Continuation of NLS Discussion Paper 92-13
Part 3 of 3
This version of the paper was split for web delivery.
3
~apter
Effects of Parentil ~aractertitics
ad
Mor
tibt
on the Re-m
“to Edu~tion,
~erience
Joseph G. Altonj i
~0~
A. W
IntrO&ctiOn
~is
chapter exaines
whether the education and experience
wages vary sys thematically with cognitive ability and with f~ily
charact.eristics that ifiluence the quantity of formal education
question is of interest for two reasons.
First, experience
mile
enomous
between
these two variables,
seeking to measure the relationship
only a few studies have ex~ined
return to these variables
little is knom
influences
education and of ttie spent in the labor market. ‘ -
of children.
me
ti=.
In particular,
of
is suqris ing in 1igh:..
the effects of..
and on the income
literature on parental effects on children’s
on the direct effect of ifierited ability and family enviro~ent
143
are
earnings and
the wage benefits
at taiment
levels’ rather than on the effects of these variables
~ii
there is an
of the fact that there is an extensive literature docwenting
parental character,istics on both the educational
chosen.
the extent to which the
depends on worker characters
about how family background
background
and education
among the two.most. ~portant. .determinants of earnings.
literature
slopes of
income focuses
on wage
On the rate of return to
education and experience. 1
Some educational
attaiment
studies have exmined
the influence of parents’ education and financial resources on years of
schooling and achievement
parental
in school, and a smaller nmber
effects on school quality. 2
whether parents and the f~ily
detemi~nts
, More generally,
little is kon
and
about the
of the value of a year of school.4
Our second re=on
education
there is little evidence on
influence the rate of return to education
in the labor market. 3
experience
However,
have ex-ined
for being interested
slopes is as a source of fmily
in family determinants
differences
of
in the demand for
1. For ~x~ple,
all Of the studies smeyed
in Griliches (lg~g) ass~e
that the rate of return to education is the same for all individuals.
2.
See
siebe~t
(1985) for a disc~~iOn
‘f this literature
and
detailed references.
3. we di~c~~ a parallel study b; AltOn.1i (1988) below.
Hauser (1973)
finds 1ittle evidence”-that father’ ~ o~cupati~n has much ef feet on the return
to education for men in a cross section analysis that permits the education
slope and intercept to va~ with father’s occupation.
Using the Kal=azoo
twins &ta, Olneck (Chapter 6 in Jencks et al. , WhO Gets Ahead?) finds little
evidence that education slopes differ by IQ or by father’s education.
4
Altonj i (1988) and Morgan and Sirageldin (1968) are examples of
studies that examine the relationship between school variables and future
earnings and provide references to a small nmber
of other studies.
There is
a large literature on race and sex differences in the rates of return to
education.
(See Cain (1986) and Welch (1973) .) However, Welch (1973) and a
recent paper by Card and Krueger (1990) are mong a handful of studies that
have looked at the effects” of schooling inputs on the returns to education.
Card and Krueger find substantial eff ic~ of student/teacher ratio, the
relative sala~ of teachers, and the length of the school tem on est~tes
of
the rates of return to education.
They rely O? variation across states and
age cOhOrts in th+se input measures and dO nOt cOntrOl fOr differences in
parental inputs. TO the extent that differences in the average education of
parents rabes
the demand for school quality in a state and has a positive
direct effect on rates of return to education, their results my overstate
the effects of school inputs on the return to education.
On the other hand,
we find little evidence below of a substantial effect of parental education on
the return to education.
144
education.
In our &b
matched fmily
set the correlations
of education
members are .46 fo~ fathers and sons,
levels between
.43 for. mothers and. sons,
.40 for’ fathers and daughters , .41 for mothers and tiughters,
brothers,
.58 for
.5.0for sisters , and .ha far brothers and sisters.
educational
attaiment
take the view that family background
primarily by influencing
the amount of education
the rate “of.return to education constant.
Most studies of
affects education
individuals
obtain, holding
However, education
such as Willis and Rosen (1979) imply that family ba:kggound
choice models
characters
that raise the rate of return to schooling may induce individuals
school longer.
of education
members
to .stay in
Perhaps the strong parental and sib.1ing correlat io~
.
arises in part because
there is a correlation
tics
in years
across family
in the economic value of education.
In this chapter we take advantage of the NV. data on .sibl ing pairs to
esttiate a simple model which measures parental effects Oh the ‘education and
experience
slopes of children.
log wage is specified
control variables.
co depend upon education,
relationship
between years of education
and the Wage.
family characteristics
of education
Since education
education
as well
are likely to
that have an independent
that omitted
will bias estimates of the return to education.
145
..
of the effect “of”IQ on the
influence on wages, there” is a strong possibility
variables
to
and mother’s
that predicC years
as the interaction between education and parent’s
depend upon unobsened
and a set of
variables , school
characteristics , and personal characteristics
We also examine the interaction
in which the
coefficients
including father?s edu=tion
and an index of faily. background
completed.
and experienc&,
We al”low the education and experience
depend upon on other variables,
education,
Cons ider a standard wage equation
family
We deal with this
by using a sibling fixed effect -to control for unobsened
co=on
to siblings.
idios~cratic
song
that are
for the family does not eliminate
Of course, controlling
differences
variables
siblings in cognitive ability and motivational
factors which have an independent effect on both wages and education.
In most
of our ,specifications, we control for a measure of IQ to reduce the
possibility
mat
the estimated
ability differerices betieen siblings will lead to biases
returns. 5
in
~ur hOpe is that most of the remaining b i~s is
absorbed by the estimate of the main effect of education on wages rather than
by the estimates
of the extent to which background variables
estimates of education.
shift the
Below we analyze the potential biases using a simple
model of wages and the demand for education.
Our main result is that the effect of the child’s IQ, father, s education,
mother’s education,
characteristics,
and index of family background,
and persoml
characters
seconda~
tics that predict years of schooling
completed have only weak influences on the relationship
wages, and between
the f-ily
labor market ~perience
background
statistically
background
me
fraework
between
education
and
In a ntiber of cases ,
direction
In view of the results,
that the effect of family background
responsible
and wages.
interactions work in the ~ong
insignificant.
school
on the education
for more than a small part of the powerful
or are
it seems unlikely
to us
slope of wages is
effect of f~ily
on years of school completed.
paper is organized as follows.
for the study.
III presents
Section I presents
Data issues are addressed
the econometric
in section II.
the empirical results, and in section IV we smarize
Section
the
5. me b~ic
aPPriach “as implemen-ted in Altonj i (1988) Using a small
s=ple
for the Panel Study of Income D~amics,
without a control for IQ. His
results are discussed below.
146
_
findings and offer some suggestions
for future research.
I - tiononetric
Raework
.
Cons ider the following
log wage equation for ..a
~youn”g woman:
,,=
(1) Wdht - ZdhtB1 +_ 2fiB2 + ZfiB3. +
is the log wage, and Z&
In (1) W&e
and father, respectively.
bbor
market experience.
characteristics
convenience,
education,
tem
The variable EDdht is education,
rates.
is uncorrelated
The education and experience
characters
For expository
‘dht
~d and “~h are indivi~l
is a transitory
fomal
seconda~
schooling
process
error
with the other right hand side
tics that are related to innate c.ognitive ~ ility,
primaq
and secOn&~
school qulity,
the amount of ttie and energy that children devote “to schooling
and
in
slopes rdh and rdh dep-end On family
quality of early childhood development,
prima~
and
a cubic specification
Finally,
specific error components;
in the eqwtion.
is
in most of our empirical work we include an
and a quadratic in experience.
variables
and ~Pdht
of the educacion
between education and experience,
that we asswe
tics Of the mother
The vector Zdht consists of other obsened
However,
component
+. Cd + Sh + .*E
and Zfi are characters
we work with a linear specification
specific and faily
background
+ r&HPdht
of the woman that affect her wage
experience profiles .
interaction
r&ED&
school,
and
and
as
the
success
infomal
147
of
parents
as
teachers at home.
aides
when
.in
in
the
Specifically,
(2)
rfi
- alXfi
+ .2X&
+ r + Vh
rdb -blxfh+bzxti+
where v
and ~
h
r+%
are .nobsemed
household
specific error components
rates of return to”education and experience.
.
One can eas ~ly generalize
correlated.
variables
Vh and ~
may be
(2) to include person specific
such as IQ scores, and we do so in the empirical work below.
variables
Using
(3)
me
affecting
(2) to substitute for rdh and rdh in (1) leads to
Wdhc - ‘dh=B~ + ZmhB2 + ZfiB3
;
+ [r + alXfi + a2Xmh ]EDdh + [r + blXfi + b2X&]~p&t
+ cd + ~h + VhEDdh + Phmpdht
+ ‘dht
Because EDti is likely to becorrelated
least squres
estimates .
eliminate
estfiation
of the above eqution
However, by dif ferencing
with the error te~s
will” lead to biased par~eter
(3) for pairs of siblings one can
the terms involving Ch from the equation.
and d’ , the difference
equation
sh and Vh,
For siblings
indexed by d
is
(4) Wdh= ~ Wd,ht - [Zdht - Zd,ht]B1
+[r+a
X
lfb
+ a2xmh]
[EDdh - EDd,h ]
+ [r + blXfi + bzxfil [~pdht
+ Sd - cd, + vh[ED&
me
- Wpd, htl
- EDd,hl + ~[mpdht
- ‘Pal, ht] + ‘dht - ud, ht
error components Vh and ph are constant within the household
148
and so
will be mcorrelated
with the explamtory
However, potent ial simultaneity
cd, if these idiospcrat
mitigate
in the above eqwtion.
problems could arise from the term cd ..-
ic components capturing differences
say, are cofielated to the differences
In a nuber
siblings d and d’ .
variables
in the education
of the specifications
in productivity,
levels chosen by
below, we try to
this problem by adding a measure Of IQ to (3) , which implies that the
difference between
the IQ score: of d and d’ will appear in (4) .
However:
is almost certainly not a perfect control, and in the next few paragraph
we
a simplified
whether
our
model of wages and the demand for
education
of the effects o“fparental education
estimates
to
IQ
we
ex~ine
on returns to
education will be subj ect to bias.
