MSc Business Administration
Digital Business Track
Suzanne ten Oever
Student number: 10563164
17
thof August 2018
Supervisor:
Dr. Alfred Zerres
Master’s Thesis
By Suzanne ten Oever
Beyond Self-Measurement of Social
Transmission:
1
Statement of Originality
This document is written by Suzanne ten Oever 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 content.
2
1. Abstract
Social transmission has become an integrated part in our lives, more and more people share content and information. For marketers it is crucial to both understand the conscious as well as the unconscious decision process to social transmission. Research on social transmission has merely focused on the conscious window, using classical self-report measurements to unravel the reasons why people share content. Neuroimaging studies provide insights into the unconscious mind and could reveal the true reasons why people share. The goal of this research is to compare classical measurements of social transmission with neuroimaging measurements in order to find out whether they measure equal concepts. This research therefore conducts an experiment comparing measures using different measurement techniques. In total 67 respondants participated in this study. There were no significant results found for the correlation between measurements. Hence, the EEG concepts could only predict a small amount of the variance on Social Transmission. Therefore this research needs to be replicated, to be able to derive significant results. This study contributed to the debate regarding how concepts, that are usually measured using self-reported measurements, can be measured using EEG. EEG model measurments are to be developed further to establish equal results as self-reported measurement models.
3
Table of Contents
1. Abstract 2
2. Introduction 4
3. Theoretical framework 6
3.1 The brain: a short introduction to its functions and structures 6
3.2 How is (persuasive) content processed and which brain areas are involved? 8
3.3 How do we make decisions? 9
3.4 Social Transmission; sharing online content 11
3.5 Measuring online social transmission 12
3.6 Emotional response 15
3.6.1 Emotional valence 15
3.6.2 Emotional arousal 16
3.7 Preference 17
3.8 Engagement 18
3.9 Model of Social Transmission 18
4. Methodology 19
4.1 Research Design 19
4.2 Pre-test & Stimuli 20
4.3 Procedure 21
4.4 Sample and data collection 24
4.5 Measurements 25 4.5.1 Social transmission 25 4.5.2 Emotional Arousal 25 4.5.3 Emotional Valence 26 4.5.4 Preferences 26 4.5.5 Engagement 27 4.5.6 Demographic variables 28
4.6 Data Preparation of EEG datasets 28
4.7 Analysis of the Event Related Potential 30
5. Results 31
5.1 Correlational analysis 32
5.2 Regression 33
5.2.1 Regression of EEG measurements 34
5.2.2 Regression of classical measurements 34
6. Conclusion and discussion 37
7. Limitations 38
8. Future research 39
9. References 41
10. Appendices 46
Appendix A – Stimuli 46
Appendix B – Protocol loaded into NIC2 48
Appendix C – Matlab Script 49
Appendix D – Presentation Script 50
Appendix E – Qualtrics experiment – self-reported measurements 52
Appendix F – Example EEG data: good vs. bad 60
4
2. Introduction
Social Transmission has become an integral part of our lives. People share online content more often. In 2007 already more than 59% of the people reported that they frequently shared content they observed online (Allsop, Basset & Hoskins, 2007), where, in 2017 this has risen to 79% (Statistica, 2017). Social media, like Facebook, Twitter and Instagram increased the speed and the ease of transmission. Though exposed to more than 3000 persuasive messages each day (Gass & Seiter, 2015), the decision to either like, share or comment is made within a split second (Rangel et al., 2008). So, how is this decision to share made and when is social transmission evoked?
Social transmission can be defined as the transmission of information from a communicator to a receiver, with the implicit or explicit goal of persuasion (Falk & Scholz, 2017). There have already been several research studies conducted handling the antecedents, processes or causes of social transmission. The decision to share content is based upon the subjective value that is derived from that decision (Falk & Scholz, 2017). The perceived costs and benefits are weighted to derive the value of choices. Information or content is shared when that information benefits the sharer or when it causes the sharer to get more intense connections within their social network (Baek, Scholz, O’Donnell & Falk, 2017). The emotional response evoked has a great influence on the computed subjective value. The emotional value given to the information or content either stimulates or de-stimulates people to share information (Health & Health, 2007; Berger, 2013).
Past research conducted on social transmission has predominantly focused on the conscious window, while the decision to share mostly takes place in the sub-consciousness, thus missing crucial information about the unconscious window. A recent study from Singer (2010) found that the brain uses only two percent of its energy on conscious activity, meaning the rest of the energy is used for unconscious processing. The enormous amount of content people get exposed to create an abundance of the content, exceeding their capacity to consume all of it resulting in the need for fast processing.
Research focussed on the conscious window often use the self-reported measurements as methodology to measure the construct of social transmission and its antecedents. Self-reported measurements do not always portray the truth. It is more difficult to define the exact processes that lead to your choice to share certain content. Therefore measurement techniques, like neuroimaging techniques, that unravel the unconscious window could help in knowing how the decision to share truly derives. Recent studies have already suggested that neural measures were better predictors of population level data than self-report measurements (Falk, Berkman & Lieberman, 2012; Berns & Moore, 2012). One method of neuroimaging studies is using Electroencephalography (EEG). EEG records electrical activity in the brain using electrodes based on the scalp. The fluctuation of electrical activity is called
5 an oscillation, and is represented in different frequencies. Different frequencies in oscillations can be observed; delta (less than 4 Hz), theta (4–8 Hz), alpha (8–12 Hz), beta (12–30 Hz), and gamma (30–100+ Hz). Each frequency or oscillation is correlated with different kinds of processes in certain brain areas. It is proven that a higher frequency is the result of more extensive processing (Mahoney, 2017).
Since social transmission includes extensive processing by computing the subjective value, it should be possible to measure the unconscious brain activity using neuroimaging techniques like EEG. The question remains, can a neuroimaging technique measure the same concepts as self-reported measurement and can brain activity measured using EEG also predict social transmission? The goal of this research is to find out whether social transmission commonly measured with self-reported measurements can also be measured with EEG techniques. The following research question derives:
How can social transmission be measured using neuroimaging techniques and how do these measurements correlate with classical self-reported measurements?
This paper makes a couple of important contributions. First, this research tries to fill the gap between existing literature about the value system of social transmission and the underly-ing responsible brain areas. Secondly, this research aims to provide correlations between clas-sical measurements of social transmission and neuroimaging techniques. Finding out whether concepts that are by definition equal also show equal outcomes on different methodologies. This research will therefore contribute to the academic debate whether neuroimaging tech-niques are also valid measurement techtech-niques, and whether real choices have similar neural and psychological consequences (Kang et al. 2011). Much literature has been focused on conscious processing, whereas this research focusses on the unconscious process of social transmission.
