Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Home Search Collections Journals About Contact us My IOPscience

Sociophysics of sexism: normal and anomalous petrie multipliers

This content has been downloaded from IOPscience. Please scroll down to see the full text.

2015 J. Phys. A: Math. Theor. 48 27FT01

(http://iopscience.iop.org/1751-8121/48/27/27FT01)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 132.239.1.231
This content was downloaded on 03/05/2017 at 14:16

Please note that terms and conditions apply.

You may also be interested in:

Zipf law: an extreme perspective


Iddo Eliazar

Brownian gas models for extreme-value laws


Iddo Eliazar

On the structure and phase transitions of power-law Poissonian ensembles


Iddo Eliazar and Gleb Oshanin

Describing the indescribable: the stationary structures of transient Markovian dynamics


Iddo Eliazar

On the generation of anomalous and ultraslow diffusion


Iddo Eliazar and Joseph Klafter

The oligarchic structure of Paretian Poisson processes


I. Eliazar and J. Klafter

Subdiffusive continuous time random walks and weak ergodicity breaking analyzed with the
distribution of generalized diffusivities
T. Albers and G. Radons

Stochastic tools hidden behind the empirical dielectric relaxation laws


Aleksander Stanislavsky and Karina Weron

Searching circular DNA strands


Iddo Eliazar, Tal Koren and Joseph Klafter
Journal of Physics A: Mathematical and Theoretical
J. Phys. A: Math. Theor. 48 (2015) 27FT01 (10pp) doi:10.1088/1751-8113/48/27/27FT01

Fast Track Communication

Sociophysics of sexism: normal and


anomalous petrie multipliers
Iddo Eliazar
Smart Device Innovation Science Team, New Devices Group, Intel Corporation,
Yakum, Israel

E-mail: iddo.eliazar@intel.com and eliazar@post.tau.ac.il

Received 27 March 2015, revised 5 May 2015


Accepted for publication 29 May 2015
Published 17 June 2015

Abstract
A recent mathematical model by Karen Petrie explains how sexism towards
women can arise in organizations where male and female are equally sexist.
Indeed, the Petrie model predicts that such sexism will emerge whenever there
is a male majority, and quantifies this majority bias by the ‘Petrie multiplier’:
the square of the male/female ratio. In this paper—emulating the shift from
‘normal’ to ‘anomalous’ diffusion—we generalize the Petrie model to a sto-
chastic Poisson model that accommodates heterogeneously sexist men and
woman, and that extends the ‘normal’ quadratic Petrie multiplier to ‘anom-
alous’ non-quadratic multipliers. The Petrie multipliers span a full spectrum of
behaviors which we classify into four universal types. A variation of the
stochastic Poisson model and its Petrie multipliers is further applied to the
context of cyber warfare.

Keywords: Petrie model, sexism, normal diffusion, anomalous diffusion,


normal Petrie multipliers, anomalous Petrie multipliers

Gender biases and sexism in technology sectors is a well known phenomenon, and an
intensively studied subject in sociology [1–6]. Indeed, examples of relevant sociological
studies cover engineering [7, 8], computing and software development [9–11], hi-tech
entrepreneurship [12–14], Silicon-Valley firms [15, 16], Bangalore’s high-tech sector [17],
Swedish information-technology firms [18], and the Israeli hi-tech industry [19, 20]. Quite
recently, computer scientist Ian Gent published in his blog a post titled: ‘The Petrie multi-
plier: Why an attack on sexism in tech is NOT an attack on men’ [21]. This blog described a
mathematical model, devised by computer scientist Karen Petrie, that proposed a rather
unexpected explanation to sexism in tech companies: the reported sexism towards females
can very well result from the male majority in these firms, rather than from mere chauvinism.
Gent’s post popularized the, so called, ‘Petrie multiplier’ perspective to the phenomenon of

