We study the role of social networks in the academic job market for graduate students of Economics. We find that the connectedness of a student’s advisor in the coauthor network significantly improves her job market outcome. We use two identification strategies and find that a) higher Eigenvector centrality of an adviser leads to her student getting placed at a better ranked institution, and b) larger distance between an adviser and an institution decreases the probability that her students are placed there. Our study sheds light on the importance of social connections in a labour market where information frictions regarding job openings are virtually absent.

1. Adviser connectedness and placement
outcomes in the economics job market
Michael E. Rose 1 Suraj Shekhar 2
1Max Planck Institute for Innovation and Competition, Munich, Germany
2Ashoka University, Sonipat, India
1

2. The Economics Job Market
“The economics jobs market . . . has its own characteristics.
Informal contacts - and phone calls by your advisors and friends -
are important . . . ”
David Colander (1997): “Surviving as a Slightly Out of Sync Economist”, in: Steven G. Medema and Warren J.
Samuels (eds): “Foundations of Research in Economics: How Do Economists Do Economics?”, Edward Elgar:
Cheltenham, UK and Northampton, MA
2

3. The Economics Job Market
“The economics jobs market . . . has its own characteristics.
Informal contacts - and phone calls by your advisors and friends -
are important . . . ”
David Colander (1997): “Surviving as a Slightly Out of Sync Economist”, in: Steven G. Medema and Warren J.
Samuels (eds): “Foundations of Research in Economics: How Do Economists Do Economics?”, Edward Elgar:
Cheltenham, UK and Northampton, MA
? Does the ‘connectedness’ (not just direct contacts) of a PhD
advisers matter for the placement outcomes of their students?
○ May reduce uncertainty regarding applicant quality Baruffaldi
et al., RP 2016
○ Student more known
○ Network as input for student’s JMP
○ Hiring (unconsciously) spend more time 2

4. Connected literatures & our contributions
1. Theory: Social networks transmit information about new
openings and applicant quality Montgomery AER 1991,
Calvó-Armengol, JET 2004
• Show empirically that at least second function exists
2. Firms more likely to hire aquantaintances of workers Hensvik
and Skans, JLE 2016; Dustman, Glitz, Schönberg and Brücker, REStud
2016; Burks, Cowgill, Hoffmann, Housman, QJE 2017
• Adopt network view with real-world network
3. Initial placement determines career Oyer, JEP 2006; Smeets,
Warzynski and Coupé, JEP 2006; Krueger and Wu, JEE 2000; Athey,
Katz, Krueger, Levitt and Poterba, AER 2016
• Highlight role of adviser’s networks in placement process
3

5. Outline
1. Adviser’s centrality → placement rank
• 2SLS; Instrument: adviser’s co-authors mean Eigenvector
centrality in network w/o adviser
• Adviser FE
2. Adviser-Placement connection → placement probability
• Shocks to social distance induced by deceased authors
• Adviser FE
4

6. Multiple different data sources
1. Econ PhD graduates from North-American universities 2000/01-2003/04
(DOI: 10.7910/DVN/ADSCLU)
• N = 3,179
2. Advisers of students
• N = 3,153
3. Placements of students
• N = 2,452
4. Rankings of placements (Tilburg method using Scopus data)
• N = 1,222
5. Co-Author network
• 266,027 publications from 363 journals
• Up to 52,237 authors
6. Faculty rosters ofEcon/Finance/Mgmt/Acc departments
• 18,310 scholars from 812 departments, 7,845 in network
• 33 deceased authors
5

7. Data: Placement network for North America
6

8. Data: Who are the advisers?
Name Students School Citations Euclid Experience
1 Andrei Shleifer 16 Harvard University 5469 1099.60 17
2 Daron Acemoglu 15 Massachusetts Institute of Technology 722 159.52 10
David E. Card 15 University of California, Berkeley 643 200.33 21
4 Roger R. Betancourt 14 University of Maryland 112 38.08 32
5 Carlos A. Végh 13 University of California, Los Angeles 307 115.65 14
John Y. Campbell 13 Harvard University 2228 516.03 17
Peter C.B. Phillips 13 Yale University 5176 1771.50 30
Ronald Andrew Ratti 13 University of Missouri 52 27.02 26
Thomas D. Willett 13 Claremont Graduate University 398 91.94 35
10 Arnold C. Harberger 12 University of California, Los Angeles 92 39.83 47
Dominick Salvatore 12 Fordham University 149 44.37 31
12 Abhijit V. Banerjee 11 Massachusetts Institute of Technology 688 335.69 12
Michael Grossman 11 City University of New York 939 259.24 29
14 John C. Haltiwanger 10 University of Maryland 528 224.11 24
Lawrence F. Katz 10 Harvard University 1722 633.56 22
Olivier Jean Blan-
chard
10 Massachusetts Institute of Technology 1864 606.88 24
Richard E. Wagner 10 George Mason University 225 102.77 37
Robert A. Moffitt 10 Johns Hopkins University 655 145.04 26
Robert M. Townsend 10 University of Chicago 561 368.18 26
Samuel Bowles 10 University of Massachusetts 804 276.27 36
7

