INTELLIGENCE  , 170191 (1977)
Intellectual Similarities within Families of Both Adopted and Biological Children* SANDRA
SCARR
and
RICHARD
A. WEINBERG
University of Minnesota The effects of genetic and environmental differences on intellectual differences among children were examined in a study of families with both biological and adopted children. IQ scores of all family members and education of natural parents were used to estimate intellectual similarities among related and unrelated persons, living together and apart. Comparisons of correlations between related and unrelated siblings produced negligible heritability values, whereas the parentchild data suggested moderate heritability for the children's IQ differences. The high mean values of the adopted children's IQ scores and the high degree of similarity among unrelated sibs suggest that IQ scores are more malleable than previously thought.
Recent interest in the social ecology of intellectual development has focused on the family context. Implicit in studies of home environments is the assumption that the development of intellectual differences among children is strongly influenced by the behavior of their parents and other family members, as well as aspects of their physical environments (Wachs, 1975). The primary emphasis in studies of family effects has been on differences between families in child rearing practices, parentchild interaction, social class variables, and the like (e.g., Baumrind, 1969). Loehlin and Nichols (1976) have indicated the lack of studies on variation within families, which may be more important in determining individual differences than betweenfamily effects. Birth order is one of the few withinfamily variables that has received any attention (Zajonc & Markus, 1975). There is little research on genetic or environmental differences among siblings or between parents and children. The comparison of adoptive and biologically related families provides a framework for studying both within and betweenfamily effects on intellectual development. Furthermore, related and unrelated persons, living together and apart, offer an opportunity to estimate genetic and environmental effects on individual variation. Ideally, one should study related and unrelated children and their parents in the same families. The present study is an investigation of the similarities in IQ scores among *The research was supported by the Grant Foundation and NICHD (HD08016). Reprints may be obtained from Dr. Sandra Scarf, University of Minnesota, Minneapolis, MN 55455. 170
INTELLECTUAL SIMILARITIESWITHIN FAMILIES
l7 [
members of families with adopted and biological children. In past adoption studies, few families had both adopted and biological children (Burks, 1928; Freeman, Holzinger, & Mitchell, 1928; Leahy, 1932; Munsinger, 1975; Skodak & Skeels, 1949), because most adoptions result from parental infertility. The present sample is unusual in several respects: first, the families adopted children from a variety of racial backgrounds; second, they adopted many children past infancy; and third, the majority also have their own biological children. The data presented in this article are family correlations, which reflect rank order resemblances among related and unrelated siblings and parentchild pairs. Resemblances among family members can be measured in two ways: means and rank orders. While similarities in the averages and distributions of scores are presumably responsive to the average values of the genotypes and environments in which the children are reared, the rank orders of scores reflect the relative values of individuals' genotypes and environments. Resemblances in average scores and distributions were reported in another paper (Scarr & Weinberg, 1976). The study of intellectual similarities is part of a larger investigation of the psychosocial functioning of transracial adoptive families. The purposes of this paper are (1) to describe the correlations in intellectual performance between parents and children, whether related or not and whether living together or not; (2) to analyze the effects of rearing together on sibling resemblance, whether the sibs are genetically related or not; and (3) to investigate the effects of selective placement of adopted children in families that resemble intellectually their natural parents. While, in the world of real families, relationshipsboth interpersonal and statisticalare full of problems, the value of a study of related and unrelated families is to clarify the roles of genetic and environmental differences in creating the intellectual diversity we observe. METHODS The Adoptive Families
The 101 participating families were recruited through the Newsletter of the Open Door Society and by letters from the Minnesota State Department of Public Welfare Adoption Unit to families with Black adopted children, 4 years of age and older, who were adopted throughout the state of Minnesota through the Lutheran Social Service and the Children's Home Society. These agencies have placed the majority of adopted Black children in the state. We were unable to ascertain how many transracial adoptive families learned about the study from the Newsletter since the mailing list of about 300 includes agencies, social workers, and interested citizens. In addition, we do not know how many of these
172
s. SCARR AND R. A. WEINBERG
families were also contacted by the State Department of Public Welfare. The support of the Open Door Society was important, however, in affirming the legitimacy of the study. The State Department of Public Welfare mailed 230 letters to transracial adoptive families. In some cases a family received more than one letter if they had adopted more than one child. The mailings of the Newsletter and the State Department of Public Welfare yielded 201 replies. Of these 65 were ineligible to participate, mostly because their children were less than 4 years of age; 28 declined to participate, 14 because they lived too far away, and 14 for various personal reasons; and 108 agreed to participate. Of the 108 volunteers, 101 families eventually took part in the study. Thus, of the 136 families known to be eligible for the study, 74% were actually studied. The 101 participating families included 145 biological children and 176 adopted children, of whom 130 were socially classified as Black (29 with two Black natural parents and 101 with one Black natural parent and one natural parent of other or unknown racial background), and 25 as white. The remaining 21 included Asian, North American Indian, and Latin American Indian children. All of the adopted children were unrelated to the adoptive parents. Adopted children reared in the same home were unrelated, with the exception of four sibling pairs and one triad adopted by the same families, who were excluded from the analyses. The sample of families live within a 150 mile radius of the Minneapolis/St. Paul, Twin Cities metropolitan area. Although nearly all of the children were adopted in Minnesota, 68 were born outside of the state. Through interstate cooperation, the child placement agencies arranged for the adoption of many Black and Indian children from other states.