1751-8113/15/27FT01+10$33.00 © 2015 IOP Publishing Ltd Printed in the UK 1


J. Phys. A: Math. Theor. 48 (2015) 27FT01

sexism. Consequently, this perspective gained a rather viral interest, as well as a notable press
coverage [22–26].
The Petrie model is as follows. Consider a firm in which m men and w women are
employed, and set r = m/w to be the firm’s male–female ratio. All employees, male and female
alike, occasionally make sexist remarks towards employees of the opposite sex. On average,
each employee is assumed to make the same number of sexist remarks. Hence, in this model
men and women are equally sexist. Also, each sexist remark made by a man is assumed to
randomly target one of the women, and vice versa. The Petrie model predicts that, on average,
the number of sexist remarks received by a representative female employee will be r2 times
greater than those received by a representative male employee. Thus, whenever there is a
male majority, sexism towards the female minority will emerge—the emergence being a
direct result of a majority bias, not of chauvinism. Ian Gent termed the quadratic r2 majority
bias ‘the Petrie multiplier’.
In this paper we present a Poisson model that generalizes the Petrie model to the case of
heterogeneously sexist employees, and a stochastic Poisson model that extends the quadratic
r2 majority bias to non-quadratic majority biases of the form cr1 + ϵ —where c is a positive
coefficient, and where ϵ is an exponent taking values in the range 0 ⩽ ϵ ⩽ 1. Namely, the
stochastic Poisson model generalizes the quadratic Petrie multiplier to an entire spectrum of
Petrie multipliers, quadratic and non-quadratic alike. Moreover, we will establish that the
Petrie multipliers display four universal types of behavior: a ‘normal’ behavior encompassed
by the exponent value ϵ = 1, and three types of ‘anomalous’ behaviors that are characterize
by the exponent ϵ.
Our approach is motivated by the paradigm shift from ‘normal’ to ‘anomalous’ diffusion.
For about a century, the predominant model of diffusion processes in the physical sciences
was Brownian motion [27, 28]. The statistics of Brownian motion are governed by the
‘normal’ Gauss distribution, and its intrinsic temporal scaling is quadratic. However, since the
nineteen seventies, a whole plethora of diffusion processes with intrinsic (non-quadratic)
power-law temporal scalings were discovered [29–31]; these non-Brownian diffusion pro-
cesses were termed ‘anomalous’ [32–34]. Contrary to their name, anomalous diffusion pro-
cesses are actually rather normal and ubiquitous in the physical sciences [35–40]. In what
follows we will explore how analogies of ‘normal’ and ‘anomalous’ diffusion in physical
systems arise in the context of sexism in human organizations.
The Poisson model adopts the setting of the basic Petrie model, and considers each
employee—independently of all other employees—to make sexist remarks according to a
Poisson process. In science and engineering Poisson processes are the fundamental statistical
methodology applied to the modeling of events that occur randomly in time [41–43], e.g. the
emissions of radioactive particles from a radioactive source, the flow of molecules to an ionic
channel on a cell membrane, the arrivals of customers to a service station, the insurance
claims filed to an insurance company, etc. Poisson processes are characterized by their rates.
Recall that if a Poisson process has rate λ then its mean number of counts during a time
interval of length t is given by λ · t .
In what follows we label the male employees by the index i = 1, … , m , and the female
employees by the index j = 1, … , w . The specific details of the Poisson model are: (i) the
sexist remarks of man i follow a Poisson process with rate αi ; (ii) the sexist remarks of woman
j follow a Poisson process with rate β j ; and (iii) the m + w Poisson processes are independent
of each other. Also, the specific details of the underlying Petrie setting are: (i) each sexist
remark made by a man targets randomly one of the w women; (ii) each sexist remark made by

2
J. Phys. A: Math. Theor. 48 (2015) 27FT01

a woman targets randomly one of the m men; and (iii) the sexist remarks are targeted
independently of the Poisson processes as well as independently of each other.
Having defined the Poisson model we pick, in an arbitrary fashion, a representative male
employee and a representative female employee. Two key results from the theory of Poisson
processes—the coloring theorem and the superposition theorem [41]—imply that: (A) the
sexist remarks received by the representative female employee follow a Poisson process with
rate

α1 + ⋯ + α m
λ female = ,
w
m α1 + ⋯ + α m
= ≔ rα¯, (1)
w m

where ᾱ is the average of the men’s rates { α1, … , αm}; (B) the sexist remarks received by the
representative male employee follow a Poisson process with rate
β1 + ⋯ + βw
λ male = ,
m
w β1 + ⋯ + βw 1
= ≔ β¯ , (2)
m w r

