-
Notifications
You must be signed in to change notification settings - Fork 0
/
trading-genetics.tex
7524 lines (5822 loc) · 609 KB
/
trading-genetics.tex
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
% Options for packages loaded elsewhere
\PassOptionsToPackage{unicode}{hyperref}
\PassOptionsToPackage{hyphens}{url}
\PassOptionsToPackage{dvipsnames,svgnames,x11names}{xcolor}
%
\documentclass[
12pt,
]{article}
\usepackage{amsmath,amssymb}
\usepackage{iftex}
\ifPDFTeX
\usepackage[T1]{fontenc}
\usepackage[utf8]{inputenc}
\usepackage{textcomp} % provide euro and other symbols
\else % if luatex or xetex
\usepackage{unicode-math} % this also loads fontspec
\defaultfontfeatures{Scale=MatchLowercase}
\defaultfontfeatures[\rmfamily]{Ligatures=TeX,Scale=1}
\fi
\usepackage{lmodern}
\ifPDFTeX\else
% xetex/luatex font selection
\setmainfont[]{Times}
\fi
% Use upquote if available, for straight quotes in verbatim environments
\IfFileExists{upquote.sty}{\usepackage{upquote}}{}
\IfFileExists{microtype.sty}{% use microtype if available
\usepackage[]{microtype}
\UseMicrotypeSet[protrusion]{basicmath} % disable protrusion for tt fonts
}{}
\makeatletter
\@ifundefined{KOMAClassName}{% if non-KOMA class
\IfFileExists{parskip.sty}{%
\usepackage{parskip}
}{% else
\setlength{\parindent}{0pt}
\setlength{\parskip}{6pt plus 2pt minus 1pt}}
}{% if KOMA class
\KOMAoptions{parskip=half}}
\makeatother
\usepackage{xcolor}
\usepackage[margin=1in]{geometry}
\usepackage{longtable,booktabs,array}
\usepackage{calc} % for calculating minipage widths
% Correct order of tables after \paragraph or \subparagraph
\usepackage{etoolbox}
\makeatletter
\patchcmd\longtable{\par}{\if@noskipsec\mbox{}\fi\par}{}{}
\makeatother
% Allow footnotes in longtable head/foot
\IfFileExists{footnotehyper.sty}{\usepackage{footnotehyper}}{\usepackage{footnote}}
\makesavenoteenv{longtable}
\usepackage{graphicx}
\makeatletter
\def\maxwidth{\ifdim\Gin@nat@width>\linewidth\linewidth\else\Gin@nat@width\fi}
\def\maxheight{\ifdim\Gin@nat@height>\textheight\textheight\else\Gin@nat@height\fi}
\makeatother
% Scale images if necessary, so that they will not overflow the page
% margins by default, and it is still possible to overwrite the defaults
% using explicit options in \includegraphics[width, height, ...]{}
\setkeys{Gin}{width=\maxwidth,height=\maxheight,keepaspectratio}
% Set default figure placement to htbp
\makeatletter
\def\fps@figure{htbp}
\makeatother
\setlength{\emergencystretch}{3em} % prevent overfull lines
\providecommand{\tightlist}{%
\setlength{\itemsep}{0pt}\setlength{\parskip}{0pt}}
\setcounter{secnumdepth}{-\maxdimen} % remove section numbering
% definitions for citeproc citations
\NewDocumentCommand\citeproctext{}{}
\NewDocumentCommand\citeproc{mm}{%
\begingroup\def\citeproctext{#2}\cite{#1}\endgroup}
\makeatletter
% allow citations to break across lines
\let\@cite@ofmt\@firstofone
% avoid brackets around text for \cite:
\def\@biblabel#1{}
\def\@cite#1#2{{#1\if@tempswa , #2\fi}}
\makeatother
\newlength{\cslhangindent}
\setlength{\cslhangindent}{1.5em}
\newlength{\csllabelwidth}
\setlength{\csllabelwidth}{3em}
\newenvironment{CSLReferences}[2] % #1 hanging-indent, #2 entry-spacing
{\begin{list}{}{%
\setlength{\itemindent}{0pt}
\setlength{\leftmargin}{0pt}
\setlength{\parsep}{0pt}
% turn on hanging indent if param 1 is 1
\ifodd #1
\setlength{\leftmargin}{\cslhangindent}
\setlength{\itemindent}{-1\cslhangindent}
\fi
% set entry spacing
\setlength{\itemsep}{#2\baselineskip}}}
{\end{list}}
\usepackage{calc}
\newcommand{\CSLBlock}[1]{\hfill\break\parbox[t]{\linewidth}{\strut\ignorespaces#1\strut}}
\newcommand{\CSLLeftMargin}[1]{\parbox[t]{\csllabelwidth}{\strut#1\strut}}
\newcommand{\CSLRightInline}[1]{\parbox[t]{\linewidth - \csllabelwidth}{\strut#1\strut}}
\newcommand{\CSLIndent}[1]{\hspace{\cslhangindent}#1}
\usepackage{subfig}
\captionsetup[subfloat]{labelformat=empty}
\captionsetup[figure]{width=5in}
\usepackage{setspace}\onehalfspacing
\usepackage{amsmath}
\usepackage{amsthm}
\usepackage{placeins}
\usepackage{etoc}
\newtheorem{prop}{Proposition}
\newtheorem{claim}{Claim}
\usepackage{array}
\usepackage{caption}
\usepackage{graphicx}
\usepackage{siunitx}
\usepackage[normalem]{ulem}
\usepackage{colortbl}
\usepackage{multirow}
\usepackage{hhline}
\usepackage{calc}
\usepackage{tabularx}
\usepackage{threeparttable}
\usepackage{wrapfig}
\usepackage{adjustbox}
\usepackage{hyperref}
\ifLuaTeX
\usepackage{selnolig} % disable illegal ligatures
\fi
\usepackage{bookmark}
\IfFileExists{xurl.sty}{\usepackage{xurl}}{} % add URL line breaks if available
\urlstyle{same}
\hypersetup{
pdftitle={Trading social status for genetics in marriage markets: evidence from Great Britain and Norway},
colorlinks=true,
linkcolor={blue},
filecolor={Maroon},
citecolor={Blue},
urlcolor={Blue},
pdfcreator={LaTeX via pandoc}}
\title{Trading social status for genetics in marriage markets:
evidence from Great Britain and Norway}
\author{Abdel Abdellaoui\thanks{Department of Psychiatry, Amsterdam UMC, University
of Amsterdam, Amsterdam, The Netherlands. Email: [email protected]},
Oana Borcan\thanks{School of Economics, University of East Anglia, Norwich,
UK. Email: [email protected]},
Pierre-André Chiappori\thanks{Department of Economics, University of Columbia,
New York. Email: [email protected]},\\
David Hugh-Jones\thanks{Corresponding author. Email: [email protected]},
Fartein Ask Torvik\thanks{Norwegian Institute of Public Health, Oslo.
