See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/16076285

Theory and Observation in Cultural Transmission

Article  in  Science · November 1982

DOI: 10.1126/science.7123211 · Source: PubMed

CITATIONS READS

283 901

4 authors, including:

Marcus W Feldman
Stanford University
766 PUBLICATIONS   40,637 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Marcus W Feldman on 05 December 2016.

The user has requested enhancement of the downloaded file.

18. J. H. Hudson, E. A. Shinn, R. B. Halley, B.
Lidz, Geology 4, 361 (1976).
19. E. M. Dniffel, Radiocarbon 22, 363 (1980).
20. G,eophys. Res. Lett. 8, 59 (1981).
21. J. F. Toggweiler, Eos 61, 984 (1980).
22. W. S. Moore and S. Krishnaswami, in Proceed-
ings of the Second International Coral Reef
Symposium (Great Barrier Reef Commission,
Brisbane, 1974), vol. 2, pp. 269-276.
23. C. Emiliani, J. H. Hudson, E. A. Shinn, R. Y. 30.
George, Science 2d2, 627 (1978).
24. R. G. Fairbanks and R. E. Dodge, Geochim. 31.
Cosmochim. Acta 43, 1009 (1979).
25. R. B. Dunbar and G. M. Wellington, Nature 32.
(London) 293, 453 (1981).
26. 1. G. MacIntyre, Atoll Res. Bull. 185, 21 (1975).
27. J. H. Hudson, personal commulnication; R. W.
Buddemeier, personal communication. 33.
28. E. M. Druffel, thesis, University of California,
San Diego (1980).
29. The carbon and oxygen isotopic ratios are de- 34.
fined as follows:
813C = 35.
13C/12Csample - 1C/' Cstandard
36.
'3C/ ndsta2Cnard 37.
relative to the Pee Dee belemnite standard
(PDB-1).
8'80 =
1°/16°sanAe - °/ °standard X 1000
80/t60standard
relative to the standard mean ocean water
(SMOW).
H. Stommel, Proc. Natl. Acad. Sci. U.S.A. 76,
3051 (1979).
W. G. Mook and E. M. Druffel, unpublished
data.
H. E. Suess, in Proceedings of the Conference
on Nuclear Processes in. Geological Settings
(Univ. of Chicago Press, Chicago, 1953), pp. 52-
56.
M. Stuiver and P. D. Quay, Earth. Planet. Sci.
Lett. 53, 349 (1981); W. F. Cain and H. E.
Suess, J. Geophys. Rks. 81, 3688 (1976).
P. P. Tans, thesis, Rijksuniversiteit en Groning-
en, Netherlands (1978).
H. Oeschger, U. Seigenthaler, U. Schlotterer,
A. Gugelmann, Tellus 27, 168 (1975).
H. E. Suess, personal communication.
R. B. Dunbar, G. M. Wellington, P. W. Glynn,
E. M. Druffel, Geol. Soc. Am. 93rdAnnu. Meet.
Abstr. 12, 417 (1980).
38. I thank J. H. Hudson and E. A. Shinn for
providing this unique coral core. I am grateful to
W. G. Mook for allowing me to publish part of
his stable isotope data for this coral core. The
technical expertise of S. Griffin, P. Yeager, D.
Sullivan, C. Hutto, and T. Linick is greatly
appreciated. I thank Drs. W. Mook, P. Wil-
liams, J. Reid, H. Oeschger, P. Brewer, W.
Deuser, L. Keigwin, B..Michel, S. Rodriguez,
B. Jenkins, P. Rhines, H. Suess, and H. Stom-
mel for enlightening discussions. I also thank D.
Beriy, M. Harvey, and T. Velasquez for their
help in preparing this manuscript. This article
consists in part of a doctoral dissertation carried
out under the direction of Professor Hans E.
Suess and submitted to the Depattment of
Chemistry, University of California, San Diego,
in March 1980. Financial support from the Na-
tional Science Foundation through grants EAR-
78-15183 and OCE-79-17652 and from the
Woods Hole Oceanographic Industries Program
4470 is gratefully acknowledged. This article is
contribution 5047 from the Woods Hole Oceano-
graphic Institution.

