Lineages of Scholars in pre-industrial Europe: Nepotism vs. - - PowerPoint PPT Presentation

Lineages of Scholars in pre-industrial Europe: Nepotism vs. Intergenerational Human Capital Transmission

David de la Croix1 Marc Go˜ ni2

1Universit´

e catholique Louvain and CEPR

2University of Vienna

CSEF-DISES, 5 February 2020

Motivation

◮ Parent-child correlations suggest low persistence of SES outcomes.

“from shirtsleeves to shirtsleeves in three generations”—Becker and Tomes (1986).

◮ Multi-generation estimates find higher persistence (Clark 2015).

◮ Children inherit highly-persistent underlying endowments. ◮ Noisily proxied by observed outcomes (e.g., income).

◮ What are these underlying endowments?

◮ Human capital, genes, and abilities. ◮ Nepotism prevalent in certain top professions.

◮ Human capital vs. nepotism have different policy implications. ◮ Econometrically, they are associated to two different biases:

measurement error and selection.

2

This paper

◮ We propose a new method to disentangle nepotism vs. human

capital.

◮ Families of university scholars and members of scientific academies

(1088–1800).

3

Human capital transmission vs. Nepotism

Ailhaud Matthaeus Matthaeus Mathon Chicoyneau Cusson Fizes Magnol Saporta Aquin Boulduc Boulduc Duverney Buxtorf Buxtorf Accorumbonus Becker Berger Bernhold Burman Burman Geuns Bontius Bobart Leyser Leyser Leyser Leyser Vallisneri Terrasson Chartier Duret Goulu Lemerre Vauvilliers Burnett Burnett Sassenus Burnett Cavendish Darwin Fuller Bicais Villars Fontaine Harpprecht Helvetius Bierling Calvert Cavendish Bartholin Bartholin Bartholin Calixt Weise Bacmeister Bacmeister Lindemann Quistorp Quistorp Quistorp Swantenius Willebrand Grape Habermann Becker Becker Eschenbach DeFerriere Cassini Cassini Cassini Cassini Cassini Cassini Cassini Bernoulli Bernoulli Bernoulli Bernoulli Bernoulli Bernoulli Bernoulli Bernoulli Bernoulli Bernoulli Bernoulli Bernoulli Bernoulli Bernoulli Euler Euler Euler Grynaeus Grynaeus Zwinger Zwinger Zwinger Zwinger Zwinger Zwinger Zwinger Furstenau Bauhin Bauhin Bauhin Stupanus Faesch Wettstein Wettstein Werenfels Battier Konig Stahelin Ryhiner Platter Hopfner Horst Platter Platter Petre Camerarius Camerarius Hyde Howard Sozzini Mesmes Deutsch Flottwell Hahn Rast Rast Sanden Sanden Sanden Hert Segers Frommann Carrere Accursio Fornelius Celsius Hydren Aurivillius Amaseo Arntzenius Alting Eyssonius Matthaeus Mulerius Wijnpersse Cremer Gorter BernidegliAntoni Busch Duising Goclenius Hombergk Hunnius Kleinschmidt Kleinschmidt Kornmann Krafft Lonicerus Lonicerus Matthaeus Matthaeus Mentzer Molther Nigidius Schroder Schroder Vultejus Waldschmiedt Zaunschliffer Bebel Brenz Cellius Gmelin Gmelin Gmelin Gmelin Arnoldi Gmelin Hafenreffer Harpprecht Baumer Bechtold Hochstetter Hoffmann Hoffmann Mauchart Mogling Mogling Mogling Mogling Mogling Osiander Ploucquet Pregizer Scheinemann Schott Schott Tafinger WeiganmeierFomann Gerhard Hamberger Hamberger Hellfeld Krause Major Posner Sagittarius Schelhammer Schenck Schmid Schroter Schroter Schroter Slevogt Slevogt Stock Voit Walch Walch Walch Wedel Wiedeburg Wiedeburg Voet Voet Albinus Bontius Gronovius Heurnius Honert Schooten Schooten Schultens Schultens Spanheim Vinnius Vitriarius Vorstius Walaeus Patin Patin Hoffmann Rittershaus Beuther Lorry Baudiss Stryk Crocius Argoli Arduino BoscDAntic Cusson Michelotti Michelotti Michelotti VaccaBerlinghieri Brendel Alefeld Boecler Dasypodius Boecler Junius Lorentz Marbach Marbach Sebiz Bohmer Bohmer Bohmer Bohmer Eichhorn Gatterer Mayer Vogel Volckamer Kortholt May Mentzer Mentzer Mollenbeck Mollenbeck Nebel Scheibler Schulz Sinold Snell Snell Thom Valentini Borcholten Crell Engelbrecht Engelbrecht Forster Horneius Kipping Meibom Meibom Meibom Meibom Remer Scheurl Schmidt Schrader Wernsdorf Wiedeburg Lidbeck Papke Stobaeus Stobaeus Burman Apinus Baier Baier Bruckmann Crell Ehrhart Ettmuller Fehr Fehr Garmann Gockel Gorter Grass Haller Heister Cambout Isenflamm Lehmann Linck Mauchart Nebel Pohl Preuss Seip Tralles Trommsdorff Trumph Valentini Valentini Vater Wedel Wepfer Widmann Brugmans Huber Schotanus Vitringa Voorda Wayen Ypey Beauvilliers BussyRabutin Colbert Riedl Dundas Franklin Monro Stewart Canterzani Gavinet Willermoz Sozzini Bignon Cureau Cambout Costeo Boivin Garcaeus Celsius Balthasar Balthasar Battus Bering Colberg Muhrbeck Gerdes Helwig Lutkemann March Mascow Mayer Overkamp Rhaw Rhaw Runge Runge Runge Stephani Westphal Monro Monro Filelfo Restout Maty Lysius Martyn Achard Lauremberg CoquebertdeMonbret Cartheuser Dacheroden Hamberger Rumpel Rumpel Suckow Trommsdorff Willemet Abrek Pacheco Adriani Adriani Akakia Malvasia Hayes Molesworth Felibien Warton Molyneux Verdugo Cresques Pelargus Cocceji Meister Cramer Cramer Franck Franck Hensler Kortholt Musaus Opitz Dunlop Schultingh Tintoretto Montesquieu Sascerides Langguth Wandal Scavenius Mauvillon Brochmand Buchwald Dillenius Horrebow Hire Hire Kirchmaier Fornier Gothofred Vogli Zoppio Serres Pryss Wishart Wishart Bourdelin Gorski Turretin Turretin Blackwell Cocchi Doni Pidoux Cardano Ettmuller Racine Leveque Freitag Freitag Frisius Pocquet Fiennes Turgot Argenville Ranchin Gonzaga Argenson Argenson Argenson Lemery Scarburgh Torre Corsini Filicaia Bentinck Tronchin Maurice Hottinger Hottinger Leger Hawenreuter Osborne Macclesfield Gezelius Moreau Moreau Zoppio Marchetti Percy Pietre Rechenberg Mercklin Douglas Gadolin Aepinus Vater Mayer Mikan Barzizza Mortimer Pohl Planta Cravetta Trommsdorff Charteris George Nymannus DegliOddi Patin Riolan Vaillant Wormius Wormius Wormius Dreier GallaratiLomeno Tibbon Tibbon Robartes Abigdor Borgognoni Aiton Sebiz Sebiz Koch Pantaleon Ribiers Broussonnet Chastelain Ranchin Maynier Sprat Yorke Yorke De Trevor Wren Wollaston Wollaston Willett Arsendo Vieussens Bellers Herault Maresius Lange Forster Trudaine Forster Forster Forster Trudaine Amelot Morand Valliere Ormesson Bouillet

