Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

Social Dynamics of Consumer Response: A Unified

Framework Integrating Statistical Physics and

Marketing Dynamics
Javier Marín (NextBrain.ai)
March 21, 2024

Abstract
Comprehending how consumers react to advertising inputs is essential for marketers aiming to optimize advertising
strategies and improve campaign effectiveness. This study examines the complex nature of consumer behaviour
by applying theoretical frameworks derived from physics and social psychology. We present an innovative equation
that captures the relation between spending on advertising and consumer response, using concepts such as
symmetries, scaling laws, and phase transitions. By validating our equation against well-known models such as

the Michaelis-Menten and Hill equations, we prove its effectiveness in accurately representing the complexity of
consumer response dynamics. The analysis emphasizes the importance of key model parameters, such as marketing
effectiveness, response sensitivity, and behavioural sensitivity, in influencing consumer behaviour. The work
explores the practical implications for advertisers and marketers, as well as discussing the limitations and future
research directions. In summary, this study provides a thorough framework for comprehending and forecasting

consumer reactions to advertising, which has implications for optimizing advertising strategies and allocating
resources.

Keywords: consumer response curve, advertising, universality principle, phase transitions, correlation length

Introduction
symmetries, scaling laws, and phase transitions provide a
Within the field of consumer behavior research, the complex promising approach to comprehending and simulating the
dynamics that underlie consumer responses to advertising have complex interactions between advertising inputs and consumer
long been a source of interest. As we explore the complex nature responses. Symmetries, commonly linked to states of equilibrium
of consumer decision-making processes, we are becoming more in physical systems, offer a perspective to examine the
interested in the comprehensive theoretical frameworks derived equilibrium and interactions inherent in consumer behavior. By
from various fields, such as physics and social psychology. analyzing the symmetries in the consumer landscape, we can
Multiple formal mathematical frameworks can be associated with detect fundamental patterns and regularities that determine the
the dynamics of physical systems. The continuous process of effectiveness of advertising and the level of consumer
transforming natural systems into formal representations engagement. Scaling laws, known for adaptability across different
provides valuable insights into the structure of the physical fields, provide a method for revealing fundamental principles that
world. This process enables us to examine formal structures from govern the dynamics of consumer response. These laws clarify the
apparently unrelated viewpoints and detect hidden similarities connection between advertising spending and consumer behavior,
between concepts. Fundamental mathematical principles like revealing how changes in one factor correspond to changes in

1
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

another, thus aiding in the creation of predictive models and

strategies. Phase transitions, which are characteristic of complex
systems undergoing critical transformations, offer valuable
insights into the sudden shifts and crucial thresholds that are

inherent in the dynamics of consumer response. By utilizing

principles derived from phase transitions, we can gain a deeper
comprehension of the non-linearities and emergent phenomena
that define the relationship between advertising interventions
and consumer decision-making processes. Using these theoretical

frameworks, our research aims to better understand the complex

connection between advertising stimuli and consumer responses
Figure 1. Utility and demand curve. Adapted from Alford, 1956.
to create comprehensive models that accurately represent the
This curves has implicit the concept of elasticity that entails a
complexity and dynamics of consumer behavior.
reduction of demanded quantity 𝑞𝑥 when the price 𝑝𝑥 increase
(Alford, 1956). The demand curve 𝐷𝑥 will follow a scaling law
expression with a negative scaling parameter:
1. Existing models for consumer response
Although often discussed in literature, there is a lack of 𝑞𝑥 = 𝐶𝑝𝑥 #$ (2)
theoretical and empirical research that provides a comprehensive

model of consumer behavior in response to advertising. The Early research studies were founded on the theory of consumer
initial modeling methods employed probabilistic approaches information, suggesting that consumers may assess the utility of
(Hauser & Wisniewski, 1982) and classical econometric a product after making sufficient purchases of it. Advertising can
approaches rooted in utility theory, such as the Marshal's model enlighten consumers about a product, prompting them to respond
(Friedman, 1949; Prest, 1949; Working, 1927). Marshall for and anticipate gaining an advantage they wouldn't have
example suggested that the equilibrium for consuming a otherwise. Advertising helps consumers by guiding them towards
commodity 𝑥 occurs when the marginal utility of 𝑥, 𝑢𝑥 equals the optimal purchases based on their preferences, and the demand
product of the price of 𝑥, 𝑝𝑥 and the marginal utility of money elasticity for a particular product rises post-advertisement
𝑢𝑚 (Alford, 1956): compared to pre-advertisement (Nelson, 1975).

𝑢𝑥 = 𝑝𝑥 × 𝑢𝑚 (1) !" !$!

!#
= 𝜖 !#
(3)

where the marginal utility of money 𝑢𝑚 is assumed to be the

%&
where is the partial derivative of quantity demanded with
inverse of money income. As money income follows a scaling law %'

respect to a change in advertising, 𝜖 is the ordinary elasticity of

relationship, the dependence of the marginal utility 𝑢𝑥 with price
%(!
𝑝𝑥 also follow a scaling law. demand, and is the partial derivative of costs of advertising
%'

In the curves shown in figure 1 the individual does not consider with respect to a change in advertising. This model is a descent

𝑥 to be important, and his utility function is additive, resulting from econometric models that evaluate product demand based on

in the given marginal utility curve 𝑈! 𝑥 for 𝑥. Two points on the its utility. The consumer response curve to advertising is also

demand curve are determined by multiplying two prices, 𝑝! 𝑥 and referred to as shape effect, particularly within the context of

𝑝" 𝑥, by the constant marginal utility of money 𝑢! 𝑚. At the Marketing Mix Modelling (Tellis, 2006). Shape effect refers to the

positions where 𝑢! 𝑥 = 𝑝! 𝑥 and 𝑢! 𝑚 and 𝑢! 𝑥 = 𝑝" 𝑥, the values variation in sales as advertising intensity increases within the

𝑞! 𝑥 and 𝑞" 𝑥 are determined, which represent two points on the same time frame, and has also implicit the concept of elasticity.

demand curve 𝐷! 𝑥. For example Tellis (2006) defined advertising elasticity, or the
elasticity of sales to advertising, as the percentage change in sales

resulting from a 1% change in advertising.

2
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

steady state response to a constant advertising rate, has the

following solution:
"#
!" $
𝑦 = $%
"# (5)
%&$%

For an impulse response the solution for the general Vidale and

Wolf equation solution can be written in the following form:

Figure 2. Linear and nonlinear response to advertising

𝛥𝑦(𝑡) = 𝑦(𝑡) − 𝑦(0)𝑒 #-+ = [𝑚 − 𝑠(0)][1 − 𝑒 #.//, ]𝑒 #-+ (6)

Shape effect can exhibit various response curves or shapes, as

illustrated in figure 2. The representation shows three distinct The steady state solution exhibits a concave shape, while the
dynamics: a linear dynamic with a specified slope, a concave
spike input solution displays a convex shape. Both strategies take
response following a scaling-law form, and an S-shaped curve. At
into account a saturation effect, however neither of these
low advertising levels in the S-shaped curve, response is modest
solutions considers a threshold in the response. Although it was
because it is overpowered by market noise. High volumes of
considered a highly comprehensive empirical model, it faced
advertising lead to lower reaction owing to saturation. Sales show
criticism for not accounting for the S-shaped dynamics in the
the highest responsiveness at moderate advertising levels. The S-
response (Little, 1979). The discussion of whether the response
shaped dynamic has the highest level of acceptance among the
curve must adhere to an S-shape or another form of shape
three, with the concave response curve coming in a close second. originated with the introduction of the initial empirical models.
Both models share the characteristic that consumer response
(Bass & Parsons, 1969; Johansson, 1973). The two currents in
reaches saturation, where it stops increasing at a given spending
the discussion agreed on the necessity of the saturation effect but
level. This a central subject for all consumer response to
disagreed on whether the model should include a possible
advertising models. S-shape curve also implies the existence of
threshold effect.
an advertising threshold at which advertising fails to create any
The S-curve model was initiated by examining the derivative of
sales response. Researchers studying shape effect analysis have
the response function (sales or market share) in relation to
explored the advertising threshold as a potential rationale for the
advertising (Johansson, 1979). According to this hypothesis, as
S-shaped curve depicted in figure 2 (Bemmaor, 1984; Hanssens more advertising is conducted, the slope of the curve transitions
et al., 2003; Huang et al., 2012; Johansson, 1979; Simon & Arndt,
from positive and increasing to positive but decreasing. By
1980; Tull et al., 1986; Vakratsas et al., 2004). Empirical models
employing this approach, the response curve can be modeled as
of advertisement dynamics include thresholds that lead to
a logistic function, where the inflection point consistently falls at
concave response curves. This partially corroborates an S-shaped
the midpoint between the starting point and the saturation level.
curve as proposed by Dubé et al. (2005) and (Feinberg, 2001). Other authors have been proposing simpler models that
However, several of these research find the presence of a threshold
encompassed both the saturation effect and the threshold. One
in customer response to be highly improbable in reality. Vidale
of the first proposals was made by Little & Lodish (1969) with a
and Wolf proposed a model that included some important
simpler equation with the advantage that did not include a
assumptions like that sales rate increases with advertising rate,
specific saturation parameter but capture both effects:
the saturation effect and also that sales fluctuates with time