technological innovations. Our concern

here is not with the comparison of muta-

Downloaded from www.sciencemag.org on November 10, 2014

tion and selection in biological and cul-
tural situations, but with another ingredi-
ent in the process of evolution-trans-
Theory and Observation in migsion. Although well studied and
quantified in biology, transmission is
Cultural Transmission poorly understood in its cultural context.
The study of quantitative aspects of cul-
-L. L. Cavalli-Sforza, M. W. Feldman tural transmission can, we believe, cre-
ate a foundation for the study of cultural
H. Cheh, S. M. Dornbusch evolution and, in the quantitative theo-
retical development upon which we have
embarked, modeling of cultural trans-
mission has a central place (2). To date
The activities, values, and behavior of selection has produced the complexity quantitative studies of cultural transmis-
an individual that are acquired through and diversity of living systems is the sion have been limited, although there
instruction or imitation will be termed cornerstone of interpretation in the bio- already exist theories, such as mathe-
"cultural." Such phenomena are not ex- logical sciences. Observed genetic varia- matical epidenliology (5), which could
clusively human (1) but are tmost highly tion is the result of interactions between augment the study of diffusion of innova-
tiorts (6). In this article we suggest some
of the possible applications of our gener-
Summary. Cultural phenomena may show considerable stability over time and al theory, in an empirical investigation of
space. Transmission mechanisms responsible for their maintenance are worthy of quantitative aspects of our general the-
theoretical and empirical inquiry; they are complex and each possible pathway has ory.
different effects on evolutionary stability of traits, as can be shown theoretically. A
survey designed to evaluate the importance of some components of cultural
transmission on a variety of traits showed that religion and politics are mostly Models of Transmission
determined in the family, a mode of transmission which guarantees high evolutionary
stability and maintenance of high variation between and within groups. Cultural transmission is the process of
acqiuisition of behaviors, attitudes, or
technologies through imprinting, condi-
developed in our species. In attempting the rules of genetic transmission, muta- tioning, imitation, active teaching and
to construct a quantitative theory for the tion, natural selection, and sampling, learning, or combinations of these. A
evolution of cultural traits we have found due to the finiteness of natural popula- quantitative theory of the evolutioh of a
many concepts from the quantitative the- tions. Each of these phenomena can, in culturally transmitted trait requires mod-
ory of biological evolution to be useful principle, be measured, and together eling who transmits what to whom, the
(2). It has oftenl been suggested (3), they allow statistical prediction of the number of transmitters per receiver,
though not widely appreciated, that the evolutionary trajectories of the geno-
evolution of cultural phenomena can be types in the population. L. L. Cavalhi-Sforza is a professor in the Depart-
viewed in a conceptual framework simi- The cultural analog of mutation in- ment of Genetics, Stanford University Medical
School, Stanford, California 94305; M. W. Feldman
lar to that of biotogical evolution, but so cludes innovation as well as random is a professor in the Department of Biological Sci-
far most analyses have been purely qual- change in the expression of traits (2). In ences, Stanford University; K. H. Chen is a profes-
sor in the Department of Mathematics, California
itative. fact, Galton (4), in explaining biological State University, Long Beach 90840; and S. M.
That the continuing process of evolu- mnutations ("sports" in domesticated Dornbusch is Reed-Hodgson Professor of Human
Biology in the Department of Sociology, Stanford
tion by random mutation and natural plants and animals) compared them to University.
SCIENCE, VOL. 218, 1 OCTOBER 1982 0036-8075/82/1001-0019$01 .00/0 Copyright © 1982 AAAS 19
their ages, and other relations between
them.
As a first step in measuring cultural
transmission it is natural to consider
discrete valued traits and the parent-
offspring relationship. When the trans-
mission is from parent to child it is
termed vertical, in agreement with usage
in epidemiology, and the natural discrete
time unit is the generation. We use hori-
zontal transmission to mean transmis-
sion between members of the same gen-
eration, and oblique for transmission
from nonparental individuals of the pa-
rental generation to members of the filial
generation. In what follows horizontal
transmission is considered in the restrict-
ed context of age peers, Transmission
from, for example, teachers or mass me-
dia has been modeled in our theoretical
studies (2).
Just as in population genetics, ecolo-
gy, and epidemiology, the dynamical
studies of these models can point to
situations in which heterogeneity (poly.
morphism) might reasonably be expect-
ed to be maintained within a population.
In addition, the use of models allows
comparison of rates of evolutionary
change in terms of mode of transmission.
For example, oblique transmission
through teachers increases homogeneity
within a population and creates greater
variation between populations in space
and time than does vertical transmission
(2, 7).
Some broad principles have emerged
from our theoretical work and can be
applied to situations in which a fixed
level of cultural mutation is present with
respect to one specified cultural trait. If
the number of transmitters per recipient
is many-to-one, the rate of evolution
(change of trait frequency with time) is
slow, and variation within and between
populations is low. An example is
oblique transmission through social class
or caste stratification in which many
transmitters, potentially even the whole
group, apply the same cultural pressure
on each of the individuals in the next
generations. This is a conservative mode
of cultural transmission. If, as in the case
of a teacher, transmission is one-to-