2 4 6 8 10 Sons' publications (in logs) 2 4 6 8 10 12 Fathers' publications (in logs)

b = 0.51 (s.e. 0.02)

2 4 6 8 10 12 Sons' publications (in logs) 2 4 6 8 10 12 Fathers' publications (in logs)

KS = 0.10 (p-val 0.00)

Preview of the Results

◮ High prevalence of nepotism in pre-industrial universities.

◮ 15% scholars’ sons are nepotic scholars. ◮ Nepotism decreases during Scientific Revolution and

Enlightenment.

◮ Nepotism more prevalent in Catholic universities.

◮ High rate of transmission of human capital (≈ 0.6):

◮ Higher than suggested by parent-child correlations. ◮ Lower than multigenerational estimates. ◮ Varies across time, institutions, fields of study...

5

Related literature

◮ Intergenerational persistence:

◮ Parent-child: Becker and Tomes (1986), Solon (1999),

Mazumder (2005), Corak (2006), Mulder et al. (2009), Black and Devereux (2011), Chetty et al. (2014), Zylberberg (2016), Bordieu et al. (2017).

◮ Multiple generations: Guell et al. (2015), Clark (2015), Clark

and Cumins (2015), Braun and Stuhler (2018), Collado et al. (2019), Lindhal et al. (2015), Adermon et al. (2018), Long and Ferrie (2018), Colagrossi et al. (2019).

◮ Top professions: Lentz and Laband (1989,1992), Dunn and

Holtz-Eakin (2000), Perez-Gonzalez (2006), Bennedsen et al. (2007), Dal Bo et al. (2009), Durante et al. (2011), Raitano and Vona (2018), Bell et al. (2018), Aina and Nicoletti (2018).

◮ Upper-tail human capital: Grief (2006), Cantoni and Yuchtman

(2014), Mokyr et al. (2002), Mokyr (2016), Squiacciarini and Voigtlander (2015), de la Croix, Doepke, Mokyr (2018).

