chapter essay and research paper

DSS In the Public Sector

Design and Implementation of Decision Support Systems in the Public Sector

By: John C. Henderson Sloan School of Management, Massachusetts Institute of

Technology Cambridge, Massachusetts

By: David A. Schiiiing Facuity of Management Sciences Ohio State University

Abstract This article examines the implications of utilizing deci- sion support systems (DSS) in the public sector based on a DSS developed and implemented for a community mental health system. The DSS includes a multiple objective (goal programming) allocation model and encompasses a multiple party decision process. The experiences and insights acquired during the develop- ment and implementation of this DSS are relevant to public sector decision support in general. The impor- tance of a DSS as a process-support aid rather than a product-oriented aid (i.e., simply providing answers) and the interaction of system architecture and the chosen design strategy are key insights. In particular, the distinction between model-oriented and data- oriented DSS does not appear to be appropriate. The public sector decision maker's concern with issues of equity requires the ability to operate in a higher dimen- sional framework than the typical spreadsheet model and there is a critical need for communication support.

Keywords: Goal programming, decision support systems, public sector.

ACM Categories: H.4.2

Introduction

Developing and implementing decision aids in the public sector is a challenging task. As Lamm [14] points out, the political process tends to pro- mote those that survive or win, not those seeking truth. Often, the essential benefit of a decision aid — a valid model — is the very element that most threatens the survival of the public deci- sion maker. It is not surprising that Brill [3] notes, "Designing a solution to a public sector problem is largely an art."

Hammond [8] suggests that it may not be suffi- cient to provide decision aids unless explicit attention is given to how these aids support effective learning. Without effective learning support dysfunctional consequences are likely to result from policy-making processes. Although Hammond argues a quasi-experimen- tal approach is a necessary condition for learn- ing, he notes that the strong quasi-rational model of inquiry represented by the application of management science techniques has had positive impact on public sector decision mak- ing. For example, management science models can help to externalize multiple objectives and, when combined with the results of quasi-experi- ments, provide an enhanced learning environment.

The need to facilitate access to decision aids as well as to support individual and organizational learning is explicitly addressed in the decision support systems literature [1]. The basic design strategy for DSS begins with an analysis of the decision process and adaptively developing a tool for the user to learn about and cope with semi-structured decisions.

Experience in DSS design has also indicated the Importance of flexibility, ease of use (at least by an intermediary), and adaptability. Design methodologies such as middle-out [16] or proto- typing [12] are explicitly directed towards achieving these characteristics. These design approaches assume there will be significant user and analyst learning in terms of both the technology and the decision process. This learn- ing is enhanced (perhaps even made possible) by developing an initial system with the char- acteristics described above. As both the user and analyst move along a learning curve, the

MIS Quarterly/June 1985 157

DSS in the Public Sector

system is adapted to support their evolving infor- mation and ieaming needs.

The trend in public sector applications of management science techniques seems consis- tent with this perspective. Public sector planning models have evolved from those that focus on efficiency to those that attempt to describe and account for conflicting objectives [18]. The appii- cation of multi-objective models in areas such as fire station location [22], police patrol scheduling [20], and water resource management [15] are some recent illustrations. Recursive frameworks [4, 11] have been proposed that use a multi- objective planning model to establish system parameters and then disaggregate these solu- tions using heuristic and simulation models in order to evaluate their impact on system opera- tions. This iterative approach is quite consistent with the adaptive design concepts proposed by DSS researchers.

Research on the application of decision support systems in the public sector has emphasized the need to address both the problems of conflicting objectives as well as the need to better support the traditional data analysis efforts of the policy analyst. Hammond notes that both forms of deci- sion aids are necessary. Providing these types of decision aids in a user friendly, adaptive mode is the objective of many current research efforts.

Decision Setting This research will focus on a decision support system designed and Implemented for the Franklin County (Ohio) Mental Health and Retar- dation Board. The Board oversees forty contract agencies which provide required community- wide mental health services. The nature of the decision process for allocation decisions is critical in this environment. There must be opportunities for various constituencies, representing diverse interests, to have an influence on complex programmatic and finan- cial decisions. Unfortunately, within this realm of complexity the decision makers are often untrained. They are chosen based on the con- stituencies and values they represent, rather than on their knowledge of the problem area or

their expertise as planners or decision makers. They serve in a voluntary mode, meeting infre- quently and typically under severe time con- straints. It is little wonder that decisions often reflect the relative power of a special interest group rather than some overall set of community goals and priorities.