{ }
where β̄ is the average of the women’s rates β1, … , βw . Consequently, the ratio of the
female rate λ female to the male rate λ male is given by
λ female α¯
ρ≔ = r 2. (3)
λ male β¯
In this paper, following the basic Petrie model, we consider the rates ratio of equation (3).
An alternative approach is to consider the rates ratio ρ˜ ≔ λ female ( λ male + λ female ). These
approaches are effectively the same, as simple transformations straightforwardly connect the
two ratios: ρ = ρ˜ (1 − ρ˜) and ρ˜ = ρ (1 + ρ).
Equation (3) manifests a Petrie multiplier with coefficient c = α¯ β¯ , and with exponent
ϵ = 1. As noted above, in the basic Petrie model all employees were considered equally
sexist. In the Poisson model this equal sexism is characterized by identical rates —
α1 = ⋯ = αm = β1 = ⋯ = βw —which, in turn, yield the basic Petrie multiplier ρ = r 2 . In
fact, it is evident from equation (3) that the basic Petrie multiplier is obtained if and only if the
average of the men’s rates coincides with the average of the women’s rates: ρ = r 2 ⇔ α¯ = β¯ .
Hence, the basic Petrie multiplier ρ = r 2 can be attained also in the case of heterogeneously
sexist employees.
In conclusion, we established that the Poisson model generalizes the basic Petrie model
by accommodating heterogeneously sexist employees, and that this heterogeneity affects the
coefficient c in a predictable manner: given the average of the men’s rates ᾱ and the average
of the women’s rates β̄ we know the precise value of c. However, the Poisson model leaves us
within the ‘quadratic realm’ of the basic Petrie multiplier—the exponent ϵ = 1. So, how can
we go beyond the exponent ϵ = 1 of the basic Petrie multiplier?
To answer this question we note that on route to equation (3) an implicit assumption was
made. Indeed, introducing the average male rate ᾱ and the average female rate β̄ we implicitly
assumed that these averages exist. But what happens in case these averages do not exist? This
fine point is exactly where matters get ‘interesting’, and to address the case of non-existing
averages we elevate from the Poisson model to the stochastic Poisson model.

3
J. Phys. A: Math. Theor. 48 (2015) 27FT01

In the stochastic Poisson model the rates themselves are assumed to be random variables:
(i) the men’s rates α1, … , αm are independent and identically distributed copies of a positive-
valued random variable A; (ii) the women’s rates β1, … , βw are independent and identically
distributed copies of a positive-valued random variable B; and (iii) the m + w random rates
are independent of each other, and conditioned on these rates the sexist remarks follow the
Poisson model. As we shall soon show, infinite-mean random rates will ‘do the job’ of taking
us beyond the exponent ϵ = 1 of the basic Petrie multiplier. First however, let us address the
case of finite-mean random rates.
Poisson processes with random rates were introduced and pioneered by Cox [44], and are
named in his honor [41]. A Cox process is characterized by its random rate, and conditioned
on its random rate it is a Poisson process. For a Cox process with a finite-mean random rate Λ
the following result holds [41]: the mean number of counts during a time interval of length t is
given by λ · t , where λ = E[Λ]. Namely, the rate λ of a Cox process is the mean E[Λ] of its
random rate Λ.
With the notion of Cox processes and their rates, as well as with the Cox-analogues of the
aforementioned coloring and superposition theorems, the Poisson results of equations (1) and
(2) are elevated to the context of the stochastic Poisson model as follows: (A) the sexist
remarks received by the representative female employee follow a Cox process with rate
⎡ α1 + ⋯ + α m ⎤
λ female = E ⎢ ⎥⎦ ,
⎣ w
1
= m E [A] = r E [A], (4)
w
where E[A] is the mean of the generic male random rate A; (B) the sexist remarks received by
the representative male employee follow a Cox process with rate
⎡ β1 + ⋯ + βw ⎤
λ male = E ⎢ ⎥,
⎣ m ⎦
1 1
= w E [B] = E [B], (5)
m r
where E[B] is the mean of the generic female random rate B. Consequently, the ratio of the
female rate λ female to the male rate λ male is given by
λ female E [A] 2
ρ≔ = r . (6)
λ male E [B ]
Equation (6) is the stochastic-Poisson-model counterpart of equation (3)—with the
means E[A] and E[B] replacing the averages ᾱ and β̄ , respectively. In the Poisson model
leading to equation (3) heterogeneity was introduced via the different deterministic rates
α1, … , αm and β1, … , βw . In the stochastic Poisson model leading to equation (6) hetero-
geneity is introduced by the statistical variability of the random rates α1, … , αm and
β1, … , βw . In the bottom line, the Poisson model and the stochastic Poisson model are
effectively synonymous—as they yield the essentially identical Petrie multipliers of
equations (3) and (6). However, the stochastic Poisson model has one major advantage over
the Poisson model: it can further accommodate the case of infinite-means random rates,
E [A] = E [B] = ∞.
To analyze the infinite-mean case, consider a Cox process with a random rate Λ, and let
p (t ) denote the probability that the process will count at least one event during a time interval
of length t. Conditioning implies that this probability is given by p (t ) = 1 − E [exp(− tΛ)].

