Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Later life success of former college student-athletes as a function of retirement from sport and participant characteristics
(USC Thesis Other)
Later life success of former college student-athletes as a function of retirement from sport and participant characteristics
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Running Head: SPORTS RETIREMENT OF COLLEGE ATHLETES 1
Later Life Success of Former College Student-athletes as a Function of
Retirement from Sport and Participant Characteristics
Erin D. Shelton
University of Southern California
SPORTS RETIREMENT OF COLLEGE ATHLETES 2
TABLE OF CONTENTS
ABSTRACT………………………………………………………………………………...
4
Chapter
1. Later Life Success of Former College Student-athletes as a Function of
Retirement from Sport and Participant Characteristics………………………....
5
The Academic Success of Student-athletes…………………………………….. 6
Models of Dropout from College………………………………………..... 6
A Conceptual Model of Academic Success for Student-Athletes………... 7
Retirement from Sport…………………………………………………………… 8
Models of Sports Retirement……………………………………………... 9
Athletic Identity vs. Academic Identity…………………………………... 12
Identity foreclosure………………………………………………..... 13
The College to Work Transition and Later Life Success……………………….. 16
The Current Study…………………………………………………………….....
20
2. METHODS…………………………………………………………………........
23
The Creation of the Longitudinal Study of
Student-athletes (LOSSA) Data…………………………………………... 23
Initial eligibility clearinghouse (IEC) and academic performance
Census (APC)……………………………………………………….. 23
Study of college outcomes and recent experiences (SCORE)……… 24
Participants……………………………………………………………….. 25
Measures………………………………………………………………….. 26
Participant and academic characteristic variables………………….. 26
Likelihood of playing professionally and identity variables……….. 27
Retirement from sport and life outcomes…………………………… 27
Analyses…………………………………………………………………... 29
Structural equation models………………………………………….. 29
Path analysis…………………………………………………………
29
3. RESULTS……………………………………………………………………….
30
Result 1: Initial Models to Assess and Adjust for Response Rates
and Bias and Establish the Factor Structure of Identity and Life
Outcomes…………………………………………………………………. 30
Assessing response rates and response bias………………………… 30
Testing the factor structure of the identity variables……………….. 35
Testing the factor structure of the life outcome variables…………... 37
Result 2: SEM to Predict Difficulty with Retirement…………………….. 39
Understanding the outcome variable of difficulty with retirement..... 40
Path analysis for retirement difficulty……………………………..... 42
Result 3: Assessing Retirement Difficulty as a Mediator………………… 47
SPORTS RETIREMENT OF COLLEGE ATHLETES 3
4. DISCUSSION……………………………………………………...………….... 56
Limitations………………………………………………………………... 62
Future Directions………………………………………………………….. 68
Suggestions for the NCAA………………………………………….
71
References…………………………………………………………………………………..
75
Tables…………………………………………………………………………………….....
86
Figures………………………………………………………………………………….......
100
Technical Appendix………………………………………………………………………...
118
Appendix A: SCORE Survey…………………………………………………… 118
Appendix B: Variance and Covariance Defined………………………………... 133
Appendix C: Structural Regression Algorithm………………………………..... 134
Appendix D: Sampling Bias……………………………………………............. 135
Appendix E: Sampling Weights………………………………………………... 136
Appendix F: Logistic Regression Algorithm…………………………………… 138
Appendix G: Pseudo R
2
and the ROC Curve…………………………………… 139
Appendix H: Odds Ratios for Logistic Regression…………………………….. 141
Appendix I: Model Fit Indices………………………………………………… 142
Appendix J: Estimation of Ordinal Variables in SEM………………….............. 144
Appendix K Testing Indirect Effects Using the Sobel Test…………….............. 145
Appendix L: Mplus Code……………………………………………….............. 146
L.1 Retirement Path Model……………………………………………….. 146
L.2 Identity CFA………………………………………………….............. 147
L.3 Life Outcomes CFA…………………………………………………... 148
L.4 Full Model Latent Variable Path Model……………………………... 149
SPORTS RETIREMENT OF COLLEGE ATHLETES 4
ABSTRACT
Existing literature focusing on collegiate student-athletes fails to connect early educational
experiences and college experiences with sport retirement and later life outcomes. The existing
body of work dealing with these issues covers academic success of student-athletes, difficulties
with retirement from sport or making the transition from sports participation for athletes in
general, and the college to work transition for students separately. College student-athletes are a
unique and growing population so expanding the literature about the experiences and outcomes
of this population is warranted. The purpose of this work was to better understand how
participant characteristics impact the retirement from sport process as well as life outcomes. The
sample consisted of N=6,636 former NCAA Division-I student athletes with data from high
school, first year of college, and 10 years after their freshman year of college. Demographics
and academic abilities were obtained from in high school and during the first year of college.
Retrospective accounts of the college experience as well as current job and life satisfaction were
measured approximately 10 years after freshman year when the participants were around an
approximate age of 31 years. Sampling bias was assessed using logistic regression and path
analysis was used for assessing the proposed hypotheses. Results indicated that likelihood of
playing professionally and academic and athletic identity were predictive of sports retirement
difficulties as well as life outcomes. Some of the relationship was mediated through sports
retirement difficulties. The variance accounted for, however, was low. This implies that
retirement difficulties can play a role in life outcomes however a more thorough investigation
into the variables of interest (i.e. sports retirement, life satisfaction, and athletic/academic
identity) is necessary to fully understand the impacts or lack thereof.
SPORTS RETIREMENT OF COLLEGE ATHLETES 5
Later Life Success of Former College Student-athletes as a Function of
Retirement from Sport and Participant Characteristics
Individuals are constantly developing and changing throughout their lifetime.
When looking at the development of an individual, it is often important and helpful to understand
what specific events led them to their current state/situation as well as where they may be headed
in the future (as discussed by Baltes, Reese, & Lipsitt, 1980). A critical juncture may surround
the transition into early adulthood. Early adulthood is a period of development that occurs
directly after adolescence and is defined by personal and economic independence, career
development, and the beginnings of serious intimate relationships and families (Santrock, 2011).
Although college “student-athletes” are purely a USA phenomena, for college-level student-
athletes, this seems to be an important time of development and an area which requires a more
thorough understanding (Wylleman, Alfermann & Lavallee, 2004; Wylleman & Reints, 2010).
Past research is rich in information about academic success of college athletes and
about the transition from sports participation to the working world that athletes endure (see
Comeaux & Harrison, 2011; National Collegiate Athletic Association, 1990; Stambulova,
Alfermann, Stalter, & Cote, 2009). There is also a large body of research looking at the college
to work/career transitions and life satisfaction or success for college students in general but not
for the special population of student-athletes specifically (Diseth, Danielsen, & Samdal, 2012;
Liu, Thomas, & Zhang, 2010, Willits, 1088). Past research on college student-athletes, however,
has failed to make any linkages between these areas or to understand how development during
earlier years impacts later life outcomes or success. In addition, none of this research accounts
for the new demands and the difficulties of what we will term here “sports retirement,” a process
that all college student-athletes inevitably face directly after college a large majority of the time.
SPORTS RETIREMENT OF COLLEGE ATHLETES 6
The purpose of this research was to synthesize the ideas in these different areas of development
and determine if they could be related to each other.
The Academic Success of Student-athletes
The formal monitoring of student-athlete success in college could be said to have
began in the 1980s with changes to the initial eligibility rules by the National Collegiate Athletic
Association (NCAA). From that period on, the NCAA would attempt to gain a better
understanding of how the academic policies put in place impacted their student-athletes (for a
more complete history, see Petr & McArdle, 2012). Several themes came up in the literature
about the factors that affect student-athlete academic success. There is a large body of research
looking at how well high school variables predict college success, namely, high school grade
point average and test scores from standardized tests (see Atkinson & Geiser, 2009; Crouse &
Trusheim, 2988; DeBeard, Spielmans, & Julka, 2004; Hoffman & Lowitzki, 2005). Much prior
research has suggested that higher GPA and higher test scores lead to more success in college
(Chissom & Lanier, 1975; Gussett, 1974; Willerman et al, 1990). This was important to take
into consideration when looking at college academic success because developmental
perspectives suggest that early life events shape later events.
Models of Dropout from College.
In one of the first organizations of its kind, Tinto (1975) suggested we should
view the process of college dropout as a series of interactions that happen over varied periods of
time in college. Individual characteristics as well as academic and social systems in the college
domain shape a student’s experiences which lead to persistence or different types of dropout.
This model attempts to take into account precollege variables and demographic indicators,
addresses different types of student commitment (i.e., their goal commitment and their
SPORTS RETIREMENT OF COLLEGE ATHLETES 7
institutional commitment), looks at the impact of academic system variables (e.g. grade
performance) and social system variables (e.g. peer group interactions), and links later
commitments to different types of dropout (see Figure 1). Tinto’s model focuses first on
individual characteristics (age, sex, ethnicity), educational experiences (GPA, SAT/ACT), and
family backgrounds (social status, values). These variables are said to have direct and indirect
impacts on academic success (Tinto, 1975). Along with these precollege factors come existing
expectations and commitments which are brought with the student-athlete into college and
influence their satisfaction with the college environment. We notice that Tinto did not provide
values for any of his paths to indicate the theoretical strength of these associations (but see
Terenzini & Pascarela, 1980).
Tinto (1975) relied on Durkheim’s (1966) classic concept of suicide when he
suggested that individual integration into the academic and social systems leads to new levels of
commitment. Tinto suggested that the higher one’s integration, the higher their commitment to
the institution and the goal of graduation. He also suggested that lower levels of commitment
lead to a higher likelihood of dropout. Research studies have shown that Tinto’s model is a
reasonable conceptual model to study dropout, except that a more explicit formalization of the
variables was needed (see Terenzini & Pascarella, 1980; Munro, 1981).
A Conceptual Model of Academic Success for Student-athletes.
Comeaux and Harrison (2011) subsequently adapted Tinto’s model to predict
academic success and included variables specific to student-athletes (see Figure 2). This model
was developed to understand the long-term process that leads to academic success or graduation
for student-athletes specifically. Like Tinto’s model, it starts with pre-college variables and
individual characteristics then moves on to commitments. The key difference in Comeaux and
SPORTS RETIREMENT OF COLLEGE ATHLETES 8
Harrison’s model is that in addition to goal commitment and institutional commitment, they
included sport commitment as an important characteristic of the individual. They felt that
previous studies had failed to include sports commitment which they considered an important
characteristic in the processes leading to various levels of academic success. In their model,
sport commitment was indexed by the amount of time and energy expended on sports
participation. Comeaux and Harrison stated that the more commitment an athlete had to their
sport, the less time they would have had to commit to other social and academic related
activities. For this model, sport commitment was seen as an important part of academic success
for student-athletes. Whether it impacted academic success positively or negatively in this
model is not known. Once a student-athlete makes it to graduation or the end of their athletic
eligibility, many are faced with the process of retiring from sport competition. This was a new
conceptualization of Tinto’s (1975) model and Comeaux and Harrison did not actually test the
validity of their model.
Retirement from Sport
Early research on sports retirement related athletic retirement to retirement from a non
athletic working career, and often viewed it as a negative life event, sometimes even a traumatic
event (e.g. analogous to the stages of dying, thanatology, or the study of the aging process, social
gerontology). More modern frameworks have focused on coping processes of student-athletes
leading to positive or negative outcomes (Schlossberg, 1981; Taylor & Ogilvie, 1994) as well as
other frameworks which have focused on a more holistic approach. The holistic approach
frameworks often included larger social factors such as cultural and sports systems factors which
may aid in more positive transitions (Stambulova et. al., 2009). In recent years there has been
increasing interest in looking at this kind of retirement as a developmental process and there has
SPORTS RETIREMENT OF COLLEGE ATHLETES 9
been a large focus on athletic identity as a mitigating factor influencing the quality and ease of
the event/process.(Wylleman & Reints, 2010; Houle, Brewer, & Kluck, 2010; Lally & Kerr,
2005; Lally, 2007; Beamon, 2012).
Models of Sports Retirement
In this area, Taylor and Ogilvie (1994) developed a conceptual model of adaptation to
retirement among athletes. It was an attempt to model the entire retirement from sport process in
college. This model starts with causes of athletic retirement such as age, injury, de-selection or
free choice. These causes lead to factors related to adaptation to retirement which include
developmental experiences as well as changes in self identity. In addition, the causes of
retirement also lead one to consider available resources such as coping skills and social support.
Factors related to adaptation and retirement and the availability of resources lead to the quality of
the athletic retirement experience. The quality of retirement can then lead to either a healthy
career transition, or to a retirement crisis such as substance abuse or social problems which then
would lead to intervention. Again, this was a conceptualization that was not empirically tested
by the researchers here.
Wylleman, Alfermann, and Lavallee (2004), attempted to explain sport retirement from a
life span perspective. To do so, they incorporated a holistic as well as development approach to
look at the process. In their model, there are four levels which occur from the ages of 10 – 35
that are considered important (see Figure 3). The first level considered is the athletic level. This
is defined by initiation of sports participation from early ages through about 12, development
from about 12 to 19, mastery from 20 to 28, and discontinuation from age 29 to just past 35.
This discontinuation can often come sooner for college athletes if they stop competing before
completing their college eligibility. The next level is the psychological level. This is defined by
SPORTS RETIREMENT OF COLLEGE ATHLETES 10
developmental changes and transitions in childhood from early ages to about 11, then on to
adolescence from 12 to about 18, followed by young adulthood and adulthood from 18 to just
past 35. The third level is the psychosocial level which is a representation of the psychosocial
changes that can occur in development relative to a person’s athletic participation. This is
characterized by parents, siblings and peers from early ages to about 13, then on to peers,
coaches, and parents from about 13 to 19, then partner and coach influences from age 20 to 28
and lastly family or coach influences from ages 28 to just past 35. The last level is the academic
vocational level. This is characterized by primary education from early ages to about 12, from
there development moves to secondary education from about 12 to18, from there it moves to
higher education from ages 19 to 23, and lastly it moves to vocational training or professional
occupation from ages 24 to beyond (Wylleman and Reints, 2010). One thing to note from this
model is that the majority of academic and athletic development occurs during the approximate
age range 18-38. During this time athletes are considered in the Mastery stage and then
Discontinuation stage of the athletic level and concurrently in the higher education, vocational
training and professional occupation stages for the academic level. From this we can infer that
the adulthood stages of development are a particularly important time in the lives of student-
athletes. In addition, this second model of retirement highlights the other transition processes
that are happening simultaneously with the transition from collegiate sports participation. It
shows the developmental process of the student-athlete along with the transition or retirement
from sport.
A third model used to explain retirement from sport and the decision making process is
what s termed the “trans-theoretical” model that emerges from health psychology (Park, Tod,
Lavallee, 2012). This model attempts to explain behavior change in an individual in any
SPORTS RETIREMENT OF COLLEGE ATHLETES 11
situation. According to this model, there are an ordered set of five stages of change everyone
must go through: include (1) pre-contemplation, (2) contemplation, (3) preparation, (4) action,
and (5) maintenance. During the pre contemplation stage (1), individuals are not considering any
changes in their current behaviors. During to the contemplation stage (2), individuals become
aware that they need to change a behavior. In the preparation stage (3), the individual is
planning to take action. And during the action (4) and maintenance stages (5), the individual
actually makes the changes and this change lasts for longer than six months. This model is often
applied to behavior change in various fields, and, in the case of the student-athlete, the authors
suggest that this model helps to explain retirement decisions related to sports. This is due to the
fact that retirement from sport can involve maladaptive behavior changes (e.g. narrow focused
daily lives due to sports participation) and the acquisition of adaptive behaviors (e.g. non-athletic
social networking). Park et. al. (2012) did a study looking the sport retirement decision making
process using the transtheoretical model. They examined the process for 12 former Korean elite
tennis players using focus groups. Their results indicated that the sport retirement decision
making process for athletes went through a series of stages similar to the pre-contemplation,
contemplation, preparation, and action stages of the transtheoretical model. This was a
qualitative study rather than a quantitative one so there are no real estimates of the strength of the
relationships between the variables or to evaluate the model and the retirement process.
While the three models presented take different approaches to the understanding of sport
retirement, they all have strong common themes which relate to the specific case of sport
retirement for student-athletes. The first common theme is sport retirement is seen as a process
where several variables impact the retirement transition. In each model there are initial variables
that impact a second set of variables and ultimately lead to the outcome. The next common
SPORTS RETIREMENT OF COLLEGE ATHLETES 12
theme, at least included in the life span perspective and the trans-theoretical perspective, is the
idea that there is a developmental aspect of life that leads to sport retirement outcomes. If we
then look at these models of retirement and compare them to the model of academic success of
student-athletes mentioned earlier, it seems that the theories can fit together nicely. From a
developmental perspective, we know that early life experiences impact academic success as well
as sport retirement, and these are both present in the model at the pre-college level. The athlete
enters college with initial levels of commitment, they are then shaped and molded by the unique
social and academic systems which then lead to the later commitments and ultimately to
academic success and sport retirement. In addition, the empirical evidence for all three models,
however, is largely lacking. These models focused on conceptualizing a new model rather than
testing the new model to see its validity and replicability. The studies focused on important
factors of the transition process but did not have information related to the directionality of the
relationships or how the variables or levels of the process specifically worked together. Despite
the lack of empirical evidence seen in the three models presented, they give insight into what
seem like logical ways to synthesize sport retirement ideas.
Athletic Identity vs. Academic Identity
When studying the research on sports retirement above and beyond models
explaining the behavior, the importance of identity in transition is often discussed. Empirical
research suggests that individuals with a stronger athletic identity have a harder time with
retiring from sport (Beamon, 2012; Webb, Nasco, Riley, & Headrick, 1998). The less a student-
athlete focuses on their academic identity, the more likely they are to have a strong athletic
identity and this can lead to difficulties when trying to make the transition and retire from sports
participation (Miron, 2010). While the relationship between academic identity and sports
SPORTS RETIREMENT OF COLLEGE ATHLETES 13
retirement is not fully understood, some recent work has indicated that more academic identity is
related to more positive academic outcomes and that low academic identity is an indicator of
potential academic trouble (Paskus, 2012). Exactly how much a strong athletic identity impacts
the transition and retirement from sports is also not well known.
Lally and Kerr (2005) did what they termed a “qualitative study” looking at career
planning and of college athlete identities. They used in-depth interviews to get an idea of the
way athletic and academic identities of the student-athletes developed over the course of their
college and athletic careers. These researchers found that not only did the career plans of these
individuals change substantially over the course of their studies, but their identities changed
substantially as well. The students had thoughts about pursuing professional careers upon
entering college; however, by the time of graduation, those thoughts had changed. They only
include eight (8) students in their study, but they stated that the more time these people spent on
activities that strengthened their academic identities, the less exclusive their athletic identities
became; The authors did point out that this may not be the case for all student-athletes. And
lastly, these students mentioned that upon entering their freshmen year of college they were not
sure of what area they wanted to major in. By the time the students were ready to graduate, they
had chosen a major and many had placed more focus on their academic identities and studies in
the latter years which may have led to easier transitions from collegiate sports (see Brown and
Hartley 1998). According to Lally and Ker (2005), students can simultaneously invest time in
their athletic identity and their academic identities and this may occur for a majority of student-
athletes. Results from this study were difficult to generalize though due to the qualitative nature
and the low sample size (N=8). So while the transition may be rather difficult, it seems that most
students can get through it successfully and may be more successful in their future (also see
SPORTS RETIREMENT OF COLLEGE ATHLETES 14
Paskus, 2012). What many fail to talk about are those individuals who do not develop their
academic identities at some point during their collegiate career, and this could lead to additional
problems in the future.
Identity foreclosure. Identity foreclosure refers to a situation where a college
student makes a commitment to an identity before they have had any time to explore other
options. In this situation, any student-athlete may put more emphasis on or exaggerate their
athletic identity, and never really explore or develop other identities such as their academic
identity (Baillie, 1993). Beamon (2012) studied how exclusive athletic identities impact the
transition of athletes. He used identity foreclosure as a theoretical framework. He also found
that an exclusive athletic identity can impact an athlete’s retirement from sports negatively and
can hinder their ability to explore alternative identity options. The environment student-athletes
are in during college can lead to stronger associations with athletic identities as well because
athletes tend to spend the majority of their time with other athletes and this leads to a kind of
social isolation (Reimer, Beal, & Schroeder, 2000). This social isolation can lead to less
academic commitment for athletes and ultimately lead to a detached view of academics while
strengthening one’s athletic identity (Adler & Adler, 1985).
Work by Beamon (2012) and Harrison, Sailes, Rotich, & Bimper (2011) have stated that
identity foreclosure is most prominent among black male athletes because they exaggerate the
role of athletics and focus most on their athletic identity rather than their academic identity. For
many impoverished athletes, playing collegiate sports is viewed as their way out of poverty or as
a tool to improve their social situation (Anderson, 2012; Beamon & Bell, 2006; Eitzen, 2001).
So when the opportunity arises to play in college, they are already clinging to their athletic
identity for survival. The issue of identity foreclosure in this population can be exacerbated
SPORTS RETIREMENT OF COLLEGE ATHLETES 15
during college years because the main focus of the student-athlete, particularly when their sport
is in season, is sports competition resulting in more hours spent on sports related activities and
often less time spent on academics or other activities (Gayles & Hu, 2009; Wolverton, 2008).
Holding on so tightly to the athletic identity also leads many student-athletes, particularly males,
to have aspirations of playing professionally (Beamon & Bell, 2006). The idea that you can
change your life outcomes and social status by participating in college sports and developing
one’s self as an elite athlete can be very appealing (Anderson, 2012; Beamon & Bell, 2006;
Eitzen, 2001).
Unfortunately, as is now clear, only a small percent of college athletes actually advance
to the professional level. According to a report written by the NCAA in 2010, of the college
students participating in sports that have professional leagues, only 9.1% of baseball players,
1.7% of football players, 1.6% of men’s soccer players, and 1.2% of men’s basketball players
actually play professionally. These percentages are even lower for women’s sports with
professional leagues (NCAA, 2010). So for those athletes with an exclusive athletic identity,
that is no academic identity, and who do not get a job in a professional league, the transition into
the work force can be quite difficult and it is possible for athletes to have foreclosed identities
well after college, if not prepared well enough while in college (Beamon, 2012).
