Testing for Racial Differences in the Mental Ability of Young Children

∗

Roland G. Fryer, Jr.

Harvard University Society of Fellows

and NBER

and

Steven D. Levitt

University of Chicago and

American Bar Foundation

February 2006

∗

We would like to thank Lawrence Katz, Kevin Murphy, Andrei Shleifer, and Lawrence H. Summers for

helpful discussions on this topic. Katherine Barghaus, Sheldon Bond, Walter Lamberson and Ethan Lieber

provided truly outstanding research assistance. Financial support for this study was provided by the

National Science Foundation, the Milton Fund at Harvard University, the Sherman Shapiro Research Fund,

and the Alphonse Fletcher Sr. Fellowship. Mailing addresses: Roland Fryer, Littauer Building, Harvard

University, Cambridge, MA 02138; Steven Levitt, Department of Economics, University of Chicago, 1126

E. 59

Street, Chicago, IL 60637. e-mail: [email protected]; [email protected].

Abstract

On tests of intelligence, Blacks systematically score worse than Whites, whereas Asians

frequently outperform Whites. Some have argued that genetic differences across races

account for the gap. Using a newly available nationally representative data set that

includes a test of mental function for children aged eight to twelve months, we find only

minor racial differences in test outcomes (0.06 standard deviation units in the raw data)

between Blacks and Whites that disappear with the inclusion of a limited set of controls.

The only statistically significant racial difference is that Asian children score slightly

worse than those of other races. To the extent that there are any genetically-driven racial

differences in intelligence, these gaps must either emerge after the age of one, or operate

along dimensions not captured by this early test of mental cognition.

Blacks in the United States have consistently scored worse than Whites on tests of

IQ and academic achievement (Shuey, 1958; Jensen, 1973, 1998; McGurk et al., 1982;

Hernstein and Murray, 1994). Among teenagers and adults, the Black-White test score

gap is typically one standard deviation in magnitude. Large racial gaps in test scores

have been found in children as young as two years old (Scott and Sinclair, 1997), and the

one standard deviation racial gap observed later in life is present by age three (Jensen and

Rushton, 2005). Even after accounting for a host of demographic and socio-economic

factors such as parental income, education, occupation, home environment, birth weight,

region, and urbanicity, a substantial Black-White test score gap generally remains.

Asians, on the other hand, tend to have systematically higher mean test scores than those

of other races (Campbell et al., 1966; Burkett et al., 1995; Rushton, 1995; Fryer and

Levitt, 2004).

Some scholars have argued that the combination of high heritability of

intelligence (typically above .7, see, for instance, Neisser, 1996) and persistent racial

gaps in test scores is evidence of genetic differences across races (Jensen, 1973, 1998;

Jensen and Rushton, 2005). As Nisbett (1998) and Phillips et al. (1998) argue, however,

the fact that Blacks, Whites, and Asians grow up in systematically different environments

makes it difficult to draw strong causal genetically based conclusions.

See, for instance, (Campbell et al., 1966; Baughman and Dahlstrom, 1968 Scarr, 1981; Kaufman and

Kaufman, 1983; Naglieri, 1986; Bracken, Insko and Sabers, 1987; Humphreys, 1988; Krohn and Lamp,

1989; Brooks-Gunn, 1993a; Brooks-Gunn, Duncan and Kelbanov, 1994, 1996; Phillips et al., 1998;

Phillips, 2000; Coley, 2002). A notable exception to this pattern emerges in the Early Childhood

Longitudinal Study Kindergarten (ECLS-K) sample of children who entered kindergarten in 1998. This

nationally representative sample differs from prior data in that the raw Black-White test score gap at the

time children enter kindergarten is substantially smaller than in most prior studies (0.64 standard deviations

in math, and 0.41 standard deviations in reading), and that the inclusion of a small number of demographic

and socio-economic controls erases the gap (Fryer and Levitt, 2004). Although through the first four years

of school, the Black-White test score gap grows substantially, and by the end of third grade these controls

no longer account for the differences in test scores across races (Fryer and Levitt, forthcoming).

Data on mental function in the first year of life represents a potentially important

piece of evidence to inform this debate.

To the extent that environmental factors play a

smaller role at early ages (or alternatively, researchers are better able to measure and

control for the environmental factors affecting infants), the presence of an early racial gap

in test scores would bolster the argument in favor of a genetic basis for racial differences.

On the other hand, an absence of racial differences in mental abilities among children age

9-12 months substantially complicates (but does not rule out) a genetic basis for a racial

IQ gap. To the extent that some aspects of adult intelligence only emerge at later stages

of development, or that these aspects go unmeasured in the early test of mental function,

the genetics story cannot be definitively rejected with these data.

Because of data limitations, prior research has not been able to directly address

this question (Jensen and Rushton 2005). Studies measuring cognitive abilities of young

children have been small-scale, rare, and based on convenience samples that are not

drawn with the goal of being nationally representative. For instance, many of the samples

involve a limited number of babies born in a particular hospital or metropolitan area

(Gravem, Ireton and Thwing, 1970; Wilson 1983), preterm infants (Rose and Wallace,

1985), those with birth weights less than 1,500 grams (Dezoete, MacArthur and Tuck,

2003), or children with rare medical conditions (McGarth et al., 2004). The Early

Childhood Longitudinal Study Birth Cohort (ECLS-B) is the first large, nationally

representative sample with measures of mental functioning (a shortened version of the

Bayley Scale of Infant Development (BSID)) for children aged one and under.

The usefulness of early tests of mental function depends critically on the assumption that these tests are

strongly related to future test scores. Measures of cognitive ability at one year of age (similar to those used

in this study) have been found to be strong predictors of test scores later in life by some researchers

(Bradley, Caldwell and Elardo, 1975; Wilson, 1983; DiLalla et al., 1990; McGarth, 2004), although other

studies report a weaker relationship (Honzik, 1983; Rose and Wallace, 1985; Kopp and McCall, 1990;

Anderson, Sommerfelt and Sonnander, 1998). This evidence is discussed in greater detail later in the

paper.

Analyzing these data, we find extremely small racial differences in mental

functioning of children aged eight to twelve months. With controls only for the child’s

age and gender, the mean White infant outscores the mean Black infant by .064 standard

deviation units – only a tiny fraction of the one standard deviation racial gap observed at

older ages. The raw scores for Blacks are indistinguishable from Hispanics and Asians,

who also slightly under perform Whites. Adding controls for socio-economic status,

home environment, and prenatal circumstances further compresses the observed racial

differences. With these covariates, we cannot reject equality in test scores across any of

the racial/ethnic groups examined, except Asians. In our sample Asian babies do slightly,

but statistically significantly, worse than babies of other races.

These findings pose a substantial challenge to the simplest, most direct, and most

often articulated genetic stories regarding racial differences in mental function. They do

not, however, preclude systematic genetic differences across races as an explanation for

later observed test score gaps if, for instance, racial gaps are concentrated in higher-order

thinking which only emerges later in life. The late emergence of racial differences in test

scores is also consistent with the existence of a gene-environment interaction such as

argued by Dickens and Flynn (2001). In their model, a positive feedback loop exists

between genes and environment, which over time serves to magnify small initial

differences when genetics and environmental circumstances are positively correlated.