9. Network and Eigenvector Centrality
1. 266,027 publications; from 1936 until T +1
• 363 journals (EconLit), data from Scopus
• Up to 52,237 different authors linked upon joint publication
• weight = number of joint publications ×0.95T−t
• Adviser’s neighbors centrality in network without adviser
8

10. Network and Eigenvector Centrality
1. 266,027 publications; from 1936 until T +1
• 363 journals (EconLit), data from Scopus
• Up to 52,237 different authors linked upon joint publication
• weight = number of joint publications ×0.95T−t
• Adviser’s neighbors centrality in network without adviser
2. Eigenvector centrality
• Measure of influence Calvó-Armengol et al. (REStud 2009);
Banerjee et al. (Science 2013); Cruz et al. (AER 2017)
EVi =
1
λ
X
j∈G(i)
EVj
8

11. Identification strategy
• Hold adviser-effects fixed across years
• Instrument adviser’s connectedness with average Eigenvector
centrality of coauthors in network w/o adviser
• Changes in the adviser’s coauthors’ centrality
• Assumptions:
1. Students do not foresee changes to their adviser’s
connectedness when matching
2. No unobserved time-variant adviser characteristics relevant
→ drop in observations to 579 students Final distribution
9

12. Summary of continuous variables
Mean SD Min. Max.
Placement score 2.74 3.76 -0.31 14.83
Adviser centrality 0.18 1.34 -0.17 8.87
Adviser’s coauthors centrality 0.06 0.52 -0.17 4.43
Adviser Euclid 235.24 357.80 2.83 2782.00
Adviser experience 20.06 8.10 3.00 47.00
Student male 0.71 0.45 0.00 1.00
PhD school rank 67.94 159.94 1.00 1193.00
Observations 579
10

13. Summary of continuous variables
Mean SD Min. Max.
Placement score 2.74 3.76 -0.31 14.83
Adviser centrality 0.18 1.34 -0.17 8.87
Adviser’s coauthors centrality 0.06 0.52 -0.17 4.43
Adviser Euclid 235.24 357.80 2.83 2782.00
Adviser experience 20.06 8.10 3.00 47.00
Student male 0.71 0.45 0.00 1.00
PhD school rank 67.94 159.94 1.00 1193.00
Observations 579
Example Euclidean Index:
p
1232 +122 +142 +402 +2352 = 268.9
10

14. Students of well-connected advisers are placed better
Adviser centrality Placement score
Adviser centrality 0.692∗∗ 0.988∗∗
(0.300) (0.403)
Adviser’s coauthors centrality 1.336∗∗∗
(0.336)
Adviser’s second neighbour 1.901∗∗∗
centrality’ (0.636)
Adviser Euclid −0.005 −0.004 0.003 0.005
(0.004) (0.004) (0.003) (0.004)
Student female 0.071 0.000 −0.565 −0.571
(0.112) (0.100) (0.589) (0.596)
PhD school rank −0.002 −0.001 −0.003 −0.003
(0.002) (0.003) (0.005) (0.006)
Adviser FE X X X X
Adviser experience FE X X X X
Field FE X X X X
Graduation year FE X X X X
N 579 579 579 579
# of advisers 194 194
Effective F 15.8 8.9
2-way cluster on PhD School and adviser 11

15. Robustness Checks
• Centrality leads show
• Adviser popularity proxied by citation trajectory change show
• School rank polynomials show
• Random adviser assignments show
12

16. Students don’t go to their advisor’s coauthors’ affiliation
2 4 6 8 10 12
1
Socialdistancetonearestplacementfacultymember
0
20
40
60
80
100
120
140
Number
of
students
Adviser;N=780
Committee;N=291
13