6

Road map

• 1. Introduction
• 2. Methods
• 3. Data
• 4. Structural estimation
• 5. Extensions
• 6. Conclusion

7

Parent-child intergenerational elasticity (b)

yi,t+1 = b yi,t + ei,t+1 ,

◮ where i families; t parents; t + 1 children; y in logs.

ˆ b yt country source 0.31-0.41 Wealth

• Agric. societies

Mulder et al. (2009) 0.48-0.59 Wealth UK Harbury & Hitchins (1979) 0.47 Earnings USA Corak (2012) 0.31-0.33 Earnings UK Corak (2012) 0.19-0.26 Earnings Sweden Jantti et al. (2006) 0.46 Education USA Hertz et al. 2007 0.35 Education Sweden Lindahl et al. 2015

→ Reversion to the mean in three generations!

more half-life

8

Measurement error

Children inherit (unobserved) endowment h which translates into (observed) outcome y imperfectly.

◮ First-order Markov process of endowments transmission:

hi,t ∼ N(µh, σ2

h) ,

hi,t+1 = βhi,t + ui,t+1

where ui,t+1 ∼ N(µu, σ2

u).

◮ Measurement error:

yi,t = g(hi,t, ǫi,t) , yi,t+1 = g(hi,t+1, ǫi,t+1)

where ǫi,t, ǫi,t+1 ∼ N(0, σ2

ǫ) and g(·) is linear.

9

Bias and proposed solutions

E(ˆ b) = β σ2

h

σ2

h + σ2 ε

:= β θ , where θ < 1

• 1. Multi-generation elasticities (Braun Stuhler 2018):

bm = Cov(yi,t, yi,t+m) Var(yi,t) = θβm

◮ Identification: b2

b1 = θβ2 θβ = β.

◮ Data: multiple generations (census).

• 2. Surname-average (Clark 2015, Clark Cummins 2015).

more

• 3. Informational content of surnames (Guell et al. 2015).

more

• 4. Kinship correlations (Collado et al. 2019).
• 5. Transition matrices (Bordieu et al. 2017, Zylberberg 2016).

10

Estimates of β

ˆ β yt Data Source 0.75 Wealth UK probate (1858-2012) Clark & Cummins (2015) 0.7–0.9 Oxbridge UK (1170-2012) Clark & Cummins (2015) 0.65 Occupation Germany, 3 gen. Braun & Stuhler (2016) 0.8 Education 2001 census, Spain Collado et al. 2018 0.74 Education EU-28, 3 gen. Colagrossi et al. (2019) 0.5–0.6 Education Germany, 4 gen. Braun & Stuhler (2016) 0.6 Education Sweden, 4 gen. Lindahl et al. (2015) 0.6 Education 2001 census, Spain Guell et al. (2014) 0.5 Earnings Sweden, 4 gen. Lindahl et al. (2015)

→ Higher persistence than parent-child correlations suggest.

11

Selection

◮ Whether observations are sampled or not may depend on

inherited endowments hi,t.

◮ Selection if data on a subgroup of the population:

◮ Testators. ◮ Top professions (e.g., doctors, civil servants).

◮ Selection in census data:

◮ Migration/early death attrition depends on ht. ◮ Lineages in same occupation more likely to be tracked. ◮ Recall bias in life-history data collected retrospectively.

12

◮ Selection indicator:

si =

• 1 if family i is used in the estimation,

0 otherwise .

◮ Parent-child elasticity in observed outcome y:

E(ˆ b) = b = θβ if Cov (si · [hi,t + ǫi,t], si · ei,t+1) = 0 , b + ξ if Cov (si · [hi,t + ǫi,t], si · ei,t+1) = 0 , where ξ = Cov (si · [hi,t + ǫi,t], si · ei,t+1) Var (si · [hi,t + ǫi,t]) .

◮ Multiple generations ratios biased if selection (nepotism)

changes across generations.

13

Model with selection

◮ First-order Markov process of endowments transmission:

hi,t ∼ N(µh, σ2

h) ,

hi,t+1 = βhi,t + ui,t+1 ,

where ui,t+1 ∼ N(µu, σ2

u).

◮ Measurement error:

yi,t = max(κ, hi,t + ǫi,t) , yi,t+1 = max(κ, hi,t+1 + ǫi,t+1)

where ǫi,t, ǫi,t+1 ∼ N(0, σ2

ǫ).

◮ Selection and nepotism:

Potential families: i ∈ I. Observed families: P = {i | hi,t > τ, hi,t+1 > τ − ν} ⊂ I .

more

14

Stationarity

Assumption: stationary human capital distribution (among potential entrants’ population).

◮ Formally, hi,t and hi,t+1 drawn from same distribution:

µu = (1 − β)µh σ2

u

= (1 − β2)σ2

h ◮ Stationary (inherited) human capital:

hi,t+1 = βhi,t + (1 − β)µh + ωi,t+1 ,

where ωi,t+1 ∼ N(0, (1 − β)2σ2

h).

