Download as pdf or txt
Download as pdf or txt
You are on page 1of 26

Family Environment and Cognitive Development: Twelve Analytic Models

Author(s): Herbert J. Walberg and Kevin Marjoribanks


Source: Review of Educational Research , Autumn, 1976, Vol. 46, No. 4 (Autumn, 1976),
pp. 527-551
Published by: American Educational Research Association

Stable URL: https://www.jstor.org/stable/80000073

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

American Educational Research Association is collaborating with JSTOR to digitize, preserve and
extend access to Review of Educational Research

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
Review of Educational Research
Fall 1976, Vol. U6y No. 4, Pp. 527-551

Family Environment and Cognitive


Development: Twelve Analytic Models

Herbert J. Walberg

University of Illinois at Chicago Circle

Kevin Marjoribanks

Oxford University*

Although educators would agree that the family environment


influences the development of children's cognitive abilities, psy
chologists and sociologists are only now beginning to understand
the specific characteristics that affect the acquisition of ability.
Obstacles to research include invalid measurements of family
characteristics and abilities, the multiplicity of confounded fac
tors in the home, school, community, and peer groups, the
expense of large-scale longitudinal studies of families in mobile
societies, and the near impossibility of major experimental ma
nipulation of family characteristics. Moreover, correlational or
causal relationships established for one group may not hold for
other times, social classes, ethnic groups, or countries.
Nonetheless, there have been some recent developments in
research that might eventually lead to policy implications on

*Now at the University of Adelaide, Australia

527

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
REVIEW OF EDUCATIONAL RESEARCH Vol. 46, No. 4

questions such as the "ideal" family environment for the foster


ing of abilities. The research has resolved some former para
doxes, produced more valid measurements and analyses, revealed
some trends with possible educational implications, and gener
ated some new problems requiring additional research. Some of
these research developments are described in the present review,
with a particular emphasis on relationships among socioeco
nomic status, family size, the sociopsychological environment of
the family, and the child's cognitive development. Mention is
made also of school, classroom, peer group, and community
factors in relation to ability. In the review, measures of aptitude,
ability, and educational achievement are generally considered,
unless otherwise noted, to index a single construct, namely,
verbal-educational ability. Although many other traits of chil
dren are of interest, most of the large scale psychological and
sociological research on the effects of family environments on
cognitive development is limited to standardized tests that
mainly tap verbal-educational ability. Accordingly, the present
review is also limited in this sense.
In what follows, twelve models for research on family environ
ments and cognitive ability are discussed in a sequence ap
proximating their historical emergence. These models are shown
in Figure 1 and referred to in the text by number.

Models 1-3: Socioeconomic Status, Family Size, and Ability


About 1925, three statistical findings confronted psychologists
in Western Europe and North America:
1. Sibsize (the number of children in the family) had been
found to correlate about -.3 with the mean ability scores of the
children in the family; that is, the larger the sibsize, the lower
the children's abilities (Model 1)?a weak relationship accounting
for 9% of the variance (the square of the correlation) and sub
ject to many exceptions.
2. Children's abilities had also been found to correlate about
.4 with parental socioeconomic status (SES). That is, SES ac
counted for about 16% of the variance in ability scores (Model 2).
3. On the average, parents of lower SES groups were found
to marry earlier than others, bear children more rapidly, and
continue having them to a later age. On hereditarian grounds,
some psychologists inferred that because of such differential re
productive rates, the national intelligence might be declining as
much as 2 or 3 points per generation (Model 3).
Several large scale surveys were carried out to investigate the
possibility of declining national ability. Samples of children in

528

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
WALBERG & MARJORIBANKS FAMILY ENVIRONMENT

1. Family size ? ability


2. Socioeconomic status - ability
3. Socioeconomic status - family size -changing national
ability
4. Socioeconomic status ? family size increasing variation
in ability
5. f(socioeconomic status, family size)-^ability
6. Family environment-global ability
7. Family environment-specific abilities
8. Ethnicity-^ environment-^abilities
9. f(socioeconomic status, family size)-environment
10. Ability,--Ability2

socioeconomic
status
family size

11. Ability, Ability2

family
school
classroom'
community
peer group

12. f (Abilities,, Environments) - Abilities2

Figure 1.
Some models for research on family characteristics and cognitive ability

relatively stable population districts in England and in all of


Scotland were obtained at two times more than 10 years apart
(Scottish Council, 1949; Emmett, 1950). To the surprise of the
original investigators, the ability level had risen slightly. An
environmental explanation was soon put forward to defend the
validity of the hereditarian inference: namely, that while the
national gene pool of ability was deteriorating, education was
improving, and the sum of the two opposing effects maintained
the level of measured ability or raised it slightly.

