The Hunt Opinion Model—An Agent Based Approach to Recurring Fashion Cycles

Rafał Apriasz; Tyll Krueger; Grzegorz Marcjasz; Katarzyna Sznajd-Weron

doi:10.1371/journal.pone.0166323

Abstract

We study a simple agent-based model of the recurring fashion cycles in the society that consists of two interacting communities: “snobs” and “followers” (or “opinion hunters”, hence the name of the model). Followers conform to all other individuals, whereas snobs conform only to their own group and anticonform to the other. The model allows to examine the role of the social structure, i.e. the influence of the number of inter-links between the two communities, as well as the role of the stability of links. The latter is accomplished by considering two versions of the same model—quenched (parameterized by fraction L of fixed inter-links) and annealed (parameterized by probability p that a given inter-link exists). Using Monte Carlo simulations and analytical treatment (the latter only for the annealed model), we show that there is a critical fraction of inter-links, above which recurring cycles occur. For p ≤ 0.5 we derive a relation between parameters L and p that allows to compare both models and show that the critical value of inter-connections, p*, is the same for both versions of the model (annealed and quenched) but the period of a fashion cycle is shorter for the quenched model. Near the critical point, the cycles are irregular and a change of fashion is difficult to predict. For the annealed model we also provide a deeper theoretical analysis. We conjecture on topological grounds that the so-called saddle node heteroclinic bifurcation appears at p*. For p ≥ 0.5 we show analytically the existence of the second critical value of p, for which the system undergoes Hopf’s bifurcation.

Citation: Apriasz R, Krueger T, Marcjasz G, Sznajd-Weron K (2016) The Hunt Opinion Model—An Agent Based Approach to Recurring Fashion Cycles. PLoS ONE 11(11): e0166323. https://doi.org/10.1371/journal.pone.0166323

Editor: Mark R. Muldoon, University of Manchester, UNITED KINGDOM

Received: August 8, 2016; Accepted: October 26, 2016; Published: November 11, 2016

Copyright: © 2016 Apriasz et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All simulation result files are available from the IDEAS/RePEc database, in the HSC Software repository (http://ideas.repec.org/s/wuu/hscode.html).

Funding: This work was partially supported by funds from the National Science Centre (NCN, Poland; URL: http://ncn.gov.pl/) through grant no. 2013/11/B/HS4/01061. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Theories of social cycles are among the earliest theories in sociology, dating back to the seminal works of Vilfredo Pareto [1]. In this paper we focus on recurring fashion cycles, which are empirically documented in many studies [2–5] and can be described within several classical theories [6–8]. Among them, probably the most known, universal and influential is the one proposed by Simmel [8, 9]. As noted by Simmel “For fashion to exist, society must be stratified, some members must be perceived as inferior or superior—or simply as worthy or unworthy of being imitated”. Certainly not all fashion cycles can be described within Simmel’s theory [9–11]. A good example is the mid-20th century vogue for blue jeans—a phenomenon that, at least at its outset, was intended as a political statement and celebration of working people rather than, say, intra-group dynamics involving self-conscious contrariness. Moreover, two-population models are not necessary to describe recurring social cycles. For example, an interesting preference model, in which individuals are described by traits and preferences, has been introduced in [12] as an alternative to status models [6, 9] or the neutral model [13]. Nevertheless, Simmel’s theory can be applied to understand many fashion cycles [12, 14, 15] and is particularly interesting from the modeling point of view, due to its simplicity.

From a psychological point of view, Simmel’s theory is based on two types of social response—conformity (a result of the tendency to imitate others) and anticonformity (a result of the tendency to distinguish ourselves from others). It is well known that both types of social response are just two faces of the same coin. They are opposites at the operational level, but at the conceptual level the responses are similar in that both indicate behavior that has been influenced by the source [16]. This immediately implies the second similarity between both types of response—they are relative, which means that the same individual may conform to one group and anticonform to another.

While conformity underlies many models of opinion dynamics, including various Ising-type models (i.e., agent-based models with agents described by a single trait that takes only binary values, for reviews see [17–19]), anticonformity is a much less considered type of social response. In the context of binary opinion dynamics models, anticonformity was probably introduced for the first time in 2004 by Galam using the notion of ‘contrarians’ [20]. Later on, this idea has been applied to other Ising-type models (see a short review in [21]). However, to the best of our knowledge, the relativity of conformity vs. anticonformity has been considered for the first time only very recently, in a simple model of two interacting cliques [22]. The model has been built on a modified version of the q-voter model [23, 24], which occurred to be very interesting both from the theoretical [25–29] and the applicative [30–33] points of view. Using Monte Carlo simulations it has been shown that the interplay between intra-clique conformity and inter-clique anticonformity leads to a bi-polarized state of the entire system if only fraction L of cross links between cliques is larger than the critical value, L* = L*(q). For smaller values of L, consensus in the entire system is possible.

From the physical point of view, conformity corresponds to ferromagnetic interactions and anticonformity to antiferromagnetic, what has been already discussed in [21]. Therefore, the model introduced in [22] reminds of the Ising model with competing ferromagnetic interactions between nearest neighbors and antiferromagnetic between next nearest neighbors [34–36]. From this point of view, the existence of the critical value of the interplay between conformity and anticonformity, below which consensus exists, is not very surprising. On the other hand, the two-community structure with competing ferro- and antiferromagnetic interactions has not been investigated before and seems to be interesting for social applications. It is also worth to recall that the two-community structure has been already investigated in the context of the Ising model in [37, 38] and the linear voter model [39], but in all these papers only ferromagnetic interactions were investigated. Here we propose a modification of a model introduced in [22] that could not be easily motivated by any physical problem but is very interesting from the social point of view—we break the symmetry between cliques.

Our modification goes in line with fashion theories that assume an inequality between groups—members of the first group conform to all individuals (we will call them “followers” or “opinion hunters”, hence the name of the model), whereas members of the second group conform only inside their own group and anticonform to members of the other group (we will call them “snobs”). As we have mentioned, our modification seems to be bizarre from the physical point of view, because it would mean that interactions between particles are non-symmetric. Therefore, to the best of our knowledge, it has never been studied before within Ising-type models. As we will show, such a modification leads to a new type of behavior, that has not been observed in [22] and is very often found in fashion [2–5]—above a critical fraction of inter-links recurring cycles occur.

Although we are aware that our model is very simple and we do not claim that it describes recurring fashion cycles in all their richness, we hope that it makes a contribution to the understanding of social cycles for which inequalities between groups are important. Moreover, in spite of its simplicity it takes into account several empirically observed facts:

Social networks display community structure—groups of nodes within which connections are dense but between which they are sparse and the number of inter-community links (edges) can be measured [40–42]. Here we use the double click approximation [39], which allows for analytical treatment and has been used also in the original model [22].
People often select cultural tastes (opinions, attitudes, behaviors, etc.) that distinguish them from members of other groups. Moreover, they abandon cultural tastes when members of other social groups adopt them [11]. Therefore we use here the idea introduced in [22], which links the type of social response with the group identity.
Empirical studies have shown that across different informal locations, freely-forming groups range in size from 2 to 7 with the average being 2.41 [43, 44]. This justifies the assumption that underlies the q-voter model that simultaneously each individual can interact only with a small group of q neighbors, instead of all his/her neighbors.
The social pressure increases with the size of the group but a small unanimous group may be more efficient in convincing others than a much larger group with a non-unanimous majority [45]. This justifies the second assumption of the q-voter model that a unanimous opinion of the q neighbors is needed to influence others.