many, cultural change is expected to be
rapid and within-population variation
low. The intermediate situations of one-
to-one or one-to-few transmission (as in
parent-tp-child or -children transmis-
sion) produce moderate rates of change
and relatively high within- and between-
population heterogeneity (Fig. 1). A spe-
cific trait may, of course, be transmitted
in more than one way by mechanisms
whose actions may not always be in
concert.
20
Table 1. Agreement of repeated observations
on the same individual: correlation between
ratings of an individual by different observers
(including self) in the six major categories of
traits (listed in Table 5). Values are the means
and raiges of correlations averaged among all
possible pairs of observers.
Correlation (r)
Category
Average Range
Religion .72 .42 to .87
Sports .58 .48 to .83
Politics .50 .22 to .77
Entertainment .35 .18 to .57
Habits .32 .12 to .51
Beliefs .17 .05 to .37
Such generalizations about rates of
cultural evolution also depend on the
time units involved. For parent-child
transmission the biological generation is
a natural time unit, but for the transmis-
sion among age peers the time unit is
shorter and the rate of change is corre-
spondingly increased. Alternatively, cul-
tural change is retarded when grandpar-
ents, elders, or oral and written tradi-
tions play key roles in the determination
of trait values (2, 7).
Survey Approach to Cultural
Transmission
We summarize an analysis of data
whose collection was suggested by the
theory of cultural transmission and
which may prove to be more generally
useful in describing the importance of
the family in cultural transmission.
There have been many studies of the
influence of parents on children in rela-
tion to political preferences (8), attitudes
toward authoritarianism (9), religiosity
(10), and other attitudes, beliefs, and
behaviors. But a number of these studies
have been criticized (11) for mode of
sample collection, poor assessment of
reliability, and lack of emphasis on group
correspondence.
For this study we distributed a ques-
tionnaire to Stanford University under-
graduates, mostly sophomores and ju-
niors who were enrolled in courses re-
quired either for the biology or human
biology major. The questionnaire was
designed to sample a variety of cultural
traits in six categories: (i) religious affili-
atiotn and attitudes, (ii) political affili-
ation and attitudes, (iii) entertainment,
(iv) beliefs about some contentious is-
sues (including superstitions), (v) habits
that are easily described without inva-
sion of privacy, and (vi) sports. Some
questions required answers of yes or no;
others required assignment to positions
on a scale (for example, a frequency
scale with answers such as more than
once a day, once a day, once a week, and
so on) and were later dichotomized (12).
Students answered the same questions
in the six categories about themselves,
their fathers, mothers, and one male and
one female friend (13). In addition, stu-
dents reported their sex, occupation of
parents, and an estimate of parental in-
come. Students also received question-
naires to mail or hand to parents and the
friends about whom they had answered
along with a letter explaining the project.
Parents and friends were asked to an-
swer the same questions about them-
selves and the students that the students
had answered. All questionnaires were
linked numerically to preserve anonym-
ity.
Only sets of questionnaires returned
by a student and both parents (N = 203)
or by a student and both friends (N = 98)
were analyzed. About 53 percent of pa-
rental pairs returned the questionnaires;
the return from both friends was 40 per-
cent. Preservation of anonymity preclud-
ed sampling of nonrespondents. These
response rates are moderately high when
compared to similar studies (8-11).
Agreement of Ratings
Each student was evaluated by either
three individuals (self and two parents or
self and two friends) or five individuals
(self, two parents, and two friends) (13).
Parents and friends were rated by them-
selves and by the student. There were
therefore at least two evaluations for
each participant. The correlation be-
tween ratings of the same person by
different people (one of whom was self)
was used as a measure of agreement
between observers.
Correlation coefficients (r) indicating
agreement for a given trait were aver-
aged after grouping the traits into the six
major categories (Table 1). Moreover,
for each trait many agreement values
were calculated and averaged, as ex-
plained below. The variation between
the correlations averaged is expressed
by their ranges. Agreements spanned
almost the whole range between zero
and one (Table 1). The ranges of agree-
ment for the various groups of traits
(which are listed individually in Table 5)
show that agreement decreased in the
following order: religion > sports > poli-
tics > entertainment > habits > beliefs
(Table 2).
There were a number of correlations,
all indicating agreement between differ-
ent Qbservers rating the same individual
SCIENCE, VOL. 218
for a given trait. Parents were rated by politics > sports/entertainment > beliefs .81 for father and mother, but only .38
themselves and their children (not by the > habits; and (iii) friend-friend, religion for friend and friend). These correlations
other parent, because of possible paren- > sports/politics > entertainment/beliefs suggest that because of the low agree-
tal separation). Thus there is a correla- > habits (15). Similarities and agree- ment for certain traits the true similarity
tion coefficient for father rated by self ments correlate highly, at least in family may be higher than that estimated by the
and father rated by child, and another groups (r = .76 for parent and offspring, raw correlation.
similar correlation for mother. For child
rated by father (or mother) and by self,
there are three possible correlations. For Ratio of number One
to few
of transmitters to Many to one One to many
each of the two friends (of different number of receivers:
(few
tco few)
sexes) there were ratings by the student
and by self, generating two other correla- tical Horizontal Social
Example: Social class or Vert