(Vidale & Wolfe, 1957). They illustrated their empirical approach

𝑟(𝑥) = 𝑟1 𝑎(1 + 𝑒 #2/ ) (7)
un the following equation:

where, 𝑥 is the exposure level, 𝑟 is the return, 𝑟1 is the return

)* *
= 𝜌𝑥 .1 − 1 − 𝜆𝑦 (4) without advertising 𝑟|𝑥 = 0, and 𝑎, 𝑏 are non-negative constants.
)+ ,

This approach can be understood as a conditional expectation of

where 𝑦 is the sales rate, 𝑥 is the advertising rate, 𝜌 is the the average fraction potential realized for a set of consumers at
response constant, 𝜆 is the decay constant, and 𝑚 is the exposure level 𝑦, denoted by 𝑟(𝑥). More recent research have
saturation sales rate. This first order differential equation, for a utilized the Hill equation to model this S-shape curve (Jin et al.,

3
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

2017). This equation, that has its origins in physiology (Hill, where Φ(𝑥) is the cumulative distribution function of the
1910) has the flexibility of capturing both, the concave and the standard normal distribution. This formula indicates that the
sigmoid effect in response curve. 89" #:
standardized sample converges in distribution to a standard
;/√3
In summary, the research conducted on consumer response
normal random variable as 𝑛 becomes large. Regardless of the
models in marketing has reached a unanimous view that customer
specific distribution of the individual random variables, the CLT
response is saturated. This concept is pervasive throughout many
ensures that the distribution of their sum or average approaches
fields of study, yet it should be acknowledged that there is limited
normality under certain conditions, facilitating probabilistic
empirical evidence to substantiate it. Nevertheless, to this day,
analysis and inference in various fields.
there remains a lack of unanimous agreement regarding the
Symmetry principles, such as group theory in physics, reveal
presence or absence of a threshold in the responses. Consequently,
underlying symmetries that govern diverse physical phenomena,
there is ongoing discourse regarding whether the curve exhibits a
leading to universal laws and conservation principles (McWeeny,
concave or sigmoidal shape.
2002). An important concept associated with symmetry is
invariance, which refers to the fundamental property of a system
2. The universality principle to remain unchanged when subjected to manipulations. The
The origins of the universality principle can be traced back to mathematical structure that underlies the study of symmetry and
the study of universal objects and phenomena that exhibit invariance is known as group theory. In physics, group theory is
common features across different mathematical contexts a powerful instrument that uses symmetry to understand and
(Batterman, 2017; Kadanoff & Wegner, 1971; Stanley, 1971). predict physical phenomena (Hamermesh, 2012; Slansky, 1981;
One of the earliest manifestations of universality can be found in Tinkham, 2003).
number theory, where prime numbers demonstrate remarkable Mathematically, transformation groups -a group of
regularities despite their seemingly random distribution. The transformations that do not modify a particular object or system-
Riemann zeta function and its connection to the distribution of allow us to study symmetry. These transformations form a group
primes exemplify this universality, as evidenced by the Riemann by satisfying the axioms of closure, associativity, identity and
hypothesis and its far-reaching implications (Katz & Messing, inverses, representing symmetries inside the system. A group 𝐺 is
1974). In probability theory, universality manifests in the described as a set plus an operation that joins any two elements
behavior of random matrices, which display similar statistical of the group. A group action on a set 𝑋 is formally defined as a
properties regardless of their specific entries. This phenomenon, function 𝛷:
known as universality in random matrix theory, has been 𝛷: 𝐺 × 𝑋 → 𝑋 (9)
extensively studied and has profound connections to various
areas of mathematics and physics, including quantum mechanics where 𝐺 is a symmetry group if its group action 𝛷 preserves the
and statistical physics. structure on 𝑋. We can explore the symmetry principle by
Mathematically, the universality principle can be formulated in examining rotational symmetry in classical mechanics
terms of limit theorems, symmetry principles, and invariant (Schattschneider, 1978). Let’s consider a physical system, such a
structures. Limit theorems, such as the central limit theorem in rigid body, that exhibits rotational symmetry. The rotational
probability theory, illustrate how certain statistical properties symmetry group for this system includes all rotations that do not
emerge universally as the size of a system grows large. alter the system's properties. This group is represented
Mathematically, the central limit theorem (CLT) can be mathematically as 𝑆𝑂(3), which stands for the special orthogonal
expressed as follows: let 𝑋! , 𝑋" , . . . , 𝑋3 be independent and group in three dimensions. The elements of the 𝑆𝑂(3) group are
identically distributed random variables with mean 𝜇 and rotation matrices that express rotations about an axis passing
!
standard deviation 𝜎. Define the sample mean 𝑋G3 = ∑345! 𝑋4 . through the origin. These matrices meet the following
3

Then, as 𝑛 approaches infinity we have the following equality: requirements: the composition of two rotations is also a rotation
(closure), rotations can be composed in any order (associativity),
89" #: the identity element corresponds to no rotation (identity) and
𝑙𝑖𝑚 𝑃 . ≤ 𝑥1 = Φ(𝑥) (8)
3→7 ;/√3
that each rotation has an inverse rotation that cancels its effect

4
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

(inverses). The symmetry principle implies that if ℒ represents nature of the system. Symmetry-breaking phenomena, such as
the laws governing the physical system, then under the action of spontaneous magnetization in ferromagnetic materials, play a

rotation 𝑅, the laws should remain unchanged. In mathematical crucial role in driving phase transitions and determining critical

terms, this can be expressed as: behavior (Castellani, 2003). These transitions are abrupt changes

in the state of a physical system, such as the transition from

ℒ(𝑅𝑥) = ℒ(𝑥) (10) liquid to gas or from ferromagnetic to paramagnetic states in

magnets. Critical phenomena occur near these phase transitions,

where x represents the state of the system and 𝑅𝑥 denotes the characterized by divergent behavior of certain physical

transformed state of the system under the rotation 𝑅. quantities, such as correlation length and susceptibility. The

Topological invariants are quantities associated with a correlation length is the scale at which the general properties of

topological space that remain unchanged under certain a material begin to deviate from its bulk properties. It is the

continuous transformations, such as deformations, stretching, or distance over which there is a significant correlation between the

bending (Carlsson, 2009). These invariants capture essential fluctuations of the microscopic degrees of freedom (such as the

geometric properties of the space that are independent of specific positions of atoms).

geometric configurations or coordinate systems. One of the most Universality categorizes different physical systems based on their

fundamental topological invariants is the Euler characteristic 𝜒 behavior near critical points, regardless of their specific

of a topological space 𝑀 (Bittner, 2004). It is defined as the microscopic details. This leads to the emergence of universal

alternating sum of the dimensions of its homology groups 𝐻4 (𝑀): scaling laws (also called power laws), which describe how physical
quantities behave as the complex system approaches criticality.

𝜒(𝑀) = ∑7 4
451(−1) 𝑑𝑖𝑚(𝐻4 (𝑀)) (11) Scaling laws can be seen as laws of nature describing complex

systems. Complex systems are made up of interconnected

The dimension of 𝐻4 (𝑀) represents the number of independent 𝑖- elements that interact to create an emergent structure. The

dimensional holes in the space, and the alternating signs ensure scaling hypothesis asserts that near critical points, physical

that the contributions of even and odd-dimensional homology quantities in complex systems exhibit a scaling behavior thar can

groups are appropriately weighted in the sum while captures the be described using power laws (Amaral et al., 1998; Plerou et al.,

cancellation of certain topological features, leading to a net 2004; Spurrett, 1999; West & Brown, 2005). For example, the

measure of the space's Euler characteristic. These invariants correlation length 𝜉 near a critical point 𝑇( can be modelled as:

remain unchanged under certain continuous transformations of

the space, providing a universal perspective on the geometric 𝜉~|𝑇 − 𝑇' |() (12)

properties of a space. These transformations include

deformations, homeomorphisms, and continuous mappings that where 𝜈 is the critical exponent characterizing the divergence of
preserve topological properties such as connectivity, the correlation length. This expression suggests that the

compactness, and orientability (Carlsson, 2009). Consider the characteristic length scale ξ diverges as the 𝑇 approaches the
" "
sphere 𝑆 and the torus 𝑇 as examples. Both have different critical point 𝑇( following a power-law behavior with a negative
shapes and topological properties, yet they share the same Euler exponent −𝜈. During phase transitions, such as the change from
characteristic (𝜒 = 2). This proves how topological invariants a liquid to a gas or the shift between ferromagnetic and
provide a universal characterization of geometric properties that paramagnetic states, particles in the system become interrelated.
is independent of specific shapes or configurations. The correlation length quantifies the extent to which correlations

between particles endure across a certain distance. The system's

correlation length tends to grow significantly when it gets closer
to critical events around a phase transition point (Gorban, 2021;

