Marc Goni is an associate professor at the Department of Economics ...
Great summary of the paper: https://www.axios.com/2022/07/02/what-h...
To see this quote come to life in a Jane Austen adaption, and see t...
Summary of the paper:
> "In this paper, I use a unique historica...
Overview of the first contribution.
> My first contribution is to...
Overview of second contribution:
> My second contribution is to sh...
For estimating causal treatment effects, this paper relies on rando...
Interesting that sorting patterns in a dating website do not differ...
> "Who attended the Season? Since it coincided with Parliament meet...
> "Who was considered a suitable match in the Season? Typically, ma...
In one season: 50 balls, 60 parties, 30 dinners and 25 breakfasts.
...
Queen Victoria was quee...
Instrumental variables (IVs) can be used to estimate causal relatio...
> Elsewhere it has been argued that elites pursue and maintain poli...
Wow
American Economic Journal: Applied Economics 2022, 14(3): 445–487
https://doi.org/10.1257/app.20180463
445
Assortative Matching at the Top of the Distribution:
Evidence from the World’s Most Exclusive Marriage
Market
†
By M G*
Using novel data on peerage marriages in Britain, I nd that low
search costs and marriage-market segregation can generate sorting.
Peers courted in the London Season, a matching technology introduc-
ing aristocratic bachelors to debutantes. When Queen Victoria went
into mourning for her husband, the Season was interrupted ( 1861–
1863), raising search costs and reducing market segregation. I exploit
exogenous variation in women’s probability to marry during the
interruption from their age in 1861. The interruption increased peer–
commoner intermarriage by 40 percent and reduced sorting along
landed wealth by 30 percent. Eventually, this reduced peers’ political
power and affected public policy in late nineteenth-century England.
(JEL C78, D83, J12, J16, N33)
It is a truth universally acknowledged, that a single man in possession of a
good fortune, must be in want of a wife.
—Jane Austen, Pride and Prejudice
I
n OECD countries, most people tend to marry those who have a similar education,
income, or social status (Chen, Förster, and Llena-Nozal 2013). Besides prefer-
ences for others like ourselves, one important determinant of marital sorting is the
matching technology: every relationship not only reects whom we choose but also
depends on whom we meet. For example, 60 percent of all married couples in the
United States met in settings where entry is restricted to similar others: at college, at
work, at a social club, etc. (Laumann etal. 1994). A robust prediction of marriage
* Department of Economics, University of Bergen (email: marc.goni@uib.no). Ilyana Kuziemko was coeditor
for this article. I thank three anonymous referees, Hans-Joachim Voth for his guidance, along with Ran Abramitzky,
Dan Bogart, Davide Cantoni, Greg Clark, Deborah Cohen, Mauricio Drelichman, Jan Eeckhout, Nancy Ellenberger,
Gabrielle Fack, Ray Fisman, Sebastian Galiani, Gino Gancia, Daniel García, Albrecht Glitz, Regina Grafe,
Alfonso Herranz, Boyan Jovanovic, Peter Koudijs, Lee Lockwood, Eeva Mauring, Joel Mokyr, Thomas Piketty,
Giacomo Ponzetto, Karl Schlag, Jaume Ventura, John Wallis, and seminar participants at Pompeu Fabra University,
Northwestern University, Paris School of Economics, University of Vienna, University of Leicester, HSE University,
Catholic University of Louvain, University of California Merced, University of Barcelona, University of Innsbruck,
Bank of Serbia, MOOD, EEA, EHS, Cliometrics, SAEe, RCI, EHA, and University of Mannheim for their com-
ments. The Cambridge Group for the History of Population and Social Structure kindly shared data. I acknowledge
the nancial support of the Heinrich Graf Hardegg Stiftung.
†
Go to https://doi.org/10.1257/app.20180463 to visit the article page for additional materials and author
disclosure statement(s) or to comment in the online discussion forum.
446 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
models is that search restricted to such settings strengthens marital sorting.
1
In turn,
assortative matching can have important implications for inequality and income
redistribution,
2
particularly at the top of the wealth distribution (Piketty andSaez
2006). Historically, marital sorting helped elites to consolidate their political power
(Puga andTreer 2014). Therefore, any matching technology that affects search
costs (e.g., online dating or students’ clubs at college) may well lead to greater sort-
ing and, hence, to greater political and economic inequality.
Testing this prediction is challenging. One reason is that, in modern marriage
markets, it is virtually impossible to isolate a particular matching technology from
other settings where courtship takes place. Recent empirical work has used speed
dating (Fisman etal. 2008) or dating websites (Hitsch, Hortaçsu, andAriely 2010).
The results are largely at odds with the theory—these matching technologies do
not seem to strengthen sorting, even when they reduce search costs and affect the
choice sets of potential mates. This discrepancy, however, may stem from the fact
that dating is very different from marriage. In most cases, dating does not reect
the long-term partnership formation at the core of search and matching theory. An
additional difculty is that because many changes to the matching technology are
recent, the long-term implications for political and economic inequality are yet to
be assessed.
In this paper, I use a unique historical setting to isolate the effect of the matching
technology on marital sorting and to evaluate some of its political-economy implica-
tions in the long run. In the nineteenth century, from Easter to August of each year,
a string of social events was held in London to help the peerage’s offspring
3
to meet
and court—the “London Season.” Courtship in noble circles was largely restricted
to London; in most cases, the only place where a young aristocrat could speak with
a girl was at a ball during the Season. Guests were carefully selected according to
social status, and the high cost involved in participating even excluded peers if they
were pressed for money. In economic terms, the Season reduced search costs for
partners and restricted the choice set of potential mates. Crucially, the Season was
interrupted by a major, unanticipated, exogenous shock: the deaths, in the same year,
of Queen Victoria’s mother and husband. As the queen went into mourning, royal
balls in the Season were canceled for three consecutive years (1861–1863). During
this period, preferences for spouses did not change, but nobles were exposed to
larger search costs and lower market segmentation. I use this large shock to identify
the effects of the Season on marital sorting and its long-term economic implications.
Specically, I combine archival and published sources to construct two novel data-
sets—one measuring attendance to the Season, another of all aristocratic marriages
in nineteenth-century Britain. This allows me to evaluate the effects of the Season
on marital sorting by title and landed wealth. To gauge the political-economy impli-
cations of marital sorting, I link my datasets to the biographies of peers elected to
the House of Commons and to local data on state-education provision.
1
Burdett and Coles (1997); Eeckhout (1999); Bloch and Ryder (2000); Shimer and Smith (2000); Adachi
(2003); Atakan (2006); Jacquet andTan (2007).
2
Kremer (1997); Cancian andReed (1998); Fernández andRogerson (2001); Fernández, Guner, and Knowles
(2005); Greenwood etal. (2014); Eika, Mogstad, andZafar (2019).
3
The peerage refers to the British aristocracy. More details are in SectionI.
VOL. 14 NO. 3 447
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
My rst contribution is to estimate a strong, plausibly causal link between search
frictions and marital sorting. I exploit the interruption of the Season ( 1861–1863)
as a quasi-experiment that raised search costs and reduced market segmentation
for some cohorts. Since the decision to marry during the interruption is potentially
endogenous, I compare cohorts who had different risks to marry in 1861–1863 based
on their age. Specically, I consider a baseline sample of peers’ daughters aged 15
to 35 in 1861. My treatment variable is a woman’s synthetic probability to marry in
1861–1863, which is based on her age in 1861–1863 and on the probability to marry
at each age in “normal times.” For example, the synthetic probability for a woman
aged 19–22 in 1861–1863 is equal to the percentage of women who married aged
19–22 in a benchmark cohort married before 1861.
4
My main estimates show that
women with a high synthetic probability to marry in 1861–1863 were more likely
to marry a commoner and less likely to marry a peer’s heir. To evaluate sorting by
landed wealth, I restrict the sample to matrimonies for which both spouses’ family
landholdings are recorded. Women with a high synthetic probability to marry in
1861–1863 sorted less by family landholdings and married husbands from poorer
families. To quantify the magnitude of these effects, I estimate an IV model where
I instrument a woman’s decision to marry during the interruption with her synthetic
probability to marry in 1861–1863. I nd that women who (exogenously) married
during the interruption were 40 percent more likely to marry a commoner, 30 per-
cent less likely to marry a peer’s heir, and married husbands 44 percentile ranks
poorer in terms of family landholdings. Finally, I present nonparametric estimates:
chi-squared tests of association reveal that higher-titled women married higher-titled
husbands only when the Season was operative—sorting by title resembles random
matching for cohorts exposed to the interruption. Similarly, Kolmogorov-Smirnov
tests show that the interruption reduced sorting by family landholdings. Altogether,
these results show that the matching technology embedded in the Season—by reduc-
ing search costs and segregating the marriage market—crucially determined sorting.
My second contribution is to show that marriage played an important role in con-
solidating the peerage as a political elite. To do so, I examine elections of Members
of Parliament (MP) at the House of Commons for 27 general elections and 97
by-elections in the late nineteenth century. I show that a woman’s marriage to a
commoner reduced her blood relatives’ probability to be elected MP in the following
years. Specically, I estimate an IV model where I instrument a woman’s probabil-
ity to marry a commoner with her synthetic probability to marry during the Season’s
interruption. I nd that, after a woman’s marriage to a commoner, her brothers were
50 percent less likely to be elected MP, and, together, they served 18 fewer years
than the brothers of women who married in the peerage. The loss of political power
was local: mostly constituencies near the family seat were affected. Not only broth-
ers but also the family heads a decade after the interruption (in the 1870s) were
affected. I also discuss historical evidence on the mechanisms behind these effects.
After a woman’s marriage to a commoner, her birth family had to mobilize con-
siderable capital to sustain her. This limited their ability to control MP elections
4
I use women born in 1815–1830 as benchmark cohort and show that results are robust to using alternative
benchmark cohorts.
448 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
by distributing favors, rents, and jobs among the local electorate. Marrying a com-
moner also reduced a family’s social prestige (Allen 2009) and a woman’s ability
to act as “power broker” on behalf of her blood relatives (Atkins 1990). Altogether,
the evidence strongly suggests a negative relationship between within-landed-elite
marriages and how contested MP elections were in late nineteenth-century England.
In contrast, the Season’s interruption increased a woman’s probability to marry a
commoner, which, in turn, reduced her family’s political power. Finally, I show that
this had important economic consequences: families who lost political power could
not effectively oppose the introduction of state education in the 1870s—a policy
otherwise subject to capture by local landowners (Stephens 1998). I use data from
Goñi (2021b) on wealth taxes set by 943 local school boards in 1872–1878. IV esti-
mates show that taxes were higher near the family seats in which a woman married
a commoner (and the family lost political power) than near the family seats in which
a woman married in the peerage (and the family retained political power).
The empirical setting I examine offers a number of advantages. First, Victoria’s
mourning generated as good as random assignment into the Season. Prince Albert’s
death and Victoria’s long mourning were unexpected (Ellis 1977; Hobhouse 1983).
Furthermore, there was high pressure to marry young: if a girl was not engaged two
or three Seasons after “coming out” into society, she was written off as a failure
(Davidoff 1973, 52). Hence, women at risk of marriage in 1861 could not sim-
ply postpone the search for a husband until the uncertain date when the Season
would resume. Second, the short nature of the interruption allows me to disentangle
search costs from preferences, as the latter unlikely changed in the short run. Third,
no centralized market emerged during the interruption. Neither could the Season
be replaced by arranging marriages, as these were not socially acceptable at the
time (Davidoff 1973, 49). Meeting was required, and the evidence suggests that
in 1861–1863, meetings took place in local marriage markets, where search costs
were large. Fourth, this setting allows me to open the “black box” of the matching
technology. In modern marriage markets, we can only guess who is on the market
and who meets whom. By contrast, the participants in the Season were well dened,
e.g., it was announced who was newly on the marriage market each year. Fifth, there
is a wealth of data to evaluate many dimensions of marital sorting, peers’ politi-
cal power, and local public goods’ provision. Sixth, in contrast to studies that use
modern-day data, this historical case study allows me to identify long-term implica-
tions of marital sorting.
This paper can be seen as integrating two literatures. Starting with the seminal
works of Gale and Shapley (1962) and Becker (1973), a large literature has studied
the determinants of assortative matching. A classic prediction is that a matching
technology that reduces search costs and limits people’s choice set increases sort-
ing.
5
Surprisingly, these well-accepted theoretical insights lack clear-cut empirical
support. Hitsch, Hortaçsu, andAriely (2010, 162) show that sorting patterns in a
dating website “absent of search frictions” differ little from those of people search-
ing off-line. Similarly, Lee (2016) suggests that lower search costs online and search
5
See references listed in footnote1.
VOL. 14 NO. 3 449
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
lters do not affect sorting. Fisman et al. (2008) nd few interracial matches in
speed dating, even though this matching technology facilitates encounters between
people of different ethnic groups. Contrary, Belot andFrancesconi (2013) nd that
meeting opportunities dominate preferences in speed dating. These mixed results
may be explained because dating does not always reect the long-term partnership
formation at the core of search theory. Several studies suggest that meeting oppor-
tunities matter for marriage. For example, Abramitzky, Delavande, and Vasconcelos
(2011) shows that male scarcity in post-WWI France reduced marital sorting.
6
Here,
I analyze a matching technology that, like online or speed dating, reduces search
costs but where outcomes are marriages. Another paper that looks at marriages in
a low–search cost environment is Banerjee etal. (2013). Their aim, however, is to
estimate the strength of same-caste preferences in India. To the extent of my knowl-
edge, my paper is the rst to provide causal evidence that a matching technology
that reduces search costs and restricts the choice set of mates generates greater sort-
ing in marriages. Hence, my rst contribution is to reconcile the empirical evidence
on matching technologies with the predictions of search theory.
7
The second literature that motivates this paper studies the persistence of elites
and institutional capture. Several studies show that elites persist by opposing inclu-
sive institutions (Sokoloff andEngerman 2000; Acemoglu 2008; Galor, Moav, and
Vollrath 2009; Allen 2009). Others have argued that household decisions such as
inheritance (Bertocchi 2006) or marriage (Puga andTreer 2014; Cruz, Labonne,
and Querubín, 2017; Marcassa, Pouyet, andTrégouët 2020) were crucial to con-
solidate elites. My contribution is to link these two strands of the literature. I show
empirically that marital sorting reinforced peers’ political power, which led to the
distortion of inclusive institutions such as state education. This suggests that mar-
riage was crucial for institutional capture in late nineteenth-century England. The
case study of the peerage is interesting for understanding inequality and elite per-
sistence.
8
The peerage was likely the most exclusive elite ever to exist. Compared to
the continental aristocracy, peers were fewer, richer, and remained in power longer
(Cannadine 1990). I present an institution—the London Season—seldom consid-
ered by economists
9
and show that it helped sustain the peerage as an unusually
small, exclusive, and rich elite.
The article proceeds as follows. Sections I and II present the historical back-
ground and the data. SectionIII establishes a plausibly causal link between search
frictions and marital sorting. SectionIV investigates the implications of sorting for
the peerage’s political power and its impact on public policy. SectionV concludes.
6
Laumann etal. (1994); Nielsen and Svarer (2009); Kaufmann, Messner, andSolis (2013) also emphasize
meeting opportunities. Differently, Bruze (2011) nds that actors sort by education even if they do not meet their
partners at school.
7
My paper also relates to the matching market design literature (Roth andSotomayor 1990).
8
For studies linking marital sorting and inequality in modern settings, see footnote2.
9
Doepke andZilibotti (2008) use the Season to illustrate the upper classes’ taste for leisure. More generally,
Allen (2009) argues that the extravagant aristocratic lifestyle was a sunk investment that allowed the peerage to
effectively rule England from circa 1550 to 1880.
450 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
I. Historical Background
This sectiondescribes the Season and who was considered a suitable match as
well as some contractual consequences of marriages in nineteenth-century Britain.
A. The London Season
The Season arose in the seventeenth century, when peers (and their families)
started to move to London every year from February to August to attend Parliament
(Davidoff 1973). In the nineteenth century, the Season developed into “the largest
marriage market in the world,” providing a string of balls and social events where
the right sort of people met (Aiello 2010).
