Microstate analysis of monastic debate
Bachelor’s Project Thesis
Jelmer Bot, s2568551, d.m.bot@student.rug.nl, Supervisor: Dr M.K. van Vugt
Abstract: Debate is an essential component of Tibetan scholastic training. This practice is notable in that it has a specific physical form, and is focused on finding out contradictions rather than convincing the opponent. This study uses a microstate analysis to investigate the development of attention in a debate and explores the differences between the roles of debate.
Given that attentional requirements may increase during the debate, a decrease in the prevalence of microstate class D (which is thought to reflect the reflexive and reorienting aspects of attention) was expected. We did not observe any change in the prevalence of microstate class D. We did observe differences in the prevalence of microstate classes between the different roles in a debate.
However, this effect may be confounded with the postures of the two debaters.
1 Introduction
Debate is an essential component of Tibetan scholastic training, it is used to develop rational and critical thinking (Dreyfus, 2008). A debate can take many forms, however, there is always at least one defender and one challenger. A debate starts when all debaters agree on the topic of the debate.
The goal of the challenger is to force the defender to contradict himself. This is achieved by propos- ing consequences of the defender’s position. A con- sequence typically consists of a claim and a rea- son: “It follows that ... because...”. The truth of the claim of a consequence does not matter, the chal- lenger commits only to the entailment of the con- sequence from the statements previously accepted by the defender.
The defender is only allowed to give one of three responses to a consequence: I accept, the reason is not established, or there is no pervasion (Drey- fus, 2008). Accepting a consequence means that the defender thinks the consequences supports his po- sition. The defender may say that the reason is not established when the reason of a consequence is not true. There is no pervasion when the reason does not entail the claim of a consequence.
This structure of debate forces the challenger to break down his arguments into single consequences and to keep track of the defender’s position. The defender on the other hand has to figure out the strategy of the challenger and try counter it. The debate ends when the defender contradicts himself
or when the challenger is unable to give a valid consequence.
Scientific research on Tibetan debate is scarce, studies that have been performed are mostly an- thropological or philosophical in nature. Instead, this study investigates the cognitive and neural as- pects of debate, exploring what happens in the brain of a debater.
Measuring brain activity during a debate is chal- lenging due to the physical nature of debate. In order to hinder the debaters as little as possible EEG measurements are used. Such measurements are typically analysed in terms of event related po- tentials. However, the debates are conducted in Ti- betan so events depending on the contents of the debate can only be recorded with the help of an observing monk. Besides, we do not understand enough of the debating practice to define mean- ingful events.
Instead, we analysed the topographical configu- rations in the data. This approach is data driven and does not rely on specific time points to time lock on. However, it is able to find out what kinds of neural patterns occur, which may give us clues about the cognitive-emotional processes that are used in debate.
Topographical configurations in EEG data de- velop discontinuously, remaining stable for 80 to 120ms and then changing rapidly to another stable configuration (Lehmann, Strik, Henggeler, Koenig, and Koukkou, 1998). Time periods in which config- urations remain stable are called microstates and
are thought to reflect distinct cognitive states. Mi- crostate analyses have consistently identified four classes of topographies in a wide range of tasks, labelled A through D (Milz, Faber, Lehmann, Koenig, Kochi, and Pascual-Marqui, 2016). Koenig, Lehmann, Merlo, Kochi, Hell, and Koukkou (1999) found that four microstate classes optimally ex- plained the variance in the topographies in their data using the cross-validation index described in Pascual-Marqui, Michel, and Lehmann (1995).
Since then the four classes have become the stan- dard (Milz et al., 2016).
The microstate class topographies as found in Koenig et al. (2002) are shown in fig. 1.1. They are characterized as follows: Class A has an orien- tation from left occipital to right frontal. Class B’s orientation is opposite, from right occipital to left frontal. Class C is oriented symmetrically from oc- cipital to prefrontal. Finally, class D’s orientation is also symmetrical from frontocentral to occipital.
Koenig et al. (2002) argues that the topogra- phy of EEG measurements are the result of the activation of underlying networks. So topograph- ical changes are the result of a change in the func-
Figure 1.1: Microstate topographies per class from Koenig et al. (2002) as seen from above.
(nose is up)
tional state of the brain. Currently the functional aspects of microstates are being studied. In a study by Lehmann et al. (1998), subjects were asked to report on their thoughts when prompted. The re- ports were classified as ‘abstract thought’ or ‘visual imagery’. They discovered that microstate class A was present just before the subjects reported hav- ing an ‘abstract thought’. Microstate class B was present before subjects reported being engaged in
‘visual imagery’. The same relation was found by Britz, Van De Ville, and Michel (2010). Further- more, they related microstate class C to a resting state network that is thought to play a role in form- ing a subjective representation of the own body.
Microstate class D was found to be correlated with reflexive and reorienting aspects of attention.