Early Adoptees and Natural Children To study family similarity, we decided to restrict the present study to three types of families:
I. Families who adopted children during the first year of life (early adoptees): Of the 176 adopted children, 65 were adopted after 12 months of age. Because early experience elsewhere can reduce the similarity of adopted children to their adoptive parents, this report includes only the 111 children adopted in the first year of life. The group included 13 children with 2 Black parents, 9 with 2 white parents, 3 Asian/Indian children, and 86 with one Black parent and one parent of other or unknown racial background. 2. Adoptive families with natural children: Of the 101 adoptive families, 72 have biological children, but they do not necessarily have earl>' adopted children as well. Correlations between natural children reared together and between parents and their own children can be compared to sibling and adoptive parentchild correlations in families that adopted children in infancy.
I N T E L L E C T U A L SIMILARITIES WITHIN FAMILIES
173
3. Adoptive families With both natural and early adopted children: Because adoptive families with natural children may differ in significant demographic or intellectual ways from other adoptive families, we calculated family similarity for both adoptive and biologically related pairs within the same families. Procedures' Most of the information was obtained directly from members of the adoptive families at the time of testing (19731975). Some additional data on the natural parents and the children's preadoption history were obtained by Minnesota State Department of Public Welfare personnel from the adoption records. Achievement and aptitude test scores were supplied by school districts for all of the schoolaged children to whom such tests had been administered.
The IQ Assessment Both parents and all children in the family over four years of age were administered an ageappropriate IQ test as part of an extensive battery of intellectual, personality, attitudinal, and demographic measures. The tests were administered in the family home during two visits by a team of trained testers. The examiners were all graduate students who had completed at least a yearlong course in psychoeducational assessment and who had participated in a training session on assessment for this study. Among the 21 examiners were 7 males and 15 females, including 2 Blacks. Testers were assigned randomly to members of the family. Race and sex of examiner were unrelated to children's or parents' IQ scores (all r ' s < .06). Eighteen testers assessed 5 or more parents and 5 or more children. The standard deviation of the mean IQ scores they each obtained was 4.0 for children and 2.8 for parents. For the 11 testers who assessed 15 or more children, the standard deviation of the scores obtained was also 4. These tester differences were normally distributed and well within the limits of sampling error for Ns of 5 to 33. Both parents and all children 16 years of age and older were administered the Wechsler Adult Intelligence Scale (WAIS). Children between 8 and 15 were given the Wechsler Intelligence Scale for Children (WISC), and children between 4 and 7 were administered the StanfordBinet Intelligence Scale, Form LM (SB). All scoring of protocols and computations of IQ scores were done by a graduate student with extensive experience in administering and scoring IQ measures. This student had no contact with the families and with the examiners except to clarify questionable responses. In no case was the scorer aware of the child's race or adoptive status.
The Adoption Records The Director of the Adoption Unit, Minnesota State Department of Public Welfare, abstracted the following information from the records of the adopted children and their families:
174
S. S C A R R AND R. A. W E I N B E R G
1. The child: a. birthdate; b. number and dates of preadoption placements, unless the child was in the adoptive home at 2 months of age; c. evaluation of the quality of preadoption placements, rated by the authors on a scale of 1 = poor to 3  good; 4 = placement only in the adoptive home; d. date of placement in adoptive home. 2. The natural parents: a. age at birth of child; b. educational level at birth of child as an estimate of intellectual functioning, since IQ scores were not available; c. occupation of mother; d. race.
Family Demographics In the interview portion of the testing session, each parent was asked his or her birthdate, last school grade completed, occupation and whether it was full time or part time, range of income, and date of marriage.
Statistical Analysis To eliminate mean and distributional differences between the adoptive parents' and the adopted and natural children's IQ scores, the scores were standardized to a mean of 0 and a standard deviation of 1 separately for parents, for adopted, and for natural children by test and within the three types of family constellations. Similarity, therefore, reflects only rank order resemblance and not similarity in mean scores. Regression analyses were applied to the parentchild IQ scores. Pearson correlations were 'calculated for sibling pairs. All parentchild and sibling data were also analyzed by intraclass correlations. When scores are standardized, the results of the various correlational and regression analyses are entirely equivalent. Therefore, only the intraclass correlations are reported in this paper. RESULTS
Family Characteristics The adoptive families who participated in the study can be characterized as highly educated, above average in occupational status, and in income. Table 1 is a summary of selected demographic characteristics of the adoptive and natural parents in the three family constellations. Generally, in all family constellations, the educational level of the adoptive parents exceeded that of the adopted children's natural parents by 3  5 years. The typical occupations of the adoptive fathers were clergyman, engineer, and teacher. Nearly half (46.5%) of the adoptive mothers were employed at least part time at the time of the study, typically as teachers, nurses, and secretaries. The median educational level of the natural parents was high school graduation, which is close to the median for that age cohort of the general population. In
I N T E L L E C T U A L SIMILARITIES WITHIN FAMILIES
1 75
TABLE 1 Income and Educational Characteristics of the Adoptive and Natural Parents of the Adopted Children by Family Constellation N
Mean
SD
Adoptive families with natural children Income
71
$15,250
$4,500
Education Adoptive mother Adoptive father
72 72
15.0 16.8
2.2 3.0
Adoptive families with early adopted children Income
73
$15,250
$4,500
Education Adoptive mother Adoptive father Natural mother Natural father
73 73 94 23
15.4 17.5 ! 2.6 12.6
2.0 2.7 1.9 1.9
Adoptive families with both natural and early adopted children Income
51
$14,750
$4,750
Education Adoptive mother Adoptive father Natural mother Natural father
51 51 41 12
15.4 17.5 12.3 12.2
2.0 2.6 1.8 .8
contrast, the mean educatiohal level of the adoptive parents was atypically high. Typical occupations of the natural mothers were office workers, nurses' aides, and students. Insufficient information was available on the occupations of natural fathers. There were no significant differences among the three family types in demographic characteristics.
