You are on page 1of 20

333-

Development and Initial Validation of a


Measure of Academic Hardiness
Lois A. Benishek
Temple University
Frederick G. Lopez
Michigan State University
Drawing upon hardiness theory and research on academic
motivation we developed and administered a measure of academic
hardiness to a large sample of high school students. In line with
theory- and research-derived expectations, results of both principal
axis factor analyses (PAF) and confirmatory factor analyses (CFA)
yielded support for a three-factor model of the construct.
Supplemental analyses indicated that, after controlling for general
academic self-concept, the challenge subscale of our novel measure
differentiated students who pursued more difficult academic
coursework and educational plans from those who did not.

Keywords: Academic hardiness, high school students, assessment,


scale development, factor analysis

Generally speaking, students who hold positive concepts of their academic


abilities are expected to perform well in school (Hansford & Hattie, 1982;
Shavelson & Bolus, 1982). After all, academic self-concepts are largely
forged from one’s direct experiences in academic contexts and should thus
reflect typical levels of motivation, effort, and achievement in this
performance domain. It is axiomatic, however, that not all students with
positive academic self-concepts respond similarly to academic
disappointments and challenges. Moreover, not all students with positive
academic self-concepts pursue opportunities to further their learning.

The authors would like to thank Betty E. Gridley and Paul A. Gore, Jr. for their
statistical consultation and Angela Bethea, Aggy Hooi, and Katherine Sapadin for their
feedback on an earlier version of this article.
Correspondence concerning this article and requests for offprints should be addressed
to Lois A. Benishek, Counseling Psychology Program, Department of Psychological
Studies in Education, 270-B Weiss Hall (265-63), Temple University, Philadelphia, PA
19122-6085. E-mail: lbenishek@mindspring.com
Copies of the Academic Hardiness Scale may be requested from Frederick G. Lopez,
Counseling Psychology Program, Department of Counseling, Educational Psychology and
Special Education, 441 Erickson Hall, Michigan State University, East Lansing, MI
48824. E-mail: flopezC~msu.edu

Published and copyright © 2001 by Psychological Assessment Resources, Inc. All rights reserved.

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


334

Indeed, studies investigating relations between academic self-concept


and subsequent achievement has produced inconsistent findings. Whereas
several investigators have observed the expected positive relations between
these constructs (Altmann & DuPont, 1988; House, 1992; Lyon & McDonald,
1990; Marsh, 1992; Muijs, 1997), others have not. For example, Byrne
(1988) found no relationship between high school students’ academic self-
concept and subsequent academic performance 7 months later. Similarly,
Bachman and O’Malley (1986) found no relationship between high school
juniors’ academic self-concept and their educational attainment 6 years
later. Although the above studies employed different measures of academic
self-concept, these measures largely emphasized student ratings of
confidence in their academic abilities and prospects as the key means of
operationalizing the construct.
These observed inconsistencies may, in part, reflect limitations with
current methods of conceptualizing and measuring academic self-concept.
That is, existing measures of academic self-concept (e.g., Reynolds, 1988) may
not be assessing important cognitive, affective, and motivational differences
that exist among students (Marsh, 1992; Marsh, Walker, & Debus, 1991). In
an effort to address this neglect and to develop a more sensitive measure of
academic self-concept, we drew from two conceptual frameworks: Kobasa’s
(1979) hardiness theory and Dweck and Leggett’s (1988) model of academic
motivation.
Hardiness Theory
Kobasa’s theory of psychological hardiness (Kobasa, 1979; Ouellette,
1993) provides a useful framework for understanding why certain students
are more willing to engage in more challenging academic coursework than
others. Hardiness theory posits that three cognitive appraisal processes
serve to buffer the deleterious effects of stressful life situations. These

cognitive processes are: commitment (i.e., perceiving one’s life activities as


valuable to self and others), challenge (i.e., perceiving change rather than
stability as an expected and normal part of life and viewing change as
beneficial to personal development), and control (i.e., perceiving oneself as
having personal control over important life events).
Hardiness theory was originally applied to understanding the relationship
between life stress and physical illness. Subsequent research, however,
has focused on hardiness’s ability to minimize such negative outcomes as
occupational burnout (Pierce & Molloy, 1990), job dissatisfaction (Hills &
Norvell, 1991), poor on-the-job performance (Westman, 1990), depression
(Rhodewalt & Zone, 1989), and maladaptive aging (Magnani, 1990).
Hardiness questionnaires have been designed to assess a variety of clinical
and nonclinical populations (e.g., Bartone, 1989; Kuo & Tsai, 1986; McNeil,
Kozma, & Hannah, 1986; Okun, Zautra, & Robinson, 1988; Sheppard &
Kashani, 1991). Curiously, however, the positive impact that hardiness
may have in academic settings has not yet been investigated, and, to date,
no one has published a psychometrically sound measure of hardiness that
is applicable to an academic setting.
Dweck and Leggett’s Model of Academic Motivation
We suspected that differences in hardiness among students should be
related to important motivational differences in students’ attitudes toward
higher learning and achievement. Dweck and Leggett’s program of research