According to [6], our model belongs to the class of true endogenous cycle models, which means that all phenomena observed on the macroscopic scale will be driven by direct interactions between agents on the microscopic scale. In the Simmel theory it is assumed that members of the elite group abandon a certain fashion because too many others, lower status individuals, adopted this fashion. However, in our model we do not need to assume this—none of the individuals has knowledge of any global, aggregated variable, such as the average preference or opinion. The macroscopic “elite’s escape” will emerge spontaneously from direct interactions between the spinsons. Stressing the endogenous nature of the model, we do not claim that exogenous factors are not important for fashion. Public media or social norms are undoubtedly very important. For example, recently it has been shown empirically based on a large-scale dataset of individual purchases of women’s shoes (16,236 transactions) across five years that women conform to new local norms (i.e. the average heel size) when moving to relatively higher status locations, but mostly ignore new local norms when moving to relatively lower status locations [10]. Our claim is that we do not need to introduce any exogenous factors to obtain recurring cycles, a fact acknowledged in the population dynamics literature [46, 47].

We will examine here the role of the number of inter-links between two groups of agents, as well as their character. Therefore, we will consider two versions of the model—quenched (parameterized by fraction L of fixed inter-links) and annealed (parameterized by probability p that a given inter-link exists; on average the number of inter-links is constant but links can change randomly over time). We would like to stress here that with respect to fashion there is no need to interpret links between individuals in a traditional way, i.e. in terms of friendship, collaboration or even acquaintance. In fact, we even do not need to know the person to be influence by him or her. Within such a general interpretation, links can denote the channels of potential influence, which justifies consideration of very large cliques and the annealed approach. Moreover, the annealed approach allows for analytical treatment, which greatly facilitates the verification of an agent-based model [48].

The main message of this paper is the following—interconnections between two unequal groups (snobs and followers) are not sufficient to produce recurring social cycles. Periodic behavior appears only above a certain critical value of interconnections, which is in a sense universal, i.e. has the same value for the fixed (quenched) and evolving (annealed) social networks. As usual, near the critical state, above which recurring cycles appear, changes in fashion are strongly irregular and therefore hard to predict.

Methods

We consider the following opinion formation model. There are 2N nodes divided into 2 cliques of size N. Each node is associated with one autonomous agent characterized at each time instant t by a dynamical binary variable S_i(t) = ±1. This variable, depending on the context, can be interpreted as a dichotomous opinion (agree/disagree) or a choice between two alternatives, e.g. old product vs. innovation, long vs. short skirt, Samsung vs. Apple, red or white wine [49]. Since agents are described by a single binary variable we will also refer to them as spinsons, a term coined by Nyczka and Sznajd-Weron [21] that reflects the dichotomous nature of an agent (spinson = spin + person).

As is common in the literature, for the macroscopic quantity that describes the state of the system at time t we choose the public opinion (alternatively interpreted as the ratio of adopted [30] or the market penetration rate [31, 49]): (1) Two cliques can interact with each other through cross links between them. The number of cross links is equal to LN², where the fraction of cross links L ∈ [0, 1] is a parameter of the model, see Fig 1. In order to clarify the description of the model let us introduce a second variable: σ_i = ±1. It defines to which of the two cliques the i-th spinson belongs. In contrast to the opinion, S_i(t), this variable is static, i.e. it does not change in time (σ_i = 1 if the i-th spinson belongs to the first group and σ_i = −1 if it belongs to the second clique). Within such a formulation we can separately calculate the opinions in each clique: (2) (3) where δ_j,k is the Kronecker delta, i.e. δ_j,k = 1 for k = j and 0 elsewhere. The summation in the above formulas is taken over the entire system, i.e. over 2N spinsons. However, the maximum value of each sum is N. This is guaranteed by the Kronecker delta: for m₁(t) only spinsons from the first clique will contribute to the sum and analogously for m₂(t) only spinsons from the second clique will contribute to the sum. Therefore we use factor 1/N in front of each sum.

Download:

Fig 1. An example of a system that consists of two interacting cliques (complete graphs).

The left clique (σ = 1) consists of “followers”, i.e. spinsons that behave like conformists with respect to both cliques, whereas the right clique (σ = −1) consists of “snobs”, i.e. individuals that behave like conformists with respect to their own group and anticonform to the other. In this example the cliques are of size N = 8, which means that the number of links inside each clique is equal to N(N − 1) = 56 and the number of all possible cross links is N² = 64. Moreover, the number of cross links is 5, which means that the fraction of cross links L = 5/64 = 0.078.

https://doi.org/10.1371/journal.pone.0166323.g001

In [22] it has been assumed that every spinson behaves like a conformist inside its own clique and anticonforms with respect to agents in the other group. Both types of social response have been described within the generalized q-voter model [21], which means that a spinson responds to the social influence only if the q-panel, which consists of q randomly chosen neighbors (spinsons that are linked to a given spinson), is unanimous. In case of conformity a spinson takes the same opinion as the q-panel, whereas in the case anticonformity it takes the opposite opinion.

This idea can be easily described using σ: if the i-th spinson is the one to be influenced (the so-called target in theories of social response [16, 32]) and the j-th spinson is the source of influence, then the i-th spinson conforms to opinion σ_i σ_j S_j(t), which equals S_j(t) if spinsons i and j belong to the same group (σ_i σ_j = 1) and −S_j(t), i.e. anticonformity, if they belong to different groups (σ_i σ_j = −1). The i-th spinson having opinion S_i(t) at time t is influenced by the group of q neighbors S₁(t), …, S_q(t) if the local average opinion: (4) If M_i(t) = 1 then S_i(t + 1) = 1 and if M_i(t) = −1 then S_i(t + 1) = −1.

Within the model proposed in [22] both cliques are equivalent. In this paper we break the symmetry between the cliques and assume that a spinson which belongs to group 1 (σ = 1) conforms to all other spinsons. This means that—using the notation from the previous paragraph—the i-th spinson always conforms to opinion S_j(t), not to σ_i σ_j S_j(t).

A model with quenched cross links

For each simulation we build a new network with the same statistical properties, i.e. two cliques of size N connected by LN² cross links. The network does not change in time, i.e. it is a “quenched network”. A single Monte Carlo update for the model with quenched links is described by the following algorithm (see Fig 2):

Randomly choose one spinson, say the i-th spinson, from the 2N spinsons in the system; goto 2.
Build the influence group by randomly choosing q spinsons from all those that are linked to the i-th spinson (q-panel): S₁, …, S_q; goto 3.
If the i-th spinson belongs to clique 1 then goto 4 else goto 5.
Calculate the average opinion of the q-panel influencing the i-th spinson: (5) goto 6.
Calculate the average opinion of the q-panel influencing the i-th spinson: (6) and goto 6.
If M_i(t) = 1 then S_i(t + 1) = 1, else if M_i(t) = −1 then S_i(t + 1) = −1, else nothing changes and goto 1.