10
The reason was that
arranged marriages were no longer acceptable so that individual choice
must be carefully regulated to ensure exclusion of undesirable partners.
Under such a system it was vital that only potentially suitable people
should mix. To meet these ends, balls and dances became the particular
place for a girl to be introduced into Society. (Davidoff 1973, 49)
Figure1 illustrates the Season’s calendar and main events. It plots 4,000 move-
ments into and out of London by Season participants, as reported in the Morning
Post in 1841. At the beginning of the year, most were out of town. Later in January,
Parliament convened, and members of the aristocracy from all over the country
moved from their family seats to London.
11
On April 20, the Queen joined in, and
the rst debutante was presented at court. Court presentations were yearly public
announcements of who was newly on the marriage market. Afterward, many events
designed to introduce bachelors to debutantes took place. It was the most crowded
time of the year in London: on May 15—the day of the royal ball at Buckingham—
over 800 families were in London for the Season. Many ladies met their future
husbands at royal parties, which were described as “mating” rituals (Inwood 1998).
The Season was over by August 12, when peers moved back to their country seats
for the shooting season. Lucy, daughter of the fourth Baron Lyttelton, followed this
calendar in 1859. Her diary describes the trip to London from her Worcestershire
family seat (May 18), the “very memorable day” of her court presentation (June 11),
and the royal ball at Buckingham (June 29).
Who attended the Season? Since it coincided with Parliament meetings, the
Season was attended by the families of members of the two Houses of Parliament.
Specically, by all hereditary peers who sat in the House of Lords (Lords), and
by those elected to the House of Commons (MPs). Peerage families without
political connections also participated in the Season (Sheppard 1977, 89–93) as
well as members of the landed gentry.
12
By centralizing the marriage market in
10
Similar arrangements in continental Europe, e.g., the Rallye in Paris, never eclipsed the Season. Continental
nobilities were too large for such meeting to be possible and did not engage in annual migrations to the capital, as
their parliaments did not meet as regularly as in Britain.
11
Sheppard (1977, 90) notes the “peripatetic existence of the ‘Fashionable World’,” suggesting that coming
from a remote family seat was not a major impediment to attend the Season.
12
The gentry was a social class consisting on landowners. Socially, they were below the peerage.
VOL. 14 NO. 3 451
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
London, the Season brought together aristocrats whose family seats were dispersed
around the country. In addition, guests to private balls in the Season were carefully
selected based on status (Davidoff 1973). Masked balls ceased because they were
gate-crashed by commoners (Ellenberger 1990, 636). Similarly, court presentations
were restricted to women sponsored by someone who had been presented before.
In 1841–1850, for example, 88 percent of women presented were from the peerage
or the gentry (Ellenberger 1990, table1). The expenses incurred in the Season also
restricted access. Few could afford to rent a house in Grosvenor Square or organize
a ball for hundreds of guests (Sheppard 1971).
13
The decline of the Season is linked to that of the peerage. Land values dropped
in the late 1870s,
14
and peers’ estates could no longer support their opulent lifestyle
(Cannadine 1990). After that, many events in the Season became public, and com-
moners were presented at court, including American nouveau riche like Consuelo
Vanderbilt (Ellenberger 1990). It was the death of the Season.
13
For example, the Duke of Northumberland spent £20,000 in the 1840 Season (Sheppard 1971), the equivalent
to $2,000,000 today (Nye 2019).
14
This was due to the nationwide fall in grain prices after the opening up of the American prairies to cultivation
and the development of steamships (Cannadine 1990).
F1. T S 1841
Notes: This gure plots 4,000 movements into and out of London of families participating in the Season of 1841, as
reported in the Morning Post. The solid line shows arrivals minus departures. Since departures are underreported,
Sheppard multiplies them by 1.6.
Source: Sheppard (1977)
Opening of Parliament
Queen arrives to London
Royal ball
−200
0
200
400
600
800
Number of families
January
February
March
April
May
June
July
August
September
October
November
December
Arrivals Departures Cumulative
End of the season
452 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
B. Choice of a Husband
Who was considered a suitable match in the Season? Typically, marriages were
not love matches but were based on eligibility. According to Davidoff (1973, 50),
“[marriage was] not so much an alliance between the sexes as an important social
denition; serious for a man but imperative for a girl. It was part of her duty to
enlarge her sphere of inuence through marriage.”
Women and men aimed to preserve their social status through marriage. That is,
peers’ daughters aspired to marry peers’ sons and vice versa. The peerage is divided
into ve titles: the most desired partners were dukes, marquesses, and earls, followed
by viscounts and barons. Next came the gentry (baronets and knights), who were
considered commoners. Titleless commoners were the least desirable. Marrying a
partner of a very different status was frowned upon (Perkin 2002, 61). Yet marrying
in the peerage was not trivial. In 1900, only 0.03 percent of people in Britain were an
aristocrat; in Europe, it was 1 percent (Beckett 1986, 35–40). The Season facilitated
encounters within this small group, leading to a high endogamy rate: in 1851–1875,
50 percent of peers’ daughters married within this 0.03 percent group.
Family landholdings also played an important role in marriage decisions. Stone
(1979, 87) argues that the aristocracy feared an “alliance with a family of lower
estate or degree than one’s own.” The strict settlement—a contract regulating peers’
inheritances—encouraged women to marry into families with large landholdings.
In detail, settlements established a widowhood pension (jointure), a yearly payment
(allowance), and capital sums (portions) for the heir’s wife and children (Habakkuk
1994, 2). These payments were raised from the family landholdings and hence, were
proportional to its size. That is, marrying an heir to large landholdings meant marry-
ing a husband with a large income as well as a large jointure, allowance, and portion
for the wife (and for her children). Marrying his brother was also desirable: even
though he was not the heir, the strict settlement established allowances and por-
tions for all the heir’s siblings.
15
For example, portions of £10,000 to £30,000 were
common for the heirs’ brothers and sisters, who typically received similar allow-
ances (Thompson 1963, 70; Perkin 2002, 67). In contrast, the income of an heir
(or the allowance of a non-heir) to small landholdings may not have met the wife’s
needs. To compensate, her birth family typically provided her considerable portions,
allowances, pin money, etc., diverting resources away from other purposes (see
SectionIVA). This compensation was agreed upon the marriage (Habakkuk 1994).
The rst Earl of Lytton gave a sense of the magnitude of the diverted resources: in
1864, he reckoned that a yearly income of £1,500 was the absolute minimum for
a married couple (Perkin 2002, 68). This is the equivalent of $170,600 today (Nye
2019).
Although some marriages involved dowries, a systematic record does not exist.
Hence, my analysis focuses on marriage outcomes rather than on the corresponding
prices. The tested prediction—that search costs reduce sorting—is robust to allow-
ing transfers between spouses (Shimer andSmith 2000; Atakan 2006). Omitting
15
Similarly, men aimed to marry women entitled to a larger allowance and portion.
VOL. 14 NO. 3 453
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
prices from the analysis is also justied because title and land reect social prestige,
which was not transferable through dowries or bride-prices (Davidoff 1973).
An important feature of the courting process was the pressure to marry young.
Women had two to three Seasons to become engaged. If they failed, they were con-
sidered “on the shelf” (Davidoff 1973, 52). Figure 2 conrms that social norms
circumscribed courting to young ages. In 1851–1875, a woman’s “market value”—
measured as her probability to marry an heir—declined after age 22. Importantly,
women of higher status could not delay their marriage longer: around age 22, the
market value of dukes’, marquesses’, and earls’ daughters equalized to that of the
lower-ranked barons’ and viscounts’ daughters. Consequently, courtship was an
intense process. Although it is unknown how many proposals women received before
accepting one, anecdotal evidence suggests that courting involved many interac-
tions. For example, Lady Nevill attended “50 balls, 60 parties, 30 dinners and 25
breakfasts” in her rst Season (Nevill 1920). In each ball, she was supposed to meet
various suitors, as decorum rules discouraged dancing more than three times with
the same suitor (Davidoff 1973, 49). Once a proposal was accepted, engagements
lasted around six months. Marriage manuals explicitly discouraged long engage-
ments. In general, marriages took place at the end of the Season.
II. Data
This section describes the datasets that I constructed for this paper. Online
Appendix A provides detailed summary statistics.
A. Attendance to the Season
The British National Archives keep the original invitations issued to parties at
Buckingham and St. James’s Palace during the Season.
16
The records cover the
period 1851–1875 and consist of circa 5,000 invitations per year. Based on these
archival documents, I created a computerized dataset with all the invitations issued
in 1851–1875 and the number of attendees at each party. Figure3 plots the number
of attendees over time. The chart reveals a huge disruption to the Season between
1861 and 1863 as a result of Queen Victoria’s mourning for the deaths of her mother
and her husband. Moreover, in 1851 attendance rates were unusually high as a result
of the Great Exhibition.
17
Royal parties were central to the Season (Davidoff 1973, 25). However, many
events took place in private houses. How well does this data reect general trends
in the marriage market? Anecdotal evidence suggests that fewer balls were orga-
nized in private houses during Queen Victoria’s mourning. While in normal years
almost all Grosvenor Square residents hosted and attended private balls (Pullar
1978), in 1862 over a third of the residents did not participate in any private ball
16
The National Archives (LC): LC 6/31-55, LC 6/127-156, and LC 6/157-164.
17
The Great Exhibition was the rst in a series of nineteenth-century World’s Fairs.
454 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
(Morning Post,cited in Wilkins 2011, 10–11).
18
In addition, Figure3 includes an
alternative measure of the Season: the debutantes presented at court, i.e., the num-
bers announced to the marriage market (Ellenberger 1990). Trends are remarkably
similar: when royal parties were well attended, more debutantes were presented. In
1862, both royal parties and court presentations were canceled.
B. Peerage Marriages
I assemble a novel dataset of peerage marriages that contains measures of sort-
ing by title and landed wealth, and the geographical origin of peerage families.
To do so, I use three data sources, two of which I have newly computerized. In
detail, I complement Hollingsworth’s Genealogical Data on the Peerage (2001)
with geo-referenced data on landholdings (Bateman 1883) and family seats (Burke
1826).
18
For example, in 1862 the Morning Post reported only two private balls for the week of July 14–20. In 1858,
seven balls took place the corresponding week (July 11–17). One possibility is that hosting a ball while the Queen
mourned would be seen as disrespectful.
F2. P M Y
Note: The sample is all 796 peers’ daughters rst marrying in 1851–1875.
0
10
20
30
Percent marrying duke, marquess, earl heir
18–20
21–23
24–26
27–29
≥30
Age at marriage
All peers’ daughters
0
10
20
30
40
18–20
21–23
24–26
27–29
≥30
Age at marriage
Dukes’, marquesses’, earls’ daughters
Barons’, viscounts’ daughters
VOL. 14 NO. 3 455
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
Hollingsworth Dataset.—This genealogical dataset covers the marriages of all
peers who died in 1603–1938 and of their offspring.
19
Hollingsworth (1964) con-
structed this dataset from peerage records, the chronicles of the family histories of
the British aristocracy.
20
The data comprise around 26,000 individuals. My baseline
sample is 644 women aged 15 to 35 in 1861 who ever married.
21
For each spouse,
I know the title and the title of the highest-ranked parent. Titles are grouped in
ve categories: (1) duke, marquess, or earl; (2) baron or viscount; (3) baronet; (4)
knight; and (5) commoner. I use this information to measure the rate of peer–com-
moner intermarriage, whether women married an heir, and sorting by title. The data-
set also lists if a title is an English, Scottish, or Irish peerage; birth order (Gobbi and
Goñi 2021); the number of children; and spouses’ date of birth, marriage, and death.
Landholdings.—I computerized new data of peers’ family landholdings based
on Bateman (1883). The book lists the great landowners in Britain and Ireland by
1876. It includes all owners of at least 3,000 acres and 1,300 owners of 2,000 acres.
I digitized the total acreage owned by families who appear both in Bateman’s book
and the Hollingsworth dataset.
22
Henceforth, a woman’s family landholdings refers
to the acreage owned by her birth family (idem for men). I focus on acreage because
19
Note that the Hollingsworth (2001) dataset excludes the gentry. Hence, I focus on the peerage—the layer for
which marriage had the highest stakes.
20
The data were redigitized by the Cambridge Group for the History of Population and Social Structure in 2001.
I am grateful to them for sharing the dataset.
21
I exclude second marriages, women married to foreigners, and members of the royal family. When I evaluate
celibacy rates, the sample is 765 because it includes women who never married.
22
I constructed the dataset in three steps. First, I digitized all 596 men who appear both in Bateman and in
Hollingsworth’s dataset. Second, I coded 353 of their wives’ birth families. This number is lower because some
coded men did not marry or married landless commoners. From these two steps, I found both spouses’ family
F3. A R P, T E
Note: The data comprise circa5,000 yearly invitations to royal parties during the Season (1851–1875).
0
200
400
600
800
Debutantes
0
2,000
4,000
6,000
8,000
Attendees
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
Ball
Children’s ball
Court reception
Concert and evening party
Breakfast and afternoon party
Debutantes presented at court (Ellenberger 1990)
456 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
the information reported is more reliable than that on land rents (Bateman 1883). To
measure sorting by landholdings, I consider the difference between spouses’ family
landholdings. Out of a potential sample of 644 women, I found the family landhold-
ings for 324 women and their husbands. Eighty-one percent of the lost observations
correspond to women marrying landless commoners. The remaining 19 percent are
women whose families (or whose husbands’ families) owned fewer than 2,000 acres
and hence, were not listed in Bateman (1883).
Family Seats.—The data on family seats are from Burke (1826). This Heraldic
Dictionary lists the seats of peerage families 34 years before the interruption of the
Season. That is, the seats where women in my baseline sample lived before mar-
rying. I geo-referenced 694 seats for 498 peerage families (some families owned
more than one seat). These seats spread over Britain and Ireland (online Appendix
FigureA6). I then link every individual in the Hollingsworth dataset to the seats
owned by her/his birth family (henceforth, family seats). This gives me the family
seat of 484 out of 644 women in the baseline sample (75 percent) and 260 out of 324
women with both spouses’ family landholdings (80 percent).
23
I use the location
of family seats to (i) assess a family’s local political power, (ii) evaluate education
provision around their seats, and (iii) control for distance to London. For these exer-
cises, the sample is restricted to women with a recorded family seat.
C. Political Power and Education Provision
To investigate the implications of the Season for political power and education
provision, I use two additional sources. Here I describe them briey; more details
are provided in the respective sectionsand in online appendixesA4 andA5.
First, I construct a new dataset on elections of Members of Parliament for the
House of Commons. To do so, I use thepeerage.com, a website that provides biog-
raphies for all members of the peerage. The biographies state whether an individual
was elected MP, the constituency, and the terms served. I hand-collected 674 biogra-
phies of the fathers and the brothers of women in my baseline sample. Of a potential
sample of 279 women with a seat in England,
24
I found the biographies of all their
fathers and brothers (except for 9 women who did not have any brothers). I also
collect the biographies of those who where family heads in the 1870s, when state
education was introduced in England. I use regular expressions to identify whether
and when an individual was elected MP, his constituency, whether he was elected in
the family seat’s county, and how many years he served.
25
My dataset spans 27 gen-
eral elections and 97 by-elections between 1776 and 1910 and covers 205 different
constituencies (see FigureA7 in the online Appendix for details).
26
landholdings for 227 couples in my baseline sample. Third, I searched the remaining spouses in the baseline sample,
nding 97 additional couples.
23
For the sample used to evaluate celibacy rates, I match 565 out of 765 women (74 percent).
24
The sample is restricted to England because only there I have education provision data.
25
thepeerage.com also lists other political posts, but, unlike for MPs, the date of appointment is not provided
systematically. This information is crucial to examine whether a family’s political power was reduced after a wom-
an’s marriage to a commoner (see SectionIVA).
26
In this sample, 20 percent of peers in the House of Lords had sons seating in the House of Commons.
VOL. 14 NO. 3 457
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
The second source I use is the Reports of the Committee of Council on Education,
digitized by Goñi (2021b). The reports list wealth taxes set by each school board
in England (1872–1878). I use data on all 943 school boards in a 10-mile radius of
387 family seats in England, i.e., the family seats of women in the baseline sample.
