Effective Design And Use Of Computer Decision Models

Computer Decision Models

Effective DesignAnd Use Of ComputerDecision Models

By: William L. FuerstMerle P. MartinDepartment of Business Analysis and

ResearchCollege of Business AdministrationTexas A&M UniversityCoiiege Station, Texas 77643

IntroductionOne of the main problems with today's computerdecision models is that, for the most part, they arewritten in computer languages which only com-puter programmers can readily understand.Managers have no idea of the details of the com-puter program, tn some organizations this defi-ciency is alleviated by the collection of a technicalstaff who. while not programmers, have enoughtechnical knowledge to learn, use, and interpretcomplex models. However, managers oforganizations without such technical experts arecompletely dependent on computer program-mers, whether such programmers are employedby the user organization or by the firm marketing apackaged model.

The purpose of this article is to provide bothdesigners and potential users of computer deci-sion models with criteria for development andevaluation of those packages. This is done bypresenting a review of the important user con-siderations which affect the success of modelusage, by presenting an evaluation of selectedfinancial simulation models, and by offering sug-gestions for possible enhancments that will serveto expand the use of computer decisions models.

AbstractComputer decison rriodels often provide useful resultsas management planning tools. However, ttiese toolsare frequently limited to firms with staffs of specialistswho can assimilate the technicai nature of the models.For other firms, the success of decision models suchas simuiation have not been demonstrated. This paperlooks at recent literature regarding decision mode/ defi-ciencies, evaluates selected financial simulation modelpackages, and suggests design needs for expandingthe use of decision models to a broader range of firms.

Keywords: Decision models. MIS design, simulation

ACM Categories: D.3, H.I

Factors of SuccessfulDecision ModeisThe user of a computer decision model isaffected by several factors as presented in Figure1 These factors determine the extent of use andrelative success of the models. The problemenvironment includes considerations such asresponse time and accuracy required to make adecision. The environment of the organization dic-tates the acceptance of model results and thededication of required resources to use the modeleffectively. As will be stressed in this article, thetechnical background and expertise of the user'sstaff affects the extent of use of a decision model,as does the user's experience with computerizedmodels. Additionally, users of a decision modelwill be constrained by hardware considerations.For example, some organizations may use com-puters with a memory so limited that use of cer-tain decision models may be impractical.

MIS Ouarterly/March 1984 17


HardwareConstraints

TechnicalBackground

andExpertise

SoftwareConstraints

USERSimulation

ModelAttributes

User Experiencewith Computerized

Models

ProblemEnvironment

OrganizationalEnvironment

Figure 1. The User's Environment

18 MIS Quarterly/March 1984


The user may also be constrained by thelanguage in which the decision model isprogrammed. Organizations that use microcom-puters or minicomputers may not find it costeffective to support the language in which themodel is written. Programs written in languagesother than BASIC or perhaps PASCAL willprobably preclude the use of such programs frommost organizations that use only microcomputers.

Finally, the specific attributes of the decisionmodel will have an effect on the extent and suc-cess of a user's involvement with that model. Thisis particularly true when the model's attributesconflict with other user factors such as technicalbackground/expertise and the problem environ-ment. Discussion of model attributes and theirconflicts with user factors will be the primaryfocus of this article.

Problems of Computer ModelsIf computer modeling is to be more widely usedas a decision support tool, then the modelinglanguage must be useful to casual users of com-puter systems, rather than only to programmersor technicians. Without the assistance oftechnical staff, firms will face a modeling environ-ment where managers are essentially the onlyusers, and only when their time allows. Casualusers may not be able or willing to recall complexdetails of a language.

For these casual users the system must providesupport to help them write or use a computermodel. Extensive error routines must be built intothe model to guide the casual user in building thatprogram [9], The language must allovw a program-mer or management modeler to describe thesystem in terms familiar to the operation beingmodeled [8], This creates a need for new. higherlevel computer languages [1 , 25] that allow thecomputer system, through complex subroutines,to deal vi ith much of the detail work. Such asystem would free the designer to deal with thecomplexities of the system to be modeled ratherthan to concentrate on the details of coding. Itwould also allow the designer more time toensure that the user is getting the informationdesired. Quite often, decision models are lessuseful because the solution presented to the user

is a bewildering volume of statistics to be inter-preted [6].

