NORC at the University of Chicago
Using Military Deployments and Job Assignments to Estimate the Effect of Parental
Absences and Household Relocations on Children’s Academic Achievement
Author(s): David S. Lyle
Source: Journal of Labor Economics, Vol. 24, No. 2 (April 2006), pp. 319-350
Published by: The University of Chicago Press on behalf of the Society of Labor
Economists and the NORC at the University of Chicago
Stable URL: http://www.jstor.org/stable/10.1086/499975
Accessed: 21-02-2018 12:49 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
NORC at the University of Chicago, Society of Labor Economists, The University of
Chicago Press are collaborating with JSTOR to digitize, preserve and extend access to Journal
of Labor Economics
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
319
[ Journal of Labor Economics, 2006, vol. 24, no. 2]
᭧ 2006 by The University of Chicago. All rights reserved.
0734-306X/2006/2402-0005$10.00
Using Military Deployments and Job
Assignments to Estimate the Effect
of Parental Absences and
Household Relocations on
Children’s Academic Achievement
David S. Lyle, U.S. Military Academy
Military deployments and job assignments provide an opportunity
to estimate the impact of parental absences and household relocations
on children’s academic achievement. Combining U.S.
Army personnel data with children’s standardized test scores from
Texas, I find that parental absences adversely affect children’s test
scores by a tenth of a standard deviation. Likewise, household
relocations have modest negative effects on children’s test scores.
Both parental absences and household relocations have the
greatest detrimental effect on test scores of children with single
parents, children with mothers in the army, children with lowerability
parents, and younger children.
I. Introduction
More than 950,000 U.S. troops deployed to Afghanistan and Iraq from
2002 through 2004. By the end of 2005, one-third of these soldiers had
I thank Daron Acemoglu, David Autor, Rozlyn Engel, Christian Hansen, Mark
Pettit, Jonathan Smidt, Casey Wardynski, Michael Yankovich, and seminar participants
at the Massachusetts Institute of Technology, the U.S. Military Academy,
and the Western Economic Association for their valuable comments and suggestions.
The views expressed herein are my own and do not purport to reflect the
position of the U.S. Military Academy, the Department of the Army, or the
Department of Defense. Contact David S. Lyle at david.lyle@usma.edu.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
320 Lyle
served two overseas tours within a span of 3 years. Lengthy and recurrent
deployments coupled with other features of military service, such as frequent
household relocations, are apt to place considerable stress on the
more than 1.4 million military families.1
Any associated feelings of anxiety,
loneliness, and uncertainty seem especially likely to affect children’s
achievement in school. As a mother recently remarked about her son, “it
affected his grades last year when he knew his father was in Afghanistan—
he worries more about daddy dying than just going away and coming
back” (Nearin and Segal 2002).
Yet, increasing episodes of parental absences and frequent household
relocations are not specific to the military. Recent trends in American family
structure suggest that children spend less time with their parents now than
they did 30 years ago. For example, according to the Current Population
Survey, in 1970, 12% of all children lived with a single parent; by 1996,
that number had increased to 28% (U.S. Census Bureau 1996). The number
of families with both parents in the work force has also increased sharply.
In 1970, only 29% of children under the age of 6 and 39% of children
under the age of 18 had both parents in the labor force; by 2000, this
number had grown to more than 61% and 68%, respectively (Haveman
and Wolfe 1993, 156; Fields and Casper 2001). Work-related travel isanother
emergent source of parental absences. In 1970, there were roughly 93 million
business trips in the United States, or approximately 1.4 annual business
trips per household. By 1997, there were more than 213 million U.S. business
trips, or about 2.1 annual business trips per household (U.S. Bureau
of the Census 2001, table 1255). Finally, Americans move frequently: U.S.
Census data show that nearly 17% of American households relocate and
6% relocate to a different county each year (U.S. Bureau of the Census
2001, table 26).
It is not hard to imagine that parental absences and household relocations
may diminish a child’s sense of security, alter a child’s level of responsibility,
and/or disrupt a child’s social networks. These and other related factors
have unpredictable educational consequences for children. For example, a
child’s academic performance could decline if the child becomes preoccupied
with feelings of loneliness. Alternatively, a child’s performance could
improve if a newfound sense of responsibility accompanies the reduced
adult supervision. Similar competing arguments exist for household relocations.
A 1994 General Accounting Office study contends that moving
disrupts a child’s social group and hurts academic performance. Other researchers,
like Piaget, have argued that exposing children to different social
environments facilitates their understanding of the world and improves
1
Deployment estimates are from Davey (2004). Military manpower estimates
are based on 2002 data from the Directorate for Information Operations and
Reports, Office of the Secretary of Defense.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Parental Absence and Children’s Academic Achievement 321
classroom performance (Piaget 1977). Therefore, determining the net effect
of these family disruptions requires an empirical analysis.
However, a number of factors complicate a simple cross-family comparison.
For example, parents may choose to be absent more or less often
or to relocate a household in response to a child’s classroom performance.
Parental absences and household relocations are also likely to alter household
income. As discussed by Haveman and Wolfe (1995), the positive
relationship between income and a child’s academic performance confounds
the predominantly negative parental absence effects found in the
literature.2
Most recent studies focus on absences due to death, divorce,
separation, or out-of-wedlock birth but fail to account for the associated
lower household income. There is even less consensus on how household
relocations affect children’s education. Heinlein and Shinn (2000) review
this literature and report 26 studies that find no effects, 19 studies that
find negative effects, and eight studies that find positive effects. They, too,
attribute the mixed findings to inadequate accounting of important factors
such as income, since lower income households tend to relocate more
frequently.3
To reduce the bias that arises from potentially endogenous parental
decisions and other unobserved determinants of children’s academic
achievement, this study exploits exogenous variation in military assignments.
Military deployments generate parental absences, and the
military’s expressed intention to move soldiers every 2–4 years produces
frequent household relocations. Using the demands of military
service as a source of variation is appealing because soldiers have little
control over the frequency and duration of absences and relocations.
Media coverage of a recent deployment quoted one soldier as saying,
“somebody else other than us [is] deciding where we will live, what
missions we will be on, and how long we will be separated [from our
families]” (Nearin and Segal 2002). I also provide evidence below suggesting
that the military assigns absences and relocations without regard
to the academic achievement of military children. Moreover, any
effects of parental absences or household relocations are not likely to
reflect a change in income because the military is made up of a relatively
homogeneous population in terms of socioeconomic status and
provides supplemental compensation for deployments and relocations.
Data restrictions and faulty estimation strategies have limited previous
attempts to identify the effect of parental absences and household relo-
2
Haveman and Wolf (1995, 1839); Taubman (1989), McLanahan and Sandefur
(1994), and Heinlein and Shinn (2000) demonstrate a positive relationship between
income and children’s academic achievement.
3
In 2000, 21% of households with an annual income of less than $25,000 moved,
and only 15% of households with an annual income of more than $25,000 moved
(Schachter 2001).
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
322 Lyle
cations using military-induced variation. In a study of Gulf War deployments,
Pisano (1992) found that sixth-grade girls performed slightly worse
in reading when a parent was deployed. However, it is difficult to draw
firm conclusions from this study, because it has only 158 observations
and the control group is not comparable to the treatment group. A study
by Marchant and Medway (1987) found no effect of military-induced
household relocations on children’s academic outcomes, but their results
are based on only 40 observations. The current study improves upon
these military-specific studies by using more than 13,000 observations on
military children and a rich set of descriptive variables.
The empirical analysis begins with reduced-form estimates of the
effect of a parent’s deployment on his or her child’s math score. Twostage
least squares (2SLS) estimates indicate that the assignment of
parental absences is uncorrelated with other potential determinants of
children’s educational achievement. The design of this experiment is
most similar to that in Angrist and Johnson (2000). Angrist and Johnson
estimate the effect of Gulf War deployments on divorce rates,
spousal employment, and children’s disability rates. They find that
deployments of male soldiers do not affect the rate of marital dissolution
but do reduce spousal employment. In contrast, female deployments
increase the rate of marital dissolution but do not affect
spousal employment. Angrist and Johnson also find no effect of military
deployments on children’s disability rates. In part, the current
study builds on their research by estimating the effect of military
deployments on children’s academic achievement.
My estimates indicate that parental absences during the contemporaneous
school year have small adverse effects on children’s test scores; the
magnitude is about a tenth of a standard deviation. Cumulative 4-year
absences also have negative effects on children’s test scores, with officers’
children experiencing a decline of as much as a fifth of a standard deviation.
For household relocations, I estimate a modest negative effect on enlisted
soldiers’ children but no significant effect on officers’ children. Other
evidence suggests that parental absences and household relocations cause
additional detrimental effects to test scores of children with single parents,
children with mothers in the army, children with parents having lower
AFQT scores, and younger children.