The Franklin County MHR Board, faced with increasing demand and an eroding resource base^ began an effort to improve the quality of their budget planning and allocation process. They identified a need to clarify and link com- munity goals to a comprehensive model for men- tal health delivery. They sought a budget pro- cess that would provide Board members with a better understanding of how specific allocations affected program level and overall community mental health system goals.

As a starting point they chose the Balanced Ser- vice System (BSS) model as the fundamental conceptualization of a mental health service system. The BSS is a model of mental dysfunc- tioning used by the Joint Commission on Accreditation of Hospitals (JCAH) to generate standards for community mental health pro- grams. In its basic form the BSS model consists of two primary service dimensions: the service function (crisis stabilization, growth, and sustenance) and the service environment (pro- tective, supportive, and natural). The function indicates the nature of the service while the environment describes where the service is pro- vided. Each of 200 possible service types are assigned to one of the cells of this two- dimensional matrix. Figure 1 depicts this matrix and includes examples of the types of services in each cell. This model satisfies requirements for a comprehensive mental health framework and also provides the basis for externalizing Board goals. As will be discussed, the goal structure addresses both specific program areas (e.g., a particular cell in the service delivery matrix) and systemwide goals (e.g., the need to balance service delivery across a range of service environments).

'The Board's allocations budget Is approximately 20 million dollars, however, projections for budget cutbacks and Inflation are significantly reducing tiiese resources whiie various need assessments indicate increasing demand for service.

158 MIS Quarterly/June 1985

DSS in the Public Sector

Crisis Stabiiization

FUNCTiON

Growth

LU

UJ

Protective

Supportive

Naturai

Sustenance Psychiatric ward of a state hospital

Twenty-four hour community emergency center

Court appointed probate screening

Private psychiatric hospital

Outpatient service at a community mental health center

Direct group counseling at the work place

Long term care in a state institution

Chronic patient deinstitutional- ization

Chronic patient living with foster family

Figure 1. The Baianced Service System Categories

The Board also recognized the need for an ade- quate decision aid. They began an effort to develop a decision support system that would: (1) provide a direct link between Board goals (as formulated using the BSS model) and allocation decisions, (2) provide a means to better under- stand the tradeoffs between goals and the impact of altering goal priorities, (3) provide the means to easily incorporate new restrictions, policies, or cost and service parameters into decisions, and (4) provide training tools for Board members. Given these needs, a DSS design and implementation effort was under- taken. The following sections describe the resulting DSS and its impact.

DSS Framework One of the basic concepts of DSS is the need for flexibility and adaptability. As many public sec- tor researchers note [3], successful public sec- tor decision aids must be able to accommodate unanticipated changes both to the structure of embedded models as well as to the nature of the user interaction. Achieving these system char- acteristics is a fundamental goal of the DSS designer. This flexibility and adaptability can be provided through a modular design. The system framework employed in this study (Figure 2) is consistent with that proposed by Sprague and Carlson [24]. It consists of three basic com- ponents; model management, data manage-

ment, and information management, and it pro- vides a user friendly interface. Each component is decoupled as much as possible and consists of a set of well-defined processing modules. This modularity minimizes the number of system interdependencles, thereby allowing most changes to be relatively localized and straight- forward. Further, various processing modules are written in a high level, analysis-oriented language (SAS). This language provides many data processing-oriented macro statements and parameterized routines which substantially reduce the time required to generate or modify particular system components, in cases where this language did not meet specific needs the module was written in Fortran or a macro com- mand language.

While initial prototyping efforts focused on the development of a mathematical model, the eventual success of the DSS depended on an effective, integrated software environment for each of the component systems. This would suggest appropriate system characteristics for DSS generators as well as give rise to questions of the validity of distinctions made in the DSS literature concerning model-oriented versus data-oriented decision support systems. To pro- vide a background for these remarks, a brief description of each component is provided.

Model Management — The model manage- ment component focuses on the generation and execution of the allocation model. A model gen-

MIS Quarterly/June 1985 159

DSS in the Public Sector

MODEL MANAGEMENT

USER

System Control

Model generation

Model execution

> DATA MAN

System Control

Transaction database

r AGEMENT

Database generation

Application database generation

> INFORMATION

V System Control

r MANAGEMENT

Information processing

Report generation

> f

Management reports

Figure 2. Decision Support System Framework

eration module translates variable definitions, system structure, and parameter estimates into an appropriate format for model execution. This module also provides a means to interface with the system transaction database. Relatively extensive changes to the model can be accom- plished by fairly simple adjustments to the model generation module.

The model execution module utilized IBM's MPS linear programming package. However, the flexibility of the model generator, combined with the capabilities of the data management component, permit the use of any appropriate linear programming software.