Suppose that
experience
‘2
(2) is equal to zero, that there is no
in equation
slope, and theat Zdht is a constant
without. much 10SS of generality,
components.
men
for all siblings.
suppose we ignore the trans ito~
AISO ,
wage
(:4) reduces to
(5) Wdh - Wd,h = [r
+
+
alxfi]
cd
-
cd,
[ED&
+
- EDd,h]
Vh[EDdh
-
EDd,hl
Let the young woman’s demand for education be
(6) ED&-
c1+c2xfi+c3sd+c4
Vh+ud
where Cl is a constant, C2 > 0, C3 .> 0, arid-C4 > 0.
from a model in which individuals
seek to”maximize
value of lifetime income, and where the education
149
One can easily derive
(6)
the present discounted
slope in the wage equation
is related to Xfi, cd, Vh, and w ~, and the interest rate used to discount
future income depends upon one or more of these variables
and on the nuber
of
years of education chosen.
Finally, asswe
(5) separately for f-i.lies “with Xfi-o and X
estimates
limit of
- EDd,hl , Cd ‘ Cd, + vh[EDdh
----------
(7) -r-+
-------.-
VaY(C3
The probability
(8)
- EDd,hl
r+
-
limit of the estimator of r + al for the X=-l
Var(C3
The differe-nce betieen
1
(8) and (7) is the probability
I Xfi-l)
limit of the OLS
are independent
(7) is an inconsistent
argment
of
(5)
and covariances
of Xfi, then the difference
from the two separate regressions
on
EDd,hl
is
------------- ,----[Sd - Cd’ ] + [Ud - ‘d, ] I Xfi-l )
It follows that if the conditional variances
me
r + al. ._
-
s~ple
that one would obtain if one used the pooled s=ple
families to estimate
express iom
[ Xfi-o)
.-: -----------------------------------
al+
esttiacor of a
The probability
[cd- ,Sd,] + ‘@d - Od, 1 I ;fi-o)
COV( [EDdh ‘ EDd,hl . :d - cd, + Vh[EDdh
.
-1.
fh
will be
the estimator of r for the Xfi-O smple
COV([ED&
Suppose one
that Xfi takes on only two values, O and 1.
is a consistent
in the coefficients
estimator of- al even thou~
estimator of r and (8) is an inconsistent
can easily be generalized
in the above
estimator
to the case in which Xfi tak@~
a variety of values .
In practice,
our specifications
include nonlinear
150
of
education
terns and
other explanatory variables.
interaction
Some specifications
include more than one
term with education, and all of our models include a measure of
experience, which was omitted from the above disc-s ion entirely.
above analysis provides
of
al
some basis for believing
But the
that the bias in the estimate
( the. itiluence of the father 8s characteristics
on the daughter’s
return
to education) will be small even if the bias in the estimate of r is large.
A Ftied Effect Approach
In practice,
(4) .
it is not efficient to work with the di Eferenced equation
Our panel data set provides more “than one obsination
individuals
in the sample.
Fmthermore,
for some househOl&
‘these circms tances a more efficient es timat ion approach
eliminate
Ch.
~is
more than one
and for others only one”””child.ti in the sample.
sibling is in the saple,
and include a separate
Ori (3) for most
In
is to work with (3)
intercept for each family- (a fixed effect) to
permits us to use the time variation
in experience
for
each individual
to identify experience
characteristics
on the experience slope even in the case of individuals who do
not have a sibling in the sample.
the &ta
on households
a separate comtmt
(9)
effects ‘and the effects of parental
It is also a convenient
with more than two children.
for each household
Wdht - Wh - (Zdht - ~)B1
way to use all of
Applying
is equivalent to esttiating
+
+ [r
+ alxfi + a2Xmhl (EDdh - EDh) + Vh(EDdh
+ [r
+ blXfi + b2Xmh] (EXPd
+ed-;
d+
OLS to (3) with
‘dht - \
151
ht
- EDh)
- sxPh) + &( ExP&t
- ExPh)
where Zh are the averages of the characteristics
of all the daughters
household h, EDh is rhe average education lev@l and. mph
market experience
coqonent,
and ~
the average
-”EDh) drop out
(ED&
the interaction,
of (9)“,but tk
experience
terns, including
When we pool
to study young men and brothers,
data for young men and young women we permit all of the coefficients
involving
and education, except the interactions between Xfi and X&
We also allow for separate intercepts
with gender.
In pooled case we are implicitly asswing
characters
the tens
remain.
We use the same” fraework
experience
labOr
of the daughters in household h, id is the average daughter
.
component Of their wages.
is the average idiospcratic
When tits for only one” daughter from household h is available,
involving
in
tics have the sme
the effects enter differently
tO varY
for males and females.
that mobse~ed
comon
f=ily
ef feet on both young men and young women.
If
for males and females, then they will not be
eliminated when we add a separate
intercept for females
to
(9).
6 consequently,
we are not sure how much emphasis to place on the fixed effecm
results based
on the pooled sample.
3.
me
me
kta
sample
Chapter
man’s
1.
Data
are based on the Young Men and Young Women cohorts of the NLS.
selection
me
or woman’s
criteria
dependent
reported
and most
variable
wage
of the variables
in all
for a given
regressions
year.
Note
are
is the
that
discussed
log
an
in
of the young
individual
6, In ~hapter 2 we find that tiere do exist gender differences in the
~ ihli~g,t “age factors on the wage
effe-cts of parental wage factors and Of ,t
rates of young” men and yotiri”g.
women.
152
may
contribute more than one obsenation
parentil education
suney
or her saple. 7
to his
The measures
(DADED and MOMED ) are based on young men and young women, s
reports, since the use of information provided by the parents
themselves would limit the smple
to only those young men and young women
whose parents are found in the mature women and older men’s cohorts.
&ta
of
Missing
on IQ scores and on mother’s and’ father’s education poses a problem
the study.
men
(31 percent of the sample for young men and
IQ is missing,
30.5 percent for young women) we ade
A1l of our equations
if data
on the
that
IQ score
all variables
involve
IQ also
are missing
and
include
is zero
that depend on IQ as zero.
a day
othewise.
variable
for young men, and 10. 9_for women) are handled
in tbe same way.
data on DADED are likely to be related to whether
present
that
Missing
(25.1 percent for young men, and 26.9 for women) and MOMED
DADED
for
is one
data
for
(13 .3 percent
.Since miss ing
the individual, s father was
in the household while he or she was growing up, we feel that it is
inappropriate
to simply eliminate cases with missing dati from. the analysis. 8
In addition to IQ, DADED, and MOMED, we constructed
educational
atta”iment variable
called ED.INDEX.
EDINDEX is a measure of the
deviation of a”child’s “expected” highest grade completed
group’s mean expected highest grade completed.
from his or her
Ie was_ constructed
predicted value from a regression Of the child’s. highest
set of parental and school quality variables
a predicted
as the
grade completed
that predict education,
perhaps influence the child’s potential wages.
Specifically,
on a
and
the parental
7
In fact. the averaze nmber
of obse~ations
uer vounc man is 4.66:
See ~his ckpter’s
Appendix fo; s~a~
statistics and
3.66 for young women.
more detailed descriptions of the young men<s and women’s data sets.
8
As we note below, most of our results are robust to restricting the
SamPle. tO individuals with nomissing
&ta on IQ, DADED., MOMED, and EDINDH.
153
variables
are
race,
encouragement
educational
parents’
to continue
whether
the natural
present
in the household
education
father
was
past
present
goal
for child
high
school,
for whether
and
at age
and
14,
indicators
for
the natural mother
child was 14 years old, for whether
when
parental
was
the mother
worked when child was 14 years old (young men were not asked this question) ,
and for whether
there were two parents
(including step - parents)
me
household when the child was 14 years old.
normalized
in the
school. quality measures
are a
school quality index, student -teacher ratio, counsel lor -student
ratio, mean teacher’s sala~,
expenditure/pupil,
average hours .of
hOmewOrk/pup il, and indicators for private school, child, s subjects most liked
and disliked,
and curriculm
vocational
We re-code
).
individual
me
type (college preparato~,
indicators
in the equation.
equation used to compute EDINDEX was estbated
young men and for young women.
Or
that are missing for a particular
variables
to zero and include missing variable
regression
comercial,
separately
For each group ED INDEX is normalized
for
to have
a
mean of zero.
me
hypothesis
underlying
influence how much education
our use of EDINDW
is that variables
that
individuals obtain also influence the. response of
wages to a year of schooling.
mere
are too many variables
that could be
related to the quality of schooling to investigate each separately.
Ve work with cubic specifications
for.experience.
We parametrize
1inear education
term is the marginal
for educa~ion and IQ and a quadratic
the models so that the coefficient
rate of return to education
individual with 12 years of education.
men
interactions
and any or all of IQ, DADED, MOMED, and EDIND~
are parametrized
so that the coefficient
154
between
are included,
on education
on the
for an
education
the equations
is the marginal
rate of
return
to etiation
100, the saple
for an inditiidual w~th
12 years
of education,
an IQ of
mean of EDINDSX (which is zero) , and a mother and father *O
each have 12 years of education. 9
Although
not show
effects include dmy
in the tables, all equations without
variables
child, s race, residence
for the year for which the wage is reported,
in an SMSA, res ihnce
in the South, two parents “in the
household when the child was age 14, in addition to nmber
experience
(more precisely, potential
years since last euolled
education
family fixed
experience,
of siblings,
calculated
as nmber
in school) 10 and experience squared,
interacted with his or her experience. 11
of
and the child, s
In the models with familY
g. AI I specifiaions
that” include interactions between education and
any or all of IQ”,”DADED, MOMED, or ED INDEX also include interactions between
education and the corresponding miss ing dwy
variable(s) .
~~
to (age -- 14.)
10. In ~reating” the experience variable we set experience
for those who never enrolled in school or had zero years of schooling.