The next section contains background information on the Texas Education
Agency (TEA 2000), a description of military assignment mechanisms,
and an overview of the data. Section III contains the empirical framework
and the identification assumptions. Sections IV and V contain the
main results for parental absences and household relocations. Section VI
concludes.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Parental Absence and Children’s Academic Achievement 323
II. Background and Data
A. Testing Children in Texas
Texas has conducted standardized testing since 1980, and the state has
one of the leading programs in the United States. In 1990, the TEA
implemented the Texas Assessment of Academic Skills (TAAS) program.
The TAAS exam certifies a student’s ability to be graduated from his or
her current grade. Schools administer the test in April and May of each
year to students in grades 3–8 and grade 10. The TAAS exam evaluates
performance in math, reading, science, writing, foreign languages, and
social studies at different points throughout a child’s progression within
the public and charter school systems. Only math scores are used in this
study because they are available for all years and all grade levels. Scores
in each subject receive a Texas Learning Index (TLI) value from 0 to 100.
The TEA normalizes TLI scores to allow comparisons across years for
the same student. For example, a student receiving a math score of 75 in
the fourth grade and a math score of 80 in the fifth grade has demonstrated
individual improvement across grades. Public and charter school students
achieving a TLI score of 70 have met the requirements for the respective
grade level and are on track to meet the required minimum score of 70
in the tenth grade. Since schools located on military installations belong
to local school districts, almost all military children living in Texas take
the TAAS exams.
B. Military Assignment Mechanisms
The army’s Human Resources Command assigns soldiers to divisionlevel
units every 2–4 years based on the “needs of the army.” This armyspecific
term captures the essence of all assignments: world events drive
army assignments. While soldiers may indicate a preference for a particular
division, the timing of the move and the assignment of a soldier to a
subordinate army unit are largely independent of a soldier’s preferences.
As illustrated in figure 1, once a soldier is assigned to a division, the
division assigns the soldier to one of several brigades, the brigade assigns
the soldier to one of several battalions, and the battalion assigns the soldier
to one of several companies.4
The “needs of the army” also determine the
missions that a soldier’s company receives, which ultimately affect how
much time a soldier spends away from his or her family. A deployment
can occur at the individual or unit level and is characterized by the soldier
leaving the base to perform some component of a larger military operation.
4
For assignment purposes, the army regards soldiers of the same rank and
occupation as equals. There are two primary division-level units in Texas, the
First Cavalry Division and the Fourth Infantry Division. This study does not
include soldiers from specialized units that are likely to deploy more often, such
as the special forces or rangers.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
324 Lyle
Fig. 1.—Assigning soldiers to units. There were 10 active duty divisions in the army at
the time of this study. Division composition varies, but this wire diagram describes the
general organization for the two primary divisions stationed in Texas, the First Cavalry
Division and the Fourth Infantry Division. For deployment purposes, soldiers in the headquarters
company are assigned to one of the three primary maneuver companies: Alpha,
Bravo, and Charlie.
During the time frame of this study, units below the brigade level from
Texas deployed to support peacekeeping operations in the Balkans, to
training and show-of-force maneuvers in the Middle East, and to humanitarian
aid missions in developing countries.
C. U.S. Army and Texas Education Agency Data
The military data are from active-duty army personnel records, dependent
records, and pay records for fiscal years 1992–98. The sample
includes children ages 6–19 who have parents serving in the active-duty
army and who were stationed in Texas in 1997 or 1998. Since the army
does not maintain reliable records on soldier absences, I use a pay supplement
(hostile fire pay) to identify soldier absences.5
The inherent incentives
associated with pay data suggest a relatively accurate assignment
of absences; soldiers will seek full compensation for their absences, and
the army will ensure that they do not receive too much. I calculate the
number of months that a soldier is deployed by dividing the sum of all
hostile fire pay received over a designated period of time by the monthly
5
Soldiers deployed for more than 7 days to designated regions around the
world receive hostile fire pay.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Parental Absence and Children’s Academic Achievement 325
hostile fire pay allotment. For the relocation variable, I define the number
of moves that a child experiences while his or her parent is in the military
as the number of army-induced moves. Other variables include child’s
gender, child’s race, military parent’s gender, military parent’s marital
status, military parent’s education level, and military parent’s AFQTscore.
Subsequently, the TEA matched test scores and grade level to the military
data for each child.6
The final data set contains observations for the
matched army and TEA data at the level of the individual child.
Table 1 displays descriptive statistics for the parental absence data. Standard
errors are in parentheses, and standard deviations are in curly braces.
Panel A compares children with an enlisted parent to children with an
officer parent. Officers’ children score 5–6 points higher on the math
section of the TAAS exam and have a smaller standard deviation in test
scores than enlisted soldiers’ children. Also, all officers have at least some
college, and officers do not have an AFQT score. Therefore, I present
the results for enlisted soldiers’ children and officers’ children separately.
Panel B of table 1 compares children with parents who deployed for
fewer than 3 months to children with parents who deployed for 3 months
or more during the “current” school year. I define the current school year
as September 1 of the previous year through March 31 of the current
year (1997 or 1998), since schools administer the TAAS exams in April
and May. Approximately 6% of children have a parent deployed for more
than 3 months during the current school year. I organize panel C identically
to panel B but stratify the data by cumulative months deployed
over a 4-year period.
Panels B and C show that a child’s math score declines as the duration
of the parental absence increases. The remaining descriptive variablesdiffer
only slightly with varying durations of absences. To explore these correlations
more closely, I present estimates from regressions of the parental
absence variable on the full set of descriptive variables in the appendix in
table A1. Although some of the correlations are statistically significant,
their magnitudes are small. Moreover, the set of all 14 descriptive variables
explains only 2%–3% of the total variation in parental absences. Nevertheless,
there are several military practices that may account for the
small correlations found in the data.
One possible explanation for why children in single-military-parent
households and children with military mothers experience fewer absences
than their counterparts is that single-parent soldiers and female soldiers
6
The army data met a minimum number of observations for each combination
of included variables to comply with the confidentiality requirements of the TEA;
data were matched by the child’s social security number. Fewer than 1,000 observations
were dropped for small cell sizes prior to the match, and there was a
40% match rate between the army and Texas data because 6–19-year-olds were
in the army data (14 grades) and the TAAS exam is only given in seven grades.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
326
Table 1
Summary Statistics for Current School Year and Cumulative 4-Year Parental Absences
B. Months Parent Deployed in
the Current School Year
C. Months Parent Deployed in
the Past 4 Years
A. Parent’s Rank Enlisted Parent Officer Parent Enlisted Parent Officer Parent
Enlisted
Parent
Officer
Parent 0–2 ≥ 3 0–2 ≥ 3 0 1–6 ≥ 7 0 1–6 ≥ 7
(1) (2) (1) (2) (3) (4) (1) (2) (3) (4) (5) (6)
TLI math score 77.14 82.47 77.21 75.91 82.56 80.91 77.41 77.17 76.05 82.73 82.38 80.59
(.11) (.18) (.11) (.50) (.18) (.83) (.13) (.32) (.43) (.20) (.51) (.71)
{12.01} {9.57} {11.97} {12.64} {9.51} {10.47} {11.87} {12.03} {12.41} {9.40} {9.81} {10.30}
Male child .51 .51 .51 .51 .51 .49 .52 .52 .48 .52 .48 .46
(.00) (.01) (.00) (.02) (.01) (.04) (.01) (.01) (.02) (.01) (.03) (.03)
{.50} {.50} {.50} {.50} {.50} {.50} {.50} {.50} {.50} {.50} {.50} {.50}
White child .44 .79 .44 .47 .79 .75 .43 .50 .40 .78 .81 .76
(.00) (.01) (.00) (.02) (.01) (.03) (.01) (.01) (.02) (.01) (.02) (.03)
{.50} {.41} {.50} {.50} {.41} {.44} {.50} {.50} {.49} {.41} {.39} {.43}
Parents married .90 .93 .90 .98 .93 .97 .89 .96 .96 .91 .99 .99
(.00) (.00) (.00) (.01) (.01) (.01) (.00) (.01) (.01) (.01) (.01) (.01)
{.30} {.26} {.30} {.15} {.26} {.16} {.31} {.19} {.21} {.28} {.12} {.12}
Father is in the army .87 .91 .87 .98 .90 1.00 .87 .98 .96 .88 .99 1.00
(.00) (.01) (.00) (.00) (.01) (.00) (.00) (.00) (.01) (.01) (.00) (.00)
{.33} {.29} {.34} {.12} {.30} {.00} {.33} {.15} {.19} {.32} {.07} {.00}
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
327
Army parent is a high school graduate .45 .44 .57 .41 .56 .51
(.00) (.00) (.02) (.01) (.01) (.02)
{.50} {.50} {.50} {.49} {.50} {.50}
Army parent has some college .55 .14 .56 .43 .13 .28 .58 .44 .49 .13 .16 .19
(.00) (.01) (.00) (.02) (.01) (.04) (.01) (.01) (.02) (.01) (.02) (.03)
{.50} {.34} {.50} {.50} {.33} {.45} {.49} {.50} {.50} {.34} {.36} {.40}
Army parent AFQT (top 40%) .27 .28 .25 .28 .26 .21
(.00) (.00) (.02) (.00) (.01) (.01)
{.45} {.45} {.43} {.45} {.44} {.41}
Elementary-age child (grades 3–6) .64 .60 .64 .68 .61 .59 .63 .66 .66 .62 .56 .58
(.00) (.01) (.00) (.02) (.01) (.04) (.01) (.01) (.02) (.01) (.03) (.03)
{.48} {.49} {.48} {.47} {.49} {.49} {.48} {.47} {.48} {.49} {.50} {.49}
Observations 11,548 2,900 10,904 644 2,742 158 8,841 1,373 818 2,294 366 213
Source.—The army data are from the Office of Economic Manpower Analysis (West Point, NY).