529

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
REVIEW OF EDUCATIONAL RESEARCH Vol. 46, No. 4

However, it was shown later that such longitudinal studies can


be highly misleading when changes in national gene pools of
ability are investigated. Spurious results are obtained because
correlations between the mean ability of siblings in the family
and the number of siblings fail to take into account the noncon
tribution of barren adults to gene pools (Cole, 1954). Anastasi
(1956) also indicated that spurious findings would result if the
research failed to start with a sample of adults with and without
children and if it did not obtain a record of their abilities and did
not ascertain fertility rates at various levels of ability. Two
research groups recognized and avoided these errors. Higgens,
Reed, and Reed (1962) sampled children and grandchildren of
nonepileptic patients in a Minnesota psychiatric hospital, and
Bajema (1963) studied the offspring of former public school
children in Kalamazoo, Michigan.
Both studies agree that adults at the highest level of IQ (above
130) have the highest average fertility rates, about three off
spring, and that feebleminded adults (IQ 55 and below) have very
low fertility. These studies suggest that for the populations
surveyed, average IQ, insofar as it is genetically determined, had
been at least stable and probably rising.

Model 4: Socioeconomic Status, Sibsize, and Increasing IQ


Variation
Another interesting though unanticipated finding (Model 4) in
both the Minnesota and Michigan studies was bimodal (double
peaked) fertility rates across the range of IQ levels. In the larger
study of 1,966 individuals (Higgens et al., 1962), for example, the
group with IQs from 56 to 85 had reproductive rates of 2.42; the
group from 86 to 115, 2.21; the group above 116, 2.60. Bajema
(1966) found bimodal fertility with respect to the number of years
of school completed. Although bimodality of these fertility rates
is not sufficient to produce bimodal IQ distributions in the
offspring generation, it is likely to increase the standard devia
tion (SD); that is, to spread the scores farther from the mean, and
to make the distribution platykurtic, or flatter than the normal
curve. Assortative mating, the tendency for like to mate with
like, also increases these tendencies. For example, B. L. Warren
(1966) found the correlation of spouses, years of education to be
about .60, and the correlation of their socioeconomic statuses (as
indexed by father's occupation) to be about .30. Since both these
variables are correlated with IQ, assortative mating could also be
increasing IQ variation from parent to offspring generations.
An analysis (Walberg, 1974) of data from an American study
that reports IQ SDs for parents and their offsprings (Higgens et

530

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
WALBERG & MARJORIBANKS FAMILY ENVIRONMENT

al., 1962) reveals a significant (p < .05) increase from 14.89 to


15.71. Unfortunately, the study started with a special population,
children and grandchildren of psychiatric patients. In longitudi
nal research in Britain the Scottish Council for Research in
Education and Population Investigation Committee (1949) found
a significant increase, from 15.48 to 16.10, in the SD of IQ from
1932 to 1947 in a nearly complete sample of population cohorts in
Scotland, and Emmett (1950) found a slightly higher increase,
from 14.21 to 15.00, in a large sample of English districts during
the same period.
It is well established in genetic theory and research that
assortative mating and bimodal fertility produce greater varia
tion in offspring. Since these mechanisms operate with respect to
IQ, years of education, and social class and since studies of
changes in verbal IQ reveal increasing standard deviations, it
appears that, insofar as it is genetically determined, IQ variation
may be increasing, and educators may be confronted with groups
of students who have more diverse abilities.
It was noted earlier that the longitudinal surveys carried out
in England and Scotland, on very large numbers of children, had
shown slight increases in IQ over time. Two somewhat compara
ble sets of data on young men have been reported in the United
States. Tuddenham (1948) found that the mean IQ of draftees for
military service rose about one standard deviation between
World War I and World War II, and Tupes and Shaycoft (Note 1)
found a half-standard deviation increase between World War II
and 1963. IQ apparently rose about a half point a year. By the
standards of a hypothetical IQ test normed early in the twentieth
century, the typical young adult male in the United States would
now rank very high in measured verbal intelligence. Better
nutrition, more verbal stimulation inside and outside the home,
additional years of higher quality schooling, and the immigrants'
mastery of English probably contributed to this rise. In addition,
the reduction of family size accompanying urbanization may
have enabled parents to concentrate their energies on fewer
children.

Model 5: Socioeconomic Status, Family Size, and Ability.


Anastasi (1956) found evidence that the correlation of sibsize
and ability is higher in samples of lower SES children than the
correlations of these variables for middle and higher SES sam
ples. Although such correlations may be compared, multiple
regression analysis indicates the relation of each independent
variable to a dependent variable in the context of the other
independent variables. For example, the weights of SES, sibsize,
and their interaction may be assessed simultaneously.

531

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
REVIEW OF EDUCATIONAL RESEARCH Vol. 46, No. 4

