The adaptive decisionmaker

John W. PayneFuqua School of BusinessDuke Uniuersity

James R. BettmanFuqua School of BusinessDuke Uniuersity

Eric J. JohnsonThe Wharton SchoolU niaersity of P ennsylaania

l$LHCaNrsRrDGEqP UNTVERSITY PRESS

2

Contingencies in decision making

Introduction

By now, our basic premise is apparent: Irrdividuals display-a great

deal of flexibility in making decisions. In their highly influential

review of behavi,oral decision theory, Einhorn and Hogarth (1981)

noted that "the most important emPirical results in the period under

review have shown the sensitivity of judgment and choice to

seemingly minor changes in tasks" (p. 61). More than a decade of

,.s.ur.h iince Einhornind Hogarth's review has strongly reafftrmed

that information processing in decision making, as in other areas ol

cognition, is highiy contingent on the demands of the task. The same

individual will use many different kinds of strategies in making a

decision, contingent ufon such factors as how information is

displayed, the nature of the response, and the complexity of the

prout.*. This chapter has two purposes. First, we briefly dehne a

number of decision strategies that have been proposed to describejudgment and choice. second, we review the research showing the

i*p-u.t of different task and context variables on the use of decision

strategies bY an individual-Before we review the literature, several aspects of our point of view

need to be made explicit. First, in order to keep the scope of the

chapter manageable, we focus on preferential decisions; inferences

"nd for.."sts receive much more limited attention. Preferential

decision probiems are typically described using three basic com-

ponents: (t) ti,. alternaiives available to the decision maker; (2)

events or contingencies that relate actions to outcomes, as well as

the associated probabilities of those events; and (3) the values

associated with the outcomes. Values are often based on multiple

objectives, such as maximizing safety and minimizing cost. These

iniormational elements, along with a goal statement (such as "choose

Contingencies in decision making 2l

the preferred alternative"), represent the task environment presentedto a decision maker. Outside the laboratory, decision problems oftenare not presented in a completed form. For example, alternativesmay not be given but may have to be generated (Keller & Ho, 1988;Gettys, Pliske, Manning, & Casey, 1987). Nevertheless, it remainsuseful to define decision problems in terms of these three basiccomponents of alternatives, uncertainties, and values.

The decision maker's internal representation of the task environ-ment is the individual's problem space (or set of knowledge states)(Newell& Simon, 1972). A typical example of a decision problem isthe seiection of an automobile. Each car is characterized by differentmileage, attractiveness, safety record, passenger capacity, and soforth. The values of some of these attributes may be known withreasonable certainty (e.g., a specific car's engine size). However, thevalue of other attributes is uncertain, such as the reiiability ordurability of a newly introduced car. Of course, the individual'srepresentation will generaily be selective and wili not include allfeatures of the car. Another example is the choice among two simplelotteries or gambles. Each lottery is dehned by a probability ofwinning or losing and a specified amount to be won or lost.

When presented to subjects, such preferential choice problems areusually "weil structured" (Langley, Simon, Bradshaw, & Zytkow,1987). That is, subjects are typically given a specified set of alterna-tives and a set of attribute values to be used in solving the problem.It is important to note, however, that preferential problems oftenhave elements that make their solution difficult. For example, con-flict is typically present, in the sense that no one option is best on allattributes of value, and conflict has iong been recognized as a majorsource of decision difhculty (Shepard, 1964). The task also may beunfamiliar in the sense that a rule for resolving the conflict cannotreadily be drawn from memory. Thus, solving decision problemsoften is not the kind of "rec ognize and calculate" process associatedwith expertise in a task domain (Chi et al., 1988). This is true evenfor the simplest laboratory decision tasks. Rather, decision makingis often characterizedby tentatiueness, search, and the use of relatiuely"weak" methods (heuristics) that are more representative of novicelikeproblem solving (Langley et al., 1987).

A second aspect of our viewpoint concerns the terms task effectsand context effects, which have often been used interchangeablyin the literature. For the DurDoses of this book. we distineuish

22 The adaptiue decision maker

between the two. Task effects describe those factors associated withthe general structural characteristics of the decision problem, includ-ing response mode, number of alternatives, number of outcomes orattributes, time pressure, information display mode, and agendaconstraints. Context effects refer to those factors associated with theparticular values of the objects in the specilic decision set underconsideration, including similarity and the overall attractiveness ofalternatives. In general, the values of context factors are moredependent on individual perceptions than the values of task factors.