Note.—Standard errors are in parentheses, and standard deviations are in curly braces. Data in panel A correspond to the data used in panel B but not to the data used
in panel C. The data-matching process does not guarantee the same observations in both the current academic year and the 4-year data sets, although the statistics presented
in panel A using the data in panel C are comparable. There are only 11,311 observations for AFQT in panels A and B and 10,891 observations for AFQT in panel C. The
army data cover children ages 6–19 with a social security number and parents in the active duty army stationed in Texas in 1997 or 1998. Dual military families are dropped.
Absence variables are constructed using army pay data over the current academic school year for panels A and B and over the past 51 months for panel C. The current
academic school year is defined as September 1 of the previous year through March 31 of the current year. Deployments correspond to a soldier receiving hostile fire pay.
The number of months absent is determined by dividing the sum of hostile fire pay over the period by the monthly hostile fire pay allowance. Children’s math scores (range
0–100) are from the Texas Education Agency (TEA) testing years 1997 and 1998 in grades 3–8 and 10. Under advisement of the TEA, scores below 35 are dropped to account
for children who did not take the exam seriously or who quit in the middle of it (this accounts for .5% of the sample and does not affect the results). Army officers must
have at least some college, and they do not take the AFQT. This study separates AFQT quintiles into two groups: top 40% and bottom 60%.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
328 Lyle
may return from their mission prematurely to address family problems.7
Returning early from a deployment could shorten the deployment length
enough to classify the absence as fewer than 3 months. The practice of
deployed units leaving a small detachment of soldiers at the home station
to care for families and process administrative actions may explain why
soldiers with more education are deployed less often than soldiers with
less education. Unexpected deployments often cause soldiers to lose tuition
money for college courses that they take during off-duty hours.
These soldiers would be likely candidates for stay-back personnel. Although
uncommon, these cases could account for the small correlations
in table A1. I examine this issue further in the next section and conduct
several robustness checks for the main results.
III. Empirical Framework
The identification strategy for this study relies on the assumption that
military-induced absences and relocation assignments provide a plausible
source of exogenous variation. The nature of military assignment mechanisms
suggests that these episodes come closer to providing a causal
interpretation of the effect of parental absences and household relocations
on children’s academic achievement than other comparisons. To investigate
this hypothesis more formally, I employ a basic linear model, using
pooled data from 1997 and 1998, with the following structure:
E p a ϩ v ϩ dD ϩ bX ϩ ␧ . (1)it 1998 it it it
Here the left-hand-side variable Eit is the TLI test score on the math
section, is a constant, and is a year dummy for 1998. The coefficienta v1998
on the dichotomous variable of interest Dit represents the effect ofd
parental absences (or household relocations) on the TLI math score of
child i in time period t (1997 or 1998). The symbol Xit denotes other
covariates, including the child’s gender, race, and grade level, as well as
the marital status of the military parent, gender of the military parent,
civilian education level of the military parent, and AFQT score of the
military parent. Since some children appear in the data in both years, I
cluster all standard errors on the individual child using Huber-White
robust standard errors.
A causal interpretation of requires Dit to be free of measurement errord
and orthogonal to other potential determinants of children’s academic
achievement as represented by . Measurement error concerns are min-␧it
imal because both the army and the individual soldiers review and update
personnel records monthly. Likewise, the nature of military assignment
mechanisms suggests that the army does not deploy or relocate soldiers
7
Angrist and Johnson (2000) find that female deployments increase the rate of
marital dissolution.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Parental Absence and Children’s Academic Achievement 329
with their children’s academic achievement in mind. However, it is possible
that the army deploys or relocates some soldiers for reasons that are
correlated with their child’s academic achievement, as explained above.
One way to address this concern is to include characteristics in Xit that
are observable to the army (from its own data) and that could potentially
be correlated with children’s educational achievement. If all relevant controls
are included and the assignment procedures discussed in Section II.B
hold, then the estimated treatment effect is even more likely to have a
causal interpretation. However, the most compelling way to test the exogenous
assignment hypothesis is with a valid instrumental variable. For
parental absences, a valid instrument is one that would be correlated with
individual parental absences and uncorrelated with other potential determinants
of children’s educational achievement.
I construct an instrumental variable for parental absences using the
soldier’s unit of assignment at the battalion level (see fig. 1). Since deployments
can occur at the level of the individual or the unit, I create
the instrumental variable by assigning a one to all those in a battalion
that had more than one-third of its soldiers deploy and a zero otherwise.
One-third of a unit is a logical choice because most battalions consist of
three main companies.8
Units deploy as part of a task force, a heterogeneous
mix of companies from different battalions. The selection of
which companies will constitute a deploying task force seems likely to
be uncorrelated with individual-level children’s educational achievement.
There should also be a positive correlation between the deployment of a
battalion and the deployment of an individual soldier; a soldier is more
likely to deploy if his or her battalion must deploy one or more of its
companies than if his or her battalion does not receive such an order.
Correlations between the instrumental variable and the characteristic
variables are presented in panel D of table A1. As in panel A of this table,
some of the variables have statistically significant correlations, but they
are small and explain only 2%–4% of the total variation in the instrumental
variable. Most of the variation in the instrumental variable comes
from division-level and brigade-level decisions as to which exact battalions
receive a deployment order. It is unlikely that divisions and brigades select
a battalion to deploy based on unobserved characteristics that are related
to the academic achievement of the children whose parents are assigned
to that unit. Therefore, instrumental variable estimates that control for
the individual-level observable characteristics available to divisions and
brigades seem especially likely to provide a useful test of the assumption
that deployment assignments are exogenous.
8
The three primary companies are Alpha, Bravo, and Charlie. The headquarters
company is divided among the three primary companies to provide maintenance
and logistical support for deployments.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
330 Lyle
IV. Empirical Results for Parental Absences
In table 2, I present ordinary least squares (OLS) and two-stage least
squares (2SLS) estimates of the impact of parental absences of 3 months
or more during the current school year on children’s math scores. Panel
A contains estimates for the children of enlisted soldiers, and panel B
contains estimates for the children of officers. Columns 1 and 2 are the
most parsimonious specifications, containing only a constant, a dummy
variable for the year 1998, dummy variables for grades 4–8 and grade 10,
and the absence variable. However, as discussed above, a causal interpretation
is more likely after conditioning on observable characteristics that
are also possibly correlated with children’s academic achievement. Therefore,
specifications in columns 3 and 4 contain child characteristics, while
those in columns 5 and 6 contain both child and parental characteristics.
Including child and parental controls alters only slightly. Since estimatesd
of remain relatively stable with the inclusion of relevant control vari-d
ables, the effect of other less relevant variables that are possibly omitted
is also likely to be quite small.9
Before discussing estimates of further, it is informative to highlightd
briefly the coefficients on the child and parental variables contained in
Xit. These estimates suggest that boys score slightly lower than girls for
enlisted soldiers’ children but not for officers’ children; whites score
higher than nonwhites; children with a male parent in the army score
higher than children with a female parent in the army; children with less
educated parents score lower than children with more educated parents;
and children with high-ability parents, as measured by the AFQT, score
higher than children with low-ability parents.