15

Road map

• 1. Introduction
• 2. Methods
• 3. Data
• 4. Structural estimation
• 5. Extensions
• 6. Conclusion

16

Database

◮ Families in 86 universities and 30 scientific academies in

Pre-industrial Europe (1088-1800).

◮ c. 1,000 lineages of scholars (969 fathers and 1,103 sons). ◮ 135 lineages with three or more generations. ◮ Scholars’ publications (library holdings in Worldcat). ◮ Original sources:

◮ Books and catalogues of universities and academies. ◮ Matched with biographical dictionaries. ◮ Worldcat.

17

Data collection example (1/2)

18

Data collection example (2/2)

Bicaise, Honoré 1590-

Overview

Works: 27 works in 58 publications in 3 languages and Genres: Quotations Roles: Author, Editor, Creator Classifications: R128, 610.14

Publication Timeline

Alternative Names Bicaise, Honoré b. 1590 Bicaisius, Honoratus. Bicaissius, Honoratus. Bicaissius, Honoratus 1590- Bicays, H. Bicays, H. 1590- Bicays, H. (Honoré), 1590-

By Posthumously by About 163… 163… 163… 164… 164… 165… 165… 165… 166… 166… 167… 167… 167… 168… 168… 169… 169… 169… 170… 170… 171… 171… 171… 172… 172… 173… 173… 173… 174… 174… 175… 175… 1

267 library holdings Bicaise, Michel active 17th century

Overview

Works: 3 works in 5 publications in 1 language and Roles: Author

Publication Timeline

Alternative Names Bicais, Michel active 17th century

By Posthumously by About 1660-1661 1661-1662 1662-1663 1663-1664 1664-1665 1665-1666 1666-1667 1667-1668 1668-1669 1669-1670 1670-1671

4 library holdings

19

0km 300km 600km

N

Institutions with 1 lineage with 2−5 with 6−25 with 26−50 with 51−100 with 101−157 Father's birth place Son's birth place

more

20

Families of scholars overtime

Fathers' reference date Frequency

1100 1200 1300 1400 1500 1600 1700 1800

50 100 150 200 250

• ●
• ●
• ●
• ●
• ●
• ●
• ●
• 1100

1200 1300 1400 1500 1600 1700 1800 2 4 6 8 10 12

Fathers' reference date publications (in logs)

21

Qualitative evidence

Human capital transmission: Jean Bauhin (1541–1613) learned very early the ancient languages and humanities. His father, Jean Bauhin, was his first master in the study of medicine and of all the underlying sciences. (Michaud 1811) Nepotism: After sixty years of teaching canon law in Salamanca, Juan Alfonso Benavente (-1478) retired in 1463. He retained his chair and his lectures were taught by substitutes, including his son Diego Alfonso de Benavente (1430-1512). Finally, on November 19, 1477, Benavente resigned to his chair on the condition that his son was firmly appointed to it. (Diccionario Biogr´ afico Espa˜ nol)

Chicoyneau

22

Descriptive evidence

value s.e.

• bs.
• A. Intergenerational correlations in pubs.

Father-son, intensive margin 0.340 0.044 601 Father-son with zero pubs. 0.228 0.013 1,103 Grandfather-grandson, intensive marg. 0.292 0.128 54

• B. Publications’ distribution

Fathers with zero pubs. 0.302 0.015 969 Sons with zero pubs. 0.375 0.015 1,103 Fathers median 4.304 0.182 969 Sons median 3.045 0.263 1,103 Fathers 75th percentile 6.707 0.107 969 Sons 75th percentile 5.940 0.114 1,103 Fathers 95th percentile 8.576 0.098 969 Sons 95th percentile 7.879 0.077 1,103 Fathers mean 3.932 0.103 969 Sons mean 3.177 0.092 1,103 Note: Publications is 1 + log of library holdings. 23

Stylized facts

Fact 1: High elasticity of publications across generations.

◮ ≈ wealth-elasticity in pre-modern societies (Mulder et al.

2009).

◮ Father-son elasticity of 0.34. ◮ Grandfather-grandson elasticity of 0.29 > 0.342 = 0.12.

Fact 2: Fathers’ publication distribution FOSD that of sons.

◮ Larger differences below the median.

more

24

Road map

• 1. Introduction
• 2. Methods
• 3. Data
• 4. Structural estimation
• 5. Extensions
• 6. Conclusion

25

Minimum distance estimation

◮ Model’s parameters: β, ν, σe, κ, µh, and σh.

more

◮ Empirical moments: j = 1, ..., 13.

more

min

p V (p) =

• j

λj ˆ mj(p) − mj σmj 2 p: parameter vector; mj: empirical moment; ˆ mj(p): simulated moment; σmj: SD of empirical moment; λj = λ > 0 for Pr(yt=0), Pr(yt+1=0), ρ(yt|>0, yt+1|>0); λj = 1 otherwise.

