Modeling Political Belief and Its Propagation, with Malaysia as a Driving Context

We discuss in this paper an agent-based simulation model that describes the propagation of political belief in Malaysia. Worldview map is used as the representational scheme for political belief. Inter-agent interaction propagates the belief throughout the agent population, subject to similarity of emotion between the interacting agents and their distances apart, and various attributes of the individual agents. Media broadcast may be used by agents in their attempt to extend their reach. Computational experiments made using the model point to its plausibility. Further, it highlights, for the ruling coalition, the importance of both a strong political propaganda machinery and a strong governance in winning the hearts and minds of the electorate.

resource competition in this paper, though such emphasis might be very relevant when considering Malaysian politics in a broader context. And as far as case studies are concerned, only a few work thus far can be classified as being under this category. Geller et al (2011) focused on the socioeconomic issues to explain how the Afghan drug industry was sustained by its political players. Gulden (2012) applied a game-theoretic framework to study the cycle of violence in the Guatemalan Civil War, and showed, through simulations, the relationship between ethnic composition and the level of political violence in a specific zone. The work as presented in this paper, being very specific to the Malaysian political experience, may be categorized as a case study. It is important to note that while a case study might not directly contribute to the fundamentals of agent-based simulation or to politics itself, it serve to demonstrate the applicability or feasibility of using simulation to study a real-world problem, illustrating its usefulness, and possibly guiding further fundamental research.

Framework
The essence of our approach is depicted in Figure 1. Agents interact with one another, and each has a number of features, pertinent among which are emotion, communication skill and influence, stubbornness, and opinions structured in the form of a worldview or mental map. Multiple opinions persists in an agent, an emotional filter determining the one which is salient. Agents interact in order to propagate its belief, locally and at times globally using Internet broadcast. More details will be exposed in the sections to follow.

Worldview map
We model the worldview of individuals in a simulated society using graphical network notation. A worldview depicts a person's thought process about how something works in the real world. It represents the surrounding world, and the relationships between its various parts. In the context of this paper, it serves a purpose similar to the cultural network analysis (CNA) in Sieck (2010Sieck ( , 2011) -it facilitates studying in detail the common perspective that members of a culture bring to a situation. The specific form of the worldview that we assume are that which we consider to be the most common in the Malaysian political context, and will be presented later. We represent a worldview map as an N-node tree. Each node, ni, in the map represents a notion or statement. Each edge, {nk, nl}, connecting between 2 nodes, nk and nl (0 ≤ k < N and 0 ≤ l < N), represents a link between two notions. It has a weight, w kl , 0 ≤ w kl ≤ 1 ,that represents the strength of the connection between the concepts. A stronger weight denotes a strong argument for the connection, while a weak weight denotes otherwise. The overall weight of a map is obtained as the average over all the weights in the map, as follows: where W is the total number of edges or weights in the map. Political conflict ensues when different sides, having different worldview maps pertaining to an issue, communicate past one another. This means that two people can experience the same situation, but pay attention to entirely different things and take away different ideas about the causes and effects of an experience. In the context of Malaysian politics, it explains the continual dichotomy between opposing sections of society viii . A network notation has two main advantages. First, it allows us to map out the thought systems of the opposing parties, allowing us to objectively understand the position of each. Secondly, the network notation, in computing terms, translates to standard tree data structure, allowing for efficient computation and simulation, enabling understanding of the macroscopic behavior that emerge in society when ideas interact.

Worldviews Assumed 3.2.1 Disclaimer and Scope
The author asserts that this paper is written with no intentional bias towards either the ruling party or the opposition in Malaysia. The goal is purely academic -to develop social simulation for the study of socio-political reality in the country and to infer insights into her political development. The maps to be shown in the next sections were derived from various sources (mostly online) and the author's personal observation in months leading to GE13, and up to a few months upon its conclusion. The author made observations through his interactions with various individuals (including family members and colleagues) with different political orientations, participation in online forums and debates, and years of following the Malaysian political scenes. Hence, the maps are illustrative in purpose, its method of derivation may be termed political ethnography, akin to the treatment in (Weiss 2013). A more rigorous approach, one which would be needed for an 'outsider' to understand a foreign society, would be something akin to Sieck's cultural map analysis, a method that incorporates eliciting of ideas from people in interviews, extraction of ideas from interview transcripts, analysis of common ideas and alignment and assembling of common ideas into complete maps.

