638440

research-article2016
EPAXXX10.3102/0162373716638440Broton et al.Working for College

Educational Evaluation and Policy Analysis

September 2016, Vol. 38, No. 3, pp. 477­–494
DOI: 10.3102/0162373716638440
© 2016 AERA. http://eepa.aera.net

Working for College: The Causal Impacts of Financial

Grants on Undergraduate Employment

Katharine M. Broton
Sara Goldrick-Rab
University of Wisconsin–Madison

James Benson
Institute of Education Sciences

One way in which financial aid is thought to promote college success is by minimizing the time students
spend working. Yet, little research has examined if this intended first-order effect occurs, and results
are mixed. We leverage a randomized experiment and find that students from low-income families in
Wisconsin offered additional grant aid were 5.88 percentage points less likely to work and worked 1.69
fewer hours per week than similar peers, an 8.56% and 14.35% reduction, respectively. Students
offered the grant also improved qualitative aspects of their work experiences; they were less likely to
work extensively, during the morning hours, or overnight. Grant aid thus appears to partially offset
student employment, possibly improving prospects for academic achievement and attainment.

Keywords: financial aid, employment, low-income college students, randomized experiment

Nearly three in four undergraduates work for Turner, 2004). Among younger students (ages =
pay while enrolled in college. Among those 16–24) attending college full-time, two in five
employed, one in five work full-time, and of work, and the majority (64% of working stu-
those working part-time, half work more than 20 dents) report working 20 or more hours per week
hours per week (Davis, 2012). The trend of (U.S. Department of Education, 2014).
increasing undergraduate employment dates Studies of financial aid show a positive effect
back a half-century, concomitant with substantial of grants and scholarships on college completion
growth in college costs and shrinkage in the pur- (e.g., Alon, 2007; Castleman & Long, 2013; S.
chasing power of need-based financial aid Dynarski, 2008; S. M. Dynarski, 2003; Goldrick-
(Bowen, Chingos, & McPerson, 2009; Goldrick- Rab, Kelchen, Harris, & Benson, 2016; Scott-
Rab, 2016). Furthermore, the earning power of Clayton, 2011b; Singell, 2004), but we know
those without a college degree has declined little about the mechanisms through which those
(Lemieux, 2008), and precarious working condi- effects arise. One common hypothesis is that
tions that make work more less certain or secure grant aid helps students cover their college costs
for employees have become more common over so that they do not have to work as much to meet
this time period (Kalleberg, 2011; Lambert, their needs (e.g., Goldrick-Rab, Harris, & Trostel,
Fugiel, & Henly, 2014). Rising undergraduate 2009; Richburg-Hayes et al., 2015). However,
employment is concentrated among full-time stu- few studies have directly examined this relation-
dents in their late teens and early 20s, rather than ship, and the results are mixed. Experimental
older adults who have worked at approximately studies of relatively modest performance-based
the same rate since 1970 (Scott-Clayton, 2012; scholarships (awarded conditional on academic
Broton et al.

progress in college) awarded to low-income stu- extensive nature of student employment is

dents often report no impact on work behaviors undermining efforts to increase degree attain-
(Barrow & Rouse, 2013; Leuven, Oosterbeek, & ment and diminish persistent and troubling
van der Klaauw, 2010; MacDonald et al., 2009; social inequalities in college completion (Bowen
Mayer, Patel, & Gutierrez, 2015; Richburg- et al., 2009; Demos & Young Invincibles, 2011;
Hayes et al., 2009; Sommo et al., 2014). Quasi- O’Sullivan & Setzer, 2014; U.S. Department of
experimental research results, however, show an Education, 2014).
inverse relationship between relatively generous To determine if grant aid can change work
amounts of merit aid (awarded based on high behaviors among college students, we use data
school performance) and the number of hours from an experimental study in which undergrad-
worked or work earnings (DesJardins, McCall, uates from low-income families were randomly
Ott, & Kim, 2010; Scott-Clayton, 2011b). assigned an offer of an additional need-based
The ability of grant aid to influence students’ grant. Their subsequent work choices were
work behaviors is of particular interest to policy- tracked using a student survey. The results are
makers and practitioners looking for ways to promising—we find that need-based grant aid
improve college success. Several quasi-experi- effectively reduced the quantity and improved
mental studies indicate that working while in col- the quality of student employment. Students who
lege is associated with lower levels of academic were offered the grant were less likely to work at
achievement (Dadgar, 2012; DeSimone, 2008; all and worked fewer hours. They were also less
Scott-Clayton, 2011a; Scott-Clayton & Minaya, likely to work extensively (i.e., 20+ hours/week)
2014; Soliz & Long, 2014; Stinebrickner & and less likely than similar peers to work during
Stinebrickner, 2003) and credit completion the morning or overnight. Further investigation
(Darolia, 2014). Working is also associated with of heterogeneity suggests that the impacts were
an interruption to students’ studies, particularly largest for first-generation students (i.e., neither
when they work long hours (Bozick, 2007; parent has a college degree) though additional
Ehrenberg & Sherman, 1987; Horn & Malizio, research is needed to confirm this subgroup
1998; Orszag, Orszag, & Whitmore, 2001; Van variation.
Dyke, Little, & Callender, 2005), and among
those who graduate, working extends their time Background and Literature Review
to degree (Bound, Lovenheim, & Turner, 2012).
However, high-quality work experiences—par- Student Employment
ticularly jobs with an on-campus location and Compared with previous generations, today’s
those connected to academic interests—can undergraduates are more likely to work and work
shield students from these negative relationships extensively. In 1960, 25% of full-time students of
and even promote college attainment (Ehrenberg traditional age (i.e., between the ages of 16 and
& Sherman, 1987; Perna, 2010; Scott-Clayton & 24 at the time of entry) worked while enrolled in
Minaya, 2014). Therefore, indicators of work college (Stern & Nakata, 1991) compared with
quality and quantity are important for under- 40% today (U.S. Department of Education,
standing students’ work experiences and how 2014). Growth in student employment began in
they might influence future outcomes. the mid-1960s and continued until the percentage
Given their relatively weaker financial of employed full-time traditional-age students
strength, students from low-income families and reached its peak at 52% in 2000 (Stern & Nakata,
those without college-educated parents are more 1991; U.S. Department of Education, 2014).
likely than their more advantaged peers to com- Growth in the percentage of students working
bine schooling with work (Roksa & Velez, 2010; extensively accounted for almost all the growth
Scott-Clayton, 2012; Walpole, 2003). Furthermore, in student employment over this time period. In
these students may be the least capable of jug- 1970, one in seven traditional-age full-time
gling work and school and benefit the most from undergraduates were working 20 or more hours
a reduction in work hours or improvement in per week, whereas today, one in four college stu-
work quality (Scott-Clayton & Minaya, 2014). dents are working extensively (U.S. Department
Thus, it may be that the high incidence and of Education, 2014).

