DEMOGRAPHY© Volume 18,Number 4 November 1981

FAMILY SIZE AND THE QUALITY OF CHILDREN*

Judith BlakeSchool of Public Health and Department of Sociology, University of California, LosAngeles, California 90024

Abstract-If couples decide to have fewer children in order to achievehigher "quality" offspring, are they correct in assuming that the quality ofchildren bears an important and inverse relation to family size? If they arecorrect, how does number of children operate to affect individual quality?This research (using U.S. whites primarily) takes educational attainment(among adults) and college plans (among youngsters) as the principalindicators of quality, but also directs some attention to measures ofintelligence. The analysis supports the "dilution model" (on average, themore children the lower the quality of each child) and indicates that onlychildren do not suffer from lack of siblings, and that other last-borns arenot handicapped by a "teaching deficit." Number of siblings (relative toother background variables) is found to have an important detrimentalimpact on child quality-an impact compounded by the fact that, whencouples are at a stage in life to make family-size decisions, most back­ground factors (however important to the quality of their children) are nolonger readily manipulable. A special path analysis of college plans amongboys uses a modification ·of Sewell's Wisconsin Model as its base. Theresults show that number of siblings is a negative influence on interveningvariables affecting college plans. In general, the research documents theunfavorable consequences for individual siblings of high fertility, even in acountry that is (at least for whites) as socially, economically, and political­ly advantaged as the United States.

Many of us have spent the preponder­ance of our careers studying the determi­nantsiof human fertility. In so far as wehave been concerned about fertility deci­sion-making within unions in modernsocieties, we have reasoned that thenumber or quantity of children competeswith two other utilities for the resourcesat a couple's disposal-the nonchild util­ities, cars, houses, vacations, etc. onwhich the couple could focus its con­sumption, and the quality of children tobe produced. We have debated about therealism of these decisionmaking modelsand how they should be specified-what

*Presented as the Presidential Address to the Pop­ulation Association of America at its annual meet­ing in Washington, D.C., March, 1981.

the ingredients should be and what goeswhere in the chain of casuality.

Today, I will not be concerned withwhether these models bear much of arelationship to how couples really thinkin making family-size decisions. Nor,will I discuss whether consumption alter­natives to children are what principallydrive couples in their family-size deci­sions. What I do want to deal with is theassumption that people have fewer chil­dren in order to have higher quality ones.And, I want to turn things around a bitand ask the following question: If this isone of the things that people have inmind when they curtail their fertility, canthey be shown, on average, to be on theright track? For research purposes, thisquestion can be broken down into three

421

1. Is the relation of family size to childquality inverse?

2. Is family size important for the quali-ty of children?

3. If it is important, what does it affect?

My research on these questions has in­volved secondary analyses of many largescale surveys that have included numberof siblings as a variable. These surveyshave never before been analyzed fromthis point of view. Here I will give youselected results-it is impossible to dis-

DEMOGRAPHY, volume 18, number 4, November 1981

the dilution model takes no account ofinversely related to the influence of reproduction on the par-

ents' life chances. Parents' socio-eco-nomic status is assumed to be uninflu­enced by reproduction. It maybe dividedamong offspring, but it is not changed bythem. An additional possible feedbackthat the dilution model leaves out is thatsocialization of siblings by each othermay be indispensible to producing highquality children. Siblings-just a few,maybe even just one-may themselvesbe a condition of child quality, We knowthat most people believe this to be thecase-they believe that the only child isdisadvantaged because of a lack of sib­lings with whom to interact (Blake, 1981;Falbo, 1977; Thompson, 1974). Howev­er, our concern here is not what peoplebelieve, but whether they are correct.

Finally, the parental dilution modelhas nothing to say about possible birth­order effects. Within any family size,does it matter whether you are first,second, and so on? Logically, the modelimplies that child quality would go downwith each successive child, but the rateof decline would taper off after the sec­ond child, because each succeeding oneexperiences proportionately less of aloss. However, if there are feedbackeffects-if older children become pseu­do-parents, and/or if sib-socializationis anecessary condition of child quality,then quality might not decline with birthorder.

I shall now tum to my attempt toanswer the three questions raised earlier:

422subsidiary ones:

1. Is family sizechild quality?

2. Is family size an important influenceon child quality?

3. And, finally, if family size is impor­tant, what does it affect?What are themechanisms by which it operates?

I will take a moment for some concep­tual orientation. Why would anyonethink that the quality of children is in­versely related to their quantity? Theobvious answer is some form of the"dilution" model of parental inputs. Inconsidering this, we need to specify,albeit schematically, just what it is thatparents put in once the child has beenborn. We can say that parents provideenvironments or settings-types ofhomes, necessities of life, cultural ob­jects (like books, pictures, music, and soon). They also provide opportunities­specific chances to engage the outsideworld or, as kids say, "to get to dothings." Finally, they provide personalattention, intervention, and teaching (ei­ther directly or by example)-what hasbeen called "treatments" (Spaeth,1976).

The dilution model says, "The morechildren, the more these resources aredivided (even taking account of econo­mies of scale) and, hence, the lower thequality of the output." It is understoodthat by child "quality" we mean someobjective measure of human capital suchas educational or occupational attain­ment. No judgement is being made aboutthe intrinsic "worth" of one person overanother.

