Effects of continuous inclusion in a virtually played ball tossing game called Cyberball on the P300 event-related-potential

A single trial analysis

Diploma Thesis 2011 137 Pages

Psychology - General



1 Abstract

2 Introduction
2.1 Introduction
2.2 Ostracism
2.3 Paradigms that Trigger the Feeling of Ostracism
2.4 External and Internal Factors that Moderate Ostracism ’ s Immediate Impact
2.5 The Need Threat Model
2.6 The Ostracism Sensitivity Threshold and Clinical Implications
2.7 Ostracism and Social Inclusion
2.8 Event-Related-Potentials
2.9 P300 Component and the Oddball Paradigm
2.9.1 P3b Latency
2.9.2 P3b Amplitude
2.10 A Social Oddball Paradigm
2.11 Single-Trial Analysis and the P300
2.12 Development of the Single Trial Analysis
2.12.1 Extraction of the Component of Interest
2.12.2 Independent Component Analysis
2.13 Aim of the Current Study
2.14 Hypothesis

3 Materials and Methods
3.1 Participants
3.2 Period of Investigation and Laboratory Environment
3.3 Cyberball Paradigm: Stimulus Sequences and Material
3.4 Experimental Design
3.4.1 Independent Variable
3.4.2 Dependent Variable
3.5 Control of Confounding Variables
3.6 EEG Recording
3.6.1 Electrode Placement and EEG Settings
3.6.2 Equipment
3.7 Test Execution
3.7.1 Participant Recruitment
3.7.2 EEG Preparation
3.8 Procedure

4 Data Analysis
4.1 EEG Preprocessing
4.1.1 Approach 1: Non-ICA correction
4.1.2 Approach 2: ICA correction
4.2 Template Matching Method
4.3 Statistics
4.3.1 Effects of Time and Group on the P300 Potential
4.3.2 Probability Models and the P300 Amplitude

5 Results
5.1 Effects of Time and Group on the P300 Potential
5.1.1 Amplitudes
5.1.2 Latencies
5.2 Probability Models and the P300 Amplitude
5.2.1 Linear Fit
5.2.2 Model of Local Target Probability

6 Discussion
6.1 Summary of Results
6.1.1 Non-ICA Corrected Data
6.1.2 ICA Corrected Data
6.2 Hypothesis
6.3 Integration of Results in the P300 Literature
6.3.1 P300 Amplitude
6.3.2 P300 Latency
6.4 Interpretation of the P300 Component Referring to an Ostracism Detection System
6.5 Summary According to the Method of Single Trial Analysis
6.5.1 The Approaches of Non-ICA and ICA Correction
6.5.2 Template Matching Method
6.6 Conclusion
6.7 Critical Review and Prospective Studies

7 References

8 Appendices


The current study investigated effects of continuous social inclusion in a virtual ball tossing game on the P300 event-related potential based on a single trial analysis. Participants played an ostensible online game called Cyberball, which served as a social modification of the classical Oddball Paradigm. During the inclusion session, ball passes to the participant became more frequent over time from almost none to 33% at the very end of the game. Each participant had to ran an inclusion and an exclusion session. According to the order of experimental conditions, two inclusion-groups encompassing 15 datasets were taken for analysis. Of special interest was the centro-parietal P3b subcomponent that emerges 220ms-500ms after stimulus onset.

The signal detection technique included template matching and cross-correlation. The data were analyzed twice: Once with the raw data, once preprocessed with the application of the Independent Component Analysis. Performances of the different approaches were compared and evaluated. Effects of target probability changes on the P300 amplitude were investigated. Further a linear fit and a model of local target probability were applied to account for P300 amplitude variations. Latencies were mainly examined to validate the methods of single trial estimations.

Group analysis of the non-ICA corrected data implied that the P300 amplitude significantly decreased over time. In the ICA corrected data only amplitude height significantly differed between groups. However, a negative slope of the linear fit in the majority of both datasets emphasized a continuous decrease of P300 amplitudes. Therefore, an inverse relationship of target probability and the P300 amplitude was replicated for a socially relevant stimulus - at least this relationship was indicated if not significantly in the ICA corrected data. The local probability fit to the single trial P300 amplitude explained ten to fifteen percent of the variance. This supported the assumption that preceding stimulus sequence is one determinant that triggers P300 amplitude behavior. Additionally, the non-ICA corrected data revealed enhanced latencies over time, which was contrary to previous findings on the relationship between target frequency and P300 latency. However, these remained difficult to explain.

Furthermore the study showed that cross-correlation and template matching are reasonable methods to estimate single trial signals. However, The applied FastICA algorithm was not satisfactory for extracting ERP components. All in all statistical and methodical limitations could account for different results from the two approaches of analysis.


2.1 Introduction

Social exclusion is a phenomenon that independent of its nature immediately leads to distress and affects people’s well-being (K. D. Williams, 2007). Additionally, people who are very sensitive to social rejection tend to overestimate the occurrence of social exclusion events and are supposed to have a biased detection of Ostracism (K. D. Williams, 2009a). However, the question if social exclusion is perceived even if it is immediately followed by a period of inclusion has never been investigated before. To assess the processing of continuous social inclusion on an electrophysiological level is of particular interest in the current study. Since the P300 event-related-potential is associated with the evaluation of relevant events, the current study will investigate the component’s behavior during the game of Cyberball.

In the following passages findings of social exclusion will be reported before eventrelated potentials and the approaches of analysis will be entered.