Another problem area concerns the user's con-fidence in the validity of the decision model. Inmany cases, this concern is justified; many com-plex models are inadequately validated becausetoo much work is required due to the detail of themodel and its output. An unverified model may notbe an adequate representation of the firm s realify[4]. Lack of model validation, of communicationbetween the user and designer, and of userunderstanding of the model results in a con-fidence trauma. The user's lack of confidence inthe model means the model will not likely be usedas a tool for solving problems or supportingdecisions [19].

An accepted precept of systems design is thatsystems have a higher probability of succeeding ifusers are involved in development. User par-ticipation in model development ensures theopening and continuation of a communicationschannel which should lead to shared under-standing. The developer of a model may notadequately understand the real world process ingeneral, or the firm's process in particular. In turn,the user does not understand the nuances of themodel and programmer logic and constraints.While participation is desirable, overcoming thiscommunication problem is time-consuming andinhibits the use of models [19], The problem iscompounded by the use of packaged computermodeling systems where the user and the vendorprogrammer are geographically separated and donot share common organizational goals.

Evaluation ofSimulation PackagesThe problems highlighted in the previous sectionlimit the use of decision model packages as deci-sion making tools, particularly as the number ofstaff technicians decreases. In an attempt toillustrate these problems, several financial simula-tion packages were evaluated. The evaluationcharacteristics used in the analysis were aimed atthe ease and effectiveness of use by decisionmakers and are supported in the literature asimportant criteria. Since the emphasis of thisstudy was to evaluate use of decision model

MIS Quarterly/March 1984 19


packages, cost was not included as a factor. Costis a more pertinent measure in acquisition deci-sions and machine-time performance evaluations.The evaluation here consists of determiningwhether the following characteristics of the finan-cial simulation packages were present,

• Interactive

• Conversational

• A natural language

• Nonprocedural

• Programmable by the user

• Modular

• Self-documenting

• Database oriented

• Graphics oriented

• Designed with user assistance

Evaluation criteria

Simulation packages should allow for human inputduring the processing cycle [3, 7, 10, 16],because more organizations, especially thosewith minicomputers and microcomputers, areusing interactive computing. Interactive packagesare not necessarily conversational [15, 22].Frustrated users continually ask "v 'hy — whycan't the computer do this, why can't it do that?"This is basically a communications problemcaused by differences between computercapabilities and human skills. Computers havevery fast processing capabilities, but must do soin a predetermined fashion. Humans, on the otherhand, are slow processors but are capable ofbeing creative and making decisions fromincomplete information. Since humans often thinkand provide input to a decision making process inan unorganized manner, systems must allow forinput instructions to be entered in an unstructuredmanner. The system must then translate thoseinputs into meaningful queries which the com-puter can process. Unfortunately, such systemsare not well-developed to date, resulting in a largegap between man and machine. Althoughsystems analysts, consultants, simulationspecialists, and programmers function to close

that gap, many organizations do not have suchspecialists permanently involved. Conversationallanguages allow the user and computer to com-municate, and permit the user to fulfill the pro-cedural requirements of the model withoutsoliciting the specialized knowledge of analysts orprogrammers.

One of the newer techniques to help close thegap is the use of natural languages, which allowusers to provide data and instructions in non-procedural, English-like commands. Naturallanguages reflect the application rather than anartificial structure based on logical principles [10,13]. These languages are further enhanced ifthey can preempt the computer's need to work ina predetermined fashion. This can be accom-plished by allowing instructions to be entered inany order in a nonprocedural fashion [20], Alanguage translater then has the task ofconverting the nonprocedural statements into acorrect procedural program for computerexecution.

