Date post: | 20-Jan-2015 |
Category: |
Documents |
Upload: | amdseminarseries |
View: | 158 times |
Download: | 0 times |
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Do Job Networks Disadvantage Women?Evidence from a recruitment experiment in
Malawi
Lori Beaman, Niall Keleher, and Jeremy Magruder
Northwestern, IPA, and UC-Berkeley
November, 2012
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Motivation
• In Malawi, as in much of the world, women aredisadvantaged in labor markets
• underrepresented in the formal sector• earn less
• There are a litany of possible explanations, e.g.
• taste-based or statistical discrimination• differences in baseline human capital• differences in preferences• differences in tenure/experience profiles• and so on
• Current policy interventions focus on closing the gendergap in educational attainment
• Question: will that be enough?
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
What about hiring processes?
• Much less research on whether the hiring process causes(dis)advantages
• About half of jobs are found through networks• In developing countries, networks are key for risk sharing,
credit in addition to labor market access
• Several advantages for employers
• relatively costless way to circulate info• (some) workers may have useful screening info about
friends and relatives (Montgomery 1991, Beaman andMagruder 2012)
• tied contracts between reference and referral may solvemoral hazard problems (Heath 2012)
• But do they disadvantage groups?
• Calvo-Armengol and Jackson (2004): the use of networkscan lead to disadvantages between groups
• Are women one of these groups?
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Women and Networks
• Priors are not so clear - potential advantages anddisadvantages.
• Could help women, if e.g. resume characteristics are scarceand hard-to-observe characteristics are more important
• Or, could leave women out: sociologists emphasizegender-homophily in networks
• necessary condition for Calvo-Armengol and Jackson(2004) mechanism
• However: as a stylized fact from observational data,women are less likely to get networked jobs
• In U.S. unemployed women are less likely to report usingtheir friends and relatives for help in search (27% of menvs. 20% of women) (Ioannides and Loury,2004)
• Based on observational data - could be confounded bydifferences in occupations, reporting choices, etc.
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Why would networks leave women out?
• It may be more costly for firms to access female referrals
• Men (or women) may not be connected to (high quality)women
• Sociology lit: women’s networks may be less organizedaround work (e.g. Smith 2000)
• Or men (or women) may have those connections, butprefer to refer men
• If it is easier to get (high quality) male referrals thanfemale referrals because of network characteristics, thencost-minimizing firms may end up hiring more menthrough referrals
• Firms may get more out of using referral systems for malehires
• References may be better able or more willing to screenmen
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Our experiment
• We conducted a recruitment experiment as part of a hiringdrive for enumerators in Malawi
• Survey firm wanted to hire more women
• Two waves: people encouraged to apply themselves andpeople then asked to make a referral
• All applicants complete skills assessment
• Competitive job between genders:
• 38% of people who apply themselves are women andperform similarly to men
• One type of position, so differences in occupational sortingcannot affect results. Reporting clear, too.
• Referral phase randomized whether applicants could referonly men, only women, or anyone, and also terms ofcontract
• Fixed finders fees or performance incentive
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Preview
• The use of referral systems disadvantages highly skilledwomen
• Only 30% of referrals (versus 38% of applicants) arewomen when people have a choice
• 2 reasons: men systematically refer men• Women’s referrals (both men and women) are less likely to
qualify
• However, when we restrict gender choices, men andwomen make references at the same rate under allcontracts regardless of which gender they must refer
• Men and women are connected to suitable men and women
• We develop and test a model to find out whichcharacteristics of networks lead to disadvantages
• Social incentives rather than productivity differences leadto disadvantages
• Screening potential of networks is maximized when menrefer men
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Outline of rest of talk
1 Describe experimental design
2 Test whether women are (dis)advantaged by referralsystems
3 Discuss a model of optimal referral choices under differentnetwork characteristics
4 Test whether men and women are connected to suitablewomen
5 Test for gender differences in network characteristics
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Our Experiment
• IPA-Malawi regularly hires a large number of enumeratorsfor several projects
• We posted fliers indicating a hiring drive at a number ofvisible places in Lilongwe and Blantyre
• Applicants were instructed to appear at a localemployment center at a specific date and time, with aresume.