tions (one for each friend). Altogether, caste influences (parent tto child) hierarchies Teacher/student
Social leaders
there were seven correlation coefficients Mass media
measuring agreement of ratings of the Rate of
same person by different observers, two cultural change: Lowest Highest
for the rating of parents, three for stu- Population Little!acceptance of Persistence of variation Rapid flux; between
dents, and two for student's friends. heterogeneity: varianits; between and (stable polymorphic pDopulation heterogeneity
within p)opulation hetero- equilibrium not uncommon); high; within population
There is a trend in the agreements gen eities are low between and within heterogeneity low
according to the relationship of observer population heterogeneities
are high
and observed (not shown in Table 1). Fig. 1. Modes and rates of cultural transmission.
Agreement between ratings of the same
person by different people (including
self) was highest when the subject ob- Table 2. Means and ranges of correlation coefficients by category for various pairs; self-ratings
served was one of the parents (r = .45, only were used.
with a large standard deviation between
traits, ±.23), less when students were Correlation (r) between responses of
scored by parents or by themselves Category Father-mother Parent-child Friend-friend
(r = .41, +.20), and least when students Aver-
and friends were rating each other age
Range Aver- age
Aver-
age
Range
(r = .33, + .18). In part the differences in
agreement may be due to the different Religion .69 .42 to .91 .57 .33 to .82 .20 .10 to .34
Politics .48 .37 to .57 .32 .09 to .52 .16 .10 to .27
periods of life of the observers. In part Sports .34 .16 to .54 .13 .04 to .20 .16 .0 to .34
they may be due to error, but biases, Entertainment .30 .08 to .58 .16 .02 to .29 .10 -.02 to .24
observer-observed interactions, and de- Habits .17 .04 to .47 .07 -.02 to .21 .05 -.25 to .22
ficiencies of the survey instrument can- Beliefs .11 .11 to .36 .09 .06 to .12 .12 .03 to .21
not be excluded. Reliability, as indicated General average .35 .22 .13
by the coefficients of agreement, affects
the confidence to be placed in the results
discussed below, especially for certain Table 3. Means of disattenuated similarities by category for various paired responses (16).
traits.
Correlation (r)
Responses Etr
Enter- e
liBef-s
Religion Sports Politics Habits
Correlation Analysis
Father-mother .83 .49 .79 .74 .49 .99
Our primary interest is the degree of Parent-child .71 .22 .61 .44 .24 .49
similarity between parent and child, be- Friend-friend .31 .33 .49 .30 .26 .68
tween parents, and between friends.
Thus, observations were first analyzed
by standard correlational methods. Self-
Table 4. Additive model of vertical transmission. The bi's are proportions of type H individuals
ratings gave values for the target correla- in progeny of the four types of parental pairs. In the last three columns is the numerical example
tions that were generally smaller than of Democratic party affiliation.
those computed from observers other
than self (14), but the differences were Example:
Additive model Democratic party
not usually statistically significant. We affiliation Ex-
were conservative and used self-ratings
in evaluating these relationships. Num- Ob- pected
Fre- Trans- ber served off
Table 2 summarizes the correlations Fa- Mo- quency mission Expected of H spring
that express overall similarities of inter- ther ther of coeffi- bv
of paren- off-
est in the six categories. The order of pair cient tal spring
correlations for traits averaged for each pairs (%)
category is similar to that for agreement; h h Po bo Po 99 25 27
for the three sets of observations the h H Pi b1 Po + aM 27 74 68
order is: (i) parent-child, religion > poli- H h P2 b2 PO1+F 14 64 51
tics > entertainment > sports > beliefs H H P3 b3 O + aF + OM 55 91 92
> habits; (ii) father-mother, religion > *For 13o = .27, AM = .41 and aF = .24, estimated by maximum likelihood; goodness of fit: X2(l) = .72, P > .20.
I OCTOBER 1982 21
 I.'- . .
42A
C)
c
. . . . .
tN00
. .
( .
*
0 ~ N m~ 0 0 o *
_-
-4
mt%0 e --fi
o* C
6
o C
,
,o

.0

rse
00 oc^
r
0o -oA
04 oo C qt
00> en
00 '
4 000 0%-- 0% 0 en 0 %O 0%ON-
r-"i
%O o oo o en
cl i
N-*
1- - -* "n
'U
-^
*
I * *

CO 0 I=0 X
41+
coQ
4)4) '0)
C .0_.
0 0 600
_ ) vNo^o c- m Q en t- I0
l
m co _
a-, en _.
_
_.
- _It No o o
o
)
crv
%n
f WI,^F
.4
. .
4
. .
.4 .el .

t
N 0 O
000 e -

0lje N..'

I
000 0 - 00 N 0%
eq
00
-4
WI) N-
O. - .
. .
Cs
0~-
._

4-

4 ) U1.
u
'27 0% 00 0 N
an
lqt en
C O 2
00 0
el -o-.
o =:
4.2. 0O -' el _U
01.
.
_ .
C .
en .
Lo .
e . .
N. .
- . .
- .

I
u
oNot
CU
a._ U.
-en mr en m
'UN N00 e4 - %0
O (_ U,-,
0
e'4
Mg4 N . .
04) . .