We must note as well that the dilutionmodel assumes that the casual arrows allgo in one direction-from the parents tothe children. The model does not assumeany feedback effects from children dur­ing the childrearing process. For exam­ple, it does not allow for the assumptionthat parents may create junior execu­tives out of older children-pseudo-par­ents on whom they can rely. Moreover,

Family Size and the Quality of Children

cuss all of the surveys in detail, or all ofthe results of anyone analysis. Theanalyses are for whites only, because ofinteractions by race. The principal sur­veys to be discussed are:

1. The National Opinion Research Cen­ter, General Social Survey, 1972-1980(pooled), approximately 8,000 menand women.

2. The 1970 National Fertility Study,approximately 4,000 women, plusdata on approximately 4,000 hus­bands.

3. The 1960 Growth of American Fam­ilies Study, approximately 2,600women.

4. The 1955 Growth of American Fam­ilies Study, approximately 2,350women.

5. A longitudinal study of 10th-gradeboys, Youth in Transition, approxi­mately 2,000 boys.

In the studies of adults, the indicatorof quality is "total years of educationachieved." For the 10th-grade boys, thedependent variable is "college plans."The independent variables, in addition tonumber of siblings, are indicators of therespondents' parents' socio-economicstatus, whether they were Catholic ornon-Catholic, Southern background,community size, respondent's age, andwhether the family was broken or intactwhen the respondent was growing up.Although these background variablesrepresent important controls, they donot, unfortunately, take full account ofpotentially negative selective influenceson respondents from one-child families.This family-size has not been a populargoal for couples, hence one would ex­pect some disproportionate negative se­lection into it-some children are single­tons because their defects led the parentsto decide to stop reproduction, or be­cause the parents themselves had physi­cal or psychological problems inimical toparenthood, or even because the parentsdid not get along well although theystayed together. Hence, we suspect that

423any given sample of only children in­cludes some disproportionate effects ofnegative selection which, although un­measured, should not be assumed to bean effect of number of siblings per se.

THE NATURE OF THE RELATIONSHIP

Let us tum first to the nature of therelation between family size and totalyears of education among adults. Fromhere on out I shall call family size "sib­size" since our point of reference is theoffspring, and its quality, as it is affectedby the number of its siblings. Does sib­size bear an inverse relation to totalyears of education? Our regression anal­yses (Figure 1) of the survey data amongboth men and women indicate, for eachsurvey, that the relation is, indeed, in­verse.

The only child does about equally wellas those from two-child families, or justslightly better, but beyond the two-childfamily, performance starts to drop off. Ibelieve that these results give quite con­sistent support to the dilution model.

Although, in this research, our "childquality" outcome variable is educationalattainment, we have necessarily paid at­tention to research on the relation ofsibsize to cognitive ability or intelli­gence. This is because intelligence isobviously an important influence on edu­cational attainment, and because a re­cent challenge to the pure dilution mod­el, by Robert Zajonc, has related to ameasure of intelligence (Zajonc and Mar­kus, 1975; Zajonc, 1976).

Considering intelligence for a moment,large-scale studies of school children inFrance (Institut National d'Etudes De­mographiques, 1973), Scotland (ThePopulation Investigation Committee andthe Scottish Council for Reserch in Edu­cation, 1949), and England (Douglas,1964; Eysenck and Cookson, 1970; Mar­joribanks, 1974) all suggest that, evenwithin social class levels, an inverse rela­tion exists between sibsize and a varietyof measures of cognitive ability. Thesedata thus support, for intelligence, the

.. ~

WOM

ENG

AF

19

55

WOM

ENG

AF

19

60

WOM

ENM

ENN

FS

19

70

WOM

ENM

ENNO

RC1

97

2-8

0

o m ~ o Cl~ ." ~ :< ~ c 3 CD

13.0

12.5

_12

.0

~11

.5

(f)

11.0

a::: ~10

.5>-

10.0

',

"....',.

;:...+

I'V

'

z13

.5Q ~ u ::> o L

JI

Figu

reI.

-To

tal

Yea

rsof

Edu

catio

nby

Sibs

ize

(Fam

ilySi

ze)

amon

gR

espo

nden

tsin

the

Gro

wth

ofA

mer

ican

Fam

ilies

Stud

ies

1955

and

1960

,th

e19

70N

atio

nal

Fert

ility

Stud

y(F

emal

eR

espo

nden

tsan

dD

eriv

edD

ata

onH

usba

nds)

,an

dth

eG

ener

alSo

cial

Surv

ey,

1972

-198

0,N

atio

nal

Opi

nion

Res

earc

hC

ente

r.R

esul

tsby

Sibs

ize

and

Bir

thO

rder

Hav

eB

een

Adj

uste

d,th

roug

hM

ultip

leC

lass

ific

atio

nA

naly

sis,

for

Par

ents

'Soc

ioec

onom

icSt

atus

,C

atho

lic-N

onC

atho

licR

elig

ion,

Res

pond

ent's

Age

,Sou

th­

em-N

onS

outh

emB

ackg

roun

d,C

omm

unity

Size

,and

Whe

ther

the

Par

ents

'Mar

riag

ew

asIn

tact

Whe

nth

eR

espo

nden

tW

asG

row

ing

Up.

Ana

lysi

sby

the

Aut

hor

isB

ased

onT

apes

from

the

Uni

vers

ityof

Wis

cons

inan

d,in

the

Cas

eof

the

NO

RC

1972

-198

0da

ta,

from

the

Inte

r-U

nive

rsity

Con

sort

ium

for

Polit

ical

and

Soci

alR

esea

rch,

see

Ack

now

ledg

men

ts.

.... !» :::J c: 3 a­ CD .. !A Z s CD 3 a­ CD .. .... -0 00 ....

Family Size and the Quality of Children

same kind of dilution effect that we haveseen to operate further down the chain ofcausality for educational attainment.

A notable exception to this supporthas been suggested by Zajonc in hisinterpretation of intelligence tests on al­most 400,000 Dutch males-survivors ofthe birth cohorts of 1944-47 in Holland(Zajonc, 1976). Zajonc emphasizes thatthe last child, and, by extension, the onlychild (who is the last child in a one childfamily), shows up poorly in those datawhen they are disaggregated by birth or­der. The data interpreted by Zajonc arefrom an article by Belmont and Marolla(1973) which, in tum, represented a re­analysis of a study by Stein, Susser,Saenger, and Marolla (1972; 1975).

As can be seen from Figure 2 the onlychild does only slightly better than thelast child in a three-child family, andwithin each family size, the results forthe last child take a sharp drop.

Zajonc believes that these data pointto the existence of a "teaching deficit"affecting last-born children-they haveno younger sibling to teach and, hence,their own learning is impaired (Zajonc,1976; Zajonc and Markus, 1975; Zajonc,Markus and Markus, 1979). In effect,Zajonc is saying that the dilution model,although in the main correct, requires usto take account of sib-socialization if it isto fit the Dutch data. Siblings do notmerely dilute parental resources; theyare resources for each other.