• Upon arrival, applicants given an id card and resumescollected
• Applicants completed a written test
• Several math problems, ravens matrices, English skillsassessment, job comprehension component, computer skillsassessment
• 2 similar versions of test to limit cheating
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Our Experiment (2)
• Following, applicants completed a practical skillsassessment
• IPA enumerators act as survey respondents, applicants actas enumerators
• To test for hard-to-observe abilities, we made a number ofincorrect answers to questions - i.e. inconsistent householdsize, implausible values for household acreage
• Actors instructed to give the right answer if the applicantspress them
• 2 versions of incorrect answers• We measure the number of traps that the applicants
caught
• Total score on all components averaged. Applicantsinformed of qualification threshold.
• Qualified individuals called for enumerator positions aspositions open
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
CA men and women arecompetitive
0.0
1.0
2.0
3ke
rnel
den
sity
est
imat
e
20 40 60 80 100CA's overall (corrected) score
Male CAs Female CAs
Figure 1: CA Ability by Gender
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Experiment: Referral Rounds
• Finally, applicants asked to make a referral
• Randomly assigned to one of following treatments:• Asked at random to make a referral who was male, a
referral who was female, or a referral who could be male orfemale
• Cross-randomized the finder’s fee:
• A fixed fee of either 1000 MWK or 1500 MWK ($6 or$10).
• A performance incentive of 500 MWK if their referral doesnot qualify or 1800 MWK if their referral does qualify
• All treatments fully blinded from the perspective ofevaluators
• Referrals attend recruitment session 3 or 4 days later.Complete same skills assessment.
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Do Referral Systems disadvantagewomen?
(1) (2) (3) (4)
All CAs Male CAsFemale
CAs
Diff: p
value
A. CA Characteristics
Fraction of CAs 100% 62% 38%
CA is qualified 53% 56% 48% 0.047
N 767 480 287
B. CA Characteristics: Made Referral, Either Gender Treatments
Fraction of CAs 100% 61% 39%
CA is qualified 57% 62% 49% 0.061
N 217 130 87
C. Referral Characteristics: Either Gender Treatments
Referral is Female 30% 23% 43% 0.002
Referral is Qualified 49% 56% 38% 0.019
Referral is Qualified Male 34% 43% 22% 0.002
Referral is Qualified Female 14% 13% 17% 0.456
N 195 117 78
Table 1: Gender Distributions of CAs and Referrals
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
A simple model
• Suppose conventional applicants (CAs) each know acollection of potential referrals, some men and women.
• Each of these potential referrals has a social transfer theywill give the applicant
• Each also has an actual quality and an observed expectedquality
• Focus on individuals the CA might actually choose:
• For each perceived probability of qualifying, the personwho maximizes social payments
• Therefore expected quality is decreasing in social payments
• Observe referral choice under two contact types: fixed feeand performance pay
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Today, graphically
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-10
-5
0
5
10
15
20
Perceived Probability of Qualifying
Soc
ial P
aym
ent
sample similar networks
Note: Diamonds: women, Circles: men
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
What we observe
• Whether someone chooses to make a referral
• For those who make a referral, we see 2 nodes in thegender-specific network for each gender:
• Characteristics of person who maximizes social incentives
• Characteristics of person who maximizes expected pay+social incentives under performance incentive contract
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Are men and women connected?
• One reason women may be disadvantaged by referralsystem is if (suitable) women are not integrated into men’snetworks
• Men and women make a decision to make a referral if theexpected payoffs are greater than 0
• Under fixed fees, this means that they know a man orwoman whose social payment is not too negative
• Under perf pay, this means that they know a man orwoman whose total package of fixed fees + expected perfpay
• A stronger question: Are there men who only knowsuitable women?
• Are men in “either” treatments more likely to return witha referral than men in “male” treatments?
• [Later] does screening behavior look different in “either”treatments versus restricted male referrals?
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Are men less likely to knowsuitable women?
(1) (2) (3) (4)
Female Treatment ‐0.004 ‐0.055 ‐0.004 ‐0.042
(0.038) (0.054) (0.050) (0.074)
Either Gender Treatment 0.014 0.017 ‐0.052 ‐0.024
(0.040) (0.055) (0.052) (0.071)
Performance Pay ‐0.148 *** ‐0.113
(0.056) (0.080)
Perf Pay * Female Treatment 0.004 ‐0.013
(0.076) (0.111)
Perf Pay * Either Treatment 0.152 * 0.086
(0.079) (0.110)