In addition to the salience of sports involvement, there are other consequences of
involvement that lead to difficult transitions or retirement from sports (Baillie, 1993). One major
consequence of sports involvement is lack of time to focus on other things not related to sports
such as one’s career options after college, and career maturity can play a big role in retirement
and transition for athletes. Linnemeyer and Brown (2010) did a study looking at the career
maturity and identity foreclosure in student-athletes and the general student body. Linnemeyer
SPORTS RETIREMENT OF COLLEGE ATHLETES 16
and Brown found that student-athletes displayed low career maturity and high identity
foreclosure. This suggests that athletics has some effect on the amount of time student-athletes
are able to spend on career related endeavors (Barber, 2008). Career maturity was defined as a
readiness to make decisions about one’s career by Savickas (1984). This may include their
attendance at career fairs or workshops where they learn to prepare resumes and other interview
etiquette (Steinbach, 2000). An additional factor that plays a role in the difficulty with transition
from sports to the work force is the number of years and hours devoted to one’s sport. Many
athletes start sports at very young ages (Baillie, 1993) and compete from those early years until
the abrupt end in college. When one devotes that much time to something or an aspect of their
life, it becomes a part of who they are and is increasingly difficult to see one’s self as something
else. When one’s athletic identity is especially strong, the abrupt ending which is usually
inevitable in collegiate sports can have negative effects on transition and retirement because they
are forced to reevaluate their situation and ultimately redefine who they are without athletics
(Barber, 2008). This often involves redefining themselves within the workforce. Because of the
conceptual nature of many models, it is difficult to estimate the amount of variance we expect to
account for by sports retirement.
The College to Work Transition and Later Life Success
Research on college student-athletes fades after looking at retirement from competition.
Webb, Nasco, Riley, & Headrick (1998) reported that while athletic identity had an impact on
retirement from sports participation it was not related to overall life satisfaction. Beyond this,
we do not have a clear picture of what happens to the majority of student-athletes after they leave
the college domain and venture out into the working class world. There is a need to understand
the full trajectory of the athlete experience from high school to the years beyond college in order
SPORTS RETIREMENT OF COLLEGE ATHLETES 17
to gauge the impact college sports participation has on life outcomes. What we can focus on
here is the body of research looking at the college to work transition faced by the general student
body and see how it might inform the current research related to student-athlete outcomes.
The majority of work on the transition from college to work and life outcomes has
suggested that there are some links between academic success and life outcomes but there are
limitations to these links. Research suggests that academics may be related to life satisfaction
while in school but is not related to life satisfaction several years after college (see Diseth,
Danielsen, Samdal, 2012; Suldo, Riley, & Shaffer, 2006; Willits1988). For example, Suldo,
Riley, and Shaffer (2066) found associations between perceptions of academic abilities and
overall satisfaction with school. They suggest that children are more likely to view their quality
of life to be high if they feel able to handle school and work and view their school experience to
be positive. This relationship supposedly continues throughout the college years. Singley, Lent,
and Sheu (2010) did a study looking at cognition, academic success and life satisfaction while in
college. Their findings suggest that social cognitive behaviors such as goal progress is related to
life satisfaction in that the more an individual feels they are making adequate progress with
personal goals, the more satisfied they are with life overall. What these two studies don’t
address is the relationship between academics and life success after the transition from school.
That is, they do not address how academic acuity relates to later life success of former athletes,
rather they focus on current life satisfaction while in school. An extension would be trying to
understand if a student athlete’s academic performance in college significantly impacts how
satisfied they are in life afterwards.
When looking at the life outcomes of college student-athletes and the process it
takes to get there, some research suggests that school environments are preparing students for
SPORTS RETIREMENT OF COLLEGE ATHLETES 18
success in later life (Willits, 1988). Willits’ theoretical work is from the developmental lifespan
perspective where adolescence is viewed as an important part of the process of human
development. Within the body of research looking at life success after college, the views are
mixed on whether academics are related to job/career success or satisfaction and global life
satisfaction (Donhardt, 2004; Liu, Thomas, & Zhang, 2010; Willits, 1988). Willits (1988) did an
empirical study where participants were measured when they were sophomores in high school
and again 37 years later. The purpose of his research was to assess whether adolescent academic
performance and other () were linked to personal success and well-being in adulthood. He
looked specifically at occupational prestige, attained income, and expressed feeling s of life
satisfaction. He found that years of formal education was strongly associated with occupational
status for both males and females (r = .654, n=539, p <.001 and r = .542, n=473, p <.001,
respectively) which they ranked using socioeconomic status scores, and education was also the
most important predictor of total family income (β= .24, n=1309, p<.001) indicating that the
more education or higher one’s education the higher one’s occupational status and family
income. Although these were longitudinal data, the reverse causal statement could be true
instead. That is, a higher occupational level could be the result of higher educational attainment,
or both could be the result of something else (i.e., higher levels of cognition).
In this 37-year panel study of over 2,000 people Willits (1988) did not find that
subjective well-being (an index of expressed life satisfaction) was related to educational
attainment. Similarly, Donhardt (2004) found the opposite to be true in his study looking at
earnings outcomes. More specifically, he was interested in the prediction of earning outcomes,
specifically the growth of earnings, for recent college graduates during the first three years after
college. Donhardt found that grade point average was not a significant predictor of earnings
SPORTS RETIREMENT OF COLLEGE ATHLETES 19
after college. It was significant in the prediction of two quarters of earning right after college but
the relationship was negative. Donhardt attributed this to students allocating time between
pursuing studies and entering the workplace. In addition, he looked at differences in earning for
high academic achievers (GPA > 3.495) and low academic achievers (GPA < 2.640). No
significant differences were found for after college earnings the first three years after graduation
for high and low academic achievers. In earlier work, Howes (1981) did a study looking at
academic success, as defined by graduation from a certificate program and graduate program and
career success as measured by promotions, job awards, salary increases etc. The participants in
her study were 96 former doctoral or 6-year certificate students who were followed over a 13-
year period. She found no significant relationship between academic success and career success.
She looked at the relationship further by assessing how test scores (GRE, GMAT, etc.) and
GPAs from college and graduate or certificate programs impacted career success and found no
relationship as well. Unfortunately this sample consists of graduate or advanced students and it is
possible that the results are hard to generalize to the population of undergraduate students (see
Sommer & Sommer, 2002, pp. 235-240)
One issue that arises when looking at research on academics and career success or
life satisfaction is that some of the findings are from many years prior and may not have had the
best operational definitions or measurement tools to accurately assess this relationship. More
recent work by Rubio, Primack, Seitzer, Bryce, Seltzer, & Kapoor (2011) proposes a model of
career success. This model was only developed to look at career success for physician-scientists
but, as stated here, it may have some implications for use with the population of NCAA student-
athletes. The model has three related components: (1) personal factors, (2) organizational
factors, and (3) career success. The personal and organizational components are supposedly
SPORTS RETIREMENT OF COLLEGE ATHLETES 20
correlated in a positive way, and these two factors then lead to career success. Personal factors
include demographics, psychosocial environment (life events, burnout, etc.), education (history,
degrees etc.), and individual personality characteristics (motivation, self efficacy, interest, etc.).
Organizational factors include institutional resources (financial, infrastructure, etc.), training
(research experience, didactic programs), relational factors (mentoring, networking), and
conflicting demands (clinical and service responsibilities). Career success includes extrinsic
success (financial success, promotion, leadership positions) and intrinsic success (job, career,
and life satisfaction).
One concern that might come up when looking at this model is that life
satisfaction is confounded with job success and both are included in a global factor or idea of
career success. This may seem problematic however past research has shown that career success
and job satisfaction are positively related to life satisfaction so it may be a logical combination
(Virick, DaSilva, & Arrington, 2009). While all of these variables do not apply to the case of the
student-athlete specifically, there are several variables present that overlap with the previous
models of academic success and retirement mentioned in earlier sections such as educational
variables and social or relational variables.
The Current Study
Because of the overlapping theories and ideas present when looking at research on
academic success, retirement from sport, and the college to career transition, it seemed that
merging these models to assess the impacts of these variables on college student-athletes would
be a useful extension of the literature. The purpose of the current study is to assess the role of
participant characteristics and early precollege/college variables in predicting later life success of
collegiate athletes while accounting for difficulties in what we term as “sport retirement.” We
SPORTS RETIREMENT OF COLLEGE ATHLETES 21
know we cannot fully evaluate sports retirement as a construct because it is a single item and
models of internal validity are not identifiable in factor analysis frameworks with a single item.
Thus, this study will (1) test how early life variables like demographics and academic
performance predict retirement difficulties (RD) and, (2) test the role RD plays in understanding
multiple life outcomes. This is done by looking at the relationship in terms of a transitional
process/progression where we model how early life variables impact RD while simultaneously
looking at the impact of RD on life outcomes with difficulty in retirement from sports
competition occurring in between (see Figure 4).
The first research question is #1: Do participant characteristics (including
demographics, academics and likelihood of playing professionally) predict retirement
difficulties? Based on past research, it is hypothesized that higher academic performances would
lead to fewer difficulties with retirement. Also, it was hypothesized that the more likely a former
athlete thought it was that they would play professionally, the more retirement difficulties they
would report. As mentioned previously, little is known about what impacts retirement
difficulties for former student-athletes specifically. Understanding the impacts of participant
characteristics on retirement difficulty is a first step to better understanding of the role retirement
might play for athletes
The second research question is concerned with merging all of the variables to get
a better understanding of how they all fit into a more transitional phase model. #2: Do
retirement difficulties mediate the relationship between participant characteristics (including
demographics, academics and likelihood of playing professionally), identity (academic and
athletic), and life outcomes (job and life satisfaction). It is hypothesized that retirement
difficulties will mediate the relationship between academics, likelihood of playing professionally
SPORTS RETIREMENT OF COLLEGE ATHLETES 22
and identity with life outcomes. Based on past research, it is hypothesized that a higher athletic
identity would negatively impact retirement (higher athletic identity would lead to more
retirement difficulties) and result is lower reported life satisfaction. In addition, it is
hypothesized that higher academic performances would predict fewer difficulties with retirement
and ultimately more positive outcomes. Based on the similarities of variables present between
the models mentioned, it is thought that merging aspects of various models would in fact allow
for the assessment of the relationship between participant characteristics, academics, identity,
and later life outcomes for student-athletes while looking at the impact of difficulties with
retirement from sport which occurs in between.
The motivations for studying and understanding the relationship between
precollege/college variables, sport retirement, identity and later life success of former student-
athletes were many. First, research on precollege/college variables and life satisfaction are
limited. Many studies have attempted to look at the relationship as it occurred with adolescents
in school settings. Second, implications from this research are far reaching. Life satisfaction is
seen as an indicator of wellbeing. Understanding the roles that participant characteristics,
identity, and academic variables play in well-being in later life is important. It can give insight
into the ways early academic experiences and college identity development impact an individual
throughout their adulthood. Also, understanding the impacts participant characteristics, early
academic experiences and identity in college have on adult life outcomes and the role that
significant transitions around that time have can help prepare any individual to make better
transitions from college to the working world.
SPORTS RETIREMENT OF COLLEGE ATHLETES 23
CHAPTER 2: METHODS
The creation of the Longitudinal Study of Student-athletes (LOSSA) Data
The data for this study come from three separate data sets which have been
merged into one using a unique set of the college student-athlete participants. The data set
includes information obtained from a large sample of student-athletes (N=24,336) who had
previously participated both while in high school and college in the Initial Eligibility
Clearinghouse (IEC). The student-athletes allowed us to merge their individual academic
outcomes data in college in what was then known as the Academic Performance Census (APC).
An additional dataset collected by the NCAA after HS Graduation was termed the Study of
Current Outcomes and Recent Experiences (SCORE). These new data were merged because they
contain retrospective college information as well as current life outcomes information on the
same college student-athletes approximately 10-12 years after the start of their freshman year.
Initial eligibility clearinghouse (IEC) and academic performance census (APC). The
IEC, now referred to as the NCAA Eligibility Center (NEC), is an entity within the NCAA with
the task of determining whether a prospective student-athlete meets the minimum academic
standards and amateurism criteria (for details, see Hosick, 2010). Their basic task of the IEC is
to determine a college student’s eligibility status for college sports participation. This branch of
the NCAA collects complete high school academic records of over N>180,000 prospective
collegiate athletes every year (Petr & McArdle, 2012). This is not thought to be overly selective
because, in order for a student-athlete to be listed as eligible for collegiate competition their first
year, they must submit official high school transcripts as well as test scores from either the SAT
or ACT. Much of the data that will be used here comes from this process. The IEC database
contains demographic information as well as information on sport, high school GPA, family
SPORTS RETIREMENT OF COLLEGE ATHLETES 24
income, number of units, ineligibility reasons, national test scores (ACT/SAT), and parent
education.
The APC is a database with college level data on Division I student-athletes. These data
were collected between 1994 and 2002 on up to about N>20,000 scholarship student-athletes in
Division I each year (data was also collected on Division II student athletes but the focus here is
on Division I athletes). This database contains information on year the student-athlete entered
college, major, athletic status, best test scores (ACT/SAT; same as in IEC), first year college
GPA and matriculation status at the end of the first year. Table 1 is a list of the IEC and APC
variables used in this study.
Study of college outcomes and recent experiences (SCORE). The SCORE
survey was developed as a longitudinal tool to study the impact of collegiate sports participation
on student-athletes during college as well as after college. It was first administered in 2006 on a
cohort of student-athletes who entered college in 1994. SCORE is still considered in its infancy
now, with the second administration occurring in 2010 on a cohort of former student-athletes
who entered college in 1996. The broad purpose of this survey was to provide NCAA a more
comprehensive view of the development of student-athletes. Table 2 presents more details on the
SCORE sections and variable descriptions used.
The SCORE data consists primarily of former college student-athletes who could
have graduated from high school in 1996; however, former athletes who were known to play
Division I sports beginning in 1996 as well as former student-athletes who were recruited to play
collegiate sports but were lost to follow up after high school were also included. The NCAA
research team used the last known addresses listed in addition to a computer service
(ACCURINT) specializing in finding current address of individuals to locate the former student-
SPORTS RETIREMENT OF COLLEGE ATHLETES 25
athletes for this study. This sampling method is a possible source of bias in the sample. Because
some last know addresses of the population may not be available or were not correct, it changes
the odds of selection of participants from the population. The NCAA research staff sent out pre-
notification post cards to approximately 25,000 former student-athletes. The post card included
an NCAA web address in addition to a unique alphanumeric code for participants to access the
survey online. Approximately four weeks after the post card was sent, NCAA sent out paper
versions of SCORE to all individuals who did not complete the survey online. The NCAA only
provided SCORE t-shirts for participation in this survey, and a selected sample of 5,000
individuals who were identified as at-risk for response based on results from the previous
SCORE received a pre-incentive pen with their survey. This sampling plan was developed by
researchers in the Data Analysis Research Network at NCAA (a consultative panel of about 20
education/psychology researchers).
Participants
The LOSSA data contains IEC/APC information on 24,336 former athletes and
IEC/APC and SCORE information for approximately 6,636 (27% of total subject pool) former
athletes who were tested approximately 10 years after high school. IEC data is obtained for
every student athlete who enters college to play sports. APC data was obtained from over 20,000
of those individuals during their first year in college. A subject pool of just over 24,000
participants with IEC and APC data were selected to receive the SCORE survey. Of those who
received the SCORE survey or who were in the subject pool, 6,636 responded to SCORE. In the
larger subject pool, the majority of participants were male (62.4%; 37.6% female), and White
(51%; 23.4% Minority or Non-White, 25.6% missing). For participants who responded to
SCORE (n=6,636), the sample had an almost even number of males and females (49% male and
SPORTS RETIREMENT OF COLLEGE ATHLETES 26
51% female). In addition, the majority of participants in the SCORE sample were white (59%.
15% minority or non-white, 26% missing). Lastly, the participants were all approximately 31
years of age at the time of SCORE testing. Sampling biases will be assessed using a variety of
methods (to be discussed).
Measures
Based on the models discussed in the previous sections, several variables were
selected to test the hypotheses proposed.
Participant and academic characteristic variables. Pre-college variables in the
conceptual model included demographics as well as academic variables. The pre-college
variables in our models included college academics as well. Family median income was used for
family background demographic information; it was measured on a continuous scale. Gender
was used in addition to Ethnicity for demographics. Educational outcomes was represented by
high school core GPA and the available (best) SAT or ACT test scores in addition to core college
GPA. These were used instead of whether they graduated or not because the majority of the
sample graduated from college (over 90%). In addition, a sport variable was included as well as
a variable to assess whether the former athlete played professionally or not. Sport obtained from
the IEC records where participants indicated the sport they intended to play upon entering
college. This sport variable was re-coded to include nine categories for easier use: (1) Football,
(2) Track, (3) Soccer, (4) Swimming, (5) Basketball, (6) Baseball, (7) Softball, (8) Volleyball
and (9) Other. These categories were determined by looking at frequencies of all three letter
sport codes on the original variable and re-categorizing based on sports with the highest
percentages of participants. All remaining participants were categorized as other. The
professional variable asked if at any point since leaving college, the participant competed at a
SPORTS RETIREMENT OF COLLEGE ATHLETES 27
professional and/or Olympic level. This variable was a dichotomous variable with 0 = “No” and
1 = “Yes”. These two variables were used in the first results section to assess bias and possible
group differences. Table 1 includes a list of these variables.
Likelihood of Playing Professionally and Identity Variables. Several
retrospective variables were used to understand how identity impacts the lives of student-
athletes. The variables used assessed at athletic as well as academic identity. The first variable
(Pq10) which was used in this section asked how likely the participant thought it was that they
would become a professional athlete upon entering college. This was measured on a 6 point
Likert type scale from 1= “very unlikely” to 6= “very likely”. This set of questions (Pq15a
through Pq15l) concerned various aspects of athletic and academic identity and were all
measured on a 6 point liker type scale from 1= “strongly disagree” to 6= “strongly agree”. The
first two variables (Pq15a and Pq15b) asked whether the participant considered them self a
dedicated athlete and student. The next two items (Pq15c and Pq15d) asked whether the
participant had many personal goals related to athletics or academics. The next items (Pq15e and
Pq15f) were concerned with whether the participant felt the need to excel in athletic or academic
pursuits to feel good about them self. The next items (Pq15g and Pq15h) asked whether the
athletic and academic experiences were important to their overall college experience. The next
two questions (Pq15i and Pq15j) asked whether the participant felt other students or professors
viewed them as more of an athlete than a student. And lastly, the final question (Pq15k and
Pq15l) asked whether the participant felt that professors discriminated or favored them because
they were athletes (these are detailed in Table 3 and Appendix A).
Retirement from sport and life outcomes. Sport retirement and life outcomes
variables were integrated into the Comeaux and Harrison model to produce a transitional model
SPORTS RETIREMENT OF COLLEGE ATHLETES 28
of development and transitions of student-athletes from college to post college. The variables
included for this portion of the model came from SCORE. It may be important to note that the
variables used were all self report measures and several were taken from the Monitoring the
Future survey (MTF; see University of Michigan, 2013). Retirement from Sport (Pq26)
measured difficulty with retiring from sports competition retrospectively. It was measured on a
Likert-type scale of 1=”not difficult at all” to 5=”very difficult.” Life outcomes were measured
by several variables in SCORE that were grouped together to address job satisfaction and life
satisfaction. There were six questions related to job satisfaction and all were measured on a five
point Likert type scale of 1=”dissatisfied” to 6=”satisfied.” The first variable (Pq51a) asked how
satisfied or dissatisfied the former athlete was with their current pay. The next question (Pq51b)
asked how satisfied or dissatisfied they were with the benefit package for their current job. The
third item (Pq51c) asked how satisfied or dissatisfied the participant was with the importance and
challenge of the work they do. The fourth variable (Pq51d) was concerned with satisfaction with
promotion and advancement opportunities at one’s place of employment. The fifth item (Pq51e)
asked how satisfied or dissatisfied they were with opportunities to use their past training or
education. And lastly, the sixth variable (Pq51f) looked at whether they were satisfied or
dissatisfied with their job security. See Table 4 for a list of variables. There were nine questions
related to life satisfaction and all were measured on a six point Likert type scale of 1=”
completely dissatisfied” to 6=”completely satisfied.” The first variable (Pq58a) asked about
satisfaction with “life as a whole.” The second variable (Pq58b) asked about “job or career”
satisfaction. The third variable (Pq58c) was concerned with satisfaction with the “current
neighborhood” in which they resided. The fourth variable (Pq58d) asked about satisfaction with
their “education level.” The fifth variable (Pq58e) asked about satisfaction with “friendships.”
SPORTS RETIREMENT OF COLLEGE ATHLETES 29
The sixth variable (Pq58f) was concerned with satisfaction with “romantic relationships or
marriage.” The seventh variable (Pq58g) asked about satisfaction with “standard of living like
housing, car, or furniture.” The eighth variable (Pq58h) asked about satisfaction with “how the
participant spends their leisurely time.” The ninth and last variable (Pq58i) asked about
satisfaction about their “spiritual or religious life” (see Table 4, and Appendix A).
Analyses
Structural equation models. The main analyses used were those fitting under
the structural equation modeling (SEM) umbrella of statistical data analytic techniques. These
models are all based on the variance and covariance associated with variables in a model
(sometimes means but not typically). More specifically, SEM modeling involves decomposing
the covariance of variables in a model. The variance of a variable describes how scores are
dispersed or spread around the mean of a variable and the covariance represents the relationship
between items or variables in unstandardized form, as opposed to a correlation which is the
association between two items in standardized form (see Appendix B for algorithms).
The broader analysis used is termed path analysis and the second main analysis is referred
to as factor analysis. Both are forms of SEM and both can be used as separate techniques or
combined into a structural regression model (Rakov & Marcoulides, 2006) and used
simultaneously depending on the research question being investigated.