2.2 Ostracism

Ostracism1, as the process of being ignored by individuals or groups (K. D. Williams, 2007), is a phenomenon found in human and non-human beings. Nearly all of us have once in our lives experienced social exclusion. According to Williams Ostracism occurs explicitly or just by being kept apart by others but it often happens without excessive explanation or explicit negative attention towards an individual (K. D. Williams, 2007). Because Ostracism has been observed in most species and from a number of perspectives, it is appropriate to consider an evolutionary view on its function and existence (K. D. Williams, 2007). In fact, Gruter and Masters describe Ostracism to appear in modern industrialized nations, in governmental, military, religious institutions and penal facilities, educational institutes and in informal groups or close relationships - thus, Ostracism can be seen as very powerful and pervasive (Gruter & Masters, 1986). Ostracism is a well-investigated phenomenon, especially in the field of Social Psychology since belonging is considered to be a fundamental requirement for security, mental health and reproductive success (Baumeister & Leary, 1995). Because of this Need to Belong, Ostracism may lead to maladaptive decisions and behaviors. The increasing frequency of irrational and socially intolerable incidents such as mass- shootings or the willingness to conduct terrorist acts has recently been associated with growing social isolation (Twenge, Baumeister, Tice, & Stucke, 2001). The observation of outrageous violence as possible consequence of social exclusion yields the growing interest on examining the phenomenon of Ostracism (K. D. Williams, 2007). To investigate the phenomenon of Ostracism and its consequences, several paradigms have been evolved that will be introduced in the next passage.

2.3 Paradigms that Trigger the Feeling of Ostracism

Research on Ostracism and social exclusion has grown fast and several paradigms have been introduced to operationalize social exclusion. In general paradigms can be categorized in two ways: Face-to-face or social paradigms and Cyber paradigms (Goodacre & Zadro, 2010).

Face-to-face paradigms involve those in which participants are ignored during the physical presence of an ostracizer. Such paradigms include e.g. ball tossing games (K. D. Williams & Sommer, 1997), conversations (Ezrakhovich, et al., 1998) or role play tasks (Zadro, Williams, & Richardson, 2005). Even if social paradigms offer high ecological validity, two main disadvantages remain. Face-to-face paradigms are rather resource- intensive, since they require the presence of participants, who play the source of Ostracism. Additionally it is difficult to guarantee standardized behavior of the confederates, thus to assure the same experimental conditions for each participant - even when ostracizers are well trained. Hence, only a small number of studies used social paradigms (Goodacre & Zadro, 2010).

Cyber paradigms on the other hand do not require the actual presence of confederates since Ostracism happens through electronic exchange with preprogrammed players. Hence they are easier to run and offer a great deal of control during the experiment. Cyber paradigms induced Ostracism via chat room discussions (K. D. Williams, et al., 2002), text-messages on cellular phones (Smith & Williams, 2004) or webcams (Goodacre & Zadro, 2010).

Williams and Jarvis established one common used Cyber paradigm called Cyberball (K. D. Williams & Jarvis, 2006). Cyberball is an ostensibly online ball tossing game in which participants believe to play with two other unknown players, who are in fact computer generated. First, the participant is included, receiving one third of all throws. Abruptly, he or she is excluded, never receiving the ball again for the rest of the game. Cyberball was designed to manipulate independent variables such as Ostracism but also to examine dependent variables of prejudice and discrimination (K. D. Williams & Jarvis, 2006).

Williams and Jarvis used post experimental questionnaires asking the participants how they felt during the game. They found significant lower levels of belonging, self-esteem, control, meaningful existence and mood. This and subsequent studies replicated these findings with strong effects ranging from 1.0-2.0 (K. D. Williams & Jarvis, 2006). These results imply that short periods of Ostracism cause a temporary state of cognitive deconstruction. Because of these consistently replicated effects the phenomenon of Ostracism has received considerable empirical attention primarily because of the crucial interest in the essential importance of belonging for human social behavior. So far a huge amount of studies has gone deep into detail into the underlying mechanisms of being ostracized (Goodacre & Zadro, 2010; Leary, Haupt, Strausser, & Chokel, 1998; Smith & Williams, 2004; K. D. Williams, Cheung, & Choi, 2000; K. D. Williams, et al., 2002; Zadro, Williams, & Richardson, 2004). Some implications will be presented in the next passage.

2.4 External and Internal Factors that Moderate Ostracism’s Immediate Impact

As mentioned above, Ostracism increases self-reported distress. Considerable research has shown that this effect is very resistant to the moderation of situational and individual factors (Goodacre & Zadro, 2010; Leary, et al., 1998; Nadasi, 1992; Smith & Williams, 2004; K. D. Williams, et al., 2000; K. D. Williams, et al., 2002).

Different paradigms show that social exclusion induces lower senses of belonging, self- esteem, control and meaningful existence independently of the source of Ostracism (Goodacre & Zadro, 2010; Smith & Williams, 2004; K. D. Williams, et al., 2000; K. D. Williams, et al., 2002). Further on, explicit and external reasons do not seem to trigger such distress. For example, Smith and Williams reported that ostracized participants were angrier and showed enhanced negative mood, even when they did not know that they were purposely excluded and even when in-group or out-group members ostracized (Smith & Williams, 2004). According to a study by Zadro and Williams, the participants displayed the same level of Ostracism, be it intentional or unintentional, even if told that their counterpart were computer-generated players (Zadro, et al., 2004).