Unless stated otherwise “a random choice” refers to a draw from a uniform distribution.

Download:

Fig 2. All possible transitions in the hunt opinion model for q = 2.

Social responses of an influenced “follower” (i.e. a spinson from the first clique; green color) are shown in the left panel, whereas the right panel presents the behavior of an influenced “snob” (i.e. a spinson from the second clique; red color). Only transitions that change the state of the system are shown.

https://doi.org/10.1371/journal.pone.0166323.g002

As usual, a single Monte Carlo step consists of 2N updates, i.e. τ = 2Nt, which means that one time unit corresponds to the mean update time of a single individual. As can be seen, we have a model in which spinsons from clique 1 follow others and therefore we call them “followers”. On the other hand, spinsons from clique 2 follow only spinsons from their own group and anticonform to agents from the other group and therefore we call them “snobs”. Within the original, symmetrical formulation in [22], the polarization was observed above the critical value of cross links, L*(q). The question is: how will the results change under the modification we make in this paper?

A model with annealed cross links

We will investigate the model with quenched links using only Monte Carlo simulations because for a quenched network we are not able to solve the model analytically. However, the theoretical description simplifies significantly if we replace the quenched network by an “annealed network”. In the latter, instead of having stable LN² links between the cliques, we introduce a probability, p, that a cross link exists. Within such a modification we are able to write down equations that describe the time evolution of the public opinion. But before doing this let us describe the algorithm for a single Monte Carlo update in the model with annealed links:

Randomly choose one spinson, say the i-th spinson, from the 2N spinsons in the system; goto 2.
The i-th spinson chooses q neighbors from the whole population (q-panel): S₁, …, S_q. Each of the q spinsons is chosen independently, i.e. a repeated choice of the same node is allowed (note, that the asymptotic results would not change if the q-panel consisted of q different spinsons). Nodes from the i-th spinson’s clique are chosen with probability 1 − p and nodes from the other clique are chosen with probability p; goto 3.
If the i-th spinson belongs to clique 1 then goto 4 else goto 5.
Calculate the average opinion of the q-panel influencing the i-th spinson: (7) goto 6.
Calculate the average opinion of the q-panel influencing the i-th spinson: (8) and goto 6.
If M_i(t) = 1 then S_i(t + 1) = 1, else if M_i(t) = −1 then S_i(t + 1) = −1, else nothing changes and goto 1.

For such a reformulated (annealed) model we are able to derive the limiting dynamical system for N → ∞ in scaled time . We first calculate transition probabilities and then derive the asymptotic dynamical system expectations. Let and be the number of nodes in state +1 at time t in clique i = 1, 2. Using the corresponding densities, , we can write down the transition probabilities:

For clique 1: (9)
and for clique 2: (10)

The factor in the above formulas arises from the fact that we have 2 cliques of equal size, hence the probability for choosing a node from one given clique is 1/2. We could also consider unequal clique sizes, but this would require an additional parameter and make the analysis more technical. Using probabilities Eqs (9) and (10) we can simulate trajectories c₁(t) and c₂(t), which can be easily converted to trajectories of the public opinion using the following relation: (11) Until now we have been dealing with random variables c_i(t), but we can also write the evolution equations of the corresponding expected values. For N → ∞ we can safely assume that random variables c_i(t) localize to the expectation values: (12) where . Taking the limit N → ∞, we get the following continuous time dynamical system: (13) The set of Eq (13) can be solved numerically. From the condition (14) we can obtain the stationary value of c₁ and c₂ and from Eq (13) the time evolution of these variables.

Results

We will present results for two models, each characterized by three parameters:

a model with quenched cross links, where L ∈ [0, 1] is the fraction of cross links, q is the number of agents in the group of influence (which varies from 2 to 7 in agreement with the empirical results [43, 44]) and N is the size of the clique (here we present results only for N = 10², 10³, 10⁴),
a model with annealed cross links, where p ∈ [0, 1] is the probability of a cross link and q and N are the same as above.

The model with quenched cross links will be investigated by Monte Carlo (MC) simulations and the model with annealed cross links by MC simulations using transition probabilities given by Eqs (9) and (10), as well as by numerical solutions of Eq (13).

It is worth noting, that parameter p can be easily related to L. In fact, from now on we will use parameter p for both models. Because p denotes the probability that a single individual will choose a neighbor from the other clique we can write: (15) where LN is the expected number of the neighbors from the other clique, and LN + N is a total number of neighbors for a single individual. Using relation Eq (15), results for both models can be easily compared if only p ≤ 0.5. For p > 0.5 the cross links are more probable than links inside the groups and L > 1, which is impossible if we assume that groups are represented by cliques (complete graphs) and we deal with unweighted links. However, it would be possible to introduce bias and assume that cross links have different weights than links inside a clique and then it would be possible to obtain p > 0.5. Again, this modification requires at least one additional parameter and makes the analysis more technical. Therefore, similarly as in the case of unequal cliques, we leave it for a future study.

We start with presenting trajectories, which represent how popularity—measured e.g. in terms of public opinion or market share of a given product—change over time. We investigate the system with two cliques of size N = 1000 and consider ordered initial conditions, i.e. m₁(0) = 1 and m₂(0) = 1, which corresponds to the situation in which everybody likes/uses the same type of product, e.g. all women wear short skirts. For the fixed size of the system both models are described only by two parameters q and p. As can be seen in Figs 3 and 4, the time evolution of the public opinion is similar in both models:

For a small probability of cross links, p < p*, both cliques are almost fully ordered in the same direction: m₁(t) > 0 and m₂(t) > 0. Moreover, opinion m₂(t) < m₁(t) fluctuates strongly around the average <m₂(t)> and m₁(t) is almost constant.
Above the critical probability of cross links, i.e. for p > p*, recurring cycles are observed. The period of the cycle decreases with p. Moreover, near the critical point the cycles are irregular and hence a change of fashion is difficult to predict.

Download:

Fig 3. Sample evolution of the opinion for quenched cross links.

Results for the snobs are presented by the red dashed lines and for the followers by the solid green lines. The probability of cross links, p, in each panel is different and increases from the top left panel to the right bottom: p = 0.15, 0.16, 0.16667, 0.17, 0.18, 0.19 (which corresponds to L = 0.176, 0.190, 0.200, 0.205, 0.220, 0.235). The size of each clique is N = 1000 for all panels.

https://doi.org/10.1371/journal.pone.0166323.g003

Download:

Fig 4. Sample evolution of opinion for annealed cross links.

Results for the snobs are presented by the red dashed lines and for the followers by the solid green lines. The probability of cross links, p, in each panel is different and increases from the top left panel to the right bottom: p = 0.15, 0.16, 0.16667, 0.17, 0.18, 0.19 (the same values as in Fig 3). The size of each clique is N = 1000 for all panels, as in Fig 3.

https://doi.org/10.1371/journal.pone.0166323.g004

There is a good theoretical reason, which needs more technical details and will not be discussed here, to believe that for N → ∞ the annealed variant of the model should converge to the quenched variant. Indeed, we can see from Figs 3 and 4 that threshold p* is roughly the same for both models. This can be seen clearly if we plot the time average of public opinion <m> as a function of p. For p > p* it should be zero, because the opinion is as often negative as positive. Indeed, we can see in Fig 5 that there is critical value of p = p*(q) above which recurring cycles appear. It is also seen that p*(q) increases with q, which has been also observed for the one-clique version of the q-voter with anticonformity [24].