Recent theory and regression analyses support Anastasi's


evidence that SES and sibsize interact and, furthermore, show
that the inverse of sibsize may provide a better prediction of
ability than the simple linear form (Model 5). Marjoribanks,
Walberg, and Bargen (1975) reasoned that, since children share
adult resources of intellectual stimulation in the family, the
mathematical relationship between sibsize and parental stimu
lation is not linear but is of a hyperbolic form involving the term,
"one divided by the number of children in the family." That is,
the amount of parental attention received by each child de
creases as the number of children in the family increases, in such
a way that with each additional child the successive decrements
in shared attention become smaller. Therefore, the expected
percentages of parental attention given children in one-, two-,
three-, four-, and five-child families would be 100, 50, 33, 25, and
20 respectively. Thus, a single child in a family may score higher
on mental ability tests because he receives all the available
parental stimulation, whereas a child with four siblings may
have lower ability scores because he receives an estimated
one-fifth of the available stimulation. Moreover, first borns may
tend to be brighter, either because some of them are single
children and receive all the available parental attention, or
because they receive 100 percent of the parental stimulation
until the second child is born, whereas later born children
usually have to share parental attention. The apparent impair
ment effect of sibsize, however, is conditioned strongly by fa
ther's occupation. In an Ontario sample of 185 11-year-old boys,
there was no effect of sibsize on ability in professional-managerial
homes, but there was an estimated difference of 40 IQ points
between single borns and those from six-child families with
fathers in unskilled occupations (Marjoribanks, Walberg, &
Bargen, 1975).
What may be important in this latter work is the analytic
method rather than the findings. A similar analysis of the ability
scores of several hundred thousand Dutch males born during
World War II failed to support the inverse sibsize relation, but
clearly revealed a negative hyperbolic relation of sibsize and
ability and an interaction of sibsize and SES (Marjoribanks &
Walberg, Note 2). Analysis of other data may show that the
form of the functional relations of ability to sibsize varies across
SES, time, and nationality.

Models 6-7: Family Environment and Ability


Although socioeconomic status is a convenient construct to
measure, it does not yield a comprehensive assessment of the
factors in the home that foster ability. Proximate, detailed

532

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
WALBERG & MARJORIBANKS FAMILY ENVIRONMENT

assessments provide for better predictions of children's ability


(Model 6). For example, an early study of English homes showed
that the quality of maternal discipline, even though statistically
controlled for SES, predicts ability, and children in intellectually
demanding homes where rewards depend on achievement tend to
score higher than others on ability tests (Kent & Davis, 1957).
Dave (1963) and Wolf (1964) in Chicago and Marjoribanks (1972) in
Southern Ontario, Canada, have shown that about half the
variance in verbal ability can be accounted for by sociopsycholog
ical assessments of the family environment. From semi
structured interviews with parents, these investigators have
rated such factors as parental encouragement of achievement,
language, intellectuality, and independence. In the Mar
joribanks (1972) study, eight family environment variables were
measured: achievement, activeness, intellectuality, indepen
dence, English language, second language, and mother and father
dominance.
A canonical analysis of the Canadian data (Walberg & Mar
joribanks, 1973) showed that different patterns of stimulation in
the family are related to the differential development of specific
abilities (Model 7). Although the first canonical variate (see
Figure 2) revealed a strong link between general parental stimu
lation and global ability (Model 6), the second variate showed
that level of father's occupation and parental encouragement of
activeness and language are related to higher verbal and reason
ing scores but lower number scores. The analysis implies that
differential environmental processes may operate selectively to
develop certain potential abilities and to leave others relatively
underdeveloped. Additional research may reveal ways in which
parents can more systematically develop specific cognitive abili
ties in their children.

Model 8: Ethnicity, Environment, and Abilities


Factor analysis (Marjoribanks, 1972) of the eight family envi
ronment variables suggests that a more parsimonious description
of the family environment might be made by using a few weighted
sums of the variables. However, Marjoribanks (1972) showed
that different socioeconomic and ethnic groups (Jewish, South
ern Italian, Canadian Indian, white Anglo-Saxon Protestant, and
French Canadian) are characterized by different patterns of the
environment variables (Model 8), and that the eight variables
lead to a richer understanding of group environmental differ
ences. Marjoribanks also showed the extent to which the family
environment accounts for ethnic group differences in the cogni
tive abilities. For spatial ability, all of the group differences were

533

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
REVIEW OF EDUCATIONAL RESEARCH Vol. 46, No. 4

Achievement press Verbal


Number 3

xlntellectual press

x Activeness press
* English press
Independence x
Occupation of father
Education of mother Reasoning
yEthlanguage
i^>f father 4
Spatial
^ * Mother dominance
x father dominance
.1

.5 -.4 7i-72-7T II

.1

-.3 Birth order

-.4 Number of children

f-.5 * Crowding ratio

x Sociopsychological 6
environment variables

Global environment indicator

Mental ability measure

Figure 2.
Canonical loadings of environmental and mental ability measures

534

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
WALBERG & MARJORIBANKS FAMILY ENVIRONMENT

accounted for by the environment, whereas the measures ac


counted for large proportions of the ethnic group differences in
verbal, number, and reasoning ability scores.