4
J. Phys. A: Math. Theor. 48 (2015) 27FT01

Thus, we obtain that: (A) the probability that the representative female employee will receive
at least one sexist remark during a time interval of length t is given by
⎡ ⎛ α1 + ⋯ + α m ⎞ ⎤
pfemale (t ) = 1 − E ⎢ exp ⎜ − t⎟ ⎥,
⎣ ⎝ w ⎠⎦
⎡ ⎛ t ⎞⎤ m ⎡ ⎛ t ⎞⎤
= 1 − E ⎢ exp ⎜ − A⎟ ⎥ = 1 − exp ⎢ −mϕ A ⎜ ⎟ ⎥ , (7)
⎣ ⎝ w ⎠⎦ ⎣ ⎝ w ⎠⎦
where ϕ A (θ ) = −ln {E [exp(− θA)]} (θ ⩾ 0 ) is the log-Laplace transform of the generic male
random rate A; (B) the probability that the representative male employee will receive at least
one sexist remark during a time interval of length t is given by
⎡ ⎛ β1 + ⋯ + βw ⎞ ⎤
pmale (t ) = 1 − E ⎢ exp ⎜ − t⎟ ⎥,
⎣ ⎝ m ⎠⎦
⎡ ⎛ t ⎞ ⎤w ⎡ ⎛ t ⎞⎤
= 1 − E ⎢ exp ⎜ − B⎟ ⎥ = 1 − exp ⎢ −wϕ B ⎜ ⎟ ⎥ , (8)
⎣ ⎝ m ⎠⎦ ⎣ ⎝ m ⎠⎦
where ϕB (θ ) = −ln {E [exp(− θB)]} (θ ⩾ 0 ) is the log-Laplace transform of the generic
female random rate B. Consequently, the ratio of the female probability pfemale (t ) to the male
probability pmale (t ) is given by
⎡ ⎛ t ⎞⎤
1 − exp ⎢ −mϕ A ⎜ ⎟ ⎥
pfemale (t ) ⎣ ⎝ w ⎠⎦
ρ (t ) ≔ = . (9)
pmale (t ) ⎡ ⎛ t ⎞⎤
1 − exp ⎢ −wϕ B ⎜ ⎟ ⎥
⎣ ⎝ m ⎠⎦
In turn, taking the time t to zero, and setting θ = t m , we obtain the limit
⎛ t⎞
mϕ A ⎜ ⎟
⎝ w⎠ ϕ A (rθ )
lim ρ (t ) = lim = r · lim , (10)
t →0 t →0 ⎛ t ⎞ θ → 0 ϕ B (θ )
wϕ B ⎜ ⎟
⎝ m⎠
Now, note that if ξ is a positive-valued random variable with a finite mean E[ξ ], and with
a log-Laplace transform ϕξ (θ ) = −ln {E [exp(− θξ )]} (θ ⩾ 0 ), then we have:
lim θ → 0 ⎡⎣ ϕξ (θ ) θ ⎤⎦ = E [ξ ]. This fact implies that, in the case of finite-means random rates,
equation (10) yields the limit
ϕ A (rθ ) (rθ ) E [ A]
lim ρ (t ) = r 2 · lim = r2 . (11)
t →0 θ→0 ϕ B (θ ) θ E [B ]
Comparing equations (6) and (11) we obtain that the rate-ratio of equation (6) coincides with
the limit of the probability-ratio of equation (10):

λ female pfemale (t )
≔ ρ = lim ρ (t ) ≔ lim , (12)
λ male t → 0 t → 0 pmale (t )

equation (12) asserts that, in the case of finite-mean random rates, the probability-ratio limit
lim t → 0 ρ (t ) is an alternative representation of the rate-ratio ρ. Therefor, if a positive limit
lim t → 0 ρ (t ) exists in the case of infinite-means random rates—then this limit can be perceived
as the infinite-mean generalization of the rate-ratio ρ. Namely, we can set ρ ≔ lim t → 0 ρ (t )

