1: \documentclass[a4paper]{article}
2: \usepackage{graphicx}
3: \usepackage{amsmath}
4: \usepackage{hyperref}
5: \usepackage{color}
6:
7:
8: \begin{document}
9:
10: \begin{center}
11:
12: {\bf \Large Opinion polarization in the Receipt-Accept-Sample model}\\[5mm]
13:
14:
15:
16: {\large Krzysztof Ku{\l}akowski }\\[3mm]
17:
18:
19:
20: {\em
21:
22: Faculty of Physics and Applied Computer Science,
23:
24: AGH University of Science and Technology,
25:
26: al. Mickiewicza 30, PL-30059 Krak\'ow, Poland
27:
28: }
29:
30:
31:
32: \bigskip
33:
34: $^*${\tt kulakowski@novell.ftj.agh.edu.pl}
35:
36:
37:
38: \bigskip
39:
40: \today
41:
42: \end{center}
43:
44:
45:
46: \begin{abstract}
47: The Zaller theory of opinion formation is reformulated with one free parameter $\mu$, which measures the largest
48: possible ideological distance which can be made by a citizen in one mental step. Our numerical results show the
49: transient effects: {\it i)} the political awareness, measured by the number of
50: received messages, increases with time first exponentially, later linearly; {\it ii)} for small $\mu$
51: correlations are present between previously and newly received messages; {\it iii)} these correlation lead to
52: a hyperdiffusion effect in the space of attitudes of messages. Citizens with small $\mu$ are more prone to extremal
53: opinions.
54:
55: \end{abstract}
56:
57:
58:
59: %\noindent
60:
61: %{\em PACS numbers:} 89.65.-s, 64.90.+b
62:
63:
64:
65: %\noindent
66:
67: %{\em Keywords:} random networks; phase transitions;
68:
69:
70:
71:
72:
73: %% #####################################################################
74:
75: \section{Introduction}
76:
77: %% #####################################################################
78:
79: It is difficult to overestimate the importance of the public opinion (PO) in our life \cite{lipp}. In all
80: political systems, PO transfers the traditional receipts of solving problems to subsequent
81: generations. Its overwhelming influence is expressed by the democratic laws, which transfer the ultimate
82: decisions to the voters. In modern societies, the role of PO is amplified by mass media
83: - "the fourth power", which is sometimes more authoritary that each of the remaining three. (Recently
84: however, the list of agents at the market of information is enlarged due to the Internet, and the monopoly
85: of the mass media is somewhat weakened \cite{cast}.) The very importance of PO was not overlooked by the
86: sociophysicists, and several models have been created to capture its dynamics - for recent reviews
87: see \cite{stau,cafl}. On the other hand, the subject has been modeled mathematically also by social
88: scientists, and their models are of particular interest for the sociophysicists, as those models can
89: be expected to have a firm sociological basis. \\
90:
91: The subject of this text is the model of the opinion dynamics, formulated by the political scientist
92: John Zaller and published for the first time in 1992 \cite{zall}. The book is cited about 1000 times
93: \cite{gosc}, but only one citation \cite{bocc} can be found in the arXiv database \cite{arxi}, where
94: sociophysicists met. The book is devoted to the explanation and the verification of the model, which
95: is clearly exposed on its first page \cite{zall}:\\
96:
97: "The ideas necessary to accomplish this integration are few and surprisingly simple.
98: The first is that citizens vary in their habitual attention to politics and hence
99: in their exposure to political information and argumentation in the media. The second
100: is that people are able to react critically to the arguments they encounter only to
101: the extent that they are knowledgeable about political affairs. The third is that
102: citizens do not typically carry around in their heads fixed attitudes on every issue
103: on which a pollster may happen to inquire; rather, they construct "opinion statements"
104: on the fly as they confront each new issue. The fourth is that, in constructing their
105: opinion statements, people make greatest use of ideas that are, for one reason or
106: another, most immediately salient to them..."\\
107:
108: Zaller termed his model 'Receipt-Accept-Sample', or RAS, and we shall use this abbreviation below in the text.
