ArticlePDF Available

Abstract and Figures

Each student faces the challenge of choosing a study program that matches his or her vocational interest. A good person-environment fit (PE fit) between student and study program influences study success and persistence, prerequisites to obtaining the desired degree. But which criterion should be used when presenting advice sets of study options to orient students toward study programs that match their vocational interests? And how long should such a list of study options be? Moving beyond existing, non-evidence-based approaches, present study sets out to develop an empirical advice set engine (EASE) to optimize the process of matching future students to fitting study options. Compared to existing, non-evidence-based alternatives, EASE shows a better balance between the number and PE fit of the options presented. EASE may be a promising way to rethink how student PE fit information can be used in student orientation and higher education research.
Content may be subject to copyright.
AN EMPIRICAL ADVICE SET ENGINE 1
From Interest Assessment to Study Orientation:
an Empirical Advice Set Engine
Stijn Schelfhout a, Bart Wille b, Lot Fonteyne c, Elisabeth Roels a, Filip De Fruyt d and
Wouter Duyck a
Date of submission: 21st of September, 2018
Accepted on 3rd of April 2019, Journal of Experimental Education
a Ghent University, Faculty of Psychology and Educational Sciences, Department of
Experimental Psychology, Henri Dunantlaan 2, 9000 Ghent, Belgium.
stijn.schelfhout@ugent.be (corresponding author, ORCID: 0000-0001-9531-304X),
wouter.duyck@ugent.be
elisabeth.roels@ugent.be
b Ghent University, Faculty of Psychology and Educational Sciences, Department of Personnel
Management, Work and Organizational Psychology, Henri Dunantlaan 2, 9000 Ghent, Belgium.
bart.wille@ ugent.be
c Department of Educational Policy, Student Counseling Office, Campus Ufo,
Sint-Pietersnieuwstraat 33, 9000 Ghent, Belgium
lot.fonteyne@ugent.be
d Ghent University, Faculty of Psychology and Educational Sciences, Department of
Developmental, Personality and Social Psychology, Henri Dunantlaan 2, 9000 Ghent, Belgium.
filip.defruyt@ugent.be
AN EMPIRICAL ADVIC E SET ENGINE 2
Masked Manuscript
Abstract
Each student faces the challenge of choosing a study program that matches his or her vocational
interest. A good person-environment fit (PE fit) between student and study program influences
study success and persistence, prerequisites to obtaining the desired degree. But which criterion
should be used when presenting advice sets of study options in order to orient students towards
study programs that match their vocational interests? And how long should such a list of study
options be? Moving beyond existing, non-evidence based approaches, present study sets out to
develop an Empirical Advise Set Engine (EASE) to optimize the process of matching future
students to fitting study options. Compared to existing, non-evidence based alternatives, EASE
shows a better balance between the number and PE fit of the options presented. EASE may be a
promising way to rethink how student PE fit information can be used in student orientation a nd
higher education research.
Keywords: person-environment fit; vocational interests; study orientatio n; RIASEC model;
empirical advice set engine
AN EMPIRICAL ADVIC E SET ENGINE 3
From Interest Assessment to Study Orientation:
an Empirical Advice Set Engine
A student’s vocational interest plays an important role in contemporary higher education.
For instance, literature indicates that a good person environment interest fit (PE fit) between a
student and a study program predicts academic achieveme nt and persistence (Tracey & Robbins,
2006; Allen & Robbins, 2010; Nye, Su, Rounds, & Drasgow, 2012; Rounds & Su, 2014; Tracey,
Allen, & Robbins, 2012). As both academic achievement and persistence are prerequisites for
graduation, students face an important decision when they have to choose a higher education
study program in pursuit of the desired degree. Especially in educational set ups with low
admission fees or high scholarships, combined with open access to (nearly) all study programs,
the possibilities towards higher education are almost limitless (OECD, 2017). As such, assisting
students in their study choice by presenting them with a manageable list of study programs - also
called an advice set - can provide a substantial support. To generate such an advice set for an
individual student, two factors need to be balanced. How many study programs should the advice
set contain? And how high should the fit quality of the advice set be? Finding a balance between
length and fit quality of the advice set would ensure that (prospective) students receive a
manageable list of suitable study programs to choose from, while the list of programs still allows
for study environment exploration (Holland, 1997). To our knowledge, educational research has
not addressed these problems directly. In fact, educational research on the use of vocational
interest and PE fit information towards study orientation has been quite scarce altogether. As a
consequence, students, student counselors and orientation tools often rely on tradition or non-
evidence based rules of thumb when establishing the length of an advice set.
The goal of the present study consists of introduc ing and exploring an Empirical Advice
AN EMPIRICAL ADVIC E SET ENGINE 4
Set Engine (EASE). This EASE will generate an advice set of appropriate length and fit quality,
for each individual student, based on empirical data of both students and study programs. At the
base of the model, we will use fine-grained methods to model the transition from a very good PE
fit to a very bad one for each individ ual student. By establishing a critical point or threshold in
the balance between the length and fit quality of the advice set, EASE will generate an advice set
for each individual student. As such, we will explore how well our EASE methodology can
balance length and fit quality of advice sets. Indeed, the model used will be refitted for each
student, providing us with a measure of internal validation. We will also compare EASE to the
more classic approaches using congruence indices, providing criterion validity at the student
level (for an overview, see Camp & Chartrand, 1992). This comparison will give us an indication
to what extent we can improve the quality of study orientation if we were to implement our
engine. Finally, as validatio n of EASE at the program level, we will check to what extent
successful students would receive their own study program as part of the EASE generated advice
set of study programs, without inflating the number of choices in this advice set.
The Importance of Vocational Interest in Student Orientation towards Higher Education
A definition of vocational interest usually incorporates a number of key features that
enable and determine higher education study orientation: direction (or prediction),
contextualization (interests have an object), stability and motivation (Rounds, 1995; Rounds &
Su, 2014; Su, Rounds, & Armstrong, 2009). First and foremost, vocational interests robustly
predict study choice (Whitney, 1969). Today, the comparison of students’ interests to study
program environments has become a key element in study orientation. Holland proposed a model
of vocational interests that enables such comparisons by using the same typology to represent
students and study programs (Astin & Holland 1961; Holland 1997). This RIASEC typology
AN EMPIRICAL ADVIC E SET ENGINE 5
takes the form of a clockwise hexagonal pattern containing six interest types or dimensions:
Realistic, Investigative, Artistic, Social, Enterprising, and Conventional (Lippa, 1998). After
decades of model development and evolution, this base concept still remains highly influential,
not in the least in the field of (higher) education (Nauta, 2010).
Second, interests are contextualized and always have an object, like an activity or an
environment (Rounds & Su, 2014). This means that students are interested in activities like
solving equations or translating a conversation, or in environments where these activities take
place, like study programs (e.g., mathematics or applied linguistics) or future occupations (e.g.,
mathematics teacher or interpreter). When constructing interest questionnaires for study
orientation, this object refers to individual study programs and their respective educational
activities. Items and scales probing students’ interests in these activities eventually lead to a
student-specific personal (P) interest profile. Since the inception of the RIASEC model, literature
has always harbored a vast set of instruments to determine such a P-profile (ACT, 2017; Arbona,
2000; Rayman & Atanasoff, 1999; SDS, 2017). For study orientation, such an instrument
typically consists of a relatively large number of items covering the spectrum of human study
related behavior. PROJECT-I, which was specifically designed for student transitio n towards a
higher education setting, is a recent and validated example of this rich assessment tradition
(Authors, 2018). Items of this instrument comprise both occupation titles (e.g., linguist scored on
the Artistic scale) and (study-related) activities (e.g., collecting quantitative and qualitative data
scored on the Investigative scale) that one enjoys, to be scored on a dichotomous yes/no scale.
The score on these items results in a personal RIASEC profile for each (future) student with
(standardized) scores on all six dimensions, ranging from 0 to 100. A set of standardized
RIASEC scores from PROJECT-I will serve as the baseline to develop the EASE methodology
AN EMPIRICAL ADVIC E SET ENGINE 6
in the present study. Apart from our specific study, EASE may however just as well be applied to
person profiles assessed by any Holland instrument other than PROJECT-I.
However, before one can compare a student to a study program, the study program profile
has to be described using the same typology as the student’s P-profile. In contrast to a P-profile,
different approaches exist to describe an environment in terms of the RIASEC dimensions using
an Environment or E-profile. An often used approach in higher education research relies on the
incumbent method (Holland, 1997). This method uses the assumption that a specific environment
is determined through the people in the environment (Schneider, 1987). In other words, applying
this assumption to a higher education setting, a study program is represented through its students.
As such, the interest profiles of students occupying a certain study program environment (the so
called incumbents) are used to determine the interest profile of that study program environment.
As an example from contemporary educational research, Allen and Robbins (2010) defined study
programs in terms of the RIASEC dimensions by averaging out the RIASEC scores of students
who demonstrated sufficiently high levels of academic achievement and persistence. By tracking
a cohort of college freshmen throughout their study curriculum, Allen and Robins (2010) showed
that students with higher levels of congruence between their personal interests and the study
program profiles (as determined through the incumbent method, based on historical data of their
predecessors) had a better chance at obtaining their degree in a timely fashion.
This last example illustrates the importance of a third key feature why vocational
interest is so important towards higher education study orientation. Vocational interests are
regarded as stable constructs (Low, Yoon, Roberts, & Rounds, 2005; Swanson & Hansen, 1988).
Students who have a good match with their study program at the beginning of their higher
education are likely to still have a good match when they graduate. This stability enables the
AN EMPIRICAL ADVIC E SET ENGINE 7
possibility of researching the predictive power of vocational interest on study results of new
students, based on their vocational interest and historical results from graduates within a specific
study program. For instance, recent meta-analytic research on almost 6,000 academic samples
has indicated that vocational interests are moderately correlated to variables indicating
performance and persistence (Nye et al., 2012). Results also showed that especially the
congruence between a person’s vocational interest and his environment was of particular
importance towards performance and persistence. This meta-analysis corrected historical views
that doubted the influence of interests on performance variables because they focused largely on
the absolute level of interest dimension scores rather than PE fit or congruence (Barrick &
Mount, 2005).