Before considering the Dutch case inmore detail, we should note that the poorshowing for the only and last child inthose data stands in marked contrast tothe results of other large surveys. Forexample, the French (Figure 3)and Scot­tish (Figure 4) surveys of school chil­dren, disaggregated by birth order, showthe only child to be highlyfavored and, ifanything, indicate an advantage for last­born children.

However, these results are for youngchildren-in Scotland the children were11, and in France they ranged from 6 to14. Hence, it can be argued that the data

425

suffer from an age-truncation bias. Forexample, when l l-year-olds are sam­pled, last-borns will naturally do betterthan first-borns in large families becauselast-born children in such families donot necessarily come from close-spacedsibsizes, but first-borns of that age mustcome from tightly spaced sibs.

For this reason, it is important tointroduce additional data that are lesssubject to such bias. The Youth in Tran­sition data set, mentioned earlier, ishelpful here. The Youth in Transitionwhite, American youngsters are aged 15and 16. Moreover, if we concentrate onsmall-to-moderate families, there is noreason to expect particular effects ofspacing, since American women havespaced their children relatively closelyeven when they had only two. The re­sults in Figure 5 come from our MultipleClassification Analysis which controlsfor parents' socio-economic backgroundand family intactness. It is evident thatthe same pattern exists as may be seen inthe Scottish and French data. Only chil­dren perform in a manner similar tothose from two-child families, and last­born children also appear to be relativelyadvantaged.

Another large study of Americanyoungsters by Hunter Breland (1974) isof interest here. Breland analyzed theNational Merit Scholarship QualificationTest scores for all participants tested in1965. Apparently no information wasavailable on parents' SES, race, or in­tactness of marriage, but it was possibleto tabulate the scores by family size andbirth order. There is a clear and statisti­cally significant negative relation ofscores with family size, except that, asmight be expected without controls forrace, socio-economic status, or familyintactness, the only child does not do aswell as first and second-borns from twoand three-child families. However, thereis no evidence of a sharp drop-off inperformance for the last child (Breland,1974, p. 1013).

Finally, since birth order as well as

426

2.5DEMOGRAPHY, volume 18, number 4, November 1981

FS2

FS32.6

3.3

3.4

6.0 1 I I I I I I I I I

1 2 3 4 5 6 7 8 9Birth Order

FS4.,

2.7 FS1- "

"FS5 '-"""-

2.8 FS6...,

\\ ,

cJ)

2.9a>~

0oena> 3.0oca>CJ'l FS9

a> 3.1-ct-l

3.2

Figure 2.-8cores on Raven Progressive Matrices Test by Sibsize (Family Size) and Birth Order among386,114 19-Year Old Dutch Males Who Were Survivors of Children Born During 1944-1947. Data from

Belmont and Marolla (1973), p. 1098.

Family Size and the Quality of Children 427

104

FS2/FS1

102FS3... • ......

100FS4"'- ~ ..

en 98<V~

0 FS5Uen<V

96oc<VCJ'I

94<V FS7- FS6C

.....-t

FS892

90 FS9

88

0' I , I , I I , I I

1 2 3 4 5 6 7 8 9Birth Order

Figure 3.-Scores on Benedetto Group Test of Intelligence by Sibsize (Family Size) and Birth Orderamong 100,000 French School Children Aged 6-14 in 1965. Data from Institut Nationale d'Etudes

Demographiques (1973), p. 67.

sibsize is available in the 1970 NationalFertility Study of adults, my MultipleClassification Analysis of this data setconsiders total years of education byboth sibsize and birth order, controlling

for all of the background variables men­tioned already. Figure 6 is for whitefemale respondents and Figure 7 is fortheir husbands.

Again, we see a very different pattern

428 DEMOGRAPHY, volume 18, number 4, November 1981

42 FS1LFS2----.

28

40FS3

V(/)Q)

FS4~

0o(f)

Q)(,) 34cQ)

Ol FS5.--Q)

+0- 32c....... FS7FS6

30

o 1 2 345Birth Order

6 7 8

Figure 4.-Scores on Group Test of Intelligence by Sibsize (Family Size) and Birth Order among ScottishSchool Children Aged 11 in 1947.Data from Population Investigation Committee and the Scottish Council

for Research in Education (1949), p. 107.

Family Size and the Quality of Children

108

from the Dutch data, and an analogousone to the results for youngsters. Single­tons are close to those from two-childfamilies, and last-borns frequently dobetter than middle children and some­times better than first-borns. Birth or­der is not, however, an important vari­able in our regressions.

429

These findings suggest that we need toreconsider the Dutch data insofar aspossible, since an analysis of almost400,000 cases cannot be ignored. In do­ing so, it is necessary to go back to theoriginal study of the young Dutchmenfrom which the results used by Belmontand Marolla, and then, derivatively, byZajonc, were generated.

The original research (Stein, Susser,Saenger and Marolla, 1975) was not con­cerned with the relation of intelligence tofamily size and birth order, nor were thedata tabulated in this fashion. Rather, itwas designed to analyze the effects ofmaternal malnutrition during pregnancyon the subsequent intellectual develop­ment of the children born to motherswho had suffered starvation during the1944-45 famine in parts of Holland. Thestudy design focused on famine and non­famine geographical areas of the coun­try, and on whether the children hadbeen carried, or born, during variousphases of the famine and post-famineperiod. The original analysis found nomeasurable effects of prenatal malnutri­tion on the intelligence of the survivingDutch men tested 18 or so years later.Then Belmont and Marolla (and laterZajonc) decided to use the Dutch datafor a very different problem-the rela­tion of intelligence to family size andbirth order.