Model 9: /(Socioeconomic status, family


size )?>Environment.
Investigations of the relationships between sibling variables
and cognitive abilities have usually found that family size (Anas
tasi, 1956; Nisbet, 1953; Maxwell, 1969; Record, McKeown, &
Edwards, 1969; Waller, 1971), crowding ratio of the family
(Fraser, 1959; Vernon, 1969; Whiteman & Deutsch, 1967) and, to a
lesser extent, birth order (Clausen, 1966; Kellaghan & Macna
mara, 1972; Schooler, 1972; Skovholt, Moore, & Wellman, 1973;
J. R. Warren, 1966) are negatively related to cognitive ability.
However, when the family environment has been used as an
explanatory variable in sibling studies, it has been defined in
terms of gross classificatory variables such as socioeconomic
status (Burton, 1968; Eysenck & Cookson, 1969; Murray, 1971;
Record et al., 1969). As a result, research has provided only a
general and sometimes contradictory picture of the interrela
tionships between sibling variables, socioeconomic characteris
tics, and those social psychological elements of the family that
are associated with cognitive performance.
Marjoribanks and Walberg (Note 2) used multiple regression to
examine the relationships of the family environment to father
occupation, father education, mother education, sibsize, the in
verse of sibsize, crowding ratio of the family, and birth order. The
environment was defined in terms of the eight variables dis
cussed earlier. The findings indicated that the most parsimoni
ous account of the variance in the environment scores is obtained
by using father's occupation and either of the sibsize variables.
Equations using the two terms account for about as much
significant variance in the environment scores as equations
containing all the measured variables, their quadratic forms (to
test for nonlinearity), and their products (to test for interactions).
The other sibling variables and socioeconomic status indicators
provide no significantly better prediction of the sociopsychologi
cal environment of the family beyond that afforded by the two
term equations.
The nature of the relationships between socioeconomic status,
family size, environment variables, and ability scores (Models 5
and 9) were examined further (Marjoribanks & Walberg, Note 2)
by plotting the regression-fitted sibsize influence on verbal abil
ity and on the environment categories at different socioeconomic
levels. In the first of the four figures that are presented here to
illustrate the relationships (see Figure 3), the regression-fitted

535

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
REVIEW OF EDUCATIONAL RESEARCH Vol. 46, No. 4

Number of
Father
children
occupation

Figure 3.
Fitted-verbal scores in relation to sibsize and father occupation :
Canadian sample

influence of sibsize on the verbal ability scores for the sample of


11-year-old Canadian boys has been plotted. The curvature of the
surface indicates the increasing influence that sibsize had on the
verbal scores as the socioeconomic level of the family decreases.
However, if the children came from small families, their per
formances were quite high regardless of their socioeconomic
status. It is not until sibsize increased beyond two, in families of
low socioeconomic status, that sibsize appeared to have a del
eterious influence on verbal ability.

536

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
WALBERG & MARJORIBANKS FAMILY ENVIRONMENT

In Figure 4 the regression-fitted influence of sibsize on the


reading scores for a sample of 11-year-old English children is
shown. The surface shows the apparently deleterious influence of
large family size on reading achievement. Even at high socioeco
nomic levels, the reading scores were low for children from
sibsizes greater than five.
Relationships between sibsize, father's occupation, and socio
psychological environment variables in the Canadian and English
samples are presented in Figures 5 and 6. The curvature of the
two surfaces indicates that within each socioeconomic level the
amount of parent-child interaction in the family decreases as the
size of the family increases. However, in the highest socioeco
nomic status groups the environment scores for the large
families are greater than the environment scores in families with
two or more children from the low socioeconomic groups.
The results of the above studies indicate that by using detailed
assessments of the family environment and by using multiple
regression models it is possible to gain an increased understand
ing of the complex nature of the relationships between socioeco

Reading
scores

Figure 4.
Fitted-reading scores in relation to sibsize and father occupation:
English sample

537

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
REVIEW OF EDUCATIONAL RESEARCH Vol. 46, No. 4

Number of Father
children occupation

Figure 5.
Fitted-environment scores in relation to sibsize and father occupation:
Canadian sample

nomic status, sibling variables, family environment, and cogni


tive ability.

Models 10-12: Longitudinal Research


Most of the research on the relationships of the family envi
ronment to ability has been restricted by designs that are
cross-sectional. Only a small number of studies have employed
longitudinal designs and have included measures of prior ability
in the research. Walberg and Marjoribanks (1974) used data that
were collected in 1964 and 1968 in an extensive survey of school
children in England (Plowden, 1967).
The 1964 sampling procedure of the Plowden study had two
stages. First, a random sample was taken from all types of

538

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
WALBERG & MARJORIBANKS FAMILY ENVIRONMENT

Family
environment

children occupation

Figure 6.
Fitted-environment scores in relation to sibsize and father occupation:
English sample

maintained primary schools in England. Then a random selection


was made of children within these schools. The total sample,
which included 3,092 children from 173 primary schools, was
divided into three age-cohorts, each of approximately 1,000 chil
dren. The average age of the senior children in 1964 was approx
imately 11, of the middle group, 8, and of the youngest group, 7.
When the three cohorts were surveyed again in 1968, the senior
cohort was in the fourth year of secondary school, the middle
cohort in the first year of secondary school, and the youngest
group in the last year of the primary school. The number of
children with complete records for both surveys was 2,350 with
approximately 800, equally divided between boys and girls, in
each of the age cohorts.
Walberg and Marjoribanks (1974) used multiple regression
models to examine the relationships of socioeconomic status
(father's occupation, family income, father's education) and sib
ling variables (sibsize, birth order) to reading achievement scores
(Model 10). Two-term equations containing the 1964 reading
scores and father's occupation accounted for as much significant
variance in the 1968 achievement scores as complex many-term
equations containing all the measured variables, their quadratic