Since some families owned several seats, the number of seats is larger than the num-
ber of women. I measure local education provision with the average tax rate in this
ten-mile radius. On average, wealth was taxed at 2.3 percent.
III. Empirical Analysis
In this section, I establish a causal link between search costs, market segmenta-
tion, and sorting. To do so, I exploit the interruption of the Season in 1861–1863 as
a quasi-experiment that raised search costs and reduced market segmentation.
A. Identication: The Interruption of the Season
On March 16, 1861, Queen Victoria’s mother died. Victoria was grief-stricken,
and her husband, Prince Albert, took over most of her duties (Hobhouse 1983).
On December 14, Albert died too. As a result of these two losses, Victoria avoided
public appearances as much as she could. From 1861 to 1863, most royal parties
during the Season were canceled (Figure3). In 1862, the Queen suspended all court
presentations, i.e., did not announce who was newly on the marriage market.
My identication strategy exploits the fact that the interruption of the Season
(1861–1863) raised search costs and reduced marriage market segmentation.
Concomitantly, nobles’ preferences for spouses were stable. Since the decision to
marry in 1861–1863 might be endogenous, I identify the effect of search costs and
market segmentation by comparing female cohorts who had different risks of mar-
rying during the interruption based on their age. Next, I dene the sample and treat-
ment and provide evidence supporting the identifying assumptions.
27
Sample and Treatment.—My baseline sample is all peers’ daughters aged 15 to
35 in 1861 who ever married. This includes women who could potentially have
married during the interruption, although with different risks, determined by their
age. My treatment variable is a woman’s synthetic probability to marry during the
interruption of the Season (1861–1863). This is based on her age in 1861–1863 and
on the percentage of women marrying at each age in “normal times” (i.e., before
the interruption). For example, consider a woman aged 20 at the start of the inter-
ruption, 21 in 1862, and 22 in 1863. Her synthetic probability to marry during the
interruption will be the probability of marrying at ages 20, 21, and 22 in normal
times. Formally, the synthetic probability ( T
t
) is
(1) T
t
= p
(
t
)
+ p
(
t + 1
)
+ p
(
t + 2
)
,
27
Admittedly, I estimate the effect of a transitory disruption of the Season; the estimated effects would be dif-
ferent if the participants thought that the Season would never resume again.
458 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
where t , t + 1 , and t + 2 index a woman’s age in 1861, 1862, and 1863, respectively,
and p
(
t
)
is the percentage of women married at age t in normal times. All p
(
t
)
are
computed from a benchmark cohort that was marriageable before the interruption
but is close to the baseline sample such that they are comparable. Specically, I use
peers’ daughters born in 1815–1830,
28
excluding those who married after 1861 or
died before age 30.
29
In SectionIIIE, I show that the results are robust to using alter-
native benchmark cohorts. Finally, note that T
t
captures the probability at birth—or
at the start of the courting process—to marry during the interruption. Hence, it is
independent of an individual’s (endogenous) marriage decisions.
30
Because social norms circumscribed courting to young ages, small age gaps led
to large differences in treatment. Figure4 illustrates this. The cohorts most exposed
to the interruption were women aged 19 to 22 in 1861. Their synthetic probability
is above 20 percent. In other words, based on marriage behavior in normal times,
one in ve women in these cohorts was expected to marry during the interruption.
Women aged 18 or 23 in 1861 have lower but considerable synthetic probabilities.
The synthetic probability rapidly declines after age 24. Nine out of 10 women aged
25 or above were not expected to marry during the interruption, likely because they
28
I choose this 15-year cohort as, on average, they married 15 years before the base sample.
29
I exclude those who died before 30 such that p
(
t
)
reects the marriage probabilities of those who participated
in the marriage market.
30
In my setting, T is preferable to the hazard rate, which measures marriage probabilities in 1861–1863 condi-
tional on remaining single—a potentially endogenous decision (online AppendixB5).
F4. S P M S’ I (1861–1863)
Notes: For a woman aged t ∈
{
15, 16, …, 35
}
in 1861, the synthetic probability to marry during the three-year
interruption of the Season is T
t
= p
(
t
)
+ p
(
t + 1
)
+ p
(
t + 3
)
, where p
(
t
)
is the probability to marry at age t in
“normal times.” All p
(
t
)
are computed from a benchmark cohort of daughters born in 1815–1830, excluding those
marrying before 1861 and dying before age30.
0
5
10
15
20
25
Synthetic probability to marry in 1861–1863 (percent)
15 20 25 30 35
Age in 1861
VOL. 14 NO. 3 459
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
were already married. Similarly, very young cohorts (aged 17 or less in 1861) were
expected to marry after the Season resumed.
Identifying Assumptions.—The identifying assumptions are that cohorts with
different treatment levels are otherwise identical, that social norms circumscribed
courting to young ages, that no centralized market replaced the Season, and that
Victoria’s mourning was the only shock to the marriage market in 1861–1863.
The rst assumption would be violated if women could select their treatment
level, i.e., their synthetic probability to marry in 1861–1863. Note, however, that this
variable is independent of a woman’s endogenous marriage decisions. It depends
only on her age, and hence, selection is not an issue—nobody can choose her age.
31
As for the second assumption, social norms circumscribed courting to young
ages (SectionI). This did not change during the interruption. First, the synthetic
probability to marry in 1861–1863 (which is based on previous cohorts) predicts
well the actual probability to marry in 1861–1863 (panel B of Table3). Second,
women did not postpone the search for a husband: the average age at marriage was
the same during the three-year interruption, three years before, and three years after
(Table1).
32
Third, if some women had postponed marriage (or had anticipated the
interruption), the pool of married women would be altered. Table1 shows that this
was not the case. The share of dukes’, marquesses’, and earls’ daughters marrying
during the interruption is identical to that in the three years before or the three years
after. Women came from a family in a similar percentile of the landholdings’ distri-
bution. The share of families in the English peerage, life expectancy, and birth order
also do not vary substantially. Fourth, it is unlikely that women (or men) anticipated
marriages or waited for the Season to resume because the timing and duration of the
interruption were unpredictable. Nobody expected Prince Albert to die in 1861: he
was only 42 and took on government duties until 1 month before his death.
33
Even
doctors failed to diagnose him with cancer, the cause of his death (Hobhouse 1983,
150). The peerage was also surprised by the length of Victoria’s mourning. They
complained that “the Queen came less and less to London, and the palace was more
and more deserted” (Ellis 1977, 361).
The third identifying assumption is that no centralized market replaced the Season
in 1861–1863. There is no mention of such a market in historical records. As dis-
cussed in SectionIIA, private balls did not replace the canceled court presentations
and royal balls. The evidence also suggests that, although Parliament met during the
Season’s interruption, the families of Lords and MPs did not move to London with
them. That is, they did not engage in an alternative, reduced version of the London
Season. For example, in 1863, the family of the sixth Earl Bathurst did not move
to London with him, even though he was an MP and his sister, aged 26, was single
(Wilkins 2011, 10). More generally, the 1841 and 1861 censuses were conducted in
31
Given that the peerage was small (Beckett 1986), that the Season centralized information, and that peerage
records published birth dates, it is unlikely that women lied about their age.
32
Similarly, online AppendixB6 shows that women who, based on their age, were at risk of marriage during the
interruption did not marry at an older age than women not exposed to the interruption.
33
For example, on November 8, 1861, Union forces intercepted the British RMS Trent. Albert intervened to
soften the diplomatic response, lowering the threat of war (Hobhouse 1983, 154).
460 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
months in which Parliament met regularly and report whether residences in London
were occupied by a family or by one individual. By focusing on the West End, where
many MPs and Lords rented a house while Parliament met, one can assess if their
families moved with them to London. In 1841, most did: 87 percent of West End’s
houses were occupied by a family. In contrast, in 1861, only a third were occupied
by a family (Wilkins 2011, 7).
In addition, the Season could not be simply replaced by arranged marriages, as
these were not acceptable by the nineteenth century (Davidoff 1973, 49). Meeting
was required, and the evidence suggests that meetings took place in local marriage
markets: Figure5 shows that, before and after the interruption, spouses came from
family seats separated by 160 miles. In contrast, in 1862 couples came from 60
miles closer. Compared to the Season, local markets were shallow and search costs
large—meeting a sizable pool of suitors would involve visits to several seats.
The nal assumption is that the Season’s interruption is the only shock to the mar-
riage market. Although Albert’s death coincides with the outbreak of the American
Civil War, the marriage market was not affected. The American economy and the
British textile sector suffered, but peers were not big investors in either (Rubinstein
1977, 115; Ventura andVoth 2015, 11). Furthermore, the textile crisis would make
industrialists (commoners) less attractive suitors—which goes against my results. In
addition, the peerage was not subject to any demographic shock in 1861–1863: the
size of the cohort was not unusual (Table1).
B. Main Estimates
To estimate the effect of the Season’s interruption (1861–1863) on marital sort-
ing, I use the following econometric specication:
(2) Pr
(
y
i,t
= 1 | 𝐗
i,t
)
= Φ
(
α + β T
t
+ 𝐗
i,t
′ δ
)
,
where i indexes women; t ∈
{
15, 16, …, 35
}
their age in 1861; and the treatment,
T
t
, is the synthetic probability to marry during the interruption (1861–1863). As
T 1—M M , , S’ I
Interrup. Before Diff. After Diff.
1861–1863 1858–1860
(1)−(2)
1864–1866
(1)−(4)
(1) (2) (3) (4) (5)
Panel A. Wife’s characteristics
Age at rst marriage
24.7 (0.6) 24.8 (0.7) −0.1 (0.9) 24.5 (0.7) 0.2 (0.9)
Life expectancy
69.0 (1.9) 64.5 (2.1) 4.5 (2.8) 69.5 (1.9) −0.5 (2.7)
Birth order (excluding heirs) 3.7 (0.3) 3.9 (0.3) −0.2 (0.4) 4.0 (0.3) −0.3 (0.4)
Duke/Earl/Marquess daughter 0.5 (0.1) 0.5 (0.0) 0.0 (0.1) 0.5 (0.1) 0.0 (0.1)
Peerage of England
0.6 (0.0) 0.5 (0.0) 0.1 (0.1) 0.6 (0.1) 0.0 (0.1)
Family acres (percentile) 47.4 (3.2) 52.6 (3.1) −5.2 (4.5) 50.6 (3.2) −3.3 (4.6)
Panel B. Cohort characteristics
Female cohort size (18–24) 261 (3.1) 261 (2.0) 0 (3.7) 262 (3.4) −1.7 (4.6)
Notes: Panel A is for all 286 peers’ daughters rst marrying in 1858–1866. “Duke/Earl/Marquess” and “Peerage
of England” are proportions. Panel B shows year-averages for the number of peers’ daughters aged 18–24. Standard
errors are in parentheses.
VOL. 14 NO. 3 461
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
described above, T
t
is based on the percentage of women marrying at a given age
in normal times. y
i,t
is a discrete outcome (e.g., married a commoner), and Φ is the
CDF of the standard normal distribution. For continuous outcomes, I estimate
(3) Y
i,t
= A + B T
t
+ 𝐗
i,t
′ Δ + ϵ
i,t
,
where Y
i,t
is the difference between spouses’ family landholdings. The coefcients
of interest, β and B , capture the effect of the interruption of the Season. The vector
X includes alternative predictors of marriage outcomes: family title, women’s birth
order, peerage of origin, and distance from family seat to London. Given that the
treatment variable varies across birth cohorts, I cluster standard errors by birth year.
I also report p-values from the bootstrap-t procedure (Cameron, Gelbach, and Miller
2008) to account for the small number of clusters ( G = 21 ).
34
Table2 reports estimates of equations(2) and (3) for the baseline sample, i.e.,
peers’ daughters aged 15–35 in 1861 who ever married.
35
The interruption of the
Season reduced marital sorting. Specically, women who—based on marriage
behavior in “normal times”—were at risk of marriage in 1861–1863 were more
likely to marry a commoner (column 1). This is consistent with the hypothesis
that the Season was crucial to prevent peer–commoner intermarriage. The inter-
ruption also reduced the probability to marry an heir (column 2). Given that in
Britain only heirs inherited titles, this was an important margin of marriage quality.
To get a sense of the magnitudes, consider two cohorts separated by a small age gap:
women aged 22 and 25 in 1861. In the absence of the marriage market disruption,
we would expect them to end up marrying similar husbands. However, the synthetic
34
The preferred cluster resampling scheme in Cameron, Gelbach, andMiller (2008), wild bootstrap, is only
suited for linear regressions. For my probit estimates in equation (2), I use the pairs resampling.
35
I exclude women marrying foreigners and members of the royal family.
F5. T I S D S’ S
Note: The sample is 68 marriages where both spouses’ family seats are in Burke (1826).
100
120
140
160
180
200
Miles
0
2,000
4,000
6,000
Attendees
1858 1859 1860 1861 1862 1863 1864 1865 1866
Royal parties (left axis)
Distance between spouses’ seats (right axis)
462 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
probability to marry in 1861–1863 was one standard deviation larger for women
aged 22 in 1861. As a result, they were 5 percent more likely to marry a commoner
and 10 percent less likely to marry an heir than women aged 25 in 1861.
The interruption of the Season also reduced sorting by landed wealth. Columns 3
to 5 restrict the sample to marriages for which Bateman (1883) lists both spouses’
family landholdings. This selected sample allows me to test whether the interruption
also affected sorting within the landed elite, that is, at the very top of the distribu-
tion. Compared to the baseline sample, some covariates are different, e.g., there are
more dukes’, marquesses’, and earls’ daughters. That said, the synthetic probability
and the actual proportion of women married in 1861–1863 are similar across sam-
ples (online Appendix A6). In other words, this sample and the baseline sample
were similarly exposed to the interruption.
In column 3, I evaluate the difference between spouses’ family landholdings.
Specically, the dependent variable is the difference between spouses’ percentile
T 2—T S’ I M O, P, OLS E
Spouses’ landholdings (rank percentile)
Married a
commoner
Married
an heir
Difference
(absolute value)
Difference
(husband − wife)
Married
down
Never
married
(1) (2) (3) (4) (5) (6)
Panel A. Baseline
Treatment
a
0.005
−0.004
0.524
−0.516
0.009 0.002
(0.002) (0.002) (0.196) (0.225) (0.003) (0.002)
[0.043] [0.033] [0.029] [0.043] [0.010] [0.227]
Controls Yes Yes Yes Yes Ye s Yes
Observations 644 644 324 324 324 765
Percent correct 66 74 — — 64 77
Mean of dep var. 0.65 0.26 29
−3
0.53 0.23
Panel B. Controlling for distance to London
Treatment
a
0.006
−0.005
0.512
−0.537
0.009 0.002
(0.002) (0.002) (0.213) (0.221) (0.003) (0.002)
[0.039] [0.021] [0.037] [0.059] [0.044] [0.282]
Distance 0.0002
−0.0001
0.028
−0.039
0.0005
−0.0003
(0.0002) (0.0002) (0.015) (0.015) (0.0002) (0.0001)
[0.502] [0.834] [0.088] [0.012] [0.074] [0.030]
Controls Yes Yes Yes Yes Ye s Yes
Observations 484 484 260 260 260 565
Percent correct 62 73 — — 65 79
Mean of dep var. 0.62 0.27 30
−5
0.55 0.22
Model Probit Probit OLS OLS Probit Probit
Notes: This table reports estimates of equations(2) and(3). The baseline sample is all peers’ daughters aged 15–35
in 1861 who ever married, excluding second marriages, women married to foreigners, and royals. Columns 3 to 5
evaluate sorting by landholdings and hence, mechanically exclude women for which Bateman (1883) does not list
both spouses’ family landholdings. Column 6 considers 765 peers’ daughters aged 15–35 in 1861, including those
who never married and excluding those who died before age 35. Controls are indicators for dukes’/marquesses’/
earls’ daughters and for English peerages, and birth order excluding heirs. Panel B also includes the distance
between the family seat and London. Hence, it restricts the sample to women with a recorded family seat. Standard
errors clustered by birth year in parentheses and p-values from the bootstrap-t procedure (Cameron, Gelbach,
andMiller 2008) are in squared brackets.
a
Synthetic probability (percent) to marry during interruption, based on marriage probabilities in normal years.
