thus, withholding information is a dominant strategy. However, the mean payoff for all group members is higher if all decide to contribute. This means that withholding information is a Pareto-‐inferior Nash
At the end of the last phase of the experiment, a random draw picked one group member to be paid his or her earnings privately in cash. This meant that all participants had a one in six chance of receiving their actual payoff. The choice to pay just one member of each group was due to personal financial constraints.
4.2 The control group
The control group did not participate in the stress test part of the experiment. They entered the third lab and were randomly assigned to a desk. The experimenter read the instructions for the knowledge sharing task and after the instructions participants answered some control questions to show they understood the task. After the instructions, the control group performed the same knowledge sharing task as the treatment group, described above.
Ideally, the experiment would control for possible fatigue effects. The participants in the treatment group engaged in about 45 minutes of work before starting the knowledge sharing task. This could have made them more tired and thus might have affected their performance on the knowledge sharing task. Other experiments using a stress test
performed a similar test on the control group without adding the social stress component to control for this possible fatigue effect. In this experiment there was limited lab time and the entire experiment had to be executed in one evening. Because of this limited time, the control group participated in the knowledge sharing part of the experiment only.