539

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
REVIEW OF EDUCATIONAL RESEARCH Vol. 46, No. 4

forms (to test for nonlinearity), and their products (to test for
interactions). Table 1 shows the powerful influence of prior
reading scores on later reading achievement. The inverse of
sibsize explained no additional variance after taking into ac
count prior reading scores and father's occupation.
In a series of multivariate studies of physics test scores and
intervening classroom environments in the United States and
Canada, Walberg (1972) found that about 75% (80% corrected for
criterion errors of measurement) of the posttest variance could
be explained by the initial test scores and the learning environ
ment measures (Model 11). In one of the most comprehensive
studies that illustrate Model 11, Keeves (1972) obtained measures
of academic achievement from tests given one year apart to
12-year-old children in the Australian Capital Territory, as well
TABLE 1
Multiple Regression of Final Reading Scores on Initial Reading Achievement,
Father's Occupation and Inverse of Sibsize.
Final Initial Father's Sibsize Multiple
Reading Statistic Reading Occupation Inverse Product R
Fourth-year b .987 .558 2.315 -.550
boys e .033 .325 2.652 .639
R;1 .722** .003 .000 .005 .852*
Fourth-year b 1.058 .847 4.344 -.936
girls e .040 .371 3.387 .786
R;1 .676** .005** .002 .012 .826*
First-year b .643 1.218 5.513 -1.087
boys e .030 .403 3.221 .813
Rf .617** .016** .002 .002 .798*
First-year b .631 .817 3.507 -.446
girls e .031 .394 3.804 .862
R ,2 .591** .015** .003 .000 .780*
Primary 6 .414 .945 4.061 -.241
boys e .028 .391 3.343 .836
Rsl .388** .034** .013 .000 .658*
Primary b .411 -.160 -5.474 1.648
girls e .030 .416 3.831 .876
#,2 .357** .015 .003 .006 .617*
Note. The statistics in the rows are the raw regression weights
errors (e), and the increment in variance explained with th
variable (R*). Relationships significant at the .01 level are i
asterisks.

540

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
WALBERG & MARJORIBANKS FAMILY ENVIRONMENT

as assessments of intervening family, peer group, and classroom


environments. The pretest and environmental measures ac
counted for approximately 74% of the posttest achievement in
mathematics and for 66% in science. The Keeves analysis may
have explained less variance than the Walberg study despite the
additional measurements of family and peer environments be
cause of less reliable measures of the environment and achieve
ment variables.
The study by Keeves is one of the first educational studies to
use path analysis in research analyzing environmental correlates
of cognitive performance. Path analysis is a technique for
hypothesizing a network of causal relationships among a set of
variables. In the path analysis example in Figure 7, the single
headed arrows are employed to portray possible causal effects,
and double-headed arrows indicate possible correlated causal
variables. The broken lines show distal measures hypothesized to
be mediated through more proximal variables. That is, the
analysis proposes for testing that family socioeconomic status
and the number of children in the family are superfluous
explanatory variables when comprehensive measures of the
family environment are taken into account (Model 12).
Once a network such as that presented in Figure 7, is con
structed, it may be tested using regression analysis. Causality
may be inferred from regression under several assumptions:
1. All variables that might affect the dependent variable
are either included in the regression equation or are
uncorrelated with the variables that are included.
2. Terms are included in the regression equation to
handle any curvilinear or interaction effects.
3. The dependent variable has no effect on the indepen
dent variable.
These assumptions are rarely met completely, but regression can
be useful in experimentally uncontrolled research. Violations of
the first assumption mean that a partial and possibly misleading
picture of causal relationships is drawn. However, the path
guided regression analysis may provoke research that includes
other hypothesized variables in subsequent equations. It may
also be shown either that hypothesized variables are not corre
lated with the independent variables or that they add no further
explanation of the variance and are thus unparsimonious or are
mediated by (share common variance with) independent vari
ables already in the model. The second assumption may be met
by including products of the independent variables to account for
interactions, and squares or other mathematical forms of the
variables to account for curvilinearity. The third assumption

541

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
scores, ,- scores

Fnyionment

Figure 7.o
Intervening environment
Path analysis diagram
Sibsize

socioeconomic.