VOL. 14 NO. 3 463
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
rank in acres, in absolute value. A value of zero indicates that both spouses’ families
are in the same percentile of the distribution; larger values indicate less sorting. I
nd that increasing the risk to marry in 1861–1863 by one standard deviation (7.3
pp) increases the difference in spouses’ family landholdings by 4 percentile ranks.
Given the sample averages, this corresponds to a 31 percent reduction in sorting.
36
This reduction in sorting should take the form of women marrying down. Relative
to men, women were pressured to marry younger and hence, were more adversely
affected by the three-year interruption of the Season. Columns 4 and 5 conrm this
hypothesis: increasing a woman’s synthetic probability to marry in 1861–1863 by
one standard deviation is associated to marrying a husband 4 percentile ranks poorer
and increases the probability to marry down by 12 percent. This nding is not spe-
cic to Victorian Britain. Low (2017) shows that women’s pressure to marry young
(due to a depreciation of their reproductive capital) may carry economic losses in
modern marriage markets.
In column 6, I examine marital rates. I consider women aged 15–35 in 1861,
whether they married or not. To avoid counting women who died at an early age
as celibate, I exclude those dying before age 35. Results show that women at risk
of marriage during the interruption were more likely to never marry, although the
effect is not statistically different from zero.
Finally, panel B reports estimates controlling for the distance between a woman’s
family seat and London. Hence, the sample is restricted to women with family seats
recorded in Burke (1826). This covariate is important because attending the Season
may have been more costly for women living further away. Although families typi-
cally rented a house in London for the entire Season, some may have stayed in their
country seats and traveled there for specic events. This covariate is signicantly
associated with sorting by family landholdings: women living further from London
sorted less by landholdings (column 3) and married down more (columns 4 and
5). That said, the main estimates are robust. After controlling for the distance to
London, I nd that women at risk of marriage during the interruption were more
likely to marry a commoner, less likely to marry an heir, sorted less by landholdings,
and married poorer husbands within the landed elite. The magnitude of the effects is
comparable to those in panel A.
Next, I perform placebo tests. I consider 40 cohorts who were on the mar-
riage market x ∈
{
10, 11, …, 50
}
years before the interruption of the Season
(1861–1863). Specically, each placebo sample and treatment is dened analo-
gously to the baseline case but x years before.
37
Then I estimate the effect of a pla-
cebo interruption between 1861 − x and 1863 − x on the probability of marrying a
commoner. Figure6 presents the results. The placebo estimates are close to zero and,
in most cases, signicantly different from the baseline estimate. Only 1 out of the 40
36
Under strict male primogeniture, the rst-born son would inherit all the family’s estates and its associated
incomes. In this case, marrying a husband from a family with few landholdings would only matter if he was the heir.
However, the peerage’s inheritance system granted younger brothers (and sisters) portions and yearly allowances
proportional to the family estates’ size (see SectionIB). Hence, marrying a non-heir from a poorer family implied
an economic loss.
37
Each placebo sample are women aged 15–35 in 1861 − x who ever married (excluding those in the baseline
sample). The placebo treatment is the synthetic probability to marry from 1861 − x to 1863 − x , based on the prob-
ability to marry at each age of women born from 1815 − x to 1830 − x .
464 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
placebo tests reports a positive, marginally signicant estimate, and its magnitude
is only half of the baseline estimate (see the placebo interruption in 1830–1832). In
other words, marriage outcomes were distorted only when the Season was actually
interrupted. This suggests that my baseline specication captures the effect of the
interruption and not any confounding factor correlated with age.
So far, I assumed that cohorts with a similar synthetic probability to marry in
1861–1863 (e.g., women aged 19 and 22 in 1861) respond similarly to the Season’s
interruption. Next, I estimate a exible specication where I include xed effects for
a woman’s age in 1861 instead of the treatment, T
t
. This allows each age cohort to
respond differently to the interruption of the Season. Formally, I estimate
(4) Pr
(
y
i,t
= 1 | 𝐗
i,t
)
= Φ
(
μ
t
+ 𝐗
i,t
′ δ
)
,
where μ
t
are xed effects for a woman’s age in 1861; t ∈
{
15, 16, …, 35
}
; and
y
i,t
indicates marrying a commoner.
Figure7 presents the results graphically. The reference cohort is women aged 16
in 1861. They were not at risk of marriage during the Season’s interruption—their
synthetic probability to marry in 1861–1863 is only around 5 percent. Likely, this
F6. P T. D V: M C
Notes: This gure presents marginal effects and 95 percent condence intervals from equation(2). Each placebo
test considers a sample and a treatment ( T ) dened analogously to the baseline case but x ∈
{
10, 11, … 50
}
years
before. The placebo sample are women aged 15–35 in 1861 − x who ever married (excluding those in the baseline
sample); T
t
placebo
is the synthetic probability to marry during a placebo interruption between 1861 − x and 1863 − x,
based on the percent of women married at each age in a previous cohort, i.e., women born between 1815 − x and
1830 − x . All estimates include the full set of controls and the distance between family seat and London. On top of
each condence interval, I report p-values for a test of equality of coefcients (baseline versus placebo).
0.06
0.01
0.01
0
0
0
0
0
0.02
0.02
0.01
0.01
0
0.01
0
0
0.01
0.01
0.12
0.14
0.41
0.23
0.25
0.2
0.31
0.22
0.08
0.1
0.12
0.13
0.17
0.08
0.03
0.09
0.16
0.11
0.1
0.21
0.17
0.19
p = 0.037
−0.015
−0.01
−0.005
0
0.005
0.01
0.015
Marginal effect of T (synthetic probability)
1851–1853
1846–1848
1841–1843
1836–1838
1831–1833
1826–1828
1821–1823
1816–1818
1811–1813
Years of placebo interruption
Baseline (1861–1863)
Placebo
VOL. 14 NO. 3 465
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
young cohort were presented at court (i.e., announced to the marriage market)
after the Season resumed. Relative to them, older women have a 25 to 30 percent
higher probability to marry a commoner. The effect declines for those aged 24 and
older and is only marginally signicant for women aged 26 in 1861. The latter,
who likely married before the interruption, also have a low synthetic probability
(around 10 percent). Overall, the gure suggests that peer–commoner intermarriage
was more prevalent for cohorts with a high synthetic probability to marry during
the interruption. Finally, the right panel shows estimates from equation (4) on the
(pooled) placebo samples described above, i.e., women on the marriage market
x ∈
{
10, …, 50
}
years before the interruption. For them, there are no visible
cohort effects: the coefcients are tightly estimated around zero for all age groups.
Again, this suggests that a woman’s birth cohort only affected her probability to
marry a commoner around the interruption of the Season.
In sum, this sectionshows that a matching technology with low search costs
and market segmentation generates sorting. The Season announced who was on
the market, created multiple settings for the opposite sexes to meet, and segre-
gated the rich from the poor. Women at risk of marriage when this matching tech-
nology was operative sorted more than women at risk of marriage when it was
interrupted.
F7. P–C I, P E A D
Notes: The left panel shows marginal effects and 95 percent condence intervals for a set of dummies indicat-
ing a woman’s age in 1861, μ
t
, in equation(4). The baseline sample are women aged 15 to 35 in 1861. The right
panel plots the corresponding estimates for a placebo test: I pool 40 cohorts of women aged 15 to 35 in 1861 − x ,
where x ∈
{
10, 11, … 50
}
, excluding women in both samples. All samples include women who ever married and
exclude second marriages, women married to foreigners, and royals. The dashed line shows the synthetic probabil-
ity to marry during the interruption, T .
5
10
15
20
25
0
−20
10
−10
20
30
40
50
60
Probability of marrying a commoner (percent)
16 17 18 19 20 21 22 23 24 25 26
Age in 1861
Baseline
5
10
15
20
25
−20
−10
0
10
20
30
40
50
60
16 17 18 19 20 21 22 23 24 25 26
Age in 1861 − x
Placebo
Fixed effects for age in 1861 (left axis)
Synthetic probability of marrying at interruption (right axis)
466 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
C. IV Estimates
So far, my estimation strategy resembles a reduced-form IV. The synthetic proba-
bility to marry in 1861–1863 captures exogenous variation in the actual probability
to marry in 1861–1863 and is regressed directly against marriage outcomes. Here, I
estimate the full-IV model. This is interesting in its own right, especially to quantify
the aggregate effect of the Season’s interruption. Formally, I treat the decision to
marry during the three-year interruption, M , as endogenous and instrument it with
T, the synthetic probability to marry in 1861–1863. In the second stage, I regress the
(instrumented) M on the marriage outcomes described above.
The identifying assumptions are that the instrument is relevant and that the exclu-
sion restriction is satised. First-stage estimates support the rst assumption. As for
the exclusion restriction, note that the instrument, i.e., the synthetic probability to
marry in 1861–1863, is based on marriage behavior in a previous cohort. Hence, it is
independent of women’s (endogenous) marriage decisions. Moreover, SectionIIIA
shows that the instrument could not affect marriage outcomes through channels
other than the Season’s interruption.
Panel B of Table3 presents rst-stage results. As before, I use my baseline sam-
ple: peers’ daughters aged 15 to 35 in 1861 who ever married. Increasing the syn-
thetic probability to marry in 1861–1863 by 1 percentage point is associated with
an increase in the actual probability to marry in 1861–1863 by 1 percentage point.
This shows that social norms and the pressure to marry young did not change during
the interruption: the probability to marry at a given age in a previous cohort (the
instrument) predicts well the timing of marriages around the interruption. The F-stat
is large for the full sample (columns 1 and 3) but falls when the sample is restricted
to 484 women with recorded family seats (even columns) and to 324 marriages with
data on spouses’ family landholdings (columns 5 to 10). To address this, I report
p-values based on Moreira (2003) conditional likelihood ratio (CLR).
Panel A reports second-stage results. The three-year interruption of the Season
had a large aggregate impact: it increased peer–commoner intermarriage dramati-
cally and reduced sorting within the landed elite. Women who (exogenously) mar-
ried during the interruption of the Season were 39–54 pp more likely to marry a
commoner and 33–45 pp less likely to marry an heir than women marrying before
and after the interruption (columns 1 to 4). Comparing estimates to sample means
suggests that the rate of peer–commoner intermarriage would have been 60–87 per-
cent larger in the absence of the Season. In addition, marrying during the interrup-
tion increased the difference in spouses’ landholdings by 45–58 percentile ranks
(columns 5 and 6). Women married husbands 44–61 percentile ranks poorer (col-
umns 7 and 8). This corresponds to marrying a husband on the twentieth instead
of the eightieth percentile (5,000 versus 31,000 acres). Overall, the probability to
marry down was 58–71 pp higher for women marrying during the interruption.
D. Nonparametric Estimation
Here I show that the interruption not only increased peer–commoner intermar-
riage but also disrupted sorting across different nobility titles. Similarly, I show
VOL. 14 NO. 3 467
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
that sorting by landholdings was distorted at various moments of the distribution
beyond the mean. To do so, I use nonparametric methods based on contingency
tables and Kolmogorov-Smirnov distribution tests. In detail, I consider my baseline
sample (women aged 15–35 in 1861) and compare a high- versus a low-treatment
cohort. The high-treatment cohort are women with a synthetic probability to marry
T 3—T S’ I M O, IV E
Married a
commoner Married an heir
(1) (2) (3) (4)
Panel A. Second stage
Married in 1861–1863 0.39 0.54
−0.33 −0.45
(0.14) (0.15) (0.13) (0.15)
Dependent variable mean 0.65 0.62 0.26 0.27
Dependent variable: Married
during interruption (1861–1863)
Panel B. First stage
Treatment
a
0.01 0.01 0.01 0.01
(0.00) (0.00) (0.00) (0.00)
Controls Yes Ye s Yes Yes
Distance London No Yes No Yes
Observations 644 484 644 484
F rst-stage 18.3 11.6 18.3 11.6
Model IV probit IV probit
Spouses’ landholdings
Difference
(absolute value)
Difference
(husband−wife) Married down
(5) (6) (7) (8) (9) (10)
Panel A. Second stage
Married in 1861–1863 45.3 58.5
−44.6 −61.4
0.58 0.71
(17.4) (23.4) (22.0) (30.0) (0.12) (0.16)
[0.04] [0.06] [0.02] [0.02] [0.00] [0.02]
Dependent variable mean 29 30
−3 −5
0.53 0.55
Dependent variable: married during interruption (1861–1863)
Panel B. First stage
Treatment
a
0.01 0.01 0.01 0.01 0.01 0.01
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Controls Yes Ye s Yes Yes Ye s Yes
Distance London No Yes No Yes No Yes
Observations 324 260 324 260 324 260
F rst-stage 9.2 4.5 9.2 4.5 9.2 4.5
Model
IV (liml) IV (liml)
IV probit
Notes: The baseline sample is all peers’ daughters aged 15–35 in 1861 who ever married, excluding second mar-
riages, women married to foreigners, and royals. Columns 5 to 10 evaluate sorting by landholdings and hence,
mechanically exclude women for which Bateman (1883) does not list both spouses’ family landholdings. Controls
are dened in Table2. Even columns also include the distance between the family seat and London and hence,
restrict the sample to women with a recorded family seat. Standard errors clustered by birth year are in parenthe-
ses. When the rst-stage F-stat is below 10, I report p-values based on Moreira’s (2003) conditional likelihood ratio
(CLR) in brackets.
a
Synthetic probability (percent) to marry during Season interruption, based on marriage probabilities in “nor-
mal times.”
468 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
during the interruption above 20 percent. This corresponds to the top quintile: the
20 percent of women with the highest synthetic probability to marry in 1861–1863.
38
Conversely, the low-treatment cohort are women below the top quintile.
I begin by showing that the interruption disrupted sorting across different titles
in the nobility. Specically, I construct a contingency table of wife’s and husband’s
title for the high- and low-treatment cohorts (Table4). The wife’s title is arrayed
across rows i . I consider barons’ and viscounts’ daughters (henceforth, BV’s daugh-
ters) versus dukes’, marquesses’, and earls’ daughters (henceforth, DME’s daugh-
ters).
39
The latter are the highest ranks of the peerage. Their husbands’ titles are
arrayed across columns j . Each cell reports observed frequencies ( O ) and expected
frequencies under random matching ( E ). Expected frequencies are E
i,j
=
n
i
× n
j
_
N
,
where n
i
is the number of counts in the i th row, n
j
is the number of counts on the j th
column, and N is the total number of counts in the table.
Table4 shows that when the Season ran smoothly, marital sorting was stronger.
Consider the low-treatment cohort in panel A: expected frequencies are similar for
DME’s and BV’s daughters.
40
That is, under random matching, they should marry
similarly. In practice, DME’s daughters married more peers’ heirs and fewer com-
moners than the lower-ranked BV’s daughters. Sorting patterns are different for the
high-treatment cohort, that is, those most exposed to the interruption of the Season:
observed and expected frequencies are similar for both DME’s and BV’s daughters.
In other words, their marriages resemble random matching.
41
Next, Table5 compares sorting patterns across cohorts using chi-squared tests of
association. First, it presents Pearson’s chi-squared test of association ( χ
2
), which
evaluates whether spouses sorted by title. Specically, the null hypothesis is that
marriages were random with respect to title. The test-statistic is
(5) χ
2
=
∑
i=1
r
∑
j=1
c
(
O
i,j
− E
i,j
)
2
___________
E
i,j
,
where O
i,j
and E
i,j
are the observed and expected frequencies in cell
{
i, j
}
, r is the
number of rows, and c is the number of columns.
For women in the low-treatment cohort, the test rejects the null hypothesis
that marriages were randomly set. In contrast, for the high-treatment cohort, the
chi-square statistic is 7 times lower, and the null of random matching cannot be
rejected. Column 3 conrms that the chi-square statistics are signicantly differ-
ent between the high- and low-treatment cohort. In other words, when the Season
worked smoothly, women sorted by title; when the Season was interrupted, marriage
resembles random matching. This result is not a by-product of the smaller sample in
the high-treatment cohort. For a 2-by-4 contingency table, the Pearson’s chi-squared
38
Specically, the high-treatment cohort are women aged 19 to 22 in 1861 (see Figure4).