Over the past decade, more neuro-scientific insight has been applied to marketing the-ory and practice. For marketing analytics, this research can contribute in finding valid measurement for effect models of marketing campaigns. Marketing departments can convince more brands to test their campaign using neuroimaging techniques while measuring equal concepts as self-reported measurements. Mostly, the impact of campaigns is measured questioning the memory of the respondent. This research may help in developing a model that can determine what customers unconsciously think of the designed content.
This research starts with creating the theoretical framework of social transmission. Then the methodology is introduced and it is learned how the concepts are measured. Next, the data analysis and results are discussed. This research finishes with an elaborative discus-sion and concludiscus-sion.
6
3. Theoretical framework
In order to understand why neuroimaging techniques could be used to measure social transmission, and measuring similar concepts as self-measurement techniques, some basic understanding of the brain and its functioning is required. After gaining basic knowledge the focus will shift to the processing of (sharable) content and the brain areas that are involved in content processing. Hence, the actual action, the final decision to share the content, will be explained and how the brain computes the concluding action. Following this, the different concepts involved in content processing and social transmission will be explained and how they can be measured using neuroimaging techniques and self-measurement techniques. .
3.1 The brain: a short introduction to its functions and structures
The human brain is a complex and extensive system, throughout the years neuroscientists try to understand how the brain works. In order to understand the results of this paper you must be acquainted with the basic structure of the nervous system and the human brain. When describing features of a structure in the brain these descriptions are always relative to the neuraxis. The neuraxis indicates an imaginary line drawn through the length of the central nervous system. The front end is referred to as anterior, the tail end as posterior. Rostral indicates a system towards the back and ventral towards the front, lateral refers to systems towards the outsides and medial refers to systems towards the middle (Reus, van der Land & Moorman, 2008; Carlson, 2014). For example, the ventromedial prefrontal cortex refers to a part of the cortex placed towards the front and in the middle of the brain.
The nervous system consists of the brain and the spinal court. The brain consists of three major divisions: the forebrain, midbrain and hindbrain. The human forebrain is the most developed structure. The forebrain consists of the telencephalon and the diencephalon. The telencephalon includes the cerebral cortex, basal ganglia and the limbic system. The cerebral cortex has a wrinkled outer layer, because of its immense size. The cerebral cortex consists of four main lobes: the frontal lobe; responsible for the motor functions, planning and reasoning, the parietal lobe; responsible for the body senses, the temporal lobe; responsible for auditory functions and the occipital lobe, responsible for the visual functions. Figure 3.1 gives an overview of the cerebral cortex and the functions per lobe (Carlson, 2014).
7
Figure 3.1. Structure of the four lobes of the brain – including functions and designation of
cortices (Ramsoy, 2014)
Figure 3.1. This figure shows the four lobes of the brain. Each lobe includes one or more cortices. The functions of the cortices is explained as well.
The basal ganglia controls your movement and coordination. The limbic system is responsible for emotions and memory, part of the limbic system is the amygdala, this region is responsible for survival instincts (e.g. the fight or flight response). The diencephalon consists of the thalamus responsible for the sensory filter, and hypothalamus, includes the hormonal system. The second major brain division is the midbrain, including the tectum, responsible for connecting inputs, and the tegmentum, which has a role in movements. The last major division is the hindbrain, this part of the brain is included in almost all mammals and consists of the cerebellum, responsible for learned skills and the pons and medulla, responsible for basic support. Figure 3.2 gives a brief overview of the three major parts, their subparts and their functions. The forebrain is the most important structure for this research. Almost every process relevant in social transmission takes place here.
8
Figure 3.2. Three major parts of the brain, their substructures and functions (Carlson, 2014).
Figure 3.2. The brain is divided into three parts, which are subdivided. Each brain area has its own properties and functions, which are also noted in the figure.
The different layers of the brain consist of millions of neurons. Neurons are the building blocks of the brain, each neuron is connected to thousands of other neurons, creating a neuronal network. A neuron is activated when a stimulus is received from the senses. The received information is passed on from one neuron to another, in this way the human brain can process the enormous amount of information. The cerebral cortex is mainly responsible for unconsciously processing the information received.
3.2 How is (persuasive) content processed and which brain areas are involved?
When processing content the brain is consciously and unconsciously working. The dual theory, originated from 1980’s, was the first to explain how the brain processes content and was further developed since then. This theory states that the processing of content takes on two routes or systems of cognition (Petty & Cacioppo, 1983; Ledoux, 1998; Kahneman, 2003) System 1 is known as the intuitive system. Content following system 1 is processed via the peripheral route. System 2 is known as the deliberate system, where content is processed via the central route. System 1 is always active and is reliant on mental operations and biases. System 2 needs a more deliberate activation, it requires attention, time and effort (Kahneman & Egan, 2011).
9 These two different routes also involve different brain areas. The central route or system 2 includes a longer path, and involves important areas of the frontal lobe (higher order cortical areas). Observing content evokes and activates a stimulus response. Information received through our vision is sent to the thalamus for filtering. The thalamus sends the information to the visual cortex for further processing. When received by the cortex the information is sent to the amygdala (part of the limbic system), this part of the brain is responsible for basal emotions, like fear or happiness. The amygdala sends information back to the cortex about the emotional value of the content. In total the central route takes less than half a second to be completed. Though, 500 ms is still too long for some defence mechanisms. Therefore the peripheral route, or system 1 is simultaneously activated. The peripheral route is faster and more efficient. Information processed via the peripheral route almost always reaches the amygdala right away, it only takes a couple of milliseconds (Reus et al., 2008). To thoroughly understand these processes a small example is given. When you see something moving, a reaction of fear is created, while coming closer to the object you see that it is nothing to fear about. Your peripheral route labelled the object as dangerous and your central route corrects this label as non-dangerous. The peripheral route is something that happens unconsciously (not knowingly aware of the process), the central route reaches a conscious outcome. The final goal of processing determines which path is seen as more important (Petty & Cacioppo, 1983). Figure 3.3 explains the content processing via the peripheral and central routes and the brain structures involved.
3.3 How do we make decisions?
The previous paragraph discussed how the brain is structured and which brain structures are involved in content processing, consciously and unconsciously. This paragraph will explain how decisions to share content are formed and how this happens unconsciously.
The cognitive neuroscience literature on decision making has focused on a limited set of brain regions, the main area’s involved in content processing are also involved in decision making, especially the frontal cortex plays an important role. During decision making the brain performs many simultaneous activities. Rangel et al. (2008) developed a framework of decision-making, which defines the processes in our decision behaviour. According to this framework five processes are followed while making a decision.
The first process is the representation phase. In this phase the information is received through the senses and travels to the corresponding cortex (occipital lobe (vision), temporal lobe (sound), parietal lobe (touch and smell)) that can processes the information. Within 100ms a reaction is elicited in these lobes, a feedforward storm to the prefrontal cortex is activated (Lamme, 2004).