Finally, as with each component of the DSS, the model management component includes pro- cessing modules to interface with the host operating system and provides for interactive dialogue with the user. This aspect, termed system control, enables much of the operation of the model management component to be rela- tively transparent to the user and provides the means to integrate this component with other parts of the system.

Data Management — The purpose of the sec- ond component, data management, is to provide the foundation for a delivery system by merging the solution database with various other data-

160 MIS Quarterly/June 1985

DSS in the Public Sector

bases (e.g., variable labels, historical data trends, etc.) in order to create an integrated solution database. From this solution database, selected application databases are extracted for use by the application programs. The applica- tion databases create significant efficiency in the subsequent information processing modules. It is important to note that this compo- nent decouples the generation and execution of the model from the generation of management information. It is, in fact, the role of a data management component to isolate changes to application programs from changes to primary data sources (in this case, changes to the alloca- tion model).

The data management component aiso pro- vides the means to access and analyze data stored in the system transaction database. As will be discussed later, this capability proved necessary for the successful implementation of the DSS. A high level language (SAS) provided efficient processing of large files^ as well as the ability to quickly adapt parameter calculations for both changes in problem structure and specific data sources.

Information Management — The information processing component creates a wide range of managerial reports. To achieve adaptability and flexibility, this component consists of a number of applications programs that operate on extrac- ted application databases. This structure per- mits modifications to a particular program or report to be localized and, therefore, greatly simplifies the adaptation of the information generation process. This component uses visual representations such as value paths and bar graphs to augment traditional tabular reports. The system allows easy manipulation of both the representation form as well as the particular for- mat via a user interface environment.

The Model The complex and political nature of the alloca- tions decision highlighted the utility of a model- based decision support system. The complexity arose not only from the great variety of allocation decisions required, but also from their inter-

'The transaction database contained over 500,000 records.

relationships. These issues were addressed by formulating a linear programming resource allocation model.

The selection of an appropriate model structure was influenced by several considerations. First, the presence of lay decision makers and other nontechnical users favored a model structure which was intuitive and, therefore, easy to understand. Second, due to the prototyp- ing/evolutionary approach used in system development, the model had to be capable of extensive elaboration. Third, the chosen struc- ture should address the multiple objective nature of the decision problem, namely the com- peting Balanced Service System categories. Finally, it was important that the model help strengthen the behavioral link between the newly adopted BSS framework and the decision maker's existing perceptions of system-wide needs.

In response to these desired characteristics a goal programming model structure was selected. Goal programming has been used and tested in a wide variety of multiple objective decision situations with sophisticated users as well as novices. Such a model structure can respond well to an evolutionary development. In addition, the multiple BSS objectives could be represented in a straightforward fashion using county-wide service needs as goal levels. By directing the Board's attention towards balanc- ing these services, the behavioral link between the BSS framework and a Board member's cur- rent cognitive model could be improved.

The model formulation follows a class goal pro- gramming structure and is discussed in detail by Henderson and Schilling [10]. While the details of this model are not germane to this article, a brief overview is provided so that the DSS and the evolving model can be discussed.

The primary decision variables reflected the amount of dollars from each funding source allocated to each service type provided by each agency. There were four different sources of funds to be accounted for, resulting in over 500 variables. Besides the budget constraints, which limited total dollars available from each funding source, restrictions were specified on the percentage increase and decrease that any

MIS Quarterly/June 1985 161

DSS in ttie Public Sector

agency's budget might change. Similarly, the total county-wide funding ievel for each service was limited in the amount that it might shift. These agency and service funding restrictions were included to insure that any allocation shifts would be politically feasible. For example, defunding an entire agency or service in a single year would be extremely difficult to implement. The Board specifically chose a strategy that would spread major funding level changes over several planning periods.

Legal restrictions were incorporated which addressed the legislative and contractual stipulations of specific funding sources. For example, federal regulations require that the proportion of federal funds to public funds must be no more than three to one.

Consistent with a goal programming approach, constraints were included that measured goal deviations and created an objective function minimizing the weighted deviations from the BSS goal levels. Since none of the goal levels was attainable, given current or foreseeable funding levels, the deviations were all one- sided. The priority weightings of the deviational variables served as a tool for identifying group conflict and consensus formation as well as a mechanism by which the group could examine alternative allocation patterns.