109: Also we accept his term 'citizen' for a social actor. The original formulation contains at least 8 free parameters.
110: This is convenient when we struggle for accordance with experimental data, but less handy for
111: somebody interested in the general behaviour of the model. Our method here is to simplify
112: the mathematical formulation of the RAS model as much as possible, preserving the sociological content of the
113: four postulates cited above. The aim of this paper is to investigate the distribution of opinions about a model
114: issue. Our approach to the Zaller model profits much from the continuous opinion dynamics, as described
115: by Deffuant et al. in 2000 \cite{defw}. There, the core idea is that other opinions are taken into account by an agent if the
116: distance from their content to his actual attitude is not larger than some threshold value. Here, the
117: distance $d$ is measured from a new message to the closest point on the plane of opinions, occupied by the agent in the past.
118: This distance is compared with the threshold parameter $\mu$; if $d<\mu$ the message is received, but it is ignored if $d>\mu$. \\
119:
120: The Zaller's mathematical formulation of the RAS model is briefly reviewed in our Section 2. In Section 3 we describe
121: its simplified version proposed in our model. Last two sections are devoted to the numerical results and their discussion.
122:
123: \section{The original mathematical formulation}
124:
125: The postulates listed above are expressed as formal axioms which constitute the RAS model \cite{zall}. On this
126: basis, a mathematical construction is built as follows:\\
127:
128: 1. Each citizen $i$ is endowed with the political awareness of a given value $W_i$.\\
129: 2. The probability of receipt a message relevant for the opinion formation increases
130: with his awareness according to a sigmoidal function $f$
131:
132: \begin{equation}
133: f(W_i;a_0,a_1)=1-\frac{1}{1+f+\exp(a_0+a_1W_i)}
134: \end{equation}
135: where $f$ - floor parameter, which marks a minimum level of reception, $a_0$ - a coefficient
136: which designates the intensity of a message, and $a_1$ - a coefficient which designates
137: strength of a relationship between awareness and reception. \\
138:
139: 3. Provided that a message is received by $i$, the probability of its acceptance decreases
140: with the awareness $W_i$ and it is given by another sigmoidal function $g$
141:
142: \begin{equation}
143: g(W_i,P_i;b_0,b_1,b_2)=\frac{1}{1+\exp(-b_0-b_1W_i-b_2P_i)}
144: \end{equation}
145: where $b_0$ - a coefficient which designates the difficulty or credibility of the message,
146: $b_1$ - a coefficient which designates the effect of awareness on resistance to persuasion,
147: $b_2$ is a coefficient which designates the effect of the predisposition on resistance to
148: persuasion, and $P_i$ measures the predisposition of $i$ to accept the message, and it depends
149: on the ideological relation of $i$ to the message content.\\
150:
151: 4. The probability of acceptance of the message and a henceforth change of attitude is equal
152: to the product $Pr= f(W_i;a_0,a_1)g(W_i,P_i;b_0,b_1,b_2)$. \\
153:
154: 5. The latter expression can be used to construct a kind of one-way Master equation for the
155: probability $Prob$ of a given opinion $Opi$, provided that at $t=0$, the baseline opinion was $Bas$
156:
157: \begin{equation}
158: Prob(Opinion)_t=Prob(Bas)+(1-Prob(Bas))*Pr*Dum_t
159: \end{equation}
160: where $Dum_t=0$ and 1 at initial and final time, respectively. The final opinion is expected to
161: be expressed in an opinion statement in a poll.\\
162:
163: 6. Additional consideration is due to a situation when the citizen is a subject of two streams of
164: opposite messages. In this case, the decision on the content the opinion statement is undertaken
165: according to opinions most immediately accessible in a citizen's mind. The probability to recall
166: a previously accepted opinion is given by yet another sigmoidal function $h$
167:
168: \begin{equation}
169: h(W_i;c_0,c_1)=1-\frac{1}{1+\exp(c_0+c_1W_i)}
170: \end{equation}
171: where the parameters $c_0$ and $c_1$ are analogous to $a_0$ and $a_1$. \\
172:
173: 7. To be specific, let us consider two opposite attitudes, say pro- and antiwar (P or A) ones
174: \cite{zall}. Let us also assume that our statistical citizen expresses some opinion statement
175: on this issue; actually, a separate expression is given in \cite{zall} for the probability of this
176: fact. The probability of the prowar opinion statement $p_p$ expressed by citizen $i$ is supposed to be
177:
178: \begin{equation}
179: p_p=\frac{S_p(i)}{S_a(i)+S_p(i)}
180: \end{equation}
181: where $S_p$ ($S_a$) is the accessibility of the prowar (antiwar) messages in the citizen's mind.