IQ Scores of Family Members As indicated in Table 2, the mean WAIS IQ scores of the adoptive parents in all three family constellations were in the high average to superior range of intellectual functioning. The distributions of scores extended from the "low average" to the "very superior," with considerable restriction of range. The scores were congruent with the very high educational level of the group. Within occupational classes, one expects a restricted range of IQ. Burr (1961) reported that within six classes, the standard deviation for IQ was 9.6 instead of the population value of 15. The variability of IQ scores of children whose fathers were found within the various occupational classes was greater (SD 14.0). The mean IQ scores (presented in Table 2) of the natural children of the
176
S. SCARR AND R. A. WEINBERG
TABLE 2 Mean IO Scores of Adoptive Family Members by Child Test and Family Constellation N
Mean
$D
Range
10.0 9.1 11.9 11.6 12.0
97139 98140 86144 92138 86144
9.7 10.6 16.7 12.4 14.0
96140 93140 81148 87150 81150
Adoptive families with early adopted children Adoptive mother's WAIS IQ Adoptive father's WAIS IQ Early adopted child's StanfordBinet IQ WISC + WAIS IQ Total IQ
71 73 92 19 111
118,5 122.3 110,3 115,2 II1,1
Adoptive families with n~ural children Adoptive mother's WAIS IQ Adoptive father's WAIS IQ Natural child's StanfordBinet IQ WISC + WAIS IQ Total IQ
71 71 47 96 143
117,7 120,2 113,8 118,1 116,7
Adoptive families with both natural and earlyadoNed children Adoptive mother's WAIS IQ Adoptive father's WAIS IQ Early adopted child's StanfordBinet IQ WISC + WAIS IQ Total IQ
Natural child's StanfordBinet IQ WISC + WAIS IQ Total IQ
50 51 56 11 67
118,7 121,5 109,0 113.6 109,8
9.4 9.7 12.5 12.8 12.6
98134 98140 86144 92133 86144
32 70 102
115,6 119,8 118.5
16.9 12.2 13.9
81148 93150 81150
adoptive families were in the high average range of intellectual functioning, as predicted by their parents' high IQ scores and their enriched home environments. The standard deviation of 14 matches Burt's (1961) finding. The mean IQ scores of the early adopted children were also in the high average range, reflecting their superior family environments. The standard deviation of 12 represents considerable restriction of range. One possible explanation for the smaller standard deviation of adopted children's IQ scores is the lack of genotypeenvironment correlation for adopted children. Another is a bias that might affect the study: the selfselection of participating families whose children have less conspicuous sibling differences. To test this hypothesis, we calculated absolute IQ differences between natural and adopted sib pairs, as shown in Table 3. Naturalnatural and adoptedadopted sib pairs do not differ in their average absolute differences in IQ scores, although their differences are smaller than one would expect among sibs in the population at large ( 13 IQ points). An argument against the selfselection by families with small sib differences is the larger than expected sibling differences for naturaladopted pairs. In fact, the average
INTELLECTUAL
177
SIMILARITIES WITHIN FAMIL1ES
TABLE 3 Mean Absolute Differences in IQ between Related and Unrelated Sibling Pairs I I vs. 2
Sib pair types
N
d
SD
l
Naturalnatural
107
l 1.4
9.2
1 v,~. :~
P
2
Adoptednatural
134
14.6
10.4
2.45
.02
3
Adoptedadopted
53
10.8
6.6
.49
.63
t2 ,.~. :~
2.49
.01
difference between all sibs in the adoptive families is 13.1 IQ points, the difference expected between sibs in all families in the population. Other kinds of selfselection are also used to criticize the results of adoption studies. Munsinger (1975) noted that obviously retarded and damaged infants are not likely to be adopted, a fact which raises the mean IQ of adoptees above the population average. This bias is slight, however: Ifall infants with eventual IQ scores of less than 60 (at most 3% of children) were eliminated from the adoption pool, the mean IQ of adoptees would be raised by only 1 IQ point. Another bias could be the selfselection of families whose children appear normal in intelligence and school work. The range of IQ scores in this study contraindicates a strong bias in this regard, since 15 of the 176 adopted children have IQ scores of 85 and below. Furthermore, since 74 per cent of those families known to be eligible did participate and the average IQ of all 176 adoptees was 106, the average IQ of children in the 26% of families who did not participate would have to be unreasonably low to explain the mean results. If we consider the sample to be composed entirely of interracial children, with white adoptees offsetting those with two Black parents, their average IQ might fall between those of Black and white children in the region, namely, 95. To obtain this figure, the nonparticipants would have to have IQ scores that average 64, or in the retarded range. This is highly unlikely for any sample of adopted children. For all the groups of children, the StanfordBinet (1972 norms') yielded a slightly lower mean score than the WISC or WAIS. Had the 1960 StanfordBinet norms been used, the average IQ scores of the children would have been 7 points higher. In families with either or both natural and early adopted children, the total IQ score of the adopted group was five to six points lower than that of the natural children, in part because a large number of natural children were old enough to take a Wechsler test. The average differences between all adopted and all natural children by test were about three points, not statistically significant differences. In the families with both natural and adopted children, the differences by test were also not significant. The average total IQ scores were statistically different in comparisons between all natural and adopted children (t 3.43, p < .01) and between natural and adopted children in families with both (t = 4.21, p < .01).