4cD V- 00 Cq e-
-OO i>t-
Q 4 - 't 0
- (A
CU
LC I
C4
en l-
00 C: 4.)

co Cs
(04
e 0.
.o a06 Q. I''U 0'IT )-N
0t O.- -
4- 00 0 00
(01-0 0 -(040(0r4 enl-' 0- 'elON -)
.00 I
S.
._
L. I
r- elf W) WI 00r oW
4) U N
N- W0 0O0 00
eO O01 t en
*' £ U.
U'.
el eel (0 0le rle
._
e~0 0
en en e
GO)

0.'

It 0 0 4)~~~~~~~4
O %D W %c W-
oo
°0cn E 04
00 -4 I0
CU
)~x % 00 (04 (04 eql enl
0% 00 00

0% 00 "it _O
m x 000 %N-O NO-0
-N'
O -O N0% 0el
U, e'a- 0t
en " C4^ 000% N rlO- 0e- NO00 _ eeO
X Xx en 00 -0
*f

oo
Cq
(0q ; rel N > 00_ N Cf0
= (0D4 0000%N -e 1'U0 ON
00N
(0400
,
a% WIe- e
co 00 U) s s0 0
0%N Wel el '* 00 'Uef)(0
(04
0
.
O7 CD oo _ co -%0%0 N:F- %0~0 %Orf)00 NI
4 C _ .4e' %0 f)000 0lOU so N
0

'el 00 0 ee N
-O O - %0
0000 00N (0

-
x
xX
Xs
N 'O 0 eN
04) SI
xx N
- Go
'00.
O-(0 N
oo en ^
CU '2F 4)
be X (0 %0^
' 4) o 00 It
.0=
=
t '0 "'D-
O _ 00
O

0. >>

I ~~~~-
~ ~ ~ ~ ~ ~ ~ ~

0 .'0 iU- _40'

>aMm t > S ; °-C tog--
X 0
j
4)0C
4)>.0~ ' 2 i ~ ~ '2o .4)444) 4) 4)4) 0 04
0~~~~~~~~~~~~~~~~

C,4 8
22E
c
rn
i 04 W~L8
04 P9 i: W ,4 CAw 06 U}C,6 A4
iLz J U,x
zJ
2 9C

..
22 * - SC~~~~~~~IENCE,YQ. 2-1l
 W
eC 0i Disattenuation (16) of correlations spring generation the proportion of H is
o u
*0 e.'1~ substantially increases many of the simi- u'= IpAb (i from 0 to 3) which may be
Cv'c larities (Table 3) and the order of impor- different from u = p3 + (PI + P2)/2, the
, U
X
tance is changed to the following: (i) frequency in the parental generation. If
Cs0N 00 o -r-
.cUCU' parent-offspring, religion/politics > be- bi is constant over generations, the tra-
Qo_ -
liefs > entertainment > sports/habits; (ii) jectories of the frequency u, of H over
father-mother, beliefs > religion/politics/ time can be given once the pairing rule pi
entertainment > sports/habits; and (iii) is specified. If parental pairs are random,
Q°04
friend-friend, belief > politics > sports Po = (1 - u)2, PI = P2 = u(l - u), and
- 00c
e1C tN4
tI O
- "o e)
-
R oa > religion/entertainment/habits. The nia- p3 = U2, or if the deviation from randon-
.
".4 .4 .
l.~ . . . .

jor change is for beliefs (including super- ness can be expressed by the simple
00
stitions), which thus show more similar- correlation between parents, rFM=
ity between parents and children than (PP3 - P1P2)/U(l - u), and this as well
00CD0 00 00Q-
raw correlations would indicate. Howev- as the b1's is assumed to be constant over
. *. .
. . . . .
CU
CO k +
a) er, the standard errors of disattenuated time, then prediction of the evolutionary
Co C e
coefficients (not shown) are much higher result is especially easy (2). Certain sets
than those of the raw ones. We will not of transmission coefficients have special
tI00o 0 Wm o It 0.
a
00 0o0-00-
. . .
. . . . . .
use disattenuation in the analysis that qualitative interest. For instance, a sim-
o_ C
follows, but it is clear that results show- ple interpretation of independence of the
ing substantial effect of disattenuation contributions from, mother and father
00 00T 0 e X00 00 0 might be considerably strengthened if (Table 4) is the additive model, and it fits
0- tI 00
O - 0O - .0
..
. . . . . . .
studied by other designs with smaller the example of Democratic party affili-
observational error. ation reasonably well. Since this additive
. . t .
I. . . .
_~ 0
model fits the data well, there is no
tI rI 0g
evidence of interaction between contri-
Analysis of Cultural Transmission butions from mother (aM) and father (aF)
4 - 0 .0 CU to vertical transmission.
. . . .
I. . . .

C)_ )
Our theory of cultural transmission (2, The superposition of other modes of
-r
d _
;^C
CC
* 7) can be used to relate familial correla- transmission on vertical can affect the
tI0 D
S
_ _

CC3 05
tions to transmission parameters. Corre- dynamics of the trait in the population.
00
. . .
r-
-. lations, however, are symmetric and do For example, suppose that after vertical
not indicate the direction or mode of transmission and before maturity the fre-
oCo-0^C transmission. Only a specially designed quency of H in the progeny is uc, I, and
OL ._CU U longitudinal survey which included all that later there is oblique transmission
lt r- r- _ cv "T' (t2,
'-3 Co protagonists in the transmission process from H ip the parental generation to h
C) CoG could define in more detail transmitters, progeny, which converts the latter at the
en 00 0 1- t W C) t-
receivers, and their actions and interac- rate f to H. Then the progeny frequency
,'
_ t d X CCQ tion. of H at maturity is u,+, = uC +I +fu,(l
Our survey had less ambitious goals - uC,+I), where u,+1 is analogous to u'
-N
W)V)
tf
00 r- -* o r- -r-
C41 CA C ) C#) 't 00C ) 0
and did not include longitudinal analysis used earlier. If, instead of oblique trans-
(17). Our study is only a first step in mission, there is horizontal transmission
relating our theory to real data and, in from Hi in the progeny to h in the prog-
_ ~ _

'IT 'It o1- -C N ° '.-_ ; CUo

t1 00 'I r- - rn0 W)
'; V particular, in distinguishing relative con- eny, then ut+I = uc,+1 +fuc,+I
CE 2
a)CCCUm
tributions of the various modes of trans- (1 - uc,+1). If Po > 0 (Table 4), trans-
F
W lt
O
W
00 t (ON X m
0 00 t1 00 u
,_

Co w <a
_
mission: vertical, horizontal, and oblique. mission to the offspring may have oc-
As mentioned above, such distinctions curred from nonparental sources, possi-
are important for understanding rates of bly as a result of horizontal and oblique
change of trait frequencies over time as transmission, and there might also have
OC e_,C well as the variation to be expected been innovation by the offspring (2). In
O X,0 _ W
00
toesl m WI
W) r
00
00 'It between and within populations. this way Po represents a maximum to the
*£ ' CC3 Consider first vertical transmission. possible contributions of oblique and
Our data are based on triads-father, horizontal transmission and can be com-
Cd mother, and one child. The following is a pared with aF + aM, a measure of the
simple model of a trait which exists in importance of vertical transmission. In
X
00>.Y
CUY
only two states, H and h, the first of the example (Table 4), vertical transmis-
which is the result of specific irreversible sion appears to be more important than
learning and the second is the naive oblique and horizontal together, since
condition. To describe vertical transmis- Po = .27 is less than aF + aM = .65. In
4)-C
sion we define pi, the relative frequency order to assess the relative role of verti-
i) -
_ ort (J
.
of each of four possible types of parental cal transmission we evaluate an index
.F o
= o
_v
.