Design of a simulation language with the user andhis/her skill limitations in mind would lead to amodel that is programmable by the user. Thisallows packaged programs to be specificallytailored to the firm without the need for hiringexpensive staff specialists.

To control complexity, modular design can beused fo organize the system into a set of increas-ingly complex modules [14, 18, 23]. Languagescan be used that are self-documenting, that is,they are comprehensive to the user without addi-tional information [5,11,21], This feature is a fur-ther enhancement of a simulation language, for itallows the user the opportunity to become well-versed in the language with a minimum amount oftraining time. Access to an external database is ofvalue to some users [17]. Graphical display ofmodel variables as they evolve allows the user anopportunity to see how a problem develops overtime. Providing flexibility during execution to allowthe user to interact with change in the model asdisplayed graphically is also an aid to building,verifying, and understanding that model [12],Finally, a model that can be designed with userassistance enhances the communicationbetween developer and user, and results inincreased user confidence and understanding[19]. While the inclusion of such user assisteddesign cart hardly be expected in the initial pro-



gramming of commercial simulation packages, itcan be provided through development of useralteration modules where report formats and pro-cess flow alternatives are selected by the user.

A look at selected packages

Financial packages were chosen in this evaluationbecause (1} they are most often the first type of"advanced" packages, beyond transaction pro-cessing packages, to be implemented, and (2)they are capable of cross-industry use. The sixpackages chosen, IFPS, FCS, Plan Master, RAL,Plancode, and IMPS were selected because theyare commonly used in industry and are well-documented. Figure 2 presents evaluation resultsof these six financial packages concerning thecriteria outlined in the preceding subsection.Although it is not desirable in this paper to discusseach package individually, it is possible fodraw some general conclusions based on thisevaluation.

All of the packages evaluated are interactive, butnone is conversational; no system prompts theuser for specific inputs. Users of these packagesmust possess prior knowledge of input pro-

cedures. The capability of submitting instructionsin a nonprocedural fashion varies among thepackages, but none support a natural languageapproach. However, of the packages examined,half allow programming by the users and provideself-documenting aids. Most packages providefor modular design to reduce complexity. Half ofthese have the capability of accessing an externaldatabase, whereas only one provides graphicaldisplays. Finally, none of the packages allows theuser to initially assist in, or to extensively modify,the design of the model.

Designing ComputerSimulation ModelsThe preceding section should not be interpretedto imply that financial simulation packages are noteffective. Indeed, one survey of users of financialsimulation planning models indicated that suchmodels were successfully implemented by middlemanagers [25]. However, this survey was con-ducted in large firms with extensive technicalstaffs possessing such titles as Financial Analyst,Planning Analyst. Project Engineer, and Cor-

Characteristics

Interactive

Conversational

Natural Language

Nonprocedural

Programmed by User

Modular

Self-Documenting

Exfernal Database

Graphical I/O

Designed withUser Assistance

IFPS

yes

no

no

yes

yes

yes

yes

yes

yes

no

FCS

yes

no

no

yes

yes

yes

yes

yes

yes

no

PLANMASTER

yes

no

no

yes

yes

yes

yes

yes

yes

no

RAL

yes

no

no

no

no

yes

no

no

no

no

IBMPLANCODE

yes

no

no

no

m

no

no

no

no

no

IMPS

yes

no

no

no

no

no

no

no

no

no

Figure 2. Characteristics of Selected Financial Packages



porate Researcher. If software packages such assimulation models are to be more widely used, ashas been computer hardware, these packagesmust be designed to allow managers, rather thantechnical staff, to effectively develop and use themodels required.

It is toward this end of allowing managers todevelop and use models that the remainder of thisarticle is directed. The next section presents thenecessary characteristics for (1) enhanced use ofsimulation models in organizations with technicalstaffs and (2) increased use in other organizationswhere technical expertise is at a premium. It isrecognized that development of such ideally user-oriented simulation models may be so costlytoday as to effectively exclude usage by the firmswith predominantly casual users. However, giventhe potential size of the commerical market forsuch user-oriented models, it is anticipated thatuser costs for acquisition should decreasesignificantly over time.