5
J. Phys. A: Math. Theor. 48 (2015) 27FT01

whenever this limit exists and is positive, and consider this limit as the generalized definition
of the rate-ratio ρ.
An analysis of equation (10) implies that a positive limit ρ ≔ lim t → 0 ρ (t ) exists if and
only if the following pair of conditions is met: (i) the log-Laplace transforms ϕ A (θ ) and ϕB (θ )
are asymptotically equivalent at zero:
ϕ A (θ )
lim = c, (13)
θ→ 0 ϕ B (θ )

where c is a positive constant; and (ii) the log-Laplace transforms ϕ A (θ ) and ϕB (θ ) are
regularly varying at zero:
ϕ A (rθ ) ϕ B (rθ )
lim = lim = r ϵ, (14)
θ→0 ϕ A (θ ) θ→0 ϕ B (θ )
where ϵ is the common regular-variation exponent of these log-Laplace transforms [45].
Consequently, combining together equation (10) and equations (13) and (14) we obtain the
following form of Petrie multipliers:

ρ ≔ lim ρ (t ) = c · r 1 + ϵ. (15)
t →0

The necessary and sufficient conditions of equations (13) and (14) were stated in terms of
the log-Laplace transforms ϕ A (θ ) and ϕB (θ ), and are hence implicit. To make these condi-
tions explicit and tangible we shall represent them in terms of the corresponding tail dis-
tribution functions: (A) that of the generic male random rate, ψA (x ) = Pr(A > x ) (x > 0 ); and
(B) that of the generic female random rate, ψB (x ) = Pr(B > x ) (x > 0 ). Tauberian theorems
will enable us to express the near-zero asymptotic behaviors of the log-Laplace transforms
ϕ A (θ ) and ϕB (θ ) in terms of the near-infinity asymptotic behaviors of the tail distribution
functions ψA (x ) and ψB (x ), respectively [45].
First, we address the Petrie coefficient c. This coefficient is given by
n
∫ ψ A (x)dx
c = lim 0n . (16)
n →∞
∫0 B
ψ ( x )d x

If the tail distribution functions ψA (x ) and ψB (x ) are integrable over the positive half-line then
c = E [A] E [B]. On the other hand, if the tail distribution functions ψA (x ) and ψB (x ) are not
integrable over the positive half-line then c = lim x →∞ ⎡⎣ ψA (x ) ψB (x ) ⎤⎦. Hence, we obtain that:
(i) in the case of finite-mean random rates, E [A], E [B] < ∞, the condition of equation (13) is
met automatically; and (ii) in the case of infinite-mean random rates, E [A] = E [B] = ∞, the
condition of equation (13) is identical to the asymptotic equivalence of the tail distribution
functions ψA (x ) and ψB (x ) at infinity. In other words, the condition of equation (13) requires
that the means of the generic random rates A and B either converge, or diverge ‘at the same
order’.
Second, we address the Petrie exponent ϵ. The Tauberian analysis implies that this
exponent takes values in the range 0 ⩽ ϵ ⩽ 1, which we split to three different scenarios: (i)
the exponent value ϵ = 0; (ii) the sub-range 0 < ϵ < 1; and (iii) the exponent value ϵ = 1.
We turn now to describe, in reverse order, these three scenarios.
The ϵ = 1 scenario. The exponent value ϵ = 1 is attained if and only if the integrals of
the tail distribution functions ψA (x ) and ψB (x ) are slowly varying at infinity:

6
J. Phys. A: Math. Theor. 48 (2015) 27FT01

ny ny
∫ ψ A (x)dx ∫ ψB (x)dx
lim 0 n = lim 0 n =1 (17)
n →∞
∫0 ψ A (x)dx n →∞
∫0 ψB (x)dx
(y > 0 ). This scenario encompasses the cases of finite-mean and of ‘almost finite-mean’
random rates, and yields quadratic Petrie multipliers. Quintessential examples of the ‘almost
finite-mean’ case are tail distribution functions that follow, asymptotically (as x → ∞), a
decay of the form ψ (x ) ∼ 1 {x [ln (x )]δ }, with exponent δ that takes values in the range
0 ⩽ δ ⩽ 1.
The 0 < ϵ < 1 scenario. Exponents in this range are attained if and only if the tail
distribution functions ψA (x ) and ψB (x ) are regularly varying at infinity, with the common
regular-variation exponent − ϵ :
ψ A (nx ) ψ B (nx )
lim = lim = x −ϵ (18)
n →∞ ψ A (n ) n →∞ ψ B (n )