C-- pairs, h x h, h x H, H x h, H x H, I - Pa/u', termed an index of vertical

C~.U and bi, the transmission coefficient or transmission (IVi), and note that, under
WCC) U).-
_ vO
0.C
probability (assumed for simplicity to be an additive model, it is 0 when vertical
cd .
VO
C! _ laf equal for the two sexes) that a mature transmission is absent and 1 when verti-
*d C=
offspring H is produced from the respec- cal transtnission is complete. At evolu-
:= mmmmmmm * cn IL
tive parental pairs (Table 4). In the off- tionary equilibrium IVT =OaF + aM un-
I 0OCTOBER 1982 23
Table 6. Analysis of goodness of fit of additive and log-linear models (based on tests in Table 5)
not distinguish from our data whether
for the 41 traits. The probabilities corresponding to each x2 are tested for agreement with the
such differences (of which this may be an
rectangtilar distribution expected for a perfect fit. The observed numbers of traits were entered
extreme example) are age-related or re-
in the corresponding classes of x2 probabilities for the additive and the log-linear model as were
flect waves of cultural change. Most
the expected numbers of traits for a perfect fit of the two models. Comparison of expected and
other traits (Table 5) seem, in terms
observed numbers of traits shows excellent agreetnent both for the additive and the log-linear
models; x2 with 5 degrees of freedom is 2.46 and 5.72 for additive and log-linear models,
either of IVT or the direct comparison of
respectively. u' and u, to be close to equilibrium.
Number of traits The sharp differences noted between
Probability the strong vertical transnmission ob-
of X2 Observed Observed served for religion and politics, and that
(additive) (log-linear) Expected for the other classes of traits, demand an
.9to 1.0 2 1 4.1 explanation. We cannot exclude errors
.7to .9 8 9 8.2 of observatioh, as discussed above, but
.Sto .7 7 12 8.2 it is possible that the traits other than
.3 to .5 11 6 8.2 religion and politics are less stable during
.I to .3 9 9 8.2
0 to .1 4 4 4.1 individual development. The result
would be lower correlations between
pairs of individuals, as with error of
measurement. Moreover,. it is natural to
der the additive model (2), A limitation ments of correlation (table 2). In gener- expect that traits which exhibit the.lower
of the data in regard to interpretation of al, the estimated values for aF and aM correlations have less well defined, and
Po is that parents and offspring are evalu- are similar to but slightly greater on perhaps later, critical periods during
ated at different ages. The ideal assess- average than the conventional partial which individuals are particularly sus-
ment of the two generations at similar correlation coefficients between father ceptible to external, rather than to paren-
ages presents obvious practical difficul- and child given mother, and mother and tal, influences.
ties. Other potential sources of oblique child given father, respectively (Table 5, A finer analysis of the nature of the
and horizontal transmission will be con- rFC.M and rMc.F). Raw correlations (Ta- observed vertica'l transmission requires
sidered later. ble 5, rFc and rMc) are clearly inadequate us to address three additional possibili-
Summary analyses of all traits (Table assessments of vertical transmission. In ties. First, the observed intergeneration-
5) include the maximum likelihood esti- particular, the IVT values (Table 5) show al transmission might. be the result of
mates and their standard errors for the that vertical transmission is often sub- components of oblique rather than truly
transmission parameters of the additive stantial. If the IVT were equal to 1, the vertical transmission, and these compo-
model, which fits the data satisfactorily trait would be maihtained in equilibrium nents might not be included in Po be-
(Table 6). For comparison, the goodness at its present frequency without other cause they are confounded with socio-
of fit of another well-known model, the modes of transmission and without con- economic stratification of parental pairs.
log linear (18), was also tested. The X2's tribution from other sources increasing To investigate this possibility we
of the three-factor, interactions for the the frequency of the trait (that is, for looked at the data on parental income
log-linear model (Table 5) for all traits P3o = 0). and occupations. These showed only
and their distribution (Table 6) show that Religious and political attitudes atd modest correlations with the traits udider
the additive and the log-linear models affiliations have the strongest vertical investigation, and partial correlations of
give similar results. An adVantage of our transmission. After disattenuation, most parent with child (given the other parent)
additive model is the natural interpreta- lower IVT values would increase consid- were hardly affected when we partialed
tion of the estimated parameters P3o, aF, erably, but the interpretation of disatten- out income and professions of father and
aM. A disadvantage is that it is .more uated coefficients is not unatnbiguous mother. At least this particular facet of
restrictive than the log-linear or logistic and statistical significance is often lost. oblique transmission seemed to be unim-
models, for example, in that the additive Averages (± standard deviations) of portant in our sample.
model may produce no fit because of the IVrs in the six categories are: religion, Second, it the transmission is truly
generation of negative expected frequen- .77 .09; politics, .61 + .11; entertain- vertical, does it have a genetic compo-
cies. This occurred for four traits (Table ment, .19 .05; beliefs, .17 .04; nent? It should be stressed that our re-
5). Two other models were fitted for sports, .12 .05; and habits, .09 .05. sults (Table 5) do not allow a clear dis-
these traits: (i) paternal transmission In terms of the relevance of our the- tinction to be made between vertical
only and (ii) maternal transmission only. ory, the values of IVT (Table 5) are transmission that is genetic, cultural, or
In the four cases of no fit with the expected to equal the sum of the esti- some mixture of these (19). It is well
additive model, the x2 values and the mates aM and aF if our sample repre- knoWn that the most powerful assess-
estimates aF and aM (Table 5) are those sents an equilibrium population, that is, ment technique uses adoption (20), data
obtained from the model that gave the if the trait frequencies in parents and for which were not available to us. How-
better fit by x2. In three cases the model children are identical, with the additive ever, Loehlin and Nichols (21), in their
was of maternal transmission only model IVT = aF + aM. Indeed the de- study of twin candidates for the National
(a( = 0) and in the other, paternal trans- gree to which this identity fails may Merit Scholarships, asked a number of
mission (aM = 0). We have explored measure the departure from equilibrium. questions sufficiently similar to ours that
only a tiny fraction of the models that Thus in the example of Democratic party direct comparison seems appropriate.
might be applied- to these transmission affiliation, IVT = .49 (Table 5) and They found that although the correla-
situations; others have been described atF + aM = .65. In this example the trait tions between twins were often high, the
(2, 7). frequencies in parents and children are differences between correlations of iden-
It is interesting to relate these statis- u = .39 and u' = .53, respectively, a sta- tical and fraternal twins for the two cate-
tics to the conmmonly employed tneasure- tisticaily significant difference. We can- gories of traits that showed the highest
24 SCIENCE, VOL. 218
vertical transmission in our data, name-
ly, religion and politics, were generally
small. This finding suggests that the
transmission 'of these traits may have
little or no genetic basis. For some traits
other than religion and politics the corre-
lations of identical and fraternal twins
showed small but not negligible differ-
ences, so that possible contributions of
biological inheritance to vertic,al trans-
mission cannot be excluded.
Third, what is the effect of horizontal
transmission? Some, of this might not be
included in PBo but be confounded with
the vertical component; for instance,
there may be influence from older sib-
lings, to the extent that the latter were
affected by parents (22). We have no
estimates for such intrafamilial transmis-
sion from the present survey. However,
we were able to obtain additional infor-
mation on horizontal transmission.
Assortative Mating and Horizontal
Transmission
Since most parental pairs in our study
had been together for at least 20 years, it
is impossible from our data to distinguish
true assortrnent at the time of marriage
from later convergence. Convergence
would involve horizontal transmission
between spouses, and its separation
from assortment at marriage would re-
Table 7. Contingency table of Democratic party affiliation of parental pairs from Table 4.
Trait
Democrat
Mother Democrat
Mother not Democrat
Total
quire a great deal of longitudinal infor-
mation (17, 23). Without attempting this
breakdown, we cap use some of our data
(Table 5) to demonstrate the present
correlations between marriage partners,
with the Democratic party affiliation as
an example (Table 7).
The association is tested by x2 (in this
example x2 = 64.84, P < .001), and the
association can be measured by rFM = XX
V (r = .58), where N is the total nuM-
ber Qf parental pairs and the sign indi-
cates the difference between the prod-
ucts of' the diagonal elements [(55 x
99) - (27 x 14)], positive in this case
(24). Father-mother 'correlations com-
puted in this way (column 1 in Table 8)
are almost all significantly greater than
zero. The order is similar tp that both for
vertical unattenuated transmission and
for agreement, namely: religion/politics
> Cntertainment/sports > beliefs > hab-
its, with beliefs and entertainmnent inter-
changed after disattenuation. Marital
correlations are on the average the high-
est (Table 2). This explains the better
Father Father not
Democrat
Total
°oa
5 27 82
14 99 113
69 126 195
correspondence between our a's (Table
5) and the partial, rather than the raw,
correlation coefficients.
The data on friends supply potential
information on horizontal transmission.
There were only a few statistically signif-
icant differences in patterns of correla-
tion between sexes, either when compar-
ing correlations of students with a friend
of the sa,me sex or a friend of the oppo-
site sex, or those of male students with
female students (Table 8). With these
results pooled, the order of uncorrected
similarities (before attenuation) is; reli-
gion > sports/entertainment > politics/
beliefs > habits. After attenuation, the
order changed substantially; beliefs >
politics > 'sports/entertainment > reli-
gion > habits. As noted, a11 transmission
correlations tend to be lower among
friends than they are among parents. The
longer time available for horizontal
transmission and the requirements for
partner selection, which are likely to be
more stringent for the choice of a mate
than for the choice of a friend, might

