You are on page 1of 9

Journal of Counseling Psychology © 2010 American Psychological Association

2010, Vol. 57, No. 2, 239 –247 0022-0167/10/$12.00 DOI: 10.1037/a0019177

Theoretical and Methodological Issues With Testing the SCCT


and RIASEC Models: Comment on Lent, Sheu, and Brown (2010) and
Lubinski (2010)

Patrick Ian Armstrong and David L. Vogel


Iowa State University

The current article replies to comments made by Lent, Sheu, and Brown (2010) and Lubinski (2010)
regarding the study “Interpreting the Interest–Efficacy Association From a RIASEC Perspective”
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

(Armstrong & Vogel, 2009). The comments made by Lent et al. and Lubinski highlight a number of
This document is copyrighted by the American Psychological Association or one of its allied publishers.

important theoretical and methodological issues, including the process of defining and differentiating
between constructs, the assumptions underlying Holland’s (1959, 1997) RIASEC (Realistic, Investiga-
tive, Artistic, Social, Enterprising, and Conventional types) model and interrelations among constructs
specified in social cognitive career theory (SCCT), the importance of incremental validity for evaluating
constructs, and methodological considerations when quantifying interest– efficacy correlations and for
comparing models using multivariate statistical methods. On the basis of these comments and previous
research on the SCCT and Holland models, we highlight the importance of considering multiple
theoretical perspectives in vocational research and practice. Alternative structural models are outlined for
examining the role of interests, self-efficacy, learning experiences, outcome expectations, personality,
and cognitive abilities in the career choice and development process.

Keywords: social cognitive career theory, Holland’s RIASEC model, interests, self-efficacy, structural
equation modeling

One of the motivations underlying our recent study examining the ing interest– efficacy correlations and the interpretation of statistical
relations between interests and self-efficacy, “Interpreting the analyses, and the importance of considering the full range of individ-
Interest–Efficacy Association From a RIASEC Perspective” (Arm- ual differences variables when conducting research on the career
strong & Vogel, 2009), is our perception that researchers evaluating choice process.
social cognitive career theory (SCCT; Lent, Brown, & Hackett, 1994) Although, as outlined below, we disagree with a number of the
often do not consider alternative theoretical perspectives. In our study, points made by Lent et al. (2010), we do agree with their position that
we presented an alternative perspective grounded in Holland’s (1959, future research may be improved by highlighting and debating the
1997) theory of RIASEC (Realistic, Investigative, Artistic, Social, methodological and theoretical considerations that have emerged
Enterprising, and Conventional) types. Using the statistical techniques from our work. We agree with Lubinski’s (2010) position that it is
of multidimensional scaling (MDS), cluster analysis, and structural important to demonstrate the incremental validity of new constructs in
equation modeling (SEM), we found evidence that interest and effi- research before their integration into career counseling interventions
cacy measures can be interpreted as components of the Holland types. but would propose adding personality measures to the list of ability
These results were contrasted with the SCCT model, which defines and interest predictors he identified. Both of these reaction articles
interests and efficacy as distinct constructs with a causal sequence of highlight topics that should be addressed in future research examining
development. In their reaction, Lent, Sheu, and Brown (2010) raised the interrelations between interests; self-efficacy; other constructs
theoretical and methodological issues with our methods, statistical such as learning experiences, personality, and cognitive abilities; and
analyses, and the interpretation of our results. In comparison, Lubinski their relative contributions to the career choice and development
(2010) noted a lack of incremental validity evidence for using self- process. Underlying these points is the more general issue of how to
efficacy measures in addition to measures of cognitive abilities and design studies that advance career counseling theory, research, and
interests for predicting career-related outcomes. In this response, we practice by considering multiple theoretical perspectives.
examine a number of these points, including the process of defining
and differentiating between constructs, assumptions underlying the
Defining and Differentiating Constructs
Holland and SCCT models, methodological issues related to estimat-
Lent et al. (2010) began their critique of our research by framing it
in terms of a well-known expression on waterfowl classification using
Patrick Ian Armstrong and David L. Vogel, Department of Psychology,
auditory and visual evidence. This analogy set the stage for their claim
Iowa State University. that our research implied that interest and self-efficacy measures are
Correspondence concerning this article should be addressed to Patrick “essentially interchangeable” (Lent et al., 2010, p. 219). However,
Ian Armstrong, Department of Psychology, Iowa State University, W237 our hypothesis that the relations between interest and efficacy
Lagomarcino Hall, Ames, IA 50010. E-mail: pia@iastate.edu measures should be interpreted in the context of Holland’s