17. Social connections and the placement process
• Not centrality, but connections and their length
• Unit of analysis: Every possible connection between adviser a
and placement k in t
• Dep. variable: Whether advisor a placed a student in k in t
• Exploit link length increase resulting from deaths of authors
14

18. Social connections and the placement process
• Not centrality, but connections and their length
• Unit of analysis: Every possible connection between adviser a
and placement k in t
• Dep. variable: Whether advisor a placed a student in k in t
• Exploit link length increase resulting from deaths of authors
Placementakt =β0 +β1IncreaseInSocialDistanceAfterDeathakt+
β2SocialDistanceBeforeDeathakt+
γ1AdviserControlskt +γ2PlacementRankkt+
a+PhDSchoolj +t +²j
14

19. Increase in social distance decreases probability of placement
Placed student
Increase in social distance −0.002∗∗∗ −0.002∗∗∗ −0.005∗∗
after death (0.000) (0.001) (0.002)
Social distance before death −0.001∗∗∗ −0.001∗∗∗ −0.002∗∗∗
(0.000) (0.000) (0.000)
Adviser Euclid 0.000 0.000 0.000
(0.000) (0.000) (0.000)
PhD school score 0.000 0.000 −0.000
(0.000) (0.000) (0.000)
Placement score 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗
(0.000) (0.000) (0.000)
Sample Full PhD prog. PhD prog. that hired
Adviser FE X X X
Adviser experience FE X X X
Graduation year FE X X X
N 406,714 238,644 89,949
# of advisers 532 532 386
Mean 0.002 0.003 0.006
Notes: SE clustered around adviser and PhD school; Constant omitted 15

20. Robustness Checks
• Control for topical overlap (distance in citation network) show
• Logistic regression show
• Random removals show
16

21. Conclusion
• Average year-on-year centrality increase in adviser’s
Eigenvector centrality improves the student’s placement rank
by about 6 ranks
• An increase in the distance between an adviser and a
university by one, decreases the probability that her student
move there declines by about 0.2 percentage points.
• "Social connections matter even in a labour market where
information frictions regarding job openings are virtually
absent"
Thank you!
17

22. Data: Students in final sample differently distributed
Initial Final
N Share N Share
Ranks 1-30 1313 41.3 363 62.7
Ranks 31-100 927 29.2 152 26.3
Ranks 101-300 488 15.4 28 4.8
Other 451 14.2 36 6.2
back
18

23. Data: Advisors w/ multiple students more prolific
0
500
1000
1500
2000
2500
3000
3500
***
Schoolscore
0
5
10
15
20
Experience
0
20
40
60
80
100
120
140
160
***
Euclideanindex
<=1 >1
#studentsplacedacademicallyindiff.years
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1e 8
***
Eigenvector
<=1 >1
#studentsplacedacademicallyindiff.years
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1e 7
***
Coauthors'meanEV
<=1 >1
#studentsplacedacademicallyindiff.years
0
2
4
6
8
10 ***
Degree
back 19

24. Centrality: Eigenvector leads
Placement score
Adviser centrality in t+1 0.107 −0.358
(0.373) (0.541)
Adviser centrality in t+2 −0.903 −0.605
(0.558) (0.691)
Adviser Euclid 0.000 0.002 0.001 0.001
(0.002) (0.004) (0.002) (0.004)
Student female −0.555 −0.628 −0.529 −0.619
(0.577) (0.631) (0.597) (0.613)
PhD school rank −0.003 −0.005 −0.004 −0.005
(0.006) (0.005) (0.005) (0.005)
Adviser FE X X X X
Adviser experience FE X X X X
Field FE X X X X
Graduation year FE X X X X
N 578 574 578 574
# of advisers 194 194 194 194
Effective F 32.1 9.2 11.2 3.6
Notes: SE clustered around adviser and PhD school; Constant omitted
back
20

25. Centrality: Controlling for time-variant adviser popularity
Placement score
Adviser centrality 0.706∗∗ 1.001∗∗
(0.298) (0.400)
Citation growth rate 96-99 4.290 3.355
(4.154) (4.642)
Citation growth past 3 years −0.138 −0.136
(0.096) (0.097)
Citation growth last year 0.223 0.227
(0.160) (0.160)
Adviser Euclid 0.003 0.005
(0.003) (0.004)
Student female −0.574 −0.580
(0.591) (0.599)
PhD school rank −0.004 −0.004
(0.005) (0.005)
Adviser FE X X
Adviser experience FE X X
Field FE X X
Graduation year FE X X
IV Direct Indirect
N 579 579
# of advisers 194 194
Effective F 15.6 8.7
Notes: SE clustered around adviser and PhD school; Constant and 1st stage
results omitted back
21

