You are on page 1of 16

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/252505427

USING VOCATIONAL INTERESTS IN CLASSIFICATION ALGORITHMS FOR THE


MILITARY

Article

CITATIONS READS

0 49

2 authors, including:

Francois J. Lescreve
Independent Researcher
10 PUBLICATIONS 6 CITATIONS

SEE PROFILE

All content following this page was uploaded by Francois J. Lescreve on 09 January 2014.

The user has requested enhancement of the downloaded file.


USING VOCATIONAL INTERESTS IN CLASSIFICATION
ALGORITHMS FOR THE MILITARY
Annemie Defranc & Francois J. Lescreve

Belgian Defense Staff


Human Resources Directorate General
Accession Policy Research & Technology Section

ABSTRACT
This paper deals with the idea of incorporating an interest measure in classification algorithms
for the military. The starting point is the formula of Maier (1955): P = f (A x M), performance
being best predicted by the interaction of ability and motivation. While abilities are
sufficiently stable over time to permit predictions, motivation to perform in military functions
is to a large extent determined by contextual factors. This hinders the prediction of
performance during selection and classification (S&C). However, some determinants of
motivation, such as interests, are known to be relatively stable. It therefore makes sense to try
to include interests measurement in S&C procedures. So, this paper digs into the assessment
of interests and their use in classification methods. First, the conceptualization of interests is
highlighted. Then the current approach and some existing inventories are described and
evaluated on their usability in a military context. Next, the classification problem is stated and
the integration of interests and abilities in computing the overall utility of assigning a
particular person to a particular job is discussed. Finally, two classification methods are
presented as illustration: the Psychometric Model and the Sequential Parallel Method. The
first method allows the use of interests expressed on a metric scale whereas the second uses
them on an ordinal scale.

INTRODUCTION

The industrial and organizational psychology abounds in research on performance and its
determinants. Ability has been commonly recognized as a good predictor of performance (e.g.
Hunter 1986; Schmidt, Ones, & Hunter, 1992). However, ability may be considered as a
necessary condition for a good performance, it is not a sufficient one: determining that a
person can do the job, does not imply that he will do the job. In 1955, Maier proposed that
performance could best be described as a multiplicative function of ability and motivation (P
= f (A x M)), such that ability is more predictive of performance at higher levels of motivation
(Vroom, 1964).
Motivation is defined by Campbell (1990) as the combined effects on behavior of three
choices: a) the choice to expend effort, b) the choice of the level of effort to expend, c) the
choice to persist at that level of effort. Thus, the multiplicative model of performance implies
that under low levels of motivation, individuals will expend less effort in a given endeavor,
resulting in a low performance regardless of ability. In contrast, when individuals are highly
motivated, ability has a great impact on performance (Sackett, Gruys, & Ellingson, 1998). A
limitation of this model lies in the fact that it only considers individuals level of motivation.
Taking the quality or type of motivation into account could improve the prediction of
performance.
2

However, three recent studies failed to find evidence for the interactive effect of cognitive
ability and trait motivation on performance (Mount, Barrick, & Strauss, 1999; Sackett et al.,
1998; Wright, Kacmar, McMahan, & Deleeuw, 1995). Hirschfield, Lawson and Mossholder
(2004) noted that the three studies used general trait-motivation constructs, rather than
context-specific conceptualizations. When Hirschfield et al. conceptualized motivation as a
context-specific tendency, it did moderate the relationship between cognitive ability and
performance, whereas general achievement motivation did not. Apparently, general trait
motivation is less likely to fully capture the extent to which individuals apply their abilities to
performing in a specific context. In this respect, Vandewalle (1997) proposed a
conceptualization of motivation at three different levels of abstraction: general, context-
specific and task-specific. General trait motivation is not connected to a particular context
and is thus relatively stable over time. Context-specific motivation refers to motivational
tendencies that are more attuned to a given context and is more malleable over time. Group
cohesion, autonomy, leadership, recognition, positive feedback, benefits and payment are
context-specific aspects that motivate individuals to perform well in their job. Task-specific
motivation denotes motivational tendencies that are connected with the task characteristics or
demands.
Two major classes of determinants of motivation can be distinguished: the first class
encompasses a wide range of external influences, such as the setting of goals, the provision of
incentives, the creation of inequities. The second class involves individual difference
variables, such as interests, values, and preferences. In this paper, we would like to focus on
interests and the potential contribution of an interest measure in a classification system. An
important quality of interests is that they are stable over long time intervals. Even at relatively
young ages of 15 or 16 years old, studies indicate that interests are stable. By age 20, the
stability of interest is obvious over test-retest intervals of 5 to 10 years, and by the age of 25,
interests display a remarkable cross-temporal stability (Johansson & Campbell, 1971). As
they do not change greatly over the years, they can be used as dependable guides predicting
job-satisfaction and hence, performance. By contrast, motivational elements determined by
contextual factors are subjected to change, and thus not very suitable for the prediction of
performance. Although interests are generally considered as good predictors of satisfaction,
they are less connected to performance. Doing something you like (e.g., playing tennis) does
not imply that you do it well. However, an interest measure combined with an ability
assessment, should resolve this issue. A similar rationale as formulated by Maier and Vroom,
can then be applied to interests: they can also be regarded as a moderator of the relationship
between ability and performance. Individuals, who are allocated to a job that does not match
their interests, are less likely to be satisfied with their work, resulting in a higher risk of
absenteeism, lower performance and attrition. In selecting and classifying applicants, one
must thus consider the balance between filling vacancies with the most qualified applicants
and taking the preferences and interests of the applicant into account. Incorporating a measure
of vocational interests in the selection and classification process would contribute to a better
person-job fit.