Observations 506 310 506 310
CA Gender Men Women Men Women
Notes
1 The dependent variable is an indicator for whether the CA makes a referral.
2 All specifications include CA visit day dummies.
Table 2: Probability of Making a Referral
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
How would different networkcharacteristics affect referral
choices
We identify four dimensions of heterogeneity:
1 Maximal Social payment received: “Closest gender”
2 Expected quality of closest person: “Quality”
3 Slope of social payment/expected quality tradeoff:“Network Shallowness”
4 Variance of actual quality relative to expected quality:“Information”
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Similar Networks
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-10
-5
0
5
10
15
20
Perceived Probability of Qualifying
Soc
ial P
aym
ent
sample similar networks
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Closer men
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-10
-5
0
5
10
15
20
Perceived Probability of Qualifying
Soc
ial P
aym
ent
higher male social payments
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Similar Networks
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-10
-5
0
5
10
15
20
Perceived Probability of Qualifying
Soc
ial P
aym
ent
sample similar networks
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Higher quality men
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-10
-5
0
5
10
15
20
Perceived Probability of Qualifying
Soc
ial P
aym
ent
higher male quality
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Similar Networks
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-10
-5
0
5
10
15
20
Perceived Probability of Qualifying
Soc
ial P
aym
ent
sample similar networks
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Shallower women
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-10
-5
0
5
10
15
20
Perceived Probability of Qualifying
Soc
ial P
aym
ent
Shallower female network
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Similar Networks
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-10
-5
0
5
10
15
20
Perceived Probability of Qualifying
Soc
ial P
aym
ent
sample similar networks
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Worse information about women
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-10
-5
0
5
10
15
20
Perceived Probability of Qualifying
Soc
ial P
aym
ent
less information about women
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Model predictions
1 Under fixed fees: only differences in closeness affect whichreferral is chosen
2 Higher quality increases returns under performance pay
• Quality (of person who gives highest social payment) isrevealed under fixed fees
3 Worse info, more shallow networks can both lead to lowerresponse to performance pay
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Closest gender, quality & socialincentives
Men may know women, but would they share opportunities?
• Prediction 1 from the model: The person referred underfixed fees is the closest person in the network
• If men are closer to men (or women), should see menreferred systematically under fixed fee - unrestrictedtreatment
• The restricted-gender fixed fee treatments also let usobserve the quality of the closest people in the network:
• If men’s networks of men are higher quality than men’snetworks of women, should see fixed fee restricted malereferrals being higher quality than fixed fee restrictedfemale
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Characteristics of closest referrals
C. Referral Characteristics: Either Gender Treatments
Referral is Female 30% 23% 43% 0.002
Referral is Qualified 49% 56% 38% 0.019
Referral is Qualified Male 34% 43% 22% 0.002
Referral is Qualified Female 14% 13% 17% 0.456
N 195 117 78
D. Referral Characteristics: Either Gender, Fixed Fee Treatments
Referral is Female 32% 25% 43% 0.042
Referral is Qualified 50% 60% 37% 0.012
Referral is Qualified Male 34% 44% 20% 0.007
Referral is Qualified Female 16% 16% 16% 0.983
N 117 68 49
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Do men refer similar male andfemale network members?
0.0
1.0
2.0
3K
erne
l den
sity
est
imat
e
20 40 60 80 100Referral's overall (corrected) score
Men referring men Men referring women
Figure 2: Men's Fixed Fee Referrals
Note: figure compares men in restricted treatments only
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
What about women’s referrals?
0.0
1.0
2.0
3K
erne
l den
sity
est
imat
e
20 40 60 80 100Referral's overall (corrected) score
Women referring men Women referring women
Figure 3: Women's Fixed Fee Referrals
Note: figure compares women in restricted treatments only
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Summary so far
• By design, we only observe clean evidence of differences insocial incentives for men or women who maximize socialincentives (who are revealed through the fixed feetreatments)
• For those people:
• Men tend to maximize men’s incentives• Low ability people tend to maximize women’s incentives
• Closest women are low ability• Closest men however are not systematically low ability
• Can conclude: at least among socially closest people, menand women have different social incentives
• Social incentives make it cheaper to (a) get male referralsfrom men and (b) use men to get high quality referrals
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Is Women’s DisadvantageProductive?