Download:

Fig 5. Dependence between the average stationary opinion and the fraction of cross links.

Average opinion as a function of the cross links in the stationary state for the followers (left panels) and the snobs (right panels) for the model with quenched cross links (upper panels) and the model with annealed cross links (bottom panels). These results do not depend on the system size and have been found to be the same for N = 10², 10³, 10⁴.

https://doi.org/10.1371/journal.pone.0166323.g005

To see the difference between the models we have to observe the shapes of the trajectories—immediately we can see that the period of the cycle for the model with quenched cross links (Fig 3) is significantly smaller than for the model with annealed cross links (Fig 4). To measure this difference quantitatively we can calculate the average time of the cycle. Near the critical point (see the top left and bottom right panels in Figs 3 and 4), the cycles are strongly irregular, which means that the period varies. If we measure the period, i.e. T such that m(t) = m(t + T), for the first cycle it will be significantly different than the period for the next cycle. However, it is seen that for larger values of p the period is almost constant and therefore the average, <T>, is a good measure of the rate of change.

In Fig 6 it can be seen that period T decreases with p and is shorter for the model with quenched cross links, if only p is distinctly larger than p*. These features have been already seen qualitatively in Figs 3 and 4. This is an interesting result from the social point of view, because it seems that presently social networks are more annealed than in the past. Nowadays we use the Internet and social media, which may result in an increasing randomness of cross links. According to our model, this should result in slower recurring fashion cycles (larger T). On the other hand, we may now expect faster recurring fashion cycles (smaller T), because the probability of interactions between communities, p, is now larger than in the past. Thus, there is a competition of two factors—more changes and randomness in the social network (annealing factor) increases the period of a fashion cycle, but a larger number of connections between communities decreases T. The question is which of the two factors wins.

Download:

Fig 6. The average time of a cycle.

Results for the model with quenched (left panel) and with annealed cross links (right panel).

https://doi.org/10.1371/journal.pone.0166323.g006

There are several more or less reliable hypotheses, including famous Laver’s law of fashion [50] or pop culture’s 40-year cycle of nostalgia, that suggest a constant value of T. If indeed T is constant then all forces/factors influencing the recurring fashion cycle must act contrary to each other and balance out. However, all these observations are merely speculations. We do not have a firm empirical evidence that presently social networks are more annealed and p increases over time. Moreover, we have not found any reliable long-time empirical data that would clearly answer whether the period of the fashion cycle decreases, increases or remains constant. Therefore collecting and analyzing relevant empirical data is the challenge for the future.

We have to note, that period T is shorter for the quenched model only for p distinctly larger than p* (compare Figs 3 and 4; see also Fig 6). However, for p close to p* the period is shorter for the annealed model. Therefore, if the social system is near the critical state we should expect more frequent changes for the annealed network. Moreover, for p → p* changes in fashion are strongly irregular and therefore hard to predict. In such a case it is even hard to speak about any period of a cycle.

Additional comments on the annealed model

As we have already mentioned, it makes no sense to consider p > 0.5 if we use a model without any bias, i.e. when all links (cross links and intra-links) have the same weight. However, if we are interested in the annealed model itself, we can consider any value of p ∈ [0, 1], which can be instructive, particularly if we take into account that present social networks are not as stable is in the past.

The annealed model has a big advantage over the quenched model—it is much simpler for analytical treatment. Here we present results of numerical solutions of Eq (13) that were obtained using the fourth order Runge-Kutta scheme. Until now we have discussed only one type of the initial conditions but using the flow diagrams in the phase space we are able to see the behavior for arbitrary initial conditions (see Fig 7).

Download:

Fig 7. The phase portraits for the model with annealed cross links and q = 2.

Blue arrows indicate the direction of flows in the phase plane and the red line is a sample average trajectory from disordered initial conditions c₁(0) = c₂(0) = 0.5001. The probability of a cross link, p, in each panel is different and increases from the top left to the bottom right panel: p = 0.08, 0.10, 0.12, 0.14, 0.16, 0.18, 0.20, 0.24, 0.30, 0.40, 0.50, 0.60. The size of each clique is N = 10⁴ for all panels.

https://doi.org/10.1371/journal.pone.0166323.g007

From Fig 7 we can see that indeed we have a critical value of p above which the limit cycle appears and the period decreases with p. However, additionally we see that below the critical point the system can reach different fixed points depending on the initial conditions. We conjecture on topological grounds and the structure of the vector field plots (Fig 7) that the saddle node heteroclinic bifurcation appears at p* [51]. It is not easy to see this in Fig 7, therefore we show schematically the behavior of the system around p = p* in Fig 8.

Download:

Fig 8. Schematic representation of the system behavior near the critical point in the phase plane.

The situation for p < p* is shown in the left panel. There are 8 fixed points: 4 stable (black) and 4 hyperbolic saddle points (red). Moreover, we have an additional unstable point at (0.5, 0.5); A1-A4 are the basins of attraction. In the right panel we see the system at the critical point p = p*. The saddle and stable fixed points merge, hence we have 4 fixed points and a so-called heteroclinic cycle. For p > p* the fixed points disappear and only the stable limit cycle is left.

https://doi.org/10.1371/journal.pone.0166323.g008

For p < p* (see the left panel in Fig 8) there are 8 fixed points (4 stable ones and 4 hyperbolic saddle points). At the bifurcation point p = p* (see the right panel in Fig 8) the unstable and stable fixed points merge, hence one has 4 fixed points and a so-called heteroclinic cycle. For p > p* the fixed points disappear and only the stable limit cycle is left. These type of bifurcations appear generically in one parameter families of flows.

Because the configuration space may be more intuitive for many readers, we additionally present the results from Fig 7 in the (c_i, t) plane; see Fig 9. In this representation we lose the information about the role of initial conditions. Instead interesting phenomena are clearly seen for large values of p—fading cycles appear and the system eventually reaches a disordered fixed point, c₁ = c₂ = 0.5, in which both products/opinions are equally popular in the system. In fact, the same phenomenon could be seen also in Fig 7, but a more zoomed-in picture would be needed (pay attention to the scale in Figs 7 and 9).

Download:

Fig 9. Average trajectories for the model with annealed cross links and q = 2.

Results for snobs are presented by the red dashed lines and for the followers by the solid green lines. The probability of cross links, p, is different in each panel and increases from the top left to the bottom right panel: p = 0.08, 0.10, 0.12, 0.14, 0.16, 0.18, 0.20, 0.24, 0.30, 0.40, 0.50, 0.60. The size of each clique N = 10⁴ for all panels.

https://doi.org/10.1371/journal.pone.0166323.g009

This behavior means that there is another critical point below which the system reaches a limit cycle and above which a stable fixed point . The appearance of an attractive limit cycle of a smooth dynamical system is actually the result of a Hopf bifurcation at the fixed point , which takes place for the critical value In the special case of a 2 dimensional system a Hopf bifurcation occurs when a fixed point—in our case the one at − looses stability at a critical parameter value in such a way that the linearized system at the fixed point has two complex conjugate eigenvalues. These eigenvalues are purely imaginary at the bifurcation value and have positive/negative real part above/below the bifurcation value. Furthermore, some non-degeneracy conditions have to be met; for details see [52]. To check if the conditions of Hopf’s theorem apply we first compute the Jacobian matrix at the fixed point

Recall that the dynamics is given by Eq (13). Some straightforward computation gives for Jacobian J at the following matrix: (16) which is of the form (17) and the eigenvalues are a − ib, a + ib. For a Hopf bifurcation to happen we need two conjugate purely imaginary eigenvalues, hence a = 0. This is equivalent to q − 1 − qp = 0 and (18) It is easy to check that all conditions of Hopf’s theorem are met and hence the loss of stability of the central fixed point creates an attractive cycle.

