Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

An experimental investigation of the impact of computer based decision aids on the process of preferential… Todd, Peter A. 1988

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-UBC_1988_A1 T62.pdf [ 17.53MB ]
Metadata
JSON: 831-1.0098241.json
JSON-LD: 831-1.0098241-ld.json
RDF/XML (Pretty): 831-1.0098241-rdf.xml
RDF/JSON: 831-1.0098241-rdf.json
Turtle: 831-1.0098241-turtle.txt
N-Triples: 831-1.0098241-rdf-ntriples.txt
Original Record: 831-1.0098241-source.json
Full Text
831-1.0098241-fulltext.txt
Citation
831-1.0098241.ris

Full Text

AN EXPERIMENTAL INVESTIGATION OF THE IMPACT OF COMPUTER BASED DECISION AIDS ON THE PROCESS OF PREFERENTIAL CHOICE by PETER A. TODD B. COM. 1983, Mc G i l l U n i v e r s i t y A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY i n THE FACULTY OF GRADUATE STUDIES Faculty of Commerce and Business Administration We accept t h i s thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA May 1988 0 Peter A. Todd, 1988 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of The University of British Columbia 1956 Main Mall Vancouver, Canada V6T 1Y3 DE-6(3/81) Abstract This research examines the impact of Decision Support Systems (DSS) on the decision making process for p r e f e r e n t i a l choice tasks. The p o t e n t i a l impact of DSS on the d e c i s i o n process i s evaluated i n terms of how the system a l t e r s the d e c i s i o n maker's cognitive load. Competing hypotheses are developed based on the possible objectives of the decision maker with respect to d e c i s i o n e f f o r t and decision q u a l i t y . One l i n e of reasoning assumes that the DSS w i l l be used i n such a way as to maximise decision q u a l i t y . The other asserts that the use of the DSS w i l l be geared towards e f f o r t conservation. These hypotheses about the impact of the DSS on the decision process are tested i n three experiments. The three studies employed concurrent verbal protocols to capture data about the decision process. In experiment 1 subjects were placed i n e i t h e r and aided or unaided decision s e t t i n g and given problems of e i t h e r f i v e or ten a l t e r n a t i v e s from which to make a choice. The r e s u l t s showed that decision strategy changed as a r e s u l t s of the use of the d e c i s i o n aid. In general, subjects behaved as e f f o r t minimisers. There were no s i g n i f i c a n t e f f e c t s r e l a t e d to the amount of information processing. Experiment 2 was s i m i l a r to experiment 1 except that subjects were given problems with e i t h e r ten or twenty a l t e r n a t i v e s . The r e s u l t s were consistent with, though stronger than those of experiment 1. Almost a l l aided group subjects used Elimination by aspects strategy while the unaided group used a Conjunctive strategy. This i s consistent with the notion of e f f o r t minimisation. There were no s i g n i f i c a n t differences i n the amount of information processing Experiment 3 was designed to test whether the r e s u l t s i n experiments 1 and 2 were a due to the tendency of decision makers to minimise e f f o r t or because i i t h e a i d was n o t p o w e r f u l enough t o i n d u c e a d d i t i v e p r o c e s s i n g . I n t h i s s t u d y t h e DSS was a l t e r e d t o b o t h i n c r e a s e t h e s u p p o r t f o r t h e a d d i t i v e d i f f e r e n c e s t r a t e g y and r e d u c e s u p p o r t f o r t h e e l i m i n a t i o n by a s p e c t s a p p r o a c h . The r e s u l t s o f e x p e r i m e n t 3 show t h a t d e c i s i o n makers t e n d t o adapt t h e i r s t r a t e g y t o t h e t y p e o f d e c i s i o n a i d s a v a i l a b l e . There i s e v i d e n c e t h a t i f a d d i t i v e s t r a t e g i e s a r e made s u f f i c i e n t l y l e s s e f f o r t f u l t o use t h e y w i l l be employed. S i m i l a r l y , when t h e degree o f e f f o r t t o f o l l o w a p a r t i c u l a r e l i m i n a t i o n s t r a t e g y i s m a n i p u l a t e d d e c i s i o n makers t e n d t o adapt i n such a way as t o m i n i m i s e e f f o r t . O v e r a l l t h e r e s u l t s o f t h e t h r e e e x p e r i m e n t s a r e c o n s i s t e n t i n d e m o n s t r a t i n g t h e a d a p t i v i t y o f d e c i s i o n makers t o t h e t y p e s o f s u p p o r t t o o l s a v a i l a b l e t o them. T h i s a d a p t i v i t y c e n t r e s a r o u n d t h e m i n i m i s a t i o n o f d e c i s i o n e f f o r t . I t appears t h a t d e c i s i o n makers a r e h i g h l y c o n s c i o u s o f t h e e f f o r t r e q u i r e d t o make d e c i s i o n s and work i n such a way as t o m i n i m i s e t h a t e x p e n d i t u r e . When f a c e d w i t h t h e use a d e c i s i o n a i d t h e y appear t o c a l i b r a t e t h e i r own d e c i s i o n e f f o r t t o t h a t p r o v i d e d by t h e d e c i s i o n a i d . There i s some e v i d e n c e t h a t s u f f i c i e n t changes i n t h e r e l a t i v e e f f o r t r e q u i r e d t o use v a r i o u s s t r a t e g i e s c a n l e a d d e c i s i o n makers t o f o l l o w more e f f o r t f u l a pproaches t h a n t h e y m ight o t h e r w i s e c o n s i d e r . The p r e c i s e n a t u r e o f t h i s e f f o r t - a c c u r a c y r e l a t i o n s h i p needs t o be s t u d i e d more c l o s e l y . The b a s i c c o n t r i b u t i o n o f t h e d i s s e r t a t i o n has been t o p r o v i d e a f o r m a l a p p r o a c h f o r t h e s t u d y o f DSS, b a s e d on c o n c e p t s drawn f r o m b e h a v i o u r a l d e c i s i o n t h e o r y and i n f o r m a t i o n p r o c e s s i n g p s y c h o l o g y . T h i s work a l s o has i m p l i c a t i o n s f o r b e h a v i o u r a l d e c i s i o n t h e o r i s t s , consumer r e s e a r c h e r s and p r a c t i c a l i m p l i c a t i o n s f o r t h e development o f DSS i n p r e f e r e n t i a l c h o i c e s e t t i n g s . i i i CHAPTER 1 - INTRODUCTION 1 CHAPTER 2 - THINKING IS HARD 11 2.0 Introduction . 11 2.1 Information processing and r a t i o n a l d ecision making 11 2.2 An overview of empirical work on information load 13 CHAPTER 3- THE COST-BENEFIT FRAMEWORK 21 3.0 Introduction 21 3.1 The perceptual view 21 3.2 The cognitive view 23 3.2.1 E f f o r t minimisation 24 3.2.2 Accuracy maximisation 25 3.2.3 Accuracy maximisation subject to an e f f o r t constraint . . . . 26 3.2.4 E f f o r t minimisation subject to an accuracy constraint . . . . 28 3.3 E f f o r t or accuracy? 31 3.3.1 Conceptual models 32 3.3.2 Simulation models 33 3.3.3 Empirical studies 34 3.4 V a l i d i t y of the cost-benefit framework 37 3.5 Reconciling the cognitive and perceptual views 38 CHAPTER 4 - A BEHAVIOURAL DECISION THEORY APPROACH TO DSS DEVELOPMENT . 41 4.0 Introduction 41 4.1 Models of p r e f e r e n t i a l choice 44 4.1.1 The additive-compensatory model 46 4.1.1.1 Formal d e s c r i p t i o n of strategy 47 4.1.1.2 Supporting the strategy 50 4.1.2 The additive-difference model 51 4.1.2.1 Formal d e s c r i p t i o n of strategy 51 4.1.2.2 Supporting the strategy 53 4.1.3 The conjunctive model 54 4.1.3.1 Description of the strategy 55 4.1.3.2 Supporting the strategy 56 4.1.4 The Elimination by aspects model 58 4.1.4.1 Formal d e s c r i p t i o n of strategy 59 4.1.4.2 Supporting the strategy 60 4.2 Summary of proposed support mechanisms 61 4.2.1 Computational support 62 4.2.2 Data storage 69 4.2.3 Information display 70 4.3 Summary 72 CHAPTER 5- PROPOSITIONS AND HYPOTHESES 73 5.0 Introduction 73 5.1 Propositions 73 5.2 An overview of dependent and independent v a r i a b l e s 79 5.3 Hypothesis development . 81 5.3.1 E f f o r t Minimisation Hypotheses 83 5.3.2 Accuracy maximisation hypotheses 93 i v CHAPTER 6- EXPERIMENTAL DESIGN & PROCEDURES 103 6.0 Introduction 103 6.1 Experimental design 103 6.1.1 Experiment 1 103 6.1.2 Experiment 2 . 105 6.1.3 Testing f or screen e f f e c t s 105 6 . 2 The task environment 106 6.3 The subject population 108 6.4 The development of the decision a i d 109 6.5 Experimental procedures 112 6.6 Data analysis plan 115 6.6.1 Basic models 115 6.6.2 Dependent va r i a b l e s 116 6.6.2.1 Computer log data 117 6.6.2.2 Verbal protocol data 118 6.6.2.3 Questionnaire data 127 CHAPTER 7: RESULTS- EXPERIMENTS 1 and 2 130 7.0 Introduction 130 7.1 Protocol r e l i a b i l i t y 130 7.2 Experiment 1- re s u l t s 133 7.2.1 Evaluation of assumptions of s t a t i s t i c a l tests 134 7.2.2 Presentation of findings 135 7.2.3 S a t i s f a c t i o n with the system 142 7.2.4 Command usage 145 7.2.5 Problem siz e 148 7.3 Discussion of experiment 1 148 7.4 Results experiment 2 156 7.4.1 Assumptions 158 7.4.2 Presentation of findings 159 7.4.3 S a t i s f a c t i o n with the system 163 7.4.4 Command usage 164 7.4.5 Problem siz e r e s u l t s 167 7.5 Experiment 2- discussion 168 7.6 Screen e f f e c t s - r e s u l t s 171 7.6.1 S t a t i s t i c a l tests 172 7.7 Screen e f f e c t s - discussion 175 7.8 Pooled r e s u l t s 175 7.8.1 Hypothesis tests 176 7.8.2 Command usage 180 7.9 Pooled data- discussion 181 7.10 Concluding comments 187 CHAPTER 8- EXPERIMENT 3:A 189 8.0 Introduction 189 8.1 Background and ra t i o n a l e for study 191 8.2 Experimental design 194 8.2.1 High EBA, low AD (conditional drop, no compare) 198 8.2.2 High EBA, high AD (conditional drop and compare) . . 199 8.2.3 Low EBA, low AD (no conditi o n a l drop, no compare) 201 8.2.4 Low EBA, high AD (no conditi o n a l drop, compare) 202 8.3 Main e f f e c t s hypotheses 204 v 8.3.1 Additive difference support hypotheses 205 8.4 The compare function 214 8.5 The task environment 218 8.6 The subject population 219 8.7 Experimental procedures . 219 8.8 Presentation of findings 221 8.8.1 The s t a t i s t i c a l model 222 8.8.2 Additive difference r e s u l t s 223 Command Usage 227 Questionnaire r e s u l t s 229 8.8.4 Elimination by aspects r e s u l t s 232 Command usage 235 Questionnaire r e s u l t s 236 8.9 Discussion 236 8.9.1 Additive difference support 239 Impact of the AD support treatment 239 Impact of compare on memory 244 Limited use of compare 246 Summary 248 8.9.2 EBA support 249 Strategy changes 249 Use of condit i o n a l drop 250 The exposure e f f e c t 251 Summary 254 8.9.3 Combined e f f e c t of EBA and AD support 255 8.10 Comparison of r e s u l t s 259 8.11 Integration of the findings 263 8.12 Concluding comments 271 CHAPTER 9 ANECDOTAL EVIDENCE 272 9.0 Introduction 272 9.1 Anecdotal protocol data 273 9.1.2 Memory functions and f o r g e t t i n g 274 9.1.2 Information load 279 9.1.3 Attitudes towards the system 284 9.1.4 Task d i f f i c u l t y 289 9.1.5 Summary of anecdotal data 292 9.2 A hybrid decision strategy 293 9.3 Conclusions 298 CHAPTER 10- CONCLUSIONS 300 10.0 Introduction 300 10.1 Overview of r e s u l t s • 300 10.1.1 Process changes 300 10.1.2 Changes i n e f f o r t and information use 302 10.1.3 Impact on memory and attention 305 10.1.4 A behavioural decision theory i n t e r p r e t a t i o n 306 10.2 Limitations of the research . 307 10.2.1 Incentives 307 10.2.2 External v a l i d i t y 309 10.2.3 S t a t i s t i c a l power 311 v i 10.3 Contributions of the research 3 1 2 10.3.1 Contributions to DSS research 3 1 3 10.3.2 Contributions to behavioural d e c i s i o n theory 315 10.3.3 P r a c t i c a l contributions 3 1 6 10.4 Directions f o r future research 3 1 7 10.5 Concluding comments 3 2 * TABLES 3 2 4 APPENDICIES 4 1 3 v i i L i s t of Tables Table 4.1 A p p l i c a b i l i t y of support functions 325 Table 5.1 Possible impacts of a i d on e f f i c i e n c y and effectiveness . . . . 328 Table 5.2 Dependent and independent v a r i a b l e s 329 Table 6.1 Experimental design 332 Table 6.3 Screen e f f e c t s design 334 Table 7.1 Experimental design- experiment 1 336 Table 7.2 Test f o r homogeneity of variance (El*) 337 Table 7.3 Proportion of information used (values are i n %) (El) 338 Table 7.4 Absolute information use (El) . 339 Table 7.5 Proportion of al t e r n a t i v e s examined i n d e t a i l (values i n percent) (El) 340 Table 7.6 Absolute number of alt e r n a t i v e s examined i n d e t a i l (max. values 5 and 10) (El) 341 Table 7.7 Variance i n a t t r i b u t e usage (El) 342 Table 7.8 Mean number of at t r i b u t e s examined per a l t e r n a t i v e (El) . . . . 343 Table 7.9 Access pattern 344 Table 7.10 Strategy assignments (El) 345 Table 7.11 Tot a l steps (El) 346 Table 7.12 Tot a l time 347 Table 7.13 S a t i s f a c t i o n with System 348 Table 7.14 Ease of use 349 Table 7.15 Command usefulness (El) 350 Table 7.16 Command use 351 Table 7.18 Problem siz e e f f e c t 352 Table 7.19 Design- experiment 2 353 Table 7.20 Tests for homogeneity of variance (E2*) 354 Table 7.21 Proportion of information used 355 Table 7.22 Absolute amount of information used 356 Table 7.23 Number of alt e r n a t i v e s examined i n d e t a i l 357 Table 7.24 Variance i n att r i b u t e s examined per a l t e r n a t i v e 358 Table 7.25 Mean a t t r i b u t e usage per a l t e r n a t i v e (E2) 359 Table 7.26 Access pattern 360 Table 7.27 Strategy assignments (E2) 361 Table 7.28 Total steps 362 Table 7.29 Tot a l time 363 Table 7.30 S a t i s f a c t i o n with System (E2) 364 Table 7.31 Command use (E2) 365 Table 7.32 Problem s i z e e f f e c t 366 Table 7.33 Amount of information used 367 Table 7.34 Number of al t e r n a t i v e s examined i n d e t a i l * 367 Table 7.35 Variance i n at t r i b u t e s examined 367 Table 7.36 Mean number of at t r i b u t e s examined 367 Table 7.37 Access pattern 367 Table 7.38 Strategy assignment 368 Table 7.39 Total steps 369 Table 7.40 Tot a l time (seconds) 369 Table 7.41 Absolute information use (pooled*) 370 Table 7.42 Mean at t r i b u t e s examined per a l t e r n a t i v e (pooled) 371 Table 7.43 Access pattern (-100 indicates pure a t t r i b u t e , +100 pure v i i i a l t e r n a t i v e processing) (pooled) 372 Table 7.44 Strategy assignments 373 Table 7.45 Strategy- a i d by size 374 Table 7.46 T o t a l time 375 Table 7.47 Tot a l steps 376 Table 7.48 Command use 377 Table 7.49 Experiment 1- summary of r e s u l t s 378 Table 7.50 Experiment 2- summary of r e s u l t s 379 Table 7.51 Pooled data- summary of r e s u l t s 380 Table 8.1 Experimental design 382 Table 8.2 Tests f o r homogeneity of variance 384 Table 8.3 Information usage 385 Table 8.4 Att r i b u t e s analysed i n d e t a i l 386 Table 8.5 Variance i n a t t r i b u t e s searched per a l t e r n a t i v e 387 Table 8.6 Mean number of att r i b u t e s searched per a l t e r n a t i v e 388 Table 8.7 Access pattern (+100 to -100) 389 Table 8.8 Strategy- Additive difference group 390 Table 8.9 Strategy- EBA group 391 Table 8.10 Proportion of dependent evaluations 392 Table 8.11 Proportion of dependent evaluations- l a s t h a l f 393 Table 8.12 Tot a l time (seconds) to complete the task . 394 Table 8.13 Tot a l steps taken to complete the task 395 Table 8.14 Command use- AD support treatment 396 Table 8.15 R e l i a b i l i t y of questionnaire constructs 397 Table 8.16 Questionnaire r e s u l t s - AD support group 398 Table 8.17 Command use- EBA support treatment 399 Table 8.18 Questionnaire r e s u l t s - EBA support group 400 Table 8.19 Summary of r e s u l t s - Additive difference 401 Table 8.20 Summary of r e s u l t s : EBA support 402 Table 8.21 Estimate of subject and system steps taken to complete the task403 ix Acknowle dgement s Many p e o p l e a s s i s t e d i n the c o m p l e t i o n o f t h i s work. I n v a r i o u s ways the f a c u l t y and g r a d u a t e s t u d e n t s i n t h e b u s i n e s s s c h o o l have h e l p e d t o shape my i d e a s and s h a r p e n t h e f o c u s o f t h e r e s e a r c h . C h r i s Wagner was r e s p o n s i b l e f o r t h e c o d i n g o f t h e system u s e d as the b a s i s f o r t h e s t u d i e s r e p o r t e d h e r e . J a s b i r S i n g h h e l p e d w i t h d a t a c o l l e c t i o n and c o d i n g o f t h e p r o t o c o l s . Each c h e e r f u l l y and k i n d l y t o o k t i m e from t h e i r own p r o j e c t s t o a s s i s t w i t h mine. Barb Weeks h e l p e d t o t r a n s c r i b e the p r o t o c o l s , an onerous d u t y a t t h e b e s t o f t i m e s . Nancy S h e l l made a v a l i a n t a t t e m p t a t i m p r o v i n g the r e a d a b i l i t y o f the f i n a l p r o d u c t . To a l l t h e s e p e o p l e and o t h e r s who a s s i s t e d i n v a r i o u s ways thank you. My committee members each c o n t r i b u t e d i n v a r i o u s ways t o making the d i s s e r t a t i o n b e t t e r and I hope t o making me a b e t t e r r e s e a r c h e r . J o h n C l a x t o n and Don Wherung b o t h f o r c e d me t o c h a l l e n g e my a a s s u m p t i o n s and q u e s t i o n a pproaches t o r e s e a r c h . T h i s has h e l p e d me t o b r o a d e n my p e r s p e c t i v e as a r e s e a r c h e r . A l D e x t e r p u t r e s e a r c h and d i s s e r t a t i o n s i n t o p e r s p e c t i v e , h e l p i n g me t o remember t h a t t h e r e i s a b i g g e r p i c t u r e t o c o n s i d e r . I z a k Benbasat was my s u p e r v i s o r b u t much more t h a n t h a t he was a mentor. I z a k r e p r e s e n t s a model o f how t o do r e s e a r c h t h a t anyone c o u l d do w e l l t o emulate. H i s c o n t i n u a l i n v e s t m e n t o f t i m e and e f f o r t i n t o my development i s a d e b t t h a t w i l l n o t be e a s i l y r e p a i d . Thank you. F i n a l l y , C o n n i e , f o r s t a n d i n g by me t h r o u g h t h e e n t i r e e f f o r t , y o u r u n f a i l i n g s u p p o r t and b e l i e f i n me makes t h i s a l l w o r t h w h i l e . x CHAPTER 1 - INTRODUCTION The focus of t h i s d i s s e r t a t i o n i s on measuring the impact of decision support systems (DSS) on the process of decision making. Decision support systems are computer based tools which augment the c a p a b i l i t i e s of decision makers and a s s i s t them i n solving semi-structured d e c i s i o n problems. There i s no s t r i c t d e f i n i t i o n of what constitutes a DSS; however, there i s some consensus on many of the common c h a r a c t e r i s t i c s of a DSS. They support rather than replace d e c i s i o n makers, and focus on decision effectiveness rather than e f f i c i e n c y (Keen and Scott Morton, 1978). By the term semi-structured problems we mean those problems f o r which there i s no known approach to a r r i v i n g at an optimal s o l u t i o n . The normative l i t e r a t u r e on DSS emphasises understanding and supporting the process of decision making i n order to enhance de c i s i o n making effectiveness (Keen and Scott Morton, 1978). Much of the empirical work i n DSS has ignored the process of decision making (Todd and Benbasat, 1987; Benbasat and Nault, 1987). With the exception of a few i n i t i a l studies (for example, Scott Morton, 1971) most research i n the DSS area has not considered the impact of DSS on decision processes. This research focusses c l o s e l y on t h i s issue. I t examines two questions: 1) How can DSS be designed to support d e c i s i o n makers? 2) How does the use of DSS impact the process of decision making? The f i r s t question r e l a t e s to system design, the second i s one of evaluation. 1 In the approximately 20 year h i s t o r y of DSS there has been an acute shortage of empirical work which examines and evaluates the impact of DSS (Hurt et a l . , 1986). At the same time the evaluation l i t e r a t u r e i s contradictory (see Benbasat and Nault, 1987 for an overview). I t has been argued that many of these contradictory r e s u l t s are due to an improper perception held by researchers about the p o t e n t i a l impact of a DSS. Todd and Benbasat (1987) have argued that there i s a need for process t r a c i n g oriented studies evaluating the possible benefits of DSS, rather than the more common input-output studies. The reasoning behind the f a i l u r e of input-output studies i s that i n r i c h , i l l - s t r u c t u r e d , problem contexts there w i l l t y p i c a l l y be much cognitive a c t i v i t y occurring between the presentation of the decision problem and i t s ultimate s o l u t i o n using a DSS. Input-output oriented research w i l l not capture these e f f e c t s , many of which might negate each other when captured as a series of outcome v a r i a b l e s . Process oriented studies have a better chance of capturing these e f f e c t s . Also, for semi-structured priblems i t may well be easier to assess the q u a l i t y of the process used by a decision maker i n evaluating a problem rather than the q u a l i t y of the outcome of a decision (Todd and Benbasat, 1987). This research examines decision a i d s 1 to support m u l t i - a t t r i b u t e , multi-a l t e r n a t i v e p r e f e r e n t i a l choice problems (Keeny and R a i f f a , 1976 provide a d e s c r i p t i o n of the c h a r a c t e r i s t i c s of such problems). These are problems where a decision maker must choose one of a number of a l t e r n a t i v e s each of which can be described by a common set of a t t r i b u t e s . Most consumer purchase 1 Throughout t h i s d i s s e r t a t i o n the term decision a i d or t o o l w i l l be used interchangeably with the term DSS. 2 d e c i s i o n s c a n be c h a r a c t e r i s e d i n t h i s manner. A l s o , b u s i n e s s d e c i s i o n s s u c h as s i t e s e l e c t i o n , s t o c k p u r c h a s e o r c a p i t a l i n v e s t m e n t p r o b l e m s ca n be c h a r a c t e r i s e d as p r e f e r e n t i a l c h o i c e p r o b l e m s . T y p i c a l l y t h e r e a r e no c o r r e c t answers o r c h o i c e s i n t h a t t h e c h o i c e i s dependent upon i n d i v i d u a l p r e f e r e n c e s . There a r e a number o f r e a s o n s f o r u s i n g t h i s p a r t i c u l a r c l a s s o f p r o b l e m s . There i s a l a r g e body o f l i t e r a t u r e r e l a t i n g t o p r e f e r e n t i a l c h o i c e problems much o f w h i c h f o c u s s e s on i n d i v i d u a l d e c i s i o n b e h a v i o u r . The s t r a t e g i e s t h a t d e c i s i o n makers employ i n t h e s e s e t t i n g s a r e w e l l documented and w e l l u n d e r s t o o d . T h i s f a c i l i t a t e s t h e e v a l u a t i o n o f d e c i s i o n maker b e h a v i o u r t o d e t e r m i n e a p p r o p r i a t e ways o f p r o v i d i n g s u p p o r t mechanisms f o r d e c i s i o n makers. The e x i s t i n g e m p i r i c a l work w i l l a l s o h e l p us t o u n d e r s t a n d and p r e d i c t how d e c i s i o n makers m i g h t make use o f d e c i s i o n a i d s d e s i g n e d t o s u p p o r t t h e s e t y p e s o f p r o b l e m s . P r e f e r e n t i a l c h o i c e problems a r e a g e n e r i c t y p e o f p r o b l e m w h i c h i s d e a l t w i t h by most p e o p l e on a d a i l y b a s i s ( e . g . , consumer p u r c h a s e d e c i s i o n s , c h o o s i n g between a l t e r n a t i v e p r o j e c t s o r i n v e s t m e n t s i n a b u s i n e s s e n v i r o n m e n t o r d e c i d i n g where t o go f o r d i n n e r ) . I t i s a c l a s s o f p r o b l e m w h i c h has been v i r t u a l l y i g n o r e d i n t h e DSS a r e a ( l i k e l y because o f i t s q u a l i t a t i v e r a t h e r t h a n q u a n t i t a t i v e o r i e n t a t i o n ) . A t t h e same t i m e , s t u d i e s o f t h i s t y p e o f p r o b l e m i n m a r k e t i n g and b e h a v i o u r a l d e c i s i o n t h e o r y have r e c o g n i s e d i m p l i c a t i o n s f o r t h e d e s i g n o f DSS b u t have n o t p u r s u e d r e s e a r c h i n t h e DSS a r e a (Payne, 1976; P a i n t o n and G e n t r y , 1985). Thus, t h e p r e f e r e n t i a l c h o i c e a r e a seems t o be one w h i c h i s r i p e f o r t h e development o f DSS. T h i s i s n o t t o s a y t h a t t h e r e a r e no systems w h i c h c u r r e n t l y s u p p o r t t h i s t y p e o f d e c i s i o n making; however, t h o s e w h i c h do e x i s t o f t e n employ v e r y 3 formal modelling techniques. Humphreys and Wisudha (1987) provide an overview of many systems which include formal models for the support of p r e f e r e n t i a l choice problems. These approaches may be d i f f e r e n t i a t e d from DSS i n the same way that t r a d i t i o n a l management science models are t y p i c a l l y separated from the body of systems r e f e r r e d to as DSS. They tend to impose a structure or approach to problem so l v i n g on the d e c i s i o n maker which i s considered "good" i n some normative sense. The DSS approach d i f f e r s i n that i t attempts to support d e c i s i o n behaviour and emphasises f l e x i b l e approaches to problem so l v i n g (Sprauge and Carlson, 1982). In addition, as w i l l be discussed i n chapter 4, some of these systems experience the same problems of user resistance that are often reported i n the management science l i t e r a t u r e . The research conducted here i s also rooted i n the notion that i n order to understand d e c i s i o n maker behaviour, p a r t i c u l a r l y with respect to the use of computer based decision support to o l s , we must develop i n s i g h t and understanding into what objectives d e c i s i o n makers hold with respect to d e c i s i o n making. T r a d i t i o n a l l y i n DSS research i t has been assumed that de c i s i o n makers wish to improve the q u a l i t y of t h e i r decisions. There i s considerable evidence i n the behavioural d e c i s i o n theory l i t e r a t u r e that t h i s may not be the case (see for example, Payne 1982). This research examines the assumption i m p l i c i t i n most DSS research, following on the advice of Weick (1983) who urged IS researchers to challenge the assumptions i m p l i c i t i n t h e i r work. For purposes of t h i s research DSS i s viewed as a mechanism which reduces cognitive load. By considering DSS i n terms of i t s impact on cognitive costs and then examining how decision makers might evaluate the cost and benefits of using a DSS, i t i s possible to make predictions of how a DSS w i l l a f f e c t 4 d e c i s i o n strategy. This follows the perspective, c u r r e n t l y i n vogue i n much of the behavioural d e c i s i o n theory l i t e r a t u r e , that d e c i s i o n makers s e l e c t i v e l y trade o f f e f f o r t and accuracy i n choosing a strategy f o r a given decision problem. In short, t h i s implies that the d e c i s i o n maker evaluates strategies both i n terms of how much e f f o r t i s required to employ the given strategy, and what the b e n e f i t of using that strategy may be. In making such assessments the d e c i s i o n maker may place d i f f e r e n t emphasis (or weight) on e f f o r t , r e l a t i v e to accuracy. These d i f f e r e n t perspectives that the d e c i s i o n maker may adopt w i l l lead to d i f