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


335

on academic motivation provided support for this expectation. In their


review of their own extensive research on academic motivation among
school-age children, Dweck and Leggett (1988) identified two distinct
cognitive-affective-behavioral patterns in academic performance and related
them to the types of goals that students were apparently pursuing. Students
who pursued performance goals attempt to establish the adequacy of their
ability and avoid demonstrating their inadequacy. Because they perceived
ability as being a fixed entity, failure on a given task resulted in feelings of
incompetence and a lack of intellectual ability (Dweck, 1986; Elliott &
Dweck, 1988; Mueller & Dweck, 1998). When faced with especially
challenging tasks or confronted with initial failure experiences, these
students demonstrated heightened vulnerability to negative affect and
lower task persistence.
Students who pursued learning goals, by contrast, viewed academic
challenges as opportunities to acquire new skills and to enhance their
competence. They view ability as being an &dquo;incremental skill&dquo; that can be
improved as one acquires knowledge and becomes more competent (Dweck,
1986; Elliott & Dweck, 1988). Failure and errors were perceived to be a
natural part of the learning process. These students reported less negative
arousal, greater levels of involvement in the task, and more persistence in
response to difficult tasks and initial failure. For example, Mueller and
Dweck (1998) found that students who received praise for their effort rather
than their intellectual ability (i.e., those in a learning-oriented condition)
were more likely to enjoy the task despite its degree of difficulty and the
likelihood of failure. In fact, these students actually preferred more difficult
tasks and did not view their lack of success as a reflection of an intellectual
deficit. By contrast, students who were who were praised for their intellect
(i.e., those in a performance-oriented condition) preferred less challenging
tasks and were less likely to persevere when they experienced failure.
Similar results have been found with students who were transitioning
into junior high, a time when academic work becomes more challenging than
it may have been in prior grades (Dweck & Sorich, 1999; Henderson &
Dweck, 1990).
The Present Investigation
Taken together, the perspectives offered by hardiness theory and research
on academic motivation suggest that, in the academic realm, many students
may choose to maintain positive self-concepts by strategically avoiding
demanding coursework that may tax their capabilities, invite failure, and
thus threaten their self-views. Using terminology consistent with hardiness
theory, the assessment of their otherwise positive self-concepts may have
overemphasized general perceptions of confidence and dedication (i.e.,
commitment) but, insufficiently incorporated the hardiness components of
&dquo;control&dquo; and &dquo;challenge.&dquo; This, in turn, may dispose them to be less resilient
to academic disappointments and failures and orient them away from
learning-based goals. By contrast, students whose academic self-concepts are
&dquo;hardier&dquo; and incorporate sufficient levels of challenge (as well as confidence
in their ability to competently control and manage failure-related distress)
should be more likely to pursue difficult learning experiences and goals. In
short, the positivity of students’ self-concept alone may not be a sufficient
indicator of these individual differences in academic behavior and motivation.
Because extant measures of academic self-concept generally do not assess

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


336

these subtle motivational features, it is plausible that measurement


limitations have substantially contributed to observed inconsistencies in the
research literature linking academic self-concept and learning outcomes.
Hence, by drawing upon both Kobasa’s hardiness theory and Dweck and
Leggett’s model of academic motivation, the present study sought to develop
an alternative measure of academic self-concept that more fully delineates
these individual differences.

Methods
Academic Hardiness Scale (AHS) development for this investigation took
place in three phases.’ In line with Kobasa’s conceptualization of
psychological hardiness, we generated a pool of 40 items reflecting
commitment, challenge, and control. After completing a random split
procedure on the overall sample of high school students, we conducted
principal axis factor analyses (PAFs) on data derived from one half of the
sample (i.e., the calibration sample) to identify the number of potential
factors underlying the measure. Finally, we conducted confirmatory factor
analyses (CFAs) on the second half of the sample (i.e., the validation sample)
to test the models identified in the PAFs. Correlations among the AHS
subscales and select variables, as well as group differences were also
explored in this phase of the study to further examine the validity of the
scores associated with the AHS.

Participants and Procedures


Participants for this study consisted of 481 students who were enrolled
at a midwestern high school. They ranged in age from 14 to 19 years (M =

16 years) and reported a mean 2.90/4.00 fall semester grade point average
(GPA). The students were predominantly Euro-American (92%), female
(52%), and living with both biological parents (61%). Sixty-three percent of
the sample had a clear career goal, and 75% were planning to attend college.
Twenty-six percent and 21% of the students’ fathers and mothers had
completed college degrees, respectively.
With the permission of school officials, we obtained participants’
cumulative GPAs and information about their actual math course enrollment
decisions during the subsequent academic semester. In addition to
completing the AHS, participants were asked to complete a battery of
measures during two regularly scheduled 45-minute class periods.
Participation in the study was voluntary; no financial incentive was provided.
A random split procedure was used to generate the two samples necessary
to conduct the two types of factor analyses completed in this investigation.
A sample of 244 students (i.e., calibration sample; 111 males; 133 females)
was used to conduct the PAFs (Study 1). The sample used to conduct the

follow-up confirmatory factor analyses (CFAs-Study 2) consisted of 237


participants (i.e., validation sample; 120 males; 117 females). No statistically
significant differences were found between the two samples with regard to
age, gender, ethnicity, year in high school, parents’ relationship status,
mother’s and father’s level of education, or fall semester grade point average.

A copy of the initial


1 measure is available upon request from the second author.

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


337

Supplemental Measures
Demographic Form
The demographic form provided basic descriptive student information
(e.g., gender, age), as well as post-high school graduation plans (i.e., whether
they anticipated enrolling in college or pursuing other work options).
Marlowe-Crowne Social Desirability Scale
(MCSD; Crowne & Marlowe, 1960)
This 33-item measure assessed students’ general &dquo;need for approval&dquo;
response motivation (Robinson, Shaver, & Wrightsman, 1991). Items (e.g.,
&dquo;I like to gossip at times.&dquo;) are scored using a forced-choice format (1 =
true or 2 = false). The 1-month test-retest reliability is .88 for MCSD scores
(Crowne & Marlowe, 1964); observed Cronbach alphas have ranged from .73
to .88 across several independent investigations. The coefficient alpha
estimate for the present study was .74.
Academic Self-Concept Scale (ASCS; Reynolds, 1988)
The ASCS is a 40-item measure of global academic self-confidence. Items
(e.g., &dquo;I consider myself to be a good student&dquo;) are rated on a 4-point scale
(1 strongly disagree to 4 strongly agree). Higher scores reflect stronger
= =

academic self-concepts. The ASCS scores have strong internal consistency


(Cronbach alpha .92) and test-retest stability (r .88) over a 2-week
= =

period within an undergraduate college sample (Reynolds, 1988). The


coefficient alpha estimate for the present study was .94. ASCS scores are
correlated (r .60) with high school GPAs (Lopez & Lent, 1992).
=