However, these data had unrecog­nized defects for research on intelligenceby family size and birth order. This co­hort of almost 400,000 men (born be­tween 1944-1947) turns out to be a de­mographically skewed sample of birthsand survivors-biases that related par­ticularly to being an only and last-bornchild. The 1944-1945 famine drasticallyaffected conceptions and, thereby, thesubsequent birth rate in starvation areasof Holland (Smith, 1947a and 1947b;Stein, Susser, Saenger and Marolla,1975). Only-child and last-child situa­were thus created among those coupleswho were most severely affected by star­vation. Furthermore, very high rates of

,5

I , I I

1 2 3 4Birth Order

Figure 5.-Scores on Quick Test ofIntelligence bySibsize (Family Size) and Birth Order among 191210th-Grade Boys. Results by Sibsize and BirthOrder Have Been Adjusted, through MultipleClas­sification Analysis, for an Index of the Parents'Socioeconomic Status, Catholic-NonCatholic Reli­gion, Community Size, and Whether the Respon­dent's Parents' Marriage Was Intact. Analysis bythe Author is Based on a Tape of the University ofMichigan, Youth in Transition Survey, see Ac-

knowledgments.

114

113

FS2~

(J) 112Q)'-0u

en FS3111

Q)ocQ)

C1110

Q)-c......109 FS4

430

13.4

13.0

DEMOGRAPHY, volume 18, number 4, November 1981

FS312.6

FS4

c12.20

+- FS50U::J

'"0 11.8w'+-0

(/) 11.4 FS6~

0(l)

>-11.0 FS7

10.6

10.2

1 2 5Order

Figure 6.-Total Years of Education by Sibsize (Family Size) and Birth Order among 3868 FemaleRespondents Over Age 23 in the 1970National Fertility Study. Results by Sibsize and Birth Order HaveBeen Adjusted, through Multiple Classification Analysis, for the Respondent's Father's OccupationalPrestige Score, Respondent's Age, Religion in Which Reared, Whether She Lived with Both Parents atAge 14, Southem-NonSouthem Background, and Community Size. Analysis by the Author is Based on

Tapes from the University of Wisconsin, see Acknowledgments.

Family Size and the Qualityof Children

14.0FS2

FS1~13.6

431

FS3

13.2

c12.80-0 FS4o

::J"0 12.4W

FS5'+-0

en 12.0~

0 FS6Q)

>-11.6

FS7

11.2

1 3 4 5Birth Order

Figure 7.-Total Years of Education by Sibsize (Family Size) and Birth Order among 3782 Husbands(Aged 25 and over) of Respondents in the 1970 National Fertility Study. Results by Sibsize and BirthOrder Have Been Adjusted, through Multiple Classification Analysis, for the Husband's Father'sOccupational Prestige Score, Husband's Age, Religion in Which Reared, Whether He Lived With BothParents at Age 14, Southern-NonSouthern Background, and Community Size. Analysis by the Author is

Based on Tapes from the University of Wisconsin, see Acknowledgments.

IS SIBSIZE AN IMPORTANT INFLUENCEON CHILD QUALITY?

The family-size decisionmaking modelunder discussion assumes not only thatincreasing sibsize is a negative influenceon the quality of children, but that it is animportant influence. Does sibsize actual­ly have an important effect on childquality outcomes? An answer to thisquestion also comes from my regressionanalyses of the surveys of adults dis­cussed so far-the National Opinion Re­search Center, General Social Survey1972-1980; the 1970 National FertilityStudy; and the Growth of AmericanFamilies Studies for 1955 and 1960. Totalyears of education achieved was the de­pendent variable (indicating "quality")and sibsize, parents' socio-economic sta­tus, Catholic-nonCatholic religionSouthern-nonSouthern background, re­spondent's age, community size, andwhether the family was intact when therespondent was growing up were thepredictors. In all of these studies, sibsizeand father's socio-economic status werethe principal variables influencing therespondent's educational attainment. Inmost cases, the relative influence of sib­size, as shown by the beta coefficients,was equal to, in one case greater than,and in one case slightly less than ourindicator offather's SES.

In terms of standardized coefficients,what this means is that, other thingsbeing equal, every increase in a standardunit of sibsize must be offset by anincrease in a standard unit of father'sSES, if the person's educational attain­ment is not to decline. Put in terms ofactual metrics instead of standardizedcoefficients-numbers of children andpoints on the occupational scale for fa­thers-in the NORC 1972-1980 data, giv­en an average score on all other variables

432 DEMOGRAPKY, volume 18, number 4, November 1981

infant and child mortality at this time, not suggest that last-borns on averagetogether with high stillbirth rates, also suffer from the lack of siblings to teach,helped determine that children would or from anything else that is very influen­grow up as singletons and last-borns- tial (Blake, 1981; Claudy, Farrell, andagain particularly in those families worst Dayton, 1979;Falbo, 1977; Falbo, 1978).hit by the famine. Children carried dur­ing the famine were selectively subject tomortality, and those born during it werefatally vulnerable (Smith, 1947a and1947b; Stein, Susser, Saenger and Mar­olla, 1975). Finally, the joint probabilityof survival of the parents decreased,since the death rate rose among adults aswell. This effect was gravest where peo­ple were hit worst by the famine.

In sum, only and last-born children inthis cohort of young Dutch men appearto have been negatively selected-s-onaverage they were more likely to cornefrom families that suffered the worst.That is how, in part, many of thembecame singletons and last-borns-theirparents had acute fecundity problemsdue to starvation, or their newly bornsiblings were stillbirths or died in infan­cy, and/or their parents were widowed,or they themselves were totally or­phaned. As a consequence of the selec­tive influence of deprivation on theseyoung people during infancy, they, likeother youngsters who have sufferedacute starvation, were more prone tocognitive deficiencies. In substantiationof this interpretation, we may note thatBelmont documents higher levels ofmental retardation and need for psychi­atric care among singletons and last­borns of all families in this study, al­though she does not interpret this finding(Belmont, 1977).

I would conclude, therefore, that theanomaly of marked intellectual deficien­cies among singletons and last-bornsfound in the Dutch data does not requirea substantive revision of the dilutionmodel. We do not need to revise it toinclude a sib-socialization component­at least on the basis of these results.Rather, most studies find only childrento be either advantaged, or no differentfrom those in two-child families, and do

Family Size and the Quality of Children

in the model, a white male with nosiblings would graduate from high schoolif his father scored 17 on the DuncanOccupational Prestige Scale, but if hehad four siblings his father would haveneeded a score of 37 in order for the sonto achieve high-school graduation. It isworth noting that, in the OccupationalChanges in a Generation analyses byFeatherman and Hauser, the coefficientsshowing the relative importance of sib­size and father's occupational status (inrelation to respondent's educational at­tainment) are similar to our analyses(Featherman and Hauser, 1978, pp. 242­243).