239
240 ARMSTRONG AND VOGEL

(1959, 1997) RIASEC types does not imply that the two constructs validity over the variance accounted for by their higher order factor.
are interchangeable. In individual-differences psychology, constructs However, it is not always the case that measures of lower order
are often classified using a hierarchical structure, including cognitive constructs in a hierarchical model will have no incremental validity
abilities (Carroll, 1993) and the five-factor model of personality over a measure of the higher order factor. Indeed, a number of studies
(Digman, 1990; Goldberg, 1993). For example, the broad personality have demonstrated that personality facet measures have incremental
traits in the five-factor model can be divided into facets (Costa & validity over broad personality trait measures (Beauducel, Lipmann,
McCrae, 1992), with warmth, gregariousness, and assertiveness clas- Felfe, & Nettelsnstroth, 2007; Bibb, Steinmayr, & Spinath, 2008;
sified as facets of extraversion; anxiety, depression, and self- Connor-Smith & Flachsbart, 2007; Ekehammar & Akrami, 2007;
consciousness classified as facets of neuroticism; and so forth. How- O’Connor & Paunonen, 2007; Paunonen & Ashton, 2001). For ex-
ever, classifying warmth, gregariousness, and assertiveness as facets ample, in a meta-analysis examining the prediction of job perfor-
of extraversion in the five-factor model of personality does not imply mance using personality measures, Dudley, Orvis, Lebiecki, and
that these facets are interchangeable. Similarly, the hypothesis that Cortina (2006) found that facet-level measures of conscientiousness
interests and self-efficacy are components of the RIASEC types does provided incremental improvements in job performance prediction
not imply their interchangeability. over broad measures of conscientiousness.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Lent et al.’s (2010) waterfowl classification analogy is also Therefore, it is possible to find that a higher order factor accounts
problematic because it marginalizes the importance of our re- for shared variance between measures but also find that that the lower
search. In fact, there is a long history in individual-differences level measures provide unique information in prediction models.
psychology of defining and differentiating constructs that appear Dudley et al.’s (2006) results suggest that using facet-level conscien-
to tap into similar underlying psychological processes on the basis tiousness measures improves the prediction of job performance. How-
of both theoretical and empirical evidence. This history can be ever, finding incremental validity evidence when using a set of mea-
traced back to Kelley’s (1927) discussion of the jingle and jangle sures does not rule out the presence of a higher order factor that
fallacies (Judge, Erez, Bono, & Thoresen, 2002). The jingle fallacy accounts for the shared variance between measures. In other words,
refers to the mistake of assuming that a single psychological finding that conscientiousness facet measures have incremental
construct is being measured on the basis of superficial similarities, validity over global conscientiousness measures is an argument for
when in fact two distinct psychological constructs are present. In using the facet-level measures but should not be interpreted as
comparison, the jangle fallacy refers to the opposite mistake of evidence that conscientiousness facets are not facets of conscien-
assuming that two distinct constructs are present on the basis of tiousness. Similarly, finding that interest and efficacy measures
superficial differences between measures. As such, we view the have incremental validity for predicting career-related outcomes
issue of examining the theoretical and measurement overlap be- (Betz & Rottinghaus, 2006; Donnay & Borgen, 1999; Larson et al.,
tween constructs as a fundamental research question. in press) provides evidence supporting the use of both interest and
To further illustrate this point, consider the results from a series of efficacy measures in career counseling but does not rule out the
four studies examining the covariance between measures of self- possibility that interests and efficacy can be interpreted as com-
esteem, neuroticism, locus of control, and generalized self-efficacy, as ponents of higher order RIASEC constructs.
well as the discriminant and incremental validity of these measures
(Judge et al., 2002). In the first study, Judge et al. (2002) conducted
a meta-analysis to estimate the population correlations between the Interpreting Holland’s Model
constructs, reporting a mean correlation of .60 between measures of
One argument against our research put forward by Lent et al.
each construct. On the basis of these results, they hypothesized that
(2010) is that Holland’s (1959, 1997) theory of RIASEC types is
the measures were indicators of a single higher order construct, which
not meant to encompass self-efficacy. Holland (1959) first pro-
they evaluated in a second study using confirmatory factor analysis
posed his theory long before Bandura (1982, 1986, 1997) devel-
(CFA). The results for the CFA presented in Study 2 indicated that a
oped social cognitive theory and the self-efficacy construct,
higher order factor could account for the relationships between mea-
thereby making it very difficult for Holland to have initially
sures, with a model including the higher order factor producing better
incorporated self-efficacy into his theory. Conversely, Holland
fit statistics than did a model without the higher order factor. In the
(1997) updated his theory after Bandura proposed social cognitive
third study, Judge et al. conducted a series of multitrait–multimethod
theory and defined the self-efficacy construct. The lack of refer-
(MTMM) analyses comparing the four measures. The results of the
ences to the self-efficacy construct in Holland’s later writing could
MTMM analyses were interpreted as supporting the presence of a
be interpreted as evidence for the position that the RIASEC types
higher order factor because the correlations between measures of the
should not include self-efficacy. However, it is also possible that
same trait using different methods tended to be of a smaller magnitude
Holland did not refer to self-efficacy when discussing the RIASEC
than the correlations between measures of different traits using the
model because, from his perspective, this construct was already
same methods. Finally, in the fourth study, Judge et al. examined
represented in the model, albeit using other terms. In fact, one of
correlations with measures of personality traits and job and life
Holland’s key collaborators1 made this very point in an article
satisfaction, finding additional evidence for the presence of a higher
reviewing the Self-Directed Search (SDS; Holland, Fritzsche, &
order factor.
The results of the Judge et al. (2002) study suggest that a higher
order factor can account for the relations between a set of conceptu- 1
In addition to the 15 journal articles listed in the PsycINFO database
ally related measures. Additionally, the results of the analyses indi- published with John Holland between 1975 and 1993, Gary Gottfredson
cated that the individual measures of self-esteem, neuroticism, locus also coauthored the Dictionary of Holland Occupational Codes (Gottfred-
of control, and generalized self-efficacy provided little incremental son & Holland, 1996).
COMMENT ON LENT, SHEU, AND BROWN (2010) AND LUBINSKI (2010) 241

Powell, 1997): “What Holland (1997) called ‘self-beliefs’ are


measured by the SDS Competencies and Self-Estimates scales.
These scales correspond to what Bandura called self-efficacy”
(Gottfredson, 2002, p. 202).
Lent et al. (2010) questioned our decision to put forward Hol-
land’s (1959, 1997) theory as an alternative to SCCT, noting that
the SCCT model provides a framework for identifying the pro-
cesses that contribute to career development, whereas the Holland
model is primarily focused on classification. Holland’s theory is
closely associated with interest measurement and occupational
classification (Campbell & Borgen, 1999; McDaniel & Snell,
1999; Rounds, 1995; Rounds & Day, 1999). However, Holland
consistently defined the RIASEC types as being part of a larger
theoretical perspective on the career development process. In par-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