While goal programming has seen numerous applications, it nonetheless has several poten- tial pitfalls. Of most concern in this application was the possibility of solution manipulation, as discussed by Harrald et al. [9]. In such a situa- tion, arbitrary bounds are added to the model in an attempt to force acceptable solutions. This activity often occurs when the model is too simplistic and unrealistic. This problem can be particularly troubling in a prototyping implemen- tation effort where both the DSS and model evolve from a simple, first-cut system. In order to inhibit arbitrary manipulation, all proposed structural changes were subjected to extensive discussion and debate with Board and staff members. Changes were introduced only if a consensus opinion existed that the modification was a fundamental policy elaboration. For example, during initial development two basic model improvements were made. It became apparent that the model ignored differences in

services based on client age and area of resi- dence. To rectify this inadequacy, constraints were added to insure that each client age group and geographic area received at least a mini- mum level of funding. In another instance, it was determined that some services (termed supple- mental) were required when, and only when, other basic services were purchased. Con- straints were then written to reflect this observa- tion. Both of these model changes were not attempts to contrive solutions but, in fact, represented evolutionary enhancements to the model which resulted from decision maker learning. In support of these conclusions it is worth noting that these modifications are still present in the model three years later.

The decision process

In the public sector, the key word is often 'pro- cess.' The means by which a decision is reached can often receive more attention than the deci- sion itself. In designing and implementing a DSS, issues of process become paramount. A common perception among users is that some of their decision making power may be sacri- ficed. For example, one of the by-products of a model-based DSS is that decision criteria must be made more explicit. Attention is then directed toward the mechanism by which these criteria are established and applied. This externaliza- tion often represents a major change for public sector decision makers.

The likelihood of successful implementation is increased as the magnitude of resultant change is decreased [13]. To this end, minimizing unnecessary process modifications is very desirable. In the case of Franklin County, the planning and allocation process involved group decision making throughout. There was strong commitment among Board members to a plan- ning process which utilized an interacting group to obtain a consenus. The Board members felt such an approach was both politically feasible and enhanced the opportunity for debate and compromise.

To avoid the pitfalls of interacting groups an estimate-discuss-estimate procedure was used to generate goal weights [6, 7]. This process calls for each committee member to review the

162 MIS Quarterly/June 1985

DSS in the Public Sector

results of a DSS analysis. Each member then assigns importance points (the sum of which equals one hundred) to the various goals^. The distribution and mean of the collective votes were tabulated and fed back to the group to stim- ulate discussion and promote conflict resolu- tion. Following this debate, a second allocation of importance points occurred. This vote- discuss-vote sequence has been shown to be effective in estimating parameters and for facilitating group consensus [6]. The average weights produced by the second voting were used as priority weights on the deviational variables in the goal programming model. The model was then solved to generate an allocation pattern and the vote-discuss-vote cycle was repeated.

This group process is the solution technique for the goal programming model. It is essentially a multi-party extension of a simple, iterative search technique for determining the appropri- ate weights of objectives. Its relatively unsophis- ticated structure is easily understood by non- technical decision makers and it blended easily into the existing group process. This simple for- mat provided an effective means to initiate the DSS prototyping effort.

As the implementation proceeded, the decision makers became quite comfortable with inter- preting the goal weights. The DSS allowed the decision makers to directly link changes in weights with changes in allocation. At latter stages in the process minority opinions (i.e., average weights based on a subset of the group members) were analyzed to further support group debate. Later, input from other consti- tuences (originally outside the process) was easily incorporated.

Results

This implementation represents a single data point and, hence, results are quite tentative. However, the study represents an actual DSS implementation, and its usage over a three-year period provides a significant opportunity to crit- ique DSS concepts. Two major insights

^Introductory training sessions emphasized the underiying assumption of an interval scale implicit in the averaging of these important points.

emerged from this study: (1) the critical relation- ship between DSS and the more traditional MIS functions, and (2) the characteristics of third generation DSS technology, particularly DSS technology applicable to the public sector.

Rockart and Flannery [19] found the desire for independence from the MIS function to be a major factor behind end user computing. This desire for independence is often associated with DSS. This research does not support the notion that DSS design will be independent of the MIS function. Specifically, a distinction between model-oriented DSS and data-oriented DSS does not appear appropriate. The DSS imple- mented in this study was conceived as model- oriented and initial development efforts empha- sized the modeling aspects of the system. And yet, experience demonstrated that the capability to link the model to the large transaction database was critical throughout the prototyping effort. We speculate that successful DSS appli- cations will generate requirements to link the DSS to the basic data processing systems in the organization. The DSS implementation signifi- cantly altered both the data definitions and the data flow associated with the Board's transac- tion data systems. This resulted in increased interdependencies between the DSS user and the MIS organization. The DSS implementation served as a catalyst to generate the commitment necessary to implement a data administration function. The structure of the model became the basis for redesign of the data collection activi- ties. While this served to help institutionalize the DSS and ensure reliable input data for the allocation model, it also created the need for end users to work closely with the MIS organization. As Board expectations for data quality increased, the credibility of the DSS became more sensitive to the database maintenance efforts of the MIS organization. Thus, the on-going success of the DSS became directly linked to the effectiveness of the MIS organization.