We set
Unfortunately, we did
it to age minus the school leaving age for other cases.
not.notice until after the paper was essentially completed that the school
leaving age is incons istent with the &ta an age and/Or the years of schooling
in a few cases. As a result, (age - experience) is less than 14 years for 1..0
percent of ~he young men, s ob.senat ions and 2.6 percent o.f the young women’s
obsemations . This explains the fact that the difference G T“able Al betieen
the maximm
of age and the maximm
of experience is 11 years for men and 8
years for women.
When we eltitiate these obsenations
and re-esttiate the young men’s
wage equation with fixed effects (Table 5, colmn
1) , the effect of the first
year of experience for an individual with 12 years of education falls from
2.65 to 2.59. The effect of education” for an individual with 1.2years falls
=
from 3“.27 to 3.08. For young women the effect of the first year Of
experience falls from .634 to .544 and effect of education falls-”from ? .48 to
7.19 (Table 7, colwn 1) .” We re-estimated all of the models reported in the
tables and found that in all cases the interactions of &ther’s
education,
mother’s education, EDINDEX, and IQ with education and with experience are not
dr~atic~y
different from those reported in the tables.
11.
In the models ~
f.tiily fixed effects, face =d nwber
of siblings
are excluded from the contro 1 variables since they should be constant within
the faily,
and therefore their effects are captured by the family fixed
effeet. For consistency when we work with the young pen !s data set, we
replace each. young man’s reports of DADED and MOMED in the interaction terms
with the average of the reports. of”all the brothers in the family.
Similar
155
fixed effects , those individuals who do not tive siblings in the smple
to identify the effects of experience,
III.
In section
mother’s
III. 1 we discuss
of education.
Resul-
the effects
education on the education slope.
determinants
various
the year, and the residence variables.
Esttiation
cons ider the effects on the education
help
of father’s education
and
In sections 111.2 and 111.3 we
slope of IQ and of the index of
In section III. k we discuss the effects of the
interact ion terns on the experience
slopes of young men and young
women.
III. 1 me
Effects of Father fs and Motherts
Ed”cation on the Education
Table 1 presents a sec of wage equations
effect included for each family.
been multiplied
for young men with a fixed
All coefficients
and stan&rd
by 100 to make the tables easier to read. 12
basis for” assessing whether
farnily background
Slope
errors have
TO provide a
has an important
influence on
education slopes , we first discuss the size of the effect of education
wages for a typical individual with 12 years of education.
On
A base line
recodings are done for the young. women Is data set, and the pooled &ta set
when family fixed effects were added. Nearly 96 % of young men who have
brothers supply a DADED report that differs by no more than one half year from
the average report over all the brothers in the family.
For mother’s
education, the figure is 94%. For young women, the corresponding percentages
are 93, and 93. Fi=lly,
for the pooled sample of young men and young women,
report
within
one
half
year
of.. the
average
83% of individuals have a DADED
over
all
his
or her
siblings,
85 .% for”MOMED.
reports,
12
Standard errors have not been corrected to account for the serial
correlation across the for a given individwl.
The standard errors for the
equations without fixed effects also ignore correlation among the errors terns
of members of the same family.
156
equation with all background. interaction effects “excluded is repOrted
colmn
1 of the tible.
the marginal
on education
is to be interpreted
effect of education when education equals 12 years.
estimated coefficient
speciftiation
cubic
The coefficient
in
me
is 3.41 with a standard error of ~ 801. 13
for IQ, the education coefficient
as
men
we add a
falls to “3.27 (c.olwn
2)
In’coltin 4 we add the interaction between
son’s
education.
small
in magtitude
education,
beween
The coefficient
coefficient
relative to the main effect of” an additional
different
from zero.
education and the son’s education
in colmn
5 has a
(standard error) of .286 (.199>”’: This re”sult is consistent
evidence is weak.
&ta
On all of DADED, MOMED, IQ, and EDINDN,
allow for both parents’
interaction
on the motherfs education
interaction
Both coefficients
In an~fforr
on
men
we
6 of Table 1) the coefficierit
increases and the father’ s becomes
are- imprecisely
estimted,
though.
to get more pretiise estimates at the cost of possible bias
from the omitted “fkily variables
education
(in colmn
with
the coefficient
rises to .730 and is significant.
interactions
with a
but the
However, when we restrict the sample to individwls
the mother, s education
negative..
year of
“The interaction
modest effect Of ’”mother’s education on the return to education,
nomissing
and the
is .039, which has the expected sign, but is
and is not si~ificantly
mother’s
father’ s..
education
that are correlated with the young man’ s
and his parents’ education, we report -timates
without
faily
fixed
13 .“ ~en
fixea effects are included, we estimate the difference in 10g
wages (tties 10.0.)
“’associatedwith inc.r~s ing education from 10 years to 12,
14, and 16 to be 6.33, 14.10, and 23.25’”(respectively).
For young women the
corresponding estimates are 15.47, 30.18, and 41.87.
men
fixed effects irk
excluded, “the estfiates are 8.82, 18.06, in”d 27.14 Fo”? “young men and 11.74, ‘-”
24.87, and 38.90” for young women.
157
,.,
effec=
in Table 2.
The esttiate of the rate of return when only parents’
for (coltin 1) is 4.39, which is in the low range of
education
is controlled
esti=tes
from other studies that do not contain detailed controls
variables.
education
(Recall that when f~ily
coefficient
The interaction be~een
3.41 ) .
was
fixed effects were controlled
and the son’s education has a coefficient
The interaction between mother’s
coefficient
father’s education
of .074 with a stantird error
of .080 with a stan&rd
error of .036.
men
fall somewhat.
imply that
parental education has a small positive effect on the relationship
and wages
additional
for young
.034.
we include both
Taken together, -the results with and without family effects
education
fur, the
education and son’s education has’ a
interactions , both coefficients
parents’
for f-ily
men.
bemeen
A point estimate of .1 implies that 4
years of parent education raises the chfid’s rate of return by .4,
which is modest relative to an overall return to education Of..4 or 5 percent.
Results
for Women
Tables
to Tables
3 and
1 =d
4 report
wage
2 for young men.
family, while Table 4 does not.
equations
Table
for young
3 includes
women
that
are
comparable
effects for each
fixed
The base line specifications
in colmn
each of the two tables indicates that the rate of return to education
higher for young women than for=_yOung men, and actually
7.78 when we control for family effects
are generally
increases
coefficients
lower when we include fmily
effects) .
Altonj i (1988) reports a similar pattern using matched
sisters from the Panel Study of Income DP~ics.
The education
fixed effects seem to indicate that unobsemed
158
is
from 6.26 to
(recall that for young men, the
education
and without
1 of
fmily
fixed
data on
slopes with
variables
that
raise
wages
for young women” are associated
while the opposite is tme
desenes
careful
peculiar pattern
positive
for young men.
investigation
in future
with
fewer
The as~et~
work.
Associated
years. of education,
is surprising
with
and
it is a
in which the ef feet of IQ (level) actually changes from
to negative when we control for fmily
through 7 of.Tables 3 and 4) .
Men
effecrs” (compare colwns
fixed effects are excluded,
2
the
coeff~clen”t an IQ is similar for young men and young women, around O. 30.14
14. ~Or yOung men, the main effecC of IQ has a positive
-
and
statistically significant effect’ on the log wage in nearly +11 specificatio~.
When family fixed effects ar~ “’~ricluded
,“the coefficients on the linear,
quadratic, and cubic terms fiply that the wages of individual in che 75th
percentile of IQ scores- are 5 .“6% higher than individuals in the 25th
percentile.
When family fixed ~feccs
are excluded, the corresponding
differential is 5.5 %. For young women we obtain a differential of .5.2 % when
ftiily fixed effects ‘“areexcluded.
However, when we add family fixed effects
we obtain a negative IQ differential equal to 5.9 %.
We do no have a full explanation for this puzzlirig result, although it
aPPears tO be due in part tO an anOmaly in the s=ple o f young women who have
sisters.
We re-estimated our equations without fixed effec~ On the sample of
young women who have sisters in the sample, since this is the sample that
identities the effect of IQ”once family ftied effects are added to the
equation.
In the basic specification in colun 2, the effect of IQ is .15
(not significantly different from zero) , which is well below value of .30 for
the full samp”le. (In models with family fixed effects, the coefficient
estimates are about equal in the -O samples .)
We also reestimated the model on a sample that excludes those young women
for whom IQ is missing and obtained a negative but statistically insignificant
coefficien~ when fixed effects are included. Years of schooling, wage rates,
and EDIND~
are systematically lower for young women for whom IQ is missing.
However, the same pattern holds for young men as well.
(Since the__lQ score
was provided by the respondent s high school, it is not surprising that it is
more likely to be missing for those who have fewer years of education. )
The
correlation between IQ and highest grade completed, the time average Of the
log wage rate, the time average of log hours worked per week (for those who
are .50, .27, .05, and .L8 respectively
worked positive hours) , and EDIND~
in
the case of young men and .b4, .31, -..04, and .46 in the case of young women.
Thus , the gender dif~erences
in the raw correlations
are small.
,.
We also investigated
the possibility
that part of the return to IQ for
young women comes through an effect on the earnings of potential
spouses.
When we substituted
family income for the log” wage rate as the dependent
variable
in the model, we obtained positive
IQ coeffi:iegts
“(multiplied” by
100) between
.01 and .31 when fixed effects” are included and a positive
coefficient
of about
.400 without
fixed effects.
159
The coefficient
(in colmn
on the interaction of daughter’s and father’s education
4 of Table 3) implies that a year of father’s education past high
school increases the daughter’s education slope by .116 with a stan&rd
of .187.
The corresponding
coefficient
interaction
tem
with mother’s education has a
These results
of .547 with a standard error of .212 .15
substantially
more powerful effect. of parental
mother’s ) on the education slope for yomg
for young men the corresponding
error
education
(especially
women than for youg
estimates were
.039 and .286. )
imply a
the
(Recall,
men.
In colmn
6 ““we”
again see the mother’s education effect becomes stronger and the father’s
weaker “hen they enter together. in the .,.eqwtionwith family fixed effects.