DEFINITIONS OF INTEREST

An extensive literature has been dedicated to the assessment of vocational interests, resulting
in the construction of numerous interest inventories. Despite this large body of empirical
research, vocational psychologists have only infrequently addressed theories and definitions
about vocational interests. This relative inattention has lead to vague and confusing
conceptual definitions of vocational interests and to a lack of consensus. The Dictionary of
3

Psychology (Reber, 1994, p. 282) defines interest as “one of the terms that sneaked unnoticed
into the psychological terminology, especially in the psychology of education. The meaning is
not very accurate and is every now and then used to denote the following: attention, curiosity,
motivation, (goal) orientedness, consciousness, appreciation and desire’. The Dictionary of
Behavioural Science (Wolman, 1973, p.199) defines interest as “an enduring attitude
consisting of the feeling that a certain object or activity is significant and accompanied by
selective attention to that object or activity.” Beyond the vagueness and inconsistency, a lot of
definitions have two claims in common: interest refers to an object or activity that attracts
attention and that yields feelings of pleasure. These two qualities are also found in definition
of Strong (1955) who provided the most influential definition of interest:
Interests are activities for which we have a liking or disliking and which we go
toward or away from, or concerning which we at least continue or discontinue the
status quo; furthermore, they may or may not be preferred to other interests and
they may continue over varying intervals of time. Or an interest may be defined as
a liking/disliking state of mind accompanying the doing of an activity or the
thought of performing the activity. (p. 138)
This definition encompasses four qualitative attributes of interest, namely attention,
pleasurable feeling, direction and activity.
Many vocational psychologists have further distinguished between “expressed” and
“measured” interests (e.g. Spokane & Decker, 1999). Expressed interests are verbal
expressions of interest in an object, activity, task or occupation, and are essentially
idiosyncratic, non-scaled and idiographic. In contrast, measured interests are captured by
means of vocational interest questionnaires and are normative, scaled and nomothetic with
regard to the scores of other respondents (Super & Crites, 1962). Both constructs reflect
vocational interests, but differ in directness and do not always agree within the same person.
When this is the case, expressed interests provide a better prediction of vocational choice than
measured interests (Bartling & Hood, 1981). They have the drawback however, to be harder
to include in a structured S&C system.
As mentioned earlier, we aim to incorporate an interest measure into classification algorithms
for the military. Consequently, a thorough inspection of the current approach in capturing
interests in the Belgian Defense, and of existing interest assessments was made, in order to
identify reliable and valid interest measures that are suitable for use in a military S&C context

CURRENT APPROACH

The current approach falls short in capturing the preferences and interests in a reliable and
valid manner. At the start of the selection, the applicant is asked to fill out an
‘autobiographical form’ that addresses studies, professional experiences, leisure activities and
sport activities. The last part of the form is designed to capture ‘measured interests’, and
contains five questions dealing with content of the function and work style (technical/not
technical, combating/ not combating, in group/alone, outdoors/indoors, and sports/ no sports).
These terms hold the risk of being interpreted differently by the applicants because of their
general meaning. For example, the applicants can interpret the work environment of a tank-
driver as ‘outdoors’, as well as ‘indoors’. At the end of the selection, applicants are asked to
order the various available trades according to their preferences. This is done during an
interview with a career counselor. However, these preferences do not appear to be stable over
time. (De Neve, 2004). A possible explanation is that the candidates answer instrumentally,
by choosing an entry or trade in function of the vacant positions and not in accordance with
4

their intrinsic interests. In short, an alternative method is required to reliably capture the
intrinsic interests and preferences of the applicants.