The relation between microstate class D and at- tention is further supported by research on the change in microstate class parameters in several conditions. Usually the coverage, duration and oc- currence of microstate classes are analysed. Cov- erage is the percentage of time all instances of a microstate class cover within a time. Duration is the mean length of all instances within a mi- crostate class in a time window. Occurrence refers to the number of instances of a microstate class in a time window. For instance the studies performed by Brodbeck, Kuhn, von Wegner, Morzelewski, Tagli- azucchi, Borisov, Michel, and Laufs (2012) and Katayama, Gianotti, Isotani, Faber, Sasada, Ki- noshita, and Lehmann (2007) who found a decrease in all parameters of microstate class D caused by sleep and hypnosis. The same effects are also found in patients diagnosed with schizophrenia (Koenig et al., 1999). Meditation on the other hand in- creases the duration and occurrence of microstate class D (Milz et al., 2016).
Other research has investigated the differences in microstate parameters caused by ageing. Four de- velopmental stages have been identified by Koenig et al. (2002). In general, the duration of microstates decreases and occurrence increases as one gets older. Microstate class C, however, is character- ized by a peak in coverage in the third develop- mental stage. The first stage, 6 to 12 years, con- tains only small differences between classes and rel- atively high durations. In the second stage, 12-16 years, the prevalence of class C increases whereas classes A and B become less prevalent. Then, in the third stage, 12-21 years, the prevalence of class
D decreases to similar values as for classes A and B. Class C becomes even more prevalent. In the final stage, 21+ years, the prevalence of classes C and B decrease and classes A and B become more prevalent.
In this paper the role of attention in Tibetan de- bate is analysed using a microstate analysis, which could help to generate hypotheses about the types of cognitive processes that may occur during this practice. Furthermore the differences between chal- lengers and defenders are analysed, in order to explore the cognitive differences between the two roles.
The prevalence of microstate class D is analysed over time within a debate in the former analysis.
We expect that the reflexive and reorienting aspects of attention become less prominent as a debate un- folds because the defender and the challenger con- stricted by the previous statements. Additionally some debaters have reported that they lose track of the world around them and are completely focussed on the debate. So given the functional relations of microstate class D (Britz et al., 2010; Koenig et al., 1999), we predict that the coverage of microstate class D decreases over time.
The last analysis aims to explore cognitive dif- ferences between defenders and challengers. Chal- lengers and defenders have distinct roles in a de- bate. Defenders have to maintain a consistent point, whereas challengers have to look at a prob- lem from all sides in order to find contradictions.
So we expect to find differences in the prevalence microstate classes but we do not have a prediction on the nature of these differences.
2 Methods
The recordings taken in 9 debates are used in this study. 2 participants were recorded in every debate.
Both a challenger and a defender were recorded in 5 debates, 2 defenders were recorded in 2 debates and 2 challengers were recorded in the final 2 de- bates. The debates took place in the science cen- tre of the Sera Jey institute in Bylakuppe, India.
7 monks were recorded, aged between 30 and 36 years old. Most participants were recorded in mul- tiple debates. All monks were highly experienced in debate.
The debates lasted between 9 and 17 minutes.
The defenders were always seated on a rug on the floor. Challengers were standing and moving around, as is typical in a debate (Dreyfus, 2008).
The challengers tried to force the defenders to con- tradict themselves by proposing consequences of the defender’s statements. One monk observed the debates, pressing a button when a notable event occurred, for instance when the debaters were in agreement or disagreement. These triggers are not used in the present study.
The EEG data was recorded with 32 electrodes per participant arranged according to the 10-10 sys- tem with a sampling rate of 500Hz. The record- ing was referenced to the mastoides. actiCAP elec- trodes were used (BrainProducts GmbH, Munich, Germany). actiCap uses active electrodes that am- plify the signal before it is sent to the recorder which reduces the interference of movement by par- ticipants.
2.1 Data processing
Data pre-processing was performed in FieldTrip (Oostenveld, Fries, Maris, and Schoffelen, 2011).
The data were rereferenced to the average reference of every participant. Power line interference was re- moved by applying a low pass filter at 50Hz. Other artifacts were corrected using an independent com- ponent analysis, components to be removed were selected by eye. The data were then down-sampled to 256Hz to reduce computation costs, FTT filtered from 2-20Hz without windowing and segmented into 2 second epochs.
The microstate analysis was performed using the microstate package of the KEY EEG Analysis Tool- box (Milz et al., 2016), which follows the proce- dure used by Koenig et al. (1999). Topographies at global field power peaks were identified and used in a modified k-means clustering algorithm, as de- scribed by Pascual-Marqui et al. (1995). The clus- tering algorithm determines four classes that max- imally explain the variance of the topographies.
These classes were then used to compute mean classes across participants and across groups. A group with the recordings of all challengers and a group with all defenders were used. Mean to- pographies were computed using a full permutation procedure that determines the solution of maximal mean correlation across the four classes. The means were computed as the first principle component of
all maps assigned to each other in the solution.
Microstate parameters were computed for each epoch by assigning the topographies of global field power peaks to one of the classes based on maxi- mal Pearson correlation. Microstates were merged when successive global field power peaks were as- signed the same class. The length of each microstate was computed as the time covered by the merged microstates, starting at the midpoint between the last map of the preceding microstate and the first map of the current microstate and ending at the midpoint between the last map of the current mi- crostate and the first map of the following mi- crostate. The start and end point of a microstate cannot be determined for the first and last mi- crostate in an epoch, so these microstates were ig- nored in the parameter computations.