However, both of the parents, education
and negative
colwns
4 and
when
family= fixed
5 in Table
4.
effects
fithough
effec”ts included, the imprecision
from
drawing
strong
conclusions
education slopes for women.
obtains quantitatively
age
interaction
dropped
from
terms become
s=ll
the analys.i~;
see
we prefer the estimates with fixed
of the estimates
shout
the effects
in that case preclude us
of pirerital education
o_n the
It is also worth noting that Altonj i (1988)
important effec-
of pireriti”leducation on the
education slopes for young men, but does not find much Of an effect for yomg
women.
In view of these mixed results , we cOnclu&
much evidence
that parental education has a strong effect on the return to
young
women.
Effects of IQ on the E&cation
Slope
education for either young men
111.2 me
A nwber
is that we do not have
or
of the specifications
include
15
an
interaction
of
the child’s
This esttiate rises to .820 when the smple
is restricted
individuals with nomissing
data on IQ; DAD ED, MOMED, and EDINDN.
160
to
---
.
education
tem
with
his
or her
IQ.
is typically negative
.010 for young men.
between
We are
su~rised
to find
(and insi~ificant)
When we add fixed effects, the coefficient
benefit more
(in percentage
III. 3 The Effec-
when fixed eff”ects =e
of EDINDM
on the E&utiOn
IQ’s
Slope
and EDINDEX and
cubic speci ficat ions of
and IQ, a ftiily fixed ef feet, and the other standard
controls
for
.The gain effect “of EDINDEX is large and positiye..(2.81) with a
standard enor
of .734, but
insignif.icant.16
coefficient
included.
terns ) from additional years of schooling.
7 of Table 1 adds the interaction of education
young men.
is
is -.085 and is highly
the level of EDINDEX to the equation containing
education
coefficient
Thus, .ye have. little evidence that those with higher
significant.
Colmn
interaction
the
with a coefficient .of -.015 to
For young women, the IQ interaction
.000 and -.057 (and insignificant)
chat
the interaction
When the family fixed ef fee-
on the educatiOn-EDINDEX
significantly
tem
different
from zero.
interaction
is negative
are excluded
and
(Table 2) , the
is .083, but it is not
(Without fiexd
effects,
the coefficient
on EDINDEX falls from 2.81 to 1.23 with a stan&rd” error of .239) .17
For young women, we obtain EDIND8X interaction coefficient
estimates
(standard errors) of .086 ( 103) when family effecti are excluded
(Table 4,
16
Omitting the education- EDINDEX interaction did not have much effect
on the coefficients on EDINDN
reported in colun
7 of the various tables.
17
Measurement error in reported education might be partially
responsible for the positive coefficient on EDINDEX.
The decline in the
EDINDM
coefficient. would be expected if the proportion of the variance in
EDIND~
tbt is across families exceeds the corresponding proportion for
education.
It.might be possible to improve upon our estimates by using the
education reports provided by relatives (e.g. , parents , brothers or sisters)
as im trments
for the education reports provided by the respondents.
161
colmn
7) .
7) and .310 (.310) when family effec~
are included
The latter indicates that a 3 year difference
associated with a difference
of almost- 1 percentage
However,
rate of return to education.
(Table 3, colwn
in predicted
education
is
point in the young woman’s
the coefficient
has a standard error of
.310, arid is therefore not precisely estimated.
In colmn
8 of Tables 1 and 2 we s-imultaneously include the titeractions
of education with IQ, D~ED,
young men.
men
IQ, and EDINDM
excluded
fixed effects are included
interactions
in the wage equations
(Table 1) the parents’
are all ins i~if icant:= men
(Table 2 ) the EDIND~
a standard
interaction
coefficient
for
education,
ftied effects are
increases
to .240 with
error of .087, while the parents I terns and the IQ interaction
remain ins ignif icant.
obsened
MOMED, and EDI~~
The same general pattern of coefficient
changes
is
in the young women’s models in Tables 3 and h.
In s~ary,
we do not have stron”g evidence that variables
related to the nmber
on the education
of years Of education
that are
completed have much of an effect
slope of wages, although the education- ED IND.U..interactio~
are typically positive and significant
in the models that exclude fixed
effects and i“iclude ill of the interaction
found to have a strong positive
terns.
The
of” EDIND~
level
effect on wages, particularly
was
in the models
including fixed effects .
It is also interesting
effect of education.
excluded the education
Table 1) .
coefficient
falh
from
(Table 2, col_*
3.78 to 3.27 when EDIND~
2 and 7) .
in the models with family fixed effects:
Evidently,
reduces the =in
For example, for young men” when fixed effects are
added as a control variable
more dramatic
to note that adding EDIND~
variables
that are positively
1.62
is
The drop is even
from 3.27. to 2.16 (see
related
to educational
.
attaiment,
wages.
and are captured by EDINDW,
direct effect on
This ..isconsistent with evidence that the education
upward by omitted variables
alternative
explanation
once EDINDH
is that measurement
is controlled
The Effects of IQ -d
for.
e=or
faily
.However, an
in education, which would
has more of an influence on the education
slope
(See footnote 17 for a further discussion. )
Paren&l
Education on ~erience
Table 5 reports estimates of the interactions between
education,
slope is biased
that are correlated with education. 18
tend to lower its coefficient,
111.4
have a positive
IQ, father’s eticat ion, and mother’s education
fixed effe”cts”included in the equations. 19
Slopes
experience
and
for young men with
In colun
1, the effecc. of
18
Screeting models emphasize that education has value in the market
because it reveals information about worker quality that is difficult for
employers to obsene
directly.
Consequently, to the extent that the variables
that are included in EDIND~ are hard for employers to obsene
directly and
are indicators of productivity, the decline in the education coefficient is
consistent with signaling
models that imply that t~e return to education is
in part a return to worker characteristics that are correlated with -- but are
not changed by- - sec On&ry and higher education.
are e~ily
However, to the extent that the variables comprising. EDIND~
obsemable,
coefficient
screening
models would seem to -predict no decrease
when EDIND=
is added as a regressor.
on” education
in the
It -y be
possible to base a test of the screening model on the relative effects on
wages of the component of education that can be predicted from information
that employers obse~e
and of the component of “education- that is
unpredicbble.
.Clearly, the problem in implementing such a test is
uncertainty about what information employers actully
can obseme.
lg,. In all equations that we repOrt, we include an interaction between
education and experience.
(The size of the coefficients and associated
standard errors of the other interaction terms are only slightly sensitive to
the omission or addition of the education-experience interaction term. ) The
coefficient on this variable is positive, large, and significant for young
men- - around .120 with a standard deviation of .030 -- indicating that 4 more
years of education raises the linear term. in the experience slope by .36 to
.51, which is large given that the linear tem in the experience slope for an
individul
with 12 years of education is 3.27 when family fixed effects are
included.
However, we do not find these effects for young women:
though the
estimated coefficients are all negative, none are significant ~see Tables 7
and 8).
163
one
additional
experience,
year
of experience
12 years
for an
of education,
individual
a father
education,
and an IQ of 100 is 2.61.
experience
in colmn
with
and mother
no previous
each
with
The interaction between
2 has a coefficient
Father’s education has a small negative and
statistically
interaction with experience and is not very sensitive
experience
and IQ (see coluns
also small, negative,
effects” (se”ecoluns
%en
.019. ‘.
However,
one controls
and between
The mother *s interaction
for family fixed effects
rises
to
interaction
the coefficient
mo ther’s education
insignificant
to whether
and mother, s education
3, 5 and 6)
slope by .34.20
term is
4, 5 and 6) .
coefficient
father’s education
IQ and
and ins ignif i~ant in the models with family fixed
we do nor control
interaction
.029
(Table 6) the IQ
with a standard error of .005, and the
term rises to .028 with a standard error of
on the interaction between experience
and
remains negative and insignificant.
Tables 7 and 8 report a parallel Set of results for young women.
coefficients
statistically
Similarly,
on the IQ-experience
The
and
ins ignif icant whether or not family fixed effects are included.
insignificant
In swag,
on
interaction are small in ma~itude
the f“ather’s and mo ther’s education
statistically
of
(standard error) of .017 (.005) ,
implying that a 20 point increase in IQ raises the experience
for the interactions between ~perience
12. years
the experience
interaction
terns are
and small in magnitude.
we do not find a strong positive effect of parental
slopes.
in some cases , negative.
—
In most cases, the point estimates
Evidently,
the relationship
betieen
education
are small, and
parental
_. ...
20
In our Smple of young men the 25th and 75th percentiles of the IQ
measure are 92 and 112 while. the mean is 101.6 and the standard deviation is
15. g. For young women, the corresponding figures are 93, 112, 102.2 and 15.2.
164
education
market
h~an
and
is not
capital
experience
general
strong
and
highly
educated
educated
consistent
or her
and
enough,
slope.
men,
it is
slopes
relationship
while
between
possibility
individual
return
that
sli~tly
the we~
IQ has
the child’s
the
a small
is steeper
between
lack
education
general
the
fla,tter for more
with
0~
only
slope
labor
between
to .influence
relationship
is co=istent
link
the
ef feet
for more
highly
parental
of a strong,
and
ability
and his
slope.
be kept
that
of the
the experience
IV. Discwsion
It should
to entering
training,
to note
(if anything)
Consequently,
prior
the strength
in on- the -job
Also,
experience
experience
given
obtiined
It is interesting
“omen. 21
education
capital
investment
slope.
on the experience
hman
in mind
family
characters
to education.
that
and
these
characteristics,
COncl~iOn
results
school
do not
mle
characteristics,
out
the
and
tics
that affect expected schooling
alter the ~
men
the effect of education on wages is nonlinear,
the
21.
%en
we pool the vounc men’s and vounc women’s samDles
and add
..
fixed effec~s we fin; th;t the exper~enc~
s“lope is” more than 3 points
lower for fe“rnales than males.
It should be kept in mind that our experience
measure is a measure of potential
experience
sdsequent
to completion
of
school ing, which may partially
explain the smaller coefficient
for females,
We also find &at
coefficients
on the interaction
between education
and
experience
is typically
.085, iugge$ti+g
a substantial
effect of education
on
the experience
slope.
This qualitative
result and the empirical
magnitude
is
consistent
with most of the evidence
of which we are aware.
