Gender, Social Networks and Performance

Ilse Lindenlaub Anja Prummer

Cambridge-INET Working Paper Series No: 2014/12

December 5, 2019
Motivation

“Loose connections are the connections you need.

It’s the No. 1 rule of business.”
— Sallie Krawcheck, owner of the global
women’s network 85 Broads

I In the US, women’s earnings in 2012 were on average 80.9% of

men’s earnings
I Part of the gender gap explained by occupational sorting
I Even within occupations, the gender gap is considerable
Motivation

“Loose connections are the connections you need.

It’s the No. 1 rule of business.”
— Sallie Krawcheck, owner of the global
women’s network 85 Broads

I In the US, women’s earnings in 2012 were on average 80.9% of

men’s earnings
I Part of the gender gap explained by occupational sorting
I Even within occupations, the gender gap is considerable
Motivation

“Loose connections are the connections you need.

It’s the No. 1 rule of business.”
— Sallie Krawcheck, owner of the global
women’s network 85 Broads

I In the US, women’s earnings in 2012 were on average 80.9% of

men’s earnings
I Part of the gender gap explained by occupational sorting
I Even within occupations, the gender gap is considerable
Motivation

“Loose connections are the connections you need.

It’s the No. 1 rule of business.”
— Sallie Krawcheck, owner of the global
women’s network 85 Broads

I In the US, women’s earnings in 2012 were on average 80.9% of

men’s earnings
I Part of the gender gap explained by occupational sorting
I Even within occupations, the gender gap is considerable
Motivation

Source
A Primer on Network Theory

(a) Degree
(b) Clustering Coefficient
E
F D A
B C
G
A E
C D B
G H
F
H
Testing it in the Data
The AddHealth Dataset

I Data on 90,118 students collected in grades 7–12 from a nationally

representative sample of roughly 140 US schools in 1994–95
I Contains info on friendship networks
I Foucs on students older than 17

I Not using data about networks in the workplace because

I There isn’t a good dataset of informal relationships
I Networks can be shaped by the work environment
I Individuals will sort themselves into occupations for which their
networking type is most beneficial

I Based on the assumption that network types are persistent. . .

I Do we believe that?
Testing it in the Data
The AddHealth Dataset

I Data on 90,118 students collected in grades 7–12 from a nationally

representative sample of roughly 140 US schools in 1994–95
I Contains info on friendship networks
I Foucs on students older than 17

I Not using data about networks in the workplace because

I There isn’t a good dataset of informal relationships
I Networks can be shaped by the work environment
I Individuals will sort themselves into occupations for which their
networking type is most beneficial

I Based on the assumption that network types are persistent. . .

I Do we believe that?
Testing it in the Data
The AddHealth Dataset

I Data on 90,118 students collected in grades 7–12 from a nationally

representative sample of roughly 140 US schools in 1994–95
I Contains info on friendship networks
I Foucs on students older than 17

I Not using data about networks in the workplace because

I There isn’t a good dataset of informal relationships
I Networks can be shaped by the work environment
I Individuals will sort themselves into occupations for which their
networking type is most beneficial

I Based on the assumption that network types are persistent. . .

I Do we believe that?
Testing it in the Data
The AddHealth Dataset

I Data on 90,118 students collected in grades 7–12 from a nationally

representative sample of roughly 140 US schools in 1994–95
I Contains info on friendship networks
I Foucs on students older than 17

I Not using data about networks in the workplace because

I There isn’t a good dataset of informal relationships
I Networks can be shaped by the work environment
I Individuals will sort themselves into occupations for which their
networking type is most beneficial

I Based on the assumption that network types are persistent. . .

I Do we believe that?
Testing it in the Data
Results

Cl. Coeff. Degree

∗∗∗
Difference −0.0618 0.0259∗∗∗
(0.0145) (0.0075)
Observations 28,259
Mean for Girls 0.13 8.017

I One caveat: no discussion of the magnitude of the effects.

I Does it really matter to have 0.02 more friends (0.2% of the sample
mean)?
Testing it in the Data
Results

Cl. Coeff. Degree

∗∗∗
Difference −0.0618 0.0259∗∗∗
(0.0145) (0.0075)
Observations 28,259
Mean for Girls 0.13 8.017

I One caveat: no discussion of the magnitude of the effects.

I Does it really matter to have 0.02 more friends (0.2% of the sample
mean)?
Testing it in the Data
Results

Cl. Coeff. Degree

∗∗∗
Difference −0.0618 0.0259∗∗∗
(0.0145) (0.0075)
Observations 28,259
Mean for Girls 0.13 8.017

I One caveat: no discussion of the magnitude of the effects.