2.1 Symmetries and invariant structures Laidler & King, 1983). As the system gets closer to the critical

Universality suggests the existence of underlying symmetries and point, the correlation length increases, a phenomenon known as

principles governing phase transitions, irrespective of the specific a divergence. A characteristic of critical phenomena and phase

5
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

transitions is the divergence of the correlation length, which Consider a system with a fixed volume 𝑉 and composition 𝑖. We
indicates the formation of long-range correlations and collective assume that the system is in thermal equilibrium with a large
behavior in the system. reservoir at a specific temperature 𝑇. The weighted partition
A scaling law can be described as a basic polynomial functional function of a system (Moyal, 1949) can be formalize as follows:
relationship of the type
#$
𝑍 = ∑4 𝑒 #%& (16)
𝑦 = 𝐶𝑥 $ (13)

where 𝑖 is the system microstate, 𝐸4 is the energy of the state 𝑖,

parametrized by a scaling parameter 𝛼, which can be positive or
𝑇 is the temperature and k is the Boltzmann factor. The term
negative, and a constant 𝐶. A change in the quantity 𝑥 leads to #$
𝑒 #%& is frequently known and the Boltzmann factor and can be
a corresponding change in the quantity 𝑦, regardless of their
considered as a weighting factor for each independent microstate
initial sizes or changing the input value 𝑥 does not alter the shape
of assembly. We can rewrite the Boltzmann factor as 𝑒 #=>$ . This
of the function's output 𝑦. This property is known as scale
factor establishes a connection between the microscopic domain
invariance since it holds true for all scales (Glattfelder, 2019). It’s
of mechanics and the macroscopic world of thermodynamics by
mathematical expression is the following:
relating the energy of the molecules in a system to the
temperature of the surrounding environment (Battaglia et al.,
𝑦 = 𝑓(𝑎𝑥) = 𝐶(𝑎𝑥)$ = 𝑎 $ 𝑓(𝑥) ~ 𝑥 (14)
2009). Boltzmann factor essentially gives us the probability of
finding the system in a particular configuration based on its
Scaling-law distributions exhibit very large occurrences and
energy and temperature. It’s important to note that 𝑍 is
observations span a wide range of magnitudes. The probability
statistically analog to the thermodynamic potential 𝐹 (Moyal,
density function of these distributions can be expressed as
1949). Consequently partition function can be used to determine
all the significant thermodynamic properties of the modeled
𝑝(𝑥) = 𝐶𝑥 #$ for 𝛼 > 0 (15)
physical system, such as internal energy, specific heat,
magnetization, magnetic susceptibility, among others. One
If 𝑝(𝑥) follows a scaling law distribution, the cumulative
specific example is the Ising model applied to ferromagnetic
distribution function 𝒫(𝑥) also follows a power law, with an
materials (Cipra, 1987). The fundamental concept is to represent
exponent 𝛼 – 1. Scaling law distribution can be observed in wide
each particle in a magnetic material as having an inherent
range of real-world complex systems like for example the size and
characteristic known as "spin" which can be classified as either
population of cities or the price moves in financial markets.
"up" or "down." The spins in the material exhibit mutual
Scaling laws can also apply across various physical systems
interaction, and their configuration influences the magnetic
undergoing phase transitions, including magnets, fluids, and
properties of the material as a whole. The model generally takes
cosmological models. Despite the differences in the microscopic
into account the interactions between adjacent spins. Spins
details of these systems, they exhibit similar scaling behaviors
exhibit alignment with adjacent spins as a result of magnetic
near critical points.
interactions. When adjacent spins align, the substance tends to

acquire magnetization, displaying ferromagnetic characteristics.

2.2 Partition function and phase The Ising model is highly dependent on temperature. At high

transitions temperatures, the arrangement of spins is disturbed by thermal

fluctuations, resulting in a reduction in the total magnetization.

Partition functions are basic quantities in statistical physics that
As the temperature decreases, the material may experience a
are used to characterize a system's thermodynamic properties
phase transition, resulting in a significant increase in
(Mandl, 1991). This function provides details about the
magnetization. When the material reaches a specific temperature
organization of energy levels and the probability of various
called the critical temperature, it can undergo sudden and
particle arrangements. Partition functions frequently incorporate
significant changes in its properties, such as its magnetization
exponential terms that represent the Boltzmann factor, which
(Newell & Montroll, 1953).
quantifies the likelihood of occupying a specific energy state.

6
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

3. The sociodynamics of consumer attention turning to conversion. Moving from generating interest
to making a purchasing choice, social proof becomes a guiding
response to advertising
force as people look to the activities of others in their social circle
Statistical physics have been already applied to social dynamics (Karasawa, 1991). The level of group cohesion, which is
by several researchers (Helbing, 2010; Jusup et al., 2022; supported by shared preferences (Greer, 2012), influences the
Weidlich, 1991). Castellano et al., (2009) extensively examined transition rate, with strong group relationships supporting
opinion dynamics, voter models, majority rule models (MR), and collective decision-making. Finally, social identity theory
social impact theory using the same principles described in (Charness & Chen, 2020; Foroudi, 2019; Karasawa, 1991)
section 2. Social dynamics replaces the concept of particles with emphasise the need of matching a product to an individual's
the concept of individuals who interact with each other in a social identity in order to accelerate the transition from interest
complex manner, from which emerge patterns, sometimes not to purchase.
easily visible, that describe the behavior of the entire system When proposing a new model to describe the underlying
(whether it is a small group or a broader social context). dynamics of consumer response, we will rely on the fundamental
However, no studies have been found that apply these basic principles described in the previous section considering as well
principles in analyzing consumer response to marketing stimuli. some specific marketing dynamics. A central concept for
Social dynamics replaces the concept of particles with the concept consumer response to advertising found in the literature is that
of individuals who interact with each other in a complex manner, consumer response tends to saturate. In our proposal, we need to
from which emerge patterns, sometimes less visible, that describe consider this effect. Regarding the idea of advertising threshold,
the behavior of the entire system (whether it is a small group or as we have seen in empirical studies demonstrate the significant
a broader social context). However, no studies have been found challenge in practically verifying their existence. On the other
that apply these basic principles in analyzing consumer response hand, the shape of the consumer's response curve will have to be
to marketing stimuli. The field of marketing can be studied determined by the presence of regularities and either local or
through the theory of social influence, although it has a global symmetries in the system. The well-known S-shape is
particularity: individuals interact with a specific advertising initially ruled out for two fundamental reasons: first, the presence
campaign while also interacting with other individuals when of an advertising threshold is very difficult to demonstrate in
developing an attitude towards a brand. Actually, this dynamics practice, and second, due to the inherent limitations of this curve
closely resembles the Ising model, where a specific ferromagnetic in most dynamics. Although these curves have been widely used
particle tends to align its spin value with that of its neighboring in econometrics and ecology (for example, Malthusian growth),
particles, while also allowing for fluctuations based on their theoretical development has proven to be very limited when
temperature changes. Perhaps for this reason, the Ising model analyzing complex models.
has been favored by many researchers when studying social

dynamics. However, there are other decisive factors in the

4. A new equation for consumer response
relationship between consumers and advertising. The preference
of a product by a social group improves the transition from curve
exposure to engagement. The intensity of an individual's Based on these premises, we propose a model that can be
affiliation with a social group influences this shift even more, represented by the following equation:

especially when the brand is well aligned with the group's beliefs
(Farivar & Wang, 2022; Spears, 2011). As individuals seek 𝑦 = 𝐶𝑥 $ (1 − 𝑒 =/ )#? (17)
approval within their social circles, social validation works as a

catalyst, driving the transition from passive exposure to active where 𝑦 represents the consumer response, 𝑥 represents the
participation (Brown & Pehrson, 2019). Group identity dynamics advertising spend, 𝐶 is a constant reflecting the overall
(Chen & Li, 2009) influence how individuals allocate attention as effectiveness of the advertising campaign, 𝛼 is the scaling
they move from attracting attention to real conversion. Within- parameter characterizing the relationship between consumer
group social comparison, together with in-group affinity response and advertising spend, 𝛽 represents the rate of change
(Wachter, 2020), play a key role for evaluating the possibility of in consumer response with respect to advertising spend that

7
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

incorporates the concept of symmetry-breaking phenomena, and role in representing the inherent effectiveness of an advertising