178
S. SCARR AND R. A. WEINBERG
THE PARENTCHILD IQ CORRELATIONS Table 4 shows the intraclass correlations of standardized IQ scores by test between the adoptive parents and their children, both'adopted and biological. Regression analyses were also done on the same data. The beta weights never differed from the intraclass correlations by more than .01 (e.g., .38 versus .37). Therefore, the regression analyses are not reported. Also not reported are any values based on Ns of 20 or fewer pairs. The correlations between the adoptive parents and their biological children were higher than those between the adoptive parents and their adopted children in every comparison, although the differences between the correlations were not usually statistically significant. Table 5 gives the same parentchild correlations based only on those families with both natural children and early adoptees. Similarly, in every case the correlations between biologically related parents and children were higher than those between unrelated parentchild pairs. Again, the differences were not generally statistically significant. If one assumes, with Jensen and Munsinger, a polygenic model in which the major (or only) source of similarity between parents and children is their shared genotypes, and one further assumes an assortative mating coefficient of .25, test reliability of .90, and no genotypeenvironmental correlation, then the predicted intraclass correlation (tO between a single parent and a child will be about .50. This correlation is predicted because a parent and a biological child share half of their segregating genes, resulting in their sharing onehalf of the genetic variance due to additive effects, none of the dominance effects, less than onequarter of any effects of epistasis, and half of the variance due to assortative mating (Jencks, 1972, pp. 274275; Jensen, 1973, p. 371; Munsinger, 1975, pp. 624625). In this model the predicted correlation between parent and child does not depend upon any environmental transmission from parent to child. In this study, the adoptive parents' IQ scores were correlated .21, and the educational levels of the natural parents of the adopted children were correlated .27. Both data are consistent with an assortative mating coefficient of .25; however, the obtained values of the parentchild correlations did not reach .50. The biological parentchild correlations in these transracial adoptive families were between . 17 and .50, with most of the values in the .30s. Because adoptive parents and their adopted children share no genes, any similarity between them could only be attributed to similarity in their environment and/or selective placement. The correlations between the unrelated parentchild pairs were found to range between .07 and .29, with the majority below .20. Under the assumptions outlined above, the biological midparentchild correlation would be about .71; that is, ~ / I / 2 (Falconer, 1960) or slightly less, if one assumes a test reliability of .90. Since the biological offspring share genes from
0
>, B
<
~~
0
~
I.U i'M ¢",1 t'~
r
O
e
oo,~0..o m
I~
r'  
~
III
0
I>,
G
~g~
~g
~
~g
.
~~
0
co 0 t
.EE
¢
Z.~
"6
~ .~=
0_ °+.~
"~
~
I1)

0
B
!
~
=~
.~.~
179
~
~
~
c
t~
tt3, ,~T
O m
m I " tt3 ,..2 ,..2 ~
I
E
S
5 tO
,.2
I
UA t
Om
t~
z
~,~
~'__.
7
B~ ¢/) rn
•
>.
II~ 1.1J "l:J e'~
m
<5 ',~"
eq
",~
0~ >.
B
OD q) r,e5
t"
i¢~
~ a ~  ~
r
(D
O L.)
t~
~.e.e~
r,
+
<
¢q
I N T E L L E C T U A L SIMILARITIES WITHIN FAMILIES
181
both parents, the combination of information in the midparent IQ should yield a better prediction of the child's IQ than either parent's score alone. The midparentoffspring correlations were around 0.50; that is, higher than the single parentchild values, but lower than the prediction from the polygenic model. The midadoptive parentadopted child IQ correlations were all between .23 and .26, about half of the midparent and natural child values.
Family Constellations Contary to Kamin's (1974) speculations, there were no consistent differences in family similarity between parents and children in families with natural or adopted children and those with both.
Sex Dil~Epvn('es in ParentChild Similarity Regressions were calculated for natural and adopted children by sex of parent and child. There was no consistent pattern of differences in beta coefficients by sex. The range of motherchild coefficients was .20.36; the range of fatherchild coefficients was .21.41. The range of sonparent coefficients was .21.36; the range for daughters was .20.41. Again, despite Kamin's (1974) comments about past adoption studies, there were no sex differences in the degree of family similarity.