This finding is consistent with other research on public sector decision making. Hammond [8], Keen and Gambino [12], and others have noted the traditional reliance of public sector analysts on descriptive data analysis to support the policy analysis process. The DSS experience supports the need for the public sector model-based DSS

MIS Quarterly/June 1985 163

DSS in the Public Sector

to provide for descriptive data analysis as well. Again, had the system been unable to easily respond to this data-intensive analysis, the implementation effort would have suffered.

The study suggests that DSS may provide increased opportunities for innovations in the MIS function. Much like the Introduction of new products requires different management and technical practices, the design and implementa- tion of a DSS requires approaches that differ from the more traditional MIS practices. Yet, if successful, the DSS creates an ever increasing dependence between the DSS end user and the MIS function. This seems particularly true in the public sector where the use of such systems may result in precedence-setting policies.

A second major insight relates to the characteristics of third generation DSS technology, particularly as they may apply to the public sector. Future public sector model-based DSS generators must address at least three needs: easy incorporation of an equity dimen- sion, enhanced data analysis capabilities, and increased communication capabilities.

As the Franklin County implementation pro- ceeded, system modifications centered around both the ability to change the model and the ability to alter information processing and basic data management modules. In many cases, changes in the model structure focused on issues of equity. As Savas [21 ] points out, there are a variety of conflicting ways to operationalize notions of equity. Initially, the model did not explicitly operationalize equity relationships. While some relationships indirectly created solutions which were more "equitable," they were not explicitly formulated to do so. For example, legal constraints which required minimum levels of services offered by agencies may have their origin in the equity notion of equal outputs.

However, as the DSS evolved the board sought to explicitiy insure equity in the allocation of funds. For example, constraints forcing distribu- tion of funds between geographical areas were added. These efforts to ensure that small, geo- graphically isolated providers received at least a minimum allocation represented Savas' equity concept of equal inputs per unit area.

Many discussions centered around developing constraints that would reflect the Board's con- cern for equal access. The ability to generate model structure, to easily test, and eventually incorporate these structural changes was an important capability. This need to consider equity issues in the public sector resulted in a technological demand for at least a three dimen- sional model. One must be able to easily accom- modate program activity, time, and equity dimensions in the model. This suggests that cur- rent automated spreadsheet modeling lan- guages, that are essentially two-dimensional, may be inadequate for end user system develop- ment in the public sector.

Previous discussions addressed the need to link the model-based DSS to the transaction system of the organization. This linkage results in the DSS user becoming an important stakeholder with regard to procedures to define, collect, and maintain elementary data. It also suggests that third generation technology must have the capa- bility to conduct a wide range of data analysis. As previously mentioned, this type of analysis has become standard practice for most public sector policy analysts. The need is to provide a single DSS that can adequately provide both modeling and data management capabilities.

Finally, third generation DSS technology must place greater emphasis on communication capabilities. This study emphasized the need for alternative modes of presentation, i.e., graphical versus tabular. The need to incorporate a graph- ics capability in DSS is widely recognized. However, this study suggests that the com- munication needs for a public sector DSS extend beyond providing for alternative modes of presentation. Public sector analysts have significant requirements for distribution of the results of their analysis. This distribution normally takes the form of reports, memos, and/or press releases. This suggests significant benefits will be gained by linking the DSS into the automated office environment. For example, data related to the model-based DSS should be easily accessible by the word processing system within the office.

The growing research findings relating to the use of adaptive design or prototyping for DSS were strongly supported by this implementation.

164 MIS Quarterly/June 1985

DSS in the Public Sector

A prototyping design strategy similar to those discussed by Keen and Gambino [12] was used to design the DSS. While this strategy proved effective, it also created high expectations on the part of the user for easy modifications. As the DSS grew in complexity, meeting these expecta- tions became difficult. The modular design of this system, which explicitly recognized a need for model management, proved crucial to meet- ing these high expectations. The implementers were able to evolve a matrix-generation lan- guage with self-contained high-level commands and flexible user interface with little or no impact on the command structure for the data or infor- mation management components. Further, each component proved necessary to the suc- cess of the implementation. These experiences suggest that while prototyping is successful as a general strategy, structured design concepts and associated design aids are quite important.