The length of each microstate was used to com- pute the parameters for each class in every epoch.
Coverage is defined as the percentage of time spent in a microstate class. Duration is defined as the mean length of all the microstates that belong to a class. Occurrence is defined as the number of oc- currences per class in an epoch.
2.2 Analysis
The coverage of microstate class D was tested for a linear regression in order to test our hypothe- sis about the role of attention in a debate. A de- crease in the prevalence of microstate class D was expected, which relates to decrease in the reorient- ing of attention. Additional models were fitted per debater to analyse individual differences.
Time was defined according formula 2.1, in or- der to compare the linear regression models across debates because debates have different lengths.
T ime = N umber of the epoch
T otal epochs in the debate (2.1) The parameters of all microstate classes were also compared between challengers and defenders to analyse the differences in prevalence of microstates between the roles. Two-way ANOVAs were used to test the influence of the role and microstate class for every parameter. A post-hoc Tukey Honest Sig- nificant Differences test determined which combi- nations actually differed.
All statistics were calculated in R version 3.0.2 (R Core Team, 2013).
3 Results
The microstate analysis was able to explain 53 ± 1.2% of the total variance of all 2 second epochs and 55 ± 1.4% of the total variance of all GFP peaks.
The resulting class topographies are shown in fig.
3.1 and are labelled as in Koenig et al. (2002).
Microstate class A is associated with ‘abstract thought’, class B with ‘visual imagery’, class C with the representation of one’s own body and class D with the reflexive and reorienting aspects of atten- tion.
3.1 Attention over time
The coverage of microstate class D did not show a significant linear regression. The linear regres- sion model had an intercept of 22% (p < .001).
The coverage decreased by 0.56% (p = 0.37) across a debate. The overall fit of the model was R2 = 9.73 × 10−5
Figure 3.1: Microstate topographies per class as seen from above. (nose is up)
Linear regression models were also fitted per par- ticipant to analyse individual differences. The sig- nificant models are visualized in fig. 3.2. Debaters 2 (9.0%), 6 (8.3%) and 11 (9.8%) had an increasing coverage for microstate class D, whereas debaters 9 (-16.6%), 13 (-19.7%), 16 (-5.2%) and 17 (-2.8%) had a decreasing coverage.
3.2 Coverage by debate role
We then asked whether the prevalence of microstate classes differed between challengers and defenders.
Whereas defenders are focused on maintaining a consistent position, challengers are focused on look- ing at the issue from many different sides and try- ing to find inconsistencies. The effect of debate role and microstate class were analysed using two-way ANOVAs. Mean coverage values per debate role and microstate class are shown in fig. 3.3.
The two-way ANOVA found all effects to be statistically significant at the .05 level except for debate role. The main effect for microstate class yielded an F ratio of F (3, 33072) = 5165.6, p <
.001, class C was the most prevalent (M = 45, SD = 23), followed by class B (M = 22, SD = 17) and class D (M = 22, SD = 16), while class A was least common (M = 11, SD = 14). The main effect for debate role yielded an F ratio of F (1, 33072) = 0.13, p = 0.72, indicating that the effect for debate role was not significant, defend- ers (M = 25, SD = 21) and challengers (M = 25, SD = 22). The interaction effect was significant, F (3, 33072) = 174.6, p < .001.
A post-hoc Tukey Honest Significant Differences test was used to test the effects of the interac- tions. Class C was significantly more prevalent in defenders (M = 78, SD = 22) than in challengers (M = 41, SD = 24) p < .001. There was no sig- nificant difference in the coverage of class D be- tween challengers (M = 22, SD = 17) and defend- ers (M = 23, SD = 16) p = .061. Class B was more prevalent in challengers (M = 25, SD = 18) as in defenders (M = 20, SD = 15) (p < .001).
The coverage of class A was higher for challengers (M = 12, SD = 15) compared to defenders (M = 10, SD = 14) p < .001.
A B C D
Microstate Class
Mean coverage (in %)
0 20 50
Challengers Defenders
Figure 3.3: Mean coverage in percent by mi- crostate class and debate role. Coverage is the percentage of an epoch that is covered by all instances of a microstate class combined.
4 Discussion
This study sought to analyse the role of attention in monastic debate, as well as the cognitive differ- ences in the roles of a debate. The linear regression analysis indicated that there is no significant de- crease in the prevalence of microstate class D over time. Microstate class D is associated with the re- flexive and reorienting aspects of attention (Britz et al., 2010).
However, significant differences in the prevalence of microstate classes A, B and C were found be- tween challengers and defenders. These findings are visualized in fig. 3.3. Microstate class A is associ- ated with ‘abstract thought’, class B with ‘visual imagery’ and class C with the representation of the own body.
4.1 Attention over time
We hypothesised that the coverage of microstate class D would decrease over time in a debate be- cause microstate class D is associated with the re- flexive and reorienting aspects of attention (Britz et al., 2010). As a debate unfolds more and more statements need to be remembered in order to pre-