ticular, Holland argued that an individual’s experiences interacting Figure 1. Traditional social cognitive career theory model of interrela-
with an environment (or learning experiences, using the SCCT tions among learning experiences, self-efficacy, interests, and outcome
term) will contribute to either stability or change in career-related expectations, as well as their prediction of vocational behaviors. LE ⫽
behaviors over time and that this developmental process can be learning experiences; SE ⫽ self-efficacy; IN ⫽ interests; OE ⫽ outcome
interpreted using the RIASEC model. Holland hypothesized that expectations; VB ⫽ vocational behavior.
congruence between the individual’s vocational personality and
the work environment will lead to positive experiences that pro-
mote the stability of career-related behaviors. Conversely, incon- through outcome expectations. In comparison, the effect of inter-
gruence will lead to negative experiences that promote changes in ests on self-efficacy is limited to an indirect mediated path through
career-related behaviors. vocational behaviors and learning experiences. When predicting
Holland (1959, 1997) also hypothesized that differentiation, the vocational behaviors, both self-efficacy and interests have direct
extent to which an individual’s vocational personality matches one paths in the SCCT model, but the overall influence of self-efficacy
of the RIASEC types, will also impact the career development is larger because this construct also has indirect paths of influence
process. Holland predicted that differentiated individuals would mediated through links between outcome expectations and inter-
react to negative experiences caused by incongruence by either ests with vocational behaviors. Self-efficacy both mediates the
changing work environments or modifying their current work influence of more distal variables on subsequent vocational behav-
environment, whereas undifferentiated individuals would change iors and also contributes to more proximate variables that also
their vocational personality to better fit the work environment (see influence outcomes. Therefore, it would appear that the SCCT
Holland, 1997, pp. 61–72). Therefore, although it is true that the model is often represented with self-efficacy as the key variable
RIASEC model is used to classify both individuals and work predicting vocational behavior.
environments (Gottfredson & Holland, 1996; McDaniel & Snell, It should be noted that our observation that self-efficacy is the
1999), Holland (1997) viewed this classification system as a foun- key construct in the SCCT model is not intended as a criticism.
dation for understanding the career development process through- Indeed, the model of interrelations among constructs in SCCT is
out the life course. As such, to suggest that Holland’s RIASEC entirely consistent with how Bandura (1982, 1986, 1997) concep-
model is merely a classification system ignores the potential utility tualized self-efficacy, and Lent et al. (1994) explicitly linked
of the theory as an alternative to the SCCT perspective on how SCCT to Bandura’s theory in the original formulation of their
career choices are shaped by the interaction between individual model. Subsequent theoretical statements and SCCT-based studies
differences and learning experiences. have maintained this link to Bandura (e.g., Lent, 2005, pp. 101–
102). Instead of criticizing SCCT for being based on Bandura’s
Interpreting the SCCT Model theory, the focus of our work is on the identification and evaluation
of alternative models to compare with SCCT. This focus is moti-
Lent et al. (2010) suggested that we misinterpreted the role of vated by our observation that many empirical studies supporting
self-efficacy in their model when we noted that self-efficacy is its the SCCT model, including the longitudinal and experimental
key construct. We are somewhat surprised by this because it is research cited by Lent et al. (2010), have focused on testing
widely accepted that self-efficacy is the key construct in Bandura’s
(1982, 1986, 1997) social cognitive theory and also in the SCCT
2
model based on Bandura’s theory. For example, Lent and Brown The full SCCT model (e.g., Lent, 2005, p. 108) includes constructs not
stated that “self-efficacy beliefs . . . are considered to be central illustrated in Figure 1. These changes were made to simplify the path
determinants of thought and action in social cognitive theory. They diagram presented in Figure 1 and facilitate comparisons with Figures 2
and 3. Constructs not illustrated include background contextual affor-
have, accordingly, received the lion’s share of attention in research
dances, person inputs (e.g., predispositions, gender, race/ethnicity, and
on SCCT” (Lent & Brown, 2006, p. 15). Additionally, as illus- disability/health status), and contextual influences proximal to choice be-
trated in Figure 1, the relative importance of self-efficacy to other haviors. Each of these constructs fall outside of the feedback loop dis-
constructs in the SCCT model can be seen by examining the cussed in this article. Additionally, the construct of vocational behavior is
relative number of paths of influence.2 Self-efficacy has a direct often divided into choice goals, choice actions, and performance domains
effect on interests and an indirect effect on interests mediated and attainments when representing the full SCCT model.
242 ARMSTRONG AND VOGEL

relatively small deviations from it (e.g., Lent et al., 2003, 2005;


Lent, Lopez, Lopez, & Sheu, 2008; Lent, Sheu, et al., 2008) and
that the model-confirming predisposition in this type of research
seems unlikely to lead to new insights. Therefore, we proposed
using Holland’s (1959, 1997) theory to specify an alternative
model to compare with the SCCT model when analyzing interest–
efficacy correlations.
Our decision to focus specifically on Holland’s (1959, 1997)
theory will hopefully not detract from the more general point that
it is important to consider alternative theoretical perspectives when
conducting research. Consider what would have happened if our
results supported the SCCT model over the proposed Holland-
based alternative, with the split between interest and confidence
scales emerging as the first dimension in the MDS analysis,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