This study also provided insight into the process used by Board members to validate the model and the DSS. This process appeared to have three distinct stages. The first stage was the acceptance of the conceptual structure for modeling a mental health system (the BSS model) and for utilizing a muti-criteria allocation model. This stage involved fairly abstract debates at the Board level concerning (1) com- parison of the BSS Model with other models of a mental health system, (2) reviewing alternative processes for obtaining information about the impact of allocation strategies, and (3) reviewing alternatives for conducting sensitivity analyses.

The second stage involved a macro-operational verification in which inputs to the model were varied and trends in output were examined to determine if the outputs of the model made intui- tive sense or could be logically accounted for. This stage resulted in structural changes to the model and helped to establish the content of several management reports.

Finally, the third stage involved validation through micro-operational sampling. This con- sisted of individuals selectively examining input data, model parameters, and then tracing out- puts at very detailed levels. Evaluations were made based on personal experience or indepen- dent data sources. For example, a Board mem- ber might ask to see the unit cost for a particular

type of childrens' service at a particular agency because he/she had been a provider in that environment. At this point, the implementation became linked to the ability to trace the origins of these parameters to the actual day-to-day transactions database. If inconsistencies were found or new formulations developed, the trans- action database had to be used to provide new input to the model. On several occasions, the transaction database was used to clarify demand characteristics or system demo- graphics that were not explicitly incorporated into the model. Had this capability been lacking, concerns about data quality would have impeded the implementation process and the DSS would not have been effective. Thus, while the model-based DSS basically operated on files extracted from a large transaction database, the ability to easily interact with the transaction database still played a major role in the imple- mentation process.

Impact on the community mental health system

It is important to mention the impact of imple- menting this DSS on the total mental health system. As noted earlier, a DSS should serve as a learning support tool that is capable of addressing both strategic and operational issues. In this study a decision to transfer the budget for mental retardation services to another community board was arrived at and justified, in part, by examining the allocation models developed for both areas. This examina- tion showed programmatic independence (e.g., no shared resources or facilities) and led to a conclusion that community level goals for these two areas were not in conflict.

Similarly, the system was used to illustrate the impact of alternative cost accounting approaches, to communicate the impact of federal fund-matching requirements, and to examine a wide range of operational issues. Its uses have evolved beyond providing direct sup- port for the allocations process. For example, extensive "what if" analysis has been per- formed in the context of contingency planning for the success or failure of a proposed tax levy.

MIS Quarterly/June 1985 165

DSS in the Public Sector

The DSS also became a focal point to revamp data collection processes, to establish new con- trols over system-wide data flows, and to create or legitimize new data requirements. As a direct result of this system implementation a com- pletely new data collection format was created and a new collection process institutionalized. This effort established new validation pro- cedures used in acquiring data as well as pro- duced a means to train providers on the BSS model of the mental health system.

The quantity of cost- and service-related data obtained from agencies was substantially increased (by nearly a factor of two) over previous years. The agencies were asked, for the first time, to indicate preferences on budget reductions, i.e., where and at what level they would place lower limits.

As might be expected, this new information management effort led to a desire for greater control over data quality and expansion of the types of data made available. Prior to the imple- mentation of the DSS, such cooperation and involvement in the collection and quality control of data were lacking. To a large extent, the DSS created a planning process that justified the effort and cost necessary to provide such data. Since these data were also used for other finan- cial and policy analysis tasks, the DSS produced a significant secondary impact on Board functions.

Finally, the process established well-defined points within the budget process where priorities were established and decisions made. This had, and will continue to have, a fundamental impact on the mechanism by which the community can influence allocations. In essence, the Board established, for the first time, direct linkages between a conceptual model of a mental health system, how such a system should function (goals), and the allocation process employed to achieve these goals.

Conclusions Generalizations cannot be made from a single data point, however, the external validity of these experiences is high in that the system was

successfully implemented and continues to be used both for allocations decisions as well as "what if" planning questions. Furthermore, these experiences appear generally consistent with the growing body of research on DSS, and thus merit the following conclusions.

First, process (the way a system or organization arrives at a decision) is critical in public sector decision making. Perhaps the most fundamen- tal conclusion of this work is the need for the management scientist to provide a process- support aid rather than a model that provides an answer. Thus, the importance of providing a range of learning and decision aids within an integrated, yet adaptive system is stressed.

Second, the emerging theory of DSS addresses a blend of design strategy, system char- acteristics, and required technological build- ing blocks. This work supports most current thinking with regard to these areas. Prototyping, as a design strategy, proved effective both in terms of defining user information needs as well as providing a mechanism to support user and analyst learning. The system characteristics of adaptability, flexibility, modularity, simplified man-machine interface, and alternative modes of presentation proved necessary to successful implementation. Thus, we find empirical support for the basic principles of DSS.