◮ Differential Evolution algorithm. ◮ We set τ = 0 and recover µu, σu from stationarity conditions.

identification

26

Identified parameters

Parameter value s.e. Intergenerational elasticity of human capital β 0.581 0.049 Nepotism ν 5.632 1.526

• Std. deviation of shock to publications

σe 0.386 0.130 Threshold of observable publications κ 2.201 0.188 Mean of human capital distribution µh 2.444 0.490

• Std. deviation of human capital distribution

σh 3.451 0.241

27

Nepotism, τ − ν = −5.63

Scholar’s son can become a scholar with a human capital:

◮ 2.4 standard deviations lower than average potential scholar. ◮ 1.6 standard deviations lower than marginal scholar.

Simulate model with identified parameters and ν = 0:

◮ 15% of scholars’ sons are nepotic scholars.

Replace nepotic scholars with average scholar:

◮ Increase in mean publications by 17.9% 28

Intergenerational human-capital elasticity, β = 0.58

◮ Higher than parent-child elasticities in wealth, earnings, education. ◮ Lower than surname-average estimates of β ≈ 0.75 (Clark 2015). ◮ Bottom-range of multigeneration (Braun & Stuhler 2018) and ICS

(Guell et al. 2015) estimates.

more (b) more (β)

29

Elasticity’s comparison

◮ Multiple-generations:

ˆ β = ˆ b2 / ˆ b1,

where yi,t+j = ˆ bj yi,t + ei,t+j.

method value s.e. N Multiple-generations ˆ β 0.84 0.12 135 Multiple-generations ˆ βA 0.86 0.11 135 Model w/o nepotism βν=τ 0.86 0.07 1,103 Baseline (full) model β 0.58 0.05 1,103

Note: ˆ βA = bG1−G3 / average

• bG1−G2, bG2−G3
• , where G1

fathers (t); G2 sons (t +2); and G3 grandsons (t +2). Boot- strapped s.e. in parenthesis.

30

Model fit

Pr(0) Q75 Q50 mean Q95

.2 .4 .6 .8 1 2 4 6 8 10 12 14

Sons

Pr(0) Q50 Q75 mean Q95

.2 .4 .6 .8 1 2 4 6 8 10 12 14

Fathers

31

Model w/o Model w/o Baseline selection nepotism model Data Parameters β 0.548 0.857 0.581 . ν 5.632 . τ −∞ . Moments Fathers with zero pubs. 0.362 0.365 0.305 0.305 Sons with zero pubs. 0.364 0.365 0.376 0.375 Median, fathers 4.408 4.457 3.502 4.302 Median, sons 4.409 4.451 3.223 3.055 75th percentile, fathers 5.971 6.030 5.513 6.721 75th percentile, sons 5.952 6.020 5.409 5.930 95th percentile, fathers 8.216 8.313 8.600 8.580 95th percentile, sons 8.209 8.218 8.557 7.874 Mean, fathers 3.681 3.705 3.518 3.912 Mean, sons 3.672 3.694 3.243 3.177 Father-son correlation† 0.349 0.357 0.348 0.349 Father-son with zero pubs. 0.215 0.215 0.175 0.228 Father-grandson correlation† 0.160 0.284 0.163 0.294 Notes: σe, κ, µh, σh not reported; † correlation on the intensive margin.

Model fit

◮ Reproduces distributional differences (Fact 2) ◮ Omitting nepotism:

◮ Fail to fit Fact 2. ◮ Bias β estimates upwards.

◮ Reproduces high father-son publication elasticity (Fact 1)

◮ Grandfather-grandson elasticity of 0.16 > 0.342 = 0.12.

33

Road map

• 1. Introduction
• 2. Methods
• 3. Data
• 4. Structural estimation
• 5. Extensions
• 6. Conclusion

34

Results over time

2 4 6 8 10 12 14 Sons' publications (in logs) 2 4 6 8 10 12 14 Fathers' publications (in logs) before 1527 1528-1625 1625-1725 1725-1800 35

Nepotism over time

β ν σe κ µh σh % nep N Before 1527 0.24 9.0 1.7 3.2

• 0.8

3.9 46.4 210 1528 to 1625 0.47 6.1 0.3 1.8 2.6 3.2 15.1 236 1626 to 1724 0.62 5.0 0.3 2.2 3.2 3.3 9.0 467 1725 to 1800 0.52 4.4 0.1 3.1 5.0 2.1 0.2 190

◮ Decrease in nepotism around the Scientific Revolution (1543–1632). ◮ Nepotism was negligible during the Enligthenment (1715–1789). 36

Human capital elasticity over time

β ν σe κ µh σh % nep N Before 1527 0.24 9.0 1.7 3.2

• 0.8

3.9 46.4 210 1528 to 1625 0.47 6.1 0.3 1.8 2.6 3.2 15.1 236 1626 to 1724 0.62 5.0 0.3 2.2 3.2 3.3 9.0 467 1725 to 1800 0.52 4.4 0.1 3.1 5.0 2.1 0.2 190

◮ Overtime, scholar lineages became more meritocratic. ◮ Does not support hypothesis that β is constant (Clark 2015). 37

Protestant Reformation

◮ Protestantism and Scientific Revolution (Merton 1938). ◮ Closure and censure in catholic institutions during

Counter-Reformation (Landes 1998).