INTEREST INVENTORIES

Before we describe and evaluate some of the major existing interest inventories, we want to
shed light on the methods used to assess vocational interests. This will be a brief overview, as
we will elaborate on the methods further when describing the interest inventories.
There are two methods to construct scales: a rational method and an empirical one. When
scales are derived by a rational method, items are written, selected and scored in accordance
with a theory. From then on items are evaluated in terms of internal (e.g. internal consistency,
high factor loadings) and external criteria (e.g. occupational membership) and how well they
represent the theory. Scales developed by the empirical method are only databased: item
selection and item scoring are based on the analysis of response data, according to an internal
or external criterion. When an internal criterion is used, scales are relatively homogeneous in
content, as they are designed to measure the same interest. On the other hand, empirical
methods using an external criterion yield scales that are heterogeneous in content and thus
difficult to interpret.
In the light of this paper, it is also useful to draw attention to a) the level of specificity of
interest measurement and b) the nature of the outcome. Interests can differ in their level of
specificity. As exemplified in the various definitions of interests, they are tied to specific
objects or activities, such as interest in tennis or interest in cars. However, broad domains of
interest have also been identified (e.g. scientific interests, artistic interests). The specificity of
interest assessment can thus cover a continuum from general to specific scales. In this respect,
homogenous scales (scales derived by a theory or by an empirical method using an internal
criterion) are more general than heterogeneous scales (using an external criterion). The
outcome of the interest measure can be metric (single values or an interest profile) or
categorical values or a rank order. In the next section, we will highlight some established
measures of interest and one still under investigation, according to their method of
development.
An example of a general, theory-based scale is John Holland’s (1959, 1997) RIASEC model.
It categorizes individuals and environments in six types or combinations of those types:
Realistic, Investigative, Artistic, Social, Enterprising and Conventional. Each personality type
is characterized by a prototypical profile of activities, interests, preferences, values and
competencies. On the basis of interest inventories founded on Holland’s theory, it is possible
to determine which type a person most resembles. Other types that are congruent with the
person’s attitudes can be determined as well. For example, a person might resemble a Social
type most, then an Investigative type, then an Artistic type and then the other types in
descending order. In this way, a personality pattern is obtained for the individual, represented
by a letter code (e.g., SIA). Such a rank ordering of the six RIASEC types only allows
measurement on an ordinal level. Moreover, the general Riasec-model fails to differentiate
between the various military functions, which are mostly Realistic in nature (e.g. Watson,
2003). A more fine-grained, differentiated interest profile is thus needed.
A well-known example of an empirically derived, heterogeneous scale is the “Occupational
Scales” in the Strong Interest Inventory (SII; Hansen & Campbell, 1985; Harmon, Hansen,
Borgen, & Hammer, 1994). These scales measure the degree to which an individual’s pattern
of interests corresponds to those of workers in a wide variety of specific occupations. They
provide the highest level of specificity in measuring vocational interests. The method of
5

developing the occupational scales was to administer the questionnaire to workers in various
occupations, who were employed at least three years in their current job and reported being
satisfied with their job. Then, the responses of a specific occupational group were compared
to those of a general reference group. Any item that was significantly preferred more of less
often by the specific occupational group than by the general reference group, was taken up in
an interest scale for that occupation. Thus, only those items that were able to differentiate the
two groups were selected. The score of an individual on a scale does not point to the
magnitude of interests, but to the extent to which the interests of the individual resemble the
interests of the specific occupational group. Heterogeneous scales have the advantage of being
less transparent for respondents, for it is not obvious which item belongs to which scale. In
this way, social desirability and instrumental responding are restricted to a minimum. As
mentioned above, the disadvantage of heterogeneous scales is their lack of psychological
meaning, due to the way in which items were selected. The only criterion is their ability to
discriminate between occupational groups and the general reference group, regardless of their
content.
An example of an empirically derived, homogenous scale is the Jackson Vocational Interest
Survey (JVIS; Jackson, 1977). Jackson used a combination of theory and empirical work to
guide item construction and created an instrument that is praised as being one of the most
carefully constructed scales available (Kline, 2005). He initially created over 3000 items,
based on the vocational literature, and relied on various statistical analyses to retain only the
most useful ones. The JVIS measures 34 basic interest scales, 26 work role scales and 8 work
style scales. Scale scores of homogenous scales in general can differ in nature (profile, metric,
rank order).
A specific interest inventory, developed for use in a military context, is currently under
investigation in the United States Navy. The inventory is termed ‘Job and Occupational
Interest in the Navy’ (JOIN) and is designed to assess applicant’s interest in specific work
activities and environments (Farmer, W.L., & Alderton, D.L., 2003; Michael, P.G.,
Hindelang, R.L., Watson, S.E., 2002; Watson, S.E., 2003). It will be used in conjunction with
the ‘Rating Identification Engine’ (RIDE), a new classification decision support system. The
main goal behind this new system is to improve the match between a recruit’s abilities,
interests and specific occupations. The JOIN model was developed such that interest profiles
for entries were based on the job evaluation of Subject Matter Experts (SME’s). This
evaluation was done by means of job interest elements, which were derived from all available
job descriptions from the Navy Enlisted Community Manager’s (ECM) website. In these job
descriptions, words were extracted that reflected the following categories: 1) job families, or
Navy community areas (i.e., aviation, construction, submarine, etc.); 2) work activity
dimensions, a process (verbs) and/or content (nouns) words; and 3) work context dimensions,
or the work environment (i.e., working indoor, working with a team, etc.). The process and
content words were connected in various combinations to form process-content (PC) pairs,
allowing participants to indicate their level of interest in the work activity (e.g., maintain-
mechanical equipment). Three pictures of personnel performing job tasks accompany each
PC-item. Preliminary analyses indicate a good internal consistency and a good usability of the
system. However, the criterion-validity still has to be established.
The approach of the United States Navy seems apt for a number of reasons. First, it allows a
differentiation between the various military functions, in contrast to general scales as
RIASEC. Second, existing theory and research suggests that it is favorable to use context- or
task-specific measures, as they recognize the variability of tendencies across different types of
activity domains, and thereby enhance the predictive validity. Thus, rather than matching a
person’s general personality type to a broad job domain, and then retrieving jobs in that
6