182: These accessibilities are given as
183:
184: \begin{equation}
185: S_p(i)=\sum_t f(W_i;a_0,a_1)g(W_i,P_i;b_0,b_1,b_2)h(W_i;c_0,c_1)
186: \end{equation}
187: where $t$ is time and the sum is performed over a given time period (say, two recent years).
188: In this sum, the constants $a_0$, $b_1$ and $b_2$ depend on time $t$, and the constants $a_0$ and
189: $b_0$ depend on the message (prowar or antiwar).\\
190:
191: 8. Some additional assumptions (all the messages of the same intensity) allow to find all the
192: coefficients from the fitting of the theoretical curves to the poll data.\\
193:
194: \section{Our formulation}
195:
196: To investigate, as we intend here, the distribution of opinions, it is necessary to postulate
197: how the opinions vary in time. On the contrary to most sociophysical approaches, the RAS model \cite{zall} does not
198: take into account direct interactions between citizens; it is only the influence of media what is taken
199: into account. This does not preclude the possibility that some citizens play the role of media, which could
200: be taken into account in future research. Here we are going to preserve the one-particle character of the
201: Zaller model. Our interest is in the time dynamics of PO with respect to a given issue.\\
202:
203: In particular we are interested in a possible sequence of events when a newborn citizen starts to hear
204: political news. His initial awareness is close to zero, but he is indiscriminative in his attitude and he
205: accepts any news as typical and normal. In this sense, his initial acceptance is large. Being a subject of a random
206: stream of messages, initially he is
207: not able to receive most of them, because they seem to him to be too sophisticated. There are some, however, apparently
208: addressed to citizens like him: full of emotions, which clearly divide the world into good and bad, expressed by
209: somebody authoritative but young. Our citizen captures such a message and learns to distinguish it from other
210: messages. His political education just started, and his awareness increases a little bit. Simultaneously, his
211: acceptance is strongly reduced. At least in the near future he will be willing to identify his opinion with this 'his first'
212: message. In physics we like to term such events as 'spontaneous symmetry breaking' \cite{ball}.\\
213:
214: To capture the evolution of the citizen's understanding, we need {\it i)} a history-dependent awareness,
215: {\it ii)} a time series of messages which vary in their content with respect to some set of issues which we consider
216: to be salient in a given society.
217: In fact, we do not need anything more to indicate, that there is a positive feedback between the political
218: attitude and the character of newly received informations. On the contrary to Ref. \cite{bocc}, we do not need a
219: Gaussian or any other distribution of awareness, as this distribution should be a result of the calculation.