f e r e n t ways of employing decision support t o o l s , the primary function of which i s to reduce the e f f o r t required to perform c e r t a i n types of analysis. A review of the l i t e r a t u r e r e l a t i n g to d e c i s i o n maker strategy s e l e c t i o n i s provided i n chapter 3 to c l a r i f y these issues. This view of DSS, which discusses the issue of cognitive costs and b e n e f i t s , was f i r s t advocated by Keen (1979) who discussed these issues i n the context of the "marginal economics of e f f o r t . " In that sense the empirical work reported here can be viewed as a t e s t of Keen's o r i g i n a l p roposition. The development of DSS i n t h i s research i s also viewed from a cognitive cost perspective. I t i s assumed that d e c i s i o n makers have a s t r i c t l y l i m i t e d supply of cognitive resources and that such l i m i t a t i o n s r e s t r i c t the approaches a d e c i s i o n maker might take to a problem. We summarise these l i m i t a t i o n s f a c i n g the d e c i s i o n maker i n the notion "thinking i s hard." This idea i s developed more f u l l y i n chapter 2. DSS development i s based on the evaluation of various behavioural s t r a t e g i e s of decision making i n a p r e f e r e n t i a l choice environment. The s t r a t e g i e s are decomposed into a s e r i e s of elementary information processes. Based upon these strategy decompositions a series of d e c i s i o n support tools 5 are proposed for the various s t r a t e g i e s . In t h i s way a DSS i s developed which d i r e c t l y supports the generally observed approaches taken by d e c i s i o n makers i n s o l v i n g p r e f e r e n t i a l choice problems. Each of the functions b u i l t into the DSS assumes one or more elementary cognitive operations from the decision maker. This i s done since i t i s assumed that i n order f o r a DSS to be useful i t must reduce the cognitive cost associated with information processing. When costs are reduced changes i n decision strategy may r e s u l t . In essence, systems designed along t h i s p r i n c i p l e may serve to increase the bounds of r a t i o n a l i t y (Taylor, 1975). This approach to DSS development i s h e a v i l y rooted i n the human information processing approach to understanding d e c i s i o n behaviour that was i n i t i a l l y formalised by Newell and Simon (1972). The system development reported here draws heavily on the ideas of Newell and Simon with respect to the decomposition and d e s c r i p t i o n of d e c i s i o n s t r a t e g i e s as s e r i e s of elementary information processes. This approach i s also grounded i n more recent work by Johnson (1979) and others who have advocated the decomposition of d e c i s i o n s t r a t e g i e s as a means f o r studying the r e l a t i o n s h i p between e f f o r t and accuracy i n d e c i s i o n making (see also Johnson and Payne, 1985). To the best of my knowledge t h i s work represents the f i r s t time that such an approach has been taken to the development of a DSS. The DSS developed i s employed i n a s e r i e s of three experiments to assess the impact of the system upon the process of d e c i s i o n making. The empirical studies use process t r a c i n g techniques, i n p a r t i c u l a r concurrent verbal protocols, to assess the impact of the d e c i s i o n aids on strategy. The f i r s t two experiments contrast the behaviour of aided and unaided d e c i s i o n makers for tasks of varying s i z e . Problems of 5, 10 and 20 a l t e r n a t i v e s are 6 c o n s i d e r e d . Each s u b j e c t s o l v e s a p r e f e r e n t i a l c h o i c e p r o b l e m ( i n v o l v i n g a p a r t m e n t s e l e c t i o n ) i n one o f t h e t h r e e p r o b l e m s i z e s e t t i n g s , e i t h e r w i t h o r w i t h o u t t h e use o f t h e d e c i s i o n a i d . W h i l e t h e y a r e s o l v i n g t h e p r o b l e m c o n c u r r e n t v e r b a l p r o t o c o l s a r e c o l l e c t e d f r o m w h i c h i n f o r m a t i o n usage and s t r a t e g y measures a r e d e r i v e d . These measures t e s t s p e c i f i c h y p o t h e s e s w h i c h were d e v e l o p e d about t h e b e h a v i o u r o f t h e d e c i s i o n makers i n u s i n g t h e system. There a r e two s e t s o f c o m p e t i n g h y p o t h e s e s : one b a s e d upon t h e a s s u m p t i o n t h a t t h e d e c i s i o n maker i s p r i m a r i l y c o n c e r n e d w i t h t h e amount o f e f f o r t t h a t i s r e q u i r e d t o make a d e c i s i o n and t h e o t h e r b a s e d upon th e a s s u m p t i o n t h a t t h e d e c i s i o n maker h o l d s d e c i s i o n q u a l i t y as a p r i m a r y c o n c e r n . These h y p o t h e s e s a r e summed up i n t h e f o l l o w i n g two p r o p o s i t i o n s w h i c h r e f l e c t the g e n e r a l manner i n w h i c h a DSS may be used: P i The use of the decision a i d to support the search process w i l l lead the decision maker to work more e f f e c t i v e l y , using more exhaustive strategies which consider more information, emphasise additive information evaluation and show less use of f i l t e r i n g and elimination strategies compared to unaided problem solvers. P i ' The use of the decision a i d to support the search process w i l l lead decision makers to work more e f f i c i e n t l y , using the least e f f o r t strategy which w i l l r e s u l t i n an acceptable solution. The r e s u l t s o f e x p e r i m e n t s 1 and 2 show t h a t d e c i s i o n s t r a t e g y changes as the r e s u l t o f use o f t h e DSS and t h a t t h e s e changes appear t o r e f l e c t a c o n c e r n w i t h e f f o r t o v e r a c c u r a c y . Such a r e s u l t i s c o n s i s t e n t w i t h much o f t h e 7 l i t e r a t u r e developed i n the behavioural decision theory area. Experiment 3 tests and extends some of the findings of E l and E2. In t h i s study a l l subjects had access to a d e c i s i o n aid; however, d i f f e r e n t versions of the a i d provide greater or l e s s e r support for the use of p a r t i c u l a r s t r a t e g i e s . I t tests the extent to which d e c i s i o n makers are adaptive to changes i n the degree of support provided for a given d e c i s i o n approach. For example, i f one strategy i s more hig h l y supported and hence made easier to use and another i s not supported, are the d e c i s i o n makers more i n c l i n e d to adopt the supported strategy, even though i n an unaided environment they may have a p r e d i s p o s i t i o n towards the unsupported strategy? The r e s u l t s of t h i s experiment show that decision makers do adapt to the degree of support provided for the various s t r a t e g i e s . The research conducted has p r a c t i c a l implications i n that i t provides a new perspective on decision support which focusses on a c t i v i t i e s , such as information s t r u c t u r i n g , rather than s p e c i f i c tasks. This notion has been advocated previously by Huber (1982) for the development of Group DSS. Development of such systems would be of i n t e r e s t to those i n the consumer research area, where l i t t l e i s known about how to apply computer based techniques to problems i n consumer choice (Painton and Gentry, 1985). In addition, t h i s work provides a rigourous method for decomposing decision processes as an input to DSS design. F i n a l l y , there are numerous examples of p r e f e r e n t i a l choice made i n organisations on a d a i l y b a s i s . For example, investment decisions, s i t e l o c a t i o n and c r e d i t granting decisions can a l l be framed as p r e f e r e n t i a l choice problems, as can v i r t u a l l y any purchase decision. Thus t h i s research has implications for marketing as well as those i n t e r e s t e d i n DSS and could 8 l e a d t o p r e s c r i p t i o n s f o r t h e d e s i g n o f DSS w h i c h w o u l d be a p p l i c a b l e t o a wi d e range o f pr o b l e m s w h i c h c u r r e n t l y r e c e i v e l i t t l e computer s u p p o r t . The d i s s e r t a t i o n i s o r g a n i s e d as f o l l o w s . C h a p t e r 2 w i l l a s s e s s the g e n e r a l l i t e r a t u r e w h i c h e s t a b l i s h e s t h e v a l i d i t y o f t h e i n i t i a l a s s e r t i o n t h a t " t h i n k i n g i s h a r d . " C h a p t e r 3 w i l l argue t h a t t h e d e c i s i o n making p r o c e s s i s managed by e m p l o y i n g a c o s t - b e n e f i t a n a l y s i s . T h i s w i l l be done by l o o k i n g a t some g e n e r a l works w h i c h make t h i s a s s e r t i o n and a l s o a t the e m p i r i c a l b ase f o u n d i n t h e p r e f e r e n t i a l c h o i c e l i t e r a t u r e . B ased upon t h i s s p e c i f i c e v a l u a t i o n o f t h e l i t e r a t u r e , we w i l l make some g e n e r a l arguments about d e c i s i o n making o b j e c t i v e s w h i c h f o c u s on t h e d e s i r e t o b a l a n c e e f f o r t and a c c u r a c y i n d e c i s i o n making. T o g e t h e r , t h e s e two c h a p t e r s t o g e t h e r r e p r e s e n t an o v e r v i e w o f t h e b e h a v i o u r a l d e c i s i o n making l i t e r a t u r e w h i c h i s r e l e v a n t t o t h i s r e s e a r c h . C h a p t e r 4 w i l l d i s c u s s t h e development o f DSS b a s e d upon a c a r e f u l a n a l y s i s o f a c t u a l b e h a v i o u r a l s t r a t e g i e s i n d e c i s i o n making. T h i s w i l l be a c c o m p l i s h e d by decomposing d e c i s i o n s t r a t e g i e s i n t o t h e i r c o n s t i t u e n t e l e m e n t a r y i n f o r m a t i o n p r o c e s s e s . G i v e n t h e s e d e c i s i o n p r o c e s s e s t h e s t r u c t u r e o f v a r i o u s DSS f e a t u r e s , t o s u p p o r t p r e f e r e n t i a l c h o i c e p r o b l e m s , c a n be s p e c i f i e d . C h a p t e r 5 w i l l p r o p o s e two d e t a i l e d s e t s o f h y p o t h e s e s w h i c h a r e i n t e n d e d t o p r e d i c t a d e c i s i o n maker's b e h a v i o u r w h i l e u s i n g t h e DSS. These h y p o t h e s e s a r e b a s e d upon t h e arguments i n c h a p t e r 3 t h a t d e c i s i o n makers a c t e i t h e r as e f f o r t m i n i m i s e r s o r a c c u r a c y m a x i m i s e r s . They a l s o t a k e i n t o a c c o u n t the d e c i s i o n making c o n t e x t w h i c h i s e s t a b l i s h e d b y t h e DSS d e v e l o p e d f r o m the i d e a s o f c h a p t e r 4. C h a p t e r 6 b e g i n s t h e e m p i r i c a l s e c t i o n o f t h e d i s s e r t a t i o n . I t p r e s e n t s , 9 i n d e t a i l , plans for two experiments which t e s t the impact of the use of a DSS designed to support m u l t i - a t t r i b u t e p r e f e r e n t i a l choice decisions on the process of d e c i s i o n making. The dependent and independent v a r i a b l e s are discussed i n d e t a i l . An o v e r a l l discussion of the experimental paradigm i s also provided. Chapter 7 presents the r e s u l t s of these experiments and discusses t h e i r implications. Results are given i n d i v i d u a l l y for each of the two experiments, the differences between the experiments are assessed and a pooling of the data i s also analysed. Chapter 8 presents an a d d i t i o n a l experiment designed to answer a s p e c i f i c question based upon the major findings of experiments 1 and 2. In p a r t i c u l a r , i t examines how decision makers respond to differences i n the type of support tools a v a i l a b l e . Chapter 9 provides an overview of anecdotal data c o l l e c t e d i n the three experiments. Evidence which supports the i n t e r p r e t a t i o n of the s t a t i s t i c a l data i s provided. A d d i t i o n a l findings r e l a t i n g to memory load, system s a t i s f a c t i o n and a l t e r n a t i v e decision strategies are also discussed. Chapter 10 presents an o v e r a l l summary of the d i s s e r t a t i o n assessing i t s contribution, points out the l i m i t a t i o n s of the work, and discusses possible areas of extension. 10 CHAPTER 2 - THINKING IS HARD 2.0 Introduction The purpose of t h i s chapter i s to provide an overview of the various l i t e r a t u r e showing the r e l a t i o n s h i p between d e c i s i o n making behaviour, information load and other task c h a r a c t e r i s t i c s . The review i s not meant to be exhaustive, but rather i s i l l u s t r a t i v e of the wide range of areas and approaches used i n the study of decision behaviour. Much of t h i s diverse work has come to the common conclusion that thinking, e s p e c i a l l y with respect to problem solving, i s a highly s t r a i n inducing a c t i v i t y . I t w i l l also show, to some extent, the ways i n which de c i s i o n makers appear to adapt themselves to cognitive s t r a i n . 2.1 Information processing and r a t i o n a l d e c i s i o n making Simon (1957) f i r s t formalized the concepts of bounded r a t i o n a l i t y and " s a t i s f i c i n g " behaviour. He asserted that due to the l i m i t e d information processing capacity of the i n d i v i d u a l i t i s impossible to make t r u l y r a t i o n a l ( i . e . , u t i l i t y maximizing)decisions i n a l l but the most s i m p l i s t i c environments. Taylor (1975) has argued that increasing an i n d i v i d u a l ' s processing capacity would expand the bounds of r a t i o n a l d e c i s i o n making. The c a p a c i t i e s of the human information processing system were summarized by Newell and Simon (1971) as: -a r e l a t i v e l y slow s e r i a l processor ( m i l l i second speed) -an extremely l i m i t e d short term memory (5-9 chunks) - v i r t u a l l y i n f i n i t e long term memory with r e l a t i v e l y f a s t , though 11 f a l l i b l e , access on the order of milliseconds, but slow storage (5-10 seconds f o r a u n i t of information). These basic l i m i t a t i o n s serve to constrain the problem s o l v i n g and decision making behaviour of i n d i v i d u a l s . Many researchers note i n p a r t i c u l a r the l i m i t a t i o n of short term memory capacity as a serious processing bottleneck ( M i l l e r , 1956). The l i t e r a t u r e reviewed below a l l tends to point to th i s c e n t r a l conclusion: d e c i s i o n makers are l i m i t e d information processors who must develop mechanisms f or coping with information load. Since these propositions have been put forward, many d e t a i l e d empirical studies have documented p r e c i s e l y the manner i n which i n d i v i d u a l processing i s l i m i t e d or d e f i c i e n t . For example, people have been found to co n s i s t e n t l y v i o l a t e the fundamental axioms of u t i l i t y theory (see Sl o v i c et a l . (1977) for a review). These axioms define the basis f o r t r u l y r a t i o n a l d e c i s i o n making. I f people were t r u l y r a t i o n a l , such v i o l a t i o n s of normative p r i n c i p l e s would not occur. However, one of the i m p l i c i t assumptions of u t i l i t y or value maximization models i s that information processing i s a c o s t l e s s a c t i v i t y . This i s c l e a r l y not the case. Decision makers are faced with an abundance of information at any given time. This i s e s p e c i a l l y true i n a managerial environment (Ackoff, 1967; Mintzberg, 1973). T y p i c a l l y managers have no problem getting information but rather have a great deal of d i f f i c u l t y f i n d i n g relevant information (Ackoff, 1967) . In addi t i o n there i s a s t r i c t l y l i m i t e d opportunity to focus a t t e n t i o n on any p a r t i c u l a r problem f o r any length of time. Given an environment that i s characterized by b r e v i t y , fragmentation and time pressure (Mintzberg, 1973) a dec i s i o n maker would require unlimited processing capacity i n order to make optimal decisions. 12 2.2 An overview of empirical work on information load Research i n the f i e l d of attention that studied i n d i v i d u a l processing c a p a c i t i e s has determined that a t t e n t i o n i s b a s i c a l l y a l i m i t e d resource (Kahneman, 1973) . There i s some question as to whether the amount of cognitive e f f o r t a v a i l a b l e for d i s t r i b u t i o n i s drawn from a general resource pool or from s p e c i a l i z e d processing modules ( A l l p o r t , 1978; Navon, 1985). However, the general conclusion that the a b i l i t y to process i s r e s t r i c t e d to a l e v e l lower than that required for the t y p i c a l d e c i s i o n problem i s not i n dispute. A stream of research by Schroder et a l . , (1967) demonstrated that increases i n environmental complexity, beyond a threshold l i m i t , l e d to a d e t e r i o r a t i o n i n the amount of information processed. Broadbent (1957) has noted that as decision makers are presented with more information they tend to r e s t r i c t the range of cues they consider s a l i e n t to the problem. Up to a c e r t a i n point such f i l t e r i n g can be attention d i r e c t i n g causing the decision maker to focus on important information; however, again, beyond a threshold such behaviour w i l l r e s u l t i n the omission of relevant information from processing. Johnson (1983) has noted that the behaviour of expert decision makers r e f l e c t s a tendency to overweigh i n d i v i d u a l cues which contain abnormal or o u t l y i n g data values while neglecting other relevant information. S i m i l a r l y , Kahneman and Tversky (1982) note that i n d i v i d u a l s tend to ignore base rate information when making p r o b a b i l i t y assessments, instead using h e u r i s t i c s such as a v a i l a b i l i t y and representativeness. Kahneman and Tversky i n showing such r e s u l t s have a somewhat d i f f e r e n t i n t e r p r e t a t i o n of them than that which w i l l be emphasised i n t h i s research. The i n t e r p r e t a t i o n embodied i n t h e i r Prospect Theory w i l l be discussed i n more d e t a i l below (Kahneman and 13 Tversky, 1979). In the mathematical modelling domain i t has also been noted that i n d i v i d u a l s are very s e l e c t i v e i n t h e i r use and weighting of cues. Given, for example, a problem i n f i n a n c i a l analysis with information describing the f i n a n c i a l p o s i t i o n of a company along 10-15 dimensions an analyst w i l l make a p r e d i c t i o n about the company which can be r e l i a b l y estimated with 2-4 cues (Mears and F i r t h , 1985). Similar r e s u l t s can be seen i n other domains as well (Dawes, 1979). In addition, there i s considerable v a r i a b i l i t y between experts and i n general they have very l i t t l e s e l f - i n s i g h t into t h e i r d e c i s i o n process (S l o v i c , 1972). This has also been noted i n V i t a l a r i (1986) with respect to systems analysts. Simple mathematical models can often be shown to outperform the d e c i s i o n maker who i s being modelled (Dawes, 1972). Another area of i n v e s t i g a t i o n which points to the l i m i t e d information processing capacity of i n d i v i d u a l s i s information load studies i n marketing. Jacoby (1974) was one of the f i r s t to point out that too much information had dysfunctional e f f e c t s on consumers. Increasing the information load leads to f i l t e r i n g which may cause relevant information to be ignored. From a DSS perspective one of the most i n t e r e s t i n g findings of t h i s research i s that increasing the number of a l t e r n a t i v e s leads to a decline i n the amount of information processed more ra p i d l y , than does an increase i n the number of a t t r i b u t e s to be considered (Malhotra et a l . , 1982; Olshavky, 1979). One study (Malhotra, 1982) indicated that performance declines were measurable when 10 or more a l t e r n a t i v e s were provided. A s i m i l a r decline i n performance due to a t t r i b u t e s requires that 15 or more be presented. Such a f i n d i n g has a c e r t a i n i n t u i t i v e appeal since the a d d i t i o n of an a l t e r n a t i v e requires an a d d i t i o n a l i t e r a t i o n through an i n d i v i d u a l ' s choice 14 model (assuming the d e c i s i o n maker does not ignore the a l t e r n a t i v e e n t i r e l y ) . Evaluating a d d i t i o n a l a t t r i b u t e s , however, requires modifications to the mental model used i n analysis for e s t a b l i s h i n g weights and c r i t e r i a relevance. However, adding a single a t t r i b u t e l i k e l y r e s u l t s i n only a minor change to the choice model. This w i l l be e s p e c i a l l y true i f the new information provided i s e i t h e r perceived to be h i g h l y c o r r e l a t e d with the e x i s t i n g a t t r i b u t e s or to have l i t t l e relevance to the task. This i s r e i n f o r c e d by the f a c t that, on average, decision makers tend to use only a small proportion of the information cues a v a i l a b l e to them (Dawes, 1979). The f a c t that a l t e r n a t i v e evaluation i s a binding c o n s t r a i n t bodes well f o r the p o t e n t i a l of developing d e c i s i o n aids to support these a c t i v i t i e s . I t e r a t i o n s of an automated or semi-automated choice model to deal with an a d d i t i o n a l set of a l t e r n a t i v e s should be r e l a t i v e l y c o s t l e s s f o r the decision maker, whereas changes to an underlying model, which supports the decision maker's problem so l v i n g strategy, would s t i l l require considerable e f f o r t and input by the i n d i v i d u a l . To summarise t h i s l i n e of research, there does appear to be a consensus that consumers can be overloaded by information sets considerably smaller than those presented i n t y p i c a l r e a l world choice problems. I t has also been found that even simple d e c i s i o n aids, such as s o r t i n g the presentation of p r i c e information, can a f f e c t a consumer's choice process (Russo, 1977). In t h i s regard the focus on developing decision aids to support p r e f e r e n t i a l choice problems appears to be u s e f u l . Decision maker responses to information load can adversely a f f e c t decision q u a l i t y (Wright, 1974). Information overload may be manifested e i t h e r i n dysfunctional behaviour, such as errors i n computations or comparisons 15 (Malhotra, 1984), or i n a reduced quantity of information processed i n order to reduce s t r a i n (Jacoby, 1984). In e i t h e r case the end r e s u l t may be a reduction i n d e c i s i o n q u a l i t y . Also, there i s l i t t l e reason to suspect that these processing d i f f i c u l t i e s , which manifest themselves c l e a r l y i n consumer problems, would not also occur i n the managerial d e c i s i o n environment. The c h a r a c t e r i s t i c s of the processor i n both cases are e s s e n t i a l l y the same. The environment faced by a manager i s l i k e l y r i c h e r i n information and the pressure to make correct decisions greater. In a d d i t i o n much managerial d e c i s i o n making i s often characterised by time pressure. Given these added pressures t h e i r a b i l i t i e s may be further impeded. The f i n a l area to be discussed which shows evidence of l i m i t a t i o n s i n information processing capacity i s the study of general m u l t i - a t t r i b u t e , m u l t i - a l t e r n a t i v e d e c i s i o n problems. A study by Payne (1976) t y p i f i e s t h i s type of work. B r i e f l y , i n three experiments subjects were presented with information about a number of hypothetical apartments described along a number of a t t r i b u t e s (such as rent, s i z e and brightness) and were asked to s e l e c t the one apartment they would most prefer for themselves. The p r i n c i p a l research objective i n the Payne study was to determine how search s t r a t e g i e s changed as a function of task complexity. In t h i s case, task complexity was operationalised as the si z e of the problem space 2 ( i . e . , the number of a t t r i b u t e s and a l t e r n a t i v e s faced i n the choice problem). Also 2 I t should be noted that i n t h i s case, and throughout the d i s s e r t a t i o n , the term problem space i s employed simply to r e f e r to the s i z e of a problem facing the d e c i s i o n maker i n terms of i t s number of a t t r i b u t e s and a l t e r n a t i v e s . This i s a somewhat loose i n t e r p r e t a t i o n of Newell and Simon's term "problem space," which i s used to indicate an e n t i r e area of cognitive space i n which a d e c i s i o n maker searches for a s o l u t i o n to a problem (Newell and Simon, 1972). C l e a r l y there i s a strong r e l a t i o n s h i p between the separate uses of the term, but they should not be viewed as s t r i c t l y equivalent. 