The stability feature also enables the validation of study orientation. The attraction-
selection-attrition model predicts that over time students will gravitate towards study programs
that match their vocational interest (Schneider, 1987). This means that successful and persistent
students become excellent incumbents for their (completed) study programs. As such,
researchers can analyze existing or new methods of study orientation by investigating to which
extent successful and persistent students would be oriented towards their original study choice
made years ago. Such criterion validity is usually measured through a hit rate, with literature
reporting numbers between 32% and 69%, depending on the interest inventory used (Burns,
2014; Donnay, 1997). Each match between a (successful and persistent) student’s study program
and the advice given through the method of study orientation is considered a hit for that study
program. Derived from this hit rate, one could also investigate how many times a program was
advised as part of an advice set. This alternative rate (or alt rate) for a study program will
directly influence the length of an advice set. Indeed, if study programs have higher alt rates,
AN EMPIRICAL ADVIC E SET ENGINE 8
students will receive advice sets with more study programs. However, one has to be wary not to
inflate the future student’s advice set with too many study programs in order to boost the validity
and usability of the instrument. Such an expansion of the advice set could overwhelm the student
with too many options and thus hinder the process of environment exploration. When validating
an instrument for study orientation, one should therefore aim at high hit rates for all study
programs, while keeping the alt rate for study programs as low as possible. As an example, if the
study program Economics has a hit rate of 81% with a 25% alt rate, it means that 81% of the
students in this study program (Economics) would receive their own study program as a part of
their advice set. This also means that 25% of the students inside and outside of Economics would
receive this choice as a part of their advice set. In this study, we will explore to which extent the
alt rate (in addition to the hit rate) provides extra information towards the validation of study
orientation. Since both concepts are measured at study program level, the external validation of
our EASE methodology will also be conducted at program level.
As a final characteristic, vocational interest can also act as motivation towards
goal attainment, as described in social-cognitive theories of vocational interest (Lent, Brown, &
Hackett, 1994; Rounds & Su, 2014). Indeed, interest in certain activities like solving equations or
translating texts can (re)direct and energize the student’s endeavors towards studying
mathematics or applied linguistics, thus creating a study environment that facilitates focus on
obtaining the desired degree. As such, the motivational component can explain why a good fit or
match between student and study program leads to academic achievement and persistence.
Fitting Students to Study Programs
Early approaches to determine the PE fit between person (like a student) and environment
(like a study program) profiles have long relied on the comparison of the highest scoring
AN EMPIRICAL ADVIC E SET ENGINE 9
dimensio ns to obtain a congruence index, also called high point coding (Brown & Gore, 1994;
Young, Tokar, & Subich, 1998). In such an index, the letters of the RIASEC dimensions for both
P and E profiles are ranked from high to low based on the dimension scores. This procedure
results in codes that describe students and programs like SAIRCE or CESAIR. By comparing the
rank and placement of the letters in P and E profile, most often only the first letters (one, two or
three dimensions at best), a categorical or ordinal measure of fit is established. As an example,
the Holland congruence index compares the highest dimensions of both P and E profiles
(Holland, 1963). If these dimensio ns are the same (for instance RIASEC vs. RSIACE) the match
between student and study program is deemed a good fit. Although these classic congruence
indices have the advantage of being user friendly and transparent, they also have limitations
(Tracey & Robbins, 2006). To give one example, too much emphasis is put on the absolute level
of the scores, whereas the relative magnitude of the interest dimensions remains underused. For
instance, both P (60, 59, 59, 20, 30, 30) and E (60, 31, 31, 20, 30, 30) profiles would result in an
equivalent three letter code (RIA) based on classic congruence indexing. However, closer
inspection of both profiles reveals substantia l differences. The P-profile displays the highest
score in the R dimension, with I and A being close seconds. In contrast, the E-profile displays a
high R score, with the I, A, E and C dimensions being at a much lower level. The previous
example also illustrates another problem. Letter coding does not provide a solution to tied
dimensio ns (De Fruyt, 2002). Indeed, following the example above, the P- and E-profile could
also have been coded RAI, instead of RIA.
As a reaction to these concerns, alternative measures of PE fit have surfaced. One of
these methods adopts a continuous approach, expressing the fit between P and E through a mere
correlation between profiles, while still being predictive of study success in the first year of
AN EMPIRICAL ADVIC E SET ENGINE 10
higher education (Tracey et al., 2012). For instance, the PE correlation fit between a profile P
(60, 59, 59, 20, 30, 30) and a profile E (60, 31, 31, 20, 30, 30) would amount to r = .62. The
example clearly shows the difference with the letter coding approach that coded both profiles as
RIA without distinctio n. Indeed, by using a correlatio n, the relative magnitude of the dimension
scores in both profiles is taken into account. The difference in elevation of the I and A d imension
of both profiles is reflected in a still high, but less than perfect correlation coefficient. This
approach has the advantage that it uses the entire profile, while also rendering a continuous
measure for further, more fine-grained analyses.
The correlation approach is also immune to the absolute height of RIASEC dimension
scores. Studies have shown that the average elevation of all dimensions does not have a direct
effect on whether or not people want to engage in a certain occupation or activity (Prediger,
1998). However, literature also indicates that within lower elevated profiles the link between PE
fit and results is even stronger (Darcy & Tracey, 2003; Tracey & Robbins, 2006). As such, study
orientation should not focus on the height of the dimensions but on PE fit between profiles to
avoid disadvantaging students with a low profile elevation. To address this problem, PE
correlation fit seems a good solution. Finally, on top of these advantages, the correlation index of
PE fit is still easy to compute and interpret (from -1 being the worst fit possible, to +1 which
represents a perfect fit) without much prior intensive data processing.
Translating PE Fit Information into Study Orientation Advice
Despite the obvious theoretical and empirical advantages, questions remain regarding the
practical implementation of this correlation approach to PE fit. Specifically, in a concrete study
orientation situation, this approach generates a series of correlations between a student’s interest
profile and a set of available study options, reflecting the transitio n from a very good PE fit to a
AN EMPIRICAL ADVIC E SET ENGINE 11
very bad one. Until now, we have no answer to the question how good the PE fit between a
student and a study program has to be before the program should be advised to that specific
student. This lack of a theoretically or empirically based objective criterion delineates a problem
that the more classic congruence indices (see above) also could not solve. Indeed, educational
literature has remained indecisive and vague how the translation from PE fit to study orientation
should be conceived. First, literature displays a multitude of congruence indices, all proposing
different rules to indicate (the degree of) PE fit, each with its own (dis)advantages. As a result,
what is deemed a good fit is only valid within the confounds of one specific index (Brown &
Gore, 1994; Camp & Chartrand, 1992; Young et al., 1998). For instance, the dichotomous
Holland index defines a good fit as a match between the highest dimensions (Holland, 1963).
And second, none of these indices provides an answer to the question of how good exactly the
PE fit between a student and a study program has to be before the program can be advised to that
specific student. In other words, there is no objective and uniform criterion, based on theory or
empirical data that allows for making a distinction between a sufficient fit and an insufficient
one. For instance, in the dichotomous Holland index described above, is it sufficient that only the
highest dimensions match in order to include it in the advice set? Or do the second and third
highest dimensions also need to match between student and program? As a result, contemporary
study orientation still has to rely on mere tradition or suboptimal, non-evidence based rules of
thumb to guide students towards fitting study programs.
Prese nt Study
How high does the fit between a student’s interests and an available study program have
to be in order to take this program into consideration as a potential study option, especially when
AN EMPIRICAL ADVIC E SET ENGINE 12
comparing this fit to that of other, also available study programs? To our knowledge, this
question has not been researched in educational literature. Since there is no evidence-based
criterion, the ideal length of a possible advice set featuring suffic iently high fitting study
programs also remains unknown. The objective of the present study is to answer both issues by
balancing them against each other. As such, we will introduce and explore EASE, an Empirical
Advice Set Engine. At its core, EASE optimizes the process of translating correlation PE fit
information into concrete study advice, using an empirically fueled engine as a base for student
friendly applications. Such a translatio n should always result in a balanced list of suggested
study programs towards environment exploration, while only containing study programs that
match a specific student’s interests to a sufficient degree. To enable this translation, we will use a
fine-grained continuous method of PE correlation fit between a specific student and a list of
available study programs, effectively modeling the transition from programs with a very bad PE
fit, to programs with a very good PE fit. By building on this transition modeling, EASE will
dispense a custom made advice set of study options to each future student individually, while
taking into account the correlation fit between the student’s profile and the entire pool of
available study options. As such, the criterion for this advice set will take the form of a minimal
fit quality or threshold, relative to the available options. Study options that demonstrate a level of
fit surpassing this threshold are included in the advice set, while the remainder of the study
options is excluded, so that they do not have to be explored or processed by the student.
It is important to note that the decision for (not) including any given program into the
advice set is always made relative to the pool of other possible study programs. As study
orientation eventually leads to making a choice of study program, it is only fair that all possible
choices are compared against each other. Ultimately, the proposed procedure should serve as the
AN EMPIRICAL ADVIC E SET ENGINE 13
baseline for data driven applications, while strengthening the quality and validity in establishing
appropriate advice sets of study options for prospective higher education students. To this extent,
the present study will explore three main research questions.
For the first question, we will test how well the novel method succeeds in balancing
advice set length and fit quality for each individual student by using two large data sets
containing student interest measures. The first data set provides us with a large sample of real-
life, successful students from different study fields, used to estimate study program interest
profiles. The second data set provides us with a sample of future students seeking actual
orientation towards fitting study options. As such we will test the following hypothesis,
H1: EASE manages to balance the length and fit quality of student advice sets.