interest and confidence scales grouped into two distinct sets of Figure 2. Mutual influence model with bidirectional link between inter-
between-type clusters, and the fit indices obtained in the SEM ests and self-efficacy, their interrelations with learning experiences and
analyses reaching cutoff criteria for the SCCT-based model but not outcome Expectations, and the prediction of vocational behaviors. LE ⫽
for the Holland-based alternative. The conclusions drawn from learning experiences; SE ⫽ self-efficacy; IN ⫽ interests; OE ⫽ outcome
these hypothetical results would be very different from what is expectations; VB ⫽ vocational behavior.
presented in our work. In particular, we would have raised
issues regarding the assumptions underlying Holland’s type
definitions and the use of measures, such as the SDS, that create Holland-based model represents the interests and self-efficacy
total scores from interest, competency, and ability self-estimate constructs as components of higher order RIASEC constructs. As
subscales. More importantly, we would have also argued that discussed in Armstrong and Vogel (2009), the primary advantage
empirical support for the SCCT model was enhanced by our of the Holland-based model illustrated in Figure 3 is its parsimony
consideration of an alternative theory. However, these results over the SCCT model, and this would also be the case when
would not have been as interesting, as it seems unlikely that the comparing it to the mutual influence model. However, as noted by
current thought-provoking debate would have occurred if our Lent et al. (2010), the parsimony advantage of using higher order
research had supported the SCCT model. Holland constructs would only be present in research measuring
interests and self-efficacy across all six of the RIASEC scales.
Alternatives to the SCCT Model Therefore, research focused on domain-specific measures of inter-
ests and self-efficacy in areas such as engineering may not benefit
A number of empirical studies have demonstrated a direct from examining the proposed Holland-based model. Conversely,
reciprocal relationship between interests and efficacy (Bonitz, this type of domain-specific research may be problematic because,
Larson, & Armstrong, in press; Nauta, Kahn, Angell, & Cantarelli, as noted by Lubinski (2010), not measuring the full range of
2002; Tracey, 2002a). In their reaction to our work, Lent et al. interests and self-efficacy beliefs is a violation of Carnap’s (1950)
(2010) acknowledged this point, noting that the relationship be- total evidence rule.
tween interests and self-efficacy “is likely to be reciprocal in
nature” (p. 220). However, they did not acknowledge that the
Linked Measures and Interest–Efficacy Correlations
traditional SCCT model does not include a reciprocal path between
interests and efficacy. As illustrated in Figure 1, in the SCCT One of the issues raised in the SCCT literature is that using
model, self-efficacy has a direct path influencing interests, but the linked interest and efficacy measures may overestimate the mag-
influence of interests on self-efficacy is indirect, mediated through nitude of correlations between the two constructs (Lent & Brown,
other constructs. In comparison, as illustrated in Figure 2, it is 2006; Lent et al., 2010). Research based on the SCCT model using
possible to specify a model with direct reciprocal links between domain-specific measures often reports interest– efficacy correla-
interests and self-efficacy. This bidirectional alternative, or mutual tions that are lower than what was reported in Armstrong and
influence model, represents a potential alternative to the SCCT Vogel (2009). For example, Table 1 lists estimates of interest–
model that could be evaluated in future research. We agree with efficacy correlations from our study, the test manuals for the Skills
Lent et al.’s observation that the empirical support for the mutual Confidence Invention (SCI; Betz, Borgen, & Harmon, 2005) and
influence model is limited to a small number of studies primarily the Campbell Interest and Skill Survey (CISS; Campbell, Hyne, &
based on college student samples. Additionally, their hypothesis Nilsen, 1992), and a number of recently published studies con-
that moderator variables may play a role in determining the pri- ducted with college students in engineering and computing majors
mary direction and magnitude of interest– efficacy influences is by Lent and colleagues (Lent et al., 2003, 2005; Lent, Lopez, et al.,
very interesting, and we agree fully with their position that these 2008; Lent, Sheu, et al., 2008). The mean interest– efficacy corre-
issues deserve further study. lation reported by Armstrong and Vogel was .64, which falls
A second alternative to the SCCT model based on Holland’s within the range of other RIASEC-based measures. In comparison,
(1959, 1997) theory is presented in Figure 3. In comparison to the the interest– efficacy correlations in studies of engineering and
mutual influence model illustrated in Figure 2, which hypothesizes computing students ranged from .42 to .46, with a study-weighted
a direct reciprocal link between interests and self-efficacy, this mean of .43. However, these differences in the magnitude of
COMMENT ON LENT, SHEU, AND BROWN (2010) AND LUBINSKI (2010) 243

cients from samples that do not represent the full range of indi-
vidual differences of the measured attributes (Hunter & Schmidt,
2004). By using a formula for correcting restriction of range
effects, it is possible to estimate what the interest– efficacy corre-
lation for all college students would be on the measures used with
engineering and computing majors. This correction is based on
estimating the ratio of variances in interest and efficacy between
the full student population and the restricted range of variances in
the samples of engineering and computing students. Therefore, the
Figure 3. Holland-based model of interrelations among learning experi- magnitude of underestimation of interest– efficacy correlations due
ences, Holland types, and outcome expectations, as well as their prediction to restriction of range attenuation is dependent on how restricted
of vocational behaviors, with interests and self-efficacy as Holland-type the sample is. If we make the modest assumption that engineering
indicator variables. LE ⫽ learning experiences; SE ⫽ self-efficacy; IN ⫽
and computing majors have above-average interest in engineering
interests; OE ⫽ outcome expectations; VB ⫽ vocational behavior.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

and computing (i.e., that they score above the 50th percentile on
This document is copyrighted by the American Psychological Association or one of its allied publishers.

these measures relative to all college students), then the corrected


interest– efficacy correlations across studies may be attributed to estimate of interest– efficacy correlation would be .623 using the
issues other than shared measurement variance. procedures outlined by Hunter and Schmidt (2004, pp. 103–109).
In particular, restriction of range effects may be present in As such, our finding of a mean interest– efficacy correlation of .64
samples obtained from specific occupations or college majors, across RIASEC categories in a general sample of college students
thereby attenuating estimates of interest– efficacy correlations rel- appears to be consistent with finding an interest– efficacy correla-
ative to studies based on unrestricted samples. The samples of tion of .43 in samples of engineering or computing majors re-
engineering students reported in Table 1 had an average SAT sponding to domain-specific measures.
mathematics score of 674.59, which corresponds to approximately The self-efficacy-based measures most frequently used in con-
the 90th percentile (College Board, 2009). It also seems likely that junction with interest measures, such as the SCI (Betz et al., 2005)
these students would represent a restricted range of interest and and CISS (Campbell et al., 1992), were developed in response to
efficacy in the engineering domain. Indeed, according to the SCCT Bandura’s work on the self-efficacy construct and the emergence
model, students with low interest and efficacy in engineering of SCCT as a model of the career choice and development process.
would be less likely to pursue engineering majors than would Aside from the statistical issue of quantifying the magnitude of
students with high interest and self-efficacy. SAT mathematics
interest– efficacy correlations, it may be essential to use this type
scores for students in the computing domains were not reported by
of linked measures when working with clients in applied settings.
Lent, Lopez, et al. (2008). However, similar to the engineering
As outlined in the SCI manual, the interpretation of jointly admin-
students, it seems likely that these students would represent a
istered interest and self-efficacy measures involves differentiating
restricted range of interest and efficacy in the computing domain.
Therefore, the samples of engineering and computing majors listed between career choices classified as being a high priority for future
in Table 1 should have, on average, interest and efficacy scores in exploration because both interest and confidence levels are high
these domains that are much higher than a general sample of and career choices that are classified as a low priority for future
college students. For this reason, and following the total evidence career exploration because both interests and confidence scores are
rule discussed by Lubinski (2010), the suggestion that our research low. When using this interpretive framework, it is also possible to
has overestimated interest– efficacy correlations needs to be eval- find cases where there are discrepancies between interests and
uated against the possibility that studies using very selective self-efficacy. When an individual has high interest but low self-
groups of students have underestimated these correlations. efficacy in a domain or low interest but high self-efficacy in a
Corrections for restriction of range effects are frequently used in domain, the interpretation is that occupations in this domain with
meta-analytic research to estimate population correlation coeffi- interest– efficacy discrepancies could become a good option for