Correction for Restriction of Range (~'d The correlations between parents and children were all depressed by the restricted range of the parents' IQ scores, as shown in Table 2. The standard deviations of the parents' scores were approximately twothirds those of the standardization population. Therefore, the intraclass correlations, corrected for restriction of range (McNemar, 1962), are also presented in Tables 4 and 5. In Table 4, the corrected correlations (?0 for all biologically related parentchild pairs were considerably higher than the uncorrected values. The single, related parentchild correlations rose to about .5 for all IQ scores, and the midparent value was .66. The adoptive parentadopted child coefficients rose to the .22 to .33 range, with a midparent value of .35. The midparentchild IQ regression for biologically related pairs is sometimes seen as an estimate of the narrow heritability of IQ scores in a sample, because it does not include variance due to assortative mating or dominance effects. The midparentchild correlation does include common environment, however, which can certainly increase the resemblance of parents and children for behavioral traits. The adopted children's IQ scores had a more restricted range than those of the natural children (SD  12.0 and 12.6 versus 14.0 and 13.9). This restriction reduced the unrelated parentchild correlations more than the related parentchild pairs. Because no reliable correction for restriction of range in two vari
182
S. SCARR AND R. A. WEINBERG
ables exists (McNemar, 1962), the further correction of the parentchild correlations was not attempted. One should bear in mind, however, that the adopted childparent correlations are probably underestimates of their true values in the population. Natural Parents' Education and the Children's IQ Scores Two sets of analyses were done to compare the natural parentchild correlations, including natural parents who were students or excluding them. Since the category "student" included parents who were attending college as well as high school, there were no differences in the educational mean or standard deviation when students were included or excluded from the analysis. There were also no differences in correlations of children's IQ scores with parents' education, whether or not the "student" category was included. Therefore, the larger sample including students is reported here. Also presented in Tables 4 and 5 are the intraclass correlations between the educational level of the adopted children's natural parents and the children's IQ scores. Although the children have never lived with their natural parents, these correlations were similar to those between the adoptive parents' IQ and their biological children's IQ scores. The single parent correlations ranged from .30 to .52, and the midparent value was .58. Educational level is not as good an estimate of intellectual level as an individual IQ score; therefore, these correlations are underestimates of the intellectual similarity between natural parents and their children who were adopted into other families. A Correction for Selective Placement To test for the effects of selective placement on the adoptive and natural parentchild correlations, an ingenious procedure was suggested by Horn, Loehlin, and Willerman (1975): correlating the education of the adoptive children's natural parents and the IQ scores of the biological children of the adoptive parents. Since the natural parents of the adopted children are neither genetically related to the natural children of the adoptive families, nor live with them, any similarity exists because of selective placementbrighter natural parents of the adopted children have been paired with brighter adoptive parents who have brighter natural children. Therefore, both the adopted and biological children reared by those parents are genetically brighter and environmentally more advantaged. The correlations between the education of the natural parents of the adopted children and the IQ scores of the biological children of the adoptive family measure the degree to which selective placement affects rank order similarity between the adoptive parents and their adopted children. The same correlations also measure the degree to which the natural parentadopted child correlations are inflated by selective placement. When the children of brighter natural parents are adopted by brighter adoptive families, their relative genotypic advantage is enhanced by the relative environmental advantages
INTELLECTUALSIMILARITIESWITHIN FAMILIES
183
provided by the adoptive home. In a sense, the adoptive family provides an environmental program for development, the rank of which is similar to that the natural parents would have provided. Thus, the adopted children come to resemble their natural parents by environmental as well as genetic means (CavalliSforza & Feldman, 1973). The correlations between the education of the natural parents of the adopted children and the IQ scores of the biological children of the adoptive parents ranged from .14 to .21, indicating some effect of selective placement. The midparent value, however, was small (.07), probably due to restriction of range. The standard deviations of the individual parent education levels were 1.9 compared to 1.5 for the midparent value.
ParentAdopted Child Correlations Corrected for Selective Placement The correlations between adoptive parents and their adopted children and between natural parents and their children who have been adopted can be corrected for the effects of selective placement. By Fisher's z transformation the parentchild correlations can be corrected for the correlation between natural parents' education and the IQ scores of the biological children of the adoptive family. The HornLoehlinWillerman correction will vary depending upon the underlying assumptions: (a) if similarity between the biological parents of the adopted children and the natural children of the adoptive parents is predominantly genetic, then the subtraction of that correlation from the natural parentadopted child correlation would not be appropriate, because the procedure underestimates the heritable correction in the scores of biologically related parents and children; the same correction, however, would be appropriate for the adoptive parentadopted child correlations; (b) if, however, the similarity between the biological parents of the adopted children and the biological children of the adoptive parents is predominantly environmental, then the correction is appropriate for biological and not adoptive correlations. Since one cannot decide which assumption is correct, it has been~suggested (Willerman, 1976) that half of the correlation between the biological parents of the adopted children and the natural children of the adoptive parents be subtracted from both biological and adoptive parentchild correlations. This correction can be interpreted to mean that both genetic and environmental factors are contributing to the similarity of the natural parents of the adopted children and the natural children of the adoptive parents. In Tables 4 and 5, the parentchild correlations which have been corrected for restriction of range (~i) have been further corrected for selective placement 0"i'). The adoptive parentadopted child correlations corrected for restriction of range and selective placement ranged from nearly zero for fatherchild pairs to moderately positive for motherchild pairs. The midparent values of.33 and. 34 were moderate. The natural parent educationadopted child IQ correlations were
184
S. SCARRANDR. A. WEINBERG
similarly reduced in magnitude as a result of the correction for selective placement. However, these correlations generally exceeded those of the adoptive parents and adopted children. Another method for correcting parentchild correlations for selective placement is path analysis. Path analysis provides standardized partial regression coefficients for a specified model. Using the data provided in Scarr & Weinberg (1976), Robert Plomin (1976) calculated the weighted average correlation between natural parents' education and adopted child's IQ as .38; the weighted average correlation between adopted parents' education and adopted children's IQ as .28; and the correlation between natural and adoptive parents' education was .22. Figure 1 is the path model, where EdNp and Edav are the educational levels of natural and adoptive parents; IQAc is the IQ level of the adopted children; g is the genetic path, and e the environmental path. Solving for paths g and e, he found that path g was .34 and path e .21. The .34 value of path g was in the range of the natural parentadopted child correlation corrected for selective placement. By the previous method this value was between .25 and .41. The .21 value for path e fails between the adoptive parentadopted child correlations o f . 11 and .26. Heritabilities Based on P a r e n t  C h i l d Correlations
The comparison of related and unrelated pairs of parents and children can provide an estimate of the proportion of genetic variance in the distribution of children's IQ scores. The degree to which the correlations of genetically related parentchild pairs exceed those of unrelated parentchild pairs is used to calculate a heritability coefficient. The heritability coefficient is an estimate of the proportion of variance in a phenotypic distribution (e.g., IQ) which is attributed to genetic differences among individuals. Parentchild correlations yield narrow heritabilities, based only on additive genetic variance. Because heritability estimates contain the unreliability of two correlation coefficients, they tend to fluc
.22
c
E d A p ~
I i e'I
FIG. 1 Pathmodelof the relationshipof naturaland adoptiveparenteducationallevelsto adoptedchildIQ(Plomin,1976).