II. The Data

The data we analyze, the Early Childhood Longitudinal Survey Birth Cohort

(ECLS-B), is a nationally representative sample of over 10,000 children born in 2001.

The first wave of data collection was performed when most of the children were between

eight and twelve months of age.

The data set includes an extensive array of information

from parent surveys, interviewer observation of parent-child interactions, and mental and

motor proficiency tests. Further details on the study design and data collection methods

are available at the ECLS website (

http://nces.ed.gov/ecls).

To measure mental proficiency, the ECLS-B uses an abbreviated version of the

BSID known as the Bayley Short Form–Research Edition (BSF-R), which was designed

to measure the development of children eight to eleven months of age in five broad areas:

exploring objects (e.g., reaching for and holding objects), exploring objects with a

purpose (e.g., trying to determine what makes the ringing sound in a bell), babbling, early

problem solving (e.g., when a toy is out of reach, using another object as a tool to retrieve

the toy), and communicating with words. A child’s score is reported as a proficiency

level, ranging from 0 to 1 on each of the five sections. These five proficiency scores

have also been combined into an overall measure of cognitive ability using standard scale

units. Most of our analysis focuses on this overall metric. The test is administered by a

trained interviewer and takes twenty-five to thirty-five minutes to complete.

Because this particular test instrument is newly designed for ECLS-B, there is no

direct evidence regarding the correlation between performance on this exact test and

outcomes later in life. There is, however, a substantial literature of longitudinal studies

relating BSID test scores of children aged eight to twelve months to tests of intelligence

later in life. Figure 1 provides a graphical representation of the correlations observed in

this prior literature, weighting the estimates of the various papers by the number of

subjects included in each study.

The horizontal axis is the age in years that the

These children will eventually be resurveyed four times between the present and first grade. At the

present time, only the first wave of data is available.

Appendix Table 1 list the full set of papers and reported correlations on which Figure 1 is based.

subsequent tests were administered and the vertical axis contains the average correlation

between the BSID score and another measure of intelligence. The figure shows that the

average correlation between BSID and future IQ scores starts very high and decreases as

children age, stabilizing with an average correlation around 0.3 at approximately five

years of age. For purposes of comparison, when older children are given achievement

scores three years apart, the correlation between scores is on the order of 0.6 (Cruse et al.,

1996).

A correlation of .3 between the BSID and future measures of IQ means that the

BSID score explains only nine percent of the variation in future test scores for a

particular individual. Even though one can explain relatively little of the within-person

variation over time in test scores, one would still expect to observe large differences in

mean test scores by race on the BSID given the prior evidence. Campbell et al. (1983),

for instance, report a correlation between maternal IQ (assessed using the Wechsler Adult

Intelligence Scale at the time of interview) and the BSID of twelve month old children as

.36. For purposes of comparison, the correlation between maternal IQ (using the same

metric) and the well-known Stanford-Binet at three years old is .39. Given the observed

one standard deviation in maternal IQ between whites and blacks, a correlation of .36

between child’s BSID and mother’s IQ would imply expected mean differences between

white and black one-year olds of .36 standard deviations.

The ECLS-B sample includes observations on 10,688 children. For 556 of these

individuals, no mental ability test was performed. These subjects are dropped from the

Other infant tests, such as the Cattell Infant Intelligence Scale, Gesell Development Schedules, or Brunet-

Lezine Development Scales, have a substantially lower correlation with later achievement than does the

BSID (Escalona and Moriarty, 1961; Karlberg et al., 1968; Birns and Golden, 1972).

Indeed, given the results cited in Jensen and Rushton (2005) regarding one standard deviation racial gaps

in IQ tests administered at age 3 and the stability in black-white differences over time, one might expect to

observe racial differences in the BSID of one standard deviation.

analysis. This is the only exclusion we make from the sample.

Throughout the analysis,

the results we report are weighted to be nationally representative using sampling weights

included in the data set.

Table 1 provides summary statistics by major racial/ethnic group in ECLS-B.

The mutually exclusive and collectively exhaustive racial/ethnic categories we report are:

non-Hispanic Whites (which we simply deem “White”), non-Hispanic Blacks (“Black”),

Hispanics, Asians/Pacific Islander (“Asian”), and other race (which combines children

characterized as Native American, mixed race, or other race). The top panel of the table

reports means and standard deviations by race on the overall measure of mental ability.

For ease of interpretation of the regressions, the overall test score has been normalized to

have a mean of zero and a standard deviation of one for the sample as a whole. Whites

score .018 standard deviations better than the sample mean on the overall mental

measure.

The next panel of the table presents basic demographic characteristics, which are

generally similar across groups. Age at testing is approximately equal across races. As

would be expected, Blacks are over represented in the South and underrepresented in the

West. Asians and Hispanics are seen in greatest numbers in the West. The fraction of

girls and boys are similar across all racial groups.

The third panel has variables capturing the home environment, including socio-

economic status (SES) quintiles, number of siblings, family structure, mother’s age, and

an interviewer rating of the effectiveness of the “parent as a teacher” based on

In cases where there are missing values for another of the covariates, we set these missing observations

equal to zero and add an indicator variable to the specification equal to one if the observation is missing

and equal to zero otherwise.

A comparison of the ECLS-B sample characteristics with known national samples such as the US Census

and the Center for Disease Control’s Vital Statistics confirms that the sample characteristics closely match

the national averages.

observation of parent-child interactions in a structured problem-solving environment.

The socio-economic status measure is constructed by ECLS and includes parental

income, occupation, and education. Whites and Asians are concentrated in the higher

ranges of the SES distribution; Blacks, Hispanics, and the “other race” category have

below average SES. Roughly 90 percent of White and Asian infants are living in

households with two biological parents compared to only 41 percent among Blacks. On

average, children in our sample have approximately one sibling; Asian children have

slightly fewer and Black kids slightly more. White and Asian mothers tend to be older.

White parents fare better than the other racial groups on the interviewer evaluation of

“parent as teacher” effectiveness.

The final panel of Table 1 presents statistics on the prenatal circumstances of the

children. Extremely low birth weight (<1,500 grams) and premature birth are most

common among Blacks and least frequent for Whites and Asians. Twins and higher

order births are much more frequent among Whites, due primarily to the greater use of

infertility techniques such as in vitro fertilization (Hamilton and McManus, 2004).

Figure 2 plots the density of BSID test scores by race. The test score distributions

for infants of different races are essentially indistinguishable visually. Note that this is in

stark contrast to the distribution of other measures of intelligence for older children in

prior studies. Figure 2 makes the main point of the analysis. In the regression analysis

that follows, we control for a wide range of other factors, but the basic conclusion

suggested by Figure 2 is unchanged.

The “parent as a teacher evaluation” is based on the Nursing Child Assessment Teaching Scale (NCATS).

The NCATS scale is composed of 73 binary (yes/no) items that are scored by trained observers. The

NCATS coding system has two main scales: the parent scale, which has 50 items, and the child scale with

23 items. The parent scale of the NCATS focuses on the caregiver’s use of a “teaching loop,” which

consists of four components: (1) getting the child’s attention, and setting up the expectations for what is

about to be done; (2) instruction giving; (3) letting the child respond to teaching; and (4) feedback, on the

child’s attempts to do the task (Nord, 2004). The total parent score can range from 0 to 50 (Yes=1, No=0).