Family .
status

Cognitive Cognitv
Time 1 Time

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
WALBERG & MARJORIBANKS FAMILY ENVIRONMENT

requires judgment; relevant prior events may often be pre


sumed to cause subsequent events rather than vice versa, and
common sense can be helpful in provisionally denying reversed
causality.
A fourth assumption of causal regressions, only now becoming
fully appreciated, is that the dependent and independent vari
ables are measured without error. This assumption is highly
questionable in educational research. Causal relations may ap
pear weak, not because they are weak, but because both causes
and effects are measured unreliably.
An attempt was made by Walberg and Marjoribanks (1974) to
satisfy the above assumptions in a path analysis of data from the
Plowden (1967) survey of English schoolchildren and from the
follow-up study (described above). In both surveys, a structured
interview schedule had been used to gather information about
the family environment and the socioeconomic status of the
children in the sample. From the home interviews, indices were
constructed that assessed the literacy of the home, parental
aspirations for the child, and parental interest in education. A
single measure of the intervening family environment was con
structed by taking the means of the two assessments of each of
the environmental indices. Then a single environment score was
developed from an equally weighted composite of the three
means. Similarly, an index of family socioeconomic status was
obtained by calculating the means of the 1964 and 1968 measures
of father occupation, father education, and family income and
then computing an equally-weighted composite of the three
means. The intervening sibsize influence was assessed in the
same manner.
The longitudinal analysis in Table 2 shows that large am
of variance in final reading achievement scores are account
by the initial reading scores and the intervening family
ronment: the major portion of the variance is explained by
achievement, but the additional variance associated with
ronment is significant in all cases. In three samples, the pr
of initial reading and intervening environment adds signifi
to the equations, and in four samples the quadratic form of in
reading makes a significant contribution. The other produc
quadratics were not significant in more than one sample.
The increments in variance associated with the inver
sibsize and family socioeconomic status are comparatively
small and are significant in only two samples. These very
increments are the apparent direct effects of these variab
final reading achievement that are unmediated by the fa
environment. More comprehensive measures of the family
ronment are likely to mediate these effects completely.

543

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
EhO > ?HWH>WO
<S o HOdo> 4^OS O

Rc2 .880* .840*


.740*

.770*

.031**
.011.005 .011 .001 .003 .649**

.000 .085** .004** .771**


.044**
AB .015.006 -.017 -.017 -.017
1.002 .008 .007
1.023 .001 1.008
.007 .005 .004

.860 .002
-.496
InverseSibsize
Intervening of

.011**
.035 .069 .009
.272.079
economic Status
Intervening .042.063.005
Family Socio

TABLE 2
Longitudinal Regressions of Final Reading Achievement on Initial Reading Scores and Measures
of the Intervening Environment

B Family
Intervening Environment
.019** .028** .014**
.320.123 .027 .412.145 .047 .274.100 .010

A Reading 1.365 .722**.854 .676**


.383.112
.617**
.731
.150 .799
1.617 .174

Initial

6 e 6 e
R>2c
R? b e Rf
Samples Statistic

Boys. Girls
YearBoys
Fourth Fourth Year
Year First

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
3>rtdw O 3> w
> Os >g
PM < o H H

.710* .560* .480*

.649* .451* .408*

.012** .014**
.013.004 .007.004.005 .019.004

.005**
.011
.005
.006 .001 .002 .006
.004
-.001

.006*
1.007.961.002 2.066 1.561 .003
1.041 1.003

.004**
.071
.156 .096.071 .001 .071 .065.001

icsofinthe
ables thevariances
sincremental rows
corrected foraccounted
variance
for each population
for by
explained
measurement areofpopulation
inerror
the
each intheorder:
variable the unstandardized
with
two(Ri2), the and
and
variables samethe
the same regression
correction.
statistic
criterion. weight
Relationships
forfigures
The (6), its lasstin
significa
the first
in two
the

.035** .051** .029**


.260.120 .012 .265.099 .083 .265.116 .046

.591** .388** .357**


.282.134
.322
.118 .699 .479 .273.133 .440

levels are indicated respectively with one and two asterisks.


R l2r
b e b e b e Rf
R?<

Girls Primary
FirstYear Boys Primary
Girls

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
REVIEW OF EDUCATIONAL RESEARCH Vol. 46, No. 4

Table 2 shows that the estimated accountable adjusted vari


ances in final reading achievement corrected for errors of mea
surement are substantial. The corrected variances are higher in
older groups and higher for boys than girls. The increasing
predictability with age and for boys is attributable to the higher
correlations of initial and final reading with age and in the
samples of boys. Environment has a small significant, positive
association with final reading, independent of initial reading in
all samples.
The regression-fitted influence of the intervening environment
on final reading achievement at different levels of initial reading
achievement for the fourth-year boys is shown in Figure 8. The
dotted lines show the regression slopes of final reading achieve
ment on environment, corrected for measurement error in both
variables for fourth-year boys scoring high (above .5 standard
deviations higher than the mean), medium, and low (below -.5
standard deviations below the mean) on initial reading. The
corrected slopes indicate how uncorrected regression can under
estimate environmental effects when there are substantial
errors in measuring the environment. On the other hand, the
corrected environmental slopes in all cases are far smaller than
the uncorrected initial reading achievement slopes; and in the
case of the first-year secondary groups, the corrected environ
mental slopes were smaller than uncorrected environmental
slopes because the correction raises the already high correlation
of initial achievement and intervening environment.
The results of this latter study suggest that adolescents may
benefit as much as younger children from a stimulating family
environment. Moreover, the larger amounts of variance ex
plained by initial achievement in reading in comparison with
that explained by the intervening four-year environments
should be a reason for patience rather than pessimism among
parents and educators. A stimulating family environment for
several years is likely to lead to some enhancement of cognitive
abilities. If the environmental stimulation in the family and peer
group were to be coordinated with classroom and school envi
ronments and sustained throughout infancy, childhood, and
adolescence, then substantial cognitive gains might be achieved.