478
 Working for College

As student employment rates have risen, high- families in the bottom income quartile report
quality labor market opportunities for those with- working sometime during the school year com-
out a college degree have plummeted (Kalleberg, pared with 63% of those from families in the top
2011). For students, there are several important quartile. On average, students from the lowest
dimensions of job quality, including wages and income quartile work 3.4 more hours per week
fringe benefits, work schedule and timing of than those from the top quartile (15.2 vs. 11.8
shifts, flexibility and autonomy, connection to hours; Scott-Clayton, 2012). Similarly, students
interests and the academic environment, and whose parents have lower levels of education are
location (Astin, 1993; Kalleberg, 2011; Lambert also more likely to work, even after controlling
et al., 2014; Presser & Ward, 2011), but prior for parental net worth. Roksa and Velez (2010)
research often only distinguishes between on- argue that undergraduates with college-educated
and off-campus work (e.g., Ehrenberg & parents are shielded from extensive labor market
Sherman, 1987; Perna, 2010; Riggert, Boyle, participation to preserve educationally condu-
Petrosko, Ash, & Rude-Parkins, 2006). This sim- cive conditions for success. College-educated
plified measure of quality assumes that on-cam- parents seem to discourage their children from
pus employers are more sensitive to students’ working during college via direct monetary
academic schedules and goals, and students are transfers and cultural capital in the form of coun-
not scheduled to work extensively. Results from seling and advice (Kalenkoski & Pabilonia,
a national survey indicate that 86% of college 2010; Roksa & Velez, 2010).
students work a “regular” job rather than in a
work–study position or assistantship (7%), and a Why Students Work. Students work during col-
small fraction work in both types (7%). Just 4% lege for a variety of reasons including gaining
of college students receive support from the labor market experience, exploring career
Federal Work-Study program, which was estab- options, meeting cultural obligations, maintain-
lished to help provide high-quality jobs to lower ing one’s identity as a worker, and earning money
income students. Correspondingly, 88% of stu- to pay for basic needs, college costs, or supple-
dents work off-campus and 9% report on-campus mental spending (Cheng & Alcántara, 2007;
work (3% work both on- and off-campus). Goldrick-Rab et al., 2009; Perna, 2010). Finan-
Moreover, fewer than one in three working stu- cial considerations play a key role in students’
dents believe that their job is related to their aca- work decisions (Bound et al., 2012; Dundes &
demic major (Perna, 2010). Marx, 2006). Working for financial reasons is
There is a great deal of variation in student particularly common among students from low-
employment, even among full-time traditional- income families; 58% of students from families
age undergraduates. A greater proportion of in the bottom income quartile report they cannot
women than men work while in college (43% vs. afford school without working compared with
35%), and racial differences indicate that Whites 30% of students from the top income quartile
are most likely to be employed (43% White vs. (Scott-Clayton, 2012). In addition to family
34% Black and 37% Hispanic). Greater shares of background, the net price of college also affects
community college students work when com- students’ need to work. Students attending col-
pared with their counterparts at 4-year colleges leges with higher net prices work more hours
(42% vs. 38%), and they are more likely to work than similar peers attending more affordable
more than 20 hours per week (70% vs. 62%). A schools (Kalenkoski & Pabilonia, 2010). Stu-
larger proportion of students attending public dents working for financial reasons, rather than
institutions work compared with their peers at interest motives such as those listed above, may
private colleges (40% vs. 34% at 4-year schools; be more likely to change their work behaviors
U.S. Department of Education, 2014). Students due to an increase in grant aid (Lobel, 1991).
from socioeconomically disadvantaged back- Students often work long hours because their
grounds are also more likely to work and work pay is low and college is expensive. The net price
extensively than their more privileged peers of college has grown as financial aid has lost
(Perna, 2010; Walpole, 2003). Among dependent “purchasing power” and sticker prices have
full-time 4-year college students, 72% from risen. When the Pell Grant was created in the

479
Broton et al.

early 1970s, it covered more than 75% of the cost any non-financial aspects of employment (e.g.,
of attending a public 4-year college for low- work experience). There are several reasons why
income students, whereas today, it covers just substituting grant aid for work might not be plau-
30%. Currently, a dependent student from a fam- sible. For example, a conservative estimate
ily in the lowest income quartile (i.e., median (Kelchen, Hosch, & Goldrick-Rab, 2014) of
annual earnings of US$21,000) would have to 4-year public college students’ unmet annual
pay 59% of her family’s total income to attend a need is US$12,300 for those from families in the
public 4-year college for 1 year, after all grant aid bottom quartile of the income distribution and
has been taken into account (Goldrick-Rab & US$16,200 for those in the third income quartile
Kendall, 2014). Because most families cannot (Goldrick-Rab & Kendall, 2014). Even the most
devote 59% of their total income to pay for col- generous grants and scholarships rarely match
lege, students often turn to work. At the mini- these estimates of financial need. Furthermore,
mum wage rate of US$7.25, this amount of students find the financial aid system compli-
unmet need translates into 33 hours of work per cated and confusing and thus have difficulty
week, year-round.1 accurately estimating their aid packages (e.g.,
Furthermore, the kinds of employment that Bettinger, Long, Oreopoulos, & Sanbonmatsu,
allowed prior generations to work their way 2012; S. Dynarski & Scott-Clayton, 2013).
through college are no longer available Finally, a small proportion of students (i.e., 7%
(Kalleberg, 2011). Among full-time undergradu- of low-income students) primarily work for non-
ates, the most common type of employment is in financial reasons (Scott-Clayton, 2012), and this
the service industry, including waiting tables may be a wise decision given potential future
(Scott-Clayton, 2012). The tipped minimum labor market payoffs (e.g., Light, 2001; Molitor
wage, common among those in the service indus- & Leigh, 2005; Ruhm, 1997).
try, has fallen in value by nearly 60% over the If the amount of grant aid is not large enough to
past 40 years (Allegretto & Filion, 2011). meet students’ full financial need or do so in a
Moreover, part-time employment has become timely manner, then students must negotiate how
more precarious with inconsistent and unpredict- much they plan to work, which loans they are will-
able hours that are often not under the control of ing to take, if they can reduce their budget, or
the employee (Kalleberg, 2011; Lambert et al., receive any additional funds from friends and
2014). Thus, today’s labor situation makes bal- family. For students already struggling to make
ancing work and school more difficult than it was ends meet, the additional grant aid may be put
a generation ago (Goldrick-Rab, 2016). toward existing unmet needs (Broton, Frank, &
Goldrick-Rab, 2014), rather than to reduce
employment. For example, when low-income stu-
Financial Aid and Labor Market
dents were asked how they spent a US$1,000
Participation
scholarship, the most common responses were to
The idea that financial aid can substitute for purchase books and school supplies (65%), pay
work is based on the economic rationale that stu- tuition and fees (18%), pay for transportation
dents need a certain amount of money to attend (7%), and pay bills (5%; Richburg-Hayes et al.,
college, and they are neutral regarding whether 2015). Other students may prefer to reduce loan
those dollars come from aid or work. This theory amounts before they cut back on work (Binder,
relies on several assumptions, including students’ Krause, Miller, & Cerna, 2015; Cha & Patel, 2010;
preferences regarding work and financial aid as Mayer et al., 2015; Miller, Binder, Harris, &
well as their ability to make ends meet and con- Krause, 2011; Patel & Valenzuela, 2013). Those
trol their work hours as preferred. If grant aid is with a normative preference against borrowing
to entirely substitute for work, the dollar value of due to poor prior experiences or cultural norms
the aid must be large enough to meet students’ may view additional grant aid as an opportunity to
full financial need, students must have the infor- act on those preferences (Boatman, Evans, &
mation to make this calculation and trust that the Soliz, 2014; Cunningham & Santiago, 2008).
aid will be delivered in a reliable and timely Students induced to change their work behav-
manner, and they have to be willing to give up iors through increased grant aid may select to

480
 Working for College

reduce their work hours. A general reduction in worked, or earnings, despite positive academic
the number of hours worked may be particularly impacts of the scholarship (Barrow & Rouse,
beneficial for those working extensively and pro- 2013; Mayer et al., 2015; Richburg-Hayes et al.,
vides students with more time to devote to aca- 2009; Sommo et al., 2014). Similarly, two inter-
demic, family, or leisure activities (Greene & national studies in which students were randomly
Maggs, 2015). Students may also use the addi- offered an additional US$250 or US$750 grant if
tional grant aid to change their work behaviors in they met certain academic benchmarks show no
ways that improve the quality of their employ- impact on students’ employment decisions (Leu-
ment experiences. For example, prior research ven et al., 2010; MacDonald et al., 2009). There
suggests that a reduction in off-campus employ- are two notable exceptions. First, results from
ment, rather than on-campus employment, may one experimental study indicate that students
promote academic success. Students may also who were randomly assigned an offer of addi-
prefer to cut back on work during particular times tional scholarship aid if they met certain aca-
of day, such as mornings and early afternoons demic benchmarks worked 3.3 more hours per
when many classes are offered or overnights for week than similar peers who typically work 9.4
better sleep. Of course, these changes in work hours (Binder et al., 2015). In another study, stu-
behaviors depend on students’ ability to meet dents who were randomized to receive an offer of
their preferences and may be difficult for those in an additional performance-based scholarship
precarious labor situations (Kalleberg, 2011; were 4.7 percentage points less likely to work
Lambert et al., 2014) or in areas with few labor anytime during their first year of college (52.1%
opportunities (Scott-Clayton, 2012). A final treatment vs. 56.8% control), but there were no
complicating factor in students’ work decisions experimental impacts on the proportion of stu-
is federal regulation that allows students to work dents currently working, the number of current
themselves out of financial aid if they earn more jobs, or the number of hours currently worked
than the protected income allowance.2 Students’ (Richburg-Hayes et al., 2015).
non-work-study earnings above a certain thresh- In a quasi-experimental analysis using a
old penalize their financial aid award (Goldrick- regression discontinuity design, DesJardins and
Rab, 2016). If a student reduces her income to colleagues (2010) report that Gates Millennial
within the protected allowance, she may actually Scholars worked 4.295 fewer hours in their fresh-
receive more financial aid (Perna, 2010). Thus, man year and 4.233 fewer hours in their junior
the premise that financial aid can simply substi- year than similar peers. Investigation of hetero-
tute for work may be more complicated than it geneous impacts yielded few differences, but it
initially seems. appears that the reduction in freshman work
hours was larger for those whose parents had
Empirical Evidence on Financial Aid and Stu- some college.3 The Gates Millennial Scholarship
dent Employment. The extant evidence on the Program targets high-achieving racial/ethnic
impact of grant aid on students’ work behaviors minority students and offers them a generous
is mixed. Several experimental studies of rela- renewable scholarship worth US$8,000 in the
tively modest performance-based scholarships first year as well as additional non-monetary sup-
report no impact on work while quasi-experi- ports. Assuming a constant linear effect, this
mental analyses of more generous merit scholar- translates into a 32-minute reduction in weekly
ships indicate that aid reduces work. In a series of work hours per US$1,000 of scholarship aid for
demonstration projects, MDRC randomly freshman. Extant research suggests that a 4.295
assigned low-income 2- and 4-year college stu- reduction in work hours enables students to take
dents to receive additional grant aid, often one more 3-credit course per year (Darolia, 2014)
US$1,000 per semester, if they met certain course or improve their GPA by 0.05 to 0.70 units
credit and grade point average (GPA) thresholds. (DeSimone, 2008; Stinebrickner & Stinebrickner,
They theorize that this grant aid can substitute for 2003). Similarly, West Virginia’s Providing Real
work, but four experimental evaluations across Opportunities for Maximizing In-State Student
five states yielded no significant impact on stu- Excellence (PROMISE) scholarship program
dents’ probability of working, number of hours awards high-achieving students full tuition and