10 The second process is called the valuation phase. This phase starts when the feedforward storm is activated. In this phase the value to the actions is assigned on the basis of the associated-action outcome, often referred to as subjective value maximization. In this process the perceived costs and benefits are weighed explicitly and implicitly to derive a common value. Two parts within the prefrontal cortex are highly associated with subjective value maximization, both the ventromedial prefrontal cortex and the medial orbitofrontal cortex integrate multiple inputs from different brain parts to form a common value signal (Delgado, 2007; Grabenhorst & Rolls, 2011; Paulus & Frank, 2003; Fellows & Farraw, 2007; Tom, Fox, Trepel & Poldrack, 2007; Plassmann, O’Doherty & Rangel, 2010). This signal offers a means for comparison between different choices on a common scale, which informs corresponding actions (Bartra et al. 2013). During this phase preferences are formed, and after 100ms a feedback loop is triggered. The neural indicators of implicit preferences have been shown to be predictive of the actual choice people make (Falk et al. 2011; Knutson et al. 2007).
When the feedback loop is activated the prefrontal cortex fires back to the cortical areas it received input from. Starting phase three, the action phase. This phase consists of choosing your option. According to Lamme (2004) not all of the information received by the prefrontal cortex are returned with the feedback loop. Only a small proportion is consciously processed by selecting an action. Though, all information received by the prefrontal cortex is saved in the brains. This means that a big proportion of the received information is not consciously processed, and is unconsciously saved.
After the selecting of an action the fourth process starts, the outcome evaluation phase. In order to make good decisions we consciously save the outcomes of our actions.
The fifth phase is the learning phase, in this process the brain has computed a reward prediction error (Schultz, 2006). The brain tracks the difference between a people’s expected outcome and the actual outcome of one’s action. An action that produced higher than expected is seen as more valuable and an action that produces less reward is devalued. Figure 3.3 shows the decision making process, including the dual process model and feedforward and feedback loop.
11
Figure 3.3. Dual process of sensory input including feedforward and feedback loop (Reus, van
der Land & Moorman, 2008).
Figure 3.3. Both the peripheral, as well as the central route are displayed in this figure. It is clearly visible that the peripheral route travels straight to the amygdala, whereas the central route travels via the visual cortex. A feedforward loop is created, where not all information is received back from the feedback loop.
The framework computed by Rangel et al. (2008) is applicable to all decision to be made, including the decision to share online content or to engage in social transmission.
3.4 Social Transmission; sharing online content
Social transmission is seen as the interpersonal communication and sharing of information between a communicator and a receiver with the implicit or explicit goal of persuasion (Falk & Scholz, 2017). Social transmission is a combination of social influence and word of mouth (Berger, 2011). Social transmission can both happen online as well as offline. This research focusses on online social transmission.
Social transmission, as well as other decision making processes, is derived from subjective value maximization (Falk & Scholz, 2017). The expected value of sharing is derived from your motivations and preferences (Wigfield, 2000). There are two main motivators influencing the expected value of social transmission. Self-enhancement is one of the main motivators. The shared information needs to be helpful to oneself, and not harmful. The shared information must shed the sharer in a positive light (Falk & Scholz, 2017). People are therefore more likely to share useful and positive information (Baek et al, 2017). Social bonding is the second main motivator. Sharing information deepens the bond between sender and receiver.
12 When sharing information the sharer always takes their receivers’ or audience current mind-sets to estimate their potential responses to shared information and to determine the impact of sharing on a conversation or relationship (Falk & Scholz, 2017).
Decisions to engage in social transmission thus involves increased activity in the brain’s value to compute the subjective value (Baek et al., 2017, Falk et al., 2013, Scholz et al., 2017). Processing content, thus always evokes an emotional response in the amygdala. This content is then processed by the higher cortical areas (prefrontal cortex) where, regardless of processing via the central or peripheral route, the subjective value is computed and preferences are formed. There is still one factor from the model untouched. In order to establish online social transmission engagement with the content is needed (Malthouse, Calder, Kim & Vandenbosch, 2016). Without engagement the created preferences do not influence the subjective value. Therefore, when measuring online social transmission, the following model can be used:
Figure 3.4. Model of Social Transmission including measurable antecedents.
Figure 3.4. Conceptual model of social transmission, where all the antecedents are measurable variables.
Hence, in order to measure whether social transmission of online content takes place the antecedents of social transmission can be measured as well. These variables can predict social transmission.
3.5 Measuring online social transmission
Both the framework developed by Rangel et al. (2008) as well as the dual process theory (Kahneman, 2003) explain that not all information is processed consciously and decisions are also not always made consciously. Despite the fact that several decades of research have shown that many important mental processes occur below the surface of consciousness (Dijksterhuis 2004; Zajonc 1980), social transmission and its predictors are still researched using explicit research methods. Explicit research methods include methods like verbal or written self-reported measurements, which leave people very limited in their ability to predict
13 their own future behaviour and to accurately identify their internal mental states (Nisbett & Wilson, 1977). Every decision that is made originates from non-conscious cognitive activity. Psychological processes, especially those which occur beyond our conscious awareness, could be better understood by analyzing the human’s brain and body responses, which can be done with neuroimaging measurements, like EEG (Ohme et al., 2010).
In order to measure constructs like social transmission either, various forms of methodologies can be used. The most commonly used methodology is self-reported measurements, others are behavioural measurements or observational measurements. Self-reported measures are measures in which respondents are asked to report directly on their own behaviours, beliefs, attitudes, or intentions (Lavrakas, 2008). These self-reported measurements are often using a questionnaire that measures constructs, like social transmission, by asking respondents opinion on different items. The items together can predict the latent construct. Self-reported measurements are a low-priced approached of obtaining data and can be easily implemented to large samples (Hoskin, 2012). Each questionnaire implemented in a study needs to be checked for reliability, to check consistency over time, and for validity, to test whether the questionnaire measures what it claims to measure. Using self-reported measurements always includes the risk of response bias, which refers to an individual’s tendency to response towards a more preferably response (Hoskin, 2012).
The two most prominent methods of neuroimaging techniques, measuring the
unconscious mind, are functional Magnetic Resonance Imaging (fMRI) and
Electroencephalography (EEG). In this paragraph these two methods will be explained and how social transmission could be measured.
The first neuroimaging technique that is discussed is the fMRI technique. An fMRI detects the increased oxygenated blood, formed when neurons take up energy to release action potentials (Matthews & Jazzard, 2003). The oxygen levels in the blood increases when there are more action potentials created by neurons, meaning there is more activity present in the neurons. An fMRI scan uses the magnetic properties of oxygen in blood to detect the exact location of the stimulus response. The magnetic field created can be detected by the fMRI scan. An fMRI therefore has an excellent spatial resolution, which is one of the main advantage of fMRI (Matthews & Jazzard, 2006; Scholte, Jolij, Fahrenfort & Lamme, 2008).