Path analysis. Typically, models compared using path analysis techniques are
concerned with the relationship between observed indicators. It is a way to test theoretical
relationships between variables (see Babin & Svensson, 2012; MacCallum & Austin, 2000;
McArdle & Prescott, 1992; Schumacker and Lomax, 1996). Path analysis uses a series of
regressions and correlations to understand this relationship, however, SEM models have an
SPORTS RETIREMENT OF COLLEGE ATHLETES 30
advantage over regular regression models in that they provide a way to take into account any
measurement error which may be present in the observed variables (Rakov and Marcoulides,
2006). It can be a powerful tool to test theories between variables when a priori hypotheses exist
(see; Schumacker & Lomax, 1996). The possibilities as far as model specification in path
analysis are many. With a three variable model, again with those variables being measured or
observed rather than latent or hypothetical, there are multiple possibilities for model
specifications. It is possible in a three variable model to have one outcome variable
(endogenous) and two predictor variables (exogenous) or one exogenous predictor and two
endogenous outcome variables. It is also possible for correlations between variables and the
estimation of means which are all things that could be considered here (see Kline, 2011; see
Appendix C).
CHAPTER 3: RESULTS
Result 1: Initial Models to Assess and Adjust for Response Rates and Bias and Establish
the Factor Structure of Identity and Life outcomes
The analyses in this section are initial analyses to assess possible issues in the data
and adjust for them. These analyses don’t actually test the proposed hypotheses, but rather are
and initial step to prepare for the following analyses which test the proposed hypotheses.
Assessing response rates and response bias. Before analyzing models using retirement
from sport, potential biases in the data needed to be accounted for and adjustments needed to be
made to the data to correct for any biases present. Biases were assessed using two different
methods. First, response or sampling bias was assessed (see Appendix D; McArdle, Petr,
Paskus, Kearns, 2007; Stapleton, 2002; United States Census Bureau, 2012). Following
McArdle, et al (2007), we looked at the response of individuals in the sample to determine if the
SPORTS RETIREMENT OF COLLEGE ATHLETES 31
participants who responded to SCORE from the subject pool of participants who received the
SCORE survey were representative of the subject pool. Using the available Race/Ethnicity to
assess the biases in response was problematic because there were a number of participants with
missing values for Race across all data sets, so in addition to Race, Sport and Gender were used.
Next prediction bias was assessed (see McArdle et. al., 2007). That is, we assessed whether key
variables, mainly demographics, were predictive of the response (0, 1) to the SCORE survey. If
demographics predicted response to the SCORE survey, it would indicate a bias in the sampling.
For example, if we had tried to predict response to SCORE using logistic regression, where 0=
“did not respond”, and 1= “responded”, and a variable such a female predicted whether or not the
individuals responded, it could indicate that there is a slight bias in the sample towards females.
In this way statistical adjustments would need to be made to adjust for the bias in the sample so
that results are more representative of the general population from which respondents came.
For example, in table 5, the number of blacks in the total sample of possible participants
was 5,410 (n
i
). Of that group, 797 responded to SCORE (n
ir
), so the response rate (RR) was
15% (797/5,410; see Appendix D). We can contrast this rate with the response rate of the total
sample, which was approximately 27%, and see that blacks (those of African-American
backgrounds) appear to be underrepresented in the available sample. Ideally, for each group we
are concerned with, a response rate similar to that of proportion of the group present in the total
sample population. While this is preferred, it seems unlikely with this type of study and data
collection method. The reason this seems unreasonable is because potential participants cannot
be forced to participate and there are other barriers that may be present that hinder response from
certain groups. To start, some addresses were incorrect at the start and it is hard to tell which
participants did not respond due to incorrect address versus some other reason. If the last known
SPORTS RETIREMENT OF COLLEGE ATHLETES 32
address of the participant was incorrect, or the most recent address was not located or incorrect,
the participant would not receive the survey. It could be that former athletes who had more
positive experiences were inclined to respond as well. Because participants cannot be forced to
respond, it is hard to insure adequate and equivalent response rates across groups prior to survey
administration.
When looking at Table 5, notice that blacks had the lowest response rate and whites had
the highest. Also note that females had higher response rates than males (36% compared to
21%), which indicates a possible overrepresentation of females in the sample. Looking closer at
male and female differences by race, we also noticed that black males had a very low response
rate compared to other male groups (13% compared to 19% - 25%) and black females tended to
have lower response rates when compared to other female groups (21% compared to 31%-39%).
Males, regardless of race, had lower response rates than the total response rate and females, with
the exception of black females, had higher response rates than the total group response rate.
Looking at response rate based on race specifically was problematic in this case. This was due to
the fact that the race variable in the data had a large amount of missing values so response rates
were likely not highly accurate and the responses were not likely representative of the actual
population. For this reason, response rates were also assessed using sport and gender (see Table
6).
Originally, the sport variable in the data had several possible codes. For the purposes of
this work, it was recorded into 9 categories (football, track, soccer, swimming, basketball,
baseball, volleyball, and other). These categories were chosen because they were the most
popular sports and generally had more participants represented, for example, football (n=5,976)
versus a fencing (n=39). When looking at Table 6, notice that men’s basketball had a very low
SPORTS RETIREMENT OF COLLEGE ATHLETES 33
response rate followed by football. All other sports had response rates above 22% with the
exception of those two sports. Again, this indicated a bias in the sample where individuals
representing those sports were underrepresented in the data. For females, the opposite was
occurred. All female sports had response rates over 27%, which was the total group response
rate, and women’s swimming and volleyball had the highest response rates that were both over
40%. This indicates that not only were females overrepresented in the population, but that two
groups of females specifically were overrepresented. This could be problematic and could
potentially lead to results that are not generalizable to the population of interest.
The next step when looking at biases in the sample was to come up with sampling
weights that could be used in future analyses with this population. Sampling weights can be
used to overemphasize data used from groups which are underrepresented in a sample and
underemphasize data from groups which are overrepresented in the sample. For example, in our
data, the sampling weights would place more importance on data from males, and specifically
male football and basketball players while placing less emphasis on data from females and
particularly female swimmers and volleyball players. This would essentially even out the sample
and act in a way that would make the data appear to be representative of the population from
which it was sampled (see Table 7 and Appendix E).
In addition to looking at sampling bias we looked at biases related to response to
SCORE using logistic regression analysis (see McArdle, 1998). Essentially, this analysis was
used to determine if participants who responded to SCORE were random or if specific
characteristics resulted in response for one group over another. If participant characteristics were
predictive of response, it could mean biases existed in the data. Logistic regression analysis (see
McArdle & Hamagami, 1994) is used to predict dichotomous outcomes such as presence or
SPORTS RETIREMENT OF COLLEGE ATHLETES 34
absence of some specific disease or, as with the current analyses, group membership. This
method is often preferred over Discriminant Analysis based methods because it is less restrictive
and much more flexible than other methods that answer the same questions (Tabachnick &
Fidell, 2007). The logistic regression procedure used here mimicked the prediction of attrition
analyses done by McArdle et al. (2007; see Appendix F). Demographic variables were used in
this analysis and included minority status (1=”minority” or non-white , 0=”white”), gender
(1=male, 0=female), income, high school core grade point average (GPA), core college GPA,
and best test score from SAT and ACT (see Table 1). The best test variable was rescaled from a
400-1600 point scale to a 0-5 point scale in order to compare it more easily to the GPA variables
that are also on a scale from 0-5.
Two logistic regression models were compared. The first logistic regression
model, Model 1, had 4 significant predictors, minority (z=-.104, p<.001), sex (z=-.129, p<.001),
high school core GPA (z=.204, p<.001), and test scores (z=.151, p=.002). In this model, whites
were more likely to respond than minorities, females were more likely to respond than males and
participants with higher GPAs and test scores were more likely to respond than those with lower
scores. The sample size for this model was n=13,569 (because of listwise deletion of cases;
L
2
=878, df=6). The first model accounted for less than 10 percent of the variation in response to
SCORE, R
2
=.092 (see Appendix G). The C statistic for this model indicated that it performed
slightly better than what would be expected by random classification, C=.660 (see Appendix G).
Model 2 was much the same as Model 1; however, the sampling weights created previously were
added to see if using that adjustment made a difference in the results. The model results were
very similar to model 1 with only slight changes (n=13,598, L
2
=971, df=6). Model 2 had the
same 4 significant predictors as Model 1 (race, gender, high school GPA, and test scores) and the
SPORTS RETIREMENT OF COLLEGE ATHLETES 35
interpretations were the same as well. In Model 2, the estimates for high school GPA, race, and
test scores went up slightly (by .022, .007, .002, respectively), while the estimate for gender went
down slightly (by .002). For both models, high school GPA was the best predictor of response,
followed by gender, race, and test scores. This was determined by looking at the descending
order of the standardized beta weights, with higher values indicating better predictors of the
outcome or response. Further interpretation of the predictors was obtained by looking at the
odds ratios of the predictor variables in the model (see Table 8; see Appendix H). The odds
ratios for the predictor variables were nearly identical for Models 1 and 2; consequently, the
interpretation of the odds for both models was identical. The confidence intervals for high
school GPA, race, gender, and test scores indicated significant odds (see Table 8). The
interpretations of the odds for continuous predictors were as follows, for a 1 unit increase in high
school GPA (i.e. a change from a 2.0 to a 3.0 GPA) the odds of responding to SCORE increased
by approximately 1.8, for a 1 unit increase in test scores (note: this is rescaled to 1-5 scale like
GPA) the odds of responding to SCORE increased by approximately 1.1. The interpretations for
dichotomous variables were as follows, white participants (students of Caucasian backgrounds)
were approximately .67 times more likely to respond than minorities participants (participants
who reported an ethnicity other than White/Caucasian), and females were approximately .62
times more likely to respond to SCORE than males. The interpretations of the odds were the
same as the original model with a bit more information about the magnitude of change.
Testing the factor structure of the identity variables. Factor analysis is a SEM
method used to determine if a set of measured variables are representative of an underlying
unmeasured factor or latent variable (see McArdle, 1996; Kline, 2011). For this section, we
looked at the variables included for academic/athletic identity to determine if they fit well
SPORTS RETIREMENT OF COLLEGE ATHLETES 36
together and whether they were representative of one or more factors. Mplus statistical software
package were used to run the analyses for this section (Muthén & Muthén, 2011). As with
previous SEM models, model fit was assessed using the likelihood ratio (L
2
) statistic, the
comparative fit index (CFI), the Tucker-Lewis index (TLI), and the root mean squared error of
approximation (RMSEA). Alternative models were explored and compared to a one-factor model
to see how the variables were best represented in the data.
The first model (Model 1.1a) for identity was a one factor model loaded with all
12 identity variables. This model provided a very poor fit to the data, L
2
= 20,973, df=54,
CFI=.276, TLI=.115, RMSEA=.266. This poor fit means that the variables in the model were
representative of more than one factor. All subsequent models will be exploration to find the
best model of identity (i.e. exploring different confirmatory models, not using exploratory factor
models). The first exploratory model, Model 1.1b, split the variables into two factors with
variables related to academics on the first factor and variables related to athletics on the second
factor. For the second model (Model 1.1b) the academic factor had four indicators and the
athletic factor had eight indicators. This model also provided a very poor fit to the data, with L
2
=
9,989, df=53, CFI=.656, TLI=.572, RMSEA=.185. Again, this was likely due to a
misrepresentation in the structure of the factors. When looking at the academic variable, the last
four variables fit poorly. These variables were all related to other’s views of the participant as an
athlete. For that reason, the third model (Model 1.1c) was a three factor model with four
indicators on each factor and the factors represented academic identity, athletic identity and how
much others viewed the participant as an athlete. This model provided a better fit than the
previous two models, with L
2
= 4,248, df=51, CFI=.855, TLI=.812, RMSEA=.122; however, it
still was not quite adequate.
SPORTS RETIREMENT OF COLLEGE ATHLETES 37
To determine if any single variables were contributing more to the misfit, the residuals
were explored. Of the variables on each factor, the variable dealing with excelling in sports for
self esteem (q15e) had a residual value that was not consistent with other items on the second
factor (it was higher than others). For that reason, it was removed from the model. The fourth
model, Model 3.1d, had three factors, two of which had four indicators and the athletic identity
which was reduced to three indicators. This fourth model fit the data better, with L
2
= 1,893,
df=41, CFI=.927, TLI=.901, RMSEA=.091. Residuals were assessed once more to determine if
the removal of any variables might be warranted, however, the values for the remaining variables
seemed adequate and thus no further models were assessed in this regard.
As with previous results, model invariance tests were run to determine whether
the final model for identity (Model 1.1d) was equivalent for Males and Females (Model 1.1e) as
well as Whites and Minorities (Model 1.1f). Both models had an average fit with the strictest
invariance constraints (equivalence of all loadings, means, intercepts, factor correlations, and
residuals across groups; see Table 9). This indicates that the factor structures of the identity
variables are equivalent for males and females as well as whites and minorities.
Testing the Factor Structure of the Life Outcome variables. In this section, we
looked at the variables included for life outcomes to see if they fit well together and were
representative of one or more factors. Again, the Mplus statistical software package was used to
run the analyses for this section (Muthén & Muthén, 2011). The same test statistics were used to
examine the goodness of fit. The first model was a one factor confirmatory model using variables
in Table 4 to model life outcomes, and alternative models were also explored and compared to
the first model to see how the variables were best represented.
SPORTS RETIREMENT OF COLLEGE ATHLETES 38
The first model run for life outcomes (Model 1.2a) was a one factor model with
15 variables. This model did not fit the data well, L
2
= 6,259, df=90, CFI=.704, TLI=.655,
RMSEA=.11. The next model fit (Model 1.2b) was a two factor model. The variables used in
the first model were asked as two separate questions in the SCORE survey. Model 1.2b will split
the variables into two factors of job satisfaction and life satisfaction based on how the items were
used in the SCORE survey. Factor one was job satisfaction and had six indicators and factor two
was life satisfaction and had nine initial indicators. This model fit the data better than Model
3.2a, L
2
= 3,934, df=89, CFI=.816, TLI=.782, RMSEA=.087. Model 3.2b was significantly better
than Model 3.2a, with δL
2
=2,325, δdf=1, p <.001, although, the overall fit was less than ideal.
To get a better idea of which variables might be contributing to misfit in the model, residuals of
each variable were considered. Residuals are the amount of variation in the observed variable
that is not explained by the latent factor or that is due to measurement error (Klein, 2011;
Raykov & Marcoulides, 2006). Looking at the normalized residuals for the indicator means we
noticed that for the first factor of job satisfaction, the normalized residuals were all similar and
small, however, for the second factor of life satisfaction, the variables dealing with job/career
satisfaction (q58b) and romantic relationships (q58f) had residuals that were not similar to other
variables and were higher indicating misfit (Hatcher, 1994). For this reason, those variables
were removed from the factor to see if the model was better represented without them. In
addition, because the first factor was concerned with job satisfaction and one of the variables
removed from the second factor is concerned with job or career satisfaction (q58b), it was kept in
the model as an indicator of the first factor dealing with job satisfaction. So, the third model
(Model 1.2c) now has two factors both of which have seven indicators. This third model fit the
data significantly better than the previous model, with L
2
= 2,283, df=76, CFI=.886, TLI=.863,
SPORTS RETIREMENT OF COLLEGE ATHLETES 39
RMSEA=.071, δL
2
=1,651, δdf=13, p <.001. . The normalized residuals were assessed once
again to see if any remaining variables were causing more misfit than others. The residuals for
variables on each factor were consistent with each other with the exception of the variable
dealing with importance and challenge of one’s job (q51c) on factor one. For this reason, in the
next model (Model 1.2d), the job importance variable was removed. Model 1.2d provided the
best fit to the data, L
2
= 1,893, df=64, CFI=.891, TLI=.867, RMSEA=.071. In addition, Model
1.2d fit the data significantly better than Model 3.2c, δχ
2
=390, δdf=12, p <.001. The normalized
residuals for the variable means were all consistent with each other on their respective factor and
no other variables were removed from the mode. The final model (see Figure 5) was a two-
factor model of life outcomes with a factor of job satisfaction and a factor of life satisfaction.
Job satisfaction has 6 indicators and life satisfaction has 7 indicators. The two factors were
correlated at r= .63 with each other (see Table 9).
Lastly, multi-group invariance models were run to assess any Gender or Ethnic
differences in the factor structure of identity and life satisfaction. The multi-group invariance
models were based on the structure used in Model 1.2d, with two factors, one which had six
variables and one which had seven. One model (Model 1.2e) was run to look at gender
invariance, and a second (Model 1.2f) was run to look a race invariance. Both models had
average fit with the strictest invariance constraints imposed (see Table 9). This indicates that the
factor structures of job and life satisfaction are equivalent for males and females as well as
whites and minorities.
Result 2: SEM to Predict Difficulty with Retirement
The variable of interest in this research was difficulty retiring from competing
seriously in one’s sport. Again, this variable was measured on a scale of 1=”not difficult at all”
SPORTS RETIREMENT OF COLLEGE ATHLETES 40
to 5 = “very difficult”. The purpose of this section was to try and get a better understanding of
whether any demographic or pre-college variables could predict later difficulties with retirement
from sport. This was assessed using path analysis (Kline, 2007; Schumacker & Lomax, 1997;
Raykov & Marcoulides, 2006). Several alternative models were compared. Path Models were
run using Mplus statistical software (Muthén & Muthén, 2011). When appropriate, model fit
was assessed following suggestions by Begozzi & Ye (2012) and using what they describe as the
recommended practical fit indexes. These indices they describe are the chi square likelihood
ratio (L
2
), root mean squared error of approximation (RMSEA), the Tucker-Lewis index (TLI;
also known as the non-normed fit index; NNFI), The comparative fit index (CFI), and
standardized or weighted root mean squared residual (SRMR/WRMR; see Appendix I). These
indices are all used because the use of a single index for model fit is not adequate and together
they are thought to collectively provide satisfactory criteria for model evaluation (Bagozzi & Ye,
2012). Descriptive statistics of the retirement difficulty outcome variable were assessed first.
These descriptives were followed by the fitting of a base model using the demographic predictors
included in the first result section. Lastly alternative models were explored that included
different variables and looked at group differences using multi group invariance tests.
Understanding the outcome variable of difficulty with retirement. When
looking at retirement difficulty for males and females, we noticed that there were some slight
visual differences (see Figure 6). Females had difficulties with retirement but males appeared to
report more difficulty than females. The mean difference of retirement difficulty for males and
females was tested using Yuen’s method for trimmed means as outlined by Wilcox (2012).
According to the test of trimmed mean differences, Males reported significantly more difficulty
with retirement (M=3.6, sd=1.2) than females (M=3.2, sd=1.3), Ty=7.6, p<.001. When this
SPORTS RETIREMENT OF COLLEGE ATHLETES 41
relationship was broken down further and retirement difficulties were assessed by gender and
sport, we noticed that the differences for retirement difficulty within gender for each sport were
very minimal (see Figures 7 and 8). When looking at the retirement from sport variable along
with other job related variables in the SCORE data, we found that several participants were still
working in jobs that were related to athletics or sports in some way. So, were individuals who
were retired from college sports competition but who had current jobs related to sports or
athletics really “retired from sports”?
In the SCORE data, about 30% of participants who responded to SCORE (n=5,975, 90%
of those who responded; missing = 661 or 10%) reported that their current job involved sports or
athletics. When broken down further and assessed separately by gender, we found that slightly
more females reported having jobs involving sports or athletic when compared to males (32%
and 29%, respectively). So the percent of males and females with sports related jobs in later life
was proportional but the question that arose was should these individuals be classified as “retired
from sport”? For these analyses, we included all former athletes who had never played
professionally in our study and treated them as retired athletes. In addition, approximately 14%
of respondents (10% male, 4% female) reported that they had competed at a professional and/or
Olympic level. When we compared the histograms for these two groups (see Figure 9) we
noticed that non professional athletes had a somewhat difficult time retiring from sports
competition while professional athletes reported having much more difficulty retiring from
sports competition (see Figure 10 for kernel density plots of this relationship; see Wilcox, 2012.
It is likely that the time passed since retirement for the individuals who responded is different for
those who played professionally and those who did not. This is partially due to the fact that
NCAA researchers did not clarify whether they wanted participants to report difficulty with
SPORTS RETIREMENT OF COLLEGE ATHLETES 42
retirement from collegiate participation and professional participation separately, and thus the
types of retirement are confounded in this one variable. In addition, respondents who thought
they were very likely to play professionally reported having a harder time with retirement from
sport than those athletes who reported they were unlikely to play professionally (see Figure 11).
Because of the differences seen for sport retirement for gender, multi-group invariance tests were
run to see if model parameters predicting retirement were invariant across groups. Because we
believed that the participants who played professionally were different from those who did not,
they were removed from all subsequent analyses. All subsequent analyses focused on those
student-athletes whose competitive career likely ended with college competition (i.e., they did
not play professionally at any point after leaving college).
Path analysis for retirement difficulty. The analyses in this section started with
base models using the same 6 variables that were used in the previous section to address
sampling bias issues. Several alternative models were also run and compared based on results
from the first two models. Lastly, multi-group invariance models were run on the final path
model to determine if the model could be represented the same for Males and Females, as well as
Whites and Minorities.
The first two models, Model 2.1a and Model 2.1b, had 6 exogenous variables, high
school GPA, test scores, race, gender, income and core college GPA, in addition to one
endogenous variable, difficulty retiring from competing (see Kline, 2011). The outcome variable
was specified as a categorical variable in the model syntax because it had five levels (1=”not
difficult at all,” 2= “not too difficult,” 3=”somewhat difficult,” 4=”difficult,” and 5=”very
difficult”). This implies it is an ordered categorical or ordinal outcome (see Joreskog, 1994;
Muthen, 1984; see Appendix J).
SPORTS RETIREMENT OF COLLEGE ATHLETES 43
In Model 2.1a there was a significant positive path from gender and significant
negative paths from Race, Test scores and self reported Income (see Table 10 and Figure12).
More specifically, Males in the survey seemed to have a harder time with retirement from sports
competition as did minorities and individuals with lower test scores and lower income. Of the
exogenous variables in the model, test score was the most important and had a negative
association (β=-.08, S.E. = .04). Model 2.1b was identical to Model 2.1a but included sampling
weights to account for the biases present in the sample. Adding the sampling weights did result
in slight changes to the variable parameter estimates, but not as much as expected. Specifically,
high school GPA became an important negative path along with the same relationship from the
other four from Model 2.1a: gender, race, test scores and income (see Table 10 and Figure 13).
The estimates for the significant paths in Model 2.1b increased from those in Model 2.1a as well.
Again, Males, Whites, and participants with lower test scores and lower family income when
growing up reported more difficulty with retirement.