Email: [email protected]} \&
Eivind Ystrøm\thanks{Norwegian Institute of Public Health, Oslo.
Email: [email protected]}}
\date{2024-06-20}
\usepackage{amsthm}
\newtheorem{theorem}{Theorem}
\newtheorem{lemma}{Lemma}
\newtheorem{corollary}{Corollary}
\newtheorem{proposition}{Proposition}
\newtheorem{conjecture}{Conjecture}
\theoremstyle{definition}
\newtheorem{definition}{Definition}
\theoremstyle{definition}
\newtheorem{example}{Example}
\theoremstyle{definition}
\newtheorem{exercise}{Exercise}
\theoremstyle{definition}
\newtheorem{hypothesis}{Hypothesis}
\theoremstyle{remark}
\newtheorem*{remark}{Remark}
\newtheorem*{solution}{Solution}
\begin{document}
\maketitle
\begin{abstract}
Under social-genetic assortative mating (SGAM), socio-economic status (SES) and
genetically inherited traits are both assets in marriage markets, become
associated in spouse pairs, and are passed together to future generations.
This gives a new explanation for persistent intergenerational inequality and
``genes-SES gradients'' -- observed genetic differences between high- and
low-SES people. We model SGAM and test for it in two large surveys from Great
Britain and Norway. Spouses of earlier-born siblings have genetics predicting
more education. This effect is mediated by individuals' own education and
income. Under SGAM, shocks to SES are reflected in the DNA of subsequent
generations, and the distribution of genetic variants in society is endogenous
to economic institutions.
\par
\textbf{Keywords:} Assortative mating, MoBa, UK Biobank.
\end{abstract}
\normalem
\section{Introduction}\label{introduction}
How families are formed, and transmit traits and assets to their
offspring, is crucial for understanding inequality and social structure.
Assortative mating in marriage markets can increase inequality between
families (Breen and Salazar 2011; Greenwood et al. 2014) and contribute to
its persistence across generations, which is surprisingly high
(Clark and Cummins 2015; Solon 2018). Wealthy families pass on
advantages to their children through both genetic inheritance and
environmental influence (Rimfeld et al. 2018; Björklund, Lindahl, and Plug 2006; Sacerdote 2011).
This paper examines a plausible aspect of marriage markets: both social
status and genetics contribute to a person's attractiveness, and as a
result, they may become associated in subsequent generations.\footnote{\emph{Social status} refers to characteristics that an individual
possesses in virtue of their social position. For example, my wealth
is a fact about me that holds in virtue of my relationship to
certain social institutions (bank deposits, title deeds et cetera).
Other examples include caste, class, income, and educational
qualifications. \emph{Socio-economic status} (SES) is a specific type of
social status which exists in economically stratified societies,
covering variables like educational attainment, occupational class,
income and wealth (e.g. White 1982).} For
example, suppose that wealth, intelligence and health are advantages in
a potential spouse. Then wealthy people are more likely to marry
intelligent or healthy people, and their children will inherit both
wealth, and genetic variants associated with intelligence or health. We
call this mechanism social-genetic assortative mating (SGAM). SGAM may
be an important channel for the transmission of inequality. It creates a
genetic advantage for privileged families, which may help to explain the
long-run persistence of inequality. At the same time, this advantage is
not a fact of biology, but is endogenous to the social structure.
Indeed, under SGAM, environmental shocks to a person's social status may
be reflected in the genetics of his or her children.
Below, we first write down a theory where attractiveness in the marriage
market is a function of both socio-economic status (SES) and genetic
variants. We show that social-genetic assortative mating in one
generation increases the correlation between SES and genetic variants in
the offspring generation. This result provides a new explanation of
\emph{genes-SES gradients} -- systematic genetic differences between high-
and low-SES people (Belsky et al. 2018; Rimfeld et al. 2018; Björklund, Lindahl, and Plug 2006). The dominant existing explanation for these
gradients is meritocratic social mobility: if a genetic variant predicts
success in the labour market, then it will become associated with high
SES and will be inherited in high-SES families. While under meritocracy,
genes causes SES, under SGAM causality goes both ways, from genes to SES
and vice versa. Also, the size of genes-SES gradients depends on
economic institutions. Under institutions which increase
intergenerational mobility, like high inheritance tax rates,
genes-SES gradients become weaker. On the other hand, an increase in meritocracy
can make them stronger. SGAM also interacts with economic institutions to
determine the level of socioeconomic inequality.
Next, using data on spouse pairs from two large genetically-informed
surveys in Great Britain and Norway, we test the hypothesis that a
person's higher social status attracts spouses with genetic variants
predicting greater educational attainment. Our genetic measure, the
polygenic score for educational attainment (PSEA), derives from
large-scale genome-wide association studies (Lee et al. 2018; Okbay et al. 2022). PSEA reflects a bundle of polygenic effects on
underlying traits, including intelligence, personality, and physical and
mental health (Demange et al. 2021). PSEA predicts, and causes,
educational attainment itself, as well as intelligence and labour market
outcomes. It is already known that humans mate assortatively on PSEA
(Hugh-Jones et al. 2016; Robinson et al. 2017; Torvik et al. 2022), which
makes it a likely candidate for detecting SGAM.