INTELLECTUAL
SIMILARITIES WITHIN FAMILIES
185
tuate due to sampling error. The estimates, therefore, should not be taken as point values. Tables 4 and 5 include three heritability estimates for each comparison of related and unrelated parentchild pairs. The first heritability estimate is 2 ( r i RPC  ri UPC), where ri RPc is the intraclass correlation of genetically related, parentchild pairs, and ri upc is for unrelated parentchild pairs. To estimate the additive genetic variance the remainder is doubled because parents and children share only half of their genes in common. The heritability estimate from the adoptive parentchild correlations obtained in this sample varied between .ll and .53, a low to moderate range. Taking assortative mating into account reduces the h 2 values by 20% but does not change the basic picture. The second heritability estimates were calculated by the same formula on the intraclass correlations corrected for the restriction in range in the parents' IQ scores (h2). These values were all higher, ranging from .18 to .76. Since the correction for restriction of range had a larger effect on the biologically related, parentchild correlations, the degree to which they exceeded those of unrelated parentchild pairs increased. The third heritability estimates were calculated by the same formula on the intraclass correlations corrected for both restriction of range in the parents' IQ scores and selective placement. When selective placement bias was eliminated, heritabilities ranged from moderate to high, including a value around 1.0. Heritability estimates were also generated using the correlations of natural parents' education and children's IQ scores. (In this case, the adopted children are in the related parentchild pairs, and the natural children of the adoptive family are unrelated to the natural parents of the adopted children.) The heritability estimates based on natural parentchild correlations were in the moderate to high range (.31 .96). Since the sample sizes for natural fathers were small, the higher heritability values were based on the smaller Ns. Since the correction for selective placement is subtracted from the natural parentadopted child correlation in the first heritability estimate (he), there is no need to calculate corrected values. A commonly used method for estimating narrow heritability is the regression of single child on midparent IQ. In Tables 4 and 5, the biologically related parentchild pairs living together yielded heritability estimates of .52 to .72 (corrected for range restriction). The related pairs living apart yielded an estimate of .58. The corrected heritability estimates (h '2) are most generalizable to the Minnesota population from which these families are sampled, because it has neither a restricted range of IQ scores nor a selective placement bias. They are, however, statistically manipulated values with inflated error possibilities. Therefore, one should not take any one figure too literally. The range of .40 to .70 most probably includes the best estimate of the heritability of children's IQ scores, based on parentchild data, in the population sampled.
186
S. S C A R R AND R. A. W E I N B E R G
SIBLING CORRELATIONS Table 6 shows the intraclass correlations for related and unrelated sibling pairs by family constellation. Pairs of adopted children reared together (A/A), pairs of adopted and natural children of the same adoptive parents (A/N), and pairs of natural siblings reared together (N/N) are included. Only the IQ scores combined across tests are presented because there were too few sibling pairs who were of similar age to take the same IQ test. The IQ scores of the genetically related siblings (N/N) correlated .42 for all natural sibs and .37 for natural sib pairs in families with early adopted children. Biological siblings share onehalf of the additive variance, onequarter of the dominance variance, onehalf of the variance due to assortative mating, and less than onequarter of the epistatic variance. Given an assortative mating coefficient of about .25 and a test reliability of .90, the sibling correlation should be about 0.55 (Jensen, 1973; Jencks, 1972). Sibling correlations should exceed parentchild correlations. The pairs of adopted and natural children in the same families were slightly less similar with a correlation coefficient of .30, a value that is surprisingly high for unrelated pairs of children. Even more astonishing are the still higher correlations of unrelated, adopted siblings. The withinfamily variance (MSw) of the early adopted sib pairs was as small as the withinfamily variance of related sibs ( .6). The betweenfamily variances were the same (1.4). Since nearly all of the genetic variance (and some of the environmental variance) in unrelated pairs occurs within families, while only half of the genetic variance (and some of the environmental variance) in related sibs occurs within families, these results imply that genetic variance has nothing to do with similarities in siblings' IQ scores. Compared to the parentchild pairs, both related and unrelated siblings have slightly smaller withinfamily differences and larger betweenfamily differences, resulting in higher intraclass correlations. The largest difference in variances, however, occurred for adopted siblings, compared to adoptive parentadopted child pairs (.62 to .81 for withinfamily effects and 1.4 to 1.2 for betweenfamily effects), Since the unrelated siblings share no more genes than the unrelated parentchild pairs, the changes in variance must mean that environmental differences between families and similarities within families have greater effects on sibling than on parentchild similarities in IQ scores.