Furthermore Ostracism’s immediate impact is not moderated by individual differences. Zadro et al. reported that differences in immediate reactions on being ostracized did not vary with individual predispositions such as social anxiety or loneliness (K. D. Williams, 2007; Zadro, Boland, & Richardson, 2006). Additionally, Ostracism causes distress regardless of the initial level of trait self-esteem (Leary, et al., 1998). Above this, findings imply that not even personality traits such as introversion or extraversion moderate immediate responses to Ostracism (Nadasi, 1992).

The broad evidence points to the conclusion that other potentially explanatory factors, like in-group or out-group membership, the ability to attribute Ostracism to no inflictive causes, and individual predispositions and personality traits are overwhelmed by simply being ignored or excluded. Williams proposed a model that attempts to predict the impact and consequences of Ostracism, which will be introduced in the following section.

2.5 The Need Threat Model

Williams combines processes, consequences and responses to Ostracism in his Need- Threat-Model described in the following (K. D. Williams, 2007; K. D. Williams, 2009b). According to his model, Ostracism is unique in that it threatens four fundamental needs, namely Belonging, Self-esteem, Control and Meaningful existence (Need to be recognized as existent). There is great experimental evidence for this Ostracism induced threat (K. D. Williams, 2009b). Further Williams distinguishes between three stages of reactions to Ostracism: (a) the reflexive stage, (b) the reflective stage and (c) the resignation stage (Zadro, et al., 2004).

Immediate responses (reflexive stage) to Ostracism trigger threats to the four fundamental needs mentioned above. Additionally the reflexive stage causes negative affects such as Anger and Sadness. Then the short-term reactions begin (reflective stage). The individual’s attention is guided to reflect the meaning and importance of Ostracism and include efforts to retrieve satisfactory levels of the four needs. Persistent Ostracism episodes diminish resources necessary to motivate the individual to reinforce threatened needs (resignation stage). Eventually, long-term consequences such as Helplessness, Unworthiness, Alienation and Depression might follow.

The first step in the entire process is the detection of Ostracism that will be introduced in the next passage.

2.6 The Ostracism Sensitivity Threshold and Clinical Implications

Social animals including human beings seem to have a specific Ostracism detection system, working as an early warning system. This serves to immediately notice social exclusion and hence to adjust ones behavior or to find adequate coping strategies (Gruter & Masters, 1986). Lancaster states that the detection of Ostracism is a fundamental mechanism of an adaptive process of all primates. Baumeister et al. claim that the Need to Belong is a powerful, fundamental, and extremely pervasive motivation (Baumeister & Leary, 1995). Being part of the group allows optimized reproduction and provides access to vital resources (Lancaster, 1986). Consequently a precognitive warning system that facilitates prevention of Ostracism helps the individual to survive.

According to Williams and colleagues the Ostracism Sensitivity Threshold (OST) influences how strong a person perceives being rejected (K. D. Williams, Forgas, J.P., & Hippel, W.V., 2005). Experimental evidence shows that people are sensitive to detect the most minor forms of Ostracism. As mentioned above, studies report that people are able to detect social exclusion even when this happens via the Internet or cell phones (Smith & Williams, 2004; K. D. Williams, et al., 2000). Williams states that this evolutionary adaptive process allows us to either correct our behavior or to search for alternative groups before more harmful effects of rejection take over (K. D. Williams, 2009b). As such this detection system is rather positive as belonging contributes to mental health. Likewise, being over sensitive in perceiving rejection implies negative consequences on mental health (Allen & Badcock, 2003; Downey & Feldman, 1996).

People who are likely to experience Ostracism are characterized by a diminished OST. They tend to overestimate the occurrence of social exclusion events and react in an excessively sensitive manner to it. This detection bias in Ostracism makes the overdetection of social exclusion likely. According to Williams this suggests that the perception of Ostracism occurs even when logically and rationally it should not. It further implies that signals that indicate Ostracism is not meaningful might be ignored or not processed. For example, variations of Cyberball showed that Ostracism was distressing even in the event of a cost applied to inclusion, while exclusion was rewarded (van Beest & Williams, 2006) or when Ostracism was triggered by computer generated players and the participants were aware of it (Zadro, et al., 2004).

Haselton’s and Buss’s Error Management Theory clarifies that such evolutionary adaptive responses are often geared towards biased detection to least threaten the survival of the individual (Haselton & Buss, 2000).

Similar to Williams’s idea of an individually sensitive OST, Downey and Feldman postulate a specific cognitive-affective processing disposition, Rejection Sensitivity (Downey & Feldman, 1996). Rejection sensitivity as a defensive motivational system emerges from repeated rejection and influences and guides appropriate responses towards rejection. According to Downey and Feldman people who are sensitive to social rejection tend to anxiously expect, readily perceive, and overreact to it.

The disposition of rejection sensitivity, in particular prioritizing detecting potential rejection at a cost to other personal and interpersonal goals is associated with features of some mental disorders including e.g. social anxiety or borderline personality disorder (Berenson, et al., 2009). Staebler and colleagues examined social exclusion in borderline personality disorder patients by using the Cyberball paradigm. Their findings implied that people with high rejection sensitivity felt excluded even in the inclusion condition and tended to underestimate the amount of ball passes compared to those with low rejection sensitivity (B. Renneberg, Herm, K., Staebler, K., Hahn, A. Roepke, S., & Lammers, C.- H., 2009).