The endogeneity of socio-economic status is the main challenge in
identifying the effect of SES on the spouse's genetic endowment. For
instance, people with high educational qualifications tend to also have
high PSEA, and as mentioned above, they may take partners based on
genetic similarity. Indeed, recent studies show strong assortative
mating on PSEA, much more than we would expect if spouses matched only
on observed measures of educational attainment (Okbay et al. 2022). To
isolate the causal link from own SES to partner's genes, we use a shock
to SES which is independent of own genetics. Specifically, we use a
person's \emph{birth order}. Earlier-born children receive higher parental
investment and have better life outcomes, including measures of SES such
as educational attainment and occupational status (Black, Devereux, and Salvanes 2011; Booth and Kee 2009; Lindahl 2008). At the same time, the facts of biology,
in particular the so-called ``lottery of meiosis'', guarantee that
siblings' birth order is independent of their genetic endowments.\footnote{Although Muslimova et al. (2020) find that PSEA and birth order
\emph{interact} to produce human capital.}
Because birth order could affect partner choice through both SES and
non-SES mechanisms, we run a mediation analysis similar to
Heckman, Pinto, and Savelyev (2013), decomposing the treatment effect into effects
of measured and unmeasured mediating variables. Specifically, we
estimate a reduced-form model with spouse polygenic scores for
educational attainment (PSEA) as the dependent variable, and own birth
order as the main independent variable. We then add in to the model
measures of own socio-economic status, including university attendance
and income. Under certain assumptions, these variables can be
interpreted as mediating the effect of birth order on spouse genetics.
In both Great Britain and Norway, later-born children have spouses with
significantly lower PSEA in the reduced-form regressions. When we add
mediators, including university attendance and/or income, the effect of
birth order shrinks substantially, becoming insignificant in Great
Britain, while the SES mediators significantly increase the spouse's
PSEA. The results are robust to the inclusion of several controls,
including non-SES mediators, and a rich set of own genetic traits. Thus,
SES appears to mediate the effect of birth order on spouse genetics. The
effects of individual mediators differ between the two countries. While
university attendance explains more than a third of the effects of birth
order in both Britain and Norway, income explains about 10\% of the
effects in Britain but has little or no independent effect in Norway.
Although our main focus is on testing the basic mechanism of SGAM, this
is suggestive evidence that in a more egalitarian society, some forms of
SES are less important to the marriage market, with long-run
implications for genes-SES gradients.
Both economists and geneticists study assortative mating. The economics
literature has typically focused on educational similarities (e.g. Pencavel 1998; Chiappori, Salanié, and Weiss 2017) or social class or
caste (e.g. Abramitzky, Delavande, and Vasconcelos 2011; Banerjee et al. 2013), but also
sorting based on age, physical traits and ethnicity
(Hitsch, Hortaçsu, and Ariely 2010). Some papers have studied substitution between
different traits.\footnote{Oreffice and Quintana-Domeque (2010) show that height and BMI are associated
with spouse earnings. Dupuy and Galichon (2014) find spouse matching on
multiple independent dimensions, including education, height, BMI
and personality. Chiappori, Ciscato, and Guerriero (2021) analyse matching on
multiple characteristics and show that a three-dimensional matching
model fits their data.} For instance, Chiappori, Oreffice, and Quintana-Domeque (2012) showed that
individuals trade off BMI for partners' income or education.
In genetics, Halsey (1958) showed that social mobility combined with
assortative mating might increase the association between genetics and
social class. Cloninger, Rice, and Reich (1979) model genetic and cultural
transmission, where assortative mating is based directly on phenotype
and culture is transmitted from parents. Assortative mating, modeled
simply as a correlation coefficient, leads culture and genetics to
be associated in offspring. Heath and Eaves (1985), following earlier
papers (Rao, Morton, and Yee 1976; Rao, Morton, and Cloninger 1979), introduce ``social homogamy'',
i.e.~assortative mating by social background. Otto, Christiansen, and Feldman (1995) extend
assortative mating to include both phenotypic and social homogamy.
More recently, interest in these topics has been revived by empirical
findings from genomics. ``Direct'' effects of individual genetic variants, estimated by
within-family studies, are different from ``indirect'' effects, i.e.~associations
found in the whole sample, and direct effects of polygenic scores can be smaller
than population-wide associations (Howe et al. 2022; A. Young et al. 2022).
Also, parental alleles which are \emph{not} transmitted to the child correlate with
child outcomes (Kong et al. 2018). Both these phenomena could be explained
by confounding from gene-environment correlation, or by assortative mating
(A. S. Young 2023; Nivard et al. 2024). Lastly, correlations between spouses'
polygenic scores for education are higher than can be explained by assortative
mating on measured phenotypic education alone (Okbay et al. 2022; Robinson et al. 2017; Torvik et al. 2022). To address this, several recent papers
papers have estimated structural models of assortative mating in family data
(Eaves et al. 1999; Torvik et al. 2022; Collado, Ortuño-Ortı́n, and Stuhler 2023; Rustichini et al. 2023). Because both cultural and genetic inheritance
proceed from parents to children, it can be hard to differentiate
them. For example, Collado, Ortuño-Ortı́n, and Stuhler (2023) derive extremely low estimates
for heritability of education, within a model in which all genetic similarity
between spouses is driven by matching either on the measured phenotype, or
on a shared cultural factor; whereas Torvik et al. (2022) estimate partner
correlation between ``true'' polygenic scores for education of 0.37, and
heritability above 50\%, in a model where environment is shared between siblings
but not across generations.\footnote{As Okbay et al. (2022) put it: ``Because the parameters of a general
biometric model cannot be separately identified from a small number of
phenotypic correlations among different types of relatives, researchers
typically have to assume that some of the parameters equal zero in order to
estimate other parameters.''} In this context, we think it is worth
taking a different approach. We cleanly identify separate environmental and
genetic contributions to assortative mating: environmental contributions using
birth order, genetic contributions by comparing polygenic scores within
siblings.\footnote{See Table \ref{tab:tbl-reversed-moba} in the appendix.}
SGAM has consequences for inequality and social mobility. Long-run estimates of
intergenerational persistence of wealth and status are surprisingly higher than
would be predicted from parent-child correlations (Clark and Cummins 2015; Barone and Mocetti 2021; Solon 2018), and distant relatives in the same
generation are also more similar than parent-child and spousal correlations
would predict (Collado, Ortuño-Ortı́n, and Stuhler 2023). Clark (2023) argues that this
can be explained by an underlying process where unobserved genetic variation
determines wealth. This requires a high degree of assortative mating. Our model
shows that genetics may itself be a mediator for the transmission of SES, via
``trading'' in marriage markets. We also show how different social and economic
institutions can affect that process. When SES is highly transmissible across
generations, this increases the long-run association between SES and genetics.