481
Broton et al.

fees to attend an in-state 2- or 4-year college. The The Fund for Wisconsin Scholars created a
scholarship is renewable for 4 years, and the process in which eligible participants were identi-
average total award is US$10,000. By exploiting fied using administrative records, randomly
discontinuities in student eligibility and program assigned via lottery, and then only notified of the
timing, Scott-Clayton (2011b) estimates that program if chosen to receive the WSG offer.
awardees earn about US$10 less per week than Twelve hundred students were randomly selected
similar peers, which translates into a 10% reduc- to receive an offer of the WSG and were sent an
tion in weekly earnings. award notification letter that they had to sign and
Overall, it is not clear if these mixed results return to receive the grant, which was then pack-
regarding the relationship between grant aid and aged and disbursed through the college’s financial
work behaviors are attributable to differences in aid office. In selecting the control group, the list of
methods, scholarship types and amounts, the par- eligible non-recipients was blocked by college to
ticular sample of students, or selection bias. facilitate the oversample of non-White students.
Almost all prior experimental analyses of the Thus, the control group is 50% larger (n = 1,800)
impact of grant aid on work show no impact, but and contains more students attending racially/eth-
these are studies of relatively modest perfor- nically diverse colleges. To account for this
mance-based scholarships where students may be research design, we use survey weights as
particularly unsure if they will receive an addi- described below. Data for this research study were
tional aid. Our experimental analysis of a more obtained independently from the program, so as to
generous need-based grant indicates that aid avoid any interaction effects that could compro-
reduces work. Thus, our findings support the mise the research or program (for more details, see
extant quasi-experimental research, which indi- Goldrick-Rab, 2016; Goldrick-Rab et al., 2016).
cates that relatively generous merit scholarships
reduce work. Furthermore, we show that this Research Questions
inverse relationship generalizes beyond a select
sample of high achievers. Can offering students from low-income fami-
lies additional grant aid induce changes in their
work behaviors? We consider average impacts on
Data and Empirical Approach the percent of students working at all, working
The Wisconsin Scholars Grant (WSG) off-campus, and working extensively. We also
estimate the average impact on the number of
The WSG is a privately funded grant, initiated total hours worked, number of hours worked off-
in 2008 and supported by a US$175 million campus, and hours worked in on-campus employ-
endowment from the Fund for Wisconsin ment. Next, we examine if the grant offer affected
Scholars, making it one of the largest need-based the time of day students worked. Finally, we
grant programs in the state (Pope, 2010).4 The investigate variation in impact by factors likely
WSG program offers students attending 4-year to influence work behaviors, including students’
colleges a US$3,500 grant per year and students ascribed characteristics (gender, race/ethnicity),
attending 2-year colleges a US$1,800 grant per type of institution attended (2- or 4-year college),
year, renewable for up to 5 years.5 To be eligible family background (parental education level and
for the grant, students must meet the following income), and prior work behavior.
criteria: Wisconsin residents who attended and
graduated from a state public high school or
Sample and Descriptive Statistics
earned a Wisconsin High School Equivalency
Diploma within 3 years of matriculating to one of The analytic sample includes 1,438 students
the state’s 42 public colleges or universities, who responded to a survey administered in the fall
where they enrolled for at least 12 credits, com- of their second year of college and completed at
pleted the Free Application for Federal Student least one of the work behavior questions.6 Nearly
Aid (FAFSA), and qualified for a federal Pell two thirds of WSLS students (n = 1,879) were eli-
Grant, while still possessing unmet need (exclud- gible to participate in the fall 2009 survey based on
ing loans) of at least US$1. WSG eligibility criteria described above and past

482
Table 1
Baseline Characteristics of Analytic Sample by Treatment Status

Characteristic Control group Treatment group p value Effect size

Response rate for survey-eligible sample (%)a 75.60 77.79 .972 NA

Female (%) 60.42 60.38 .991 −0.001
Targeted racial/ethnic minority (%)b 22.58 23.13 .813 0.019
Average age (years) 18.30 18.40 .543 0.034
Married (%) 1.34 0.99 .570 −0.188
Has dependent child (%) 3.47 3.16 .750 −0.058
First-generation college student (%)c 56.05 56.43 .895 0.010
Father holds a college degree (AA or higher; %) 25.99 27.67 .534 0.052
Mother holds a college degree (AA or higher; %) 34.23 36.01 .538 0.047
Financially dependent on parents (%) 94.20 94.98 .531 0.092
Average expected family contribution (US$) 1,407 1,541 .250 0.068
Parent(s)’ adjusted gross income (US$) 27,109 29,079 .061 0.105
Parent(s)’ investment income (US$)d 5,160 4,151 .210 −0.068
Student’s adjusted gross income (US$) 3,141 3,200 .780 0.016
Student’s investment income (US$) 279 213 .385 −0.052
Eligible for Simplified Needs Test (%) 57.99 55.07 .294 −0.072
Enrolled in 4-year college (%) 63.71 64.17 .861 0.012
First year in college with no prior enrollment (%)e 99.63 99.84 .469 0.520
Sample size 818 620

Note. Data come from students’ 2008 FAFSA except for race/ethnicity, which are self-reported on a survey. No imputation is
performed for missing data items. Effect sizes are calculated according to What Works Clearinghouse (2014). For information
regarding randomization among the full WSLS study sample, see Goldrick-Rab (2016) and Goldrick-Rab, Kelchen, Harris, and
Benson (2016). FAFSA = Free Application for Federal Student Aid; WSG = Wisconsin Scholars Grant; WSLS = Wisconsin
Scholars Longitudinal Study.
a
The fall 2009 survey-eligible sample includes 1,879 students (1,082 in the control group and 797 in the treatment group) and is
based on WSG eligibility criteria and past survey participation.
b
Targeted minority groups include African American, Latino, Southeast Asian, Native American, and multiracial. “Targeted”
refers to a policy of the University of Wisconsin System.
c
First generation means that neither parent has a college degree.
d
Parent investment income had few extreme values with undue influences and were therefore trimmed at the 99th percentile
(Osborne & Overbay, 2004).
e
The large effect size on the no prior college enrollment variable likely reflects the fact that both groups are so close to 1 and the
standard deviation is small rather than a meaningful difference between groups.

survey participation; eligibility for participation targeted racial or ethnic minority.8 The average
was independent of treatment status (p > .05). The age is 18.3 years, 64% attend a 4-year college,
overall response rate was 76.53% (77.79% treat- and 95% are financially dependent on their par-
ment group and 75.60% control group) indicating ents. Their parents’ average adjusted gross
low overall and differential attrition (What Works income is approximately US$28,000, and the
Clearinghouse, 2014).7 Examination of baseline average expected family contribution to college
characteristics indicates no statistically significant expenses is nearly US$1,500. Almost all the stu-
(p < .05) differences between the treatment and dents reported that they are first-time enrollees in
control groups in the analytic sample. Given the college and 56% are first-generation college stu-
internal validity, observed differences in outcomes dents, meaning that neither parent has a college
between the treatment and control groups are degree. Because the full WSLS study sample was
attributable to the WSG offer (Table 1). not eligible to participate in the fall 2009 survey,
Six in 10 students in the analytic sample are the analytic sample is not representative of the
women and nearly one in four identify as a full study sample. Students in the analytic sample