50 100 150 200 250 300 350 Frequency 2 4 6 8 10 12 14 Publications (in logs) Catholic institutions (903 scholars) Protestant institutions (623 scholars)

38

Human capital elasticity over time

β ν σe κ µh σh % nep N Protestant 0.40 5.1 0.1 1.7 4.8 2.6 2.8 532 Catholic 0.74 5.8 0.8 2.2

• 0.9

4.0 31.0 357

◮ Protestant-catholic gap in scientific production could be associated

with nepotism.

39

Other heterogeneity

β ν σe κ µh σh % nep N

• A. Field of study (of fathers)

Lawyer 0.64 7.4 1.0 2.6

• 0.3

3.9 29.0 296 Physician 0.46 7.6 0.4 2.4 2.5 3.2 14.9 331 Theologian 0.53 3.8 0.3 1.2 4.5 2.6 3.2 138 Scientist 0.70 6.0 0.2 1.7 3.0 3.7 11.3 170

• B. Son’s nomination date

After father’s death 0.51 4.6 0.3 2.1 3.1 3.2 11.5 463 Before father’s death 0.67 6.7 0.4 1.9 2.3 3.7 15.1 453 Before age 30 0.63 6.8 0.4 2.4 4.3 2.8 4.3 307 After age 30 0.39 3.7 0.3 2.4 4.5 2.4 2.0 460

◮ Nepotism most prevalent in law faculties. ◮ Does not support hypothesis that β is constant (Clark 2015). ◮ Nepotism stronger if son nominated during father’s lifetime. ◮ Result not driven by age of nomination.

different university

40

Road map

• 1. Introduction
• 2. Methods
• 3. Data
• 4. Structural estimation
• 5. Extensions
• 6. Conclusion

41

Conclusions

◮ New method to disentangle nepotism vs. human capital.

◮ Intergenerational correlations and distributional differences. ◮ Omitting selection (nepotism) can bias persistence estimates.

◮ Data on families of scholars in Europe (1088–1800). ◮ Two factors explain intergenerational persistence:

◮ Nepotism: 15% scholars’ sons are nepotic scholars. ◮ High rate of transmission of human capital (β = 0.6).

◮ Importance for UTHC production:

◮ Removing nepotism would increase scientific output by 17.9% ◮ Nepotism remains high until Enlightenment. ◮ Nepotism more prevalent in Catholic institutions.

42

Standard model of intergenerational transmission

yi,t+1 = byi,t + (1 − b)¯ y + ui,t+1 , where:

◮ y: observed proxy for social status (e.g., wealth, income,

stature) in logs,

◮ b: intergenerational elasticity of y, ◮ ut: i.i.d. N(µu, σ2 u).

In the long run: limt→∞ Var[ht] = σ2

u

1 − b2 → Long-run inequality Var[ht] increases with b. → Elasticity over m generations is bm. → Full social mobility in few generations if b is low.

back

43

Half-life

yi,t+1 = b yi,t + ei,t+1 ,

◮ Number of generations until the gap with the mean halves:

t 1

2 = − ln(2)

ln(|b|) , → Depends negatively on b.

◮ Example, b = 0.5 (US earnings, Corak 2006) implies t 1

2 = 1:

Exam 1/2 the gap closed after 1 generation. Exam 3/4 the gap closed after 2 generations. Exam ...

back

44

Surname-average method

Clark (2015), Clark and Cummins (2015)

yi,t+1 = b yi,t + ei,t+1 , ¯ yk,t+1 = bA ¯ yk,t + ¯ ek,t+1 ,

◮ where k: rare surnames (or siblings). ◮ ¯

ek,t+1 = 0 if k independent of e.

◮ ˆ

bA = ¯ yk,t+1 ¯ yk,t = ¯ hk,t+1 ¯ hk,t = β.

◮ Data: repeated cross-section (with surnames). ◮ Criticism: group effects (Chetty et al. 2004, Guell et al. 2018).

back

45

Informational content of surnames (ICS)

Guell, Rodriguez-Mora, and Telmer (2015)

yi,k = γ′

LXi,k + b′L + ui,k ,

yi,k = γ′

FXi,k + b′F + ui,k ,

ICS = R2

L − R2 F . ◮ Population variance > Variance within surname. ◮ Structural estimation of β using ICS. ◮ Data: census cross-section (with surnames).

back

46

Model

◮ First-order Markov process of endowments transmission across

generations (Clark 2015, Braun & Stuhler 2018).