domain, an argument is made for specific profile comparison with respect to context-specific
activities, as is the case in JOIN. Third, as noted by Farmer and Watson, developing an
interest measure from job interest elements based on SME evaluation has the advantage that it
is very efficient when an entry is added or modified. Job interest elements can be easily added
and subtracted from a job profile, and entirely new job interest elements can be added to the
system based on SME input.
However, the use of pictures in the JOIN inventory raises some questions about validity. A
picture contains ample information, which can produce a different impact and a different
interpretation on the applicants. Moreover, it is difficult to outline all disturbing variables.
In deciding what method is most appropriate (theory-driven or data-driven and heterogeneous
or homogenous scales), Savickas and Spokane (1999) advise to consider the needs of the
population. If the population consists of individuals who require only preliminary direction to
guide their exploration in the vocational world, it can be clear that those individuals would not
be served with a large number of empirically derived scales. On the other hand, if the
population consists of individuals who are differentiating between various occupations that lie
all in the same interest domain, empirically derived scales may be indicated, because of their
subtlety.
Lastly, we would like to point to the fact that it is not recommended to simply integrate the
outcome of an interest inventory into classification algorithms without consulting the test-
taker first. Most interest inventories are designed to guide and advise the client in his/her
choice. The results of an interest inventory serve as an initiation to further reflect on
vocational choice. It can be clear that the results of an interest questionnaire should always be
discussed with the applicant, before incorporating the interest outcome into classification
algorithms.
As it was pointed out earlier, the assessment of vocational interests can result in a variety of
measurements. The assessment can yield:
• a profile describing the person’s interests (e.g. RIASEC);
• a metric value expressing how well an entry matches the person’s interests;
• a rank order expressing how well different entries match the person’s interests.
In the next part, we’ll focus on the use of this data in classification algorithms for the military.
We’ll first briefly sketch the classification problem and advocate that this is the preferred
method to make enlistment decisions. Next, we’ll discuss how vocational interests can be
helpful for classification purposes and finally, we’ll illustrate this using two different
classification methods.

THE CLASSIFICATION PROBLEM

Traditional selection methodology aims at maximizing Person-Job fit (P-J fit). For example,
when we consider a group of applicants for a single entry, we’ll usually try to assess how well
each of these applicants ‘fit’ for the job. The P-J fit can for instance be expressed by means of
a composite score that is made up of relevant test-scores weighed according to a multiple
regression in which job performance is the dependent variable. Once these composite scores
are available for all applicants, they can be sorted and the persons with the highest predicted
performance can be hired.
7

In the military context, things are a bit more complicated. This is primarily due to the fact that
many entries are available and that most applicants are willing to accept more than just one of
them. Technically spoken, this is a multi-entry, multi-applicant situation. The consequences
are that P-J fits must be computed for all combinations of applicants and entries they’re
prepared to accept. Assuming that the required qualities differ among the entries and indeed
that the entries vary in the extent in which they match a person’s interests, the P-J fit of a
person will be different for each entry. The problem to solve can then be seen as an applicants
by entries matrix containing P-J fits. This matrix contains the information needed to decide
which applicant is to be accepted for each entry. Given the usually gigantic number of
possible solutions, some sophisticated algorithm is practically required to solve this problem.
There must of course be a good reason to proceed in such a way. The reason is the quality of
the solution. Compared to other –widespread- solutions, the classification method sketched in
previous paragraph yields far better quality (Lescreve, 2002).