Finally, we view decision making as consisting of such interrelatedsubprocesses as information acquisition, evaluation of information,and the expression of a decision. The literature review focuses onthose task and context influences that result in (1) a change in thesalience and use of information in the environment, andlor (2) achange in the processes used to combine information into a judg-ment or choice.l For example, a change in response mode mayresult in the same evaluation (combination) strategy being usedbut with attention focused on different information. Alternatively,the change in response mode may result in the use of a differentevaluation process. Thus, observed contingent decisions may resultfrom the effect of a task or context variable on the acquisition(salience)of the information used or on the strategy used to combinethe information or both. Mellers, Ord6frez, and Birnbaum (1992)have recently emphasized a similar distinction between the encoding(salience) of information and the integration (evaluation) of multipleitems of information in their examination of preference reversals (seethe section on response mode effects in this chapter).

Decision strategies

Before considering how individuals flexibly use various strategiesin response to different decision tasks, we outline some of the more

t Goldstein and Busemeyer (1992) have suggested that rather than affecting eitherthe encoding or combination of attribute information, task and context factorsimpact the criterion used by a decision maker to decide whether the evidence infavor ofone alternative or another is suflicient to warrant an expressed preference.Although we do not doubt that the use of a criterion for deciding is sometimespart of the decision-making process, we feel that the evidence presented here isoverwhelmingly supportive of the use of multiple decision strategies rather thanuse ofa single criterion-dependent strategy. Further discussion ofthis point canbe found in chaDter 3.

Contingencies in dectsion nmking 23

common strategies used. A variety of such decision-making strategieshas been identified, and descriptions of some of these strategies andtheir properties are given here. Each strategy can be thought of asa method (a sequence of operations) lor searching through thedecision problem space. That search may reflect information aboutsuch aspects as the relative importance of an attribute - weight orsalience (e.g., safety is more important that comfort); cutoif valuesspecifying a minimal acceptable level for attributes (e.g., the gasmileage cannor be less than 20 miles per gallon); and differentialpreferences across attr ibute levels (e.g., a loss oi$10 hurts worse thanthe pleasure of a gain of $10). Search is often serective, and differentstrategies limit the amount or type of information processed invarious ways.

Before examining the specihc strategies, however, some generalaspects of decision processes need to be addressed. First, as notedearlier, decision problems often involve conflict among values,because no one option best meets all of our objectives. Some of thedecision strategies used by people can be thought of as conflictconfronting and others as conflicr avoiding (Hogarth, 19g7). That is,some decision processes confront and resoive confljct by consideringthe extent to which one is willing to trade off more of one valuedattribute (e.g., economy) for less of another valued attribute (e.g.,safety). other strategies do not as explicitly confront and resolvetradeoffs among valued attributes. Second, particular evaluationstrategies can either be used alone or in combination with otherstrategies. some typical combinations are discussed after theindividual strategies have been presented. Third, strategies can beeither constructed on the spot or their use could be planned a priori(see chapter 5 for further details). For example, Bettman (1979, p. 33)has suggested that "choice heuristics may not be stored in theirentirety in memory, but may exist only as fragments - subpartswhich are put together constructively at the time of making adecision." Fourth, strategies differ in both how effortful they areto use and how accurate they are likely to be (see chapter 3). Forexample, a heuristic that only considered information on one attri-bute (e.g., the lexicographic heuristic)might be less erfortful and lessaccurate for some types of decisions than a heuristic that examineda larger proportion of the available information.

24 The adaptiue decision maker

Descriptions of choice Processes

The weighted additiue (WADD) rule, The weighted additive ruleconsiders the values of each alternative on all the relevant attributesand considers a// the relative importances or weights of the attributesto the decision maker. Further, the conflict among values is assumedto be confronted and resolved by explicitly considering the extentto which one is willing to trade off attribute values, as reflected bythe relative importances or weights. A rule like WADD involvessubstantial computational processing of the information. TheWADD rule deveiops an overall evaluation of an alternative bymultiplying the weight times the attribute value for each attributeand summing these weighted attrifute values over all attributes. Itis assumed that the alternative with the highest overall evaluationis chosen. Given that the weighted additive rule processes all of therelevant probiem information and resolves conflicting values explicitlyby considering tradeoffs, the WADD rule (or some variant of it) isoften viewed as a normative procedure for dealing with preferentialdecision problems of the type considered in this book (see, e.g.,Keeney & Raiffa, 1916).