◮ Measurement error:

◮ Endowments (human capital) translated into observed

• utcomes (publications) with a noise.

◮ Selection:

◮ Population of potential scholars. ◮ Selection based on endowments. ◮ Nepotism: selection criterium different for sons of insiders.

47

Intergenerational transmission of human capital

◮ Generation t (fathers) human capital endowment:

ht ∼ N(µh, σ2

h) . ◮ Generation t + 1 (sons) inherit human capital endowment:

ht+1 = βht + ut+1 ,

where β: intergenerational elasticity of human capital. where u: random ability shock, i.i.d N(µu, σ2

u)

48

Selection and nepotism

◮ Population of potential scholars from families i ∈ I. ◮ Selection depends of human capital, hi > τ. ◮ Nepotism: different selection for scholars’ sons, hi,t+1 > τ −ν. ◮ The set P of scholar lineages (father and son) is:

P = {i | hi,t > τ, hi,t+1 > τ − ν} ⊂ I , where nepotism implies ν > 0.

49

Publications and measurement error

◮ Human capital used to produce scientific knowledge. ◮ Human capital (unobserved) translates imperfectly into

publications (observed): yi,t = max(κ, hi,t + ǫi,t) , yi,t+1 = max(κ, hi,t+1 + ǫi,t+1)

where yi: publications (in logs). where ǫi,t, ǫi,t+1 ∼ N(0, σ2

ǫ): shocks affecting pubs.

where κ: minimum yi to observe publications today.

50

Stationarity

◮ Assumption: stationary human capital distribution (among

potential scholars).

◮ Formally, hi,t and hi,t+1 drawn from same distribution. ◮ This implies:

µu = (1 − β)µh σ2

u

= (1 − β2)σ2

h ◮ Human capital:

hi,t+1 = βhi,t + (1 − β)µh + ωi,t+1 ,

where ωi,t+1 ∼ N(0, (1 − β)2σ2

h). back

51

Covered institutions: top ten

Institution Cntry Dates N Main Source University of Bologna ITA 1088 157 Mazzetti (1847) Royal Society of London UK 1660 71 www.royalsociety.org/ University of Avignon FRA 1303-1793 58 Laval (1889), Teule (1887), Fournier (1892), Barjavel (1841), Duhamel (1895) University of T¨ ubingen GER 1476 42 Conrad (1960) Leopoldina GER 1652 37 www.leopoldina.org/ University of Basel CHE 1460 34 Herzog (1780), Rep.Acad.Germ. University of Montpellier FRA 1289-1793 30 Dulieu (1975, 1979, 1983) University of Padova ITA 1222 27 Pesenti (1984), Facciolati (1757) University of Jena GER 1558 27 G¨ unther (1858) University of Pavia ITA 1361 27 Raggi (1879)

back

52

Chicoyneau dynasty

Michel Chicoyneau (1626-1701)

• Prof. Montpellier

1659-1701 Gaspard Chicoyneau (1673-1693)

• Prof. Montpellier

1691-1693 Fran¸ cois Chicoyneau (1672-1752)

• Prof. Montpellier

1693-1752 Acad´ emie des Sciences 1732-1752 Michel-Aim´ e Chicoyneau (1670-1691)

• Prof. Montpellier

1689-1691 Fran¸ cois Chicoyneau (1702-1740)

• Prof. Montpellier

1731-1740 Jean-Fran¸ cois Chicoyneau (1737-1758)

• Prof. Montpellier

1752-1758 Data source: Dulieu, 1983 back

53

M¨

• gling family

Daniel M¨

• gling (1546-1603)
• Prof. in Heidelberg & T¨

ubingen Medicine Johann Ludwig M¨

• gling (1585-1625)
• Prof. in T¨

ubingen Medicine Johann Ludwig M¨

• gling (1613-1693)
• Prof. in T¨

ubingen Medicine Johann David M¨

• gling (1650-1695)

Law

• Prof. in T¨

ubingen Johann Friedrich M¨

• gling (1690-1766)
• Prof. in T¨

ubingen & Giessen Law Jakob David M¨

• gling (1680-1729)
• Prof. in T¨

ubingen Law Jacob Friedrich M¨

• gling (1708-1742)
• Prof. in T¨

ubingen Law

back

Quantile-quantile plot

2 4 6 8 10 12 Sons' publications (in logs) 2 4 6 8 10 12 Fathers' publications (in logs)

Kolmogorov-Smirnov (Fathers vs. Sons) = 0.102∗∗∗ (p-value = 0.000)

back

55

Moments used in estimation

value s.e.

• bs.
• A. Intergenerational correlations in pubs.