Since the balance between student’s advice set length and PE fit quality is a key feature of this
study, we will also compare EASE to advice sets generated by classic congruence indices, such
as the Holland index discussed above, providing criterion validity at the student level.
H2: EASE displays a better balance between student advice set length and fit quality than
classic congruence indices.
Finally, we will test the validity of the EASE methodology at the program level, by exploring
how many persistent and successful students would receive their own study program as part of
their advice set. We also deem it worth investigating whether receiving the correct study
program as part of the advice set does not needlessly inflate the length of the advice set. We will
thus compare the EASE generated advice sets to those generated through classic indices.
H3: EASE generated advice sets have higher validity than those generated by classic
congruence indices by displaying a better balance between hit rate and alt rate.
Me thod
AN EMPIRICAL ADVIC E SET ENGINE 14
Data sets. All data were primarily gathered in function of a large, university-wide
longitudinal project to enhance study orientation and study success among (future) students at a
Western-European university (Shanghai top 100) with eleven diverse faculties. From this project,
two obtained datasets were used, D1 (N1 = 4,892; 66% female) and D2 (N2 = 7,063; 61% female).
D1 features the scores on the RIASEC questionnaire PROJECT-I (Authors, 2018) from 3rd
bachelor and master students, assessed in the period between August 2013 and September 2015.
Students were recruited from 62 study programs with on average 78 students for each program
and a wide variety in student numbers (SD = 80.20). These students all met the conditions of
academic success and perseverance by completing the first two years of their study program (see
Allen & Robbins, 2010). Only students who indicated that they would consider choosing the
same study program again were included (97%). For each study program, the scores of all
successful students or incumbents were averaged out, following the procedure of Allen and
Robbins (2010). This operation provides us with an E-profile for each of the 62 study programs.
These programs and their E-profile will function as possible study options for the current
investigation. D2 contains the RIASEC interest scores of future students (16 to 18 years old) on
the verge of making the transitio n towards higher education. Interest assessments were
conducted using a freely available, online version of PROJECT-I in the period between January
2014 and September 2015 (see APPENDIX A). Highly irregular (for instance, scores of 0 and
100 on all dimensions) and incomp lete profiles were excluded from the analyses (2%). All
entries were rescaled analogous to D1. There was no overlap between D1 and D2.
Procedures.
EASE. Using the P-profiles of 7,063 future students and the E-profiles of 62 study
programs, we will apply the EASE methodology to each student individually. As we are looking
AN EMPIRICAL ADVIC E SET ENGINE 15
for a way to model the transition from a very good PE fit to a very bad one for each student
individually, we have to correlate the student RIASEC profile (six dimensions) with each of the
62 study program RIASEC profiles (the same six dimensio ns). Such a correlation is a measure of
PE fit quality. Table 1 shows an example for a single random student, ranking the fit quality of
the student with the available study options from high to low. Each study option with a specific
fit quality for an individual student is tied to a number of possible study options. This options
variable indicates how big the advice set of the student would be, if the corresponding fit quality
would act as the threshold (including all programs at or above its fit quality value) for making
the advice set. Exploring the relation between the (PE) fit quality and the number of options for
this student even further, we observe a linear trend between both variables. This trend indicates
that the distribution of fit quality within one student could approximate a uniform distribution,
resulting in a gradual transition from very good to very bad fitting study options. We will test
this approximatio n towards a uniform distribution for each student. Moreover, Figure 1 also
shows that a high fit quality is tied to a low number of options and vice versa. This fit
quality/options combinatio n reflects the balance between the length and the minima l fit quality
of the possible advice set for each student, formally defined as
    (1)
Balance has a single purpose: by finding the best possible balance for a student, we will be able
to determine the optimal threshold for that student, weighing the number of study options against
the minima l fit quality for study options in the advice set. As such, study options with a PE fit
equal to or above this threshold will make up the proposed advice set.
AN EMPIRICAL ADVIC E SET ENGINE 16
In order to find this optimal threshold, we introduce EASE. Its purpose will consist of
finding the optimal threshold for each student separately, based on the balance variable. Further
inspection of table 1 shows that the balance variable rises to more than ten and then goes down
again. In other words, it displays the larger part of a symmetrical and inverted, U-shaped curve,
with the turnover point (the point where the rise stops and the descent begins) somewhere near
the ten point mark of the balance variable. This turnover point is the equivalent of the vertex of a
parabola and corresponds to our intended threshold. In other words, the vertex represents the
point of ideal balance between a sufficient PE fit and an acceptable length of the advice set. If we
can connect the vertex to the corresponding fit quality value, we have our threshold value for the
advice set makeup. As we are looking for a PE correlation fit quality, and the balance variable
displays an inverted U-shape curve, we propose a quadratic linear regression of fit quality on
balance using the functional form of a parabola,
          (2)
to model the balance curve. As such, parameters , and need to be estimated while
represents the residual variance. Expressing fit quality as a function of balance allows us to
estimate the -coordinate of the vertex through its parameters by using
 
   (3)
AN EMPIRICAL ADVIC E SET ENGINE 17
and as such obtain our optimal correlation threshold to reflect the ideal balance between length
and fit quality of the advice set. All options above the computed optimal threshold are deemed of
good enough fit quality and they will be included in the optimal and student-specific advice set.
However, as the parabola is estimated through a regression, there will always be a margin of
error. This margin of error could result in inflated or deflated thresholds, illegitimately
discarding or including study options to form the advice set. Considering we are advocating the
principle of self-directed search, we deem it more important to keep borderline valid options in
the advice set in opposition to discarding the less valid ones. To ensure EASE does not discard
these valid options, we establish the actual threshold at the lower end of the threshold’s
confidence interva l. Because the optimal threshold is based on parameter estimations, we use
parameter confidence intervals (CI) to establish its own CI. In doing so, we take a conservative
approach and use the upper and lower parameter bounds rendering the widest interval. Finally,
the explained variance (R²) of the quadratic regression provides us with a measure of how well
the model fits the data. In other words, the EASE model fit will provide us with an estimate of
how well the EASE methodology managed to balance advice set length and fit quality for a
specific student. In sum, we define EASE as a quadratic linear regression, fueled by the model of
a very good PE fit to a very bad PE fit between a student and a set of possible study programs,
enabling the construction of an actual threshold for each individual student, which ultimately
results in a balanced advice set of appropriate length and sufficient PE fit for each (future)
student. In order for EASE to work, we do make the assumption that the PE fit values between a
student and a pool of study programs entirely cover the correlatio n continuum. This assumption
has to be tested.
AN EMPIRICAL ADVIC E SET ENGINE 18
Congruence indices comparisons. As a test of the last two hypotheses, the EASE
generated advice sets of study programs for each student are compared against more classic
methods of constructing advice sets based on congruence indices, such as the letter congruency
index discussed above. As these congruence indices all have specific features, we choose to
include three classic indices, adapted or combined from the dichotomous first letter agreement
index (Holland, 1963) and the two-letter agreement index (Healy & Mourton, 1983).
For the first comparison (H2), the EASE data and letter method data are acquired from D1
and D2. For the EASE data, the procedure is identical to the one described above. For the
congruence indices using letter methods, i.e. 1L, (one-letter), 2L (two-letter), and 1+2L (one-and
two-letter combination) the procedure for making advice sets is conducted as follows. Study
programs are included in the 1L advice set if the future student and study option profiles have the
same highest scoring RIASEC dimension. For instance, a study program with E-profile code
ECISAR (e.g., economics) would be included in the advice set of a student with P-profile
ERCIAS. Study programs are included in the 2L advice set if the two highest dimensions from
the study program profile reoccur in the three highest dimensions from the future student profile.
For instance, a study program with E-profile code EC ISAR (like economics) would be included
in the advice set of a student with P-profile ERCIAS. Study programs are included in the 1+2L
advice set if the conditions of 1L or 2L are met.
For the second comparison (H3), the data is acquired from D1 (successful and persistent
students) and the procedure for both the EASE application and the letter methods is identical to
the procedure from the first comparison, with one exception: the profiles of the (successful and
persistent) students are also drawn from D1.
Me asures and analyses.
AN EMPIRICAL ADVIC E SET ENGINE 19
PE fit distribution. Before we apply EASE to the data, we have to verify to which extent
the PE fit distribution for each student and the pool of study options approximates a uniform
distribution. A good approximation indicates a gradual coverage of the correlation continuum,
with a linear transitio n from very good to very bad fitting study options. The approximation is
measured through an R², as the result of a regression of options on fit quality (or vice versa).
EASE application. Figure 2 gives an example of an EASE application for a random
student. As the regression of the parabola model has to be carried out for each of the 7,063
students, analyses will report the range, mean and standard deviation across all students of the
following variables: linear fit (R²) (measuring how good the engine manages to balance the
length and fit quality of the engine), the optimal and actual correlation threshold and the advice
set size and average fit quality.