Math Anxiety (Betz, 1978)


The Fennema-Sherman Math Anxiety Scale, as revised by Betz (1978), was
used to assess students’ degree of emotional discomfort and distress
regarding math-related performances. Each of the 10 items on this measure
(e.g., &dquo;Mathematics makes me feel uneasy and confused&dquo;) are rated on a 5-
point scale (1 strongly disagree to 5 strongly agree). Ratings are summed
= =

to produce a total score, with higher scores reflecting higher math anxiety.
The Cronbach alpha estimate obtained in this study was .90.
Mathematics Self-Efficacy (Lopez, Lent, Brown, & Gore, 1997)
Students’ confidence in their ability to solve math problems characteristic
of the math course in which they were currently enrolled (e.g., algebra,
geometry) was assessed using a 10-item math course-specific self-efficacy
scale (0 no confidence to 9
=
complete confidence). Higher scores are
=

indicative of greater levels of math self-efficacy. The development of these


scales is described in detail elsewhere (i.e., Lopez et al., 1997). The Cronbach
alpha coefficient in this study was .84.
Mathematics-Related Interests (Lopez & Lent, 1992; Lopez et al., 1997)
This scale assessed students’ degree of interest in each of 20 math- and
science-related activities (e.g., solving a math puzzle) using a 3-point scale
(1 dislike, 2 indifferent, 3 like). Ratings are summed to produce a total
= = =

math interest score, with higher scores reflecting stronger interests. Lopez
and Lent (1992) reported a Cronbach alpha estimate of .91 for this scale’s
scores. The Cronbach alpha estimate in this study was .90. Scale scores

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


338

correlated positively with measures of math self-efficacy and perceived


math relevance in a sample of high school students (Lopez & Lent, 1992).

Construct Definitions and Initial Item Development of the


Academic Hardiness Scale
In keeping with both Kobasa’s (1979) three-component model of
psychological hardiness and with Dweck and Leggett’s (1988) distinction
between performance-based goals and learning goals, an effort was first made
to define the commitment, challenge, and control dimensions of academic
hardiness and to generate items representing these components. Given the
specific nature of the academic context, some accommodations to both
perspectives were made in arriving at the following initial definitions. The
academic hardiness measure contained unique items that represent a
conceptual synthesis of both theories.
Commitment
Commitment (Fl) was defined as students’ reported willingness to expend
consistent effort and to engage in personal sacrifices in order to achieve
academic excellence, irrespective of the content or demands of individual
courses, instructors, or personal interests. Highly committed students were
conceived as exhibiting personal dedication and involvement with all their
courses, with less committed students showing inconsistent and more
conditional involvement. Fifteen items were originally written to tap this
component of academic hardiness (e.g., &dquo;Regardless of the class I’m in, I try
to do my best work&dquo; and &dquo;I find it hard to get real excited about my classes&dquo;).

Challenge
This component (F2) was defined aspurposeful efforts to seek
the students’
out difficult academic coursework and experiences and to justify such
actions as inherently important for personal learning. In this regard,
Kobasa’s notion of challenge was extended to address risk-taking in the
pursuit of learning goals. Eleven items were developed to assess this
component. Positively keyed challenge items stress accepting the risks of
personal failure or disappointment as a necessary and even desirable aspect
of the learning process (e.g., &dquo;Taking difficult classes is the best way to
improve one’s knowledge&dquo;). Negatively keyed items, by contrast, emphasize
the avoidance of demanding academic experiences that might threaten
personal confidence and self-esteem (e.g., &dquo;I don’t see the purpose of taking
a class if I’m not totally confident that I can do well in it&dquo;).

Control
This component (F3) was defined as students’ beliefs that they possessed
the capacity to achieve desired educational outcomes through personal
effort and through effective emotional self-regulation in the face of academic
stresses and disappointments. In essence, we specifically extended Kobasa’s
emphasis on internal control to include one’s perceived ability to directly cope
with academically related threats to self-esteem, to manage distressing
affect and personal doubts, and to respond resiliently to failure experiences.
Fourteen items were originally written to assess this component. An example
of a positively keyed item is, &dquo;Getting good grades is as much a matter of
effort as it is of ability&dquo;; an example of a negatively keyed item is, &dquo;I find it
difficult to bounce back from academic disappointments.&dquo;

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


339

The initial version of the AHS consisted of 40 statements designed to


assess the three academic hardiness components. Each statement was
answered using a 4-point Likert scale (1 completely false, 4 completely
= =

true) with higher scores indicating a greater degree of each component.


Composite scores were calculated by combining the three component scores.
Study 1-Principal Axis Factor Analyses Process
Initial Construct Analyses
Correlations Among Composite and Component Scores
The correlations among the three AHS components were .48 (F1 with F2),
.32 (Fl with F3), and .49 (F2 with F3) for the 40-item measure, suggesting
some degree of overlap among the components. These values, however,
were low enough to indicate that the components were measuring different
constructs. The greatest amount of shared variance was 24% for the
challenge (F2) and control (F3) components. As was expected, the correlations
between each of the components with the overall composite score were
higher in magnitude than the correlations among the components themselves
(r for Fl .84, F2 .79, F3 .71).
= = =

Item-Total Correlations
Examination of the item-total correlations indicated that 13 AHS items
were not correlating adequately with the overall construct (i.e., values < .30
indicate less than 10% shared variance with the overall construct). These
items were deleted from the measure before proceeding with the factor
analysis. The resulting measure consisted of 6 control, 14 commitment,
and 7 challenge items, yielding a 27-item measure of academic hardiness.
Internal Consistency Reliability Estimates and Correlations
The coefficient alpha estimates for the 27-item measure were .89, .82, .73,
and .91 for the AHS commitment, challenge, and control components, and
the composite score, respectively. Although the value associated with control
component scores was the weakest, the values associated with all the
constructs’ scale scores were within an acceptable range (Nunnally, 1978),
indicating that the items were adequately assessing common constructs.
Correlations among these subscales were .54 (Fl with F2), .50 (Fl with F3),
and .52 (F2 with F3); the subscales correlated .91 (Fl), .79 (F2), and .73 (F3)
with the composite score.