• If employers encourage referral hires, they likely gainsomething from their use
• One thing which has been emphasized is screening (e.g.Montgomery (1990), Beaman and Magruder (2012))
• If employees see hard to observe characteristics, canimprove outcomes for employer
• If men (women) are less able to screen women, it may leadto employers discouraging female referrals
• From the model: CAs will screen if and only if they havegood information, and networks are not too shallow
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-10
-5
0
5
10
15
20
Perceived Probability of Qualifying
Soc
ial P
aym
ent
less information about women
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Low info and screening
• Low info makes the tradeoffs “steeper” - it becomes moreexpensive and more infeasible to find very high qualityreferrals
• Essentially, most referral probabilities of qualificationpushed towards 1/2
• this increases the payoffs to referring someone who youthink is relatively low ability under perf pay incentives
• and decreases the payoffs to referring someone who youthink is relatively high ability under perf pay
• Empirically, if men (women) have lower ability to screenwomen, should observe a smaller increase in performancein response to perf pay
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
(1) (2) (3) (4)
Female Referral Treatment ‐0.030 ‐0.190 ** 0.068 ‐0.181
(0.062) (0.083) (0.081) (0.113)
Either Gender Treatment 0.071 ‐0.231 *** 0.227 *** ‐0.242 **
(0.066) (0.082) (0.084) (0.107)
Performance Pay 0.267 *** 0.021
(0.093) (0.122)
Perf Pay * Female Treatment ‐0.248 * ‐0.022
(0.127) (0.171)
Perf Pay * Either Treatment ‐0.383 *** 0.032
(0.132) (0.169)
Observations 390 227 390 227
CA Gender Men Women Men Women
Notes
1 The dependent variable is an indicator for the referral qualifying.
2 All specifications include CA visit day dummies.
Referral Qualifies
Table 4: Referral Performance
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Screening Results
• Men can screen men
• Men cannot screen women (or, at least, won’t at theselevels of incentives)
• Allowing the option to refer women eliminates thescreening premium
• Suggests that employers who want to maximize screeningmay discourage men from making female referrals.
• Some evidence that difference is info and not shallowness:men are more likely to make a low quality referral underperf pay-female treatments than under fixed fee-female
• Women show less ability to screen men or women overall
• Some not quite sig evidence that they can screen women
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Screening with choice of genderWhy is there a lower screening premium when we allow eithergender?
• Under performance pay, one maximizes the sum of socialincentives and expected perf incentive
• This makes the theoretical effect ambiguous
• Considering either gender in general allows you to “buy”quality with giving up a lower amount of social incentives
• Increases both chance that you observe someone who hasa high chance of qualifying and gives you OK socialincentives
⇒ May increase performance premium
• Also ↑ chance that you observe someone who has an OKchance of qualifying but gives you great social incentives
⇒ May decrease performance premium
• Happens, in particular, when info is bad about one gender
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
What exactly is being screened?
• Have much richer data than is being used here - detailedassessments of different referral characteristics
• Men are screening in some ways across a broad category ofcharacteristics
• Women are screening, too -
• significantly, women screen women on language scores andcognitive skills.
• Women screen men on survey experience
• The former is probably more valuable as screening foremployers. May be a role for encouraging female referralsof women
• still, if employers use referrals for screening, biggest returnsare to get men to refer men
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Survey
exp
Tertiary
Education
Math
Score
Language
Score
Ravens
score
Computer
Score
Practical
Exam Score
Feedback
points
(1) (2) (3) (4) (5) (6) (7) (8)
‐0.033 0.045 ‐0.017 ‐0.115 ‐0.092 0.062 1.033 3.003 ***
(0.069) (0.074) (0.142) (0.207) (0.194) (0.371) (0.661) (1.044)
0.040 0.072 0.009 0.087 0.089 0.623 1.378 ** 1.856 *
(0.072) (0.077) (0.148) (0.215) (0.203) (0.387) (0.689) (1.089)
Performance Pay 0.080 0.067 0.134 ‐0.005 0.230 0.943 ** 0.496 1.883
(0.080) (0.085) (0.164) (0.238) (0.224) (0.428) (0.757) (1.197)
‐0.075 0.025 ‐0.259 ‐0.027 ‐0.293 ‐0.915 ‐0.950 ‐2.443
(0.108) (0.116) (0.223) (0.325) (0.305) (0.583) (1.026) (1.622)
‐0.165 ‐0.083 ‐0.065 ‐0.169 ‐0.367 ‐0.856 ‐1.768 * ‐3.371 **
(0.113) (0.121) (0.232) (0.338) (0.318) (0.607) (1.069) (1.696)