483
Table 2
Baseline Characteristics of Full, Survey-Eligible, and Analytic Samples

Full study Survey-eligible Analytic

Characteristic sample samplea sample p value

Assigned to treatment (%) 40.00 41.12 41.15 .217

Female (%) 58.69 57.74 60.40 .010
Targeted racial/ethnic minority (%)b NA 23.27 22.80 NA
Average age (years) 19.53 18.36 18.34 .000
Married (%) 3.98 0.98 1.20 .000
Has dependent child (%) 13.07 4.07 3.34 .000
First-generation college student (%)c 57.29 56.22 56.21 .282
Father holds a college degree (AA or higher; %) 26.21 26.45 26.66 .243
Mother holds a college degree (AA or higher; %) 33.93 35.00 34.94 .212
Financially dependent on parents (%) 82.39 94.09 94.52 .000
Average expected family contribution (US$) 1,314 1,435 1,462 .015
Parent(s)’ adjusted gross income (US$) 23,220 27,422 27,916 .000
Parent(s)’ investment income (US$)d 4,031 7,050 4,743 .011
Student’s adjusted gross income (US$) 4,784 3,224 3,165 .000
Student’s investment income (US$) 359 347 252 .150
Eligible for Simplified Needs Test (%) 61.87 58.85 56.80 .000
Enrolled in 4-year college (%) 50.00 60.94 63.90 .000
First year in college with no prior enrollment (%) 86.78 99.78 99.72 .000
Sample size 3,000 1,879 1,438

Note. Data come from students’ 2008 FAFSA except for race/ethnicity, which are self-reported on a survey. No imputation is
performed for missing data items. The reported p value compares the analytic sample with the full study sample. FAFSA = Free
Application for Federal Student Aid; WSG = Wisconsin Scholars Grant.
a
The fall 2009 survey-eligible sample includes 1,879 students (1,082 in the control group and 797 in the treatment group) and is
based on WSG eligibility criteria and past survey participation.
b
Targeted minority groups include African American, Latino, Southeast Asian, Native American, and multiracial. “Targeted”
refers to a policy of the University of Wisconsin System.
c
First generation means that neither parent has a college degree.
d
Parent investment income had few extreme values with undue influences and were therefore trimmed at the 99th percentile
(Osborne & Overbay, 2004).

are more likely to be younger, female, financially and how many hours they worked in off-campus
dependent on their parents, first-time college and on-campus employment.9 From these mea-
enrollees, and from more economically advan- sures, we created a total hours worked variable and
taged families. Students in the analytic sample two indicator variables: one noting any off-campus
are also less likely to be married, have dependent work and one noting extensive work if students
children, or attend a 2-year college (Table 2). worked 20 or more total hours in the past week.
Thus, the results may not be generalizable to the Finally, employed students were asked to indicate
sample of Wisconsin students described above. if they had worked during any of the following
blocks of time: 8:00 a.m. to 12:00 p.m., 12:00 p.m.
Survey Measures to 6:00 p.m., 6:00 p.m. to 10:00 p.m., 10:00 p.m. to
2:00 a.m., or 2:00 a.m. to 8:00 a.m.
Measures of undergraduate labor force partici-
pation are based on student self-reports to survey
Analytic Plan
questions in fall 2009 and are based on work mea-
sures used in national studies (e.g., American An intent-to-treat analysis is used to estimate
Community Survey). Students were asked if in the the experimental effect in which students offered
last week they had a job where they worked for pay the WSG are compared with students who would

484
 Working for College

have been offered it if selected during random baseline characteristics, WSGi × Z i is a vector of
assignment.10 The equation is interaction terms between the WSG indicator and
individual-level characteristics, and εi is an error
yi = α + β (WSGi ) + γX i + εi , (1) term. Interaction terms include the following indi-
vidual-level characteristics: gender, race/ethnicity,
where yi is a student outcome, WSGi is an indi- college institutional sector, first-generation college
cator of an offer of the WSG, X i is a vector of student status, family income, and prior work
individual-level baseline controls, and εi is an behavior.13 To conserve space, only statistically
error term. Ordinary least squares regression is significant interaction terms are reported in the
used to estimate impacts on continuous outcome text. To aid in interpretation, we present predicted
variables including total hours worked in the past means and probabilities for each subgroup (Ai &
week as well as total hours worked in on- and off- Norton, 2003).
campus employment. Logistic regression is used
to estimate the effect of the grant offer on dichot- Limitations
omous outcome measures, including working at
all, working off-campus, working extensively, or There are several limitations in this study.
during certain 4- to 6-hour blocks of time through- The sample is not nationally representative and
out the day. Individual-level baseline control vari- only includes full-time, traditional-age students
ables serve to increase precision and include from low-income families attending a public
expected family contribution, financial depen- college in the state of Wisconsin. Older adults,
dency status, parents and students adjusted gross including those who have been primarily in the
income, parents and students investment income, workforce and returned to college are excluded
first-time college enrollment, eligibility for sim- and warrant additional study. Our analyses are
plified needs test, marital status, and parental sta- limited to intent-to-treat estimates to cleanly
tus.11 Inverse probability weights are used in all exploit the exogenous variation in the lottery
analyses due to unequal assignment probabilities design; analyses incorporating non-random
among students across schools.12 The treatment compliance would likely result in slightly larger
impacts are reported as changes in the number of impacts. Finally, work behaviors are self-
hours worked for continuous outcomes and as reported on a survey during students’ second
percentage point differences for dichotomous year of college. We proxy for work quality using
outcomes. In addition, the magnitudes of the measures of work location, work shift, and evi-
treatment impacts are reported as effect sizes to dence of extensive work hours. Longer term
aid the reader in substantive interpretation (Lipsey impacts are not examined.
et al., 2012). The standardized mean difference,
Hedges’s g, is reported for continuous outcomes, Findings
and the Cox index, which is a comparable mea- Among students not offered the WSG, 68.72%
sure, is used for dichotomous outcomes (What report working in the prior week with an average
Works Clearinghouse, 2014). In addition, we of 11.78 hours worked (Table 3). The proportion
report unadjusted impact estimates in the appen- of working students in our sample is 28 percent-
dix as a reference for the reader (Table A1). age points higher than a national estimate of the
Following the analysis of average impacts, percentage of full-time college students (ages =
heterogeneous treatment effects are estimated by 16–24) working in 2009 (U.S. Department of
introducing interaction terms. The equation is Education, 2014). This likely reflects the fact that
our sample only includes students from low-
yi = α + β (WSGi ) + γX i + δZ i
(2) income households who attend a public college
+ θ (WSGi × Z i ) + εi or university. Half of control group students work
off-campus and indeed, this is where they spend
where yi is a student outcome, WSGi is an indica- most of their work hours (10.03 hours off-cam-
tor of an offer of the WSG, X i is a vector of indi- pus vs. 1.74 hours on-campus). Three in 10 stu-
vidual-level baseline controls described above, dents in the control sample work extensively,
Zi is a vector of individual-level whereas nationally in 2009, one in four full-time

485
Table 3
Wisconsin Scholars Grant Impact on Undergraduate Labor Market Participation

Outcome Control Treatment group p value Effect size n

Working at all (%) 68.72 −5.88 .029* 0.159 1,397

Total hours worked (in the last week) 11.78 −1.69 .013* 0.144 1,383
On-campus hours worked 1.74 −0.33 .151 0.083 1,386
Off-campus hours worked 10.03 −1.37 .047* 0.114 1,392
Working off-campus (%) 49.92 −4.36 .124 0.107 1,392
Working extensively (20+ hours; %) 30.21 −5.17 .042* 0.159 1,383
Working 8:00 a.m.–noon (%) 54.97 −7.67 .029* 0.190 916
Working noon–6:00 p.m. (%) 71.81 2.34 .456 0.072 916
Working 6:00 p.m.–10:00 p.m. (%) 59.42 3.05 .370 0.079 916
Working 10:00 p.m.–2:00 a.m. (%) 16.44 −0.67 .797 0.031 916
Working 2:00 a.m.–8:00 a.m. (%) 14.26 −6.40 .005** 0.408 916