( x > 0 ). This scenario manifests the case of ‘heavy-tailed’ random rates, and yields Petrie
multipliers that are super-linear and sub-quadratic. Quintessential examples of this scenario
are tail distribution functions that follow, asymptotically (as x → ∞), a decay of the form
ψ (x ) ∼ 1 {x ϵ [ln (x )]δ }, with non-negative exponent δ.
The ϵ = 0 scenario. The exponent value ϵ = 0 is attained if and only if the tail dis-
tribution functions ψA (x ) and ψB (x ) are slowly varying at infinity:
ψ A (nx ) ψ B (nx )
lim = lim =1 (19)
n →∞ ψ A (n ) n →∞ ψ B (n )

( x > 0 ). This scenario manifests the case of ‘super heavy-tailed’ random rates, and yields
linear Petrie multipliers. Quintessential examples of this scenario are tail distribution
functions that follow, asymptotically (as x → ∞), a decay of the form ψ (x ) ∼ 1 [ln (x )]δ ,
with positive exponent δ.
Equation (15), together with the results following it, provide a comprehensive analysis of
the majority bias in the context of the stochastic Poisson model. As stated at the beginning of
the paper, the stochastic Poisson model extends the basic Petrie multiplier ρ = r 2 of the basic
Petrie model to the general Petrie multipliers of equation (15). Note that the basic Petrie
model is a special case of the stochastic Poisson model, attained by setting the generic random
rates A and B to be deterministic and identical: A = B = λ , where λ is an arbitrary posi-
tive rate.
The Tauberian analysis of the stochastic Poisson model establishes four universal types
of behaviors—one ‘normal’ and three ‘anomalous’:

• Normal: finite-mean random rates and quadratic Petrie multipliers.


• Mildly Anomalous: infinite-mean random rates and quadratic Petrie multipliers.
• Anomalous: infinite-mean random rates and Petrie multipliers that are sub-quadratic and
super-linear.
• Wildly Anomalous: infinite-mean random rates and linear Petrie multipliers.
A tabular classification of the four behavioral types is given in table 1. The mildly
anomalous behavioral type manifests the fine borderline between normal and anomalous
Petrie multipliers: on the one hand, the random rates have infinite means, and hence the
behavior is ‘anomalous’; on the other hand, the infinite means are not ‘infinite enough’, and

7
J. Phys. A: Math. Theor. 48 (2015) 27FT01

Table 1. Tabular classification of the four behavioral types. The columns specify the
mean behavior of the generic random rates A and B. The rows specify the exponent ϵ of
the Petrie multipliers.
E [A], E [B] < ∞ E [A] = E [B] = ∞
ϵ=1 Normal Mildly anomalous
0<ϵ< 1 — Anomalous
ϵ=0 — Wildly anomalous

hence the Petrie multipliers display a ‘normal’ quadratic behavior. Note that the more
‘anomalous’ the behavior of the random rates—i.e. the greater the statistical variability of the
random rates—the smaller the majority bias. In particular, to obtain the mildest majority bias
—characterized by linear Petrie multipliers—we need apply the wildest type of random rates.
Let us now compare the Petrie multipliers to diffusion. To that end consider the cor-
nerstone Continuous Time Random Walk (CTRW) model [32, 46, 47]: a random walk over
the integers, with waiting times that are independent and identically distributed copies of the
random variable A (or B). The CTRW describes motion on the microscopic scale, and
following an appropriate spatio-temporal scaling it yields a diffusion process on the macro-
scopic scale: D (t ) (t ⩾ 0 ). The spatio-temporal scaling is quadratic and is invariant with
respect to the waiting times if and only if they possess a finite mean, E[A] < ∞. The
diffusion’s propagation is quantified by its Mean Square Displacement, which follows a
temporal power-law: E[ ∣ D (t ) ∣2 ] = d · t ϵ , where d is a positive diffusion coefficient (con-
tingent on the spatio-temporal scaling), and where ϵ is the abovementioned exponent (con-
tingent on the generic waiting time A). The ϵ = 1 scenario manifests Brownian-type diffusion,
the 0 < ϵ < 1 scenario manifests sub-diffusion [48–51], and the ϵ = 0 scenario manifests
super-slow diffusion. The analogies between the Petrie multipliers and diffusion are self-
evident.
Last, we emphasize that variations of the stochastic Poisson model can apply to contexts
other than sexism. As an example, consider modeling a cyber war held between a ‘male state’
and a ‘female state’. The male state employs mh hackers, and has ms virtual sites; the female
state employs wh hackers, and has ws virtual sites. Each hacker attacks the virtual sites of the
opposite state, and a successful malware-infection of a ‘female site’ by a ‘male hacker’ is the
equivalent of a sexist remark made by a male employee towards a female employee, and
vice–versa. In this context, assuming that the cyber war began at time t = 0, we set: (A)
pfemale (t ) is the probability that a representative virtual female site is malware-infected by
time t; and (B) pmale (t ) is the probability that a representative virtual male site is malware-
infected by time t. We further set rh = m h wh to be the male–female hackers ratio, and assume
that the parameters ms and ws tend to infinity—while maintaining the positive limit
rs = lim m s, ws →∞ ( m s ws ) of male–female sites ratio. Then, in the limit m s, ws → ∞, the ratio of
the female infection probability pfemale (t ) to the male infection probability pmale (t ) is given
by