aF

Political position
V
.4 I ,/ A Significant for maternal transmission
V Slqnlflcant for paternal tranamisalon
Breakfast
V Significant for both transmissions
V Enjoy parties 0 No significance for either transmission
BaseballV

Museum
.3 F V
OGood friends Menqberof party
V Voter Swim Democrat
ESP Football VA end church
V ,2 V7 /
O Jog
Tennis Attend movies
Coffee Check bill A
A TV C
Margarine/butter
0
.1 -Serious A AMusic Republican
movies0 A A,Pray
Light movieso o Ability/luck A
Protestant A
X UFOQ Salt use
Milk with dinnero Health jo9 TeaoAHoroscope Jewish
I I ,l O OAdventure movles a
-.3 -.2 -. 1 .1 .2 .3 .4 .5 Cat'holic
ahol

Morning/evening QaMper
0 0OGood
0
OGod correspondent 0 Lucky number
coresponentO Last-minute person
-.1
Fig. 2. Coefficients of paternal and maternal transmission (cxM and aF) estimated for 41 traits on the basis of data shown in Table 5 from an
additive model of vertical transmission. The 450 line corresponds to equal influence of father and mother; tM and aF are statistically significant if
their departure from zero is greater than twice their estimated standard errors. The standard errors are obtained from the maximum likelihood
estimation, and the test of significapce is approximatq, especially if the pumber of informative (H x h and h x H) parental pairs is small.
I OCTOBER 1982 25
contribute to the almost threefold differ- space. We have suggested ways to use case, for example, transmission may oc-
ence in average correlations (Table 2). survey data for the operation and mea- cur in both directions, as with many
The lower agreement (r = .33 for friends surement of somne components of trans- infectious diseases. The strong mother-
and r = .45 for parents) partially ex- mission. The analysis here has been cen- father correlations may likewise have a
plains the lower correlations between tered on dichotomous traits but can in component of reciprocal horizontal
friends than between parents. Even so it principle be extended to polychotomous transmission. No statistical measure of
is likely for many of these traits that and continuous traits (2). association can indicate bona fide causa-
vertical transmission is stronger than The models have their roots in our tion (in our study, transmission). Yet if
horizontal. evolutionary theory, but also seem to be an individual, at the end of a period of
in excellent agreement with certain more socialization and education, resembles
conventional modes of analysis which its mentors, some process of transmis-
Discussion may not specify a transmission model: sion (conscious or subconscious) must
partial correlations and log-linear analy- have been going on.
Our quantitative theory of cultural sis of contingency tables. We find strong vertical cultural trans-
evolution suggests that information on That our transmission models are ade- mission for political tendencies and reli-
the rules of cultural transmission is criti- quate to explain the data does not pre- gion. This is not unexpected. Significant
cal to prediction of variation between clude several other potentially important vertical transmission is also present for a
and within populations over time and possibilities. In the parent-offspring number of other traits represented in our
categories of entertainment and sports,
superstitions arid beliefs, and customs
and habits. In some cases mothers are
Table 8. Horizontal transmission and assortment of traits. Correlations are between a students' important transmitters, and in others fa-
father and mother (rFM), between a male student and a male friend (rBB), between a female
student and a female friend (rGG), between a female student and a male friend (rBG), and thers appear to do the transmitting (Fig.
between a male student and a female friend (rGB). Self-ratings were used -throughout. 2), but for all traits the data are compati-
Significance of r is coded as follows: 1, P < .05; 2, P < .01; and 3, P < .001. The significance ble with zero interaction between the
of heterogeneity refers to the correlations between friends (rBD, rOG, rBG, and rGB), and P is two parental components. We have
coded as before.
pointed out the important distinction be-
Het- tween vertical cultural and vertical bio-
Correlations and significance ero-
Trait
logical transmission. It is interesting that
gene-
the traits in our study that exhibit the
rFM P rBB P rGG P rBG P rGB P ity
strongest evidence of vertical transmis-
sion cannot have an important biological
R 1 .62 3 .38 1 .46 3 .11 .36 2 component.
R 2 .39 3 .19 .30 1 .11 .39 2
R 3 .82 3 .28 -.06 .18 .27 1 It is likely that the ages of the subjects
R4 .93 3 .45 1 .20 .42 1 .20 in surveys such as ours will affect the
R 5 .66 3 .11 .14 .19 -.12 results. It is likely also that the variojis
P1 .34 3 -.10 .22 -.14 .05 forms of contact change with age, as
P 2 .41 3 .18 .25 -.02 .03
P 3 .62 3 .07 .01 .37 1 .07 does sensitivity to the reception of cul-
P4 .57 3 -.03 .02 .10 .08 tural transmission (25). Both are likely to
P 5 .39 3 .32 1 .14 .18 .04 be trait-specific, and individual differ-
SI .32 3 .25 .25 -.03 .17 ences may exist. These considerations
S2 .55 3 .00 .00 .29 .35 1 together with the fact that the mode of
S 3 .14 1 .03 .25 -.14 .24
El .32 3 .34 1 .22 .28 .17 cultural transmission is presumably also
E 2 .46 3 .12 .08 .27 .17 culturally transmitted suggest that varia-
E 3 .19 2 .43 2 .33 1 .18 .28 1 tion between and within populations
E 4 .31 2 .41 1 .32 1 .32 1 .12 could be strongly influenced by age-re-
E 5 .39 3 .00 -.01 .00 .38 1 lated variables.
E 6 .36 3 -.14 -.19 -.14 -.04
E 7 .14 1 .03 .12 .03 .07 We realize that our sample is smnall
E 8 .50 3 .25 .15 .18 .43 2 and, like most survey samples, biased. It
E9 .13 .10 -.09 -.09 .28 1 may be possible to apply methods of
ElO .18 1 -.38 1 .05 -.23 .05 analysis such as we have described to
E I1 .16 1 -.09 .17 .05 -.04
H 1 .12 .00 .19 .05 .10 larger data sets, from various cuitures,
H 2 .18 .04 .19 .14 .23 and to longitudinal data as well as single
H 3 -.07 .45 -.14 -.36 -.23 3 time points. It will be important to study
H 4 .12 .65 2 -.02 .06 .07 3 the relative influence of many mecha-
H 5 .06 -.05 .01 .45 .01 I nisms of cultutral transmission. Longitu-
H6 .44 3 .11 .06- .02 .11
H7 .01 -.49 -.04 .33 .06 2 dinal observations will be invaluable in
H8 .27 2 -.39 -.05 .30 -.26 2 assessing the constancy and direction of
H9 .55 3 -.10 -.15 -.22 -.05 influence, possible age effects, and criti-
H 10 .11 .00 .11 .13 .02 cal periods. Even in their absence, our
B 1 .12 ~ .15 .18 .24 .17
transmission tnodels augment correla-
B 2 .28 3 .16 .10 -.02 -.06
B 3 .32 3 -.02 .05 -.02 -.24 tional analysis and in so doing provide a
B 4 .19 2 .01 -.08 -.12 .00 better perspective from which to evalu-
B 5 .09 -.04 .04 .30 .09 ate transgenerational comparisons and to
B 6 .21 2 .03 .34 1 .02 .24
B 7 .34 3 -.27 -.21 .07 .09 mnake predictions about cultural evolu-
tion.
26 SCIENCE, VOL. 218
 Other potential applications of this ap- 1975), chapter 6] summarizes and analyzes re-
search on transnission.
proach to the study of cultural transmis- 11. R. W. Connell, Public Opin. Q. 36, 321 (1972).
sion may be found in linguistics, psy- 12. Those questions whose answers were positions
on a scale were dichotomized at the mean. This
chology, anthropology, sociology, and conservative approach made the analysis rea-
communication sciences. Perhaps the sonably distribution free and allowed compari-