The PR Worldview
The worldview map held by a 'typical' PR supporter is, based on the author's observation, as depicted in Figure 2. The primary points in this world view are that BN rules are marked by the following: 1. Poor governance a. Crime; Crime -robbery, snatch theft -has been reaching an epidemic level, especially in the cities ix . The police has been weak and ineffective, being unable to do anything much about this.
b. Middle-class trap; Government policies has caused large brain drain. And due to the brain drain, b. Under BN, cronyism has been rampant, and the police force corrupted xiv . c. BN has been too tainted by corruption scandals. Anti-corruption agencies have been acted mainly on small fish, leaving the big ones swimming free. Even if there are cases involving big names, it is all just for show, not a true indication of the government's seriousness in combating corruption xv .   Lee 2010). Note that the dichotomy between the two views, consequential when an Islamist side partners with a secular side, is a known weak spot for the opposition xxi .
8. Election Committee (EC) fraud: Various complaints of cheating have been aired -intentional blackouts allowing for change or addition of ballot boxes, dubious electoral rolls, electorates carrying false IC, failure of supposedly indelible ink, and phantom voters (voters brought in from other district to vote for the ruling party in a particular district) xxii . In fact, the PR has been sophisticated in reinforcing the worldview in Figure 2 through the manipulation of affect, the political role of which has been well-documented (Glaser andSalovey 1998, Brader andCorrigan 2006). PR succeeded in deploying catchphrases that sum the essence of their drive. The message was conjured to stir up visceral association with its cause in fighting an 'unjust' and 'incompetent' government.

The BN Worldview
In the mind of a typical BN supporter, the dominant worldview is as shown in Figure 3. The primary points in this worldview are that BN should continue to rule as:

Development in Malaysia, under BN, is directed towards Vision 2020, with the Economic Transformation
Programme as the primary driver. Further, the government has made it a point to fulfill whatever promise, social and such, made to the people (Najib's "Janji Ditepati" or "Promises Fulfilled" slogan) xxvi . 6. UMNO offers a modernist form of Islam, that counters PAS's radicalism xxvii . 7. The government acknowledges that there have been mistakes before, in the form of scandals and such committed by BN politicians and ministers. However, the government has been taking actions to correct the mistakes xxviii .
8. The BN government through its 1Malaysia social initiative, has been reaching out to all regardless of race, color and creed. Through nationwide initiatives such as B1RM, the government has shown that it cares for all xxix . Further, arguments have been put forth as to why PR cannot or should not be allowed to take over the government: 9. Anwar Ibrahim should not be made the prime minister, as he is sexually immoral, has vested foreign interest, and champions his causes through rallies and chaos xxx . 10. PR cannot be the government, as all it is good at is in demonizing the government, and has no concrete alternative to show for all its criticism and condemnation.
11. DAP is chided for its Chinese chauvinism, and its pro-republic stanch (opposed to the monarchy system in Malaysia) xxxi .
12. PAS offers policies which are not compatible or practical with the multiracial, multi-religious context in Malaysia. Further, they have brought no visible progress to Kelantan, the only state in Malaysia where they have been ruling uninterrupted for the past two decades xxxii . In reinforcing its worldview, it is interesting to note the primary affect targeted by BN campaigners has been 'fear for the unknown', a powerful political tool (Hatemi 2013). Malaysia has never been ruled by any party other than BN ever since independence, and the specter of the nation being torn apart once PR is in power proved to be overwhelming for many.