Goal of Simulation iVIodelsFigure 3 presents the user-oriented goal, criteria,and factors of simulation models. Statement andmeasurement of a goal are problems when deal-ing with the impact of a simulation model on mak-ing successful decisions. No optimal solution isgiven to the user as is the case with optimizationmodels. Rather, inputs to the simulation model aremanipulated to determine relative effects on theoutcome, providing the manager with increasedinsight into, rather than a solution to, the problemat hand. Measurement of such increased insightis not well-defined. Thus, the most appropriategoal concerns the effective use of the model as adecision-making aid.

This goal can be broken down into three criteria:user understanding, ease of use, and modelresponsiveness. Each of these criteria can be fur-ther subdivided into specific factors of the goal ofa user-oriented simulation model.

User understanding

The model must be understood by users who donot possess technical familiarity, skills, or staff.Modular progreun design has been shown to be

beneficial. In addition, communication betweenthe user and the model must be based onmethods and techniques that best impartknowledge of complex systems. To achieve atrue self-documenting system, the most effectivemedia of documentation must be selected. Fur-ther experimental research must be conducted asto what is the most effective media. Care must betaken in research, and in practice, to differentiatebetween programmer and user documentation.Finally, it is generally accepted that user involve-ment in systems design increases understandingand confidence in the system being implementedThis can be developed by providing for a develop-ment module that allows the user to define, in hisown terms, specialized process sequences,specific report formats, and user/mode! protocol.User involvement in the design process can beenhanced by bringing together the modeldeveloper and prospective model user to;

• give the developer a better under-standing of the user's specific environ-ment,

• allow the user to better understand andthus have more confidence in themodel to be implemented, and

• jointly tailor the generalized simulationpackage to meet the specificrequirements of the user.

For many organizations, providing such adevelopment module is a practical alternative to amodel that is programmable by the user.

Ease of use

The simulation model must also be easy to use. Inmany organizations, the user is likely to betechnically unsophisticated. This simulation modelmust be interactive because the unsophisticateduser must be put at ease when using and supply-ing data for the model. Thus, the language shouldbe conversational, allowing for ongoing com-munication between the model and user. Thiscommunication should:

• provide continuous feedback so that auser knows that an input entry has hadsome effect,

• be consistent from one point of themodel's process to the next.



GOAL

Effective Use ofSimulation Model

CRITERIA

User Understanding

Ease of Use

Model Responsiveness

—

—

—

—

FACTORS

Modularity

Self-Documentation

User Assisted Design

Interactive

Conversational Language

Natural Language

Non-Procedural Language

"What if" Capability

Graphics

Summary Information

Figure 3. Effective Use of Simulation Models



minimize requirements for generallyunreliable human memory,

be specifically designed, in terms oflanguage and context, for that specialaudience of casual users,

sustain user orientation to the extentthat the user always knows at whatstage the simulation model is operating[26],

Communication between the user and the simula-tion model must be biased toward the user. Thatcommunication should be natural (as close aspossible to the user's language) and non-procedural (entry of instruction in nonspecifiedsequence). The objective is to allow the user tocommunicate to the simulation model in a mannerconsistent with the user's natural thinking pro-cesses. The burden then shifts to the model pro-grammer. This programmer must explain theuser's language to the computer rather than forc-ing computer convenient languages on the user.This shift in responsibility does not preclude con-struction of the model with procedural program-ming languages such as FORTRAN, BASIC,SLAM or GPSS, However, the use of suchlanguages must be transparent to the user. Theuser need not know nor care what language wasused to construct the model that communicateswith him in natural terms.

Model responsiveness

Obviously, a computer simulation model shouldbe responsive to its user. However, the charge ofthis responsiveness changes for organizationswithout technical staffs. These organizationsdon't have persons trained in both solving andtolerating inscrutable and complex computermodels. For simulation models, a "what if"capability is the essence of the models'usefulness. However, relatively unsophisticatedusers do not know that such a capability is possi-ble. The model programmer cannot be content towait on the user's request for sensitivity analysis,but must actively persuade the user to seekresults to alternative scenarios. This maynecessitate linkings to an external databasewhether or not such linking is requested by theunsophisticated user.