EASE application versus classical 1L, 2L and 1+2L methods. Two comparisons are
made. The first one will compare the balance between average advice set size and fit quality of
both methods at the student level (H2). The second comparison will compare the balance
between the hit rate and alt rate of study programs (H3). To control for the substantial differences
in student numbers across study programs (see above), we use percentages (instead of absolute
numbers) to ensure each study program has the same weight. For each comparison separately,
the EASE results will be projected onto an interpolation of the results from the classic
congruence indices (i.e., 1L, 2L and 1+2L methods). As such, Figure 3 and Figure 4 show two
polynomial interpolations, each consisting of two linear equations (depicted in full). These linear
equations connect the results from 1L with 2L and 2L with 1+2L. Figure 3 depicts the relation
between advice set size and fit quality of a student advice set. A congruence index method with a
lower advice set fit quality is tied to a higher advice set size: students receive more study
AN EMPIRICAL ADVIC E SET ENGINE 20
programs in their advice set, that consequentially fit worse. Figure 4 shows the relation between
hit rate and alt rate of study programs. A congruence index method with a higher hit rate is tied
to a higher alt rate. In other words, by increasing the number of options each student receives (alt
rate), the chance rises they will also receive their own program as a part of the advice set (hit
rate). By projecting the EASE results onto the interpolation of the classic methods (dotted line),
hypothetical values can be established. We can now use a two-sided, one sample t-test to test
whether these hypothetical values significantly differ from the observed EASE values to
investigate if EASE (vs. classic congruence indices) indeed manages to obtain a better balance
between the size and fit quality of a student advice set, and the hit rate and alt rate of study
programs respectively. For an interpretation of the average differences, we will also report a
Cohen’s d effect size, with 0.01 = very small effect, 0.20 = small effect, 0.50 = medium effect,
0.80 = large effect ,1.20 = very large effect, 2.00 = huge effect (Sawilowsky, 2009).
Results
PE fit distribution. Figure 1 already hinted that the transition of PE fit within a student
from a very good fitting study program to a very bad one is a very gradual and continuo us
process, following a uniform distribution. Formally, we tested this assumption for each
prospective student, with an average regression R² across students of .97 (SD = .03), and a range
from .84 to .99.
Hypothesis 1. Our first aim was to test how well EASE would be able to balance advice
set length and fit quality. Our EASE methodology proved to be well capable of balancing advice
set length and fit quality, as indicated by high levels of explained variance. Indeed, the quadratic
regression of balance on fit quality resulted in a linear fit with an average of .99 (SD = .01),
ranging from .86 to .99 across the (prospective) student sample. This high level of explained
AN EMPIRICAL ADVIC E SET ENGINE 21
variance resulted in an accurate estimation of the optimal correlatio n threshold, which ranged
from r = .14 to r = .58 (M = .46, SD = .06). The subsequent actual threshold ranged from r = .11
to r = .53 (M = .44, SD = .06). The width of its confidence interval ranged from .02 to .15 (M =
.06, SD = .02). The advice sets were constructed based on the actual threshold for each
prospective student separately. Figure 5 shows the distribution of the student advice set sizes. A
student thus received an average advice set of almost 18 study options (M = 17.91, SD = 5.37),
which is about 29% of the complete pool of 62 study options. Also, about 98% of the students
received an advice set ranging from 7 to 28 options. Figure 6 shows the distribution of the
(average) fit quality of the student advice sets. The fit quality of study options in an advice set
was on average very high, with M = .69 (SD = .09) and a range in a right-skewed distribution
from r = .18 to r = .87. Also, about 96% of the students had an advice set with an average fit
quality of r = .50 or better. There were no advice sets with zero options. This means that all
students received at least one possible study option as part of their advice set. Considering the
combined results of the analyses regarding our first hypothesis, we decide to accept H1.
Comparison EASE and congruence indices.
Hypothesis 2. Our second aim was to establish whether EASE displays a better balance
between advice set size and fit quality than the classical approaches. Figure 3 clearly indicates
the EASE results are above the interpolatio n line of the classic congruence indices. This
deviatio n from the interpolation line already suggests that EASE manages to balance student
study advice set size and program fit quality better than the classic congruence indices. By
projecting the EASE values onto the interpolation line we can obtain hypothetical values to
formally test the difference between EASE and the interpolation of the classic congruence
indices on both advice set size and fit quality. Note that the congruence indices dispensed a
AN EMPIRICAL ADVIC E SET ENGINE 22
varying number of zero sized advice sets (i.e., students who would receive no valid options). The
1L, 2L and 1+2L methods respectively rendered 401 (6%), 22 (< 1%) and 3 (< 1%) of such zero
sized advice sets
1
. Average fit quality values were computed by excluding the results of zero
sized advice sets.
EASE generated student advice sets display an average size of 17.91 options. Inserting
this value into the interpolation projects an average student advice set fit quality of r = .57. This
is the fit quality that the classical letter methods would generate for an advice set size of 17.91.
However, EASE generated student advice sets with a much better average fit quality of r = .69,
compared to the interpolated r = .57. We can test this difference by using a two sided, one
sample t-test. The difference between the observed EASE value and the projected interpolation
value proved to be significant, t(7062) = 106.54, p < .001, Cohen’s d = 1.27. This means that an
equal advice set size for EASE and the classic congruence indices results in a very large
difference in advice set fit quality, with EASE scoring r = .12 above the level of the interpolation
line. Also, EASE generated advice sets display an average explained variance of 48% (i.e., .69²),
while the classic congruence indices only predict an explained variance of 32% (.57²). In other
words, EASE explains 16% more variance concerning the relation between a student’s P-profile
and his or her advice set of study programs (E-profiles) compared to the interpolatio n of the
classic congruence indices (at equal advice set size).
By keeping advice set size constant, EASE yields better fit quality. It’s also possible to
reverse this rationale: does EASE generate larger advice sets, while still keeping a constant fit
quality? EASE generated student advice sets display an average fit quality of r = .69. Inserting
1
As an explanation, none of the 62 study programs had a RIASEC code starting with C, whereas
students exist for which this is the dominant RIASEC dimension.
AN EMPIRICAL ADVIC E SET ENGINE 23
this into the interpolation equation yields a projection that is off the chart. As such, we propose
to take a conservative approach and adopt the 1L edge value of 12.49 (advice set size) as an
overestimation of the EASE projected value, as the actual interpolation would result in an even
smaller sized advice set. However, EASE generated advice sets with a size of 17.91 options,
compared to the interpolated 12.49. A two-sided, one sample t-test revealed a significant
difference between this EASE advise set size and the conservative projection on the interpolation
line, t(7062), p < .001, Cohen’s d = 1.01. This means that EASE can maintain the same
(maximal 1L) fit as the classic congruence indices, while rendering larger advice set sizes, with a
difference of about 5.42 options. In sum, the tested projections of EASE on the interpolation line
of the classic congruence methods indicate that EASE manages to outperform these classic
methods in balancing advice set size and fit quality. As a consequence, we decide to accept H2.
Hypothesis 3. Our third aim was a test of the validity of the advice sets at the level of the
study program. We investigated whether successful students received their own study program
more often by using EASE over classic congruence indices through a higher hit rate, without
inflating the advice set through a higher alt rate. Figure 4 clearly indicates the EASE results are
below the interpolation line of the classic congruence indices. The deviatio n from the
interpolation line already suggests that EASE will have a higher hit rate at a lower alt rate. To
formally test the difference between the classic congruence indices and the EASE methodology,
we projected the observed EASE values onto the interpolation line of the classic congruence
indices to obtain hypothetical values.
Through the EASE method, study programs receive an average hit rate of .81. Projecting
this .81 on the interpolation line of the classic congruence indices renders an alt rate of .35.
However, EASE displays an alt rate of .27. A two-sided, one-sample t-test revealed EASE has
AN EMPIRICAL ADVIC E SET ENGINE 24
indeed a lower alt rate than the interpolation line, t(61) = -6.04, p < .001, Cohen’s d = 0.77. This
means that an equal hit rate of .81 for both EASE and the classic congruence indices, results in a
large difference in alt rate, with EASE scoring .08 lower than the classic congruence indices. In
other words, EASE improves (lowers) the alt rate of study programs with 23% compared to the
classic congruence indices, at an equal hit rate of .81. In sum, through the use of EASE,
programs have to appear less often in advice set to achieve the same hit rate.
Analogous to hypothesis 3, we can also reverse this rationale. What happens with the hit
rate if we keep the alt rate constant? The EASE method generates an average alt rate in study
programs of .27. Projecting this alt rate on the interpolation line renders a hit rate of .68. EASE
however, displays a hit rate of .81. A two-sided, one-sample t-test revealed EASE has indeed a
lower alt rate than the interpolation line, t(61) = 8.82, p < .001, Cohen’s d = 1.12. This means
that an equal alt rate of .27 for both EASE and the classic congruence indices results in a very
large difference in hit rate, with EASE scoring .13 higher than the classic congruence indices. In
other words, EASE improves (strengthens) the hit rate of study programs with 19% compared to
the classic congruence indices, at an equal alt rate of .27. In other words, if one would present
study programs equally often in study advice through both methods, EASE will yield a higher hit
rate than classical methods. To summarize, the tested projections of EASE on the interpolation
line of the classic congruence methods indicate that EASE manages to outperform these classic
methods in validity by demonstrating a better balance between hit rate and alt rate of study
programs. As a consequence, we decide to accept H3.
Discussion
AN EMPIRICAL ADVIC E SET ENGINE 25
Assisting (prospective) students in their study choice by orienting them towards a set of
study programs that really matches their interest is of great importance to enhance study success
and persistence in higher education (Tracey & Robbins, 2006; Allen & Robbins, 2010; Nye et
al., 2012; Rounds & Su, 2014; Tracey et al., 2012). Until now, extant educational research
remained indecisive and vague on how to translate PE fit into study advice. In the past, students,
scholars and counselors relied on non-evidence based rules of thumb and a plethora of
congruence indices, each with their own flaws and fortes, to establish goodness of fit (Brown &
Gore, 1994; Camp & Chartrand, 1992; Healy & Mourton, 1983; Holland, 1963; Nye, et al.,
2012; Young et al., 1998). Also, literature did not harbor an objective criterion to decide how
well exactly the student’s interests had to match a study program in order for the program to be
eligible as a part of the advice set of study programs presented to a specific student. As a
consequence, the ideal length of such a custom made advice set also remained unknown. This
crux in educational literature is quite surprising as we have argued that vocational interest and
PE fit are of capital importance towards higher education study orientation through the features
of prediction, contextualization, stability and motivation (Lent et al., 1994; Low et al., 2005;
Nauta, 2010; Rounds, 1995; Rounds & Su, 2014; Swanson & Hansen, 1988; Su et al. 2009;
Whitney, 1969). In order to translate PE fit into study advice, the present study proposes the
EASE (Empirical Advise Set Engine) methodology. EASE empirically generates an
individualized advice set of study programs that is sufficiently large and of good fit quality for
each future student. In doing so, the engine balances the number of study programs in the advice
set versus the minimal fit quality required for such a study program to enter the advice set. At its
base, EASE uses the benefits of the fine-grained PE correlation fit measure to model the
transition from a very good PE fit to a very bad PE fit between any given student and a set of
AN EMPIRICAL ADVIC E SET ENGINE 26
study options (Allen & Robbins, 2010; Tracey et al., 2012). By finding the ideal balance between
the number of study options and minimal PE fit, a correlation threshold is generated for each
student. Study programs with a PE fit (regarding the specific student) above the threshold are
added to the advice set and presented to the student as part of the final advice set, while the other
options are no longer taken into account as programs fitting the student’s interests.