Correction for Restriction of Range (ti) Because the standard deviation of the adopted children's IQ scores was restricted by 20%, as shown in Table 2, corrected intraclass coefficients (~'i) were calculated for those correlations involving adopted children (A/A and A/N). After correction, as shown in Table 6, the correlations of unrelated children were even higher.
,< Z ,<
$ Z Z a..<
~
Z Z
eo
z o
E .gl
.~
S
¢
,,,b5 en
,?:
o
0 ,....
S
&
0
8 0
.~
0
.5
¢m
8 ~Y
187
O'
©
188
s. SCARR AND R. A. WEINBERG
One cause of the exceptionally high IQ correlations between adopted siblings may be similarity in their natural parents' education and preplacement histories. If agencies match natural and adoptive parents for two children adopted into the same family, they also create a correlation in background variables between the unrelated sibs. Table 7 gives the correlations of background variables among adopted siblings reared in the same homes. Although three of the four correlations in background characteristics were not statistically significant, there is a suggestion that selective placement has increased the A/A correlations to some extent. Although there was no obvious way to correct for any effects of the selective placement of two adopted children in the same family, one should bear in mind that the adoptedadopted sibling correlations should be a little lower. Heritabilities Based on Sibling Correlations
Heritability estimates calculated from comparisons of the correlations of related and unrelated sibling pairs were low or negligible. When the intraclass correlations were corrected for restriction of range in the adopted children's IQ scores, the heritability estimates (/i z) were zero. If the natural children's IQ scores were corrected for a slight restriction of range (SD  14), there would be no essential change in these results. DISCUSSION The traditional biometrical approach to calculating heritability estimates, based on related and unrelated parentchild and sibling data, yielded conflicting results. Whereas the parentchild correlations suggested moderate heritabilities for children's IQ scores, the sibling data yielded negligible values. Jencks (1972) and ScarrSalapatek (1974) have noted the same trend in data from past adoption studies. Although few pairs of unrelated siblings had been studied, their IQ correlations were too high to yield heritability estimates in the same range as parentchild and twin comparisons. The results of this study confirm previous suspicions that the IQ scores of unrelated siblings are nearly as TABLE 7 Early Adopted Sibs: Correlations in Placement Histories and Natural Mothers' Education
Length of time in home
N(pairs) 52
r .34
p .01
Number of placements
37
.21
.10
Quality of placements
37
.21
.10
Natural mother's education
36
.14
.21
I N T E L L E C T U A L SIMILARITIES WITHIN FAMILIES
189
similar as those of natural sib pairs. Rather than force an immediate biometrical solution onto the combined parentchild and sibling data (see Eaves & Jinks, 1974; Jencks, 1972; ScarrSalapatek, 1974) we prefer to puzzle about the lack of fit.
Sibling and ParentChild Data We propose that the major explanation for the unusually high correlations between unrelated sibs lies in their common rearing environments. Children adopted in the first year of life spend their developing years together, ~hether they are related or not. The relative advantages of their common environmentin home, neighborhood, schoolcan create strong pressures toward similar intellectual performance. Parents and children do not share a common rearing environment. Parents' relative intellectual level is fairly constant in adulthood. While their intelligence influences the rearing environment they provide for their related and unrelated "offspring," the parents are probably not greatly influenced by it. Thus, the relative similarity of related and unrelated children to their parents may depend almost equally on withinfamily environments and on genetic relatednessthe degree to which the children genotypically respond to the environment afforded by the parents. In the case of unrelated siblings, the withinfamily IQ variance was reduced to levels similar to those of related parentchild and natural sibling pairs (  . 6 ) . Since the genetic variance between unrelated siblings is nearly 100%, even given a slight degree of selective placement, the correlations in their IQ scores must come from their rearing environments. Not all children in the same family are treated in the same way by their parents, however, Parents can provide "compensatory programs" for children who threaten to intellectually lag behind other siblings in the family. Therefore, the withinfamily environments for adopted sib pairs may appear quite different to an observer but produce more similar intellectual outcomes than identical treatments of genetically different children would produce. In our interviews with the parents there were many anecdotal accounts of special tutoring and enrichments offered to adopted children who were suspected not to be intellectually as proficient as other children in the family. Several families with adopted children in the low average range of IQ scores were engaged in periodic evaluations of their progress in raising the child's intellectual level (with reported success, we might add). Similar compensatory programs were not offered to other siblings in the families. Nor were most families engaged in such active interventions. The general effects of successful parental efforts to alter some adopted children's intellectual levels would be to reduce the withinfamily IQ variance for unrelated sib pairs. The IQ rank of compensated children could change with
190
S. S C A R R A N D R. A. W E I N B E R G
respect to uncompensated ones in other families, thereby increasifig the betweenfamily variance. The parents in our sample who compensated their lower IQ adopted children did not themselves have higher IQ scores than other parents. Therefore, parentchild correlations would be unaffected by the parents' extra environmental manipulations, even if their efforts decreased the IQ variance among their unrelated children. Because this was not a longitudinal study, we cannot verify the claimed effectiveness of some parents' attempts to produce higher IQ levels in their adopted children. Nor can we demonstrate that these efforts explain, even partially, the high correlations among unrelated siblings. All that can be said is that the reported parental interventions, if successful, would produce the effects observed in the study.