There thus seems to be little questionthat, among whites, even when majorcultural, period, and socio-economicbackground variables are taken into ac­count, increasing sibsize is an importantnegative influence on a person's educa­tonal attainment. Moreover, our resultssuggest strongly that there is no familysize that is too small for the productionof quality. The only child, pronatalistideology to the contrary notwithstand­ing, is shown in study after study to be asadvantaged, or more so, as children fromtwo-child families, and clearly more ad­vantaged than children from larger sib­sizes.

Hence, we seem to have an answer totwo of our questions. If people believethat they can trade off child quantity forchild quality they are, indeed, on theright track; and sibsize is an importantinfluence on the quality of children. Infact, sibsize is more important than ourregression coefficients indicate. This isbecause, at the time of the onset ofchildbearing (at the time when family­size decisions are most crucial), a cou­pIe's family size is still a decision vari­able, whereas a major share of the socio­economic status they will transmit totheir offspring has been determined. So,choosing their family size is a way inwhich all parents can still affect the qual­ity of their children, regardless of therelative importance of their socio-eco-

433

nomic position which is, or rapidly onthe way to becoming, a fait accompli.

THE INFLUENCE OF SIBSIZE ONINTERVENING VARIABLES

Although it is important to establishthe nature of the relation between sibsizeand a child's educational attainment, andit is also necessary to assure oneself thatsibsize is not a relatively trivial influenceon achievement, a primary focus of ourinterest is in specifyingjust what it is thatsibsize affects. In what manner doessibsize impinge on variables interveningbetween parental background and educa­tional outcomes?

Fortunately, there is a distinguishedresearch literature documenting suchvariables. This work, pioneered by Wil­liam Sewell and various colleagues (Sew­ell and Shah, 1967; Sewell and Shah,1968a; Sewell and Shah, 1968b; Sewell,Haller, and Portes, 1969; Sewell, Haller,and Ohlendorf, 1970; Sewell, 1971; Sew­ell and Hauser, 1972; Sewell and Hauser,1975: Sewell, Hauser, and Featherman,1976; Sewell and Hauser, 1980) has notuntil very recently (Sewell, Hauser, andWolf, 1980) included sibsize as a predic­tor. However, the basic "Wisconsinmodel" has shown the significance ofspecifying the influence of parental back­ground (occupational and/or educationalachievement) on intervening variablessuch as perceived parental pressure foracademic achievement, child's ability,and child's grades as these impinge onthe child's educational aspirations (and,ultimately, on educational attainment).

I have adapted the Wisconsin model toinclude sibsize and to disaggregate "set­tings" (a 19-item index of the culturaland physical richness of the home envi­ronment) from an overall parental socio­economic index. The purpose here is toattempt to measure how parents' socio­economic level is actually expressed inthe home environment, and takes advan­tage of a critique of the Wisconsin modelby Joe Spaeth (1976). The parents' levelof social status is indicated by their edu-

PA

RE

NT'

SE

DU

CA

TIO

N

SIB

SIZ

E

GR

AD

ES

Fig

ure

8.--

Sch

emat

icP

ath

Dia

gram

ofE

ffec

tsof

Sel

ecte

dV

aria

bles

onC

olle

gePl

ans

amon

g10

thG

rade

Boy

s,U

nive

rsit

yo

fMic

higa

n,Y

outh

inT

rans

itio

nda

tata

pe.

.. ~ C m ~ o a ~ ." :r: .:< ~ c :I CD ... !» :s c :I a­ CD .. ~ z ! :I i" .. ... ;g ...

Fig

ure

9.-

Path

Dia

gram

ofth

eE

ffec

tson

Col

lege

Plan

s(Y

)of

Par

ents

'E

duca

tion

(XI)

,S

ibsi

ze(X

2),

Set

ting

s(X

3),T

reat

men

ts(}

4),

Abi

lity

(X~),

Mon

eyS

aved

(~),

and

Gra

des

(X7

).A

naly

sis

byth

eA

utho

ris

Bas

edon

aT

ape

ofth

eU

nive

rsit

yof

Mic

higa

n,Y

outh

inT

rans

itio

nS

urve

y.

'I' ~.

.:c 5!! = a a- f D e.. ~. 9.. 2 0:: a; :I t VI

.r:o.

w '"

Tab

lel.

-Co

rrel

atio

nM

atri

x.V

aria

bles

inPa

thA

naly

sis

ofY

outh

inT

rans

ition

10th

Gra

deB

oys

Xl

X 2X 3

X 4X 5

X 6X 7

PA

RE

NT'

SED

UC

ATIO

NX

l1.

0000

c m

SIB

SIZ

EX 2

-0.1

558

1.00

00~ 0 0

SETT

ING

SX 3

0.32

82-0

.129

91.

0000

~ "V

X 40.

2056

1.00

00:::I

:TR

EATM

ENTS

0.22

79-0

.163

8.:< <

0.19

980.

2133

1.00

000

AB

ILIT

YX 5

0.32

59-0

.209

6C 3 CD

SAVE

DM

ONEY

X 60.

1950

-0.2

014

0.21

410.

1400

0.13

051.

0000

... !D

0.20

82-0

.156

50.

1468

0.21

620.

3881

0.20

021.

0000

:::sG

RADE

SX 7

c 3 a-CO

LLEG

EPL

ANS

Y0.

2806

-0.1

623

0.22

480.

3644

0.32

060.

1792

0.32

51CD .. ~ Z ~ CD 3 a- CD .. ... -0 G

O ...

"TI

Tab

le2.

-Dec

ompo

siti

onof

Eff

ects

ina

Path

Mod

elof

Col

lege

Plan

s:10

thG

rade

Boy

s,Y

outh

inD

Tra

nsiti

on~o -< '"

DEPE

NDEN

TPR

EDET

ERM

INED

TOTA

LIN

DIRE

CTEF

FECT

SVI

ADI

RECT

N°

CDVA

RIAB

LEVA

RIAB

LEEF

FECT

X 2X 3

X 4X 5

X 6X 7

EFFE

CTD :::

ID

.. -::rSI

BSI

ZEX 2

PARE

NT'S

EDX

l-.