𝛾 is the critical exponent that captures the behavior near a phase channel or campaign, regardless of advertising spend.
transition in consumer response. Next, we will provide a rationale 3. The term (1 − 𝑒 =/ ) introduces a non-linear
for each of these parameters: adjustment to the scaling behavior introduced by 𝐶𝑥 $ ,
1. The term 𝐶𝑥 $ is the same than in equation 13 and incorporating the symmetry-breaking effects observed in phase
introduce the property of scale invariance, meaning that varying transitions. Symmetry-breaking occurs when a system transitions
the value of the consumer response function argument 𝑥, the from a symmetric state to an asymmetric one, often triggered by
shape of the function 𝑦 is preserved. Alpha, 𝛼 represents the external factors or critical thresholds. This term is equivalent to
scaling parameter that influences the overall shape and a partition function as shown in section 2.2. As advertising spend
magnitude of the consumer response curve. It implies that a increases 𝑥, this term gradually shifts the curve, reflecting
change in the quantity 𝑥 leads to a corresponding change in the changes in consumer behavior. In our equation, beta represents

quantity 𝑦, regardless of their initial sizes or changing the input the rate of change in consumer response with respect to
value 𝑥 does not alter the shape of the function's output 𝑦. advertising spend and does not illustrate how the change evolves

Although this parameter does not directly illustrates how with increasing investment directly. Instead, beta quantifies the

advertising investment affects sales, it does determine how steep sensitivity or responsiveness of consumer response to changes in
or flat the curve is. A higher alpha value indicates a steeper curve, advertising spend. A positive beta value indicates that consumer

where small changes in advertising spend lead to significant response changes rapidly in response to variations in advertising
changes in consumer response, while a lower alpha value spend, while a negative beta value suggests that consumer

corresponds to a flatter curve, where larger changes in advertising response will reach a saturation point from where response 𝑦 does
spend are needed to produce noticeable effects on consumer not increase when increasing spend 𝑥. The parameter can be
behavior. It is important to clarify that this parameter is distinct referred to as “Response Sensitivity Coefficient” as highlight the
from the channel influence obtained by attribution models. The role of beta in quantifying the sensitivity of consumer response
parameter alpha can take on values ranging from 0 to 1. to changes in advertising spend. Beta can have positive and

Advertising response 𝑦 tends to grow linearly with spend 𝑥 when negative values. For 𝛽 < 0 the term 𝐶𝑥 $ (1 − 𝑒 =/ ) will have a
𝛼 = 1, and it is constant and equal to 𝐶 when 𝛼 = 0. However, decreasingly growing behavior with a saturation level with the
=/ #?
we must consider the impact of the additional term (1 − 𝑒 ) term 𝑒 =/ illustrating an exponential decay. Furthermore, it is
on this behavior, as it has the potential to alter the linear and important to note that the variable 𝛽 cannot assume a value of
constant response based on the values of the parameters 𝛽 and zero, as this would result in the equation 𝑦 = 0 for all values of
𝛾. We can call this parameter “Marketing Sensitivity Index” as 𝑥. For 𝛽 > 0 consumer response does not show a diminishing
illustrates how consumer response is affected by changes in returns effect
advertising expenditure. 4. Gamma, 𝛾 represents the critical exponent
2. The parameter 𝐶 represents the overall effectiveness characterizing the divergence of the correlation length of
or impact multiplier of the advertising channel. It reflects the consumer behavior (equation 12). This parameter quantifies how
inherent effectiveness of the channel in generating consumer the correlation length of consumer behavior changes as critical
response, irrespective of the advertising spend, while the channel thresholds are approached. A higher gamma (more negative)
influence obtained from attribution models measures the overall value indicates a more rapid increase in the correlation length of
effectiveness of a particular advertising channel. Higher 𝐶 values consumer behavior as critical points are reached, suggesting that
indicate channels with greater inherent effectiveness. A suitable consumer behavior becomes more interconnected and influenced
designation for parameter alpha could be "Response Scaling by external factors as critical thresholds are approached. A larger
Factor" or "Marketing Effectiveness Index.". These designations correlation length indeed suggests that the impact of the
highlight the importance of alpha in scaling the relationship advertising campaign extends over a larger group or cluster of
between advertising spending and consumer response. Parameter consumers due to a shift from individual responses to the
𝐶 could be referred to as "Baseline Effectiveness Constant" or emergence of interconnected audience clusters (critical point).
"Consumer Response Baseline." These designations highlight 𝐶's Conversely, if its value is low, it suggests that customers behave

8
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

more independently and disengaged from one another than Because

collectively. When considering the correlation length in consumer
response dynamics we can establish similarities with percolation ¬m1 − 𝑒 =/ )? ~ (1 − 𝑒 =@/ )#? n
theory (Essam, 1980). Large correlation length indicates that the

influence of the advertising campaign reaches a wider range of then

interconnected consumers, resembling the way percolation theory
explains the emergence of large connected clusters in random ¬m𝐶𝑎 $ 𝑥 $ (1 − 𝑒 =@/ )#? ~ 𝑥n
systems. For parameter gamma, a suitable designation could be
"Response Dynamics Exponent" or "Behavioral Sensitivity By changing the value of the function's argument (𝑥), the shape

Index." These designations emphasize the role of gamma in of the function 𝑦 is not maintained due to the inclusion of a

quantifying the responsiveness and dynamics of consumer symmetry-breaking effect with the term (1 − 𝑒 =/ )#? .
behavior to changes in advertising stimuli. Since gamma (𝛾) is

the exponent of the term (1 − 𝑒 =/ ), when 𝛾 = 0, (1 − 𝑒 =/ )? = 5. Experiments with data

1 and the consumer response curve will have the form 𝑦 = 𝐶𝑥 $
5.1 Material and methods
been symmetric and following a scaling law. This will lead to a
concave shape curve with no saturation point. To validate our equation, we have used three dummy datasets

It is important to note here that saturation point is not the same containing investment data in media as independent variables

than tipping points introduced by 𝛽 critical points introduced by and one dependent variable. Firstly, we have employed a

𝛾. Tipping point refer to a critical or spend quantity (𝑥) Marketing Mix Modelling methodology to calculate the impact of

meanwhile the critical point refers to consumer interaction media on the Key Performance Indicator (KPI) as a measure of

amount. To find the tipping point where the curve transition consumer response. We have utilized the Lightweight MMM

from a symmetric state to an asymmetric one we have to Python library developed by the Google research team with the

calculate the first derivative of function 𝑓(𝑥). We can write this configuration given in Table 1. Based on the attribution model

derivative as follows (see appendix): results, we utilized the Python library SciPy for curve fitting (see
table 3) to obtain the various parameters for each of the models

)* described in table 2. To validate our equation, we have calibrated

= 𝐶𝛼𝑥 $#! (1 − 𝑒 =/ )#? + 𝛾𝛽𝑒 =/ (1 − 𝑒 =/ )#?#! (18)
)/
the data obtained from the attribution model using three
different equations: our proposal, the Michaelis-Menten model,
)*
Tipping point will occur when = 0, so:
)/ and the Hill equation (table 2). The validation will be conducted
by comparing various metrics for the fitting of curves obtained
0 = (1 − 𝑒 =/ )? 𝐶𝛼𝑥 $#! + −𝛾𝛽𝑒 =/ (1 − 𝑒 =/ )?#!
with different models.

In order to obtain a numerical solution for the equation 18, we

Parameters Values
can use iterative techniques such as Newton's method to find the model_name 'carryover'
solution (Battiti, 1992). We can prove that equation 17 is not seasonality_degrees 4

scale invariant. The probability distribution 𝑝(𝑥) does not follows acceptance_p 0.85
number_warmup 2000
a scaling law, and the cumulative distribution function 𝒫(𝑥) does
samples 2000
not neither. With equation 15 as

Table 1: Attribution model parameters used to calculate the channel

𝑦 = 𝐶𝑥 $ (1 − 𝑒 =/ )#? influence of different datasets in the experiment (https://lightweight-
mmm.readthedocs.io/en/latest/)

we can rewrite as

𝑓(𝑎𝑥) = 𝐶(𝑎𝑥)$ (1 − 𝑒 =@/ )#?