16 of i n t e r e s t was the question of how much information was processed (or accessed) as a function of size of the problem space. The findings indicated that the percentage of a v a i l a b l e information considered declined both as the number of a t t r i b u t e s and the number of al t e r n a t i v e s increased. In addition, f o r small information sets a constant amount of information was searched f o r each a l t e r n a t i v e . For larger a l t e r n a t i v e sets, a v a r i a b l e amount of information was accessed f o r each a l t e r n a t i v e . The i n i t i a l study was r e p l i c a t e d with some minor modifications to the experimental design and y i e l d e d e s s e n t i a l l y the same r e s u l t s . Decision makers tend to use additive strategies to analyse small problems and el i m i n a t i o n strategies f o r larger problems. These r e s u l t s have also been r e p l i c a t e d i n other task domains. Olshavsky (1979) extended Payne's work by examining varying l e v e l s of information under two d i f f e r e n t types of choice s i t u a t i o n . One, which i s describable by "many, t e c h n i c a l l y simple, dichotomous a t t r i b u t e s " and another, with "many t e c h n i c a l l y complex, multichotomous valued a t t r i b u t e s . " Findings reported c l e a r l y indicate that the strategy employed i s a function of task complexity. As the s i z e of the problem space increases subjects move from single-stage s t r a t e g i e s which e x p l o i t v i r t u a l l y a l l a v a i l a b l e information to multi-stage s t r a t e g i e s which emphasize f i l t e r i n g i n the e a r l y p o r t i o n and a d e t a i l e d analysis only f o r the reduced search space. The key determinant of information search i n t h i s study appears to be the number of a l t e r n a t i v e s searched rather than the number of a t t r i b u t e s , a f i n d i n g consistent with those i n the consumer behaviour l i t e r a t u r e c i t e d previously. Olshavsky also found that f o r larger search spaces the use of s i m p l i f y i n g h e u r i s t i c s generally l e d to le s s time per u n i t of information examined. In general, t h i s study 17 confirms the findings of Payne (1976) and extends them to problems of varying task complexity. Biggs et a l . , (1985) examined the e f f e c t of changing both s t r u c t u r a l task v a r i a b l e s (e.g., task size) and contextual v a r i a b l e s (e.g., s i m i l a r i t y between a l t e r n a t i v e s ) on the d e c i s i o n strategy adopted by bank loan o f f i c e r s making c r e d i t granting decisions. The contextual manipulation introduces a new aspect to complexity. The more s i m i l a r two a l t e r n a t i v e s are the more processing i s l i k e l y to be required f o r d i s c r i m i n a t i o n . This work i s p a r t i c u l a r l y germane i n that i t involves business decisions being made i n a r e a l i s t i c context by actual loan o f f i c e r s and thus has high external v a l i d i t y . Results with respect to problem s i z e i n Biggs et a l . (1985) were consistent with the studies c i t e d previously. In terms of s i m i l a r i t y , i t was found that a greater proportion of the a v a i l a b l e information was searched as the a l t e r n a t i v e s became more s i m i l a r . This i s consistent with the notion that d i f f e r e n t i a t i n g between objects becomes more d i f f i c u l t ( i . e . , requires greater cognitive e f f o r t ) when differences are small. The common conclusion which can be drawn from t h i s l i n e of research i s that the information processing strategy adopted i n p r e f e r e n t i a l choice tasks v a r i e s with the s i z e of the problem space being considered (size i s often used i n these studies as a surrogate for complexity). As task s i z e increases l e s s information i s processed, more f i l t e r i n g and elimination type s t r a t e g i e s are adopted. I t i s also important to note that these strategies w i l l be invoked when problems are s t i l l r e l a t i v e l y small (e.g., s i x a l t e r n a t i v e s by four a t t r i b u t e s ) . In r e a l i s t i c d e c i s i o n settings, where much more information i s t y p i c a l l y a v a i l a b l e , d e c i s i o n makers would be even more s e r i o u s l y constrained and the tendency towards the use of elimination strategies and s i m p l i f y i n g h e u r i s t i c s 18 w o u l d be r e e n f o r c e d . On t h e whole t h e r e i s a c e n t r a l c o n c l u s i o n t h a t emerges f r o m t h e many s t u d i e s c o n d u c t e d i n a v a r i e t y o f domains and f o r many d i f f e r e n t p u r p o s e s : the f a c t that "thinking i s hard." When f a c e d w i t h s i t u a t i o n s w h i c h demand s i g n i f i c a n t r e s o u r c e s f o r c o m p l e t e e n u m e r a t i o n , i n d i v i d u a l s r e s o r t t o s t r a t e g i e s w h i c h f i l t e r i n f o r m a t i o n and d r a s t i c a l l y r e d u c e t h e amount o f work r e q u i r e d t o p r o d u c e a s o l u t i o n . T h i s i s a c l e a r i n d i c a t i o n t h a t d e c i s i o n makers e i t h e r by d e s i g n o r from n e c e s s i t y a r e h i g h l y c o g n i z a n t o f t h e amount o f e f f o r t t h e y expend on p r o b l e m s o l v i n g . The f i n d i n g s f r o m t h e s e s t u d i e s r a i s e t h e i s s u e o f t h e n a t u r e o f the mechanism w h i c h c a u s e s t h e s e s t r a i n r e d u c i n g a c t i v i t i e s t o t a k e p l a c e . A number o f p r o p o s i t i o n s have been p u t f o r t h s u c h as p e r c e p t u a l p r o c e s s i n g l i m i t s and t h e c o s t o f t h i n k i n g (Payne, 1982). The e x p l a n a t i o n w h i c h w i l l be d e a l t w i t h i n some d e t a i l h e r e i s t h e c o s t - b e n e f i t a p p r o a c h w h i c h argues g e n e r a l l y t h a t d e c i s i o n s a r e made about p r o b l e m - s o l v i n g s t r a t e g i e s and p r o c e d u r e s b a s e d upon th e m a r g i n a l r e t u r n t o be e x p e c t e d f r o m c o n t i n u e d s e a r c h . I n g e n e r a l t h e r e i s t h o u g h t t o be a t r a d e o f f between t h e amount o f e f f o r t r e q u i r e d t o implement a s t r a t e g y and t h e a c c u r a c y o f t h e c h o i c e made. One o f t h e p r i n c i p a l q u e s t i o n s t h a t we w i l l d e a l w i t h i n t h e n e x t c h a p t e r i s how d e c i s i o n makers make t h i s c o s t b e n e f i t assessment. Do t h e y v a l u e e f f o r t o r d e c i s i o n q u a l i t y more h i g h l y ? I n o t h e r words i s t h e d e c i s i o n maker more i n t e r e s t e d i n making e f f e c t i v e d e c i s i o n s o r e f f i c i e n t d e c i s i o n s ? T h i s q u e s t i o n i s o f p a r t i c u l a r i n t e r e s t i n t h e d e s i g n and development o f DSS s i n c e the p r i n c i p l e a s s u m p t i o n w h i c h u n d e r l i e s much o f t h e DSS l i t e r a t u r e i s t h a t d e c i s i o n makers a r e c o n c e r n e d w i t h e f f e c t i v e n e s s and w i l l use a d e c i s i o n a i d t o i n c r e a s e t h e e f f e c t i v e n e s s o f t h e i r d e c i s i o n making (Keen and S c o t t M o r t o n , 19 1978) . On the other hand many of the claims r e l a t i n g to the u t i l i t y of DSS r e f e r more often to t h e i r contribution to e f f i c i e n c y i n d e c i s i o n making (Keen, 1981). I f we consider the spreadsheet, one of the most commonly employed de c i s i o n aiding t o o l s , i t should become cl e a r that i t s primary advantage i s that i t reduces the computational load f a c i n g d e c i s i o n makers. In t h i s sense such tools are time savers. Reduction i n computational e f f o r t , r e s u l t i n g i n a time savings to the decision maker, may be t r a n s l a t e d into further effort, being invested i n problem solving. This may then lead to more e f f e c t i v e , well informed d e c i s i o n making. On the other hand these time savings may be used f o r other a c t i v i t i e s i n which case the d i r e c t b e n e f i t from the d e c i s i o n a i d would not be apparent. The question which must then be addressed i s how i s t h i s savings i n time and e f f o r t u t i l i s e d by the d e c i s i o n maker? Does the de c i s i o n maker have an o v e r a l l o r i e n t a t i o n towards increasing the q u a l i t y of decisions or i s there a desire to s i m p l i f y the d e c i s i o n process i n order to minimise the e f f o r t associated with d e c i s i o n making. In order to p r e d i c t how a d e c i s i o n maker w i l l make use of a computer based d e c i s i o n a i d we must have some understanding of the answers to the questions r a i s e d above. In chapter 3 empirical evidence i s reviewed which helps us to b e t t e r understand these issues. 20 CHAPTER 3- THE COST-BENEFIT FRAMEWORK 3.0 Introduction Normative models of r a t i o n a l d e c i s i o n making behaviour, based on notions of u t i l i t y or value maximization, have been found to have les s than adequate power as d e s c r i p t i v e theories of choice behaviour. Recently there have been attempts to remodel the concepts of economic man to f i t into the framework of man as a l i m i t e d capacity information processor. In general, the thrust of t h i s work has been to develop more robust models which describe the behaviour of i n d i v i d u a l s making decisions. The research conducted i n these areas can be broken into two groups: one taking a perceptual approach to d e c i s i o n making, the other a cognitive approach. 3 In t h i s chapter we review these two approaches and focus i n d e t a i l on the l i t e r a t u r e r e l a t i n g to the cognitive approach. We also review the evidence of the r e l a t i v e importance of e f f o r t and accuracy i n d e c i s i o n making. 3.1 The perceptual view Prospect theory (Kahneman and Tversky, 1979) i s the best example of a perceptual model of decision behaviour. I t argues that there i s a single, probably hardwired, approach to the evaluation of a l t e r n a t i v e s based on a generalized u t i l i t y model. Changes or variance i n d e c i s i o n outcomes are a t t r i b u t a b l e to the mechanism which i s used to e d i t the i n i t i a l representation of the problem. The way i n which the problem i s represented i s a f f e c t e d by 3 Others (see Junngermann, 1985) have r e f e r r e d to these as the p e s s i m i s t i c and o p t i m i s t i c approaches. 21 c o n t e x t u a l v a r i a b l e s ( f o r example, whether t h e p r o b l e m i s f r a m e d i n terms o f g a i n s o r l o s s e s ) . The e d i t i n g phase s e r v e s t o s i m p l i f y t h e p r o b l e m r e p r e s e n t a t i o n i n o r d e r t o f a c i l i t a t e t h e s u bsequent e v a l u a t i o n p r o c e s s . These p r o c e s s e s a r e t h o u g h t t o be g u i d e d by t h e b a s i c l i m i t a t i o n s o f t h e human p e r c e p t u a l p r o c e s s i n g system. These systems t e n d t o o p e r a t e " a u t o m a t i c a l l y " and a r e n o t under t h e d i r e c t , c o n s c i o u s c o n t r o l o f t h e i n d i v i d u a l . One way w h i c h p r o s p e c t t h e o r y w o u l d e x p l a i n why d e c i s i o n b e h a v i o u r does n o t c o n f o r m t o n o r m a t i v e t h e o r y i s because t h e a p p a r a t u s b e i n g u s e d t o s u p p o r t t h e d e c i s i o n p r o c e s s i s n o t geared-up f o r i m p l e m e n t a t i o n o f t h e n o r m a t i v e model. F o r example, t h e human p e r c e p t u a l s y s t e m a t t e n d s most r e a d i l y t o changes o r d i f f e r e n c e s w h i l e t h e u t i l i t y o r v a l u e models r e q u i r e p r o c e s s i n g i n terms o f a b s o l u t e m agnitudes. As a r e s u l t , p r o c e s s d i f f e r s between the p r e d i c t i o n s o f t h e n o r m a t i v e model and e m p i r i c a l o b s e r v a t i o n s . To t h e e x t e n t t h a t t h e p r o s p e c t t h e o r y model i s v a l i d i t w o u l d p r e d i c t d e c i s i o n b e h a v i o u r s w h i c h a r e l a r g e l y i n v a r i a n t a c r o s s s u b j e c t s . I n d i v i d u a l s w o u l d g e n e r a l l y be unaware o f and u n a b l e t o c o n t r o l t h e s t r a t e g i e s t h e y use ( o r i n v o k e ) f o r d i f f e r i n g s i t u a t i o n s . Such i n v a r i a n c e i s shown by Kahneman and T v e r s k y i n numerous s t u d i e s where v e r y l a r g e p r o p o r t i o n s o f d e c i s i o n makers have been shown t o change t h e i r r e s p o n s e s t o q u e s t i o n s , i n o p p o s i t e d i r e c t i o n s , s i m p l y due t o f r a m i n g e f f e c t s . I f p r o s p e c t t h e o r y i s a c o m p l e t e and v a l i d model o f d e c i s i o n b e h a v i o u r , t h e n v e r y l i t t l e p r o g r e s s i s l i k e l y t o be made i n terms o f a i d i n g o r a s s i s t i n g d e c i s i o n making. I f i n d i v i d u a l s do n o t have c o n s c i o u s c o n t r o l o v e r t h e mechanisms w h i c h t h e y use d u r i n g p r o b l e m s o l v i n g , t h e n no amount o f t r a i n i n g o r a s s i s t a n c e w i l l f a c i l i t a t e t h e improvement o f d e c i s i o n making. D e c i s i o n b e h a v i o u r may be m a n i p u l a t e d i n t h i s c a s e by a c o n s c i o u s f r a m i n g o f t h e 22 problem to invoke c e r t a i n processes from the d e c i s i o n maker, but support tools would be v i r t u a l l y worthless. Given t h i s inference we w i l l not consider prospect theory or the perceptual view i n more d e t a i l . I t provides no useful avenues for the exploration of new mechanisms f o r supporting the decision maker but rather leads to the conclusion that d e c i s i o n makers cannot be supported. This may be why some dec i s i o n making researchers have r e f e r r e d to t h i s as the p e s s i m i s t i c approach (Junngermann et a l . , 1985). 3.2 The cognitive view An a l t e r n a t i v e view of decision making i s provided by those who consider strategy to be under the conscious c o n t r o l of the i n d i v i d u a l . The use of a p a r t i c u l a r strategy i s based on some form of "cost-benefit" evaluation. Decision makers presumably contrast the amount of cognitive e f f o r t required to implement a p a r t i c u l a r strategy with the expected benefits associated with the p a r t i c u l a r strategy. The benefits of the various s t r a t e g i e s are t y p i c a l l y measured as the l i k e l i h o o d of that strategy leading to a good d e c i s i o n or an accurate response (Payne, 1982). Given values f o r cognitive e f f o r t and d e c i s i o n accuracy* a trade-off i s made. The assumption i s that i d e a l l y d e c i s i o n makers would l i k e to maximise the q u a l i t y of t h e i r decisions while at the same time minimising cognitive e f f o r t . However, to the extent that these two objectives are t y p i c a l l y c o n f l i c t i n g , some form of trade-off i s required (Johnson and Payne, 1985). The cost-benefit framework does not specify p r e c i s e l y how decision makers view the nature of the trade-off between e f f o r t and accuracy. Trade-offs are not n e c e s s a r i l y equivalent. Savings of a single u n i t of e f f o r t may be worth more, or l e s s , than a s i n g l e u n i t s a c r i f i c e i n 4 The terms de c i s i o n accuracy and decision q u a l i t y w i l l be used interchangeably. 23 terms of d e c i s i o n q u a l i t y or accuracy. The c e n t r a l question i s : when evaluating trade-offs between e f f o r t and accuracy do d e c i s i o n makers place primary emphasis on 1) e f f o r t reduction? 2) increase i n accuracy? or, 3) are they i n d i f f e r e n t between the two? Unfortunately, the l i t e r a t u r e i n t h i s area i s scant; however, e x i s t i n g evidence w i l l be reviewed below. The r e l a t i v e importance of each component i n the trade-off evaluation w i l l have implications f o r the way an i n d i v i d u a l w i l l use a d e c i s i o n a i d . Consider that there are four possible objectives that a d e c i s i o n maker may have: 1) minimise e f f o r t 2) maximise accuracy 3) maximise accuracy subject to an e f f o r t c o n s t r a i n t 4) minimise e f f o r t subject to an accuracy constraint. These l a t t e r two views simply t r y to express the f a c t that a d e c i s i o n maker may place r e l a t i v e l y more emphasis on accuracy or on e f f o r t . This emphasis, as we s h a l l see, has implications for decision behaviour. Each of these four p o s s i b i l i t i e s w i l l be reviewed but only the l a t t e r two conditions are p a r t i c u l a r l y i n t e r e s t i n g and defensible. The f i r s t two w i l l be examined b r i e f l y f o r the sake of completeness. 3.2.1 E f f o r t minimisation I t i s p l a u s i b l e that i n absence of other incentives that i n d i v i d u a l s are simply e f f o r t minimisers. They w i l l s e l e c t a d e c i s i o n strategy which minimises 24 the amount of cognitive e f f o r t required to complete a given task. In t h i s case there would be no c r i t e r i a imposed on the q u a l i t y of the end r e s u l t i n completing a task. Thus, for example, i n faci n g a choice task, completion could be defined as making a s e l e c t i o n from the p o t e n t i a l choice set. Completion of a task r e q u i r i n g the generation of an a l t e r n a t i v e would be defined as the generation of a possible course of a c t i o n without regard for the q u a l i t y or number of a l t e r n a t i v e s put f o r t h . In short, the idea of e f f o r t minimisation implies that strategy s e l e c t i o n would be undertaken without regard f o r d e c i s i o n q u a l i t y . Presumably, the same strategy would always be used and that would be an e f f o r t minimising strategy, such as the conjunctive (CONJ) or elimination by aspects (EBA) approaches i n p r e f e r e n t i a l choice problems. However, taken to the extreme such an objective would l i k e l y be evidenced by widespread use of simple random s e l e c t i o n s t r a t e g i e s or strategies basing choice on the examination of a single c r i t e r i o n . The f a c t that empirical studies show that d e c i s i o n makers are adaptive would seem to indicate that they are not simply e f f o r t minimisers (Payne, 1976). At the very l e a s t i t would appear that there i s a minimum l e v e l of accuracy that must be maintained and that perhaps s t r a t e g i e s are manipulated to maintain t h i s balance (Bettman et a l . , 1986). 3.2.2 Accuracy maximisation I t i s also conceivable that d e c i s i o n makers would l i k e to make the best possible d e c i s i o n i n a l l cases and that they s e l e c t s t r a t e g i e s to achieve t h i s end. In f a c t , t h i s i s what a normative de c i s i o n models assumes, decision makers wish to make optimal decisions. V i o l a t i o n s of normative u t i l i t y and 25 preference models have been documented s u f f i c i e n t l y to argue that i n d i v i d u a l s do not act i n a u t i l i t y maximising manner (Slovic et a l . , 1977). This i s l i k e l y due i n part to simple l i m i t s on information processing w i t h i n which the d e c i s i o n maker must operate. Information processing i s not a costless a c t i v i t y , and as a r e s u l t i n d i v i d u a l s cannot at a l l times work to maximise the q u a l i t y of t h e i r solutions. I f t h i s were the case then, once again, we would expect to see inv a r i a n t d e c i s i o n behaviour with only very complex additive s t r a t e g i e s being used. The preceding two l i n e s of argument have given the extreme cases. In fa c t strategy s e l e c t i o n must be viewed as being somewhat more complex. Individuals c l e a r l y are adaptive i n the strategies they employ. For simple problems they tend to use l i n e a r s trategies which most c l o s e l y approximate normative models; fo r larger problems simpler h e u r i s t i c s are used which s a c r i f i c e some accuracy i n d e c i s i o n making for reduced e f f o r t . Contingent d e c i s i o n behaviour i s one of the most s i g n i f i c a n t and robust findings i n behavioural d e c i s i o n research over the l a s t ten years (Bettman et a l . , 1986). The f a c t that such behaviour occurs c o n s i s t e n t l y a t t e s t s to the f a c t that d e c i s i o n makers are focussing on trade-offs between multiple objectives. I t i s argued that these objectives are to minimise e f f o r t and maximise accuracy (Johnson and Payne, 1985). The extent to which e f f o r t and accuracy are valued i n t h i s trade-off i s explored below. 3.2.3 Accuracy maximisation subject to an e f f o r t c o n s t r a i n t One l i n e of reasoning would argue that i n d i v i d u a l s wish to maximise the q u a l i t y of t h e i r decisions but must do so within the confines of l i m i t e d information processing capacity. There i s a f i x e d pool of cognitive e f f o r t which must be a l l o c a t e d to various a c t i v i t i e s (Kahneman, 1973). I t seems 26 reasonable to assert that e f f o r t w i l l be expended i n such a way as to optimise constrained d e c i s i o n making. This i s e s s e n t i a l l y Simon's bounded r a t i o n a l i t y argument. Decision makers s i m p l i f y a problem space and then make an optimal d e c i s i o n within that s i m p l i f i e d space. This argument would imply that the trade-offs demonstrated between e f f o r t and accuracy are simply due to information load or overload considerations. The l i t e r a t u r e reviewed i n the previous section c l e a r l y indicates that information load i s one of the causes of contingent behaviour. The findings i n the m u l t i - a t t r i b u t e search studies could c e r t a i n l y be interpreted i n t h i s manner (Payne, 1976; Olshavsky, 1979; Biggs et a l . , 1986). Certain strategies which work well f o r simpler problems would require too much e f f o r t i n a larger, more complex problem and as a r e s u l t cannot be employed. Nevertheless, under t h i s condition we would expect to see d e c i s i o n makers use strategies which are moderately more e f f e c t i v e but require considerably more e f f o r t , provided that the a d d i t i o n a l expenditure of e f f o r t was not beyond some cognitive threshold. E s s e n t i a l l y t h i s l i n e of reasoning would imply that decision makers place a high value on obtaining a good, or correct s o l u t i o n , and are w i l l i n g to expend a large amount of e f f o r t to do so. Output, or work put into a problem, i s e s s e n t i a l l y bounded by the cognitive c a p a c i t i e s of the d e c i s i o n maker. In short, the d e c i s i o n maker i s effectiveness oriented. What would be the impact of a decision a i d i n supporting problem so l v i n g where strategy s e l e c t i o n was guided by t h i s p r i n c i p l e of constrained optimisation? F i r s t , consider that the function of a well designed decision a i d should be to reduce cognitive s t r a i n on the d e c i s i o n maker. I t does t h i s by automating some of the computation and storage requirements that would otherwise be l e f t i n the hands of the d e c i s i o n maker. I f the i n d i v i d u a l ' s 27 objective i s to make the best possible decision, given l i m i t e d processing capacity, i n an aided environment the d e c i s i o n maker would invest geater e f f o r t into the task and p o s s i b l y achieve a higher q u a l i t y r e s u l t . The primary impact would be to increase decision effectiveness. Strategies or approaches which were simply too taxing for the unaided d e c i s i o n maker would be used i n aided s e t t i n g s . Thus we would expect to see the use of more complex strategies which more c l o s e l y conform with normative models. For example, the decision maker with a d e c i s i o n a i d r e a d i l y combining p r o b a b i l i t i e s and outcomes as well as s e l e c t i n g gambles based upon equaprobable rules would l i k e l y use an expected value strategy more frequently, provided he or she understood that t h i s was the "correct" approach. This would also lead to the more complete and consistent use of information for a given problem. In short, a d e c i s i o n maker who i s guided by the accuracy maximisation r u l e would use a d e c i s i o n a i d to implement more complex st r a t e g i e s , converting the expanded pool of e f f o r t into a higher q u a l i t y s o l u t i o n for the given problem. 3.2.4 E f f o r t minimisation subject to an accuracy constraint Information load, or overload, i s c l e a r l y not the only f a c t o r which influences strategy s e l e c t i o n (Beach and M i t c h e l l , 1978; Christensen-Szlanski, 1980; Johnson & Payne, 1985; Shugan, 1980). In f a c t there i s evidence which tends to i n d i c a t e that savings i n e f f o r t are more important to d e c i s i o n makers than improving decision q u a l i t y . This would lead to the a s s e r t i o n that a d e c i s i o n maker's true objective i s to minimise e f f o r t subject to an accuracy constraint. In t h i s case i t i s argued that d e c i s i o n makers are much more s e n s i t i v e to the amount of work which goes into a problem than they are with the q u a l i t y of the r e s u l t i n g decisions. Such trade-offs may, of course, also 28 be a f f e c t e d by a v a r i e t y of other conditions such as the importance of the decision. For example, the d e c i s i o n to purchase a home may be much more important than the decision to rent an apartment. In t h i s case the decision maker would be i n c l i n e d to place a higher l e v e l c o n s t r a i n t on the q u a l i t y of the decision. Within the confines of t h i s constraint, however, the decision maker would s t i l l be expected to be more s e n s i t i v e to increases i n e f f o r t than to adjustments i n decision q u a l i t y beyond the minimal constrained l e v e l . Thus, the whole issue of the l e v e l of incentive provided to the d e c i s i o n maker may most e a s i l y be interpreted as an input into the way the d e c i s i o n maker determines the i n i t i a l constraint on d e c i s i o n q u a l i t y . From the set of d e c i s i o n s t r a t e g i e s which meets t h i s minimum acceptable l e v e l of q u a l i t y the d e c i s i o n maker w i l l then be more s e n s i t i v e to e f f o r t considerations i n s e l e c t i n g an actual strategy for use i n the problem so l v i n g process. This implies that given a choice between modifying e i t h e r v a r i a b l e , as might be the case when using a decision a i d which reduces cognitive load, the d e c i s i o n maker would opt for the l e a s t e f f o r t route rather than converting the freed resources into extra work on the problem which may lead to better q u a l i t y decisions. Thus, when using a d e c i s i o n aid, we would expect that the d e c i s i o n maker would use the strategy which provided an acceptable s o l u t i o n with the smallest possible expenditure of e f f o r t . I f a d e c i s i o n a i d were to automate a ser i e s of s t r a t e g i e s , reducing the cognitive e f f o r t f o r each but maintaining t h e i r r e l a t i v e degree of d i f f i c u l t y or s t r a i n , we would anticipate that the d e c i s i o n maker would continue to u t i l i s e the same strategy as used i n an unaided environment. For a decision a i d to induce change i n t h i s case, i t must a l t e r the e f f o r t rankings or r e l a t i o n s h i p between various s t r a t e g i e s . For example, i f d e c i s i o n makers were i n d i f f e r e n t between the a p p l i c a t i o n of a 29 conjunctive or e l i m i n a t i o n by aspects strategy i n making a choice between a l t e r n a t i v e s , a d e c i s i o n a i d which made the EBA approach easier to implement might r e s u l t i n a s h i f t i n strategy. In short, i n a d e c i s i o n aided environment which equally supported a l l strategies, not changing the e f f o r t r e l a t i o n s h i p s between them, there would be no s h i f t strategy. S h i f t i n g of strategies would require that the net d i f f e r e n c e i n the e f f o r t required to use a c e r t a i n strategy be p o s i t i v e . The net d i f f e r e n c e i n e f f o r t can be assessed by determining the t o t a l number of cognitive operations assumed by the d e c i s i o n a i d and subtracting the number of units of e f f o r t required to implement the function using the a i d . The l a t t e r value can be thought of as the cost of using the system. The net d i f f e r e n c e proposed here i s e s s e n t i a l l y the same as Keen's notion of "the marginal economics of e f f o r t " (Keen, 1979) . I t points out that the two ways to create differences between the e f f o r t required to implement various s t r a t e g i e s with a d e c i s i o n a i d are to 1) Provide tools with d i f f e r i n g l e v e l s of power i n terms of the degree to which they assume cognitive operations f o r the decision maker or, 2) Provide interface techniques i n which c e r t a i n functions are inherently more d i f f i c u l t to employ than others. Of course, i n v i r t u a l l y a l l problems approach 1) would be preferable to approach 2), however there are s i t u a t i o n s where approach 2) would be necessary. Consider, for example, the d i f f i c u l t y of t r y i n g to get a decision maker to use an additive strategy over an e l i m i n a t i o n strategy i n a choice task. I t may not be possible to develop aids for the a d d i t i v e strategy which make i t easier to use than an elimination approach, even i f the l a t t e r was unsupported. In these cases i t may be to the system designer's advantage to 30 a c t u a l l y put impediments i n the way of the use of e l i m i n a t i o n s t r a t e g i e s , thus helping to t i p the e f f o r t d i f f e r e n t i a l i n favour of the a d d i t i v e approach. This argument points to the f a c t that i n developing d e c i s i o n a i d i n g tools we must consider both the costs and benefits that those tools are l i k e l y to provide and that t h i s net difference i s important. This issue w i l l be developed further i n chapter 4. To summarise, there i s evidence and l o g i c a l argument to support both the reasoning that d e c i s i o n makers are e f f o r t minimisers and that they are constrained accuracy optimisers. A c l o s e r examination of some of the l i t e r a t u r e r e l a t i n g to e f f o r t accuracy trade-offs w i l l help to shed further l i g h t on which of these assertions i s correct. 