In addition, participants with lower high school core GPAs had more difficulty with
retirement. The variables in these two models only accounted for about 3-4% of the variance in
the outcome variable (Model 2.1a β=.03 and Model 2.1b β =.04) indicating that there were other
variables not used in this analysis which account for the observed variance in retirement
difficulties. Although the use of sampling weights altered the results slightly between models,
they were not used in subsequent analyses because the changes in the results were very marginal,
only increasing the variance accounted for in the outcome by half of a percent. This could be
due to the fact that the group differences are not severe enough or large enough to make a
difference or it could mean that the variables used are not adequate in predicting response
resulting in weights that are not effective.
SPORTS RETIREMENT OF COLLEGE ATHLETES 44
The next set of models run were multi-group models which tested the invariance of the
path model across gender (Males vs. females) and race (minorities vs. whites). Group invariance
models started with the most restrictive forms. Constraints were relaxed for models in which the
strictest form of invariance was not suitable, resulting in poor model fit.
The parameter estimates from each exogenous (predictor) variable to the endogenous
(outcome) variable were forced to be equivalent across groups. In addition, the thresholds of the
ordinal outcome variable were forced to be equivalent across groups as well. That is, the z-score
cut point of the outcome variable, where a participant would move from one classification to
another, was forced to be the same for both groups rather than allowing different cut points of the
outcome variable for each group. This was done to specifically test whether the cut points or
transition points from one group level to another were equivalent across groups.
Model 2.1c was a multi-group invariance model to look at equivalence of the model for
Males and females. This model had five variables (one less than the base models because gender
was used for the grouping variable and thus not included in the model): high school GPA, test
scores, race, core GPA, and income (see Table 11). Model 2.1c fit the data well, χ2 = 12, df=9,
CFI=.92, TLI=.91, RMSEA=.01. The variance accounted for in the model for females was
R
2
=.03 and for Males it was R
2
=.03. To compare nested models, delta change statistics were
used (Sprenkle & Piercy, 2005). This statistic is simply the difference (delta or δ) between the
chi square (δL
2
) value and the degrees of freedom (δdf) for two nested models (Bagozzi & Yi,
1988). The next model, Model 2.1d, was a multi-group invariance model to look at equivalence
of the model for whites and minorities. Like the other models, this was a complete invariance
model with equivalent regression paths as well as equivalent thresholds across groups. This
model also had five variables (again, one less than the base models because race was used for the
SPORTS RETIREMENT OF COLLEGE ATHLETES 45
grouping variable and thus not included in the model): high school GPA, test scores, gender, core
GPA, and income. Model 2.1d did not fit the data well, L
2
= 29, df=9, CFI=.56, TLI=.51,
RMSEA=.03 (see Table 11). The variance accounted for within the model for minorities was
only R
2
=.04 and for whites it was only R
2
=.03. The fit statistics indicated that some parameters
in the model were not equivalent across groups. The next model, Model 2.1e, was run with the
threshold constraints relaxed across groups. In other words, the thresholds for each group were
freely estimated. This model fit the data better, L
2
= 10, df=5, CFI=.89, TLI=.78, RMSEA=.02
(See Table 11). The variance accounted for in the model did not change after relaxing the
threshold constraints, minority R
2
=.04 and whites R
2
=.03.
The next set of models added an additional predictor variable of likelihood of
playing professionally. This was done in order to assess the impact of perceived likelihood of
playing professionally on retirement difficulty after taking into account demographic variables.
Model 2.2a was a baseline model that mimicked Model 2.1a. This set of models was run without
weights. This decision was based on earlier results which indicated the inclusion of weights was
not significantly helpful. Model 2.2a included seven variables: high school core GPA, test
scores, core GPA, gender, race, family income, and the likelihood of playing professionally.
This addition of likelihood of playing professionally contributed an additional five percent of the
total variance accounted for in retirement difficulties, R
2
=.08 (See Table 11). Race (β=-.07,
p<.001) and income (β=-.07, p<.001) were negatively associated with retirement difficulties, and
likelihood of playing professionally (β=.23, p<.001) was positively associated with retirement
difficulties in this model. Whites and individuals with lower income reported having more
difficulty with retiring from competing seriously in their sport, as did individuals who thought
they were more likely to play professionally.
SPORTS RETIREMENT OF COLLEGE ATHLETES 46
The next model (Model 2.2b) was a multi-group factorial invariance model looking at the
equivalence of the retirement difficulty path model across Males and Females. In Model 2.2b,
all regression paths as well as thresholds were constrained to be equal across groups. The model
fit well, L
2
= 11, df=10, CFI=.99, TLI=.99, RMSEA=.01 (See Table 11), which was an
indication that the levels of the outcome variable as well as the paths leading to retirement were
equivalent across groups. Race, income, and likelihood of playing professionally were also
significant in this multi-group model. The next model, Model 2.2c, assessed invariance of the
retirement model across minority and white groups. Again, all variable paths and thresholds
were constrained to be equal. This model fit the data poorly, L
2
= 58, df=10, CFI=.76, TLI=.71,
RMSEA=.05. In this model, race and family median income were negatively associated with
retirement difficulties for both groups. Minority participants with lower high school GPAs and
lower family income were as likely as whites with lower high school GPAs and lower family
income to report having more difficulty with retiring from sports competition (β=-.05, p<.03 and
β=-.06, p<.001). In addition, likelihood of playing professionally was positively related to
retirement difficulties. Minorities who thought they were more likely to play professionally were
as likely as whites who thought they were likely to play professionally to report more difficulty
with sports retirement (β=.22, p<.001). An additional model was run, Model 2.2d, in which the
equality constraints on the thresholds relaxed across groups. This model fit the data better, L
2
=
11, df=21, CFI=.92, TLI=.85, RMSEA=.04, indicating that while the paths from each variable to
the outcome variable of retirement were equally predictive for minorities and whites but the
points at which a minority participant moves from one group to the other was not the same for
whites. Because this model did not impact the predictors, the same variables which were
important in the previous model were significant predictors for this model. High school core
SPORTS RETIREMENT OF COLLEGE ATHLETES 47
GPA, income, and likelihood of playing professionally were equally predictive for whites and
minorities. This model accounted for 10% of the variance in retirement for minorities (R
2
=.10)
and slightly less of the variance in retirement for whites (R
2
=.07). From this, we concluded that
likelihood of playing professionally was an important variable in understanding the retirement
difficulties of former student athletes.
Result 3: Assessing Retirement Difficulty as a Mediator
This section expanded on the previous two sections assessed weather retirement
difficulty mediated the relationship between participant characteristics and identity with life
outcomes. Path analysis was used to model the relationship between participant characteristics
(i.e. demographics and high school tests and GPA), retirement from sports competition, and later
life outcomes for participants who did not go on to play professional sports. More specifically,
this model tested, for example, whether there was a direct relationship between the early
participant characteristic and identity on later life outcomes or whether that relationship was
filtered through retirement difficulty. As with previous models, sampling weights were not used
since they did not add much in the modeling process. The first model (Model 3.1) had seven
exogenous variables (high school core GPA, first year college GPA, test scores, gender, race,
income and the self-rated likelihood of going pro), one endogenous measured variable
(retirement difficulty), and two endogenous latent variables, one with six indicators and the other
with seven indicators (job satisfaction and life satisfaction; see Figure 14). Retirement difficulty
in this first model was entered as continuous rather than categorical. Since it is ordered, it was
treated like a true continuous variable rather than a categorical variable. The purpose was to
compare the model with a continuous retirement variable to one with a categorical retirement
variable. Model 3.1 had a mediocre fit, L
2
= 2,011, df=166, CFI=.86, TLI=.84, RMSEA=.06. Of
SPORTS RETIREMENT OF COLLEGE ATHLETES 48
the seven exogenous variables predicting retirement difficulty, four were significant (p < .05).
Likelihood of going pro (β=.21) was positively associated with retirement difficulties and
minority (β= -.06), high school GPA (β-.06) and test scores (β=-.05) were negatively associated
with retirement difficulties. These variables, as with before, accounted for less than 10% of the
variance in retirement difficulty (R
2
=.06) suggesting that there were additional variables which
also predict a large proportion of the remaining variance.
It is possible that additional measures in the SCORE data set could contribute or it may
be that the important predictors were measured in SCORE. This result could be an indication
that personal characteristics are not very predictive of later retirement difficulty or it could be an
indication that the use of a single indicator to understand retirement is insufficient to truly
understand the relationship between the variables. If the reason for the lack of prediction is the
former, it would suggest retirement difficulties could be related to other social and athletic
variables which happen during the collegiate competition years. In addition, retirement difficulty
was a significant predictor of both job and life satisfaction (β=-.04 and β=-.20) and had a
negative association with both, however, it accounted for miniscule amounts of variation in life
outcomes (R
2
=.002 and R
2
=.01).
An alternative model, Model 3.2, was also run to see if the retirement difficulty variable
was better represented as an ordered categorical variable (see Muthen, 1984) rather than
continuous. This model did provide a better fit to the data, L
2
= 1266, df=166, CFI=.925,
TLI=.914, RMSEA=.044, indicating that representing retirement as a categorical variable was
better than representing it as continuous. For this reason, subsequent models will specify
retirement difficulty as an ordered categorical variable. In this alternative model, only two
exogenous variables were significant predictors of retirement difficulties: likelihood of going pro
SPORTS RETIREMENT OF COLLEGE ATHLETES 49
(β=.23) had a positive association and High School GPA (β=-.11; see Table 12) had a negative
association. Based on the standardized beta estimates, the more likely an athlete thought it was
that they would play at a professional level the more difficulty they had with retirement.
Additionally, participants with lower high school GPAs had more difficulty with retirement than
those with higher GPAs. These two variables also accounted for less than 10% of the variation
in retirement difficulty (R
2
=.09). In addition, retirement difficulty was a significant negative
predictor of life outcomes (β=-.11, p<.001) however, it was not a significant predictor of job
satisfaction (β=-.03, p=.112). This indicates that former athletes, who reported having had less
difficulty with retiring from competing seriously in their sport, also reported having more
satisfaction with life. As with other variables in the model, retirement difficulties only accounted
for very minimal amounts of variance in job satisfaction (R
2
=.001) and life satisfaction
(R
2
=.011). Based on the first two models, it was determined that retirement difficulty was best
represented as a categorical variable, thus all subsequent models analyzed retirement as
categorical.
The next set of models was run to determine how well predictor characteristic
variables predicted not only retirement difficulties, but also some key life outcomes. The third
alternative model, Model 3.3, included participant characteristic variables as direct predictors of
the life outcome variables in addition to direct predictors of retirement difficulties (see Figure
15). This model provided a good fit to the data (L
2
= 1,429, df=152, CFI=.913, TLI=.892,
RMSEA=.05), however, it was significantly worse than the previous model (δχ
2
=148, δdf=12, p
<.001), indicating that the earlier model should be preferred (see Table 11). The fourth
alternative model (Model 3.4) was run to see if the paths from retirement difficulty were
necessary or rather it was run to see if retirement difficulty was useful in understanding life
SPORTS RETIREMENT OF COLLEGE ATHLETES 50
outcomes. For this model, the paths from retirement difficulty to job and life satisfaction were
forced to zero indicating no relationship between those variables, and paths from each exogenous
predictor variable to retirement difficulty and each outcome factor remained in the model (see
Figure 16). This model also provided an adequate fit to the data (L
2
= 1,360, df=152, CFI=.918,
TLI=.899, RMSEA=.05) which was significantly better than Model 3.3 (δχ
2
=69, δdf=2, p <.001).
Based on the result, it appeared that with the seven exogenous variables in the model, retirement
difficulty was not needed or useful in understanding the relationship between participant
characteristics and life outcomes (see Table 11).
Based on the results from Model 3.4, likelihood of playing professionally and gender had
positive associations with retirement difficulties and race, test scores, and high school GPA had
negative associations with retirement difficulties. Participants who reported higher perceived
likelihood of playing professionally had more difficulty with retirement (β=.03) in addition to
males (β=.02), whites (β= -.06), and individuals with lower test scores (β=-.05) and high school
GPAs (β=-.06). These four variables accounted for less than 10% of the variation in retirement
difficulty (R
2
=.07). In addition, race had a positive association with job satisfaction and high
school GPA had a positive association with job satisfaction. White participants (β=-.06) and
participants with higher high school GPAs (β=.14) had higher job satisfaction, accounting for 3%
of the variation in this outcome variable. Lastly, gender, race, and test scores had negative
associations with life satisfaction while high school GPA, family median income, and likelihood
of playing professionally had positive associations with life satisfaction. Female (β=-.06) and
white (β=-.13) participants reported having more life satisfaction as well as individuals with
higher high school GPAs (β=.14) but lower test scores (β=-.12), higher family income (β=.06),
SPORTS RETIREMENT OF COLLEGE ATHLETES 51
and those who felt they were less likely to play professionally (β=.04), accounting for 5% of the
variation in life satisfaction.
One final alternative model was run using the seven exogenous variables (Model 3.5).
This model was run to try and obtain a more simplistic model by using only those factor items on
the outcome variables that were the most important. For the job satisfaction variable, two items
had lower values than all other items and thus were removed (q51b and q51f) leaving the four
indicators with the highest loadings in the model. For life satisfaction, four items out of the
seven had lower loadings than the others and thus were removed from the model (q58c, q58d,
q58e, and q58i) leaving the three items with the highest loadings in the model. The point of this
model was to see if the model fit as well with only the strongest factor indicators included. A
simpler model would be easier to replicate in future studies and would lead to a far simpler
interpretation. This fifth model, Model 3.5, fit the data well, L
2
= 562, df=55, CFI=.932,
TLI=.896, RMSEA=.05, indicating that the factors are represented well with fewer indicators or
that the removal of indicators did not result in poor model fit. For all subsequent models, the
outcome factors were used in the reduced item form (see Table 12).
The first set of models in the section assessed the impact of participant
characteristics on retirement difficulties and life outcomes without the effects of identity
included. The next two models included the identity factors that were assessed in result three:
athletic identity, academic identity, and others views of athletic identity. These next models
assessed the impact of identity on the retirement process for former athletes. The first model,
Model 3.6, was a full path model with the seven exogenous variables from the previous models
as well as the three additional exogenous latent predictor identity variables, retirement
difficulties, and the two life outcome factors. The first model assessed the relationship between
SPORTS RETIREMENT OF COLLEGE ATHLETES 52
these variables with all exogenous predictors leading to retirement difficulty as well as the life
outcome factors. This model had a mediocre fit to the data, L
2
= 3417, df=250, CFI=.83,
TLI=.79, RMSEA=.06, with an acceptable RMSEA value but with CFI and TLI values that are
low (Bagozzi & Yi, 2012, p21).
The next model, Model 3.7, was run as an alternative model to the previous to assess
whether retirement difficulty was useful in understanding the relationship between participant
characteristic and life outcomes. In this model, the paths from retirement to life outcomes were
fixed to zero meaning that retirement had no impact on life outcomes and that all exogenous
variables were impacting retirement and life outcomes. This model had a mediocre fit, L
2
=
3462, df=252, CFI=.82, TLI=.79, RMSEA=.06, and was significantly worse than the previous
model (Model 3.6), δχ
2
=45, δdf=2, p <.001, indicating that some portion of the relationship
between the participant characteristics and life outcomes was impacted by retirement difficulties
(see Table 12). In Model 3.6, athletic identity (β=.199) was a positive predictor and academic
identity (β=-.062) was a negative predictors of retirement difficulties however, others views of
one’s identity was not predictive of any outcome in the model. The more an individual identified
with being an athlete, the more difficulty they had with sport retirement, and the less an
individual identified with being a student the more difficulty they had with sport retirement.
Also, likelihood of playing professionally was a positive predictor of retirement difficulties and
high school GPA and test scores were negative predictors of retirement difficulties. Individuals
who felt they were more likely to play professionally had more difficulty with retirement (β=.22)
as did individuals with lower high school GPAs (β=-.06) and test scores (β=-.05), and
participants who were white (β=-.06). Athletic identify and academic identity were positive
predictors of job and life satisfaction. Participants who identified more as an athlete as well as
SPORTS RETIREMENT OF COLLEGE ATHLETES 53
those who identified more with being a student reported being more satisfied with their job
(β=.10 and β=.17 ,respectively) and life (β=.13 and β=.13). Retirement difficulties had a
negative association with life outcomes. Participants who reported more difficulty with retiring
from sports competition also reported less satisfaction with job and life (β=-.06 and β=-.12).
Higher test scores and high school core GPAs were positive predictors and resulted in higher job
satisfaction (β=.01 and β=.15). Test scores was a negative predictor of life satisfaction and high
school GPA was a positive predictor of life satisfaction. Lower test scores but higher high
school GPAs resulted in higher life satisfaction (β=-.13 and β=.13). In addition, gender was
negatively related to job satisfaction while likelihood of playing professionally was positively
related. White individuals and those who thought it was more likely they would play
professionally were more satisfied with their job (β=-.10 and β=.05). Gender was negatively
related to life satisfaction and family median income was positively related to life satisfaction.
White participants were also more likely to report higher life satisfaction (β=-.14) as were those
individuals who reported higher family median income values (β=.05). The addition of the
identity variables added approximately five percent of the variation accounted for in retirement
difficulties (R
2
=.11), job satisfaction (R
2
=.08), and life satisfaction (R
2
=.10; see Figure 17 and
Table 13).
Because Model 3.6, which had all exogenous variables predicting retirement as
well as life outcomes in addition to retirement difficulty predicting life outcomes, represented the
relationship between all of the variables the best, further analyses were needed to fully
understand the role retirement difficulties played in this process. In this model, retirement
difficulties were viewed as a variable that mediated the relationship between the participant
characteristics and the outcome variables. Mediation describes the relationship between a set of
SPORTS RETIREMENT OF COLLEGE ATHLETES 54
variables such that the impact of one variable on another is transmitted through a third variable
called a mediator. Mediation allows researchers to understand how a third variable might impact
the relationship between two other variables, as was done here (MacKinnon, Fairchild, and Fritz,
2007; see Appendix K). Each indirect effect of the participant characteristics on life outcomes
through retirement difficulty was tested (see Table 14). For job satisfaction, there were
significant indirect effects from athletic and academic identity, race, and likelihood of playing
professionally. Higher student identity led to lower retirement difficulty which led to higher job
satisfaction while lower student identity led to more retirement difficulty and less job
satisfaction. Higher athletic identity led to more difficulty with retirement and lower job
satisfaction while low athletic identity led to less retirement difficulty and higher job satisfaction.
Whites reported having more retirement difficulties which lead to less job satisfaction and
minorities had less retirement difficulties which lead to more job satisfaction. Having a higher
perceived likelihood of playing professionally led to more difficulty with retirement from sport
that then leads to less job satisfaction while lower perceived likelihood of playing professionally
leads to less difficulty with sport retirement and more job satisfaction.
For life satisfaction, there were significant indirect effects from athletic and academic
identity, high school GPA, race, test scores, and likelihood of playing professionally. Higher
student identity led to fewer retirement difficulties that led to more life satisfaction while lower
student identity led to higher retirement difficulty leads to less life satisfaction. Higher athletic
identity led to more retirement difficulty leads to less life satisfaction and lower athletic identity
led to fewer retirement difficulties and more life satisfaction. Higher high school GPA led to
less retirement difficulty that led to more life satisfaction, and lower high school GPA led to
more retirement difficulty which led to less life satisfaction. Whites had more retirement
SPORTS RETIREMENT OF COLLEGE ATHLETES 55
difficulty which led to less life satisfaction and minorities had less retirement difficulty which led
to more life satisfaction. Higher test scores led to less retirement difficulty which then resulted
in more life satisfaction and lower test scores led to more retirement difficulty which resulted in
less life satisfaction. Higher perceived likelihood of playing professionally led to more
retirement difficulties which then led to less life satisfaction and lower perceived likelihood of
playing professionally led to less retirement difficulty which led to more life satisfaction.
Lastly, in order to rule out potential group differences in the final model, multi-
group invariance models were run. These models were run for gender and race and mimicked
the model specification from Model 3.6 above. All invariance models started with the strictest
form of constraints, where all paths and estimates were forced to be equivalent across groups. If
the model fit with the strictest form or constraints, no further models were assessed because
complete invariance between groups was achieved. If the model fit poorly with the strictest form
of invariance, constraints were released as needed to assess what level of invariance was present.
As mentioned before, lack of invariance can be a sign that there are interactions between
variables and that the model works differently for Males and Females, or for Whites and
Minorities.
The first two multi-group models, Model 3.8 and Model 3.9, assessed invariance for
males and females based on the full path model where the exogenous predictor variables
predicted both retirement difficulties as well as the life outcome factors, and where the
exogenous variables predicted life outcomes but retirement difficulties did not (i.e. the paths
from retirement difficulty to life outcomes was fixed to zero). The model with all paths allowed
to be estimated fit the data well and was the better of the two models, χ
2
= 2,769, df=559,
CFI=.874, TLI=.871, RMSEA=.05. This indicated that the model was best when retirement was
SPORTS RETIREMENT OF COLLEGE ATHLETES 56
mediating the relationship between participant characteristics and life outcomes and that it
worked the same for males and females. The next two models, Model 3.10 and Model 3.11,
were run to assess the invariance of Model 3.6 for race. These two models (Model 3.10 and
3.11) mimicked Models 3.8 and 3.9 with the exception of the race and gender variables being
switched so that the models looked at invariance of minorities versus whites. As with the
previous invariance models, Model 3.10, where the relationship between participant
characteristics and life outcomes is mediated by retirement difficulty, fit better than the model
with no retirement prediction on life outcomes, L
2
= 2072, df=559, CFI=.875, TLI=.875,
RMSEA=.04. As with earlier models looking at invariance for gender, this model worked the
same for minorities and whites indicating invariance between the groups (see Figure 17 for final
model).