III. Racial Differences in the Mental Ability of Young Children

Our empirical approach involves weighted least squares estimation of equations

taking the general form:

iii

XRTESTMENTAL

εβ

+Γ+=

∑

(1)

where i indexes individuals and r corresponds to the racial group to which an individual

belongs. MENTAL TEST reflects either the overall test score, or one of the underlying

components of the test – depending on our particular specification. The vector X captures

a wide range of possible control variables, and ε is an error term. Also included in all

specifications are interviewer-fixed effects, which adjust for any mean differences in

scoring of the test across interviewers.

The basic results for the coefficients on the race variables are presented in Table

2. The omitted race category is “Whites,” so the other race coefficients are relative to

that omitted group. Each column reflects a different regression. The first column

includes only interviewer-fixed effects. As in the raw data, Blacks, Hispanics, Asians,

and “Other” races slightly under perform Whites. Only for Blacks can one reject the null

of no difference at the .05 level.

As one moves from left to right in Table 2, the set of covariates is progressively

expanded. The coefficients on the other covariates are not shown in the table, but full

estimation results can be found in Appendix Table 2. Column (2) adds controls for age at

which the test is administered and the gender of the child. Because the age the test is

Each of the 13 regions was staffed by one field supervisor and between 14 and 19 interviewers, for a total

of 256 field staff (243 interviewers), who conducted an average of 42 child assessments each. The number

of interviews per interviewer range from 1 to 156. Almost all interviewers assessed children from different

races (James et al., 2004).

taken is such an important determinant of test performance, we adopt an extremely

flexible, non-parametric functional form for age (including age in days linearly, age in

days squared, all the way up to a quintic in age). The R-squared in the regression jumps

substantially with the inclusion of the age and gender controls (from .14 in column 1 to

.68 in column 2); which is mainly due to the age controls. A child taking the test at age

eleven months is predicted to perform .66 standard deviations better than the same child

when tested at age nine months, according to our estimates. Girls outscore boys by .06

standard deviations on average, a difference that persists throughout all our

specifications. Including these controls improves the performance of Whites versus other

races because Whites are slightly younger on average when tested.

Column (3) adds indicators for the family’s SES quintile. The SES variable (not

shown in the main table) enters with the expected sign, i.e. higher SES children score

higher on the test, but the magnitude of the effects are small (a top quintile SES child

outscores a bottom quintile child by .08 standard deviations. Inclusion of the SES

variable shrinks the coefficient on Black and “other race.”

Adding a range of other controls for a child’s home environment (family

structure, mother’s age, number of siblings, and region dummies) shrinks the gap

between Whites and each of the other races. The only race coefficient that remains

statistically significant is the underperformance by Asians of roughly .06 standard

deviations. Adding a control for the interviewer assessment of “parent as teacher” in

column (5) further shrinks all of the race coefficients.

The final column adds controls for a range of prenatal condition variables (birth

weight, premature birth, and multiple births). The coefficient on Black becomes very

close to zero; the other race variables are not greatly affected. All the point estimates are

substantively small: the largest gap, between Asians and Whites, is only .048 standard

deviation units. After Whites, the group performing best is Blacks, although one cannot

reject equality between Blacks and any of the other groups.

IV. Sensitivity Analysis and Extensions of the Basic Analysis

Although on average we observe only minor differences across races in mental

test performance, important differences in sub-groups of the population remain a

possibility. Table 3 presents a wide range of estimates to test the sensitivity of our basic

results. Each row of the table presents the race coefficients from a separate regression.

In each case, the specification uses the full set of controls included in the final column of

the preceding table.

The top row of the table simply reproduces the baseline estimates from column

(6) of Table 2. The next row presents unweighted estimates. The subsequent rows

segment the table by gender, SES quintile, family structure, region, urbanicity, and birth

weight. Of the 104 coefficients presented, twelve are statistically significant at the .05

level; almost all of these correspond to underperformance by Asians.

Evidence exists that there are racial differences in early motor skills, with Blacks

outperforming other races and Asians lagging (Malina, 1988; Colombo, 1993; Toy et al.

2000). In our data, both Blacks and Asians outperform Whites on the test of motor skills.

If the early life test of mental abilities depends on both sensorimotor and intellectual

skills, precocity in the former might mask a deficiency in the latter among Blacks.

Therefore, we report results controlling for an individual’s performance on the test of

motor skills. The black coefficient increases from -.002 to -.034 with the inclusion of

motor skills, though the absolute magnitude of the difference continues to be extremely

small. All other racial groups remain essentially unchanged.

The final sensitivity test we report corresponds to the age at which the test of

mental function is administered. In prior studies, the correlation between BSID tests and

later IQ scores were higher when the BSID was administered to older infants (Bayley,

1955). Thus, if true, racial differences in intelligence are being masked by limitations of

the BSID to ascertain these differences for children at young ages, then we would expect

the racial differences in our sample to be smaller among the youngest children sampled

and larger among the older children sampled. The final rows of Table 3 provides no

evidence of this for Blacks, although for the “other race” category, the gaps are steadily

rising versus Whites with age.

Investigating the individual components underlying the composite mental test score

The overall test score is made up of five underlying components: exploration,

exploration with purpose, babbling, problem solving, and word use. Table 4 presents

regression estimates on each of these parts of the test. Each column corresponds to a

different component of the test. The mean level of proficiency and standard deviation in

that mean are reported at the bottom of each column. Note that we have not transformed

the proficiency scores on the sub-tests to have mean zero and standard deviation one, so

the estimates in this table are not directly comparable to those of the two preceding

tables. Because there is so little variation on some of the sub-tests, transforming each of

the tests to have the same standard deviation across children will artificially inflate test

score differences on the sections with little variation (i.e., explores objects and uses

words). Only the race coefficients are shown in the tables. The specifications include the

full set of controls, mirroring column (6) of Table 2. Nineteen of the twenty race

coefficients are negative, implying that in virtually all of the cases the omitted group

(Whites) shows greater proficiency. The magnitudes of the estimates, however, are quite

small: less than one-half a percentage point difference in proficiency in almost all cases.

In only one instance (less babbling done by Asians infants) is the gap between another

group and Whites statistically significant at the .05 level. The only positive estimate in

the table is associated with greater (but not statistically significantly different) babbling

by Blacks. Thus, there is little in the underlying test components to alter the conclusions

suggested by the composite measure of mental function.

V. Discussion

The debate over racial differences in intelligence is among the most divisive in

the social sciences. Utilizing a newly available, nationally representative data set with

measures of mental function among children before their first birthday, we find little

evidence of systematic racial differences. Some substantively small, but statistically

significant differences are present in the raw data. Including controls for age, socio-

economic status, home environment and prenatal environment largely erase these small

differences.

Comparing the magnitude of differences observed in these data to prior studies of

children, teens, and adults puts into perspective how small the observed differences are in

this analysis of infants. Phillips et al. (1998), which investigates test outcomes among

kindergarteners in the early cohorts of the Children of the National Longitudinal Survey

of Youth (NLSY), finds some of the smallest published racial gaps. They report a raw

Black-White gap of over one standard deviation, which shrinks to one-third of a standard

deviation with the inclusion of myriad controls. Those gaps are an order of magnitude

larger than what we find among infants.