The OLS estimates of indicate that children whose parents are absentd
for 3 months or more score approximately one point lower than children
whose parents are absent fewer than 3 months during the current academic
school year. The size of the effect is equivalent to a tenth of a standard
deviation decline in the test scores of enlisted soldiers’ children. I use the
2SLS estimates in columns 2, 4, and 6 to test the hypothesis that militaryinduced
parental absences are exogenous, or rather that the orthogonality
assumption for OLS holds for the specification in equation (1). The firststage
results for the 2SLS specifications are shown at the bottom of the
respective columns. They are statistically significant and have the anticipated
sign. Additionally, the instrumental variable explains approximately 60%
of the variation in individual absences. Comparisons of the OLS and 2SLS
point estimates reveal only small differences, and a Hausman specification
9
Altonji, Elder, and Taber (2000) show that this is a reasonable informal test
that indicates the degree to which small correlations between the variable of
interest and other unobservable characteristics may bias estimates.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Table 2
OLS and 2SLS Estimates for Current School Year Parental Absences
A. Enlisted Parent B. Officer Parent
OLS 2SLS OLS 2SLS OLS 2SLS OLS 2SLS OLS 2SLS OLS 2SLS
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)
Absent ≥ 3 months Ϫ.914 Ϫ.839 Ϫ1.027 Ϫ.993 Ϫ1.133 Ϫ1.196 Ϫ1.573 Ϫ2.192 Ϫ1.413 Ϫ1.739 Ϫ1.231 Ϫ1.497
(.511) (.650) (.506) (.644) .506) (.643) (.823) (1.082) (.839) (1.106) (.820) (1.081)
Male child Ϫ.795 Ϫ.795 Ϫ.766 Ϫ.766 Ϫ.209 Ϫ.210 Ϫ.228 Ϫ.229
(.284) (.284) (.286) (.286) (.466) (.466) (.459) (.459)
White child 3.536 3.536 2.687 2.687 3.667 3.662 3.437 3.434
(.281) (.281) (.303) (.303) (.596) (.596) (.597) (.597)
Parents married .254 .255 Ϫ.012 Ϫ.011
(.475) (.475) (.758) (.758)
Father is in the army 2.299 2.302 1.255 1.268
(.480) (.481) (.747) (.748)
Army parent is a high school graduate Ϫ.952 Ϫ.950
(.294) (.294)
Army parent has some college Ϫ2.005 Ϫ1.989
(.639) (.636)
Army parent AFQT (top 40%) 2.213 2.213
(.321) (.321)
First stage .748 .748 .746 .866 .867 .863
(.016) (.016) (.016) (.029) (.029) (.029)
Partial first-stage R2
.60 .59 .59 .61 .61 .59
Observations 11,548 11,548 11,548 11,548 11,311 11,311 2,900 2,900 2,900 2,900 2,900 2,900
Note.—Dependent variable is Texas Learning Index score for mathematics. Standard errors (in parentheses) account for clustering at the individual-child level because
some children appear in both years. All regressions contain a constant and dummies for the year 1998 and grades 4–8 and 10. Other controls added as indicated. Differences
in sample sizes within a panel are a result of some enlisted soldiers missing AFQT scores. The instrument used in the 2SLS estimates is constructed from the unit of assignment.
The unit is considered “deployed” when at least one-third of the unit is deployed during the academic school year. Hausman test statistics have the following p-values: panel
A—cols. 1 and 2 p .84, cols. 3 and 4 p .99, and cols. 5 and 6 p .86; panel B—cols. 1 and 2 p .32, cols. 3 and 4 p .60, and cols. 5 and 6 p .67. See note in table 1 for
additional sample description.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
332 Lyle
test fails to reject the null hypothesis that the orthogonality condition holds
for all OLS specifications in table 2.10
In table 3, I provide an additional robustness test to determine if children
with single military parents, children with mothers in the army, and
children with less educated parents affect the OLS estimates of . Ford
comparison purposes, column 1 in each panel contains the estimates from
column 5 in the corresponding panels of table 2. Samples in column 2 do
not include children with a single military parent or children with a mother
in the army. In column 3, I restrict the sample to only children with a
parent who has some college. In column 4, I restrict the sample to children
with married parents, a father in the military, and a military parent with
some college. Point estimates of from the restricted samples in columnsd
2, 3, and 4 are similar to point estimates from the full sample in column
1. Therefore, the small correlations of absences with marital status, gender
of the military parent, and education level of the military parent do not
appear to be driving the estimates of in the full sample.d
Based on the results from these robustness tests in table 3, combined
with the specification tests in table 2 and the description of the military’s
assignment mechanism in Section II.B, there is strong evidence to support
my claim that military deployments offer a plausible source of exogenous
variation.11
Accordingly, I proceed with the remainder of the analysis on
parental absences using OLS to estimate the model in equation (1).
Table 4 is similar to table 2 except that the variable of interest is cumulative
4-year parental absences. This approach tests for the presence
of aggregate or long-term effects of parental absences. I stratify the cumulative
parental absences by three categories: 0 months, 1–6 months,
and 7 months or more over the previous 4 years. Using the model in
equation (1), I include dummy variables in Dit for children with a parent
absent 1–6 months and for children with a parent absent 7 months or
more, making children with a parent not absent over the past 4 years the
omitted category.
As in table 2, estimates of are only slightly sensitive to the inclusiond
of additional control variables. The specification in column 3 contains the
full set of observable characteristics. Estimates in column 3 of panel A
10
See the table 2 note for the Hausman test p-values. All are insignificant at
the 95% confidence level.
11
As an additional falsification exercise, I test how parental absences in 1998
affect academic achievement in 1997. I find that parental absences in 1998 have
no measurable effect on academic achievement scores in 1997 (0.187 points with
a standard error of 1.30). By way of comparison, parental absences in 1998 adversely
affect 1998 achievement scores by 1.26 points (standard error of 1.25).
While small sample size mitigates the conclusiveness of this exercise, the point
estimates suggest that the effect of absences is contemporaneous and not present
across years.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Table 3
Robustness Test: OLS Estimates for Current School Year Parental Absences
A. Enlisted Parent B. Officer Parent
(1) (2) (3) (4) (1) (2) (3) (4)
Absent ≥ 3 months Ϫ1.133 Ϫ1.082 Ϫ1.317 Ϫ1.302 Ϫ1.231 Ϫ1.173 Ϫ1.158 Ϫ1.148
(.506) (.512) (.765) (.759) (.834) (.854) (.932) (.961)
Male child Ϫ.766 Ϫ.620 Ϫ.640 Ϫ.534 Ϫ.228 Ϫ.169 Ϫ.393 Ϫ.381
(.286) (.312) (.380) (.420) (.459) (.495) (.497) (.542)
White child 2.687 2.551 2.493 2.287 3.437 3.440 3.708 3.831
(.303) (.329) (.410) (.452) (.597) (.667) (.658) (.755)
Parents married .254 .550 Ϫ.012 .385
(.475) (.611) (.758) (.793)
Father is in the army 2.299 2.212 1.255 1.035
(.480) (.613) (.747) (.748)
Army parent is a high school graduate Ϫ.952 Ϫ.949
(.294) (.319)
Army parent has some college Ϫ2.005 Ϫ2.093
(.639) (.647)
Army parent AFQT (top 40%) 2.213 2.342 2.668 3.037
(.321) (.352) (.421) (.467)
R2
.05 .05 .06 .06 .05 .05 .05 .04
Observations 11,311 9,488 6,252 5,103 2,900 2,552 2,503 2,164
Sample restricted to children with married/fathers in the army No Yes No Yes No Yes No Yes
Sample restricted to children whose parents have some college
(enlisted) or have a college degree (officer) No No Yes Yes No No Yes Yes
Note.—Dependent variable is Texas Learning Index score for mathematics. Standard errors (in parentheses) account for clustering at the individual-child level because some
children appear in both years. All regressions contain a constant and dummies for the year 1998 and grades 4–8 and 10. Other controls added as indicated. See note in table
1 for additional sample description.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
334 Lyle
Table 4
OLS Estimates for Cumulative 4-Year Parental Absences
A. Enlisted Parent B. Officer Parent
(1) (2) (3) (1) (2) (3)
Absent 1–6
months Ϫ.401 Ϫ.632 Ϫ.633 Ϫ.289 Ϫ.388 Ϫ.482
(.381) (.376) (.377) (.585) (.578) (.580)
Absent ≥ 7
months Ϫ1.602 Ϫ1.517 Ϫ1.514 Ϫ2.149 Ϫ2.079 Ϫ2.127
(.515) (.506) (.504) (.819) (.815) (.819)
Male child Ϫ.796 Ϫ.748 Ϫ.159 Ϫ.181
(.289) (.289) (.468) (.462)
White child 3.568 2.798 3.574 3.351
(.286) (.307) (.592) (.596)
Parents married .566 .057
(.496) (.770)
Father is in the
army 1.625 1.249
(.513) (.750)
Army parent is a
high school
graduate Ϫ.778
(.299)
Army parent has
some college Ϫ1.740
(.641)
Army parent
AFQT (top
40%) 2.103
(.325)
Observations 11,032 11,032 10,891 2,873 2,873 2,873
R2
.02 .05 .05 .02 .05 .05
Note.—Dependent variable is Texas Learning Index score for mathematics. Standard errors (in parentheses)
account for clustering at the individual-child level because some children appear in both years.