CHAPTER 4: DISCUSSION
The purpose of the current study was to better understand how Retirement from
Sports (RS) for former student-athletes was predicted and how it might impact later life
outcomes. To be more specific, this study attempted to understand how participant
characteristics (demographics) predicted the difficulty or ease at which a former NCAA Division
I student-athlete retired from competing seriously in their sport, particularly for those who do not
continue on into a professional career. In addition, this study was also concerned with
understanding the impact, if any, that RS difficulties had on later life outcomes like job and life
satisfaction while mediating the relationship between participant characteristics and identity with
life outcomes. The first hypothesis assessed whether participant characteristics like
demographics, academics, and perceived likelihood of playing professionally impacted
retirement difficulties. The second hypothesis was concerned with whether retirement
SPORTS RETIREMENT OF COLLEGE ATHLETES 57
difficulties mediated the relationship between participant characteristics and identity on life
outcomes. Based on past research, it was hypothesized that higher academic performances
would lead to fewer difficulties with retirement. Also, based on past research, it was also
hypothesized that a higher athletic identity would negatively impact retirement (higher athletic
identity would lead to more retirement difficulties) and result is lower reported life satisfaction
and that higher academic performances would lead to fewer difficulties with retirement and
ultimately more positive outcomes
The results were assessed in three steps, with the last step incorporating the
previous three and evaluating a full path model with participant characteristics, RS difficulty,
and life outcome factors. The first set of results were preliminary analyses which assessed
whether sampling biases were present in the sample and established the factor structure of
identity and life outcomes. Results from these initial analyses were used to test the hypotheses in
the following result sections. For SCORE, the participant pool contained approximately 25,000
participants and was considered to be representative of the population of student-athletes. The
percentage of participants who actually responded to the SCORE survey out of the 25,000 was
about 30% (6,000). Because of this, it was important to understand whether the participants who
actually responded were also representative of the population of student-athletes so that results
could be more easily generalized to the population. The findings from result one indicated that
there were biases present and that the participants who actually responded were not
representative of the subject pool. This also meant that the SCORE respondents would not be
representative of the general population of student-athletes. In addition, individuals with lower
test scores and high school grades, as well as males were less likely to respond to SCORE.
Women athletes tended to be overrepresented and male athletes were underrepresented in the
SPORTS RETIREMENT OF COLLEGE ATHLETES 58
sample. This was especially true for male football and basketball players. The low variation
accounted for is the reason the use of sampling weights was not actually necessary for this
analysis. While response was based on measured participant characteristics, it was not to a large
enough extent to change the results of the analyses.
The factor structure of identity items as well as life outcome items was assessed
next to determine if the variables formed solid factors and whether the factors were
unidiminsional or multidimensional (see. McDonald, 1981; Waugh & Chapman, 2005). The
variables used came directly from the self-report SCORE survey. The first factor models
assessed the structure of identity. The final model was a three factor model of identity with
academic, athletic, and others view of one identity as factors. The final model had a total of 11
indicators (there were originally 12 but one was removed due to its misfit in the model). The
factor structure was decent and was also invariant between males and females as well as between
minorities and whites. The next set of factor models were used to assess job and life satisfaction.
The goal was to determine if life outcomes represented one factor or multiple factors as was
suggested in the SCORE survey, as well as determine if the variables formed two solid factors or
if the model could be further optimized to address life outcomes (i.e. adding more than two
factors). From these analyses, we learned that the variables did form two separate factors and
that a few variables did not fit into the model well and thus were removed. The overall model
was average and this was true when differences were assessed for gender and race. More
specifically, the factor model was equivalent for males and females as well as for minorities and
whites.
The second results section assessed how well participant characteristics accounted
for difficulties with retiring from competing seriously in one’s sport. This was done using path
SPORTS RETIREMENT OF COLLEGE ATHLETES 59
models rather than traditional regression models to be consistent with the models to follow and
the results from this section were incorporated into a final path analysis model. These models
were a first step in building the larger path model and so were run as path models themselves.
The models were run in two sets, one set with only the participant characteristic variables, and a
second set with the addition of a variable assessing the perceived likelihood of playing
professionally. From these models, we learned that participant characteristic variables accounted
for a small amount of variation in retirement (less than 5%) and that higher perceived likelihood
of playing professionally was predictive of more sport retirement difficulty. Perceived
likelihood of playing professionally accounted for an additional 4% of the variation in retirement
difficulty when added to the model, however, these variables together accounted for less than
10% of the variation in retirement difficulty.
Alternative models suggested that the sampling weights only affected the high school
GPA variable and even when it became significant in weighted analyses, it still had the least
amount of impact in the model. The weights did not add much to the model and didn’t result in
major changes to the estimates. In addition, variables seem to predict equally well for males and
females but this was less so for minorities and whites. In particular, minorities and whites had
very different levels of difficulty and test scores were more predictive for whites than minorities
with the opposite interpretation for the groups. Whites with lower test scores had more difficulty
retiring and minorities with higher scores had a harder time retiring. This was an interesting
result because past research suggests that minorities underperform when compared to whites on
standardized tests. This result appears to imply that minorities struggle with retirement whether
they have high or low test scores because their high scores still tend to be below average
SPORTS RETIREMENT OF COLLEGE ATHLETES 60
(Freedle, 2003). This may also be due to the fact that minorities, in general, had a harder time
retiring from competition than whites.
The third and final results section assessed the mediation of retirement difficulty
by adding all of the pieces together. This full path model assessed how well participant
characteristics, likelihood of playing professionally, and identity predicted retirement difficulties
and how well retirement difficulties then predicted life outcomes. Several alternative models
were run, first with only participant characteristics and likelihood of going pro and then with the
addition of the identity variables. This was done to assess the impact identity had above and
beyond participant characteristics. The best model with identity variables removed was one
where all exogenous variables predicted both retirement and life outcomes but where retirement
did not predict life outcomes. However, when identity was added to the model, this relationship
was the opposite. With identity in the model, the best model was one where all exogenous
variables predicted retirement and life outcomes and where retirement also mediated the
relationship between the predictors and life outcomes. The results found present a somewhat
complex view of how these variables work together, suggesting that retirement difficulties did
regulate at least some of the relationship between earlier life variables and later life outcomes.
The results from this section supported the hypothesis that RS played a role in the transitional
process for former athletes. While the result was positive, this model still accounted for no more
than 10% of the variation in life outcomes. In addition, this process is the same for males and
females as well as minorities and whites.
The findings partially supported the first hypothesis. Higher perceived likely of
playing professionally resulted in more retirement difficulties but high school academic
characteristics were not predictive of retirement difficulties. In addition, Whites and individuals
SPORTS RETIREMENT OF COLLEGE ATHLETES 61
with lower income reported having more difficulty with retiring from competing seriously in
their sport. The findings also partially supported the hypothesis that retirement difficulties
mediated the relationship between participant characteristics and identity with life outcomes.
Higher athletic identity led to more difficulty with sport retirement and ultimately to less job and
life satisfaction. Lower academic identity led to more retirement difficulty and lower job and life
satisfaction. Higher perceived likelihood of playing professionally led to more retirement
difficulty and less job and life satisfaction. Lastly, higher high school core GPA lead to fewer
difficulties with retirement and ultimately more life satisfaction. The relationship between high
school core GPA and job satisfaction was not mediated by retirement difficulty.
Although the findings are significant, they are fairly small, but they can be a step
forward in understanding the life outcomes for NCAA student-athletes. This work has shed
some light on the importance of different aspects of the transitional process for student-athletes
and specifically highlighted the importance of the sport retirement experience for student-
athletes who do not go on to play professionally. Additionally, these analysis gave an idea of
how pre-existing characteristics and identity development function together in the developmental
process. The findings from this study can help NCAA research and athletic department officials
get a better understanding of which athletes are at risk of negative experiences and outcomes and
if they can identify them, it is possible to implement interventions to try and maximize the
number of positive experiences and positive outcomes.
Understanding how long adjustment to the new circumstances may take and how athletes
cope with the process might be a key to understanding what happens after college. Sinclair and
Orlick (1993) did a study assessing transitions from sport for athletes. They developed an athlete
retirement questionnaire which was concerned with understanding the transitional experiences of
SPORTS RETIREMENT OF COLLEGE ATHLETES 62
athletes (specifically high-performance athletes). It was a 34-item instrument self report measure
which addressed things like coping strategies, retirement decisions, and difficulties with the
transition. They found that in their sample it could take anywhere from 6 months to a year or
more for many athletes to feel they had completely adjusted to their life outside of athletics. In
addition, many of the participants reported that it was difficult lose the social aspect of their sport
and they were worried about jobs and finances after retirement. These authors make some
suggestions for helping athletes make a positive transition: seminars or workshops dealing with
retirement and adjustment issues, provide a resources center, and maintain contact with retired
athletes. Some work has been done looking at research on coping with losses and grieving and
suggests that similar principals can be applied to the case of the student athlete (see Baillie,
1993). In this situation, the grieving is for the loss of competition or sport and athletes may go
through what Kubler-Ross (1969) identified as the stages of adjustment: denial, anger,
bargaining, depression, and acceptance.
Limitations
The current findings can take us a step closer to understanding the sports
retirement process for D-I student athletes, and while the minimal findings are an indication that
the self-report variables are not highly related, it can also be an indication of other issues related
to the data and sample and may signify that improvements are necessary to really understand this
process. The minimal findings could be due to issues that arise when using survey data, they
could be caused by the sample that was obtained, or it could be that better variables or measures
are necessary to really understand sports retirement for this population.
Past research in psychology has focused on the benefits and limitations of survey
research. Sommer and Sommer (2002) note that mail surveys, as used partially in the presented
SPORTS RETIREMENT OF COLLEGE ATHLETES 63
research, are useful when researchers want to cover a large area of the population quickly while
keeping costs related to travel and labor down. Mail surveys allow for standardization and a
level of anonymity that is not possible with face-to-face interviews. A problem, however, is that
mail surveys often result in low response rates, as seen with the SCORE data (30%). Sommer
and Sommer (2002) note that a survey sent to randomly selected households from a public
directory will yield a response rate somewhere between 10-40 percent. One might think that the
rates would be higher in a specialized population that is smaller, however, that is not what was
observed with the SCORE survey of D-I student athletes. Any response rates can lead to biases
in the sample, and this was probably observed in the SCORE data.
Within the survey research literature, there is a growing interest in understanding
the effectiveness of internet surveys in relation to traditional paper and pencil methods. The
results, however, have been mixed. Some research has suggested that internet surveys are more
effective than paper pencil methods, increasing response rates and reducing costs (Kiernan,
Kiernan, Oyler, Gilles, 2005). Other research suggests that response rates for paper and pencil
administration and web- based survey are similar, noting the use of pre-notification postcards as
necessary to increase response (Kaplowitz, Hadlock, & Levine, 2004; Hart, Brennan, Sym,
Larson, 2009). Alternatively, other researchers found lower response rate for web based surveys
when compared to paper and pencil versions (Shih & Fan, 2008; Yetter & Capaccioli, 2010). A
meta analysis by Shih and Fan (2010) looked at 39 articles comparing web based and paper
pencil surveys. Overall, they found there was a higher response rate for mail based paper and
pencil surveys. They also noted that age played a significant role in response rate differences.
Older individuals tended to respond more to paper pencil surveys while younger, college aged
individuals tended to be more responsive to the web based surveys (Shih & Fan, 2010;
SPORTS RETIREMENT OF COLLEGE ATHLETES 64
Kaplowitz, Hadlock, & Levine, 2004; Yetter & Capaccioli, 2010). Incentives were also provide
in the form of a t-shirt and for a selected group of high-risk non-responder, they provided an
additional incentive of a pen. This is important to consider because the methods used to collect
the data for the SCORE 2010 administration was a mix of internet and paper and pencil surveys,
in addition to notification post cards. The use of online surveys methods may or may not have
contributed to the low response rate for the SCORE survey. It is also important to note that more
paper and pencil surveys were returned than online surveys in the second SCORE administration
(in the first administration in 2008, only paper and pencil surveys were used). Using both
methods still yielded less than ideal response rates. For data that are more representative and
more easily generalizable, researchers at NCAA may want to re-think their data collection
methods or even their instrument. The use of internet for survey research may keep the costs
down, but it may have hindered their response rates as well. Better incentives may be helpful
such as a small monetary award after the survey is completed or being entered into a raffle for an
item. It is unclear whether the incentives used (a t-shirt and a pen) were adequate for the sample
population with an average age range of 29-31. Possible research into appropriate incentives and
most useful incentives would be beneficial.
An additional concern with survey research is related to biases that often occur
with self report measures. Reis and Gable (2000) explain that with self report measures when
there is retrospection, participants have to retrieve events from memory in order to answer the
questions and whatever details seem most relevant to them are selected. They note that features
relevant to the question (i.e. satisfaction) must be recalled and used in such a way as to make a
decision for response to the survey item. They state that instead of this complex process being
carried out fully, participants use cognitive shortcuts to answer questions quickly and this leads
SPORTS RETIREMENT OF COLLEGE ATHLETES 65
to varying levels of accuracy. Common issues with retrospective survey items consist of
forgetting on the participant’s part, lack of salience, confusion about the reference period,
recency effects, and influences of current state of mind (Bailar, 1989; Reis & Gable, 2000;
Sommer & Sommer, 2002). Forgetting is one of the major problems that needs to be addressed.
As time passes, it is natural for the participant to forget details about feelings or events from the
specified time period. This bias only gets worse over time. In addition, what is important for the
researchers may not seem important to the participant and therefore may not be memorable and
may be lacking salience for the participant. Sometimes there can also be confusion about the
reference period. In a situation like this, participants have a hard time separating events before
or after a certain point, for instance, participants may have a hard time remembering details of
retirement more than 2 years after the event or may not remember certain details of college that
happened 2 years prior to graduation. Additionally, current mood may impact the way a
participant answers retrospective survey items, for example, answering how satisfied the
participant was with their college experience 5 years after the occurrence can be impacted by the
current mood either positively or negatively. Some work has been done to examine ways to
improve these issues in retrospective survey research. One method to overcome issues discussed
is to narrow the reference period. Because the longer the time between the event and the
questions the more one forgets, using shorter reference periods, for example six years after
freshman year or one year after college graduation can help get more accurate reports or
retirement difficulty (Bailar, 1989; Sommer & Sommer, 2002). For more accurate accounts of
college level variables, data should be collected during the college years or during the
development of identity during college rather than after the fact. This is not always possibly
however and can be especially difficult to do on such a large scale for several years as well as
SPORTS RETIREMENT OF COLLEGE ATHLETES 66
expensive. But it may be more plausible to monitor some aspects of the graduation and
retirement process at shorter time intervals leading up to the traditional SCORE survey which is
10 years after freshman year. Self report measures are also subject to social desirability biases.
John and Benet-Martinez (2000) note that researchers have argued that social desirability limits
the usefulness of self report measures. The researchers focus on two types of social desirability:
impression management and self-deceptive enhancement. Both types of desirability result in the
participant trying to be viewed more favorably. This could be of particular interest in this case
since it involves former student athletes reporting responses to the entity that largely regulated
their collegiate sports career. While the participants are assured their responses are confidential,
this may not deter them from responding in a more desirable way.
Also, while the survey weights used in the analyses presented here appeared to
have little impact, it is possible that the current sample has lead to some of the minimal findings
here. As mentioned previously, we know that biases existed within the sample. For instance, we
know that individuals who graduated made of the majority of the sample (95%). So the sample
consisted mostly of successful student athletes with regards to college. It is likely that the
variables are not highly related for successful college graduates but may be more related for
those individuals who did not graduate. Unfortunately, it is impossible to tell from the current
SCORE data.
Lastly, a major focus of this study was on the role of sport retirement and
specifically difficulties with retiring from sport competition. Unfortunately, retirement was
assessed with a single item making it difficult to understand it with any depth. For one,
retirement is more complex than a single item implies and individuals may perceive retirement
from sport differently. Using multiple items to measure retirement would ensure that
SPORTS RETIREMENT OF COLLEGE ATHLETES 67
misunderstanding on one item did not impact the entire scale (Kane & Radosevich, 2011). In
addition, multiple item measures allow for the understanding of a concept at more specific levels
than one item allows. For example, asking how hard it was to retire from sport competition does
not address what aspects of the retirement process were hard. It could be that not seeing
teammates every day is what made sport retirement difficult for athletes and not that competition
ended or vice versa. Asking about retirement difficulties with one item confounds difficulty and
doesn’t allow for a thorough assessment of causes or sources of difficulty that may relate to the
retirement process (Kane & Radosevich, 2011). Also, results from multiple item measures can
be more consistent over time. This is crucial for researchers at the NCAA if they plan to use
these items or study retirement of their athletes in depth. Using a measure that is consistent and
reliable over time can help researchers get more accurate accounts of the RS process for student
athletes from year to year. A more accurate and thorough measure of retirement is necessary to
better understand the role it plays in the lives of student-athletes (Spector, 1992). RS was not
considered as a construct, and certainly not the most important construct.
In addition, with a single item measure of a complex construct, measurement error is
typically very high. As was demonstrated here, having multiple items allows researchers to use
different tools to then assess research questions more accurately such as factor analysis. It is
imperative that researchers at NCAA operationally define what they mean by retirement from
competition. Doing this will help guide future research in this area by giving more specific ideas
and guidelines related to what the important aspects of sports retirement are for them. For
example, the question in the SCORE survey asks how hard it was to retire from competing
seriously in one’s sport, but they don’t state what retirement means exactly. Retirement in this
context could mean no longer playing college sports but a former athlete could still be playing
SPORTS RETIREMENT OF COLLEGE ATHLETES 68
for fun in coed or adult leagues. And if they take their adult leagues seriously, they may not
consider themselves retired at all. It would be helpful to know what is meant by retirement and
what level of retirement each participant is at. For example, are they retired completely from
playing altogether or are they playing for fun. Another question they may be important in
understanding the retirement process is when exactly they retired. If retirement is defined as no
longer playing one’s sport or choice, if might help to know how long it took for them to stop
competing all together (including for fun). For example, retirement for someone who stopped
with college may have been harder while dealing with the transition from college to work as well
but a former athlete who was still playing for fun while making the college to work transition
may report less difficulty with retirement from competing once finished or vice versa. Asking
more questions related to retirement and being more specific with question wording can greatly
improve the understanding of the participant when answering questions related to retirement on
SCORE and can increase the understanding of the researchers at NCAA with regards to
retirement for student athletes. The motivation behind further research in this area is to get a
deeper understanding of what happens to student athletes after college and collegiate sports
participation. The NCAA has been very diligent in monitoring student athletes during college
but has failed to monitor the impact of NCAA collegiate sports participation has beyond.
Because past research has suggested there are negative outcomes related to sports retirement, it
seems like a practical idea to gauge whether sports participation in college leads to negative or
position outcomes, and if so, wheat factors specifically contribute to negative versus positive
outcomes. Knowing this can help athletic coaches and staff as well as other school officials and
the NCAA maximize the position experiences and outcomes of their student athletes in and
beyond college. It can potentially help ensure a positive transition into the workforce
SPORTS RETIREMENT OF COLLEGE ATHLETES 69
Future Directions
Despite some difficult limitations, this work is a stepping leading to further
understanding the process of retirement from sports (RS) in the life of an NCAA student-athlete
as well as a catalyst for making improvements to research methods for understanding this
process better in the future. There are several directions that research in this area can go from
here. One important direction for this research is to extend what was done here is to include
additional concepts from the longitudinal process models of Tinto (1975) and Comeaux and
Harrison (2011; see Figure 18). In these models, commitments and integration into the system
played a large role in success. Tinto suggests that the higher one’s integration, the higher their
commitment to the institution and the goal of graduation. He also suggests that lower levels of
commitment led to a higher likelihood to dropout. Comeaux and Harrison suggest that sports
commitment is also important in addition to social and academic commitments and goals. These
were not assessed by the current models but may play an important role in understanding the
impact of sport retirement on life outcomes for student-athletes. The current model accounted
for no more than 10% of the variation in retirement difficulties and life outcomes meaning that
90% of the variation is unaccounted for, or is measurement error. It is important to get a better
understanding of the factors that impact this relationship so that athletic personnel can better
prepare students for successful transitions out of college and set them up for successful futures.
Future research should focus on developing an adequate measure to assess sports
retirement from collegiate sports. This would first require some work to get a proper operational
definition of what is meant by sports retirement and this is not an easy task. For one, retirement
can mean different things to different people and there can be different types or levels of
retirement. For example, one may retire from playing sports collegiately but may be playing for
SPORTS RETIREMENT OF COLLEGE ATHLETES 70
fun in adult leagues and thus doesn’t consider themselves completely retired or gauges the
difficulty differently because they are still involved in their sport. This can also be true for
individuals working in careers related to their sports or sports in general. Because of this, it is
important to understand how retired someone is (i.e. still playing for fun, not playing or working
in sports related field, etc.) to get a better understanding of how retirement impacts life
outcomes. It may be that different types of retirement led to different outcomes such as
individuals who are still playing for fun or who have jobs in sports report less difficulty with
retirement and thus have more positive life outcomes than those who have zero involvement in
their sport. It might add to the understanding if reasons for retirement were explored as well. It
could be that someone was cut from their team and thus retired, it may be injury related, or it
could just be eligibility ran out. Either way, different reasons for retirement may lead to different
outcomes. Understanding whether student athletes who retire from competition for injury
reasons have more negative outcomes can help better prepare those individuals for their
departure. This may be by providing counseling services to help them cope or helping them with
job preparation, either way it would be helpful and could led to more positive outcomes for
student athletes. Continued research in the area of sports retirement and life outcomes of
collegiate athletes is warranted to expand the knowledge of the impact collegiate sports
participation can have on outcomes for athletes. The current research sheds light on some big
issues facing NCAA researchers in the quest for understanding the role retirement difficulties
play. Taylor and Olgive (1994) state that sports retirement may not lead to negative outcomes or
reactions for all athletes and that it really depends on the quality of adaptation. This could be
critical to understanding this process for collegiate student athletes and this particular set of
results. Because retirement can have negative effects psychologically, emotionally, and
SPORTS RETIREMENT OF COLLEGE ATHLETES 71
physically (Hallden, 1965, as cited in Taylor and Olgive, 1994) it is important to continue
research in this area to determine if the findings in this paper are generalizable to the entire
population, or if this process and the relationship between the variables examined here have the
same relationship for the missing parts of the sample (i.e. non responders, transfers, non
graduates, etc).
In addition, continued research can help inform researchers in other areas about
the role of retirement difficulties or some activity on transitions out of college and later life
satisfaction. For instance, this type of research may help understand the process of retirement for
music majors who do not peruse a musical career past college but instead enter the workforce.