3.3 E f f o r t or accuracy? There are three l i n e s of research which we can examine to help untangle the r e l a t i v e importance of e f f o r t and accuracy to d e c i s i o n makers. None of the evidence can be considered conclusive; however, taken together i t does tend to in d i c a t e that i n d i v i d u a l s are highly conscious of the e f f o r t they put into making decisions and are r e l a t i v e l y less concerned with optimising decision q u a l i t y . The three areas of work we w i l l examine are 1) conceptual models of cost b e n e f i t trade-offs (Beach and M i t c h e l l , 1978; Shugan, 1980); 2) simulation work on the e f f o r t and accuracy of various decision h e u r i s t i c s (Thorngate, 1980; Johnson and Payne, 1985; Bettman et a l . , 1986); and 3) empirical studies examining strategy s e l e c t i o n i n choice problems (Russo and Dosher, 1983; Bettman and Kakkar, 1976; Jarvenpaa, 1988) 31 3.3.1 Conceptual models Shugan (1980) develops a conceptual model of strategy s e l e c t i o n which accounts f o r expenditures of cognitive e f f o r t . He argues that the f a i l u r e to see s i g n i f i c a n t use of u t i l i t y or value maximising st r a t e g i e s i s due to the f a c t that those strategies f a i l to consider the "cost of thinking" as an input to strategy s e l e c t i o n . He asserts that i n s e l e c t i n g s t r a t e g i e s people are attempting to minimise e f f o r t subject to a constraint on the q u a l i t y of the decision. He expresses t h i s constraint i n terms of an e x t e r n a l l y s p e c i f i e d confidence l e v e l f or making a correct decision. Strategies are evaluated i n terms of the amount of e f f o r t required to meet the s p e c i f i e d l e v e l of decision q u a l i t y . The l e a s t e f f o r t f u l strategy which provides an acceptable decision w i l l be chosen. Shugan (1980) also recognises that e f f o r t may be a function of task c h a r a c t e r i s t i c s such as s i m i l a r i t y between a l t e r n a t i v e s or importance of the decision. In general, the model i s consistent with the idea that i n d i v i d u a l s act to minimise e f f o r t subject to some constraint on the q u a l i t y of the decision. S i m i l a r l y Beach and M i t c h e l l (1978) have developed a model designed to spe c i f y the contingencies which may e f f e c t the s e l e c t i o n of decision s t r a t e g i e s . They argue that various c h a r a c t e r i s t i c s of the d e c i s i o n maker and the task w i l l influence strategy s e l e c t i o n . For example, task conditions such as complexity, novelty, ambiguity, or s i g n i f i c a n c e of outcomes can e f f e c t the desire to make an accurate choice or the amount of e f f o r t to be focussed on the problem. At the same time the general c a p a b i l i t i e s of the decision maker may influence h i s or her desire to work on a problem and may also impact the type of s o l u t i o n that w i l l be accepted. I t should be noted that cognitive 32 load i s but one aspect of the model. Beach and M i t c h e l l state that the d e c i s i o n maker i s motivated to s e l e c t the l e a s t e f f o r t f u l strategy which w i l l provide an acceptable s o l u t i o n given the various contingencies. This c l e a r l y involves the notion of a trade-off between the desire to make a good decision and the c o n f l i c t i n g desire to minimise e f f o r t expended on the problem. The implied d e c i s i o n r u l e i s to minimise e f f o r t subject to an accuracy constraint. 3.3.2 Simulation models Thorngate (1980) conducted a simulation study to evaluate the usefulness of a v a r i e t y of h e u r i s t i c s which can be employed to make decisions under r i s k . He examined the q u a l i t y of the decisions r e l a t i v e to an expected model. The h e u r i s t i c s were tested under various l e v e l s of information load. The r e s u l t s of the simulation indicated that the h e u r i s t i c s performed r e l a t i v e l y well and at the same time required much less e f f o r t than the expected value model. In t h i s sense Thorngate (1980) considers the h e u r i s t i c s e f f i c i e n t . They provide good, though not optimal, solutions and do so while imposing r e l a t i v e l y few cognitive demands on the decision maker. He argues that an i n d i v i d u a l ' s tendency to ignore or misuse p r o b a b i l i t i e s may simply be the r e s u l t of a trade-off, where the increase i n cognitive e f f o r t to move from a good to optimal s o l u t i o n i s not considered j u s t i f i e d . The use of the h e u r i s t i c s may be motivated by the f a c t that the optimal model provides l i t t l e b e n e f i t r e l a t i v e to the use of h e u r i s t i c s while, at the same time imposing a large cost. Johnson and Payne (1985) conducted a simulation study which r e p l i c a t e d and extended the work of Thorngate. While the l a t t e r focussed on the q u a l i t y of decisions produced by the h e u r i s t i c s , Johnson and Payne modeled both e f f o r t 33 and accuracy of h e u r i s t i c s i n an attempt to understand the nature of the trade-off between the two. They also simulated the performance of the h e u r i s t i c s over a wide v a r i e t y of task conditions by varying the number of a l t e r n a t i v e s , the number of outcomes for each a l t e r n a t i v e , variance i n p r o b a b i l i t y l e v e l s , and the presence or absence of dominated a l t e r n a t i v e s i n the choice set. The r e s u l t s of the simulation suggest that there i s indeed a trade-off between the e f f o r t involved i n using some h e u r i s t i c s and the accuracy of responses that they provide. Also, some h e u r i s t i c s are r e l a t i v e l y stable i n terms of the e f f o r t they require as information load increases. The findings also i n d i c a t e that decision makers need to be adaptive i n order to use h e u r i s t i c s to e f f e c t i v e l y reduce e f f o r t while at the same time maintaining a high degree of accuracy. No s i n g l e h e u r i s t i c performs well under a l l conditions. However, h e u r i s t i c s can be highly accurate while saving the d e c i s i o n maker sub s t a n t i a l e f f o r t r e l a t i v e to the normative approach. Johnson and Payne argue that these trade-offs are at l e a s t a p a r t i a l explanation for the contingent d e c i s i o n behaviour found i n many studies (see Payne, 1982 for a review). This work also reinforces the notion that d e c i s i o n makers are e f f o r t conscious. H e u r i s t i c s have been developed which gr e a t l y reduce e f f o r t while at the same time s a c r i f i c i n g l i t t l e i n terms of d e c i s i o n q u a l i t y . The key from the d e c i s i o n maker's point of view i s to chose the appropriate h e u r i s t i c to minimise e f f o r t while at the same time maintaining reasonable l e v e l s of accuracy. The empirical evidence to be reviewed below suggests that t h i s i s i n large measure the case. 3.3.3 Empirical studies 34 Russo and Dosher (1983) i n t h e i r study of binary choice problems have also asserted that i n d i v i d u a l s make deliberate choices of d e c i s i o n s t r a t e g i e s based upon a trade-off between error and e f f o r t . They claim that these trade-offs show a greater a t t e n t i o n to e f f o r t reduction than they do to accuracy maximisation. Christensen-Szalanski (1980) also makes s i m i l a r arguments. Ind i v i d u a l d e c i s i o n making i s adaptive and r e f l e c t s a trade-off between the costs and ben e f i t s of making a decision. Further trade-offs are not simply due to information load considerations since external incentives can impact the amount of e f f o r t put into a problem (Christensen-Szalanski, 1980). Bettman et a l . (1986) demonstrate that decision makers are h i g h l y adaptive i n the types of s t r a t e g i e s they s e l e c t , considering both e f f o r t and accuracy. People perform as i f they are t r y i n g to maintain a r e l a t i v e l y constant l e v e l of de c i s i o n q u a l i t y while minimising the e f f o r t they expend i n f i n d i n g a so l u t i o n . Bettman and Kakkar (1977) also show r e s u l t s which indi c a t e that i n d i v i d u a l s consider e f f o r t minimisation important. Jarvenpaa (1987) has also demonstrated that d e c i s i o n makers use strategies that allow them to process information i n the l e a s t e f f o r t f u l manner. At the same time i t was noted that d e c i s i o n makers would change strategies i f accuracy f e l l below a c e r t a i n threshold. These findings on information presentation are consistent with S l o v i c ' s (1972) claim that d e c i s i o n makers tended to use information only i n the form presented. This notion, labeled concreteness, was also based on the b e l i e f that d e c i s i o n makers are highly conscious of the degree of e f f o r t that they put into a problem. To summarise these arguments, there i s evidence, though i t i s f a r from being conclusive, that i n d i v i d u a l s solve problems i n such a way as to maintain reasonable l e v e l s of accuracy while at the same time t r y i n g to minimise 35 e f f o r t . In t h i s case we expect decision makers to s e l e c t adaptively from a r e p e r t o i r e of strategies i n order to reach good, though not n e c e s s a r i l y optimal solutions while at the same time minimising expenditures of e f f o r t . In a s i m i l a r review of a v a i l a b l e evidence Stone (1987) also concluded that the primary objective of a decision maker i s to minimise expenditures of e f f o r t . Decision q u a l i t y i s viewed as r e l a t i v e l y l e s s c r i t i c a l . Cognitive cost i s a prime impediment to the use of complex de c i s i o n s t r a t e g i e s . Changing the cognitive costs associated with c e r t a i n d e c i s i o n s t r a t e g i e s may well a l t e r the t r a d i t i o n a l trade-off between e f f o r t and accuracy as well as the r e l a t i v e weights the d e c i s i o n maker places on them. In t r y i n g to understand and evaluate the impact of decision aids, which are designed to reduce cognitive cost, i t w i l l be u s e f u l to examine both the e f f o r t minimisation and accuracy maximisation perspectives. To t h i s end hypotheses w i l l be developed, which p r e d i c t both types of performance by i n d i v i d u a l s using a d e c i s i o n aid, i n chapter 5. These hypotheses, which w i l l i n some cases be c o n f l i c t i n g , can then be examined i n l i g h t of the empirical r e s u l t s presented l a t e r to determine which l i n e of reasoning i s more defensible. This should a s s i s t i n allowing us to better p r e d i c t and understand the impact of d e c i s i o n aids on the decision process. The next chapter w i l l attempt to provide some i n s i g h t into the impact of d e c i s i o n aids on processing which i s c o n t r o l l e d by an effort-accuracy trade-o f f . P r i o r to t h i s analysis i t i s necessary to b r i e f l y address two issues. 1) What i s the o v e r a l l v a l i d i t y of the cost-benefit approach? 2) Can the perceptual and cognitive views be reconciled? 36 3.4 V a l i d i t y of the cost-benefit framework Payne (1982) provides ample evidence f o r the basis of a cost-benefit approach. Much of the l i t e r a t u r e presented i n chapter 2 i s also most e a s i l y i nterpreted i n l i g h t of the cost-benefit argument. As w e l l , the review j u s t presented which focusses more p r e c i s e l y on the nature of such trade-offs provides evidence of a d a p t i v i t y which i s consistent with the cost-benefit framework. The simple f a c t that strategies change systematically with changes i n such task v a r i a b l e s as s i m i l a r i t y and complexity provides evidence of adaptive mechanisms. Further, protocol data, such as Payne (1976), provide evidence that subjects are aware of the f a c t that they are adapting strategies to various task conditions. In t h i s sense, the cost-benefit argument has a great deal of empirical support i n addition to face v a l i d i t y . Also such a theory provides a great deal of comfort to those who wish to perpetuate the b e l i e f of man as a r a t i o n a l decision maker (March, 1977). Actions can be interpreted as economically r a t i o n a l i f the "cost of thinking" i s incorporated. This f a c t alone ensures that the cost-benefit approach w i l l have a large number of advocates. One basic d i f f i c u l t y with the cost-benefit argument i s that which i s faced by any theory claiming the existence of a d r i v i n g or c o n t r o l l i n g routine, namely the problem of i n f i n i t e regression (Einhorn and Hogarth, 1981). I f the choice of strategy based upon cost-benefit i s i t s e l f considered a decision problem, how i s t h i s evaluation controlled? Einhorn and Hogarth (1981) provide the most complete c r i t i q u e of such explanations and express the concern that post hoc any r e s u l t can be interpreted as a function of an effort-accuracy trade-off. Payne (1982) argues that there i s l i k e l y not a s i n g l e d e c i s i o n system, but rather there are multiple i n t e r a c t i n g systems. 37 Another system may control the high l e v e l evaluation or operate p a r a l l e l to i t , being responsible for the execution of the strategy s e l e c t i o n program. Also, Johnson and Payne (1985) argue that the problem of i n f i n i t e regression may be overcome i f we speculate that the use of c e r t a i n s t r a t e g i e s under c e r t a i n conditions i s a learned automatic behaviour. Further, such theories which propose cognitive mechanisms should generally be considered to describe a system which functions as i f i t were performing i n the s p e c i f i e d manner. Few people would argue these p a r a l l e l s l i n k d i r e c t l y to the implementation l e v e l . In t h i s respect there may be a major conceptual flaw i n considering the cost-benefit approach as the single mechanism responsible f o r invoking d e c i s i o n s t r a t e g i e s . I t appears to be reasonably robust as a model of how i n d i v i d u a l s behave i n making decisions as has been demonstrated by the empirical work of Payne, Biggs, Olshavsky and others as well as the simulation work of Johns on and Payne (1985j 1986) and Thorngate. In addition, the actual implementation of the cognitive mechanisms which support the cost-benefit argument while of great i n t e r e s t to cognitive psychologists i s not p a r t i c u l a r l y germane to the study of applied d e c i s i o n making and decision support. In essence i t can be argued that i f the cost-benefit concept helps to p r e d i c t the way i n d i v i d u a l s behave, then we have a u s e f u l e m p i r i c a l l y grounded basis for developing decision support systems, at l e a s t for a p a r t i c u l a r class of problems. Such a base i s notably absent from much e x i s t i n g work i n the DSS f i e l d . 3.5 Reconciling the cognitive and perceptual views Though the cost-benefit model i s reasonably robust, i t does not explain the invariance i n performance across a wide range of tasks c i t e d i n the 38 general l i t e r a t u r e on h e u r i s t i c s (Kahneman et a l . , 1982). On a surface examination the two approaches seem somewhat contradictory; however, on closer inspection the two views may be reconciled. The work conducted i n e s t a b l i s h i n g the perceptual model tends to focus on elementary information processing, making s p e c i f i c judgments most often under conditions of binary choice. The cognitive approach generally studies information search and problem s t r u c t u r i n g i n m u l t i - a t t r i b u t e , m u l t i - a l t e r n a t i v e s e t t i n g s . The majority of the a c t i v i t i e s i n these problems surround f i l t e r i n g and s t r u c t u r i n g a l t e r n a t i v e s . These problems are not d i r e c t l y comparable to the unidimensional, r i s k y choice problems accounted f o r by prospect theory. In general terms there would be no reason to expect that the c o g n i t i v e mechanisms described i n prospect theory generalize to choices between m u l t i - a t t r i b u t e choice problems (Kahneman and Tversky, 1979). The f i n a l comparisons or judgments between a p a i r of a l t e r n a t i v e s on a given a t t r i b u t e may be automatically executed using a routine s i m i l a r to that described by prospect theory. This does not imply that the choice of an i n i t i a l search strategy i s not a t t r i b u t a b l e to a cost-benefit computation nor does i t lead to the conclusion that there i s a need to choose between perceptual and cognitive approaches (Payne, 1982). Each would appear to operate at d i f f e r e n t l e v e l s of problem so l v i n g and possibly consciousness. Individuals appear to have co n t r o l over the choice of search strategies and which items w i l l be compared. When making d i r e c t pairwise evaluation of preferences i t may be that a mechanism such as prospect theory i s invoked. In t h i s sense, for the development of d e c i s i o n aids we may argue that i t w i l l be more f r u i t f u l to work i n areas such as s t r u c t u r i n g i n a m u l t i - a t t r i b u t e environment rather than i n t r y i n g to change evaluation procedures which are hardwired. 39 The n e x t c h a p t e r o u t l i n e s some s t r a t e g i e s f o r d e c i s i o n making i n m u l t i -a t t r i b u t e , m u l t i - a l t e r n a t i v e s e t t i n g s and p u t s f o r t h some mechanisms t o s u p p o r t t h o s e p r o c e s s e s b o t h by co m p l e t e a u t o m a t i o n o f t h e p r o c e d u r e and by p r o v i d i n g s p e c i f i c t o o l s t o augment human i n f o r m a t i o n p r o c e s s i n g power. 40 CHAPTER 4 - A BEHAVIOURAL DECISION THEORY APPROACH TO DSS DEVELOPMENT 4.0 I n t r o d u c t i o n The p r e c e d i n g c h a p t e r s have a r g u e d t h a t i n d i v i d u a l s f i n d d e c i s i o n making a complex, s t r a i n i n d u c i n g t a s k . Much o f t h i s d i f f i c u l t y has been a t t r i b u t e d t o t h e d e c i s i o n maker's l i m i t e d c a p a c i t y f o r i n f o r m a t i o n p r o c e s s i n g . Under d i f f e r e n t c o n d i t i o n s o f i n f o r m a t i o n l o a d and d e c i s i o n i m p o r t a n c e , d i f f e r e n t p r o c e s s i n g s t r a t e g i e s w i l l be a d o p t e d w h i c h w i l l i n f l u e n c e f i n a l c h o i c e s . S t r a t e g i e s a r e t h o u g h t t o be d e t e r m i n e d by t h e a p p l i c a t i o n o f a s e t o f c o s t -b e n e f i t r u l e s w h i c h a s s e s s t r a d e - o f f between e f f o r t and a c c u r a c y . T h i s c o s t -b e n e f i t n o t i o n may p r o v i d e t h e b a s i s f o r t h e development o f DSS w h i c h t r u l y augment d e c i s i o n makers. Presumably one way t o improve a c c u r a c y o f f i n a l c h o i c e s w o u l d be t o r e d u c e t h e e f f o r t t h a t i s r e q u i r e d t o implement t h o s e s t r a t e g i e s w i t h h i g h e r p r o b a b i l i t i e s o f good o r a c c u r a t e outcomes. I t i s p r o p o s e d h e r e t h a t d e c i s i o n s u p p o r t systems (DSS) c o u l d p r o v i d e a mechanism f o r a t t a i n i n g t h i s r e s u l t . T y p i c a l l y a DSS f o c u s s e s on t h e e v a l u a t i o n o f a l t e r n a t i v e s p r i o r t o c h o i c e . DSS a r e g e n e r a l l y a p p l i e d t o s e m i - s t r u c t u r e d p r o b l e m s o f a q u a n t i t a t i v e n a t u r e . Problems a r e s e m i - s t r u c t u r e d i n t h a t t h e r e i s no known p r o c e d u r e o r a l g o r i t h m f o r o b t a i n i n g an o p t i m a l s o l u t i o n o r t h e v a r i a b l e s u s e d as i n p u t t o t h e a l g o r i t h m a r e p r o b a b i l i s t i c . I n t h i s d i s s e r t a t i o n o u r f o c u s i s on i l l - s t r u c t u r e d problems where no o p t i m a l a l g o r i t h m e x i s t s , namely m u l t i -a t t r i b u t e , m u l t i - a l t e r n a t i v e p r e f e r e n t i a l c h o i c e p r o b l e m s . One o f the c h a r a c t e r i s t i c s o f p r e f e r e n t i a l c h o i c e problems i s t h a t t h e v a l u e o f t h e i n p u t 41 data i s generally r e l i a b l e and d e t e r m i n i s t i c . In t h i s sense we are dealing with problems of decision making under c e r t a i n t y . However, due to differences i n i n d i v i d u a l preferences, no two i n d i v i d u a l s w i l l n e c e s s a r i l y make the same f i n a l choice nor w i l l they follow the same procedure i n making a choice. There are numerous paths that can be taken to reach a common s o l u t i o n (Simon, 1982). In the p r e f e r e n t i a l choice l i t e r a t u r e a number of s t r a t e g i e s have been observed. Without a d e t a i l e d knowledge of an i n d i v i d u a l ' s preferences, i t i s not possible f o r anyone to make a guaranteed s a t i s f a c t o r y choice f o r the decision maker. Also there i s no generally accepted approach f o r information processing i n these sett i n g s . These problems d i f f e r from t y p i c a l DSS a p p l i c a t i o n settings since much of the data used i s q u a l i t a t i v e . The normative model used to solve these problems i s preference or value theory which i s c l o s e l y r e l a t e d to expected u t i l i t y theory; however, as we have in d i c a t e d previously, s i g n i f i c a n t deviations from t h i s model occur i n r e a l world problem settings. One approach to developing DSS for such applications i s to implement models which extract estimates of preferences from the d e c i s i o n maker and then proceed to suggest choices based upon those estimates. There i s a great deal of l i t e r a t u r e i n t h i s area dating back to Edwards (1961). More recent examples of such systems are described by Humphreys and McFadden (1980) and Humphreys and Wisudha (1987). The r e a c t i o n of users to formal systems for supporting p r e f e r e n t i a l choice has been mixed and t h e i r use has been f a r from widespread (Dickmeyer, 1984). John et a l . (1983) found that such systems were judged by users as i n f e r i o r to an analyst performing the same task, though the q u a l i t y of the outcomes was not d i f f e r e n t between the system and analyst. In t h i s respect systems which 42 u t i l i z e s o p h i s t i c a t e d techniques for e l i c i t i n g and computing preferences based upon u t i l i t y theory may have low user acceptance. Other experimental evidence also points to the reluctance of i n d i v i d u a l s to accept the outcomes of de c i s i o n analysis (Christen and Somet, 1980). F i s c h o f f (1981) has suggested that the solutions suggested by d e c i s i o n a n a l y t i c techniques may not be accepted because they are output from an incomprehensible process with no i n t u i t i v e appeal and are d i f f i c u l t to j u s t i f y . These arguments seem to be supported by the f a c t that formal choice models have found l i t t l e use i n p r a c t i c a l management s i t u a t i o n s . Another possible explanation for the l i m i t e d a p p l i c a t i o n of such models may be that the cognitive cost associated with the e x p l i c i t i d e n t i f i c a t i o n of preference functions i s quite high. Keen (1979) has argued that i f the perceived b e n e f i t of the t o o l does not exceed i t s cost, i t w i l l not be used. This may be one of the d i f f i c u l t i e s inherent i n complex systems that perform preference determination. Given that preference and u t i l i t y modelling approaches have not been widely used or p a r t i c u l a r l y successful, i t may be that other techniques which are developed to support and augment the actual behaviour of d e c i s i o n makers would be of i n t e r e s t . These systems would conform to one of the c e n t r a l tenets of DSS, namely that the system should support, not replace the i n d i v i d u a l d e c i s i o n maker. The introduction of l i m i t e d support models i n these areas can be compared to the use of DSS i n problem domains where optimal management science models f a i l e d to achieve successful implementation (Keen and Scott-Morton, 1978). In applying d e c i s i o n strategies very few people e x p l i c i t l y employ such techniques as preference evaluation or expected value when evaluating d i f f e r e n t a l t e r n a t i v e s (see, Biggs, 1978 and Troutman and Shanteau, 1976 for 43 evidence with respect to p r e f e r e n t i a l choice and Kahneman et a l . (1982) for an overview of r e s u l t s of p r o b a b i l i s t i c evaluations). As argued i n the previous se c t i o n choices between p a i r s of r i s k y a l t e r n a t i v e s are l i k e l y governed by a mechanism such as that described by prospect theory. Furthermore, such mechanisms are presumably hardwired and i n v a r i a n t to task demands. In t h i s regard supporting evaluation at t h i s l e v e l may be a l e s s than f r u i t f u l endeavour. The s e l e c t i o n of strategy for s t r u c t u r i n g information a c q u i s i t i o n and use, however, i s thought to be c o n t r o l l e d by a conscious cost-benefit strategy. Thus i t might seem reasonable that the development of decision aids which would reduce the cognitive load of implementing various strategies could encourage more thorough analysis. 