All regressions contain a constant and dummies for the year 1998 and grades 4-8 and 10. Other controls
added as indicated. Differences in sample sizes within a panel are a result of some enlisted soldiers missing
AFQT scores. Parental months absent are the number of months absent out of the past 51 months. See
note in table 1 for additional sample description.
reveal a marginally significant decline in math scores of 0.63 points for
children with a parent deployed between 1–6 months during the last 4
years relative to children with no parental deployments. There is also a
significant decline of 1.5 points for children with a parent deployed 7 or
more months during the previous 4 years relative to children with no
parental deployments. Panel B contains the estimates for officers’ children.
Like the enlisted estimates, the effect for children with parents absent 1–6
months over the past 4 years remains small and insignificant, but children
with parents absent 7 months or more experience a decline of 2 points
in test scores (a fifth of a standard deviation).
A comparison of the 1.13 point decline associated with parental absences
of 3 months or more in the current school year (table 2) with the 0.63
point decline associated with parental absences of 1–6 months in duration
over the past 4 years (table 4) suggests that the detrimental effect of a
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Parental Absence and Children’s Academic Achievement 335
parental absence may lessen over time. Furthermore, the larger 1.51 point
decline associated with a longer parental absence over the past 4 years as
compared to estimates of shorter parental absences suggests that absences
of longer duration have larger adverse effects on children’s test scores.
Since certain characteristics of children and their parents can exacerbate
these effects, I include interaction specifications for children of enlisted
soldiers in table 5.12
Panel A compares absence estimates for children with
a father in the army relative to children with a mother in the army. For
all of my measures of parental absences, the absence of a mother has an
adverse effect that is at least as strong as the absence of a father, and in
most cases, the effect is a much stronger one. Since only 13% of the
children in this sample have a mother in the army, the standard errors on
these estimates are relatively large. Only the estimate for children with a
mother who is deployed for 7 months or more over the last 4 years is
statistically significant. Estimates in column 3 indicate that children with
a father in the army who is deployed for 7 months or more during the
past 4 years experience a 1.36 point decline in test scores relative to children
with no parental deployments over the past 4 years. In contrast,
children with a mother in the army who is deployed for 7 months or
more experience a 5.07 point decline in test scores relative to children
with no parental deployments over the past 4 years.
Panel B of table 5 contains similar specifications that compare children
with married parents to children with a single parent. Single parents make
up only 10% of the sample, so again the standard errors are large. However,
point estimates have a pattern similar to those in panel A. Children
with an absent single parent score worse than children with an absent
married parent.
Comparisons for children with parents in the top 40% and the bottom
60% of the AFQT distribution are in panel C. In all cases, there is no
statistically significant decline in academic achievement for children with
parents in the top 40% of the AFQT distribution who experience a parental
absence. Yet, there is a significant negative effect for children with
parents in the bottom 60% of the AFQT distribution. This finding further
supports the army’s use of the AFQT score as a measure of potential
success in the armed forces: the households of soldiers who have higher
AFQT scores are better able to handle the parental absences associated
with a military vocation.
Finally, panel D contains evidence that parental absences have larger
adverse effects on younger children than older children. Jensen, Martin,
and Watanabe (1996) find similar results for behavioral responses of young
children with parents deployed during the Gulf War. Weisenberg et al.
12
I do not present results for officers’ children because officers do not have
AFQT scores and the sample is too small to produce informative estimates.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Table 5
OLS Interacted Current School Year and Cumulative 4-Year Parental Absence Specifications
for Children of Enlisted Soldiers
Deployed ≥ 3 Months in
the Current School Year
(1)
Deployed 1–6 Months
in the Past 4 Years
(2)
Deployed ≥ 7 Months
in the Past 4 Years
(3)
A. Father in the army versus mother in the army:
Father is in the army Ϫ1.111 Ϫ.594 Ϫ1.357
(.507) (.384) (.512)
Mother is in the army Ϫ2.478 Ϫ1.570 Ϫ5.073
(5.167) (1.805) (2.593)
B. Married parent versus single parent:
Parent is married Ϫ1.026 Ϫ.573 Ϫ1.489
(.511) (.385) (.517)
Parent is single Ϫ5.513 Ϫ2.075 Ϫ1.957
(3.237) (1.715) (2.185)
C. Army parent has high AFQT versus army parent has low AFQT:
Parent in top 40% of AFQT Ϫ.016 Ϫ.286 Ϫ.925
(.959) (.698) (.938)
Parent in bottom 60% of AFQT Ϫ1.503 Ϫ.754 Ϫ1.679
(.589) (.444) (.587)
D. Elementary-level versus secondary-level children:
Elementary-age child (grades 3–6) Ϫ1.444 Ϫ.654 Ϫ1.858
(.621) (.453) (.639)
Secondary-age child (grades 7, 8, and 10) Ϫ.478 Ϫ.599 Ϫ.866
(.862) (.621) (.770)
Observations 11,311 10,891 10,891
Note.—Dependent variable is Texas Learning Index score for mathematics. Standard errors (in parentheses) account for clustering at the individual-child level because some
children appear in both years. All regressions contain a constant and dummies for the year 1998, grades 4–8 and 10, child’s gender, child’s race, parent’s marital status, military
parent’s gender, military parent’s education level, and military parent’s AFQT quartile. There are no officer results presented because officers do not have AFQT scores and
the sample is too small to produce informative estimates. See note in table 1 for additional sample description.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Parental Absence and Children’s Academic Achievement 337
(1993) also observe a larger adverse effect among younger Israeli children
during the Gulf War.
V. Empirical Results for Relocations
In this section, I estimate the impact of household relocations on children’s
academic achievement. I use an approach parallel to that in the
previous section, but here Dit from equation (1) represents the number
of moves a child has made since the parent has been in the military. Table
6 contains summary statistics presented by rank of the parent in panel A
and by number of military-induced moves in panel B. Similar to the
parental absence section, comparisons across the relocation categories reveal
small correlations with a few of the relevant descriptive variables.
Unfortunately, in this case I do not have an instrumental variable to test
the assumption that relocation assignments are exogenous. However, the
small degree of correlation between the relocation variables and the other
observable variables shown in appendix table A2 supports the assertion
that the “needs of the army” and not the personal characteristics of soldiers
determine relocations.13
Interpreting causal estimates of from the modeld
in equation (1) requires that military-induced relocations be orthogonal
to other potential determinants of children’s academic achievement.14
As
in the parental absence section, various robustness tests support this
assumption.
Table 7 contains the main estimates for relocation effects. Estimates for
enlisted soldiers’ children are in panel A, while estimates for officers’
children are in panel B. Dummy variables for children who experience
three or four moves and children who experience five or more moves
make up Dit, so the omitted category is children who experience fewer
than three moves. Consistent with the identification assumption, estimates
of in columns 1–3 are only modestly sensitive to the inclusion of ad-d
ditional control variables. Moreover, dropping children with a single parent,
a mother in the army, a parent with a low AFQT score, or a parent
13
The observable variables in table A2 explain less than 3% of the overall
variation in relocations for the 3–4 moves category. The grade level of the child
explains most of the 11% variation in relocations for the 5-or-more moves category
because older children are more likely to have moved more times.
