2.1 PCA and the contact network
Individuals live in a square matrix formed by
cells with periodic boundary conditions (the top edge of the matrix contacts the bottom edge, and the right-hand edge contacts the left-hand edge – the toroidal surface). A network model, used by Schimit and Monteiro,
40 where each cell of the PCA lattice corresponds to an individual and an edge between two cells represents a social contact, configures the neighborhood of each cell. Each cell has social contact with a neighborhood defined as the square matrix of size
centered on such a cell, where
r is the maximum radius in which a connection can be made. Each cell makes
m connections with other cells pertaining to its neighborhood matrix. The cells with a Moore radius equal to
i form the layer
i. The probability
of creating a connection between a cell and any cell pertaining to the layer
i of its neighborhood matrix is given by
, where
. Consider the case
, where we have
,
, and
, that is, the connection probability between a cell and any of the eight surrounding cells composing layer 1 is 50%, between a cell and any of the 16 cells composing layer 2 is 33%, and so on. Two or more connections between the same two cells are allowed. Such a random network has connections mainly locally connected and presents a “high” clustering coefficient and “small” average shortest path length, like a human social contacts network.
40,41 Here, PCA is modeled using a lattice of
n = 200,
N = 40,000 cells (individuals). Each individual connects to
neighbors inside a radius
.
2.2 Consumer market in PCA
Individuals must decide if they will buy certain product. In this stage, his status is Consumer. He may buy a product due to being influenced by neighbors, marketing, and external situation (good and bad economic situation). From the moment he buys the product, he becomes a Buyer. In this state, he may influence his neighbors about their decision to buy or not. If a Buyer has not been satisfied about the product, he is now a Regretful and does not positively influence his neighbors. All Regretful individuals may die and, in his/her place, a Consumer is born, keeping the individuals number
N constant. A state transition diagram is shown in
Figure 1.
Pb is the probability of a Consumer buying the product and becoming a Buyer and is given by the following:
where v is the buyer neighbors of each consumer and C is the total connections of each individual, making v/C the influence of the neighborhood. Moreover, k(t) is a function of external factors and m(t) is a function of marketing operations in the population. Constants α, β, and γ represent the influence of neighborhood, external factors, and marketing operations, respectively. Finally, Pr is the constant probability of a Buyer regretting the purchase, Pn is the constant probability of a Buyer become a Consumer (either dying or having to buy the product again for some reason), and Pc is the probability of a buyer individual returning to be a consumer due to the necessity of buying another unit of the same product. Cells states are simultaneously updated at the end of each iteration.
2.3 Consumer market in ODEs
Consider that individuals are homogeneously distributed over space. Then, the PCA model can be represented by the following ODE model:
where a is the buying rate constant and b is the regret rate constant. When a Buyer individual returns to be a Consumer due to the necessity for buying another unit of the same product, such transition is expressed by constant rate c. Moreover, when a Regret individual either loses the disappointment about the product or dies, it is expressed by the constant rate e (which is called the death rate constant in epidemiological models). Because dC(t)/dt+dB(t)/dt+dR(t)/dt = 0, the total number of individuals remains constant, then: C(t) + B(t) + R(t) = N. Stationary solutions (C*, B*, R*) (where C*, B*, and R* are constants satisfying dC(t)/dt = 0, dB(t)/dt = 0 and dR(t)/dt = 0 for any instant t), are as follows:
and
where
Stability analysis of
Equations (2)40 reveals that the buyers-free stationary state given by
Equation (2) is asymptotically stable if
R0<1 and unstable if
R0>1; the state in which buyers remain in the population given by
Equation (3) is stable if
R0>1 and unstable if
R0< 1. In epidemiological modeling,
R0 is a bifurcation parameter called the basic reproduction number, that is, the number of new infected individuals from the influence of a single infected individual.
17 Here, it reveals if satisfied buyers will remain in the population, helping to spread information about the product and keeping sales active.