The above results mean that for p < p_bif(q) a fixed point is repelling and the system reaches a limit cycle, whereas for p > p_bif(q) a fixed point is attracting and the system eventually reaches this point. This is clearly seen in Fig 10. For p = p_bif − ϵ the system reaches a limit cycle and for p = p_bif − ϵ the system reaches a fixed point . Moreover, values of p_bif agree with the derived analytical Formula (18), i.e. , , , .

Download:

Fig 10. Average trajectories for the model with annealed cross links and several values of q.

Each column corresponds to a different value of q, from left to right q = 2, 3, 4, 5. Results for the snobs are presented by the red dashed lines and for the followers by the solid green lines. Upper panels correspond to the probability of cross links p = p_bif(q) − ϵ, middle panels to p = p_bif(q) and bottom panels to p = p_bif(q) + ϵ, where ϵ = 0.01. Note the difference in ranges on the vertical axis that corresponds to c—in the upper panels the range is much larger, i.e. [0, 1] than in the middle and bottom panels. Values of p_bif, agree with the derived analytical Formula (18), i.e. (first column), (second column), (third column), (fourth column). The size of each clique is N = 10⁴ for all panels.

https://doi.org/10.1371/journal.pone.0166323.g010

Conclusions

With some exceptions [12, 13], in most models of fashion and fads two types of social response are taken into account—conformity and anticonformity, sometimes referred to as imitation and distinction [9] or coordination and anti-coordination [53]. Some models take into account also the influence of certain exogenous factors, like popularity of the product [54]. We are aware that exogenous factors including popularity or advertisement are important in creating fashion, but it seems that they are not necessary, what has been shown in many other models [12, 13, 53]. Our aim was not to propose the final and complete model of the recurring fashion cycles but rather to show that a small modification of the model without cycles [22] can lead to the appearance of a periodic behavior. Moreover, we wanted to examine what is the role of stable connections between cliques. Therefore we have considered two versions of the same model—quenched (parameterized by fraction L of fixed inter-links) and annealed (parameterized by probability p of inter-links). For p ≤ 0.5 we have derived the relation between parameters L and p that allows to compare both models and show that the critical value of inter-connections is the same for the quenched and annealed schemes. However, the period of the cycles has been found to be shorter for the quenched model, i.e. for static social networks, if p is distinctly larger than p*. On the other hand, if p is close to p* then the period is shorter for the annealed model. Moreover, near the critical point the cycles are irregular and a change of fashion is difficult to predict.

In case of the annealed model we were able to write down equations describing the evolution of the system and provide deeper theoretical analysis. We conjectured on topological grounds that the so-called saddle node heteroclinic bifurcation appears at p*. Moreover, we have shown analytically that for p ≤ 0.5 there is another critical point below which the system reaches a limit cycle and above which a stable fixed point . The appearance of an attractive limit cycle of the smooth dynamical system is actually the result of a Hopf bifurcation at the fixed point , which takes place for the critical value . This result means that if the influence from the other clique is large enough, i.e. p > p_bif, there is a possibility of a stale-mate state, in which there is no fashion.

We are aware that bringing the model closer to reality would require further development. For example, one could consider cliques of different sizes. Another modification, also mentioned in the paper, could be to assign weights to links (for example ω for cross links and 1 − ω for intra-links). Within such a modification, it makes sense to compare both models also for p > 0.5. However, both modifications require additional parameters and would make the analysis more technical. Moreover, they do not provide many more insights. Therefore, for the sake of clarity of presentation, they were left for a more specialized journal.

Author Contributions

Conceptualization: TK KSW.
Formal analysis: TK.
Investigation: RA GM.
Methodology: TK.
Software: RA GM.
Supervision: KSW.
Validation: RA GM.
Visualization: RA.
Writing – original draft: KSW.
Writing – review & editing: RA TK GM KSW.