Her#ability Estimates Heritability estimates are not safely generalized from the sibling data to the general population. Adoptive parents have special investments in their adopted children, or they would not have bothered to adopt. In the case of transracial families, there are even more keenly felt responsibilities for the intellectual (and other) development of their children. In other words, the families in this study were, we feel, highly invested in the intellectual success of all of their children, natural and adopted. We believe that siblings reared in these families have had intensive and extensive experiences that have pushed all of them toward similar levels of intellectual precocity. This is hardly typical of the entire Minnesota population. Heritability estimates, based on the parentchild data, are probably less biased against genetic variance. Intellectual differences among the parents did correlate with intellectual differences among their children, more highly for natural than adopted "offspring." In addition, the natural parent data provide a crucial check on the degree of selective placement and the heritability estimates from adoptive parent data. It is remarkable that the natural parentadopted child correlations were only slightly smaller in magnitude than the biologically related, adoptive parentnatural child pairs, even though the former do not live together while the latter do, and the educational levels of the natural parents are not as good an estimate of intellectual level as the IQ scores of the adoptive parents. Since the natural parentchild correlations yield heritability estimates comparable to the adoptive parentchild values, more credence should be placed on the range of h 2 values (.4 to .7) estimated from parentchild data. There are still some biases in these data, both for and against genetic variance, as noted in the results. On balance, we concluded that they provide reasonably coherent support for the moderate heritability of 1Q scores in a racially mixed sample of children in Minnesota.
I N T E L L E C T U A L SIMILARITIES WITHIN FAMILIES
19 
ACKNOWLEDGMENTS We are grateful for the assistance of Louise CarterSaltzman, Harold Grotevant, Margaret Getman, Marsha Sargrad, Patricia Webber, Joanne Bergman, William Thompson, and Carol Nelson.
REFERENCES Baumrind, D. Authoritarian vs. authoritative parental control. Adolescence, 1969, 3 (1), 120. Burks, B. S. The relative influence of nature and nurture upon mental development. A comparative study of foster parentfoster child resemblance and true parenttrue child resemblance. 27th Yearbook of the National Society for the Stud), of Education, 1928, Pt. 1, 219316. Burt, C. Intelligence and social mobility. British Journal of Statistical Psychology, 1961, 14, 324. CavalliSforza, L. L., & Feldman, M. W. Cultural versus biological inheritance: Phenotypic transmission from parents to children (a theory of the effect of parental phenotypes on children's phenotypes). American Journal of Human Genetics, 1973, 25, 618637. Eaves, J. J., & Jinks, J. L. IQ and inequality. Nature, 1974, 248, 287. Falconer, D. S. Introduction to quantitative genetics. New York: Ronald Press, 1960. Freeman, F. N., Holzinger, K. J., & Mitchell, B. C. The influence of environment on the intelligence, school achievement, and conduct of foster children. 27th Yearbook of the National Society for the Study o f Education, 1928, Pt. 1, 103217. Horn, J. M., Loehlin, J. C., & Willerman, L. The Texas adoption project. In L. CarterSaltzman (Chair), Adoption studies: Similarities of children to adoptive and natural parents. Symposium presented at the meeting of the Behavior Genetics Association, Texas, 1975. Jencks, C. Inequalio': A reassessment of the effect of family and schooling in America. New York: Basic Books, 1972. Jensen, A. R. Educability and group differences. New York: Basic Books, 1973. Kamin, L. J. The science and politics oflQ. Hillsdale, N.J.: Lawrence Erlbaum Assoc., 1974. Leahy, A. M. A study of certain selective factors influencing prediction of the mental status of adopted children. Journal of Genetic Psychology, 1932, 4 1 , 2 9 4  329. Loehlin, J. C., & Nichols, R. C. Heredity, environment, and personality: A study of 850 twins. Austin: University of Texas Press, 1976. McNemar, Q. Psychological statistics. Wiley: New York, 1962. Munsinger, H. The adopted child's IQ: A critical review. Psychological Bulletin, 1975, 82, 623659. Plomin, R. Personal communication, July 1976. ScarrSalapatek, S. Some myths about heritability and IQ. Nature, 1974, 251,463464. Scarr, S., & Weinberg, R. A. 1Q test performance of Black children adopted by wl~ite families: American Psychologist, 1976, 31, 726739. Skodak, M., & Skeels, H. M. A final followup study of one hundred adopted children. Journal of Genetic Psychology, 1949, 75, 85125. Wachs, T. Proximal experience and early cognitiveintellectual development: 1. The physical environment. Unpublished manuscript, 1975. Willerman, L. Personal communication, July 1976. Zajonc, R. B., & Markus, G. B. Birth order and intellectual development. Psychological Review. 1975, 82 (1), 7488.