9
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

Proposed equation 𝑦 = 𝐶𝑥 ' (1 − 𝑒 () )*+ advertising (OOH) spend is the lowest, suggesting a response
Michaelis-Menten
equation
curve that is reasonably flat and that spend increase does not
𝑉,-) 𝑥
𝑦=
(López et al., 2000; 𝑘, + 𝑥 convert to revenue generation. The marketing sensitivity
Raaijmakers, 1987)
Hill’s equation 1 index (𝛼), does not have a linear relationship with the channel
𝑦=
𝑘 .
(Hill, 1910) 1 + " 𝑥- $ influence obtained from an attribution model due to the

significant impact of the customer response baseline. For

Table 2: Equations used for fitting consumer response curve
instance, a high marketing effectiveness (𝛼) value indicates a high
level of channel importance if the consumer response baseline is
also high. OOH exhibits the highest positive response sensitivity
Parameters Values
method 'L-BFGS-B' values (𝛽), suggesting that from a certain spend level the
options:
outcome is constant. The response sensitivity (𝛽) value for Print
'disp' True
'maxiter' 1E+6 ads spend is the lowest and negative, indicating that consumer
tol 1E-6
response achieves saturation rapidly. The behavioral sensitivity
index (𝛾), for both TV and Google search spend indicate a
Table 3: Curve fitting method using python SciPy library

(https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.minimize reasonably low level, implying that we are targeting several

.html) audiences who are acting independently. This suggests the
presence of several different small clusters, as per the concepts of

5.2 Results percolation theory. For Print ads and Facebook spend, behavioral

Table 4 shows the results obtained for dummy data 1. We can sensitivity index (𝛾) values are high, indicating a significant
increase in the correlation length of consumer behavior near
observe that channel Facebook has the highest consumer
critical points. This implies that consumers becomes more
response baseline (𝐶) , which suggests that it is more effective in
interconnected and behave as an homogeneous audience.
creating consumer response. In other words, when we have to
start from zero, its response curve grows at the highest rate.
Marketing Response Behavioral Consumer Channel
Google's spend has the lowest (𝐶) value, indicating a Effectiveness, Sensitivity, Sensitivity, Response RoAS influence
𝛂 𝛃 𝛄 Baseline, 𝐂 (%)
comparatively lower level of intrinsic response when compared to
Online 1 0.228485 0.010439 0.075213 5078.6 9.65 10.43
other channels, as well as lower initial curve growth. Offline 1 0.343378 -0.164161 0.000041 5122.7 8.97 8.08
Offline 2 0.041432 0.082051 0.222679 13836.5 62.69 1.55
Offline 3 0.191907 -0.012182 0.009226 13883.8 2.37 20.22
Marketing Response Behavioral Consumer Channel Offline 4 0.034323 0.006126 0.145001 35052.3 86.49 4.63
Effectiveness, Sensitivity, Sensitivity, Response RoAS influence
𝛂 𝛃 𝛄 Baseline, 𝐂 (%) Offline 5 0.377626 0.00229 0.068685 6245.7 29.40 29.93

TV spend
0.165 -0.072 0.0000 54997.3 0.929 5.62
(offline)
Table 5: Equation fitting for dummy data 2
Out-of-home
advertising 0.018 0.286 0.0083 77805.3 0.451 2.01
spend (offline)
Tables 5 and 6 shows the results obtained for dummy data 2 and
Print adds
0.048 -1.000 1.0000 77556.3 1.341 2.07
spend (offline) dummy data 3. As we previously discussed, we can see from these
Google search
0.150 0.004 0.0043 37877.0 1.865 4.55 tables that the relationship between the channel's influence and
spend
Facebook the combination of the Marketing Sensitivity Index and the
0.045 -0.011 1.0000 90718.6 0.748 2.64
spend
Consumer Response Baseline is proportional. However, those two

parameters separately do not exhibit proportionality with

Table 4: Equation fitting for dummy data 1
influence. For example in table 5 we see that channel Offline 3
has the highest influence (20.22%). Marketing effectiveness index
TV spend has the highest marketing effectiveness index (𝛼),
of this channel is 0.1919, lower that for example Offline 1 that
suggesting that small changes in TV advertising spending could
has an alpha value of 0.3434. In this case Offline 1 has a consumer
lead to significant changes in consumer response. It implies that
response baseline of 5078.6, much lower than Offline 3 (13883.8).
with increasing the spend we can expect an important revenue

increase. The marketing effectiveness (𝛼) value for out-of-home

10
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

The curves obtained from applying the proposed equation to our The Behavioral Sensitivity Index (gamma) values exhibit a
dummy datasets are shown in figures 3, 4 and 5. The different significant range, spanning from 0 to 1. The impact of this

tables show the value of Response Sensitivity (𝛽) is not parameter on the response curve is not straightforward and needs

consistently negative. A negative beta value implies a diminishing to be evaluated in relation to percolation theory. Consider a

returns effect, where the curve first increases rapidly (lower 𝑥) network in which consumers are represented as nodes, and the

but gradually slows down until it approaches saturation. connections between them represent interactions or influences

Positive values of response sensitivity (𝛽) indicate the absence related to advertising. Percolation theory can be applied to study

of a saturation point, and the curve will continue to grow to a the propagation of advertising messages through connections and

certain extent based on that value without decreasing its growth their impact on customer behavior.

rate. Unlike the other models used, our equation does not

necessarily imply that the consumer's response will saturate. If a)

we wanted that to be the case, we would simply need to ensure

that the value of beta is always negative. However, it has been
deemed interesting for this value to have the possibility of not
saturating, with the advantage that by not forcing saturation (as

is the case with the Michaelis-Menten equation with 𝑦 = 𝑉,@/ ),

the response curve will grow to a lesser extent and be more
realistic. This is clearly illustrated in figures 3, 4, and 5. The
curves obtained using the Hill and Michaelis-Menten equations
grow to far higher values than those obtained with the proposed
equation. This inevitably leads to a poorer fit of the curve at b)
small spend values (𝑥), as these other equations primarily focus

on fitting higher values of the response (𝑦) in order to find the

saturation point. Our equation solves this problem by
incorporating a term equal to a partition function with a
parameter that has a similar effect to the Boltzmann factor
(equation 6).

Marketing Response Behavioral Consumer Channel

Effectiveness, Sensitivity, Sensitivity, Response RoAS influence
𝛂 𝛃 𝛄 Baseline, 𝐂 (%)

Brand Search 0.068966 0.889698 0.140121 1.638052 0.23869 0.38

Partnerships 0.514752 0.19758 0.835328 1.78229 0.83772 0.97 c)
TV 0.667243 0.327049 0.027788 1.477909 0.34771 2.14
Programmatic 0.457689 0.984439 0.297694 2.877326 1.16897 2.4
Magazines 1 0.635695 -0.007573 0.000298 3.758777 1.71482 1.04
Magazines 2 0.178672 -0.999983 0.934603 11.67936 6.64410 0.95
Magazines 3 0.140763 0.854003 0.299656 41.06679 11.5662 5.76
Advertorials 0.39016 0.42328 0.324954 1.476066 0.34659 1.39
Sponsorship 0.594036 0.061015 0.004442 3.627537 1.21638 1.99
Business Events 1 0.685643 0.059398 0 9.5255 3.40482 2.62
Business Events 2 0.671039 -0.016889 0.009344 17.30644 6.11824 4.01
Business Events 3 0.337004 0.001819 0.000008 2.489279 0.44001 0.51
Out Of Home 0.492034 0.308437 0.215954 1.934648 0.37178 1.53
Advertising

Figure 3: Curves obtained with dummy data 1. a) Proposed equation, b)

Table 6: Equation fitting for dummy data 3
Michaelis-Menten equation, c) Hill’s equation

11
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

In equation 16, the parameter Behavioral Sensitivity Index (𝛾) c)

quantifies the behavior of our system in proximity to this

threshold. Below this threshold, there are clusters of connected

nodes that are not sufficiently large to cover the full network.
When the threshold is exceeded, a large interconnected cluster

forms, covering a substantial part of the network. Regarding

advertising, this threshold signifies the minimum amount of
exposure or acceptance required for a specific product or concept

to achieve broad popularity within a population. When

consumers engage with advertising messages, they have the
Figure 4: Curves obtained with dummy data 2. a) Proposed equation, b)
potential to create groups of influence that are based on common Michaelis-Menten equation, c) Hill’s equation
qualities, preferences, or social relationships. These clusters
denote cohorts of consumers who are inclined to adopt
comparable attitudes or behaviors in reaction to advertising.
Cluster formation occurs when particular advertising efforts a)

powerfully resonate with specific audiences.

In our experiments the value of behavioral sensitivity index (𝛾)

illustrates if our audience is fragmented or if it’s acting as a large
group. Channels with larger behavioral sensitivity index (𝛾)
values can include mainstream television channels, popular social
media platforms, widely-read newspapers or magazines, and
influential websites, and has the potential of easily spread brand
messages.
b)

a)

c)

b)

Figure 5: Curves obtained with dummy data 3. a) Proposed equation, b)

Michaelis-Menten equation, c) Hill’s equation

12
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

With intercept: in section 1. This implies that when the spend is zero, the
Hill’s equation Michaelis-Menten Proposed equation
response is also zero. But previous information that 𝑦 = 0 when
F- F- F-
𝐫𝟐 p-val 𝐫𝟐 p-val 𝐫𝟐 p-val
pval pval pval
Ch1 0.000 0.955 0.955 0.001 0.680 0.680 0.001 0.728 0.728
𝑥 = 0 is not enough to completely explain regression through the
Ch2 0.087 0.000 0.000 0.102 0.000 0.000 0.002 0.546 0.546
Ch3 0.042 0.003 0.003 0.043 0.003 0.003 0.041 0.003 0.003
origin (RTO) as a more suitable fit. The primary distinction
Ch4 0.346 0.000 0.000 0.356 0.000 0.000 0.258 0.000 0.000
Ch5 0.004 0.363 0.363 0.004 0.373 0.373 0.009 0.174 0.174
between the two models lies in the fact that the regression via
Avg 0.096 0.264 0.264 0.101 0.211 0.211 0.062 0.290 0.290
the origin model has a higher coefficient of determination (𝑟 " ). A
higher value of 𝑟 " does not necessarily indicate a superior fit,
RTO model:

Hill’s equation Michaelis-Menten Proposed equation

because it could be possible the model is overfit.
F- F- F-
𝐫𝟐 p-val 𝐫𝟐 p-val 𝐫𝟐 p-val
pval pval pval
Ch1 0.329 0.000 0.000 0.311 0.000 0.000 0.382 0.000 0.000
Ch2 0.336 0.000 0.000 0.315 0.000 0.000 0.501 0.000 0.000 With intercept:
Ch3 0.359 0.000 0.000 0.337 0.000 0.000 0.495 0.000 0.000
Ch4 0.785 0.000 0.000 0.787 0.000 0.000 0.741 0.000 0.000 Hill’s equation Michaelis-Menten Proposed equation
Ch5 0.471 0.000 0.000 0.470 0.000 0.000 0.554 0.000 0.000 F- F- F-
𝐫 𝟐 p-val 𝐫 𝟐 p-val 𝐫𝟐 p-val
Avg 0.456 0.000 0.000 0.444 0.000 0.000 0.535 0.000 0.000 pval pval pval
Ch1 0.000 0.842 0.842 0.000 0.922 0.922 0.000 0.874 0.874
Ch2 0.209 0.000 0.000 0.182 0.000 0.000 0.147 0.000 0.000
Table 7: Metrics for curve fitting in dummy data 1. The upper table displays Ch3 0.006 0.446 0.446 0.000 0.943 0.943 0.000 0.958 0.958
Ch4 0.037 0.049 0.049 0.042 0.037 0.037 0.047 0.026 0.026
the outcome of an Ordinary Least Squares (OLS) regression analysis including Ch5 0.093 0.002 0.002 0.075 0.005 0.005 0.063 0.010 0.010
Ch6 0.002 0.669 0.669 0.003 0.590 0.590 0.003 0.575 0.575
an intercept term (standard regression), whereas the subsequent table Ch7 0.105 0.001 0.001 0.155 0.000 0.000 0.153 0.000 0.000
Ch8 0.048 0.024 0.024 0.046 0.028 0.028 0.042 0.037 0.037
showcases the findings obtained when the intercept term is excluded
Ch9 0.054 0.017 0.017 0.087 0.002 0.002 0.110 0.001 0.001
(Restricted Total OLS regression, RTO). Channels are in order: TV, OOH, Ch10 0.092 0.002 0.002 0.097 0.001 0.001 0.100 0.001 0.001
Ch11 0.194 0.000 0.000 0.212 0.000 0.000 0.210 0.000 0.000
Print ads., Google search and Facebook Ch12 0.205 0.000 0.000 0.372 0.000 0.000 0.365 0.000 0.000
Ch13 0.003 0.603 0.603 0.005 0.484 0.484 0.005 0.465 0.465
Avg 0.081 0.204 0.204 0.098 0.232 0.232 0.096 0.227 0.227

With intercept:
RTO model:
Hill’s equation Michaelis-Menten Proposed equation
F- F- F- Hill’s equation Michaelis-Menten Proposed equation
𝐫𝟐 p-val 𝐫𝟐 p-val 𝐫𝟐 p-val
pval pval pval F- F- F-
𝐫𝟐 p-val 𝐫𝟐 p-val 𝐫𝟐 p-val
Ch1 -inf 0.000 1.000 0.033 0.396 0.396 0.096 0.140 0.140 pval pval pval

Ch2 0.064 0.234 0.234 0.140 0.071 0.071 0.101 0.131 0.131 Ch1 0.321 0.000 0.000 0.312 0.000 0.000 0.332 0.000 0.000
Ch3 0.016 0.558 0.558 0.338 0.003 0.003 0.002 0.846 0.846 Ch2 0.349 0.000 0.000 0.329 0.000 0.000 0.332 0.000 0.000
Ch4 0.551 0.000 0.000 0.548 0.000 0.000 0.555 0.000 0.000 Ch3 0.308 0.000 0.000 0.386 0.000 0.000 0.392 0.000 0.000
Ch5 0.138 0.074 0.074 0.057 0.260 0.260 0.016 0.559 0.559 Ch4 0.792 0.000 0.000 0.788 0.000 0.000 0.788 0.000 0.000
Ch6 0.334 0.003 0.003 0.318 0.004 0.004 0.319 0.004 0.004 Ch5 0.324 0.000 0.000 0.330 0.000 0.000 0.338 0.000 0.000
Avg 0.184 0.145 0.311 0.239 0.122 0.122 0.181 0.280 0.280 Ch6 0.175 0.000 0.000 0.167 0.000 0.000 0.174 0.000 0.000
Ch7 0.251 0.000 0.000 0.301 0.000 0.000 0.274 0.000 0.000
Ch8 0.490 0.000 0.000 0.484 0.000 0.000 0.508 0.000 0.000
Ch9 0.527 0.000 0.000 0.500 0.000 0.000 0.502 0.000 0.000
RTO model:
Ch10 0.140 0.000 0.000 0.141 0.000 0.000 0.144 0.000 0.000
Hill’s equation Michaelis-Menten Proposed equation Ch11 0.087 0.002 0.002 0.080 0.003 0.003 0.086 0.002 0.002
F- F- F- Ch12 0.362 0.000 0.000 0.484 0.000 0.000 0.435 0.000 0.000
𝐫𝟐 p-val 𝐫𝟐 p-val 𝐫𝟐 p-val
pval pval pval Ch13 0.216 0.000 0.000 0.214 0.000 0.000 0.222 0.000 0.000
Ch1 0.926 0.000 0.000 0.851 0.000 0.000 0.893 0.000 0.000 Avg 0.334 0.000 0.000 0.347 0.000 0.000 0.348 0.000 0.000
Ch2 0.388 0.001 0.001 0.447 0.000 0.000 0.419 0.000 0.000
Ch3 0.306 0.004 0.004 0.089 0.147 0.147 0.278 0.007 0.007
Ch4 0.975 0.000 0.000 0.976 0.000 0.000 0.972 0.000 0.000
Table 9: Metrics for curve fitting in dummy data 3. The upper table displays
Ch5 0.089 0.147 0.147 0.216 0.019 0.019 0.355 0.002 0.002
Ch6 0.945 0.000 0.000 0.939 0.000 0.000 0.903 0.000 0.000 the outcome of an Ordinary Least Squares (OLS) regression analysis including
Avg 0.605 0.025 0.025 0.586 0.028 0.028 0.637 0.001 0.001
an intercept term (standard regression), whereas the subsequent table
showcases the findings obtained when the intercept term is excluded
Table 8: Metrics for curve fitting in dummy data 2. The upper table displays
(Restricted Total OLS regression, RTO). Channels are in order: Brand
the outcome of an Ordinary Least Squares (OLS) regression analysis including
Search, Partnerships, TV, Programmatic, Magazines 1, Magazines 2,
an intercept term (standard regression), whereas the subsequent table
Magazines 3, Advertorials, Sponsorship, Business Events 1, Business Events 2,
showcases the findings obtained when the intercept term is excluded
Business Events 3 and Out Of Home Advertising
(Restricted Total OLS regression, RTO). Channels are in order: Online 1,

Offline 1, Offline 2, Offline 3, Offline 4 and Offline 5

Based on the 𝑟 " values in our RTO models, it appears unlikely

Tables 7, 8 and 9 displays the outcome of an Ordinary Least that the model is overfit in any of the datasets (only in dataset

Squares (OLS) regression with an intercept, which is the usual 2 we reach values of 𝑟 " > 0.9 for Ch4 and Ch6). Furthermore,

regression method and without an intercept, known as the it is evident that while the predictor coefficient in the RTO model

Restricted Total OLS (RTO) regression. It is important to note consistently shows statistical significance (𝑝 < 0.01), it is not

that when 𝑥 is equal to zero, 𝑦 should also be equal to zero. We necessarily statistically significant in all the regular OLS models.

assume that there is no threshold in the variable 𝑦 as discussed Finally, additional test can be conducted by comparing the