HOW VOCATIONAL INTERESTS CAN HELP

To function properly, the classification algorithms must process information expressing how
desirable it is to allocate a particular applicant to a specific entry. That information
encompasses two basic facets: skills and motivational aspects. As we’ll see in following
section, algorithms are available to either use these facets together or separately.
Let’s elaborate a bit on the first option. In that, the aptitude estimate and the expected fit in
terms of interests are integrated to a new value. That value is a single quantification
expressing how desirable it is to allocate a person to a job. Reducing all available information
to a single value has the advantage of simplifying the classification problem and allowing the
use of very powerful algorithms. In practice, as we’ll see, this seems the better option for
exactly that reason. It is the power of the algorithm searching for the best solution that makes
the difference.
In the second option, skills and motivational aspects are not integrated but treated jointly. This
means that for each applicant-entry combination, two data are available: one expressing the
person’s ability to perform well and one expressing how the person’s interests match the
entry. Since the two are not merged, no explicit assumptions regarding the way they should be
integrated are needed (multiplicative, additive…). In practice however, classification
algorithms need to balance the relative importance of both aspects in some way and this may
prove to be problematic (see example in next section).
As we pointed out earlier, the outcome of interest’s measurement can take different shapes: a
profile, a single metric value or a rank order. In each case, these measurements express the
applicant’s interests for a particular entry. We would not recommend the use of an interests-
profile for classification purposes. A profile always includes a series of values whereas in the
classification context, we’re only interested in a single quantification to estimate the M factor
in P = f(A x M). In addition, algorithms using only one value are easier and more powerful.
So, in the event the interests’ assessment results in a profile, we would advocate to reduce it
to a single metric value. Different techniques are available to do so. In essence, these are
based on the match between an applicant’s profile and a profile established, either empirically
or by means of subject matter experts, for the entry. We’ll now have a closer look at two
classification systems. The first one uses interests on a metric scale while the second uses
them on an ordinal scale.
8

TWO EXAMPLES

The Psychometric Model


The Psychometric Model is a smart classification tool used for the classification of the
Belgian NCOs from 1995 to 2003. Although it would bring us too far to describe the tool
exhaustively in this paper1, some comprehension of its principles is necessary for our purpose.
The tool is meant for batch classification and requires a pool of applicants and a set of jobs.
Identical jobs are clustered in ‘trades’ or ‘entries’. Each trade can contain several identical
jobs. On the applicant’s side, the Psychometric Model uses selection variables (both metric
and categorical) and preferences towards the different trades. For each entry, a minimum
and/or maximum value can be set for each metric variable together with a weight. That weight
can be compared to a beta weight used in multiple linear regressions. An initial payoff is
computed for each applicant-trade combination as a weighted sum using the applicant’s
standardized metric variable scores and the weights given to those variables for the trade.
Applicants not meeting the set minima or maxima for a trade are rejected for that particular
trade.
For each entry, for each categorical variable, a coefficient can be set for each class. Such a
coefficient will then be used multiplicatively to adapt the initial payoff. Coefficients larger
than one will be set for classes that are highly appropriate for the considered entry. For
instance, a person who took training in mechanics (and belongs to that class) could get an
increase of his payoff of 10% (coefficient = 1.1) for a trade such as ‘Vehicle & Equipment
Maintenance Apprentice’. Coefficients smaller than one will have the opposite effect.
Coefficients zero will reject an applicant for the particular entry. For instance, a candidate
who hasn’t a specific security clearance (and thus belonging to a class ‘no clearance’ on a
variable ‘security status’) can be rejected for an entry like ‘Signals Intelligence Analysis
Apprentice’ by setting the coefficient for this combination to zero.
In a third step, the payoffs will be adapted again according to the applicant’s preference
towards the trade and the importance that is given to respecting the choices of the applicants.
In accordance with Maier’s view, the integration of ability and preferences in computed in a
multiplicative way. After that third step, we obtain payoff values that represent the utility of
assigning the applicant to a particular trade.
In the last step, we put all these payoff values in a matrix containing the applicants as row
headers and the jobs as column headers. The matrix is squared by adding dummy jobs (or
dummy persons) with payoff values set to zero. Then, by using an operational research
algorithm derived from the well-known Hungarian method, we link each applicant to a job in
such a way that the sum of the payoffs of the applicants for the jobs they’re assigned to is
maximal. By doing so we capitalize as much as possible on the differential aptitudes and
interests of the applicant pool to satisfy the different requirements of the different entries.
The integration of metric, categorical and interests data is achieved by following formula:

u v w
Yij = ( ∑ βmj.Xim)( ∏ γcij)( ∏ [((Xijp/XMaxjp).ϕpj)+(1-ϕpj)])
m= 1 c =1 p =1

1
For more detailed descriptions of the Psychometric Model, please refer to the bibliography. Copies of the
papers can be made available. Contact the authors for that purpose.
9

Yij is the payoff-value of person i for job j;


m (1 to u) represent the metric variables;
βmj is the weight given to variable m for job j;
Xim is the score of person i on variable m;
c (1 to p) represent the categorical variables;
γcij is the coefficient given for job j to the category of variable c to which person i belongs;
p (1 to w) represent the variables concerning the preferences;
Xijp is the expressed preference of person i for job j on variable p;
XMaxjp is the scale maximum of Xijp. The reason why this is required, is to obtain a maximum value of 1
for the expression Xijp/XMaxjp ;
ϕpj is the weight given to preference variable p for job j;