Although damaging to the hypothesis that genetic differences are at the root of

racial gaps in intelligence, the results of our analysis do not preclude a possible role for a

genetic contribution to racial differences in intelligence for a number of reasons. First,

one could reasonably argue that the control variables we include in the regression

analysis are themselves partly genetically determined. By controlling for factors such as

socio-economic status and birth weight (which systematically differ across races), we

may indirectly be parsing out important channels through which genetics are operating.

The fact that the raw differences in test performance across races are so small, however,

makes this argument largely moot.

A second possible argument one can make against our findings is that the

particular form of the BSID used in ECLS has not been proven to correlate with future

measures of intelligence. The relationship between other forms of the BSID and later test

scores calls this argument into question. Nonetheless, the answer to this inquiry cannot

be definitively determined in this data set at the present time. In principle, however,

retesting the subjects as they age provides a means of resolving this question.

A third argument in defense of the genetic story would be one in which the racial

differences are concentrated in higher order thinking (or general intelligence, “g”, see

Jensen, 1998) which may not yet have emerged among one year olds. Unlike the

argument in the preceding paragraph, it is not clear that such a hypothesis is easily tested

even with the passage of time.

As noted earlier, one possible explanation for the growing racial gap with age is a genetics-environment

interaction as proposed by Dickens and Flynn (2001).

References

Acheson, Shawn and Molfese, Victoria J. “Infant and Preschool Mental and Verbal

Abilities: How Are Infant Scores Related to Preschool Scores?” International

Journal of Behavioral Development, 1997, 20(4), pp. 595-607.

Andersson, Helle W.; Sommerfelt, Kristian and Sonnander, Karin. “Gender and its

contribution to the prediction of cognitive abilities at 5 years.” Scandinavian

Journal of Psychology, 1998, 39, pp. 267–74.

Barnard, Kathryn E.; Bee, Helen L.; Clark, Barbara; Eyres, Sandra J.; Gray, Carol

A.; Hammond, Mary A.; Snyder, Charlene and Spietz, Anita L. “Prediction of

IQ and Language Skill from Perinatal Status, Child Performance, Family

Characteristics and Mother-Infant Interaction.” Child Development, 1982, 53(5),

pp. 1134-56.

Baughman, Earl and Dahlstrom, Grant W. Negro and White Children: A

Psychological Study in the Rural South. New York: Academic Press, 1968.

Bayley, Nancy. “On the Growth of Intelligence.” American Psychologist, 1955, 10,

pp.805-18.

Bethel, James; Green, James L.; Kalton, Graham and Nord, Christine. Early

Childhood Longitudinal Study, Birth Cohort (ECLS–B), Sampling. Volume 2 of

the ECLS-B Methodology Report for the 9-Month Data Collection, 2001–02.

Washington, DC: U.S. Department of Education, NCES, 2004.

Birns, Beverly and Golden, Mark. “Prediction of intellectual performance at 3 years

from infant tests and personality measures.” Merrill Palmer Quarterly, 1972, 18,

pp. 53-58.

Bracken, Bruce A.; Insko W. and Sabers, D. “Performance of Black and White

Children on the Bracken Basic Concept Scale.” Psychology in Schools, 1987,

24(1), pp. 22–27.

Bradley, Robert; Caldwell, Bettye M. and Elardo Richard. “The Relation of Infants’

Home Environment to Mental Test Performance from Six to Thirty-six Months: A

Longitudinal Analysis.” Child Development, 1975, 46(1), pp. 71-76.

Brooks-Gunn, Jeanne; Duncan, Greg J.; Klebanov, Pamela K. and Sealand, Naomi.

“Do Neighborhoods Influence Child and Adolescent Development?” American

Journal of Sociology, 1993a, 99(2), pp. 353–95.

Brooks-Gunn, Jeanne; Duncan, Greg J. and Klebanov, Pamela K. “Economic

Deprivation and Early-Childhood Development.” Child Development, 1994, 65

(2), pp. 296–318.

__________. “Ethnic Differences in Children’s Intelligence Test Scores: Role of

Economic Deprivation, Home Environment and Maternal Characteristics.” Child

Development, 1996, 67(2), pp. 396–408.

Burkett, John; McMillen, Marilyn; Owings, Jeffery and Pinkerton, Daniel B.

“Statistics in Brief: Making the Cut: Who Meets Highly Selective College

Entrance Criteria?” Washington, DC: National Education Longitudinal Study

(NELS),1995.

Butler, Bruce V.; Goffeney, Barbara and Henderson, Norman B. “Negro-White,

Male-Female Eight-month Developmental Scores Compared with Seven-Year

WISC and Bender Test Scores.” Child Development, 1971, 42, pp. 595-604.

Campbell, Frances A.; Nicholson, Janet E. and Ramey, Craig T. “The Predictive

Power of the Bayley Scale of Infant Development and the Stanford-Binet

Intelligence Test in a Relatively Constant Environment.” Child Development,

1973, 44, pp. 790-95.

Campbell, Ernest Q.; Coleman, James S.; Hobson, Carol J.; McPartland, James;

Mood, Alexander M.; Weinfeld, Frederic D. and York, Robert L. Equality of

Educational Opportunity. Washington D.C.: U.S. Government Printing Office

(GOP), 1966.

Campbell, Frances A.; MacPhee, David; Ramey, Craig T. and Yeates, Keith O.

“Maternal IQ and home environment as determinants of early childhood

intellectual competence: Developmental analysis.” Developmental Psychology,

1983, 19, pp. 731-39.

Chan Toy, Colleen; Deitz, Jean; Engel, Joyce M. and Wendel, Susan. “Performance

of 6-Month-Old Asian American Infants on the Movement Assessment of Infants:

A Descriptive Study.” Physical & Occupational Therapy in Pediatrics: A

Quarterly Journal of Developmental Therapy, 19(3/4), pp.5-23, 2000.

Coley, Richard J. An Uneven Start: Indicators of Inequality in School Readiness. Policy

Information Report. Educational Testing Service. Princeton, NJ: Educational

Testing Service (ETS), 2002.

Colombo, John. Infant Cognition: Predicting Later Intellectual Functioning.

Newbury Park, CA: Sage Publications Inc., 1993.

Cruse, Carol L.; Dumont, Ron; Price, Linda and Whelley, Peter. “The Relationship

between the Differential Ability Scales (DAS) and the Wechsler Intelligence

Scale for Children--Third Edition (WISC-III) for Students with Learning

Disabilities.” Psychology in the Schools, 1996, 33(3), pp. 203-09.

Dezoete, Anne J.; MacArthur, B.A. and Tuck, B. “Prediction of Bayley and Stanford-

Binet Scores with a group of very low birthweight children.” Child Care, Health

& Development, 2003, 29(5), pp. 367-72.

Dickens, William T.; Flynn, James. “Heritability Estimates Versus Large

Environmental Effects: The IQ Paradox Resolved.” Psychological Review, 2001,

108(2): 346-369.