Likewise there appears to exist a strong link between social rejection and depression. A social risk theory of depression states that substantial Ostracism leads to a hypersensitivity to signals of social threat (Allen & Badcock, 2003). Rather than retrieve satisfactory levels of thwarted needs, chronically rejected people tend to send others signals that they do not wish risky interactions to avoid total exclusion. The same assumption was made for extremly lonely people. Eventually this hypersensitivity leads to helplessness and alienation (Cacioppo & Hawkley, 2005)

This result provides support for the assumption that individual dispositions, cognitive and personality factors might affect the Ostracism Sensitivity Threshold mentioned above. A diminished OST might intensify the perception of one being excluded even when included and eventually lead to a permanent over detection of Ostracism (K. D. Williams, 2009a). The detection and processing of social inclusion is of particular interest in the current study.

2.7 Ostracism and Social Inclusion

The phenomenon of Ostracism is well investigated. However, the question if the feeling of exclusion remains stable if Ostracism is followed by immediate social inclusion has never been considered before. If an Ostracism Detection System actually exists, is it just as sensitive for detecting social inclusion? It is not yet clear whether Ostracism is detected at all, if it is immediately followed by a period of inclusion or if inclusion is rather detected as such. According to Williams Ostracism causes powerful reactions - perhaps it therefore might overshadow subsequent inclusion (K. D. Williams, 2009a).

The retrieval from social exclusion is of crucial importance as it plays a fundamental role in the speed of psychological recovery. As mentioned above, social exclusion is associated with several mental disorders such as depression or borderline personality disorder (Allen & Badcock, 2003; Cacioppo & Hawkley, 2005; B. Renneberg, Herm, K., Staebler, K., Hahn, A. Roepke, S., & Lammers, C.-H., 2009) and automatically leads to cognitive distress (Goodacre & Zadro, 2010; Nadasi, 1992; Smith & Williams, 2004; K. D. Williams, et al., 2000; K. D. Williams, et al., 2002; Zadro, et al., 2004) - the strong link between Ostracism and mental wellbeing that appears to exist raises the idea that social inclusion must be of crucial importance to promote people’s wellbeing and the prevention of mental health risks.

According to Williams participants recover from Ostracism a few seconds after the Cyberball game without debriefing or distraction (K. D. Williams, 2009b). It seems as if participants cognitively cope to the just experienced form of exclusion. Further, if Ostracism happens mistakenly, it is likely for the participants to recover quickly at the point when the mistake is discovered (Zuckerman, Miserandino, & Bernieri, 1983). Nevertheless, individual differences moderate recovery from Ostracism, e.g. studies report that social anxiety or high rejection sensitivity result in problems with selfregulation (Ayduk, Gyurak, & Luerssen, 2008; Oaten, et al., 2008).

But how do people perceive social information such as inclusion in a new and unfamiliar social context? Gardner and colleagues found differences in the perception of a social environment depending on a person’s sense of belonging (Gardner, Pickett, & Brewer, 2000). They reported that people with a decreased sense of belonging, either momentarily or chronically, perceived the social environment in a very different way - more precisely they were characterized by a better memory for social events compared to people who were not experiencing a threat to their belonging needs. Therefore the sensitivity to social information, such as the perception of social inclusion, depends on an individual’s current level of belongingness needs (Gardner, et al., 2000). This promotes the idea that social inclusion perhaps might not specifically be observed or particularly detected since inclusion does not threat the belonging needs.

Similar, research has shown that expectancies towards an event guide one’s interpretation of it (Bruner, 1957). It might be considered, that the degree of one’s concern or the expectancy towards rejection will influence a person’s perception to inclusion. Chronic or automatic expectation of exclusion therefore might lead to an underestimation of inclusion similar to the concept of a diminished OST mentioned above.

The question of how such a detection system works is of particular interest in this study with special attention on the processing of inclusion. The current study is ought to investigate this processing on a neuronal level in a healthy group of people. The current study presents an EEG study further investigating the P300 event-related-potential that refers to index cognitive processing. Event-related-potentials and especially the P300 component are presented in the following chapter.

2.8 Event-Related-Potentials

Event-related-potentials (ERPs) describe electrophysical responses to internal or external stimuli. ERPs become visible in an electroencephalogram (EEG) - a method to measure electrical activity through the scalp. The EEG reflects a vast number of ongoing brain processes as event-related activities appear additionally to spontaneous brain activity, which emerges independently of any stimuli.

illustration not visible in this excerpt

Figure 2-a: frequency spectrum of spontaneous EEG activities. Adapted by Malmivou (Malmivuo & Plonsey, 1995).

ERPs are used to study sensory or cognitive processes in the brain as they reflect electrophysiological correlates of complex information processing. To separate spontaneous brain activity from event-related activity, all EEG signals are usually averaged together within a specific time window after an incoming event. This provides an enhanced signal-to-noise ratio - random brain activity is thereby averaged out and relevant brainwaves with several positive and negative peaks result (Coles, 1996).

One segment of a waveform is called a component. Different components are linked to different information processing stages and different parts in the brain depending on how long after an event the potential lasts in its latency (Birbaumer & Schmidt, 2002). Latency more precisely describes the delay between stimulus-onset and peak amplitude. A component’s amplitude describes the difference between mean pre-stimulus baseline voltage and voltage of the largest peak of the ERP waveform in its specific time window.

illustration not visible in this excerpt

Figure 2-b: schematic illustration of ERP components. The component ’ s label refers to the positivity (P) or negativity (N) of its amplitude and the approximate value of milliseconds it emerges after stimulus onset. Adapted by Birbaumer (Birbaumer & Schmidt, 2002).

In the following the P300 component and its method of investigation will be described as they are of particular interest for the current study.

2.9 P300 Component and the Oddball Paradigm

The P300 was first reported by Sutton and colleagues in 1965 (S. Sutton, Braren, Zubin, & John, 1965). It describes a late positive component peaking around 200ms-700ms after stimulus onset (S. Sutton, et al., 1965). It is evoked by the application of a seldom task- relevant stimulus.