References

1. Pareto V. The Mind and Society, translated from Trattato di Sociologia Generale (1916) Arthur Livingston ed. Harcourt 1935
2. Richardson J, Kroeber AL. Three centuries of women’s dress fashions a quantitative analysis. Anthropological Records 5:2, University Of California Press, Berkeley and Los Angeles 1940.
3. Robinson DE. Fashions in Shaving and Trimming of the Beard: The Men of the Illustrated London News, 1842-1972 American Journal of Sociology 1976 Mar; 81(5):1133–1141.
- View Article
- Google Scholar
4. Belleau BD Cyclical fashion movement: Women’s day dresses: 1860–1980. Clothing and Textiles Research Journal 1987; 5:15–20.
- View Article
- Google Scholar
5. Curran L. An analysis of cycles of skirt lengths and widths in the UK and Germany, 1954–1990. Clothing and Textiles Research Journal 1999; 17(2):65–72.
- View Article
- Google Scholar
6. Callahan G, Hoffmann A. (2015) Two-Population Social Cycle Theories Available at SSRN: http://ssrn.com/abstract=2560270 or http://dx.doi.org/10.2139/ssrn.2560270.
7. Smith A. The Theory of Moral Sentiments 1790. The Theory of Moral Sentiments. Library of Economics and Liberty. 28 October 2012. http://www.econlib.org/library/Smith/smMS5.html.
8. Simmel G. Fashion International Quarterly 1904 Oct; 10(1): 130–155 reprinted in American Journal of Sociology 1957 May; 62(6):541–558.
- View Article
- Google Scholar
9. Benvenuto S. Fashion: Georg Simmel Journal of Artificial Societies and Social Simulation 2000 Mar; 3(2).
- View Article
- Google Scholar
10. Galak J, Gray K, Elbert I, Strohminger N. Trickle-Down Preferences: Preferential Conformity to High Status Peers in Fashion Choices. PLoS ONE 2016; 11(5): e0153448 pmid:27144595
- View Article
- PubMed/NCBI
- Google Scholar
11. Berger J, Heat C. Who drives divergence? identity-signaling, outgroup dissimilarity, and the abandonment of cultural tastes. Journal of Personality and Social Psychology 2008; 95(3):593–607. pmid:18729697
- View Article
- PubMed/NCBI
- Google Scholar
12. Acerbi A, Ghirlanda S, Enquist M. The Logic of Fashion Cycles. PLoS ONE 2012; 7(3): e32541 pmid:22412887
- View Article
- PubMed/NCBI
- Google Scholar
13. Bentley RA, Lipo CP, Herzog HA, Hahn M Regular rates of popular culture change reflect random copying. Evolution and Human Behavior 2007; 28:151–158.
- View Article
- Google Scholar
14. Miller CM, Mcintyre SH, Mantrala MK. Toward formalizing fashion theory. Journal of Market Research 1993; 81(5):1133–1141.
- View Article
- Google Scholar
15. Krawczyk MJ, Dydejczyk A, Kułakowski K. The Simmel effect and babies’ names Physica A 2014; 395:384–391.
- View Article
- Google Scholar
16. Nail PR, Di Domenico SI, MacDonald G. Proposal of a double diamond model of social response. Review of General Psychology 2013; 17:1–19.
- View Article
- Google Scholar
17. Castellano C, Fortunato S, Loreto V, Statistical physics of social dynamics. Rev. Mod. Phys. 81, 591–646 (2009)
- View Article
- Google Scholar
18. Galam S, Sociophysics: A Physicist’s Modeling of Psycho-political Phenomena. New York Springer (2012)
19. Sornette D, Physics and financial economics (1776–2014): puzzles, Ising and agent-based models. Reports on Progress in Physics 2014; 77:062001. pmid:24875470
- View Article
- PubMed/NCBI
- Google Scholar
20. Galam S, Contrarian deterministic effects on opinion dynamics: the hung election scenario. Physica A 2004; 333:453–460
- View Article
- Google Scholar
21. Nyczka P, Sznajd-Weron K. Anticonformity or Independence? -Insights from Statistical Physics. Journal of Statistical Physics 2013;151:174–202.
- View Article
- Google Scholar
22. Siedlecki P, Szwabiński J, Weron T. The interplay between conformity and anticonformity and its polarizing effect on society. Journal of Artificial Societies and Social Simulation 2016; 19(4); http://jasss.soc.surrey.ac.uk/19/4/9.html. )
- View Article
- Google Scholar
23. Castellano C, Munoz MA, Pastor-Satorras R. Nonlinear q-voter model. Physical Review E 2009; 80:041129.
- View Article
- Google Scholar
24. Nyczka P, Sznajd-Weron K, Cislo J. Phase transitions in the q-voter model with two types of stochastic driving. Phys. Rev. E 2012; 86: 011105. pmid:23005366
- View Article
- PubMed/NCBI
- Google Scholar
25. Timpanaro AM, Prado CPC Exit probability of the one-dimensional q-voter model: Analytical results and simulations for large networks Physical Review E 2014; 89(5):052808
- View Article
- Google Scholar
26. Timpanaro AM, Galam S Analytical expression for the exit probability of the q-voter model in one dimension Physical Review E 2015; 92(1):012807
- View Article
- Google Scholar
27. Mobilia M Nonlinear q -voter model with inflexible zealots Physical Review E 2015; 92(1):012803
- View Article
- Google Scholar
28. Mellor A, Mobilia M, Zia RKP, Characterization of the nonequilibrium steady state of a heterogeneous nonlinear q-voter model with zealotry EPL 2016; 113(4):48001
- View Article
- Google Scholar
29. Jedrzejewski A Pair approximation for the q-voter model with independence on complex networks arXiv:1607.07307 [physics.soc-ph]
30. Kowalska-Pyzalska A, Maciejowska K, Suszczynski K, Sznajd-Weron K, Weron R. Turning green: Agent-based modeling of the adoption of dynamic electricity tariffs. Energy Policy 2014; 72:164–174
- View Article
- Google Scholar
31. Maciejowska K, Jedrzejewski A, Kowalska-Pyzalska A, Weron R Impact of social interactions on demand curves for innovative products Acta Physica Polonica A 2016; 129(5): 1045–1049
- View Article
- Google Scholar
32. Nail PR, Sznajd-Weron K. The Diamond Model of Social Response within an Agent-Based Approach Acta Physica Polonica A 2016; 129(5):1050–1054
- View Article
- Google Scholar
33. Byrka K, Jedrzejewski A, Sznajd-Weron K, Weron R. Difficulty is critical: The importance of social factors in modeling diffusion of green products and practices. Renewable and Sustainable Energy Reviews 2016; 62:723–735
- View Article
- Google Scholar
34. Elliott R Phenomenological discussion of magnetic ordering in the heavy rare-earth metals Phys. Rev. 1961; 124 (2): 346–353
- View Article
- Google Scholar
35. Bak P Commensurate phases, incommensurate phases, and the devil’s staircase Reports on Progress in Physics 1982; 45 (6): 587–629
- View Article
- Google Scholar
36. Selke W The ANNNI model—Theoretical analysis and experimental application Physics Reports 1988; 170 (4): 213–264
- View Article
- Google Scholar
37. Suchecki K, Hołyst JA Ising model on two connected Barabasi-Albert networks Physical Review E 2006; 74(1):011122 pmid:16907075
- View Article
- PubMed/NCBI
- Google Scholar
38. Suchecki K, Hołyst JA Bistable-monostable transition in the Ising model on two connected complex networks Physical Review E 2009; 80(3):031110 pmid:19905065
- View Article
- PubMed/NCBI
- Google Scholar
39. Sood V, Antal T, Redner S. Voter models on heterogeneous networks. Physical Review E 2008; 77: 041121
- View Article
- Google Scholar
40. Newman MEJ. Fast algorithm for detecting community structure in networks. Phys. Rev. E 2004; 69:066133
- View Article
- Google Scholar
41. Newman MEJ. Modularity and community structure in networks. PNAS 2006 Jun; 103(23):8577–8582 pmid:16723398
- View Article
- PubMed/NCBI
- Google Scholar
42. Lancichinetti A, Kivela K, Saramaki J, Fortunato S. Characterizing the community structure of complex networks. PLoS One 2010; 5(8):e11976 pmid:20711338
- View Article
- PubMed/NCBI
- Google Scholar
43. Desportes JP, Lemaine JM. The sizes of human groups: An analysis of their distributions. In Canter D, Jesuino JC, Soczka L, Stephenson GM (Eds.), Environmental social psychology (pp. 57–65). Dordrecht, Netherlands: Kluwer Academic (1988) https://doi.org/10.1007/978-94-009-2802-2_5
44. Ruback RB, Collins RT, Koon-Magnin S, Ge W, Bonkiewicz L, Lutz CE. People transitioning across places: a multimethod investigation of how people goto football games Environment and Behavior 2013; 45(2):239–66.
- View Article
- Google Scholar
45. Bond R. Group size and conformity. Group Process Intergroup Relat 2005;8:331–54.
- View Article
- Google Scholar
46. Murray JD. Mathematical Biology Springer-Verlag 1989, Heidelberg.
47. Szwabiński J. Density outbursts in a food web model with closed nutrient cycle Physica A 2013; 392:3589–3600.
- View Article
- Google Scholar
48. Rand W, Rust RT. Agent-based modeling in marketing: Guidelines for rigor J. of Research in Marketing 2011 Jun; 28(3):181–193.
- View Article
- Google Scholar
49. Bouzdine-Chameeva T, Galam S. Word-of-mouth versus experts and reputation in the individual dynamics of wine purchasing. Adv. Complex Syst. 2011; 14:871–885.
- View Article
- Google Scholar
50. Steele V. Encyclopedia of clothing and fashion Charles Scribner’s Sons, Thomson Gale 2005
51. Shilnikov LP, Shilnikov AL, Turaev DV, Chua LO. Methods of Qualitative Theory in Nonlinear Dynamics. Part II Singapore 2001: World Scientific.
52. Marsden J, McCracken M. Hopf Bifurcation and its Applications. Springer-Verlag, New York 1976 https://doi.org/10.1007/978-1-4612-6374-6
53. Cao Z, Gao H, Qu X, Yang M, Yang X. Fashion, Cooperation, and Social Interactions. PLoS ONE 2013; 8(1) pmid:23382799
- View Article
- PubMed/NCBI
- Google Scholar
54. Denrell J, Kovacs B The Effect of Selection Bias in Studies of Fads and Fashions. PLoS ONE 2015; 10(4): e0123471 pmid:25886158
- View Article
- PubMed/NCBI
- Google Scholar