Conclusion
Our review indicates that refined measures of the family
environment and the use of complex statistical models increase
our understanding of the relationships between socioeconomic
status, sibling variables, family environment, and cognitive de
velopment. For subsequent investigations of the relation of

546

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
WALBERG & MARJORIBANKS FAMILY ENVIRONMENT

Final
reading
achievement

Figure 8.
Regression surf ace for relations between family environment and reading scores:
English sample

environments to the development of children's traits, we suggest


the following equation:

h,h = 2 b> Ihh * + 2 bk E\,li2 + 2 bjk {!%) {E*k,hJ


where Ij9 is a given individual characteristic measured at time
t2 and where Ij,t represents the characteristic measured at an
earlier time tt. The first term on the right side of the equation is a
weighted composite of a number of antecedent individual charac
teristics (the asterisk indicates that the indicators are in the

547

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
REVIEW OF EDUCATIONAL RESEARCH Vol. 46, No. 4

optimal mathematical form; for example, linear, quadratic, or


logarithmic, for prediction); the second term is a similar inter
vening environmental composite; and the last term is the
weighted composite of products of antecedent characteristics and
environment. Research based on the proposed generalized equa
tion would allow for the multiplicity of possible causes in the two
domains, and the possibility that initial characteristics interact
with environment measures, determining later characteristics.
In future research on the relations between family environ
ments and individual development, measurement of both socio
psychological environments and behaviors can be further im
proved. Also, if new statistical and methodological techniques,
such as correcting relationships for unreliability of measures,
testing different mathematical forms of relations, and using
longitudinal analyses, are used, then it is likely that our under
standing of the causal relations between environmental condi
tions and behavioral development will be increased.
However, even if verbal-educational ability, the mental trait
that we know most about, was the only one that concerned a
nation or a family, far more research would be required to
provide a solid basis for national or family policy. And we
certainly agree with Gottesman and Erlenmeyer-Kimling (1971)
that for future policy developments there is a need to develop
multiple indices of individual value and not to be concerned with
the maximization of a single trait. Compared to the variety of
agricultural, business, manufacturing, and health indexes, some
issued periodically for many decades, data on children's traits,
precious though they are, are indeed spotty and unreliable and
have not been thoroughly analyzed. Obviously, we need to learn
much more about how children's traits are changing and how we
might change them more deliberately and effectively.
Lastly, the research reviewed above (and two impressive
studies, Zajonc, 1976, and Cicirelli, in press, that appeared too
recently to be included here) suggest that family environment
measures should be considered for inclusion in experimental and
correlational studies of educational effects. In experiments they
would serve as potent covariates, remove extraneous variance
from achievement, and allow more precise estimates of the
educational effects in question. They may be critical in correla
tional studies because the investigator is in the unenviable
position of having to claim all causal variables are in the regres
sion in order to make a causal inference about an unrandomized
treatment effect; and family environment is an obvious correlate
of the usual criterion tests. (This is not to say that other problems
of imputing causation from statistically-controlled correlation
would be solved.) Comprehensive, proximal interview assess

548

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
WALBERG & MARJORIBANKS FAMILY ENVIRONMENT

ments are preferable to socioeconomic status, family size, or


questionnaire measures of such things as the number of books
and magazines in the home. Better measures require more
planning and effort but may be a more worthwhile investment
than a large number of cases. Moderate-sized studies with care
ful measures of the home environment are likely to yield more
insight and statistical control than large superficial surveys.

Reference Notes
1. Tupes, E., & Shaycoft, M. Normative distribution of AQE aptitude indexes for
high school age boys. Lackland, Texas; U.S. Air Force Base Technical Docu
mentary Reports, 1964.
2. Marjoribanks, K., & Walberg, H. J. Birth order, family size, social class and
intelligence. Unpublished manuscript, 1975.

References
Anastasi, A. Intelligence and family size. Psychological Bulletin, 1956,55, 187-209.
Bajema, C. J. Estimation of the direction and intensity of natural selection in
relation to human intelligence by means of the intrinsic rate of natural
increase. Eugenics Quarterly, 1963,10, 175-187.
Bajema, C. J. Relation of fertility to educational attainment in a Kalamazoo
public school population: A follow-up study. Eugenics Quarterly, 1966, 13,
306-315.
Burton, D. Birth order and intelligence. Journal of Social Psychology, 1968, 76,
199-206.
Cicirelli, V. G. Sibling structure and intellectual ability. Child Development, 1976,
in press.
Clausen, J. A. Family structure, socialization, and personality. In L. W. Hoffman
& M. L. Hoffman (Eds.), Review of child development research. New York:
Russell Sage Foundation, 1966.
Cole, L. The population consequences of life history phenomena. Quarterly
Review of Biology, 1954,20, 103-137.
Dave. R. The identification and measurement of home environmental process
variables related to educational achievement. Unpublished doctoral disserta
tion, University of Chicago, 1963.
Emmett, W. G. The trend of intelligence in certain districts in England. Popula
tion Studies, 1950, 3, 324-337.
Eysenck, H. J., & Cookson, D. Personality in primary school children; 3-family
background. British Journal of Educational Psychology, 1969, UO, 117-131.
Fraser, E. Home environment and the school. London: University of London
Press, 1959.
Gottesman, I. I., & Erlenmeyer-Kimling, L. A foundation for informed eugenics.
Social Biology, 1971, 18, 51-58.
Higgins, J. V., Reed, E. W., & Reed, S. C. Intelligence and family size: A paradox
resolved. Eugenics Quarterly, 1962, 9, 84-90.
Keeves, J. P. Educational environment and student achievement. Stockholm:
Almquist & Wicksell, 1972.
Kellaghan, T., & Macnamara, J. Family correlates of verbal" reasoning ability.
Developmental Psychology, 1972, 7, 49-53.
Kent, N., & Davis, D. R. Discipline in the home and intellectual development.
British Journal of Medical Psychology, 1957, 30, 27-33.