Social transmission has already been researched using fMRI techniques. The study of Scholz et al. (2017) combined neuroimaging data with shares of articles form the New York Times. A neurocognitive framework of the underlying sharing decisions was tested, where they tried to translate this into population-level virality. Consistent with studies looking at overall decision and value-making processes (Delgado, 2007; Grabenhorst & Rolls, 2011; Paulus & Frank, 2003; Fellows & Farraw, 2007), they found that the Ventral Striatum (VS, part of the basal ganglia) and the ventromedial pre-frontal cortex (part of the higher cortical areas) were
14 involved in creating the subjective value. Within these studies, the level of activation in brain areas responsible for self-relatedness and social bonding influenced the activity in the brains’ valuation system. Information that elicits greater brain response in self-, social-, and in turn value-related systems were more likely to be shared. So, the overall value of information sharing was predictive of large-scale sharing dynamics. This study thus confirmed that the same brain area’s involved in overall decision making, are also involved in social transmission. An fMRI studies have an excellent spatial resolution, which means that the location of the brain activity is more exact than other neuroimaging techniques. Though, one of the downsides is its poor temporal resolution, the complicatedness and the costs of the measurements. Therefore, this study did not find the reaction time, and exact response to a stimuli. An EEG study results in quicker response, are easier to replicate and costs less to perform. An EEG can detect the level of activity more precisely than an fMRI, therefore an EEG can detect when a response is evoked (Ramsoy, 2014).
EEG measures the firing neurons that act upon a stimulus. To be precisely the synchronized activity of pyramidal (= neurons facing vertically towards the top of the brain) in cortical brain regions is measured using EEG. When neurons fire, they release postsynaptic potentials. When these postsynaptic potentials occur at the same time in synchrony they sum up and generate an electrical field. This electrical field can be measured with an EEG, by attaching electrons to the scalp of the subject (Mahoney, 2017). The fluctuation of electrical activity is called an oscillation, and is represented in different frequencies. These frequencies vary slightly dependent on individual factors, stimulus properties and internal states. Research classifies these frequencies based on specific frequency ranges. They are classified into five categories: delta (less than 4 Hz), theta (4–8 Hz), alpha (8–12 Hz), beta (12–30 Hz), and gamma (30–100+ Hz). The beta and alpha bands are prone to be involved in decision making: the lower frequency bands (alpha) are associated with less complex brain activities (like emotional processing) and the higher frequency bands are associated with more complex brain activities, like cognitive processing (the beta bands). Beta oscillations are thus a sign of more complex brain activities. Therefore studies try to locate these oscillations in order to see where these complex brain activities originate from and if they have any predictive power (Boksem & Smith, 2015; Davidson, 2004; Khushaba, 2013). While temporal resolution is one of the downsides of fMRI, it is one of the strengths of an EEG study. An EEG study can measure the exact moment in time at which an action-potential is activated and one responses to a stimulus. Since it takes 300ms for the conscious window to open, the response time is crucial in the decision making process (Olofsson, Nordin, Sequiera & Polich (2008). The different antecedents of Social Transmission have already been proven to be measurable using EEG measurements (Boksem & Smith, 2015; Davidson, 2004; Khushaba, 2013; Schaffer et al., 1983).
15
3.6 Emotional response
The sensory input travels via the amygdala, evoking an emotional response. The response of emotions can be categorised into two dimensions: valence and arousal. Emotional valence is the degree of positive versus negative feelings that are evoked by the content (Davidson, 2004). Emotional arousal is seen as a state of mobilisation. Low arousal is characterised by relaxation or deactivation, high arousal increases action related behaviour (Davidson, 2004).
3.6.1 Emotional valence
The evoked emotional valence is classically measured using self-report measurements. Self-report measurements of current emotional feelings are more valid than self-Self-reported measurements made somewhat distant in time (Davidson, 2004). One of the most used and valid self-reported measurement of emotional valence is the Positive and Negative Affect Schedule (PANAS). This scale was developed in 1988 by Watson, Clark and Tellegan, the researchers extracted 60 terms correlated with emotional valence and took the 10 items that correlated most on either positive or negative affect (Watson, Clark & Tellegan, 1988). The emotional valence that cause social transmissive behaviour has already been widely studied using self-reported measurement and the PANAS scale. In the context of most forwarded emails, the emails elicited the following emotions: ‘happy’, ‘brightens’, ‘excited’, ‘connected’, and ‘inspired’, though ‘pass-along’ emails were evoked emotions such as: ‘irritated’, ‘disappointed’, and ‘overwhelmed’ (Phelps, Lewis, Mobilio, Perry & Raman, 2004). Most liked video campaigns often evoked the following six primary emotions: ‘surprise’, ‘joy’, ‘fear’, ‘anger’, ‘sadness’ and ‘disgust’ (Dobele et al., 2007; Texeira, 2012). In which surprise and any emotion in combination with surprise increased the social sharing (Dobele et al., 2007). Forwarding behaviour was found to be linked to the desire for fun and social connection, messages that are classified as impactful, motivational, funny or inspirational were more likely to be forwarded (Brown et al., 2010; Phelps et al. 2004).
Emotional Valence can be measured with EEG using the asymmetry model proposed by Davidson et al. (1979). According to this model the level of valence could be measured using the difference between the left and the right hemispheres. They propose that the left prefrontal cortex is involved in approach behaviour, which is often linked to positive emotions, whereas the right prefrontal cortex is involved in withdrawal behaviour from aversive stimuli, which is often linked to negative emotions (Ramirez, Rafael & Vamvakousis, 2012). Activity of more beta oscillations in the right prefrontal cortex thus indicate negative valence, and activity of more beta oscillations in the left prefrontal cortex indicate positive valence. Research indicates that valence exerts influence between 100 and 300 ms (Olofsson et al., 2008). Comparing the hemispherical activation is thus linked to detecting positive or negative valence.
16 Both the self-measurement scales as well as the neuroimaging measurement measure the same concept of emotional valence. The following hypothesis arises:
H1: If emotional valence measured using self-measurement techniques is positive, then the emotional valence measured using neuroimaging techniques should correlate, and more beta oscillations should be detected in the left prefrontal cortex.
3.6.2 Emotional arousal
Emotional arousal is classically measured also using self-reported measurements. One of the most widely used scale is the Pleasure, Arousal and Dominance (PAD) scale. This scale was originally developed by Mehrabian and Russel (1974) and was validated by other researches since. As said, arousal also forms in a dimensions, while low arousal or deactivation is characterized by relaxation, high arousal or activation is characterized by activity (Heilman, 1997). Using self-reported measurement the excitatory state of arousal has been shown to increase action related behaviours. According to Berger and Milkman (2012), since sharing information also requires an action this excitatory activation of content increases the likelihood of that content to be shared. Psychological arousal triggers the autonomic nervous system and boosts social transmission (Berger, 2013).