To explore the possibilities of our EASE methodology, we presented three research
questions. (1) How well does the EASE methodology succeed in balancing the length and fit
quality of a student advice set? (2) How do the EASE generated advice sets compare to sets
generated with more traditional congruence (letter) indices? (3) How valid is the EASE
methodology?
As an answer to the first question, EASE displays a remarkable ability to balance length
and fit quality of student advice sets by determining an empirical PE fit threshold for each
individual student, through the use of the parabola model. This threshold leads to varied student
advice sets of about 18 study programs, with about 98% of the prospective students receiving an
advice set between 7 and 28 choices, leaving ample room for study environme nt exploration
(Holland, 1997; Gottfredson & Holland, 1975). Our study also includes a number of validation
mechanisms for the parabola model. Indeed, the model fits to all student profiles individually,
while also providing two forms of criterion validity at the student and study program level,
addressed in research questions two and three respectively.
Indeed, as an answer to the second question , EASE presents student advice sets that are
qualitatively better than those generated with the classical congruence indices. For instance, an
EASE advice set of about 18 study programs delivers study advice to future students that
explains 48% of the variance in the relation between the student’s P-profile and the advice set of
AN EMPIRICAL ADVIC E SET ENGINE 27
study programs. This variance level is 16% higher than the level achieved by the classic
congruence indices. Also, about 98% of all students received an advice set with a fit quality of r
= .50 or better.
And finally, as an answer to the third question, our EASE methodology shows strong
criterion validity for study programs; about 81% of all successful students received their own
study choice as part of their EASE generated advice set. Comparing the EASE hit rate to the
range reported in literature (i.e., 32% to 69%), our method seems to be more accurate than using
classic methods of making the PE fit (Burns, 2014; Donnay, 1997). Moreover, EASE also
outperforms a combination of congruence indices by displaying a hit rate that is 19% higher at
equal alt rates (Holland, 1963; Healy & Mourton, 1983). The results from this last question also
show the incremental research value of the alt rate when validating study orientation tools. For
sure, high hit rates in study orientatio n are important to ensure validity, but not at all cost. Good
study orientation should also monitor whether the alt rates are not needlessly inflating the
student’s advice set: if too many less fitting programs are suggested, the process of environment
exploration will suffer (Holland, 1997; Gottfredson & Holland, 1975). Classic congruence
indices may present a strong hit rate or a low alt rate. But EASE has a better balance between
both, with an alt rate that is 23% lower than those of the classic congruence indices, measured at
equal hit rates. As such, EASE presents the right programs to the right students, without having
to present programs too often to achieve that.
In sum, the exploration of our three research questions clearly shows the classical
congruence indices still produce acceptable results concerning fit quality and validity of the
generated advise sets in order to orient students towards higher education. However, when
comparing these results to those obtained through the EASE methodology, EASE provides each
AN EMPIRICAL ADVIC E SET ENGINE 28
student individually with better fitting and valid study options. On top of better quality and
validity of the advice set, the EASE methodology also presents a number of positive features that
the congruence indices fail to reproduce. Indeed, when generating advice sets for future students,
EASE ensures an orientation advice set of at least one study option for each student, while the
congruence indices cannot provide orientation for up to 6% (i.e. the number of zero-sized advice
sets for the 1L method) of all future students. As such, EASE has a better answer (vs. the classic
methods) to the absence of programs with a dominant C -dimension. Next, EASE succeeds in
establishing an objective, data driven and student specific criterion that allows to identify study
programs that should be part of the student specific advice set orientation (vs. study programs
that should be discarded). Finally, EASE establishes this criterion while comparing all available
study programs against each other. This comparison seems only fair when considering study
orientation should ultimately lead to making a choice between study programs.
Theoretical Implications
As an important theoretical addition to the structure of PE fit, we established that the
transition from a very good PE fit to a very bad PE fit is apparently a very gradual and
continuous process for each individual student. The correlation approach thus provides a
continuous, fine-grained measure for modeling the structure of PE fit as an approximated
uniform distribution. This also means that the parabola estimated for each individual student has
properties that find their origin in the uniform distribution of PE fit. Though these properties
were not intended as such, they are a direct consequence of the empirically observed PE fit
distribution and they will influence the length and quality of the advice set. For instance, EASE
uses the symmetry about the parabola vertex to make the distinction which programs are suitable
for the student and which are not. Moreover, this distinction is made more clear cut as the
AN EMPIRICAL ADVIC E SET ENGINE 29
programs are gradually distributed across the course of the parabola. However, EASE does not
use the full course of the parabola. Indeed, EASE does not aim for advice sets equal in length to
everyone. Instead, each student should receive a list of options based on the fit of his profile to
the pool of available programs. By using only a part of the parabola course including the vertex,
EASE also succeeds in balancing the number of suggested options. As such, EASE renders
advice sets of study options that are large enough for the intended self-exploration, without
inflating the advice sets to unworkable lengths.
Moreover, this research does not have to limit itself to the domain of education. We
speculate that the structural uniform distribution of PE fit as shown in this study could also be
present in other settings like work or hobbies, effectively paving the way for the introduction of
EASE in these settings as well. As such, we advocate further research on the distribution of PE
fit between a student and study programs. We also advise to always explore this uniform
distribution when using the EASE methodology.
Practical Implications and Limitations
The analyses above show that EASE offers a good method to offer prospective students a
list of suggested programs, that is not too short or long, and that fits their interests well. The
practical implicatio ns for study orientation towards higher education of the proposed
methodology are tied to a number of boundary conditions that deserve further attention. A first
and obvious requirement is that there are interest profiles of both (future) students and study
programs. Although in the present study these two types of profiles resulted from the same
interest instrument (i.e. PROJECT-I, Authors, 2018) which was administered to future students
as well as successful students, this is not essential. The only prerequisite is that both personal and
environmental profiles are commensurate measures, consisting of the same number of
AN EMPIRICAL ADVIC E SET ENGINE 30
conceptually related (e.g., RIASEC) dimensio ns and thus it both mathematically possible and
conceptually reasonable to compute the correlatio ns between both profiles as a commensurate
assessment. It should be clear from the above that these correlations form the basis of the EASE
method. Any assessment can make use of the EASE methodology, as long as the compared
measures of person and environment are commensurate.
A second requirement consists of a sufficiently large pool of study options. The present
exploration already showed that a set of 62 options is sufficient to extract a very stable advice
set. The high amounts of explained variance in the EASE application do seem to suggest that
even a smaller pool of study options could enable balancing advice set length and fit quality. The
question remains how small the pool of study options can become while still keeping the PE
correlation fit continuum sufficiently covered. This needs to be clarified in future research, while
at the same time asking the question if such a small pool needs an advice set to begin with.
A third requirement is of course a (set of) student RIASEC profiles to apply EASE and
generate advice sets. For individual student orientatio n, the data of a single future student is
sufficient to construct a distribution pattern and apply our EASE to that specific student
generating a valid and precise advice set, containing an appropriate and sufficient number of
study programs.
A final requirement consists of (data fueling) EASE itself. In this exploration, we have
provided but one possible configuration, defining balance as the (simple) product of options and
fit quality in order to pinpoint a correlation threshold. Other setups might require adaptations like
weighting the components, if one or the other would be more important in a specific context.
Future Research and Applications
EASE has the potential to fuel an orientation tool for centers of higher education like
AN EMPIRICAL ADVIC E SET ENGINE 31
colleges or universities that harbor an extensive set of (diverse) study programs. Automating this
engine through an online applicatio n can reach a vast number of (future) students to fill out any
RIASEC questionnaire. This will enable the entire EASE procedure by meeting the mentioned
requirements. By featuring any RIASEC questionnaire, data can be gathered on the profiles of
both actual and future students. Actual students will act as incumbents effectively rendering
study program profiles, while the profile of future students is run through the engine to generate
appropriate advice sets towards study choice. Advice sets can take a form similar to Table 1,
listing appropriate programs instead of a number of options, (not) including the PE fit through a
fit quality for purposes of further exploring the advice set. We also refer to APPENDIX B that
contains a full example for one student, featuring both the EASE execution code and practical
application of the algorithm.
Results from the current set of analyses already suggest that the presented EASE
methodology has the potential to significantly advance our understanding of the concept of PE fit
and how it can be applied in practice. As such, it would also be highly beneficia l to use these
data from automated online applications to facilitate this process of ongoing research. Indeed,
additional research on this method is desired, especially on the properties of EASE across
different instruments and contexts. A correlation fit can be used independently of the featured
instrument, as long as it is possible to establish a correlatio n between a profile P and E. In theory,
this makes our method appropriate for PROJECT-I, UNIACT, Self - Directed Search or any
other Holland-based instrument as long as it features all six RIASEC scales (ACT, 2017;
Arbona, 2000; Authors, 2018; Gottfredson & Holland, 1975; Nauta, 2010; Rayman & Atanasoff,
1999; SDS, 2017). It is worthwhile to compare said instruments on variables such as fit quality
and advice set size.