Exactly how people think of "weights" within the context of theWADD rule is the subject of investigation. There is some evidencethat weights are sometimes local in interpretation, in that the relativeweights reflect the ranges of attribute values across the alternativesin the choice set - that is, the greater the range, the greater theimportance of the attribute (Goldstein, 1990). At other times, theweight given to an attribute seems to be interpreted by subjects moreglobally, for example, safety might always be viewed as much moreimportant than costs, without much consideration of local rangesof values (Beattie & Baron, 1991). Another issue related to weightsis whether the influence of the weights on preferences reflects anadding or averaging process. In an averaging model, the weightsare constrained to sum to one; that is, they are normalized (seeStevenson, Busemeyer, & Naylor, 1990, for a discussion of addingvs. averaging models).

Two strategies related to the WADD rule may be used in makingdecisions under risk, the expected value and expected utility rules.The expected value rule involves multiplying the value (i.e., mone-tary amount) of each possibie outcome of a lottery by its probabilityof occurrence. These products of the values and probabilities are

Contingencies in decision making Zs

then summed over the outcomes to arrive at the expected value ofthe lottery. This mult iplyin-e and summing process is assumed to berepeated lor aii the lotteries in a choice set. It is further assumedthat the lottery or gambie with the highest EV will be chosen. Theexpected utility rule differs from rhe EV rule in that the utility ofeach outcome is substituted for its monetary value. This valuation(utility assessmenr) aspect of the EU rule expands the domain towhich tbe EU rule applies beyond monetary gambles; it may alsorequire addjtional processing effort. However, the general process-ing assumptions of both modeis are very similar.

The EV rule, and especialiy the EU role, are also viewed asnormative rules for choice. Thus, in the literature one can see theEV and EU rules used as both proposed descriptions of actualbehavior and as normative pr.r.riptions for behavior. However,while people sometimes make decisions in ways consistent withprocedures like the WADD, EV, and EU rules, more often peooleappear to make decisions usin,e simpler choice processes (heuristics).some oi the more common choice heuristics are described in thischapter. Each heurisric represents a different method ior simplifyingsearch through the decision problem space by limiting the amountof information processed and/or making how information is processedeasler.

The equal weight ( EQW ) heuristic. This processing strategy ex_amines all the alternatives and all the attribute values foi eachalternative. However, the equal weight strategy simplifies decisionmaking by ignoring information about the relative importance orprobabiiity of each atrdbute. An overall value for each alternativeis obtajned by simply summing the varues for each attribute for thatalternative. This assumes that the attribute vaiues are expressed, orcan be expressed, on a common scale of value. Hence this heuristicis a special case of the weighted additive rule. The equal weightrule has been advocated as a highly accurate simplification of thedecision-making process (Dawes, 1979; Einhorn & Hogarth, 1975).A variation of this rule that has been advocated for use in riskychoice is the equiprobable procedure, in which probability in-formation is ignored and the alternative with the highest averagepayoffselected (Thorngate, 1980). o. Huber (1999)references empiii-cal work documenting the use of an equal weight heuristic in .r isky choice.

26 The adaptiue decision maker

The satisJicing (SAT) heuristic. Satisficing is one of the oldestheuristics identified in the decision-making literature (Simon, 1955).With this strategy, alternatives are considered one at a time; in theorder they occur in the set. This heuristic compares the value of eachattribute of an alternative to a predelined cutoff level, often thoughtof as an aspiration level. If any attdbute value is below the cutoff,then that alternative is rejected. The hrst alternative that has valuesthat meet the cutoffs for all attributes is chosen. If no alternativespass all the cutoffs, the cutoffs can be relaxed and the process re-peated, or an alternative can be randomly selected. An implicationof the satisficing heuristic is that choice will be a function of theorder in which a decision maker evaluates alternatives. That is, ifalternative A and alternative B both pass the cutoffs, then whetherA or B is chosen will depend on whether A or B is evaluated first.There will be no comparison of the relative merit of arternative A ascompared with alternative B. A variation of this procedure is theconjunctive modei proposed by Coombs (1964), Dawes (1964), andEinhorn (1970).

The lexicographic ( LEX ) heuristic. The lexicographic proceduredetermines the most important attribute and then examines thevalues of all alternatives on that attribute. The alternative with thebest value on the most important attribute is selected. If two alter-natives have tied values, the second most important attribute isconsidered, and so on, until the tie is broken. For example, aconsumer may always buy the cheapest brand. Sometimes the LEXstrategy includes the notion of a just-noticeable difference (JND). Ifseveral alternatives are within a JND of the best alternative on themost important attribute (or any attributes considered subsequently),they are considered to be tied (Tversky, 1969). This version of theLEX rule is sometimes called lexicographic-semiorder (LEXSEMI).