Note. The following covariates are included in the models: expected family contribution, parents and students adjusted gross
income, parents and students investment income, dependency status, first-time enrollment, eligibility for simplified needs test,
marital status, and parental status. Sensitivity checks indicate that the inclusion of father’s college status does not substantively
or significantly change the findings; point estimates are somewhat larger with the inclusion of father’s college education status.
Effect sizes are calculated according to What Works Clearinghouse (2014).
†
p < .10. *p < .05. **p < .01.

young adult students reported working 20 or points less likely to work mornings between 8:00
more hours weekly (U.S. Department of a.m. and noon (p < .05; effect size = 0.19) and 6.40
Education, 2014). The most popular time for stu- percentage points less likely to work overnight
dents to work is between noon and 6:00 p.m. between 2:00 a.m. and 8:00 a.m. (p < .01, effect
(71.81%) and more than half work in the morn- size = 0.41).
ings (8:00 a.m.–noon) or evenings (6:00 p.m.–
10:00 p.m.). Late night (10:00 p.m.–2:00 a.m.)
Variation in Impacts
and overnight (2:00 a.m.–8:00 a.m.) shifts are
less common, but 16.44% and 14.26% of stu- Investigation of heterogeneous impacts sug-
dents report working during these times in the gests that main effects differ by level of parental
past week, respectively. education with treatment effects concentrated
Results indicate that offer of the WSG changed among those whose parents do not have a college
both the quantity and quality of undergraduate stu- degree. Due to the large number of interactions
dent employment. Students offered the WSG are tested, however, this finding may be the result of
5.88 percentage points less likely to work at all, a chance and should be interpreted with caution.
reduction of 8.56% (p < .05; effect size = 0.16). Results presented in Table 4 indicate a statisti-
Treatment group students reduced their total cally significant interaction between first-genera-
weekly work hours by 1.69 hours or 14.35% (p < tion college student status and WSG offer on the
.05; effect size = 0.14). This reduction came probability of working extensively (p < .05) and
through a 1.37 hour or 13.66% decline in the num- the number of hours worked off-campus (p < .05).
ber of hours worked off-campus (p < .05; effect Thirty-five percent of first-generation college stu-
size = 0.11) and a 0.33 hour or 18.97% decline in dents in the control group are predicted to work
the number of hours worked on-campus (p = .15; extensively compared with 24.01% of first-gener-
effect size = 0.08). The grant offer also reduced ation students offered the additional grant aid—a
the proportion of students working extensively difference of 11.14 percentage points or a 31.69%
(20+ hours) by 5.17 percentage points or 17.11% reduction. Among those with a college-educated
(p < .05; effect size = 0.16). Finally, offer of the parent, 25.61% of students in the control group
WSG influenced the time of day in which students and 25.44% of students in the treatment group are
worked. Treatment students are 7.67 percentage predicted to work more than 20 hours per week.

486
 Table 4
Subgroup Variation in WSG Treatment Impacts

Off-campus Working
Total hours On-campus hours Working extensively
Interaction model Working at all p value worked p value hours worked p value worked p value off-campus p value (20+ hours) p value

WSG −0.110 .550 −0.782 .458 −0.249 .471 −0.583 .578 −0.079 .668 −0.253 .239
(0.183) (1.052) (0.345) (1.048) (0.184) (0.214)
WSG × Female −0.276 .262 −1.463 .288 −0.139 .762 −1.265 .361 −0.158 .503 −0.008 .977
(0.246) (1.375) (0.459) (1.385) (0.236) (0.268)
WSG −0.371 .009 −1.976 .015 −0.448 .105 −1.498 .069 −0.227 .092 −0.309 .039
(0.143) (0.808) (0.276) (0.822) (0.134) (0.149)
WSG × Minoritya 0.354 .213 1.416 .365 0.256 .621 1.019 .514 0.244 .374 0.315 .309
(0.284) (1.561) (0.518) (1.561) (0.274) (0.309)
WSG −0.280 .177 −2.078 .105 −0.362 .124 −1.788 .165 −0.148 .438 −0.265 .171
(0.208) (1.282) (0.235) (1.288) (0.191) (0.193)
WSG × FourYearCollege 0.021 .933 0.622 .676 0.042 .918 0.649 .663 −0.057 .813 −0.016 .952
(0.255) (1.487) (0.408) (1.490) (0.241) (0.264)
WSG −0.341 .073 −1.034 .306 −0.732 .059 −0.291 .776 0.029 .874 −0.009 .964
(0.190) (1.009) (0.387) (1.020) (0.181) (0.207)
WSG × FirstGenerationb 0.001 .996 −2.041 .149 0.769 .124 −2.819 .048 −0.474 .052 −0.538 .049
(0.256) (1.414) (0.500) 1.427 (0.244) (0.273)
WSG −0.162 .350 −1.414 .165 −0.417 .143 −0.976 .347 −0.079 .629 −0.202 .261
(0.173) (1.018) (0.285) (1.037) (0.163) (0.180)
WSG × HigherIncomec −0.203 .403 −0.591 .664 0.174 .705 −0.821 .552 −0.201 .383 −0.125 .628
(0.242) (1.363) (0.460) (1.379) (0.230) (0.259)
WSG −0.277 .046 −1.828 .018 −0.495 .050 −1.361 .081 −0.134 .294 −0.270 .053
(0.139) (0.769) (0.252) (0.780) (0.128) (0.139)
WSG × NoWork_HSd 0.107 .741 1.709 .294 0.941 .187 0.868 .588 −0.247 .477 0.431 .325
(0.324) (1.628) (0.713) (1.603) (0.347) (0.437)
Sample size 1,397 1,383 1,386 1,392 1,392 1,383
(continued)

487
 Table 4 (continued)

488
Working 8:00 Working Working 6:00 Working 10:00 Working 2:00
Interaction model a.m.–noon p value noon–6:00 p.m. p value p.m.–10:00 p.m. p value p.m.–2:00 a.m. p value a.m.–8:00 a.m. p value

WSG −0.552 .025 −0.016 .952 0.337 .184 0.003 .993 −1.512 .001
(0.246) (0.263) (0.253) (0.316) (0.472)
WSG × Female 0.361 .233 0.222 .503 −0.309 .319 −0.093 .817 1.191 .029
(0.302) (0.332) (0.310) (0.399) (0.545)
WSG −0.447 .007 0.103 .578 0.243 .154 0.069 .750 −0.779 .006
(0.166) (0.185) (0.170) (0.215) (0.281)
WSG × Minoritya 0.566 .104 0.191 .634 −0.439 .215 −0.699 .215 0.426 .488
(0.348) (0.402) (0.354) (0.564) (0.614)
WSG −0.057 .802 0.648 .013 0.234 .308 0.004 .990 −0.660 .052
(0.225) (0.260) (0.230) (0.300) (0.339)
WSG × FourYearCollege −0.425 .146 −0.860 .009 −0.174 .560 −0.092 .817 −0.028 .953
(0.292) (0.330) (0.298) (0.398) (0.482)
WSG −0.464 .042 0.331 .205 0.441 .062 0.041 .895 −0.944 .012
(0.228) (0.261) (0.236) (0.313) (0.374)
WSG × FirstGenerationb 0.235 .439 −0.421 .223 −0.493 .116 −0.085 .837 0.435 .390
(0.304) (0.345) (0.314) (0.412) (0.506)
WSG −0.401 .048 −0.034 .880 0.075 .719 −0.168 .562 −0.572 .088
(0.203) (0.224) (0.209) (0.289) (0.335)
WSG × HigherIncomec 0.163 .567 0.304 .342 0.095 .746 0.236 .547 −0.193 .692
(0.285) (0.320) (0.293) (0.393) (0.486)
WSG −0.364 .019 0.194 .272 0.112 .484 −0.131 .529 −0.780 .003
(0.155) (0.176) (0.159) (0.209) (0.258)
WSG × NoWork_HSd 0.295 .515 −0.214 .672 0.194 .673 0.847 .236 0.893 .374
(0.454) (0.506) (0.459) (0.714) (1.004)
Sample size 916 916 916 916 916