Principal Axis Factor Analyses


Principal axis factor analyses (PAF) using an oblique rotation were
completed on the 27-item AHS (N 244) using SPSS (Marija J.
=

Norusis/SPSS, Inc., 1990). PAF is the exploratory analysis of choice given


that the communality estimates that result from PAF are more accurate than
PCA when the questionnaire of interest contains less than 35 items (Gorsuch,
1983). An oblique rotation was chosen for this analysis given that the
initial inter-scale correlations suggested that the factors would be correlated
with one another.
Five a priori criteria were used to identify the substantive factors.
These criteria included (a) factors with eigenvalues greater than 1.0,
(b) examination of the scree plot, (c) the magnitude of the item loadings,

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


340

(d) the presence of co-loadings with the other resulting factors, and (e)
the conceptual meaningfulness of the factors (Stevens, 1992; Tabachnick &
Fidell, 1996).
The initial PAF results suggested that both a two- and three-factor model
should be explored further. The PAFs were completed again, forcing two- and
three-factor solutions. Only factor loadings of .40 or greater were considered
to be significant and used to interpret the factors (Stevens, 1992). Tables 1
and 2 contain the factor loadings (obtained from the pattern matrix) and the
communalities for the two- and three-factor models, respectively. Values
associated with the pattern matrix are preferable to those provided in the
structure matrix given that an oblique rotation is used and these values tend
to be less inflated (Tabachnik & Fidell, 1996).
The two-factor model accounted for 37% of the explained variance, with
the resulting statistically significant factor loadings ranging from .40 to
.87 (M =
.64). Nine items had factor loadings less than .40. The internal
consistency reliability estimates for Factor 1, Factor 2, and the composite
score were .86, .80, and .84, respectively. The two factors correlated .46
with one another, indicating that they shared 21% of the variance with
one another. The two factors correlated .89 and .86 with the AHS
composite score.

Factor 1 consisted of 10 commitment items and appeared to measure


students’ commitment to academic excellence, as exhibited by their
investment in and appreciation for challenging educational activities. Factor
2 consisted of three control and five challenge items and appears to represent
students’ willingness to undertake demanding academic coursework (i.e., a
mastery-oriented attitude) and confidence in their ability to perform
academically.
The three-factor model accounted for 42% of the explained variance.
Statistically significant factor loadings resulting from this model ranged from
.40 to .84 (M .59). This model contained four items with factor loadings less
=

than .40. The internal consistency reliability estimates for this model were
.85, .78, .69, and .84 for the three factors and the composite score,
respectively. The factors showed relatively low correlations with each other
(range .32 to .39). At most, this indicates 15% shared variance, suggesting
=

that the three factors were assessing relatively different aspects of


academic hardiness. The three factors correlated .88, .79, and .72 with the
composite score.

Factor 1 consisted of 10 commitment items, Factor 2 consisted of 7


challenge items, and Factor 3 consisted of 3 control and 3 commitment
items. Similar to the 2-factor model, the first factor appears to be a measure
of commitment to academic pursuits. The second factor can be interpreted
as a challenge or mastery-oriented learning style (i.e., task approach versus
avoidance). Finally, Factor 3 assesses intrinsic motivation to perform well
academically (e.g., interest, ability, motivation, effort) and affective reactions
to academic performance (i.e., ego strength). It reflects students’ ability to
control the degree of effort given to academic pursuits, as well as their
ability to manage any negative reactions that might occur when they are
faced with difficult academic experiences.
In addition to their psychometric properties, both the two- and the three-
factor models were deemed worthy of further investigation for conceptual
reasons. Overall, the three-factor model closely paralleled Kobasa’s

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


341

Note. N = 244. Comm = Commitment, Chall = Challenge, Cont = Control. Bolded values
indicate that the item is loading > .40 on its respective factor. Underlined items indicate
that the item was inversely scored.

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


342

(continued)

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


343

Table 2 (continued)
Academic Hardiness Scale Factor Loadings and
Communalities From Principal Axis Factor Analysis

Note. N =244. Cont Control, Com Commitment, Chall Challenge. Bolded values
= = =

indicate that the item is loading > .40 on its respective factor. Underlined items indicate
that the item was inversely scored.

conceptualization of hardiness which is to be expected given that her theory


served as the foundation for our item development. The two-factor model was
also conceptually meaningful despite the fact that items from each of the
three hypothesized subscales were interdispersed with each other on these
two factors.

Study 2-Confirmatory Factor Analyses and


Validity Analyses Process
The AHS was revised prior to beginning this phase of the study. Nine items
with factor loadings less than .40 in the PAF were deleted from the measure,
yielding a 18-item questionnaire.
Confirmatory Factor Analyses
Confirmatory factor analyses using EQS 5.7 (Bentler & Wu, 1998) were
completed on the two- and three-factor models, as well as single-factor
model (i.e., null model) in which each of the items was assumed to assess
one broad, global factor. CFA is an appropriate method for evaluating the
construct validity of a measure (Rahim & Magner, 1995) and is used to
determine &dquo;whether a particular factor model fits the data better than
others&dquo; (Bryant & Yarnold, 1995; p. 119).
Analyses were completed on data derived from the validation sample of
high school students (N 237). The analyses were based on covariance
=

matrices, maximum likelihood estimates, and listwise deletion methods.