If so, institutional reforms that increase \emph{intergenerational mobility}, like
mass education or inheritance taxation, may affect not only economic but genetic
inequality. Conversely, an increase in \emph{economic meritocracy} increases the
long-run association between SES and genetics,\footnote{See Proposition \ref{prop-gamma}.} posing the problem raised by
M. Young (1958) and more recently Markovits (2019): meritocracy may be
self-limiting or even self-undermining.
In terms of cross-sectional inequality, the conventional wisdom is that it is
increased by assortative mating on SES (Fernández and Rogerson 2001; Breen and Salazar 2011; Greenwood et al. 2014). But that depends what else people
assort on. As we show below\footnote{See Figure \ref{fig:pic-heritability-inequality}.}, if the same genes that are relevant in marriage
markets also affect economic outcomes, then an increase in the role of genes
vis-à-vis SES in marriage markets may increase economic inequality: it makes
households more unequal in genetics, and these are passed on to their children
with high reliability.
SGAM can also explain a large body of evidence for cross-sectional
associations between genetics and social status. For example: from twin
studies, the heritability of occupational class and educational
attainment, i.e.~the proportion of variance explained by genetic
differences between individuals, is around 50\% (Tambs et al. 1989).
Genome-wide Complex Trait Analysis (GCTA) shows that the family
socio-economic status of 2-year-old children can be predicted from their
genes (Trzaskowski et al. 2014). Children born into higher-income families
have more genetic variants predicting educational attainment
(Belsky et al. 2018). Adoption studies show that both post-birth
environment and pre-birth conditions (genetics and prenatal environment)
contribute to the transmission of wealth and human capital (e.g. Björklund, Lindahl, and Plug 2006). There is also a genes-SES gradient in genetic
predictors of health. DNA-derived scores predicting several health
outcomes are associated with regional economic deprivation
(Abdellaoui et al. 2019). The correlation between education and health
may be mediated by shared genetic causes (Amin, Behrman, and Kohler 2015; Boardman, Domingue, and Daw 2015). Family SES correlates with several health-related
polygenic scores (Selzam et al. 2019), and genetic variants associated
with SES may explain the genetic correlations between many mental health
outcomes (Marees et al. 2021).
SGAM shows how marriage markets can lead high SES to be associated with
different genetic variants, i.e.~it can explain genes-SES gradients. The
standard explanation for these gradients is returns to human capital in
labour markets, also known as meritocratic mobility. Higher-ability
parents reap higher market returns, and they may then pass both higher
socio-economic status and their genes to their children, leading to an
association between the two (Belsky et al. 2018).\footnote{Belsky et al. (2018) offer three reasons for the association between
education-linked genetic variants and SES, but do not consider SGAM.} This mechanism depends
on the level of meritocracy in social institutions
(Branigan, McCallum, and Freese 2013; Heath et al. 1985): in a society where social status
was ascribed rather than earned, it could not take effect. Indeed, after
the fall of communism in Estonia, the heritability of SES increased,
presumably because post-communist society allowed higher returns to
talent (Rimfeld et al. 2018). By contrast, SGAM does not require meritocracy.
Even when social status is entirely ascribed, it can still become
associated with certain genetic variants, so long as their associated
phenotypes are prized assets in marriage markets. Since meritocracy is
historically rare, while assortative mating is universal, this suggests
that genes-SES gradients are likely to be historically widespread.
Lastly, we contribute to a literature in economics that examines the
relationship between genetic and economic variables.
Benjamin et al. (2011) and Benjamin et al. (2024) are reviews. Several
recent papers use polygenic scores, in particular polygenic scores for
educational attainment (e.g. Barth, Papageorge, and Thom 2020; Papageorge and Thom 2020; Ronda et al. 2020). Barban et al. (2021) use PSEA as an instrument for
education in a marital matching model. These papers, like much of the
behavior genetics literature, take genetic endowments as exogenous and
examine how they affect individual outcomes, perhaps in interaction with
the environment. We take a different approach by putting genetics on the
left hand side of the estimating equation. Assortative mating and cultural
inheritance are social processes, so we think there are good prospects for
social scientists to contribute to understanding how genetic variants get
distributed in society -- what geneticists call ``stratification'' and ``dynastic
effects''.
The observations behind SGAM are not new. That status and physical
attractiveness assort in marriage markets is a commonplace and a
perennial theme of literature. In the Iliad, powerful leaders fight over
the beautiful slave-girl Bryseis. In Jane Austen's novels, wealth,
attractiveness and ``virtue'' all make a good match. Marx (1844)
wrote ``the effect of ugliness, its repelling power, is destroyed by
money.'' The literature on mate preference from evolutionary psychology
(Buss and Barnes 1986; Buss 1989; Buss and Schmitt 2019) confirms that
attractive mate characteristics include aspects of social status (``high
earning capacity,'' ``professional status'') as well as traits that are
partly under genetic influence (``intelligent,'' ``tall,'' ``kind,''
``physically attractive''). Despite this, to our knowledge, few papers have examined
the socio-economic consequences of assortative
mating between SES and genetics.\footnote{Specifically, Halsey (1958) and Rustichini et al. (2023).} In particular, we are the first to show
how SGAM interacts with institutional variables to affect economic inequality,
mobility and associations between genes and SES, and the first to cleanly
identify an environmental effect on spouse genetics.
\section{Model}\label{model}
People in the marriage market have two characteristics:
\(x=\left( x_{1},x_{2}\right)\), drawn from a normal distribution
\[
\mathcal{N}
\left(
\begin{array}{c}
0 \\
0%
\end{array}%
,%
\begin{array}{cc}
s^{2} & \sigma \\
\sigma & S^{2}%
\end{array}%
\right).
\]
We interpret \(x_1\) as a genetic measure, for example of genetic variants predicting
height, physical attractiveness, health or intelligence. \(x_2\) is a measure of
socio-economic status, such as income or wealth, or social status more generally
(we sometimes use ``wealth'' as a shorthand). The correlation between
\(x_1\) and \(x_2\) is
\[
Corr = \frac{\sigma }{sS} < 1.