The Simulation Framework
In essence, the simulation framework is influenced by Zaller's (1992) Receive-Accept-Sample (RAS) model; a person keeps multiple views in its head, what kept being determined by her past consideration. At any particular moment, when its opinion is being called for, that which is salient at the moment will be retrieved and effective. We adapt the RAS model to fit in with Malaysian political reality, in particular, as mentioned before, reality as observed on the ground. Hence, in our simulation, an agent has both positive and negative schema in its mind -"UMNO/BN is bad", denoted Sb, and "UMNO/BN is good", Sg. The former is an anti-establishment position, and the latter pro-government. Both are latent in an individual. The scheme which is dominant, that is has more weight, determines whether the agent is a PR or a BN supporter. An emotional filter determines the scheme that is salient at the moment. In fact, in a number of studies (Schemer et al 2007, Healy et al 2009, emotion has been shown to be an important determinant of political orientation. While most of the studies were conducted in the context of developed nations, the author noted that it plays a significant role among the Malaysian electorates too, though not quite the same way as with an American electorate for example. A Malaysian with a high stress level, frustrated at the bureaucracy, angry and disappointed, for example with perceived ethnically-driven unfairness or at the ever degrading public transport system, is likely to detach himself from the establishment and side with an opposition party even though he or she might not actually fully endorse the party's position. He or she is likely to start reading only opposition media channels from then on, searching for government's wrong-doing to convince himself or herself that the government is 'bad'. Hence, the negative schema emerges especially in times of 'stress'. Still, the positive schema might re-emerge in "better days". We further note that the 'stress' experienced by an agent may not be real, in the sense that it might not come from actual experiences, but it may in fact be imparted by propagandist media. In fact, the Internet can be used for both information and intentional misinformation. News articles, especially in the Internet, can be articulated to invoke certain emotion and circulated with little consideration for truth, with sometimes adverse effects on the emotion of the electorate. Hence, while the Internet certainly makes it easier for the activists involved to pursue their goals (Farrell 2012, Aday et al. 2010, Conover et al. 2010, through lowering of the cost of certain kind of collective actions, it allows as well for easier manipulation of the masses, and the materials produced in the words of (Drezner and Farrell 2008) can have a "knock-on consequence for the media and for politics". And hence the distaste for it by some government figure, not just in Malaysia, but elsewhere as well, and notably the Turkish Prime Minister, Recep Tayyip Erdogan quipped, "There is now a scourge that is called Twitter. The best examples of lies can be found there. To me, social media is the worst menace to society." We therefore associate with each agent an emotional state, e, -the level of emotion that it has for the government. The higher the emotional state, the happier the agent is with the authority. This is further illustrated below in Figure  4. As shown in the figure, we assume for each agent an emotional value between 0 and 1. A value above 0.5 indicates a pro-BN emotion, while that below an anti-BN (or pro-PR) feeling. A value of exactly 0.5 can of course be construed as a totally neutral feeling.
Further, we associate a certain stress level in each of the areas populated by agents. The stress level denotes the environment under which the agent is living, and this stress affects the agent's emotional state. In our experiments, we assume two primary areas -urban and rural. Assuming a Malaysian scenario (might be true anywhere else), urban areas are more populated, and have a higher stress factor, and rural areas are relatively 'stress-less'. The extent to which environmental stress affects an agents emotional state depends on its 'stubbornness'. Stubborn agents will maintain its happiness or unhappiness despite the change in the environmental stress. Agents are spatially distributed, each with a position {x,y}. And every agent attempts to influence other agents. Different agents have different level of influence, infl, measured by a certain radius of influence. Influential agents, to be understood as politicians or individuals with high political interest and engagement, have a higher radius of influence compared to others (the general public). In the actual Malaysian context, they would attempt to assert influence through various medium (Weiss 2013) -flags, posters, banners, billboards, political rally, walkabouts in markets, neighborhoods, etc., religious services, dinners, and so on. Further, we note the role of the Internet in allowing global communication -broadcasting to the entire population rather than just within a localized target; hence, in the simulation, at each interaction cycle, if an agent, with a probability bi, uses the media, its influence is set to be arbitrarily large. When an agent, the propagator, attempt to influence another agent, whether locally or globally (through Internet broadcast), the chances of engagement depends on the similarity of emotional status between the two. An agent happy with the government is likely to engage with another of similar emotion, and similarly with an agent unhappy with the government. Consequently, a stressful agent is not likely to respond to a 'positive' influence -BN-oriented agent or view BN material, and likewise an agent already happy in "the way things work" is not likely to make itself interact with 'negative' material -PR-oriented agent or believe in PR material. Hence, emotion is assumed to affect attention, a relation documented in a number of works (for example Vuilleumier et al. 2003, Blanchette & Richards 2009). We further associate the notion of mobility with agents. Most agents are static; that is, they will stick to their spatial location, a situation akin to actual society where individuals have work and family to attend to and politics are not central in their life. But some agents are politically-centered individuals or "opinion leaders" (Lazarsfeld et al 1944), and they will move from one place to another in a direction {d x , d y }, wherever they see a chance to spread their influence. Preferential mobility together with selective openness to political discourse, facilitated as noted by (Sunstein 2002) by the Internet, lead people to become more "inward looking and extreme". The schema that the propagator may engage the receiver, either S b or S g. , depends on the emotional states of the two. If both have the same emotional orientation, they will interact using the schema that matches its emotion. For example, if both have PR-oriented emotion, then the schema chosen will be Sb. If they have different emotional orientation, then the active schema will be the one that matches the schema of the agent with the stronger emotion. If the emotional strengths are equal, the schema chosen will be randomly decided upon with equal probability. With the schema decided upon, the issue, {nk, nl}, within the chosen schema, is then randomly chosen, and the outcome of the interaction probabilistically depends on the communication skill, commsk, of each. If the receiver loses, it will be drawn closer to the opinion of the propagator. The actual amount of increase or decrease of conviction of the receiver on a particular issue depends on its emotional state. Specifically, if the interaction is over S b , the receiver will be more inclined to reduce w kl if it is in a good emotional state, and to increase the value On the other hand, if the interaction is over S g , the receiver will be more inclined to reduce w kl if it is in a bad emotional state, and to increase the value if otherwise. It should be noted that due to the way we compute the overall weight in a map, as the average over all the weights in the map, the smaller the map, the greater would be the effect of any adjustment of weight. A simple example is as follows: Compare a map, S0, with just a single weight, that is with just 2 nodes, to a map, S1, with a hundred weights. A change of ᐃw in the single weight in S0 would affect its overall weight more than the same change on a single weight in S1. Hence, the more complex a map, which means the more complex the predisposition, the slower will be the impact of any change of weight on it, a property in line with Zaller's resistance axiom (1992): People tend to resist, to some extent, arguments that are inconsistent with their political predispositions. Finally, it is important to note that in our simulation, we assume that any agent, even the staunchest supporter or party strongman, can be converted to the opposing side. While studies such as that done by Norris et al (1999) asserted that due to selective exposure and attention (or confirmation bias), at best, campaign can only reinforce, rather than change partisanship, to "reduce defections from the ranks", the goal of many campaigns in Malaysia has been exactly that -to increase defections from the ranks. In fact, defections can involve even the most hardcore supporters xxxiii ; individuals who used to be top party officials, or even top government leaders, have been known to defect to opposing parties. Hence, any agent, even the mobile ones, can be converted to the opposing side. In all, an agent, ai, has the following key attributes:   After each time step, we perform the following updates:  each agent, ai, is moved according to its accumulated motion direction, diri  the emotion, ei, for ai is perturbed based on the current environment stress factor, stress. The adjustment, incr, is randomly proportionate to the inverse of the stubbornness, stubi, of the agent, as shown below: emo = 1 -stress emoDiff = emo -e i incr = random fraction * (1 -stub i ) e i = e i + incr * emoDiff A brief note on the implementation: We implemented the algorithm above in the C language. While the majority of social simulation work appears to rely on higher-level tools such NetLogo (Gilbert and Troitzch 2005), we chose a lower-level language in order to take full advantage of the multicore platform where we run the experiments. Running in parallel on the multicore platform, we were able to execute thousands of runs of the simulation in a reasonable amount of time. The experiments are described in the next section.