Measurement error was specified as uncorrelated in each of the models. The
items were permitted to load on the factor suggested by the results of the
PAF analysis and were constrained to zero on the other factors. Each latent
variable was scaled to the first indicator by fixing it to 1.0. Resulting factor
loadings and the amount of variance accounted for by each item are found
in Table 3.
A number of goodness-of-fit indexes were used to evaluate the degree to
which the competing models fit the data. These indexes included the chi-
square statistic (J6reskog & S6rbom, 1984), the chi-square to degrees of

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


344

Table 3
Academic Hardiness Scale Factor Loadings Associated
With the Confirmatory Factor Analysis

Note. N = 237. Cont = Control, Com = Commitment, Chall = Challenge. Underlined


items indicate that the item was inversely scored. The path estimate in parenthesis did
not reach statistical significance.

freedom ratio (x2/df ), the non-normed fit index (NNFI; Bentler & Bonnet,
1980), the comparative fit index (CFI; Bentler, 1990), the adjusted goodness-
of-fit index (AGFI; J6reskog & Sorbom, 1984), and the root mean squared
of the residuals (RMR). Nonsignificant chi-square values, x2/df ratios less
than 2.0, NNFI and CFI values greater than .90, AGFI values greater than
.80, and RMR values less than .10 typically indicate that the model is
adequately representing the data (Byrne, 1989; Cole, 1987; Marsh & Hocevar,
1985). However, it is sometimes acceptable to have fit indexes that are less
than .90 (Hoyle & Panter, 1995).

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


345

There are limitations associated with certain goodness-of-fit indexes.


For instance, the chi-square statistic is influenced by large sample sizes and
is likely to produce a statistically significant result regardless of the degree
of fit (Bentler & Bonnet, 1980). Furthermore, statistically significant chi-
square values result from models containing a large number of variables,
high degrees of freedom (Bryant & Yarnold, 1995), and when there are
violations of multivariate normality (West, Finch, & Curran, 1995). Given
the problems associated with these commonly used fit indexes, the fit of the
alternative models was examined by completing chi-square difference tests
in order to directly compare the resulting chi-square values associated
with these models (Hoyle & Panter, 1995; Loehlin, 1992; Ullman, 1996). This
procedure is preferable to simply comparing changes in the incremental fit
indexes (e.g., CFI) when evaluating the fit of nested models (Hoyle &
Panter, 1995).
There were numerous indications from the CFA analyses that the three-
factor model best represented the data (Table 4). First, the fit indexes are
better for the three-factor model in comparison to either the null (i.e., one-
factor) model or the two-factor model (i.e., x2 244.063, p < .001; x2/df = =

1.863; NNFI .906; CFI .919; AGFI .867; RMR .066). Second, results
= = = =

from the chi-square difference tests indicate that the three-factor model
provided a significantly better fit to the data in comparison to the other two
models tested in this analysis (i.e., null model vs. three-factor model chi-
square difference 367.338, df difference 4, p < .001; two-factor model vs.
= =

three-factor model chi-square difference 66.966, df difference 2, p < .001).


= =

Third, examination of more basic psychometric properties also provides


support for this model. The three factors correlated .41 (Fl with F2), .07 (Fl
with F3), and .47 (F2 with F3) and .87, .76, and .44 with the composite score.
These findings indicate that the shared variance among these factors is at
most 22%, indicating that the factors are assessing similar but not redundant
concepts. The coefficient alpha estimates for the factors and the composite
are .85, .80, .64, and .85, respectively. These values indicate that the scores
associated with the first two factors have good internal consistency reliability,
and the reliability for the scores associated with the third factor is modest.
Finally, this model is conceptually more meaningful than the two- and the
one-factor models.
Two unexpected findings emerged from the confirmatory factor analysis.
First, the LaGrange indicators suggested that the fit of the model would be
improved by allowing one item to load on two factors. The final model
includes this co-loading. Second, one item did not reach statistical
significance. A comparison of the means, standard deviations, and item-total
correlations for this item in both data sets identified no notable differences
in this item between data sets, suggesting that this phenomenon was likely
to have been a random event.

Validity Analyses
We were also interested in exploring the relationship between the AHS
subscales with select demographic and math-related indexes. We chose to
focus on high school mathematics because national statistics indicate that
negative attitudes and expectations about continuing math study
predominate among K-12 learners (Eccles, 1986). Furthermore, 50% of all
American high school students discontinue taking math after the 10th

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


346

W
<a
4)
r4
1$4
#
z
t)
.p4

0a)
&dquo;0
«
t)
-.1~
0-)
0
cue
1-4
4)
10
0
~ ~
p 4) i CL)
’32
Cd
E4A
...

E-4 h
4’.’
M
a)
u
0’&dquo;
10
a
0-4
-4a

E
4~
o
M
CO
4)
14
&dquo;0
o
0
15

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


347

grade (National Association of Educational Progress, 1988; National Center


for Educational Statistics, 1984; NCES) and only 7.5% of 17-year-old
students are proficient in higher levels of mathematics, such as geometry,
algebra, and beginning statistics (NCES, 1998). These findings suggest
that investigating the relationship between mathematically related variables
(e.g., math self-efficacy, math anxiety, and math interests) and academic
hardiness may help us to better understand these negative attitudes,
expectations, and behaviors.
Correlational Analyses
As part of a preliminary validational effort, we examined associations of
our AHS subscale scores (i.e., Fl= commitment; F2 challenge; F3 control) = =

with several other demographic, self-report, and objective academic


performance measures that were part of our larger participant database. We
believed that it was important to determine the extent to which social
desirability might be impacting students’ responses to the questionnaire. We
found little or no correlation between social desirability and the subscale
scores (r with F1 =
.27, p < .01; no statistically significant relationships with
F2 or F3). The 9% overlap between the first academic hardiness factor and
social desirability (.272) indicates that there is no reason to believe that
students were misrepresenting themselves on this factor to the extent that
it would confound the results of this study.
As expected, AHS subscale scores were only moderately correlated with
global academic self-concept (ASCS) scores [.51(F1), .51(F2),.50(F3), p < .01],
indicating that while the AHS subscales share common variance with ASCS
scores (as might be expected), they are not tapping entirely redundant
constructs. It is also interesting to note that while the correlations between
the AHS subscales [r .15(F1), p < .05; r .31(F2), p < .01; r .14(F3),
= = =

p < .05] and the ASCS scale (r .20, p < .01) with students’ post-high
=

school plans were statistically significant, the AHS challenge scale was
most highly correlated with the anticipation of enrolling full-time in college
(r .31). On the other hand, relative to the three AHS subscale scores [r
= =