14
I draw additional support for the exogenous relocation assignment claim from
Angrist and Johnson (2000, 43). They assert, “The nature of duty assignments
varies considerably, and families have little control over the timing of moves or
the location of the next job.” While soldiers may submit a preference for a duty
location, in this study it is the number of relocations that serves as the source of
variation. As long as the frequency of moves is uncorrelated with duty location
preferences and the previous assignment locations, the identifying assumption
holds.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
338
Table 6
Summary Statistics for Household Relocations
B. Number of Moves
A. Parent’s Rank Enlisted Parent Officer Parent
Enlisted
Parent
Officer
Parent 0–2 3–4 ≥ 5 0–2 3–4 ≥ 5
(1) (2) (1) (2) (3) (4) (5) (6)
TLI math score 77.20 82.56 78.19 77.45 76.20 83.25 82.60 82.25
(.12) (.19) (.26) (.17) (.22) (.47) (.30) (.29)
{11.95} {9.60} {11.36} {11.87} {12.35} {9.33} {9.86} {9.41}
Male child .52 .52 .48 .52 .52 .51 .50 .54
(.00) (.01) (.01) (.01) (.01) (.03) (.02) (.02)
{.50} {.50} {.50} {.50} {.50} {.50} {.50} {.50}
White child .43 .82 .45 .44 .41 .87 .80 .81
(.00) (.01) (.01) (.01) (.01) (.02) (.01) (.01)
{.49} {.39} {.50} {.50} {.49} {.34} {.40} {.39}
Parents married .93 .97 .92 .93 .94 .93 .96 .98
(.00) (.00) (.01) (.00) (.00) (.01) (.01) (.00)
{.25} {.18} {.28} {.25} {.23} {.26} {.19} {.12}
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
339
Father is in the army .92 .95 .84 .92 .95 .85 .96 .98
(.00) (.00) (.01) (.00) (.00) (.02) (.01) (.00)
{.28} {.21} {.37} {.27} {.22} {.36} {.19} {.13}
Army parent is a high school
graduate .45 .56 .44 .40
(.00) (.01) (.01) (.01)
{.50} {.50} {.50} {.49}
Army parent has some college .55 .12 .44 .56 .60 .05 .12 .13
(.00) (.01) (.01) (.01) (.01) (.01) (.01) (.01)
{.50} {.32} {.50} {.50} {.49} {.22} {.33} {.34}
Army parent AFQT (top 40%) .25 .26 .27 .22
(.00) (.01) (.01) (.01)
{.43} {.44} {.44} {.41}
Elementary-age child (grades 3–6) .64 .60 .87 .67 .45 .75 .67 .48
(.00) (.01) (.01) (.01) (.01) (.02) (.01) (.02)
{.48} {.49} {.33} {.47} {.50} {.43} {.47} {.50}
Observations 10,206 2,502 1,965 5,023 3,218 389 1,093 1,020
Source.—The army data are from the Office of Economic Manpower Analysis (West Point, NY).
Note.—Standard errors are in parentheses, and standard deviations are in curly braces. There are only 10,121 observations for AFQT. The army data cover children ages
6–19 with a social security number and parents in the active duty army stationed in Texas in 1997 or 1998. Dual military families are dropped. The number of moves equals
the number of moves that a child makes since the parent has been in the army. Children’s math scores (range 0–100) are from the Texas Education Agency (TEA) testing years
1997 and 1998 in grades 3–8 and 10. Under advisement of the TEA, scores below 35 are dropped to account for children who did not take the exam seriously or quit in the
middle of it (this accounts for .5% of the sample and does not affect the results). Army officers must have at least some college, and they do not take the AFQT. This study
separates AFQT quintiles into two groups: top 40% and bottom 60%.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Table 7
OLS Estimates for the Number of Household Relocations
A. Enlisted Parent B. Officer Parent
(1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
3–4 moves Ϫ.376 Ϫ.334 Ϫ.703 Ϫ.977 Ϫ1.391 Ϫ.344 Ϫ.083 Ϫ.194 Ϫ.395 Ϫ.145
(.354) (.352) (.354) (.382) (.514) (.613) (.613) (.617) (.651) (.676)
≥ 5 moves Ϫ1.237 Ϫ1.081 Ϫ1.487 Ϫ1.576 Ϫ1.854 Ϫ.210 .033 Ϫ.104 Ϫ.195 Ϫ.140
(.416) (.411) (.416) (.442) (.591) (.638) (.637) (.654) (.670) (.697)
Male child Ϫ.759 Ϫ.730 Ϫ.647 Ϫ.470 Ϫ.481 Ϫ.491 Ϫ.447 Ϫ.577
(.300) (.300) (.318) (.424) (.508) (.500) (.518) (.558)
White child 3.469 2.533 2.515 2.210 3.763 3.648 3.635 3.925
(.297) (.323) (.334) (.458) (.700) (.699) (.720) (.801)
Parents married .810 Ϫ.064
(.604) (1.259)
Father is in the army 2.410 2.210
(.604) (1.065)
Army parent is a high school graduate Ϫ1.011 Ϫ.978
(.308) (.325)
Army parent has some college Ϫ2.007 Ϫ1.969
(.718) (.718)
Army parent AFQT (top 40%) 2.201 2.356 3.041
(.346) (.360) (.472)
R2
.02 .04 .05 .05 .06 .02 .04 .05 .05 .04
Observations 10,206 10,206 10,121 9,038 4,887 2,502 2,502 2,502 2,355 2,066
Sample restricted to children with married/fathers
in the army No No No Yes Yes No No No Yes Yes
Sample restricted to children whose parents have
some college (enlisted) or have a college degree
(officer) No No No No Yes No No No No Yes
Note.—Dependent variable is Texas Learning Index scores for mathematics. Standard errors (in parentheses) account for clustering at the individual-child level because
some children appear in both years. All regressions contain a constant and dummies for the year 1998 and grades 4–8 and 10. Other controls added as indicated. Differences
in sample sizes within a panel are a result of some enlisted soldiers missing AFQT scores and dropping children with single parents, mothers in the army, and low education
levels where indicated. For panel A, omitting children whose parents had low AFQT scores does not affect the results. See note in table 6 for additional sample description.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Parental Absence and Children’s Academic Achievement 341
with only a high school degree has little effect on the point estimates in
columns 4 and 5.
The specification in column 3 of panel A contains the full set of relevant
control variables for the entire sample of enlisted soldiers’ children. Estimates
of indicate that children who move three or four times scored
0.7 points lower than children who move fewer than three times. Children
who move five or more times score 1.5 points lower than children who
move fewer than three times. For children of enlisted soldiers, these estimates
imply that the number of relocations has an increasingly adverse
effect on academic achievement. In contrast, parallel estimates for officers
in panel B are close to zero and insignificant. While a detailed analysis
of differences between the children of enlisted soldiers and those of officers
is beyond the scope of this article, these findings suggest that some
dimension of an officer’s family, perhaps greater parental education or
higher income, mitigates the adverse effects of relocations.
There is, however, a potentially confounding issue with regard to measuring
relocation effects on standardized test scores. A common observation
on standardized tests is that students become more proficient at
taking the exams over time. The construction of the relocation variable
allows for the possibility that children who have moved fewer times could
also have spent more time in Texas. Thus, the estimates may be picking
up a time-in-Texas effect rather than a relocation effect. For about 60%
of the sample, I am able to establish the number of concurrent years that
a child has taken the TAAS exams. Therefore, in table 8, I include control
variables for time in Texas to determine if they affect the main estimates
found in table 7.
For comparison purposes, column 1 of table 8 contains estimates from
the specification in column 3 of table 7. In column 2, the sample contains
only children for whom I can ascertain the number of TAAS exams taken.
Comparisons of columns 1 and 2 in panel A reveal that the point estimate
for three or four moves is about half the size of the estimate in column
1, while the point estimate for the five or more moves category changes
only slightly. Estimates in column 3 indicate a strong time-in-Texas effect.
A child who takes the TAAS exam a second time scores 3.12 points higher
than the first time he or she took the TAAS exam. By the third time a
child takes the TAAS exam, he or she scores 5.17 points higher than the
first time. I include estimates from specifications containing both the
relocation variables and the time-in-Texas controls in column 4. Comparing
estimates in column 4 to those in columns 2 and 3 reveals no
significant correlations between the number of moves and the time in
Texas.