AN EMPIRICAL ADVIC E SET ENGINE 32
Similarly, EASE offers the ability to explore to which extent and under which form the
EASE method can be applied to contexts other than education as the results from the uniform
distribution approximation seem to indicate. For instance, given the centrality of interests in
many aspects of professional career development, we deem it worthwhile to examine to which
extent this threshold method may also be applied in actual working contexts. The EASE method
could help in generating advice sets consisting of job profiles which can then be linked (e.g., by
labor agencies) to the interest profiles of individual job seekers.
Conclusion
Person-enviro nment interest fit is an important predictor of higher education performance
and persistence. Nevertheless, little progress has been made over the past years in charting
student PE fit distribution and in developing methodologies to translate PE fit informatio n into
valid and workable study advice. The method proposed in the current work introduces a novel
way of translating PE fit into student orientation. Compared to more traditional and mainly
convention-based congruence index approaches to PE fit and study orientation, this new
methodology ensures the creation of advice sets, balanced in length (to enable environment
exploration) and fit quality (in terms of correlation PE fit). In sum, EASE may be a promising
way to rethink how student PE fit information can be used in both fundamental research and
practical applications regarding student orientation and higher education research
AN EMPIRICAL ADVIC E SET ENGINE 33
References
ACT, Inc. (2017). The ACT Interest Inventory technical manual. Retrieved October, 23, 2017,
from http://www.act.org/content/dam/act/unsecured/documents/
Allen, J. & Robbins, S. (2010). Effects of InterestStudy program Congruence, Motivation, and
Academic Performance on Timely Degree Attainment. Journal of Counseling
Psychology, 57(1), 23-35. doi:10.1037/a0017267
Arbona, C. (2000). Practice and research in career counseling and development 1999. Career
Development Quarterly, 49(2), 98-134. doi:10.1002/j.2161-0045.2000.tb00554.x
Astin, A. W., & Holland, J. L. (1961). The environmental assessment technique: A way to
measure college environments. Journal of Educational Psychology, 52(6), 308316.
doi: 10.1037/h0040137
Authors (2018). Masked for review.
Barrick, M. R., & Mount, M. K. (2005). Yes, personality matters: Moving on to more important
matters. Human Performance, 18(4), 359372. doi: 10.1207/s15327043hup1804_3
Brown, S. D., & Gore, P. A. Jr. (1994). An evaluation of interest congruence indices:
Distribution characteristics and measurement properties. Journal of Vocational Behavior,
45, 310-327. doi:10.1006/jvbe.1994.1038
Burns, S. T. (2014). Validity of Person Matching in Vocational Interest Inventories. The Career
Development Quarterly, 62(2), 114-127. doi: 10.1002/j.2161-0045.2014.00074.x
Camp, C. C., & Chartrand, J. M. (1992). A comparison and evaluation of interest congruence
indices. Journal of Vocational Behavior, 41, 162-182.
Counseling Psychology, 45, 204-211. doi:10.1037%2F0022-0167.45.2.204
AN EMPIRICAL ADVIC E SET ENGINE 34
Darcy, M., & Tracey, T.J.G. (2003). Integrating Abilities and Interests in Career Choice:
Maximal versus Typical Assessment, Journal of Career Assessment, 11(2), 219-237.
doi: 10.1177/1069072703011002007
De Fruyt, F. (2002). A person-centered approach to PE fit questions using a multiple trait
model. Journal of Vocational Behavior, 60(1), 7390. doi: 10.1006/jvbe.2001.1816
Donnay, D. (1997). E. K. Strong’s legacy and beyond: 70 years of the Strong Interest
Inventory. The Career Development Quarterly, 46(1), 222. doi: 10.1002/j.2161-
0045.1997.tb00688.x
Gottfredson, G.D., & Holland, J.L. (1975). Vocational choices of men and women comparison
of predictors from self-directed search. Journal of Counseling Psychology, 22(1), 28-34.
doi:10.1037/h0076150
Gottfredson, G.D., & Holland, J.L. (1991). The Position Classification Inventory: Professional
manual, Odessa, FL: Psychological Assessment Resources.
Healy, C. C., & Mourton, D. L. (1983). Derivatives of the Self-Directed Search: Potential clinical and
evaluative uses. Journal of Vocational Behavior, 23, 318-328.
Holland, J. L. (1963). Explorations of a theory of vocational choice and achieve ment. II. A four-
year prediction study. Psychological Reports, 12, 547-594.
Holland, J.L. (1997). Making vocational choices: A theory of vocational personalities and work
environments (3rd ed.). Odessa, FL: Psychological Assessment Resources.
Lent, R. W., Brown, S. D., Hackett, G. (1994). Towards a Unifying Social Cognitive Theory of
Academic Interest, Choice and Performance. Journal of Vocational Behavior, 45(1), 79-122.
doi: 10.1006/jvbe.1994.1027
AN EMPIRICAL ADVIC E SET ENGINE 35
Lippa, R. (1998). Gender-related individual differences and the structure of vocational interests:
the importance of the people-things dimension. Journal of Personality and Social
Psychology, 74(4), 996-1009. doi:10.1037/0022-3514.74.4.996
Low, K. S. D., Yoon, M., Roberts, B. W., & Rounds, J. (2005). The stability of vocational
interests from early adolescence to middle adulthood: A quantitative review of
longitud inal studies. Psychological Bulletin, 131(5), 713737.
Nauta, M.M. (2010). The development, evolution, and status of Holland's theory of vocational
personalities: reflections and future directions for counseling psychology. Journal of
Counseling Psychology, 57(1), 1122. doi:10.1037/a0018213
Nye, C.D., Su, R., Rounds, J., & Drasgow, F. (2012). Vocational Interests and Performance. A
Quantitative Summary of over 60 Years of Research. Perspectives on Psychological
Science, 7(4), 384-403. doi:10.1177/1745691612449021
OECD (2017). Retrieved from http://www.oecd.org/edu/ doi:10.1787/888932461199
Prediger, D. J. (1998). Is interest profile level relevant to career counseling? Journal of
Counseling Psychology, 45(2), 204-211. doi:10.1037/0022-0167.45.2.204
Rayman, J., & Atanasoff, L. (1999). Holland’s theory and career interventio n: The power of the
hexagon. Journal of Vocational Behavior, 55(1), 114-126. doi:10.1006/jvbe.1999.1701
Rounds, J. B. (1995). Vocational interests: Evaluation of structural hypotheses. In D. Lubinski &
R. V. Dawis (Eds.), Assessing individual differences in human behavior: New concepts,
methods, and findings (pp. 177232). Palo Alto,CA: Consulting Psychologists Press.
Rounds, J., & Su, R. (2014). The Nature and power of Interests. Current Directions in
Psychological Science, 23(2), 98-103. doi: 10.1177/0963721414522812
AN EMPIRICAL ADVIC E SET ENGINE 36
Sawilowsky, S (2009). New effect size rules of thumb. Journal of Modern Applied Statistical
Methods, 8(2), 467-474.
Schneider, B. (1987). The people make the place. Personnel Psychology, 40, 437-453.
doi:10.1111/j.1744-6570.1987.tb00609.x
SDS (2017). Self-Directed Search, a part of Aspira 360. Retrieved October, 23, 2017, from
http://www.self-directed-search.com/
Swanson, J. L., & Hansen, J. I. C. (1988). Stability of Vocational Interest over 4-year, 8-year and
12-year Intervals. Journal of Vocational Behavior, 33(2), 185-202.
Su, R., Rounds, J., & Armstrong, P. I. (2009). Men and things, women and people: A meta-
analysis of sex differences in interests. Psychological Bulletin, 135(6), 859-884.
doi:10.1037/a0017364
Tracey, T.J.G., & Robbins, S.B. (2006). The interest-study program congruence and college
success relation: A longitudinal study. Journal of Vocational Behavior, 69(1), 64-89.
doi:10.1016/j.jvb.2005.11.003
Tracey, T.J.G., Allen, J., & Robbins, S.B. (2012). Moderation of the relation between person-
environment congruence and academic success: environmental constraint, personal
flexibility and method. Journal of Vocational Behavior, 80(1), 38-49.
doi:10.1016/j.jvb.2011.03.005
Whitney, D. R. (1969), Predicting from expressed vocational choice: a review. The Personnel
and Guidance Journal, 48(4), 279286. doi:10.1002/j.2164-4918.1969.tb03318.x
Young, G., Tokar, D. M., & Subich, L. M. (1998). Congruence revisited: Do 11 indices
differentially predict job satisfaction and is the relation moderated by person and situation
variables? Journal of Vocational Behavior, 52, 208-223. doi:10.1006/jvbe.1997.1587
AN EMPIRICAL ADVIC E SET ENGINE 37
Figures and Tables
Table 1. Balance between PE fit quality and number of possible study options (length) of the
advice set.
fit quality
options
balance
-0.87
62
-54.1
-0.85
61
-52.09
-0.83
60
-49.61
0.22
28
6.1
0.26
27
6.95
0.26
26
6.8
0.32
25
7.92
0.37
24
8.9
0.39
23
8.87
0.49
22
10.74
0.5
21
10.58
0.51
20
10.18
0.54
19
10.21
0.58
18
10.49
0.59
17
9.97
0.64
16
10.26
0.68
15
10.16
0.7
14
9.79
0.72
13
9.39
0.77
12
9.26
0.79
11
8.71
0.82
10
8.16
0.84
9
7.58
0.9
8
7.17
0,90
7
6.31
0.91
6
5.46
0.92
5
4.62
0.96
4
3.86
0.98
3
2.94
0.98
2
1.96
0.99
1
0.99
AN EMPIRICAL ADVIC E SET ENGINE 38
Figure 1. Scatterplot of options and fit quality for a random student.