DiLalla, Lisabeth F.; Cypher, Lisa H.; Fagan, Joseph F.; Fulker, David W.; Haith,

Marshall M.; Thompson, L.A.; Plomin, Robert and Phillips, Kay. “Infant

Predictors of Preschool and Adult IQ: A Study of Infant Twins and Their

Parents.” Developmental Psychology, 1990, 26(5), pp. 759-69.

DiLalla, Lisabeth F.; Lovelace, Laneel and Molfese, Victoria J. “Perinatal, Home

Environment and Infant Measures as Successful Predictors of Preschool cognitive

and Verbal Abilities.” International Journal of Behavioral Development, 1996,

19(1), pp. 101-19.

Enright, Mary K.; Jaskir, John and Lewis, Michael. “The Development of Mental

Abilities in Infancy.” Intelligence, 1986, 10, pp. 331-54.

Escalona, Sibylle K. and Moriarty, Alice. “Prediction of school age intelligence from

infant tests.” Child Development, 1961, 32(3), pp. 597-605.

Feldman, Judith F.; McCarton, Cecelia; Rose, Susan A. and Wallace, Ina.F.

“Information Processing at 1 Year: Relation to Birth Status and Developmental

Outcome During the First 5 Years.” Developmental Psychology, 1991, 27(5), pp.

723-37.

Fryer, Roland and Levitt, Steven. "Understanding the Black-White Test Score Gap in

the First Two Years of School." Review of Economics and Statistics, 2004 86, pp.

447-64.

__________. “The Black-White Test Score Gap Through Third Grade.” American Law

and Economic Review (forthcoming).

Gannon, D.R. “Relationships between 8-Mo. Performance on the Bayley Scale of Infant

Development and 48-Mo. Intelligence and Concept Formation Scores.”

Psychological Reports, 1968, 23, pp. 1199-205.

Gravem, Howard; Ireton, Harold and Thwing, Edward. “Infant Mental Development

and Neurological Status, Family Socioeconomic Status and Intelligence at Age

Four.” Child development, 1970, 41(4), pp. 937-45.

Hamilton, Barton H. and McManus, Brian P. “Competition, Insurance, and Quality in

the Market for Advanced Infertility Treatment.” Mimeograph. St. Louis, MO:

Washington University, John M. Olin School of Business, 2004.

Hernstein, Richard J. and Murray, Charles. The Bell Curve: Intelligence and Class

Structure in American Life. New York: Free Press, 1994.

Honzik, Marjorie P. “Measuring mental abilities in infancy: The value and limitations,”

in M. Lewis, ed., Origins of Intelligence. 2nd ed. New York, NY: Plenum, 1983,

pp. 67-105.

Humphreys, Lloyd G. “Trends in Levels of Academic Achievement of Blacks and Other

Minorities.” Intelligence, 1988, 12(3), pp. 231–60.

Jensen, Arthur R. Educability and Group Differences. New York: Free Press, 1973.

__________. The G Factor: The Science of Mental Ability. Westport, CT: Praeger, 1998.

Jensen, Arthur R. and Rushton, J. Philippe. “Thirty Years of Research on Race

Differences in Cognitive Ability.” Psychology, Public Policy, and Law, 2005,

11(2), pp. 235-94.

Karlberg, P.; Klackenberg, G.; Engström, L.; Klackenberg-Larsson, L.;

Lichtenstein, H.; Stensson, J. and Svennberg, I. “The development of children

in a Swedish urbancommunity: A prospective, longitudinal study: parts l-VI.”

Acta Paediatrica Scandinavica Supplement, 1968, 187.

Kaufman, Alan S., and Kaufman, Nadeen L. K-ABC: Kaufman Assessment Battery for

Children. Circle Pines, MN: American Guidance Services, 1983.

Kopp, Claire B. and McCall, Robert B. “Predicting later mental performance for

normal, at-risk, and handicapped infants,” in Paul B. Baltes and Orville G. Brim,

Jr., eds., Life-span development and behavior. Vol 4. New York: Academic Press,

1990, pp. 31-61.

Krohn, Emily J. and Lamp, Robert E. “Current Validity of the Stanford-Binet Fourth

Edition and K-ABC for Head Start Children.” Journal of Psychology, 1989,

27(1), pp. 59–67.

Malina, Robert M. “Racial/ethnic variation in the motor development and performance

of American children.” Canadian Journal of Sport Sciences, 13(2), pp.136-43,

1988.

McGrath, Ellen; Bellinger, David C.; Newburger, Jane W.; Wypij, David and

Rappaport, Leonard A. “Prediction of IQ and achievement at age 8 years from

neurodevelopmental status at age 1 year in children with D-transposition of the

great arteries.” Pediatrics, 2004, 114 (5), pp. 572-76.

McGurk, Frank C. J.; Osborne, Robert T.; Osborne, Travis R.; and Shuey, Audrey

M., eds. The Testing of Negro Intelligence. Vol. 2. Athens, GA: The Foundation

for Human Understanding, 1982.

Myrianthopoulos, Ntinos C.; Naylor, Alfred F. and Willerman, Lee. “Intellectual

Development of Children from Interracial Matings: Performance in Infancy and at

4 Years.” Behavior Genetics, 1974, 4(1), pp. 83-90.

Naglieri, Jack A. “WISC-R and K-ABC Comparison for Matched Samples of Black and

White Children.” Journal of Social Psychology, 1986, 24(1), pp. 81–88.

Neisser, Ulric; Boodoo, Gwyneth; Bouchard, Thomas J.; Boykin, A. Wade; Brody,

Nathan; Ceci, Stephen J.; Halpern, Diane F.; Loehlin, John C.; Perloff,

Robert, Sternberg, Robert J. et al. “Intelligence: Knowns and unknowns.”

American Psychologist, 1996, 51, pp. 77-101.

Nisbett, Richard E. “Race, Gentics and IQ,” in Christopher Jencks and Meredith

Phillips, eds.,

The Black-White test score gap. Washington, DC: The Brookings

Institute, 1998, pp. 86-102.

Nord, Christine; Andreassen, Carol; Branden, Laura; Dulaney, Rick; Edwards,

Brad; Elmore, Anne; Flanagan, Kristin Denton; Fletcher, Philip; Green,

Jim; Hilpert, Richard et al. Early Childhood Longitudinal Study, Birth Cohort

(ECLS-B), User’s Manual for the ECLS-B Nine-Month Public- Use Data File and

Electronic Code Book. Washington, DC: U.S. Department of Education, NCES,

2004.

Phillips, Meredith; Brooks-Gunn, Jeanne; Crane, Jonathan; Duncan, Greg J. and

Klebanov, Pamela K. “Family Background, Parenting Practices, and the Black-

White Test Score Gap,” in Christopher Jencks and Meredith Phillips, eds.,

The

Black-White test score gap. Washington, DC: The Brookings Institute, 1998, pp.

103-45

Phillips, Meredith. “Understanding Ethnic Differences in Academic Achievement:

Empirical Lessons from National Data,” in David W. Grissmer and Michael J.

Ross, eds., Analytic Issues in the Assessment of Student Achievement.

Washington, DC: U.S. Department of Education, National Center for Education

Statistics (NCES), 2000, pp 103-32.

Rose, Susan A. and Wallace, Ina F. “Visual recognition memory: a predictor of later

cognitive functioning in preterms.” Child Development, 1985, 56 (4), pp. 843-52.