1558

-.15

58CD

----

----

----

() cSE

TTIN

GSX 3

PARE

NT'S

EDXl

.328

2.0

126

----

----

--.3

156

a ~oSI

BSIZ

EX 2

-.08

07-.

0807

0--

----

----

--- n

TREA

TMEN

TSX 4

PARE

NT'S

EDXl

.227

9.0

205

.207

4:.

----

----

--a:: .. CD

SIBS

IZE

X 2-.

1315

-.13

15:::

I--

----

----

--A

BILI

TYX 5

PARE

NT'S

EDXl

.325

6.0

251

.023

4.0

243

----

--.2

528

SIB

SIZE

X 2-.

1627

---.

0060

-.01

54--

----

-.14

13

SETT

INGS

X 3.0

744

----

----

----

.074

4

TREA

TMEN

TSX 4

.117

3--

----

----

--.U

73

GRAD

ESX 7

PARE

NT'S

EDXl

.208

2.0

198

.017

7.0

324

.082

6--

--.0

557

SIB

SIZE

X 2-.

1272

---.

0046

-.02

05-.

0462

----

-.05

59

SETT

INGS

X 3.0

560

----

--.0

243

----

.031

7

TREA

TMEN

TSX 4

.156

4--

----

.038

3--

--.U

81

ABIL

ITY

X 5.3

267

----

----

----

.326

7~ w "00

1

Tab

le2.

-(C

onti

nued

)

.a:..

w 00

DEPE

NDEN

TVA

RIAB

LEPR

EDET

ERM

INED

VARI

ABLE

TOTA

LEF

FECT

X 2X 3

INDI

RECT

EFFE

CTS

VIA

X 4X 5

X 6X 7

DIRE

CTEF

FECT

COLL

EGE

PLAN

SY

C mPA

RENT

'SED

Xl.2

806

.018

9.0

314

.060

7.0

499

.005

2.0

094

.105

1~ 0

SIB

SIZE

X 2-.

1215

-.00

80-.

0385

-.02

79-.

0077

-.00

95-.

0299

C)

--~ ."

SETT

INGS

X 3.0

991

----

--.0

146

.007

3.0

054

.071

3% .:< <

TREA

TMEN

TSX 4

.293

1--

----

.023

1.0

028

.020

1.2

471

0 C 3A

BILI

TYX 5

.197

3.0

012

.055

6.1

405

CD--

----

--... !JO

MONE

YSA

VEL)

X 6.0

504

----

----

----

.050

4:;, c 3

GRAD

ESX 7

.169

9.1

699

rr--

----

----

--CD .. ~ Z 0 < CD 3 rr CD .. ... -0 00 ...

Family Size and the Quality of Children

cational attainment. "Treatments" is ameasure of the youngster's perception ofactive personal encouragement by par­ents for college attendance. The child'sability is indicated by a combined indexof cognitive achievement tests, andgrades are those for the previous year. Ihave also included a variable relating towhether the youngster said he had or hadnot saved money, on the assumption thatthis should have a relationship to thedependent variable, college plans. Thedata base is the University of Michiganstudy of 10th-grade boys, Youth in Tran­sition.

The schematic path diagram in Figure8 indicates the main features of the theo­retical model. My hypothesis is that theoverall importance of sibsize inheres vir­tually entirely in its indirect effects onthe intervening variables. In otherwords, there seems to be no theoreticalreason why sibsize should have a directeffect on educational aspirations (or, forthat matter, on educational attainment).In the model, parents' educational leveland sibsize are predetermined. Thehome environment (settings) and thechild's perception of parental encourage­ment to go on to college (treatments) areassumed to be antecedent to ability,money saved, and grades. My reason forplacing perception of parental encour­agement prior to ability is that I thinkparental encouragement doubtless mate­rially affects measured ability, and al­though encouragement is also assuredlystimulated by ability, it probably is influ­enced more by nonobjective factors thanby testing results. Hence, in a recursivemodel, I think it can be argued that oneerrs less than by assuming that encour­agement depends on whether the child isof a certain demonstrable level of intelli­gence.

Figure 9 shows the path diagram, Ta­ble 1 (the correlation matrix), and Table2 (the decomposition of effects in theoverall path analysis). A number ofpoints are worth noting here. First,among the intervening variables, "treat-

439

ments" has the most important total anddirect effect on college plans. Second, ofthe two predetermined variables as­sumed to influence treatments (parents'education and sibsize), sibsize has a neg­ative influence on "treatments" that isalmost half as large as the positive effectof parents' educational level (coefficientfor sibsize is - .0385 and for parents'education level, .0607). Third, the nextmost important intervening variable isability and, in this case, too, sibsize has anegative effect that is almost half as greatas the positive effect of parents' educa­tionallevel (the coefficient for sibsize is- .0279and for parents' educational levelit is .0499). Sibsize also has a negativeeffect on ability that is slightly largerthan the positive effect of "treatments."For the remaining variables, sibsize hassome negative effects on every variablein the model.

Interestingly, the analysis suggeststhat "settings"-cultural and physicaladvantages in the home-are not nearlyas important for academic aspirations asis the child's perception of academicencouragement by parents-"treat­ments." This indicates that parents can­not very successfully offset the negativeeffect of sibsize on "treatments" by sim­ply providing "settings." In effect, chil­dren are more motivated by parentalattention, interaction and encourage­ment, than by passive environments ofcultural and material "advantage." Inso­far as increasing sibsize dilutes thatuniquely important personal input, it hasa major negative consequence for whichit is apparently difficult to compensate.