A common method to investigate changes in the P300 component is the O ddball Paradigm. In this paradigm target stimuli with rare occurrence probability are presented within a series of non-target stimuli. Subjects are asked to only react to the target stimuli. The standard stimuli usually do not require any responses. Stimuli are often of auditory or visual nature. First shown by Sutton et al. specific target signals, that occur seldom within a series of standard stimuli and require a special cognitive or motor response (button pressing, counting etc.) elicit an increased P300 component (S. Sutton, et al., 1965). Thus the component’s major function somehow seems to be categorizing and evaluating incoming stimuli. However, its true nature is still the matter of a running debate and remains unknown (Gerloff, 2004).

The P300 component emerges predominantly centro-parietal and is divided into two subcomponents, the P3a and P3b (Gerloff, 2004).

The P3a component, formerly known as the novelty P3, can be found in more central and frontal regions. It reflects the earlier one of the two subcomponents. Its latency lasts between 220ms and 320ms after stimulus onset and is evoked by novel stimuli. The P3a also emerges, even if stimuli targets are not actively perceived. For this reason it is interpreted as a correlate of an automatic orientation reaction to any stimulus (Gerloff, 2004).

The P3b, formerly known as the classic P3 (Gerloff, 2004; Nieuwenhuis, Aston-Jones, & Cohen, 2005), is found especially over parietal brain areas. Its latency varies between 300-700ms after stimulus onset and its amplitude is relatively large with about 10-20 microvolt (Polich, 2007). The P3b component has been associated with the processing of infrequent, more improbable or surprising events. As the current paper focuses on the P3b subcomponent the terms P3b, P3 or P300 will be used synonymously if not otherwise noted. In the following passages previous findings on the nature of the P3b potential are reported.

2.9.1 P3b Latency

P3b latency is suggested to index the duration of mental processes reflected by the component.

According to a study by Sutton and colleagues, P3b latency is determined by the time it takes to resolve uncertainty or ambiguity about the occurrence of a stimulus (Samuel Sutton, Tueting, Zubin, & John, 1967). Another major influence of the P3b latency is task demand. McCarthy and Donchin reported P3b latency to be longer for simpler tasks. This suggested that P3b latency is likely to reflect the time it takes to detect a target (McCarthy & Donchin, 1981). Additional findings associate P3b latency to the duration of stimulus evaluation. Thereby increasing stimulus complexity is associated with longer P3b latencies (Emanuel Donchin & Coles, 1988; Emanuel Donchin & Coles, 1998; Johnson, 1986; Kutas, McCarthy, & Donchin, 1977). Latest findings report a correlation between changes in local context and P3b latency. Context changes, particularly targets turning into non-predictive targets, are linked to enhanced P3b latencies (Fogelson, et al., 2009).

2.9.2 P3b Amplitude

P3b amplitude is suggested to reflect the intensity of mental processes (Kok, 2001). Johnson presented a triarchic model in which he introduced three major variables that influence P300 amplitude (Johnson, 1986).

First there is information transmission defined as the amount of stimulus information received by a person relative to the actual information the stimulus contains. Information transmission is affected by resource allocation (Kok, 2001; Polich, 2007). According to Johnson, reduced information transmission results in smaller P3b amplitudes (Johnson, 1986). Similarly, Kahneman and colleagues found that reduced processing capacity and diminished attentional resources lead to smaller amplitudes. On the other hand, if task conditions require small amounts of attentional resources P300 amplitudes are relatively high with shorter latencies (Kahneman, 1973). Kok therefore assumed the amount of diminished P3b amplitude to be a measure of required processing capacity or working memory (Kok, 2001).

The second major influence of the P300 amplitude is stimulus meaning, which encompasses variables such as stimulus relevance, task demand and stimulus complexity. P3b amplitude increases with stimulus value or relevance to the task (E. Donchin, 1981; Johnson, 1986). Furthermore if stimuli are of significant value (e.g. emotionally valent) larger P3b amplitudes are elicited (Johnston, Miller, & Burleson, 1986). Additionally, findings report that the P3b amplitude decreases with increasing task complexity (Polich, 1987). Similar stimulus complexity affects the P3b component. Stimulus complexity thereby refers to the effort to evaluate or categorize the target. Findings associate increasing stimulus complexity to enhanced P3b amplitudes (Emanuel Donchin & Coles, 1988; Johnson, 1986; Kutas, et al., 1977).

The third major influence of the P3b component is referred to as subjective probability. Subjective probability, however, is a function of many variables including global probability and local probability with respect to the preceding stimulus sequence. Global probability or a priori probability refers to the target’s frequency relative to the number of standard stimuli (Duncan-Johnson & Donchin, 1977). Duncan-Johnson et al. reported that the priori probability of a target stimulus is inversely related to the P300 amplitude. Stimuli that occur less frequently are associated with enhanced amplitudes whereas stimuli occurring more frequently are linked to diminished P3 amplitudes (Duncan-Johnson & Donchin, 1977). Further studies replicated these findings of larger amplitudes for rather improbable stimuli (E. Donchin, 1981; Johnson, 1986). Local probability refers to the likelihood of a target in a specific sequence of stimulus presentation. Duncan-Johnson reported that the sequence of immediately preceding stimuli affected P3b amplitude independently from the targets of global probability (Duncan-Johnson & Donchin, 1977). Particularly larger amplitudes are observed when a non-target stimulus precedes a target stimulus. Subsequent studies replicated findings on this correlation of the number of non-targets foregoing a target stimulus and the P300 amplitude (Curry & Polich, 1992; Ford, Duncan-Johnson, Pfefferbaum, & Kopell, 1982; Craig J. Gonsalvez, et al., 1995; Squires, Wickens, Squires, & Donchin, 1976). Finally, effects of global and local probability on P300 amplitude seem to be mediated by differences in target-to-target interval (C. L. Gonsalvez & Polich, 2002).