549

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
REVIEW OF EDUCATIONAL RESEARCH Vol. 46, No. 4

Marjoribanks, K. Ethnic and environmental influences on mental abilities.


American Journal of Sociology, 1972, 78, 323-337.
Marjoribanks,.K., & Walberg, H. J. Family environment: Sibling constellation and
social class correlates. Journal of Biosocial Science, 1975, 7, 15-25.
Marjoribanks, K., Walberg, H. J., & Bargen, M. Mental abilities: Sibling constella
tion and social class correlates. British Journal of Social and Clinical
Psychology, 1975, U, 109-116.
Maxwell, J. Intelligence, education and fertility: A comparison between the 1932
and 1947 Scottish surveys. Journal of Biosocial Science, 1969,1, 2A7-271.
Murray, C. The effects of ordinal position on measured intelligence and peer
acceptance in adolescence. British Journal of Social and Clinical Psychology,
1971,10, 221-227.
Nisbet, J. D. Family environment and intelligence. Eugenics Review, 1953, 45,
31-40.
Plowden, B. Children and their primary schools. London: H. M. Stationery Office,
1967.
Poole, A., & Kuhn, A. Family size and ordinal position: Correlates of academic
success. Journal of Biosocial Science, 1973, 5, 51-59.
Record, R. G., McKeown, T., & Edwards, J. H. The relation of measured intelli
gence to birth order and maternal age. Annals of Human Genetics, London,
1969, S3, 61-69.
Schooler, C. Birth order effects: Not here, not now! Psychological Bulletin, 1972,
78, 161-175.
Scottish Council for Research in Education and Population Investigation Com
mittee. The trend of Scottish intelligence. London: University of London Press,
1949.
Skovholt, T., Moore, E., & Wellman, F. Birth order and academic behaviour in
first grade. Psychological Reports, 1973, 32, 395-398.
Tuddenham, R. D. Soldier intelligence in World Wars I and II. American Psychol
ogist, 1948, 3, 54-56.
Vernon, P. E. Intelligence and cultural environment. London: Methuen, 1969.
Walberg, H. J. Social environment and individual learning: A test of the Bloom
model. Journal of Educational Psychology, 1972, 63, 69-73.
Walberg, H. J. Optimization reconsidered. In H. J. Walberg (Ed.), Evaluating
educational performance: A sourcebook of methods, instruments and exam
ples. Berkeley, Calif.: McCutchan, 1974.
Walberg, H. J., & Marjoribanks, K. Differential mental abilities and home
environment: A canonical analysis. Developmental Psychology, 1973, 9, 363
368.
Walberg, H. J., & Marjoribanks, K. Social environment and cognitive develop
ment: Toward a generalized causal analysis. In K. Marjoribanks (Ed.),
Environments for learning. London: National Foundation for Educational
Research Publications, 1974, Pp. 259-273.
Waller, J. H. Differential reproduction: Its relation to IQ test score, education and
occupation. Social Biology, 1971, 18, 122-136.
Warren, B.L. A multiple variable approach to the assortative mating problem.
Eugenics Quarterly, 1966,13, 285-290.
Warren, J. R. Birth order and social behaviour. Psychological Bulletin, 1966, 65,
38-59.
Whiteman, M., & Deutsch, M. Social disadvantage as related to intellective and
language development. In M. Deutsch, I. Katz, & A. R. Jensen (Eds.), Social
class, race and psychological development. London: Holt, Rinehart & Winston.
1967.
Wolf, R. M. The identification and measurement of home environmental process
variables that are related to intelligence. Unpublished doctoral dissertation,
University of Chicago, 1964.
Zajonc, R. B. Family configuration and intelligence. Science, 1976,192, 227-236.

550

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms
WALBERG & MARJORIBANKS FAMILY ENVIRONMENT

AUTHORS
HERBERT J. WALBERG. Address: University of Illinois at Chicago Circle, Box
4348, Chicago, Illinois 60680. Title: Professor of Education. Degrees: B.E.,
Chicago State College; M.E., University of Illinois; Ph.D., University of
Chicago. Specialization: Social psychology of education, measurement,
evaluation and statistical analysis.
KEVIN MARJORIBANKS. Address: University of Adelaide, Adelaide 5001 S.A.,
Australia. Title: Professor of Education. Degrees: B.S., University of Sydney;
M.A.T., Harvard University; Ph.D., University of Toronto. Specialization:
Sociology of education and the family, the study of environments for learning.

551

This content downloaded from


110.37.71.115 on Mon, 29 Apr 2024 11:47:06 +00:00
All use subject to https://about.jstor.org/terms

You might also like