Rushton, Philippe J. “Race and crime: international data for 1989-1990.” Psychological

Reports, 1995, 76(1), pp. 307-12.

Scarr, Sandra. Race, Social Class and Individual Differences in I.Q. Mahwah, NJ:

Lawrence Erlbaum Associates, 1981.

Scott, R. and Sinclair, D. “Ethnic-related cognitive profiles of black and white pre-

school children.” Homo- Journal of Comparative Human Biology, 1997, 28,

pp.116-20.

Shuey, Audrey M. The Testing of Negro Intelligence. Lynchburg, VA: J.P. Bell

Company, 1958.

Wilson, Ronald S. “The Louisville Twin Study: Developmental Synchronies in

Behavior.” Child Development, 1983, 54(2), pp. 293-316.

Figure 1

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

11.523456789

Age (years)

Correlation with 8-12 month BSID-MDI

Notes: See Appendix Table 1 for a list of sources from which the correlations are based. Each is a (sample size) weighted average of

the correlations found across all relevant studies.

Figure 1: The Correlation between Bayley Scores and Later IQ

Figure 2: The Distribution of Bayley Scores, by Race

Variables All Races White Black Hispanic Asian Other

Cognitive Development

Mental function composite score 0.0002 0.0182 -0.0258 -0.0185 -0.0513 0.0065

(0.9996) (0.9426) (0.8951) (0.9156) (0.9551) (1.1528)

Demographics

age 10.2677 10.2379 10.2920 10.2983 10.3397 10.3285

(1.9362) (1.7914) (1.7777) (1.822) (1.8556) (2.4305)

female 0.4892 0.4848 0.4855 0.4987 0.4708 0.5130

(0.6835) (0.6401) (0.6232) (0.627) (0.6593) (0.8661)

region:

northeast 0.1684 0.1818 0.1575 0.1470 0.1974 0.1428

(0.5513) (0.5355) (0.4856) (0.4609) (0.6072) (0.7097)

midwest 0.2223 0.2818 0.2092 0.1076 0.1583 0.2527

(0.5511) (0.558) (0.4997) (0.3649) (0.4284) (0.7413)

south 0.3675 0.3601 0.5569 0.3147 0.1876 0.2906

(0.6566) (0.6132) (0.6237) (0.5854) (0.5275) (0.8)

west 0.2418 0.1763 0.0765 0.4307 0.4567 0.3140

(0.5779) (0.4788) (0.3479) (0.6199) (0.651) (0.7526)

Home environment

SES quintiles:

1st quintile (lowest) 0.2000 0.0850 0.3424 0.3789 0.1021 0.1753

(0.5346) (0.3471) (0.5894) (0.6086) (0.4213) (0.6472)

2nd quintile 0.1999 0.1572 0.2408 0.2633 0.1417 0.2622

(0.5391) (0.4543) (0.5325) (0.5529) (0.5695) (0.7867)

3rd quintile 0.1999 0.2087 0.2091 0.1817 0.1436 0.2097

(0.5456) (0.5174) (0.511) (0.482) (0.4305) (0.6977)

4th quintile 0.2005 0.2624 0.1289 0.1146 0.1538 0.2047

(0.5589) (0.569) (0.4212) (0.3985) (0.4298) (0.6962)

5th quintile (highest) 0.1997 0.2868 0.0788 0.0615 0.4588 0.1480

(0.5554) (0.5849) (0.339) (0.3013) (0.6531) (0.5923)

number of siblings 0.9848 0.9568 1.1316 0.9899 0.7783 0.9754

(1.4761) (1.2918) (1.543) (1.46) (1.2742) (1.8726)

biological mother and biological father

present

0.7892 0.8833 0.4111 0.7901 0.9212 0.7322

(0.5352) (0.4051) (0.6171) (0.5085) (0.4085) (0.7824)

one biological parent present 0.1938 0.0960 0.5793 0.1950 0.0754 0.2504

(0.5155) (0.3716) (0.6185) (0.4939) (0.4058) (0.7737)

one biological parent and one non-

biological parent present

0.0114 0.0142 0.0067 0.0094 0.0015 0.0106

(0.1455) (0.1508) (0.1007) (0.1109) (0.0349) (0.1715)

other parental configuration 0.0056 0.0065 0.0030 0.0054 0.0020 0.0069

(0.1042) (0.0979) (0.0578) (0.1097) (0.0424) (0.056)

mother's age 27.2587 28.3941 25.1252 26.0201 29.4158 25.8535

(8.3082) (7.6037) (7.5755) (7.334) (7.2416) (10.6263)

parent as teacher score 29.6979 31.2523 28.4809 27.3848 27.0006 29.9106

(17.4207) (15.4744) (16.1934) (16.909) (19.1218) (23.2064)

Table 1: Summary statistics

Prenatal environment

birthweight:

< 1500 0.0127 0.0102 0.0240 0.0121 0.0072 0.0154

(0.0445) (0.0348) (0.0635) (0.0419) (0.054) (0.0749)

> 1500 & < 2500 0.0621 0.0557 0.0998 0.0550 0.0651 0.0601

(0.1934) (0.1616) (0.2524) (0.1773) (0.2618) (0.2609)

> 2500 & < 3500 0.5451 0.5123 0.6023 0.5669 0.6832 0.5458

(0.6776) (0.6391) (0.5985) (0.6153) (0.6079) (0.8588)

3500 or more 0.3796 0.4212 0.2732 0.3656 0.2445 0.3778

(0.6737) (0.6367) (0.5794) (0.608) (0.5815) (0.8539)

Percent premature 0.1167 0.1016 0.1685 0.1169 0.0948 0.1534

(0.3677) (0.3186) (0.412) (0.3492) (0.3246) (0.5675)

Days premature (conditional on being

premature)

21.1239 21.0457 22.7998 20.2129 17.4876 21.4877

(23.6229) (24.2518) (23.5968) (19.9711) (13.4793) (32.8074)

multiple birth:

singleton 0.9668 0.9624 0.9693 0.9768 0.9786 0.9452

(0.1007) (0.0997) (0.091) (0.0749) (0.1175) (0.1813)

twin 0.0298 0.0347 0.0300 0.0214 0.0197 0.0252

(0.0927) (0.0925) (0.0882) (0.0718) (0.1125) (0.113)

higher order 0.0018 0.0028 0.0007 0.0007 0.0016 0.0005

(0.0296) (0.0337) (0.0214) (0.0102) (0.0332) (0.0118)

NOTES: The entries are means and standard deviations of child-level data for those children in ECLS-B who

do not have missing values for test scores. Test scores are a mental composite score, normalized to have a mean

of 0 and a standard deviation of 1. The category “White” includes only non-Hispanic Whites. Precise

definitions of the variables are provided in the data appendix. The SES measure incorporates information on

parental education, occupational status, and family income. Days premature is conditional on being premature.

The total number of children in the sample who receive a positive weight in the estimation is 10,132. In all

cases, sample weights provided with ECLS are used in the calculations.