220: Also, we consciously do not contribute to the discussion if a given political orientation is correlated
221: with the awareness \cite{bio}. Instead, we will show that a citizen can be randomly trapped by a series of
222: messages close to each other till the time when his attitude is firmly established.\\
223:
224: The mathematical formulation of our version of the RAS model is then as follows:\\
225: \noindent
226:
227: 1. At the beginning, the political awareness of each citizen is equal to one. Later it is represented by the number
228: $n(t)$ of different opinions/messages received till time $t$. Each citizen starts with one received message,
229: placed at the centre of coordination.\\
230:
231: 2. Each new message is represented as a point on a plane, with coordinates $(x,y)$. The plane plays the role of the
232: space of main attitudes with respect to some salient issues, for example 'safety vs freedom' and 'free trade vs
233: welfare state'. Simultaneously, each message is represented by a point at this plane. The actual value of the
234: dimensionality of this space is of minor importance, except the condition that it is larger than one. When the
235: dimensionality $D$ is two or larger, the size of the $(D-1)$-dimensional circumference of the occupied area increases
236: with its size. This means, that the ability to receive new messages increases with the awareness. \\
237:
238: 3. The ability to receive a new message is a decreasing function $f$ of the distance $d$ between the point $(x,y)$
239: and the closest point received in the past. For simplicity we adopt the threshold function $f=\Theta(\mu-d)$ \cite{defw}.
240: Here $\mu$ is a parameter, which can be roughly interpreted as the capacity to receive new messages. Receipt of a new message
241: is equivalent to a visit at the point $(x,y)$ assigned to this message. The starting point is placed at the
242: coordination center.\\
243:
244: 4. The spatial distribution of received messages can be used to calculate the probabilities of the opinion statements in
245: a similar way, as Zaller did in Eq. (5). In our case, the probabilities are calculated from the political contents
246: of the previously received messages, i.e. the position of the points which represents
247: the messages. For an exemplary issue defined by the $y$ axis as the boundary between the opinions 'pro' and 'anti',
248: the message weight is just its $x$ coordinate. Having chosen $x>0$ as 'pro' and $x<0$ as 'contra' we can calculate the
249: respective probability 'pro' as $p$
250:
251: \begin{equation}
252: p=\frac{\sum_{x>0}x(t)}{\sum \mid x(t)\mid}
253: \end{equation}
254: \\
255:
256: 5. Other issues can be visualised as other axis, not necessarily orthogonal to the plane of salient issues, defined above.
257: Opinions on those other issues can be formed on the basis of the projection of the new axis to the 'salient' plane.\\
258:
259: 6. Once a citizen has a given attitude 'pro' along the OX axis, i.e. $p>0.5$, this attitude can be neutralized by
260: receipt a given number of messages with $x<0$. It is natural to set $p=0$ as a neutral attitude. Then, the number
261: of messages to neutralize $p$ can be roughly evaluated as $(2p-1)n(t)$.\\
262:
263: It seems to this author, that this formulation fulfils the content of
264: the postulates of Zaller's theory. As mentioned above, the distribution of the awareness
265: appears as a natural consequence of the procedure listed above. As we are going to demonstrate,
266: the area around received messages varies from one citizen to another. Simultaneously, the circumference
267: of this area is a measure of the amount of ideas which can be accessible by the citizen in a near future.
268: A good school is where young minds are gradually fed by new ideas, without prejudices towards
269: this or other direction. On the contrary, if a citizen is indocrinated by only one idea,
270: he is not able to receive anything else; in our model, such brainwashing is equivalent to the
271: case when subsequent messages either are close to the current position, or too far from it to
272: be received. Actually, the stream of messages we encounter in our life is not completely random,
273: but it depends on our intellectual environment. At the moment, however, we do not construct a
274: theory of the whole society, then our analysis is limited to citizens and does not embrace
275: the media.
276:
277: We hope that the arguments given in the preceding paragraph allow to state that the first and the
278: second postulate of Zaller's theory: that the citizens vary in their awareness, and that
279: they are able to take into account only those ideas which they are knowledgeable about. The essence
280: of the third and fourth postulates is captured in the point that the opinion statement is formulated
281: on the basis on the recent trajectory, but separately for each new issue formulated in a poll.