.30(F1), p < .01; r .38(F2), p < .05; r .31(F3), p < .01], ASCS scores were
= =

somewhat more strongly correlated with students’ cumulative high school


GPAs (r = .52, p < .01).
Among the math-related variables, statistically significant relationships
were found between the commitment (r = .27, p < .01) and challenge (r .34, =

p < .01) subscales with students’ interest in math activities. Math anxiety
was moderately correlated with the challenge (r -.45, p < .01) and control =

(r -.41, p < .01) subscales and to a lesser degree with the commitment
=

(r -.16, p < .05) subscale. Statistically significant relationships were also


=

found among the three subscales with math self-efficacy [r .24(F1), .45(F2), =

.30(F3), p < .05] with the strongest relationship existing between math
self-efficacy and the challenge subscale.
A statistically significant relationship was found between gender and
the commitment subscale (r = .31, p < .01), but not with the challenge and
control subscales. The direction of the correlations indicated that high
school girls obtained higher commitment scores than their male counterparts.

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


348

College-Bound Versus Non-College-Bound Differences


To explore whether statistically significant differences in AHS subscale
scores existed between college-bound and non-college-bound students which
could not be explained by differences in global academic self-concept (ASCS),
students were divided into college-bound (n 170) and non-college-bound =

(n 57) groups on the basis of their expressed post-high school graduation


=

plans. A series of three univariate analyses of covariance were conducted on


the three AHS subscales using ASCS as the covariate. A Bonferroni
correction for Type I error (p < .05/3 .017) was used. The groups differed
=

with regard to their challenge scores (F 13.97, p < .001, ,,2 .06), with
= =

significantly higher scores reported by the college-bound group.


Continued Versus Discontinued Math Course Enrollment
We also explored whether students’ decision to pursue additional math
coursework was related to their AHS scores. Students were divided into
either continuer (n =
170) or non-continuer (n 49) groups on the basis of =

their actual math enrollment decisions during the subsequent academic


semester. Three one-way analyses of covariance were conducted on the
AHS subscale scores after controlling for students’ ASCS. To control for
Type I error, an alpha level of .017 was also used in these analyses. Results
identified one statistically significant between group difference. Once again,
those students who pursued additional math coursework reported higher
challenge scores than did those students who did not pursue additional
math coursework (F 7.59, p < .01, il2 .03).
= =

Discussion
The present study sought to develop a self-report instrument of academic
hardiness, the Academic Hardiness Scale (AHS). Our findings provide
preliminary support for the reliability and validity of the scores associated
with this measure. A principal components factor analysis was conducted
on 27 of the 40 original items, and the results largely upheld our item
construction efforts. Support was found for both a two- and three-factor
model. However, the three-factor model accounted for a larger amount of the
variance and had greater conceptual appeal than the two-factor model.
Confirmatory factor analyses were conducted in the second phase of this
study. Results from this analysis indicated that, in comparison to the one-
and two-factor models, the three-factor model best represented the data and
was conceptually sound.

The combined results of the exploratory and confirmatory analyses


identified the three subscales as: commitment and dedication to academically
difficult pursuits; investment in mastery-oriented learning endeavors (i.e.,
challenge); and ability to control both effort and affect directed toward
academically difficult activities. These subscales parallel those suggested
by Kobasa’s (1979) conceptualization of personality hardiness.
Each of the subscales scores associated with the final model demonstrated
reasonable levels of internal consistency reliability given that the
development of the AHS is in its early stages. The magnitude of the
correlations among the subscales indicate that the measure was assessing
relatively independent constructs. The results from the three-factor PAF and
the three-factor CFA identified, at most, 15% and 22% overlap between
the subscales, respectively, indicating that the subscales are not redundant

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


349

with one another. Somewhat unexpectedly, our scale development procedure


resulted in a more delimited measure of control than was originally
conceived. Several original control items which assessed students’ ability
and effort attributions of their academic success were not retained in the
final model.
All three AHS subscales were significantly and moderately correlated with
global academic self-concept. These results are consistent with our
expectation that global measures of academic self-concept may largely tap
self-views of academic confidence and dedication, but may not adequately
assess specific student motivations to pursue difficult coursework. The
magnitude of the relationship between the academic hardiness subscales and
global academic self-concept indicate that, although these variables are
certainly related to one another, they are not redundant constructs.
The relationships identified between the challenge subscale with other
variables of interest in this study indicate that this characteristic may be
useful in understanding how students perceive academically challenging
coursework. For instance, the pattern of correlations among the academic
hardiness subscale scores and the academically oriented measures suggests
that the challenge subscale (range = .38 to .45) may play a more salient role
than either the commitment (range .16 to .30) or the control subscale
=

(range .00 to .41).