[ref1] 1. Pareto V. The Mind and Society, translated from Trattato di Sociologia Generale (1916) Arthur Livingston ed. Harcourt 1935

[ref2] 2. Richardson J, Kroeber AL. Three centuries of women’s dress fashions a quantitative analysis. Anthropological Records 5:2, University Of California Press, Berkeley and Los Angeles 1940.

[ref3] 3. Robinson DE. Fashions in Shaving and Trimming of the Beard: The Men of the Illustrated London News, 1842-1972 American Journal of Sociology 1976 Mar; 81(5):1133–1141.
View Article
Google Scholar

[4] View Article

[5] Google Scholar

[ref4] 4. Belleau BD Cyclical fashion movement: Women’s day dresses: 1860–1980. Clothing and Textiles Research Journal 1987; 5:15–20.
View Article
Google Scholar

[7] View Article

[8] Google Scholar

[ref5] 5. Curran L. An analysis of cycles of skirt lengths and widths in the UK and Germany, 1954–1990. Clothing and Textiles Research Journal 1999; 17(2):65–72.
View Article
Google Scholar

[10] View Article

[11] Google Scholar

[ref6] 6. Callahan G, Hoffmann A. (2015) Two-Population Social Cycle Theories Available at SSRN: http://ssrn.com/abstract=2560270 or http://dx.doi.org/10.2139/ssrn.2560270.

[ref7] 7. Smith A. The Theory of Moral Sentiments 1790. The Theory of Moral Sentiments. Library of Economics and Liberty. 28 October 2012. http://www.econlib.org/library/Smith/smMS5.html.

[ref8] 8. Simmel G. Fashion International Quarterly 1904 Oct; 10(1): 130–155 reprinted in American Journal of Sociology 1957 May; 62(6):541–558.
View Article
Google Scholar

[15] View Article

[16] Google Scholar

[ref9] 9. Benvenuto S. Fashion: Georg Simmel Journal of Artificial Societies and Social Simulation 2000 Mar; 3(2).
View Article
Google Scholar

[18] View Article

[19] Google Scholar

[ref10] 10. Galak J, Gray K, Elbert I, Strohminger N. Trickle-Down Preferences: Preferential Conformity to High Status Peers in Fashion Choices. PLoS ONE 2016; 11(5): e0153448 pmid:27144595
View Article
PubMed/NCBI
Google Scholar

[21] View Article

[22] PubMed/NCBI

[23] Google Scholar

[ref11] 11. Berger J, Heat C. Who drives divergence? identity-signaling, outgroup dissimilarity, and the abandonment of cultural tastes. Journal of Personality and Social Psychology 2008; 95(3):593–607. pmid:18729697
View Article
PubMed/NCBI
Google Scholar

[25] View Article

[26] PubMed/NCBI

[27] Google Scholar

[ref12] 12. Acerbi A, Ghirlanda S, Enquist M. The Logic of Fashion Cycles. PLoS ONE 2012; 7(3): e32541 pmid:22412887
View Article
PubMed/NCBI
Google Scholar

[29] View Article

[30] PubMed/NCBI

[31] Google Scholar

[ref13] 13. Bentley RA, Lipo CP, Herzog HA, Hahn M Regular rates of popular culture change reflect random copying. Evolution and Human Behavior 2007; 28:151–158.
View Article
Google Scholar

[33] View Article

[34] Google Scholar

[ref14] 14. Miller CM, Mcintyre SH, Mantrala MK. Toward formalizing fashion theory. Journal of Market Research 1993; 81(5):1133–1141.
View Article
Google Scholar

[36] View Article

[37] Google Scholar

[ref15] 15. Krawczyk MJ, Dydejczyk A, Kułakowski K. The Simmel effect and babies’ names Physica A 2014; 395:384–391.
View Article
Google Scholar

[39] View Article

[40] Google Scholar

[ref16] 16. Nail PR, Di Domenico SI, MacDonald G. Proposal of a double diamond model of social response. Review of General Psychology 2013; 17:1–19.
View Article
Google Scholar

[42] View Article

[43] Google Scholar

[ref17] 17. Castellano C, Fortunato S, Loreto V, Statistical physics of social dynamics. Rev. Mod. Phys. 81, 591–646 (2009)
View Article
Google Scholar

[45] View Article

[46] Google Scholar

[ref18] 18. Galam S, Sociophysics: A Physicist’s Modeling of Psycho-political Phenomena. New York Springer (2012)

[ref19] 19. Sornette D, Physics and financial economics (1776–2014): puzzles, Ising and agent-based models. Reports on Progress in Physics 2014; 77:062001. pmid:24875470
View Article
PubMed/NCBI
Google Scholar

[49] View Article

[50] PubMed/NCBI

[51] Google Scholar

[ref20] 20. Galam S, Contrarian deterministic effects on opinion dynamics: the hung election scenario. Physica A 2004; 333:453–460
View Article
Google Scholar

[53] View Article

[54] Google Scholar

[ref21] 21. Nyczka P, Sznajd-Weron K. Anticonformity or Independence? -Insights from Statistical Physics. Journal of Statistical Physics 2013;151:174–202.
View Article
Google Scholar

[56] View Article

[57] Google Scholar

[ref22] 22. Siedlecki P, Szwabiński J, Weron T. The interplay between conformity and anticonformity and its polarizing effect on society. Journal of Artificial Societies and Social Simulation 2016; 19(4); http://jasss.soc.surrey.ac.uk/19/4/9.html. )
View Article
Google Scholar

[59] View Article

[60] Google Scholar

[ref23] 23. Castellano C, Munoz MA, Pastor-Satorras R. Nonlinear q-voter model. Physical Review E 2009; 80:041129.
View Article
Google Scholar

[62] View Article

[63] Google Scholar

[ref24] 24. Nyczka P, Sznajd-Weron K, Cislo J. Phase transitions in the q-voter model with two types of stochastic driving. Phys. Rev. E 2012; 86: 011105. pmid:23005366
View Article
PubMed/NCBI
Google Scholar

[65] View Article

[66] PubMed/NCBI

[67] Google Scholar

[ref25] 25. Timpanaro AM, Prado CPC Exit probability of the one-dimensional q-voter model: Analytical results and simulations for large networks Physical Review E 2014; 89(5):052808
View Article
Google Scholar