282: Here we do not concentrate much on the sampling process, when a new issue is formulated. According to Zaller, opinions
283: on such new issues are formulated on the basis of other, most salient issues. To refer to an opinion
284: on a new issue, a relation should be determined between this new issue and the old (salient) one.
285: We imagine that such a relation can be constructed in a three-dimensional space; if a new axis is orthogonal
286: to the old ones, nothing can be stated on a new issue. However, if their angle is different from $\pi /2$,
287: information on the distribution of opinions can be drawn by means of a geometrical projection.
288:
289: The selected form of the function $f$ is adopted from the Deffuant model. In general, any decreasing function
290: $f(x)$ fulfils the condition that messages more distant are less likely to be received. On the other hand,
291: one of the postulates by Zaller indicates that some messages can be received but some others can not,
292: and that the ability to receive them increases with the awareness of the receiving citizen. In our formulation,
293: the awareness increases with the area around previously received messages. When this area is larger, more new
294: messages can be received.
295:
296: \section{Numerical results}
297:
298: In a generic case, the awareness of a citizen increases with time at first exponentially, later it becomes a linear
299: function of time. An example of the time dependence of the awareness is shown in Fig. 1. At the later
300: stage all incoming messages appear to be at the already known area.
301: Numerically, the distance between a new point and some already received points is shorter than the threshold $\mu$;
302: then, the new point is also received. This means that the number of received messages increases by one
303: at each time step. At the earlier stage, however, the time to receive a new message is longer. Namely,
304: the probability to receive the second message is equal to $s=\pi \mu^2/4$, where the factor 1/4 comes from the fact
305: that new messages appear as points placed randomly at the square $2\times 2$. Then the probability $P(n)$
306: to pass $n$ new mesages until the receipt of the $n+1$-th one and the increase of the awareness
307: to $W=2$ is $P(n)=s(1-s)^n$. The time to receive the third message depends on where the first received point
308: is placed, and should be averaged over its possible positions. Similarly the distributions of waiting time to get
309: subsequent messages depend on the positions of the points which refer to previously received messages.
310: The numerical calculations indicate, that these functions also decrease exponentially with $n$, and the rate
311: of decrease is larger for higher values of the awareness. The plots are shown in Fig. 2. This means in particular,
312: that the mean time to receive a new message decreases with the awareness.
313:
314: \begin{figure}
315: \begin{center}
316: \includegraphics[scale=0.45,angle=-90]{ff1.ps}\\
317: \end{center}
318: \caption{Example for the time dependence of the awareness $W(t)$.
319: }
320: \end{figure}
321:
322: \begin{figure}
323: \begin{center}
324: \includegraphics[scale=0.45,angle=-90]{h2.ps}\\
325: \end{center}
326: \caption{The probability distributions $P(n)$ to wait $n$ timesteps until an increase of the awareness $W$ to the values 3, 4, 5, 6 and 7.
327: }
328: \end{figure}
329:
330: Being (socio)physicists, we are tempted to investigate the geometrical aspect of the correlation. For this purpose
331: we calculate the time dependence of the square $r^2$ of the sum of coordinates of the messages
332: received till a given time against time. Note that in this way a point is taken into account several times,
333: if no messages were received after the message which it represents. Subsequent points are not correlated in two
334: cases: first, if the whole plane is already densely occupied by previously received messages, and second - for
335: large $\mu$, where any new message is received, despite its position. In the latter case, the slope of $r^2$ against
336: time should be 1 in the log-log scale in accordance with the diffusion law. Our numerical results
337: indicate, that there are two transient times for small $\mu$: $t_1(\mu)$ and $t_2(\mu)$. For $t<t_1(\mu)$ the slope is
338: larger than 1 and reaches 2.5. For $t_1(\mu)<t<t_2(\mu)$ the slope is smaller than 1. Finally, for $t>t_2(\mu)$ the slope is
339: 1 for each $\mu$. When $\mu$ increases, both $t_1(\mu)$ and $t_2(\mu)$ decrease and merge.