pfemale (t )
ρ (t ) ≔ = c · rh · rsϵ, (20)
pmale (t )

where the coefficient c and the exponent ϵ are as in equation (15). This cyber-war model
exemplifies how the stochastic Poisson model, and its resulting Petrie multipliers, can be
modified and applied to areas other than the sociophysics of sexism. Yet another

8
J. Phys. A: Math. Theor. 48 (2015) 27FT01

sociophysical example of the application of stochastic Poisson modeling is in the context of


the, so called, ‘matchmaking paradox’ [52, 53].
In conclusion, the recently introduced ‘Petrie multiplier’—which provides a majority-
bias explanation to sexism in Tech companies—received notable viral interest and media
coverage. Powered by concepts and tools from the theory of diffusion we presented in this
paper a rather profound generalization of the basic Petrie multiplier ρ = r 2 . Namely, we
applied statistical physics of normal and anomalous diffusion to comprehensively model and
analyze the sociophysics of sexism, obtained Petrie multipliers of the form ρ = cr1 + ϵ , and
classified these Petrie multipliers to four universal behavioral types. This interdisciplinary
approach demonstrates the potency of theoretical physics with regard to the modeling of
timely sociological phenomena.

Acknowledgments

The author is most thankful to Dr Yael Itzhaki (Netta Career Development Center, Israel) and
Hadass Ben Eliyahu (Tel Aviv University, Israel) for their kind assistance with the relevant
sociology literature [1–20]. Also, the author is grateful to Dr Oren Tapiero (Concordia
University, Canada) for bringing the press item [26] to his attention—thus instigating the
research that led to this paper.

References

[1] Ranson G and Reeves W J 1996 Gender Soc. 10 168


[2] McKay S C 2006 Gender Soc. 20 207
[3] Wasburn M H and Miller S G 2007 J. Coll. Stud. Retention: Res. Theory Pract. 9 205
[4] Baron J N, Hannan M T, Hsu G and Kocak O 2007 Work Occup. 34 35
[5] Laznjak J, Sporer Z and Svarc J 2011 Gender Technol. Dev. 15 175
[6] Riain S O 2013 Contemp. Sociol.: J. Rev. 42 863
[7] Hacker S, Smith D E and Turner S M 1990 Doing it the Hard Way: Investigations of Gender and
Technology (Boston: Unwin Hyman) pp 11–126
[8] Faulkner W 2000 Soc. Stud. Sci. 30 759
[9] Edwards P N 1990 Signs: J. Women Cult. Soc. 16 102
[10] Wright R 1996 C. Cheng, Masculinities in organizations (London: Sage) pp 77–96
[11] Benn E R 2007 Gender Work Organ. 14 312
[12] Ahl H J, Czarniawska B and Hopfl H 2002 Casting the Other: the Production and Maintenance of
Inequalities in Work Organizations (London: Routledge) pp 52–67
[13] Lewis P 2006 Gender Work Organ. 13 453
[14] Welbourne T M, Cycyota C S and Ferrante C J 2007 Group Organ. Manage. 32 524
[15] Cooper M 2000 Qual. Sociol. 23 379
[16] Shih J 2006 Gender Soc. 20 177
[17] Clark A W and Sekher T V 2007 Gender Technol. Dev. 11 285
[18] Peterson H 2007 Gender Work Organ. 14 333
[19] Frenkel M 2008 Gender Work Organ. 15 352
[20] Swed O and Butler J S 2015 Armed Forces Soc. 41 123
[21] Gent I 2013 The Petrie multiplier: Why an attack on sexism in tech is NOT an attack on men 13
October 2013, personal blog: (http://iangent.blogspot.co.uk/2013/10/the-petrie-multiplier-why-
attack-on.html)
[22] Robertson A 2013 Sexism in tech jobs explained by Dundee University lecturer 16 October 2013
The Courier: (http://www.thecourier.co.uk/news/local/dundee/sexism-in-tech-jobs-explained-
by-dundee-university-lecturer-1.142196)
[23] Chart D 2013 The Petrie multiplier 20 October 2013 personal blog: (http://www.davidchart.com/
2013/10/20/the-petrie-multiplier/)