Several theories have been put forward on the cognitive processes that the P3b is ought to reflect. Squires added the variables global probability, local probability and stimulus relevance to one broader factor of expectancy (Squires, et al., 1976). He developed a quantitative model in which waveform changes are related to stimulus expectancy changes. If a repetitive pattern of stimuli is disrupted and therefore expectations elicited by the stimulus history are broken, enhanced P3b amplitudes result (Squires, et al., 1976). Donchin (1981) stated that the P300 component predicts the memorability of events (E. Donchin, 1981). He brought up a concept of surprise, similar to Donchin and Coles Context-Updating-Theory. According to the Context Updating Theory, processes start updating stimulus representation after each stimulus, whereas larger amplitudes of the P300 indicate greater changes of context. The P300 amplitude is therefore suggested to index activity that is needed to update mental representations of stimulus context (Emanuel Donchin & Coles, 1988).

So far, studies on P300 showed two major results implying that the P300 positivity occurs when stimuli are task-relevant and further that amplitude behavior is inversely related to a targets probability. Consequently, the P300 is not associated with the processing of physical characteristics of a stimulus, but is linked to psychological and cognitive processing. The current study focuses on the P300 behavior during a modification of Williams Cyberball game (K. D. Williams & Jarvis, 2006). During the game the participant will continuously be involved, by increasing the frequency of ball possessions. Of special interest will be whether the P300 component reflects the increase of game participation and therefore can be seen as being associated with the processing of social inclusion or not. The precise method of investigation will be described in the following passage.

2.10 A Social Oddball Paradigm

A common method of investigating changes in the P300 component is the O ddball Paradigm (see 2.5). There are some variations of the Oddball paradigm, such as Waltzing Oddball2 introduced by Verleger and Berg (Verleger & Berg, 1991). Through this paradigm, Verleger and Berg assessed the effects of probability and task relevance of target tones on the P300 ERP. Amplitudes evoked by a standard Oddball paradigm were compared to those evoked using the Waltzing Oddball paradigm. Findings suggested that a parietal P3 was not only affected by the probability and task relevance of target stimuli but also by the frequency of occurrence of the stimuli, more precisely the sequence of tones. Another variation of the oddball paradigm to elicit P300 effects would be the single stimulus paradigm (Polich & Margala, 1997). The single stimulus paradigm presents target stimuli but no standard stimuli. Thereby target stimulus probability is manipulated by inter-target-intervals. Polich and Margala investigated effects to target probability on the P300 components using an oddball task compared to using a single stimulus paradigm. P300 from the single-stimulus paradigm produced almost identical probability effects as those observed from the oddball task suggesting that the single stimulus task appears to elicit P300 components in the same fashion as the traditional oddball paradigm does (Polich & Margala, 1997). Thus, different modifications of the Oddball paradigm provided detailed insights and predictions of the determination of the P300 properties.

However, previous studies only investigated P300 changes using simple sensory stimuli of visual or auditory nature. The current study focuses on P300 amplitude as a response to more complex stimuli, namely of socially relevant quality. The present study presents a modification of William’s Cyberball game (K. D. Williams & Jarvis, 2006) which resembles the characteristics of the classic Oddball-Paradigm. The socially relevant target, however, refers to a ball pass to the participant. Previous studies of our research group (publication in prep.) have shown that this variation elicits the P300 effect. However the present study ought to replicate the P300 effect for a socially relevant event, based on a single trial analysis. Within continuous inclusion in the Cyberball game, local probability of the target increases and thus should result in inverse behavior of the P3b amplitude (Duncan-Johnson & Donchin, 1977). A model of probability calculations that accounts for P3b amplitude changes could eventually provide access to the electrophysical processing of social inclusion.

2.11 Single-Trial Analysis and the P300

For a better understanding of the coupling between evoked potentials and ongoing brain processes, the current study conducts a Single Trial Analysis. Using this method amplitudes and latencies of ERP components are extracted, and each trial is analyzed separately.

Subsequent studies have shown that simply averaging EEG signals leads to considerable loss in the amount of information (T. P. Jung, et al., 2001; S. Makeig, et al., 2002). Any changes in response amplitude, latency, or phase characteristics from trial to trial are suppressed after conventional averaging (Lange, Pratt, & Inbar, 1997). Thus, Single Trial Analysis seems more adequate for studying the dynamics of brain activation. So far several studies conducted Single Trial Analysis in order to investigate trial-by-trial fluctuations of the P300 component - with partly ambiguous results.

Mars et al. investigated how stimulus probabilities are learned trial-by-trial rather than on average. A quantitative model of surprise accounted for ongoing changes in the P300 component (Mars, et al., 2008). Holm et al. examined the role of reaction time and preceding stimulus sequence in P300 trial-by-trial variations. Analyzing single trial P300 signals showed that the number of preceding standard stimuli significantly affected latency, whereas they did not affect amplitude. The conventional averaging method did not capture such dynamics in latency variations and overestimated the influence of preceding stimulus sequences on amplitudes (Holm, Ranta-aho, Sallinen, Karjalainen, & Muller, 2006). Bonala et al. conducted a Single Trial Analysis on P300 data in order to investigate local target probability-induced amplitude changes (Bonala, Boutros, & Jansen, 2008). Stimulus sequences only affected P300 amplitude as such, if less than three non-targets preceded a target stimulus. However, for more than two non-targets preceding a target stimulus the ability of producing a P300 was reduced.