340: In this way, we observe a kind of cross-overs in time from the hyperdiffusion to the subdiffusion and then to the normal
341: diffusion. These results are presented in Fig. 3. The plots are averaged over 1000 trajectories. The results are the same
342: if we limit the area where new points appear to a circle with radius equal to 1.0.
343:
344:
345: \begin{figure}
346: \begin{center}
347: \includegraphics[scale=0.45,angle=-90]{arry.ps}\\
348: \end{center}
349: \caption{Time dependence of the squared sum of positions of all received messages for $\mu$ = 0.25, 0.3, 0.4, 0.6 and 1.5.
350: The slope of the curves except the one for $\mu$=1.5 is initially larger than 1.
351: }
352: \end{figure}
353:
354: A characteristic feature of the model is that subsequently received messages are correlated in their ideological content.
355: In our geometrical realisation this correlation is expressed as the correlation between positions of subsequently received messages.
356: Once a new message is selected, the area around received messages gets widened towards a given direction. This raises the possibility of
357: correlations between subsequent messages. Then, we calculate this correlation for the direction $x$ - the results are shown in Fig. 4. We use the
358: expression for non-stationary correlations, i.e.
359:
360: \begin{equation}
361: W_x=\frac{\langle x(t)x(t+1)\rangle-\langle x(t)\rangle \langle x(t+1)\rangle}{\sqrt{Var(x(t)) Var(x(t+1))}}
362: \end{equation}
363: \\
364:
365: This correlation is larger for small values of $\mu$. When the whole area is covered, the correlation
366: disappears. It is obvious, that the correlation lifetimes increase when $\mu$ decreases. The results should not depend on the selected
367: direction; the $OX$ axis is chosen for simplicity of the parametrization. The effect of correlations is analogous to the viscous fingering in the snowflake formation \cite{fract}, where randomly attached molecules increase the probability of attaching further molecules at the same area, and the curvature of the surface increases.
368:
369: \begin{figure}
370: \begin{center}
371: \includegraphics[scale=0.45,angle=-90]{fr4.ps}\\
372: \end{center}
373: \caption{Time dependence of the correlation $Wx$ between two subsequently received points (messages) for $\mu$ = 0.4, 0.6, 0.8, 1.0 and 1.5.
374: }
375: \end{figure}
376:
377:
378: In Fig. 5 we show the probability distribution $P(p)$ of these probabilities at a not-too-long time moment, i.e. after receipt of 500 messages.
379: As we see, the obtained plot is close to the Gaussian distribution for $\mu >0.4$, but its shape deviates from Gaussian for smaller $\mu$ because of the limitation of the variable $p$ to the range (0,1). None of these plots shows a fat tail. Note that different
380: trajectories contribute to the average with their different numbers of received messages. We also checked that in time, the distribution gets narrow;
381: the time dependence of the variance of $P(p)$ decreases, as shown in Fig. 6. This narrowing can be interpreted as follows: as citizens get
382: more complete information, their opinions are clarified. The final mean probability $<p>=0.5$; this follows from the initial symmetry of the
383: distribution, which is preserved by the dynamics.
384:
385: \begin{figure}
386: \begin{center}
387: \includegraphics[scale=0.45,angle=-90]{norm.ps}\\
388: \end{center}
389: \caption{The probability distribution $P(p)$ that the opinion statement of randomly selected agent is equivalent to accept a given issue with
390: probability $p$. The curves are obtained after 500 messages are received, for $\mu$=0.2, 0.25, 0.35 and 1.5 from the widest to the most narrow curve,
391: respectively, and averaged over $10^5$ citizens.
392: }
393: \end{figure}
394:
395: Let us remind that the factor $\mu$ describes the largest possible distance between messages, from the previously to the one newly received.
396: It is somewhat surprising that opinions of the citizens which understand less swiftly,. i.e. of those with smaller $\mu$, is temporarily more wide.