Table 1
Published Estimates of Interest–Efficacy Correlations

R Domains measured Sample Source

.70 Campbell Interest and Skill Survey Orientation Scales Employed adults (n ⫽ 4,842) Campbell, Hyne, & Nilsen (1992)
.68 RIASEC types Undergraduate students (n ⫽ 331) Betz, Borgen, & Harmon (2005)
.64 RIASEC types Undergraduate students (n ⫽ 608) Armstrong & Vogel (2009)
.61 RIASEC types Employed adults (n ⫽ 362) Betz, Borgen, & Harmon (2005)
.46 Engineering Engineering students (n ⫽ 487) Lent et al. (2005)
.45 Engineering Engineering students (n ⫽ 328) Lent et al. (2003)
.42a Engineering Engineering students (n ⫽ 209) Lent, Sheu, et al. (2008)
.42 Computing disciplines Computing students (n ⫽ 1,208) Lent, Lopez, Lopez, & Sheu (2008)
a
Mean interest– efficacy correlation across time points in a longitudinal study. RIASEC ⫽ Realistic, Investigative, Artistic, Social, Enterprising, and
Conventional.
244 ARMSTRONG AND VOGEL

future exploration if the interest or self-efficacy deficit can be do something and being able to do it well, especially under more
modified. difficulty performance conditions, are somewhat different matters”
A recent study by Tracey (in press) illustrates the potential (p. 221). The implication here is that confidence measures are not
utility of using linked interest and self-efficacy measures to predict optimal measures of self-efficacy, as the threshold for endorsing a
career-related outcomes. Using a sample of 2,145 adults who confidence item would be lower than the threshold for endorsing
completed the Personal Globe Inventory (PGI; Tracey, 2002b), the an equivalent self-efficacy item. We find this argument problem-
effects of interest and efficacy congruence with expressed career atic because Lent et al. did not cite any empirical research to
choices were examined as predictors of career certainty. Overall, support their assumed distinction between confidence and self-
Tracey (in press) found that both interest-choice and efficacy- efficacy measures.
choice congruence were related to career certainty, with higher In our MDS analyses of interest and efficacy, we found that a
levels of congruence linked to higher levels of certainty. Although two-dimensional solution produced a structure consistent with
efficacy congruence did not provide any incremental validity for Holland’s RIASEC model and that adding a third dimension to the
predicting certainty over interest congruence, there was a signifi- analyses produced a separation between interest and efficacy mea-
cant interest– efficacy interaction effect. When there was agree-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

sures that improved model fit by 4.3%. Lent et al. (2010) misin-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

ment between interests and efficacy, there was also a strong terpreted our discussion of this result as implying that this effect is
positive relation between congruence and career certainty, but in “somewhat trivial” (p. 222). In fact, we do not see this result as
cases of interest– efficacy discrepancies, there was little relation being trivial. Improving model fit by 4.3% when adding a third
between congruence and certainty. The PGI requires individuals to dimension separating interest and efficacy scales demonstrates that
rate the same items on both their interests and confidence. There- there are differences in how individuals respond to interest and
fore, we can assume that interest– efficacy discrepancies reflect the efficacy measures. However, because the contribution of this third
attitudes of the respondent and are not artifacts of differences in dimension is small relative to the first two dimensions, our results
item content. In comparison, when using interest and efficacy suggest that interest– efficacy distinctions should be interpreted
measures that were not designed to be used together for a joint within the context of the RIASEC model. Similarly, Lent et al.
administration in either research or applied settings, the results misinterpreted our hierarchical clustering results, noting that “sep-
may be confounded by item content differences. arate self-efficacy and interest subclusters did emerge within each
RIASEC type” (Lent et al., 2010, p. 222). Although this observa-
Other Methodological Considerations tion is technically correct, it ignores the fact that our cluster
analysis, as based on Gati’s (1979, 1991) model, is focused on
Lent et al. (2010) raised a number of specific criticisms regard- evaluating between-type clusters.3
ing how we framed our research question, performed our statistical When reviewing our SEM analyses, Lent et al. (2010) were
analyses, and interpreted our results. They were critical of our use critical of our inclusion of interest and efficacy response factors in
of the Alternate-Forms Public Domain RIASEC marker scales our Holland-based model and also raised questions regarding the
(AFPD; Armstrong, Allison, & Rounds, 2008) in our research, utility of our model for predicting vocational outcomes. We agree
noting that there is limited validity evidence for these measures. with their observation that using method factors is a novel ap-
Although there is structural and convergent validity evidence for proach for interest– efficacy research but would point out that this
the AFPD scales (see Armstrong, Allison, & Rounds, 2008, pp. technique is well established in other areas of individual-
291–293), we agree that additional research on the validity of these differences psychology (Eid & Diener, 2006). Additionally, our
measures is warranted. Additionally, it will be important to repli- MDS analysis demonstrated that there were differences in how
cate our results for the AFPD scales with the commercially avail- individuals respond to interest and efficacy measures. Therefore,
able interest and efficacy measures that are used in applied set- any model fit to this data must account for differences in how
tings. Lent et al. also made the point that RIASEC-based measures individuals respond to the two types of measures. This is done
may be less ideal than domain-specific measures when examining within the SCCT framework by modeling interest and efficacy
the relations between interests and self-efficacy. We agree some- measures as indicators of separate constructs. In our alternative
what with this point, noting that domain-specific measures have model, which hypothesizes that the shared variance between in-
been developed for interests and self-efficacy (Betz et al., 2003) terests and efficacy can be accounted for by Holland-based con-
and that these measures have incremental validity over broad structs, the logical place to account for differences between the two
RIASEC measures (Donnay & Borgen, 1996). However, similar to
what is found with personality facets (e.g., Dudley et al., 2006), the
3
use of these domain-specific measures does not rule out the pres- In Gati’s (1979, 1991) model, the six RIASEC types form three
ence of higher order factors because basic interest scales can be between-type clusters: (R, I), (A, S), and (E, C). In Armstrong and Vogel
conceptualized as facets of the Holland types (Tracey & Rounds, (2009), the four scales for each RIASEC type (activity interest, occupa-
1995). tional interest, activity confidence, and activity confidence) each formed a
Researchers working with confidence measures have generally distinct within-type cluster, and these six type clusters then formed the
three between-type clusters predicted by Gati’s model. These results dem-
assumed that the underlying attribute being measured is self-
onstrate that the interest– efficacy distinction is secondary to the RIASEC
efficacy (Rottinghaus, Larson, & Borgen, 2002). However, Lent et typology. Conversely, if separate between-type clusters had formed for
al. (2010) were critical of our operational definition of the self- interest and efficacy scales (e.g., if the R and I confidence scales formed a
efficacy construct, arguing that there is an important distinction separate cluster from the R and I interest scales), then the results would
between an individual’s level of confidence to perform a task and suggest that the RIASEC typology is secondary to the interest– efficacy
his or her self-efficacy to perform a task, noting that “being able to distinction.
COMMENT ON LENT, SHEU, AND BROWN (2010) AND LUBINSKI (2010) 245