\]
People's attractiveness is given by
\[
i\left( x\right) =ax_{1}+\left( 1-a\right) x_{2}
\]
where \(a \in [0, 1]\) is a parameter reflecting the relative importance of
genetics to wealth in the marriage market.\footnote{Note that since the variance of the shocks to \(x_1\) and \(x_2\)
(see below) has been normalized to 1, \(a\) also reflects this variance. That
is, a large variance of SES shocks (compared to genetic shocks) translates into
\(a\) being large.} If \(a = 0\),
marriage markets are highly inegalitarian, such that only SES matters. If \(a =
1\), marriage markets are economically egalitarian and only genetics matter. We
expect realistic societies to fall between these extremes, with \(0 < a < 1\).
Then, both genes and SES matter to attractiveness, and as a result,
social-genetic assortative mating (SGAM) takes place.\footnote{This model does not assume that people match \emph{directly} on
genetics, which most observers agree would be unlikely. Instead we assume that
genetics may contribute to an attractive phenotype which is matched on.}
Attractiveness \(i\) is
distributed \(N(0,\sigma_{I}^{2})\), where
\[
\sigma _{I}^{2}=a^{2}s^{2}+\left( 1-a\right) ^{2}S^{2}+2a\left( 1-a\right)\sigma.
\]
People form matches with transferable utility, where the surplus for a match
between \(x\) and \(y\) is \(S(i(x), i(y))\) such that \(\partial^{2}S/\partial
i\partial j > 0\), i.e.~\(S\) is supermodular. As a result there is positive
assortative mating on attractiveness: \(x\) matches with \(y\) only if they are at
the same quantile of attractiveness, i.e.~if \(i(x_{1},x_{2}) = i(y_{1},y_{2})\).
Within attractiveness quantiles, matching is random. This is the
SGAM mechanism.
We also consider random matching as a benchmark to compare against SGAM. Under
random matching, the distribution of couples' characteristics is normal with
mean 0 and covariance matrix
\[
\mathbb{C}\left(
\begin{array}{c}
x_{1} \\
x_{2} \\
y_{1} \\
y_{2}%
\end{array}%
\right) =\allowbreak \left(
\begin{array}{cccc}
s^{2} & \sigma & 0 & 0 \\
\sigma & S^{2} & 0 & 0 \\
0 & 0 & s^{2} & \sigma \\
0 & 0 & \sigma & S^{2}%
\end{array}%
\right). \allowbreak
\]
Our first proposition shows that if SGAM is taking place, i.e.~if \(0 < a < 1\), then
there is a positive correlation between one partner's wealth and the other
partner's genetics.
\begin{proposition}\label{prop-couples-SGAM}
Under SGAM, the distribution of couples' characteristics is normal, with mean 0
and covariance matrix
\begin{equation}\label{cov-couples-SGAM}
\mathbb{C}\left(
\begin{array}{c}
x_{1} \\
x_{2} \\
y_{1} \\
y_{2}%
\end{array}%
\right) =\allowbreak \left(
\begin{array}{cccc}
s^{2} & \sigma & A^{2} & AC \\
\sigma & S^{2} & AC & C^{2} \\
A^{2} & AC & s^{2} & \sigma \\
AC & C^{2} & \sigma & S^{2}%
\end{array}%
\right) \allowbreak
\end{equation}
where:
\begin{align*}
A &= \frac{as^{2}+\left( 1-a\right) \sigma }{\sqrt{a^{2}s^{2}+\left(
1-a\right) ^{2}S^{2}+2a\left( 1-a\right) \sigma }} &= \frac{as^{2}+\left( 1-a\right) \sigma }{\sigma_I}; \\
C &= \frac{a\sigma +\left( 1-a\right) S^{2}}{\sqrt{a^{2}s^{2}+\left(
1-a\right) ^{2}S^{2}+2a\left( 1-a\right) \sigma }} &= \frac{a\sigma +\left( 1-a\right) S^{2}}{\sigma_I}.
\end{align*}
In particular, the covariance between $x_2$ and $y_1$, $AC$, is positive if
either $x_1$ and $x_2$ are already correlated ($\sigma > 0$) or if they are
uncorrelated ($\sigma = 0$) and the attractiveness parameter $a$ is strictly
between 0 and 1.
\begin{proof}
See Appendix.
\end{proof}
\end{proposition}
We consider the distribution of couples' wealth. Under random matching this has
mean \(0\) and variance \(2S^2\). Under SGAM, the variance is:
\[
V(x_{2}+y_{2}) = 2S^{2} + 2C^{2} \ge 2S^{2}
\]
with strict inequality if \(a < 1\) or \(\sigma > 0\). The variance is decreasing in
\(a\) and equals \(4S^2\) if \(a = 0\). Thus, SGAM increases cross-sectional inequality,
but less so than pure matching on wealth.
\subsection{Children}\label{children}
All couples have the same number of children. A child's characteristics are
given by:
\begin{eqnarray}
x_{1}^{\prime } &=&\frac{\tau }{2}\left( x_{1}+y_{1}\right) +\varepsilon
\label{Chil} \\
x_{2}^{\prime } &=&\frac{\theta }{2}\left( x_{2}+y_{2}\right) +\eta
\nonumber
\end{eqnarray}
where \(x\) and \(y\) are the child's parents, and \(\varepsilon\) and \(\eta\) are
independent normal random shocks with mean \(0\) and variance \(1\).
Parameter \(\tau \approx 1\) reflects genetic inheritance. Under standard biological
assumptions \(\tau = 1\) and characteristics show no regression to the mean. In our
model this leads the variance of \(x_1\) to grow without limit over generations.
In reality, we expect \(\tau < 1\) because very extreme characteristics are
selected against, a process known as stabilizing selection
(Schmalhausen 1949; Sanjak et al. 2018).
Parameter \(\theta \in [0, 1]\) reflects inheritance of SES. Unlike \(\tau\) it may vary
between societies. \(\theta\) is high when there is high intergenerational
transmission of SES. Thus, \(\theta\) captures social and economic institutions
that affect this intergenerational transmission, from taxation and
public education to hereditary nobility. If we interpret \(x_2\) narrowly as
wealth, \(1 - \theta\) can be thought of as the rate of inheritance tax.
For the time being, we assume that a person's genetic endowment has no impact on
their SES. Technically, thus, \(x_{2}^\prime\) does not directly depend on
\(x_{1}^\prime\). In a meritocratic society we would expect adult SES to partly depend
on genetics. We show that even absent meritocracy, correlations between \(x_1^\prime\) and
\(x_{2}^\prime\) can arise. In an extension below, we relax this assumption and allow
meritocracy.