=

Further support for the concurrent and predictive validity of the AHS
subscales emerged when between group differences were explored. In
comparison to their counterparts, college-bound students and those who
pursued additional math coursework reported statistically significant higher
challenge scores. Although the effect sizes for each of these findings were
small, the results indicate that challenge plays an important role in
understanding certain academic intentions and behaviors even after
accounting for academic self-concept.
Research Limitations and Future Directions
One challenge associated with developing a new measure such as the
AHS is the difficulty associated with obtaining scores that are reliable and
valid. Our preliminary findings provide support for conducting additional
psychometric investigations on the AHS. Items should be added, revised,
and/or deleted to enhance the reliability of the measure’s subscale scores and
to ensure that a similar number of items are assessing each of the
dimensions. This is particularly relevant to the control subscale. Test-
retest reliability data should also be obtained to determine the stability of
the construct over time. Effort should be made to further evaluate the
validity of the AHS scores, as well as the possibility of a monomethod bias
(i.e., To what extent do teachers’ perceptions of students’ academic hardiness
differ from students’ self-perceptions?). The generalizability of academic
hardiness to other populations (e.g., races/ethnicities; different age groups)
also needs to be explored given that participants in this study were
predominantly White high school students.
Once the psychometric properties associated with the three-factor measure
have been strengthened, causal models can be developed to clarify the
strength of the relationships associated with each of the AHS factors in
predicting academic outcomes. It may also be useful to clarify the
relationship between academic hardiness and other somewhat similar

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


350

cognitively oriented constructs. One such construct is defensive pessimism


(Cantor & Norem, 1989; Norem & Cantor, 1986a, 1986b), a cognitive strategy
used by individuals to buffer themselves against threats to self-esteem
when they are faced with risky or challenging situations. Defensive
pessimism is similar to academic hardiness in that control of personal
affect appears to play an instrumental role in how individuals perform in
academically risky situations.
In summary, the current findings suggest that the AHS shows promise in
producing scores that are reliable indicators of the cognitive, affective, and
motivational aspects of academic hardiness among high school students. The
AHS, and the challenge subscale in particular, distinguished both a college-
bound orientation and persistence in difficult academic coursework more
effectively than did the measure of global academic self-concept employed
in the current investigation. Whether this relatively greater sensitivity is
observed when the AHS is compared to other measures of academic self-
concept remains to be investigated. To the extent that future research
yields similar findings, the AHS may be a useful predictor of long-term
achievement-related motivation and outcomes in high-demand academic
contexts or in the pursuit of challenging educational goals.

References
Altmann, H. A., & DuPont, S. F. (1988). The relationship between academic self-concept
and academic achievement. Canadian Journal of Counseling, 22(3), 170-174.
Bachman, J. G., & O’Malley, P. M. (1986). Self-concepts, self-esteem, and educational
experiences: The frog pond revisited (again). Journal of Personality and Social Psychology,
50, 35-46.
Bartone, P. T. (1989). Predictors of stress-related illness in city bus drivers. Journal of
Occupational Medicine, 31, 657-663.
Bentler, P. M. (1990). Comparative fit indices in structural models. Psychological Bulletin,
107, 238-246.
Bentler, P. M., & Bonnet, D. G. (1980). Significance tests and goodness-of-fit in the
analysis of covariance structures. Psychological Bulletin, 88, 588-606.
Bentler, P. M., & Wu, E. J. C. (1998). EQS for Windows: Version 5.7. Encino: CA:
Multivariate Software
Betz, N. (1978). Prevalence, distribution, and correlates of math anxiety in college students.
Journal of Counseling Psychology, 25 , 441-448.
Bryant, F. B., & Yarnold, P. R. (1995). Principal-components analysis and exploratory and
confirmatory analysis. In L. G. Grimm & P. R. Yarnold (Eds.), Reading and understanding
multivariate statistics (pp. 99-136). Washington, DC: American Psychological Association.
Byrne, B. M. (1988). Self-concept/academic achievement relations: An investigation of
dimensionality, stability, and causality. Canadian Journal of Behavioral Science, 18, 173-186.
Byrne, B. M. (1989). A primer of LISREL: Basic applications and programming for
confirmatory factor analytic models. New York: Springer-Verlag.
Cantor, N., & Norem, J. K. (1989). Defensive pessimism and stress and coping. Social
Cognition, 7
(2), 92-112.
Cole, D. A. (1987). Utility of confirmatory factor analysis in test validation research.
Journal of Consulting and Clinical Psychology, 24, 584-594.
Crowne, D. P., & Marlowe, D. (1960). A new scale of social desirability. Journal of
Educational Psychology, 24, 249-254.
Crowne, D. P., & Marlowe, D. (1964). The approval motive. New York: Wiley.
Dweck, C. S. (1986). Motivational processes affecting learning. American Psychologist, 41,
1040-1048.
Dweck, C. S., & Leggett, E. L. (1988). A social-cognitive approach to motivation and
personality. Psychological Review, 95, 256-273.

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


351

Dweck, C. S., & Sorich, L. (1999). Mastery-oriented thinking. In C. R. Snyder (Ed.),