397: This effect is visible in Fig. 5, where the distribution of probabilities is more wide for smaller value of $\mu$. The reason is the correlation between subsequently received messages, which is larger for smaller $\mu$.
398:
399:
400:
401: \section{Discussion}
402:
403: The assumption most essential here is the geometrical-like distance between messages, determined by their political content. Next assumption
404: correlates the distance between a new message and previously received messages with the receipt of the new message. The latter seems natural,
405: the former can seem doubtful. However, separating out the geometrical considerations, we are left with the conjecture that messages
406: with the content far from anything previously received are received less likely. This conjecture seems to be quite natural.
407:
408: It is possible to continue this kind of reasoning. Our main goal is to demonstrate, that opinions of people endowed
409: with smaller factor $\mu$ are more spread, i. e. there are more extremal opinions, than for the people with larger $\mu$.
410: To falsify this theory one should demonstrate that the result should be the opposite, i.e. the capacity to receive new
411: informations is positively correlated with the extremal opinions. It seems to this author that this is not the case.
412: On the other hand, in our model new messages appear in a limited area. Then, the model is appropriate to a given area
413: of knowledge (here the square $2\times 2$), which sooner or later will be covered by received messages. We think that
414: capable citizens spread their fields of interest beyond the boundaries which limit thoughts of less endowed people.
415: Then in new areas everybody is a freshman at least at the beginning of their exploration.
416:
417: Our result that opinions of people with smaller $\mu$ are more spread is akin to the result that opinions of people
418: with smaller threshold parameter are more spread (Figures 1 and 2 in Ref. \cite{defw}). Although the outcome of both
419: results is the same, their origin is different. The Authors of Ref. \cite{defw} assumed the homogeneous initial distribution
420: of opinions. There, smaller threshold means smaller adaptation and smaller mobility in the opinion space. As a consequence,
421: the distribution remains wide. In our case, the initial distribution is concentrated at the centre. Smaller $\mu$ means
422: smaller adaptation and smaller mobility, but larger correlations. As a consequence, the distribution gets wider.
423:
424:
425: \begin{figure}
426: \begin{center}
427: \includegraphics[scale=0.45,angle=-90]{tg8.ps}\\
428: \end{center}
429: \caption{The variance of the probability distribution $P(p)$ against the capacity parameter $\mu$, for various times of calculation: from 50 to 500 timesteps.
430: }
431: \end{figure}
432:
433: We found that the awareness, measured by the number of received messages, increases with time exponentially, later linearly. The seemingly exponential part reveals a substructure, which can be of interest for statistically oriented minds, but its existence
434: relies on the assumption that messages can be represented by discrete points. It is possible that their representation as, say, extended
435: excitations in a network of ideas could be more fruitful. Other result are indications of a hyperdiffusion in the space of messages. The effect is transient and it disappears when the considered area becomes known. In this transient time and for small $\mu$, positive correlations are found between the content of subsequently received messages. The transient time decreases with $\mu$; more capable citizens are quicker.
436:
437: One of the conclusions by Zaller was that the acceptance decreases with the awareness. In our formulation, the equivalent of non-acceptance is
438: the number of messages necessary to neutralize a given attitude. For a given $p$, this number increases with the awareness $n$, as remarked
439: in point 6 of our formulation. This point does not preclude the mean decrease of non-acceptance with time, as shown in Fig. 6. However, more
440: detailed discussion of this question should take into account some initial distribution of attitudes. Here we concentrate on the influence
441: of $\mu$ on the opinion spread; then we purposefully neglect this initial distribution, placing initially each citizen at the coordination centre.
442:
443: We feel that the weak part of our formulation is that the messages previously received do not lead in the model to
444: a coherent picture, but they stand by in minds till the end of the simulation. A partial remedy could be to add
445: another parameter of forgetting old messages in time, as was done for example in the Bonabeau model \cite{bona}.
446: Zaller proposed to take into account only some last part of the simulation. This effect is under investigation.