4.1 Models of p r e f e r e n t i a l choice This s e c t i o n i s intended to describe, i n formal terms, some basic d e s c r i p t i v e models of information search that are employed i n p r e f e r e n t i a l choice problems. Providing formal d e f i n i t i o n s for these models w i l l point d i r e c t l y to the types of mechanisms which need to be incorporated into a d e c i s i o n a i d designed to support d e c i s i o n making i n a p r e f e r e n t i a l choice s e t t i n g . Some design considerations which can be drawn from t h i s analysis w i l l be presented i n section 4.2. Four de c i s i o n strategies commonly employed i n m u l t i - a t t r i b u t e , multi-a l t e r n a t i v e , p r e f e r e n t i a l choice problems are 5: a d d i t i v e compensatory (AC), additive d i f f e r e n c e (AD), conjunctive (CONJ) and e l i m i n a t i o n by aspects (EBA) (Payne, 1976; Biggs, 1978; Biggs et a l . , 1985; Jarvenpaa, 1988). Each of 5 For s i m p l i c i t y these w i l l now be r e f e r r e d to as " p r e f e r e n t i a l choice problems." 44 these e n t a i l the use of various elementary cognitive operations i n order to carry out an examination of a l t e r n a t i v e s and make a choice between them. U t i l i s i n g these strategies w i l l place varying l e v e l s of cognitive load on the d e c i s i o n maker, with the additive strategies (AC,AD) being more d i f f i c u l t than the e l i m i n a t i o n or f i l t e r i n g s t rategies (CONJ, EBA). In order to derive mechanisms for supporting these types of s t r a t e g i e s i t i s u s e f u l to review how each can be implemented and i d e n t i f y the elementary information processes (EIPs) involved. EIPs are processes which we would generally consider to be atomic, for example a s i n g l e s h i f t i n attention, reading a s i n g l e data point, comparing two points, or r e t r i e v i n g an element of information from long term memory (Johnson and Payne, 1985). These are t y p i c a l l y processes which occur at the m i l l i - s e c o n d l e v e l . The idea behind evaluating the s p e c i f i c EIPs i s to t r y to derive support mechanisms based upon an understanding of how decision makers a c t u a l l y behave i n s o l v i n g problems. This i s i n some sense at odds with the more t r a d i t i o n a l normative approach which s p e c i f i e s how d e c i s i o n makers should behave i n order to obtain desired outcomes. However, the f a i l i n g of many normative approaches to d e c i s i o n making i s that they do not d i r e c t l y take into account the s p e c i f i c c a p a b i l i t i e s and l i m i t a t i o n s of the decision maker. By t r y i n g to b u i l d support based upon de s c r i p t i v e models, we may overcome t h i s problem and derive systems which a l l e v i a t e the problems t y p i c a l l y f a c i n g a d e c i s i o n maker such as l i m i t e d short term memory and r e l a t i v e l y slow s e r i a l processing speed. The models can be evaluated and c l a s s i f i e d along two basic dimensions 1) compensatory versus noncompensatory and 2) independent versus dependent Models which evaluate a single a l t e r n a t i v e at a time allowing low ratings 45 on one a t t r i b u t e to be o f f s e t by high ratings on other a t t r i b u t e s are s a i d to be compensatory. Those which require minimum threshold l e v e l s f o r s p e c i f i c elements do not permit low values to be o f f s e t by high values on other a t t r i b u t e s ; these models are considered noncompensatory. The use of a p r o t o t y p i c a l compensatory search strategy i s manifested by the f a c t that a constant amount of information i s searched f o r each a l t e r n a t i v e . Noncompensatory strategies imply f i l t e r i n g or e l i m i n a t i o n r e s u l t i n g i n v a r i a b l e amounts of information being searched f o r each a l t e r n a t i v e . The second major c r i t e r i a f o r d i s c r i m i n a t i n g between decision-making approaches i s a form of independence. Independence, i n t h i s case, implies that analysis and comparisons are not made across a l t e r n a t i v e s u n t i l an o v e r a l l measure of each a l t e r n a t i v e has been established. Independence i s suggested by searches which are interdimensional, i . e . , the basic u n i t of analysis i s the a l t e r n a t i v e and a l l information i s examined f o r a given a l t e r n a t i v e before moving on to the next one. Only one a l t e r n a t i v e i s evaluated at a time. Strategies which are not independent are demonstrated by searches which occur across a l t e r n a t i v e s , comparing a t t r i b u t e values. Such searches are s a i d to be intradimensional. 4.1.1 The additive-compensatory model Analysis using the Additive-Compensatory Model implies that a subject would examine one a l t e r n a t i v e at a time along a l l relevant a t t r i b u t e s . During the evaluation each a t t r i b u t e would be assigned a weight. A f t e r completing the evaluation a score for each a l t e r n a t i v e would be derived summing the product of the a t t r i b u t e value and weight; the a l t e r n a t i v e with the highest sum would be chosen. Thus, the Additive-Compensatory strategy i s 46 compensatory; a l l a t t r i b u t e s are evaluated, weighted and summed allowing low values on some at t r i b u t e s to be o f f s e t by high values on other a t t r i b u t e s . Since each a l t e r n a t i v e i s examined i n i t s e n t i r e t y and no comparisons are made u n t i l an o v e r a l l evaluation has been a r r i v e d at, the model exhibits independence. In terms of cognitive load t h i s i s a r e l a t i v e l y demanding model fo r the de c i s i o n maker to employ without the use of any form of de c i s i o n aids. I t involves the use of a l l a v a i l a b l e information and considerable processing i s associated with each item of information examined. 4.1.1.1 Formal d e s c r i p t i o n of strategy More formally the operators used to invoke t h i s model could be described as follows: Move to a l t e r n a t i v e j at random from the candidate set. L00P1 I f number of nonevaluated a t t r i b u t e s f o r chosen a l t e r n a t i v e = and number of nonevaluated a l t e r n a t i v e s > <ji move to next a l t e r n a t i v e j+1 LOOP 2 I f the number of nonevaluated a t t r i b u t e s f o r the chosen a l t e r n a t i v e > 9 move to the next a t t r i b u t e i read the value of the a t t r i b u t e a i J ( a t t r i b u t e i f o r a l t e r n a t i v e j ) r e t r i e v e the weight (wA) associated with the given a t t r i b u t e i combine the a t t r i b u t e value and weight g i v i n g v^a^ r e t r i e v e current t o t a l (Sum(w ia i j)) add the weighted value to t o t a l f o r current a l t e r n a t i v e store new t o t a l 47 Go to LOOP 2 Go to LOOP 1 I f number of nonevaluated a l t e r n a t i v e s = 0 r e t r i e v e scores f o r each a l t e r n a t i v e (j) compare a l l scores s e l e c t the a l t e r n a t i v e with highest score 6 This i s a minimal complexity version of the AC model i n that i t assumes the existence of a preference structure (Keeney and Ra i f f a ) which can be re t r i e v e d . The weights are stored i n long term memory (LTM) which implies that f o r each a t t r i b u t e a r e t r i e v a l must take place; such a c t i v i t i e s are r e l a t i v e l y demanding. Storage of a un i t of information into LTM takes on the order of ten seconds; r e t r i e v a l i s a m i l l i s e c o n d operation but i s subject to a r e l a t i v e l y high error rate. For t h i s model a t t r i b u t e weights, unless they e x i s t p r i o r to analysis, would have to be created and stored once and r e t r i e v e d many times during the evaluation of a set of a l t e r n a t i v e s . Also a running t o t a l of the score f o r a given a l t e r n a t i v e would have to be maintained and updated as would a vector of f i n a l scores f o r each a l t e r n a t i v e . The values of the a t t r i b u t e s are read from an external source which i s a reasonably low cost function, taking on the order of 0.3 seconds. Since a l l information i s 6 In t h i s case the score i s a preference value and the high p o s i t i v e value r e l a t e s to the t o t a l preference score. I t i s e n t i r e l y p ossible that t h i s high preference score was a r r i v e d at by the aggregation of values on at t r i b u t e s where lower values would r e s u l t i n a higher l e v e l of preference. In l a t e r model the > sign i s used to indicate t h i s same preference r e l a t i o n s h i p . Thus expression such as a ^ > c^ should be read as a i s prefered to c rather than a i s greater than c since i n some cases smaller values w i l l be preferable to l a r g e r ones. 48 examined using t h i s strategy, the issue of ordering alternatives i s not important. Combining the attribute value and weight i s generally assumed to be a m u l t i p l i c a t i v e operation (however, any method of varying complexity could be used). The current t o t a l value for the alternative must be retrieved (again, presumably from LTM) and the new value added to i t (such processes take on the order of 0.3 - 0.5 seconds). The new t o t a l i s then stored i n LTM. The continual summation of the updated t o t a l minimizes the number of accesses to memory and makes this a reasonably e f f i c i e n t form of the AC model. This set of instructions w i l l be executed for each attribute of the given alternative u n t i l i t i s completely evaluated, at which time attention i s sh i f t e d (via a MOVE command) to the next alternative. When a l l alternatives have been evaluated, i t i s necessary to retrieve the t o t a l for each alternative. The scores must be compared. The precise nature of t h i s mechanism would be dependent upon the number of alternatives. A minimal cost comparison could be conducted i f a l l scores could be read into short term memory (STM) for processing. The simplest approach may be to retrieve two scores, compare them, reta i n the highest and retrieve another. This process would continue u n t i l a single alternative remains. A l t e r n a t i v e l y , i t i s conceivable that on completing the evaluation of alternative n+1, i t i s compared d i r e c t l y to the t o t a l for the previous alternative n and the better score i s retained. This procedure would not be t o t a l l y consistent with the pure AC model, but would be a p r a c t i c a l means by which a decision maker could reduce some of the storage and processing demands associated with the implementation of this strategy. In a l l of t h i s analysis we are assuming that the decision maker does not have recourse to any external support for the problem solving process either i n terms of external storage mechanisms or 49 tools to support the processing and i n t e g r a t i o n of information. 4.1.1.2 Supporting the strategy Given an algorithm for processing i n such a manner i t i s possible to s p e c i f y several components of the d e c i s i o n procedure which could be supported. I f the vector of weights f o r each a l t e r n a t i v e was a v a i l a b l e , along with a mechanism to place a l l a t t r i b u t e s on a common scale of measurement, i t would be a r e l a t i v e l y simple task to automate the e n t i r e d e c i s i o n process. This i s exactly what a value function (Keeney and R a i f f a , 1976) attempts to do, formalise and implement a preference structure f o r making trade-offs under c e r t a i n t y . I t i s also possible to develop more l i m i t e d support f o r t h i s processing strategy by a l l e v i a t i n g the computational and storage demands placed on the d e c i s i o n maker. Attention to and reading of a t t r i b u t e s into STM must remain with the d e c i s i o n maker. However, storage of weights and t o t a l s could e a s i l y be accommodated i n a computer-based d e c i s i o n a i d . S i m i l a r l y , a v a r i e t y of basic mathematical functions could be provided f o r s p e c i f y i n g the combination of weights and values and addition of t o t a l s . Once a vector of t o t a l scores i s obtained i t would be a simple matter to s o r t the outcome scores and s e l e c t the highest value. These functions are a l l r e l a t i v e l y easy to implement and would a l l e v i a t e the d e c i s i o n maker's computational burden. However, a l l of these support mechanisms presuppose an a b i l i t y to weight the various a t t r i b u t e s , something which i s seldom done i n an e x p l i c i t manner. Even i n the case where weights are unavailable, some basic support mechanisms could be developed; namely, simple rank ordering of a t t r i b u t e s and of a l t e r n a t i v e s to f a c i l i t a t e evaluation and choice. Also i n terms of data display, an i n d i v i d u a l may perform better when each a l t e r n a t i v e i s presented 50 i n d i v i d u a l l y to avoid the p o s s i b i l i t y of d i s t r a c t i o n from i r r e l e v a n t data. In general, without f o r c i n g a s p e c i f i c a t i o n of a value function there i s l i t t l e that can be done to support t h i s mode of analysis. 4.1.2 The ad d i t i v e - d i f f e r e n c e model The Additive-Difference (AD) model (Tversky, 1969) i s used when an i n d i v i d u a l makes pairwise comparisons between s p e c i f i c a t t r i b u t e s f o r two a l t e r n a t i v e s . These comparisons are evaluated f o r each a t t r i b u t e and the r e s u l t s are weighted and summed to produce an o v e r a l l evaluation. This i s a compensatory model as addition of the weighted differences w i l l permit low values to be o f f s e t by higher values on other a t t r i b u t e s . The model i s not independent since comparisons are made between a l t e r n a t i v e s . When one set of the comparisons f o r two a l t e r n a t i v e s i s complete, dominant a l t e r n a t i v e i s retained while the other i s rejected. The retained a l t e r n a t i v e i s then compared against the next a l t e r n a t i v e . 4.1.2.1 Formal d e s c r i p t i o n of strategy A d e t a i l e d d e s c r i p t i o n of the EIPs involved i n using the AD model i s as follows: Move to an a l t e r n a t i v e j (at random) from candidate set Drop j from candidate set Store j as current-preferred-alternative j * LOOP 1 I f candidate set > 0 Move to a l t e r n a t i v e k from candidate set Drop k from candidate set Move to a t t r i b u t e i 51 LOOP 2 I f a t t r i b u t e set > 0 read a t t r i b u t e value a t J read a t t r i b u t e value a i k subtract aLi from a i k giving difference (d±) r e t r i e v e weight (wt) for a t t r i b u t e ( a ^ combine di and wx giving widi  add w±dA to t o t a l difference score 0 w ^ store new t o t a l difference fi w ^ move to next a t t r i b u t e i + 1 Go to LOOP 2. Retrieve t o t a l difference I f t o t a l difference > 0 eliminate a l t e r n a t i v e j store a l t e r n a t i v e k i n current-preferred else eliminate a l t e r n a t i v e k Go to L00P1. Select j * as desired a l t e r n a t i v e Like the AC model t h i s model assumes that processing includes a set of con t r o l routines to monitor the status of the analysis and execution routines to a c t u a l l y make choices between a l t e r n a t i v e s . Storage requirements f o r the outcome of analysis are minimal since at any time only the current preferred a l t e r n a t i v e i s being c a r r i e d . Information about previously eliminated a l t e r n a t i v e s can be disregarded. Contrast t h i s with the AC model which 52 requires the storage of each outcome since no comparison i s formally made u n t i l a l l information has been evaluated. Thus from a con t r o l point of view t h i s model i s l e s s complex than the AC model. 4.1.2.2 Supporting the strategy Computational support would involve a function to take the difference between the values f o r each ati, a i k p a i r . I f a l l a t t r i b u t e s are numeric, t h i s d i f f e r e n c e function simply involves subtraction. I f the v a r i a b l e s are represented i n a q u a l i t a t i v e form, a more complex mechanism might be necessary. Other f a c i l i t i e s would be e s s e n t i a l l y the same as those i n the AC model i n terms of storage and r e t r i e v a l of weights and accumulation of t o t a l s . In terms of information display i t may be appropriate to have only two al t e r n a t i v e s shown at any time. A f a c i l i t y to eliminate a dominated a l t e r n a t i v e would l i k e l y be b e n e f i c i a l . Again the minimal support i n terms of task s t r u c t u r i n g could be provided without the e x p l i c i t use of a value function. Another s t r u c t u r i n g mechanism which may be u s e f u l i n implementing t h i s strategy would be the a b i l i t y to move a l t e r n a t i v e s together. Considering the general representation of a p r e f e r e n t i a l choice problem i n matrix form, the a b i l i t y to move rows or columns would l i k e l y ease some of the s t r a i n involved i n comparing a l t e r n a t i v e s . In a computer-based system t h i s would be e s p e c i a l l y important i f a l t e r n a t i v e s i , j could not be simultaneously displayed on a sing l e screen. Thus, basic task s t r u c t u r i n g support aids may prove us e f u l even without recourse to implementing value functions which act to formalise the preference structure. The previous two models, AC and AD, can be characterized by the f a c t that they lead to a complete analysis of any information set. AD i s s l i g h t l y less 53 demanding to implement since i t s storage requirements are not as large. Support f o r both those models involves developing techniques f o r assessing preferences i n order to make value trade-offs and aggregation of values. Although some basic elements can be introduced which may support task s t r u c t u r i n g these are r e l a t i v e l y minor. However, i t may be possible to encourage i n d i v i d u a l s to spend more time conducting t h i s analysis by f a c i l i t a t i n g the e l i m i n a t i o n strategies that are t y p i c a l l y adopted to pare down large problem spaces p r i o r to d e t a i l e d a n a l y s i s . We w i l l now turn to two e l i m i n a t i o n models, conjunctive and elimination, by aspects. Neither of these models assumes the use of d e t a i l e d preference structures for purposes of i n t e g r a t i o n nor do they generally r e s u l t i n a complete search of the problem space. As a r e s u l t they are thought to be less c o g n i t i v e l y demanding s t r a t e g i e s . 4.1.3 The conjunctive model The conjunctive model s p e c i f i e s that a l t e r n a t i v e s are evaluated against some minimum acceptance l e v e l along a l l a t t r i b u t e s . Any a l t e r n a t i v e which f a i l s to meet one of the threshold l e v e l s set f o r any of the a t t r i b u t e s w i l l automatically be dropped from consideration. Each a l t e r n a t i v e i s evaluated independently u n t i l i t i s e i t h e r eliminated from consideration or the evaluation i s completed. Evaluation generally stops when an a l t e r n a t i v e i s found which s a t i s f i e s a l l minimum c r i t e r i a . This makes the order i n which a l t e r n a t i v e s are searched extremely important to f i n a l s e l e c t i o n . This model i s obviously noncompensatory since a l t e r n a t i v e s which are below thresholds on a p a r t i c u l a r a t t r i b u t e are immediately excluded from further consideration. This i s a basic c h a r a c t e r i s t i c of a l l f i l t e r i n g models. Each evaluation i s 54 independent; no comparisons are made between a l t e r n a t i v e s . This independence assumes that the threshold values are ex t e r n a l l y generated and are not dependent upon values of actual a l t e r n a t i v e s contained i n the choice set. 4.1.3.1 Description of the strategy More formally, analysis under a conjunctive model would proceed as follows: LOOP 1 I f number of nonevaluated a l t e r n a t i v e s = 0 reset threshold values (c) store new threshold values (c) move to a l t e r n a t i v e j (at random) from candidate set drop j from candidate set else move to a l t e r n a t i v e j (at random) from candidate set drop j from candidate set LOOP 2 I f number of nonevaluated a t t r i b u t e s = 0 for a l t e r n a t i v e j . se l e c t a l t e r n a t i v e j stop, else move to next a t t r i b u t e read a ^ re t r i e v e threshold value cL  compare a ^ to cL i f aLi > cL go to LOOP 2 55 else eliminate a l t e r n a t i v e j go to loop 1. This model i s i n t e r e s t i n g i n a number of respects. F i r s t , i t makes no demand on long term memory for the storage of the r e s u l t s of a n a l y s i s . I t also requires no t r a n s l a t i o n or conversion of data values from the current representation. The concept of preference s t r u c t u r i n g i s very weak i n t h i s model. I t i s s i m i l a r to the idea of l e x i c o g r a p h i c a l ordering (Keeney and R a i f f a ) . S t r a i n reduction from using the conjunctive model i s f a c i l i t a t e d i n two ways. F i r s t , each a t t r i b u t e evaluation i s r e l a t i v e l y simple r e q u i r i n g at most three EIPs i n contrast to nine for AC and seven f o r AD. Second, t h i s strategy leads to s a t i s f i c i n g i n that the f i r s t a l t e r n a t i v e which meets the minimum c r i t e r i a i s chosen; thus, less information i s l i k e l y to be processed i n t o t a l . The only condition i n which every element i n the search space would be examined i s i f a l l a l t e r n a t i v e s were rejected based upon the l a s t a t t r i b u t e a v a i l a b l e f o r each a l t e r n a t i v e , an u n l i k e l y outcome. (One might also assume that a "smart" system, human or machine, would adapt to t h i s condition by beginning the evaluation with the v a r i a b l e which has been most c r i t i c a l to the d e c i s i o n over a number of t r i a l s . ) 4.1.3.2 Supporting the strategy I n t e r e s t i n g l y , though t h i s model i s r e l a t i v e l y simple from a computational point of view, considerable support can be provided f o r i t . While e x p l i c i t conversion to u t i l i t i e s i s seldom done and i s d i f f i c u l t to evoke, i t would l i k e l y be reasonably easy to have people s p e c i f y threshold c r i t e r i a f o r 56 various a t t r i b u t e s . Once these have been e l i c i t e d i t would be r e l a t i v e l y t r i v i a l to provide an automatic f i l t e r i n g mechanism to make eliminations based on t h i s strategy. Also, such an approach could permit d e c i s i o n makers to break away from the s a t i s f i c i n g implications of the model by presenting a set of f e a s i b l e a l t e r n a t i v e s rather than choosing the f i r s t acceptable so l u t i o n . For the f e a s i b l e set i t would be possible to r e s p e c i f y the c r i t e r i a vector to recompute a new, presumably reduced, f e a s i b l e set. M u l t i p l e i t e r a t i o n s of such a strategy would lead to much more r e f i n e d solutions, at reasonably low cost to the d e c i s i o n maker. A l t e r n a t i v e l y , such an approach could be used to f i l t e r the number of a l t e r n a t i v e s to a more manageable l e v e l so that the more comprehensive additive strategies could be invoked. I t i s possible that the savings i n cognitive e f f o r t which r e s u l t from f a c i l i t a t i n g the elimination stage could lead to a r e l a t i v e l y greater expenditure of resources on the subsequent additive stage of problem solving. This has the p o t e n t i a l to improve the q u a l i t y of f i n a l choices. The f a c t that r e l a t i v e l y complex choice problem are completed i n stages has been reported by Biggs (1978), Olshavsky (1979) and Bettman and Park (1984). I t has also been indi c a t e d that the stages move from eli m i n a t i o n to additive evaluation. P a r t i a l support could be provided f o r the conjunctive strategy i n a number of ways. Most obvious would be the use of a command to delete a l t e r n a t i v e s from the set of a v a i l a b l e choices. This would s i m p l i f y the tracking of a l t e r n a t i v e s during analysis, i n essence acting as a memory ai d . In addition one might wish to reorder c r i t e r i a based on t h e i r p o t e n t i a l to reduce the search space (or t h e i r importance to the d e c i s i o n maker) to examine the most relevant c r i t e r i a f i r s t . Also, simple sort f a c i l i t i e s may help to more quickly i d e n t i f y promising a l t e r n a t i v e s for further evaluation. 57 Information display for the conjunctive model would probably be best given by presenting one a l t e r n a t i v e at a time. The c r i t e r i a vector might also be presented to allow f o r the manual assessment of d i f f e r e n c e s . Further, i t may be i n t e r e s t i n g to develop a system which "learns" the c r i t e r i a values over time by examining a subject's behaviour. The system could then provide e x p l i c i t feedback about the subject's d e c i s i o n behaviour, p o i n t i n g out possible inconsistencies. Once the procedure had s t a b i l i z e d , automatic f i l t e r i n g could be implemented. In short, though the conjunctive strategy i s one of the le s s demanding decision strategies there i s considerable p o t e n t i a l to support and, indeed, augment such a strategy at l i t t l e cost to the decision maker. 