These individuals may experience similar reactions to the ending of their musical career and the
re-identification in the workforce. So this research reaches far beyond just that of the student
athlete retirement and can have implications for many transitions faced by college students in
today’s society who are faced with the abrupt end to the college career and the activities, clubs,
and social activities associated with it.
Suggestions for the NCAA. The immediate next steps would be to offer
recommendations to the NCAA as to prepare for their next administration of SCORE. The first
recommendations I would mention would be to seriously consider the variables being used.
Specifically the ones that are intended to represent factors (identity, job satisfaction, and life
satisfaction) are weak. Specifically regarding life and job satisfaction, it seems these variables
were taken from the Monitoring the Future (MTF; see University of Michigan, 2013) survey as a
way to obtain a comparison sample of non-athletes (supposedly to compare responses). One
issue that arises is that there is no documentation as to how these items on MTF were originally
chosen back in 1975. There certainly have been better measures developed over the past 35 plus
SPORTS RETIREMENT OF COLLEGE ATHLETES 72
years to better understand life satisfaction (refer to Pavot & Diener) and it seems that any
structure of the scale items from MTF have not really be tested at all. My own results seem to
show that the items used in the SCORE survey seem to be a subset of the full satisfaction scale
used on MTF – and this is partly why the factor structure of satisfaction was not very strong. For
some reason, the happiness or life satisfaction items were assessed using 24 items in MTF while
SCORE only used 9. Thus, I recommend the inclusion of a short scale specifically developed
and tested to reliably measure life satisfaction. For example, Pavot and Diener (1993) developed
the Satisfaction with Life Scale which is thought to measure “life as a whole.” The scale is made
up of five items to which the participant is to agree or disagree: (1) in most ways my life is close
to my ideal, (2) the conditions of my life are excellent, (3) I am satisfied with my life, (4) so far I
have gotten the important things I want in life, (5) if I could live my life over, I would change
almost nothing. Responses range from 1= “strongly disagree” to 7= “strongly agree”. They
measure has been used by a good number of researchers and has shown good reliability and
stability over retesting periods. The benefits of replacing the current scale with one similar to
that listed above are the documented reliability and validity, as well as the concise nature of it.
It’s almost half of the length of the current scale and considering the length of the SCORE
survey as a whole; a few less items might help. In the same way that better scales probably
exists for measuring life satisfaction, there are possibly more reliable scales to measure job
satisfaction and athletic versus academic identity. These options should be more deeply
explored.
Also, another immediate recommendation is to develop a retirement difficulty
scale to be used on the next administration of SCORE. In order to rule out retirement difficulties
as a predictor of negative life outcomes or pathways, it needs to be explored more thoroughly.
SPORTS RETIREMENT OF COLLEGE ATHLETES 73
For starters, I would recommend looking at other measures of retirement not related to sports
specifically. The Health and Retirement Study (HRS; see University of Michigan, 2013) which
was first launched in 1992 is a longitudinal study concerned with work and health transitions of
older adults. Particularly, they are interested in attitudes and beliefs about retirement, as well as
reasons for retirement, how people retire, and stressors related to retirement. They have several
questions related to basic retirement that might be helpful for NCAA researchers trying to better
understand the process or sports retirement for student athletes. They ask questions related to the
level of retirement (i.e. completely retired, partly retired, not at all retired), time of retirement
(month and year), whether it was their decision or whether it was forced, how much they had
thought about retirement, how much they discussed it with others (family, spouse/significant
other, friends, coworkers, etc.), quality or satisfaction with post retired life compared to pre
retired life and thoughts about positive or negative aspects of retirement (i.e. no pressure and
more free time versus being bored or missing people). These questions seem reasonable and
seem like they can be adapted to represent sports retirement for student athletes. In addition,
they can also be useful in representing later life work retirement much farther down the road as
was intended if NCAA decided they wanted to do so. I think that doing a better job of
understanding the role sports retirement plays for student athletes would give the NCAA
researchers a deeper understanding of the outcomes or paths these students take.
Lastly, because of the biases mentioned which researchers encounter with
retrospective surveys, it is recommended that the NCAA consider multiple sampling occasions
for the SCORE administration. Sampling student athletes during their first year after college
graduation or college exit may result in more accurate reports of initial retirement questions. For
example, asking how much a former athlete thought about retirement and how much they
SPORTS RETIREMENT OF COLLEGE ATHLETES 74
prepared for retirement or even how they feel or felt about retirement shortly after it actually
happened will likely yield more accurate reports than asking 5-6 years after the fact. In addition,
NCAA can re-sample those individuals 5-10 years after to see how the former student athletes
adjusted to life after collegiate sports and compare retrospective accounts of the sports retirement
process to reports taken at the actual time of retirement. By doing this, they can see how ideas
about retirement changes over time and weather life outcomes impact the remembered retirement
difficulty/process. Considering the recommendations listed would be a great next step to fully
understanding sports retirement for NCAA student athletes and in addition to life outcomes.
SPORTS RETIREMENT OF COLLEGE ATHLETES 75
References
Adler, P. & Adler, P. A. (1985). From idealism to pragmatic detachment: The academic
performance of college athletes. Sociology of Education, 58 (4), 241-250.
Anderson, E. D. (2012). Using the master tools: Resisting colonization through colonial sports.
The International Journal of the History of Sport, 23 (2), 247-266.
Atkinson, R. C. & Geiser, S. (2009). Reflections on a century of college admissions tests.
Educational Researcher, 38 (9), 665-676.
Babin, B. J. & Svensson, G. (2012). Structural equation modeling in social science research:
Issues of validity and reliability in the research process. European Business Review,
24(4), 320-330.
Bagozzi, R. P. and Yi, Y. (1988). On the evaluation of structural equation models. Journal of the
Academy of Marketing Science. 16(1), 74-94.
Bagozzi R. P. & Yi, Y. (2012). Specification, evaluation, and interpretation of structural equation
models. Journal of the Academy of Marketing Science, 40, 8-34.
Bailar, B. A. (1989). “Information Needs, Surveys, and Measurement Errors.” In Panel Surveys,
edited by D. Kasprzyk, G. J. Duncan, G. Kalton, and M. P. Singh, pp. 1-24. New York:
Wiley.
Baillie, P. H. (1993). Understanding retirement from sports: Therapeutic ideas for helping
athletes in transition. The Counseling Psychologist, 21 (3), 399-410.
Baltes, P. B., Reese, H. W., & Lipsitt, L. P. (1980). Life –span developmental psychology.
Annual Review of Psychology, 31, 65-110.
SPORTS RETIREMENT OF COLLEGE ATHLETES 76
Barber, P. (2008, July 1). Game over: For many athletes, the excitement of college
graduation is tinged with melancholy, uncertainty of life after sports. The Press
Democrat, p. C1.
Beamon, K. (2012). “I’m a baller”: Athletic identity foreclosure among African-American
former student-athletes. Journal of African American Studies, 16, 195-208.
Beamon, k. & Bell, P. A. (2006). Academics versus athletics: An examination of the effects of
background and socialization on African American male student-athletes. The Social
Science Journal, 43, 393-403.
Brown, C., & Hartley, D. L. (1998). Athletic identity and career maturity of male college student
athletes. International Journal of Sports Psychology, 29, 17-26.
Chissom, B. S. & Lanier, D. (1975). Prediction of first quarter freshman GPA using SAT scores
and high school grades. Educational and Psychological Measurement. 35, 461-463.
Comeaux, E. & Harrison, C. K. (2011). A conceptual model of academic success for student-
athletes. Educational Researcher, 40(5), 235-245
Crouse, J. & Trusheim, D. (1988). The case against the SAT. Public Interest, 93, 97-110.
DeBeard M. S., Spielmans, G. I., & Julka, D. L. (2004). Predictors of academic achievement and
retention among college freshmen: A longitudinal study. College Student Journal, 38(1),
66-77.
Diseth, A., Danielsen, A. G., & Samdal, O. (2012). A path analysis of basic need support, self-
efficacy, achievement goals, lie satisfaction and academic achievement level among
secondary school students. Educational Psychology, 32(3), 335-354.
Donhardt, G. L. (2004). In search of the effects of academic achievement in post graduation
earnings. Research in Higher Education, 45(3), 271-284.
SPORTS RETIREMENT OF COLLEGE ATHLETES 77
Durkheim, E. (1966). Suicide. New York: The Free Press.
Eitzen, D. S. (2001). Sports in contemporary society. New York: NY, Worth Publishers.
Freedle, R. O. (2003). Correcting the SAT’s ethnic and social-class bias: A method for re-
estimating SAT scores. Harvard Educational Review, 73(1), 1-43.
Gussett, J. C. (1974). College entrance examination board scholastic aptitude test scores as a
predictor for college freshman mathematics grades. Educational and Psychological
Measurement, 34, 953-955.
Gayles, J. G. & Hu, S. (2009). The influence of student engagement and sport participation on
college outcomes among division I student-athletes. The Journal of Higher Education, 80
(3), 315-333.
Hanley, J. A. and McNeil, B. J. (1982). The meaning and use of the area under a “receiver
operating characteristic (ROC) curve,” Radiology, 143, 29–36.
Harrison, L., Sailes, G., Rotich, W. K., & Bimper, A. (2011). Living the dream or
awakening from the nightmare: race and athletic identity. Race, Ethnicity, and
Education, 14, 91–103.
Hart, A. M., Brennan, C. W., Sym, D., and Larson, E. (2009). The impact of personalized pre-
notification on response rates to an electronic survey. Western Journal of Nursing
Research. 31(1), 17-23.
Hatcher, L. (1994). A step-by-step approach to using SAS for factor analysis and structural
equation modeling. Cary, NC: SAS Institute Inc.
Hoffman, J. L. & Lowitzki, K. E. (2005). Predicting college success with high school grades and
test scores: Limitations for minority students. The Review of Higher Education, 28 (4),
455-474.
SPORTS RETIREMENT OF COLLEGE ATHLETES 78
Hosick, M. B. (2010, November 11). NCAA eligibility center up to the task. . Retrieved from
http://www.ncaa.org/wps/wcm/connect/public/NCAA/Resources/Latest News/2010 news
stories/November/NCAA Eligibility Center up to the task
Houle, J. L., Brewer, B. W., Kluck, A. S. (2010). Developmental trends in athletic identity: A
two-part retrospective study. Journal of Sport Behavior, 33(2), 146-159.
Howes, N. J. (1981). Characteristics of career success: An additional input to selecting
candidates for professional programs. Journal of Vocational Behavior, 18, 277-288.
John, O. P. & Benet-Martinez, V. (2000). Measurement: Reliability, construct validation, and
scale construction. In H. T. Reis & C. M. Judd (Eds.), Handbook of Research Methods in
Social and Personality Psychology (pp. 339-369). New York: Cambridge University
Press.
Joreskog, K. G. (1994). On the estimation of polychoric correlations and their asymptotic
covariance matrix. Psychometrika, 59 (3), pp. 381-389.
Joreskog, K. (1994). Structural equation modeling with ordinal variables. Multivariate Analysis
and its Applications. 24 (297-310).
Kane, R. L. & Radosevich, D. M. (2011). Conducting health outcomes research. Sudbury, MA:
Jones & Bartlett Learning.
Kaplowitz, M. D., Hadlock, T. D., and Levine, R. (2004). A comparison of web and mail
survey response rates. The Public Opinion Quarterly. 68(1), 94-101.
Kiernan, N. E., Kiernan, M., Oyler, M. A., and Gilles, C. (2005). Is a web survey as
effective as a mail survey? A field experiment among computer users. American
Journal of Evaluation. 26, 245-252.
SPORTS RETIREMENT OF COLLEGE ATHLETES 79
Kline, R. B. (2011). Principles and practice of structural equation modeling. New York, NY:
Guilford Press.
Kubler-Ross, E. (1996). On death and dying. New York, NY: Macmillian.
Lally, P. S. (2007). Identity and athletic retirement: A prospective study. Psychology of Sport
and Exercise, 8, 85-99.
Lally, P. S. & Kerr, G. A. (2005). The career planning, athletic identity, and student role identity
of intercollegiate student-athletes. Research Quarterly for Exercise and Sport, 76 (3),
257-285.
Linnemeyer, R. M. & Brown, C. (2010. Career maturity and foreclosure in student-athletes, fine
arts students, and general college students. Journal of Career Development, 37, 616-634.
Liu, X., Thomas, S., & Zhang, L. (2010). College quality, earnings, job satisfaction: Evidence
from recent college graduates. Journal of Labor Research, 31, 183-201.
MacCallum, R. C. & Austin, J. T. (2000). Applications of structural equation modeling in
psychological research. Annual Review of Psychology, 51, 201-226.
MacKinnon, D. P., Fairchild, A. J., & Fritz, M. S. (2007). Mediation analysis. Annual Review of
Psychology. 58, 593-914.
McArdle, J. J. (1998). Contemporary statistical models for examining test bias. In J. J. McArdle
& R. W. Woodcock (Eds.), Human cognitive abilities in theory and practice (pp. 157-
195). Mahwah, NJ: Lawrence Earlbaum.
McArdle, J.J. (1996). Current directions in structural factor analysis. Current Directions in
Psychological Science, 5 (1), 11-18.
SPORTS RETIREMENT OF COLLEGE ATHLETES 80
McArdle, J. J., Fisher, G.G., & Kadlec, K.M. (2007). Latent variable analysis of age trends in
tests of cognitive ability in the Health and Retirement Study, 1992-2004. Psychology and
Aging, 22 (3). 525-545.
McArdle, J. J. & Hamagami, F. (1994). Logit and multilevel logit modeling of college
graduation for 1984-1985 freshman student-athletes. Journal of the American Statistical
Association, 89(427), 1107-1123.
McArdle, J.J. Petr, T., Paskus, T. P., Kearns, A. (2007). Educational attainments and outcomes
from a longitudinal survey of NCAA Division I student-athletes ten years after college.
Unpublished manuscript.
McArdle, J.J. & Prescott, C.A. (1992). Age-based construct validation using structural equation
modeling. Experimental Aging Research, 18 (3), 87-115.
McDonald, R. P. (1981). The dimensionality of tests and items. British Journal of Mathematical
and Statistical Psychology, 62, 1067-1074
Miron, P. (2010). Role identity and its implications in the athlete’s personal development.
Timisoara Physical Education and Rehabilitation Journal, 3 (5), 7-12.
Mulaik, S. A., James, L. R., Van Alstine, J., Bennett, N., Lind, S. & Stilwell, C. D. (1989).
Evaluation of goodness-of-fit indices for structural equation models.
Psychological
Bulletin, 105 (3), 430-445.
Munro, B. H. (1981). Dropouts from higher education: Path analysis of a national sample.
American Educational Research Journal, 18(2), 133-141.
Muthen, B. (1984). A general structural equation model with dichotomous, ordered categorical
and continuous latent variable indicators. Psychometrika, 49(1), 115-132.
SPORTS RETIREMENT OF COLLEGE ATHLETES 81
Muthén, B.O. (1998-2004). Mplus Technical Appendices. Los Angeles, CA: Muthén & Muthén
Muthen, B. O., Du Toit, S., & Spisic, D. (1997) Robust inference using weighted least squares
and quadratic estimating equations in latent variable modeling with categorical
and continuous outcomes.
Muthén, L. K., & Muthén, B. O. (1998-2011). Mplus User's Guide. Sixth Edition. Los Angeles,
CA: Muthén & Muthén.
Nagelkerke, N. J. D. (1991), “A Note on a General Definition of the Coefficient of
Determination,” Biometrika, 78, 691–692.
National Collegiate Athletic Association (1990). NCAA Academic Performance Study.
Research Report 90-01. Overland Park, KS.
Park, S., Tod, D., & Lavallee, D. (2012). Exploring the retirement from sport decision-making
process based on the transtheoretical model. Psychology of Sport and Exercise,
13, 444-
453.
Pascarella, E. T. & Terenzini, P. T. (1991). How College Affects Students. San Francisco: Jossey-
Bass.
Paskus, T. S. (2012). A summary and commentary on the quantitative results of current NCAA
academic reforms. Journal of Intercollegiate Sport, 5, 41-53.
Pavot, W. & Diener, E. (1993(. Review of the satisfaction with life scale. Psychological
Assessment, 5 (2), 164-172.
Petr, T.A., & McArdle, J.J. (2012). Academic research and reform: A history of the
empirical basis for NCAA academic policy. Journal of Intercollegiate Sport.
2012(5), 27-40.
SPORTS RETIREMENT OF COLLEGE ATHLETES 82
Preacher, K. J., & Leonardelli, G. J. (2012). Calculation for the sobel test: An interactive
calculation tool for mediation tests. Retrieved from
http://quantpsy.org/sobel/sobel.htm
Raykov, T. & Marcoulides, G. A. (2006). A first course in structural equation modeling.
Mahwah, NJ: Earlbaum.
Reis, H. T., & Gable, S. L. (2000). Event sampling and other methods for studying everyday
experience. In H. T. Reis & C. M. Judd (Eds.), Handbook of Research Methods in Social
and Personality Psychology (pp. 190-222). New York: Cambridge University Press.
Riemer, B. A., Beal, B., & Schroeder, P. (2000). The influences of peer and university culture on
female student-athletes’ perceptions of career termination, professionalization, and social
isolation. Journal of Sport Behavior, 23 (4), 364-378.
Rubio, D. M., Primack, B. A., Switzer, G. E., Bryce, C. L, Seltzer, D. L., & Kapoor, W. N.
(2011). A comprehensive career-success model for physician-scientists. Academic
Medicine, 86 (12), 1571-1576.
Santrock, J.W. (2011). A topical approach to life-span development. McGraw-Hill.
SAS institute Inc. (2009). SAS Online Doc 9.2. Cary, NC: SAS Institute Inc.
Satorra, A. (2000). Scaled and adjusted restricted tests in multi-sample analysis of moment
structures. In Heijmans, R.D.H., Pollock, D.S.G. & Satorra, A. (eds.), Innovations in
multivariate statistical analysis. A Festschrift for Heinz Neudecker (pp.233-247).
London: Kluwer Academic Publishers.
Satorra, A. & Bentler, P. M. (2001). A scaled difference chi-square test statistic for moment
structure analysis. Psychometrika, 66(4), 507-514.
SPORTS RETIREMENT OF COLLEGE ATHLETES 83
Savickas, M. L. (1984). Career maturity: The construct and its measurement. The Vocational
Guidance Quarterly, 32, 222-231.
Schlossberg, N. K. (1981). A model for analyzing human adaptation to transition. The
Counseling Psychologist, 9(2), 2-18.
Schumacker, R. E. & Lomax, R. G. (1996). A beginner’s guide to structural equation modeling.
Mahwah, NJ: Earlbaum.
Shih, T. & Fan, X. (2008). Comparing response rates from web and mail surveys: A
meta- analysis. Field Methods. 20, 249-272.
Singley, D. B., Lent, R. W., & Sheu, H. B. (2010). Longitudinal test of a social cognitive model
of academic and life satisfaction. Journal of Career Assessment, 18, 133-146.
Sobel, M. E. (1982). Asymptotic intervals for indirect effects in structural equations models. In
S. Leinhart (Ed.), Sociological methodology 1982 (pp.290-312). San Francisco: Jossey-
Bass.
Sommer, R. & Sommer, B. (2002). A practical guide to behavioral research: Tools and
techniques. New York, NY: Oxford
Spector, P. (1992). Summated rating scale construction: An introduction. Hillsdale, NJ:
Lawrence Erlbaum.
Sprenkle, D. H. & Piercy, F. P. (2005). Research methods in family therapy. (2
nd
ed., pp.439-
441). New York, NY: Guilford Press.
Stambulova, N., Alfermann, D., Statler, T., & Cote, J. (2009). ISSP position stand: Career
development and transitions of athletes. Journal of Sport and Education Psychology, 7,
395-412.
Steinbach, P. (2000). Life after athletics: Recognizing the needs of a "special population,"
SPORTS RETIREMENT OF COLLEGE ATHLETES 84
UNLV launches a career fair for student-athletes. Athletic Business Magazine,
Retrieved from http://www.unlvrebels.com/genrel/041300aaa.html
Suldo, S. M., Riley, K. N., & Shaffer, E. J. (2006). Academic correlates of children and
adolescents’ life satisfaction. School Psychology International, 27, 567-582.
Taylor, J. & Ogilvie, B. C. (1994). A conceptual model of adaptation to retirement among
athletes. Journal of Sport and Exercise Psychology, 6, 1-20.
Terenzini, P. T. & Pascarella, E. T. (1980). Toward the validation of Tinto’s model of college
student attrition: A review of recent studies. Research in Higher Education, 12(3), 271-
282.
Tinto, V. (1975). Dropout from higher education: A theoretical synthesis of recent research.
Review of Educational Research, 45(1), 89-125.
University of California Los Angeles : Statistical Consulting Group. (2013). Analyzing data:
Path analysis. http://www.ats.ucla.edu/stat/mplus/seminars/introMplus_part2/path.htm
(accessed February 7, 2013).
University of Michigan. (2013, April 09). Monitoring the future: A continuing study of American
youth. Retrieved from http://www.monitoringthefuture.org/
University of Michigan. (2013). Health and retirement study: A longitudinal study of health,
retirement, and aging. Retrieved from http://hrsonline.isr.umich.edu/index.php
United States Census Bureau (2012, December 17).Response rates definitions. Retrieved from
http://www.census.gov/acs/www/methodology/response_rates_definitions/
Virick, M., DaSilva, N., & Arrington, K. (2010). Moderators of the curvilinear relation between
extent of telecommuting and job and life satisfaction: The role of performance outcome
orientation and worker type. Human Relations, 63, 137-154.
SPORTS RETIREMENT OF COLLEGE ATHLETES 85
Waugh, R. F. & Chapman, E. S. (2005). An analysis of dimensionality using factor analysis
(true-score theory) and rasch measurement: What is the difference? Which method is
better? Journal of Applied Measurement, 6(1), 80-99.
Webb, W. M., Nasco, S. A., Riley, S., & Headrick, B. (1998). Athlete identity and reactions to
retirement from sports. Journal of Sport Behavior, 21 (3), 338-362.
Wilcox, R. (2012). Modern statistics for the social and behavioral sciences. Boca Raton, FL:
Taylor & Francis Group.