The IQ. experience interaction
has a coefficient
of .013 (with a s tanda?d
error of .004) when we use the pooled sample and include family effects,
indicating
that. a 10 PO int increase in IQ will raise the experience
slope by
me
estimate
i* .021 with a stan&rd
error of .004 when
between
.05 and .08.
family effects are” ekci~did. ”
Finally,
the coefficients
on the experience -father’. s education,
and
experience -motherl s education
interactions , which are a main focus of our
are small in magnitude
and never statistically
significant
when we
analysis,
pool the young yen’s and young women, s &ta.
fmily
165
..
ex ante return must take account of the fact that educational
uncertain.
Variables
outcomes
are
that affect years of schooling may raise the ~
return even if they have nO effe”ct or a small negative effect on the response
.
TO see this, note that if
of wages to a given nmber
of years of schooling.
most of the return to college is associated with graduation,
then variables
that lower the value of a college degree but increase the prob~ility
grad-tion
conditional
of
on starting college may raise the ex ante rate .of
return to starting college. 22
A simple way to investigate
which do not include education
the interaction
these issues is to estimate wage equations
squared and cubed as _regressOrs.
terms should “get credit” for differences
return to education associated with different education
In this case,
in the average
Tables 9 and
levels.
10 report the results of wage regressions which included linear specifications
of education.
The results are basically
similar to the estimates
—
above. 23
In this paper we have explored the possibility
experience
and
an
discussed
that education
slopes of wage equations are influencedby
index
of family
background
personal characteristics
variables,
IQ, parental
school characters
that predict years of education
and
education,
tics , and
completed.
result is that the effect of the child?s IQ, father, s education,
Our main
motherts
—
22. See “ei~b”rOd (lg62) for the initial discussion
Of this distinction
and
Altonj i (1989) for an empirical analysis of ex ante rates of return using kta
from the High School Class of 1972. He finds that favorable fmily backgrowd
and individual characteristics raise the ex an te rate of return to starting
college even though his methodology asswes
that wage response to education is
the sae for all individuals.
23 ~ese
results are OnIY suggestive, since variables that have Only
small effects on average returns may have large effects on margi=l _retuns.
TO address these issues, it would be necessa~
to adopt the methodology of
AltOnji (1989) , which is beyond the scope of this paper.
166
.-
education, and index of faily
and personal characteristics
background,
between education
labor market ‘experience and wages.
background
school characteristics,
that predict years of schooling completed have
only weak influences on the relationship
between
seconda~
In a nmber
and wages , and
of ‘cases, the f-ily
interactions work in the wrong direction or are statistically
insignificant;
In view of the results, it seems unlikely
effect” of f-ily
hackgiound
to us that the
on the etication slope of wages i.s.Eespons ible for
more than a small pafi of the powerful affect of family background
on years of
school completed.
A substantial
in a nmber
estimates
research agenda remains.
.ofcases,
when f~ily
particularly
are imprecise.
First, we wish to emphasize
Furthermore,
that
fixed” effects ari included, our
the findings for young men are at
variance with results in Altonj i (1988) using a sample of young men from the
PSID.
We plan to replicate our analysis using tbe most recent data from the
PSID, which contains large saples
of siblings.
and Krueger suggest that educatioml
education
slopes.
evidence
inputs can have large effects On
and peer characters
in measured educational
that parental characters
relationship
recent results of Card
On the other hand, the literature of ,schoo ling achievement
suggests that fatily” background
important variables
me
tics are the most
achievement.
tics in particular
There k
strong
have “a strong
to the level of earnings and to the titiber of years. of schooling.
It remains to be the seen whether
and experience
they have a substantial
slopes in wage eqtitions ,
167
ef Feet on education
Tale
me
1
Effects of IQ and Parental Education
on Wage hvels tid Education Slopes
Depen&nt
V.ri*le:
bg
-al
Houly
Yomg
Uage
(1967 Mllzs)
Men
Eqmtions
wifi Faily Fixed Effectsl
(Coefficients and Standard Errors Have Been Multiplied
Explanatog
Variables
Educationz (based
on a cub i.c
specification)
(2)
“3..41
(.801)
3.27
(.834)
.117
(.028)
Education x
Experience
--
--
Education
x DmED2
. . .=.
..
Education
x EDIND~
--
--
.-.
..269(.115)
“.3.15
(.132)
--
--
--
.010
(.040)
-.180
(.187)
.-
-.216.
(.189)
.438
.286
(.199). (.229)
--
.406
(.232)
--
-.367
(.238)
-.355
( .265)
.226
(.116)
.255”
(.116)
..226
(.133)
--
2.81
(.734)
2.75
(:754)
--
--
.“277
.216 (..115) ( 116)
--
-.
1.84
(1.26)
.121
(.028)
--
---
2.16
(.889)
.117
(.028)
--
--
3.8b
(S.12)
(8)
.120
(.028)
.119.
(.028)
.039
(.164)
Education
x MOMED2
IQ3 (based on.
a cubic
specification)
.4.b3
(1.03)
“.118
(.028)
.000
(.040)
,(7)
(5).. . (6)
2.27
2.69
(.990). (1.08)
.117
.117
(.0.28) (.028)
--
Education
x IQ
(4).
(3)
(1)
by 100)
--
EDIND=4
NOTES :
1.
In addition
to the variables
reported,
all regressions
contain
the following
control v=riable$:
Y... d.~i.s,
.xPeri.nc E (de fin.d as the number of yeacs since Last enrolled
in school) . experience
squared,
and indi. at.zs for reside”..
i“ an SMSA a“d i“ the sou Lh, two parents
in the household
when
child w.* 14 year= old, a“d indicators
representing
missing
variable
report%
the
2.
Education
i. d-fined
.S hishest
grade oompleted
min”n 12. =. its coefficient
i= read .s the
additional
wage ao. zuing to a yeas of education
past high school.
DADED is :he average
report ovex all
brotbers
in the f~ily,
rather than the Individual
youns man, $ report,
of father$s
.ducasion
mtius 12;
moth* z,* education
(MOMED) ia constructed
similarly.
3.
IQ is defined
a% tha
4.
EDINDEX
is an index
of education.
5.
The
typical
individual,
of paxe”tal
.e&r.ssiom
had
s reported
imflue”ce
N- 19298,
R2-
IQ score
a“d
.74,
school
minus
quality
and ~SE-
168
100.
.263.
factor*
“hich
predict
number
of Y...s
Table 2
me
Depednt
Effecti of IQ and Parental Educacion
On Wage bvels and Education Slopes
Vari~le:
bg
=1
Hourly Wage
Men
Yomg
ExplamtO~
Variables
(1967 Mllars)
Eqmtions
witiout F-ily Fixed .Effectsl
(Coefficients and Standard Errors Have Been Multiplied
by 100)
(8)
(2)
(3)
(4)
(5)
Educationz (based
.-4.39
on a cubic
specification)
(.409)
3.78
(.415)
2.9.5
(.458)
3.94
(.450)
3.98
(.435)
3.21
4.02
(245.4) (.439)
Education x
Experience
.172
(.032)
.174
.173
.032)” (.032)
.173
(.032)
.174
‘“.1.68
.17”1
(.”032) ““(:032) (.032)
(1)
.173
: -- (.032)
(6)
(7) .
2.68
(.4g6)
Education
x IQ
.-..—...--
Education
x DADED
--
--
Education
x MOMSD
--
--
Education
x SDIND~
--
--
--
.274
(.038)
.334
(.044)
.275.
(.039)
.272 “’ :“;273
““:248
.319
(.039)
(.039.) (.03”9J (.044)
DADED2
-.256
(.108)
-.285
(.108)
-.267
(.108)
-.335
(.123)
-.3.1.2 -’:319
-.:346
(.109) ““C:”124) (.109)
-:372 “-”-” ““
(..124)
MOMED2
.607
(.118)
,553
(.118)
.537
(.118)
.535
(.118)
.49”2
(.124)
.“513
(.118)
.421
(.126)
-.
--
=1 .23
(.239)
.986
(.244)
IQ3 (based on
a cubic
specification)
.007
.013)
--
--
.074
(.-034)
..-
---
--
.050
(.037)
.-
.080
.064
(.036) ““(.040)
--
......
------
.----
.>477
(.126)
-----------
EDIND~4
-.015
(.013)
--’ ”--
--
.054
(.037)
---=
.064
( .040>
.(::;;)’- ~::;)
NOTES:
1..._
1“ additiont. the variablesreported,all tegres. ions contain the following control variable. :
year du-ieo
, experience
(d. fined a. the number of years since last .Drolled
in school) , experi.uce
for rata, residence
squared,
number of sibling. . and indicators
in an SMSA and i“ the South, two Parent.
i“ the household
when the child was 14 years old, and indicate, a rePrasentims
missins
variable
rePo, ts .
2.
Ed”cation
is defined
as highemt
grade compl.t@d
minus
additional
wage s.cruins
to a y.ar of education
past high
ed.. ati.n (MO~D)
father, = education
(DADED) a“d ❑ other,%
3 ... IQ is defined
4,
EDINDEX
as the
i. an index
individual,
.f parental
s reported
i“flue”ce
IQ storm
and
12.. so its coefficient
ix x.ad .= the
school.
A similar
interpretation
holds
.Oeffi. i.”ts.
minus
100.
school quality fact..% “hich predict n.~ex of yea=s
of education.
5.
The tYPiCal regression
had
N=
19298,
R2-
for
.29,
169
and
~SE=
.386.
Table 3
me
Effects of IQ and Parental Education
On Wage hveb
and Ed=tion
Slopes
kpenknt
VSi&le:
hg
-
Houly
YOug
Wage
(1967 Dollars)
Women
Equations wifi F=ily
Fixed Effectsl
(Coefficients and Standard Emor: Have Been Mul~iplied
Expla-tO~
Variables
Education2 (based
on a cubic
spectiication)
Education x
Experience
(1)
(2)
7.78
7.48
(.9.62)
(.g74)
-.o&9
(.043)
-.047
(.043)
Education
x IQ
--
Education
x DmED2
--
(5)
““”””(”3) (4)
6.58
7.00
9.44
8.57
(1.13)
(1.13)
(1.15)
(1.22)
-.045
(.043)
-.