4.1.4 The E l i m i n a t i o n by aspects model The f i n a l d e c i s i o n model to be considered i s the Elimination-by-Aspects (EBA) model (Tversky, 1973). The EBA model assumes that an i n d i v i d u a l weights a l l relevant a t t r i b u t e s according to t h e i r importance. The a t t r i b u t e s are then sampled according to t h e i r weights and compared f o r a l l a l t e r n a t i v e s against some threshold l e v e l . Any a l t e r n a t i v e which does not meet the threshold l e v e l i s eliminated and a l l remaining a l t e r n a t i v e s are evaluated against the next a t t r i b u t e sampled. The sampling i s accomplished by choosing an a t t r i b u t e from the set with a l i k e l i h o o d according to i t s weight or importance as an evaluation c r i t e r i a (for example, i n choosing an apartment rent i s l i k e l y to be a more highly weighted f a c t o r than brightness and as a r e s u l t i s more l i k e l y to be chosen f i r s t as an e l i m i n a t i o n c r i t e r i a ) . This procedure i s i t e r a t e d u n t i l the best possible a l t e r n a t i v e i s ascertained. Since the model involves elimination of any a l t e r n a t i v e which i s below some 58 s p e c i f i e d l e v e l on a given a t t r i b u t e , i t i s noncompensatory. Since the analysis occurs across a l t e r n a t i v e s , i t i s not independent. EBA i s very s i m i l a r to the conjunctive model except that where the conjunctive model can be characterized as a depth f i r s t search analysing i n d i v i d u a l a l t e r n a t i v e s , EBA proceeds breadth f i r s t across a l t e r n a t i v e s . 4.1.4.1 Formal d e s c r i p t i o n of strategy Formally the implementation of a pure EBA strategy would proceed as follows: LOOP 1 I f number of al t e r n a t i v e s > 1 move to an a t t r i b u t e a A from the set of a t t r i b u t e s r e t r i e v e threshold value cL (associated with a t t r i b u t e a t) Else Select remaining a l t e r n a t i v e . LOOP 2 I f number of nonevaluated a l t e r n a t i v e s f o r a t t r i b u t e aL > 0 move to a t t r i b u t e value a ± j ( a t t r i b u t e i f o r a l t e r n a t i v e j ) read aii compare a t j to cL i f aLi < C i eliminate a l t e r n a t i v e j go to LOOP 2 Else go to LOOP 1 Note that t h i s model assumes that the i n d i v i d u a l has ordered the c r i t e r i a according to i t s importance and selects them i n t h i s way. The process involved 59 i n formulating t h i s rank order i s ignored; i n the same way f o r the additive models we d i d not consider the development of preference structures. Similar to the conjunctive model, EBA puts no demands on LTM f o r the storage of r e s u l t s other than the need to remember which a l t e r n a t i v e s are s t i l l a c t i v e l y being considered. F i n a l s e l e c t i o n occurs when only one a l t e r n a t i v e remains i n consideration, i . e . , choice occurs by default. The computational complexity involved i n analysis i s exactly the same as that i n the .conjunctive model--the only difference being i n the flow of c o n t r o l . Assuming a vector of threshold or c r i t e r i o n values i t would be possible to implement a f u l l y automated version of EBA which would be very s i m i l a r to the multiple i t e r a t i o n conjunctive model described e a r l i e r . The f a c t that i n t e r n a l processing occurred down a t t r i b u t e s (conjunctive) or across a l t e r n a t i v e s (EBA) would be of l i t t l e consequence i f analysis proceeded to narrow the choice set to a sin g l e a l t e r n a t i v e . 4.1.4.2 Supporting the strategy P a r t i a l support of t h i s d ecision strategy could be provided by the mechanisms described below. A procedure to rank order a t t r i b u t e s according to t h e i r importance w i l l a i d i n problem s t r u c t u r i n g . The s e l e c t i o n of a t t r i b u t e s i s presumed to be c o n t r o l l e d by sampling from a d i s t r i b u t i o n i n which the p r o b a b i l i t y of choosing the given a t t r i b u t e s i s a function of the weight or importance that the d e c i s i o n maker attaches to the a t t r i b u t e . Such a scheme can be roughly approximated by rank ordering the a t t r i b u t e s from most to l e a s t important. A procedure to eliminate or r e t a i n a l t e r n a t i v e s contingent upon comparison to some threshold l e v e l would also support t h i s strategy, reducing the need to 60 d i r e c t l y invoke comparison and e l i m i n a t i o n operations. This would work i n a simpler (from a human point of view) fashion than the same techniques i n the conjunctive model since only a si n g l e c r i t e r i a value rather a vector would be required to conduct a si n g l e t e s t . This would allow f o r p a r t i a l f i l t e r i n g along a number of c r i t i c a l dimensions p r i o r to engaging i n a more complete an a l y s i s . In terms of information display i t i s probably best to see a l l a l t e r n a t i v e s simultaneously, though perhaps f o r only a si n g l e a t t r i b u t e at a time. Of course, f o r problems of reasonable s i z e (for example, greater than ten a l t e r n a t i v e s ) presentation on a si n g l e CRT screen becomes i n f e a s i b l e . 4.2 Summary of proposed support mechanisms Table 4.1 outlines the set of d e c i s i o n aids associated with the decision models described i n the previous section. Most of the support applies, not s u r p r i s i n g l y , to computational areas, where the computer has a d i s t i n c t advantage i n terms of both speed of a p p l i c a t i o n and consistency. Some support can be provided through data storage, though on the whole t h i s i s l i k e l y the most l i m i t e d area (though i t should be noted that the r e l a t i v e l y large storage capacity and f a s t r e t r i e v a l speed of computer based systems provide the opportunity to consider a far wider range of a l t e r n a t i v e s than i s t y p i c a l for p r e f e r e n t i a l choice problems). The support provided v i a information display i s l i k e l y to have considerable impact, as there i s evidence that d i f f e r e n t displays promote the use of d i f f e r e n t search s t r a t e g i e s (Bettman and Kakkar, 1977; Bettman and Zins, 1979; Jarvenpaa, 1987). The two columns, a p p l i c a b i l i t y and cost, i n Table 4.1 attempt to provide some i n d i c a t i o n of the r e l a t i v e strength of support that each functions 61 provides, as well as the l i k e l y l e v e l of cost to the de c i s i o n maker of employing that function were i t b u i l t into a DSS. This s e c t i o n w i l l b r i e f l y discuss the various support tools and provide a r a t i o n a l e f o r the assessments made i n Table 4.1. I t should be noted that these cost measures are simply based upon the researcher's best guess and are not grounded i n any s o l i d empirical or t h e o r e t i c a l base. I t i s f o r t h i s reason that the ratings are given only on a r e l a t i v e l y crude nominal scale. The discussion w i l l focus on the three areas mentioned previously, namely support v i a computational aids, storage aids, and information presentation. 4.2.1 Computational support Functions f o r the extraction of preference structures would apply only to the AD and AC models as these are the only ones which require conversion from current base values using a value function for purposes of in t e g r a t i o n . I f a preference function can be accurately derived, i t i s a r e l a t i v e l y straightforward matter to apply the function to a set of a l t e r n a t i v e s and choose one that maximises the value of the preference function. Presumably these functions are the key to making choices f o r the addit i v e s t r a t e g i e s . In t h i s regard t h e i r a p p l i c a b i l i t y or " u t i l i t y " i s high. At the same time, the cost to the user of going through a procedure designed to extract a u t i l i t y function i s l i k e l y r e l a t i v e l y high. People are forced to think about problems i n a manner which i s not natural f o r them ( i . e . , i n terms of differences between gambles) . For many people t h i s can lead to r e j e c t i o n of the system (Humphreys and McFadden, 1980). In addi t i o n these procedures are time consuming and complex, thus r e q u i r i n g a high l e v e l of commitment and cognitive e f f o r t by the dec i s i o n maker. Ultimately, the need to make e x p l i c i t , 62 processes which take place unconsciously, i f at a l l , may well be more c o g n i t i v e l y demanding than implementing such a strategy without a decision ai d . In t h i s regard, the use of such support functions would l i k e l y be most appropriate i n s i t u a t i o n s where: 1. the d e c i s i o n i s l i k e l y to be highly r e p e t i t i v e and the preference function can be assumed constant over time, and 2. the d e c i s i o n i s a highly important, one time occurrence (e.g. the choice of s i t e s f or LNG plants (Keeney, 1980)). For more mundane, everyday problems the l e v e l s of e f f o r t expended would l i k e l y not be compensated by perceived increases i n d e c i s i o n q u a l i t y . In short, d e c i s i o n aids of t h i s nature, while perhaps hi g h l y applicable to the problem, are simply too c o g n i t i v e l y demanding to be implemented i n p r a c t i c a l s i t u a t i o n s . Such an a s s e r t i o n could perhaps be tested by embedding such preference e l i c i t a t i o n functions into a system which contained support f o r other decision s t r a t e g i e s as well and observing the usage rates of various functions. Manipulations of task v a r i a b l e s , such as s i g n i f i c a n c e , would also be of i n t e r e s t i n terms of t h e i r i n t e r a c t i o n with s t r a t e g i e s used. Arithmetic operators are applicable to AD and AC f o r the maintenance of running t o t a l s and for taking differences between a t t r i b u t e p a i r s . Since no arithmetic computation i s performed i n e i t h e r conjunctive or EBA s t r a t e g i e s , t h i s set of functions i s not applicable. There i s one arithmetic computation per a t t r i b u t e i n the additive compensatory strategy and two f o r additive d i f f e r e n c e , taking the difference between the a t t r i b u t e p a i r and adding t h i s d i f f e r e n c e to the t o t a l f or the current comparison; thus, t h i s function could provide a moderate l e v e l of support. Presumably, the l e v e l of support would be an increasing function of the s i z e of the problem space. Since most people 63 are aware of how basic arithmetic operations function and are used to automated versions of these procedures the cost of use would be low. A comparison function i s useful i n the AD strategy f o r making pairwise assessments between a l t e r n a t i v e s . In both e l i m i n a t i o n based strategies comparison would be against an e x t e r n a l l y determined standard. This function would presumably be a control point with a s p e c i f i c a c t i o n (e.g., eliminate a l t e r n a t i v e ) being taken based on the r e s u l t of the comparison. Such a function i s r e l a t i v e l y easy to implement and i s often i m p l i c i t l y embedded i n other functions such as eliminate or f i l t e r . In t h i s sense the use of t h i s mechanism i s v i r t u a l l y c o s t l e s s . Total cost of the compare operation i s dependent upon the model with which i t i s associated. For the conjunctive and AD s t r a t e g i e s one compare operator needs to be invoked f o r each a t t r i b u t e or a t t r i b u t e p a i r considered. In the case of EBA, one compare can be applied, from the d e c i s i o n maker's point of view, simultaneously across a l t e r n a t i v e s ; t h i s may r a d i c a l l y s h i f t the cost-benefit tradeoff, p a r t i c u l a r l y for large problems. A S e l e c t i o n function could be used with the AC strategy to choose, from the stored t o t a l s , the r e s u l t with the highest preference value. Such an approach i s r e l a t i v e l y easy to use and thus of r e l a t i v e l y low cost. The b e n e f i t s accruing are also low with a p a r t i c u l a r advantage coming f o r very large a l t e r n a t i v e sets. For the conjunctive and EBA s t r a t e g i e s , s e l e c t i o n may be applicable as a mechanism for determining ranges of values which may provide input for e s t a b l i s h i n g the i n i t i a l threshold values. Rank ordering of a t t r i b u t e s on a l t e r n a t i v e s i n terms of importance may be u s e f u l i n a l l models. In absence of complete support f o r the AC or AD s t r a t e g i e s , the arrangement of a l t e r n a t i v e s i n order of preference may help to 64 focus analysis. In the elimination strategies rank-ordering along a t t r i b u t e s may be h e l p f u l to focus the exploration on more c r i t i c a l a t t r i b u t e s at the s t a r t of a n a l y s i s . The c r i t i c a l a t t r i b u t e can be thought of as one which more quickly reduces the search space. Such an approach i s r e l a t i v e l y c o s t l y to the d e c i s i o n maker since i t would require an e x p l i c i t statement as to the importance of various a t t r i b u t e s . This i s s i m i l a r to, though les s demanding than, the expression of a preference structure based on a value function. I t requires only that the decision maker be able to order preferences f o r the various a t t r i b u t e s . Trade-off values between a t t r i b u t e s need not be considered. Elim i n a t i n g dominated a l t e r n a t i v e s should perhaps be the f i r s t pass taken i n any choice problem. I t has l i m i t e d value, since f o r any r e a l world multi-a t t r i b u t e problem, complete dominance i s u n l i k e l y to occur. As the number of a t t r i b u t e s increases, dominance i s even les s l i k e l y to be a p a r t i c u l a r l y u s e f u l t o o l f or reducing the search space. The e l i c i t a t i o n of a threshold vector i s s i m i l a r to, though much simpler than, determining a preferences structure which s p e c i f i e s precise trade-offs between a t t r i b u t e s . Here the decision maker need only consider what the minimum acceptable value of a t t r i b u t e i w i l l be, not the value of a t t r i b u t e i r e l a t i v e to a t t r i b u t e j . Since the decisions involved i n s p e c i f y i n g thresholds are much coarser than those required to s p e c i f y trade-offs, they should be simpler to implement. The use of threshold vectors has high value i n e i t h e r of the f i l t e r i n g strategies as i t can form the basis of an automated s e l e c t i o n mechanism. Determining the precise nature of the threshold values i s l i k e l y to be r e l a t i v e l y demanding upon de c i s i o n makers. One strategy may be to extract a band of values for each a t t r i b u t e rather than a s p e c i f i c point 65 estimate. Also, f o r di s c r e t e a t t r i b u t e s , such as c o l o r or s i z e , either acceptable or unacceptable values could be enumerated. Another p o s s i b i l i t y would be to prompt the decision maker by asking a serie s of binary questions about each a t t r i b u t e to determine t h e i r c r i t e r i a values. This simply aids the de c i s i o n maker i n searching through the range of values a v a i l a b l e f o r a given a t t r i b u t e . Pairwise comparisons could also be used i n t e s t i n g f o r inconsistencies i n assessments to help the de c i s i o n maker a r r i v e at a "truer" set of preferences. This could be accomplished by showing the de c i s i o n maker pa i r s of m u l t i - a t t r i b u t e a l t e r n a t i v e s and observing the trade-offs being made. Such decisions may show where i n i t i a l expressions of thresholds are being v i o l a t e d and may indicate why. Obviously, the more complex such procedures become the more c o s t l y they are for the decision maker. F i l t e r i n g based on threshold values would be a powerful t o o l for elimi n a t i n g a l t e r n a t i v e s with l i t t l e o v e r a l l cost to the user. The function would simply scan the o r i g i n a l data base and eliminate a l t e r n a t i v e s based on established c r i t e r i a . This function could be e a s i l y invoked with l i t t l e c ognitive cost (assuming the threshold vector i s established) and would have a high payoff i n terms of i t s a b i l i t y to r e l i e v e the user of the burden of many comparison-elimination operations. The possible drawback of t h i s approach i s the p o t e n t i a l user uneasiness at giving over a large degree of c o n t r o l , f o r at l e a s t the i n i t i a l stages of the choice process, to an automated system. Their wi l l i n g n e s s to do t h i s would l i k e l y be a function of t h e i r confidence i n the i n i t i a l vector of thresholds which they had established. Individuals comfortable with such procedures may also be those who look favorably on u t i l i t y based analysis. A more conservative approach to supporting the EBA and conjunctive 66 s t r a t e g i e s would be the use of an eliminate (or drop) on co n d i t i o n function. In t h i s case, the decision maker would speci f y a s i n g l e value f o r a single a t t r i b u t e and the system would scan and f i l t e r the database. Such an approach i f c a r r i e d through f o r a l l a t t r i b u t e s would achieve the same r e s u l t as an automatic f i l t e r . The cognitive cost of such an approach would be higher since i t requires multiple executions of the command, once f o r each a t t r i b u t e i n the search space for which the decision maker can s p e c i f y a threshold value. Another, even more basic approach to supporting the noncompensatory st r a t e g i e s , i s to provide a sort f a c i l i t y . By s o r t i n g the a l t e r n a t i v e s on a p a r t i c u l a r a t t r i b u t e , a d e c i s i o n maker who i s manually scanning the data could know when the point i s reached where a l l remaining a l t e r n a t i v e s are acceptable (or unacceptable) without a c t u a l l y examining these values. This i s b a s i c a l l y a weak form of support for the e l i m i n a t i o n procedures. Its value i n a s s i s t i n g the choice process i s somewhat lower than ei t h e r the f i l t e r or drop functions as i t does not e x p l i c i t l y reduce the search space. Rather, i t indicates to the d e c i s i o n maker which set of a l t e r n a t i v e s are acceptable, assuming the d e c i s i o n maker has a set of c u t - o f f values i n mind. The cognitive load attached to using sort as the basis f o r an elimination strategy i s c e r t a i n l y high compared to other f i l t e r i n g mechanisms with the compare and eliminate functions being assumed by the d e c i s i o n maker. However, t h i s simple s t r u c t u r i n g of the problem eliminates the need f o r the examination of a t t r i b u t e values which ei t h e r meet or f a i l to meet the desired c r i t e r i a . A f i n a l feature which may be useful i n supporting any of the strategies where a l t e r n a t i v e s are dropped from consideration p r i o r to choice i s a restore function. The use of a p a r t i c u l a r strategy, for example invoking a s e r i e s of 67 e l i m i n a t i o n (drop) functions, could lead to the e l i m i n a t i o n of a l l a l t e r n a t i v e s . I f a choice must be made from t h i s set, then i t i s necessary to be able to restore a l t e r n a t i v e s . This obviously provides l i t t l e assistance i n a c t u a l l y making a choice; i t does, however, protect the d e c i s i o n maker from l o s i n g desired or necessary information. S i m i l a r l y , a t t r i b u t e s which may have been eliminated as unimportant e a r l y i n the dec i s i o n process may need to be restored to make a f i n a l choice, p a r t i c u l a r l y i n s i t u a t i o n s where there i s a high degree of s i m i l a r i t y between a l t e r n a t i v e s . A l l of the functions above are intended to provide support to the decision maker during the choice process. They do t h i s by assuming the r e s p o n s i b i l i t y f o r one or more of the cognitive operations involved i n implementing a strategy. In a l l cases, f i n a l choice i s to be l e f t with the d e c i s i o n maker. Two of the functions, u t i l i t y preference and f i l t e r , could provide a single f i n a l choice to the dec i s i o n maker; however, backtracking from these should be possible to present a set of most desirable a l t e r n a t i v e s . Most of the functions which are both valuable and easy to implement serve to reduce the size of the search space. I t i s assumed that the f i n a l search among a reduced set of a l t e r n a t i v e s requires the judgement of the de c i s i o n maker and w i l l not be automated. Attempts to d i r e c t l y model or support t h i s process require the use of a preference e l i c i t a t i o n approach which has proven to be both cumbersome, time-consuming and unappealing f o r many de c i s i o n makers. The objective i n developing support mechanisms i n t h i s area should be on supporting the i n i t i a l s t r u c t u r i n g of the problem and i n reduction of the search space to f a c i l i t a t e the f i n a l i n d i v i d u a l judgement. By a l l e v i a t i n g the burden placed on the dec i s i o n maker to reduce the search space, i t i s conceivable that the freed resources could be applied to conducting a more 68 thorough analysis of the remaining a l t e r n a t i v e s . On the other hand, the development of a series of tools such as those described above may simply increase d e c i s i o n making e f f i c i e n c y . 4.2.2 Data storage Aside from a set of functional c a p a b i l i t i e s f o r the s t r u c t u r i n g and manipulation of data within the context of the search space,' a d d i t i o n a l support may be possible by a l l e v i a t i n g storage demands placed on the decision maker's long term memory. Registering information i n long term memory i s one of the more demanding elementary information processing tasks with storage time on the order of tens of seconds (Newell and Simon, 1971). R e t r i e v a l of information from long term memory i s reasonably f a s t but f a l l i b l e . The storage of intermediate r e s u l t s would be of value i n both additive s t r a t e g i e s . Both presumably carry an accumulated score f o r the current, active a l t e r n a t i v e (or p a i r of al t e r n a t i v e s i n the case of AD) . In addition AC would maintain a l i s t of t o t a l scores f o r each a l t e r n a t i v e i n the search space. Automatic storage of the r e s u l t s of the a n a l y t i c a l functions used above, i n t h i s case the preference function and arithmetic operators, r e l i e v e s the d e c i s i o n maker of what i s e s s e n t i a l l y a bookkeeping task. This may serve to free resources for more task oriented tasks. I t i s a simple matter to store the preference function or the threshold vector once they have been s p e c i f i e d . These values can be automatically integrated into the various functions. Hence, the s t r a i n of using t h i s f a c i l i t y i s very low. In f a c t such features may be completely transparent to the users of the system. S i m i l a r l y , the storage of eliminated a l t e r n a t i v e s r e l i e v e s the decision 69 maker of the need to r e c a l l information which may be needed i n the event that an e l i m i n a t i o n strategy r e s u l t s i n no s o l u t i o n . In general, the storage functions w i l l have low a p p l i c a b i l i t y i n terms of t h e i r impact on the choice process. At the same time, since such f a c i l i t i e s can be automatically provided without any intervention by the d e c i s i o n maker, they are very low i n terms of cognitive cost and thus l i k e l y to have a p o s i t i v e impact on the decision process. 4.2.3 Information display The f i n a l area where we can consider ways to support the various s t r a t e g i e s i s i n the area of information display. For e i t h e r the AC or conjunctive model evaluations are performed independently f o r each a l t e r n a t i v e along i t s a t t r i b u t e s . For t h i s reason, i t i s l i k e l y that display of a l l information about one a l t e r n a t i v e at a time would be preferable. Though the strategy could be employed equally i f multiple a l t e r n a t i v e s are present, these may serve to d i s t r a c t a t t e n t i o n and could lead to errors i n a n a l y s i s . Two data display f a c i l i t i e s may be useful for dealing with the AD model i n a computer supported environment. F i r s t , because the strategy implies the pairwise evaluation of a l t e r n a t i v e s along a t t r i b u t e s i t i s desirable to f u l l y d i s p l a y two a l t e r n a t i v e s at a time. This minimises the p o t e n t i a l f o r error due to d i s t r a c t i o n and the possible reading of i n c o r r e c t values into the evaluation. The second issue associated with the AD model i s the a b i l i t y to move a l t e r n a t i v e s so that adjacent p a i r s are being compared. Such a f a c i l i t y i s p a r t i c u l a r l y necessary i n a computer-based environment since l i m i t a t i o n s i n screen s i z e mean that only small problems can be displayed on a s i n g l e screen. Windowing back and f o r t h to examine a l t e r n a t i v e s may put a severe s t r a i n on 70 short term memory. Thus, the function i s l i k e l y to be p a r t i c u l a r l y valuable since short term memory i s thought to be an important con s t r a i n t on human information processing. For the EBA model analysis presumably proceeds across a l t e r n a t i v e s f o r a given a t t r i b u t e . This implies that the desired data d i s p l a y involves the complete presentation of values f o r a given a t t r i b u t e . In order to implement each of these presentation formats f o r the given strategy i t i s necessary to know a p r i o r i which strategy i s to be used. One could simply ask the decision maker, though t h i s i s not p a r t i c u l a r l y s a t i s f y i n g since i t i s u n l i k e l y that he or she w i l l be able to specify an approach to the problem. Also, i t must be remembered that these strategies w i l l l i k e l y evolve over the course of a choice exercise (Bettman and Park, 1981; Olshavsky, 1979); thus, no sing l e presentation format w i l l be appropriate. A simple recommendation to cope with t h i s problem i s to have t o t a l l y f l e x i b l e presentations where decision makers can open or close p a r t i c u l a r rows or columns i n the search space. This allows the presentation to adapt to evolving strategies over time. Feedback mechanisms could also be adapted into the DSS to help point out inconsistencies, e s p e c i a l l y i n the form of i n t r a n s i t i v i t y , which may a r i s e for a given choice problem. Feedback i s not fundamental to the support of the dec i s i o n process but may serve to make dec i s i o n makers more aware of t h e i r performance and the biases inherent i n the way they process information. Remus and Kotteman (1986) reports that such cognitive feedback has been useful for DSS users working on production scheduling problems. 