26. Centrality: Controlling for PhD school rank polynomials
Adviser centrality Placement score
Adviser centrality 0.690∗∗ 0.685∗∗
(0.298) (0.300)
Adviser’s coauthors centrality 1.336∗∗∗ 1.334∗∗∗
(0.336) (0.336)
Adviser Euclid −0.005 −0.005 0.003 0.003
(0.004) (0.004) (0.003) (0.003)
Student female 0.071 0.069 −0.571 −0.577
(0.111) (0.112) (0.590) (0.587)
PhD school rank −0.002 −0.006 −0.000 −0.012
(0.002) (0.004) (0.013) (0.017)
PhD school rank2 0.000 0.000 −0.000 0.000
(0.000) (0.000) (0.000) (0.000)
PhD school rank3 −0.000 −0.000
(0.000) (0.000)
Adviser FE X X X X
Adviser experience FE X X X X
Field FE X X X X
Graduation year FE X X X X
N 579 579 579 579
# of advisers 194 194
Effective F 15.8 15.7
Notes: SE clustered around adviser and PhD school; Constant omitted
back
22

27. Centrality: Random adviser assignments
0 10 25 50 75 100
Shareofestimations(in%)
0.00
0.05
0.10
0.50
1.00
Estimated
p
value
0 10 25 50 75 100
Shareofestimations(in%)
0.00
0.05
0.10
0.50
1.00
Estimated
p
value
Randomisation within field Randomisation matching distribution
back
23

28. Distance: Controlling for topical overlap (citation distance)
Placed student
Increase in social distance −0.002∗∗∗ −0.002∗∗∗ −0.005∗∗
after death (0.000) (0.001) (0.002)
Social distance before death −0.001∗∗∗ −0.001∗∗∗ −0.002∗∗∗
(0.000) (0.000) (0.000)
Citation distance −0.001∗∗∗ −0.001∗∗ −0.002∗∗∗
(0.000) (0.000) (0.001)
Adviser Euclid 0.000 0.000 0.000
(0.000) (0.000) (0.000)
PhD school score 0.000 0.000 −0.000
(0.000) (0.000) (0.000)
Adviser FE X X X
Adviser experience FE X X X
Graduation year FE X X X
N 403,061 237,652 89,710
# of advisers 531 531 385
Mean 0.002 0.003 0.006
Notes: SE clustered around adviser and PhD school; Constant omitted
back
24

29. Distance: Logistic regression
Placed student
Placed student
Increase in social distance −1.599∗ −1.606∗ −1.178
after death (0.928) (0.928) (0.958)
Social distance before death −0.531∗∗∗ −0.394∗∗∗ −0.457∗∗∗
(0.041) (0.046) (0.057)
Adviser Euclid 0.000 0.000 0.000
(0.001) (0.001) (0.001)
PhD school score 0.000 0.000 −0.000
(0.000) (0.000) (0.000)
Placement score 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗
(0.000) (0.000) (0.000)
Sample Full PhD prog. PhD prog. that hired
Adviser FE X X X
Adviser experience FE X X X
Graduation year FE X X X
N 396,743 232,807 89,712
# of advisers 519 519 386
Mean 0.002 0.003 0.006
Notes: SE clustered around adviser and PhD school; Constant omitted
back
25

30. Distance: random removals
Placed student
Increase in social distance 0.002 0.006 0.017
after random death (0.002) (0.004) (0.013)
Social distance before death −0.001∗∗∗ −0.001∗∗∗ −0.002∗∗∗
(0.000) (0.000) (0.000)
Adviser Euclid 0.000 0.000 0.000
(0.000) (0.000) (0.000)
PhD school score 0.000 0.000 −0.000
(0.000) (0.000) (0.000)
Placement score 0.000∗∗∗ 0.000∗∗∗ 0.000∗∗∗
(0.000) (0.000) (0.000)
Sample Full PhD prog. PhD prog. that hired
Adviser FE X X X
Adviser experience FE X X X
Graduation year FE X X X
N 406,714 238,644 89,949
# of advisers 532 532 386
Notes: SE clustered around adviser and PhD school; Constant omitted
back
26