These studies provide evidence that trial-by-trial features of stimulus sequences might influence the P300 component, whereas fluctuations of amplitude and latency are only obtained using Single Trial Analysis. Looking into the dynamic structure of cognitive functions, Single Trial Analysis may provide access to a better understanding of different processes and ongoing changes of the brain.

2.12 Development of the Single Trial Analysis

2.12.1 Extraction of the Component of Interest

Typically, the ERP signals are not readily apparent in the EEG record and special procedures are necessary to detect them. Common signal detection techniques are e.g. peak picking or cross-correlation (Gratton, Kramer, Coles, & Donchin, 1989; Smulders, Kenemans, & Kok, 1994).

Peak picking methods extract values of maximum amplitudes in a predefined time window. To estimate single trial amplitudes, Mars et al. created averages for each participant. P300 estimates were then extracted 60ms around the time point of maximum peak of the average (Mars, et al., 2008).

Alternatively a template of the expected response waveform can be computed. To determine the presence of similar responses, the acquired template is cross-correlated to single-trial ERPs for all time points in which the component is expected. The point of maximum correlation is taken as the amplitude value of the component (Gratton, et al., 1989). A main advantage of this approach is that only signals are detected which resemble the expected shape of the component. This again reduces the risk to extract simple background noise (Arpaia, Isenhart, & Sandman, 1989). Templates of, for example the average component can be created - underlying the assumption that brain responses appear in a consistent fashion that is captured to a certain extent by averaging. Bagshaw et al. created the average P300 component and cross-correlated this template with each single trial signal. If signals passed the correlation criteria, they were extracted as P300 signal (Bagshaw & Warbrick, 2007). According to Gratton, the signal detection technique of cross-correlation seems to be a reliable and reasonable method of estimating single trial responses (Gratton, et al., 1989).

However, one common limitation of deriving information from EEG records remains, i.e. the temporal and spatial overlap of scalp potentials. Such a spatiotemporal overlap makes it difficult to precisely measure ERP components (Groppe, 2008). Various preparatory techniques have been developed to provide more accurate ERP estimates, such as Wavelet Denoising or Independent Component Analysis. In the following section the Independent Component Analysis will be described briefly.

2.12.2 Independent Component Analysis

One popular method of extracting ERP signals is the Independent Component Analysis (ICA). ICA - as a particular method of blind source separation (BSS)3 - is a method that separates a set of mixed signals into its individual source signals. The underlying assumption of BSS and therefore ICA is, that source signals are based on independent physical processes and therefore are statistically independent from each other. The principle of ICA is commonly illustrated with the help of the Cocktail Party Problem: several people are talking simultaneously and a few microphones record the speech signals. The recording of each microphone is an example of a mixture of the independent voices. An important aspect to be considered is, that the number of estimated source signals is identical with the number of different mixtures of a set of source signals - analogous to a factor analysis (Stone, 2005).

The cocktail party problem also applies to EEG recordings. At each electrode location, the recorded signal can be considered as a neuronal mixture of underlying independent components.

illustration not visible in this excerpt

Figure 2-c: main procedure of the Independent Component Analysis illustrated with the Cocktail Party Phenomenon. People (source signals) speak into microphones (x/sensors) and a non-linear algorithm (W) unmixes signals to recover source signals (u). Adapted by Mouraux (Mouraux, 2008).

The objective of ICA is to find a matrix that allows the recovery of the original source signals. The signals are separated, so that statistical independence is maximized. Independence is thereby defined as the minimization of mutual information and the maximization of non-Gaussianity as a measure of statistical independence (Stone, 2005).

In general the ICA algorithms are designed in order to determine which temporally independent and spatially fixed activations make up a time-varying response. It follows the assumption that the summation of electrical activities that emerge from different brain areas is linear. Therefore ERPs are decomposed into a linear sum of activations of independent sources. Thus the time-course of the sources is maximally independent and scalp distribution is spatially fixed.

illustration not visible in this excerpt

Figure 2-d: shows main procedure of Independent Component Analysis analogous to a factor analysis. Adapted from Makeig (Scott Makeig, Jung, Bell, Ghahremani, & Sejnowski, 1997).

There are several ICA algorithms to identify independent sources and efficiently eliminate spontaneous EEG or ocular artifacts. These are spatial algorithms that use contrast functions to measure spatial independence. Temporal ICA algorithms however use temporal independencies to achieve source separation. Spatio-temporal algorithms use both, spatial and temporal features for source estimations (Gómez-Herrero, 2005).

One benefit of ICA is that it offers a cleaner and less ambiguous measure of ERPs (Groppe, 2008). Thus ICA reveals interesting information on brain activity by providing access to its independent components.

illustration not visible in this excerpt

Figure 2-e: mixed measured signals are separated into clean independent components through the ICA algorithm. Adapted by Jung (A. Jung, 2008).

