Professional Documents
Culture Documents
net/publication/229450777
CITATIONS READS
10 1,043
3 authors:
11 PUBLICATIONS 35 CITATIONS
University at Buffalo, The State University of New York
398 PUBLICATIONS 7,738 CITATIONS
SEE PROFILE
SEE PROFILE
Rajan Batta
University at Buffalo, The State University of New York
189 PUBLICATIONS 6,216 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Colin G Drury on 24 May 2019.
ABSTRACT
Personnel assignment problems have traditionally been solved using operations research (OR) meth-
ods. The purpose of the research was to demonstrate the value of taking a systems view of person-
nel assignment beyond the usual OR models. Human factors methodologies and research methods
were used to describe and analyze the components of the personnel assignment process at a printing
company. These included macroergonomics, human–computer interaction, the skills–rules–
knowledge (SRK) framework, hierarchical task analysis, decision ladders, and abstraction decom-
position spaces. This approach facilitated assessments for function allocation and work system design,
which are argued to be necessary for successful implementations of any solution for personnel
assignment problems. © 2006 Wiley Periodicals, Inc.
1. INTRODUCTION
Correspondence to: Karen S. Holness, Department of Industrial Engineering, University at Buffalo, Buffalo,
New York 14260. E-mail: ksh_ny@yahoo.com
285
286 HOLNESS, DRURY, AND BATTA
optimization models, there is little or no discussion of how the input data was collected
and verified. There are also few descriptions of how the solutions were applied and inte-
grated into the workplace from a systems and organization design standpoint. The liter-
ature review in Section 2 gives more detail. Instead, emphasis is placed more on using
real organization data to validate the developed models. This practice points to a larger
issue in model development. Specifically, it raises the question, “What organizational or
process design factors do OR practitioners actually consider when developing these mod-
els?” In fact, the ability to assess the impact of a designed solution on an organization
may influence the choice of variables, constraints, solution approach, and ultimately solu-
tion implementation. Although these specific issues are not addressed here, they are a
necessary part of future PA research.
Nonetheless, accomplishing solution implementation would require additional tech-
niques beyond optimization model creation and verification. Questions such as, “What
should the model be capable of doing?” or “Who is the best person to use the model to
generate assignments?” or “How does the introduction of a model change the way the job
is performed?” cannot be answered with OR alone. It is in the best interest of a company
paying for an OR model solution for their PA process to understand the implications of
using the model and the resulting changes to the PA process. It is also important to fully
understand the impact the new process will have on the other functions in the organiza-
tion that both provide inputs to it and use its outputs. Addressing these issues requires
analysis that surpasses creating a spreadsheet and training someone in the company how
to use it. Human factors (HF) methodologies, coupled with the traditional OR solution
creation and validation, can be used to facilitate the successful implementation and inte-
gration of the solution into an organization. Specifically, macroergonomics (ME), which
includes function allocation, takes a broad, systems approach to work system design. It
seeks to optimize both the technical and social aspects of a process by looking at its
primary functions, variances, and social structure. Also, human–computer interaction (HCI),
which provides several techniques and methods to facilitate computer interface design, is
a necessary segment of any algorithmic or software solution.
2. LITERATURE REVIEW
x i, j ⫽ 0 or 1 for all i, j
where i is the job, j is the resource (person or machine), ci, j is the cost or benefit of
assigning job i to resource j. x i, j is equal to 1, if job i is assigned to resource j and equal
to 0 otherwise. si, j is the size or amount of resource j ’s capacity used if job i is assigned
to j. bj is the total capacity of j.
The objective function optimizes the value of the assignments, x i, j . The variable ci, j is
negative if it is a cost factor and positive if it represents a benefit. Constraint (1) ensures
that each job receives exactly one resource. Constraint (2) ensures that the total capacity
or capability of each resource is not exceeded. Different versions of this assignment model
have been used, along with various heuristics and algorithms, to model different person-
nel assignment scenarios.
Journal articles in the OR field describe the development, testing, and validation of
different optimization methods for a variety of PA cases in companies with small to large
workforces (Holness, 2003). The majority of the PA cases in the OR literature describes
two common modeling steps: (a) identifying the variables and constraints for creating
assignments, and (b) studying a process where the assignments will be utilized.
1972), while other authors simply reviewed the use of optimization models, methods,
algorithms, and heuristics to solve allocation problems (Gardenfors, 1973; Gavish & Pirkul,
1991; Ross & Zoltners, 1979;).
2.2.2. Human resources studies. There were few PA examples in the human
resources literature as it was defined above. The majority of authors focused on man-
power planning. For example, Bergmann and Martin (1987) studied the use of various
strategies and policies for recruitment, including the use of simple linear models and
sensitivity analysis. Other authors investigated the use of models to determine the num-
ber of people required to meet current or forecasted demand, or to actually forecast demands
for human resources (Bechet & Maki, 1987; Niehaus, 1988). Methods such as regression,
Markov models, goal programming, linear programming, and simulation were recom-
mended in these cases.
Singell and Lillydahl (1996) studied how faculty members allocated their time to four
tasks: teaching, research, service, and leisure. A random utility (multinomial, grouped-
data logit) model was created, using predetermined utility values to determine “how indi-
vidual and institutional differences affect time allocation decisions.” Doyle (2002)
performed a questionnaire study of how managers choose subordinates for participation
in organizational change. Some authors focused on the conceptual implications of select-
ing domestic employees for overseas assignments or creating assignments for expatriates
to ensure their quality of working life and work life balance (Bolino & Feldman, 2000;
Torbiorn, 2002). However, no mathematical or optimization models were used in these
cases.
3. RESEARCH STRATEGY
The goal of this research was to show the value of combining “soft system” methodolo-
gies from human factors studies with the “hard system” mathematical formulations and
validation from OR to increase the chances of successful implementations in organiza-
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
PERSONNEL ASSIGNMENT 289
and is particularly useful for function allocation. These classifications were used in the
current research because it was necessary to determine what steps could be performed by
a computer instead of a human operator.
For example, Roberts and Gawron (1986) used the SRK framework to distinguish
between operator and manager functions as part of the development of a cockpit auto-
mation technology program. The framework was used in a decision aid for assigning
tasks, particularly those tasks that were not easily distinguishable as either an operator or
a manager function. All functions were classified as operations (skill-based), decisions
(rule-based) or problem formulations (knowledge-based) using a predetermined set of
assumptions and decision criteria. Once categorized, it was then easier to determine which
were primarily for operators (skill) versus managers (knowledge). The decision tasks
were assigned to either operators or managers, depending on the capabilities of the par-
ticular user and the machine used for the task. In summary, using this framework allowed
the authors to assess not only human-to-human task allocations but human–machine assign-
ments as well.
Human–computer interaction logically follows function allocation. Human–computer
interaction facilitates the design, testing, and evaluation of the actual interfaces between
humans and machines or computers. There are established characteristics of an effective
interface design, which include consistency, familiarity, feedback, and error correction
(Shneiderman, 1998). These are necessary to ensure that an interface does not provide
additional complexity to the execution of the function of interest. It is also important that
the interface minimizes the stress of its human users, whether novices or experts. Human–
computer interaction methodology also guarantees that usability tests are carried out to
ensure alignment with its intended purpose, in addition to the creation of documentation
and training plans.
In terms of personnel assignment, OR solutions typically involve the creation of com-
puter software to run the optimization models and their associated heuristics. It is safe to
assume that users of these models may not be the OR practitioners who created them.
Therefore, the type of interface a shift supervisor would use to allocate jobs would be
very different from that used by an OR practitioner. This is especially true if changes to
the model by supervisors are required when she or he uses the software. In this respect, it
is important that the interface is appropriate to the class of user, which implies the need
for HCI. It is also critical to ensure that the software is used by the appropriate persons for
the function, which involves function allocation. Finally, the macroergonomics process
can help ensure approval and support of new software from all levels in an organization.
Macroergonomics also guarantees that those directly impacted by its output will be con-
sidered in the design of the solution.
The research was conducted in the pressroom department of a printing company. The
pressrooms run 5 days a week, with three 8-hour shifts per day. Running the presses on
the weekends and extending shifts to 12 hours as needed is considered overtime. There
were between 300 and 350 total hourly employees working in the pressrooms’ depart-
ment at the time of the study. It should be noted that at this company, PA was called
Manning and the primary person who worked on PA was called the Manning Clerk. How-
ever, in this article, this person will be referred to by the more generic term of PA Clerk.
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
PERSONNEL ASSIGNMENT 291
4.1. Participants
One of the manufacturing managers at this plant was the key stakeholder for the study. He
granted permission to interview himself, the pressrooms’ PA Clerk, and one pressrooms’
supervisor—all of whom were actively involved in the personnel assignment process cur-
rently used at the company. There was only one pressrooms’ PA Clerk, who handled the
assignment functions for all of the presses on all three shifts. He was a fulltime union
employee with over 20 years of service, who also worked on the presses on the weekends.
In the remainder of this article, the PA Clerk will be referred to as “he” although it is
certainly possible for this position to be filled by either a man or a woman in other
companies.
was desired to map the existing, primarily manual, PA process as a user-independent pro-
cess. This allowed the identification of possible objective functions, key model assump-
tions, variables, constraints, and criteria–goals for a model. Second, these methods provided
the data to define the kind of support structure necessary for the successful implementa-
tion and use of a software solution.
To model the PA process, flow charts and hierarchical task analysis (HTA) spread-
sheets were created using the data from the interviews and the observations made. Because
of the level of detail and influencing factors in the procedures used, a listing of process
steps in a spreadsheet proved to be a more efficient method of presenting the data than a
traditional HTA diagram. Therefore, flowcharts were created to present only the macro
view of each element of the PA process. Spreadsheets were then created for each element
of the PA process. They contain numbered lists of the actions taken in the process with the
following descriptors for each action:
• Classification: The action was classified as skill (S), rule (R) or knowledge (K).
• Day: The day of the week the action was performed
• Tool/Method: What was used to execute the action (computer, paper, conversation,
etc.)?
• Input: What information was required to do the action?
• Procedure: Actual steps taken
• Decision: What decisions, if any, had to be made to perform the action successfully?
• Variables for Rules: What variables affect the choice of rule to make the decision?
• Rules: What rules, if any, were used to make a decision?
• Output: What resulted from the execution of the action?
• Input for: What depended on the results from the current action?
• Comments: Additional information about the execution of this action
With the results from the flowcharts and the HTA spreadsheets, decision ladders were
created as a supplement to the SRK classifications. This was done to assess the algorith-
mic nature of the process steps. Decision ladders were chosen because “the decision lad-
der provides a generic template for identifying the demands associated with particular
control tasks. As a result, it allows us to identify the information requirements that must
be satisfied. Regardless of the role allocation architecture that is adopted, these require-
ments must be satisfied if task goals are to be reliably and consistently attained” (Vicente,
1999). For this reason, the decision ladders generically map information-processing activ-
ity, making the diagrams user-independent. This is particularly useful for software devel-
opment because it allows for the additional analysis of information sources and their
interaction.
Having two PA stages is unusual when compared to the OR literature. All assignment
models assume that a labor pool already exists and the solution simply matches people to
jobs. Although a set of employees does indeed exist at this company, generating a PA
labor pool separately is deemed necessary for two reasons. First, department PA strives to
merely adjust the current week’s assignments for the next week, instead of simply “wip-
ing the slate clean.” It attempts to consider each employee for each department only when
deemed necessary. Second, the size of the labor pool changes considerably from week to
week, due primarily to vacations. Both of these issues will be discussed later in more
detail.
5.1.1. Department PA decision ladders. There were two main decisions in the
department PA process. The first was made solely by the PA Clerk; it is depicted in the
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
294 HOLNESS, DRURY, AND BATTA
Figure 2 Department personnel assignment (PA) process decision ladder (rectangles ⫽ informa-
tion processing activities, circles ⫽ states of knowledge, shading ⫽ not relevant to this decision).
assignment changes or not. The decision process begins at the “observe” step, where the
PA Clerk mentally matches a person’s current assignment and work characteristics with
that of the future staffing needs of the person’s current department. The resulting state of
knowledge is an assessment of whether the person remains where they currently are or is
needed more in another department. The “define task” activity is where the PA Clerk
decides what department the person will move to, using the criteria from the observation
step. The PA Clerk uses his knowledge of the presses, the kinds of jobs that are run, and
the person’s capabilities to come up with a “best” fit or feasible solution. The resulting
state of knowledge is the department the person will work in.
When the decision is made, the person’s assignment is noted on the PA board—a phys-
ical structure mounted on a wall with plastic leaves that flip up. The board is divided into
sections primarily by each press department and by employee job classification: Jour-
neymen (JM), Assistants (ASST), and Junior Helpers (JH). Journeymen are considered
the most experienced. Each employee has a card, which can be left in its current spot or
be moved to a new location on the board. If it is moved, the change is also noted in a
computer program used for PA.
Each card is color-coded by employee type: white for JM, pink for ASST, and blue for
JH. This coding scheme helps the PA Clerk easily tell the difference between the cards
without taking time to read its contents. This becomes particularly important when the
cards are rearranged on the PA board. The color-coding of the cards, as a preattentive
feature, enables the PA Clerk to easily find the cards when searching the board and to
place cards in their correct locations. Color codes are notably useful for both searching
and identification tasks because they provide a salient feature that “catches the eye” and
allow for parallel, instead of serial, searches (Sanders & McCormick, 1993; Treisman,
1986). As a result, the time it takes to complete the task decreases.
When one person’s assignment is complete, the PA Clerk repeats the process for the
remainder of the names on the employee list, which is organized by job classification.
There are 300 to 350 employees. It is important to note that while he goes through the
names, he also compares each one with any changes previously made. This is done to
make sure he accommodates as many first-choice shift preferences as possible. For this
reason, he may revisit a previous name on the list and change his or her assignment.
A red card is placed at the end of the classification or pressroom–department block
listing on the PA board, either at an angle or straight, to signify that a press department
has all the required people or is short, respectively. Like the color-coded employee cards,
the red card provides another visual cue to aid in the searches performed on the board.
Both the color and the angle of this card provides a preattentive feature that enhances the
PA Clerk’s parallel processing, which refers to the initial collection and organization of
information that “pops out” of a display. With parallel processing, a person can then selec-
tively pay attention to certain items for additional evaluation (Wickens & Hollands, 2000).
This feature attracts the attention of the PA Clerk and enables him to quickly draw con-
clusions about the state of the system.
This decision ladder illustrates that the PA Clerk relies primarily on information assess-
ments and evaluation criteria to decide on assignment changes. No interpretation or
evaluation of each person’s current assignment in terms of the overall goal of Monday
PA (to create the PA pool) is required. Because of this, it was concluded that the process
was primarily skill- and rule-based. However, for a novice in the PA Clerk position, it is
believed that the task would be primarily rule-based with more controlled, rather than
automatic, processing. This belief is justified by the fact that a high level of familiarity
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
296 HOLNESS, DRURY, AND BATTA
with the employees and the presses contributes to the PA Clerk’s ability to quickly
evaluate each person’s status. This can be attributed to a type of recognition-primed
decision-making (RPD), whereby “the decision maker in essence “recognizes” the pat-
tern of cues in a problem as one that matches a template of typical cues in prior expe-
rience, and thereby rapidly categorizes (diagnoses) the situation, without going through
the time- and effort-consuming processes of cue weighting and integration” (Wickens
& Hollands, 2000). The PA Clerk’s expertise also allows him to mentally track the
changes he makes and compare them with those made for other employees. This ability
would certainly be difficult for a novice and would most likely require specific decision
rules and training.
The second decision involves resolving deficiencies. Once the PA Clerk generates a PA
pool for each department that satisfies as many production requirements as possible, he
then determines whether the department is short (called deficiencies) or if there are extra
people. If there are extra people, voluntary layoffs and bump down requests are granted.
Bump downs allow a person to work in a position of a lower classification level. If there
are deficiencies, however, a rule-based decision logic is executed, mainly by members of
the scheduling department and a pressroom department superintendent, who is also the
first shift supervisor for that department. The PA Clerk does not directly handle decisions
regarding deficiencies, although he may contribute to the discussion. The options for han-
dling deficiencies involve posting 12-hour shifts and requesting upgrades. Upgrades are
the opposite of bump downs and allow an employee to work in a higher classification job.
The process of handling deficiencies was primarily a rule- and knowledge-based process
because of the level of evaluation and consideration given to the consequences of the
shortage options. It is important to note that although the task is knowledge-based due to
the level of problem solving, it involves common problems with a predetermined set of
solutions. It can be said, then, that the level of uncertainty is small, but large enough to
require discussion and consensus. The knowledge-based facet to this task presents a chal-
lenge to automating it. Because of the level of communication and personal interaction
involved, simply rerunning an OR model with different constraint values may not be
sufficient. Further discussion may be required.
In summary, the majority of the steps taken in the department PA process involve infor-
mation gathering and assessment, which are primarily skill- and rule-based. Those that
are rule-based utilize simple decision rules. The PA Clerk uses very little problem solving
when rearranging the cards on the PA board and formulating the final PA pool. Resolving
deficiencies is primarily a rule-based process, again with simple, yet specific decision
rules. Here, the use of the decision ladder and the SRK framework highlighted the level
of complexity used in this process and the possibility for automation. Because the steps in
this process, in the novice or expert case, involve cue processing and use of simple deci-
sion rules, they are prime candidates for algorithm development.
5.1.2. Press PA decision ladders. The main decision made by the supervisors in
the Monday press PA process is to assign employees of different classification levels to
each press by considering the type of job being run on the press. Like the PA Clerk’s
decision process for determining assignment changes, these decisions do not require any
knowledge-based behavior. The comparisons made between each employee for assign-
ment on a particular press are primarily skill- and rule-based comparisons. No decision
ladders were made for making last minute changes because all of them were determined
to be rule-based.
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
PERSONNEL ASSIGNMENT 297
6. MACROERGONOMICS ANALYSIS
for either Monday to Friday or Overtime PA. Figure 3 shows a unit operation diagram for
this process.
As shown in the diagram, all of the functions are carried out in both the department and
press PA processes. The key personnel are the PA Clerk and the shift supervisors, respec-
tively. The first two functions are the information gathering and assessment steps, which
have been determined to be skill- and rule-based steps. The last two functions, creating
and modifying the base schedule, involve skill-, rule-, and knowledge-based behavior.
These functions currently utilize the most person-to-person interactions to complete. The
result of this process is a dynamic schedule, which is subject to change throughout its use.
Therefore, it can be concluded that the first two functions could be automated and the
output used by a human operator for the last two functions. Also, because the last two
involve problem solving, they could be performed mainly by humans, with some low
levels of automation for data presentation and organization. But adding automation does
not simply mean replacing the PA Clerk with a computer or letting a software program
create the PA pool and crew assignments. Considerations must be made of the potential
effects of eliminating the PA Clerk position and/or replacing the supervisors. Decisions
must also be made regarding the primary user of the software output to create and modify
the base schedule. In fact, a variety of organizational design factors comes into play with
the redesign of this system. It is worth examining each of these factors in detail and their
relationship to the throughput of the system.
6.1.1. Human–human interactions. Taylor and Felten (1993) state that “the social
system can be seen as not merely work or friendship behavior, but rather the source of
adaptability and flexibility in coping with variances in the product and with changes in
the system’s complex environment” (p. 115). Following this reasoning, it is important to
understand the social system used in a process to capture the impact of automation on
how the social system functions. This understanding will also facilitate job restructuring,
in terms of formalization, centralization, etc. Thus, it is worth taking the time to consider
these interactions when planning for the PA process.
6.1.2. Levels of abstraction. Apart from the functional interactions in the PA pro-
cess, there are system-wide interactions that impact how it operates. To put into perspec-
tive the organizational variables and constraints that impact the overall PA process, it is
helpful to look at them from different levels of abstraction. The abstraction–decomposition
space, as described by Vicente (1999) and originally developed by Rasmussen (1986), is
ideal for analyzing work domains in sociotechnical systems. Figure 4 displays the overall
PA process (Monday and Overtime) from different levels of abstraction (rows), from the
perspective of the different people involved in the process (columns). These levels illus-
trate the means–ends and part–whole relationships between the different elements.
The levels of abstraction illustrate that the successful completion of the entire process
depends on two key factors—scheduling information and press employee availability. It
was discovered during the HTA creation that high variability exists in the scheduling
process. This variability could cause daily changes to the press assignments, which, in
turn, resulted in the need for weekend production. The need for overtime was also com-
pounded by the fact that mandatory work was not enforced during the week. It was enforced
on the weekends, however, after volunteers were requested. Employee availability was
greatly affected by the predominately senior workforce (by years of service) and the union
policy on vacations by seniority. Other factors such as high attrition among the lower
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
PERSONNEL ASSIGNMENT 299
classification employees, informal skill tracking, and career progression between classi-
fication levels were also contributing factors to the formation of the PA pools.
Any changes to the level of automation in this PA process must address the two factors
of scheduling information and press employee availability. Such considerations would be
virtually impossible to model as constraints in an OR model. Therefore, an assessment
must be performed on the information technology and social infrastructure to resolve
these factors, which are necessary to perform the PA tasks with automation.
7. RECOMMENDATIONS
7.1.1. Social system questions. There are two classes of questions that must be
addressed when considering the addition of an OR model to this process: the PA Clerk
position and human–human function allocation. The following is only a sample of the
questions that may arise:
7.1.2. Technical system questions. The first and most important technical ques-
tion to ask is, “What should be automated?” The design of an OR model or other software
aid will depend on which functions are chosen for automation. Also, it was previously
mentioned that separating a PA process into parts is not done in the literature because the
OR models assume that the labor pool has already been created or will be created by the
model. How the OR model is developed will also depend on any other changes made for
the PA process, like changes in employee availability during the week through mandatory
assignment. Of course, such changes may not be feasible or cost effective but worth con-
sidering nonetheless.
If it is decided to create an OR model to either generate the PA pool or create press
assignments, the model inputs must be decided upon. A basic assignment model formu-
lation assumes that there is a one-to-one matching between person and job and ignores
the fact that each employee can have multiple skills. This is also true of the current depart-
ment PA process, which does not consider upgrades or bump downs when the PA pool is
first generated. They are considered only if there or shortages or extra people. However,
considering different equipment qualifications without assigning employees multiple jobs
complicates the model considerably. Actually, it may be necessary to quantify the fre-
quency of upgrades and bump downs granted each week in order assess the feasibility of
including this factor in the model. The model is not impossible, just more complicated.
Applying the correct heuristic or algorithm would be the task of an OR practitioner. How-
ever, the basic function of the model must be decided upon by the organization. This must
be done with the understanding that, depending on how the model is created, its use may
require different skills from those currently used in the process.
However, regardless of how complicated the model turns out to be, the basic model
does require some key pieces of information: an objective function or functions, a value
or cost variable to evaluate each assignment, and an assignment variable. The objective
function specifies what needs to be optimized. It is possible to have multiple objective
functions in the same model if multiple variables are required to solve the assignment
problem. The value or cost variable reflects the desired association for each assignment.
For example, a value or cost could reflect crew potential, with optimal values compared
to a crew rating score. In another example, it may reflect the cost or value to the depart-
ments of assigning a person there. However, many others are possible. The type of value
or cost will determine whether the model solves a minimization or maximization prob-
lem. Finally, the assignment variable is usually associated with each job-to-employee
pairing. Because there are multiple characteristics for each employee and multiple jobs
for each department and press, the indices associated with the assignment variable can be
numerous. An alternative to multiple indices is multiple variables and multiple objective
functions.
Nesting classification and seniority under employee is one way to capture multiple
characteristics. Franz, Baker, Leong, and Rakes (1989) used nesting in their model that
assigned medical staff to different clinics within different regions in a geographic area.
This method allowed them to index the total number of employees of different job
classifications per clinic district. Figure 5 shows an example for the printing company’s
PA process. All employees would be categorized by their primary classification, which
is their highest classification to date. Then, under each classification, the employees are
listed by seniority. This is, in fact, how the employee lists used by the PA Clerk are
currently structured. However, they also list which equipment each person is qualified
to work on. This additional qualification could be characterized under a different index.
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
PERSONNEL ASSIGNMENT 303
Of course, the ultimate decision about these indices would have to be made by the
organization.
Table 1 summarizes some possible criteria for an optimization model of this PA process
as it currently functions. The most important listing in this table, however, is the practice
of starting the PA process using the previous week’s PA pool and simply making adjust-
ments to it. Most of the OR models in the literature do not follow this assumption; they
essentially start with a clean slate. It may not be feasible for this company to start at this
point. But, if a clean slate is desirable, this process change must be properly facilitated.
For example, one way to ensure minimal changes to the previous schedules, if desirable
to the employees is to include a cost term for each change from the previous week.
Regarding the input of employee availability, it is important to note that it is possible
to run a model using a pool for employees known to be available. However, this would
require an initial screening of employees, similar to what the PA Clerk currently does. It
is also possible to equate either the cost variable or the assignment variable to zero to
reflect the unavailability of certain employees. Another alternative is to create dummy
press jobs in the model to assign unavailable employees. This would allow the model to
assign each person somewhere. However, either option would require some change of the
model structure and code to reflect unavailability. The capability of making such a change
would have to be decided upon while defining the requirements of the new work design.
The constraints, like the inputs, must be also be decided upon. There are hard con-
straints, which cannot be violated and soft constraints, which do not have grave conse-
quences if violated. For example, suppose the model was designed to form crews and
includes generating the PA pool. Then, a hard constraint would allow for the evaluation of
the “best possible” pairing of each employee to a task, using inputs that most likely assign
values to the pairings of certain employees. Such pairings would be based on each person’s
classification and qualifications, as well as the supervisor’s assessments of optimal pair-
ing. For this reason, it would be critical to track the supervisors’ ratings of individual
performance. However, this constraint would most likely be soft because of the addi-
tional policies on 12-hour shifts, upgrades, and bump downs. Realistically, it would be
difficult to satisfy this constraint without accounting for the possibility that a person could
refuse a “best possible” assignment because these options are considered voluntary under
these policies.
To summarize, careful consideration must be made when deciding what will be auto-
mated, what the variables, constraints, etc. will be and what the model output will look
like. This again ties back to how the model will be used to create and modify the base
schedule for a redesigned PA process. Consideration must also be given to the way the job
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
TABLE 1. Possible Operations Research (OR) Model Criteria for the Personnel Assignment (PA) Processes
304
Monday • Maximize coverage cijkm • Employee classification • Work staffing levels • Start with the current week’s schedule,
PA by classification, x ijkm (JM, ASST, JH) required per union contract PA pool and crew assignments (as is) or
seniority and Pijkm • Employee seniority for each classification start from scratch.
qualification Others (years in position, • Total # JM assigned to dept • Labor pool exists
• Maximize or years at company) or press ⫽ Total # JM • Press production schedule and labor
minimize cost or • Employee qualification needed for the dept or press. demand exists
value of the (1–3 levels) Same for ASST and JH • Consider shift preferences
assignment • Employees shift • Use as much of total labor • Assign all available slots to all available
• Maximize shift preferences supply as possible employees
preferences • Employee availability • Each worker is assigned • Assignments done by classification,
• Others, as deemed • Employee request for somewhere (Use a dummy with seniority and qualification within
HOLNESS, DRURY, AND BATTA
DOI: 10.1002/hfm
PA above • Employee number of • Employees considered by total of works
works and refusals and refusals.
year-to-date • First week of the year all works and
• Employees taking the refusals are zero. Employee lists are
14th day off ordered by seniority.
Note. cijkm ⫽ cost or value of assigning the ith employee (classification and seniority nested under employee) to press j during shift k. x ijkm ⫽ 1 if employee i is assigned to press
j during shift k, 0 otherwise. Pijkm ⫽ the preference of employee i of being assigned to press j during shift k (accommodates shift preferences). i ⫽ employees by primary
classification. j ⫽ seniority—years of service in plant or in current primary classification. k ⫽ qualification (maybe qualification for certain department or press—some indexed
code for each department and each press type and job type), e.g., 1 ⫽ 132 JM, 2 ⫽ 132 ASST1, etc. s ⫽ shift ⫽ $1,2,3%. m ⫽ press, which can be indexed like the qualifications.
PERSONNEL ASSIGNMENT 305
will be carried out by whoever generates the output. The benefits of this level of careful
planning and analysis are numerous because ultimately, it will help to guarantee the suc-
cess of the new system.
At this point, it is worth noting some of the human–computer interaction (HCI) impli-
cations of using an OR model. It is assumed that the person(s) using the model would not
be familiar with OR. Therefore, some kind of user interface and software would be required
to enable whoever does the PA to make necessary changes to the model (e.g., input or
constraint values, etc.). At the base level of functionality, the model interface should have
a familiar format (e.g., Windows environment) and allow for data input, access, and inte-
gration from a variety of sources. Finally, it must accommodate both novice and expert
users. Beyond these basics, any further recommendations would depend on the details of
what the model/software will be expected to do.
8. CONCLUSIONS
This research demonstrated the value of including a systems view when analyzing a tra-
ditional operations research problem. This view provides additional guidelines for a suc-
cessful implementation of an OR model. The knowledge elicitation questions asked during
data collection and the consequent analysis of the answers using HTA resulted in the
mapping of both the process steps and the relationships between these steps. More impor-
tantly, the use of these methods made three different analyses possible. First, the HTA
captured the interactions among personnel and the types of communication methods they
employed. Second, the HTA facilitated the classification of the steps in terms of skill-,
rule-, or knowledge- (SRK) based behavior, according to the Rasmussen (1986) criteria.
Third, it highlighted the decisions made in the process, which were further analyzed using
decisions ladders. The ladders allowed for an extended SRK analysis of the decisions
themselves. Such information, at this level of detail, would not have been captured in a
simple flowchart of the process. The preliminary ME study showed the areas in the PA
process that required additional consideration and analysis because of the potential impact
that an OR solution would have on its performance. It also allowed for an extraction of
possible objective functions, variables, model inputs, constraints, and assumptions in terms
of organizational goals, social factors and communication systems.
Future research on the use of human factors research methods as well as systems or
organizational perspective for personnel assignment problems is wide and far-reaching.
For example, it may be worth doing a strategy study like those conducted in the organi-
zational behavior literature, to test the assumption that an OR model is better equipped to
Human Factors and Ergonomics in Manufacturing DOI: 10.1002/hfm
306 HOLNESS, DRURY, AND BATTA
evaluate multiple variables, constraints, and objectives than does the human operator.
Additionally, if an OR model is developed for this company, it would certainly be worth
studying how the OR practitioner goes about developing the model, in terms of integrat-
ing social as well as technical factors. Finally, rigorous tests on the value added by a
human factors approach can be assessed. This can be done by comparing specific perfor-
mance metrics of completed test cases with and without the use of human factors, such as
employee involvement, implementation of the OR solution, or sustained use after imple-
mentation. Such research could lead to the formulation of methodologies to enhance the
usability of OR solutions for personnel assignment problems.
REFERENCES
Bechet, T.P., & Maki, W.R. (1987). Modeling and forecasting focusing on people as a strategic
resource. Human Resource Planning, 10(4), 209–217.
Bergmann, T.J., & Martin, G.E. (1987). Optimal large-scale manpower recruitment policies. Human
Resource Planning, 10(2), 93–101.
Bolino, M.C., & Feldman, D.C. (2000). Increasing skill utilization of expatriates. Human Resource
Management, 39(4), 367–379.
Childs, M., & Wolfe, H. (1972). A decision and value approach to research personnel allocation.
Management Science, 18(6), B269–B278.
Dean, B.V., Denzler, D.R., & Watkins, J.J. (1992). Multi-project staff scheduling with variable resource
constraints. IEEE Transactions on Engineering Management, 39(1), 59–71.
Doyle, M. (2002). Selecting managers for transformational change. Human Resource Management
Journal, 12(1), 3–15.
Eason, K. (1988). Information technology and organizational change. London: Taylor and Francis.
Franz, L.S., Baker, H.M., Leong, G.K., & Rakes, T.R. (1989). A mathematical model for sched-
uling and staffing multiclinic health regions. European Journal of Operational Research, 41,
277–289.
Gardenfors, P. (1973). Assignment problems based on ordinal preferences. Management Science,
20(3), 331–340.
Gavish, B., & Pirkul, H. (1991). Algorithms for the multi-resource generalized assignment prob-
lem. Management Science, 37(6), 695–713.
Gomar, J.E., Haas, C.T., & Morton, D.P. (2002). Assignment and allocation optimization of par-
tially multiskilled workforce. Journal of Construction Engineering and Management, 128(2),
103–109.
Hendrick, H.W., & Kleiner, B.M. (2002). Macroergonomics: Theory, methods and applications.
Mahwah, NJ: Erlbaum.
Holness, K.S. (2003). Personnel assignment from a human factors perspective. In Proceedings
of the Human Factors and Ergonomics Society 47th Annual Meeting (pp. 1394–1398). Santa
Monica, CA: Human Factors and Ergonomics Society.
Langholtz, H., Gettys, C., & Foote, B. (1993). Resource allocation behavior under certainty, risk
and uncertainty. Organizational Behavior and Human Decision Processes, 54, 203–224.
Langholtz, H., Gettys, C., & Foote, B. (1994). Allocating resources over time in benign and harsh
environments. Organizational Behavior and Human Decision Processes, 58, 28–50.
Langholtz, H., Gettys, C., & Foote, B. (1995). Are resource allocations anticipated in resource
allocation tasks? Organizational Behavior and Human Decision Processes, 64(3), 274–282.
Langholtz, H.J., Ball, C., Sopchak, B., & Auble, J. (1997). Resource allocation behavior in com-
plex but commonplace tasks. Organizational Behavior and Human Decision Processes, 70(3),
249–266.
Langholtz, H.J., Ball, C., Sopchak, B., & Auble, J. (1998). Resource allocation strategies: A verbal
protocol analysis. Organizational Behavior and Human Decision Processes, 76(1), 70–88.
Niehaus, R.J. (1988). Models for human resource decisions. Human Resource Planning, 11(2),
95–107.
Parasuraman, R., Sheridan, T.B., & Wickens, C.D. (2000). A model for types and levels of human
interaction with automation. IEEE Transactions on Systems, Man and Cybernetics, 30(3),
286–297.
Rardin, R.L. (1998). Optimization in Operations Research. Upper Saddle River, NJ: Prentice Hall.
Rasmussen, J. (1986). Information processing and human machine interaction: An approach to cog-
nitive engineering. New York: North Holland.
Roberts, D.P., & Gawron, V.J. (1986). Operator/Manager classification as a precursor to function
allocation. In Proceedings of the Human Factors and Ergonomics Society 30th Annual Meeting
(pp. 857–860). Santa Monica, CA: Human Factors and Ergonomics Society.
Ross, G.T., & Zoltners, A.A. (1979). Weighted assignment models and their application. Manage-
ment Science, 25(7), 683– 696.
Sanders, M.S., & McCormick, E.J. (1993). Human factors in engineering and design. New York:
McGraw Hill.
Shadbolt, N., & Burton, M. (1995). Knowledge elicitation: A systematic approach. In J.R. Wilson
& E.N. Corlett (Eds.), Evaluation of human work (pp. 406– 440). London: Taylor and Francis.
Shneiderman, B. (1998). Designing the user interface: Strategies for effective human computer
interaction. Reading, MA: Addison Wesley.
Singell, L.D., & Lillydahl, J.H. (1996). Will changing times change the allocation of faculty time?
Journal of Human Resources, 31(2), 429– 449.
Taylor, J.C., & Felten, D.F. (1993). Performance by design: Sociotechnical systems in North Amer-
ica. Englewood Cliffs, NJ: Prentice Hall.
Torbiorn, I. (2002). Staffing for international operations. Human Resource Management Journal,
7(3), 42–52.
Treisman, A.M. (1986). Properties, parts, and objects. In K.R. Boff, L. Kaufman, & J.P. Thomas
(Eds.), Handbook of perception and human performance. New York: Wiley.
Vicente, K. (1999). Cognitive work analysis. Mahwah, NJ: Erlbaum.
Villeda, R., & Dean, B.V. (1990). On the optimal safe allocation and scheduling of a workforce in
a toxic substance environment. IEEE Transactions on Engineering Management, 37(2), 95–101.
Warner, D.M., & Prawda, J. (1972). A mathematical programming model for scheduling nursing
personnel in a hospital. Management Science, 19(4), 411– 422.
Waterson, P.E., Older Gray, M.T., & Clegg, C.W. (2002). A sociotechnical method for designing
work systems. Human Factors, 44(3), 376–391.
Wickens, C.D., & Hollands, J.G. (2000). Engineering psychology and human performance (3rd
ed.). Upper Saddle River, NJ: Prentice Hall.