Experimental Results and Discussion
The central political plot in the period of time surrounding GE13, as described early in the paper, pit an embattled BN against a PR side manned by more competent politicians. Hence, we ran simulation experiments to explore various scenario within that context. The key agents manipulated in the experiments were the mobile, politicallycentered agents -ntBN and proBN. The stubbornness values, and the initial values for emotion and positive opinions (which are proportional to the edge weights in the positive schema, and inversely to that in the negative schema), shown in Table 1, are chosen to differentiate between the two sides. The population size for both groups were variable, ranging in the experiment from 100 to 500. In conforming to the desired context, ntBN agents in general were assumed to have better communication skills compared to proBN; the average communication skills for proBN ranged with a value no more than 60 (out of 100), while that for ntBN at a value always higher than that for proBN. The general population, regardless of their political inclination, PR-(oppType) or BN-inclined (proType), had relatively low communication skill. The positive opinions of both oppType and proType were assumed to be completely random, with a uniform distribution in the range between 0 and 1. As can be seen in the table, there were more proType compared to oppType in the general population, with an approximate ratio of 60 to 40 xxxiv . The differences between proType and bnType were namely where stubbornness and emotion were concerned, factors that we might loosely here attribute to political genes (Hatemi 2012). In fact, aside from differences due to background (Hing and Ong 1987, page 137 for example), BN-type and PR-type individuals, as with liberals and conservatives in American politics (Amodio et al. 2007), can sometimes be recognized based on differences in outlook or thinking style xxxv , with the opposition-type individuals often perceived as being more likely to be more motivated (Weiss 2013) and entrenched in their position xxxvi (hence the higher stubbornness value). In spite of the predisposition, it is important to note that, subject to the right setting, each individual may incline herself to either side, in the course of a political campaign. For all continuous random attribute values used in the experiment, the value for each agent was chosen randomly using a Gaussian distribution that centered on the stated values, and with a standard deviation 10% of the values. The total population was fixed at 10,000 and the number of time-step was limited to 50, indicating the time constraint under which either camp had to win over the other side.
The size and population distribution for urban and rural areas were shown in Table 2. As indicated in the table, more of the population was within the smaller urban area, creating an area with a population density much higher than that for the rural area. The figures in the table more or less reflected the actual figures for Malaysia xxxvii . We performed 6 sets of experiments, each set comprising of 1000 runs. The factors being manipulated are indicated in Table 3; infl corresponding to the influence level of the agents, bcast its broadcasting probability and governance corresponding to lower stress level in urban (citystress) and village area (vlgstress). In experiment 1, we made it be such that ntBN always had higher infl and bcast, and BN governance was poor (citystress being always more than 0.5). In experiment 2a, we made it be such that proBN always have higher infl compared to that for ntBN. In experiment 2b, we increased BN's advantage by making proBN's bcast average being always higher than that for ntBN. We repeated the sequence of experiments in experiments 3a, 3b and 4, but with BN governance being always good in these rounds (both citystress and vlgstress less than 0.5). The results, to be elaborated upon in the subsections to follow, are shown in Table 4.