39
The Hollingsworth dataset groups titles into DME versus BV to distinguish high versus low peerage titles. In
online AppendixB3, I construct contingency tables with ve rows, i.e., dukes’, marquesses’, earls’, viscounts’, and
barons’ daughters. Results are robust.
40
Note that to calculate the expected frequencies for marriages with commoners ( E
1,1
and E
2,1
), I only consider
commoners who married into the peerage.
41
Random matching predicts 51 percent low-treatment DMEs to marry commoners. In practice, only 42 percent
did so. In the high-treatment cohort, we expect 53 and observe 49 percent of such marriages.
VOL. 14 NO. 3 469
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
test requires no cells with zero count and an expected cell count of ve or more in
at least six cells. Both conditions are satised. Furthermore, the likelihood ratio test,
which is accurate for small samples, conrms the results.
42
Table5 also presents Kendall’s rank correlation coefcients ( τ
b
). The coefcient
is based on the number of concordances ( Q ) and discordances ( D ) in paired obser-
vations. Two marriages are concordant (discordant) if the woman with higher title
42
Formally, the likelihood ratio statistic is χ
LR
2
= 2
∑
i=1
r
∑
j=1
c
O
i,j
· ln
(
O
i,j
_
E
i,j
)
.
T 4—C T
Husband’s rank at age 15
Commoner Gentry Peer’s son Peer’s heir Observations
Panel A. Low-treatment cohorts (T < eightieth percentile)
a
Wife
Baron/Viscount’s daughter
O 161 34 20 49 264
E 135.8 32.8 25.2 70.2
Duke/Earl/Marquess’s daughter
O 108 31 30 90 259
E 133.2 32.2 24.8 68.8
Observations 269 65 50 139 523
Panel B. High-treatment cohorts ( T ≥ eightieth percentile)
a
Wife
Baron/Viscount’s daughter
O 34 11 5 10 60
E 31.7 8.9 5.5 13.9
Duke/Earl/Marquess’s daughter
O 30 7 6 18 61
E 32.3 9.1 5.5 14.1
Observations 64 18 11 28 121
Notes: The baseline sample is all peers’ daughters aged 15–35 in 1861 who ever married, excluding second mar-
riages, women married to foreigners, and royals (Observations = 644). Cells report observed (O) and expected fre-
quencies under random matching: E =
(
n
i
× n
j
)
/ N , where n
i
is the i th row’s counts, n
j
is the j th column’s counts,
and N is the table’s total counts.
a
T: Synthetic probability to marry in 1861–1863, based on marriage probability in normal times.
T 5—T I S T, N E
Low-treatment High-treatment Difference
(1) (2) (1) − (2)
Pearson’s chi-squared, χ
2
24.6 3.5 21.1
(0.000) (0.320) (0.000)
Likelihood ratio, LR − χ
2
24.9 3.6 21.3
(0.000) (0.315) (0.000)
Kendall’s rank correlation, τ
b
0.20 0.11 0.09
(0.000) (0.191) (0.164)
Observations 523 121 644
Notes: The baseline sample is peers’ daughters aged 15–35 in 1861 who ever married, exclud-
ing second marriages, women married to foreigners, and royals (Observations = 644). Low-
(high-) treatment cohorts are women with a synthetic probability to marry in 1861–1863 below
(above) the eightieth percentile. τ
b
ranges between −1 (negative) and +1 (positive assortative
matching). Column 3 converts stats into Spearman correlation and uses Fisher’s Z transfor-
mation (Rosenberg 2010). The distribution of τ
b
is bootstrapped; p-values are in parentheses.
470 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
of the two married the husband with higher (lower) title of the two. If the women’s
or men’s titles are the same, the pair is tied. Formally,
(6) τ
b
=
Q − D
______________________________________
√
_______________________________
(
N
(
N − 1
)
/ 2 − t
wom
)
(
N
(
N − 1
)
/ 2 − t
men
)
,
where N is the number of marriages; and t
wom
=
∑
i
t
i
(
t
i
− 1
)
/ 2 , where t
i
is the
number of ties in woman’s title i ( t
men
is calculated analogously). Kendall’s coef-
cient ranges between −1 (negative association) and +1 (positive association).
In the low-treatment cohort, the Kendall’s rank correlation is positive and signi-
cantly different from zero: higher-titled women married men with higher titles. In
other words, there was positive assortative matching. This result vanishes when the
Season was interrupted: Kendall’s rank correlation is halved and not signicantly
different from zero for the high-treatment cohort. A one-sided test rejects the null
hypothesis that Kendall’s rank correlation was higher for the high-treatment cohort.
That said, I cannot reject the null that the high- and low-treatment have the same
Kendall’s rank correlation, with a p-value of 0.16.
Note that these estimates consider the baseline sample of women aged 15–35 in
1861 who ever married. This excludes unmarried individuals and men who failed to
marry a peer’s daughter. In online AppendixB4, I construct extended contingency
tables with these populations and show that the nonparametric results are robust.
Finally, I present nonparametric estimates for the effect of the interruption on
sorting by landholdings. By construction, the sample is restricted to women mar-
rying in the landed elite—i.e., those for which Bateman (1883) lists both spouses’
family landholdings. I use a Kolmogorov-Smirnov test to compare sorting by land-
holdings for the high- versus low-treatment cohorts. The test statistic is
(7) K-S
n,m
= sup
y
| H
n
(
y
)
− L
m
(
y
)
|,
where H and L are the cumulative distribution functions for the high- and
low-treatment cohort; n and m are the sizes of each cohort; and y is the measure of
sorting: the difference between spouses’ percentile rank in landholdings.
Figure8 presents the results. Panel A shows the difference between spouses’ per-
centile ranks, in absolute value. When the Season worked smoothly, spouses were
similar in terms of landholdings. For example, 50 percent of the marriages in the
low-treatment cohort were between spouses ranked 20 percentiles away. In con-
trast, among high-treatment women, only 30 percent married husbands within 20
percentiles. Overall, the Kolmogorov-Smirnov test shows that spouses’ ranks were
more similar in the low- than in the high-treatment cohort. In other words, women
at risk of marriage during the Season’s interruption sorted less by landholdings. The
evidence also suggests that after the interruption, sorting returned to its previous
levels: sorting patterns are similar for low-treatment women who were “younger”
and “older” than the high-treatment cohort. That is, for women who courted, respec-
tively, after and before the interruption.
The disruption in sorting patterns is mostly driven by women marrying down.
Panel B shows the difference between husband and wife in percentile ranks.
VOL. 14 NO. 3 471
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
Hence,negative values correspond to women marrying poorer husbands. While 20
percent of high-treatment women married husbands 50 percentiles poorer, only 6
percent of low-treatment women did so. The Kolmogorov-Smirnov test conrms
that the distributions are different. In other words, that women with a higher risk to
marry during the interruption married down more.
E. Robustness and Extensions
I perform several robustness checks and extensions of the analysis. This section
briey describes them; the detailed results are available in the online Appendix.
Treatment.—A woman’s synthetic probability to marry during the interruption is
based on (i) her age in 1861–1863 and (ii) the probability to marry at each age in a
benchmark cohort who married in normal times. So far, I have used peers’ daughters
born in 1815–1830 as benchmark cohort. TablesB1 andB2 in the online Appendix
show that my results are robust to using alternative benchmark cohorts. I report
estimates of equations(2) and(3) where I dene the treatment using ve alterna-
tive benchmark cohorts: women born in 1810–1825, born in 1820–1835, married
in 1845–1860, married in 1840–1855, and married in 1835–1850. As before, these
exclude women married after 1861 and dead before age 30.
Nonparametric Estimates for Marriage Cohorts.—In Section IIID, I assigned
women to the high- and low-treatment group based on their risk to marry during
the interruption. That is, based on their age cohort. Online AppendixB2, instead,
compares marriage cohorts: women marrying during the three-year interruption
(treatment) versus women marrying three years before (control). Although marriage
F8. S L, N E
Notes: The sample are peers’ daughters aged 15–35 in 1861, for which Bateman (1883) lists both spouses’ fam-
ily landholdings (Observations = 324). “ High-treatment” cohorts are women with a synthetic probability to marry
in 1861–1863, T , above the eightieth percentile (Observations = 64). “ Low-treatment” comprises women with
T < eightieth percentile (Observations = 256). The latter is subdivided into “younger” (Observations = 64) and
“older” than the high-treatment cohort (Observations = 192).
0
0.2
0.4
0.6
0.8
1
Cumulative
0 20 40 60 80 100
|wife's percentile rank of acres − husband's
percentile rank of acres|
Panel A
Marrying
down
0
0.2
0.4
0.6
0.8
1
Cumulative
−100 −50 0 50 100
Difference in acres’ percentile
rank (husband − wife)
Panel B
High versus low K-S: 0.23 (p-value 0.005)
High versus younger K-S: 0.26 (
p-value 0.021)
High versus older K-S: 0.24 (
p-value 0.005)
High-treatment (T ≥ 80 percentile)
Low-treatment (
T < 80 percentile)
Older
Younger
High-treatment
(
T ≥ 80 percentile)
Low-treatment
(
T < 80 percentile)
Older
Younger
K-S: 0.19 (
p-value 0.036)
K-S: 0.16 (
p-value 0.292)
K-S: 0.20 (
p-value 0.029)
472 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
decisions are potentially endogenous, this exercise shows that the estimated effects
are driven by women who actually married in 1861–1863. In other words, com-
paring marriage or age cohorts yields similar results. This further suggests that the
pressure to marry young was high and that women could not endogenously select to
marry during the interruption, before, or after.
Results for Men.—In online AppendixC, I estimate the effect of the Season on
men. I rst provide evidence that the interruption is a valid instrument for peers’
sons: those who married in 1861–1863 were neither negatively selected nor the
black sheep in their families.
43
That said, male aristocrats were not so pressured
to marry young as women. Hence, it is less obvious how to dene the treatment
and sample or whether to base it entirely on ages in 1861. To overcome this, I use
marriage cohorts and exploit an additional source of disruption to the Season over
a longer time window: changes in the size of the marriageable cohort. Specically,
I estimate an instrumental variables model for peers and peers’ sons marrying in
1851–1875. The treatment variable capturing the Season’s intensity in each year is
the number of attendees at royal parties. In the rst stage, I instrument attendance
with an indicator for the interruption and with the size of the marriageable cohort
(the number of peers’ daughters aged 18–24). The second stage regresses a man’s
marriage outcome on attendance at the Season in the year of his marriage.
The biggest threat to identication is if the size of the cohort also affected local,
decentralized marriage markets, which emerged around peers’ seats during the
months when the Season was inactive. This scenario is unlikely. Gautier, Svarer,
and Teulings (2010) and Botticini and Siow (2011) show that local, decentralized
marriage markets are typically not subject to increasing returns to scale. That is,
they are not affected by the size of the cohort.
44
As for the relevance assumption, the
instrument accounts for substantial variation in Season’s attendance: increasing the
cohort by one individual increased attendance by 80 people (see online Appendix
TableC2).
A higher attendance to the Season strengthened sorting among male aristocrats.
It reduced peer–commoner intermarriage, although the effects are weaker than for
women (online Appendix TableC2).
45
The effects are larger for sorting by landed
wealth: increasing attendance at the Season by 5 percent—250 additional attend-
ees—reduced spouses’ difference in acreage by 0.6–0.7 percentile ranks (online
Appendix TableC3).
46
I also show that the Season increased sorting by an index
based on the rst principal component of title, acreage, land rents, antiquity of the
43
The percentage of male heirs and of higher peerage ranks married in 1861–1863 is very similar to that in the
three years before and after. Also, family xed effect estimates show that peers’ sons who married in 1861–1863 did
so at the same age as their brothers who married in normal times. This refutes the possibility that, within peerage
families, some sons delayed marriage until the Season resumed, while others (the “black sheep”) selected to marry
during the interruption.
44
In addition, the Sargan test cannot reject the instruments’ exogeneity, and I show that the bias would be small
for slight violations of the exclusion restriction (see online AppendixC5).
45
Online Appendix TableC2 also reports estimates for women. Results are consistent with those of SectionIIIB,
giving credibility to the econometric specication used here.
46
The effect is stronger than for marrying commoners because of market depth. The Season, a meeting technol-
ogy, was indispensable to meet the fewer great landowners’ daughters.
VOL. 14 NO. 3 473
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
lineage, the peerage of the title, and the ownership of woods. Finally, by central-
izing the marriage decisions in London, the Season matched spouses from distant
geographical origins.
47
Increasing Returns to Scale (IRS).—In online AppendixC4, I use the framework
described above to show that the Season displayed IRS: the output of the matching
function (i.e., sorting) increases more than the proportional change in inputs (i.e.,
attendees). Whether the matching function displays IRS or not has important eco-
nomic and demographic implications. For example, under IRS, fertility booms will
echo over time: a “boom” cohort may encounter partners more easily, marry earlier
on, and hence, have more children. Previous studies have estimated returns to scale
in marriage markets by comparing the city and the countryside (Gautier, Svarer, and
Teulings 2010; Botticini andSiow 2011). Instead, I consider a matching technology
that not only pooled singles together but explicitly facilitated their courtship. In this
respect, my results provide better insights for increasingly common matching tech-
nologies, e.g., dating websites.
IV. Implications: Elite Capture and Public Goods
So far, I have shown that the interruption of the Season increased women’s mar-
riages to commoners. This sectionexamines the corresponding political-economy
implications. First, I examine elections of Members of Parliament and show that a
woman’s marriage to a commoner reduced her blood relatives’ political power in
the following decades. Next, I show that families who lost political power could
not effectively oppose state-education provision in the 1870s—a policy otherwise
subject to capture by local landowners (Stephens 1998).
A. Marital Sorting and Political Power
Elsewhere it has been argued that elites pursue and maintain political power
by opposing inclusive institutions,
48
by exploiting their economic power (Baland
and Robinson 2008), and through personal ties—i.e., marrying strategically to pre-
vent entry by newcomers (Marcassa, Pouyet, and Trégouët 2020). For example,
Cruz, Labonne, andQuerubín (2017) show that candidates for public ofce in the
Philippines are disproportionately drawn from central families in the marriage net-
work. Similarly, Puga andTreer (2014) nd that the wealthiest Venetian merchants
used marriage alliances to monopolize the galley trade and to block political com-
petition. The Season was no different. Attending it and marrying in the peerage was
crucial for political achievement. Lady Aberdeen wrote that the Season “was a part
of the very life of the people who had the largest stake in the country (…) Nobody
could well come to the front without participating in it to some degree” (Ellenberger
1990, 637).
47
Online AppendixC4 evaluates match quality, proxied by consanguinity and fertility.
48
Sokoloff andEngerman (2000); Acemoglu (2008); Allen (2009); Galor, Moav, andVollrath (2009).
474 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
Did the interruption of the Season and the increase in women’s marriages to com-
moners affect the political power of peerage families? To evaluate this question, I
use the synthetic probability to marry during the interruption as an instrument for
a woman’s marriage to a commoner. I then look at whether a woman’s marriage
to a commoner affected her blood relatives’ probability to be elected Member of
Parliament in the House of Commons. I focus on MPs because being elected is
a good proxy for an individual’s overall political power and inuence on public
policy. In normal circumstances, landed aristocrats controlled MP elections by dis-
tributing favors, rents, and jobs among the local electorate—especially before the
introduction of the secret ballot in 1872 (Baland andRobinson 2008). According
to Edward Stanley, “When any man attempted to estimate the probable result of a
county election in England, it was ascertained by calculating the number of the great
landed proprietors in the county and weighing the number of occupiers under them”
(Baland and Robinson 2008, 1738).