Neuroimaging studies can also be used to measure emotional arousal. EEG can detect the activity in the prefrontal cortex related to either a high level or low level of arousal. A high level of arousal is often characterized by a higher level of beta oscillations and a lower level of alpha oscillations in the prefrontal cortex. According to Ramirez et al. (2012) the level of arousal could be indicated by the alpha/beta ratio in the frontal part of the brain. Arousal exerts influence shortly after emotional valence exerts influence, arousal exerts influence from 200 ms and later (Olofsson et al., 2008).
Both the self-measurement scales as well as the neuroimaging measurement measure the same concept of emotional arousal. Leading to the following hypothesis:
H2: If emotional arousal measured using self-measurement techniques is high, then the emotional arousal measured using neuroimaging techniques should correlate, and also be high and more beta oscillations should be present in the frontal cortex.
3.7 Preference
As explained, after the emotional response is evoked the subjective value is formed, which then forms the preferences. The preference for the content is seen as the following: “for content is seen as the liking of any content above other content” (In Cambridge Dictionary, n.d.).
17 Preferences are predictors of the actual choice people make (Falk et al. 2011; Knutson et al. 2007). Preference is a more complex mechanism than emotional value. The measuring of preferences is therefore also more complex. There are two common ways that preferences can be measured, either it is measured by means of a conjoint analysis, or by computing a ranking. A conjoint analysis is typically used to reveal preferences for certain attributes of an entity (which can be a product, or content). Consumers try to evaluate the product attributes and levels separately, whereas the total utility of a product is computed. A conjoint method looks at different attributes, whereas a ranking score evaluates the total entity compared to other entities. This method is more relevant when comparing different entities (Eggers & Sattler, 2011). Since this study tries to reveal the content that causes the most visible response, compared to other content, a ranking score is used.
Neuroimaging studies already looked at where preferences originate from and how they are formed. Studies found that Beta oscillations are often originated from brain areas involved in computing of the subjective value, i.e. most beta oscillations originated from the vmPFC (Hlinka et al. 2010; Mantini et al. 2007; Marco-Pallares et al. 2008). In contrast, lower frequencies were related to losses and other negative outcomes (Cavanagh, Cohen & Allen, 2009; Cavanagh et al., 2010; Cohen, Elger & Ranganath, 2007; Marco-Pallares et al. 2008; Van de Vijver, Ridderinkhof & Cohen, 2011). It was also found that one of the roles of beta oscillations might be synchronizing neural populations over a longer distance, in order to connect the different brain areas involved in valuation and reward processing (Berns et al. 2001; Marco-Pallares et al. 2008). Lucchiari and Pravettoni (2012) recently observed that beta activity seems to be modulated by the experience of pleasure associated with a favourite brand, whereas theta modulation seems to reflect the lack of this experience. Research from Boksem and Smith (2015) found that beta oscillations in the medial prefrontal cortex are related to individual preference. The higher the amplitude of EEG oscillations in the beta frequency range (16–18 Hz) during viewing of a movie trailer, the higher participants ranked that particular movie relative to the other movies for which they viewed trailers. Adding the EEG data to the preference model significantly increased the explained variance of the preference for movies (Boksem & Smith, 2015). Another study that looked at preferences and choice using EEG was the study of Khushaba (2013), where they examined choice options for crackers and different preferences. They showed that there was a clear phase synchronization between the left and the right frontal and occipital lobe indicating inter-hemispheric communication during a choice task. Beta oscillations in the frontal cortex should thus indicate preferences for the content shown. Both the self-reported measurement as well as the neuroimaging technique measure the same concept of preference. Therefore the following is suspected.
18 H3: If preferences for the content measured using self-reported measurement is high, then preference measured using EEG should correlate, and more beta oscillations in the frontal cortex should be detected.
3.8 Engagement
As said, In order to establish social transmission, engagement with the content is needed (Malthouse et al., 2016). Engagement is seen as being involved or absorbed into the content. A higher level of engagement results in more attention to the content. Classically engagement is measured using self-reported measurement scale. Engagement has been widely researched in combination with different contexts. Therefore engagement does not have a commonly used scale variable.
When using neuroimaging techniques to measure engagement, frontal lateralization was found to be useful. Higher oscillation band power in left versus right frontal cortex indicated positive feelings, engagement and motivation (Davidson, 2004; Schaffer et al., 1983). Especially gamma oscillations are predictors for engagement. Engagement is computed with a higher order consciousness and thus more brain activity is needed to be engaged. Therefore gamma oscillations are often present while being engaged. Higher gamma band power in left versus right frontal cortex reflected a person’s monetary approach-avoidance tendencies, to either engage or withdraw (Davidson, 2004). Again, both the classical as well as the neuroimaging techniques are used to measure the same concept. Therefore the following is suspected:
H4: If the engagement measured using self-measurement techniques, then engagement measured using EEG techniques should correlate and more gamma oscillations in the left frontal cortex should be detected
3.9 Model of Social Transmission
Previous paragraphs explained that the antecedents of social transmission can be measured using either self-reported measurements or EEG techniques. When the antecedents correlate, and hypothesis 1 to 4 are true, than the antecedents measure equal concepts. The following is then suspected:
H5: If the antecedents of social transmission measured with EEG techniques correlate with the self-reported measurements, then the antecedents measured with EEG techniques should be able to predict social transmission.
19
Figure 3.5. Final conceptual model of Social Transmission, including all measurable variables.
Model 1:
Model 2:
Figure 3.5. Social transmission is both predicted by Emotional Value, and Emotional Arousal. It is not possible for a model to have two independent X-variables of emotion. Therefore the model computed in figure 3.4 is separated into one model using the variable ‘emotional valence’ and one model predicting social transmission using ‘emotional arousal’.
4. Methodology
This chapter elaborates on the data and methodology used to study the research question and test the hypothesized effects. First, the overall experimental setting will be explained, thereafter the variables and data preparation will be explained separately.
4.1 Research Design
The purpose of this study is to determine the correlation between two different methodologies, and testing the model for social transmission with those two different methodologies. The correlation is tested by measuring respondents using self-reported and using EEG. The same respondents are thus measured twice. Therefore this study uses an experimental within
20 subjects design. An experimental design is often chosen to establish causal relationships. In order to test whether the EEG measures also predict social transmission a causal relationship needs to be established between the variables used in the experiment.
4.2 Pre-test & Stimuli
In this study the concept of social transmission is investigated, how this propagates in an online environment. Therefore, this research uses a combination of online images. The online images used are also known as memes. Memes often circulate in an online environment. Memes are conceptualized as the following: “An Internet meme is a piece of culture, typically a joke, which gains influence through online transmission” (Davidson, 2012). Memes are therefore also related to the concept social transmission.