13
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

standard error of the coefficients for both models (tables 19 and Michaelis-Menten and Hill equations offer an accurate method of
11). It is readily apparent that the coefficient values for the RTO adjusting the data while ignoring any 𝑦 points corresponding to
model are considerably higher (and positive) than those obtained 𝑥 = 0. On the other hand, our proposed equation effectively
using the normal OLS model. As a result, the standard error for incorporates them, showing a distinct curve shape. The curve
the RTO model is also smaller, since the standard errors values generated by the proposed equation first shows nearly
for both models are relatively. For example, in Ch1 the coefficient perpendicular growth until it reaches a certain threshold, at
result for the typical ordinary least squares (OLS) regression is - which point it takes on a convex shape. However, the Hill and
0.02, with a corresponding standard error of 0.125. However, the Michaelis-Menten equations does not allow for a curve with this
RTO model yields a larger value for Ch1's coefficient, which is shape able to capture this all this data points in the y-axis when
3.050 with a standard error of 0.379 (table 11). 𝑥 = 0. If we had used a standard OLS regression model for

comparison (where 𝑦 ≠ 0 when 𝑥 = 0) the Michaelis-Menten

Int. Coef. Int. Res.
𝐫𝟐 Intercept Coef std std p-
Coef. F-
Std
equation should yield better results for all datasets.
p-val pval
error error val error
Ch1 0.000 4.120 -0.02 0.11 0.125 0.000 0.874 0.874 0.947
Ch2 0.147 12.509 -1.54 0.53 0.401 0.000 0.000 0.000 4.147
Ch3 0.000 41.122 -0.03 2.63 0.645 0.000 0.958 0.958 19.408
Ch4 0.047 13.914 1.320 1.93 0.558 0.000 0.026 0.026 7.210
Ch5 0.063 14.536 -1.0 0.72 0.469 0.000 0.010 0.010 5.447
Ch6 0.003 49.772 -0.14 0.80 0.129 0.000 0.575 0.575 6.655
Ch7 0.153 1,258.0 8.32 33.28 1.098 0.000 0.000 0.000 273.557 5. Conclusions and future work
Ch8 0.042 11.465 -0.69 0.67 0.334 0.000 0.037 0.037 4.076
Ch9 0.110 21.486 -1.49 1.58 0.468 0.000 0.001 0.001 6.857
Ch10 0.100 175.638 -4.48 7.71 1.487 0.000 0.001 0.001 69.030 In this study, we explored the dynamics of consumer response to
Ch11 0.210 641.963 -13.5 29.14 1.723 0.000 0.000 0.000 227.343
Ch12 0.365 5.157 0.96 0.16 0.093 0.000 0.000 0.000 1.255 advertising stimuli through the lens of various mathematical
Ch13 0.005 64.637 -0.72 3.27 0.980 0.000 0.465 0.465 29.329

frameworks borrowed from physics and social psychology. By

Table 10: Statistics for a standard OLS model with dummy data 3
integrating these frameworks into our analysis, we aimed to
better understand the intricate relationship between advertising
𝐫𝟐 Coef Coef. std error p-value F-pvalue Res. Std error
Ch1 0.332 3.050 0.379 0.000 0.000 2.185 inputs and consumer behavior. Our investigation revealed several
Ch2 0.332 4.038 0.457 0.000 0.000 7.487
Ch3 0.392 7.397 0.501 0.000 0.000 28.266
key insights:
Ch4 0.788 5.305 0.258 0.000 0.000 8.685
Ch5 0.338 4.527 0.503 0.000 0.000 9.083
Ch6 0.174 7.586 0.261 0.000 0.000 20.370 By leveraging principles from physics, such as symmetries, scaling
Ch7 0.274 45.096 2.570 0.000 0.000 580.797
Ch8 0.508 3.708 0.298 0.000 0.000 6.767 laws, and phase transitions, along with concepts from social
Ch9 0.502 4.597 0.282 0.000 0.000 11.969
Ch10 0.144 11.038 1.329 0.000 0.000 105.336
psychology, we were able to construct a comprehensive
Ch11 0.086 18.328 1.960 0.002 0.002 358.377
Ch12 0.435 3.360 0.156 0.000 0.000 2.671
Ch13 0.222 9.393 0.705 0.000 0.000 41.673 theoretical framework for analyzing consumer responses to
advertising. These frameworks provided valuable perspectives on
Table 11: Statistics for RTO model with dummy data 3
understanding the underlying dynamics of consumer behavior.
We proposed a novel equation to model the relationship between
After evaluating the appropriateness of the RTO model, we can
advertising spending and consumer response and validated it
conclude that the curves fitted using our suggested equation
against established models such as the Michaelis-Menten and Hill
(equation 17) have a superior match in all datasets. The proposed
equations. Through the validation process, we demonstrated the
equation yields an average 𝑟 " value of 0.535 for dummy data 1,
efficacy of our proposed equation in capturing the nuances of
while the Hill and Michaelis-Menten models yield average values
consumer response dynamics. Our analysis highlighted the
of 0.456 and 0.444, respectively. The average 𝑟 " values for
significance of model parameters, such as marketing effectiveness,
dummy data 2 are as follows: 0.637 for the suggested equation,
response sensitivity, and behavioral sensitivity in shaping the
and 0.586 and 0.605 for the Hill and Michaelis-Menten models,
dynamics of consumer response. These parameters provided
respectively. Finally, for the dummy data 3, the average values
additional insights into the effectiveness of different advertising
of 𝑟 " are 0.348 for the suggested equation and 0.334 and 0.347 for
channels and the behavior of consumer responses.
the Hill and Michaelis-Menten model, respectively. The
The findings from our study have practical implications for
discrepancy in the accuracy of the three provided equations
advertisers and marketers. By understanding the interplay
might be traced to certain characteristics of the input data.
between advertising spending and consumer behavior,
Figures 3, 4, and 5, especially the first two, show an important
practitioners can optimize their advertising strategies to
amount of data points down the y-axis at 𝑥 = 0. Both the
maximize the impact on key performance indicators (KPIs) and

14
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

return on advertising spend (RoAS). Moreover, our insights into

the behavior of different advertising channels can inform
allocation decisions and resource allocation for advertising
campaigns.

Despite the contributions of our study, it is important to

acknowledge its limitations. The dummy datasets used for
validation may not fully capture the complexity of real-world
advertising scenarios. Future research could involve the analysis
of actual advertising data to further validate and refine our

models.

15
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

References Castellano, C., Fortunato, S., & Loreto, V. (2009). Statistical

physics of social dynamics. Reviews of Modern Physics,
81(2), 591–646.
Alford, R. F. G. (1956). Marshall’s Demand Curve. Economica,
https://doi.org/10.1103/REVMODPHYS.81.591/FIGUR
23(89), 23. https://doi.org/10.2307/2551267
ES/22/MEDIUM
Amaral, L. A. N., Buldyrev, S. V., Havlin, S., Salinger, M. A.,
Charness, G., & Chen, Y. (2020). Social Identity, Group
& Stanley, H. E. (1998). Power Law Scaling for a System
Behavior, and Teams. Annual Review of Economics,
of Interacting Units with Complex Internal Structure.
12(1), 691–713. https://doi.org/10.1146/annurev-
Physical Review Letters, 80(7), 1385.
economics-091619-032800
https://doi.org/10.1103/PhysRevLett.80.1385
Chen, Y., & Li, S. X. (2009). Group Identity and Social
Bass, F. M., & Parsons, L. J. (1969). Simultaneous-Equation
Preferences. American Economic Review, 99(1), 431–457.
Regression Analysis of Sales and Advertising. Applied
https://doi.org/10.1257/AER.99.1.431
Economics, 1(2), 103–124.
Cipra, B. A. (1987). An Introduction to the Ising Model. The
https://doi.org/10.1080/00036846900000024
American Mathematical Monthly, 94(10), 937–959.
Battaglia, O. R., Bonura, A., & Sperandeo-Mineo, R. M.
https://doi.org/10.1080/00029890.1987.12000742
(2009). A pedagogical approach to the Boltzmann factor
Dubé, J. P., Hitsch, G. J., & Manchanda, P. (2005). An
through experiments and simulations. European Journal
empirical model of advertising dynamics. Quantitative
of Physics, 30(5), 1025. https://doi.org/10.1088/0143-
Marketing and Economics, 3(2), 107–144.
0807/30/5/011
https://doi.org/10.1007/S11129-005-0334-2/METRICS
Batterman, R. W. (2017). Philosophical Implications of
Essam, J. W. (1980). Percolation theory. Reports on Progress
Kadanoff’s Work on the Renormalization Group. Journal
in Physics, 43(7), 833. https://doi.org/10.1088/0034-
of Statistical Physics, 167(3–4), 559–574.
4885/43/7/001
https://doi.org/10.1007/S10955-016-1659-9/METRICS
Farivar, S., & Wang, F. (2022). Effective influencer marketing:
Battiti, R. (1992). First- and Second-Order Methods for
A social identity perspective. Journal of Retailing and
Learning: Between Steepest Descent and Newton’s
Consumer Services, 67, 103026.
Method. Neural Computation, 4(2), 141–166.
https://doi.org/10.1016/J.JRETCONSER.2022.103026
https://doi.org/10.1162/NECO.1992.4.2.141
Feinberg, F. M. (2001). On continuous-time optimal advertising
Bemmaor, A. C. (1984). Testing alternative econometric models
under S-shaped response. Management Science, 47(11),
on the existence of advertising threshold effect. Journal
1476–1487.
of Marketing Research, 21(3), 298–308.
Foroudi, P. (2019). Influence of brand signature, brand
Bittner, F. (2004). The universal Euler characteristic for
awareness, brand attitude, brand reputation on hotel
varieties of characteristic zero. Compositio Mathematica,
industry’s brand performance. International Journal of
140(4), 1011–1032.
Hospitality Management, 76, 271–285.
https://doi.org/10.1112/S0010437X03000617
https://doi.org/10.1016/J.IJHM.2018.05.016
Brown, R., & Pehrson, S. (2019). Group Processes: Dynamics
Friedman, M. (1949). The Marshallian demand curve. Journal
within and Between Groups (3rd ed.). John Wiley &
of Political Economy, 57(6), 463–495.
Sons.
Glattfelder, J. B. (2019). The Semantics of Symmetry,
Carlsson, G. (2009). Topology and data. Bull. Amer. Math.
Invariance, and Structure. Frontiers Collection, Part
Soc., 46(2), 255–308.
F1071, 65–92. https://doi.org/10.1007/978-3-030-03633-
https://www.ams.org/journals/bull/2009-46-02/S0273-
1_3/FIGURES/1
0979-09-01249-X/
Gorban, A. N. (2021). Transition states and entangled mass
Castellani, E. (2003). On the meaning of symmetry breaking.
action law. Results in Physics, 22, 103922.
Symmetries in Physics: Philosophical Reflections, 321–
https://doi.org/10.1016/J.RINP.2021.103922
334.