Although women could not become MPs, the interruption of the Season and
women’s marriages to commoners could affect MP elections through several chan-
nels. One possibility is that a woman exposed to the interruption became poorer
because she married a commoner or a husband with smaller family landholdings
(and hence, with a smaller allowance). Since the husband could not provide her
with the life of comfort she was accustomed to, her birth family had to step in. They
mobilized considerable capital (portions, allowances, etc.) to sustain her, divert-
ing resources away from their local electorate. Another related possibility is that
a woman’s marriage to a commoner reduced her birth family’s social prestige and
hence, its political power. Some noblewomen played the role of “power brokers,”
bringing “together politicians for the informal social contact which could make or
break a career” (Atkins 1990, 45). Marrying a commoner limited a woman’s role
as power broker and her capacity to promote her kin’s political career through such
informal social contact. Similarly, Allen (2009) argues that many public ofces
were appointed through patronage, not merit. A woman’s marriage to a commoner
could exclude her kin from these appointments and/or limit her kin’s capacity to
allocate jobs (Allen 2009, 306). This, in turn, could reduce the family’s grasp over
the electorate.
To evaluate these issues empirically, I consider 279 women in my baseline sam-
ple (aged 15–35 in 1861) with a family seat in England. I restrict the sample to
England because only there I have state-education data for SectionIVB. To measure
the political power of a woman’s family, I check whether her brothers were elected
MP and how many years they served. I also look at whether they were elected
MP in the family seat’s constituency. This proxies for local political power, which
could affect policies implemented locally—e.g., the introduction of state education
(Stephens 1998). I also evaluate the political power of the head of a woman’s birth
family (henceforth, family head). Specically, I assess whether the family head in
the 1870s—when state education was introduced—was elected MP and how many
years he served.
FigureA8 in the online Appendix provides evidence that a woman’s marriage to a
commoner reduced her birth family’s political power. It considers the sample of peer-
age families described above and reports the number of brothers elected MP before
VOL. 14 NO. 3 475
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
and after their sister’s marriage. The thin blue line (thick red line) is for women who
married in the peerage (married a commoner). Before the marriage, both groups had
the same number of MPs. Ten and 20 years after it, however, the number of MPs was
much lower for families in which a woman married a commoner.
My main specication is an instrumental variables model. The rst stage uses the
interruption of the Season to capture exogenous variation in a woman’s probability
to marry a commoner, M . Specically, I model M as in SectionIII:
(8) M
i,j,t,s
= β
M
T
t
+ 𝐗
i,j,t,s
′ δ
M
+ ν
i,j,t,s
M
,
where i indexes a woman in the baseline sample, j her birth family, t her age in 1861,
and s her family seat. The treatment, T
t
, is the synthetic probability to marry during
the Season’s interruption (see equation(3) for details). In the second stage, β
P
cap-
tures the effect of marrying a commoner on family j ’s political power:
(9) P
i,j,t,s
= β
P
M
ˆ
i,j,t,s
+ 𝐗
i,j,t,s
′ δ
P
+ ν
i,j,t,s
P
,
where P is one of the six measures of political power described above. To isolate the
effect of marriage on political power, P is restricted to MP elections after i ’s mar-
riage. The index s allows me to evaluate whether a member of family j was elected
MP in the county where s is located. The vector X includes alternative predictors
of political power: family j ’s title, the distance from seat s to London, the number
of brothers,
49
and county characteristics from Hechter (1976) (percent working in
manufacturing, log income p.c., percent voting conservative in the general elections
of 1885, percent of nonconformists, and religiosity). I also include a covariate cap-
turing family j ’s previous political power. This is dened as P but considering only
the MP elections of i ’s father before i ’s marriage.
50
This covariate controls for the
unlikely possibility that, as Parliament met in 1861–1863, the daughters of MPs
moved to London with them and attended private balls during the interruption (see
SectionIIIA). The unit of observation in the regressions is a woman. Since some of
the observations are sisters, I also report p-values clustered by family.
Table6 presents IV estimates of equations (8) and (9). The rst stage conrms
that women with a high synthetic probability to marry during the interruption were
more likely to marry a commoner. Note, however, that the sample size is smaller than
in SectionIII and, hence, the F-statistic is low.
51
To address this, I report p-values
based on Moreira’s (2003) conditional likelihood ratio.
Panel A shows that, by increasing women’s marriages to commoners, the inter-
ruption of the Season reduced the political power of some peerage families in the
following decades. Specically, after a woman’s marriage to a commoner, her broth-
ers were 50 percent less likely to be elected MP (column 1) and, together, they
49
I exclude brothers who died before reaching majority (age 21) and hence, who could not be elected MP. To
account for large families, I also include a quadratic term.
50
Using the MP elections of i ’s brothers before i ’s marriage would understate a family’s previous political
power, e.g., for women with many brothers under age 21—the age of majority.
51
This is because I restrict the sample to women with seats in England and include county covariates correlated
with MP elections (second stage) but less relevant in the rst stage.
476 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
served 18 fewer years (column 2) than the brothers of women who married in the
peerage. The loss of political power was local: the brothers of women who married
a commoner were 47 percent less likely to be elected MP (column 3) and served 11
fewer years (column 4) in the county where their (birth) family seat was located.
Since women typically moved away to live with their husbands, why would their
birth families’ local power be affected? Even if women moved away, marrying a
commoner would force her birth family to divert resources away from local elector-
ates (see discussion above). In addition, during the Season’s interruption, marriages
became more local (Figure 5). In other words, women probably married “local”
commoners, reducing her birth family’s local prestige. Admittedly, a local marriage
can help to consolidate the family’s estates and, hence, to tighten the family’s grasp
over the electorate. That said, Mingay (1963) argues that by the late nineteenth cen-
tury, local marriages were not used for estate consolidation.
T 6—W’ M C H F’ P P, IV E
Dependent variable: Family’s political power after woman’s marriage
Any
brother
is MP
All
brothers’
MP years
Any
brother is
local MP
All
brothers’
local years
Family
head
is MP
Family
head’s
MP years
(1) (2) (3) (4) (5) (6)
Panel A. Second stage
Woman married
−0.54 −18.40 −0.47 −11.00 −0.49 −7.83
a commoner
[0.004] [0.033] [0.037] [0.025] [0.006] [0.019]
{0.025} {0.015} {0.102} {0.010} {0.007} {0.013}
Woman married a peer/peer’s son
ref. ref. ref. ref. ref. ref.
Mean of dependent variable 0.42 4.76 0.23 1.84 0.18 1.39
Dependent variable: Woman married a commoner
Panel B. First stage
Treatment 0.008 0.008 0.008 0.008 0.010 0.010
(synthetic prob.)
a
(0.004) (0.004) (0.004) (0.004) (0.004) (0.004)
Woman birth order Yes Yes Yes Ye s Yes Ye s
F-statistic 3.4 3.4 3.4 3.4 3.3 3.2
Observations 270 270 270 270 279 279
Baseline controls Yes Ye s Yes Yes Ye s Yes
County controls Yes Ye s Yes Yes Ye s Yes
Number of brothers Yes Ye s Yes Yes No No
Political power before Yes Yes Ye s Ye s Yes Yes
Model IVprobit IV IVprobit IV IVprobit IV
Notes: The sample is 279 women in the baseline sample (i.e., aged 15 to 35 in 1861) with a family seat in England.
Panel A presents estimates for the effect of a woman’s marriage to a commoner on the political power of her broth-
ers (columns 1 to 4) and of the family head in the 1870s (columns 5 and 6). Columns 1 to 4 report 270 observations
because 9 women had no brothers. In odd columns, the dependent variable indicates if any brother was elected MP
(column 1), elected MP in the family seat’s county (column 3), or if the family head was elected MP. In even col-
umns, the dependent variable is the corresponding number of years served as MP. Baseline controls are indicators
for duke/marquess/earl’s families and for English titles, and distance from family seat to London. County controls
are percent working on manufacturing, income p.c., percent voting conservative in the 1885 General elections, per-
cent nonconformists, and religiosity. When a family owns seats in different counties, I take the average. Number of
brothers excludes brothers who died before age 21 and includes a quadratic term. “Political power before” is iden-
tical to the dependent variables but considers only the MP elections of fathers before a woman’s marriage. Panel A
reports CLR p-values (Moreira 2003) in square brackets and CLR p-values clustered by family in curly brackets;
panel B reports standard errors in parentheses.
a
Synthetic probability (percent) to marry during interruption, based on marriage probabilities in normal times.
VOL. 14 NO. 3 477
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
Finally, I show that those who were family heads in the 1870s also lost political
power. After a woman’s marriage to a commoner, the head of her birth family was
49 percent less likely to be elected MP and served 8 fewer years than the family
heads of women who married in the peerage.
In online AppendixB8, I show that peerage families also lost political power as a
result of marriages with a family of lower estates. I estimate the IV model described
above where M is based on the difference in spouses’ family landholdings—which
mechanically restricts the sample to those marrying in the landed elite. I nd that,
after a woman married down by family landholdings, her brothers were 42 percent
less likely to be elected MP and served 9.8 fewer years in the county where their
(birth) family seat was located (see online Appendix TableB10). Altogether, these
ndings illustrate a negative relationship between within-landed-elite marriages,
landownership, and how contested MP elections were in late nineteenth-century
England.
Admittedly, nineteenth-century Britain saw numerous political changes, includ-
ing three Reform Bills extending the franchise. That said, it is unlikely that these
changes can explain away my results. FigureA7 in the online Appendix plots the
number of sampled individuals elected MP over time. That is, the elections of fathers
and brothers of women in my baseline sample (i.e., aged 15 to 35 in 1861). In this
sample, only the 1832 Reform Act signicantly altered MP elections.
52
After the
interruption of the Season, there is no evidence of a downward trend, even though
the Reform Acts of 1867 and 1884 enfranchised the urban male working class and
agricultural laborers. In other words, my sample and estimates are likely to capture
the effect of the interruption of the Season and women’s marriages to commoners
and not broader political changes in the nineteenth century.
Finally, I explore the effect of men’s marriages on political power. Here I briey
describe the results; details are in the online Appendix. First, I consider the pos-
sibility that the brothers of women exposed to the interruption were exposed to it
themselves. Online AppendixB9 shows that this does not explain my estimates. I
estimate the effect of a woman’s marriage to a commoner on her brothers’ political
power, limiting the latter to those who married before the interruption. Results are
robust. In fact, the birth year of women and their brothers does not overlap exces-
sively, i.e., their exposure to the interruption is different. Next, online AppendixB10
shows that a man’s marriage to a commoner also reduced his birth family’s political
power and that the effects are qualitatively similar for older and younger brothers.
This is difcult to reconcile with the possibility that my results are driven by a sec-
ular decline in peers’ political power.
52
The 1832 reform reapportioned seats in Parliament favoring cities and industrial areas and abolished most
rotten boroughs. Hence, my measures of political power after a woman’s marriage do not incorporate elections in
rotten boroughs. In fact, family seats in my sample were only 8.27 miles from an enfranchised constituency (see
online Appendix TableA4 and discussion).
478 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
B. Effects on Public Good Provision
So far I have shown that the interruption of the Season increased women’s mar-
riages to commoners, which, in turn, reduced the political power of some peerage
families. Here I evaluate the impact of these changes on public policy.
The introduction of state education in England in the 1870s provides a good test-
bed to evaluate this question. Education provision was decentralized and hence,
subject to capture by peers with local political power (Stephens 1998). Specically,
state schools were built and run with funds from rates—wealth taxes raised by local
school boards in each poor law district and borough. Since taxes were levied on land
property, local elites were expected to pay for most of the new schools. However,
peers galvanized into “a furry of activity to ward to the dread intrusion of a School
Board” (Thompson 1963, 208). They took over school boards and brought taxes
down, especially where they held local political power (Goñi 2021a). Peers with
political power encouraged the election of board members favorable to their inter-
ests. The election system of board members facilitated this: First, because only those
with a rent or land valued at £10 or above could elect board members. Second,
because elections were based on cumulative voting, which can favor interest groups
such as the landed aristocracy (Stephens 1998). Interestingly, women were eligi-
ble to sit on school boards. Yet in 1870, only seven women were elected (Hollis
1987).
53
Another related possibility is that peers with local political power could
effectively lobby elected school board members to set low education taxes.
I test the hypothesis that, by affecting peers’ political power, the interruption of
the Season and women’s marriages to commoners reduced the peerage’s inuence
over school boards. Schematically, this can be summarized as
Season
interruption
⇒
Women’s
marriages to
commoners
⇒
Families’
political
power
⇒
School
boards’ state
education
My identication strategy is an instrumental variables approach that compares
education provision in the vicinity of different peers’ family seats. Specically, I
compare family seats in which a woman married a commoner (and, hence, the fam-
ily lost political power) to family seats in which a woman married in the peerage
(and, hence, the family retained political power).
54
The interruption of the Season
provides exogenous variation in a woman’s probability to marry a commoner and
hence, in her birth family’s political power. To illustrate my strategy, consider
Bineld and Cassiobury, the seats of Baron Kinnaird and Earl Essex (Figure 9).
These seats are only 22.5 miles apart. That is, they are subject to similar local con-
ditions that may affect education provision. In addition, the two seats belong to
families with similar status and past political inuence. Only their marriage patterns
53
Hence, my results cannot be explained by women being/not being elected to a school board.
54
Online AppendixB7 shows that the distance from a woman’s birth-family seat to London is not signicantly
associated to her probability to marry a commoner (correlation 0.058, p-value 0.335) or to the size of the family
landholdings (correlation −0.39, p-value 0.616).
VOL. 14 NO. 3 479
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
differ. Olivia Kinnaird grew up in Bineld. She was 22 when the Season was inter-
rupted and married a commoner. In the following decades, none of her brothers were
elected MP and her father, who was the family head in 1870, did not hold any public
position in England. In contrast, Adela Capel, from Cassiobury, was 33 in 1861,
attended the Season before its interruption, and married an earl. Her father, who
was the family head in 1870, had been a famous supporter of Robert Peel and was
offered the position of Lord Lieutenant of Ireland. In sum, these different marriages
affected the political power of the respective families and hence, their capacity to
undermine state education in the 1870s. On average, school boards in a 10-mile
radius of Bineld, the seat of Olivia Kinnaird, taxed wealth at 4.2 percent. By con-
trast, near Cassiobury, the average tax rate was only 1.5 percent.
Formally, I estimate the relation between women’s marriages to commoners,
their birth families’ political power, and education provision in two steps. First, I
document a reduced-form effect of women’s marriages to commoners on education
provision. To do so, I estimate an IV model where the rst stage takes the form of
equation (8); i.e., I use a woman’s synthetic probability to marry in 1861–1863,
T, as an instrument for her probability to marry a commoner M . The second stage
is
(10) E
i,j,t,s
= β
e
M
ˆ
i,j,t,s
+ 𝐗
i,j,t,s
′ δ
e
+ ν
i,j,t,s
e
,
where i indexes a woman in the baseline sample, j her birth family, t her age in 1861,
and s her birth family seat. E captures the provision of state education. Specically,
E is the average tax rate set by school boards in a ten-mile radius of seat s between
1872 and 1878. The coefcient β
e
captures the reduced-form effect of i ’s marriage
to a commoner on education provision around her family seat s .
Second, I show that this reduced-form effect is explained by the loss of political
power associated with marrying a commoner. Again, the interruption of the Season
is the source of exogenous variation: in the rst stage, I directly use a woman’s
synthetic probability, T , as an instrument for her birth family’s political power, P :
(11) P
i,j,t,s
= η T
t
+ 𝐗
i,j,t,s
′ θ + ξ
i,j,t,s
.
In the second stage, β
E
captures the effect of peerage families’ political power on
education provision:
(12) E
i,j,t,s
= β
E
P
ˆ
i,j,t,s
+ 𝐗
i,j,t,s
′ δ
E
+ ν
i,j,t,s
E
.
All estimates are based on my baseline sample: women aged 15–35 in 1861 who
ever married. As before, I measure the political power of a woman’s birth family
with the MP elections of her brothers. For education provision, I use the reports of
the Committee of Council on Education, digitized by Goñi (2021b). I look at wealth
taxes from 943 school boards located within 10 miles of the family seats in my sam-
ple. Since education was provided locally, the unit of observation in my regressions
is a family seat (and the area around it). That is, I consider the 387 family seats of
women in the baseline sample. Since some families owned more than one seat, the
480 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
number of observations is larger than in the previous section.