Although the results for time in Texas are empirically striking, they are
difficult to interpret. One possible interpretation is that these estimates
are picking up an effect of children becoming more familiar with the
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Table 8
OLS Estimates for Household Relocations and Time in Texas
A. Enlisted Parent B. Officer Parent
(1) (2) (3) (4) (1) (2) (3) (4)
3–4 moves Ϫ.703 Ϫ.481 Ϫ.495 Ϫ.194 .039 .115
(.354) (.431) (.428) (.617) (.730) (.729)
≥ 5 moves Ϫ1.487 Ϫ1.268 Ϫ1.036 Ϫ.104 Ϫ.150 Ϫ.068
(.416) (.522) (.521) (.654) (.790) (.786)
Child is taking TAAS exam for the
second time 3.116 3.064 1.715 1.715
(.411) (.412) (.563) (.565)
Child is taking TAAS exam for the
third time 5.171 5.121 .599 .569
(.562) (.563) (1.049) (1.043)
R2
.05 .05 .07 .07 .05 .05 .06 .06
Observations 10,121 6,462 6,462 6,462 2,502 1,675 1,675 1,675
Note.—Dependent variable is Texas Learning Index scores for mathematics. Standard errors (in parentheses) account for clustering at the individual-child level because some
children appear in both years. All regressions contain a constant and dummies for the year 1998, grades 4–8 and 10, child’s gender, child’s race, parent’s marital status, military
parent’s gender, military parent’s education level, and military parent’s AFQT quintile (enlisted only). The data used in table 7 were merged by cells with a panel data set created
specifically for this test. Cells consist of year, grades, child’s gender, child’s race, parent’s marital status, military parent’s gender, military parent’s education level, and military
parent’s AFQT quintile (enlisted only), relocation status, and math score. Identical cells were dropped (630) from the premerged test data. Since only data from 1997 and 1998
are used, a child can take the TAAS exam at most three times. Differences in sample size between col. 1 and cols. 2–4 are a result of matching with the duration data. All
observations with a score prior to 1996 or that have an initial score in tenth grade are dropped, since the length of time in Texas cannot otherwise be determined. See note in
table 6 for additional sample description.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Parental Absence and Children’s Academic Achievement 343
Table 9
OLS Interacted Household Relocation Specifications
Moved
3–4 Times
(1)
Moved
≥ 5 Times
(2)
A. Father in the army versus mother in the army:
Father is in the army Ϫ.854 Ϫ1.476
(.374) (.432)
Mother is in the army .675 Ϫ2.600
(1.029) (1.439)
B. Married parent versus single parent:
Parent is married Ϫ.770 Ϫ1.450
(.369) (.427)
Parent is single .130 Ϫ2.282
(1.174) (1.434)
C. Army parent has high AFQT versus low AFQT:
Parent in top 40% of AFQT Ϫ.698 Ϫ.121
(.606) (.698)
Parent in bottom 60% of AFQT Ϫ.698 Ϫ1.891
(.422) (.481)
D. Elementary-level versus secondary-level children:
Elementary-age child (grades 3–6) Ϫ.636 Ϫ1.979
(.379) (.486)
Secondary-age child (grades 7, 8, and 10) Ϫ.530 Ϫ.694
(.920) (.927)
Observations 10,121 10,121
Note.—Dependent variable is Texas Learning Index score for mathematics. Standard errors (in parentheses)
account for clustering at the individual-child level because some children appear in both years.
All regressions contain a constant and dummies for the year 1998, grades 4–8 and 10, child’s gender,
child’s race, parent’s marital status, military parent’s gender, military parent’s education level, and military
parent’s AFQT quartile. There are no officer results presented because officers do not have AFQT scores
and the sample is too small to produce informative estimates. See note in table 6 for additional sample
description.
TAAS exams over time. An alternative explanation is that prior locations
may have had lower educational standards than Texas schools; in that
case, these estimates reflect a pure increase in human capital development
or a catching-up effect. It is also possible that children who live in Texas
for a longer duration may perform better on the test because of increased
stability from remaining in one location longer. Untangling this issue is
not possible with the data used in this study; however, the results raise
important questions for future research.
Finally, table 9 contains estimates from interacted specifications (as in
table 5) to determine how household relocation effects vary across different
households. The interpretation of these findings is consistent with
that discussed in the parental absence section. Estimates in panel A show
that children with a father in the army who move five or more times
score 1.5 points lower than children who move fewer than three times.
However, children with a mother in the army who move five or more
times score 2.6 points lower than children who move fewer than three
times. Likewise, among children who move five or more times, estimates
in panel B reveal a 0.83 point difference in test scores between children
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
344 Lyle
with single parents and children with married parents. In panel C, children
with parents in the top 40% on the AFQT experience no significant effect
from an increased number of moves. Meanwhile, children with parents
in the bottom 60% on the AFQT score nearly two points lower when
they move five or more times relative to children who move less than
three times. Also, panel D shows that younger children score worse than
older children who move five or more times.15
VI. Interpretation and Conclusion
Although using the military to evaluate the effect of parental absences
and household relocations on children’s academic achievement is appealing
from an empirical perspective, the unique nature of military service
also requires a careful interpretation of the findings. For example, lessons
from this study may not generalize because of differences between the
underlying military and civilian populations. At the time of this study,
the military was manned with an all-volunteer force and only accepted
applicants who met minimum mental, physical, and medical standards.
Episodes of parental absences and household relocations in the military
may also differ from those in the civilian sector in terms of duration,
frequency, danger, and available support structures. However, a study by
Hiew (1992) provides some evidence that findings from a military population
may apply, at least in part, to the civilian population. Hiew (1992)
directly compares military and civilian parental absences by estimating
the effects of deployments in the Canadian military as well as employment-induced
separation of fathers in Japan on children’s behavioral outcomes.
He finds that both military and civilian parental absences result
in comparable stress levels, behavioral problems, and correspondingly
poor attitudes toward academic achievement for elementary-age children.
Apart from the comparability between military children and civilian children,
the question of how military labor force requirements affect military
families is a timely issue of first-order importance for the military and
society as a whole. The U.S. Department of Defense is the second-largest
employer in the United States, and approximately 60% of the country’s
2.4 million active-duty and reserve soldiers have children.16
Little is known
about how these children will respond to the longer, widespread, recurrent,
and increasingly hazardous deployments of their military parents necessi-
15
In table A3 in the appendix, I provide estimates from specifications that
include both parental absence and household relocation variables in the same
regression to test for possible correlations. Comparing estimates in cols. 1–4
reveals very little correlation between parental absences and household relocations.
I also test for correlations between parental absence and time in Texas in cols.
5–7 and find no evidence of a substantive correlation.
16
Estimates are based on 2002 data from the Directorate for Information Operations
and Reports, Office of the Secretary of Defense.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Parental Absence and Children’s Academic Achievement 345
tated by the ongoing Global War on Terror. Understanding these effects
also has implications for assessing the costs of conscription, a policy that
has repeatedly resurfaced over the course of history in times of war.
A related issue is whether the effects of parental absences and household
relocations should be seen as economically important. At first glance, the
fairly small magnitude of the negative effects found in this study suggests
that parental absences and household relocations had little impact on the
educational achievement of military children in the late 1990s. However,
since human capital development is a building process, a child who falls
behind in one year may fall further and further behind with subsequent
years of education. A small educational setback in the third grade could
become quite substantial by the twelfth grade. Although the scope of this
study precludes a complete assessment of this issue, a natural avenue for
future research would employ a longitudinal study to explore how absences
and relocations affect children’s academic achievement over time.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
346
Appendix
Table A1
Covariate Correlations with Parental Absences
A. Deployed ≥ 3