Experiment Set 1: Resourceful ntBN vs Less Resourceful proBN
The scenario is that of highly skilled and stubborn PR engaged agents, ntBN, up against less resourceful (lower influence range and broadcasting probability) BN agents, proBN, in a context where BN governs poorly. As shown in Table 1 (Exp-1), PR wins emphatically. Two representative cases are shown in Table 5      While it may appear from the table that the number of agents (ntBN or proBN), popsize, played a significant role in determining the outcome, the value was actually randomly generated, and ntBN popsize was greater than proBN popsize in 499 out of the 1000 runs. Figure 5 shows for run 1 the number of conversion to PR made by each of the first 1000 agents during the first simulation iteration. A negative value indicates a conversion to the other side, i.e. to BN. Figure 6 shows the PR voters population dynamic during the first iteration. As can be seen in the figures, the key politically-active agents -ntBN and proBN -, forming the first few hundred agents (along the horizontal axis), played the major role in determining the dynamics of electoral conversion. Similar graph profiles are noted for the different runs in different experimental cases, and these will not be reproduced.
In reality, the PR routing of BN as depicted in Experimental Set 1 might be reflected in what Najib referred to as the 'Chinese tsunami' xxxviii in GE13, an en bloc voting by the Chinese against the ruling coalition. MCA being at a low point was seen to have to failed to campaign effectively among the Chinese population at a time when the community was seen to peak in its agitation and frustration, rightly or wrongly, with the government.