Note. This table presents the WSG coefficient and WSG × Characteristic Coefficient for each of the six interaction models (i.e., gender, race/ethnicity, college institutional sector, first-generation college student status,
family income, and prior work behavior). Standard errors are shown in parentheses. The following covariates are included in each model: expected family contribution, parents and students adjusted gross income, parents
and students investment income, dependency status, first-time enrollment, eligibility for simplified needs test, marital status, and parental status. Maximum sample size is reported. No imputation is performed for missing
data items. Interaction terms with p values below .10 are bolded to aid in interpretation. WSG = Wisconsin Scholars Grant.
a
Minority includes African American, Latino, Southeast Asian, Native American, and multiracial. These groups are “targeted” racial/ethnic minorities according to University of Wisconsin System policy.
b
First generation means that neither parent has a college degree.
c
Higher income is operationalized as those with a parental adjusted gross income above the mean. The results are not sensitive to the operationalization of higher income.
d
No work in high school is operationalized as those who indicated they typically worked 0 hours during their senior year in high school on a student survey (16%). The results are not sensitive to the operationalization
of work behavior in high school. As a sensitivity test, we also examined if the impact of the grant offer varied by students’ work behavior in fall of their first year of college (after they had been notified of the WSG but
typically had not received any grant aid), and it did not.
 Working for College

The reduction in work hours occurred through a college achievement and attainment. Indeed,
decline in the number of hours worked off-cam- research shows that the offer of the WSG
pus. First-generation college students in the con- improved students’ academic outcomes, includ-
trol group are predicted to work 11.53 hours ing on-time bachelor’s degree completion for
off-campus per week while those in the treatment 4-year college entrants (Goldrick-Rab et al.,
group are predicted to work 8.42 hours, a differ- 2016).
ence of 3.11 hours or a 26.97% reduction. Among Furthermore, the impacts of the grant offer
those with a college-educated parent, students in appear to be larger for some of the most disad-
the control group are predicted to work 8.81 hours vantaged students although we cannot be sure
off-campus compared with 8.51 hours among that this subgroup variation is not the result of
those in the treatment group. Treatment impacts statistical chance. First-generation students typi-
regarding working at all, on-campus work, or dur- cally work more than their peers with a college-
ing particular shifts did not systematically vary by educated parent, but the WSG offer entirely
parental education level (Table 4). offset this advantage in extensive and off-campus
In addition, results indicate subgroup varia- work. Thus, the grant offer may have enabled
tion by gender and college institutional sector first-generation students to make work decisions
regarding working during different times of day. on par with students who have a college-educated
Further examination of the data, however, does parent.
not suggest any clear or systematic pattern of The magnitude of the reduction in work hours
subgroup variation in changes to shift work. Men is very similar to prior quasi-experimental esti-
offered the grant were 12.48 percentage points mates. DesJardins and colleagues (2010) suggest
less likely to work overnight (4.30% WSG vs. a 32-minute reduction in weekly work hours per
16.78% control) whereas women were 3.19 per- US$1,000 of merit scholarship aid. Our impact
centage points less likely to work between 2:00 translates into a 35-minute reduction in weekly
a.m. and 8:00 a.m. (9.79% WSG vs. 12.98% con- work hour per US$1,000 of additional need-
trol; p < .05). The grant offer also induced 2-year based grant aid.14 Furthermore, Scott-Clayton
college students to work between noon and 6:00 (2011b) suggests a US$4.04 reduction in weekly
p.m. (80.04% WSG vs. 67.81% control) whereas earnings per US$1,000 of merit scholarship aid
it reduced the probability of working afternoons and assuming minimum wage work, our results
for 4-year college students (70.29% WSG vs. indicate a US$4.25 weekly reduction per
74.48% control; p < .05; Table 3). US$1,000 of grant aid.15 Prior research suggests
that a 1.69-hour reduction in weekly work
induces students to take one additional course
Discussion credit per year (Darolia, 2014) or improve their
Using a randomized experiment, we find that GPA by 0.02 to 0.27 units (DeSimone, 2008;
the offer of a private need-based grant induced Stinebrickner & Stinebrickner, 2003).
undergraduates to change their work behaviors. The effect sizes reported in our study (e.g.,
Students offered the grant were less likely to 0.15) are often interpreted as substantively
work at all and reported working fewer hours. meaningful by education researchers (Lipsey
They also changed their work experiences in et al., 2012), but students offered additional
important ways: treatment students were less grant aid did not come close to substituting a
likely to work extensively and reduced the num- dollar of aid for a dollar of work earnings. On
ber of hours they worked in off-campus employ- average, treatment students in our sample could
ment. In addition, these students changed the have used the grant aid to buy themselves out of
time of day in which they worked, resulting in a US$96 in weekly work earnings or 13.27 hours
smaller proportion working overnight or during of work per week at the minimum wage rate.16
the morning, likely allowing them a greater Instead, students report working less than 2
opportunity to sleep and attend classes. These fewer hours per week. This means that students
quantitative and qualitative changes in work were either unable to use the grant aid to buy
behaviors are consistent with changes recom- themselves out of work or preferred not to do so.
mended by theory and practice to improve There is some evidence that the grant crowded

489
Broton et al.

out loans due to rules about financial aid packag- Scholars and policymakers hypothesize that
ing. Specifically, to include the grant in students’ financial aid can buy students out of work, and
financial aid packages, sometimes loans had to our findings support this theory. However, grant
be reduced. This limited the extent to which the aid is not a simple dollar-for-dollar substitute for
grant could help students have more cash-in- work earnings. Instead, students pay for college
hand to avoid working (Goldrick-Rab et al., through a complex combination of grant aid,
2016). Prior work also indicates that the finan- loans, gifts, work, and cost-cutting measures.
cial need of low-income students is understated Furthermore, changes in work behavior may be
through a systematic overestimate of the an important pathway through which financial
expected family contribution and underestimate aid affects college success, but it is not likely the
of living costs (Goldrick-Rab, 2016; Kelchen only mechanism, and more research is needed to
et al., 2014). Certainly, a sizable minority of stu- understand the full causal chain of events. To the
dents in our sample are struggling to make ends extent that working while in college is a key
meet and report trouble affording basic needs mechanism in the reproduction of social advan-
like food and housing, so these students may tage, the results suggest that need-based grant aid
have applied the additional grant aid to unmet may be a promising path for promoting more
needs (Broton et al., 2014). equitable college outcomes.

Appendix
Table A1
Wisconsin Scholars Grant Unadjusted Impact on Undergraduate Labor Market Participation

Outcome Control Treatment group p value Effect size n

Working at all (%) 68.64 −5.27 0.048* −0.143 1,436

Total hours worked (in the last week) 11.73 −1.61 0.017* −0.137 1,420
On-campus hours worked 1.75 −0.31 0.175 −0.077 1,423
Off-campus hours worked 9.96 −1.31 0.056† −0.109 1,431
Working off-campus (%) 49.55 −3.93 0.160 −0.096 1,431
Working extensively (20+ hours) (%) 29.83 −4.79 0.056† −0.146 1,420
Working 8:00 a.m.–noon (%) 54.14 −7.53 0.030* −0.183 942
Working noon–6:00 p.m. (%) 71.83 1.88 0.544 0.058 942
Working 6:00 p.m.–10:00 p.m. (%) 59.25 3.71 0.273 0.095 942
Working 10:00 p.m.–2:00 a.m. (%) 16.14 0.00 0.999 0.000 942
Working 2:00 a.m.–8:00 a.m. (%) 13.99 −5.84 0.008** −0.367 942

Note. Models do not include any covariates. Effect sizes are calculated according to What Works Clearinghouse (2014).
†
p < .10. *p < .05. **p < .01.

Authors’ Note Declaration of Conflicting Interests

All opinions and errors are those of the authors. The author(s) declared no potential conflicts of inter-
James Benson worked on this project as a post-doc- est with respect to the research, authorship, and/or
toral fellow at the University of Wisconsin–Madison. publication of this article.
This article was written when James Benson was
employed by the University of Wisconsin–Madison, Funding
and it does not necessarily reflect the views of the
The author(s) disclosed receipt of the following finan-
Institute of Education Sciences, where he is cur-
cial support for the research, authorship, and/or publi-
rently employed.
cation of this article: The Bill and Melinda Gates
Foundation, Great Lakes Higher Education Guaranty
Acknowledgments Corporation, Institute for Research on Poverty,
The authors thank Robert Kelchen for feedback. Spencer Foundation, William T. Grant Foundation,