(1) (2) (3) (4) (5)

(6)

black -0.061 -0.063 -0.037 -0.026 -0.022 -0.002

[0.028]* [0.017]** [0.018]* [0.019] [0.019] [0.019]

hispanic -0.013 -0.046 -0.015 -0.02 -0.011 -0.012

[0.028] [0.016]** [0.017] [0.017] [0.017] [0.017]

asian -0.026 -0.068 -0.063 -0.072 -0.056 -0.048

[0.035] [0.020]** [0.020]** [0.020]** [0.020]** [0.020]*

other -0.013 -0.039 -0.024 -0.026 -0.023 -0.016

[0.039] [0.025] [0.025] [0.025] [0.025] [0.024]

controls:

age, gender, region yes yes yes yes yes

SES yes yes yes yes

home environment yes yes yes

parents' score yes yes

prenatal conditions yes

Observations 10132 10132 10132 10132 10132 10132

R-squared0.140.680.680.680.680.

Robust standard errors in brackets

* significant at 5%; ** significant at 1%

Table 2: Estimating Group Differences in the Mental Function Composite Score

Dependent variable: Standardized Mental Function Composite Score

NOTES: The dependent variable is the mental composite score, which are normalized to have a mean of 0 and a

standard deviation of 1 in the full, unweighted sample. Non-Hispanic Whites are the omitted race category, so

all of the race coefficients are gaps relative to that group. The unit of observation is a child. Standard errors are

in parentheses. Estimation is done using weighted least squares, using sample weights provided in the data set.

In addition to the variables included in the table, indicator variables for children with missing values on each

covariate are also included in the regressions. All regressions include interviewer fixed effects.

Table 3: Sensitivity Analysis of Group Differences in Mental Function Composite Score

black hispanic asian other

baseline -0.002 -0.012 -0.048 -0.016

[0.019] [0.017] [0.020]* [0.024]

unweighted 0.020 0.000 -0.061 0.007

[0.014] [0.013] [0.015]** [0.015]

males only -0.018 0.007 -0.024 -0.041

[0.028] [0.023] [0.030] [0.039]

females only 0.021 -0.012 -0.073 0.004

[0.024] [0.025] [0.027]** [0.029]

SES quintile 1 -0.046 -0.042 -0.143 0.001

[0.049] [0.044] [0.062]* [0.059]

SES quintile 2 -0.045 -0.059 -0.095 0.028

[0.035] [0.036] [0.055] [0.066]

SES quintile 3 -0.006 0.000 -0.019 -0.060

[0.042] [0.036] [0.047] [0.043]

SES quintile 4 0.028 0.022 -0.108 -0.028

[0.045] [0.039] [0.050]* [0.042]

SES quintile 5 0.043 -0.008 -0.017 -0.034

[0.051] [0.048] [0.035] [0.045]

if both biological parents present 0.009 -0.020 -0.034 0.009

[0.024] [0.019] [0.020] [0.027]

if a biological parent not present 0.000 0.021 -0.168 -0.043

[0.035] [0.039] [0.075]* [0.049]

if northwest -0.010 -0.010 0.018 -0.090

[0.045] [0.046] [0.060] [0.088]

if midwest 0.113 0.033 0.019 0.029

[0.041]** [0.040] [0.048] [0.044]

if south -0.053 -0.037 -0.070 -0.031

[0.026]* [0.027] [0.035]* [0.042]

if west 0.063 -0.005 -0.076 -0.009

[0.052] [0.030] [0.031]* [0.036]

if urban, inside UA 0.011 0.010 -0.037 -0.013

[0.021] [0.019] [0.021] [0.026]

if urban, outside UC -0.050 -0.072 -0.015 -0.030

[0.082] [0.054] [0.076] [0.072]

if rural -0.034 -0.117 -0.045 -0.041

[0.057] [0.069] [0.224] [0.053]

if normal birthweight -0.008 -0.012 -0.047 -0.023

[0.020] [0.018] [0.021]* [0.026]

if moderately low birthweight -0.005 -0.030 -0.079 -0.043

[0.041] [0.040] [0.059] [0.063]

if very low birthweight 0.071 0.009 -0.212 -0.079

[0.046] [0.054] [0.149] [0.101]

controlling for motor skills -0.034 -0.005 -0.066 -0.037

[0.018] [0.016] [0.018]** [0.023]

age < 9 months -0.059 -0.046 -0.054 0.037

[0.038] [0.033] [0.038] [0.048]

9 months <= age < 10 months -0.005 0.000 -0.019 0.002

[0.025] [0.024] [0.025] [0.032]

10 months <= age < 11 months 0.024 -0.022 -0.040 -0.002

[0.034] [0.032] [0.035] [0.047]

11 months <= age < 12 months 0.003 0.014 -0.105 -0.044

[0.064] [0.063] [0.070] [0.120]

age >= 12 months -0.014 0.039 -0.038 -0.071

[0.072] [0.063] [0.066] [0.081]

NOTES: Specifications in this table are variations on those reported in column (6) of Table 2. Only the race

coefficients are reported. The top row simply reproduces the baseline results in column (6) of Table 2. The

remaining rows correspond to different weights, socio-economic quintiles and other particular subsets of the

data. For further details of the baseline specification, see the notes to Table 2.

explores objects

purposefully babbles early problem uses words

black -0.0007 -0.0012 0.0022 -0.0027 -0.0033

[0.0005] [0.0039] [0.0047] [0.0046] [0.0039]

hispanic -0.0006 -0.0039 -0.0021 -0.0013 -0.0013

[0.0004] [0.0039] [0.0043] [0.0041] [0.0033]

asian -0.0004 -0.0054 -0.0128 -0.0088 -0.0048

[0.0004] [0.0044] [0.0053]* [0.0047] [0.0030]

other -0.0002 -0.0015 -0.0031 -0.0041 -0.0031

[0.0005] [0.0051] [0.0064] [0.0061] [0.0046]

controls:

age, gender, region, mother's age yes yes yes yes yes

SES yesyesyesyesyes

home environ yes yes

parents' score yes yes

prenatal cond yes yes

mean of depe .0906 0.0317

sd of depend .2172) (0.1399)

Observations 0132 10132

R-squared 0.64 0.44

Table 4: Estimating Group Differences in the Underlying Components of the Mental Function Composite Score

ment yes yes yes

yes yes yes

itions yes yes yes

ndent variable 0.9938 0.9058 0.5471 0

ent variable (0.0145) (0.1555) (0.268) (0

10132 10132 10132 1

0.21 0.43 0.73

NOTES: Entries are unadjusted mean scores on specific components of the mental composite score. They are

proficient probability scores, which are constructed using IRT scores and provide the probability of mastery of

a specific skill set.