55
To address this, I
report p-values clustered by family. The sample is restricted to England because
only there I have education provision data.
Table7 reports the results. First-stage estimates in panel B conrm my previous
ndings: increasing a woman’s synthetic probability to marry during the interrup-
tion increased her probability to marry a commoner (column 1). It also decreased
her brothers’ probability to be elected MP (column 2), to be elected MP in the
family seat’s county (column 4), and how many years her brothers served as MP
(columns 3 and 5) after her marriage. The family heads in the 1870s were similarly
affected (columns 6 and 7). The magnitudes are large. For example, every percent-
age point increase in a woman’s risk to marry in 1861–1863 reduced by 1 pp her
family head’s probability to be elected MP. That said, the F-statistics are low. To
address this, panel A reports CLR p-values adjusted for weak instruments.
56
Panel A reports second-stage estimates. Column 1 documents a reduced-form
effect of peer–commoner intermarriage on education provision. Wealth taxes for
education were 2.66 percentage points higher near the seats of families in which a
55
Of the 279 women in the baseline sample with a family seat in England, 2 had 4 seats, 22 had 3 seats, 58 had
2 seats, and 197 had 1 seat. Hence, the total number of seats is 387.
56
The Lagrange multiplier K and J overidentication tests are consistent with CLR p-values.
F9. I S E P’ F S, E
Notes: This map shows the average tax rate set by school boards near Cassiobury and Bineld in 1872–1878. It also
shows all sampled seats in England and the school boards in a ten-mile radius. Shapeles for England provided by
the Historic Counties Trust, Historic County Borders Project.
Tax rates by school boards
0.4 percent
10.2 percent
Peer’s family seat
Woman married a peer
Woman married a commoner
10-miles radius around seat
County borders
Berkhampstead (2.5%)
Great Gaddesden (2.6%)
Hemel Hampstead (1.4%)
Chesham (0.4%)
Edgeware (0.5%)
Kingsbury (1.8%)
Dorney (2.6%)
Reading (3.4%)
Earley (4.2%)
Windlesham (10.2%)
Cove and Hawley (0.4%)
Binfield, seat of Baron Kinnaird
Cassiobury, seat of Earl Essex
0 5 10 miles
250 50 100 miles
VOL. 14 NO. 3 481
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
woman (exogenously) married a commoner than near seats of families in which a
woman married in the peerage. The effect is economically meaningful. Given that
the average tax in the sampled school boards is only 2.3 percent, the estimated effect
amounts to doubling tax rates.
T 7—D I S E, IV E
Average tax rate for education within ten miles of family seat (percent)
Dependent variable
(1) (2) (3) (4) (5) (6) (7)
Panel A. Second stage
Woman married a commoner 2.66 — — — — — —
(0.007)
[0.035]
Any brother is MP —
−1.69
— — — — —
(0.002)
[0.015]
All brothers’ MP years — —
−0.12
— — — —
(0.033)
[0.063]
Any brother is local MP — — —
−8.20
— — —
(0.023)
[0.049]
All brothers’ local MP years — — — —
−0.27
— —
(0.008)
[0.033]
Family head is MP — — — — —
−1.86
—
(0.017)
[0.055]
Family head years MP — — — — — —
−0.62
(0.043)
[0.12]
Political power in family seat after woman’s marriage
Woman
married a
commoner
Any
brother
is MP
All
brothers’
MP years
Any
brother is
local MP
All
brothers’
local years
Family
head
is MP
Family
head’s
MP years
Panel B. First stage
Treatment
a
0.008
−0.008 −0.170 −0.004 −0.085 −0.009 −0.053
(0.027) (0.001) (0.006) (0.086) (0.018) (0.001) (0.046)
[0.056] [0.003] [0.008] [0.053] [0.008] [0.005] [0.050]
Birth order Yes Yes Ye s Yes Yes Yes Yes
F-stat 4.33 10.0 6.16 10.74 5.85 1.68 3.16
Observations 387 374 374 374 374 387 387
Baseline co. Yes Yes Ye s Yes Yes Yes Yes
County co. Ye s Yes Yes Yes Ye s Yes Yes
Number of brothers No Ye s Yes Ye s Yes No No
Pol. before No Yes Ye s Yes Ye s Yes Yes
Notes: This table reports estimates of equations(8) and (10) (column 1) and equations(11) and (12) (columns2–7).
The sample is women in the baseline sample with a family seat in England. The unit of observation is a family seat
(and the ten miles around it). Some women had several seats, hence, Observations = 387 is larger than before. In
panel A, variables indicate if a woman i living in seat s married a commoner (column 1), if any of her brothers was
elected MP (column 2), elected MP in the family seat’s county (column 4), or if the family head in the 1870s was
elected MP (column 6). Columns 3, 5, and 7 consider the corresponding number of years as MP. Columns 1 to 4
report fewer observations because nine women had no brothers. Baseline and county controls, “Number of broth-
ers,” and “Pol. before” are dened in Table6. Parentheses report p-values (rst stage) and CLR p-values adjusted
for weak IV (second stage). Brackets report CLR p-values adjusted for family clusters.
a
Synthetic probability (percent) to marry during interruption, based on marriage probabilitie in normal times.
482 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
Columns 2 to 7 show that this reduced-form effect is the result of a loss of
political power associated with women’s marriages to commoners. Wealth taxes
for education were lower near the seats where peers had retained political power
than near the seats where they had lost it after a woman’s marriage to a commoner.
Specically, taxes were 1.69 pp lower near seats where any brother was elected MP
(column 2). Every additional year served as MP reduced taxes by 0.12 pp (column
3). These effects are larger when I narrow the focus on local political power (col-
umns 4 and 5): if any brother was elected MP in the family seat’s county, taxes for
education were 8.2 pp lower. Every additional year as local MP reduced taxes by
0.27 pp, twice the effect in column 3. Finally, families who lost political power in
the 1870s—when the policy was implemented—were also less likely to capture
school boards and push taxes down: near the seats in which the 1870s’ family head
was elected MP, education taxes were 1.9 pp lower (columns 6 and 7).
Differences in tax rates across school boards could be explained by local
socioeconomic conditions. To address this, all regressions include county-level
covariates that are potentially correlated with education provision: income per
capita; religiosity; and the percentage of employment in manufacturing, of
nonconformists, and of conservative vote in the 1885 general election. I also con-
trol for the distance between family seats and London—which may be correlated
with attendance to the Season. Online Appendix B11 performs three additional
robustness checks: First, I include xed effects for seats in the same 50-by-50
miles grid cell. That is, I estimate the effects using variation within close-by geo-
graphic areas. Second, I exclude school boards in cities to show that the effects
are not driven by urbanization. Third, I relax the assumption that peerage fami-
lies inuenced school boards in a ten-miles radius around their seats. To do so, I
pair every seat with each of the 1,433 school boards in England. I then dene the
dependent variable as the weighted average tax rate, where weights decay expo-
nentially by the distance between school boards and seats.
So far, I have argued that, after a woman’s marriage to an aristocrat, her birth fam-
ily retained political power and hence, could effectively capture local school boards
and undermine state education. Next, I consider two alternative mechanisms. One
possibility is that, after a woman’s marriage to an aristocrat, the budget constraint
of her birth family was relaxed, allowing them to set lower taxes without affecting
the funds raised for education or the quality of state schools. Online Appendix B11
shows that this was not the case. I nd very similar results when I use the total edu-
cation funds raised from taxes instead of the tax rate: near the family seats in which
a woman married an aristocrat, the average school board not only set lower tax rates
but also raised 1.7 percent fewer funds for education than near seats of families in
which a woman married a commoner. Education funds were also lower where peers
had retained political power than where they had lost it after a woman’s marriage to
a commoner (online Appendix TableB14).
57
Another possibility is that families in which a woman married an industrialist—the
main supporters of state education—became more sympathetic to the husband’s plight
57
Additional evidence from Goñi (2021a) suggests that funds raised from taxes are correlated with county-level
measures of state-education quality (see FigureB5 in the online Appendix).
VOL. 14 NO. 3 483
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
and changed their political preferences. This could affect institutional outcomes
requiring a broad coalition in Parliament.
58
Empirically, this hypothesis is hard to
evaluate, as I do not observe whether the husband was an industrialist or his political
preferences. Alternatively, I can assess the political preferences of husbands in the
peerage. In online AppendixC4, I show that the Season’s interruption did not alter
sorting by political preferences: peers in liberal (conservative) clubs continued to
marry the daughters of liberal (conservative) club members. In other words, it is
unlikely that the interruption altered preferences for redistribution, education provi-
sion, etc., at least among those marrying in the peerage.
59
Altogether, the evidence suggests that the interruption of the Season (exoge-
nously) increased marriages between peers’ daughters and commoners. As a conse-
quence, peerage families lost political power, limiting their ability to take over local
school boards and undermine the introduction of state education in the 1870s.
V. Conclusion
In nineteenth-century Britain, from Easter to August each year, the children of
the nobility engaged in a whirlwind of social events—the Season. From presen-
tations at court to royal parties, the objective was to pull together the right sort of
suitors and to aid in their courtship. This paper shows that by reducing search costs
and segregating “undesirable” suitors, this institution crucially contributed to mar-
ital sorting. To establish causality, I focus on three years during which the Season
was interrupted by the deaths of Queen Victoria’s mother and husband. Women
who were at risk of marriage when the Season was interrupted were 30percent
less likely to marry peers’ heirs and 40 percent more likely to marry commoners.
Within the landed elite, sorting by landholdings decreased by 30 percent, and
women married husbands 44 percentile ranks poorer. These ndings reconcile
the empirical evidence with the theoretical search literature
60
by showing that a
matching technology with low search costs and market segmentation can generate
sorting.
The interruption of the Season halted social interactions within the peerage
temporarily. By affecting marriage decisions, however, it had long-lasting con-
sequences. Using data on elections of Members of Parliament, I show that a
woman’s marriage to a commoner halved her blood relatives’ probability to be
elected MP, especially for local constituencies near the family seat. This, in turn,
affected the introduction of state education by local school boards in the 1870s.
Education investments were twice larger near the domains of families in which
a woman married a commoner (and hence, the family lost political power) than
near the domains of families in which a woman married a peer (and hence, the
58
Similarly, Jha (2015) argues that, in the 1640s, common nancial interests facilitated a broad coalition favor-
ing Parliament supremacy across otherwise different groups.
59
In addition, results are not driven by the late-1870s fall in land values (see SectionI). Land values fell due
to a nationwide fall in grain prices. My identication strategy exploits local variation in education provision within
England and hence, is not affected by nationwide trends.
60
Burdett andColes (1997); Eeckhout (1999); Bloch andRyder (2000); Shimer and Smith (2000); Adachi
(2003); Atakan (2006).
484 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
family retained political power). These ndings have important implications: they
show empirically that, in late nineteenth-century England, marital sorting played
an important role for elite consolidation (Puga andTreer 2014; Marcassa, Pouyet,
andTrégouët 2020), distorted public goods’ provision and the emergence of inclu-
sive institutions.
61
The marriage market embedded in the London Season echoes with some increas-
ingly popular matching technologies. These typically lower search costs but can
also restrict the choice sets for partners through search lters and customized rec-
ommendations.
62
Whether such matching technologies will lead to greater sorting
is an intriguing question for future research.
REFERENCES
Abramitzky, Ran, Adeline Delavande, and Luis Vasconcelos. 2011. “Marrying Up: The Role of Sex
Ratio in Assortative Matching.” American Economic Journal: Applied Economics 3 (3): 124–57.
Acemoglu, Daron. 2008. “Oligarchic Versus Democratic Societies.” Journal of the European Economic
Association 6 (1): 1–44.
Adachi, Hiroyuki. 2003. “A Search Model of Two-Sided Matching under Nontransferable Utility.”
Journal of Economic Theory 113 (2): 182–98.
Aiello, D. 2010. “The London Season.” Last modied November 22, 2010. Accessed May 9, 2011.
http://www.thehistorybox.com.
Allen, Douglas W. 2009. “A Theory of the Pre-modern British Aristocracy.” Explorations in Economic
History 46 (3): 299–313.
Atakan, Alp E. 2006. “Assortative Matching with Explicit Search Costs.” Econometrica 74 (3): 667–80.
Atkins, Peter J. 1990. “The Spatial Conguration of Class Solidarity in London’s West End 1792–
1939.” Urban History 17: 36–65.
Baland, Jean-Marie, and James A. Robinson. 2008. “Land and Power: Theory and Evidence from
Chile.” American Economic Review 98 (5): 1737–65.
Banerjee, Abhijit, Esther Duo, Maitreesh Ghatak, and Jeanne Lafortune. 2013. “Marry for What? Caste
and Mate Selection in Modern India.” American Economic Journal: Microeconomics 5 (2): 33–72.
Bateman, John. 1883. The Great Landowners of Great Britain and Ireland. London, UK: Harrison.
Becker, Gary S. 1973. “A Theory of Marriage: Part I.” Journal of Political Economy 81 (4): 813–46.
Beckett, J.V. 1986. The Aristocracy in England 1660–1914. Oxford, UK: Blackwell.
Belot, Michèle, and Marco Francesconi. 2013. “Dating Preferences and Meeting Opportunities in Mate
Choice Decisions.” Journal of Human Resources 48 (2): 474–508.
Bertocchi, Graziella. 2006. “The Law of Primogeniture and the Transition from Landed Aristocracy to
Industrial Democracy.” Journal of Economic Growth 11 (1): 43–70.
Bloch, Francis, and Harl Ryder. 2000. “Two-Sided Search, Marriages, and Matchmakers.” Interna-
tional Economic Review 41 (1): 93–115.
Botticini, Maristella, and Aloysius Siow. 2011. “Are There Increasing Returns to Scale in Marriage
Markets?” IGIER Working Paper 395.
Bruze, Gustaf. 2011. “Marriage Choices of Movie Stars: Does Spouse’s Education Matter?” Journal
of Human Capital 5 (1): 1–28.
Burdett, Ken, and Melvin G. Coles. 1997. “Marriage and Class.” Quarterly Journal of Economics 112
(1): 141–68.
Burke, John. 1826. A General and Heraldic Dictionary of the Peerage and Baronetage of the United
Kingdom. London, UK: Henry Colburn.
Cameron, A. Colin, Jonah B. Gelbach, and Douglas L. Miller. 2008. “Bootstrap-Based Improvements
for Inference with Clustered Errors.” Review of Economics and Statistics 90 (3): 414–27.
61
Sokoloff andEngerman (2000); Acemoglu (2008); Allen (2009); Galor, Moav, andVollrath (2009).
62
For example, dating websites like “Ivy Date” or “nChooseTwo” (Harvard University, MIT, and Boston
University) aim at matching top university students; Gray and Farrar, an exclusive dating service in Britain, only
admits those who can afford a (cheapest) fee of £15,000.
VOL. 14 NO. 3 485
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
Cancian, Maria, and Deborah Reed. 1998. “Assessing the Effects of Wives’ Earnings on Family
Income Inequality.” Review of Economics and Statistics 80 (1): 73–79.
Cannadine, David. 1990. The Decline and Fall of the British Aristocracy. New Haven, CT: Yale Uni-
versity Press.
Chen, Wen-Hao, Michael Förster, and Ana Llena-Nozal. 2013. “Demographic or Labour Market
Trends: What Determines the Distribution of Household Earnings in OECD Countries?” OECD
Journal: Economic Studies 1: 179–207.
Committee of Council on Education. 1872–1878. Report of the Committee of Council on Education
(England and Wales), with appendix. London: Eyre and Spottiswoode.
Cruz, Cesi, Julien Labonne, and Pablo Querubín. 2017. “Politician Family Networks and Electoral
Outcomes: Evidence from the Philippines.” American Economic Review 107 (10): 3006–37.
Davidoff, Leonore. 1973. The Best Circles: Society, Etiquette, and the Season. London, UK: Croom
Helm.
Doepke, Matthias, and Fabrizio Zilibotti. 2008. “Occupational Choice and the Spirit of Capitalism.”
Quarterly Journal of Economics 123 (2): 747–93.