(6)
(7)
(8).
6.53
(1-37)
6..78
(1-01)
-.047
(.043)
--
--
-—.
.116
(.187)
--
-7099
(.200)
--
-.057
(.050)
(.046)
..--—-,.=
—--
--
--
Education
x MOMED2
--
Education
x EDINDH
IQ3 (based on
a cubic
specification)
-. .=..,--
--
-. 2a4
_-
,--
(.132)
-.276
(.146)
--
--
--
.5.47
(.212)
-.29a
(.133)
-.331
“(.133)
—-----
---
= ---- lM
(.205)
-..
.711
(.234)
.310
(.310)
.480
(.344)
-.316
(.132)
- .2a8
(.146)
2771
(.a20)
2.97
(.a32)
.633
(.226)
--
-,336
(.133)
.—.
EDIND~G
._
-.044
(.043)
-.047
-.046
(.043). (.043)
-.046
(.043)
..- . 000
by 100)
.
NOTES:
1. In addition to the variables r.port.d, all r.sre%sioms contain the followins control variables:
y.. r d“mi..
, experience
(defined
. . the number of years .ince last .nrolled
in .chool).
.Xp. rienc.
for
r.. ide”.. i“ an SMSA and 1“ &he South.
sw.r.d,
and indicator
two Parents
in the household
when
old, and indicators
representing
missim~
variable
reports.
child was 14 years
2.
Education
ia defined
as highest
gzade completed
minus 12, so its coefficient
is re=d as the
additional
wase accruing
to a yea, of education
past high school.
DADED is the average zep.rt over
.ather than the individual
young Woman- s report,
.f father, s .duca Li.n minus
sister, in the family,
(MOMED) i.
constructed
simila.ly.
mother, . education
3.
4.
of
5.
IQ is defined
ED INDEX
as the
is an index
Individual,.
of parental
reported
i“flu.n..
IQ *core
and
school
minus
resxession
had
N- 14320,
R2-
.72,
quality
and WSE=
170
all
12;
100.
education.
The typical
the
.283.
factors
“hich
predict
numbex
of
yea=.
Table 4
me
Wpen&nt
Effects of IQ ad Parentil Education
on Wage kvels and Education Slopes
V~i*le:
Lg
Real HOUly
Wage
(1967 ~llars)
Women
YOwg
Equations titiout F-ily
Fixed Effectsl
(Coefficients-”fid Standard Errors Have Been Multiplied
ExplanatO~
Variables
(l).
(3)
(2)
(4)
(5)
(6)
by 100)
(7)
(8)
Educationz (based
on a cubic
specification)
6.26
(.502)
5.78 ““”5.04
(.503)
(.551)
5.1.5
(.537)
-“5.04
5.37
(.541)
(.523)
5.14
(.520j
3.73
(.594)
Education x
Experience
-,002
(.043)
-.022
(.042)
-.015
(.042)
-.023
(.042)
-.021
(.042)
.-.021
(.042)
-.030
(.042)
-.025
(.042)
-.085
(.015)
--
-.
--
-.089..
(.016)
--
-.072
(.047)
--
Education
x IQ
--
.,. —--
Education
x DADED
--
.-
-.119
(.042)
Education
x MOMED
--
.-
--
--
--
-..
--
. .
.296
(.042)
....398
(.045)
..302”.: .300
(.042)
(.042)
.303
.042)
“.259
.368
(.’043) (.045)
..007 -“.139
(.122”) (.122)
-.137
(.122)
.049
(.134)
-“”.
100
(.123)
.028
.135)
-.212
(.122)
.“307
(.137)
.“290
(.136)
.258
(.137)
.344
(.139)
Education
x EDIND~
IQ3 (based on
a cubic
specification)
DADED2
.-
.“260
(.136)
.-
--
--
~~~-.081
(.044)
-.027
(.049) ._ -----
. . —-.
---
-.107
(.047)
..__, .-,.
ED.INDm4
.086
(.103)
-.023
.(..050)
.379
(.111)
-.io4
(.135)
.316 “ .187 ““- .259
(.”141] (.137)
(.141)
----- 1.64
(.283)
1.54
(.283)
NOTES :
1.
In addition
to the variables
reported,
all regressions
contain
the fo.llowi”g control
variables:
dumie$
, experience
(defined
as the nmber
of yea.. since last enrolled
in school) , exp. rie”c.
squared,
“umber of siblings , and indicators
for race, residence
i“ an SmA
a“d in the south,
two parents
in the household
when the child was 14 years old, and indicators
representing
❑issing variable
.ep. rtx.
Y.ar
2.
Education
is defined
as hishemt
grad- completed
minus
sdditi.nal
was. ac. ruimg to a year of education
past hish
father, , education
(DADED) and ❑others s education
[MO~D)
3.
lQ is defined
. . the
4,
EDINDEX
is an index
of education.
5.
The
typical
individual,
of parental
regie$,ion
s reported
influence
“hod N-
14320,
IQ s.ora
a“d school
R2-
.24,
171
12, . . its coefficient
is resd as the
school.
A similar
interpretation
holds
.Oafficiants.
minus
100.
quality
a“d
for
factors
WSE-
.376.
which
predict
““mber
.f yea. %
Table 5
me
Dep-&nt
Effects of IQ =d
on Wage hvels
=d
Variable:
bg
Parental Education
~erience
Slopes
Real Houly
Yomg
(1967 ~ll=s)
Men
Eqmtions
with F=ily
Ftxed EffectsT
(Coefficients and Standard Errors Have
Been Multiplied by 100)
ExplanatO~
Variables
(1)
Educati0n2 (based
On a cubic
specificat%oti)
(Experience)z
(4)
(3)
(2)
(5)
(6)
:.
Experiences
“
Education
x Experience
‘3.31
3.49
(.8L1)
(.840)
3.21
(.841)
2.65
(.299)
2.66
“2.6k
(.308)
(.305)
2.61
2.61.
.2.62
(.304)
(....307)(.314)
-.071”
(.010)
-.0.70
(:010)
-.072
(.010)
-.072
-.071
(.010). (.910)
.117
(.028)
.091
(.031)
.123
(.030)
.Ufi
(.030)
--
.017
(.005)
--
--
--
.-
-.034
--
Experience
~~
( .020)
--
Experience
x MOMED2
-------
.269,_
.136
(.115)
(.121)
IQ4 (based on a
cub ic specification)
3.27
3.44
(..843) (.847)
3.27
(.834)
Experience
x IQ
x DADED2
Uage
.272
(.115)
.127
.105
(.031). (.033)
--
==-. 028
(.022)
. .-.023
(.021)
.272
(.li5)
-.071
(.Q1O)
-.013(.024)
.018
(.005)
-.033
(.023)
.-.019
(.024)
.273
.135
(.115 ).. (.121”)
NOTES :
1. -In addition
t. the variables
reported,
.11 regxessio”s
contain the following
c.”trol
variables:
year dumies,
and indicators
for residence
i“ an SMSA and in tha South,
two
par~mts
in th. hOusehOLd
when the =hild w.. 1~ years O1d. and indicators
x*Pr*samting
mis. i”g va. iable reports.
2.
Education
is de fimad as highest
grade comF1eted
mi””s 12, so it= c.efficient
ix read as
the addit50na1
wage accr”im6
to a year of education
pa%t high school.
DADED is the averase
report over all brothers
in the family, rathar than the individual
young man, . report,
of
father, . education
minus 12;
mother> s education
[MOMED) is c.n=tructed
similarlY.
3.— Experience
is measured
enrolled
in school.
b,
IQ is definad
5.
The
typical
as the
regression
as pobontial
imdivid”al,
had
expesien.
s reported
N = 19298,
R2 =
e, that
IQ score
is , “umber. of years
minus
.74, and WSE
172
100.
=
.263.
since
last
—
Table 6
The Effects of IQ md Parental Education
On Wage tivels =d ~erience
SlopDepen&nt
V~iable:
bg
E-
Howly
Yowg
Wage (1967 Dollars)
Men
Equations without Family Fixed Effects 1
(Coefficients and Stantird Errors Have
Been Multiplied by 100)
ExplanatO~
Variables
(1)
(2)
(3)
(4)
3.78
(.415)
4.27
(.428)
3.96
(.423)
3.7a
( .423)
3.a9
4.28
(.“4”26) (.435)
1.35.
(,260)
1.4a
(.26a)
1.49.
(.267)
1.35
(.264)
L.45
.1.51
( .“26a) (.274)
.002
(.012)
.002
(.012)
.003
(.012)
.003”” ‘-”.00L
“(.0”12) (.012)
.172
(.032)
.110
(.035)
.152
(.033)
.173
(.033)
Experience
x IQ
--
.029
(.005)
--
--
.-
Experience
x DADED
-----
-
.02a
(.019)
--
.047
(.021)
.035
.021)
Experience
x MOMED
...-
“-..026
(.020)
-.049..
(=:022j
.056
.022)
Educationz (based
on a cubic
. .. :.
specification)
Experience
(Experience)z
.
Education
x E~erience
---
IQ4 (based on a
cub ic specif icat ion)
.274
.(.03a)
.039
(.osa)
.277
(.03a)
DADED2
-.285 .“”-.2?4
(.lea)
(.lea)
--.4aa
(.193)
MOMED2
.553
(.118)
.552.
.sic
(.118)
(.118)
.“”160
.11’0
(.034)
(.036)
.02a
(.005)
.“273
.275
(.03a) (.03aj
.043
(.058)
-.2a9.
i“.67k “”- ..57a
‘(.109)- (.211)
(.212)
.a20
(.217)
1.02
(.235)
1.
In addition
to the variables
reported,
all regressions
.Ontai”
the f.llowins
variables : Year du-ies,
““mber of siblings,
and indicators
for rata, roaide”ce
14
yea=.
old,
and in the South, two parents
in the household
when the chi Ld “as
indicators
Xepre=e”ting
missing
variable
reports .
2.
Ed.cation
the additional
interpretation
l.oa
(.235).