The same situation is frue for graphics. Theunsophisticated user may not be aware of thesystem's graphics capability and how graphicscan help in representing data. For those users, acertain amount of graphics should be auto-matically provided. The increasingly sophisticateduser should have the option to suppressautomatically generated graphics.

Finally, there is the need for summary rather thandetailed information. The content of summaryinformation is a subject needing increasedresearch efforts. Should there be medians ratherthan means? Should there be ranges rather thanstandard deviations? These ^ e among a myraidof questions which need to be answered.

SummaryThe use of computer decision modeis in general,and simulation models in particular, can providemanagers with a valuable tool for decision making.Unfortunately, the use of these models has beenlimited, especially in organizations with small or notechnical staffs. In order to expand the use ofmodels there must be greater attention todeveloping designs that are truly non-technicaland user-oriented, A combination of conversa-tional, natural, and nonprocedural communicationinterfaces seems appropriate, tn addition, themodel must be fully understood and accepted bythe user if it is to be effectively used. Researchinto media that contributes to such understandingand acceptance is essential. Other techniquessuch as modular and user-assisted design willcertainly narrow the gap between thetechnologically designed computer decisionmodel and the technologically uncertainmanagers. Finally, graphics, access to an exter-nal database, and summary statistics would beuseful enhancements once language andunderstanding characteristics have beensatisfied.

References[1 ] Annino, J,S, and Russel, C. "The Ten Most

Frequent Causes of Simulation AnalysisFailure and How to Avoid Them," Simula-


Computer Decision Modeis

tion. Volume 32, Number 6, June 1979,pp, 137-138,

[2] Ballard, B,W. and Bierman, A.W. "Program-ming in Natural Language; "NLC as a Pro-totype," Proceedings of the 1979 AnnualConference of the Association for Com-puter Machinery, New York, New York,1979, pp. 228-237,

[3] Beguin, F,M. and Probst. A.R, "InteractiveSimulation Models With User Guidance forStrategical Planning," Proceedings of theSummer Computer Simulation Conference,Society of Computer Simulation, Chicago,Illinois, July, 1977,

[4] Bekey, G,A, "Models and Reality: SomeReflections on the Art and Science ofSimulation," Simulation. Volume 29,Number 5, November 1 977, p, 161,

[5] Bowker, D. "An Interactive Aid to FinancialPlanning," Journal of the OperationalResearch Society, Volume 31 ,Number 10, October 1980, pp, 883-887.

[6] Bruggere, T.H, "Business Graphics: Mak-ing Computers Work For Management,"Proceedings IEEE Computer Software andApplications Conference. 1978, p. 744.

[7] Card, S,K., Moron, TP. and Newell, A."The Keystroke-Level Model for User Per-formance Time With Interactive Systems,"Communications of the ACM, Volume 23,Number 7, July 1980, pp, 396-410.

[8) Craig, G.D, "A Simulation System for Cor-porate Planning," Long Range Planning,Volume 13, Number 5, October 1980,pp, 43-56,

[9] Cuff, R.N. "On Casual Users," Interna-tional Journal of Man-Machine Studies,Volume 12, Number 2, February 1980,pp, 63-187,

[10] EDP Solutions. Datapro Research Corpora-tion, Delran, New Jersey, 1979,pp. E98-100, 146,

[11] Frankowski, EN, and Franta, W,R, "A Pro-cess Oriented Simulation Model Specifica-tion and Documentation Language,"Software Practice and Experience.Volume 10, Number 9, September 1980,pp, 721-742.

[12] Hurrion, R.O,, Seeker, R,J,R. "VisualInteractive Simulation; An Aid to DecisionMaking," QMEGA. Volume 6, Number 5,1978, pp. 419-426.