AN EMPIRICAL ADVIC E SET ENGINE 39
Figure 2. EASE regression for a random student. Scatter plot data points are depicted in hollow.
The quadratic regression is drawn in full.
AN EMPIRICAL ADVIC E SET ENGINE 40
Figure 3. Projection of EASE on the letter method interpolation of the relation between the size
and fit quality of student advice sets.
AN EMPIRICAL ADVIC E SET ENGINE 41
Figure 4. Projection of EASE on the letter method interpolation of the relation between hit rate
and alt rate.
AN EMPIRICAL ADVIC E SET ENGINE 42
Figure 5. Distribution of advice set size.
AN EMPIRICAL ADVIC E SET ENGINE 43
Figure 6. Distribution of advice set fit quality (based on the correlations between the P and E
RIASEC profiles).
... The theoretical notion that people are attracted to, and perform better in environments that fit with their personal characteristics is a contemporary cornerstone of organizational psychology (Nye, Perlus et al., 2018;van Vianen, 2018), with ample applications in both work (Nye et al., 2012) and higher education (Schelfhout et al., 2019, Schelfhout et al., 2021a. In his original seminal framework, Schneider (1987) formulated three mechanisms that describe how people (1) are attracted to an environment to achieve fit, (2) are selected into an environment because of (perceived) fit and (3) possibly also leave an environment in case of experienced misfit (i.e., attrition). ...
... Assessing PE interest fit is a well-established practice in study orientation (Nauta, 2010) and such assessment can assist in remedying these high rates of failure through study counseling (Schelfhout et al., 2021a). For instance, students that have to choose a specific study program (or major) in higher education would benefit to know if a specific program would fit their vocational interests. ...
... For instance, correlation fit and Euclidean distance are two continuous and more recent measures that have had some merit in charting the beneficial effects of PE interest fit (Tracey et al., 2012). Correlation fit is a technique that does not compare the highest scores between person and environment, but relies on the scoring pattern of the six dimensions (Schelfhout et al., 2021a). Calculating the correlation between the RIASEC scores of the person and the RIASEC scores of the environment compares both profile patterns and thus provides a continuous measure of PE interest fit. ...
Article
Polynomial regression is a proven method to calculate person-environment (PE) interest fit between the RIASEC (realistic, investigative, artistic, social, enterprising and conventional) interests of a student and the RIASEC profile of a study program. The method has shown much larger effects of PE interest fit on academic achievement than earlier approaches in literature. However, the polynomial regression method in its current form only focuses on establishing the regressed interest fit (RIF) of a population of students with their study environments, in order to observe how large the general impact of PE interest fit can become on academic achievement. The present study (N = 4407 across n = 22 study programs) further validates this method towards new applications by theoretically deriving two measures of RIF that only affect a single environment like a study program. Analyses show that the use of RIF for a single study environment results in an even stronger positive relation between PE interest fit and academic achievement of r = 0.36, compared to r = 0.25 for the original polynomial regression method. Analyses also show that RIF for one environment can be used to generate interpretable and reliable RIASEC environment profiles. In sum, RIF for a single (study) environment is a promising operationalization of PE interest fit which facilitate both empirical research as well as the practical application of interest fit in counseling settings.
... In higher education, a good interest fit predicts study success and persistence (Burns, 2014;Donnay, 1997;Päßler & Hell, 2012;Nye et al., 2012;Rounds & Su, 2014;Schelfhout et al., 2021;Schelfhout et al., 2019). In other words, students that fit their study choice have a higher chance of graduating. ...
... Strictly speaking however, predicting study choice in a regression still does not coincide with predicting study choice in actual behavior if both predictors as well as the criterion are questioned at roughly the same time. We therefore opted to use a second, independent data set of former successful and persistent students to construct the program RIASEC profiles by using these former students as incumbents (Allen & Robbins, 2010;Schelfhout et al., 2019;Schelfhout et al., 2021). For the present study, student PE interest fit is therefore determined by comparing the data of incoming students to the data of former senior students. ...
Article
STEM (Science, Technology, Engineering and Mathematics) enrolments in higher education are declining while the STEM gender gap of female underrepresentation seems to widen. The present study addresses both issues by exploring how the fit between a student's vocational interests and the STEM field contributes to a (non-) STEM study choice. Data was collected in the unique setting of an open access and low cost higher education system, which allowed for study of vocational interests without unwanted influence of admission conditions. Specifically, we assessed the interest fit of N = 9162 first-year Belgian university students with (1) the STEM field (i.e., STEM fit) and (2) their specific program of choice (i.e., program fit). Results indicated STEM fit indeed predicted STEM study choice, with a stronger effect in female students. Results also indicated that female students showed a better specific program fit. In order to promote student STEM enrolment and address the gender gap, the present study therefore advocates a gender-specific approach to attract more students with appropriate interest profiles.
... Thus, one should consider that interest assessments may propagate gender differences in occupations -especially in non-traditional ones like Ludwikowski et al. (2019) discuss. Career advisers should be sensitized for such issues, and advice set engines, like Schelfhout et al. (2019a) describe, should be programmed, respectively. ...
Article
Full-text available
Grounding on Holland’s RIASEC model of vocational interests and the respective assumptions on person-environment fit (congruence), this paper focuses on how congruence is related to study outcomes, especially students’ persistence, performance, and satisfaction. The paper distinguishes the measure of congruence with respect to social congruence (SOC) (interest fit with the study mates) and aspirational congruence (ASP) (interest fit with the occupation aspired) and also distinguishes the effects of congruence for gender and six different study areas including Science, Technology, Engineering, Mathematics (STEM), medicine, economics, education, and languages. The paper analyses 10,226 university freshmen of the German National Educational Panel Study (NEPS) and follows them longitudinally with respect to their study outcomes. The results show that students’ persistence was more related to SOC than to ASP, especially for male students. Furthermore, SOC was particularly important for students in STEM areas. Regarding performance, however, ASP was more important. Here, we notably found correlations for STEM subjects with a balanced proportion of female students. Regarding satisfaction, mainly marginal correlations could be found. The results indicate conceptual differences between social and aspirational congruence as well as specific effects for gender and study area. While research might take this into account by specifically developing their models for different study areas, career counseling may reflect on the different significance of the interest-based person-environment fit for different study areas. Initiatives for raising young people’s participation in STEM should therefore specifically focus on students that have high chances to develop interest profiles that are congruent to STEM rather than students who show profiles which already indicate a low congruence.
... The trade fair follow-up is the fourth part of the interview with the focus on evaluation and cost-effectiveness analyses as well as the wishes regarding the linking of the trade fair location, city/region and the education fair. [10][11][12] The expert questionnaire for other stakeholders is structured in a similar way, but is not broken down into the above sections. Relevant questions were taken from each section on the organisation of educational fairs, future development and the interaction between the educational fair and the venue. ...
Article
Full-text available
The aim of this article is to consider why exhibitors and visitors participate in educational fairs and whether these kinds of trade fairs are important for the professional and study orientation of young professionals. The main content of the article is to demonstrate the relevance of educational fairs and their opportunities and to show a basis for decisionmaking for exhibitors on how to determine whether a cost-intensive participation in an educational fair is worthwhile. In principle, the analysis of current literature has shown that education fairs take place in large numbers, especially in Germany, and there is a desire for such presence and direct contact with target groups, especially under the current limitations of the Corona pandemic. Based on research using expert interviews and an online survey, results were generated as to what information provides, why exhibitors and visitors use educational fairs. In conjunction with the literature research, criteria of exhibitors and visitors were compared for participation in the educational fair and then the decision criteria were merged into a pestel analysis. Finally, necessary decisions were summarised on educational fair planning, exhibition execution and educational fair analysis. Educational fairs give exhibitors the possibility of direct contact with target groups and feedback. For visitors, educational fairs provide an overview of offers, possibly new impetus for educational offers or by practical testing the consolidation of the career wish. It is to be expected that educational fairs will remain an important marketing tool in the future, possibly establishing hybrid formats.
Article
Student fail rates in the first year of open access academic higher education can become dramatically high. The present study in Flanders, Belgium examines how performance on program-specific basic skillsets can identify students at risk at the start of their curriculum in 21 bachelor programs (N = 6,624), months before actually failing their exams or dropping out. Results identify up to 58% of the students prone to failure at relatively lower error rates while still adhering to the principles of higher education equity. In practice, institutions and counselors can use the methodology of this study to identify at-risk students and offer them reorientation and remediation trajectories, preventing failure. Future applications towards more restricted or selective international education systems are discussed.
Article
Full-text available
In the Netherlands and Flanders, vocational interest inventories are frequently used to address (study) career dilemmas. In contrast to their popularity in practice, in the Dutch language region, there is relatively little research looking at vocational interest(s) (inventories). This artide introduces the Spherical Model of Vocational Interests and the Dutch translation of the Personal Globe Inventory(PGI; Tracey, 2002), a measure for this model. The Spherical Model adds Prestige interests as a third dimension of vocational interests to the traditional two-dimensional circumplex. Additionally, the Spherical model splits the traditional circumplex in eight, rather than six, interest domains. The quality of the Dutch PGI was investigated with 12 samples. The psychometric qualities of the full Dutch PGI and its short version appeared to be acceptable to exemplary: the items of the scales largely fitted with the appropriate scales, the scales correlated according to the expected circumplex order, and the reliabilities were acceptable. The largest gender difference was found on the People-versus-Things dimension. Additionally, younger and more educated people scored higher on Prestige interests. Future research could further the understanding of the content of the Prestige dimension and how this dimension affects (study) career processes and outcomes.