Experiment Set 2a: BN improves Local Propaganda Machinery
The situation in this set was similar to that in Experiment Set 1, but with proBN's radius of influence being always greater than that for ntBN. This would in reality be achievable for BN through greater investment in campaign (propaganda) resources -more party workers, longer campaigning period, more banners, posters, more door-to-door visit, more gifts, etc. As shown in Table 4 (Exp-2a), while BN managed to increase the number of times it winsaverage of 584 votes per run up from 448 in the previous case, in general, it was still routed by PR. In spite of the greater infl, the less charismatic (lesser commsk) proBN in the absence of good BN governance, managed to gain only a little more ground for BN. Two representative cases are shown in Table 6. Note that in run 1, though BN won, it won thinly compared to PR in run 2. Further, note that PR won big in run 2 even though rural stress (vlgstress) was only 33. Plots of PR conversion and population dynamics for run 2 are shown in Figure 9 and 10.   Table 7. For run 1, note that even though proBN's bcast value was high, BN won only narrowly compared to PR for run 2. The margin with which BN is able to improve its win in this experimental set confirms the significance of the Internet in determining electoral swing, as per the discussion in Section 1.   Table 9. In run 1, PR edged past BN in spite of its smaller population size. In run 2, a strong BN size with an almost equal bcast measure overran PR.

Experiment Set 4: Repeat Experiment 2b but With Better Governance
Finally, we consider experimental runs where BN had both better campaign and media machinery, and recorded better governance. In this experimental set, proBN always had infl and bcast than that for ntBN, and the stress levels in both the urban and the rural areas were always less than 50 (out of 100

Insight from the Computer Experiments
As noted in (Sodaro 2004, page 71), "the purpose of a model is not to represent reality perfectly but to enable us to understand reality by allowing us to compare it against some standard or pattern". In fact, though a simulation that is only loosely based on actual parameters cannot be expected to mirror the reality exactly, from the results, we can draw the following conjecture: A ruling political party (or coalition) with agents less skilled in communication can maintain its position provided: 1. it provides good living condition or good governance 2. its agents make good use of the media (or media policy), especially the Internet, to reach out to the general public further than the opposing parties. To conclusively prove the conjecture, field work and empirical analysis would certainly be required, and this is beyond the scope of the paper. However, there are evidences hinting at its truth. BN had its biggest electoral victory since 1978 in the 2004 11 th General Election, a year marked by former prime minister, Abdullah Badawi, as "a transitional period, a time of high hopes" xxxix , and also the year when Malaysian Internet usage was only 42.3 percent (compared to 65.8 percent in 2012). Considering the Internet to be the enabler for the opposition in Malaysia, it is evident that the "feel good" factor together with the opposition's lack of media access then allowed a convincing win for BN. A further example is to the south of the country, in Singapore, where Shiang Nee xl noted that the ruling PAP seems to have recognized that for it to maintain its support base, it needs taking "a harder line against dissent and resisting pressure for political reforms, but giving out more economic benefits for voters that count the most -the middle class". "Harder line" here inevitably implies censuring of Internet media.

Limitation and Future Work
One limitation of the work is that we assume that everyone has access to the Internet. While the importance of the Internet has been highlighted by many in the literature and in the political circles, and confirmed in the simulation experiments, in actuality, not everyone in Malaysia has easy access to the medium. Those in the deep rural areas are not likely to tune to it, even if they do have the opportunities to do so, either out of cost concern or out of a simple preference for traditional, low-technology lifestyle. Hence, future work should consider low-technology group of agents. Also, we could have conducted a statistically more rigorous data analysis of the experimental results. For example, it would be interesting to study the statistical correlation between the various parameters manipulated with in the experiments. We feel however that such analysis would be meaningful only when tied to an actual field study. Further, we note that the agent-based simulation as presented here can be extended to capture more of the political complexity in Malaysia. The role of political networking for example can be better captured in the simulation, that being a significant factor especially in the rural electorate (Weiss 2013). Finally, though the use of worldview maps makes explicit the structure of the opinion of individual agents, to make the simulation more realistic, there should be an individual bias in the way in which an agent chooses and communicates on a topic when it is trying to influence another agent. It would then be interesting to see how the semantic of the opinion can possibly affect the simulation results.

Conclusion
We have presented an agent-based simulation model that describes the propagation of political perception or belief in Malaysia. Worldview map is used as the token that represents political belief. Computational experiments were made using parameters inspired by the situation "on the ground". The results obtained do point to the plausibility of the model, at least in the qualitative sense. In essence, the work shows that it is possible to capture the dynamics of Malaysian politics in the form of social simulations.