Observations 386 390 390 390 390 390 383 382
Notes
1 The dependent variable is an indicator for the referral qualifying.
2 All specifications include CA visit day dummies.
Table 5: Screening of Male CAs on Different Characteristics
Perf Pay * Female
Treatment
Perf Pay * Either
Treatment
Female Referral
Treatment
Either Gender
Treatment
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Survey expTertiary
Education
Math
Score
Language
Score
Ravens
score
Computer
Score
Practical
Exam Score
Feedback
points
(1) (2) (3) (4) (5) (6) (7) (8)
0.032 0.151 ‐0.332 ‐1.140 *** ‐0.435 ‐0.627 0.972 2.152
(0.091) (0.110) (0.216) (0.342) (0.270) (0.538) (0.963) (1.349)
0.040 0.017 ‐0.189 ‐0.246 ‐0.172 ‐0.139 0.015 0.879
(0.086) (0.104) (0.205) (0.324) (0.256) (0.509) (0.910) (1.274)
Performance Pay 0.264 *** 0.143 ‐0.400 * ‐0.465 ‐0.175 0.419 1.832 * 1.604
(0.098) (0.119) (0.234) (0.370) (0.293) (0.582) (1.056) (1.479)
‐0.320 ** ‐0.292 * 0.402 1.330 ** 0.551 0.232 ‐2.164 ‐2.134
(0.138) (0.166) (0.326) (0.515) (0.408) (0.811) (1.468) (2.055)
‐0.270 ** ‐0.052 0.368 0.500 ‐0.260 ‐0.372 ‐1.625 ‐4.511 **
(0.136) (0.164) (0.323) (0.510) (0.403) (0.802) (1.448) (2.027)
Observations 226 227 227 227 227 227 222 222
Notes
1 The dependent variable is indicated in the column heading.
2 All specifications include CA visit day dummies.
Table 6: Screening of Female CAs on Different Characteristics
Female Referral
Treatment
Either Gender
Treatment
Perf Pay * Female
Treatment
Perf Pay * Either
Treatment
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Conclusions
• Stylized Fact: women are less likely to receive job referralsthan men (from data in US and Europe)
• Using a recruitment experiment in Malawi, we confirmthat women are disadvantaged by referral systems
• Men choose not to refer women, when given the choice
• Women choose women at about the population average,but make on average low quality referrals
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Conclusions: Economics
• We test several network constraints that could drive thisresult
• Men and women are equally likely to be connected to menand women
• Men are closest to men, but have high quality male andfemale contacts
• Women are not socially closer to one gender than theother, but have low quality networks of women
• Men can screen men well, cannot screen women; womencan screen both men and women to a lesser extent
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Conclusions: Policy
Permitting women’s disadvantage in referral rates has threebenefits to employers:
• It is lower cost for men to refer men than for men to referwomen (since social incentives are higher)
• It is lower cost to get high quality referrals if men aremaking referrals
• Screening benefits of referral systems are maximized whenmen are encouraged to refer only men
• All in all, a hard problem to solve
• Current policies to address gender gap - such as investingin girls’ education - will not be enough to overcome this
• Maybe a role for quota systems in hiring policy
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Comment on Attrition• 80% of applicants make a referral• Reference rate is always really similar across genders,
different across treatments
• Differences in referral quality across gender, withintreatment can be taken at (close to) face value for thosewho make referrals
• Difference in referral quality across treatment will be thecombined effect of some attrition + population averagechoices
• For employers (and to understand actual trends inreferences), the net effect (including attrition) is therelevant dimension in any event
• Implications for e.g. ability to screen are the same ifindividuals attrit because they know their options are bad
• We also simulate the model and recover the samepredictions on the attrition decision and results withinmade referrals
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Can work experience explainresults?
• Men are more likely to have worked at a survey firm in thepast than women
• Working at a survey firm may both enhance your networkand give you better information
• While it does not affect any of the interpretations - ordisadvantages women face - it may be an underlyingmechanism
• We find no differential response among people who haveworked at a survey firm in the past.