Thesis
De overgang naar hoger onderwijs is een beproeving voor elke Vlaamse student. Inderdaad, omdat Vlaanderen een open toegang heeft tot hoger onderwijs, kan elke student met een diploma secundair onderwijs aan bijna elke opleiding beginnen. Om het vooropgestelde diploma te halen van de gekozen opleiding, dient een student twee taken tot een goed einde te brengen. De student dient een haalbare studiekeuze te maken. En de student dient te slagen in examens om op schema te blijven om het begeerde diploma te halen. Deze taken zijn niet zo eenvoudig als ze lijken. De data in deze dissertatie geven aan dat slechts 36% van de eerstejaarsstudenten erin slaagt om alle opleidingsonderdelen tot een goed einde te brengen om zo op schema te blijven om tijdig het beoogde diploma te behalen. Om dit onrustwekkende cijfer te verbeteren heeft de Universiteit Gent het SIMON-project (Study Skills and Interest MONitor) gestart. SIMON is erop gericht om studenten die dreigen te falen te (her)oriënteren naar een meer haalbaar studieprogramma vooraleer ze hun examens effectief falen, met verlies van tijd en middelen tot gevolg. In deze dissertatie worden de PAKSOC (praktisch, analytisch, kunstzinnig, sociaal, ondernemend en conventioneel) studie interesses van een student gebruikt om de impact van het SIMON project te vergroten door in de literatuur een aantal openstaande vragen te onderzoeken omtrent studiekeuze en studiesucces. Om dit te bewerkstelligen heb ik in deze dissertatie de uitvoering en resultaten besproken van vier empirische studies. Zo heb ik onder meer gevonden dat de fit tussen de interesses van een student en een set van studieprogramma’s kan worden benaderd via een uniforme distributie. Deze distributie kan dan worden gebruikt als de basis voor een Empirische Advies Set Engine, of ook wel EASE. EASE verstrekt gepersonaliseerde studieoriëntatie voor elke student, gebaseerd op een objectief criterium dat toelaat de lengte en de fit van de set met voorgestelde programma’s te balanceren. Deze balans is superieur aan deze die wordt gegenereerd door meer klassieke indices van interessefit, die trouwens ook worden gebruikt in SIMON. Dergelijke studieoriëntatie kan van cruciaal belang zijn in bepaalde gespecialiseerde gebieden. Als dusdanig heb ik ook onderzocht hoe de interessefit van studenten en studieprogramma’s kan bijdragen tot een economisch belangrijke STEM (wetenschap, technologie, ingenieur en wiskunde) studiekeuze. De resultaten hiervan laten duidelijk zien dat vrouwelijke studenten een betere interessefit hadden met hun gekozen (STEM en niet-STEM) programma in vergelijk met mannelijke studenten. Vrouwelijke STEM - studenten hadden ook een betere interessefit met het STEM veld in vergelijking met hun mannelijke collega’s. STEM studiekeuze en de genderkloof (mannelijke meerderheid) in het STEM veld werden verklaard door een model dat alle PAKSOC dimensies bevatte, naast wekelijkse uren wiskunde in het secundair onderwijs, en de fit met het STEM veld. Een mannelijke STEM keuze was meer gerelateerd aan uren wiskunde in het secundair, terwijl een vrouwelijke STEM keuze meer gerelateerd was aan de fit met het STEM veld. Naast studiekeuze behandelt de huidige dissertatie ook studiesucces. Omtrent dit studiesucces, heb ik ook een verandering voorgesteld in methodologie. Als dusdanig spitst de huidige dissertatie zich toe op identificeren van studenten die dreigen te falen in hun gekozen studieprogramma. Hiertoe heb ik mij vooral gericht op het voorspellen van resultaten van individuele studenten, en niet op het verklaren van populatievarantie in studiesucces, zoals het meestal gebeurt in de literatuur. Deze methodologie valideert ook een set (niet-) cognitieve predictoren voor identificatie van falende studenten. Wat betreft deze identificatie, heb ik ook de mogelijkheid onderzocht om minder strenge vals-positieve (succesvolle studenten die worden geïdentificeerd als falend) ratio’s te gebruiken. Specifiek voor studie-interesses heb ik een aanwezigheidsgraad gevonden van 24% in de identificatiemodellen. Dit betekent dat studie interesses voorkwamen in 24% van de (programma-) specifieke modellen om studiesucces te voorspellen. Dit was de derde meest impactvolle predictor, na studieantecedenten en cognitief vermogen. De relatie tussen studie-interesses en studiesucces wordt ook beïnvloed door de omgeving. Resultaten laten zien dat programma’s een lage diversiteit hebben in de studie interesses van studenten die het programma hebben gekozen. Populaties met een hogere diversiteit werden trouwens gelinkt aan hogere gemiddelde gecontroleerde motivatie en lagere gemiddelde autonome motivatie. In het algemeen was een hogere diversiteit over programma’s ook gelinkt aan betere gemiddelde studieresultaten. Bij een aantal programma’s met een zeer specifiek interessepatroon (hoge sociale dimensie, lage praktische dimensie) observeerde ik echter het omgekeerde effect. Ik vond ook dat de interessediversiteit in programma’s een sterkere invloed had op studiesucces dan individuele interessefit. Om te besluiten, stel ik dat in deze dissertatie, de empirische resultaten en de specifieke operationalisatie van de PAKSOC dimensies en interessefit een uniek perspectief (open toegang) bieden op studie-interesses en hun effect op studiekeuze en studiesucces. Oriëntatie naar een interessante studiekeuze wordt gebaseerd op een objectief criterium: hoe goed moet de fit zijn tussen de interesses van een student en het profiel van een programma? Oriëntatie naar haalbare studiekeuzes wordt gebaseerd op het identificeren van falende studenten door het voorspellen van studiesucces, terwijl er nog altijd wordt rekening gehouden met de specifieke set up van het onderwijssysteem met open toegang. Deze dissertatie stelt studieadviseurs ook in staat deze bevindingen onmiddellijk in de praktijk te brengen.
Article
Full-text available
A new, Holland-based Interest Inventory is proposed, intended to facilitate the transition from secondary to tertiary education. Specific interest items were designed to grasp activities that are prevalent during tertiary studies, including an Academic-track-scale to assist in the choice between academic and vocational-oriented programs. Interest profile descriptions are complemented by a list of matching study programs. Data from 3,962 students were analyzed to evaluate the underlying circumplex structure, the criterion validity of the Academic-track-scale and the study program RIASEC codes. It is concluded that the assessment and feedback tools are promising instruments to facilitate the transition to tertiary education.
Article
Full-text available
Research for more than 60 years has shown that entry into occupations can be predicted from scores on interest inventories at a rate better than chance (Donnay, ). The psychometric scoring methodologies used today by a majority of vocational interest inventories were developed in the 1920s and 1960s. Researchers are challenged with improving the theory and science behind vocational interest inventories to align them with current vocational constructions. In this study, validity comparisons were made between person matching and standard scoring based on 5,143 medical students who had taken a vocational interest inventory and had entered their medical residency. Person matching was found to improve differentiation between occupational groups and increase the amount of information offered in the scoring report; in addition, it could potentially increase occupational group assignment to advance vocational interest inventory validity.
Article
Full-text available
Despite their significance to both individuals and organizations, interests are often misunderstood, and their predictive power is often overlooked. In this article, we discuss the nature of interests, describe several key features of interests, and, contrary to the received knowledge of many, explain how interests can be used to predict career and educational choice, performance, and success. Finally, we discuss the continuity of interests across the life span and explain how evidence of stability supports conceptualizations of interests as being distinct dispositions rather than simply extensions or workplace instantiations of basic personality traits.
Article
The present study investigated long-term stability of vocational interests in a sample of 409 subjects tested with the Strong-Campbell Interest Inventory (SCII; Hansen & Campbell, 1985) as college freshmen in 1974 and retested 12 years later in 1986. In addition, 204 of the subjects also were tested 4 years after their freshman year. Interest stability was determined by computing a Pearson product-moment correlation coefficient, for each subject, between her or his test and retest SCII profiles. Results indicated that (a) there was a remarkable degree of interest stability over all three time intervals; (b) individual differences in stability also were apparent over the three intervals; (c) the stability coefficients were significantly related to self-ratings of stability, and were significantly higher than correlations based on randomly matched profiles; and (d) five methods of operationally defining stability produced somewhat different results in terms of characteristics of the coefficient distributions; however, the different methods resulted in similar rank-orderings of individuals.
Article
The present article focuses on ways in which structural information from the ability and interest domains can be usefully integrated to inform career development and choice processes as well as allied research. Examining the structural models in these two domains, in addition to the distinction between maximal and typical assessment, the authors suggest that self-assessments of ability/self-efficacy have greater utility than assessments of maximal ability for informing career exploration and choice. Given this, they note several ways that self-ratings of ability can be integrated with interest information in clinical settings. In the second section, they focus on the definition of the general factor in vocational interest data. The authors hypothesize that this factor can be interpreted as interest flexibility and is particularly salient as a moderator in the person-environment fit-career outcome relation. They then propose a number of testable hypotheses relevant to this relation.
Article
The current study first longitudinally examines the validity of person–environment (P–E) RIASEC congruence, adopting a job analytic method to assess the environment and using P–E difference scores to compute congruence, to predict a range of intrinsic career outcomes—including job satisfaction, skill development, work involvement, and perceived stress—in a sample of 401 college seniors involved in the labor market. Second, the incremental validity of an alternative conceptualization of the P component (i.e., the Five-Factor Model of personality [FFM]) is examined over and beyond RIASEC congruence scores. Congruence across RIASEC types significantly predicted job satisfaction and skill development, but FFM traits significantly and substantially contributed to the prediction of all intrinsic career outcomes. Finally, a person-centered perspective to P–E questions is introduced, demonstrating that cluster analysis of individuals' FFM scores resulted in two clusters (i.e., internalizers/externalizers and resilients) showing very different positions on the labor market and reporting differing initial career outcomes. It is concluded that RIASEC P–E fit congruence studies should be complemented with FFM assessment to increase predictive validity. Applications of an FFM-based, person-centered approach in I/O and P–E fit research and practice are discussed.