IQ is defined
The
control
in an SMSA
.nd
is defined
a= highest
grade completed
minus 12, . . it= coefficient
i. read
wage accruing
to a y.ar of education
past high school.
A =imilar
holds for fatther, s education
(DADED) and mother, s education
(MOMEDl
3.
Experience
is measured
enrolled
i“ school.
5.
“.003”
(.012)
—
NOTES :
4.
(6)
(5)
tyPical
as the
as potential.
individual,
regression
had
experience,
s reported
N
-
19298,
that
IQ score
R2
-
is, number
minus
.29,
1.73
and
of years
100.
WSE
-
.384.
since
last
.S
Tale
me
Depen&nt
7
Effecti of IQ and Parental Education
on Wage bvels
md ~erience
Slopes
V~i~le:
kg
-1
Houly
YOwg
Wage (1967 ~llars)
Women
—
Equations tifi Fmily Fixed Effectsl
(Coefficients and Standard Errors Have
Been Multiplied by 100)
Explanatoq
Variables
(1)
(2)
(3)
(4)
(s)
(6)
7.48
(.974)
7.31
(.986)
7.63
(.9S0)
7.60
(.9S3)
7.66
(.9S5)
7.47
(.993)
Experience
.634
(.410)
.509
(.415)
.726
(.415)
.672
(.416)
.715. . .5.9.3
(.418)
(.422)
(Experience)z
.006
(.014)
.006
(.014)
.007
(.014)
.006
( :014)
.007”
(.014)
.008
(.014)
Education
x Experience
-.047
(.043)
-.023
(.048)
-.06.5
(.045)
-.061
(.047)
-.067
(.047)
-.040
(.050)
--
--
-- -. .?001
(.007)
..031
(.026)
--
.023
.027
(.029). (.030)
Educationz (based
on a ctiic
specification)
Experience
x IQ
-- .--.
0.03
(..007)
Experience
X DADED2
--
-----
Experience
x MOMED2
.. ..
—--
--
.037
(.029)
.021
(.033)
.019.
(.033)
-.315
(.143)
-.282”
(.132)
-..280
(.132)
-.280
(.132)
-.293
(..144)
IQ& (based on a
cub ic specification)
NOTES:
-.284
(.132)
..-.
1. In additi.nt. the variablesrep.rted,all res.assio”*contain
v.riables:
year d-i..
, a“d
P.=.nk.
in the household
when
missing
variable
report=
indicators
the child
for
“as
the followins
control
re. id. ”.a in an SMSA and in the South,
LWO
rePresenti”s
14 year. old, and indicators
2.
Education
is defined
a$ highest
srade completed
minus 12, s. its c.efficimme
is read as
the additional
wage .ccruin&
to a yea. of education
past high school.
DADED is the averase
tePort
over all brothers
in the family, rathez than the individual
young ❑a”, s repert,
of
father’s
education
minus
12;
mother, . ed”cati.n
(MOMED) is Qonstructad
similarly.
3.
Experience
is maas”red
anrolled
in school.
4.
10 is defined
5.
The
typical
as the
regression
as potential
individual,
had
experience,
s reported
N = 14320,
R2 -
thae
Iq minus
is, nmber
100.
.72, and MSE
174
-
,2s3,
ot years
since
Lass
‘-
T
Tale
8
The Effects of IQ ad
on Wage Wvels
-d
Depen&nt
V=i*le:
Paren-1
~erience
tig Wal
Houly
YOmK
Education
SIOpWage
(1967 Mllars)
Women
Equations without F-ily
Fixed Effectsl
(coefficients and Standard Errors Have.
Been Multiplied by 100)
ExplanatO~
Variables
(1)
Educati0n2 (based
on a cubic
spec=ication)
(2)
(4)
(3)
5..79
5.78
‘(.503) (.521)
5.85
(.515)
(5)
(6)
5.87
5.89
(.523.) (.525)
5.87
(.536)
Experience
-.640
(.349)
-.733
(.356)
-.615-.588
(..355) (.354)
1.57g
(.357)
-.684
(.362)
(Experience)2
.030
(.016)
.030
(.016)
.031
(.0”16)
.030
(.016)
.030
(.016)
.0”30
(.016)
Education
x Experience
-.022
(.042)
-.021
(.046)
-.029
(.044)
-.032
(.045)
-.034
“(.045)
--
--
Experience
xIQ
--
Experience
x DADED
--
Experience
-------
.012
--
.012
.(.006)
(.007)
--
.006
(.023)
.
:-
-.004
(.02”6)
-.009
(.027)
..-
.029
(.026)
.027
.024
(.”029) (.029j
.298
(.042)
MOMED
x
-.02”9
(.047)
IQ& (based on a
cub ic specification)
.296
(:042)
.175
(.074)
.297
(.042)
.297
(.042)
DADED2
-.139
(.122)
-.137
(.122)
-.182
(.246)
-.130
-.074
(.1.22) (.273)
-.027
(.276) ----
MOMED2
.260
(.136)
.?64
(.137)
.255
(.137)
-.036 “-- .03.1
(.283)
(.312)
.008
(.313)
..178
(.075)
NOTES :
1.
In addition
to the variables
raported,
all .egresmions
COntai”
the following
v.riabl.s:
Y*.. dumies.
n.tier of siblings,
and indicators
far ra.. , tesidence
and in the South,
two parents
in the household
when the child was 14 years old,
indicators
zepzese”timg
missing
variable
reports
2.
Education
the additional
interpretation
is defined
as hishest
gzade completed
wage accruing
to a year of education
holds f., father,.
education
(DADED)
3.
Experience
is meas”c.d
enzolled
in school.
4.
IQ i= de finid
5.
The
tYPical
aB the
as potential
regression
had
N -
minus 12, so it. coefficient
is read
Past high school.
A similar
and mother,,
education
(MO~D)
experience,
imdi.vid. al, s zeport”ed
1&320,
COnt. ol
in an SMSA
and
that
IQ
score
R2
=
is, nmber
minus
.24.
175
and
of years
100.
WSE
=
,377,
=ince
last
as
Table 9
The Effects of IQ -d Parental Education
On Uage kvels and Education Slopes
kpen&nt
Vari*le:
tig -al
Houly
YOmg
Wage (1967 bllas)
Men
Equations with =near
Specification of EducatiOnl
(Coefficients and Standard Errors Have Been Multiplied by 100)
Explanatory
Variables
Without Faily
(1)
(2)
.With Family Fixed Effects
Fixed Effects
(4).
:-(3) ““
(6)
(5)
(8)
(7)
Edmati0n2
(based
on a linear
specification)
3.86
(.295)
3.99
(.287)
3.97
(.284)
3.22
( .321)
3.26
.670)
3.57
(.645)
2.95
3.66
(..605) (.566)
Education x
Experience
.115
(.027)
.163
.158..
(.027)
(.027)
.166
(.029)
.108
.027)
.111
(.027)
.112
(.027)
Education
x IQ
.001
( .012.)
~.-
--
.025
.038)
---
--
.085
(.030)
..
-.
.219
--
Education
x DADED3
. .
. .
.090
(.033)
Education
x EDI~H
--
--
--
--.
--
--
.082
(.158)
.092
.249
.281
.277
(.03.8) (.038). (.039)
DADED2
-.270
(.lo8j
-.355.
(.121)
-.315
(.109)
-.347
(.109.)
.546”
(.118)
.538
(.118)
.485
(.123)
.514
(.118)
1.23
(.227)
--——— ---ED INDm5
of control
variables
fixed
with a family
are
Education,
DADED, and MOMED
read .- the additional
waga
3.
The
---:
.408
(.157)
(.050)
.318
(.043)
rather
--
. .
IQq (based
on a cubic
specification)
2.
--
( .134)
Education
x MOMEDS
NOTES :
1.
Fox a list
specifications
.110
(.028)
used in each
effect, Table
regression,
4 for those
.276
(.129)
.313
.306
(.113). (.114)
-..
---
..
--
--
--
--
--
---.——
---2.35
(.706)
‘----
see footnote
without.
1 in Table
3 fox
are defined
.. highest
grad. completed
minu% 12, =. their
ace. uing to a year of education
past high school.
.Patio”%
with family fixed effects “.. the averaged
bxothexs,
than the individual
young man, . corra=ponding,
rmports.
4.
5.
re”ports OC DADED
the
coefficient=
amd
IQ is defined
as the individual- s reported
Iq score minus 100.
ED INDEX is a“ index of parental
influence
and school quality
factons which predict
number
of education.
6.
.384.
The ty~ical
zeEression
without
fixed effects had N= 192$8. R2- .29, and tiSE=
Th. typical
reE2ession
with fixed effect.
had N- 19298, R2- .74, and WSE=
.263.
176
.260
(.115)
of MO%D,
of
y.a.s
---
TAle
The Effects of IQ md
On Wage hvels ad
kpen&nt
VariAle:
hg
10
Parental Education
Edu~tiOn
Slopes
Real Howly
YOmg
Women
Equations with Ltie~
Specification of EducatiOnl
(Coefficients and Standard Errors Have Been Multiplied by 100)
Explanatory
Variables
Wi&Out
F~ily
(1)
(2)
E&cationz
(based
on a linear
specification)
7.07
(.371)
Education x
Experience
-.152
-.136
(.035”) “(.034)
Education
x IQ
-.071
(.015)
6.49
(.355)
--
Fixed Effects
(4)”
(3)’
.-
-.064
(.039)
Education
x MOMED3
--
--
Education
x EDIND~
--
--
(5)
(6)
6.52”
(.353)
5.46
(.398)
5.71
(.780)
5.J8
(.779)
-.123”.
(.035)
-.062
(.038)
-.051
(.043)
....-.050
,“(. OLT)
--
(8)
(7)
6.85
C.753)
5.20
(.724)
_.. .
-.
...057...
-(:”175)
--
--
--
--
.337
( . 19.2)
--
---
-... .=
““-.O1O
-...--
-.005
(.040)
---
-.