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Competition
• Niederle and Vesterlund (2007) find that women areaverse to competition relative to men
• Making a reference involves introducing the employer to apotential competitor for the job
• May have an incentive to refer someone bad (though, amarginal incentive for an informed decision maker -referral is one additional applicant among many)
• May have been particularly salient in our context, asapplicants not yet hired
• However, certainly a relevant incentive in on-the-jobreferrals, too
• Again, suggests a mechanism, without affectinginterpretations or policy prescriptions
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Cross-randomization
• We cross-randomized a treatment designed to make thecompetitiveness more salient
• CAs were told the qualification threshold was either
1 Absolute: scoring better than 602 Relative: scoring in the top half of applicants
• We hypothesize that the relative treatment makes thecompetition more salient (since CAs compete directly withreferrals to be in the top half)
• (admittedly, somewhat weak test)
• Look just at fixed fee referrals to isolate social incentives
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
(1) (2) (3) (4) (5) (6)
Dependent VariableCA
Qualifies
Referral
Qualifies
Referral
Qualifies
CA
Qualifies
Referral
Qualifies
Referral
QualifiesCompetitive Treatment 0.021 0.072 0.052 0.014 0.090 0.227
(0.062) (0.069) (0.121) (0.086) (0.095) (0.165)
Female Treatment 0.094 ‐0.024
(0.116) (0.177)
Either Treatment 0.175 ‐0.160
(0.123) (0.169)
Competitive * Female 0.007 ‐0.263
(0.166) (0.236)
Competitive * Either 0.103 ‐0.142
0.176 (0.236)
Observations 287 232 232 166 133 133
CA Gender Men Men Men Women Women Women
Appendix Table 3: Competition incentives among fixed fee referrals
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
0.2
.4.6
.8R
efer
ral i
s F
emal
e
20 40 60 80 100CA's overall (corrected) score
Referrals of Male CAs Referrals of Female CAs
Figure 2: Gender choice in referrals, by CA performance
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
.2.4
.6.8
1R
efer
ral's
qua
lific
atio
n ra
te
20 40 60 80 100CA's overall (corrected) score
Referrals of Male CAs Referrals of Female CAs
Figure 3: Referral qualification rate, by CA performance
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
0.2
.4.6
.8R
efer
ral q
ualif
ies
20 40 60 80 100CA's overall (corrected) score
Men referring women, fixed Men referring men, fixedMen referring women, perf Men referring men, perf
Figure 6: Referral Qualifies , by Male CA performance
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
0.2
.4.6
.8R
efer
ral q
ualif
ies
20 40 60 80 100CA's overall (corrected) score
Women referring women, fixed Women referring men, fixedWomen referring women, perf Women referring men, perf
Figure 7: Referral Qualifies , by Female CA performance
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Social Payments and Qualification
• Possible (reasonable?) that social payments increase withqualification in the ambient network
• Referrals give you better social transfers if they get the job
• Consistent with our modelling assumptions
• No assumption made about the joint distribution ofαgj ,Qg
j in the ambient network
• Selection rule still leads to decreasing relationship amongnon-dominated referrals
• However, may change interpretation of social payments
• Incentives aligned with employer• differences in quality expectations may lead to women’s
disadvantage if men expect men to be higher quality,women have wrong quality expectations
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Unbiased Expectations of Quality
• Model assumed εgj was mean 0 - allowed us to estimate Qg1
• Already showed that men’s fixed fee referrals of men ARENOT higher ability than men’s fixed fee referrals of women
• And women’s (low quality) fixed fee referrals ARE NOTthe highest quality people they know (they know highquality men)
• So, if CA’s have unbiased expectations: can conclude thatexpectations of quality ARE NOT source of women’sdisadvantage
• But, expectations of quality could be biased
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Biased Expectations of Quality
• If expected social incentives increase in expected referralqualification and expectations are biased (for now, againstwomen)
• Incentives to refer a qualified person are still strictly largerunder perf
• Would expect to see even more men referred under perf(We don’t)
• Would expect to see men restricted to refer women attritmore under perf (We don’t)
• Moreover, some evidence that social incentives are notstrongly correlated with expected referral performance
• Men referring other men are choosing not to refer the bestmen they know under fixed
• Men do respond to incentives
• Similar argument holds for women referring low abilitypeople.
Do JobNetworks
DisadvantageWomen?
BKM
Motivation
Experiment
Set-up
Main Result
Theory
Networkstructure
Connections
HeterogeneousNetworks
SocialIncentives
Screening
Screening
Either Genderversus Restricted
Conclusions
Bonus Slides
Comment 1
Comment 2
Comment 3
Comment 4
Comment 5
Selection rule even with positive relationship