[69] View Article

[70] Google Scholar

[ref26] 26. Timpanaro AM, Galam S Analytical expression for the exit probability of the q-voter model in one dimension Physical Review E 2015; 92(1):012807
View Article
Google Scholar

[72] View Article

[73] Google Scholar

[ref27] 27. Mobilia M Nonlinear q -voter model with inflexible zealots Physical Review E 2015; 92(1):012803
View Article
Google Scholar

[75] View Article

[76] Google Scholar

[ref28] 28. Mellor A, Mobilia M, Zia RKP, Characterization of the nonequilibrium steady state of a heterogeneous nonlinear q-voter model with zealotry EPL 2016; 113(4):48001
View Article
Google Scholar

[78] View Article

[79] Google Scholar

[ref29] 29. Jedrzejewski A Pair approximation for the q-voter model with independence on complex networks arXiv:1607.07307 [physics.soc-ph]

[ref30] 30. Kowalska-Pyzalska A, Maciejowska K, Suszczynski K, Sznajd-Weron K, Weron R. Turning green: Agent-based modeling of the adoption of dynamic electricity tariffs. Energy Policy 2014; 72:164–174
View Article
Google Scholar

[82] View Article

[83] Google Scholar

[ref31] 31. Maciejowska K, Jedrzejewski A, Kowalska-Pyzalska A, Weron R Impact of social interactions on demand curves for innovative products Acta Physica Polonica A 2016; 129(5): 1045–1049
View Article
Google Scholar

[85] View Article

[86] Google Scholar

[ref32] 32. Nail PR, Sznajd-Weron K. The Diamond Model of Social Response within an Agent-Based Approach Acta Physica Polonica A 2016; 129(5):1050–1054
View Article
Google Scholar

[88] View Article

[89] Google Scholar

[ref33] 33. Byrka K, Jedrzejewski A, Sznajd-Weron K, Weron R. Difficulty is critical: The importance of social factors in modeling diffusion of green products and practices. Renewable and Sustainable Energy Reviews 2016; 62:723–735
View Article
Google Scholar

[91] View Article

[92] Google Scholar

[ref34] 34. Elliott R Phenomenological discussion of magnetic ordering in the heavy rare-earth metals Phys. Rev. 1961; 124 (2): 346–353
View Article
Google Scholar

[94] View Article

[95] Google Scholar

[ref35] 35. Bak P Commensurate phases, incommensurate phases, and the devil’s staircase Reports on Progress in Physics 1982; 45 (6): 587–629
View Article
Google Scholar

[97] View Article

[98] Google Scholar

[ref36] 36. Selke W The ANNNI model—Theoretical analysis and experimental application Physics Reports 1988; 170 (4): 213–264
View Article
Google Scholar

[100] View Article

[101] Google Scholar

[ref37] 37. Suchecki K, Hołyst JA Ising model on two connected Barabasi-Albert networks Physical Review E 2006; 74(1):011122 pmid:16907075
View Article
PubMed/NCBI
Google Scholar

[103] View Article

[104] PubMed/NCBI

[105] Google Scholar

[ref38] 38. Suchecki K, Hołyst JA Bistable-monostable transition in the Ising model on two connected complex networks Physical Review E 2009; 80(3):031110 pmid:19905065
View Article
PubMed/NCBI
Google Scholar

[107] View Article

[108] PubMed/NCBI

[109] Google Scholar

[ref39] 39. Sood V, Antal T, Redner S. Voter models on heterogeneous networks. Physical Review E 2008; 77: 041121
View Article
Google Scholar

[111] View Article

[112] Google Scholar

[ref40] 40. Newman MEJ. Fast algorithm for detecting community structure in networks. Phys. Rev. E 2004; 69:066133
View Article
Google Scholar

[114] View Article

[115] Google Scholar

[ref41] 41. Newman MEJ. Modularity and community structure in networks. PNAS 2006 Jun; 103(23):8577–8582 pmid:16723398
View Article
PubMed/NCBI
Google Scholar

[117] View Article

[118] PubMed/NCBI

[119] Google Scholar

[ref42] 42. Lancichinetti A, Kivela K, Saramaki J, Fortunato S. Characterizing the community structure of complex networks. PLoS One 2010; 5(8):e11976 pmid:20711338
View Article
PubMed/NCBI
Google Scholar

[121] View Article

[122] PubMed/NCBI

[123] Google Scholar

[ref43] 43. Desportes JP, Lemaine JM. The sizes of human groups: An analysis of their distributions. In Canter D, Jesuino JC, Soczka L, Stephenson GM (Eds.), Environmental social psychology (pp. 57–65). Dordrecht, Netherlands: Kluwer Academic (1988) https://doi.org/10.1007/978-94-009-2802-2_5

[ref44] 44. Ruback RB, Collins RT, Koon-Magnin S, Ge W, Bonkiewicz L, Lutz CE. People transitioning across places: a multimethod investigation of how people goto football games Environment and Behavior 2013; 45(2):239–66.
View Article
Google Scholar

[126] View Article

[127] Google Scholar

[ref45] 45. Bond R. Group size and conformity. Group Process Intergroup Relat 2005;8:331–54.
View Article
Google Scholar

[129] View Article

[130] Google Scholar

[ref46] 46. Murray JD. Mathematical Biology Springer-Verlag 1989, Heidelberg.

[ref47] 47. Szwabiński J. Density outbursts in a food web model with closed nutrient cycle Physica A 2013; 392:3589–3600.
View Article
Google Scholar

[133] View Article

[134] Google Scholar

[ref48] 48. Rand W, Rust RT. Agent-based modeling in marketing: Guidelines for rigor J. of Research in Marketing 2011 Jun; 28(3):181–193.
View Article
Google Scholar

[136] View Article

[137] Google Scholar

[ref49] 49. Bouzdine-Chameeva T, Galam S. Word-of-mouth versus experts and reputation in the individual dynamics of wine purchasing. Adv. Complex Syst. 2011; 14:871–885.
View Article
Google Scholar

[139] View Article

[140] Google Scholar

[ref50] 50. Steele V. Encyclopedia of clothing and fashion Charles Scribner’s Sons, Thomson Gale 2005

[ref51] 51. Shilnikov LP, Shilnikov AL, Turaev DV, Chua LO. Methods of Qualitative Theory in Nonlinear Dynamics. Part II Singapore 2001: World Scientific.

[ref52] 52. Marsden J, McCracken M. Hopf Bifurcation and its Applications. Springer-Verlag, New York 1976 https://doi.org/10.1007/978-1-4612-6374-6

[ref53] 53. Cao Z, Gao H, Qu X, Yang M, Yang X. Fashion, Cooperation, and Social Interactions. PLoS ONE 2013; 8(1) pmid:23382799
View Article
PubMed/NCBI
Google Scholar

[145] View Article

[146] PubMed/NCBI

[147] Google Scholar

[ref54] 54. Denrell J, Kovacs B The Effect of Selection Bias in Studies of Fads and Fashions. PLoS ONE 2015; 10(4): e0123471 pmid:25886158
View Article
PubMed/NCBI
Google Scholar

[149] View Article

[150] PubMed/NCBI

[151] Google Scholar

Figures

Abstract

Introduction

Methods

A model with quenched cross links

A model with annealed cross links

Results

Additional comments on the annealed model

Conclusions

Author Contributions

References