I Does it really matter to have 0.02 more friends (0.2% of the sample
mean)?
Theory Model
Setting

I Undirected network g of N workers

I At each period t, two of those workers i, j ∈ {1, . . . , N} are selected
at random to complete a project
I Whether the project is successful depends on their exerted effort,
their network structure and past project outcomes
Theory Model
Setting

1. Worker selection
I At the beginning of each period, two workers are drawn at random
from the set of workers to complete a project
I Two workers can only complete their project successfully if there
exists a direct link between them
I If not, the project fails with certainty, leading to zero payoff

I Probability of being selected for a project and being partnered with a

directly connected worker
2di
si =
N(N − 1)
Theory Model
Setting

1. Worker selection
I At the beginning of each period, two workers are drawn at random
from the set of workers to complete a project
I Two workers can only complete their project successfully if there
exists a direct link between them
I If not, the project fails with certainty, leading to zero payoff

I Probability of being selected for a project and being partnered with a

directly connected worker
2di
si =
N(N − 1)
Theory Model
Setting

2. Information
I State of the world at every period
(
θh with probability q
θ=
θl with probability 1 − q

with associated values of the project 2vh and 2vl , vh > vl

I Each worker receives a signal xi ∈ {0, 1}, where 1 indicates the high
state, and
1 1
Pr(xi = 1 | θh ) = p > , Pr(xi = 1 | θl ) = 1 − p <
2 2
Theory Model
Setting

2. Information
I State of the world at every period
(
θh with probability q
θ=
θl with probability 1 − q

with associated values of the project 2vh and 2vl , vh > vl

I Each worker receives a signal xi ∈ {0, 1}, where 1 indicates the high
state, and
1 1
Pr(xi = 1 | θh ) = p > , Pr(xi = 1 | θl ) = 1 − p <
2 2
Theory Model
Setting

2. Information
I Workers can also observe the signals of all other nodes they are
directly or indirectly connected to
I ni is the number of signals i receives
I yi is the number of high signals i receives
I yi = yj if i and j are directly or indirectly connected

I Bayesian updating:

π(yi ) = Pr(θh | yi )vh + (1 − Pr(θh | yi ))vl

Theory Model
Setting

2. Information
I Workers can also observe the signals of all other nodes they are
directly or indirectly connected to
I ni is the number of signals i receives
I yi is the number of high signals i receives
I yi = yj if i and j are directly or indirectly connected

I Bayesian updating:

π(yi ) = Pr(θh | yi )vh + (1 − Pr(θh | yi ))vl

Theory Model
Setting

3. Choice of effort
I Cost of effort c(ei ) = kei2
I Success probability function f (ei , ej )
Theory Model
Setting

4. Repeated game
I The game is played repeatedly
I To simplify, assume t = 1, 2

I If the project is not successful, the link between the two team
members deteriorates
I But the links with common friends also deteriorate!
I Probability of doing a project in period 2 by someone affected by
period 1’s failure
Cij
rij =
di
Theory Model
Setting

4. Repeated game
I The game is played repeatedly
I To simplify, assume t = 1, 2

I If the project is not successful, the link between the two team
members deteriorates
I But the links with common friends also deteriorate!
I Probability of doing a project in period 2 by someone affected by
period 1’s failure
Cij
rij =
di
Theory Model
Setting

4. Repeated game
I The game is played repeatedly
I To simplify, assume t = 1, 2

I If the project is not successful, the link between the two team
members deteriorates
I But the links with common friends also deteriorate!
I Probability of doing a project in period 2 by someone affected by
period 1’s failure
Cij
rij =
di
Theory Model
Results

Proposition
1. A higher degree (more information) increases first period effort only if the state
is high
2. A higher clustering (more peer pressure) increases first period effort
independently of the state of the world
3. Unless uncertainty vanishes, a worker with more information but less peer
pressure adjusts his effort to the expected project value better than a worker
with less information and more peer pressure
Theory Model
Results

Proposition
1. A higher degree (more information) unambiguously increases first and second
period wages only if the state is high in both periods
2. The effect vanishes as uncertainty vanishes
3. Higher clustering (more peer pressure) increases the first period wage
independently of the state but has an ambiguous effect on second period wage

Proposition
1. Wage Dynamics: If a C-worker has a lower first period wage than a D-worker,
then he also expects a lower wage in the second period, even if second period
uncertainty vanishes
2. Comparative Advantage: C-Workers hold a comparative advantage in
environments with less uncertainty
Theory Model
Results