Memes were selected form an online meme database, from the following website:
www.knowyourmeme.com. Know Your Meme is a website and video series which uses wiki software to document various Internet memes and other online phenomena, such as viral videos, image macros, catchphrases, internet celebrities and more. It also investigates new and changing memes through research, as it commercializes on the culture. The release dates of the memes were from the year 2017.
In order to establish stimuli that evoke a clear response by participants a pre-test was conducted. The pre-test helps with measuring the validity of the scale questions. The pre-test consisted of 48 participants of which 26 completed the entire test. Respondents were picked according to a convenience sample of the researcher. In total 63 meme pictures were shown to the participants. Each participant viewed 5 of the 63 pictures. Of the total 63 pictures, 14 different pictures were only shown once, these pictures were deleted from the sample. The remaining 49 pictures were assessed upon their emotional arousal, emotional valence, engagement, and their preference. For each individual variable, the picture was given a score. In the end, the score was added up per picture. A selection of 25 images was made upon the leftover 49 images. The top performing pictures were used in the final experiment. A diversity of images was used to make sure subjects weren’t already aware of most of the pictures. Appendix A shows the total images used for this experiment
4.3 Procedure
The experiment took place in the lab of the Amsterdam Business School. After arriving at the lab, participants received an informed consent, all participants must agree to the informed consent in order to proceed with the EEG. In total the study had a duration of 30 minutes. The EEG study consisted of 5 minutes and the self-reported measurement study consisted of 10 minutes. There was a preparation time needed of 15 minutes.
21 Participants started with the EEG study. Participants were asked to take place behind the computer. EEG recordings are done with electrode arrays. The electrical data is caught from sensors placed on the scalps surface. The EEG electrodes are mounted in elastic caps to make recordings easier. This cap has fixed locations of the electrodes. The most commonly used system is the 10-20 system (Mahoney, 2017). The 10-20 system is describes the locations of the electrodes, the electrodes are placed at 10% and 20% along lines of longitude and latitude. In this study the 10-10 system will be used, which is an extension of the 10-20 system. More electrodes can be placed using this system. In the 10-20 system, electrode names begin with one or two letters indicating the general brain region or lobes where the electrode is placed (Fp = frontopolar; F = frontal; C = central; P = parietal; O = occipital; T = temporal) (Oostenveld & Praamstra, 2001). Each participant needed to put on the cap including the electrodes. The neoprene caps were available in different sizes, each participant received a cap of their own size and the electrodes were placed upon these caps. EEG was recorded from a 10 active dry electrode Enobio from Neuroelectrics (company developing EEG for commercial use). Electrodes were placed, only on the eight channels that were needed to test our hypothesis: AF3, F3, AF4, F4, Fz, and Cz, and two electrodes as a reference. The electrode positions will be explained in the variable section. In contrast to most neuroimaging studies, this study makes use of dry electrodes. Dry electrodes are developed to measure brain activity without the use of conductive gel. Conductive gel is often used for better conductance of the electrodes measuring the brain activity. Using dry electrodes makes the process go much faster. Figure 4.1 shows the electrodeposition used in this study using the 10-10 system.
Figure 4.1. The 10-10 system for placement of electrodes in an EEG study
Figure 4.1. The 10-10 system indicates where electrodes can be placed in an EEG study. The electrodes indicated red are used in this study.
22 EEG measurements can detect small muscle movements. In this study, the frontal part of the brain is measured the electrodes are also capable of measuring the muscle movements of the forehead, including muscle movement in the eyes and eyebrows (such as blinks). In order to correct the recordings on muscle or eye movements an electrooculogram (EOG) was recorded. In total two electrodes were placed on the forehead. One was placed above the eyebrow for recordings of horizontal EOG. One for vertical EOG, which was placed on the temple, in line with the pupil (Mahoney, 2017). Since the electrical signals are very small, the recorded data is digitized and sent to an amplifier. The amplifier used in this paper is the Enobio from the company Neuroelectrics. Once the data is amplified, it can be displayed as a time series of voltage values. Voltage describes the electrical pressure or the electrical potential. Most software programs measuring EEG data can show the data in a continuous flow of voltages. EEG recordings are often described in frequency and power. Frequency is the speed of an oscillation expressed in Hertz (Hz). Power is the amount of energy in a frequency band typically expressed as squared amplitude. The EEG and EOG signals were sampled at a rate of 500 Hz, meaning that 500 samples are taken in one second. The EEG recordings always reflect the voltage between the site electrode and the ground electrode. The ground electrode is circuited in the amplifier. Since this ground electrode can always introduce some noise to the data, a reference electrode is used. Using the reference electrode cancels out the noise introduced by the ground electrode. The reference electrode must be placed on a presumably inactive zone, therefore a two electrodes are placed on the earlobes. There is almost no activity on the earlobes, meaning that the noise caught by the reference electrode is no brain activity (Mahoney, 2017). These electrode positions are called the EXT and CMS electrodes. In total there are thus 10 electrodes placed. All the equipment used in this study can be found in Figure 4.2.
23
Figure 4.2. Equipment used in a neuroimaging study
Figure 4.2. All equipment used in this study was provided by the company Neuroelectrics.
During each study, the placements of the electrodes was loaded into the software program from Neuroelectrics called NIC2. This software makes sure that the electrodes are connected to the device and the placements of the electrodes are correctly read. In order to make sure the software connects the right electrode to the right position in the 10-10 system a protocol was made. This protocol explains to the software which electrodes are used and were they are placed. The protocol also runs for a certain amount of minutes until the experiment was finished. In total the protocol ran for 5 minutes. The protocol used is shown in Appendix B. Matlab is used to store and process the data, we use an extension of Matlab, called MatNIC. MatNIC is able to connect NIC2 to Matlab. The protocol is loaded and run by commands given to MatNIC (Appendix C). In order to present the stimulus Presentation was used. This software is able to talk to both NIC2 and Matlab. It is a neurobehavioral software, specially made for neuroscience studies. Presentation also makes it possible for respondents to precede to the next stimuli image by clicking their left mouse-button. When the participants would click the mouse in order to precede the experiment a marker would be sent to NIC2 and Matlab, to show at which time point the respondents were shown to next picture. The stimuli are presented in a loop in a specific order. This order has been presented to make the script easier. This can affect the external validity of the experiment. The 25 different pictures that were used were presented using a self-made script. The specific script can be found in Appendix D. The EEG
24 experiment thus starts when the protocol is loaded in Matlab and the stimuli is presented in Presentation. A set-up of the experiment is shown in Figure 4.3.
Figure 4.3. Set-up of the experiment
Figure 4.3. This figure shows the set-up of the experiment. The experiment starts at point 1 and ends at point 9. Point 5 and 6 are steps within the process that start at the same time.