490
 Working for College

Wisconsin Center for the Advancement of 11. These baseline control variables are included
Postsecondary Education, and an anonymous donor because they have an effect size difference greater
provided funding for this study, conducted in partner- than 0.05 standard deviations (see Table 1; What
ship with the Fund for Wisconsin Scholars, the Higher Works Clearinghouse, 2014). Father holds a college
Educational Aids Board, the University of Wisconsin degree is not included in the covariate adjusted model
System, and the Wisconsin Technical College System. due to missing data, but was included as a sensitiv-
The authors also thank the Institute of Education ity check and did not significantly or substantively
Sciences Grant R305B090009 and National Science change the findings; results from that model were less
Foundation Graduate Research Fellowship Grant conservative.
DGE-1256259 for research support. 12. The use of survey weights is designed to ensure
internal validity. As an external validity sensitiv-
Notes ity check, we also ran the models with school fixed
effects. The results are substantively and statistically
1. Calculation based on 59% of US$21,000 =
similar.
US$12,390 in unmet need. (US$12,390 / 52 weeks) /
13. Because we estimate six potential types of sub-
US$7.25 = 33 hours per week.
group variation for each of 11 outcomes, we make 66
2. The Working Student Act (S.2796) is legislation
statistical comparisons. We would expect about three
that has been introduced by Senator Tammy Baldwin
statistically significant results by chance. Thus, we
to enact an across-the-board 35% increase in the
also conduct a visual analysis of the data to determine
income protection allowance for students. This means
the overall pattern of findings. Still, caution is war-
that a dependent student can earn up to US$2,191
ranted in interpreting these results.
more before their income affects their financial aid
14. Calculation based on the weighted average value
award (The Working Student Act, 2014).
of the grant for the analytic sample (US$2,886.30)
3. Note that DesJardins, McCall, Ott, and Kim
assuming a constant linear effect.
(2010) made 54 comparisons, and two were statisti-
15. Scott-Clayton (2011b) reports a US$9.55
cally significant, which is what we would expect by
reduction in weekly earnings per US$2,366 in grant
chance. “So some caution is warranted when interpret-
aid, on average in Years 2 through 4. Our calculation
ing these findings” (p. 472).
is based on the average grant amount of US$2,886.30
4. More information on the Fund for Wisconsin
and a US$7.25 minimum wage rate. These are esti-
Scholars is available at www.ffws.org.
mates, as we do not have wage data.
5. The renewable nature of the grant is conditional
16. For this calculation, we define a school year as
on Pell Grant eligibility, full-time enrollment, and sat-
30 weeks in conjunction with University of Wisconsin
isfactory academic progress. The grant is transferable
(UW) System Regent policy.
among all public 2- and 4-year colleges in Wisconsin.
6. Most students completed all or none of the work
behavior questions; there was 1% item missingness. References
7. The overall attrition rate is 23.47%, and the dif- Ai, C., & Norton, E. C. (2003). Interaction terms in
ferential attrition rate is 2.19%. This level of over- logit and probit models. Economics Letters, 80,
all and differential attrition meets the conservative 123–129.
boundary for “low attrition” according to the What Allegretto, S. A., & Filion, K. (2011, February).
Works Clearinghouse (2014). Waiting for change: The $2.13 federal submini-
8. Targeted minority groups include African mum wage (Briefing Paper 297). Washington, DC:
American, Latino, Southeast Asian, Native American, Economic Policy Institute & Center on Wage and
and multiracial. “Targeted” refers to a policy of the Employment Dynamics.
University of Wisconsin System. Alon, S. (2007). The influence of financial aid in leveling
9. Due to a few outliers, the off-campus work hours group differences in graduating from elite institutions.
measure was trimmed to 40 hours/week and the on- Economics of Education Review, 26, 296–311.
campus work hours measure was trimmed to 20 hours/ Astin, A. (1993). What matters in college? Liberal
week. This affected less than 2% of off-campus cases Education, 79(4), 4–15.
and less than 1% of on-campus cases (Osborne & Barrow, L., & Rouse, C. E. (2013). Financial incen-
Overbay, 2004). tives and educational investment: The impact
10. Approximately, 90% of students who were of performance-based scholarships on student
offered the Wisconsin Scholars Grant (WSG) received time use. Chicago, IL: Federal Reserve Bank of
the grant in the 2008–2009 academic year, according Chicago.
to the Fund for Wisconsin Scholars. We report work Bettinger, E. P., Long, B. T., Oreopoulos, P., &
outcomes from fall 2009. Sanbonmatsu, L. (2012). The role of application

491
Broton et al.

assistance and information in college decisions: Davis, J. (2012, October). School enrollment and
Results from the H&R Block FAFSA experiment. work status: 2011 (American Community Briefs,
The Quarterly Journal of Economics, 127, 1205– ACSBR/11-14). Washington, DC: U.S. Census
1242. Bureau.
Binder, M., Krause, K., Miller, C., & Cerna, O. Demos, & Young Invincibles. (2011). The state
(2015). Providing incentives for timely progress of young America: Economic barriers to the
toward earning a college degree: Results from a American dream. New York, NY: Demos.
performance-based scholarship experiment. New DeSimone, J. S. (2008, May). The impact of employ-
York, NY: MDRC. ment during school on college student academic
Boatman, A., Evans, B., & Soliz, A. (2014). Applying performance (NBER Working paper 14006).
the lessons of behavioral economics to improve the Cambridge, MA: National Bureau of Economic
federal student loan programs: Six policy recom- Research.
mendations. Indianapolis, IN: Lumina Foundation. DesJardins, S. L., McCall, B. P., Ott, M., & Kim, J.
Bound, J., Lovenheim, M. F., & Turner, S. (2012). (2010). A quasi-experimental investigation of
Increasing time to baccalaureate degree in the how the Gates Millennium Scholars program is
United States. Education Finance and Policy, 7, related to college students’ time use and activities.
375–424. Educational Evaluation and Policy Analysis, 32,
Bowen, G., Chingos, M., & McPerson, S. (2009). 456–475.
Crossing the finish line: Completing college at Dundes, L., & Marx, J. (2006). Balancing work and aca-
America’s public universities. Princeton, NJ: demics in college: Why do students working 10 to
Princeton University Press. 19 hours per week excel? Journal of College Student
Bozick, R. (2007). Making it through the first year of Retention: Research, Theory & Practice, 8, 107–120.
college: The role of students’ economic resources, Dynarski, S. (2008). Building the stock of college-
employment, and living arrangements. Sociology educated labor. Journal of Human Resources, 43,
of Education, 80, 261–285. 576–610.
Broton, K., Frank, V., & Goldrick-Rab, S. (2014). Dynarski, S. M. (2003). Does aid matter? Measuring
Safety, security, and college attainment: An inves- the effect of student aid on college attendance and
tigation of undergraduates’ basic needs and insti- completion. The American Economic Review, 93,
tutional response. Madison: Wisconsin HOPE Lab. 279–288.
Castleman, B. L., & Long, B. T. (2013, July). Looking Dynarski, S., & Scott-Clayton, J. (2013). Financial aid
beyond enrollment: The causal effect of need-based policy: Lessons from research. Future of Children,
grants on college access, persistence, and gradua- 23, 67–91.
tion (NBER Working Paper 19306). Cambridge, Ehrenberg, R. G., & Sherman, D. R. (1987).
MA: National Bureau of Economic Research. Employment while in college, academic achieve-
Cha, P., & Patel, R. (2010). Rewarding progress, ment, and postcollege outcomes: A summary of
reducing debt: Early results from the performance- results. The Journal of Human Resources, 22,
based scholarship demonstration in Ohio. New 1–23.
York, NY: MDRC. Goldrick-Rab, S. (2016). Paying the price: College
Cheng, D. X., & Alcántara, L. (2007). Assessing work- costs, financial aid, and the betrayal of the
ing students’ college experiences: A grounded the- American dream. Chicago, IL: University of
ory approach. Assessment & Evaluation in Higher Chicago Press.
Education, 32, 301–311. Goldrick-Rab, S., Harris, D. N., & Trostel, P. A.
Cunningham, A. F., & Santiago, D. A. (2008). Student (2009). Why financial aid matters (or does not) for
aversion to borrowing: Who borrows and who college success: Toward a new interdisciplinary
doesn’t. Washington, DC: Institute for Higher perspective. In J. Smart (Ed.), Higher education:
Education Policy and Excelencia in Education. Handbook of theory and research (Vol. 24, pp. 1–
Dadgar, M. (2012, June). The academic consequences 45). New York, NY: Springer Science & Business.
of employment for students enrolled in commu- Goldrick-Rab, S., Kelchen, R., Harris, N., & Benson,
nity college (CCRC Working Paper No. 46). New J. (2016). Reducing income inequality in higher
York, NY: Community College Research Center at education: Experimental evidence on the impact
Columbia University. of financial aid on college completion. American
Darolia, R. (2014). Working (and studying) day and Journal of Sociology.
night: Heterogeneous effects of working on the Goldrick-Rab, S., & Kendall, N. (2014). Redefining
academic performance of full-time and part-time college affordability: Securing America’s future
students. Economics of Education Review, 38, with a free two year college option. Indianapolis,
38–50. IN: Lumina Foundation.