Eeckhout, Jan. 1999. “Bilateral Search and Vertical Heterogeneity.” International Economic Review
40 (4): 869–87.
Eika, Lasse, Magne Mogstad, and Basit Zafar. 2019. “Educational Assortative Mating and Household
Income Inequality.” Journal of Political Economy 127 (6): 2795–2835.
Ellenberger, Nancy W. 1990. “The Transformation of London ‘Society’ at the End of Victoria’s Reign:
Evidence from the Court Presentations.” Albion: A Quarterly Journal Concerned with British Stud-
ies 22 (4): 633–53.
Ellis, G.A. 1977. “Buckingham Palace.” In Royal Places of England, edited by Robert S. Rait. Lon-
don, UK: Hollings.
Fernández, Raquel, and Richard Rogerson. 2001. “Sorting and Long-Run Inequality.” Quarterly Jour-
nal of Economics 116 (4): 1305–41.
Fernández, Raquel, Nezih Guner, and John Knowles. 2005. “Love and Money: A Theoretical and
Empirical Analysis of Household Sorting and Inequality.” Quarterly Journal of Economics 120
(1): 273–344.
Fisman, Raymond, Sheena S. Iyengar, Emir Kamenica, and Itamar Simonson. 2008. “Racial Prefer-
ences in Dating.” Review of Economic Studies 75 (1): 117–132.
Gale, David, and Lloyd S. Shapley. 1962. “College Admissions and the Stability of Marriage.” Ameri-
can Mathematical Monthly 69 (1): 9–15.
Galor, Oded, Omer Moav, and Dietrich Vollrath. 2009. “Inequality in Landownership, the Emergence
of Human-Capital Promoting Institutions, and the Great Divergence.” Review of Economic Studies
76 (1): 143–79.
Gautier, Pieter A., Michael Svarer, and Coen N. Teulings. 2010. “Marriage and the City: Search Fric-
tions and Sorting of Singles.” Journal of Urban Economics 67 (2): 206–18.
Gobbi, Paula E., and Marc Goñi. 2021. “Childless Aristocrats: Inheritance and the Extensive Margin
of Fertility.” Economic Journal 131 (637): 2089–2118.
Goñi, Marc. 2021a. “Landed Elites and Education Provision in England: Evidence from School
Boards, 1870–99.” Unpublished.
Goñi, Marc. 2021b. “Landed Elites and Education Provision in England. Evidence from School
Boards, 1870–99.” Unpublished data (accessed January 2018).
Goñi, Marc. 2022. “Replication data for: Assortative Matching at the Top of the Distribution: Evidence
from the World’s Most Exclusive Marriage Market.” American Economic Association [publisher],
Inter-university Consortium for Political and Social Research [distributor]. https://doi.org/10.3886/
E140921V1.
Greenwood, Jeremy, Nezih Guner, Georgi Kocharkov, and Cezar Santos. 2014. “Marry Your Like:
Assortative Mating and Income Inequality (Revised Version).” American Economic Review: Papers
and Proceedings 104 (5): 348–53.
Habakkuk, John. 1994. Marriage, Debt, and the Estates System. Oxford, UK: Oxford University Press.
Hechter, M. 1976. “U.K. County Data, 1851–1966.” UK Data Archive Database. http://dx.doi.
org/10.5255/UKDA-SN-430-1.
Historic Counties Trust. n.d. “Historic County Borders Project.” [Data collection]. http://www.county-
borders.co.uk (accessed January
2018).
Hitsch, Gunter J., Ali Hortaçsu, and Dan Ariely. 2010. “Matching and Sorting in Online Dating.”
American Economic Review 100 (1): 130–63.
Hobhouse, Hermione. 1983. Prince Albert: His Life and Work. London, UK: Hamish Hamilton.
486 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2022
Hollingsworth, T.H. 1964. “The Demography of the British Peerage.” Supplement to Population Studies.
Hollingsworth, T.H. 2001. “Hollingsworth Genealogical Data on the British Peerage.” Redigitized
from handwritten original index sheets and archived by the Cambridge Group for the History of
Population and Social Structure (accessed March 2011).
Hollis, Patricia. 1987. Ladies Elect: Women in English Local Government, 1865–1914. New York, NY:
Oxford University Press.
Inwood, Stephen. 1998. A History of London. London, UK: Papermac.
Jacquet, Nicolas L., and Serene Tan. 2007. “On the Segmentation of Markets.” Journal of Political
Economy 115 (4): 639–64.
Jha, Saumitra. 2015. “Financial Asset Holdings and Political Attitudes: Evidence from Revolutionary
England.” Quarterly Journal of Economics 130 (3): 1485–1545.
Kaufmann, Katja Maria, Matthias Messner, and Alex Solis. 2013. “Returns to Elite Higher Education
in the Marriage Market: Evidence from Chile.” Unpublished.
Kremer, Michael. 1997. “How Much Does Sorting Increase Inequality?” Quarterly Journal of Eco-
nomics 112 (1): 115–39.
Laumann, Edward O., John H. Gagnon, Robert T. Michael, and Stuart Michaels. 1994. The Social
Organization of Sexuality: Sexual Practices in the United States. Chicago, IL: University of Chi-
cago Press.
Lee, Soohyung. 2016. “Effect of Online Dating on Assortative Mating: Evidence from South Korea.”
Journal of Applied Econometrics 31 (6): 1120–39.
Low, Corinne. 2017. “A ‘Reproductive Capital’ Model of Marriage Market Matching.” Unpublished.
Marcassa, Stefania, Jérôme Pouyet, and Thomas Trégouët. 2020. “Marriage Strategy among the Euro-
pean Nobility.” Explorations in Economic History 75: 101303.
Mingay, G.E. 1963. English Landed Society in the Eighteenth Century. London, UK: Routledge and
Kegan Paul.
Moreira, Marcelo J. 2003. “A Conditional Likelihood Ratio Test for Structural Models.” Economet-
rica 71 (4): 1027–48.
National Archives. n.d. “Records of the Lord Chamberlain and other ofcers of the Royal Household (LC).”
LC
6/31-55, LC 6/127-156, and LC6/157-164.
Nevill, Dorothy. 1920. The Reminiscences of Lady Dorothy Nevill. London, UK: Nelson and Sons.
Nielsen, Helena Skyt, and Michael Svarer. 2009. “Educational Homogamy: How Much Is Opportuni-
ties?” Journal of Human Resources 44 (4): 1066–86.
Nye, Eric W. 2019. “Pounds Sterling to Dollars: Historical Conversion of Currency.” Accessed August
2019. https://www.uwyo.edu/numimage/currency.htm.
Perkin, Joan. 2002. Women and Marriage in Nineteenth-Century England. London, UK: Routledge.
Piketty, Thomas, and Emmanuel Saez. 2006. “The Evolution of Top Incomes: A Historical and Inter-
national Perspective.” American Economic Review 96 (2): 200–205.
Puga, Diego, and Daniel Treer. 2014. “International Trade and Institutional Change: Medieval Ven-
ice’s Response to Globalization.” Quarterly Journal of Economics 129 (2): 753–821.
Pullar, Philippa. 1978. Gilded Butteries: The Rise and Fall of the London Season. London, UK:
Hamish Hamilton.
Rosenberg, Michael S. 2010. “A Generalized Formula for Converting Chi-Square Tests to Effect Sizes
for Meta-Analysis.” PLoS ONE 5 (4): e10059.
Roth, Alvin E., and Marilda A. Oliveira Sotomayor. 1990. Two-Sided Matching: A Study in Game-The-
oretic Modeling and Analysis. Cambridge, UK: Cambridge University Press.
Rubinstein, W.D. 1977. “Wealth, Elites, and the Class Structure of Modern Britain.” Past and Present
76 (1): 99–126.
Sheppard, F.H.W. 1971. London, 1808–1870: The Infernal Wen. London, UK: Secker and Warburg.
Sheppard, F.H.W. 1977. “The Social Character of the Estate: The London Season in 1841.” In Sur-
vey of London: Volume 39: The Grosvenor Estate in Mayfair, Part 1 (General History), edited by
F.H.W. Sheppard, 89–93. London, UK: Athlone Press for the Greater London Council.
Shimer, Robert, and Lones Smith. 2000. “Assortative Matching and Search.” Econometrica 68 (2):
343–69.
Sokoloff, Kenneth L., and Stanley L. Engerman. 2000. “Institutions, Factor Endowments, and Paths of
Development in the New World.” Journal of Economic Perspectives 14 (3): 217–32.
Stephens, W.B. 1998. Education in Britain, 1750–1914. New York, NY: St. Martin’s Press.
Stone, Lawrence. 1979. The Family, Sex and Marriage in England 1500–1800. Harmondsworth, UK:
Penguin.
VOL. 14 NO. 3 487
GOÑI: ASSORTATIVE MATCHING AT THE TOP OF THE DISTRIBUTION
thePeerage.com. n.d. “A Genealogical Survey of the Peerage of Britain as well as the Royal Families of
Europe.” http://thepeerage.com.
Thompson, F.M.L.
1963. English Landed Society in the Nineteenth Century. London, UK: Routledge
and Kegan Paul.
Ventura, Jaume, and Hans-Joachim Voth. 2015. “Debt into Growth: How Sovereign Debt Accelerated
the First Industrial Revolution.” NBER Working Paper 21280.
Wilkins, Kathryn. 2011. “‘The Most Exclusive Village in the World’: The Utilization of Space by the
Victorian Aristocracy during the London Season.” Urban History Review/Revue d’histoire urbaine
40 (1): 5–16.
Wow
Marc Goni is an associate professor at the Department of Economics in University of Bergen. Among other topics, he is interested in Economic history, marital assortative matching, public economics, and inter-generational inequality.
For estimating causal treatment effects, this paper relies on randomization: the random assignment of women into the mourning period of Queen Victoria.
Instrumental variables (IVs) can be used to estimate causal relationships when controlled experiments are not feasible, as in this case. They are commonly used in economics and epidemiology, among other fields.
More background on instrumental variables here: https://en.wikipedia.org/wiki/Instrumental_variables_estimation
Interesting that sorting patterns in a dating website do not differ from sorting patterns offline!
> A classic prediction is that a matching technology that reduces search costs and limits people’s choice set increases sorting. Surprisingly, these well-accepted theoretical insights lack clear-cut empirical support. Hitsch, Hortaçsu, and Ariely (2010, 162) show that sorting patterns in a dating website “absent of search frictions” differ little from those of people searching offline.
Great summary of the paper: https://www.axios.com/2022/07/02/what-happened-when-the-rich-stopped-intermarrying
Queen Victoria was queen from June 1837 until her death in 1901.
Here is more background on her: https://en.wikipedia.org/wiki/Queen_Victoria
> "Who was considered a suitable match in the Season? Typically, marriages were not love matches but were based on eligibility. According to Davidoff (1973, 50),
“Marriage was not so much an alliance between the sexes as an important social
definition; serious for a man but imperative for a girl. It was part of her duty to enlarge her sphere of influence through marriage.” Women and men aimed to preserve their social status through marriage. That is, peers’ daughters aspired to marry peers’ sons and vice versa. The peerage is divided into five titles: the most desired partners were dukes, marquesses, and earls, followed by viscounts and barons. Next came the gentry (baronets and knights), who were considered commoners. Titleless commoners were the least desirable."
Longer explanation of peerages in the UK: https://en.wikipedia.org/wiki/Peerages_in_the_United_Kingdom
> "Who attended the Season? Since it coincided with Parliament meetings, the Season was attended by the families of members of the two Houses of Parliament. Specifically, by all hereditary peers who sat in the House of Lords (Lords), and by those elected to the House of Commons (MPs). Peerage families without political connections also participated in the Season (Sheppard 1977, 89–93) as well as members of the landed gentry."
To see this quote come to life in a Jane Austen adaption, and see the styles of the times, is something else: https://www.youtube.com/watch?v=Z-k4mKlZ2UY&t=62s
In one season: 50 balls, 60 parties, 30 dinners and 25 breakfasts.
How many potential suitors would meet?
> Elsewhere it has been argued that elites pursue and maintain political power by opposing inclusive institutions,48 by exploiting their economic power (Baland
and Robinson 2008), and through personal ties—i.e., marrying strategically to pre-vent entry by newcomers (Marcassa, Pouyet, and Trégouët 2020). For example, Cruz, Labonne, and Querubín (2017) show that candidates for public office in the Philippines are disproportionately drawn from central families in the marriage net-work. Similarly, Puga and Trefler (2014) find that the wealthiest Venetian merchants used marriage alliances to monopolize the galley trade and to block political com-petition. The Season was no different. Attending it and marrying in the peerage was crucial for political achievement. Lady Aberdeen wrote that the Season “was a part of the very life of the people who had the largest stake in the country (…) Nobody could well come to the front without participating in it to some degree”
Overview of second contribution:
> My second contribution is to show that marriage played an important role in consolidating the peerage as a political elite. To do so, I examine elections of Members of Parliament (MP) at the House of Commons for 27 general elections and 97 by-elections in the late nineteenth century. I show that a woman’s marriage to a commoner reduced her blood relatives’ probability to be elected MP in the following years. Specifically, I estimate an IV model where I instrument a woman’s probability to marry a commoner with her synthetic probability to marry during the Season’s interruption. I find that, after a woman’s marriage to a commoner, her brothers were 50 percent less likely to be elected MP, and, together, they served 18 fewer years than the brothers of women who married in the peerage. The loss of political power was local: mostly constituencies near the family seat were affected. Not only brothers but also the family heads a decade after the interruption (in the 1870s) were affected. I also discuss historical evidence on the mechanisms behind these effects. After a woman’s marriage to a commoner, her birth family had to mobilize considerable capital to sustain her. This limited their ability to control MP elections.
Summary of the paper:
> "In this paper, I use a unique historical setting to isolate the effect of the matching technology on marital sorting and to evaluate some of its political-economy implications in the long run. In the nineteenth century, from Easter to August of each year, a string of social events was held in London to help the peerage’s offspring3 to meet and court—the “London Season.” Courtship in noble circles was largely restricted to London; in most cases, the only place where a young aristocrat could speak with a girl was at a ball during the Season. Guests were carefully selected according to social status, and the high cost involved in participating even excluded peers if they were pressed for money. In economic terms, the Season reduced search costs for partners and restricted the choice set of potential mates. Crucially, the Season was interrupted by a major, unanticipated, exogenous shock: the deaths, in the same year, of Queen Victoria’s mother and husband. As the queen went into mourning, royal balls in the Season were canceled for three consecutive years (1861–1863). During this period, preferences for spouses did not change, but nobles were exposed to larger search costs and lower market segmentation. I use this large shock to identify the effects of the Season on marital sorting and its long-term economic implications. Specifically, I combine archival and published sources to construct two novel data-sets—one measuring attendance to the Season, another of all aristocratic marriages in nineteenth-century Britain. This allows me to evaluate the effects of the Season on marital sorting by title and landed wealth. To gauge the political-economy implications of marital sorting, I link my datasets to the biographies of peers elected to the House of Commons and to local data on state-education provision."
Overview of the first contribution.
> My first contribution is to estimate a strong, plausibly causal link between search frictions and marital sorting. I exploit the interruption of the Season ( 1861–1863)
as a quasi-experiment that raised search costs and reduced market segmentation for some cohorts. Since the decision to marry during the interruption is potentially endogenous, I compare cohorts who had different risks to marry in 1861–1863 based on their age. Specifically, I consider a baseline sample of peers’ daughters aged 15 to 35 in 1861. My treatment variable is a woman’s synthetic probability to marry in 1861–1863, which is based on her age in 1861–1863 and on the probability to marry at each age in “normal times.” For example, the synthetic probability for a woman aged 19–22 in 1861–1863 is equal to the percentage of women who married aged 19–22 in a benchmark cohort married before 1861. My main estimates show that women with a high synthetic probability to marry in 1861–1863 were more likely to marry a commoner and less likely to marry a peer’s heir."
A quasi-experiment is essentially an empirical study that does not use random assignment but can still estimate the causal impact of an intervention on a target population. Endogenous here is a term used in economics to describe a variable that has an internal cause/origin (as opposed to exogenous). It is necessary to understand what he means by synthetic probability above to follow the results later.