Willits, F. K. (1988). Adolescent behavior and adult success and well-being: A 37-year panel
study. Youth Society, 20, 68-87.
Wolverton, B. (2008). Athletes’ hours renew debate over college sports. The Chronicle of Higher
Education, 54, A1- A23.
Wylleman, P., Alfermann, D., & Lavallee, D. (2004). Career transitions in sport. European
perspectives. Psychology of Sport and Exercise, 5, 7-20.
Wylleman, P. & Reints, A. (2010). A lifespan perspective on the career of talented and elite
athletes: Perspectives on high-intensity sports. Scandinavian Journal of Medicine &
Science in Sports, 20, 88-94.
Yetter, G. & Capaccioli, K. (2010). Differences in responses to web and paper surveys
among school professionals. Behavior Research Methods. 42I(1), 266-272.
SPORTS RETIREMENT OF COLLEGE ATHLETES 86
Table 1
Initial Eligibility Clearinghouse and Academic Performance Study Variables
Variable Description N Min Max Mean SD
Male Gender (0=female, 1=male) 24336 0 1 0.62 0.48
Minority Race (0=white, 1=minority) 18100 0 1 0.35 0.47
HSCGPA High school GPA 23942 1 5 3.07 0.64
TestRC Best ACT/SAT score 23942 -0.8 5 2.13 0.55
Coregpa Core college GPA 23321 0.99 5 2.99 0.77
fmin Family median income 23000 0 150 43.00 16.40
SportRC Sport re-coded 24336 0 9 4.3 3.0
SPORTS RETIREMENT OF COLLEGE ATHLETES 87
Table 2
SCORE Section and Variable Descriptions
Section # Section Name Example of Items
1 College and Sports Experiences Years participation, professional
athlete aspirations and thoughts,
athletic awards and effects of athletes
experience on academics
2 College Educational Experiences Degrees earned, major chosen, final
GPA, and satisfaction with the college
experience.
3 Current Career and Work Experiences Annual earnings and feelings about
current job
4 Health and Well-being Happiness and life satisfaction, sports
related injuries and medical
conditions, and drug/alcohol use.
5 Daily Life Experiences
Social support, significant others and
friends, and being successful in life
6 Background Information Marital status, gender, race and
number of children
SPORTS RETIREMENT OF COLLEGE ATHLETES 88
Table 3
Likelihood of Playing Professionally and Identity Variables from SCORE (minimum =1 and
maximum =6 for all except Pq11 which is 0 = “No” and 1= “Yes”)
Variable Description
Available
N
Mean SD
Pq10 When you entered college, how likely
did you think it was that you would
become a professional athlete?
6399 2.69 1.50
Pq11 At any point since leaving college, have
you competed at a professional and/or
Olympic level?
6636 .14 .35
Pq15a I consider myself a dedicated athlete 6376 5.41 0.83
Pq15b I consider myself a dedicated student 6370 4.98 1.10
Pq15c I had many personal goals related to my
sport
6359 5.08 1.00
Pq15d I had many personal goals related to my
academics
6360 4.90 1.10
Pq15e I needed to excel in athletic pursuits to
feel good about myself
6324 4.34 1.30
Pq15f I needed to excel in academic pursuits to
feel good about my self
6358 4.42 1.20
Pq15g My sports experiences were an important
part of my overall college experience
6350 5.43 0.92
Pq15h My academic experiences were an
important part of my overall college
experience
6362 4.97 1.00
Pq15i I felt other students viewed me as more
of an athlete than as a student
6333 4.12 1.40
Pq15j I felt that my professors viewed me as
more of an athlete than as a student
6348 3.44 1.40
Pq15K I felt that some of my professors
discriminated against me because I was
an athlete
6346 2.75 1.60
Pq15l I felt that some of my professors favored
me because I was an athlete
6361 2.79 1.40
SPORTS RETIREMENT OF COLLEGE ATHLETES 89
Table 4
Retirement Difficulty and Life Outcome Variables
Variable Description
N Min Max Mean
SD
Pq26 How difficult was it for you to retire
from competing seriously in your
sport?
5997 1 5 3.34 1.3
Pq51a Satisfaction with pay 5969 1 5 3.83 1.2
Pq51b Satisfaction with benefit package 5945 1 5 3.97 1.2
Pq51c Satisfaction with importance and
challenge of your work
5961 1 5 4.25 1.0
Pq51d Satisfaction with opportunities for
promotion and advancement
5952 1 5 3.78 1.2
Pq51e Satisfaction with opportunities to use
past training and education
5951 1 5 4.06 1.0
Pq51f Satisfaction with job security 5953 1 5 4.13 1.1
Pq58a Satisfaction with your life as a whole
these days
6558 1 6 5.1 .79
Pq58b Satisfaction with your job/career 6176 1 6 4.47 1.0
Pq58c Satisfaction with the neighborhood
where you life
6551 1 6 4.97 .98
Pq58d Satisfaction with you education level 6554 1 6 5.08 .95
Pq58e Satisfaction with your friendships 6544 1 6 5.17 .83
Pq58f Satisfaction with your romantic
relationships
6321 1 6 5.29 1.1
Pq58g satisfaction with your standard of
living
6551 1 6 5.05 .95
Pq58h Satisfaction with the way you spend
your leisure time
6549 1 6 4.93 .94
Pq58i Satisfaction with your
spiritual/religious life
6447 1 6 4.83 .97
SPORTS RETIREMENT OF COLLEGE ATHLETES 90
Table 5
Response rate by Race and Gender
Group Sampled Responded
Response
Rate
Blacks 5,410 797 15%
Whites 15,598 4,857 31%
Asians 467 121 26%
Hispanics 877 204 23%
Black Males 4,111 520 13%
White Males 8,931 2,248 25%
Asian Males 264 57 22%
Hispanic Males 556 106 19%
Black Females 1,299 277 21%
White Females 6,667 2,609 39%
Asian Females 203 64 32%
Hispanic Females 321 98 31%
Males 14,172 3,013 21%
Females 8,696 3,103 36%
SPORTS RETIREMENT OF COLLEGE ATHLETES 91
Table 6
Response Rates by Sport
Group Sampled Responded
Response
Rate
M football 5,962 1,071 18.0%
M track 1,648 394 23.9%
M soccer 996 231 23.2%
M swimming 687 200 29.1%
M basketball 1,430 212 14.8%
M baseball 1,843 440 23.9%
M volleyball 51 14 27.5%
M other 2,409 545 22.6%
W track 1,670 533 31.9%
W soccer 1,234 454 36.8%
W swimming 821 340 41.4%
W basketball 1,160 346 29.8%
W softball 928 334 36.0%
W volleyball 838 344 41.1%
W other 2,266 785 34.6%
Total 23,943 6,243 26.1%
Missing 393 393 100.0%
Total 24,336 6,636 27.3%
SPORTS RETIREMENT OF COLLEGE ATHLETES 92
Table 7
Creation of Sampling Weights
Group
Sample
Proportion
Possible
Response
Rate
(RR=n/N)
Response
Bias
(RR/n%)
Sampling
Weight
n
Weights
summed
to = n
Total Sample
24,336
(100%)
6,636
(27.2%)
M Football 24.5% 16.1% 0.66 1.52 1,071 1625.73
M Track 6.8% 5.9% 0.88
1.14
394 449.38
M soccer 4.1% 3.5% 0.85 1.18 231 271.59
M Swimming 2.8% 3.0% 1.07 0.94 200 187.33
M Basketball 5.9% 3.2% 0.54 1.84 212 389.94
M Baseball 7.6% 6.6% 0.88
1.14
440 502.55
M Volleyball 0.2% 0.2% 1.01 0.99 14 13.91
M Other 9.9% 8.2% 0.83 1.21 545 656.89
W Track 6.9% 8.0% 1.17 0.85 533 455.38
W Soccer 5.1% 6.8% 1.35 0.74 454 336.49
W Swimming 3.4% 5.1% 1.52
0.66
340 223.87
W Basketball 4.8% 5.2% 1.09 0.91 346 316.31
W Softball 3.8% 5.0% 1.32 0.76 334 253.05
W Volleyball 3.4% 5.2% 1.51 0.66 344 228.51
W Other 9.3% 11.8% 1.27 0.79 785 617.90
Missing 1.6% 5.9% 3.67 0.27 393 107.16
TOTALS 100.0% 100.0% 15.94 15.33 6,636 6636.00
SPORTS RETIREMENT OF COLLEGE ATHLETES 93
Table 8
Odds Ratios for Demographic Variables in the Longitudinal Prediction of Response (0,1) to
SCORE
Model 1
Model 2
Variable
Odds Ratio
Estimates
95% Wald
Confidence
Limits
Odds Ratio
Estimates
95% Wald
Confidence
Limits
HscGPA 1.8 1.7 2.1 1.9 1.6 2.0
Minority 0.7 0.6 0.7 0.7 0.6 0.7
Male 0.6 0.6 0.7 0.6 0.6 0.7
Income 1.0 1.0 1.0 1.0 1.0 1.0
CoreGPA 0.9 0.9 1.0 0.9 0.9 1.0
Test
scaled
1.2 1.1 1.3 1.2 1.1 1.3
SPORTS RETIREMENT OF COLLEGE ATHLETES 94
Table 9
Model Fit Statistics for Identity and Life Outcome Factor Models
Model N L
2
df CFI TLI RMSEA
1.1a-1 factor identity model 5491 20973 54 0.28 0.12 0.27
1.1b-2 factor identity model 5491 9989 53 0.66 0.57 0.19
1.1c-3 factor identity model 5491 4248 51 0.86 0.81 0.12
1.1d-3 factor identity model minus q15e 5491 1893 41 0.93 0.90 0.09
1.1e-Inentity invariance model for gender 5491 2660 118 0.90 0.90 0.09
1.1f-Identity invariance model for race 4094 1752 118 0.91 0.92 0.08
1.2a-1 factor life outcomes model 5699 6259 90 0.70 0.66 0.11
1.2b-2 factor life outcomes model 5699 3934 89 0.82 0.78 0.09
1.2c-2 factor life outcomes minus q58b and f 5699 2283 76 0.89 0.86 0.07
1.2d-2 factor life outcomes minus q51c 5699 1893 64 0.89 0.87 0.07
1.2e-Life outcomes invariance for gender 5699 3022 168 0.88 0.88 0.08
1.2f-Life outcomes invariance for race 4255 2316 168 0.87 0.88 0.08
SPORTS RETIREMENT OF COLLEGE ATHLETES 95
Table 10
Parameter Estimates for Models Predicting Retirement Difficulties with and without Weights
Model 2.1a Categorical no
Weights
Model 2.1b Categorical with Weights
Variable Est Std Est Sd Error P Est Std Est Sd Error P
HscGPA -0.096 -0.053 0.049 .05 -0.107 -0.06 0.051 .03
Gender 0.128 0.063 0.036 0 0.137 0.067 0.036 0
Race -0.134 -0.051 0.047 .003 -0.141 -0.056 0.046 <.05
Test scores -0.149 -0.078 0.042 0 -0.158 -0.083 0.044 0
CoreGPA -0.026 -0.018 0.034 0.44 -0.024 -0.017 0.035 .50
Fmin -0.043 -0.074 0.01 0 -0.043 -0.075 0.01 0
SPORTS RETIREMENT OF COLLEGE ATHLETES 96
Table 11
Model Fit Statistics for Path Models Predicting Difficulty With Retiring from Sports Competition
Model N χ
2
df CFI TLI RMSEA R
2
2.1a-Base model categorical 3846 0 0 1.00 1.00 0.00 0.03
2.1b- Base model with
weights categorical
3846 0 0 1.00 1.00 0.00 0.04
2.1c-Multi group model for
gender categorical
3846 12 9 0.92 0.91 0.01 ~
females 2155 6 ~ ~ ~ ~ 0.03
males 1691 6 ~ ~ ~ ~ 0.03
2.1d-Multigroup model for
race
3414 29 9 0.56 0.51 0.03 ~
minority 722 22 ~ ~ ~ ~ 0.04
white 3124 6 ~ ~ ~ ~ 0.03
2.1e-Multigroup model for
race with free thresholds
3414 10 5 0.89 0.78 0.02 ~
minority 722 8 ~ ~ ~ ~ 0.04
white 3124 2 ~ ~ ~ ~ 0.03
2.2a-Baseline model adding
pro variable
3842 0 0 1 1 0 0.078
2.2b-Complete invariance
model for gender adding pro
variable
3842 11 10 0.994 0.993 0.008 ~
females 2151 6 ~ ~ ~ ~ 0.066
males 1691 6 ~ ~ ~ ~ 0.081
2.2c-Complete invariance
model for race adding pro
variable
3842 58 10 0.76 0.71 0.05 ~
minority 722 47 ~ ~ ~ ~ 0.1
white 3120 11 ~ ~ ~ ~ 0.07
2.2d-Complete invariance
model for race adding pro
variable with free thresholds
3842 21 6 0.92 0.85 0.04 ~
minority 722 17 ~ ~ ~ ~ 0.1
white 3120 4 ~ ~ ~ ~ 0.07
SPORTS RETIREMENT OF COLLEGE ATHLETES 97
Table 12
Fit Statistics for Full Path Models
Model N L
2
df CFI TLI RMSEA
3.1 Continuous Retirement Variable
3917 2111 166 0.862 0.843 0.06
3.2 Categorical Retirement Variable
3917 1266 166 0.93 0.91 0.04
3.3 7 Exogenous to all Endogenous
3917 1429 152 0.913 0.89 0.05
3.4 Zero Paths from Retirement to
Outcomes 3917 1360 154 0.918 0.9 0.05
3.5 Reduced Item Outcomes
3917 562 55 0.93 0.9 0.05
3.6 Adding Identity Predictors
3920 3417 250 0.83 0.79 0.06
3.7 Zero Paths from Retirement to
Outcomes 3920 3462 252 0.82 0.79 0.06
3.8 MGFI Gender
3920 2769 559 0.87 0.87 0.05
3.9 MGFI Gender & Zero Paths from
Retirement to Outcomes
3920 2819 561 0.87 0.87 0.05
3.10 MGIF Race
3920 2072 559 0.88 0.88 0.04
3.11 MGFI Race and Zero Paths from
Retirement to Outcomes
3920 2753 561 0.87 0.87 0.05
SPORTS RETIREMENT OF COLLEGE ATHLETES 98
Table 13
Variance Accounted for in Outcome Variables of the Full Path Model
Model R
2
F1 R
2
F2 R
2
q26
3.1 Continuous Retirement Variable
0.002 0.01 0.06
3.2 Categorical Retirement Variable
0.001 0.01 0.09
3.3 7 Exogenous to all Endogenous
0.05 0.03 0.07
3.4 Zero Paths from Retirement to Outcomes
0.03 0.05 0.07
3.5 Reduced Item Outcomes
0.03 0.04 0.07
3.6 Adding Identity Predictors
0.08 0.1 0.11
3.7 Zero Paths from Retirement to Outcomes
0.08 0.08 0.11
SPORTS RETIREMENT OF COLLEGE ATHLETES 99
Table 14
Mediation Analysis Results
Path a b s
a
b
a
z S.E. p
F3 -> RD -> F1 -0.065* -0.06 0.021 0.018 2.09 0.002 0.04*
F4 -> RD -> F1 0.207* -0.06 0.021 0.018 -2.86 0.004 <.01*
hscGPA -> RD -> F1 -0.114* -0.06 0.050 0.018 1.82 0.004 0.07
male -> RD -> F1 0.043 -0.06 0.037 0.018 -1.01 0.002 0.31
minority -> RD ->F1 -0.169* -0.06 0.046 0.018 2.24 0.005 0.03*
test -> RD -> F1 -0.089* -0.06 0.043 0.018 1.96 0.003 0.09
coreGPA -> RD -> F1 -0.008 -0.06 0.035 0.018 0.20 0.002 0.84
fmin -> RD -> F1 0.000 -0.06 0.001 0.018 0.00 <.001 1.00
pro -> RD -> F1 0.166* -0.06 0.013 0.018 -2.91 0.003 <.01*
F3 -> RD -> F2 -0.065* -0.12 0.021 0.019 2.45 0.003 0.01*
F4 -> RD -> F2 0.207* -0.12 0.021 0.019 -4.88 0.005 <.01*
hscGPA -> RD -> F2 -0.114* -0.12 0.050 0.019 2.04 0.007 0.04*
male -> RD -> F2 0.043 -0.12 0.037 0.019 -0.11 0.005 0.29
minority -> RD ->F2 -0.169* -0.12 0.046 0.019 3.04 0.007 <.01*
test -> RD -> F2 -0.089* -0.12 0.043 0.019 1.91 0.006 0.06
coreGPA -> RD -> F2 -0.008 -0.12 0.035 0.019 0.20 0.004 0.84
fmin -> RD -> F2 0.000 -0.12 0.001 0.019 0.00 <.001 1.00
pro -> RD -> F2 0.166* -0.12 0.013 0.019 -5.15 0.004 <.001*
Note: a=the unstandardized regression estimate for the prediction of retirement, b is the
unstandardized regression estimate for the prediction of life outcomes from retirement,
s
a
is the standard error of a, and s
b
is the standard error of b.
F1=Job satisfaction
F2=Life satisfaction
F3=Academic identity
F4=Athletic identity
RD=Retirement Difficulties
*=significant at p< .05
SPORTS RETIREMENT OF COLLEGE ATHLETES 100
Figure 1. Tinto’s Conceptual Model of College Dropout.
SPORTS RETIREMENT OF COLLEGE ATHLETES 101
Figure 2. Comeaux & Harrison’s Model of College Student-athlete Academic Success
Comeaux E , Harrison C K
EDUCATIONAL RESEARCHER
2011;40:235-245
Copyright © by American Educational Research
Association
SPORTS RETIREMENT OF COLLEGE ATHLETES 102
Figure 3. A Lifespan Perspective of the concept of “ Sport Retirement.” (Wylleman et. al., 2004)
SPORTS RETIREMENT OF COLLEGE ATHLETES 103
Figure 4. A Transitional Model of the impacts of Demographics, Academic Performance, and
Identity on Life Outcomes While Accounting for Difficulties Retiring from Sports Competition
for Former Student-athletes.
Job
Satisfaction
Life
Satisfaction
Identity
Academic
Performance
Difficulty
Retiring
Demographics
SPORTS RETIREMENT OF COLLEGE ATHLETES 104
Figure 5. Two Factor Model of Life Outcomes. (Notes: N=5,699; Values are Standardized
estimates)
SPORTS RETIREMENT OF COLLEGE ATHLETES 105
Figure 6. Histograms of Difficulty Retiring from Competition for Males and Females.
Retirement Difficulty for Females
Dificulty Retiring from Competition for Females
Frequency of Observations
1 2 3 4 5
0 200 400 600 800
Retirement Difficulty for Males
Dificulty Retiring from Competition for Males
Frequency of Observations
1 2 3 4 5
0 200 400 600 800
SPORTS RETIREMENT OF COLLEGE ATHLETES 106
Figure 7. Histograms of Difficulty Retiring from Competition for Males by Sport.
Football Players
Frequency
1 2 3 4 5
0 200 400 600 800
Male Track Athletes
Frequency
1 2 3 4 5
0 200 600
Male Soccer Players
Frequency
1 2 3 4 5
0 200 600
Male Swimmers
Frequency
1 2 3 4 5
0 200 600
Male Basketball Players
Frequency
1 2 3 4 5
0 200 600
Baseball Players
Frequency
1 2 3 4 5
0 200 400 600 800
Male Volleyball Players
Frequency
1 2 3 4 5
0 200 600
Males from Other Sports
Frequency
1 2 3 4 5
0 200 600
SPORTS RETIREMENT OF COLLEGE ATHLETES 107
Figure 8. Histograms of Difficulty Retiring from Competition for Females by Sport.
Female Track Athletes
Frequency
1 2 3 4 5
0 200 400 600 800
Female Soccer Players
Frequency
1 2 3 4 5
0 200 400 600 800
Female Swimmers
Frequency
1 2 3 4 5
0 200 400 600 800
Female Basketball Players
Frequency
1 2 3 4 5
0 200 400 600 800
Softball Players
Frequency
1 2 3 4 5
0 200 400 600 800
Female Volleyball Players
Frequency
1 2 3 4 5
0 200 400 600 800
Females from Other Sports
Frequency
1 2 3 4 5
0 200 400 600 800
SPORTS RETIREMENT OF COLLEGE ATHLETES 108
Figure 9. Histograms of Difficulty Retiring from Competition for Athletes who have Competed
at a Professional and or/ Olympic Level and Those who have Not.
Difficulty with Sports Retirement
Non Professional/Olympic Athletes
Frequency
1 2 3 4 5
0 200 600 1000 1400
Difficulty with Sports Retirement
Professional/Olympic Athletes
Frequency
1 2 3 4 5
0 50 100 150 200 250
SPORTS RETIREMENT OF COLLEGE ATHLETES 109
Figure 10. Kernel Density Plots of Difficulty Retiring from Competition for Athletes who have
Competed at a Professional and or/ Olympic Level and Those who have Not.
SPORTS RETIREMENT OF COLLEGE ATHLETES 110
Figure 11. Histograms of Difficulty Retiring from Competition by Likelihood of Becoming a
Professional Athlete.
Difficulty with Sports Retirement
Very Likely
Frequency
1 2 3 4 5
0 50 100 150
Difficulty with Sports Retirement
Likely
Frequency
1 2 3 4 5
0 50 100 150
Difficulty with Sports Retirement
Somewhat Likely
Frequency
1 2 3 4 5
0 100 200 300
Difficulty with Sports Retirement
Somewhat unlikely
Frequency
1 2 3 4 5
0 50 150 250
Difficulty with Sports Retirement
Unlikely
Frequency
1 2 3 4 5
0 100 300
Difficulty with Sports Retirement
Very Unlikely
Frequency
1 2 3 4 5
0 100 300
SPORTS RETIREMENT OF COLLEGE ATHLETES 111
Figure 12. Path Diagram of the Model of Participant Characteristics Predicting Retirement
Difficulty without Using Sampling Weights. (Notes: N=3,846; Values are raw maximum
likelihood estimates with Standardized values in parenthesis. The asterisk indicates significant
paths).
SPORTS RETIREMENT OF COLLEGE ATHLETES 112
Figure 13. Path Model of Participant Characteristics Predicting Retirement Difficulty with
Weights. (Notes: N=3,846; Values are raw maximum likelihood estimates with Standardized
values in parenthesis. The asterisk indicates significant paths).