The second part of the study consisted of an experimental study programmed in Qualtrics. While in the first study the participants viewed all the images/stimuli used in this study the classical measurement methods take a longer period of time to complete per image. Therefore, instead of showing all 25 images and assessing them on the self-measurement scales, just 4 images were reviewed. Each participant rated the pictures according to the measurement scales. The experiment can be found in Appendix E.
4.4 Sample and data collection
As explained, this study consisted of two parts. The same participants were used in the EEG measurements as well as the classical measurements. The sample of the study (N= 67) consisted of only students from the Amsterdam Business School. The students could receive credits for completing the entire study. We used a student sample for the cost-effectiveness and ease of use. All were students and most of them employed part-time. Participants were aged between 17 and 23 years old (M = 19.78; SD = 0.29). In total 27 males (40.3%) and 40 females (59.7%) participated. None of the participants were deleted from the self-measurement sample, they all finished the survey. All of the participants used social media, so none of the participants needed to be excluded from the cases.
25
4.5 Measurements
This section explains the operationalization of all variables used in the experiment. The survey flow of the pre-test and of the main experiment can be found in Appendix E. In the present study social transmission is used as the main dependent variable. Arousal, valence, preferences, and engagement are used as the independent variables. Each measurement section will first explain how the variable is measured using self-reported measurement techniques, followed by how they are measured using EEG neuroimaging techniques. As explained the brain activity is measured using the EEG Enobio device and is presented in respective band power. The electrode positions explained in this section correspond to the 10-10 system explained in the procedure.
4.5.1 Social transmission
Social transmission was measured by asking respondents’ willingness to share. After analysing the pre-test it was seen that using the question: “would you share this image on your own social platform”, did not perform quite well. Respondents were often not willing to share the image. Hence, the action of liking the image, or tagging a friend in the caption of the image occurred more often. Verbal feedback was given by the respondents that it was more common for the images to be liked or tagged, than to share the image. Therefore the willingness to share was operationalized into three different parts: liking; tagging and sharing. According to (Falk et al., 2011) likes might be a precursor for social transmission. However, the complexity of social transmission makes it impossible to detect with the EEG, therefore the dependent variable can only be assessed upon self-reported measurements.
4.5.2 Emotional Arousal
As previously mentioned, for emotion research the alpha and beta bands are the particular bands of interest. The level of arousal is determined by computing the ratio of the alpha and the beta band waves as recorded by the EEG. The EEG signal for arousal is measured in four different locations, located in the pre-frontal cortex: AF3, AF4, A3, and A4. High arousal is characterized by a large number of beta waves and low activity of alpha waves, since Beta waves are associated with higher brain activity and alpha waves with relaxation. Thus the beta/alpha ratio is an indicator of the arousal state of a person (Matlovic, 2017). Arousal can be measuring using the strength of the beta waves divided by the strength of the alpha waves in the electrodes located in the prefrontal cortex.Therefore, the following formula can be used to measure the level of arousal:
26 As scale variable for arousal, the Pleasure, Arousal and Dominance (PAD) model was used. The PAD scale was originally developed by Mehrabian and Russel (1974). This model describes three emotional dimensions; pleasure, arousal, and dominance, and is used to describe perceptions of physical environments. For this study, only the variables linked to arousal are used. Arousal reflects the extent to which the environment stimulates the individual (Hall, Elliott & Meng, 2017). In total the Arousal scale consists of 5 items, which are assessed on a 5-point Binominal scale. Participants choose the aroused feeling that they are most compelled to, an example of an item is choosing between ‘stimulated’ or ‘relaxed’. Cronbach's alpha of this scale = 0.842.
4.5.3 Emotional Valence
The level of valence is measured by comparing the activation level in the different hemispheres. Research showed that the difference in activation between the cortical hemispheres was an indication of either a positive state or a negative state. Left frontal inactivation is an indicator of a withdrawal response, which is often linked to a negative emotion. On the other hand, right frontal inactivation may be associated with an approach response or positive emotion. Since high alpha activity was an indication of low brain activity and high beta activity was an indication of higher brain activity, an increase in alpha activity and decrease of beta activity in the different hemispheres can be an indication of cortical inactivation. Electrodes F3 and F4 are located in the prefrontal cortex, which play a crucial role in emotion regulation. Using the following formula the alpha and beta power in the different hemispheres is detected, which detects valence:
Valence = α (F4)/ 𝛽 (F4) − α (F3)/ 𝛽 (F3)
As scale variable for arousal, the Positive and Negative Affect Schedule (PANAS) from Watson et al. (1988) was used. This scale consists of 20 items that describe feelings and emotion on an affected level. Subjects rate their score per item on a 5-point Likert scale, from Strongly Agree to Strongly Disagree. This scale includes items which employ either positive or negative emotions, such as ‘Interested’, ‘Upset’ or ‘Nervous’. Cronbach's alpha of Positive Valence scale = 0.885. And Chronbach’s alpha of Negative Valence scale = 0.846.
4.5.4 Preferences
Preferences are also computed in the frontal part of the brain. While computing your preference, both hemispheres are working and computing the same preference signal. Both Boksem and Smith (2015), as well as Khushaba et al. (2013), found that the electrode placed in the middle of the prefrontal cortex could be used to measure preferences. Since the computation of preference needs a higher brain activity, the detection of beta waves was an indication of a higher preference (Boksem & Smith, 2015; Khusaba et al., 2013). The
27 preferences for the stimuli is measured using higher amplitudes of beta waves around the FCz electrode. A higher power of beta waves indicates a higher preference. Unfortunately the Enobio EEG device misses the location of the FCz signal. For this reason the electrodes parallel from the FCz are used. The difference between the beta activity in the Cz and Fz electrode indicated the beta activity in the FCz electrode (Mahoney, 2017). The following formula is used to calculate the power of the beta activity in the Fz and Cz electrodes:
Preference = 𝛽 (Fz)/ 𝛽 (Cz)
As mentioned in the theoretical part, preferences are often assessed by ranking the presented stimuli. In the research of Boksem and Smith (2015), the preference of the movies was computed by ranking the movie trailers from least favoured to most favoured. In this study, due to time restrictions and easiness, the overall ranking score of one image is asked, when comparing the score, the pictures can be placed from least preferred (1) to most preferred (4). Each picture thus receives a score from 1 to 11.