Because the ODE model is a mean-field approximation for the PCA model, parameters
a, b, c, and
e can be estimated from PCA simulations by the following expressions obtained from
Equation (1):
where ΔB(t)C→B/Δt is the increase per time step of buyer individuals due to the buying process; ΔR(t)B→R/Δt is the increase per time step of regret individuals due to the regret process; ΔC(t)B→C/Δt is the increase per time step of consumer individuals due to the necessity for buying another unit of the same product; and ΔC(t)R→C/Δt is the increase per time step of consumer individuals due to regret individuals that either lose the disappointment about the product or die. Assume that the probability of state transition at each iteration can be calculated using the relative frequency of its occurrence, then
ΔB(t)C→B/[ΔtC(t)];
ΔR(t)B→R/[ΔtB(t)];
ΔC(t)B→C/[ΔtB(t)];
ΔC(t)R→C/[ΔtR(t)]. Consequently:
where Sv is the number of consumer individuals with v connections to buyer neighbors. The expression for c comes from the fact that a buyer individual is firstly tested to establish if it will need to buy another unit of the same product (with probability Pr); if not, then it is tested to establish if the individual loses the disappointment about the product or dies.
Therefore, product sales parameters
b, c, and
e of linear terms in
Equation (1) are related to the probabilities of regret, loss of disappointment, and death. The value of
a is related to average buying probability
, and the term
aC(
t)
B(
t) gives the sales rate of the product.
In order to obtain similar evolutions of
C(
t),
B(
t), and
R(
t) in the ODE and PCA approaches, parameters
a, b, c, and
e can be estimated from simulations with the PCA model, then used in
Equations (1), which are numerically solved. Average values of these four parameters are calculated by taking into account the last 20 time steps of PCA simulation (when the system has already reached its permanent regime).
Figure 2 shows the dynamical behavior for PCA (
Figure 2(a)) with parameters: for
n = 200,
α = 0.83,
β = 0,
γ = 0,
Pr = 60%,
Pc = 30%,
Pn = 10% with initial conditions
C(0) = 99.5%,
B(0) = 0.5%, and
R(0) = 0%; and for the ODE (
Figure 2(b)) with parameters:
a = 0.716,
b = 0.666,
c = 0.127,
e = 0.098 with initial conditions
C(0) = 0.995,
B(0) = 0.005, and
R(0) = 0%. By using those parameters,
R0 = 0.984, confirming that for values
R0 < 1, product sales are not successful.
Figure 3 shows the dynamical behavior for PCA (
Figure 3(a)) with parameters:
n = 200,
α = 0.5,
β = 1,
γ = 1,
Pr = 5%,
Pc = 10%,
Pn = 10% with initial conditions
C(0) = 99.5%,
B(0) = 0.5%, and
R(0) = 0%; and for ODEs (
Figure 3(b)) with parameters:
a = 1.312,
b = 0.050,
c = 0.0.095,
e = 0.101 with initial conditions
C(0) = 0.995,
B(0) = 0.005, and
R(0) = 0%. By using those parameters,
R0 = 9.042, confirming that for values
R0 > 1, product sales are successful, with about 60% of the population pertaining to buyer state in the permanent regime. It is worth noting that a different parameter set has been tested, and the results were dynamically the same, that is, the system output is always limited and small variations in input parameters culminate in a small variation in results.
Figure 4 shows the sales rate (that is,
) in function of parameters
α, β, and
γ; for
α data,
β and
γ are fixed
β = γ = 0.05 and
α = 0, 0.1, 0.2, …, 2.0, considering an average of 10 simulations for each value. The same is valid for
β and
γ data. Note that when sales rates have low values, it is better to invest in word of mouth advertising, since it returns betters results. When the sales rate increases, the external situation (and traditional marketing, since it has the same influence in
Pr) variation leads to higher sales rates. The professor has to understand such variation in order to compose scenarios for the players.
When
α and
β vary for
α = 0, 0.05, 0.1,…, 2.0, and
β = 0, 0.05, 0.1,…, 1.0 (
γ = 0.1), the sales rates are as presented in
Figure 5, where, for each pair of parameters, the average of 10 simulations is considered. Note that, again, it is possible to see that the better external scenario is more efficient than word of mouth advertising, since for equals variations of
α and
β, the sales rate is higher for the
β variation.
Figure 6 presents sales rates for the variation of probabilities
Pr and
Pc (
Pr = 0, 0.05, 0.1, …, 2.0 and
Pc = 0, 0.05, 0.1, …, 1.0), where, for each pair of parameters, the average of 10 simulations is considered. Note that investment to increase the probability of the buyer needing to buy another product (
Pc, which increases
) is more efficient than reducing the probability of the buyer regretting the purchase (
Pr). Although those probabilities are pre-fixed in the business game, it is possible to let the professor manage them, if product quality is considered as a variable of the game.