Months in the Current
School Year
B. Deployed 1–6
Months in the Past
4 Years
C. Deployed ≥ 7
Months in the Past
4 Years
D. One-Third of
Unit Deployed ≥ 3
Months
Enlisted
Mean
Officer
Mean
Enlisted
(1)
Officer
(2)
Enlisted
(1)
Officer
(2)
Enlisted
(1)
Officer
(2)
Enlisted
(1)
Officer
(2)
Male child .515 .513 .000 Ϫ.003 .005 Ϫ.018 Ϫ.010 Ϫ.017 Ϫ.002 Ϫ.004
{.500} {.500} (.004) (.008) (.007) (.014) (.006) (.011) (.004) (.007)
White child .439 .788 .004 Ϫ.012 .024 .012 Ϫ.012 Ϫ.013 .002 Ϫ.025
{.496} {.409} (.005) (.010) (.008) (.017) (.006) (.014) (.005) (.010)
Parents married .902 .928 .021 .003 .031 .043 .019 .022 .020 Ϫ.004
{.297} {.258} (.005) (.013) (.010) (.017) (.007) (.012) (.005) (.013)
Father is in the army .872 .908 .043 .051 .082 .110 .042 .071 .050 .045
{.334} {.290} (.004) (.008) (.008) (.013) (.006) (.008) (.004) (.008)
Army parent is a high school graduate .448 .022 .053 .018 .024
{.497} (.004) (.008) (.006) (.005)
Army parent has some college .549 .137 .061 .009 .026 .056
{.498} {.344} (.016) (.020) (.019) (.015)
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
347
Army parent AFQT (top 40%) .275 Ϫ.004 Ϫ.008 Ϫ.014 .001
{.446} (.005) (.009) (.007) (.005)
Child in fourth grade .169 .156 .000 Ϫ.001 Ϫ.008 .033 .015 Ϫ.010 Ϫ.009 .001
{.375} {.363} (.008) (.015) (.010) (.019) (.008) (.016) (.008) (.014)
Child in fifth grade .160 .150 Ϫ.007 .000 Ϫ.014 .033 .011 Ϫ.012 Ϫ.010 Ϫ.005
{.366} {.357} (.008) (.015) (.012) (.022) (.009) (.018) (.008) (.013)
Child in sixth grade .152 .152 Ϫ.012 .011 Ϫ.027 .023 .003 Ϫ.014 Ϫ.015 .005
{.359} {.359} (.008) (.016) (.012) (.022) (.009) (.018) (.008) (.014)
Child in seventh grade .141 .138 Ϫ.012 .002 Ϫ.016 .065 .000 .007 Ϫ.015 .005
{.348} {.345} (.008) (.016) (.012) (.024) (.009) (.020) (.008) (.014)
Child in eighth grade .125 .136 Ϫ.016 .009 .031 .038 Ϫ.005 Ϫ.005 Ϫ.018 .007
{.331} {.342} (.016) (.023) (.019) (.015)
Child in tenth grade .096 .122 Ϫ.014 .000 Ϫ.015 .032 .010 Ϫ.012 Ϫ.022 .004
{.294} {.327} (.009) (.016) (.013) (.023) (.010) (.019) (.009) (.015)
Intercept .029 .010 Ϫ.003 Ϫ.051 .008 .008 .040 .029
(.007) (.015) (.012) (.020) (.010) (.019) (.008) (.014)
R2
.03 .02 .02 .02 .01 .01 .04 .02
Observations 11,311 2,900 11,311 2,900 10,891 2,873 10,891 2,873 11,311 2,900
Note.—Dependent variable is dummy variable for absent during period. Standard deviations (in curly braces in first two columns) are those of the means of all characteristic
variables. Standard errors (in parentheses in remaining columns) account for clustering at the individual-child level because some children appear in both years. All regressions
contain a year dummy for the year 1998. Means and standard deviations are calculated using data from panel A. See note in table 1 for sample description.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Table A2
Covariate Correlations with Household Relocations
Enlisted
Mean
Officer
Mean
A. Moved 3–4
Times
B. Moved ≥ 5
Times
Enlisted
(1)
Officer
(2)
Enlisted
(1)
Officer
(2)
Male child .516 .517 .009 Ϫ.024 .017 .030
{.500} {.500} (.012) (.023) (.010) (.022)
White child .428 .818 .003 Ϫ.040 Ϫ.016 Ϫ.006
{.495} {.386} (.013) (.030) (.011) (.028)
Parents married .932 .966 Ϫ.023 Ϫ.173 Ϫ.043 .083
{.251} {.182} (.026) (.065) (.022) (.051)
Father is in the army .918 .953 .044 .181 .159 .211
{.274} {.211} (.024) (.060) (.019) (.044)
Army parent is a high school graduate .446 Ϫ.012 Ϫ.042
{.497} (.012) (.010)
Army parent has some college .554 .116 .031 .048
{.497} {.320} (.035) (.034)
Army parent AFQT (top 40%) .249 .042 Ϫ.051
{.432} (.014) (.013)
Child in fourth grade .169 .159 .063 .020 .086 .066
{.375} {.366} (.015) (.032) (.011) (.028)
Child in fifth grade .162 .150 .089 Ϫ.044 .173 .177
{.369} {.357} (.017) (.037) (.013) (.033)
Child in sixth grade .153 .153 .062 Ϫ.048 .210 .238
{.360} {.360} (.018) (.036) (.014) (.034)
Child in seventh grade .141 .135 .023 Ϫ.135 .313 .341
{.348} {.342} (.018) (.039) (.015) (.035)
Child in eighth grade .127 .139 Ϫ.037 Ϫ.135 .404 .320
{.333} {.347} (.019) (.037) (.016) (.035)
Child in tenth grade .095 .122 Ϫ.030 Ϫ.146 .422 .340
{.293} {.327} (.020) (.039) (.018) (.036)
Intercept .445 .562 .029 Ϫ.113
(.027) (.070) (.022) (.055)
R2
.01 .03 .11 .08
Observations 10,121 2,502 10,121 2,502 10,121 2,502
Note.—Dependent variable is dummy variable for household relocation effect. Standard deviations (in curly
braces in first two columns) are those of the means of all characteristic variables. Standard errors (in parentheses
in remaining columns) account for clustering at the individual-child level because some children appear in
both years. All regressions contain a year dummy for the year 1998. See note in table 6 for sample description.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
Parental Absence and Children’s Academic Achievement 349
Table A3
Specification Test for Parental Absence and Household Relocation
Correlations
Enlisted Parent
(1) (2) (3) (4) (5) (6) (7)
Absent ≥ 3
months -1.133 Ϫ.592 Ϫ.642 .113 .017
(.507) (.605) (.605) (.681) (.673)
3–4 moves Ϫ.561 Ϫ.559
(.352) (.352)
≥ 5 moves Ϫ1.204 Ϫ1.212
(.412) (.412)
Child is taking
TAAS exam
for the second
time 3.073 3.073
(.389) (.389)
Child is taking
TAAS exam
for the third
time 4.776 4.775
(.535) (.535)
R2
.05 .05 .05 .05 .05 .07 .07
Observations 11,311 10,306 10,306 10,306 7,178 7,178 7,178
Note.—Dependent variable is Texas Learning Index score for mathematics. Standard errors (in parentheses)
account for clustering at the individual-child level because some children appear in both years.
All regressions contain a constant and dummies for the year 1998, grades 4–8 and 10, child’s gender,
child’s race, parent’s marital status, military parent’s gender, military parent’s education level, and military
parent’s AFQT quintile (enlisted only). See table 8 note for a description of the matching procedure. See
notes in tables 1 and 6 for additional sample description.
References
Altonji, Joseph, Todd Elder, and Christopher Taber. 2000. Selection on
observed and unobserved variables: Assessing the effectiveness of Catholic
schools. NBER Working Paper no. 7831, National Bureau of Economic
Research, Cambridge, MA.
Angrist, Josh, and John Johnson. 2000. Effects of work-related absences
on families: Evidence from the Gulf War. Industrial and Labor Relations
Review 54, no. 1:41–58.
Davey, Monica. 2004. The new military life: Heading back to the war.
New York Times, December 20.
Fields, Jason, and Lynne Casper. 2001. American families and living arrangements.
Current Population Reports, U.S. Census Bureau, Washington,
DC.
Haveman, Robert, and Barbara Wolfe. 1993. Children’s prospects and
children’s policy. Journal of Economic Perspectives 7, no. 4:153–74.
———. 1995. The determinants of children’s attainments: A review of
methods and findings. Journal of Economic Literature 33, no. 4:1829–78.
Heinlein, Lisa M., and Marybeth Shinn. 2000. School mobility and student
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms
350 Lyle
achievement in an urban setting. Psychology in the Schools 74, no. 4:
349–57.
Hiew, C. 1992. Separated by their work: Families with fathers living apart.
Environment and Society 24, no. 2:206–25.
Jensen, Peter, David Martin, and Henry Watanabe. 1996. Children’s response
to parental separation during Operation Desert Storm. Journal
of the American Academy of Child and Adolescent Psychiatry 35, no.
4:433–41.
Marchant, Karen, and Fred Medway. 1987. Adjustment and achievement
associated with mobility in military families. Psychology in the Schools
24:289–94.
McLanahan, Sara, and Gary Sandefur. 1994. Growing up with a single
parent: What hurts, what helps. Cambridge, MA: Harvard University
Press.
Nearin, Lynn, and Robert Segal. 2002. Military families prepare for possible
war with Iraq. National Public Radio, November 11.
Piaget, Jean. 1977. Moral judgment: Children invent the social contract.
In The essential Piaget, ed. H. Gruber and J. Jacques Vone`che. New
York: Basic.
Pisano, Mark Charles. 1992. The children of Operation Desert Storm: An
analysis of California Achievement Test Scores in sixth graders of deployed
and nondeployed parents. PhD diss., Campbell University.
Schachter, Jason. 2001. Geographic mobility. Current Population Reports,
U.S. Census Bureau, Washington, DC.
Taubman, Paul. 1989. Role of parental income in educational attainment.
American Economic Review 79, no. 2:57–61.
Texas Education Agency. 2000. Texas Student Assessment Program: Technical
digest. Austin: Texas Education Agency.
U.S. Bureau of the Census. 1980, 1990, 2001. Statistical Abstract of the
United States. Washington, DC: U.S. Government Printing Office.
U.S. Bureau of the Census. 1996, 1999, 2000. Current Population Reports
from the CPS. Washington, DC: U.S. Government Printing Office.
U.S. General Accounting Office. 1994. Elementary school children: Many
change schools frequently, harming their education. GAO/HEHHS-94-
45, 1–55. Washington, DC: Health, Education, and Human Services
Division.
Weisenberg, M., J. Schwarzwald, M. Waysman, Z. Solomon, and A. Klingman.
1993. Coping of school age children in the sealed room during
scud missile bombardment and postwar stress reactions. Journal of Consulting
and Clinical Psychology 61:462–67.
This content downloaded from 194.228.79.120 on Wed, 21 Feb 2018 12:49:55 UTC
All use subject to http://about.jstor.org/terms