Father-son, intensive margin 0.358 0.042 549 Father-son with zero pubs. 0.229 0.013 1,034 Father-grandson, intensive margin 0.295 0.129 54

• B. Publications’ distribution

Fathers with zero pubs. 0.303 0.015 906 Sons with zero pubs. 0.370 0.015 1,034 Fathers median 4.355 0.199 906 Sons median 3.172 0.265 1,034 Fathers 75th percentile 6.713 0.105 906 Sons 75th percentile 5.934 0.119 1,034 Fathers 95th percentile 8.500 0.076 906 Sons 95th percentile 7.845 0.074 1,034 Fathers mean 3.927 0.105 906 Sons mean 3.198 0.094 1,034 Note: Publications is 1 + log of library holdings.

back

56

Parent-child elasticities b

ˆ b yt country source 0.31-0.41 Wealth

• Agric. societies

Mulder et al. (2009) 0.48-0.59 Wealth UK Harbury & Hitchins (1979) 0.47 Earnings USA Corak (2012) 0.31-0.33 Earnings UK Corak (2012) 0.19-0.26 Earnings Sweden Jantti et al. (2006) 0.46 Education USA Hertz et al. 2007 0.35 Education Sweden Lindahl et al. 2015

back

57

Estimates of the “true” β

ˆ β yt Data Source 0.75 Wealth UK probate (1858-2012) Clark & Cummins (2015) 0.7–0.9 Oxbridge UK (1170-2012) Clark & Cummins (2015) 0.65 Occupation Germany, 3 gen. Braun & Stuhler (2016) 0.8 Education 2001 census, Spain Collado et al. 2018 0.74 Education EU-28, 3 gen. Colagrossi et al. (2019) 0.5–0.6 Education Germany, 4 gen. Braun & Stuhler (2016) 0.6 Education Sweden, 4 gen. Lindahl et al. (2015) 0.6 Education 2001 census, Spain Guell et al. (2014) 0.5 Earnings Sweden, 4 gen. Lindahl et al. (2015)

back

58

Model

◮ First-order Markov process of endowments transmission:

hi,t ∼ N(µh, σ2

h) ,

hi,t+1 = βhi,t + ui,t+1 ,

where ui,t+1 ∼ N(µu, σ2

u).

◮ Measurement error:

yi,t = max(κ, hi,t + ǫi,t) , yi,t+1 = max(κ, hi,t+1 + ǫi,t+1)

where ǫi,t, ǫi,t+1 ∼ N(0, σ2

ǫ).

◮ Selection and nepotism:

Potential families: i ∈ I. Observed families: P = {i | hi,t > τ, hi,t+1 > τ − ν} ⊂ I .

back

59

Identification (1/2)

Benchmark:

τ τ-ν

.1 .2

• 6 -5
• 1 0

10

fathers sons κ

.12 .25

• 1 0

5 10 15

fathers sons corr = .45

6 12

son

6 12

father

5 10 15

son

5 10 15

father

Measurement error:

τ τ-ν

.1 .2

• 6 -5
• 1 0

10

fathers sons κ

.12 .25

• 1 0

5 10 15

fathers sons corr = .25

6 12

son

6 12

father

5 10 15

son

5 10 15

father

Nepotism:

τ τ-ν

.1 .2

• 6 -5
• 1 0

10

fathers sons κ

.12 .25

• 1 0

5 10 15

fathers sons corr = .473

6 12

son

6 12

father

5 10 15

son

5 10 15

father

Identification (2/2)

Distributional differences in outcomes across generations:

◮ Nepotism: ↑ ◮ Measurement error: ∅

Variance of sons’ outcomes relative to their fathers’:

◮ Nepotism: ↑ ◮ Measurement error: ∅

Information of father-son correlation on human-capital transmission:

◮ Nepotism: ↑ ◮ Measurement error: ↓

back

61

Fathers and sons in different universities (moments)

different Moments baseline university Fathers with zero pubs. 0.305 0.157 Sons with zero pubs. 0.375 0.094 Median, fathers 4.302 5.577 Median, sons 3.055 6.407 75th percentile, fathers 6.721 7.041 75th percentile, sons 5.930 7.39 95th percentile, fathers 8.580 8.727 95th percentile, sons 7.874 8.495 Mean, fathers 3.912 4.88 Mean, sons 3.177 5.627 Father-son correlation† 0.349 0.327 Father-son with zero pubs. 0.228 0.058 Father-grandson correlation† 0.294

• 0.022

Fathers (N=255) and sons (N=277) in different universi-

• ties. 84.5% also in baseline sample.

62

Fathers and sons in different universities (estimates)

different Parameter baseline university

• Intergen. elasticity of human capital

β 0.581 0.715 Nepotism ν 5.632 0.005

• Std. dev. of shock to publications

σe 0.386 2.117 Threshold of observable publications κ 2.201 0.79 Mean of human capital distribution µh 2.444 3.942

• Std. dev. of human capital distribution

σh 3.451 2.971 % nepotism 15.04% 0.02%

back

63