Components can either be rejected (such as ocular movements) or extracted, as in the current study the P300 component. Independent Component Analysis and the P300

Debener and colleagues investigated the role of task-relevance on the novelty P3. Of special interest was, whether a spatiotemporally overlapping P3b feature could account for the predicted amplitude of the novelty P3 in response to target as compared to nontarget novel stimuli. Applying the Independent Component Analysis, using the INFOMAX algorithm4 data-analysis of the IC’s in comparison to initial analysis of scalp ERP revealed that amplitude height in novelty P3 could partially be explained due to the spatiotemporal overlap of novelty P3 and P3b. The novelty P3 was elicited by both task- relevant and task-irrelevant stimuli, in contrast to findings of the initial ERP analysis. Thus, task irrelevance was no longer interpreted as a necessary feature of stimuli to elicit a novelty P3. Therefore ICA applied to single-trial EEG data could successfully decompose the spatiotemporally overlapping ERPs of novelty P3 and P3b into a range of underlying EEG processes.

This finding underlines the problem of spatiotemporally overlapping ERP signals and indicates the usefulness of the ICA (Debener, Makeig, Delorme, & Engel, 2005).

Due to the growing interest for ICA applications, the current paper aims to evaluate the benefits of this method. The FastICA algorithm was chosen to perform source separation. Hyvarinen established this algorithm and its function is based on a spatial algorithm derivative on a non-quadratic non-linearity (Hyvarinen, 1999). The FastICA is a common method to identify artifacts and interference from mixed signals. Great advantages of the FastICA algorithm in particular are its speed and its simplicity to use. Several simulations as well as applications on real-life data have validated the contrast function and algorithm and its application offered satisfactory performance (Zarzoso & Comon, 2006). For example the FastICA algorithm has been compared favorably with neural-based adaptive algorithms and principal component analysis (PCA), whereas most ICA algorithms were found to outperform (Zarzoso & Comon, 2006). Li and colleagues applied the FastICA algorithm to EEG data and found 76.67% accuracy on single trial P300 response identification (Li, Sankar, Arbel, & Donchin, 2009). This gave evidence that the ICA algorithm can accurately separate independent components from noisy signals. Therefore it seems reasonable to apply the FastICA algorithm on EEG signals to identify the P300 component and to extract single-trial signals.

However, the validation of the resulting components remains a primary issue since the true ERP is unknown. This paper aims to compare and evaluate performances of two single trial approaches, namely Independent Component Analysis as a preparatory method and a single trial signal detection in the raw data.

2.13 Aim of the Current Study

The current paper aims to investigate the behavior of the P300 component on a single trial basis when the probability of a social target event changes unexpectedly.

Of special interest are two different approaches of the Single Trial Analysis and their advantages or disadvantages. The data will be analyzed in the following way: Once the raw data will be used to extract P300 signals (non-ICA corrected data) and once the FastICA algorithm will be applied on the dataset (ICA corrected data) to eventually extract P300 components. The study is meant to evaluate how additional data correction using the ICA provides more information compared to analyzing the raw data.

The methodical approach will be conducted as follows: based on the signal detection technique evaluated by Gratton this study creates an average template that will be crosscorrelated with each single trial separately (Template Matching Method) (Bagshaw & Warbrick, 2007; Gratton, et al., 1989). To this aim results of single trial estimates of the non-ICA corrected data and ICA corrected data will be compared.

Although P300 amplitudes will be of particular interest, the current study also takes into account the analysis of latencies - primarily to gain further validation of the Template Matching Method and due to the comparison of the performances of the two Single Trial Analysis approaches. The hypothesis will be verified or denied for each approach separately. Finally, the approaches will be compared and evaluated according to their performances.

According to Williams’ Cyberball studies mentioned above, the investigators of the current study assume that with continuous inclusion the subject will register the increasing participation in the game over time. Concerning Donchins Context-Updating- Theory (Emanuel Donchin & Coles, 1988) and Squires model of expectation, the subject will adjust his/her expectancy towards the event when he/she gets the ball. Over the course of time this event will occur more frequently. Because target probability and P300 are inversely related, it will be expected that P300 amplitudes decrease over time as the frequency of the relevant event increases. Hence, significant continuous changes in the P300 amplitudes over the course of time are suggested and a linear function to fit for P300 amplitude changes of each participant. As mentioned above, the P300 amplitude is especially affected by relevant events. Thus only significant effects for the “Self”-events (participant gets the ball) are expected. Changes in amplitude or latency for “Other”- events (other players have the ball) are eliminated from the analysis.

Finally, the relationship between the P300 amplitude and local target probability is of particular interest. A model of local target probability will be proposed to investigate its relation to single trial P300 amplitude changes. Since the P300 amplitude and local target probability are inversely related, the correlation is expected to be negative (DuncanJohnson & Donchin, 1977).


1 The term “Ostracism” refers to an Athenian democratic process in which citizens were expelled from the city for ten years. Williams revisited the term of ostracism and uses it synonymously to the terms rejection and social exclusion, as we will in the current paper if not otherwise noticed.

2 Waltzing Oddball: high tones are the target stimuli but only those that are preceded by two other high tones.

3 Blind Source Separation is the separation of mixed signals into a set of signals with only very little information about either the respective source-signals or the mixing process. The major assumption of BSS is that source signals are uncorrelated. Procedures of BSS encompass e.g. Principal Component Analysis, Independent Component Analysis and Dependent Component Analysis.

4 The INFOMAX algorithm exploits temporal independence for BSS, and reveals spatial fixed and temporally independent components


ISBN (eBook)
ISBN (Book)
File size
4.8 MB
Catalog Number
Institution / College
Free University of Berlin
Cyberball Ostracism P300 ICA Inclusion Exclusion Single Trial Independent Component Analysis Odball



Title: Effects of continuous inclusion in a virtually played ball tossing game called Cyberball on the P300 event-related-potential