492
 Working for College

Greene, K. M., & Maggs, J. L. (2015). Revisiting the Miller, C., Binder, M., Harris, V., & Krause, K.
time trade-off hypothesis: Work, organized activi- (2011). Staying on track: Early findings from a
ties, and academics during college. Journal of performance-based scholarship program at the
Youth and Adolescence, 44, 1623–1637. University of New Mexico. New York, NY: MDRC.
Horn, L. J., & Malizio, A. G. (1998). Undergraduates Molitor, C. J., & Leigh, D. E. (2005). In-school work
who work: National Postsecondary Student Aid experience and the returns to two-year and four-
Study 1996. Washington, DC: U.S. Department of year colleges. Economics of Education Review, 24,
Education. 459–468.
Kalenkoski, C. M., & Pabilonia, S. W. (2010). Parental Orszag, J. M., Orszag, P. R., & Whitmore, D. M.
transfers, student achievement, and the labor sup- (2001). Learning and earning: Working in college.
ply of college students. Journal of Population Newton, MA: Upromise.
Economics, 23, 469–496. Osborne, J. W., & Overbay, A. (2004). The power
Kalleberg, A. L. (2011). Good jobs, bad jobs: The rise of outliers (and why researchers should always
of polarized and precarious employment systems in check for them). Practical Assessment, Research
the United States, 1970s to 2000s. New York, NY: & Evaluation, 9(6), 1–12.
Russell Sage Foundation. O’Sullivan, R., & Setzer, R. (2014). A federal work
Kelchen, R., Hosch, B. J., & Goldrick-Rab, S. (2014). study reform agenda to better serve low-income
The costs of college attendance: Trends, variation, students. Washington, DC: Young Invincibles.
and accuracy in institutional living cost allow- Patel, R., & Valenzuela, I. (2013). Moving forward. Early
ances. Madison: Wisconsin HOPE Lab. findings from the performance-based scholarship
Lambert, S. J., Fugiel, P. J., & Henly, J. R. (2014). demonstration in Arizona. New York, NY: MDRC.
Precarious work schedules among early-career Perna, L. (2010). Understanding the working college
employees in the U.S.: A national snapshot. student: New research and its implications for pol-
Chicago, IL: EINET Employment Instability, icy and practice. Sterling, VA: Stylus Publishing.
Family Well-Being, and Social Policy Network at Pope, E. (2010). An overview of student financial aid.
the University of Chicago. Madison: Presentation to the Wisconsin Legislative
Lemieux, T. (2008). The changing nature of wage inequal- Council Study Committee on Student Financial
ity. Journal of Population Economics, 21, 21–48. Aid Programs.
Leuven, E., Oosterbeek, H., & van der Klaauw, B. Presser, H. B., & Ward, B. W. (2011). Nonstandard
(2010). The effect of financial rewards on stu- work schedules over the life course: A first look.
dents’ achievement: Evidence from a randomized Monthly Labor Review, 134(7), 3–16.
experiment. Journal of the European Economic Richburg-Hayes, L., Brock, T., LeBlanc, A., Paxson,
Association, 8, 1243–1265. C. H., Rouse, C. E., & Barrow, L. (2009).
Light, A. (2001). In-school work experience and the Rewarding persistence: Effects of a performance-
returns to schooling. Journal of Labor Economics, based scholarship program for low-income par-
19, 65–93. ents. New York, NY: MDRC.
Lipsey, M. W., Puzio, K., Yun, C., Hebert, M. A., Richburg-Hayes, L., Patel, R., Brock, T., De la Campa,
Steinka-Fry, K., Cole, M. W., . . . Busick, M. D. E., Rudd, T., & Valenzuela, I. (2015). Providing
(2012). Translating the statistical representation more cash for college: Interim findings from the
of the effects of education interventions into more performance-based scholarship demonstration in
readily interpretable forms. Washington, DC: California. New York, NY: MDRC.
Institute of Education Sciences, National Center Riggert, S. C., Boyle, M., Petrosko, J. M., Ash, D., &
for Special Education Research. Rude-Parkins, C. (2006). Student employment and
Lobel, S. A. (1991). Allocation of investment in work higher education: Empiricism and contradiction.
and family roles: Alternative theories and implica- Review of Educational Research, 76, 63–92.
tions for research. The Academy of Management Roksa, J., & Velez, M. (2010). When studying school-
Review, 16, 507–521. ing is not enough: Incorporating employment in
MacDonald, I. H., Malatest, R., Assels, R., Baround, models of educational transitions. Research in
R., Gong, L., Bernstein, L., . . . Greenwood, J. Social Stratification and Mobility, 28, 5–21.
(2009). Foundations for success project: Final Ruhm, C. J. (1997). Is high school employment
impacts report. Victoria, British Columbia, consumption or investment? Journal of Labor
Canada: R.A. Malatest & Associates. Economics, 15, 735–776.
Mayer, A. K., Patel, R., & Gutierrez, M. (2015). Four- Scott-Clayton, J. (2011a). The causal effect of federal
year effects on degree receipt and employment work-study participation: Quasi-experimental evi-
outcomes from a performance-based scholarship dence from West Virginia. Educational Evaluation
program in Ohio. New York, NY: MDRC. and Policy Analysis, 33, 506–527.

493
Broton et al.

Scott-Clayton, J. (2011b). On money and motivation: Van Dyke, R., Little, B., & Callender, C. (2005).
A quasi-experimental analysis of financial incen- Survey of higher education students’ attitudes to
tives for college achievement. Journal of Human debt and term-time working and their impact on
Resources, 46, 614–646. attainment. Bristol, UK: Higher Education Funding
Scott-Clayton, J. (2012). What explains trends in labor Council for England.
supply among U.S. undergraduates? National Tax Walpole, M. (2003). Socioeconomic status and col-
Journal, 65, 181–210. lege: How SES affects college experiences and
Scott-Clayton, J., & Minaya, V. (2014). Should stu- outcomes. The Review of Higher Education, 27,
dent employment be subsidized? Conditional 45–73.
counterfactuals and the outcomes of work-study What Works Clearinghouse. (2014). Procedures and
participation. New York, NY: Center for Analysis standards handbook version 3.0. Washington, DC:
of Postsecondary Education and Employment at U.S. Department of Education.
Columbia University. The Working Student Act. (2014). Available from
Singell, L. D. (2004). Come and stay a while: Does www.baldwin.senate.gov
financial aid effect retention conditioned on enroll-
ment at a large public university? Economics of
Education Review, 23, 459–471. Authors
Soliz, A., & Long, B. T. (2014). The causal effect of Katharine M. Broton is a doctoral candidate in
federal work-study on student outcomes in the Ohio the Department of Sociology at the University of
public university system. New York, NY: Center Wisconsin–Madison. Her areas of research include
for Analysis of Postsecondary Education and education, inequality, and poverty. Her current work
Employment Conference at Columbia University. focuses on the experiences of college students from
Sommo, C., Boynton, M., Collado, H., Diamond, J., low-income and poor families.
Gardenhire, A., Ratledge, A., . . . Weiss, M. J.
(2014). Mapping success: Performance-based Sara Goldrick-Rab is professor of educational
scholarships, student services, and developmental policy studies and sociology at the University of
math at Hillsborough Community College. New Wisconsin–Madison and founding director of the
York, NY: MDRC. Wisconsin HOPE Lab, the nation’s only translational
Stern, D., & Nakata, Y. -F (1991). Paid employment research laboratory aimed at finding effective ways to
among U.S. college students: Trends, effects, and make college affordable. She has directed the
possible causes. The Journal of Higher Education, Wisconsin Scholars Longitudinal Study since 2008.
62, 25–43. Her latest book, Paying the Price: College Costs,
Stinebrickner, R., & Stinebrickner, T. R. (2003). Financial Aid, and the Betrayal of the American
Working during school and academic performance. Dream, will be published by the University of Chicago
Journal of Labor Economics, 21, 473–491. in September 2016.
Turner, S. (2004). Going to college and finishing col-
JAMES BENSON worked on this project as a post-
lege. Explaining different educational outcomes. In
doctoral fellow at the University of Wisconsin-Madison.
M. Hoxby (Ed.), College choices: The economics of
where to go, when to go, and how to pay for it (pp.
13–62). Chicago, IL: University of Chicago Press. Manuscript received August 5, 2015
U.S. Department of Education. (2014). Digest of edu- Revision received December 16, 2015
cation statistics. Washington, DC: Author. Accepted February 7, 2016

494