Third, model selection and formulation need careful attention. The model structure must match the problem structure, but it must also support decision maker understanding. Institu- tionalization of a DSS that uses a complex model is facilitated when the model serves both as an analytic tool and as a conceptual model. Care must also be taken to circumvent model usage traps. In a prototyping/evolutionary environment, the analyst must closely scrutinize model modifications in order to avoid the temp- tation of solution manipulation and to insure model integrity. Failure to do so can invalidate the entire DSS while still appearing (at least to the untrained eye) to perform correctly.

Fourth, the distinction between model-oriented DSS and data-oriented DSS, and the notion of independence for DSS users, does not appear appropriate given these experiences. This sys- tem was conceived as a model-oriented DSS and Initial efforts were directed toward the

166 MIS Quarterly/June 1985

DSS in the Public Sector

modeling aspect of the system. And yet exper- ience demonstrated that the capability to link the model to a large transaction database was very important. DSS users cannot remove them- selves from the need to examine, verify, and communicate fundamental data. To be effective the DSS has to provide the means to access this elementary data in a timely fashion. We specu- late that this will become a feature of most successful model-oriented DSS generators. That is, to be successful there will be pressure to link the DSS to the basic data processing of the organization. From a management perspective, this will result in a need to better coordinate DSS and MIS design efforts.

Fifth, the importance of addressing equity issues is stressed. The notion of equity in public policy is, in itself, a major research issue. This study suggests future DSS technology must enable the user to incorporate equity dimen- sions as well as activity and time dimensions. This indicates that public sector applications require a DSS generator that extends beyond the two dimensional framework currently repre- sented by financially oriented DSS generators. The fact that DSS generators in the public sector must be at least three-dimensional increases demands for flexibility and sophisticated forms of presentation.

Finally, the benefits of DSS are difficult to assess a priori. The benefits of DSS will include such issues as support of organization change, sup- port of individual learning and improved man- agement of the technological growth of the organization. This work indicated that significant system-wide impact occurred and should be explicitly recognized in the evaluation of the success or failure of the DSS effort. The DSS affected fundamental areas such as learning, organizational development, data processing, and decision process framing. It influenced user learning by providing the means to investigate the complexity of the problem in a systematic manner. It affected organizational development by unfreezing positions and attitudes concern- ing both the mission of the Board as well as the structure of the allocation process. It affected data processing by providing the felt need and political support necessary to revamp data col- lection procedures and to increase the quality and integrity of their database. Finally, it pro-

vided the means to frame the decision as one of tradeoffs between goals rather than increases or decreases in specific budget line items. This not only changed the allocation decision process but helped to institutionalize a goal-oriented planning process.

References

[1] Alavi, M. and Henderson, J.C. "An Evolu- tionary Strategy for Implementing a Deci- sion Support S y s t e m , " Management Science, Volume 27, Number 1 1 , November 1981, pp. 1309-1323.

[2] Alter, S.L. Decision Support Systems: Cur- rent Practice and Continuing Challenges, Addison-Wesley, Reading, Massachu- setts, 1980.

[3] Brill, E.D., Jr. "The Use of Optimization Models in Public Sector P l a n n i n g , " Management Science, Volume 25, Number 5, May 1979, pp. 413-422.

[4] Cohon, J. and Marks, D. "Multiobjective Screening Models and Water Resources Investments," Water Resources Research, Volume 9, Number 4, 1973, p. 826.

[5] Delbecq, A. and Van De Ven, A.H. "A Group Process Model for Problem Identification and Program Planning," Journal of Applied Behavioral Science, Volume 7, Number 4, July-August, 1971, p. 466.

[6] Delbecq, A.L., Van De Ven, A.H. and Gustaf- son, D.H. Group Techniques for Program Planning: A Guide to Nominal Group and Delphi Process, Scott Foresman and Co., Glenview, Illinois, 1975.

[7] Gustafson, D.H., Shukia, R.M., Delbecq, A.L. and Walster, G.W. "A Comparative Study of Differences in Subjective Likelihood Estimates Made by Individuals, Interacting Groups, Delphi Groups and Nominal Groups," Organizational Behavior and Human Performana, Volume 9, Number 2, April 1973, pp. 280-291.

[8] Hammond, K.R. "Toward Increasing Com- petence of Thought in Public Policy Forma- t i o n , " Judgment and Decision in Public Policy Formation, Kenneth R. Hammond, (ed.) AAAS Selected Symposium, Boulder, Col- orado, 1980.