4.5.5 Engagement
Engagement is measured using the frontal lateralization between electrode F3 and F4, measuring the gamma oscillations. Gamma waves are correlated with higher order consciousness (which includes engagement) and frontal lateralization can be used for testing engagement (Astolfi et al., 2008; Yilmaz et al., 2014). Higher gamma band power in left vs. right frontal cortex indicated positive feelings, engagement and motivation, it reflects a person’s momentary approach-avoidance tendencies, to either engage or withdraw. Recent evidence suggests that frontal lateralization can indeed be used for testing of respondents’ engagement when confronted with media ads, physical products, and services. Values of Engagement show how involved people are while looking at the stimulus. When this value is low, people tend to drift off with their minds and their attention will have lost. The log of the gamma scores is used to also smoothen the data. Therefore engagement can be measured using the following formula:
Engagement = Log (F4 Gamma/F3 Gamma)
To assess engagement on a scale variable the Customer Engagement (CE) scale of Cheung et al. (2011) is used, this scale measures the engagement level of customers from different target groups on different platforms (online and offline). For this reason this scale can also be used in an online social media environment, and therefore can also be implemented in this experiment. Items in this scale either assessed absorption of the content or dedication to the content. These items were rewritten to measure the engagement level in a social media environment only. After rewriting the items a Principal Component Analysis was performed. This scale consists of 7 items which employ the absorption of the social media content and dedication of the social media content. The items are assessed using a 5-point Likert scale
28 from ‘Strongly Agree’ to ‘Strongly Disagree’ as suggested by Cheung et al. (2011). The KMO measure verified the sampling adequacy for the analysis, KMO =0.78. Bartlett’s test of sphericity X2 (21) = 653.49, p<0.001. The 7 items on Engagement formed two components with an Eigenvalue higher than one, the first component explained 51.91% of the total variance in Engagement. Items 1 and 2 were therefore deleted from the scale in order to compute one scale variable of Engagement. Table 1 shows the factor loadings after the rotation. The Engagement scale has a high reliability, with Chronbach’s Alpha = 0.87.
Table 4.1. PCA Engagement items
Component 1 2 Item 1 0.03 0.92 Item 2 0.11 0.95 Item 3 0.21 0.84 Item 4 0.98 0.04 Item 5 0.90 0.05 Item 6 0.85 0.17 Item 7 0.87 0.20
Note. Only component 1 was used in this study, deleting item 1 and 2.
4.5.6 Demographic variables
During the study, some questions about demographic variables were collected. Respondents were asked about their gender, divided in male/female, measured on a nominal scale, age, measured on a ratio scale, and employment, measured on an ordinal scale. Besides the demographic variables, some general questions about their usage of social media were asked, including how many times they post, like, share or tag friends on social media. Questions concerning their social media usage might explain variance in the eventual outcomes.
4.6 Data Preparation of EEG datasets
This section will explain the data preparation of the self-measurement model and the EEG model. From the total 67 participants that participated in the EEG study, 11 participants needed to be excluded, due to technical problems with the EEG device during the experiment, as participants with thick hair or sweaty foreheads could cause channels/electrodes to be disconnected with the EEG device.
Data extracted from the EEG channels consists of a lot of noise, muscle movement and other artefacts (Mahoney, 2017). Therefore the data that is extracted needs to be cleaned before data processing. The frequency bands that are of interest in this study are as explained the alpha, beta and gamma bands. Any frequency bands that are measured below or above these frequency can therefore be deleted from the data, this is called removing the baseline.
29 After the baseline was removed from the 56 participants an Independent Component Analysis (ICA) ran through the data. ICA applied to EEG-data has proven to be an efficient method for separating artefacts from clean data (Delorme, Sejnowski & Makeig, 2007). An ICA is a statistical measurement in Matlab transforms multidimensional vector into components that are statistically independent. It searches for a linear transformation that minimizes the statistical difference between components. This way the different participants can be compared to each other as well as with their channels.
After the ICA was performed each data set of the respondents was checked visually by the researcher. Data sets that consisted of linear drift were excluded from the dataset. Drift is also seen as an artefact. Drift can be noticed when data is visually moving towards a downwards spiral or upwords spiral. In total 14 respondents needed to be excluded, because of too much drift in the data. In Appendix F an example of ‘good’ drift vs. ‘bad’ drift is shown. Following this the data needed to be cut in what is called epochs: the data needed to be cut in epochs of 0 to 500 ms. In this way, each epoch represents a response to one of the 25 stimuli presented. Data was cut by means of the marker from the mouse click that was sent to the data using Presentation and MatNIC. In total, the data per respondent would consist of 25 epochs, because in total 25 pictures were shown. In total 5 participants had more than 25 epochs, and thus clicked more than 25 times on the left mouse-button. This caused uncertainty of which image they saw. Therefore these respondents also needed to be excluded. In Appendix G an overview can be found of the total respondents and which ones are still left. Therefore the EEG study consisted of 37 respondents, of which the mean age was 19.8 (SD=1.15), and 57.1% female and 42.9% male. The loss in the number of respondents might be caused by the following reasons. First of all, this research did not make use of performing a dry run experiment. The behavioural responses were not tested in advance on errors. Secondly, the processing and analysing of the EEG data do not have specific steps, there is a wide range of approaches. Thirdly hand or leg muscle artefacts may cause a slow drift in the EEG amplitude. Respondents were asked to sit still, but this could explain the number of drift in the data.
For the computation of the variables the band waves needed to be extracted from each data set and its epochs. Therefore a Butterworth filter was used, this filter extracts the specific EEG frequency bands (Butterworth, 1930). To estimate the EEG connectivity across different electrode positions the following bands were calculated: alpha band (8–12 Hz), the beta band (13–30 Hz) and gamma band (31–50 Hz). After filtering the data it could be checked whether there was an evoked response to the presented stimuli per frequency band, therefore an Event Related Potential Analysis (ERP) was performed
30
4.7 Analysis of the Event Related Potential
An ERP is a time-domain analysis that calculates whether the potential evoked was positive (presented a positive polarity) or negative (non-active). An ERP looks at the response evoked by the stimuli and the data will give an overview of the different potentials evoked in a response at which time frame. There are two main components within an ERP study, the N represents the negative polarity and the P the positive polarity, followed by a number that indicated the latency. The alternating polarity shows the fluctuation in frequency per ms. When the waves are differentiating in power, this implicates a different frequency oscillation. The first response after a stimulus is shown, is already visible after 100 ms. The first arousal, valence, preference or engagement response will already be visible after 300 ms. Therefore, the analysis of the different electrodes is discussed with ERP’s from 0 to 500 ms. The ERP waveform is formed by averaging across the epochs within each group per electrode (Cohen, 2014). The ERP analysis is given per position and band wave. Figure 4.4 represents the mean evoked response per electrode. The y-axis shows the potential, which is either positive or negative. The x-axis shows the time in milliseconds. The fluctuations in time from positive to negative represent an activated response in different frequencies. The ERP analysis gives a fast overview of the brain processing per position. From these graphs it can be concluded that the stimulus indeed evoked a response in all electrodes used to test the hypotheses. The gamma band ERP signals either show a high polarity (in Cz) or a low polarity (in Fz). This could be an indication of artefacts in the data. This needs to be taken into consideration when making conclusions about the gamma band power.
Figure 4.4. The ERP signals per electrode position
Alpha Bands