Appendix Table 1: Literature on the Relationship Between Bayley Scores and Future IQ

Author(s)

Age of Bayley

(months)

Future IQ Test and Age (years) Correlation N

Gannon, 1968

8 Stanford-Binet at age 4 0.13 371

Butler, Goffeney and

Henderson and, 1971

8 Wechsler Intelligence Scale for Children at age 7 0.19 626

Ireton, Gravem and Thwing,

1970

8 Stanford-Binet at age 4 0.25 500

Myrianthopoulos, Naylor and

Willerman, 1974

8 Stanford-Binet at age 4 0.22 129

Bayley, 1955

9 & 12 Mean 16-18 IQ 0.32, 0.30 45

Ramey et al., 1973

9 to 12 Stanford-Binet at age 3 0.71 11

9 & 12 Stanford-Binet at age 3 0.38 340

Wechsler Preschool and Primary Scale of Intelligence at ages 4, 5, and

0.29, 0.35,

0.33

340

Wilson, 1983

Wechsler Intelligence Scale for Children at ages 7, 8, and 9 0.27, 0.27,

0.27

340

Bradley, Caldwell and

Elardo 1975

12 Stanford-Binet at age 3 0.32 77

Barnard et al.1982

12 Stanford-Binet at age 4 0.21 156

Enright, Jaskir and Lewis,

1986

12 Stanford-Binet at age 3 0.12 116

DiLalla et al., 1990

12 Stanford-Binet at age 3 0.32 40

Rose et al., 1991

12 Stanford-Binet at ages 3 and 4 0.30, 0.22 40

DiLalla, Lovelace and

Molfese, 1996

12 Stanford-Binet at ages 3 and 4 .17, .14 94

Acheson and Molfese, 1997

12 Stanford-Binet at ages 3, 4 and 5 .14, .15, .06 89

NOTES: To generate relevant literature, Pubmed, EBSCOhost and JSTOR were searched for all years. Keywords used for the search included, “Bayley Scale of

Infant Development,” “BSID, “predictive power,” “correlation,” and “IQ” – which generated a list of roughly 2,000 references. Among these, we selected

according to the following criteria. (1) the test of infant cognition was the Bayley Scale of Infant Development (BSID); (2) The “future” test age was

administered more than 2 years after the infant measure was taken (correlations at 12 and 18 months were obtained from Wilson 1983); (3) the initial BSID had

to be administered between 6 and 12 months of age; (4) the sample had to be “representative” – we omitted studies with subjects afflicted by major health issues

(e.g., heart disease, Down syndrome, mental retardation) or severe complications at birth (i.e. pre-term, mother used drugs or alcohol); and (5) only studies done

in the United States were used. The final compilation contained 14 studies.

(1) (2) (3) (4) (5) (6)

Black -0.061 -0.063 -0.037 -0.026 -0.022 -0.002

[0.028]* [0.017]** [0.018]* [0.019] [0.019] [0.019]

Hispanic -0.013 -0.046 -0.015 -0.02 -0.011 -0.012

[0.028] [0.016]** [0.017] [0.017] [0.017] [0.017]

Asian -0.026 -0.068 -0.063 -0.072 -0.056 -0.048

[0.035] [0.020]** [0.020]** [0.020]** [0.020]** [0.020]*

Other -0.013 -0.039 -0.024 -0.026 -0.023 -0.016

[0.039] [0.025] [0.025] [0.025] [0.025] [0.024]

Age 488.089 483.958 498.736 505.248 504.52

[110.641]** [110.111]** [109.414]** [108.913]** [107.444]**

Age^2 -96.648 -95.941 -98.855 -100.032 -100.028

[21.419]** [21.318]** [21.180]** [21.086]** [20.798]**

Age^3 9.525 9.465 9.751 9.856 9.867

[2.062]** [2.053]** [2.039]** [2.031]** [2.002]**

Age^4 -0.467 -0.464 -0.478 -0.483 -0.484

[0.099]** [0.098]** [0.098]** [0.097]** [0.096]**

Age^5 0.009 0.009 0.009 0.009 0.009

[0.002]** [0.002]** [0.002]** [0.002]** [0.002]**

female 0.061 0.061 0.063 0.061 0.069

[0.011]** [0.011]** [0.011]** [0.011]** [0.011]**

Socioeconomic Status Quintiles:

2 0.024 0.021 0.018 0.021

[0.018] [0.018] [0.018] [0.018]

3 0.076 0.067 0.055 0.052

[0.018]** [0.019]** [0.019]** [0.019]**

4 0.081 0.073 0.054 0.047

[0.019]** [0.021]** [0.021]** [0.021]*

5 (highest) 0.079 0.074 0.048 0.042

[0.020]** [0.022]** [0.023]* [0.022]

Number of Siblings:

1 -0.066 -0.064 -0.063

[0.013]** [0.013]** [0.013]**

2 -0.079 -0.079 -0.07

[0.017]** [0.016]** [0.017]**

3 -0.135 -0.134 -0.117

[0.025]** [0.025]** [0.025]**

4 -0.14 -0.13 -0.114

[0.044]** [0.044]** [0.044]**

5 -0.102 -0.081 -0.042

[0.059] [0.059] [0.057]

6 -0.122 -0.126 -0.136

[0.077] [0.075] [0.073]

Family configuration:

Single biological parent -0.019 -0.017 -0.013

[0.017] [0.017] [0.017]

Biological parent and other parent -0.039 -0.051 -0.039

[0.053] [0.053] [0.052]

Appendix Table 2: Estimating Group Differences in the Mental Function Composite Score

Dependent variable: Standardized Mental Function Composite Score

Other parent type -0.044 -0.051 0.002

[0.110] [0.108] [0.111]

Midwest -0.038 -0.035 -0.031

[0.048] [0.047] [0.047]

South 0.043 0.058 0.06

[0.056] [0.056] [0.055]

West 0.016 0.02 0.011

[0.063] [0.063] [0.063]

Mother's age -0.14 -0.118 -0.07

[0.669] [0.666] [0.660]

Mother's age^2 0.012 0.01 0.005

[0.047] [0.047] [0.046]

Mother's age^3 (*100000) -43.527 -35.908 -13.402

[160.510] [159.846] [158.698]

Mother's age^4 (*100000) 0.744 0.598 0.146

[2.674] [2.665] [2.648]

Mother's age^5 (*100000) -0.005 -0.004 0

[0.017] [0.017] [0.017]

Mother's age missing -0.506 -0.41 -0.22

[3.718] [3.697] [3.664]

Parentscore -0.269 -0.3

[1.226] [1.220]

Parentscore^2 0.015 0.017

[0.080] [0.080]

Parentscore^3 (*100000) -38.583 -48.659

[256.303] [256.023]

Parentscore^4 (*100000) 0.485 0.669

[4.012] [4.014]

Parentscore^5 (*100000) -0.002 -0.004

[0.025] [0.025]

Mparentscore -1.741 -1.897

[7.321] [7.253]

Birthweight:

< 1500 -0.355

[0.090]**

>= 1500 & < 25000 -0.051

[0.087]

>= 2500 & < 3500 0.068

[0.086]

>= 3500 0.132

[0.086]

Days Premature:

0 0.103

[0.060]

7 0.032

[0.066]

14 -0.022

[0.069]

21 0.075

[0.070]

28 -0.097

[0.078]

35 0.005

[0.076]

42 0.005

[0.087]

49 -0.061

[0.076]

56 0.045

[0.085]

63 -0.059

[0.075]

70 -0.291

[0.083]**

77 -0.365

[0.074]**

Singleton birth -0.001

[0.060]

Twin birth -0.077

[0.060]

Triplet or more birth -

Constant 0.013 -982.802 -973.364 -1,002.51 -1,015.15 -1,012.18

[0.014] [227.402]** [226.276]** [224.953]** [224.352]** [221.337]**

Observations 10132 10132 10132 10132 10132 10132

R-squared0.140.680.680.680.680.71

NOTES: See notes to Table 2. The columns in this table report the full regression results of

specifications (1) through (6) in Table 2.