[13] Kaplan, S.J, "On the Difference BetweenNatural Language and High Level QueryLanguages," Proceedings of the 1978Annual Conference of the Association forComputing Machinery. New York, NewYork, 1978, pp. 27-38.

[14] Kiviat, P,J, "Requirements for an Interac-tive Modeling and Simulating System," inMulti /Access Computing. Rosenthal, P,H.and Mish, R.K, ed,, Hayden Book Co. Inc,Rochelle Park, New Jersey, 1974,pp. 53-61,

[15] Kupka. I. & Wilsing, N. ConvsrsationaiLanguages, John Wiley & Sons, New York,New York, 1980, pp. 98-107,

[16] Ledgard, H,, Whiteside, J.A,, Singer, A,and Seymour, W, "The Natural Language ofInteractive Systems," Communications ofthe ACM. Volume 23, Number i oOctober 1980, pp. 556-563,

[17] Lewis, J.M, "A Comparative Analysis ofFinancial Modeling Languages," Pro-ceedings of the 1978 Winter SimulationConferences, New York, New York, 1 978,pp. 45-53,

[18] Linger, R,C, "The Management of SoftwareEngineering Part III: Software Design Prac-tices," IBM Systems Journal, Volume 19,Number 4, 1980, pp, 432-450.

[19] Lyneis, J,M. "Achieving OrganizationalChange Through Corporate Modeling,"Proceedings of the 1977 Summer Com-puter Simulation Conference. La Jolla,California, 1977, pp. 596-601.

[20] Miller, L.A, "Natural Language Program-ming: Styles, Strategies, and Contrasts,"IBM Systems Journal. Volume 20,Number 2, 1981, pp. 184-215,

[21) Oren, T.I, and Zeigler, B,P. "Concepts forAdvanced Simulation Methodologies,"Simulation, Volume 32, Number 3,March 1979, p, 69,

[22] Sheppard, S, "Control Issues in theDevelopment of a Conversational Simula-tion Language," unpublished Ph,D. disser-tation, University of Pittsburgh, Pittsburgh,Pennsylvania, 1 977,

[23] Schneiderman, B. and Mayer, R, "Syntactic/Semantic Interactions in ProgrammerBehavior: A Model and ExperimentalResult," International Journal of Computerand Information Sciences. Volume 8,Number 3, June 1979, pp, 219-238.



[24] Wagner, G,R, "A Survey of Uses and Usersof a Contemporary Planning System,"EXECUCOM Systems Corporation, Austin.Texas, 1977,

[25] Winograd, T. "Beyond ProgrammingLanguages," Communications of theACM, Volume 22, Number 7, July 1979,pp. 391-401,

[26] Simpson, H. "A Human-Factors StyleGuide for Program Design," Byte.Volume 7, Number 4, April 1982,pp. 108-132,

About the Authors

WiUiam L. Fuerst is an Assistant Professor ofBusiness Computing Science in the College ofBusiness Administration at Texas A&M University.He received his D.B.A. from Texas Tech Univer-sity in management information systems. Hismajor areas of interest include design, implemen-tation, and use of computerized systems.

behavioral factors affecting MIS success, end-user computing, and computer simuiation. He hasconsuited with both large and small organizationsregarding a variety of computer-related matters.He is a member of AIDS and DPMA, and haspublished in MIS Quarterly and DecisionSciences, among others.

Merle P. Martin is a Lecturer in the College ofBusiness Administration at Texas A&M University.He received his MS from Stanford University instatistics. He has been Manager of TechnicalOperations for the Aiaska Court System,Associate Professor of Quantitative Systems forthe University of Alaska, Anchorage and a Lt. Col-onel in the United States Air Force. His majorareas of interest include design, impiementation,and use of computerized systems, end-user com-puting and computer simulation. He is a memberof DPMA and has published in the Journal ofSystems Management.


Effective Design And Use Of Computer Decision Models

Documents

Transcript of Effective Design And Use Of Computer Decision Models