A consequence of using a lexicographic-semiorder decision rureis that a person may exhibit intransitivities in preferences in whichX > Y, Y > Z, and Z > X. The following example decision problem,adapted from Fishburn (1991), illustrates that potential. professorP is about to change jobs. She knows that if two offers are far aparton salary (e.g., more than $ 10,000 apart), then salary will be thedetermining factor in her choice. otherwise, the prestige of theuniversity will be dominant. She eventually receives three offers,described in part as follows:

Contingencies in decision ntaking

Salary Prestige

X $65,000 LowY $ 50,000 HighZ $58,000 Medium

She prefers X to Y on the basis oi the better salary of X. BecauseI and Z are less than $ 10,000 apart in salary, she prefers Y to Z onthe basis of the greater prestige. She also prefers Z to X on the basisof prestige. Thus, X > Y, Y > Z, and Z > X, an intransitive patternof preferences. The general assumption is that rationality in choicerequires transitivity in preferences, although Fishburn ( 1 99 1) presentssome arguments why it may be reasonable for people to sometimesviolate transitivity.

The elimination-by-aspects (EBA) heuristic. First described byTversky (1.972), an EBA choice strategy begins with determinationof the most important attribute (Tversky actually assumed that theattribute is selected probabiiisticaily, with the probability that anattribute is selected being a function of its weight or importance).Then, the cutoff vaiue for that attribute is retrieved, and allalternatives with values for that attribute below the cutoff areelimjnated. One can interpret this process as rejecting or eliminatingalternatives that do not possess an "aspect"; the "aspect" is dehnedas having a value on the seiected attribute that is greater than orequal to the cutoff level. The EBA process continues with the secondmost important attribute, then the third, and so on, until onealternative remains. Note that while an EBA process violates theidea that one should use all relevant information in making adecision, it does reflect rationality in the ordered use of the attributes.This "partial" rationality in processing characterizes most choiceheuristics.

The majority of conJirming dimensions (MCD) heuristtc. Describedby Russo and Dosher (1983), the MCD heuristic invoives processingpairs of alternatives. The values for each of the two alternatives arecompared on each attribute, and the alternative with a majority ofwinning (better) attribute values is retained. The retained alternativeis then compared with the next alternative among the set ofalternatives. The process of pairwise comparison repeats until allalternatives have been evaluated and the hnal winning alternativehas been identified.

27

28 The adaptiue decision maker

The MCD heuristic is a simplified version of a more general

model of choice called the additive difference (ADDIF) model

(Tversky, 1969). In that processing strategy, the alternatives are

.otnpu.LO directly on each dimension, and the difference between

the subjectiu. u"iu"s of the two alternatives on that dimension is

determined. Then a weighting function is applied to each difference

and the results are summed over all dimensions to obtain an overall

relative evaluation of the two alternatives. Under Some condi-

tions, the additive difference rule and the WADD rule will produce

identical preference orderings, although the two rules differ in some

aspects oi processing (see Tversky, 1969, for a further discussion

of how th; ADDIF and WADD models are related). The MCD

heuristic simplifres the additive difference model both by ignoring

attribute weignts and by coding the attlibute differences in a binary

fashion, so that only the direction of the difference, but not its

magnitude, is considered.A variation on the additive difference process, proposed by

Aschenbrenner, Bockenholt, Albert, and Schmalhofer (1986), is to

process the attribute differences sequentially, accumulating the

summed differences until the summed advantage of one option over

another exceeds some criterion value. Bockenholt, Albert, Aschen-

brenner, and Schmalhofer (1991) suggest that the criterion value may

reflect the balance the decision maker desires between the effort

invoived in a decision process and the quality of the choice process.

The frequency of good and bad features ( FRQ) heuristic. Alba and

Marmoistein (1987) suggest that decision makers may evaluate or

choose alternatives based simply upon counts of the good or bad

features the alternatives possess. To implement this heuristic, a

p".ron would need to devilop cutoffs for specifying good and bad

ieatures. Then the decision -aker would count the number of such

features. Depending upon whether the decision maker focused on

good features, bad fiatures, or both, different variants of the heuristic

iould arise. Note that this heuristic could be viewed as the

application of a voting rule to multiattribute choice, where the

attributes can be viewed as the voters.