measures is in method factors. However, we agree with Lent et al. cess beyond determining the appropriate level of employment for
that this alternative model needs to be evaluated by examining its an individual.
prediction of vocational outcomes. To the extent that the shared Although we agree with Lubinski’s (2010) position that cogni-
variance between interests and efficacy measures is important for tive ability measures need to be included in vocational research
predicting outcomes, then our model may be preferred due to its and practice, we propose adding personality to his list of estab-
parsimony. Conversely, if the differences in how individuals re- lished constructs that predict career-related outcomes. In a meta-
spond to the two types of measures are more important, then the analysis of longitudinal studies examining personality traits, cog-
SCCT model should emerge as the better solution in future re- nitive abilities, and socioeconomic status (SES) for predicting
search comparing models. mortality/longevity, divorce, and educational and occupational at-
tainment, Roberts, Kuncel, Shiner, Caspi, and Goldberg (2007)
found that the magnitude of effects for personality was comparable
A Broader Individual-Differences Perspective to those of SES and cognitive abilities. Personality traits were also
found to have incremental validity in predicting these important
Lubinski (2010) synthesized a wide range of theoretical, empir- life outcomes above cognitive abilities. Another recent study found
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

ical, and epistemological sources in support of the position that


This document is copyrighted by the American Psychological Association or one of its allied publishers.

that interest and self-efficacy measures had incremental validity


researchers working within the SCCT framework need to incor- over personality traits for predicting the current academic majors
porate measures of cognitive abilities into their study designs. We of college students (Larson et al., in press). Overall, there is
find ourselves in agreement with this point. As noted by Lubinski, empirical evidence demonstrating that interests, cognitive abilities,
cognitive ability scores have implications for the level of employ- personality, and self-efficacy can be used to predict to academic
ment that an individual is best suited for, and occupations that and career-related behaviors and outcomes. However, as noted by
require higher levels of cognitive ability tend to be of higher Lubinski, until research is conducted examining the incremental
prestige than occupations that require lower levels of cognitive validity of using all of these constructs in the same study, there
ability. The importance of evaluating the incremental validity of remain unanswered questions regarding their relative importance.
self-efficacy over cognitive abilities is illustrated in a recent meta-
analysis by Judge, Jackson, Shaw, Scott, and Rich (2007) exam-
Summary and Conclusions
ining the prediction of job performance using self-efficacy mea-
sures after accounting for the effects of general mental ability, One of the key objectives of the research presented in Arm-
personality, and previous work experience. The results of the strong and Vogel (2009) was to encourage the examination of
analyses demonstrated that self-efficacy was a significant predictor alternative theoretical perspectives in vocational research. We
of job performance. However, self-efficacy did not have any believe that the testing of alternative models enhances understand-
incremental validity over the other measures. Judge et al. inter- ing of career-related theory and practice, irrespective of which
preted their results as indicating that Bandura (1997) overstated the specific model is supported by the data. In our research, we have
importance of self-efficacy as a predictor of work-related out- found evidence that a Holland-based interpretation of the relation-
comes. ships of interest and efficacy may be a useful alternative to the
When representing prestige or occupational level in models of SCCT model. However, Lent et al. (2010) raised a number of
the world of work, these constructs are often conceptualized as a issues with our findings on the basis of theoretical and method-
third dimension that is orthogonal to the interest-based data–ideas ological points regarding our research methods, statistical analy-
and things–people dimensions that underlie Holland’s (1959, ses, and interpretation of results. Although we have disagreed with
1997) RIASEC types (Rounds, 1995; Rounds & Day, 1999). For a number of the specific points made by Lent et al., overall, we
example, the spherical model of interests (Tracey & Rounds, 1996) agree with their position that future research may be improved by
includes prestige as a third dimension, and Gottfredson and Hol- highlighting and debating the methodological and theoretical con-
land (1996) proposed using complexity. These conceptualizations siderations that have emerged from our work. We also find our-
imply that once an appropriate level of occupational attainment is selves in agreement with Lubinski’s (2010) position that it is
set by the individual’s ability level, the career choice process is important to demonstrate the incremental validity of new con-
determined more by interests and other noncognitive factors. How- structs in research before their integration into career counseling
ever, there is an emerging body of empirical research on integrat- interventions. Overall, demonstrating the importance of consider-
ing individual-differences measures into trait complexes (Acker- ing alternative models, including the need to examine the interre-
man & Heggestad, 1997; see also Armstrong, Day, McVay, & lations between interests and self-efficacy in relation to other
Rounds, 2008) that has demonstrated links between abilities and constructs within these models, remains one of the essential ob-
the RIASEC model of interests. Additionally, Lubinski (2010) jectives in our work.
presented longitudinal data from Project Talent illustrating differ-
ences in verbal, spatial, and mathematical abilities for individuals References
who completed either a bachelor’s degree, master’s degree, or
Ackerman, P. L., & Heggestad, E. D. (1997). Intelligence, personality, and
doctorate in nine academic disciplines. Spatial and verbal abilities
interests: Evidence for overlapping traits. Psychological Bulletin, 121,
were implicated in the choice of academic discipline, with high 219 –245.
levels of spatial ability, relative to verbal ability, linked to careers Armstrong, P. I., Allison, W., & Rounds, J. (2008). Development and
in the STEM (science, technology, engineering, and mathematics) initial validation of brief public-domain RIASEC marker scales. Journal
fields and the opposite ability pattern linked to non-STEM fields. of Vocational Behavior, 73, 287–299.
As such, cognitive abilities may influence the career choice pro- Armstrong, P. I., Day, S. X., McVay, J. P., & Rounds, J. (2008). Holland’s
246 ARMSTRONG AND VOGEL