Subsequent to this analysis, WilliamSewell, in a personal communication,has called my attention to the 18-yearfollow-up.study of the original Wisconsinhigh school seniors on which most of hiselaboration of the Wisconsin model wasbased (Sewell, Hauser, and Wolf, 1980).This follow-up analysis includes sibsizefor the first time and involves a consider­able expansion of the original model. Ofinterest, here, however, is the fact that,

440 DEMOGRAPHY, volume 18, number 4, November 1981

with regard to college plans as a depen­dent variable, the results are very similarto our path analysis of the Youth inTransition data (Sewell, Hauser, andWolf, 1980, p. 566). Parental encourage­ment is the most important influence inthe Wisconsin data on men, and amongthe predetermined variables affecting pa­rental encouragement, sibsize ranksclose to father's occupational status inimportance (the coefficient for sibsize is- .106 and for father's occupational sta­tus it is .150).

SOME IMPLICATIONS OF THISRESEARCH

Since my discussion is a brief previewof an extensive research undertaking, letme close now with some selected conclu­sions about and implications of the workoutlined here.

First, I think it documents that, inso­far as people generally think in terms of adilution model, they are right. Moreover,at least with regard to the variables stud­ied here, there seems to be, on average,no important positive feedback from thesocialization by siblings of each other.That is, neither do last-boms suffer fromnot having a sibling to teach, nor do theygain in important ways from having olderbrothers and sisters. This is suggested bythe fact that the regressions we did oneach sibsize separately, which includedbirth order as a variable, indicated that itwas the least important predictor in theequations and, frequently, not even sta­tistically significant. These results, forintelligence and educational attainment,appear to extend to other dependentvariables in our work such as life satis­faction, social adjustment, alienation andoccupation achievement (Blake, 1981).

Findings may be different where sib­lings in the same family are studied. But,the information so far available on unre­lated people implies that the effect ofsibsize far transcends any effects of birthorder, and that unless one had some verycogent theoretical reasons for studyingbirth order, the empirical results would

hardly spur one on. Reviews of some ofthe problems involved in research on theeffects of birth order may be found inAdams, 1972; Clausen, 1965; Schooler,1972;and Terhune, 1974.

Second, by extension, our researchindicates that the only child is not disad­vantaged because of the absence ofbrothers and sisters. Although singletonsdo not show up in all of the studies asmore advantaged than children fromtwo-child families, and sometimes theyappear to do slightly less well, I person­ally believe that this is because of selec­tion, as I have already discussed. In allstudies (with adequate controls) but theDutch one, only children do very wellindeed and, typically, surpass those fromfamilies of more than two siblings. Theseresults thus suggest that, if couples wishto have only one child, it seems reason­able to expect no deleterious conse­quences. Indeed, my work on education­al level is further bolstered by separateanalyses of other characteristics of theonly child (Blake, 1981;Claudy, Farrell,and Dayton, 1979; Falbo, 1977 and1978).

Third, these findings imply that peo­ple's family-size decisions, in the con­text of the typical life cycle, are the mostimportant of the background influenceson child quality we have studied. Notonly can sibsize be shown to competestrongly with parental SES on a statisti­cal basis, but we must remember that, bythe time couples are deciding how manychildren they want, their socio-economicstatus is typically no longer very readilyaffected by choice. Equally, if we con­sider other variables in our model, al­though one is better off having had Jew­ish than Catholic parents when it comesto educational attainment, such religiousdifferences are not readily rearranged asa couple is about to embark upon repro­duction. A young Catholic couple nor­mally does not have the option of sud­denly taking on all the unmeasuredcharacteristics associated with beingJewish. This couple does have the choiceof not having six children, however.

Family Size and the Quality of Children

Fourth, our model of the interveningvariables that sibsize affects indicatesthat it may not be so easy to avoid thenegative consequences of large families,even if one is well-off. On average, chil­dren from large families have less ability,lower grades, and apparently receiveless academic encouragement from theirparents, even when parental backgroundand "advantages" in the home are con­trolled. Large families are considerablymore deleterious to a child's educationalattainment than are broken homes, sug­gesting that, on average, a simple nos­trum of familistic traditionalism-stablemarriages and high fertility-is not theticket to producing quality offspring.

It is interesting to ask why the effect ofbroken homes is not greater, if the dilu­tion model is valid. One reason may bethat, although a mate (typically a hus­band, since most children of brokenhomes were with their mothers) presum­ably contributes to the personal inter­ventions and treatments that are so im­portant in childbearing, he also makesdemands himself. He is a diluter of amother's attention and personal re­sources, as well as a contributor to child­rearing. Moreover, a certain amount ofthe theoretical advantage of unbrokenhomes may be drained off in negotiationand transaction costs between the par­ents as they attempt to cope with child­rearing jointly. It may thus be that theadvantages of single parenthood partiallyoffset the disadvantages, whereas on av­erage there appear to be no similarlyoffsetting compensations in the case oflarge versus small families.

Finally, although it is inappropriate toextrapolate the conclusions of this re­search to developing countries where thefamily structures are so different fromour own, it seems inconceivable that theresults are totally irrelevant either. ThefindingsI have discussed relate to a periodwhen, for whites at least, the UnitedStates was the most open, richest, andsociologically and politically advantagedcountry in the world. Yet, even underthese relatively ideal conditions, coming

441

from a large family had negative effectsonthe quality of the people we have studied.It is hard to imagine that, in countriesmuch less favored, differences in kinshipstructure can offset the deleterious impacton human capital of large family size.Hence, one practical advantage of re­search such as this may be to add to theall-too-modest empirical literature on theindividual consequences of high fertility­a literature that will not necessarily influ­ence the bulk of people in developingcountries, but may have some effect ontheir policy-makers. Personally, I doubtwhether, on average, any country, or anyethnic or racial group, can meet the hu­man capital demands of the modem worldand simultaneously maintain a statisticalnorm of large families.

ACKNOWLEDGMENTS

The research on which this paper isbased began as a result of work on theonly child funded by NICHHD contractHD82802. The research has also beenfunded, in part, by the Fred H. BixbyFoundation through the instrumentalityofthe Fred H. Bixby Chair. I am gratefulto Sandra Rosenhouse and Jorge DelPinal for help with data processing andstatistical analysis, and to Elizabeth Ste­phenson of the Data Archives Library,Institute for Social Science Research atUCLA for her assistance in acquiringand making available the data tapes ofthe 1955 and 1960 Growth of AmericanFamilies Studies and the 1970 NationalFertility Study; and to the Inter-Univer­sity Consortium for Political and SocialResearch (based at the University ofMichigan) for the data tapes of the Na­tional Opinion Research Center, GeneralSocial Survey 1972-1980 in OSIRIS for­mat, as well as for the tapes of thelongitudinal survey, Youth in Transition.