Combined strategies. Individuals sometimes use combinations of

strategies. Typicaliy, combined decision strategies have an initial

phase, wheri poor-alternatives are eliminated, and then a second

Conttngencies in decision ntaking

phase examining the remaining alternatives in more detai l (payne,1976). One such combined heuristic that is frequently observed indecision behavior is an elimination-by-aspects plus weighted additivestrategy. EBA is used to reduce the number of alternatives to somesmall number (e.g., two or three), and then a weighted additive ruleis used to seiect among those remaining alternatives.

other heurisrics. Several even simpler heuristics also have beenproposed. A frequent strategy for choice o[ this sort is the habitualheuristic: choose what one chose last time. A related heuristic,mentioned in chapter 1 and suggested by P. Wright (1975), is alfectreferral. An individuai simpiy elicits a previously formed evaruationfor each alternative from memory and selects the most highiyevaluated alternative. No detailed attribute information is considered.Note that both of these heuristics are only relevant for repeatedchoices.

General properties of chotce strategtes

The strategies we have discussed are just some of those proposed todescribe choice behavior. These strategies have come from a nurnberof discipiines and have been described using very different kinds offormaiisms. As a result, in order to compare and contrast strategiesfor choice, researchers have often described them using fairly broadand -elobal characteristics. Several of these characteristics are con-sidered next.

Contpensatory Dersus noncompensatory. A central distinction amongstrategies is the extent to which they make tradeoffs amongattributes. Decision strategies (such as weighted additive) that maketradeoffs are called compensatory strategies, whereas strategies(such as lexicographic) that do not make tradeoffs are called non-compensatory. The key to this distinction is the ability of a goodvalue on one attribute to make up for bad values on other attributes.A lexicographic strategy is noncompensatory because a bad valueon the most important attdbute will ensure that an aiternativewould never be chosen, no matter how good it is on another attri-bute. A weighted additive model is compensatory because goodvalues on one attribute can offset bad vaiues on another. Finally.some rules, Iike the majority of conhrming dimensions (MCD) rule,

29

30 The adaptiue decision maker

are partially compensatory in that the total number of advantagesfor an aiternative does matter, but the relative sizes of the advantages

do not.This distinction between compensatory and noncompensatory

rules is related to how a strategy deals with conflict. Compensatoryrules confront conflict, whereas noncompensatory rules avoid it.

Hogarth (1987) has suggested that peopie hnd making explicittradeoffs emotionally uncomfortable. Thus, decision makers mayavoid strategies that are compensatory not only because they aredifijcult to execute (cognitive effort) but also because they requirethe explicit resolution of diffrcult value tradeoffs (conflicts)2.

Consistent nersus selectiue processing. A related aspect of choicestrategies is the degree to which the amount of processing isconsistent or selective across alternatives or attributes. That is, is thesame amount of information examined for each alternative or attri-bute, or does the amount vary? In general, it has been assumed thatmore consistent processing across alternatives is indicative of a morecompensatory decision strategy (Payne, 1976). Consistent pro-cessing sometimes invoives examination of all information for everyalternative and attribute. A more variable (selective) processingpattern, on the other hand, indicates a strategy of eliminating alter-natives or attributes using only part of the information available,without considering whether additional information might changethe decision.

Amount of processing, A third general processing characteristic isthe total amount of processing. A key distinction among decisionrules is whether they explicitly ignore potentially relevant informa-tion in solving a decision probiem, and thus reduce the amount ofinformation processed, or attempt to process all relevant informa-tion. Whether processing is consistent or not, the total amount ofinformation examined can vary, from quite cursory to exhaustive.For some strategies, such as EBA, lexicographic, and satisficing,the total amount of information processed is contingent upon theparticular values of the alternatives and cutoffs.

2 As an example of this avoidance of tradeoffs, Gregory, Kunreuther, Easterling,and Richards (1991) note that the unwill ingness ofpeople to consider trading offincreases in environmental risks for money (economic benehts) is one reason whythe siting of hazardous waste disposal facilities is so controversial.

Contingencies in decision ntaking 3l

A lt ernatiue-hased uer sus attribute-b ased processing. A fourth aspecrof processing concerns whether the search and processing of alter-natives proceeds across or within attributes or dimensions. Theformer is often called holistic or alternative-based and the latterdimensional or attribute-based processing. In alternative-based pro-cessing, multiple attributes of a single alternative are consideredbefore information about a second alternative is processed. Incontrast, in attrjbute-based processing, the values of several alter-natives on a single attribute are processed before information abouta second attribute is processed. Russo and Dosher (1983) suggestthat attribute-based processing is cognitively easier.