MIS Quarterly/June 1985 167

DSS in the Pubiic Sector

[9] Harrald, J., Leotta, J., Wallace, W.A. and Wendell, R.E. "A Note on the Limitations of Goal Programming as Observed in Resource Allocation for Marine Environmen- tal Protection," Navai Research Logistics Quarteriy, Volume 25, Number 4,1978, pp. 733-739.

[10] Henderson,J.C. and Schilling, D.A. "Design and Implementation of Decision Support Systems in the Public Sector," College of Administrative Science Working Paper Series, WPS 81-77, Ohio State University, 1981.

[11] Henderson, J . C , Showalter, M.A. and Kra- jewski, L.H. Jr. "An Integrated Approach for Manpower Planning in the Service Sec- tor, Qmega, Volume 10, Number 1, January 1982.

[12] Keen, P.G.W. and Gambino, T.J. "Building A Decision Support System: The Mythical Man-Month Revisited," in Buiiding Decision Support Systems, J.F. Bennett (ed.) Addison-Wesley, Reading, Massachusetts, 1982.

[13] Keen, P.G.W. and Scott Morton, M.S. Deci- sion Support Systems: An Organizational Perspective, Addison-Wesley, Reading, Massachusetts, 1979.

[14] Lamm, R.A. "The Environment and Public Policy," Judgment and Decision in Pubiic Poiicy Formation, K.R. Hammond, (ed.) AAAS Selected Symposium, Boulder, Col- orado, 1980.

[15] Major, D. and Lenton, R. Multiobjective Multimodei Riverbasin Pianning: The MiT- Argentina Project, Prentice-Hall, Englewood Cliffs, New Jersey, 1978.

[16] Ness, D. "Interactive Systems: Theories of Design," Joint Wharton/ONR Conference Interactive Information and DSS, University of Pennsylvania, Philadelphia, Penn- sylvania, November 1975.

[17] Pressman, J. and Wildavsky, A. Implemen- tation: How Great Expections in Washington Are Dashed in Oakland, University of Califor- nia Press, Oakland, California, 1973.

[18] ReVelle, C.S., Bigma, D., Schilling, D.A., Cohon, J.A., and Church, R. "Facility Loca- tion Analysis: A Review of Context-Free and EMS Models," Health Services Research, Summer 1977, pp. 129-147.

[19] Rockart, J.F. and Flannery, L.S. "The

Management of End-User Computing," Pro- ceedings of The Second International Con- ference on Information Systems, E. McLean (ed.), Boston, Massachusetts, December 7-9, 1981, pp 351-363.

[20] Saladin, B. A Methodology for the Allocation of Police Patrol Vehicies, Unpublished Doc- toral Dissertation, Ohio State University, Col- umbus, Ohio, 1980.

[21] Savas, E.S. "On Equity in Providing Public Service," Management Science, Volume 24, Number 8, April 1978, pp. 800-808.

[22] Schilling, D.A., Elzing, D.J., Cohon, J.A. and ReVelle, C S . "Design and Analysis of Loca- tion and Relocation Alternatives in a Fire Protection System," Proceedings of IEEE Modeling and Simulation Conference, Pitts- burgh, Pennsylvania, May 1980.

[23] Schilling, D.A., ReVelle, C S . and Cohon, J.L. "An Approach to the Display and Analysis of Multiobjective Problems," Socio and Economic Planning Science, Volume 17, Number 2, 1983, pp. 57-63.

[24] Sprague, R.H. and Carlson, E.D. Buiiding Ef- fective Decision Support Systems, Prentice- Hall, Englewood Cliffs, New Jersey, 1982.

About the Authors

John C. Henderson is an Associate Professor of Management Science at the Sloan School of Management, and a member of the Center for Information Systems Research, M.I. T. Prior to join- ing the faculty of M.I. T, he served on the faculty at Wharton (University of Pennsylvania), Ohio State, andFiorida State universities. He has wide consulting experience and has served in the pubiic sector as Staff Director for the Joint Select Com- mittee on Electronic Data Processing, Fiorida Legisiature. Dr. Henderson has taught and pub- iished extensiveiy in the fieids of decision support systems, MIS design and impiementation, and the strategic impacts of information technoiogy. He serves on the editoriai boards of Management Science, MIS Ouarterly, Systems, Objectives and Solutions, and Office: Technology and People.

David A. Schilling is an Associate Professor of Management Science at Ohio State University. He received a Ph.D. in systems anaiysis/opera- tions research from the John Hopkins University.

168 MIS Quarterly/June 1985

DSS in the Public Sector

His research interests include multicriteria deci- sion maktng, model-based decision support systems, group decision making, and facility location analysis. He has published articles in Management Science, Decision Sciences, European Journal of Operations Research, and Omega, among others.

MIS Quarterly/June 1985 169