Coping: The psychology of what works (pp. 232-251). New York: Oxford University Press.
Eccles, J. (1986). Gender roles and women’s achievement. Educational Researcher, 95,
15-19.
Elliott, E. S., & Dweck, C. S. (1988). Goals: An approach to motivation and achievement.
Journal of Personality and Social Psychology, 54, 5-12.
Gorsuch, R. L. (1983). Factor analysis (2nd ed.). Hillsdale, NJ: Erlbaum.
Hansford, B. C., & Hattie, J. A. (1982). The relationship between self and self-
achievement/performance measures. Review of Educational Research, 52, 123-142.
Henderson, V., & Dweck, C. S. (1990). Achievement and motivation. In S. Feldman &
G. Elliott (Eds.), At the threshold: The developing adolescent (pp. 308-329). Cambridge, MA:
Harvard University Press.
Hills, H., & Novell, N. (1991). The examination of hardiness and neuroticism as potential
moderators of stress outcomes. Behavioral Medicine, Spring, 31-38.
House, J. D. (1992). The relationship between academic self-concept, achievement-related
expectancies, and college attrition. Journal of College Student Development, 33, 5-10.
Hoyle, R. H., & Panter, A. T. (1995). Writing about structural equation models. In R. H.
Hoyle (Ed.), Structural equation modeling: Concepts, issues, and applications (pp. 158-176).
Thousand Oaks, CA: Sage.
J&ouml;reskog, K. G., & S&ouml;rbom, D. (1984). LISREL VI user’s guide. Mooresville, IN: Scientific
Software.
Kobasa, S. C. (1979). Stressful life events, personality, and health: An inquiry into
hardiness. Journal of Personality and Social Psychology, 37, 1-11.
Kuo, W. H., & Tsai, Y. (1986). Social networking, hardiness, and immigrants mental
health. Journal of Health and Social Behavior, 27, 133-149.
Loehlin, J. C. (1992). Latent variable models: An introduction to factor, path, and structural
analysis (2nd ed.). Hillsdale, NJ: Erlbaum.
Lopez, F. G., & Lent, R. W. (1992). Sources of mathematics self-efficacy in high school
students. Career Development Quarterly, 41 , 3-12.
Lopez, G., Lent, W., Brown, D., & Gore, P. A., Jr. (1997). Role of social-cognitive
F. R. S.
expectations in high school students’ mathematics-related interest and performance. Journal
of Counseling Psychology, 44
(1), 44-52.
Lyon, M. A., & McDonald, N. T. (1990). Academic self-concept as a predictor of achievement
for a sample of elementary school students. Psychological Reports, 66, 1135-1142.
Magnani, L. E. (1990). Hardiness, self-perceived health, and activity among independently
functioning older adults. Scholarly Inquiry for Nursing Practice: An International Journal,
, 171-184.
4
Marija J. Norusis/SPSS, Inc. (1990). SPSS base system user’s guide. Chicago, IL: SPSS, Inc.
Marsh, H. W. (1992). Content specificity of relations between academic achievement and
academic self-concept. Journal of Educational Psychology, 84 (1), 35-42.
Marsh, H. W., & Hocevar, D. (1985). Application of confirmatory factor analysis to the study
of self-concept: First and higher order factor models and their invariance across groups.
Psychological Bulletin, 97, 562-582.
Marsh, H. W., Walker, R., & Debus, R. (1991). Subject-specific components of academic self-
concept and self-efficacy. Contemporary Educational Psychology, 16, 331-345.
McNeil, K., Kozma, A., & Hannah, E. (1986). Measurement of psychological hardiness in
older adults. Canadian Journal on Aging, 5 (1), 43-48.
Mueller, C. M., & Dweck, C. S. (1998). Intelligence praise can undermine motivation and
performance. Journal of Personality and Social Psychology, 75, 33-52.
Muijs, R. D. (1997). Predictors of academic achievement and academic self-concept:
A longitudinal perspective. British Journal of Educational Psychology, 67(3), 262-277.
National Assessment of Educational Progress. (1988). The mathematics report card: Are
we measuring up? Princeton, NJ: Educational Testing Service.

National Center for Educational Statistics. (1984). High school and beyond: A national
longitudinal study for the 1980s. Washington, DC: Government Printing Office.

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016


352

National Center for Educational Statistics. (1998). Digest of education statistics 1998.
Washington, DC: Government Printing Office.
Norem, J. K., & Cantor, N. (1986a). Anticipatory and post hoc cushioning strategies:
Optimism and defensive pessimism in "risky" situations. Cognitive Therapy and Research,
(3), 347-362.
10
Norem, J. K., & Cantor, N. (1986b). Defensive pessimism: "Harnessing" anxiety as
motivation. Journal of Personality and Social Psychology, 52, 1208-1217.
Nunnally, J. C. (1978). Psychometric theory. New York: McGraw Hill.
Okun, M. A., Zautra, A. J., & Robinson, S. E. (1988). Hardiness and health among women
with rheumatoid arthritis. Personality and Individual Differences, 9, 101-107.
Ouellette, S. C. (1993). Inquiries into hardiness. In L. Goldberger & S. Breznitz (Eds.),
Handbook of stress: Theoretical and clinical aspects (pp. 77-100). New York: Free Press.
Pierce, C. M. B., & Molloy, G. N. (1990). Psychological and biological differences between
secondary school teachers experiencing high and low levels of burnout. British Journal of
Educational Psychology, 60, 37-51.
Rahim, A. M., & Magner, N. R. (1995). Confirmatory factor analysis of the Styles of
Handling Interpersonal Conflict: First-order factor model and its invariance across groups.
Journal of Applied Psychology, 80, 122-132.
Reynolds, W. M. (1988). Measurement of academic self-concept in college students. Journal
of Personality Assessment, 52, 223-240.
Rhodewalt, F., & Zone, J. B. (1989). Appraisal of life change, depression, and illness in hardy
and non-hardy women. Journal of Personality and Social Psychology, 56 , 81-88.
J.
Robinson, P., Shaver, P. &
R., Wrightsman, L. S. (1991). Measures of personality and
social psychological attitudes. San Diego, CA: Academic Press.
Shavelson, R. J., & Bolus, R. (1982). Self-concept: The interplay of theory and methods.
Journal of Educational Psychology, 47, 3-17.
Sheppard, J. A., & Kashani, J. H. (1991). The relationship of hardiness, gender, and
stress to health outcomes in adolescents. Journal of Personality, 59, 747-768.
Stevens, J. (1992). Applied multivariate statistics for the social sciences (2nd ed.). Hillsdale,
NJ: Erlbaum.
Tabachnick, B. G., & Fidell, L. S. (1996). Using multivariate statistics (3rd ed.). New
York: Harper Collins.
Ullman, J. B. (1996). Structural equation modeling. In B. G. Tabachnick & L. S. Fidell
(Eds.), Using multivariate statistics (pp. 709-811). New York: Harper Collins.
West, S. G., Finch, J. F., & Curran, P. J. (1995). Structural equation models with nonnormal
variables: Problems and remedies. In R. H. Hoyle (Ed.), Structural equation modeling:
Concepts, issues, and applications (pp. 56-75). Thousand Oaks, CA: Sage.
Westman, M. (1990). The relationship between stress and performance: The moderating
effect of hardiness. Human Performance, 3 (3), 141-155.

Downloaded from jca.sagepub.com at PENNSYLVANIA STATE UNIV on February 20, 2016

You might also like