Formation of eualuations. The strategies differ in terms of whetheror not an overall evaluation for each aiternative is explicitly formed.In the equal weight or weighted additive rules, for example, eachalternative is given a score that represents its overall evaluation. onthe other hand, rules such as lexicographic or EBA eliminate somealternatives and select others without directly forming an overallevaluation.

Quantitatiue t)ersus qualitatiue reasoning. Finally, the strategies alsodiffer in terms of the degree of quantitative versus qualitativereasoning used. Some strategies include quantitative reasoningoperations. For exampie, the equal weight method involves asumming of values, and the frequency heuristic requires counts. Theweighted adding rule includes the even more quantitative operationof multipiying two values. In contrast, the reasoning contained inthe other strategies described previously is more qualitative innature. That is, most of the operations for a strategy such as EBAinvolve simple comparisons of values. Tversky et al. (1988) make asimilar distinction between qualitative and quantitative thinking.Hegarty, Just, and Morrison (1988) also have recently exploredstrategy differences that involve a distinction between qualitativeand quantitative reasoning in making inferences about mechanicalsystems.

The various decision strategies or rules we have described re-present different combinations of these generalproperties. Table 2.1characterizes each of the major strategies in terms of five of theseproperties. Consistent with our conception of strategies as par-ticular sequences of mental and elfector operations (see chapter 1),

'qt

EE> , 9

; ( €

c ) h o

i l o

< i .q o

: P

6) l l

B a r

( )x

=3- / t rI! .E

oox . <

9 . =E :. - x

9 > ,

t t *

' , t l

n r l

c>

x .9x

o i l

* X' - H

i l i O

n t =

< 5\ ( q

> qt l

s<'<. a

zzz)) )zzt^J t^l t^t t^l -l

v v v v v v v v

J F J F J F F J

U ( J Q o c n a l ) ( )

zz>

U Q U Z Z Z O O

Q F ri L x E e}?8 i l ; f ,>E

o . ^> =' - = u= Y : >'!: > c€

L ^ . - ^

: 4 - c d - J< , ^ t a - /( t v 4 v

.9a i . 2

c o F e

fr"Et

!i 6 )x @q x

F - ti 3 ? . =. 6 r =- > :

' - ^ L ^

: F . . o J

: g ; S

7 - Q( J ; ( }

.= rD ."== > ( )v r \ =U \ ' ;

E.FI,

zt v

i = h

2 . 4C ) T -o . a ) Xt r > =( )98

oq

!

()

.4

v)

v)

Lq)

L

l\

e.i

,-

Contingencies in decision ntaking

Table 2.2. Operations used by choice heuristics

Heuristics comparisons Eiiminations concatenationsWADDADDIFEQWEBASATLEXMCD ++FRQ ++

a aJ J

++

++++ +++

+++++.Lr T

+

+'i -r

++++

Note: within each type of operation, the number of pluses indicates therelative extent to which each heuristic utilizes that type of operation. Forkey to abbreviations, see Table 2.i .

the various strategies can also be classified in terms of differentcombinations of operations. For example, the wADD rule obviouslyinvolves operations that concatenati the values of two or moreattributes, for example, adding two values together. The lexico-graphic decision rule, on the other hand, does not include explicitconcatenation operations, but does involve many comparisons ofone alternative's value on an attribute against the value of anotheralternative on the same attribute. Further, some rules, such as EBAo1 sAT, inciude explicit elimination operations. Table 2.2 offers aclassification of each of the major decision strategies in terms of theextent to which such operations as comparisons, eliminations, andconcatenation processes are utilized. In chapter 3, we will expandupon the set of operators used to describe the various strategies inorder to develop a measure of decision effort.

These strategies may be differentially available across individuals.A person's repertoire of strategies may depend upon many factors,such as cognitive development, experience, and more formai trainingand education- By 12 years of age, for example, children know many!t]f!gS_ies, but they apply them iess consisrenily than aduits (Kiayman,1985). Individuals may aiso acquire new strategies through .*p.ri.n..with a broad variety of choice tasks. Finally, formal i.u.hing .unprovide individuals with new strategies they can use (e.g., Larrick,Morgan, & Nisbett, 1990).