71 4.3 Summary This chapter makes some assertions about the value of providing various types of d e c i s i o n aids to p r e f e r e n t i a l choice problems. I t has been argued that the l a r g e s t b e n e f i t can be derived from providing computational procedures which take over r e s p o n s i b i l i t y f o r a number of the elementary information processing tasks facing a d e c i s i o n maker. Associated with each of these procedures i s a cost of use or i n t e r a c t i o n f o r the d e c i s i o n maker. Following from the cost b e n e f i t argument made i n chapter 3 the b e n e f i t s that accrue from lowering the cognitive load must be greater than the cost of invoking the procedure. This i s one possible explanation as to why highly valuable procedures such as u t i l i t y assessment are not widely employed. Obviously, i t i s also necessary to point out that most of the above are a ser i e s of deductions drawn from a basic argument about how i n d i v i d u a l s process information. The recommendations as to what should be included i n a DSS for p r e f e r e n t i a l choice problems are not drawn from a c o l l e c t i o n of empirical work i n t h i s area as there simply i s no such body of l i t e r a t u r e . The next chapter w i l l examine, i n general terms, how a decision maker might employ a decision a i d which incorporates many of the e f f o r t reducing p r i n c i p l e s developed i n the t h i s chapter. I t w i l l propose a se r i e s of testable hypotheses which p r e d i c t the behaviour of decision makers with the DSS under two sets of assumptions, one focussing on accuracy maximisation and the other taking an e f f o r t minimisation perspective. 72 CHAPTER 5- PROPOSITIONS AND HYPOTHESES 5.0 Introduction In t h i s chapter a set of propositions and two sets of d e t a i l e d competing hypotheses are developed to describe the p o t e n t i a l impact of a DSS, designed according to the s p e c i f i c a t i o n s i n Chapter 4, on the process of decision making. The hypotheses are derived by examining both the e f f o r t and accuracy perspectives outlined i n Chapter 3. These hypotheses which are tested i n the experiments described i n Chapter 6 and reported i n Chapter 7. 5.1 Propositions The components of the decision a i d recommended i n the previous section are drawn from a basic theory of cognitive processing. I t i s argued that a l t e r i n g the load on the processor may change the way decisions are made. These changes i n process may r e f l e c t a focus on e i t h e r e f f o r t or accuracy i n d e c i s i o n making as outlined i n Chapter 3. This may lead to more information use and p o s s i b l y improved de c i s i o n processes i n p r e f e r e n t i a l choice s i t u a t i o n s . A l t e r n a t i v e l y the impact of the system may simply be evidenced by more e f f i c i e n t processes. Of course, whether or not a system would a l t e r d e c i s i o n making i n choice problems, and what impact such a system might have on information processing, i s an open question which requires empirical i n v e s t i g a t i o n . The focus of t h i s work i s on how the use of d e c i s i o n aids change decision processes and a l t e r cognitive load. I t i s not c l e a r exactly how the use of 73 these tools w i l l impact the choice process. The way i n which hypotheses are formed, i n large measure, depends on the p a r t i c u l a r perspective taken about the objectives and the a d a p t a b i l i t y of d e c i s i o n makers. While i t i s reasonably well established that decision makers are adaptive, making trade-offs between e f f o r t and accuracy, i t i s unclear how each of these i s valued. That i s , does the d e c i s i o n maker p r i m a r i l y focus on minimising e f f o r t , or i s the primary emphasis on making the best possible decision? The cost-benefit framework does not p r e c i s e l y s p e c i f y the r e l a t i o n s h i p between various problem c h a r a c t e r i s t i c s , strategy s e l e c t i o n , e f f o r t and accuracy. This implies that we may develop hypotheses i n which e f f o r t i s the key determinant of strategy and also l o g i c a l l y , hypotheses which emphasise accuracy as the key to strategy s e l e c t i o n . These two sets of hypotheses w i l l have very d i f f e r e n t implications for the impact of the d e c i s i o n aid. Hypotheses centered around accuracy w i l l p r i m a r i l y be effectiveness oriented, while those centred around e f f o r t w i l l be e f f i c i e n c y r e l a t e d . In f a c t , since the cost-benefit framework has been developed from an empirical base which manipulated costs by increasing them, i t i s d i f f i c u l t to be c e r t a i n how the d e c i s i o n maker w i l l react to a reduction i n costs for implementing p a r t i c u l a r s t r a t e g i e s . I f we assume that decision makers operate r a t i o n a l l y w i t h i n the confines of the cost of information processing and place a primary emphasis on d e c i s i o n q u a l i t y , we would l i k e l y i n f e r that use of the d e c i s i o n a i d w i l l encourage the use of more complex strategies f o r a given d e c i s i o n s e t t i n g . Such " r a t i o n a l i t y " i s often assumed by those who support the cost-benefit model (see for example Junngermann, 1985; Johnson and Payne, 1986; March, 1977). From t h i s viewpoint the following general p r o p o s i t i o n could be made. 74 PI The use of a decision aid to support search strategies w i l l lead to the use of more exhaustive strategies which consider more information, emphasise additive information search and show use of f i l t e r i n g and elimination strategies compared to unaided problem solving. E s s e n t i a l l y PI asserts that the cost-benefit concepts form a u s e f u l basis f o r the design of computer based decision support systems i n a p r e f e r e n t i a l choice environment. DSS can be used to support d e c i s i o n making effectiveness. Such an a s s e r t i o n i s i n t u i t i v e l y appealing. I t indicates that as c a p a c i t i e s to process information are increased decision makers w i l l respond by doing a more thorough a n a l y s i s . A l t e r n a t i v e l y one might argue: PI' The use of a decision aid to support search strategies w i l l lead decision makers to work more effi c i e n t l y , using the least effortful strategy available that w i l l result i n an acceptable solution. This i s the e f f o r t minimisation argument. I f t h i s were the case, then the only outcome of developing DSS might be to increase d e c i s i o n e f f i c i e n c y . While improvement of both e f f i c i e n c y and effectiveness are u s e f u l goals, the focus i n DSS has c l e a r l y been on effectiveness (Keen and Scott-Morton, 1979) . Neither PI nor PI' f u l l y account for a l l the l i t e r a t u r e that was discussed i n Chapter 3. Each has a reasonable body of empirical support. Some would argue that e f f o r t considerations are paramount (see f o r example Stone, 1987; Russo and Dosher, 1983). However empirical evidence such as that provided by Payne (1976) and others indicates that more comprehensive st r a t e g i e s are employed f o r smaller problems while f i l t e r i n g s t r a t e g i e s are used f o r the larger, more demanding tasks. This i s d i r e c t support f o r p r o p o s i t i o n PI i n that comprehensive additive strategies are used when the d e c i s i o n maker i s not under heavy cognitive s t r a i n and he or she only reverts to elimination 75 s t r a t e g i e s as s t r a i n increases. Since both propositions can be supported based upon the available l i t e r a t u r e , hypotheses based upon both the e f f o r t minimisation and accuracy maximisation perspectives are developed. P r i o r to a d e t a i l e d presentation of the hypotheses i t would be useful to discuss the implications of the r e s u l t s r e l a t i n g to the two propositions. I t should be noted that PI and PI' cannot be treated as mutually exclusive; that i s i t i s possible that a given d e c i s i o n a i d w i l l contribute to both effectiveness and e f f i c i e n c y . Consider as an example the c a l c u l a t o r . Such a device allows the decision maker to perform complex computations very accurately, thus enhancing effectiveness over t r a d i t i o n a l p e n c i l and paper or mental approaches. At the same time, for any reasonably complex computation the use of a c a l c u l a t o r probably w i l l also r e s u l t i n more e f f i c i e n t ( i e . , faster) computation. In t h i s case the t o o l impacts on both dimensions. I f d e c i s i o n makers tended to favour one of these two a t t r i b u t e s of d e c i s i o n making, then they would perceive e i t h e r e f f i c i e n c y or effectiveness as the primary b e n e f i t of using a c a l c u l a t o r , though i n f a c t i t helped on both dimensions. Thus i n evaluating the impact of the d e c i s i o n a i d we should bear i n mind that i t may e f f e c t only e f f o r t , only accuracy, or improve both simultaneously. Table 5.1 shows a l l possible empirical r e s u l t s r e l a t i n g to the two propositions. We s h a l l review each of these four p o s s i b i l i t i e s i n turn. Outcome 1 r e f l e c t s the idea that the d e c i s i o n a i d works to improve effectiveness but not e f f i c i e n c y . This i s the t r a d i t i o n a l DSS argument. While some authors mention the possible impact of DSS on e f f i c i e n c y , they argue that t h i s i s l a r g e l y tangential and not the key focus of a DSS. A system should be designed to make the decision maker more e f f e c t i v e . I t should improve both the 76 outcome and process of decision making (Keen and Scott-Morton, 1978) . Such a r e s u l t would provide strong support f o r PI and give no r e a l support f o r PI' . In essence t h i s implies that a decision maker reinvests the cognitive resources freed when a DSS i s used i n order to improve the decision e f f e c t i v e n e s s . Outcome 2 s p e c i f i e s that the use of the a i d improves e f f i c i e n c y but does not improve effectiveness. Here the DSS helps the i n d i v i d u a l to make decisions more quickly, using l e s s resources. DSS use does not n e c e s s a r i l y r e s u l t i n better q u a l i t y decisions. This would support PI' and would not support PI. This i s a r e s u l t which i s often c i t e d i n case studies and other anecdotal reports of DSS use (see f o r example Scott-Morton, 1971 and Keen, 1981). The stated objective of DSS development may centre around eff e c t i v e n e s s , but the perceived b e n e f i t often appears to be e f f i c i e n c y . This may not be due to flawed DSS design per se but could be due to the underlying objectives of the deci s i o n maker. We often i m p l i c i t l y assume that, given the opportunity, people want to make better decisions. However, as the evidence presented i n Chapter 3 indicates people may be equally concerned with e f f o r t and e f f i c i e n c y and may promote these objectives over effectiveness. Thus, though a DSS may be capable of supporting improvements i n both areas, the dec i s i o n maker may opt f o r the le a s t e f f o r t approach, using the dec i s i o n a i d p r i m a r i l y as a work saving device. Outcome 3 implies the best of a l l possible worlds, providing support for both PI and PI'. This i s the r e s u l t noted i n our c a l c u l a t o r example above. I t may be expected to occur i n c e r t a i n "high leverage" s i t u a t i o n s where the dec i s i o n a i d a c t u a l l y a l t e r s the basic e f f o r t accuracy trade-off, making the more e f f e c t i v e approaches to the problem easier to implement than the 77 h e u r i s t i c s or manual approaches that were previously employed. In t h i s case even the e f f o r t minimiser would choose to use the more e f f e c t i v e s t r a t e g i e s as they would also be the more e f f i c i e n t approaches. Natural occurrence of such opportunities may be rare since i t seems i n t u i t i v e l y that there i s a natural trade-off between e f f o r t and accuracy. F i n a l l y , outcome 4 s p e c i f i e s that the d e c i s i o n a i d may have no impact on e i t h e r e f f i c i e n c y or effectiveness r e l a t e d c r i t e r i a , with no support for e i t h e r PI or PI'. Such a f i n d i n g would be c l e a r l y undesirable. I t s occurrence could most probably be explained i n one of two ways. The f i r s t i s simply that the d e c i s i o n was improperly designed and d i d not a c t u a l l y support the problem f o r which i t was tested. This i s an issue of construct v a l i d i t y . A decision a i d should appropriately operationalise support e i t h e r f o r the domain or theory on which i t i s based. I t may also be possible that the d e c i s i o n a i d i s a u s e f u l one but that users have not properly recognised how to e x p l o i t i t for the problem at hand. From an experimental point of view t h i s i s simply an issue of i n t e r n a l v a l i d i t y . The t r a i n i n g of subjects i n the use of the system i s necessary to ensure that they are f a m i l i a r with i t ' s functions. At the same time as part of t h i s t r a i n i n g i t may be desirable to point out the linkages between the system and various decision approaches or leave t h i s i n the hands of the user. At any rate, i t i s e s s e n t i a l that subjects understand the functions of the system and that s u f f i c i e n t checks are i n s t i t u t e d to ensure that the system i t s e l f does not impede problem solving. A second p o t e n t i a l explanation for outcome 4 i s that while the system may provide adequate support f o r some problem solving approach, strategy or theory, the approach i t s e l f may be a poor one. Any r e s u l t s i n t h i s case may be p e r f e c t l y legitimate and i n t e r e s t i n g from a t h e o r e t i c a l or research perspective i n that we would be 78 a b l e t o r u l e o u t a p a r t i c u l a r a p p r o a c h t o d e c i s i o n s u p p o r t . U n f o r t u n a t e l y such r e s u l t s a r e n o t p a r t i c u l a r l y u s e f u l f r o m a p r a g m a t i c p o i n t o f v i e w and a r e i n t h a t sense u n d e s i r a b l e . The f o u r outcomes s p e c i f i e d above r e p r e s e n t t h e p o s s i b l e c o n c l u s i o n s o f t e s t i n g b a s e d upon t h e p r o p o s i t i o n s P I and P I ' . Each has d i f f e r e n t i m p l i c a t i o n s f o r o u r p e r s p e c t i v e on t h e i n t e r a c t i o n between t h e d e c i s i o n maker and t h e d e c i s i o n a i d . As we w i l l see i n t h e c o n t e x t o f t h i s p a r t i c u l a r d e c i s i o n a i d , most o f t h e b e h a v i o u r s a s s o c i a t e d w i t h e f f e c t i v e n e s s o r i e n t e d p e r f o r m a n c e a r e e x c l u s i v e o f t h o s e w h i c h w o u l d be l i n k e d w i t h a d e c i s i o n maker who f o c u s s e s on e f f i c i e n c y . Thus i t i s a n t i c i p a t e d t h a t outcome 3 d e t a i l e d above i s u n l i k e l y i f t h e b a s i c t h e o r e t i c a l p e r s p e c t i v e o f t h e e f f o r t a c c u r a c y t r a d e - o f f i s c o r r e c t . A t t h e same ti m e t h e t e s t i n g done d u r i n g system development and t h e d i r e c t l i n k a g e t o b e h a v i o u r a l s t r a t e g i e s o f p r e f e r e n t i a l c h o i c e s h o u l d h e l p t o e n s u r e t h a t any r e s u l t s c e n t e r i n g a r o u n d outcome 4 would be due t o problems w i t h t h e u n d e r l y i n g t h e o r y o f e f f o r t a c c u r a c y t r a d e - o f f s . 5.2 An o v e r v i e w o f dependent and i n d e p e n d e n t v a r i a b l e s The dependent and i n d e p e n d e n t v a r i a b l e s f o r t h i s s t u d y a r e l i s t e d i n T a b l e 5.2. Two e x p e r i m e n t s a r e p r o p o s e d w h i c h w i l l examine t h e i m p a c t o f d e c i s i o n s u p p o r t i n t h e p r e f e r e n t i a l c h o i c e e n v i r o n m e n t . The i n d e p e n d e n t v a r i a b l e s a r e b r i e f l y o u t l i n e d h e r e . The p r i n c i p a l i n d e p e n d e n t v a r i a b l e o f i n t e r e s t i s the a v a i l a b i l i t y o f a d e c i s i o n a i d . A l s o o f i n t e r e s t i s how t h e d e c i s i o n a i d i n t e r a c t s w i t h p r o b l e m s i z e ( i . e . , t h e number o f a l t e r n a t i v e s f r o m w h i c h the d e c i s i o n maker must c h o o s e ) . Thus, t h e s i z e o f t h e p r o b l e m space i n terms o f th e number o f a l t e r n a t i v e c h o i c e s w i l l a l s o be m a n i p u l a t e d . The development o f t h e d e c i s i o n a i d i s d i s c u s s e d f u l l y i n C h a p t e r 6 and 79 the commands included i n the decision a i d are described i n Appendix 6.3. At t h i s point we w i l l b r i e f l y describe the o p e r a t i o n a l i s a t i o n of the a i d so that the hypotheses can be properly understood. The d e c i s i o n a i d developed f o r t h i s research follows reasonably c l o s e l y the design guidelines developed i n Chapter 4. The system includes a subset of the commands summarised i n Table 4.1. For the most part the elimination strategies are well supported, v i a such commands as drop, c o n d i t i o n a l drop, close and c o n d i t i o n a l close. Less support i s provided for the additive strategies; however, functions such as move, reorder and c a l c u l a t e have a c l e a r relevance to these approaches. In short the aided d e c i s i o n makers receive a system which has a series of commands b u i l t into i t which support various aspects of p r e f e r e n t i a l choice d e c i s i o n s t r a t e g i e s . I t i s not claimed that the system f u l l y automates, or supports these s t r a t e g i e s . Rather, the f a c i l i t i e s provided help to reduce the l e v e l of cognitive e f f o r t associated with the use of these strategies compared to the unaided treatment. In order to maintain equivalence between the treatment groups, the unaided group also uses a microcomputer based system. However, i n t h i s case the subject i s provided only with a command to access information ( i . e . , the open command). None of the other manipulation commands are a v a i l a b l e to the unaided d e c i s i o n maker. A f i n a l v a r i a b l e of i n t e r e s t pertains to the presentation of information. This v a r i a b l e deals with the number of display screens a d e c i s i o n maker needs to view i n order to examine a l l a v a i l a b l e information. Though t h i s v a r i a b l e has not been developed above, i t can be viewed as an extension of the complexity associated with problem s i z e . When a l l a l t e r n a t i v e s or a t t r i b u t e s of a problem cannot be viewed simultaneously, the problem becomes more d i f f i c u l t as an a d d i t i o n a l memory load i s placed upon the d e c i s i o n maker. Also 80 i n order to assess the e f f e c t s of problems of various s i z e , i t becomes necessary to move from sin g l e to multiple screen displays. The manipulations of screen s i z e conducted i n E l and E2 help to con t r o l f o r these e f f e c t s . In terms of dependent v a r i a b l e s , we are concerned with d e c i s i o n strategies and information use rather than outcomes. The primary r a t i o n a l e f o r t h i s i s that i n p r e f e r e n t i a l choice s i t u a t i o n s determination of an optimal s o l u t i o n depends on an i n d i v i d u a l ' s preference structure and thus i s extremely hard to measure. This i s not unusual i n the DSS domain. We have argued previously that i t i s e s s e n t i a l to focus on process i n order to determine the impact of a DSS i n a p a r t i c u l a r d ecision s e t t i n g (Todd and Benbasat, 1987). The experiments to be proposed here w i l l focus on information use and d e c i s i o n strategy. Measurements w i l l be taken i n three ways: computer logs of subject i n t e r a c t i o n with the DSS, concurrent verbal protocols, and a post experimental questionnaire. The data c o l l e c t e d w i l l allow f o r a determination of the amount of information being used and the type of strategy employed. (The precise nature of the dependent variables i s out l i n e d i n se c t i o n Chapter 6. At t h i s point we w i l l merely specify the hypotheses which are drawn from the development of Chapters 3 and 4 above). 5.3 Hypothesis development Two sets of hypotheses which are intended to p r e d i c t the impact of the deci s i o n a i d on information processing strategy are developed. The two groups of hypotheses are rooted i n the notion that d e c i s i o n makers make trade-offs between e f f o r t and accuracy i n choice tasks (see Chapter 3). One set of hypotheses takes the perspective that the i n d i v i d u a l acts to minimise e f f o r t subject to a constraint on the o v e r a l l q u a l i t y of the 81 s o l u t i o n . That i s , t h e d e c i s i o n m a k e r w i l l a d o p t t h e l e a s t e f f o r t f u l s t r a t e g y w h i c h l e a d s t o a n a c c e p t a b l e s o l u t i o n . Such a d e c i s i o n maker w i l l be refered to as the " e f f o r t minimiser" throughout the discussion. The second set of predictions i s based upon the premise that the decision maker acts as a constrained optimiser, t r y i n g to achieve the best possible s o l u t i o n within the confines of l i m i t e d information processing capacity. In short, t h e d e c i s i o n m a k e r w i l l s e l e c t a f e a s i b l e s t r a t e g y w h i c h y i e l d s t h e b e s t p o s s i b l e s o l u t i o n . This type of d e c i s i o n maker w i l l be r e f e r e d to as the "accuracy maximiser." I t should be noted i n discussing these two perspectives that we are t a l k i n g about p r o t o t y p i c a l modes of processing by i n d i v i d u a l s . While there i s empirical evidence to support both perspectives, not enough i s known to adequately determine whether one of these two processing approaches i s an i n v a r i a n t c h a r a c t e r i s t i c of a l l decision makers, i s contingent upon task c h a r a c t e r i s t i c s or, i s an i n d i v i d u a l difference v a r i a b l e . The discussion of the hypotheses focusses on the two perspectives as dichotomous and absolute. This i s not n e c e s s a r i l y the case. I t does, however, f a c i l i t a t e d i s c u s s i o n and t e s t i n g i n absence of a more r e f i n e d view of d e c i s i o n maker's actual objectives. Both of these perspectives on choice processes were elaborated i n Chapter 3. The two approaches lead to some d i s t i n c t predictions about the behaviour of i n d i v i d u a l s with a d e c i s i o n aid. The hypotheses w i l l be elaborated upon with reference to the l i t e r a t u r e discussed i n Chapter 3 and the d e c i s i o n a i d o u t l i n e d i n Chapter 4. The hypotheses are stated around the dependent measures of information use and d e c i s i o n process. Generally, main e f f e c t s are stated f o r the d e c i s i o n a i d 82 v e r s u s no a i d c o n d i t i o n . Where a p p r o p r i a t e , i n t e r a c t i o n s a r e s t a t e d f o r t h e d e c i s i o n a i d w i t h e i t h e r p r o b l e m s i z e o r t h e s c r e e n s i z e . T h i s r e s e a r c h f o c u s s e s on t h e i m p a c t o f t h e d e c i s i o n a i d ; t h u s , main e f f e c t s due t o p r o b l e m s i z e o r s c r e e n s i z e a r e n o t h y p o t h e s i s e d . The e f f e c t s o f p r o b l e m s i z e w i l l be t e s t e d m e r e l y as a v a l i d i t y check, t o see i f t h e r e s u l t s commonly f o u n d i n the p r e f e r e n t i a l c h o i c e l i t e r a t u r e c a n be r e p l i c a t e d . The s c r e e n v a r i a b l e a r o s e due t o f i n d i n g s w h i l e p i l o t t e s t i n g t h e system t h a t i n d i v i d u a l s w o u l d c u t problems down u n t i l t h e y c o u l d v i e w a l l a l t e r n a t i v e s on a s i n g l e s c r e e n . I n t h i s sense we have, a t p r e s e n t , no t h e o r e t i c a l i n t e r e s t i n s c r e e n s i z e and our i n v e s t i g a t i o n o f main e f f e c t s due t o t h i s v a r i a b l e w i l l be p u r e l y e x p l o r a t o r y . I n e s s e n c e , t h e n , t h e s c r e e n v a r i a b l e c a n be t r e a t e d as an e x p e r i m e n t a l c o n t r o l w h i c h w i l l f a c t o r out c o n f o u n d i n g w h i c h may r e s u l t from p r o b l e m s i z e m a n i p u l a t i o n s . 5.3.1 E f f o r t M i n i m i s a t i o n Hypotheses Each group o f h y p o t h e s e s w i l l be p r e s e n t e d i n d i v i d u a l l y . A t t h e end o f the p r e s e n t a t i o n t h e two groups w i l l be summarised and c o n t r a s t e d . The h y p o t h e s e s f o c u s on two a r e a s . The f i r s t s i x h y p o t h e s e s c e n t r e a r o u n d changes i n i n f o r m a t i o n use and p r o c e s s w h i c h a r e e s s e n t i a l l y e f f e c t i v e n e s s o r i e n t e d . The f i n a l two h y p o t h e s e s , H7 and H8, a r e c o n c e r n e d w i t h p r o c e s s e f f i c i e n c y . HI I n f o r m a t i o n use One o f t h e k e y measures o f t h e c o m p l e t e n e s s o f t h e d e c i s i o n s t r a t e g y i s i n f o r m a t i o n u s e . Problems w h i c h c o n t a i n a f i x e d number o f a l t e r n a t i v e s each d e s c r i b e d by a number o f a t t r i b u t e s a l l o w f o r an assessment o f t h e p r o p o r t i o n o f a v a i l a b l e i n f o r m a t i o n u s e d by t h e s u b j e c t s . 83 A d e c i s i o n maker who i s guided p r i m a r i l y by e f f o r t considerations w i l l tend to use the l e a s t e f f o r t f u l strategy which provides a reasonable s o l u t i o n . One of the key c h a r a c t e r i s t i c s of such a strategy i s that i t w i l l involve f i l t e r i n g and r e s t r i c t e d a ttention to cues so that not a l l information i s considered. Thus, we would expect that i n an unaided s e t t i n g a de c i s i o n maker would use a r e l a t i v e l y small proportion of a v a i l a b l e information. Would the de c i s i o n a i d have any impact on information use i n t h i s setting? The a i d i s designed to d i r e c t l y support much of the f i l t e r i n g a c t i v i t i e s u t i l i s e d i n the EBA and conjunctive s t r a t e g i e s . I t i s r e l a t i v e l y less e f f e c t i v e at supporting those strategies which would lead to complete information use ( i . e . , AC and AD). In essence the a i d may widen the gap between additive and elimination strategies by making the elimination s t r a t e g i e s r e l a t i v e l y more a t t r a c t i v e . By t h i s reasoning we would p r e d i c t that the hypothesis: HI: There w i l l be no di f f e r e n c e i n the amount of information used by aided and unaided subjects, would not be rejected. Individuals following a l e a s t e f f o r t approach would tend to f i l t e r with or without the aid. The l o g i c a l a l t e r n a t i v e to t h i s hypothesis i s put f o r t h i n the next section under accuracy maximisation. In that case we would expect that aided d e c i s i o n makers w i l l process more information and HI would be rejected. H2 A i d by s i z e i n t e r a c t i o n on information use We do not expect any change i n the proportion of information used as a function of problem siz e and decision a i d . From previous research (Payne, 1976; Olshavsky, 1979; Biggs et a l . 