The Sequential Parallel Method


The starting point of the ‘Sequential Parallel Assignment Method’ is quite similar to the one
in the Psychometric Model. For each entry, a payoff is computed to predict the ability part of
the equation. The preference of the applicants for the different entries is expressed on an
ordinal scale. Once all applicants completed the selection procedure, a table is made for each
entry. The tables contain three fields: the person identification, the payoff and the rank of
their preference for the table-entry (a value 1 indicates that the entry is the first choice of the
applicant, 2 is his/her second choice, etc). The tables are sorted in descending order of the
payoff. When an applicant is not interested at all in the entry, zero replaces his/her payoff.
The next figure shows an example of such tables.
Entry A Entry B

Table 1 Table 2

The Sequential Parallel method then considers the number of vacancies per entry. If n is the
number of vacancies, then the method will assign persons to the trade if they are among the n
best ranked persons for the entry and the entry is their first choice. Say for instance that there
are 4 vacancies for Entry A and 4 for Entry B in our example in the figure above. For Entry
A, the four best ranked applicants would be assigned to the entry while for Entry B, nobody
would since the four best ranked persons did not have Entry B as first choice. This would be
done for all entries.
Once a person is assigned to a job, his or her record is deleted in the other tables. In our
example for instance, we can see that person with Person-ID 84061018758 (second ranked in
Entry A and first in Entry B) is assigned to entry A for this entry is his/her first choice and
10

s/he is among the 4 (because there are 4 vacancies) best ranked for the entry. Keeping his/her
record in entry B is then pointless and therefore it can be removed. When considering the 4
(because there are also 4 vacancies for this entry) best-ranked persons for entry B, we now
also have to consider the person who was the 5th best ranked for the entry. In the event this
entry is his/her first choice (which it isn’t in our example), s/he would get entry B.
This procedure is continued as long as persons can be assigned to their first choice.
Then, after adapting the number of vacancies per entry (set to the original number minus the
number of persons assigned to the entry in the run considering first choices), the second
choice is considered in a similar way. Then the third choice is reviewed and so on.
In this method, it can happen that an applicant is assigned to his/her second choice for
instance because at that time s/he doesn’t qualify for the entry of his/her first choice. If,
during the classification process, due to the deletion of the records of persons who were
assigned to an entry of a higher choice, this person’s first choice becomes available for
him/her, then the person will be assigned to his/her first choice while the vacancy of his/her
second choice will be made available again for other persons. The graphic below shows the
logic of the method.

For Rank = 1 to # Entries (review each rank)


Run 1
For Entry = 1 to # Entries (review each entry)
For Person = 0 to # Available jobs (review first ranked persons)
If PersonRank <= Rank and Payoff > 0 then
If Person already assigned then delete assignment and reopen job
assign Person to Entry
delete Person for Entries where PersonRank > PersonRank
end if
Next Person
Next Entry
If there was an assignment during this run, do another run
Next Rank
Graph 1
The method stops as soon as there are no more vacancies or all choices have been reviewed.

DISCUSSION

While both methods use the applicants’ interests or choices to reach a classification solution,
it is useful to point a few differences out. We already mentioned that the Psychometric Model
(PsyMod) uses metric values denoting the applicant’s preferences while the Sequential
Parallel method (SeqPar) uses the ranks of the preferences. This means SeqPar is more
flexible since it can be used for rank-data or a conversion of metric to ordinal data. PsyMod
requires metric data. The only limitation of SeqPar is that the method doesn’t allow ties. On
the other hand, it is clear that an ordinal scale doesn’t give adequate information about the
perceived distance between successive ranks. An example from the NCO recruiting will
illustrate this point.
11

Person A Person B
Entry Preference Entry Preference
Infantry 99 Navy Radio technician 99
Armor 98 Air Force Electronics 80
Artillery 97
Transportation 80
Table 3