Proposition
1. A higher degree (more information) unambiguously increases first and second
period wages only if the state is high in both periods
2. The effect vanishes as uncertainty vanishes
3. Higher clustering (more peer pressure) increases the first period wage
independently of the state but has an ambiguous effect on second period wage

Proposition
1. Wage Dynamics: If a C-worker has a lower first period wage than a D-worker,
then he also expects a lower wage in the second period, even if second period
uncertainty vanishes
2. Comparative Advantage: C-Workers hold a comparative advantage in
environments with less uncertainty
Facts Explained by the Model

1. Wage and performance gaps between men and women are especially
large within occupations and tasks characterized by uncertainty
2. More men than women choose occupations with high earnings
volatility
3. Having women in the network is particularly beneficial high up in the
organizational hierarchy
4. During recessions men’s unemployment exceeds women’s
unemployment
5. The beginning of the career is the most decisive period for the
gender wage gap formation
Facts Explained by the Model

1. Wage and performance gaps between men and women are especially
large within occupations and tasks characterized by uncertainty
2. More men than women choose occupations with high earnings
volatility
3. Having women in the network is particularly beneficial high up in the
organizational hierarchy
4. During recessions men’s unemployment exceeds women’s
unemployment
5. The beginning of the career is the most decisive period for the
gender wage gap formation
Facts Explained by the Model

1. Wage and performance gaps between men and women are especially
large within occupations and tasks characterized by uncertainty
2. More men than women choose occupations with high earnings
volatility
3. Having women in the network is particularly beneficial high up in the
organizational hierarchy
4. During recessions men’s unemployment exceeds women’s
unemployment
5. The beginning of the career is the most decisive period for the
gender wage gap formation
Facts Explained by the Model

1. Wage and performance gaps between men and women are especially
large within occupations and tasks characterized by uncertainty
2. More men than women choose occupations with high earnings
volatility
3. Having women in the network is particularly beneficial high up in the
organizational hierarchy
4. During recessions men’s unemployment exceeds women’s
unemployment
5. The beginning of the career is the most decisive period for the
gender wage gap formation
Facts Explained by the Model

1. Wage and performance gaps between men and women are especially
large within occupations and tasks characterized by uncertainty
2. More men than women choose occupations with high earnings
volatility
3. Having women in the network is particularly beneficial high up in the
organizational hierarchy
4. During recessions men’s unemployment exceeds women’s
unemployment
5. The beginning of the career is the most decisive period for the
gender wage gap formation
Discussion

I They justify their hypothesis empirically using a dataset that is not

entirely relevant for the question they are trying to answer
I And their results are not even that strong!
I Completely ad-hoc model for what they want to show
I It’s hard to draw any useful policy conclusions without a causal
identification
I Network structures are most likely endogenous

I What do you think

I About the paper?
I About their hypothesis?
Discussion

I They justify their hypothesis empirically using a dataset that is not

entirely relevant for the question they are trying to answer
I And their results are not even that strong!
I Completely ad-hoc model for what they want to show
I It’s hard to draw any useful policy conclusions without a causal
identification
I Network structures are most likely endogenous

I What do you think

I About the paper?
I About their hypothesis?
Discussion

I They justify their hypothesis empirically using a dataset that is not

entirely relevant for the question they are trying to answer
I And their results are not even that strong!
I Completely ad-hoc model for what they want to show
I It’s hard to draw any useful policy conclusions without a causal
identification
I Network structures are most likely endogenous

I What do you think

I About the paper?
I About their hypothesis?
Discussion

I They justify their hypothesis empirically using a dataset that is not

entirely relevant for the question they are trying to answer
I And their results are not even that strong!
I Completely ad-hoc model for what they want to show
I It’s hard to draw any useful policy conclusions without a causal
identification
I Network structures are most likely endogenous

I What do you think

I About the paper?
I About their hypothesis?
Discussion

I They justify their hypothesis empirically using a dataset that is not

entirely relevant for the question they are trying to answer
I And their results are not even that strong!
I Completely ad-hoc model for what they want to show
I It’s hard to draw any useful policy conclusions without a causal
identification
I Network structures are most likely endogenous

I What do you think

I About the paper?
I About their hypothesis?
Discussion

I They justify their hypothesis empirically using a dataset that is not

entirely relevant for the question they are trying to answer
I And their results are not even that strong!
I Completely ad-hoc model for what they want to show
I It’s hard to draw any useful policy conclusions without a causal
identification
I Network structures are most likely endogenous

I What do you think

I About the paper?
I About their hypothesis?
Thank you!