We can now calculate the covariance matrix for
\(x^\prime = (x_{1}^{\prime }, x_{2}^{\prime })\) under SGAM as:
\begin{eqnarray}
\mathbb{C} &=&\left(
\begin{array}{cc}
1 & 0 \\
0 & 1%
\end{array}%
\right) +\left(
\begin{array}{cccc}
\frac{\tau }{2} & 0 & \frac{\tau }{2} & 0 \\
0 & \frac{\theta }{2} & 0 & \frac{\theta }{2}%
\end{array}%
\right) \allowbreak \left(
\begin{array}{cccc}
s^{2} & \sigma & A^{2} & AC \\
\sigma & S^{2} & AC & C^{2} \\
A^{2} & AC & s^{2} & \sigma \\
AC & C^{2} & \sigma & S^{2}%
\end{array}%
\right) \allowbreak \left(
\begin{array}{cc}
\frac{1}{2}\tau & 0 \\
0 & \frac{1}{2}\theta \\
\frac{1}{2}\tau & 0 \\
0 & \frac{1}{2}\theta
\end{array}%
\right) \nonumber \\
&=&\left(
\begin{array}{cc}
\frac{1}{2}A^{2}\tau ^{2}+\frac{1}{2}s^{2}\tau ^{2}+1 & \frac{1}{2}\theta
\sigma \tau +\frac{1}{2}AC\theta \tau \\
\frac{1}{2}\theta \sigma \tau +\frac{1}{2}AC\theta \tau & \frac{1}{2}%
C^{2}\theta ^{2}+\frac{1}{2}S^{2}\theta ^{2}+1%
\end{array}%
\right) \allowbreak \label{cov-children-SGAM}
\end{eqnarray}
\newline
We now explore two issues. First, under SGAM, genetic characteristics are no
longer exogenous; because of assortative matching, they are (partly) socially
determined. In particular, even if genetics and SES are uncorrelated among
parents, the expected genetic endowment of the child is positively related to
parental SES. Second, as a result, in the long run a correlation appears between
traits; that is, high SES people inherit genes that are attractive in marriage
markets.
Regarding point 1, we compute the expected genetic characteristic of the
child, conditional on parental wealth:
\[
\mathbb{E}\left[ \frac{\tau }{2}\left( x_{1}+y_{1}\right) +\varepsilon \mid
x_{2}=v,y_{2}=w\right]
\]
Given the symmetry of the model, this conditional
expectation only depends on the parents' total wealth, i.e.~\(v+w\).
\begin{claim}\label{claim-E-children-RM}
Under random matching, the expected genetic endowment of the children is
proportional to the parents' SES and to the covariance between SES and genetics
for the parents. In particular, if $\sigma = 0$ (i.e. genetics and SES are
uncorrelated for the parents), then the expected genetic endowment of the
children does not depend on parental SES.
\end{claim}
\begin{claim}\label{claim-E-children-SGAM}
Under SGAM, if $\sigma = 0$ (i.e. genetics and
SES are uncorrelated for the parents), then the expected genetic endowment of
the children is linearly increasing in parental SES. The relationship
increases with the ratio of genetic variance to SES variance, is zero for
$a = 0$ or $a = 1$, and is highest for intermediate values of $a$.
\end{claim}
Next, we study the correlation between children's traits 1 and 2 as a function
of \(\sigma\), the covariance of parents' traits. We first consider the
general case, then concentrate on \(\sigma = 0\), i.e.~when traits are initially
uncorrelated.
\begin{claim}\label{claim-corr-children-RM}
Under random matching, the correlation between characteristics is smaller
for children than for parents. In particular, if genetics and SES are
uncorrelated for the parents, then they are uncorrelated for the children.
\end{claim}
\begin{claim}\label{claim-corr-children-SGAM}
Under SGAM, if genetics and SES are uncorrelated for the parents, then they
are positively correlated for the children so long as $0 < a < 1$. The
correlation is increasing in $\theta$.
\end{claim}
Whether characteristics are more or less correlated for children than for
parents depends on whether the initial correlation between parents'
characteristics is larger or smaller than the asymptotic one, derived below.
Figure \ref{fig:pic-intuition} shows the intuition behind the model. Parents
match on downward-sloping attractiveness isoquants given by \(a x_1 + (1-a) x_2 = u\).
Their children are in between them on both dimensions. This compresses the
distribution along the attractiveness isoquants, which leads to a
positive correlation between genetics and SES. The correlation between \(x^\prime_1\)
and \(x^\prime_2\) is 0 when \(a = 0\) or \(a = 1\), because then spouses don't trade off
SES for genes. It is highest for intermediate values of \(a\).
\begin{figure}
{\centering \subfloat[Parents\label{fig:pic-intuition-1}]{\includegraphics{trading-genetics_files/figure-latex/pic-intuition-1} }\subfloat[Children\label{fig:pic-intuition-2}]{\includegraphics{trading-genetics_files/figure-latex/pic-intuition-2} }
}
\caption{Theory. The shaded area is the population distribution. Parents (solid circles) match along attractiveness isoquants (dotted lines). Children (hollow circles) are between them. As a result, the children's distribution is squeezed along attractiveness isoquants, and $x_1$ and $x_2$ become associated.}\label{fig:pic-intuition}
\end{figure}
These results show that SGAM can lead to a genes-SES gradient, i.e.~a positive
correlation between genes and SES. Also, the strength of the genes-SES
correlation is affected by economic institutions, as captured in \(\theta\).
When \(\theta\) is high, the genes-SES correlation is high too.
We now consider the asymptotic distribution of \(x_1\) and \(x_2\) when the matching
process is repeated over many generations. As we would expect, our main results
continue to hold.
\begin{proposition}\label{prop-asymptotics-RM}
Under random matching, the dynamics converges to a stationary distribution that
is normal with mean zero and covariance matrix
\[
\mathbb{C}\left(\begin{array}{cc}
x_1 \\
x_2
\end{array}
\right)
=
\left(
\begin{array}{cc}
\frac{2}{2-\tau ^{2}} & 0 \\
0 & \frac{2}{2-\theta ^{2}}%
\end{array}%
\right) \allowbreak
\]%
In particular, the traits are asymptotically uncorrelated and children's
expected genetic endowment is independent of parents' wealth.