1985) we know that the proportion of 84 information used i s inversely r e l a t e d to problem s i z e . Since the d e c i s i o n a i d i s used only i n a manner to conserve e f f o r t and supports the type of stra t e g i e s that are known to be employed i n these settings, we would not expect i t to lead to a wider gap i n information use as problem s i z e increases. For a l l problems the a i d should allow the dec i s i o n maker to consider the same amount of information with less e f f o r t . Thus, H2: T h e r e w i l l be no i n t e r a c t i o n b e t w e e n t h e u s e o f t h e d e c i s i o n a i d a n d p r o b l e m s i z e w i t h r e s p e c t t o i n f o r m a t i o n u s e , w i l l not be rejected under the assumption that the i n d i v i d u a l s primary concern i s with e f f o r t minimisation. The alternate hypothesis i s , again presented under the accuracy maximisation section discussed next. In that case we would expect an i n t e r a c t i o n e f f e c t which would show that the di f f e r e n c e i n the amount of information processed, between the aided and unaided groups, w i l l increase with problem s i z e . H3 Number of a l t e r n a t i v e s analysed i n d e t a i l The number of a l t e r n a t i v e s examined i n d e t a i l i s another measure of the completeness of information searched and can be i n d i c a t i v e of changes i n strategy. More exhaustive, additive strategies r e s u l t i n a more complete examination of a l t e r n a t i v e s while f i l t e r i n g approaches imply that only a l i m i t e d number of a l t e r n a t i v e s w i l l be searched i n d e t a i l . I f the dec i s i o n maker i s acting as an e f f o r t minimiser, then we would expect the to see the use of f i l t e r i n g s t rategies and a tendency to examine as l i t t l e information as i s necessary to f i n d an acceptable s o l u t i o n . Since the de c i s i o n a i d provides strong support f o r f i l t e r i n g , i t may a c t u a l l y discourage the more e f f o r t f u l d e t a i l e d examination of a l t e r n a t i v e s which t y p i c a l l y occurs 85 towards the end of the choice process. At some point i n the choice process the d e c i s i o n maker w i l l l i k e l y begin an additive examination of a l t e r n a t i v e s (Olshavsky, 1979) . The unaided e f f o r t minimiser i s expected to work i n t h i s manner. The aided d e c i s i o n maker may f i n d that the use of the a i d a c t u a l l y widens the e f f o r t gap between additive and e l i m i n a t i o n oriented processing. I f e l i m i n a t i o n strategies become r e l a t i v e l y easier, the d e c i s i o n maker be i n c l i n e d to continue elimination-type processing through to the end of the exercise. This would r e s u l t i n a reduction i n the number of a l t e r n a t i v e s examined i n d e t a i l . To t h i s end we p r e d i c t that hypothesis: H3: The use of the decision a i d w i l l have no eff e c t on t h e number of alternatives examined i n d e t a i l , w i l l be rejected. Aided decision makers, focussing on e f f o r t minimisation, w i l l tend to examine fewer a l t e r n a t i v e s i n d e t a i l than t h e i r counterparts i n the unaided group. An a l t e r n a t i v e to t h i s hypothesis i s presented i n section 5.3. under accuracy maximisation. H4 A t t r i b u t e usage Another measure of information use and consistency i n search focusses on the examination of a t t r i b u t e s across the various a l t e r n a t i v e s . The two relevant measures are: 1) Variance i n the number of a t t r i b u t e s examined across a l t e r n a t i v e s , 2) The average number of a t t r i b u t e s considered per a l t e r n a t i v e . Additive strategies are characterised by consistent, low variance use of information across a l t e r n a t i v e s . Elimination s t r a t e g i e s generally r e s u l t i n h i g h l y v a r i a b l e use of information across a l t e r n a t i v e s . Under heavy information loads we expect that the d e c i s i o n maker w i l l tend to reduce the 86 number of cues which are relevant to the problem, leading to reductions i n the average number of a t t r i b u t e s considered. The e f f o r t minimiser i s l i k e l y to use strategies which reduce the amount of information which must be considered i n order to make a d e c i s i o n about a p a r t i c u l a r a l t e r n a t i v e . This w i l l tend to lead to v a r i a b l e information use across a l t e r n a t i v e s . This behaviour w i l l not l i k e l y be contingent on the d e c i s i o n a i d since the a i d f a c i l i t a t e s the use of f i l t e r i n g approaches. Thus we would expect the hypothesis: H4a) There w i l l be no d i f f e r e n c e i n the variance i n a t t r i b u t e use across a l t e r n a t i v e s between the aided and unaided groups would not be rejected. Both groups w i l l use s t r a t e g i e s that lead to a r e l a t i v e l y high average variance; however, we expect no d i f f e r e n c e between the groups. The alternate hypothesis i n t h i s case would be based upon the reasoning that aided decision makers would be more i n c l i n e d to use additive s t r a t e g i e s , which would r e s u l t i n more constant information use per a l t e r n a t i v e and hence lower variance. In terms of the average number of a t t r i b u t e s considered per a l t e r n a t i v e the e f f o r t minimiser would be expected to s e l e c t i v e l y consider a t t r i b u t e s , ignoring those which are considered to be r e l a t i v e l y unimportant. The use of the a i d greatly reduces the e f f o r t associated with examining a t t r i b u t e s through the use of c o n d i t i o n a l drop and sort commands. For a d e c i s i o n maker who i s attempting to minimise e f f o r t we would expect that such savings would be r e f l e c t e d i n increased e f f i c i e n c y rather than i n increased information use, thus the hypothesis: H4b) There w i l l be no differences i n the average number of a t t r i b u t e s considered by the aided and unaided groups, 87 w i l l not be rejected. Again, the l o g i c a l alternate hypothesis would be based upon the reasoning that the use of the d e c i s i o n a i d w i l l encourage accuracy maximisation and hence lead to a larger number of a t t r i b u t e s being examined per a l t e r n a t i v e . This notion i s developed more f u l l y i n the next section. H5 Decision strategy While the f i r s t three measures out l i n e d above focus on the extent to which information i s processed, subsequent measures and hypotheses r e l a t e s p e c i f i c a l l y to the nature of the strategy employed i n the search process. Strategies are determined from the analysis of verbal protocols c o l l e c t e d while the subject solves the problem. The exact procedeures are described more f u l l y i n section 6.6.2.2. The measures taken with respect to information use i n HI through H4 are also associated with determining d e c i s i o n strategy. R e c a l l that e f f o r t minimisers use the l e a s t e f f o r t f u l strategy which i s expected to lead to an acceptable s o l u t i o n . Thus they would be expected to make s i g n i f i c a n t use of f i l t e r i n g s t rategies i n making a choice. The two dominant f i l t e r i n g or noncompensatory strategies employed i n p r e f e r e n t i a l choice problems are Conjunctive and EBA. As argued above the d e c i s i o n a i d provides s u b s t a n t i a l support for the f i l t e r i n g approaches through elimination and sort commands. Previous research has indicated that i n d i v i d u a l s w i l l tend to use such strategies f o r problems of the s i z e and type employed i n t h i s research. Thus, i t i s expected that the unaided d e c i s i o n maker, whether we take the e f f o r t or accuracy perspective, i s expected to use e i t h e r EBA or CONJ. The possible exception to t h i s would be i n the f i v e a l t e r n a t i v e problem where we might expect to see some use of additive s t r a t e g i e s by unaided d e c i s i o n makers. 88 The nature of the information presentation may cause the unaided group to favour conjunctive processing over EBA. Both st r a t e g i e s i f applied with a consistent set of thresholds would l i k e l y lead to the same s o l u t i o n (see Chapter 4 ) . However, i n the unaided s e t t i n g there may be d i f f e r e n t l e v e l s of e f f o r t associated with the two s t r a t e g i e s . Once a row or column i s opened i n the unaided s e t t i n g i t i s v i s i b l e on the screen f o r the duration of the exercise. The presence of a l l information on the screen at once could be viewed as d i s t r a c t i n g . There would be multiple information cues competing for attention, r e q u i r i n g the decision maker to expend e f f o r t d i s c r i m i n a t i n g between a l t e r n a t i v e s . In order to avoid t h i s problem the i n d i v i d u a l may begin to focus on a l t e r n a t i v e based search f a i r l y e a r l y i n the exercise. Such a pattern of information access would c l e a r l y favour a conjunctive evaluation. In order to manage information access to avoid problems with d i f f e r e n t i a t i o n between a t t r i b u t e values at the perceptual l e v e l while s t i l l employing the EBA strategy would require the unaided subject to access information on a c e l l by c e l l b a s i s . C l e a r l y , t h i s i s more e f f o r t f u l than opening whole rows or columns and would be avoided by an i n d i v i d u a l attempting to minimise e f f o r t . At the same time, assuming e f f o r t minimisation, the aided subject may be i n c l i n e d to favour EBA over the Conjunctive strategy. This i s due to the f a c t that one of the system's more powerful features, the c o n d i t i o n a l drop command, d i r e c t l y , and e x c l u s i v e l y , supports the EBA strategy. There i s no d i r e c t p a r a l l e l support for conjunctive processing. 7 Thus, from an e f f o r t point of view EBA has some advantage over Conjunctive i n the aided d e c i s i o n s e t t i n g . 7 The lack of t h i s l e v e l of support for the Conjunctive strategy occurs p r i m a r i l y because i t would require a function which guided the d e c i s i o n maker through the s p e c i f i c a t i o n of a threshold vector which could then be applied over the e n t i r e d e c i s i o n matrix. Such high l e v e l , or macro, forms of support tend to lead the user and are not being considered i n t h i s research. 89 Also since i n the aided s e t t i n g information presentation i s f l e x i b l e , rows and columns can be closed as well as opened, the forces which might cause the unaided group to use conjunctive processing are much les s compelling here. Thus, from the perspective of e f f o r t minimisation i t i s expected that the hypothesis: H5: There w i l l be no d i f f e r e n c e i n strategy between the aided and unaided groups w i l l be rejected. The alternate hypothesis that both groups w i l l tend to use the l e a s t e f f o r t f u l strategy, EBA, for the aided subjects and Conjunctive for the unaided subjects w i l l receive support. H6 Changes i n strategy during the task I t i s not unusual f o r d e c i s i o n makers to change s t r a t e g i e s over the course of a problem. Bettman and Park (1981) have noted s h i f t s from e l i m i n a t i o n based to a d d i t i v e based strategies as the d e c i s i o n maker narrows i n on a candidate s o l u t i o n . Russo (1977) points out that such changes are necessary since u l t i m a t e l y the s e l e c t i o n of an a l t e r n a t i v e i s required. The e a r l i e r i n the d e c i s i o n process such s h i f t s occur, the more complete the evaluation i s l i k e l y to be. For the e f f o r t minimiser we would expect such s h i f t s to occur r e l a t i v e l y l a t e i n the d e c i s i o n process, i f at a l l . Additive s t r a t e g i e s are inherently more e f f o r t f u l than elimination strategies and would not be used by an e f f o r t minimiser unless the noncompensatory approaches do not y i e l d an acceptable s o l u t i o n . The s h i f t to additive strategies would not l i k e l y be influenced by the d e c i s i o n aid, although to the extent that t h i s strategy supports e l i m i n a t i o n and widens the e f f o r t gap between the e l i m i n a t i o n and additive 90 s t r a t e g i e s , we may expect to see some instances of pure e l i m i n a t i o n type st r a t e g i e s i n the aided group. Based upon t h i s reasoning we would expect that the hypo the s i s: H6 The decision aid wil l not impact the point at which decision makers switch from elimination to additive strategies, w i l l not be rejected. The aided decision maker who i s attempting to minimise e f f o r t would not, as a r e s u l t of savings from the aid, conduct more d e t a i l e d evaluation. Rather, they would be expected to make decisions more e f f i c i e n t l y . In t h i s case the l o g i c a l alternate hypothesis i s that the aided d e c i s i o n maker w i l l conduct more additive evaluation towards the end of the exercise. H7 T o t a l Steps Most of the hypotheses s p e c i f i e d to t h i s point have centred around how d e c i s i o n maker's problem so l v i n g processes and information use might be impacted by the d e c i s i o n a i d . To capture the e f f o r t dimension two hypotheses which measure e f f o r t expended on the decision process are presented. I t i s i n the area of e f f o r t that the a i d would impact the e f f o r t minimiser. One measure of e f f o r t expended on the problem i s the t o t a l number of steps taken to reach a s o l u t i o n to the problem, where a step i s measured as a single elementary cognitive operation such as a comparison between two values, reading a si n g l e value or eliminating an a l t e r n a t i v e . Johnson and Payne (1985) have argued that i t may be appropriate to consider each elementary process as consuming the same amount of e f f o r t . Thus the t o t a l number of references made to an item of information i n the protocol might be considered a step. A s i m i l a r measure of e f f o r t has been used previously by Bettman et a l . (1986). As discussed previously the decision a i d automates a number of cognitive 91 processes normally required f o r the d e c i s i o n maker to evaluate choice problems. The question then i s whether these steps are reinvested into the problem. The t o t a l value of steps incorporates the e f f o r t of the a i d and the d e c i s i o n maker. The aided e f f o r t minimiser i s expected to use the system as a substitute f or personal e f f o r t . Thus we would expect that the hypothesis: H7: There w i l l be no d i f f e r e n c e i n the t o t a l number of steps taken to make a choice by the aided and unaided d e c i s i o n makers, w i l l not be rejected. The aided decision maker attempting to minimise e f f o r t would complete the choice task i n the same number of steps as the unaided d e c i s i o n maker. H8 T o t a l time Time taken to complete the problem i s also a u s e f u l surrogate f o r the amount of e f f o r t expended on the problem. Presumably, the use of cognitive operators takes time; thus time i s a reasonable i n d i c a t o r of e f f o r t expenditures. I t i s not a precise measure of e f f o r t since i t i s impossible to be c e r t a i n that an i n d i v i d u a l i s engaged i n cognitive a c t i v i t y during an e n t i r e problem so l v i n g session. Thus time measures may tend to overestimate a c t i v i t y . However, we would not expect that time would have d i f f e r e n t impacts on the various experimental treatments. This implies that differences i n e f f o r t , as measured by elapsed time, should be preserved. One of the d i f f i c u l t i e s associated with using time as a measure i s that we cannot d i s t i n g u i s h between the l e v e l of e f f o r t and time d i r e c t l y r e l a t e d to the task, versus that which may be di r e c t e d towards the understanding and use of the system. While the system interface i s reasonably simple, minimising 92 command entry time, i t may be that the dec i s i o n maker expends considerable time t r y i n g to s e l e c t appropriate functions from the system to apply to the problem. This means that we must cautiously i n t e r p r e t r e s u l t s associated with time. For the e f f o r t minimiser we would a n t i c i p a t e savings i n e f f o r t to be p a r a l l e l e d by savings i n time taken to complete the task. This i s e s s e n t i a l l y the argument that a DSS contributes to dec i s i o n making e f f i c i e n c y . I t i s expected that the hypothesis: H8: There w i l l be no differe n c e i n t o t a l time taken by the aided and unaided groups, w i l l be rejected. Aided decision makers attempting to minimise e f f o r t , w i l l spend le s s time than t h e i r counterparts without the de c i s i o n a i d . 5.3.2 Accuracy maximisation hypotheses The following hypotheses are a l l based upon the notion that the decision maker emphasises accuracy i n dec i s i o n making when making trade-offs between e f f o r t and accuracy. This i s a c e n t r a l , though untested, assumption behind the development of DSS. Such systems t r a d i t i o n a l l y have the goal of augmenting the dec i s i o n makers effectiveness (Keen and Scott-Morton, 1978) . This i m p l i c i t l y assumes that the dec i s i o n maker w i l l expend reductions i n cognitive load on further problem dire c t e d e f f o r t which w i l l presumably lead to increased d e c i s i o n effectiveness. HI Information use Consider the dec i s i o n maker who i s guided by the desire to maximise de c i s i o n q u a l i t y subject to an e f f o r t constraint. The unaided de c i s i o n maker 93 would be expected to use predominantly f i l t e r i n g type s t r a t e g i e s consistent with the previous l i t e r a t u r e (e.g., Payne, 1976). The aided d e c i s i o n maker i s l i k e l y to make greater use of additive type s t r a t e g i e s . The reasoning behind t h i s i s as follows. The aided decision maker may i n i t i a l l y employ the decision a i d f o r f i l t e r i n g and elimination. To the extent that such procedures are well supported by the aid, a decision maker would expend fewer cognitive resources on t h i s phase of the decision. I t would be expected then that the decision maker, t r y i n g to optimise decision q u a l i t y , subject to an e f f o r t constraint, would expend the saved resources on a more thorough analysis of the problem. This would lead to greater information use. Even i f the d e c i s i o n maker employed e x c l u s i v e l y elimination type strategies we might expect looser f i l t e r s and search past the discovery of an i n i t i a l s a t i s f a c t o r y s o l u t i o n would also lead to greater information use i n the aided case. In short, under the assumption that the decision maker emphasises d e c i s i o n q u a l i t y we would expect hypothesis HI to be rejected. Subjects with the d e c i s i o n a i d would use a greater proportion of a v a i l a b l e information than those with out the ai d . Thus we would expect HI: there w i l l be no d i f f e r e n c e i n information use between the aided and unaided groups, w i l l be rejected. H2 A i d by s i z e i n t e r a c t i o n with information use Under the assumption that the d e c i s i o n maker attempts to maximise accuracy, we would assert that as the problem s i z e grows, the d e c i s i o n maker w i l l come under greater s t r a i n and processing of information w i l l become inc r e a s i n g l y l i m i t e d . The decision a i d makes the f i l t e r i n g approach much more 94 s t r a i g h t f o r w a r d and i s n o t a f f e c t e d by p r o b l e m s i z e . C o n s i d e r f o r example t h a t t h e d e c i s i o n maker implements a c o n d i t i o n a l e l i m i n a t i o n . I n t h e u n a i d e d s e t t i n g t h e amount o f e f f o r t r e q u i r e d f o r c o m p a r i s o n and e l i m i n a t i o n r i s e s w i t h p r o b l e m s i z e . U s i n g t h e d e c i s i o n a i d t h e d e c i s i o n maker w o u l d e x p e r i e n c e no s u c h change i n e f f o r t as t h e s y s t e m a u t o m a t i c a l l y e x e c u t e s t h e p r o c e d u r e b a s e d upon a command w h i c h does n o t change w i t h i n c r e a s e s i n t h e number o f a l t e r n a t i v e s . The same ca n be s a i d o f p r o c e d u r e s s u c h as s o r t and c a l c u l a t e . I n t h i s r e g a r d we w o u l d e x p e c t t h e a c c u r a c y m a x i m i s e r t o be l e s s overwhelmed by l a r g e r p r o b l e m s s i n c e o p e r a t i o n s c o n d u c t e d w i t h t h e a i d a r e i n v a r i a n t i n terms o f changes i n e f f o r t w i t h r e s p e c t t o changes i n p r o b l e m s i z e . Thus, we w o u l d e x p e c t , H2: T h e r e w i l l be no i n t e r a c t i o n between t h e u s e o f t h e d e c i s i o n a i d and p r o b l e m s i z e i n terms o f i n f o r m a t i o n u s e w i l l be r e j e c t e d . As t h e s i z e o f t h e p r o b l e m i n c r e a s e s t h e d i f f e r e n c e between the a i d e d and u n a i d e d groups s h o u l d w i d e n as t h e a i d becomes more u s e f u l and t h e e f f o r t d i s c r e p a n c y between t h e a i d e d and u n a i d e d groups t o implement the same o p e r a t i o n s becomes l a r g e r . H3 Number o f a l t e r n a t i v e s examined i n d e t a i l I f we c o n s i d e r a d e c i s i o n maker who a c t s as an a c c u r a c y m a x i m i s e r , p r e s u m a b l y t h e more d e t a i l e d s e a r c h a r g u e d f o r under H i and H2 above wou l d a l s o l e a d t o t h e e x a m i n a t i o n o f more a l t e r n a t i v e s i n d e t a i l . One a t t r i b u t e o f t h e a r o u s e d d e c i s i o n maker i s t h a t he o r she b e g i n s t o l i m i t cue usage, thus i g n o r i n g p o t e n t i a l l y r e l e v a n t i n f o r m a t i o n ( B r o a d b e n t , 1958). I f t h e d e c i s i o n a i d s e r v e s t o l i m i t c o g n i t i v e s t r a i n , we m i g h t e x p e c t t h e d e c i s i o n maker t o f o c u s on a w i d e r range o f cues t h u s i n c r e a s i n g t h e number o f a l t e r n a t i v e s 95 examined i n d e t a i l . Also to the extent that the a i d f a c i l i t a t e s f i l t e r i n g , the investment of e f f o r t i n t h i s stage of problem so l v i n g w i l l be r e l a t i v e l y low for the aided d e c i s i o n maker. When the d e t a i l e d search which i s more t y p i c a l of the l a t t e r stages of problem so l v i n g begins, the remaining pool of resources w i l l be r e l a t i v e l y larger allowing more work at t h i s stage. Thus, fo r the accuracy maximiser we would expect H3: The use of the d e c i s i o n a i d w i l l have no e f f e c t on the number of a l t e r n a t i v e s examined i n d e t a i l , w i l l be rejected; aided d e c i s i o n makers w i l l examine a greater number of al t e r n a t i v e s i n d e t a i l than unaided de c i s i o n makers. H4 A t t r i b u t e evaluations For the accuracy maximiser the reduction i n information load due to the use of the dec i s i o n aids i s expected to lead to a more complete evaluation of a t t r i b u t e s . To the extent that the dec i s i o n maker converts h i s or her expanded pool of cognitive resources into the a d d i t i o n a l use of addit i v e processing, we would expect to see reduced variance i n the number of a t t r i b u t e s examined across the a l t e r n a t i v e s . Thus: H4a: There w i l l be no differ e n c e i n the variance i n a t t r i b u t e use across a l t e r n a t i v e s between the aided and unaided groups, w i l l be rejected. Also, as argued above the decision a i d greatly reduces the e f f o r t of comparing a t t r i b u t e values to thresholds. In t h i s respect we would expect that the aided d e c i s i o n maker w i l l examine a greater number of a t t r i b u t e s . In t h i s case f o r the accuracy maximiser we would expect that H4b: There w i l l be no di f f e r e n c e i n the average number of a t t r i b u -96 tes examined per alternative between the aided and unaided groups, w i l l be rejected. H5 Decision strategy In general, we expect the accuracy maximiser w i l l u t i l i s e the most e f f o r t f u l strategy a v a i l a b l e within the confines of h i s or her l i m i t e d information processing capacity. Note that t h i s assumes that the increased e f f o r t required to use more exhaustive strategies r e s u l t s i n a payoff i n terms of increased d e c i s i o n effectiveness. Johnson and Payne (1985) argue that t h i s i s not an u n r e a l i s t i c assumption. Unaided de c i s i o n makers i n the decision s e t t i n g under i n v e s t i g a t i o n here have t y p i c a l l y been observed to use f i l t e r i n g s t r a t e g i e s (see f o r example Payne, 1976). The possible exception to t h i s for the proposed studies may be the f i v e a l t e r n a t i v e case where decision makers may be able to u t i l i s e compensatory st r a t e g i e s even i n the unaided s e t t i n g . I f t h i s were the case then we might expect to see l i t t l e d ifference i n strategy between the aided and unaided settings f o r the f i v e a l t e r n a t i v e case. The aided d e c i s i o n maker would also be expected to use compensatory s t r a t e g i e s , using the l i m i t e d system features f o r supporting these processes such as the move, reorder, sort and c a l c u l a t e commands. In the 10 and 20 a l t e r n a t i v e treatments we expect that the d e c i s i o n maker w i l l be under a r e l a t i v e l y heavy cognitive load n e c e s s i t a t i n g the use of f i l t e r i n g s t r a t e g i e s . Use of the DSS i n t h i s case r e l i e v e s some of the cognitive load on the decision maker by automating some of the elementary information processes associated with the use of various choice procedures. In essence, t h i s acts to expand the pool of resources which the d e c i s i o n maker 97 can commit to the problem. As a r e s u l t we would expect that these freed resources w i l l be applied to conducting further search and evaluation. Given a goal of ac