In this example, Person A tells us on a metric scale from 1 to 99 that there is little difference
between his three highest choices. Person B on the contrary indicates that to him, there is an
important difference between his two choices. He tells us that he wouldn’t be very satisfied if
he was assigned to his second choice whereas for Person A, it doesn’t really matter which of
his three first choices he gets (but he wouldn’t be happy with the fourth). The Sequential
Parallel method only looks at the ranks and therefore has less information. The question at
stake is whether Person A wouldn’t be more satisfied when he’s assigned to his third choice
than Person B would be when assigned to his second choice? The Psychometric Model works
with the metric preferences and will consequently give precedence to the assignment of
Person A to any of his first three choices compared to the assignment of Person B to his
second choice.
A second difference related to the applicants’ preferences, is the way in which these are
processed. The Sequential Parallel method doesn’t integrate aptitudes and preferences. The
method sorts the applicants for each trade in descending order of their aptitudes and assigns
them taking the rank order of their choices into account as was explained earlier. The focus of
the algorithm in SeqPar alternates between ability and choice and very well may yield
counter-intuitive assignments. PsyMod integrates aptitudes and preferences for each trade.
This originates from Maier’s idea that the level of performance results from the product of
aptitude and motivation. The integrated value of aptitude and preference specifies the utility
of assigning a person to a particular job. That is the value that is used by the Psychometric
Model in the search of an optimal solution for the classification problem.
A third facet of the topic is that of the importance that is given to the respect of the applicants’
preferences. The role that is played by the preferences in the Sequential Parallel method is
fixed. It is known that there is an interaction between the respect of the preferences and the
respect of aptitudes. If the recruitment manager wants to maximize aptitudes, there is only
little room to respect the applicant’s choices. Conversely, if we want to satisfy the applicants,
this will most probably happen at the expense of aptitude. This is illustrated by the next graph.
12

Quality versus Preferences


Recruitment: Flemish NCO Level 2, 2000
750 100

95
725

Mean Preference
90
Mean Payoff

700
Sequential Parallel Assignment (85,66)
85

675
80

Sequential Parallel Assignment (665,58)


Preference
650 75
0,0 0,2 0,4 0,6 0,8 1,0 Payoff

Preference weight in Psychometric Model

Graph 2
This graph shows for the same recruitment, the average aptitude (left ordinate) and the
average preference (right ordinate) for different levels of preference weights in the
Psychometric Model. In addition, the graph shows the fixed level of average payoff and
average preference yielded by the Sequential Parallel method (horizontal lines).
The graph clearly shows the interaction between aptitude and preference levels. If one wants
to maximize the aptitude, this will happen at the expense of the respect of preferences and
vice versa. The Sequential Parallel method doesn’t allow varying the importance given to the
preferences.
Based upon empirical evidence, a usual value given to the preference weight is 0.6. When this
value is set for the considered recruitment, you can see that the Psychometric Model performs
better both for the respect of preferences and for yielding higher aptitudes.

CONCLUSION

This paper started with recognizing that while ability is a good predictor for performance, it is
not a sufficient condition. Motivational elements moderate the effect of ability on
performance. These motivational elements are known to be determined to a large extent by
contextual factors. Yet, cross-temporal motivational elements such as interests can be
measured and therefore prove useful in performance prediction. For the purpose of
classification, two sets of data are needed: one pertaining to the ability (invariable over the
entries) and another one related to the applicant’s interests (possibly different for each entry).
We illustrated ways to incorporate these interests in classification algorithms, whether the
measurement yields metric or ordinal values.

REFERENCES

• Alley, W. E., Recent advances in classification theory and practice. In M. G. Rumsey, C.


B. Walker & J. Harris (Eds)., Personnel selection and classification. Hillsdale, New
Jersey, Lawrence Erlbaum Associates, Publishers. 1994.
13

• Bartling, H.C., & Hood, A.B. (1981). An 11-year follow-up of measured interest and
vocational choice. Journal of Counseling Pscyhology, 28, 27-35.
• Campbell, J. P. (1990). Modeling the performance prediction problem in industrial and
organizational psychology. In M. D. Dunnette & L. M. Hough (Eds.), Handbook of
industrial and organizational psychology (Vol. 2, pp. 687-732). Palo Alto, CA: Consulting
Psychologists Press.
• Darby, M., Grobman, J., Skinner, J. & Looper, L., The Generic Assignment Test and
Evaluation Simulator. Human Resources Directorate, Manpower and Personnel Research
Division, Brooks AFB, 1996.
• De Neve, G. (2004). Ontwikkelen van een interessevragenlijst toepasbaar op de
instroomfuncties binnen Defensie. Niet gepubliceerde licentiaatsverhandeling. Vrije
Universiteit Brussel, faculteit voor psychologie en educatiewetenschappen.
• Farmer, W.L., & Alderton, D.L. (2003). Vocational Interets Measurement in the Navy –
JOIN. Procedings of the 45th Annual Conference of the International Military Testing
Association.
• Hansen, J.C., & Campbell, D.P. (1985). Manual for the SVIB-SCII (fourth ed.). Stanford,
CA: Stanford University Press.
• Hardinge, N. M., Selection of Military Staff. In International Handbook of Selection and
Assessment., Edited by Neil Anderson and Peter Herriot, Wiley, 1997. p 177-178.
• Harmon, L.W., Hansen, J.C., Borgen, F.H., & Hammer, A.L. (1994). Strong Interest
Inventory: Applications and technical guide. Palo Alto, CA: Consulting Psychologists
Press.
• Hirschfeld, R.R., Lawson,L., & Mossholder, K.W. (2004). Moderators of the Relationship
Between Cognitive Ability and Performance: General Versus Context-Specific
Achievement Motivation. Journal of Applied Social Psychology, 34, 2389-2409.
• Holland, J.L. (1959). A theory of vocational choice. Journal of counseling Psychology, 6,
35-45.
• Holland, J.L. (1997). Making vocational choices: a theory of vocational personalities and
work environment (3rd edition). Odessa, Florida: Psychological Assessment Resources,
Inc.
• Hunter, J.E. (1986). Cognitive ability, cognitive aptitudes, job knowledge and job
performance. Journal of vocational behaviour, 29, 340-362.
• Jackson, D. N. (1977). Manual JVIS Jackson Vocational Interest Survey. Port Huran,
Michigan: Sigma Assessment Systems.
• Johansson, C.B., & Campbell, D.P. (1971). Stability of the Strong Vocational Interest
Blank for men. Journal of applied psychology, 55, 34-36.
• Keenan, T,. (1997) Selection for Potential: The Case of Graduate Recruitment. in
International Handbook of Selection and Assessment., Edited by Neil Anderson and Peter
Herriot, Wiley, 1997. p. 510.
• Kroeker, L. & Rafacz, B., Classification and assignment within pride (CLASP): a recruit
assignment model., US Navy Personnel Research and Development Center, San Diego,
CA, 1983.
14