447: Even more essential process
448: disregarded here as in Ref. \cite{zall} is the contact between people. This mechanism of the opinion formation was
449: discussed by many sociophysicists \cite{sznajd,krahe,weis,gala}; we recommend \cite{cafl} as the current review.
450: The next challenge is to build the interpersonal mechanism into
451: the RAS theory. For this purpose, the Deffuant model \cite{defw,weis} seems most natural. This model relies on
452: the assumption that we react to opinions of other people if they are not too far from our initial beliefs.
453: Our formulation of the RAS theory explores the same idea.
454:
455: \bigskip
456:
457: {\bf Acknowledgements.} I am grateful to two Anonymous Referees for their helpful remarks. The calculations were performed in the ACK Cyfronet, Cracow, grants No. MNiSW /SGI3700 /AGH /030/ 2007 and MNiSW /IBM BC HS21 /AGH /030 /2007. The text is dedicated to my wife Magda.
458:
459:
460: \bigskip
461:
462:
463: %\section*{Acknowledgements}
464:
465:
466:
467: \begin{thebibliography}{99}
468: \bibitem{lipp} W. Lippmann, {\it Public Opinion}, Harcourt, Brace and Co., New York 1922.
469: \bibitem{cast} M. Castells, {\it The Rise of the Network Society}, Blackwell Publ., Cambridge, MA 1996.
470: \bibitem{stau} D. Stauffer, {\it Sociophysics Simulations II: Opinion Dynamics}, in {Modeling Cooperative
471: Behavior in the Social Sciences}, Eds. P. L. Garrido, J. Marro and M. A. Munoz, AIP Conf. Proc., Vol. 229,
472: Melville, NY, 2005, p. 56.
473: \bibitem{cafl} C. Castellano, S. Fortunato and V. Loreto, {\it Statistical physics of social dynamics}, arXiv:0710.3256.
474: \bibitem{zall} J. R. Zaller, {\it The Nature and Origins of Mass Opinion}, Cambridge UP, Cambridge 2005.
475: \bibitem{gosc} http://scholar.google.pl/
476: \bibitem{bocc} N. Boccara, {\it Models of opinion formation: influence of opinion leaders}, arXiv:0704.1790.
477: \bibitem{arxi} http://arxiv.org/
478: \bibitem{defw} G. Deffuant, D. Neau, F. Amblard and G. Weisbuch, {\it Mixing beliefs among interacting agents}, Adv. in
479: Complex Systems 3 (2000) 87.
480: \bibitem{ball} Ph. Ball, {\it Critical Mass: How One Thing Leads to Another}, Arrow Books Ltd, London 2005.
481: \bibitem{bio} D. M. Amodio, J. T. Jost, S. L. Master and C. M. Yee, {\it Neurocognitive correlates of
482: liberalism and conservatism}, Nature Neuroscience, 10 (2007) 1246.
483: \bibitem{fract} J. Feder, {\it Fractals}, Plenum Press, New York 1988.
484: \bibitem{bona} E. Bonabeau, G. Theraulaz and J-L. Deneubourg, {\it Phase diagram of a model of self-organizing hierarchies},
485: Physica A 217 (1995) 373.
486: \bibitem{sznajd} K. Sznajd-Weron, {\it Sznajd model and its applications}, Acta Phys. Pol. B 36 (2005) 2537.
487: \bibitem{krahe} R. Hegselmann and U. Krause, {\it Opinion dynamics and bounded confidence: models, analysis and
488: simulation}, JASSS 5 (2002).
489: \bibitem{weis} G. Weisbuch, G. Deffuant and F. Amblard, {\it Persuasion dynamics}, Physica A 353 (2005) 555.
490: \bibitem{gala} S. Galam, {\it Sociophysics: A review of Galam models}, arXiv:0803.1800.
491: \end{thebibliography}
492:
493:
494: \end{document}
495:
496: