MANAGEMENT SOENCE

Vol. 34, No. I. Januaiy 1988

PruaeilmU.SA

EVALUATING PROJECT SCHEDULING AND DUE

DATE ASSIGNMENT PROCEDURES: AN
EXPERIMENTAL ANALYSIS*
JOHN DUMOND AND VINCENT A. MABERT
Air Force Institute of Technology, Wright-Patterson Air Force Base, Dayton, Ohio 45402
School of Business, Indiana University, Bloomington, Indiana 47405
Managers of construction projects, maintenance activities, auditing contracts, software
shops, etc. are fiequentty faced with the task of establishing a new project's due date, which must
Mmpete with other projects already in pro^«ss or expected (forecasted) to start in the future.
The study reported here addresses the problem of establishing due dates for projects which
r«]uire limited r«ources, in an environment where new projects arrive continuously and
randomly over time. A set of procedures is developed which set each project's due date when it
arrives using infonnation about the new project, current projects, and available resources. The
due date setting procedures are tested via simulation with four activity scheduling heuristics that
control the as^nment of resources to specific activities of available projects. A second test
demonstrates the performance of the due date procedures, where a portion of arriving projects
havetoeirdue dates established by external forces beyond management's control. Performance
measures of project mean completion time, project mean lateness, project standard deviation of
lateness, and total tardines (sum of all projects' tardy time) were collected for evaluation.
This study presents a number of important results for managers interested in scheduling
projects and setting due dates. First, using more information concerning the current work in
prepress, available resources, and activity precedent relationships provides a better due date
estimate for a new projert. Second, afinitescheduling procedure (called SFT) consistently gives
better due date estimates than simpler aggr^ate procedures. Third, when some project due
dates are set externally, due date performance deteriorates. However, when SFT is combined
with a due date oriented activity scheduling rule, due date performance deterioration is less.
Fourth, the effort, measured by CPU time, for SFT to estimate a good due date depends upon
the ratio of activity resources required to total resources available, rather than the number of
activiti« across all projects. Andfifth,similarities and differences between the results observed
in this study and past due date job shop scheduling research are reviewed.
(PROJECT MANAGEMENT—RESOURCE CONSTRAINTS; PRODUCTION SCHEDUL-
ING; SIMULATION—APPLICATIONS)

1. Introduction
Since the building of the pyramids, manners have continually tackled the job of
managing large projects when limited resources have been present. Managing these
projecte requires a number of complex planning and difficult control decisions. Plan-
ning covers decisions such as establishing activity/project completion due dates, deter-
mining available resources (timit^ and quantity), ^timating activity durations, defin-
ing precedent relationships, etc. Control decisions focus on determining which activi-
ties have hi^est priority, assigning available resources to specific activities, comparing
sch^ukd finish times to actual times, etc.
Ehiring the last 20 yraus, a number of researchers have focused their attention on
various facets of the static resource constrained project mans^ement scheduling prob-
lem, where » i ^ e and/or multiple projects exist. In the single project setting, only
%^vitks within this project are competing for available resources, and r^earchers have
used heuristics (Cooper 1976, Davis and Patterson 1975, Wiest 1967) and optimization

* Acc^ted by David G. I^mneabring, former Departmental Editor, i«ceived June 20,1985. This paper has
been with the »ithors 6 months for 3 revisioiis.
101
0025-1909/88/3401/0101$01.25
Capyi%M e 1988. The Institiite of MaaKcmeat SE^eoces
 102 JOHN DUMOND AND VINCENT A. MABERT

techniques (Davis and Hddom 1971, Patterson 1984, Patterson and Roth 1976, Talbot
and Patterson 1978, Thesen 1976) to develop the best schedule. The presence of multi-
ple projects causes competition to exist between activities on different projects and a
number of studies (Fendley 1968, Kurtulus and Davis 1982, Patterson 1973, 1984,
Pritsker et al. 1969, Wiest 1967) have also been conducted.
The above-mentioned research a^umes that all projects are known at once, yielding
a static situation where complete knowled^ exists of the demand for the resources over
time. However, if new projects come into existence over time then a dynamic condition
persists and uncertainty is present as to the demand for the resources in the future.
Our literature review indicates only one published work dealt with the dynamic
multiple project problem. Thomas and Coveleski (1973) describe the use of a network
based computer program known as the Long R a i ^ Planning System (LRPS) used at
Babcock & Wilcox (B&W) Corporation. The prc^ram is used to schedule the manu-
focturing and installation of nuclear generating equipment. The system uses proprietary
heuristics tofinitelysch»lule thousands of activities from different projects through up
to 150 constrained resource.
The multi project-dynamic atuation repr^ents a common busings situation for
both manufacturing and service firms. For example, software development houses and
data processing departments routinely need to schedule the designing, programming,
testing and implementing of major hardware/software projects. Accounting firms fre-
quently schedule teams of staff members to conduct audits for various clients at differ-
ent times. Other examplesfiromthe con^ruction and publishing industry could also be
cited to indicate thefi:«quencythis project mana^ment structure occurs in practice.
It is obvious that many oiiganizations are faced with managing multiple projects
requiring multiple resources. One key planning decision they face involves the estab-
lishment of a completion (promise) date that is attainable, recognizing new projects will
be added to the existing set and compete for limited resources. With the project's due
date estaUished, activity control decisions (scheduling) need to be made on the assign-
ment of resources to minimize deviations from the established due date. This study
describes the first systematic evaluation of the project activity scheduling-due date
assignment decisionreportedin the literature.
Hiis paper pr«ents an evaluation and analysis of a set of activity scheduling heuris-
tics and project due date assignment procedures usii^ computer emulation. The test
scenario is broken into two phas^ where the first environment has management in
comt^ete control of estaUishing all project due dat». Under this controlled setting, five
heuristic Kheduling and four due date asdgnment imxiedures are t^ted.
The second analyas relaxes the complete control assumption and h ^ man^ement
partially controllii^ due dates. A set of teste are conducted whrae 10% to 50% of project
due dates are ^jedfied by external source. This evaluation i»x}vi(fes a test ofconditions
commonly encountered by j»acticing managers.
In the next section the project scheduling heuristics are present^ follow^ by a
(tecription of the due date asdgnment imxedur^. §4 describ^ tiK e^qperimental ctesign
employed and assumptions estabfi^H^ for this study. §S pre^nts an an^ysis of
the comi^ete and partial <x>ntr(rf expoimen^ with Ihe final section summ^irizing the
results of tiiis study and the potentitd impact of dififerent diw date setting procedure.

2. Schednli^ H(»rarties
T t e fdlowing iBcgect sdwduling heuristics aUocate tiie constraiiKd a^^il^^ re-
sousces b^ed on a {mcHitiiKd li^ (rf'the competii^ activiti^ from one or mcne {Hojects.
A rcvkw (rf* ti» Uterature iiuiki^es Hiat some iHti^ct K t o l i ^ i ^ teuristks p^(Hin
hetiet than oti^s in minimising prated comi^etion times in a sta&: environment. If
 EVALUATING PROJECT M : H E D U L I N G 103

one considers the dynamic Kiheduling problem as a series of static, multi-project sched-
uling problems, updating the system each time a new project arrives, then those heuris-
tics can be used in a dynamic environment. Five scheduling heuristics are presented in
this experiment, two that do not consider the due date and three that are sensitive to the
due date. The two non-due date oriented heuristics evaluated were first in system-first
served (FIFS) and shortest activity from shortest project. Because this study concerns
the dynamic environment where meeting due dates is an important criterion, three due
date oriented heuristics were also evaluated: minimum slack based on due date
(MINSLKIDD]), and minimum late finish time based on due date (MINLFT[DD]),
and shortest activity from shortest project based on due date (SASP[DD]). The five
scheduling heuristics are described in greater detail below:'
2.1. First in System, First Served (FIFS)
This procedure is representative of scheduling heuristics found in many dynamic
scheduling environments. The project activity which has been in the system longest
receives the available resources. The index used to determine an activity's priority is
based on the arrival time of the project, with ties broken randomly. The activity's
priority index is calculated whenever a new schedule is developed.
/(nFS)y = Min {tad where (1)

tat = time of arrival of project /,

j = set of competing activities,
I{SH)ij = index value for activity; on project / using scheduling heuristic SH.
2.2. Shortest Activity from Shortest Project (SASP)
Since a number of scheduling studies (job shop and project mans^ement) reported in
the literature indicated good performance with a shortest process time type rule, SASP
is also tested in this study. The basic SASP rule was found effective in the static,
multi-project problem studied by Kurtulus and Davis (1982). In the unmodified ver-
sion the index used to determine an activity's priority is based on the critical path time
plus the activity's duration. The index is given by the following equation.
/(SASP)o = Min {d,j + CPTIME,) where (2)

CPTIME, = critical path time remaining for project i,

dij = duration of activity j for project i.
2.3. Minimum Slack—Based on the Due Date (MINSLKIDD])
This procedure is an extension of the baac minimum slack heuristic designed for the
static environment which determines the project's completion date as the last activity's
earlyfinishtime (Davis and Patterson 1975). This modified heuristic uses the project's
initial due date set at time of receipt to establi^ activity late start times and computes
the ^as^ accordingly. As a project apprcraches its due date the slack in activities
approaidh zero and as the project becomes late the slack for aQ activities become
n^ative, pving hi^ier priority to n^ative activities. The activity in<tex is computed by

( y , l l y ) y ) wtere (3)
' H m stndy oify leporKfive(MINSLKPX)}, MINLFItlX)], SASPtDD}, HFS, sad SASP) heuristics. n»
unmodified v t m a u of MINSLK and MINLFT teuriitics were tested in (DmmHid 198S) aadfimndto be
infiam pH^MiMrs to tte ino^tod sA in meetii% es^aMidMd due dates. Ther^ne, ^ mtthors decided not to
nidniie tinw lesui^ suKe they add fi^ to tte dtscHS&XB.
 104 JOHN DUMOND AND VINCENT A. MABERT

LST[DDJy = LST of activity j based on project I's established due date,

LST,; = late start time of activity^ of project / based upon project j's best completion
time,
EST,7 = early start time of activity 7 of projart /.

When a new project arrives to the system it is given a due date (using one of the due
date rules described below). This due date become the latefinishtime (LFT) of the last
activity and is noted as LFT[DD]. Based on this date a LST[DD],7 is calculated
(LFr[DD]y - dij) for each activity of the project. Additionally, the LSTy is determined
using standard critical path methodolc^. If the project is ahead of schedule then
LST[DD],; will be greater th?.n LSTy and the slack (LST,; - ESTy) will be equal to or
greater than zero. If the project is late then LST[DD],y will be less than LSTy and the
slack (LST[DD]y - EST[DD]y) will be negative.
2.4. Minimum Late Finish Time—Based on the Due Date (MINLFT[DD])
This procedure is an extension of the minimum late fiinish time heuristic. This
modified version uses the project's established due date or the currently computed late
finish time of the project's last activity as the index. The activity with the earliest
adjusted LFT is given the priority. Because of this, it resembles the "earliest due date"
scheduling rule found in job shop scheduling.
= Min (Min (LFTy, LFr[DD]y) where (4)

Tij = late finish time of activity j of project /,

LFT[DD]y = LFT of activity 7 based on project i's due date.
2.5. Shortest Activity from Shortest Project—Based on the Due Date (SASP[DD])
SASP[DD] is a compound heuristic, combining MINSLK[DD] and SASP, and com-
parable to a truncated SPT rule. In the modified version the index is based on the SASP
index or on the MINSLKJDD] index if the activity's slack is n^adve. (A negative slack
indicate the activity/project is late.)
f/(MSLKD)y if /(MSLKD),, < 0,
/(SASPD)y = Min I ' ' (5)
'^ [(dij + CFTIMEi) otherwise, where
CPTIME, = critical path time remaining for project /,
dij = duration of activity 7 of project /,
/(MSLKD)y = MINSLK(DD] index given by equation (2).
Equation (5) givesfirstpriority to all activities that are late (n^ative slack). Once all
late activities have b ^ n allocated resources, then the remaining available resources are
allocated by the SASP index.

3. Doe Date AssapnofHut Rnk»

No puMished projo:t management research has addresed the setting of due dates,
but w ^ has been ojnducted in the related area of job shop sd^uling (Eilon and
Oiowdhury 1976, Eilon and Hodgson 1967, Ragatz and Mabert 1984, Weeks 1979).
Job shop research suiqx>rts the contention that when more infonnation is used in
estaldishing due dates, the ratimates are more attainaide (Ragatz and Mabert 1984).
This con<^)t could be ^ ^ e d in a laoject environment in a similar fa^on. Sin(% this
rKearch is exfdoratory in ti^ fieM of project scheduling, job shc^ sdieduling tiraearch
im>vide5 a convenient Parting pc»nt in 4e£nii% a number of iq[^n»diK. Four due date
rules are ^veloped bdow, wiia« latter rules increase in ^
 EVALUATING MIOJECT SCHEDUUNG 105

3.1. Mean Flow Due Date Rule (FLOW)

This simple rule, which is easy to implement and frequently found in practice, does
not directly consider any of the characteristics of the project or resource requirements,
but represents jiist a fixed estimate of expected completion time. It is given by the
current date plus the historical mean (average) completion time of a project.
DD, = TNOW + F where (6)
DD, = due date of project /,
TNOW = present date,
F = historical mean completion time per project.
3.2. Number ofActivities Due Date Rule (NUMACT)
This rule uses information on one of the project's characteristics in setting the due
date, similar to the job shop's "number of operations" rule (Eilon and Chowdhury
1976). The due date is established as the present date plus a constant, "ki", times the
number of activities in the new project. Now, ki can be viewed as the average time
required to complete each activity, and is estimated by using historical data. NUMACT
is given by the following:
DD, = TNOW + ki* NUMACT, where (7)
NUMACT, = number of activities in project /,
ki = a parameter representing the expected duration per activity.
3.3. Critical Path Time Due Date Rule (CPTIME)
The critical path estimates the time to complete the project ifresourceswere uncon-
strained. Sin(% it considers the technolc^cal constraints associated with the project and
the durations of the critical activities, it is comparable to the "total processing time"
assignment rule in the job shop literature. Since resoutx^s are constrained, an adjust-
ment is made for the CPTIME rule by multiplying the critical path time by a delay
factor. This adjusted value becomes an estimate of the project's flow time.
DDi = TNOW+fc2•CPTIME, where (8)
CPTIME, = oitical path time of project /,
^2 = a parameter representing the expected delay.
3.4. Scheduled Finish Time Due Date Rule (SFT)
This rule finitely schedules, using one of the schedtding heuristics, the new project
into the system along with all other active projects before setting the due date. The
schedule which is developed at this point establishes start and stop times for each
activity of each project. The scheduled finish time of the last activity of the arriving
project is an estimate of the completion date and could serve as an excellent due date if
no new projects arrive. Since new projects continue to arrive the originally scheduled
finishtimeis often not met, thus requiring the addition of a delay factor to the flow time
estimate.
The SFT procedure USM three steps to set the due date for a new project. They sue:
1. Temporarily set the due date of the new incoming project as the present date plus
the critical imth time of tiie new project, ignorii^ resource needs.
2. Finitely sdiedule the remainii^ activitii^ of all older projects plus all the activiti^
of the arriving project using the ai^rcqmate scheduling heuristic (FIFS, SASP[DD],
SASP, MIN1JT[DD3 or MINSLiqDD]) currently imptemented.
3. Set tt» permanent due date oftiie newly arrived imjgect as the {»esent date i^us ^3
time ^ e^mated comfri^kin time for ths new im^^ct, which is (SFT(£)j - TNOW).
 106 JOHN DUMOND AND VINCENT A. MABERT

DD, = TNOW + k3(SFT(E)i - TNOW) where (9)

^3 = a parameter representing the expected delay factor,
SFT(E)i = the estimated Scheduled Finish Time for project / after loading.
The SFT due date setting rule finitely schedules, determining feasible early activity
start times based upon resource constraints, the new project before setting the due date
and therefore considers the current conditions of the system. Conrequently, the same
project arriving at different times but using the same heuristic would be given two
different estimates of completion times. That is, if the system is relatively empty a
shorter due date is established than if the system is relatively full.

4. Experimental Design
4.1. Research Steps
To explore the performance of the four due date procedures and five scheduling
heuristics, the study was divided into two phases: complete control and partial control
of due date setting. The complete control phase provides an opportunity to determine
the relative performance of the four due date setting rules and the five scheduling
heuristics under one set of environmental conditions. It is assumed that management
fully controls the setting of all due dates, that the parameters used to set the due dates
are satisfactorily determined, and that the projects vary considerably in their character-
istics. A two-factor factorial design is used to analyze the combined effects of due date
assignment rules and scheduling heuristics.
The partial control test, described in greater detail later, provides a better under-
standing of due date rule performance when less than 100 percent control is present. It
explores the sensitivity of the results to externally set due dates over a wide range of
conditions.
The complete control experiment is preceded by a pre-test which establishes the
parameters (fc's) for the due date setting rules of NUMACT, CPTIME, and SFT. The
"k" values are ^nsitive to many different factors (for example, characteristics of the
project stream, the scheduling heuristic, resource levels, arrival rate, etc.) and are,
therefore, unique for each combination of scheduling heuristic and due date setting
rule used.
A pilot simulation run is used to estimate initial ^'s in stage one, using each of the five
different scheduling heuristics and the SFT due date rule with ^ 3 = 1 . These observa-
tions provide an estimate of mean completion time (MCT), which is similar to knowing
MCT from historical observations. Based upon these data, the initial fc's for rules
NUMACT, CPTIME, and SFT are:
NUMACT: ik, = MCT/(mean activities per project), (10)
CPTIME: k2 = MCT/(mean critical path time per project), (11)
SFT: ^3 = MCT/(MCT - mean lateness), where (12)
mean lateness = (2f-i TC, - SFT(£:),)/p,
TC, = time of completion of project /,
SFT(E)i = initial due date estimate of project / when it arrived,
p = total number of projects.
When the scheduling teuristic us« due date data in %tting priorities then some
interaction between k and the mean completion time may exi^. Conseqi^ntly, otto'
vsiaes of k were examined. St^e two involve a sy^ematic search of ctfher k values
around the i n i ^ estimates to ttetermine if corruption is iwcsssary. Values of k which
poduce near zero l a t e n t are good e^imates of the comi^ion tim». Several repUca-
 EVALUATING PRCWECT SCHEDULING 107

dons of ^^roximately half the normal length simulations usii^ k values somewhat
hi^er and «}mewhat lower than the analytical k value were tested. The initial estimates
for ki, used with NUMACT, and ^2, used with CPTIME, were determined (based on
mean lateness), while minor adjustments to ki, used with SFT, were required for
MINSLK[DD], MINLFTPD], and SASP[DD] to get mean lateness to approximately
zero. (See Dumond 1985 for complete analysis of'fc* values.)
4.2. Simulation Model and Test Environment
The simulation model is programmed in the SLAM (Pdtsker and P^den 1979)
software langus^e, usii% discrete event logic. As each new project arrives it is assigned a
due date and scheduled into the system. The simulation is event oriented and advances
the clock to the next event—^activity start, activity stop, project completion, new project
arrival, or end of emulation. The model simulates a dynamic environment in which
there is a continuous flow of new projects, arriving stochastically, which must be
comi^eted. As each new project arrives it is given a due date based on one of the four
due date rules and scheduled into the system using one of thefivescheduling heuristics.
The schedule developed by the heuristic establishes start and stop times for all activities.
Since duration times are assumed deterministic, the current schedule is maintained
until a new project arrives. At that point, the new project is given a due date and a new
schedule is developed and implemented for all activities.
Upon completion of a project a number of perfonnance criteria are coUected.^ They
are:
1. Project Mean Completion Time (days),
2. Project Mean Lateness (days),
3. Standard Deviation of Mean Lateness (days),
4. Total Tardiness—sum of all tardy days for all projects.
An observation (replication) consisted of the simulating of 2000 days of operation.
During the 2(XK) days approximately 250 projects arrived to the system following a
uniform distribution with a mean interarrival rate of eight days. Each project consisted
of a number of activities,rangingfrom 6 to 49 with a mean of 24, requiring from one to
three different resources simultaneously. Activity resource requirements ranged from

^a. M M B comj^etion time is cateulated as follows:

[ 2 (tci - ta,)]/p where

/-I

tc, = time of completion of project i,

tai = time of arrival of pnqect i,
p = number of tHOJects.
b. Mean lateness is calculated as follows:

[X(tci-dd,)]/p where
/-I
ddt >= (hie date of i»oject i.
c Tte standard deviation of lateness is adculated as follows:

(-1

d. Total tardiness is osculated as fdlows:

t-ddi if Uh>ddi,
otiiennae.
 108 JOHN DUMOND AND VINCENT A. MABERT

zero to eight units simultaneously for the three resource cat^ories. The mean critical
path time is 31.4 days, ranging from 8 to 78 days. The projects used in this experiment
were selected from ones used in other project scheduling research and are contained in
the Patterson monc^raph (1982). The specific projects used were Problems 7,9,10,13,
14, 20, 31, 37, 44, 54, 59, 61, 63, 70, 73, 92, 97, 98, 101, and 104.
The resource levels chosen for this experiment were selected to attain an overall
resource utilization of 85.35%.^ Two resources categories had 34 units of resource
available per day, with the third having 39 units per day. The three types of resources
used in this study were assumed to be fixed at those levels (i.e. no temporary changes
were permitted toreflectovertime, subcontracting, etc.).
The experiment was conducted using a CDC 855 series computer. The execution
time for each observation varied depending on the specific treatment levels and the
random number seed used to generate the project arrivals. The average execution time
per observation was 106.9 CPU seconds.

5. Experimental Results
5.1. Complete Control Environment
The analysis of the scheduling heuristics and due date assignment procedures is based
on four performance criteria, where 15 replications are collected for each treatment
combination. Project completion time, the dominant criterion found in the static
resource-constrained project scheduling research, resulted in approximately the same
valuesreflectedin the mean completion time data shown in Table 1, with SASP being
slightly lower than the rest. Three of the five heuristics consider the project due date
when scheduling the constrained resources and therefore the project completion times
are affected by the due date rule used to set the project's due date. The two heuristics,
FIFS and SASP, do not consider due dates and consequently have the same mean
completion time r^ardless of the due date rule used. The mean completion times for all
heuristics and due date rules are not significantly different using the Scheffee multiple
range at the 0.05 level.
The remaining three criteria are due date performance measures that focus on differ-
ences between each project's actual completion date and its promised due date. As can
be seen in Table 2 the mean lateness values are approximately zero for all combinations
of scheduling heuristics and due date procedures. These data confirm that the estimate
"k" parameters were correct in attaining the zero lateness goal.
The standard deviation of lateness, shown in Table 3, indicates how well the schedul-
ing heuristics and due date rul^ meet the promise due date. It is apparentfi-omTable 3
that significant differences at the 0.001 level are present for the heuristic rules, with
SASP causing larger lateness poformancs based on overall means across all due date
rules. An ANOVA test at the 0.001 level on the due date rule used indicates that
significant differences are also pre^nt here. The data indicate that the FLOW rule
performs the poorest and the SFT rule performs significantly better than the other three
at the 0.01 level using the Scheffee test. It is interesting to note the tremendous reduc-
tion in laten^s variability, at least half, withtihteSFT due date rule.

' Resource utilization is defined as the amount of actual resource us^ge (three types) ova- the time period
(2000 (biys simulated) and cdcuktted by ti» following equation:

amount trf'resource i' used in paiod t,

= amount (tfrescNBce i avaikl^ p o - ^ .
 EVALUATING PROJECT SCHEDULING 109

TABLE I
Mean Completion Time (in days) (Cdi Standard Deviation)

Due Date Rule

Heuristic FLOW NUMACT CPTIME SFT Row Mean

FIFS 52.9 52.9 52.9 52.9 52.9

(2.42) (2.42) (2.42) (2.42)
MINSLKtDD] 54.5 52.1 51.7 53.3 52.9
(2.74) (2.39) (2.07) (2.58)
MINLFTIDD] 53.8 51.2 50.7 52.0 51.9
(2.60) (2.25) (2.18) (2.52)
SASP[DD] 54.6 52.0 51.5 51.8 52.5
(2.76) (2.23) (1.99) (2.14)
SASP 49.1 49.1 49.1 49.1 49.1
(2.64) (2.64) (2.64) (2.64)
Column Mean 52.9 51.4 51.3 51.8

TABLE 2
Mean Lateness (in days) (Cell Standard Deviation)

Due Date Rule

Heuristic FLOW NUMACT CPTIME SFT Row Mean

HFS 0.85 0.26 0.48 0.33 0.48

(2.42) (2.46) (2.49) (0.10)
M1NSLK[DD] 0.49 0.49 0.55 0.78 0.57
(2.74) (2.42) (2.13) (0.64)
MINLFT[DD] 0.72 0.76 0.50 0.26 0.55
(2.60) (2.28) (2.25) (0.79)
SASPIDD] 0.50 0.38 -0.06 0.32 0.29
(2.76) (2.28) (2.05) (0.62)
SASP 0.03 0.29 0.60 0.34 0.32
(2.41) (2.71) (2.33) (1.06)
Column Mean 0.51 0.44 0.50 0.40 0.43

TABLE 3
Standard Deviation of Lateness (in days)

Due Date Rule

Heuristic FLOW NUMACT CPTIME SFT Row Mean

FIFS 29.29 21.32 23.09 2.13 18.96

MINSLK[DD] 2S.93 18.93 17.60 7.25 18.18
MINIJT[DD1 28.34 17.99 16.31 7.80 17.61
SASI^DD] 29.45 19.68 18.32 7.69 18.78
SASP 52.61 39.25 34.08 29.27 38.80
Column Mesm 33.72 23.43 2L88 10.82 22.46

Turning to the total tardin^s measure shown in Table 4, it is aiq;»rent that ths SFT
dm date rule inovides oinsderaUy better results than the other tiuee due date r u l ^
^i«4ik^ is confirm^ by the Sdieffee test Significant diflfeiencK also east b^tvraen SASP
110 JOHN DUMOND AND VINCENT A. MABERT

TABLE 4
Total Tardiness (in days)

Due Date Rule

Heuristic FLOW NUMACT CPTIME SFT Row Mean

FIFS 3107 2190 2305 210 1953

MINSLK(DD} 3193 2034 1837 851 1979
MINLFT[DD) 3168 1972 1700 840 1920
SASPIDD] 3254 2114 1849 826 2011
SASP 3857 2547 2047 1689 2534
Column Mean 3315 2171 1947 883 2079

and the remaining four scheduling heuristics tested, with no overall difference between
FIFS, MINSLK[DD], MINLFT[DD], and SASP[DD]. The results here confirm what
was observed with the standard deviation of lateness measure.
Overall, the results indicate that there isrelativelylittle or no difference between the
three due date oriented heuristics, MINSLK[DD], MINLFT[DD], and SASP[DD].
Additionally that FIFS performs similar to them, with the major exception being when
FIFS is used with SFT due date rule. This combination, FIFS/SFT, resulted in signifi-
cantly better values on all the due date critieria of standard deviation of lateness and
total tardiness. SASP provided the lowest mean completion time, while it also had the
poorest due date performance, as measured by total tardiness and lateness variability.
5.2. Partial Control Results
The previous experiment assumed that management had full control over the setting
of due dates; althou^ a desirable condition, it is not typical. Frequently, management
may be able to set the due dates for most of the projects, but some projects arrive which
must be completed as soon as possible, and other projects arrive that can be given the
"back burner" treatment.
This experiment assumes that projects fall into one of three priority categories: (1)
normal, (2) h i ^ , or (3) low. Management controls the setting of due dates for the
normal priority projects not for the high or low priority projects. The high and low
priority projects are assumed to arrive with their due dates already set. For this test, the
high priority projects arrive with their due dates set as the eariiest possible completion
time which is the critical path time. The low priority projects arrive with their due dates
set at three times the project's critical path time. The &ctor of three was chosen based

TABLE 5

Proportion
Due Dates Mean Comi^etion Mean Lateness Lateness Std Dev. Total Tardiness
Set
External CPTIME SFT CPTIME SFT CFTIME SFT CPTIME SFT

0 51.7* 52.5 0.32 0.41 19.3 6.2 1923 676

0.1 51.9 52.6 -1.30 -1.23 21.9 13.8 1851 750
0.2 51.9 52.6 -2.95 -3.06 25.5 18.4 1845 838
OJ 51.9 52.8 -3.69 -3.72 26.6 21.9 1878 1017
0.4 52.1 52.7 -4.75 -4.92 28.4 24-7 1889 1117
0.5 52.4 52.7 -5.75 -5.99 30.1 27.3 1882 1249
52.0 52.6 -3.02 -3.08 25.3 18.7 1879 941

t. Each data entry is the ttyengie performance for all four scheduling heuristics
 EVALUATING PROJECT SCHEDUUNG 111

CPTIME Due Date Procedure

Mean Completion Time
Time ((Iay8)

0,1 0.2 0.3 04 0.5

Proportion Set Externally
FIFS -t- MINSLK(DD) - * - MINLFT(DD) SASP(DD)

SFT Due Date Procedure

Mean Completion Time
Time (days)

0.1 0,2 0,3 0,4 0.5

Proportion Set Externally
FIFS < - tw«NSLK(DD) - » - K«NLFT{raD) - SASP(DD)

ExHmrr l

on the £act that "normal" projects took about l.S times their critical path time to
comjriete. Doubling of the normal provides dgnificant project ^bck for low piority
projects.
Normal iHojects, under maimgement control, have their due dates set by either SFT
or CPTIME, ^ u d i were selected for further study based upon two reasons. First, SFT
w ^ deariy ta& superior perftmna durii^ the comjriete control test, with CPTIME the
best of ^ e remaining three. And %cond, these two inxxedures immde an important
contrast in tiie df^ce of stqjlmtication, with SFT estimatii^ due dates based upon
current ^^stem 1(KM]S and CPTIME usa% an i
Tl» focus crf'^s ^cpaimoit is to assesstiieperfcHinance of SFT and CPTIME as tte
 112 JOHN DUMOND AND VINCENT A. MABERT

CPTIME Due Date Procedure

Mean Lateness
Time (days)

0.1 0.2 0.3 0.4 0.5

Proportion Set Externally
FIFS • + - M1NSLK{DD) -*- MINLFT{DD) SASP(DD)

SFT IXie Date Procedure

Mean Laten^s
Time (days)

0.0 0.1 0.2 0.3 0.4 0.5

Proportion Set Externally
- * - Fire - + - MtNSLK(DD) -*- MINLFT(DD) - a - SAa»(DD)
ExHiBn'2

proportion of projects whc»e due dates are %t extenmlly incr^ees, using a three factor
expmment design. One &ctor is the portion of project that arrive with externally set
due dates—this factor has six treatments (0,0.1,0.2,0.3,0.4,0.5; where 0.1 equals 10%,
0.2 eqirals 20%, etc.). Ilie second factor is tte scl^uUi^ heuri^c &clor at four
treatment leveb (FIFS, MINSLKIDD], MINLFTIDDI, and SASP[DD1). The third
factor is ths two due date (SFT and CPTIME) setting rules u s ^ to set tt^ dm dates for
tise "nomuil'' {at33«^
Wheo mam^ment does oot have full control over tibe setti^ of due dates some
deteru>ratioa in pedmmsaxx is antidpated. The a n a l ; ^ of this experiment is bi^ed oo
the four performance criteria discu^ed earlier. TaUe 5 presents a summary of
CPTIME's and SFFs performance for the four criteria, lunros tiie ax settings of 0 to 50
 EVALUATING PROJKT SCHEDUUNG 113

CPTIME Due Date Procedure

Lateness Standard Deviation
Time (days)

0.1 0.2 0.3 0.4 0.5

Proportion Set Externally
FIFS -t- MINSLK(C») -*- MINLFT(DD) " SASP(DD)

SFT Due Date Procedure

Lateness Standard Deviation
Time (days)

0.0 0.1 0.2 0.3 0.4 0.5

Proportion Set Externally
•— FIFS -t- MtNSLK(DO) -•- k«NLFT(DD) - » - SA3P(DD)

EXHIKT3

percent of project due dates set externally. Tlw data are the avera^ of the four heuristic
proc^ures. Iitde change can be noted in mean completion time for different levels of
control. However, one notes significant changes for the otiier thrx measure. To obtain
a b^ter ua^rstandii^ of th»e rraults, we use graphical analysis by individual schedul-
ing toiristicrute.
Exhibit 1 pr^ents a plot of mean comi^tion time. U»ng either CPTIME or SFT due
d i ^ rule to set the doe dates ofUbe internally controUed projects the mean comidetion
timra see \iitually una£fe(^ed bytfiefS'i^xHtioQ of projects that have ti^ir due dates set
externally. We con,^ctiB« that tiiis ocoirs because resource are never intentionaUy left
Kile. All the l»»ui^cs allocate the resources wteiever they are availa^and sufl^^nt
to me^ the ^maods of tlK next activity on tlm prioritized list. For due date oriented
 114 JCWN DUMOND AND VINCENT A. MABERT

CPTIME Due E^te Procedure

Total Tardiness
Time (day8)
2600

0.1 0.2 0,3 0,4 0,5

Proportion Set Externally
FIFS MNSLK(DO) MINLFT(DD) SASP{t»)

SFT Due Date Procedure

Totsd Tardiness
Time (days)
2600

0.0 0.1 0,2 0,3 0,5

Proportion Set Extwnally
FIFS MNSLKfCX}) ti«NtFT(DD)

ExiiBaT4

heuristics this means that if dl the high and norm^ priority projects are scheduled ami
reKHirc^ are still avaiiUiUe then "bsKsk burner" projects will be wc^ed cm. Conse-
quently, m«m comptetion time for these teuristics vias relatively insemitive to the
portion of {Hcgeds arrivii^ with ex^nally set doe d^es. For tbs noodue daie oriented
t»uriAic, FIFS, the mean con^^etiontimeis al»(du^y unaffected by the externally set
due d^es—^«4»e^ar h ^ priority w low {Hicmty.
Tte meaa lates^s r ^ t s {dotted in Exhibit 2 r ^ c t the fuiA that tte "back bamer"
due d^es w^e hoosex than tte "hot, h ^ pskmty" ^te <Mes v/&e t ^ t As mixe and
more isojects arrived with eitemaUy set due dates tte meui latenras decteseed. This
occiared li^wtho'tl» inlonudly contndted {mjecte had tlieir dt» (telra s^ by CPTIME
or SFT.
 EVALUATING PROJECT SCHEDULING 115

The effect of externally set due dates on the distribution of laten^s is apjaient in
Exhibit 3. The overall perfonnance deteriorates as the number of projects which arrive
with externally set due dates increases. This occurs whether management uses the
CPTIME or SFT rule to set the due dates of the project it controls. The most dramatic
deterioration occurs in the FIFS/SFT combination whichreliesheavily on the setting of
"good" due dates since FIFS does not "chase" the established due date.
Exhibit 4 indicates the results r^arding total tardiness are slightly different, depend-
ing on the due date rule used to set the internally established due dates. When using the
CPTIME rule to set the dates the results indicate that neither the due date oriented
heuristics nor FIFS are grossly affected as more projects arrive with externally set due
dates, where the overall effect on total tardiness is only very slight.
Overall, when using the SFT rule to set the due dates of the internally controlled
projects, the results continue to be far better than when using CPTIME, al^ough there
is some deterioration. The due date oriented scheduling heuristics experience a slow rise
in total tardin^s which reflects their ability to adapt to the established due dates, no
matter what the source. A more substantial deterioration is apparent in the results
regarding the FIFS heuristic. The remarkable perfonnance of the FIFS/SFT combina-
tion with regard to the total tardiness measure is eroded as fewer and fewer projects
have their due dates set by management.

6. OHiclnsioiis
In general, it was found that ^ e three due date oriented heuristics—MINSLK[DD],
MINLFr[DD], SASP[DD]—performed close across all due date criteria, while SASP
and FIFS were often significantly different from the others. Also, the three due date
oriented heuristics were found to berelativelyinsensitive to various levels of externally
set due dates when the SFT due date setting rule is used to set the due dates for the
remaining projects. On the other hand, FIFS rapidly deteriorates in perfonnance when
a greater proportion of dates are set externally.
The FIFS/SFT combination provided the most inter^ting results in all the experi-
ments. Althoi^ the FIFS heuristic is not due date oriented, it still performed very well
on the due date measure. When used with the SFT due date setting rule, established
due dates were based upon good knowle^e of system l c ^ conditions. Consequently,
the SFT rule provides very good estimates of the project completion time which tends
to be met by the FIFS heuristic, when all due date setting is controlled by the FIFS/SFT
combination. The FIFS heuristic, by ignoring due dates, does not suffer from the
"nervousnras'** found in the paforman<x of the due date oriented heuristics. However,
when complete due date control is not present, FIFS can not adjust to externally set
due dates.
Since a number of the due date rules reported here are adapted from the job ^ o p
scheduling literature, it is ^qnopriate to compare and contrast r^earch results. Table 6
summarizes some of tiiis work by the four due date rules: 1. Comtant Allowance
(FLOW), 2. Number of Operations (NUMACT), 3. Total Work/Process Time
(OTIME), and Complete Status Information (SFT). Your authors selected the perfor-
mance criterion of tardiness to compare, since it vms the common stati^ic among the
studi«i review^ and a imoiary mea»ire of due date perfonnance.
lilce all comparisons, caution must be ^vised for a number of reasons. First, not all
i<^ ^ o p sd^duiing teuristics map directly with this r^xirted study. For example. Fust
O>me-First S«ve (FCFS), Siack P& C^X3»tu>n (S/OP), and l^iortest Process Time
(SPT) are not equi^^fent to HFS, MINSLjqDD], and SASP respectively. Second, tiie

<n one owoAw activity to aaoAo'aad bade agaia as wixk ptognsaei.

116 JOHN DUMOND AND VINCENT A. MABERT

TABLE 6
Due Date Rule/Scheduling Heuristic Tardiness Performance Summary

No. of
Author Machines Scheduling Heuristic Ranking Based on Tardiness

I. Constant Allowances Due Zkcie Rule:

Bida!r&Bertrand(1981) 1 nFS*>''SPT
Weeks (1979) 8-24 SPT S S/OP
Dumond&Mabert - SASP > SASP[DD1 > J J j J J ^ ^ ^ j > FIFS

2. Number of(^)erations Due Date Rule:

Ragatz & Mabert (1984) 9 S/OP > FCFS > SPT
Dumond & Mabert — SASP > HFS > SASP[DD] > MINSLK[DD1 > MINLFT[DD]
3. Total Work/Process Time Due Date Rule:
Baker &Bertrand (1981) 1 MSLK > SPT
Eilon & Chowdhury (1976) 5 FCFS>SPT
Elvers 8 S/OP > EDD fe HFS > SPT
Weeks (1979) 8-24 SPT S S/OP
Ragatz & Mabert (1984) 9 FCFS > S/OP > SPT
Dumond & Mabert — FIFS > SASP > ^ ^ D ^ [ * ° ' > MINLFTIDD]

4. Complete Information Due Date Rule:

Ragatz & Mabert (1984) 9 SPT > FCFS > S/OP
_ MINLFT[DD]
Dumond & Mabert
SASP > SASPIDD] > FIFS
MINSLK[DD]

* SPT = Shortest Process Time; S/OP = Slack Per Operation; FCFS = First Come First Serve; EDD
= Earliest Due Date; MSLK = Minimum Slack
*'A> B indicates heuristic A had greater tardiness than B.A'&B indicates Heuristic A had equal or greater
tardiness than B.AS B indicates no dominance in reported results.

shop environments varied widely in characteristics from number of machine centers to

the types of distributions controlling arrival and service rates. And third, the due date
rules of CPTIME (similar to Total Work/Process Time) and SFT do not have exact
counterparts.
Recognizing these limitations. Table 6 does allow one to see a number of interesting
points about scheduling heuristics. First, no one heuristic performs best across all
studies and due date setting combinations. For example, SPT provides the lowest
tardiness in a number of cases when combined with Constant Allowance, Number of
Operations, or Tot^ Work/Process Time due date rules. When more work centers are
present, S/OP seems to be as good or better than SPT. Second, when FIFS/FCFS type
heuristics are used with Total Work/Pro<»ss Time due date rules, the greatest tardiness
is experienced. Third, altiioi^ SPT performs very well in the job ^ o p environment
with Constant Allowance, Number of Operations, and Total Work/Process Time, ife
project management SASP cousin routinely performs the poorest. Fourth, there seems
to be a shift in the heuristic rankings as one moves from naive to more s(^histi(ated due
date planning lc^c. This study and one reported by Ragatz and Mabert (1984), which
u ^ a complex regre^on model to estimate due dates, indicate ihat SPT^ASP type
heuristics had the p o o r ^ perfcMinance. Artd ifth, the studies by Ragatz and Matot
(1984), Eilon and Chtnvdhury (1976), and this one confirm the c o n c ^ of uang more
detail infixrmation (shop status data in job shops cnr resource as^nments/pr«»dent
constraints in project management) to estaJdish and more clc^dy m ^ t due dates.
ThisreviewnolK some of the diflferetices in perf(mnance of the sdheduling heuristics
tested. Davis (1973) has noted certain amiiarities between the ^ b diop and project
 EVALUATING PROJECT SCHEDUUNG 117

man^ement problem, but there are suffident differences in research results to indicate
that they may not be equivalent. For example, projects have parallel branches of
activities requiring multiple resources, with splitting/joining of branch paths, while job
shop research has been viewed as serial routings and single resource requirements. A
systematic study would be appropriate to evaluate the equivalence of the environments
and the use of the same planning/control logic.
This study has provided an initial evaluation of due date assignment procedures,
when using various scheduling heuristics. The results suggest the SFT due date rule
provides very good results for all test settings, since it incorporates more detailed
infonnation into the due date decision than the more a^r^ate planning factor ap-
proach of CPTIME. The benefits from SFT are not without some costs, as compared to
CPTIME. When a new project arrives to the system, the calculation of the appropriate
due date using CPTIME is given by equation (6). The calculation time required to do
this by a high speed computer is just a few milli-seconds, and attempts to accurately
measure this value in the study were fruitless. On the other hand, SFTfinitelyschedules
all activities to determine a promise due date, requiring more computing effort. In this
study, it required an average of 0.424 seconds of CPU time to develop one feasible
schedule for a newly arrived job.
The computing requirements for SFT were not excessive for the small to medium
size projects scheduled in this study. However, some practidng managers often en-
counter projects with activities exceeding a thousand and computing effort due to size
may be of concern. Our research indicates that computing effort changes in a modest
linear manner with project size (measured by number of activities), as long as the same
level of utilization (resource requirements/resources available) is maintained. However,
when this ratio changes, a major shift occurs in CPU requirements. For example, the
resource levels were increased and decreased by 6.5% from the main factor settings in
this study to evaluate the effect on computing. These changes resulted in a 38% decrease
and a 106% increase, respectively, in computing time for SFT to provide a feasible
schedule. This suggests an exponential relationship between computing time and re-
source utilization.
Presenting a realistic cost-benefit analysis of using a SFT type planning system would
be impractical in this experimental investigation. However, one canreviewsome infor-
mation from the Babcock & Wilcox (B&W) experience,^ and draw some comparisons.
It cost B&W $225,000 to initially design and program their system, with an additional
$200,000 per year for staff and computer operations (Buffa and Miller 1979, p. 681).
Being able to more accurately predict completion dat^ of projects allowed B&W to
reduce costly escalation costs, which were as much as $ 100,000 per week. Thomas and
Coveleski (1973, p. 28) note $360,000 in savings in one case due to escalation charges
and the use of LRPS. Other benefits were also noted by the use of the system, such as
identifying bottleneck resources early in the planning process for close monitoring.
Also, the system helped guide the appropriate investment of millions of dollars of
capital equipment and the avoidance of costly equipment additions that provided little
benefit.
The documented B&W experience indicates that planning systems using SFT type
logic provide tremendous finandal benefits; while this study presents a sy^matic
comi^rison of different due date planning and activity scheduling approaches. Our
research illustrate that SFT provides superior performance over more aggr^ate plan-
ning factor api»^(»ches (i.e. FLOW, NUMACT, and CPTIME) in setting due dates. The
reduction in latene^ variability and tardines by a factor of two or more by SFT, as