16
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

Greer, L. L. (2012). Group cohesion: Then and now. Small Psychology, 30(4), 293–307.
Group Research, 43(6), 655–661. https://doi.org/10.1111/J.2044-8309.1991.TB00947.X
Hamermesh, M. (2012). Group theory and its application to Katz, N. M., & Messing, W. (1974). Some consequences of the
physical problems. Courier Corporation. Riemann hypothesis for varieties over finite fields.

Hanssens, D. M., Parsons, L. J., & Schultz, R. L. (2003). Inventiones Mathematicae, 23(1), 73–77.
Market response models: Econometric and time series https://doi.org/10.1007/BF01405203/METRICS
analysis (Vol. 2). Springer Science & Business Media. Laidler, K. J., & King, M. C. (1983). The development of
Hauser, J. R., & Wisniewski, K. J. (1982). Dynamic Analysis of transition-state theory. Journal of Physical Chemistry,
Consumer Response to Marketing Strategies. 87(15), 2657–2664.

Management Science, 28(5), 455–486. https://doi.org/10.1021/J100238A002/ASSET/J100238A

https://doi.org/10.1287/MNSC.28.5.455 002.FP.PNG_V03
Helbing, D. (2010). Quantitative sociodynamics: stochastic Little, J. D. C. (1979). Aggregate advertising models: The state
methods and models of social interaction processes. of the art. Operations Research, 27(4), 629–667.
Springer Science & Business Media. Little, J. D. C., & Lodish, L. M. (1969). A media planning
Hill, A. V. (1910). The possible effects of the aggregation of the calculus. Operations Research, 17(1), 1–35.
molecules of hemoglobin on its dissociation curves. J. López, S., France, J., Gerrits, W. J. J., Dhanoa, M. S.,
Physiol., 40, iv–vii. Humphries, D. J., & Dijkstra, J. (2000). A generalized

Huang, J., Leng, M., & Liang, L. (2012). Recent developments Michaelis-Menten equation for the analysis of growth.
in dynamic advertising research. European Journal of Journal of Animal Science, 78(7), 1816–1828.

Operational Research, 220(3), 591–609. https://doi.org/10.2527/2000.7871816X

https://doi.org/10.1016/J.EJOR.2012.02.031 Mandl, F. (1991). Statistical physics (Vol. 14). John Wiley &
Jin, Y., Wang, Y., Sun, Y., Chan, D., & Koehler, J. (2017). Sons.
Bayesian Methods for Media Mix Modeling with McWeeny, R. (2002). Symmetry: An introduction to group
Carryover and Shape Effects. theory and its applications. Courier Corporation.
https://ai.google/research/pubs/pub46001. Moyal, J. (1949). Stochastic processes and statistical physics.
Johansson, J. K. (1973). A generalized logistic function with an Journal of the Royal Statistical Society. Series B
application to the effect of advertising. Journal of the (Methodological), 11(2), 150–210.
American Statistical Association, 68(344), 824–827. Nelson, P. (1975). The economic consequences of advertising.
https://doi.org/10.1080/01621459.1973.10481431 The Journal of Business, 48(2), 213–241.
Johansson, J. K. (1979). Advertising and the S-curve: A new Newell, G. F., & Montroll, E. W. (1953). On the Theory of the

approach. Journal of Marketing Research, 16(3), 346– Ising Model of Ferromagnetism. Reviews of Modern
354. Physics, 25(2), 353.

Jusup, M., Holme, P., Kanazawa, K., Takayasu, M., Romić, I., https://doi.org/10.1103/RevModPhys.25.353
Wang, Z., Geček, S., Lipić, T., Podobnik, B., Wang, L., Plerou, V., Gopikrishnan, P., Gabaix, X., & Stanley, H. E.
Luo, W., Klanjšček, T., Fan, J., Boccaletti, S., & Perc, (2004). On the Origin of Power-Law Fluctuations in

M. (2022). Social physics. Physics Reports, 948, 1–148. Stock Prices. http://arxiv.org/abs/cond-mat/0403067
https://doi.org/10.1016/J.PHYSREP.2021.10.005 Prest, A. R. (1949). Some Experiments in Demand Analysis.
Kadanoff, L. P., & Wegner, F. J. (1971). Some Critical The Review of Economics and Statistics, 31(1), 33.
Properties of the Eight-Vertex Model. Physical Review https://doi.org/10.2307/1927192
B, 4(11), 3989. Raaijmakers, J. G. W. (1987). Statistical Analysis of the

https://doi.org/10.1103/PhysRevB.4.3989 Michaelis-Menten Equation. Biometrics, 43(4), 793.

Karasawa, M. (1991). Toward an assessment of social identity: https://doi.org/10.2307/2531533
The structure of group identification and its effects on Schattschneider, D. (1978). The Plane Symmetry Groups: Their
in-group evaluations. British Journal of Social Recognition and Notation. The American Mathematical

17
 Social Dynamics of Consumer Response: A Unified Framework Integrating Statistical Physics and Marketing Dynamics

Monthly, 85(6), 439–450. West, G. B., & Brown, J. H. (2005). The origin of allometric
https://doi.org/10.1080/00029890.1978.11994612 scaling laws in biology from genomes to ecosystems:
Simon, J. L., & Arndt, J. (1980). The shape of the advertising towards a quantitative unifying theory of biological
response function. Journal of Advertising Research. structure and organization. The Journal of Experimental

Slansky, R. (1981). Group theory for unified model building. Biology, 208(Pt 9), 1575–1592.
Physics Reports, 79(1), 1–128. https://doi.org/10.1242/JEB.01589

https://doi.org/10.1016/0370-1573(81)90092-2 Working, E. J. (1927). What Do Statistical “Demand Curves”

Spears, R. (2011). Group Identities: The Social Identity Show? The Quarterly Journal of Economics, 41(2), 212–
Perspective. Handbook of Identity Theory and Research, 235. https://doi.org/10.2307/1883501

201–224. https://doi.org/10.1007/978-1-4419-7988-9_9
Spurrett, D. (1999). Complexity and post-modernism:

Understanding complex systems. South African Journal

of Philosophy, 18(2), 258–274.
https://doi.org/10.1080/02580136.1999.10878187
Stanley, H. E. (1971). Phase transitions and critical phenomena
(Vol. 7). Clarendon Press, Oxford.
Tellis, G. J. (2006). Modeling Marketing Mix. The Handbook of
Marketing Research, 506–522.
https://doi.org/10.4135/9781412973380.N24

Tinkham, M. (2003). Group theory and quantum mechanics.

Courier Corporation.
Tull, D. S., Van Wood, R., Duhan, D., Gillpatrick, T.,
Robertson, K. R., & Helgeson, J. G. (1986). “Leveraged”
Decision Making in Advertising: The Flat Maximum

Principle and Its Implications. Journal of Marketing

Research, 23(1), 25–32.
https://doi.org/10.1177/002224378602300103
Vakratsas, D., Feinberg, F. M., Bass, F. M., & Kalyanaram, G.
(2004). The Shape of Advertising Response Functions
Revisited: A Model of Dynamic Probabilistic Thresholds.
Marketing Science, 23(1).
https://doi.org/10.1287/MKSC.1030.0035

Vidale, M. L., & Wolfe, H. (1957). An operations-research

study of sales response to advertising. Operations
Research, 5(3), 370–381.
Wachter, S. (2020). Affinity Profiling and Discrimination by
Association in Online Behavioral Advertising. Berkeley

Technology Law Journal, 35.

https://heinonline.org/HOL/Page?handle=hein.journals/
berktech35&id=389&div=&collection=
Weidlich, W. (1991). Physics and social science — The
approach of synergetics. Physics Reports, 204(1), 1–163.

https://doi.org/10.1016/0370-1573(91)90024-G

18