RIASEC model as an integrative framework for individual differences. performance: Examining the intercorrelations and the incremental va-
Journal of Counseling Psychology, 55, 1–18. lidity of narrow traits. Journal of Applied Psychology, 91, 40 –57.
Armstrong, P. I., & Vogel, D. L. (2009). Interpreting the interest– efficacy Eid, M., & Diener, E. (2006). Handbook of multimethod measurement in
association from a RIASEC perspective. Journal of Counseling Psychol- psychology. Washington, DC: American Psychological Association.
ogy, 56, 392– 407. Ekehammar, B., & Akrami, N. (2007). Personality and prejudice: From Big
Bandura, A. (1982). Self-efficacy mechanisms in human agency. American Five personality factors to facets. Journal of Personality, 75, 899 –925.
Psychologist, 37, 122–147. Gati, I. (1979). A hierarchical model for the structure of interests. Journal
Bandura, A. (1986). Social foundations of thought and action: A social of Vocational Behavior, 15, 90 –106.
cognitive theory. Englewood Cliffs, NJ: Prentice-Hall. Gati, I. (1991). The structure of vocational interests. Psychological Bulle-
Bandura, A. (1997). Self-efficacy: The exercise of control. New York, NY: tin, 109, 309 –324.
Freeman. Goldberg, L. R. (1993). The structure of phenotypic personality traits.
Beauducel, A., Leipmann, D., Felfe, J., & Nettelsnstroth, W. (2007). The American Psychologist, 48, 26 –34.
impact of different measurement models for fluid and crystallized intel- Gottfredson, G. D. (2002). Interests, aspirations, self-estimates, and the
ligence on the correlation with personality traits. European Journal of Self-Directed Search. Journal of Career Assessment, 10, 200 –208.
Gottfredson, G. D., & Holland, J. L. (1996). Dictionary of Holland occu-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Psychological Assessment, 23, 71–78.


pational codes (3rd ed.). Odessa, FL: Psychological Assessment Re-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Betz, N. E., Borgen, F. H., & Harmon, L. W. (2005). Skills Confidence


Inventory Manual (rev. ed.). Palo Alto, CA: Consulting Psychologists sources.
Press. Holland, J. L. (1959). A theory of vocational choice. Journal of Counseling
Betz, N. E., Borgen, F. H., Rottinghaus, P., Paulsen, A., Halper, C. R., & Psychology, 6, 35– 45.
Harmon, L. W. (2003). The expanded Skills Confidence Inventory: Holland, J. L. (1997). Making vocational choices: A theory of vocational
Measuring basic dimensions of vocational activity. Journal of Voca- personalities and work environments (3rd ed.). Odessa, FL: Psycholog-
tional Behavior, 62, 76 –100. ical Assessment Resources.
Betz, N. E., & Rottinghaus, P. J. (2006). Current research on parallel Holland, J. L., Fritzsche, B. A., & Powell, A. B. (1997). The Self-Directed
measures of interests and confidence for basic dimensions of vocational Search technical manual. Odessa, FL: Psychological Assessment Re-
sources.
activity. Journal of Career Assessment, 14, 56 –76.
Hunter, J. E., & Schmidt, F. L. (2004). Methods of meta-analysis (2nd ed.).
Bibb, T., Steinmayr, R., & Spinath, B. (2008). Personality and achievement
Thousand Oaks, CA: Sage.
motivation: Relationship among Big Five domains and facet scales,
Judge, T. A., Erez, A., Bono, J. E., & Thoresen, C. J. (2002). Are measures
achievement goals, and intelligence. Personality and Individual Differ-
of self-esteem, neuroticism, locus of control and generalized self-
ences, 44, 1454 –1464.
efficacy indicators of a common core construct? Journal of Personality
Bonitz, V. S., Larson, L. M., & Armstrong, P. I. (in press). Interests, self
and Social Psychology, 83, 693–710.
efficacy, and choice goals: An experimental manipulation. Journal of
Judge, T. A., Jackson, C. L., Shaw, J. C., Scott, B. A., & Rich, B. L. (2007).
Vocational Behavior.
Self-efficacy and work-related performance: The integral role of indi-
Campbell, D. P., & Borgen, F. H. (1999). Holland’s theory and the
vidual differences. Journal of Applied Psychology, 92, 107–127.
development of interest inventories. Journal of Vocational Behavior, 55,
Kelley, T. L. (1927). Interpretation of educational measurements.
86 –101.
Yonkers-on-Hudson, NY: World Book.
Campbell, D. P., Hyne, S. A., & Nilsen, D. L. (1992). Manual for the
Larson, L. M., Wu, T. F., Bailey, D. C., Gasser, C. E., Bonitz, V. S., &
Campbell Interest and Skill Survey. Minneapolis, MN: National Com-
Borgen, F. H. (in press). The role of personality in the selection of a
puter Systems.
major: With and without vocational self-efficacy and interests. Journal
Carnap, R. (1950). Logical foundations of probability. Chicago, IL: Uni- of Vocational Behavior.
versity of Chicago Press. Lent, R. W. (2005). A social cognitive view of career development and
Carroll, J. B. (1993). Human cognitive abilities. New York, NY: Cam- counseling. In S. D. Brown & R. W. Lent (Eds.), Career development
bridge University Press. and counseling: Putting theory and research to work (pp. 101–127).
College Board. (2009). SAT percentile ranks, 2009 college-bound seniors: Hoboken, NJ: Wiley.
Critical reading, mathematics, and writing percentile ranks by gender Lent, R. W., & Brown, S. D. (2006). On conceptualizing and assessing
and ethnic groups. Retrieved from http://professionals.collegeboard social cognitive constructs in career research. Journal of Career Assess-
.com/profdownload/SAT-Percentile-Ranks-by-Gender-Ethnicity- ment, 14, 12–35.
2009.pdf Lent, R. W., Brown, S. D., & Hackett, G. (1994). Toward a unifying social
Connor-Smith, J. K., & Flachsbart, C. (2007). Relations between person- cognitive theory of career and academic interest, choice, and perfor-
ality and coping: A meta-analysis. Journal of Personality and Social mance. Journal of Vocational Behavior, 45, 79 –122.
Psychology, 93, 1080 –1107. Lent, R. W., Brown, S. D., Schmidt, J., Brenner, B., Lyons, H., &
Costa, P. T., Jr., & McCrae, R. R. (1992). Revised NEO Personality Tristman, D. (2003). Relation of contextual supports and barriers to
Inventory (NEO-PI–R) and NEO Five-Factor Inventory (NEO-FFI) pro- choice behavior in engineering majors. Journal of Counseling Psychol-
fessional manual. Odessa, FL: Psychological Assessment Resources. ogy, 50, 458 – 465.
Digman, J. M. (1990). Personality structure: Emergence of the five-factor Lent, R. W., Brown, S. D., Sheu, H., Schmidt, J., Brenner, B. R., Gloster,
model. Annual Review of Psychology, 41, 417– 440. C. S., . . . Treistman, D. (2005). Social cognitive predictors of academic
Donnay, D. A. C., & Borgen, F. H. (1996). Validity, structure, and content interests and goals in engineering: Utility for women and students at
of the 1994 Strong Interest Inventory. Journal of Counseling Psychol- historically Black universities. Journal of Counseling Psychology, 52,
ogy, 43, 275–291. 84 –92.
Donnay, D. A. C., & Borgen, F. H. (1999). The incremental validity of Lent, R. W., Lopez, A. M., Lopez, F. G., & Sheu, H. (2008). Social
vocational self-efficacy: An examination of interest, self-efficacy, and cognitive career theory and the prediction of interests and choice goals
occupation. Journal of Counseling Psychology, 46, 432– 447. in the computing disciplines. Journal of Vocational Behavior, 73,
Dudley, N. M., Orvis, K. A., Lebiecki, J. E., & Cortina, J. M. (2006). A 52– 62.
meta-analytic investigation of conscientiousness in the prediction of job Lent, R. W., Sheu, H., & Brown, S. D. (2010). The self-efficacy–interest
COMMENT ON LENT, SHEU, AND BROWN (2010) AND LUBINSKI (2010) 247