\end{proposition}
\begin{proposition}\label{prop-asymptotics-SGAM}
Under SGAM, for $\theta <1$ and $\tau <1$, the dynamics converge to a stationary
distribution that is normal with mean zero and covariance matrix
\[
\mathbb{C}\left(\begin{array}{cc}
x_1 \\
x_2
\end{array}
\right)
=
\left(
\begin{array}{cc}
\bar{s}^{2} & \bar{\sigma} \\
\bar{\sigma} & \bar{S}^{2}%
\end{array}%
\right)
\]
Moreover, the asymptotic correlation between characteristics,
$corr = \bar{\sigma}/\bar{s}\bar{S}$, is non-negative, positive for $0 < a < 1$,
increasing in $\theta$ and increasing then decreasing in $a$.
The coefficient of parents' wealth on children's genetics is also positive
for $0 < a < 1$.
For $\theta = 1$, the dynamics diverge and $\bar{S}^{2}$ goes to $+\infty$; for
$\tau = 1$, the dynamics diverges and $\bar{s}^{2}$ goes to $+\infty$.
\end{proposition}
\begin{figure}
{\centering \includegraphics{trading-genetics_files/figure-latex/pic-asymptotic-corr-1-rgl}
}
\caption{Long-run correlation between genetics $x_1$ and SES $x_2$, by weight of genetics in spouse matching ($a$) and strength of inheritance of SES ($\theta$). $\tau$ = 0.95.}\label{fig:pic-asymptotic-corr}
\end{figure}
Figure \ref{fig:pic-asymptotic-corr} plots the asymptotic squared correlation
between \(x_1\) and \(x_2\). In genetic terms, this is the heritability of SES. It is
maximized for intermediate levels of \(a\). Note that both \(\bar{S}^{2}\) and
\(\bar{\sigma}\), as well as the correlation between characteristics and the
conditional expectation of genetics given wealth, are increasing in \(\theta\).
Higher transmission of SES increases inequality of SES in equilibrium
(unsurprisingly), but also the heritability of SES.
\subsection{Extensions}\label{extensions}
We consider three extensions. First, the relative attractiveness of genes and
SES might differ for men and women. Our basic result extends to this setup.
\begin{claim}\label{claim-men-women-different}
Suppose that men's and women's attractiveness is given by
\begin{align*}
i(x) &= ax_1 + (1-a)x_2, \\
j(y) &= by_1 + (1-b)y_2
\end{align*}
respectively, with $0\le a \le 1$, $0 \le b \le 1$. Then if $\sigma = 0$, children's
characteristics $x^\prime_1$ and $x^\prime_2$ will be positively correlated
unless $a = b = 0$ or $a = b = 1$. The correlation is increasing in $\theta$.
\end{claim}
Interestingly, the \(x_1\)-\(x_2\) correlation is highest when \(a\) and \(b\) are most
different from each other. So gender differences in what counts as attractive
make the effects of SGAM stronger. Intuitively, if one sex only assorts on
SES while the other sex only assorts on genetics, this induces a very reliable
correlation between genes and SES in couples, since (e.g.) every high-SES male
is matched for sure with a high-genetics female.
Second, in modern meritocracies, people's adult SES depends not just
on their parents' social status and on chance, but also on their own effort and
skills, which might be related to their genetics. So, let
\begin{align}
x^\prime_1 &= \tau \frac{x_{1} + y_{1}}{2} + \varepsilon \nonumber \\
x^\prime_2 &= \gamma x^\prime_1 + \theta \frac{x_{2}+y_{2}}{2}+\eta \label{eqn-gamma}
\end{align}
where \(\gamma > 0\) represents the effect of own genetics on own SES. \(\gamma\)
can be thought of as measuring meritocratic mobility, but alternatively as the
degree of overlap between genes that matter in marriage markets and in labour
markets. The basic result continues to hold, and also, \(\gamma\)
increases the correlation between genes and SES; a highly meritocratic
society may in the long run lead to a steep genes-SES gradient.
\begin{proposition}\label{prop-gamma}
Under SGAM and equation \eqref{eqn-gamma}, if genetics and SES are
uncorrelated for the parents, then they
are positively correlated for the children so long as $0 < a < 1$ or
$\gamma > 0$. The correlation is increasing in $\gamma$. Also, so
long as $\gamma > 0$ and either $0 < a < 1$ or $\sigma > 0$, the coefficient
of parents' wealth on children's wealth exceeds $\theta$.
\end{proposition}
Figure \ref{fig:pic-heritability-inequality} plots computed equilibrium values
of two interesting outcomes: economic inequality, i.e.~the standard deviation of
\(x_2\), and heritability of SES, i.e.~the squared correlation between \(x_1\) and \(x_2\).
Meritocratic mobility \(\gamma\), which opens a pathway from highly transmissible
genetics to SES, always increases both inequality and heritability
(cf. Rimfeld et al. 2018). The effect of \(a\) is more complex. At low levels of
meritocracy, increasing \(a\) decreases inequality by reducing assortative mating
by SES, and intermediate values of \(a\) increase heritability by assorting
SES with genetics. As meritocratic mobility increases, it acts as a multiplier
for the effects of \(a\) on both outcomes, and \(a\) starts to increase both
heritability and inequality, for almost all values. Again, this is because
\(a\) increases assortation by highly transmissible genetics which in turn
affect SES. In particular, consider a broad social ``modernization'', in which
the direct transmission of SES and the role of SES in matching decrease, while
meritocratic mobility increases since careers become open to talents. If the
decrease in \(\theta\) does not outweigh the increase in \(a\) and \(\gamma\), both
equilibrium heritability and inequality of SES may increase.
\begin{figure}
{\centering \includegraphics[width=1\linewidth]{trading-genetics_files/figure-latex/pic-heritability-inequality-1}
}
\caption{Equilibrium inequality (standard deviation of $x_2$) and heritability (squared correlation of $x_1$ and $x_2$) for different values of assortative mating $a$, meritocratic mobility $\gamma$ and transmission of SES $\theta$. $\tau = 0.95$. The x axis is scaled differently between the left and right plots.}\label{fig:pic-heritability-inequality}
\end{figure}
Third, we consider non-normal distributions of \(x_1\) and \(x_2\), non-normal
shocks \(\varepsilon\) and \(\eta\), and non-linear attractiveness functions. Suppose
\begin{equation}\label{non-linear-f}
i(x) = f(ax_1, (1-a)x_2)
\end{equation}
with \(f\) strictly increasing in both its arguments. Our sole condition on the
distribution of \(x\) is that not everybody with attractiveness \(i\) is both
genetically and socially identical. In particular, this allows for discrete
distributions, like some kinds of social status for \(x_2\), and monogenic or
oligogenic attractiveness phenotypes for \(x_1\).
\begin{proposition}\label{prop-non-normal}
Let attractiveness be given by \eqref{non-linear-f}. Let $(x_1, x_2)$ have
any distribution such that a positive measure of the population has
$i(x) = i$ where the conditional distribution of $(x_1, x_2)$ given $i(x) = i$ is
non-degenerate. Let $\eta$ and $\varepsilon$ be mean 0 and independent of
$x$ and each other. If genetics and SES are uncorrelated for the parents, then
the correlation among children is non-negative, and strictly positive if
$0 < a < 1$.