sons with the model of Table 4. It is also
most exciting prospect is that of being practically insensitive to inadequacy of scales, a
major problem with graded data, although there
able to use observations on transmission is a cost in lost information. If graded scales
to predict variation between individuals were used, analysis by linear structural equa-
tions could have been employed [L. L. Cavalli-
and populations, over space and time. Sforza and M. W. Feldman, Genetics 90, 391
(1978); C. R. Cloninger et al., Am. J. Hum.
Referee and Nots Genet. 31, 366 (1979); S. Karlin, Theor. Pop.
Biol. 15, 308 (1979)]. Some of the pitfalls of this
1. J. M. Davis, in Imitation: A Review and Critique approach have recently been stressed by A. S.
in Perspectives in Ethology, P. Bateson and P. Goldberger [in Genetic Epidemiology, N. W.
Klopfer, Eds. (Plenum, New York, 1973), pp. Morton and C. S. Chung, Eds. (Academic Press,
43-72; B. G. Galef, Jr., Adv. Study Behav. 6, 77 New York, 1978), pp. 195-222], by M. W.
(1976); D. Mainardi, in preparation; P. C. Mun- Feldman and L. L. Cavalli-Sforza [in Genetic
dinger, Ethol. Sociobiol. 1, 183 (1980). Analysis of Common Diseases: Applications to
2. L. L. Cavalli-Sforza and M. W. Feldman, Cul- Predictive Factors in Coronary Disease, C. F.
tural Transmission and Evolution: A Quantita- Sing and M.- Skolnick, Eds. (Liss, New York,
tive Approach (Princeton Univ. Press, Prince- 1979), pp. 203-227], and by S. Karlin [in ibid.,
ton, N.J., 1981). pp. 497-5201.
3. R. W. Gerard, C. Kluckhohn, A. Rapoport, 13. Father and mother were those regarded by the
Behav. Sci. 1, 6 (1956); L. L. Cavalli-Sforza, in student as having that role. The first time the
Mathematics in the Archaeological and Histori- questionnaire was used it was sent to the parents
cal Sciences, F. R. Hodson, D. G. Kendall, P. and friends of the student, producing five evalu-
Tautu, Eds. (Edinburgh Univ. Press, Edin- ations of the student. Because of the low return
burgh, 1971), pp. 535-541; D. T. Campbell, in from friends the first year, when the question-
Social Change in Developing of Evolutionary naire was given to students in the same classes I
Theories, H. R. Barringer, A. I. Blanksten, R. year later one randomly chosen set obtained
W. Mach, Eds. (Schenkman, Cambridge, parental evaluations and another set obtained
Mass., 1965), pp. 19-49; W. H. Durham, Hum. those offriends. Each set produced three assess-
Ecol. 4, 89 (1976). ments of the student. The parents and friends
4. F. Galton, Natural Inheritance (Macmillan, collections were each pooled over the 2 years.
London, 1889). In the second year, 11 more questions on cus-
5. N. T. J. Bailey, The Mathematical Theory of toms and habits were added, and assessments
Infectious Diseases (Hafner, New York, 1973). are therefore based on smaller totals. Minor
6. E. M. Rogers, Diffusion of Innovations (Collier- differences between total numbers for the vari-
Macmillan Canada, Gait, Ontario, 1962); ous traits reflect blank entries.
and F. F. Shoemaker, Communication of Inno- 14. Issues concerned with self-report have generat-
vations-A Crosscultural Approach (Macmillan, ed theoretical and methodological controver-
New York, 1971). sies: see D. Bem [in Advances in Experimental
7. L. L. Cavalli-Sforza and M. W. Feldman, Social Psychology, L. Berkowitz, Ed. (Academ-
Theor. Pop. Biol. 4, 42 (1973); M. W. Feldman ic Press, New York, 1972), vol. 6, pp. 2-62] and
and L. L. Cavalli-Sforza, Ann. Hum. Biol. 2, D. P. Crowne and M. W. Stephens [in The Self
215 (1975); Theor. Pop. Biol. 9, 238 (1976); in in Social Interaction, C. Gordon and K. J.
"Proceedings of the 41st session of the Interna- Gergen, Eds. (Wiley, New York, 1968), vol. 1,
tional Statistical Institute, New Delhi" (Bull. pp. 145-154].
Int. Stat. Inst. 2, 151 (1977)]1 15. The slash (for example, in sports/politics and
8. M. K. Jennings and R. G. Niermi [The Political other cases) indicates that the categories are
Character of Adolescence (Princeton Univ. close in the ordering.
Press, Princeton, NJ., 1974)] present an exten- 16. One method of removing the effect of attenua-
sive analysis from a social psychological per- tion is to compute the ratio similarity: agree-
spective. a ment. Thus, suppose that the observed value in
9. S. Scarr, Race, Social Class, and Individual one individual is e and in a second is q and that
Differences in I.Q. (Erlbaum, Hillsdale, N.J., each is the sum of two independent random
1981). variables e = x + e,, 1 = y + a2 where el and
10. R. L. Johnstone [Religion and Society in Inter- a2 are independent errors having variances Oj2
action (Prentice-Hall, Englewood Cliffs, N.J., and a22, and x and y, with variances aU2 and
I OCTOBER 1982
View publication stats
cy are realizations of a "true" value for
the trait. Then the observed correlation r is
equalJ to r r, where r = ax,a,,/[a 2 + a,2)
(ay2 + a22)]I/ yaverages the intraclass correla-
tion or fraction of variance due to x and y. We
use the agreement between observers as an
estimate of r, so that the division of the observed
correlation by the agreement can be regarded as
an estimate of the true correlation. If self-ratings
were actually accurate, however, this procedure
would result in an overestimate. On the other
hand, raw correlations from self-ratings only
were not higher on the average than those based
on ratings by others. This contrasts with the
expectation if self-ratings were substantially
more accurate.
17. A special application of our methods in the
longitudinal analysis of smoking behavior in
married couples has been made by R. A. Price,
K.-H. Chen, L. L. Cavalli-Sforza, M. W. Feld-
man, Social Biol., in press.
18. Y. M. M. Bishop, S. E. Fienberg, P. W. Hol-
land, Discrete Multivariate Analysis (MIT
Press, Cambridge, Mass., 1975).
19. K. K. Kidd, R. C. Heimbuch, and M. A. Rec-
ords [Proc. Natl. Acad. Sci. U.S.A. 78, 606
(1981)] used a procedure similar to the log-linear
model in testing for parental transmission of
dichotomous data.
20. L. Cavalli-Sforza and M. Feldman, Am. J.
Hum. Genet. 25, 618 (1973); in Genetic Epidemi-
ology, N. E. Morton and C. S. Chung, Eds.
(Academic Press, New York, 1978); H. Grote-
vant, S. Scarr, R. Weinberg, J. Pers. Soc.
Psychol. 35, 667 (1977).
21. J. C. Loehlin and R. C. Nichols, Heredity,
Environment and Personality (Univ. of Texas
Press, Austin, 1976).
22. A parallel survey on Taiwan students (K.-H.
Chen, L. Cavalli-Sforza, M. Feldman, Hum.
Ecol., in press) included information on older
and younger brothers and sisters. It suggested
that siblings are sometimes more influential than
parents, and the parents' influence is often me-
diated through them.
23. B. K.- Eckland, Eugen. Q. 15, 71 (1968); R. J.
Garrison, V. E. Anderson, S. C. Reed, ibid., p.
113; R. Lewontin, D. Kirk, J. Crow, ibid., p.
141.
24. This produces an estimate whose expectation
indicated earlier by m or rFM is the usual mating
correlation introduced by S. Wright [Genetics 6,
144 (1921)].
25. J. P. Scott, Ed. Benchmark Papers in Animal
Behavior, vol. 12, Critical Periods (Dowden,
Hutchinson & Ross, Stroudsburg, Pa., 1978).
26. We are grateful to F. B. Christiansen, L. Cron-
bach, and M. Hannan for their critical com-
ments on an early draft, and to B. Aschenbren-
ner and M. Langer for assistance in the data
collection. Supported in part by Boys Town
Center for the Study of Youth Development at
Stanford University and NIH grants 10452,
20467, and 20816.
27