REFERENCES

Adams, Bert N. 1972. Birth Order: A CriticalReview. Sociometry 35:411-439.

Belmont, Lillian. 1977. Birth Order, IntellectualCompetence, and Psychiatric Status. Journal ofIndividual Psychology 33:97-104.

Belmont, Lillian and Francis A. Marolla. 1973.

442 DEMOGRAPHY, volume 18, number 4, November 1981

Birth Order, Family Size, and Intelligence. Sci­ence 182:1096-1101.

Blake, Judith. 1981. The Only Child in America:Prejudice versus Performance. Population andDevelopment Review 7:43-54.

Breland, Hunter. 1974. Birth Order, Family Con­figuration, and Verbal Achievement. Child De­velopment 45:1011-1019.

Claudy, J. G., W. S. Farrell, Jr., and C. W.Dayton. 1979. The Consequences of Being anOnly Child: An Analysis of Project Talent Data.Palo Alto, California: American Institutes forResearch.

Clausen, John A. 1965. Family Size and BirthOrder as Influences upon Socialization and Per­sonality: Bibliography and Abstracts. NewYork: Social Science Research Council.

Douglas, J. W. B. 1964. The Home and the School.London: Macgibbon and Kee.

Eysenck, H. J. and D. Cookson. 1970. Personalityin Primary School Children: 3-Family Back­ground. British Journal of Educational Psycholo­gy 40:117-131.

Falbo, T. 1977.The Only Child: A Review. Journalof Individual Psychology 33:47-61.

--. 1978. Sibling Tutoring and Other Explana­tions for Intelligence Discontinuities of Only andLast Borns. Journal of Population 1:349-363.

Featherman, D. L. and R. M. Hauser. 1978. Op­portunity and Change. New York: AcademicPress.

Institut National d'Etudes Demographiques, 1973.Enquete Nationale sur Ie Niveau Intellectuel desEnfants d'Age Scolaire. Cahier No. 64. Paris:Presses Universitaires de France.

MaIjoribanks, Kevin. 1974. Academic Achieve­ment: Family Size and Social Class Correlates.In: K. Marjoribanks (ed.), Environments forLearning. Windsor, Berks: NFER Publishing.

Population Investigation Committee and the Scot­tish Council for Research in Education. 1949.The Trend of Scottish Intelligence. London:Universitv of London Press Ltd.

Schooler, Carmi. 1972. Birth Order Effects: NotHere, Not Now! Psychological Bulletin 78:161­175.

Sewell, William H. 1971. Inequality of Opportunityfor Higher Education. American SociologicalReview 36:793-809.

Sewell, William H. and Robert M. Hauser. 1972.Causes and Consequences of Higher Education:Models of the Status Attainment Process. Amer­ican Journal of Agricultural Economics 54:851­861.

--. 1975. Education, Occupation and Earn­ings: Achievement in the Early Career. NewYork: Academic Press.

--. 1980.The Wisconsin Longitudinal Study ofSocial and Psychological Factors in Aspirationsand Achievements. In Alan C. Kerckoff (ed.),Research in Sociology of Education and Social­ization. Greenwich, Conn.: JAI Press.

Sewell, William H. and Vimal P. Shah. 1967.

Socioeconomic Status, Intelligence and the At­tainment of Higher Education. Sociology of Edu­cation 40:1-23.

--. 1968a. Social Class, Parental Encourage­ment and Educational Aspirations and Achieve­ments. American Journal of Sociology 73:559­572.

--. 1968b. Parents' Education and Children'sEducational Aspirations and Achievements.American Sociological Review 33:191-209.

Sewell, William H., Archibald O. HaIler, andGeorge W. Ohlendorf. 1970. The Educationaland Early Occupational Status Attainment Proc­ess: Replication and Revision. American Socio­logical Review 35:1014-1027.

Sewell, William H., Archibald O. HaIler, and Ale­jandro Portes. 1969. The Educational and EarlyOccupational Attainment Process. American So­ciological Review 34:82-92.

Sewell, William H., Robert M. Hauser, and DavidL. Featherman. 1976. Schooling and Achieve­ment in American Society. New York: Academ­ic Press.

Sewell, William H., Robert M. Hauser, and WendyC. Wolf. 1980.Sex, Schooling, and OccupationalStatus. American Journal of Sociology 86:551­583.

Smith, Clement A. 1947a. Effects of MaternalUndernutrition upon the Newborn Infant in Hol­land (1944-45). The Journal of Pediatrics 30:229­243.

--. 1947b. The Effect of Wartime Starvation inHolland Upon Pregnancy and Its Products.American Journal of Obstetrics and Gynecology53:599-608.

Spaeth, Joe L. 1976. Cognitive Complexity: ADimension Underlying the SocioeconomicAchievement Process. In William Sewell, RobertM. Hauser, and David L. Featherman (eds.),Schooling and Achievement in American Socie­ty. New York: Academic Press.

Stein, Zena, Mervyn Susser, Gerhart Saenger, andFrancis Marolla. 1972. Nutrition and MentalPerformance. Science 178:708-713.

--. 1975. Famine and Human Development:The Dutch Hunger Winter of 1944-1945. NewYork: Oxford University Press.

Terhune, K. W. 1974. A Review of the Actual andExpected Consequences of Family Size, Reportto The Center for Population Research, NationalInstitute of Child Health and Human Develop­ment.

Thompson, V. D. 1974. Family Size: Implicit Poli­cies and Assumed Psychological Outcomes.Journal of Social Issues 30:93-124.

Zajonc, R. B. 1976. Family Configuration andIntelligence. Science 192:227-236.

Zajonc, R. B. and Gregory B. Markus. 1975. BirthOrder and Intellectual Development. Psycholog­ical Review 82:74-88.

Zajonc, R. B., H. Markus, and G. B. Markus.1979.The Birth Order Puzzle. Journal of Person­ality and Social Psychology 37:1325-1341.