• Lescrève, F., A Psychometric Model for Selection and Assignment of Belgian NCO’s in
Proceedings of the 35th annual conference of the Military Testing Association. US Coast
Guard, . 1993 p. 527-533.
• Lescrève, F., The Selection of Belgian NCO’s: The Psychometric model goes operational.
in Proceedings of the 37th annual conference of the International Military Testing
Association. Canadian Forces Personnel Applied Research Unit. 1995, p. 497-502.
• Lescrève, F., The Psychometric model for the selection of N.C.O.: a statistical review.
International Study Program in Statistics, Catholic university of LEUVEN, 1996
• Lescrève, F., Data modeling and processing for batch classification systems. in
Proceedings of the 39th Annual Conference of the International Military Testing
Association, Sidney, 1997
• Lescrève F., Immediate assessment of batch classification quality. In Proceedings of the
37th annual conference of the International Military Testing Association., 1998, Internet:
www.internationalmta.org
• Lescrève F., Equating distributions of aptitude estimates for classification purposes. In
Proceedings of the 40th annual conference of the International Military Testing
Association., 2001.
• Lescrève F., Improving Military Recruit Quality Through Smart Classification
Technology. Final Report of an International Collaborative Research Project Sponsored
by the US Office of Naval Research, October 2002
• Maier, N.R.F. (1955). Psychology in industry (2nd ed.). Boston: Haighton-Mifflin.

• Michael, P.G., Hindelang, R.L., & Watson, S.E. (2002). JOIN: Job and Occupational
Interest in the Navy. Proceedings of the 44rd Annual Conference of the International
Military Testing Association.
• Mount, M. K., Barrick, M. R., & Strauss, J. P. (1999). The joint relationship of
conscientiousness and ability with performance: Test of the interaction hypothesis.
Journal of Management, 25, 707-721.
• Reber, A.S. (2002). Woordenboek van de psychologie. Uitgeverij Bert Bakker.
(Dictionary of psychology).
• Sackett, P. R., Grays, M. L., & Ellingson, J. E. (1998). Ability-personality interactions
when predicting job performance. Journal of Applied Psychology, 83, 545-556.
• Savickas, M.L. & Spokane, A.R. (1999). Vocational interests. Meaning, measurement and
counselling use. Davies-Black publishing. Palo Alto, California.
• Schmidt, F. L., Ones, D. S., & Hunter, J. E. (1992). Personnel selection. Annual Review
of Psychology, 43, 627-670.
• Spokane, A.R., & Decker, A.R. (1999). Expressed and measured interests. In M.L.
Savickas & A.R. Spokane (Eds.), Vocational interests (pp. 211-233). Palo Alto, CA:
Davies-Black.
• Strong, E.K., Jr. (1955). Vocational interests 18 years after college. Minneapolis:
University of Minnesota Press.
• Super, D.E., & Crites, J.O. (1962). Appraising vocational fitness (2nd ed.). New York:
Harper & Bros.
15

• VandeWalle, D. (1997). Development and validation of a work domain goal orientation


instrument. Educational and Psychological Measurement, 57, 955-1015.
• Vroom, V.H. (1964). Work and motivation. New York: Wiley.
• Watson, S.E. (2003). Job and occupational interest in the navy. Proceedings of the 45th
Annual Conference of the International Military Testing Association.
• Wolman, B.B. (1973). Dictionary of behavioural science. New York: Van Nostrand
Reinhold.
• Wright, P. M., Kacmar, K. M., McMahan, G. C., & Deleeuw, K. (1995). P = f(MXA):
Cognitive ability as a moderator of the relationship between personality and job
performance. Journal of Management, 21, 1129-1139.

View publication stats

You might also like