.229
(.067)
IQb (based
on a cubic
specification)
.382
(.045)
DADED2
-.-121
“(. 122 )
.021
:133”)
.121
.123”j
MOMED2
.282
(.137)
.261
.137)
.139)
.316
(.032)
---
--
.312
(-042)
.306
of control variable,
with a family fixed
.-
are
used in each
effect, Table
.259
(.043)
-.214
(.122)
.i84
---
-- ..=
.227
(.198)
.=.300
-.270
-.342
-.337
( .145) ““-(
:’13’2) (“.133) ( 132)
--
--
--
--
-,-
.
--
--
:-----
(.137)
1.53
(:276)
EDINDH5
2.
Education,
D~ED,
and MOMED
are read as the additional
wage
Witi Family Fixed Effects
(.044)
Education
x DADED3
NOTES:
For a list
1.
.p*cificatioms
Wage (1967 ~llars)
regression,
4 for those
‘-
see footnote
without.
-“ -:.. ---.”
‘-
1 in Table
3 for
da f.ined a. highest
grade Completed
minus 12, s. their
to
a y.ar
of education
past high school.
“;:;:9)..
the
coefficients
accruing
The equation%
with family fixed effects u%= Ehe averaged
sixter, , rePorts
tather than the individual
yo”ns wom.n,~
c.rresp.”dimg
reports
3.
of DADED
12 and
IQ is deti”ed
as the individual, s reported
IQ score ❑inus 100.
influence
and school quality
factors which predict
number
EDINDEX
is an index of parental
education.
of
6.
The tYPical
resrefi, ion without
fixed effects had N- I&320, E2- .24, and RmE.376.
Thm typical
.egrnssion
“ith fixed ef~..ts
had N- 14320. R2=
,7*, ~n* ~SE=
.z~~.
4.
5.
177
of MOMED,
of
year.
Appendk
Stizistics for tie Yoag
Sv~
~-
Table Al
&n’s
md
Yomg
Women’s Dati Se=
Yom
Mm
Wmm
----—------—————————-
V~ihlel
Std
s-l.
Size
Me=
----------——-—
m“
Dev
—-—----—
w
—--------
s-la
SW
M*m
size
———-——
D-
0. 65s
L4320
LU20
———-—
&
M=
0.432
-0.914
3.193
2.403
0.000
9948
14.839
46.000
158.000
19298
0.457
-0.883
5.188
12.958
1S=8
2.786
0.000
18.000
IQ Score
101.127
13304
15.4L2
50.000
158.000
FatherS s
9.709
14451
3.732
0.000
18.000
9.844
10462
3.855
0.000
18.000
10.143
16736
3.258
0.000
18.000
10.202
12758
3.176
0.000
18.000
29.057
19298
3.704
24.000
39.000
28.262
14320
3.307
Z&.000
37.000
9.085
18877
4.988
0.000
28.000
9.441
14303
4.500
2.000
29.000
0.219
19298
2.124
-6.164
4.957
0.182
1&320
1.766
-5.986
4.U3
3.317
19164
2.5s5
0.000
18.000
3.593
14264
2.628
0.000
16.000
0.832
19256
0,374
0.000
1.000
0.821
14319
0.383
0.000
1.000
0.421
0.000
1.000
0.292
14320
0.455
0.000
1.000
0.627
14314
0.49s
0.000
1.000
14314
0.449
0.000
1.000
1.109
Log Sourly
w-e
Utez
Ei&hest Grade
=.638
18.000
C~leted
102.3=
Edu.ation
Mother, s
Educati~
Ue
Y.=,
of
~*zienoe3
m1ma4
Nher
of
Sfiuw=
Tw
P-ents b
Eo.seh.ti at a.
14?
Black?
0.230
Resid-ce
0.399
19297
0.490
0.000
1.000
0.700
19297
0.458
0.000
1.000
in So.ti?
Ee.idmce
i“ m?
------------------------------Notes :
1.
Mis Sk
values
Y-
mitted
-.
frm
is ❑ea.ued
2.
w%.
3.
~eri.n..
i.
in
ED1~~
me
ptential
sqle
size
is
19298 f.=
y-
mm,
14320 for
1967 &ll=s.
m.asued
-
,,ptential
individuals never euolled
4.
all ..I..htio”s.
men.
is ~
index of f~~
qerience,,.
.. tie.
in school ox.raporti”g zero ye~s
batigromd
v~i~les,
sch~l
P=.diet Years of education.
178
of yews
since last -rolled
of edacatim.
ch~acteristi.z,
e~eri~ce
md
i“ .*-1.
e~als
persmal
=.
For fiose
ti”us 14 ye a.,.
.b~a.tezistics
tias
Appendix Table ~
of Distribution of ObsenatiO~
S-V
P.,.ent
.~tiihut.d
.bsenation*.
me
~ere
U.
=ith.r
=erasa
md
five or six obse~atio~;
ntier .f .bsewatiom
per y-
3764 brothen sets in *.
average “her
of obse~ations
Youm
Wmm4 s Data Set:
~er.
are 14320 absematims
from 1968 to 1982.
❑=
is 4.66, mile
men, s tit. set, incltii”s
per brofier set (includiw
prwid~
waen
tirowh
tie =de
brothers , md
siwlet-s
twelve
is fi-
3423, or 90.9 Prc=t,
by 3907 individuals in tie Y.mg
Fifty- three per. ent .f the y-
Men’s md
21 p,. ent centributed s-n
, 31, or 0.8 Percent, =ets of tie.
8.2 Percent, sets of two brothem
me
yomg
ti Yowg
obse~ations.
singletqs,
1 set of far
) is 5.13, xe
Medim
309, or
brofiers
and mode -.
Wmen>s data set. s.-i”.
the Period
..ntrib”ted one, two or ttie. obs emations;
31
ad 16 percent cmtitiutti
six tb..a
eleven
or five obsematims;
of .bse-ati..%
per YOUW w
is 3.66, ~ila tia mde is two
Percent contributed either f.u
obsenations,
~. aerage me,
.bs e~ationz.
~ere
... 3571 sister sees b
or 7.5 pr.mt,
~e
sets of tw
avezage nhez
obsematims,
the young waa,s
of .bseHati-s
a“d the ❑de
data set, includiw
3269; or 91.5 percent, siwletim,
sisters, 32. or 0.9 percent, sets of tire. sisters, and 1 sat of fo~
per =ist~
is tm
set (ticLdtig
.bse_tions.
me
-i-
siwlemns
269,
iister~.
) is 4.01, the median is four
COntrib.tim
of my
,i,ce, set is 19
obs .Hatims.
Pooled Data Se&:
~ere
are 33618 obsanations
Prec.diw
~ere
YOW
ti the p~led
men’s and yo~g
are 8039 brother-sistet
ad
of obse~ati.ns
as given i“ the
sets in the pooled data set, i“cl.ding 5367, or 81.7 percent sinsleco”s.
are 990 sets, .r 15,1 perce”c, of tw
P.rcent, .f four .ibUnss.
data set with tie distmb”tim
wwen, s amries.
@ibli.gs, 178 sets, . . 2.7 parcmt,
5 seti of six siblings.
179
nere
of tire. siblings, 32 sets, .r 0.5
--
References
ti Chapter 3
Altonj i, Joseph G. “The Effects of Fmily Background and School
Characteristics on Education and Labor Market Outcomes. ” Mimeo,
Northwestern University, December 1988.
Altonj i, Soieph G. ‘Controlling for Personal Characteristics, School and
when Estimating the
Comunity
Characters tics, and High School Curriculm
1989.
Return to Education. “ Mimeo, NOrthwes tern University, Novmber
Bound, John, Zvi Griliches, and BrOnT
H. Hall. “wages, SchOOling and IQ of”
Internat i~nal
DO the F=ily Factors Differ?”
Brothers and Sisters:
Economic Review27
(FebW~
1986) : JJ - 105.
A
Cain, Glen G. , !.~e E~OnOmi~ AmIyS is Of hbor Market Discrimination:
Sumey, ” in O. Ashenfelter and R. Layard (e~. ) , Handbook of Labor
Economics.
hstertim:
Elsevier Science Publishers, 1986.
“Does School Quality Matter?
Retu?ns to.
Card, David, and Alan Krueger.
E&cation
and the Characteristics of Public Schools in the United States”.“
Mimeo, Princeton University, April 1990.
“More on Brothers. ” in KinQrnetriC5:
Chamberlain, Gaq and Zvi Griliches.
DetemiD~nts
of Socioeconomic Success within and between Families, edited
bsterdam:
North-Holland, 1.9.7J.
by Paul Taubman.
Griliches , Zvi, “Sibling Models and Data in Economics: Begimings
of a
Suney, “ Journal of Political Economv 87 (October 19J9) : S3J - S64.
Griliches, Zvi, “Est&ating
Problems. ” Econometrics
the Returns to Schooling: Some Econometric
45 (January 19J7) : 1-.22...
Hauser, Robert M. 1’
Socioeconomic .Background and Differential Returns to
~
e Imuacts of
Education. “ In Does
CO Ie e Matter?
Hi~her Education, L.C. Solmon and P .J. Taubman, e&.
New York: Academic
Press, 19J3 .
McElroy, Marjorie, “The Joint Determination
Work: The Case of Young Men” JOU
nal
~
283-316.
of Household Membership and Market
conomics 3 (July 1985) :
Morgan, Jmes
and Ismail Sirageldin.
“A Note on the Q=lity
Dime= ion in
Education. ” Journal of Political Economy 76 (September 1968) : 1069- 10JJ.
Olneck, Michael R.
“The Effed’ts of Education. “
Gecs Ahead ? Vew York:
Basic _BOOks ,.1979.
In C. Jencks, —.
et al . m
Siebert, U. Stanley, “Developments in the Economics of Huan
LabOur Economics,
London and New York: Longman, 1985.
180
Capi-1.
“ In
WeisbrOd,
~y0
Burton A.
‘r
Education and Investment in Huan Capital. “
Supplement, 70 (October 1962) : .106- 123.
Welsh, Finis.
“Black-White Differences’ in Returns to Schooling. “
m
ev -ew 63 (December 1973) : 893- 907.
~
Journal of
~erican
Willis, Robert J. , and Shewin Rosen. “Education and Self - Selection. “ Journal
~
87 (October 1979): S7- s36.
181