9
J. Phys. A: Math. Theor. 48 (2015) 27FT01

[24] Petrie K 2013 Attack on sexism not an attack on men 27 November 2013 The Scotsman: (http://
www.scotsman.com/news/attack-on-sexism-not-an-attack-on-men-1–3209552)
[25] Massey B 2014 The Petrie multiplier 5 February 2014 personal blog: (http://fob.po8.org/
node/622)
[26] Butterworth J 2015 A mathematical model of oppression: the Petrie multiplier 8 March 2015
the Guardian: (http://www.theguardian.com/science/life-and-physics/2015/mar/08/the-petrie-
multiplier?CMP=fb_gu)
[27] Gardiner C 2004 Handbook of Stochastic Methods (New York: Springer)
[28] van Kampen N G 2007 Stochastic Processes in Physics and Chemistry 3rd edn (Amsterdam,
North-Holland)
[29] Scher H and Lax M 1973 Phys. Rev. B 7 4502
[30] Shlesinger M F 1974 J. Stat. Phys. 10 421
[31] Scher H and Montroll E W 1975 Phys. Rev. B 12 2455
[32] Metzler R and Klafter J 2000 Phys. Rep. 339 1
[33] Sokolov I M and Klafter J 2005 Chaos 15 026103
[34] Klafter J and Sokolov I M 2005 Phys. World 18 29
[35] Sancho J M, Lacasta A M, Lindenberg K, Sokolov I M and Romero A H 2004 Phys. Rev. Lett. 92
250601
[36] Sancho J M, Lacasta A M, Lindenberg K, Sokolov I M and Romero A H 2005 Phys. Rev. Lett. 94
188902
[37] Khoury M, Lacasta A M, Sancho J M and Lindenberg K 2011 Phys. Rev. Lett. 106 090602
[38] Eliazar I and Klafter J 2011 Ann. Phys. 326 2517
[39] Eliazar I and Shlesinger M F 2013 Phys. Rep. 527 101
[40] Metzler R, Jeon J H, Cherstvy A G and Barkai E 2014 Phys. Chem. Chem. Phys. 16 24128
[41] Kingman J F C 1993 Poisson Processes (Oxford: Oxford University Press)
[42] Cox D R and Isham V 2000 Point Processes (Boca Raton, FL: CRC Press)
[43] Streit R L 2010 Poisson Point Processes (New York: Springer)
[44] Cox D R 1955 J. R. Stat. Soc. B 17 129
[45] Bingham N H, Goldie C M and Teugels J L 1987 Regular Variation (Cambridge: Cambridge
University Press)
[46] Montroll E W and Weiss G H 1965 J. Math. Phys. 6 167
[47] Esposito M and Lindenberg K 2008 Phys. Rev. E 77 051119
[48] Magdziarz M, Weron A and Weron K 2007 Phys. Rev. E 75 016708
[49] Magdziarz M, Weron A, Burnecki K and Klafter J 2009 Phys. Rev. Lett. 103 180602
[50] Weron A and Magdziarz M 2009 Europhys. Lett. 86.6 60010
[51] Meroz Y and Sokolov I M 2015 Phys. Rep. 573 1
[52] Eliazar I and Sokolov I M 2010 J. Phys. A: Math. Theor. 43 055001
[53] Sokolov I M and Eliazar I 2010 Phys. Rev. E 81 026107

10

You might also like