\end{proposition}
Other extensions are possible. We assumed that all couples have the same number
of children. If fertility increased with \(x_1\) or \(x_2\), we would expect this to
reduce the variance of traits in the children's generation and possibly also
their covariance. Here, matching preferences, summarized by the \(a\) parameter,
are exogenous; it would be natural to model \(a\) as an equilibrium outcome.
For example, if parents care about their children's wealth, \(a\) might decrease
in \(\theta\) and increase in \(\gamma\). Indeed, below we see suggestive evidence that
income is valued differently in Norwegian and British marriage markets.
\subsection{Discussion}\label{discussion}
The meanings of both social status, and ``good genes'' in the
marriage market, are likely to vary across societies. Social status could
encompass variables like social class or caste; ethnic identity in
``ranked'' ethnic systems; or in modern societies, SES, including wealth, income
and occupation. Regarding genetics, standards of physical attractiveness, and
other genetically-influenced characteristics which make someone a ``good match'',
vary across societies and over time. The central prediction of the model is that
whatever those characteristics, in the long run they will become correlated with
SES.
Recent empirical work shows high persistence of SES over time, in particular at
the top. Clark (2023) argues that this could be explained by unobserved
genetic variation.
Proposition \ref{prop-gamma} shows that if genes affect own wealth directly,
under assortative mating, the regression coefficient of parents' wealth on own
wealth exceeds the ``direct'' coefficient \(\theta\), because parents' wealth
correlates with parents' genetics and via that with own wealth. Thus,
regressions of wealth on wealth may include the effect of unobserved genetic
variation. This may be a confound due to pre-existing gene-SES correlation (if
\(\sigma > 0\)). But under SGAM (\(0 < a < 1\)) it can also be a mediating
variable, since changes in someone's wealth may indeed affect the identity of
their spouse, hence the genetics of their offspring, and from that their
offspring's adult wealth.
The converse also holds: regressions of children's characteristics on their
genetics alone risk overestimating the effect of genetics, by confounding it
with the effects of correlated socio-economic status. Recent work in genetics
has shown this. Polygenic scores for educational attainment have smaller effects
in between-sibling regressions, where between-family variation in SES is
partialled out and where genetic variants are guaranteed to be randomly
allocated, than in regressions which pool the whole sample (Howe et al. 2022).
Parents' genetic variants which are \emph{not} passed on to children predict
children's characteristics, partly due to social stratification in the geneticists'
sense of non-random mating (Kong et al. 2018; A. S. Young 2023).
The model predicts variation in the effects of SGAM. In particular, in
``caste societies'' where there is complete endogamy within social status
groups, there is no scope for SGAM, because marriage partners do not trade off
genetics for social status (\(a = 0\)). Also, the association between genes and
SES is increased by the institutional variable \(\theta\), which captures
intergenerational persistence of SES. This implies that policy has long-run
effects on biosocial structure: reducing \(\theta\) not only increases
intergenerational mobility, but reduces the correlation of genes with SES, and
hence the bias of what Harden (2021) calls the ``genetic lottery''.
Conversely, reforms that increase meritocracy (\(\gamma\), Proposition
\ref{prop-gamma}) may strengthen the genes-SES gradient.
\section{Data and methods}\label{data-and-methods}
The central insight in our model is that higher SES and good genes
assort in the marriage market. We wish to test this directly, i.e.~to
test whether \(0 < a < 1\) in the attractiveness equation \[
i(x) = a x_1 + (1-a) x_2
\] where \(x_2\) is social status and \(x_1\) is genetic endowment. Consider
the effect of a change in \(x_2\) holding \(x_1\) constant. If \(a = 1\) then
this will not change \(i(x)\) and therefore will not change the expected
characteristics of the spouse. So, if we regress spouse's \(x_1\) on own
\(x_2\), and reject the null of no effect, we can reject \(a = 1\).\footnote{Conceivably, if \(a = 0\) but there is a pre-existing correlation
between \(x_1\) and \(x_2\) in the population, then an increase in own
\(x_2\) will increase spouse's expected \(x_2\) and therefore spouse's
expected \(x_1\), even though the latter does not enter the
attractiveness equation. We can separately test the null that
\(a = 0\) by regressing spouse's \(x_2\) on own \(x_1\), holding own \(x_2\)
constant. Existing work has already linked own genetics to spouse's
SES, e.g.~education, so we focus on the other direction and treat
this direction as a robustness check below.}
We use data from two sources: Great Britain and Norway. This allows us
to check our basic result in two different societies, and also to make
(tentative) comparisons between them. Our Great Britain data comes from
the UK Biobank, a study of about 500,000 individuals born between 1935
and 1970 (Bycroft et al. 2018). The Biobank contains information on
respondents' genetics, derived from DNA microarrays, along with
questionnaire data on health and social outcomes. The Biobank does not
contain explicit information on spouse pairs. We categorize respondents
as pairs if they had the same home postcode on at least one
occasion;\footnote{A typical UK postcode contains about 15 properties.} both reported the same homeownership/renting status,
length of time at the address, and number of children; attended the same
UK Biobank assessment center on the same day; both reported living with
their spouse (``husband, wife or partner''); and consisted of one male and
one female. We also eliminate all pairs where either spouse appeared
more than once in the data. This leaves a total of
35,682 pairs.\footnote{In the appendix, we test the validity of our matching process by
counting the proportion of pairs who had a shared genetic child, in
a subsample of the data. We also check whether any misidentified
pairs might have biased our results, by constructing a dataset of
``known fake pairs''.}