relationship and RIASEC type: Which is figure and which is ground? Rounds, J. (1995). Vocational interests: Evaluating structural hypotheses.
Journal of Counseling Psychology, 57, 219 –225. In D. J. Lubinski & R. V. Dawis (Eds.), Assessing individual differences
Lent, R. W., Sheu, H., Singley, D., Schmidt, J. A., Schmidt, L. C., & in human behavior: New concepts, methods, and findings (pp. 177–232).
Gloster, C. S. (2008). Longitudinal relations of self-efficacy to outcome Palo Alto, CA: Davies-Black.
expectations, interests and major career goals in engineering students. Rounds, J., & Day, S. X. (1999). Describing, evaluating and creating
Journal of Vocational Behavior, 73, 328 –335. vocational interest structures. In M. L. Savickas & A. R. Spokane (Eds.),
Lubinski, D. (2010). Neglected aspects and truncated appraisals in voca- Vocational interests: Meaning, measurement, and counseling use (pp.
tional counseling: Interpreting the interest– efficacy association from a 103–133). Palo Alto, CA: Davies-Black.
broader perspective. Journal of Counseling Psychology, 57, 226 –238. Tracey, T. J. G. (2002a). Development of interests and competency beliefs:
McDaniel, M. A., & Snell, A. F. (1999). Holland’s theory and occupational A 1-year longitudinal study of fifth- to eighth-grade students using the
information. Journal of Vocational Behavior, 55, 74 – 85. ICA–R and structural equation modeling. Journal of Counseling Psy-
Nauta, M. N., Kahn, J. H., Angell, J. W., & Cantarelli, E. A. (2002).
chology, 49, 148 –163.
Identifying the antecedent in the relation between career interests and
Tracey, T. J. G. (2002b). Personal Globe Inventory: Measurement of the
self-efficacy: Is it one, the other, or both? Journal of Counseling Psy-
spherical model of interests and competence beliefs. Journal of Voca-
chology, 49, 290 –301.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

tional Behavior, 60, 113–172.


O’Connor, M. C., & Paunonen, S. V. (2007). Big Five personality predic-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Tracey, T. J. G. (in press). Relation of interest and self-efficacy occupa-


tors of post-secondary academic performance. Personality and Individ-
tional congruence and career choice certainty. Journal of Vocational
ual Differences, 43, 971–990.
Paunonen, S. V., & Ashton, M. C. (2001). Big Five factors and facets and Behavior.
the prediction of behavior. Journal of Personality and Social Psychol- Tracey, T. J. G., & Rounds, J. (1995). The arbitrary nature of Holland’s
ogy, 81, 524 –539. RIASEC types: A concentric-circles structure. Journal of Counseling
Roberts, B. W., Kuncel, N. R., Shiner, R., Caspi, A., & Goldberg, L. R. Psychology, 42, 431– 439.
(2007). The power of personality: The comparative validity of person- Tracey, T. J. G., & Rounds, J. (1996). The spherical representation of
ality traits, socioeconomic status, and cognitive ability for predicting vocational interests. Journal of Vocational Behavior, 48, 3– 41.
important life outcomes. Perspectives in Psychological Science, 2, 313–
345.
Rottinghaus, P. J., Larson, L. M., & Borgen, F. H. (2002). The relation of Received February 4, 2010
self-efficacy and interests: A meta-analysis of 60 samples. Journal of Revision received February 10, 2010
Vocational Behavior, 62, 221–236. Accepted February 10, 2010 䡲

You might also like