' Buffit and Miikr (1979, iq>. 663-695) {»esem an exteasive case study on the Loag R a ^ nanning System
at 'BtSxock uul ^^kox Coipotxivm. Also, personal cmvasatkms daring January 1986 wUh B&W ^aff have
coafirmed tias iofiHrmation to be acramte.
 118 JOHN DUMOND AND VINCENT A. MABERT

shown in Tables 4 and 5, indicates that the benefits from a SFT type planning system
would be substantial.
Biblit^praphy
BAKER, K . AND J. W. BERTRAND, "An Investigation of Due-Date Asagnment Rules witii Constrained
Tightness,"/. Oper. Management, 1, 3 (February 1981), 109-120.
BUFFA, E. AND J. MILLER, Production-Inventory Systems: Planning and Control, 3rd. ed., Richatd D. Irwin,
Homewood, IL, 1979.
COCM>ER, D. F., "Heuristics for Scheduling Resourt^-Constrained Projects: An Experimental Investigation,"
Managemem Sci., 11,11 (July 1976), U 86-1194.
D A V S , E . W., "Project Scheduling Under Resource Constraints—Historical Review and Categori2ation of
Procedures," AIIE Trans., 5, 4 (December 1973), 297-313.
AND G. E HEIDORN, "An Algorithm for Optimal Project Scheduling Under Multiple Resource
Coastmtas," Management Set, 17,12 (August 1971), B-803-B-816.
AND J. H. PATTERSON, "A Comparison of Heuristic and Optimum Solutions in Resource-Con-
strained ftoject Scheduling," Managemem Sci., 21,8 (April 1975), 944-955.
DOERSCH, R. H. AND J. H. PATTERSON, "Scheduling a PtoH«t to Maximize its Present Value: A Zero-One
I^ogrammii^ Ansoach," Managemem Sci., 23,8 (April 1977), 882-889.
DUMOND, JCNHN, "Evaluation of Due Date Assignment Rules for Dynamically Arriving Projects," Ph.D.
EKssertation, Indiana University, Bloomington, 1985.
EiLON, S. AND I. G. CHOWDHURY, "Due Daies in Job Shop Scheduling," Imemat. J. Production Res., 14,2
(1976), 223-237.
AND R. M. HODGSON, "Job Shops Scheduling with Due Dates," Imemat. J. Production Res., 6, 1
(1967), 1-13.
ELVERS, D . , "Job Shop Dispatdiing Rules Using Various Delivery Date Setting Criteria," Production and
Invemory Managemem, 14,4,62-69.
FENDLEY, L. G . , "Toward the £>evelopment of a Comirtete Multiproduct Scheduling System," J. Industrial
Engineering, 19, 10 (1968), 505-515.
KURTULUS, I. AND E . W . DAVIS, "Multi-Project Schedulii^ Categorization of Heuristic Rules Perfor-
mance," Managanem Sci., 28, 2 (February 1982), 161-172.
MASON, A . T . AND C . L. MOOEHE, "A Bnmdt and Bound Algorithm for Minimizing Cost in Project
Sd»edvi&t^" Managemem Sci., 18, 4 (December 1971), B-158-B-173.
PATTERSON, J. H., "Alternative Methods of Project Scheduling With limited Resources," Naval Res. Logist.
Quart., 2 0 , 4 (1973), 767-784.
, "Exact and Heuristic Solution Procedures for the Constrained Resource Project Scheduling Prob-
tem," Vols. I-IV, a monograph, November 1982.
, "A Comparison (rf Exact Amnoadies for Solving the Multi{de Constrained Resources, Project
Scheduling PaMem," Managemem Sci., (July 1984), 854-867.
AND W. D. HuffiR, "A Horizon-Varying, Z«x)-One Approach to ftoject Scheduling," Management
Sci., 20,6 (Fetauary 1974), 990-998.
AND G. W. ROTH, "Scheduling a Project Under Mtittii^e Resource Constraint: A Zero-One Program-
mii% An>roadi," AIIE Trans., (Decembn 1976), 445-455.
PRmocER, A. A. B. AND C D. VEOIOEH, Introduction to Simulation and SLAM, idm Wiley & Sons, Inc., New
YoA, 1979.
, L. J. WATTERS, AND P. M. WOLFE, "Multiiwoject SdMsduUi% With limited Resources: A Zero-One
Programmii% Ansoadi," Managemem Sci., 16,1 (September 1969), 93-108.
RAGATZ, G . L. AND V. A. MABERT, "A Emulation Analyses of £>ue Date Ass^nment Rules," / . Oper.
Managerrmtt, 5, I (No^«mb« 1984), 27-37.
SCHRAOE, L., "S(4vii% Resource-Coostnined Netwwk Prot^ms l ^ Imididt Enumeration-NoniHeemptive
Case," (^er. Res^ 10, (1970), 263-278.
Snuscm, J. P., E W. DAVIS MO) B . M . KHUMAWALA, " M U W ^ Resource—Constrained SdieduUi^ Using
ftanch and Bound," AIIE Trans., 10,3 (SejAembn 1978), 252-259.
TALBOT, F . B . , "Rnource-Con^rained Project Sdiedulii% With Time-Raource Tr^e-oflfe Hie Non-
pfeempUve Case," Mam^ment Sci., 28,10 (OcteUsa 1982), 1197-1210.
AND J. H . PATT^tSCWI,"/
," Al<tfu«emen{&t., 24, U (July 1978), 1163-1174.
A., "HeurBtk Sdieduling (^Activities Under Precedence Restrictions," Mam^emem ScL, 2 3 , 4
onbR 1976), 412-422.
E. A^a^ D. COVELESCI, I t a n s i n g Nistear Equqanmt Manu&ctnrii^," Imetfiu^, 3, 3 (May
1973), 18-29,
J. K., "A Smutaokm Stady <rf Rre&Hd^ Due-D^es," Mant^mem Sd., 25, 4 (A|ml 1979),
363-373.
WBBST, J. D., "A Ifeiiriak Modd to Sdiedidiag Ijoge Projects witt Limited Sesouices," i t / o f ^ ^
13,6 (FaKuary 1967), B35»-B377.