SPORTS RETIREMENT OF COLLEGE ATHLETES 113
Figure 14. Full Path Model Assessing the Relationship between Participant Characteristics and
Life Outcomes while Accounting for the Impact of Retirement Difficulties. (Notes: N=3,917;
Values are standardized estimates. The asterisk indicates significant paths).
SPORTS RETIREMENT OF COLLEGE ATHLETES 114
Figure 15. Path Model with Participant Character Variables Predicting Retirement Difficulties
and Life Outcomes (Notes: N=3,917; Values are standardized estimates. The asterisk and bold
lines indicate significant paths).
SPORTS RETIREMENT OF COLLEGE ATHLETES 115
Figure 16. Path Model with Participant Character Variables Predicting Retirement Difficulties
and Life Outcomes and the Paths from Retirement Difficulties to Life Outcomes Fixed to Zero
(Notes: N=3,917; Values are standardized estimates. The asterisk and bold lines indicate
significant paths).
SPORTS RETIREMENT OF COLLEGE ATHLETES 116
Figure 17. Path Model with Participant Character Variables Predicting Retirement Difficulties
and Life Outcomes.
SPORTS RETIREMENT OF COLLEGE ATHLETES 117
Figure 18. A proposed path diagram assessing the longitudinal relationship between high school
academics, college outcomes, retirement from sport and life outcomes of former student-athletes
mimicking that of Tinto (1975) and Comeaux & Harrison (2011).
SPORTS RETIREMENT OF COLLEGE ATHLETES 118
Technical Appendix
Appendix A: SCORE Survey
SPORTS RETIREMENT OF COLLEGE ATHLETES 119
SPORTS RETIREMENT OF COLLEGE ATHLETES 120
SPORTS RETIREMENT OF COLLEGE ATHLETES 121
SPORTS RETIREMENT OF COLLEGE ATHLETES 122
SPORTS RETIREMENT OF COLLEGE ATHLETES 123
SPORTS RETIREMENT OF COLLEGE ATHLETES 124
SPORTS RETIREMENT OF COLLEGE ATHLETES 125
SPORTS RETIREMENT OF COLLEGE ATHLETES 126
SPORTS RETIREMENT OF COLLEGE ATHLETES 127
SPORTS RETIREMENT OF COLLEGE ATHLETES 128
SPORTS RETIREMENT OF COLLEGE ATHLETES 129
SPORTS RETIREMENT OF COLLEGE ATHLETES 130
SPORTS RETIREMENT OF COLLEGE ATHLETES 131
SPORTS RETIREMENT OF COLLEGE ATHLETES 132
SPORTS RETIREMENT OF COLLEGE ATHLETES 133
Appendix B: Variance and Covariance Defined
The variance is defined as follows
(1)
where X
i
represents the ith exogenous variable in the model,
represents the mean of item X ,
and N is the sample size. So the variance is a function of the sum of each value of a variable
minus the mean of that variable squared, divided by the sample size minus 1.
The covariance represents the relationship between items or variables in unstandardized form, as
opposed to a correlation which is the association between two items in standardized form. It can
be specified as follows
(2)
where X and Y are variables in the model and N is the sample size. The covariance then, is the
sum of each value of X minus its mean multiplied by the sum of each value of Y minus its mean all
divided by the sample size minus one. In SEM, the variances and covariances are collectively assessed
in terms of matrices. For a model with three variables (x
1
, x
2
, and x
3
), the variance/covariance matrix can
be defined as follows
(3)
Where variances for each variable are represented on the diagonal axis and covariances are represent on
the off diagonal.
SPORTS RETIREMENT OF COLLEGE ATHLETES 134
Appendix C: Structural Regression Algorithm
If we look at a simple two variable model of means and covariance with X being the exogenous
predictor and Y being the endogenous outcome variable, and a constant of one included for the
mean, the structural regression of this model can be specified with the following
(4)
where
is the intercept term or the value of the outcome variable Y when the predictor variable
is zero,
is the slope term interpreted as the amount of change in the outcome Y for every
single unit difference in the measured predictor X, and e is the unobserved residual or the part of
the outcome variable Y that is not accounted for by the predictor X. Each model has a set of
structural expectations for the variance and covariance as well as for the mean if included. The
difference between the estimated parameters and the covariance and mean parameters or the
observed statistics is the misfit of the model. Because all of the parameters in this model are
obtained from the observed variables there is no misfit of the model (Mulaik, James, Van
Alstine, Bennett, Lind, & Stilwell, 1989). This does not however mean that the model fits, it just
implies that the model cannot be evaluated using the fit statistics provided.
SPORTS RETIREMENT OF COLLEGE ATHLETES 135
Appendix D: Sampling Bias
The sampling biases can be looked at with a series of simple calculations and tables. First,
response rates by race and gender are assessed (see table 5). Response rates can be calculated
using a simple equation
(6)
where RR represents the response rate for a particular group of participants, n
i
represents the
sample size for a particular group of participants, and n
ir
represents the number of participants
who responded to SCORE for that particular group.
SPORTS RETIREMENT OF COLLEGE ATHLETES 136
Appendix E: Sampling Weights
Sampling weights can be calculated using a sequence of equations
(7)
Where the P is the population proportion for a particular group, n
i
is the sample size for that
group in the population, and N is the sample size for the entire population. Sample proportion or
response rate is also needed for sampling weights and is represented in equation 7 above. The
sampling weight is a function of the population proportion and the sample proportion or response
rate
(8)
Where W is the sampling weight, S is the response rate or sample proportion, n
i
is the sample
size of a specified group in the population, n
ir
is the number of participants in group i who
responded to the survey, P is the population proportion, and N is the total population. A
sampling weight of 1 would indicate a group that is neither overrepresented nor underrepresented
in the population. Weights above 1 indicate groups that are underrepresented in the population
and thus would count more in analyses and weights lower than 1 indicate groups that are
overrepresented in the population and thus would count less in analyses.
Table 7 shows the population proportion, sample proportion, response bias, sampling
weights and n for male and female sport groups. If we look at the first line of information for
men’s football, we see that male football players represent 24.5% of the total population
sampled, they responded to SCORE at a rate of approximately 16% (which again is lower than
the total population response rate), their response bias is .66 indicating underrepresentation in the
population of responders, their sampling weight is 1.52 which is above one and indicates that
scores from this group would be counted more in analyses. In the table, the sample size (n) and
SPORTS RETIREMENT OF COLLEGE ATHLETES 137
weights (summed to n) show the amount of male football players who initially responded and
what the sample size would be using the weight. So in analyses, results with weights would be
calculated as though there were approximately 600 more male football players in the sample.
Essentially, with the weights calculated, the sample would resemble the population originally
sampled rather than the sample that actually responded to the survey. See Stapleton, 2002.
SPORTS RETIREMENT OF COLLEGE ATHLETES 138
Appendix F: Logistic Regression Algorithm
A logistic regression model with multiple predictors can be expressed with the following
equations:
(9)
(10)
where the probability that Y=1 is referred to as Y
and 1-Y
is the probability that Y=0, r is the
linear regression equation where
is the constant,
are the regression coefficients (β
1
, β
2
,
β
3
,…β
k
), and X
1-k
are the predictors (X
1
, X
2
, X
3
,…X
k
),. The ln symbol refers to the natural log
and the natural log of the odds (the probability of being in one group divided by the probability
of being in the other group) is equal to the linear regression equation (see Stevens, 2009;
Tabachnick & Fidell, 2007; Hosmer & Lemeshow, 2000; Long, 1997).
SPORTS RETIREMENT OF COLLEGE ATHLETES 139
Appendix G: Pseudo R
2
and the ROC Curve
The R
2
statistic is used to determine variance accounted for in the outcome variable by the
predictor variables and is also a useful tool for comparing alternative models. For logistic
regression, the R
2
is calculated with the following equation
(11)
Where
is an adjusted value which was proposed by Nagelkerke (1991) and is a function of
and
, which were proposed by Cox and Snell (1989, pp208-209). In this equation, n is the
sample size, is the likelihood of the intercept only model, and
is the likelihood of the
specified model. Here,
gives the maximum value of
. This is a value less than one for
discrete models. The adjusted coefficient,
, can achieve a maximum value of one. Larger
values are an indication of a better model. The C statistic is a measure of specificity and
sensitivity and represents the area under a receiver operator curve (ROC). It is a way to
determine how well a model does at predicting membership in a particular outcome category. In
this case, the C statistic represents how well the predictors do at classifying or differentiating
between participants who responded and those who did not. The C statistic for the area under the
receiver operator curve is defined by the following equation
(12)
Where c is a value from the concordance index, t is the total number of pairs with different
responses,
is the total number of pairs that are concordant and
is the total number of pairs
that are discordant (Hanley & McNeil, 1982). Concordant pairs refers to a situation where a pair
of observations has different responses and the response with the lower ordered response has a
lower predicted mean score than the response with the higher ordered response. A pair is
SPORTS RETIREMENT OF COLLEGE ATHLETES 140
discordant when the lower ordered response has a higher predicted mean and the higher ordered
response has a lower predicted mean. It is considered a tie if the pair is neither concordant nor
discordant (SAS Institute Inc., 2009). This statistic generally ranges from .5 to 1 where a model
with random classification results in C= .5 and a perfect model results in C=1.
SPORTS RETIREMENT OF COLLEGE ATHLETES 141
Appendix H: Odds Ratios for Logistic Regression
If we expand on logistic regression equations 9 and 10 in Appendix F, odds ratios for the
predictors would be defined as follows
(13)
where Y is the outcome that the odds ratio is based on, X is a dichotomous predictor variable
with values of X=0 and X=1, and as with previous models,
is the constant and
is the
regression coefficient. The odds ratio, unless otherwise specified, represents the change in the
odds ratio for a 1 unit increase in the predictor. For this data, it represented the change in the
odds of response for a 1 unit increase in the predictor variable. Confidence intervals around the
odds ratio which contain 1 were considered to be non significant and those not containing 1 were
considered significant odds (see SAS Institute Inc, 2009).
SPORTS RETIREMENT OF COLLEGE ATHLETES 142
Appendix I: Model Fit Indices
The likelihood (L
2
) is estimated as a function of the difference between the proposed model and
baseline model covariance matrices, as well as the sample size. If the residuals are normally
distributed this L
2
2
with appropriate dfs. Lower values indicate a better
fitting model. The CFI assess increments in model fit compared to the baseline model. It’s a
function of the difference between the chi-square and degrees of freedom for the baseline model
minus the difference between the chi-square and the degrees of freedom for the proposed model,
all divided by the difference between the chi-square and degrees of freedom for the baseline
model
(14)
Values closer to one on the CFI indicate better fit. The TLI is a function of the difference
between the ratio of chi-square to degrees of freedom for the baseline model and the proposed
model, divided by the ratio of chi-square to degrees of freedom for the baseline model minus 1
(see equation 10)
(15)
This index has a penalty for adding parameters. Values closer to 1 indicate better fit. The
RMSEA is a function of the ratio of chi-square to degrees of freedom as well as the sample size
(16)
Values closer to zero indicate better fit (Kline, 2011).
Nested models were compared following the work of Satorra & Bentler 2001; also see Satorra,
2000) using the scaled chi-square difference test. This can be used with robust models in Mplus
SPORTS RETIREMENT OF COLLEGE ATHLETES 143
(MLM and MLR). Mplus uses a correction for models run with continuous variables. To
calculate the difference, the correction cd needs to be calculated using the following equation
(17)
Where d0 is the degrees of freedom for the nested model, c0 is the scaling factor for the nested
model, d1 is the degrees of freedom in the comparison model, and c1 is the scaling correction
factor for the comparison model. Next, the Satorra –Bentler test of the difference between
likelihoods is calculated as follows
(18)
Where t0 is the robust likelihood value for the nested model, c0 is the scaling correction factor
for the nested model, t1 is the likelihood value for the comparison model, c1 is the scaling
correction factor for the comparison model, and cd is the scaling correction for the difference.
For models with categorical variables in Mplus which use the WLSMV estimator, the
DIFFTEST option is specified to calculate the difference test. If the difference between two
nested models is significant, it is interpreted that one model is significantly better or worse than
the other and the better model is preferred.
SPORTS RETIREMENT OF COLLEGE ATHLETES 144
Appendix J: Estimation of Ordinal Variables in SEM
The estimation of ordinal variables in SEM is different than for continuous variables because
mainly ordinal variables do not have units of measurement or origins and according to Joreskog
(1994), their means, variance, and covariance have no meaning. In models with ordinal items,
polychoric correlations (method to obtain correlations between ordinal variables where the
estimate is based on what the correlation would be if categories were on a continuous scale;
(Joreskog, 1994) or some other type of correlation used for ordered data are estimated and
parameters are estimated using a weighted least squares (WLS) estimator rather than maximum
likelihood (ML) which is typically used with continuous outcomes. Because the ordinal outcome
variable has different levels, they are viewed as thresholds in the output. The number of
thresholds is one minus the number of levels. So, for an ordinal variable on a five point likert
scale, there would be four thresholds provided in output. The thresholds are points that
discriminate one group or level from another. They are estimated as z scores (standardized
values) and points at which when exceeded, define a participant as in the next category. The first
model (Model 2.1a) was run without sampling weights and Model 2.1b was run using the
sampling weights created in result 1. Because this is a path model with no latent variables or
factors, the structural relationship is measured rather than measurement relationships and this
model assumes perfect measurement of the observed variables and (University of California Los
Angeles, 2013). This model has zero degrees of freedom because the number of estimated
parameters is the same as the number of known parameters. For this reason, fit statistics are
thought by some not to be very helpful (Kline, 2006, p.157, 196).
SPORTS RETIREMENT OF COLLEGE ATHLETES 145
Appendix K: Testing Indirect Effects Using the Sobel Test
One popular way that indirect effects or mediation effects are tested is by using what is now
termed the Sobel test (Sobel, 1982). The Sobel test can be calculated with the following
equation
(19)
This expression was clearly defined by other much earlier (i.e., the quadratic form of the
product), and requires the model to fit the data Where a is the unstandardized regression estimate
for the relationship between the predictor and the mediator, b is the unstandardized regression
estimate for the relationship between the mediator and the outcome variable,
is the standard
error of a and
is the standard error of b (most recently used by Preacher & Leonardelli, 2012).
SPORTS RETIREMENT OF COLLEGE ATHLETES 146
Appendix L: Mplus Code
L.1_ Retirement Path Model
DATA:
File = "F:\Dissertation-USC\Data\Retirement Diss data.dat";
VARIABLE:
NAMES =
id id2 coregpa ethnic fammedin hscgpa income source
q26 q49 q50 q51a q54a q57 q58a q58b sex responded
sport male minority testrc testrc2 weight q11 q10
q58c q58d q58e q58f q58g q58h q58i q51b q51c q51d q51e q51f fmin;
USEVARIABLES =
q26 hscgpa male minority testrc coregpa fmin q10;
USEOBSERVATIONS = q11 LE 0;
MISSING = all (-99);
CATEGORICAL = q26;
ANALYSIS:
MODEL:
q26 ON hscgpa male minority testrc coregpa fmin q10;
OUTPUT: SAMPSTAT standardized stdyx;
SPORTS RETIREMENT OF COLLEGE ATHLETES 147
L.2 _Identity CFA
VARIABLE:
NAMES=
id id2 coregpa ethnic fammedin hscgpa income source
q26 q49 q50 q51a q54a q57 q58a q58b sex responded
sport male minority testrc testrc2 weight q11 q10
q58c q58d q58e q58f q58g q58h q58i q51b q51c q51d q51e q51f
q83a q83b fmin retrc q15a q15b q15c q15d q15e q15f q15g q15h
q15i q15j q15k q15l;
USEVARIABLES = q15a q15b q15c q15d q15f q15g q15h q15i q15j q15k q15l ;
USEOBSERVATIONS = q11 LE 0;
grouping = male (0=female, 1=male);
MISSING IS = all (-99);
ANALYSIS:
MODEL:
F1 BY q15b* q15d q15f q15h ;
F2 BY q15a* q15c q15g;
F3 BY q15i* q15j q15k q15l;
F1@1; [F1@0];
F2@1; [F2@0];
F3@1; [F3@0];
F1 WITH F2 (p1);
F1 WITH F3(p2);
F2 WITH F3(p3);
q15a(u1); q15b(u2); q15c(u3); q15d(u4);
q15f(u5); q15g(u6); q15h(u7); q15i(u8);
q15j(u9); q15k(u10); q15l(u11);
MODEL FEMALE:
F1 WITH F2(p1);
F1 WITH F3(p2);
F2 WITH F3(p3);
q15a(u1); q15b(u2); q15c(u3); q15d(u4);
q15f(u5); q15g(u6); q15h(u7); q15i(u8);
q15j(u9); q15k(u10); q15l(u11);
OUTPUT:tech1 tech4 stdyx SAMPSTAT STANDARDIZED residual;
SPORTS RETIREMENT OF COLLEGE ATHLETES 148
L.3_Life Outcomes CFA
VARIABLE:
NAMES =
id id2 coregpa ethnic fammedin hscgpa income source
q26 q49 q50 q51a q54a q57 q58a q58b sex responded
sport male minority testrc testrc2 weight q11 q10
q58c q58d q58e q58f q58g q58h q58i;
USEVARIABLES=
q50 q51a q54a
q58a q58b q58c q58d q58e q58f q58g q58h q58i q57;
MISSING = all (-99);
CATEGORICAL ARE q50 q57 q51a q58a q58b q58c q58d q58e q58f q58g q58h q58i;
ANLYSIS:
estimator=WLSMV;
MODEL:
F1 BY q50* q54a q51a;
F1@1; [F1@0];
F2 BY q58a* q58b q58c q58g ;
F2@1; [F2@0];
F3 BY q58d* q58e q58f q58h q58i q57;
F3@1; [F3@0];
F1 WITH F2 F3;
OUTPUT: SAMPSTAT standardized stdyx;
SPORTS RETIREMENT OF COLLEGE ATHLETES 149
L.4_Full Model Latent Variable Path Model
VARIABLE:
NAMES ARE
id id2 coregpa ethnic fammedin hscgpa income source q26 q49 q50 q51a q54a
q57 q58a q58b sex responded sport male minority testrc testrc2 weight q11 q10 q58c q58d
q58e q58f q58g q58h q58i q51b q51c q51d q51e q51f q83a q83b fmin retrc q15a q15b
q15c
q15d q15e q15f q15g q15h q15i q15j q15k q15l;
USEVARIABLES = q51a q51d q51e q58a q58b q58g q58h q26 hscgpa male minority
testrc coregpa fmin q10 q15a q15b q15c q15d q15f q15g q15h q15i q15j q15k
q15l;
USEOBSERVATIONS = q11 LE 0;
MISSING = all (-99);
CATEGORICAL = q26 ;
ANALYSIS:
MODEL:
F1 BY q51a* q51d q51e q58b;
F2 BY q58a* q58g q58h;
F1@1; [F1@0];
F2@1; [F2@0];
F3 BY q15b* q15d q15f q15h ;
F4 BY q15a* q15c q15g;
F5 BY q15i* q15j q15k q15l;
F3@1; [F3@0];
F4@1; [F4@0];
F5@1; [F5@0];
q26 ON hscgpa male minority testr coregpa fmin q10 F3 F4 F5;
F1 F2 ON q26 hscgpa male minority testrc coregpa fmin q10 F3 F4 F5 ;
F1 F2 WITH F3@0 F4@0 F5@0;
OUTPUT: tech1 tech4 stdyx SAMPSTAT STANDARDIZED residual;
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Exploring the academic success of black male former student-athletes and their experiences with academic support upon re-entry to college
PDF
Walking away from the game: a phenomenological study on student-athletes career transition
PDF
A functional use of response time data in cognitive assessment
PDF
Personal and environmental influences on preparing for transition from sport: a social cognitive approach to exploring the collegiate experiences of Black former NFL players
PDF
Family conflict, negative mood, and adolescents' daily problems in school
PDF
Latino/a college student-athletes: Influences on recruitment, enrollment and degree completion
PDF
Sources of stability and change in the trajectory of openness to experience across the lifespan
PDF
Native Hawaiian student success in the first-year: the impact of college programs and practices
PDF
Life transitions, leisure activity engagement, and cognition among older adults
PDF
The Awareness Podcast: a look into how sports can negatively affect a student-athlete's mental health
PDF
Applying adaptive methods and classical scale reduction techniques to data from the big five inventory
PDF
Oppression of remedial reading community college students and their academic success rates: student perspectives of the unquantified challenges faced
PDF
Discovering the sources of impact of college on LGBTQ students' identity development and mapping those experiences
PDF
Effects of AT-1 receptor blockers on cognitive decline and Alzheimer's disease
PDF
Attaining success: how African American college students persist, engage and graduate from a moderately selective institution: a case study
PDF
Fulfillment of a regional college promise: lessons from the first year, first cohort
PDF
Mechanisms of the effect of involuntary retirement on older adults' health and mental health
PDF
College academic readiness and English placement
PDF
Predictors of success in meeting the high school mathematics and English requirements for college entrance: a longitudinal study
PDF
Evaluating social-cognitive measures of motivation in a longitudinal study of people completing New Year's resolutions to exercise
Asset Metadata
Creator
Shelton, Erin D.
(author)
Core Title
Later life success of former college student-athletes as a function of retirement from sport and participant characteristics
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Psychology
Publication Date
07/02/2013
Defense Date
04/15/2013
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
athletic identity,college athletes,college athletics,latent variable path modeling,life after sports,life outcomes,life satisfaction,OAI-PMH Harvest,sports psychology,sports retirement
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
McArdle, John J. (
committee chair
), Gatz, Margaret (
committee member
), John, Richard S. (
committee member
), Rueda, Robert (
committee member
)
Creator Email
erinshelton2@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-282924
Unique identifier
UC11293898
Identifier
etd-SheltonEri-1731.pdf (filename),usctheses-c3-282924 (legacy record id)
Legacy Identifier
etd-SheltonEri-1731.pdf
Dmrecord
282924
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Shelton, Erin D.
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
athletic identity
college athletes
college athletics
latent variable path modeling
life after sports
life outcomes
life satisfaction
sports psychology
sports retirement