Decisions under Time Pressure: How Time Constraint Affects Risky Decision Making

ORGANIZATIONAL BEHAVIOR AND HUMAN DECISION PROCESSES

Vol. 71, No. 2, August, pp. 121–140, 1997ARTICLE NO. OB972717

Decisions under Time Pressure: How TimeConstraint Affects Risky Decision Making

Lisa Ordonez

University of Arizona

and

Lehman Benson III

University of Arizona

Subjects rated the attractiveness of and judged maximum buy-ing prices for gambles which had some probability of winninga dollar amount, otherwise winning nothing. Change-of-processtheory (Mellers, Chang, Birnbaum, & Ordonez, 1992; Mellers,Ordonez, & Birnbaum, 1992) asserts that decision makers multi-ply probability and amount information when stating buyingprices but add this information when reporting attractivenessratings. When subjects were placed under time constraint, how-ever, some subjects’ ratings were consistent with a multiplicativecombination process. This result only occurred when these sub-jects performed the rating task under time constraint and hadperformed the buying price task in the previous set of trials.These subjects were less likely to engage in cognitive tasks, asmeasured by the Need for Cognition Scale (Cacioppo, Petty, &Kao, 1984). Apparently, the extra cognitive demands of the timeconstraint caused these subjects to use the same strategyemployed in the previous task. When the time constraint wasremoved, these subjects appeared to switch back to an additivestrategy. These changes in information processing produced pre-dictable patterns of preference reversals. q 1997 Academic Press

This research was supported by a grant to the first author from the University of Arizona SmallGrants Program. We thank Lee Beach, Terry Connolly, Leamon Crooms, Barbara Mellers, AmnonRapoport, and two anonymous reviewers for comments on earlier drafts. In addition, we thankAnup Kuzmiyil and Vaishali Ghiya for programming assistance.

Address reprint requests to Lisa Ordonez, Management and Policy, School of Business andPublic Administration, University of Arizona, Tucson, AZ 85721. E-mail: [email protected].

121 0749-5978/97 $25.00Copyright q 1997 by Academic Press

All rights of reproduction in any form reserved.

122 ORDONEZ AND BENSON

INTRODUCTION

Overview

Researchers have typically studied risky decision making in situations thatallow the decision maker an unlimited amount of time to perform the task.However, many real world decisions are made under some form of time con-straint. Air traffic controllers must make quick decisions when directing airtraffic to avoid an accident. Police officers must quickly decide whether situa-tions warrant the use of force. A driver, upon seeing a yellow light, must decidewhether to speed up through the intersection or hit the brakes to stop beforethe red light appears.

Do decision makers use the same strategies when making risky decisionswith and without time constraint? Are there individual differences in responsesto time constraint? How do decision makers react when a deadline has beenmet or when a time constraint has been removed? This paper investigatesthese issues.

Research suggests that decision makers tend to speed up execution of theirdecision strategies or switch to simpler strategies when under time constraint(Edland & Svenson, 1993; Johnson, Payne, & Bettman, 1993; Svenson & Ben-son, 1993a, 1993b; Smith, Mitchell & Beach, 1982; Wright, 1974). Furthermore,when faced with high levels of time constraint, decision makers tend to relymost heavily upon negative information (Wright, 1974). Moreover, decisionmakers under time constraint either filter information that is used or omitcertain information from consideration altogether (Miller, 1960).

Time Constraint

As noted in Benson and Beach (1996), many researchers select time con-straint levels arbitrarily, with no rationale. In these studies, we first timedsubjects on the task of interest and then selected as our constraint the executiontime that was one standard deviation below the mean. If times are normallydistributed, this forces about 84% of the subjects to perform the task fasterthan normal. However, setting a time constraint is not enough to ensure thatsubjects feel time pressure (Svenson & Benson, 1993; Benson, 1993). Timeconstraint exists whenever there is a time deadline, even if the person isable to complete the task in less time. Time pressure indicates that the timeconstraint induced some feeling of stress and created a need to cope with thelimited time. Thus, it is possible to have time constraint but no time pressure.

Change-of-Process Theory and Preference Reversals

Preference reversals were first demonstrated by Lichtenstein and Slovic(1971) and Lindman (1971). Subjects were offered pairs of gambles, matched onexpected value. Each pair contained a $ bet (a gamble with a small probability ofwinning a moderate amount) and a P bet (a gamble with a high probability ofwinning a small amount). They found that changing the task from choice to

DECISIONS UNDER TIME PRESSURE 123

pricing (buying or selling) resulted in systematic reversals in preference. Inchoice, subjects tended to choose the P bet over the $ bet. However, in thepricing task, the same subject often priced the $ bet higher than the P bet.The result, clearly, is problematic for researchers interested in measuring pref-erence.

Goldstein and Einhorn (1987) and Mellers et al. (1992a, 1992b) found prefer-ence reversals in the buying price and attractiveness rating task combination.They showed that subjects stated higher buying prices for the $ bet over theP bet, but rated the P bet as more attractive than the $ bet. Reversals havealso been found due to changes in point of view (Birnbaum & Stegner, 1979)and due to violations of monotonicity (Birnbaum & Sutton, 1992).

Several theories have been proposed to explain preference reversals(Goldstein & Einhorn, 1987; Mellers, Chang, Birnbaum, & Ordonez, 1992;Mellers, Ordonez, & Birnbaum, 1992; Tversky, Sattath, & Slovic, 1988). Mellerset al. (1992a, 1992b) presented results which were consistent with their“change-of-process” theory but which could not be explained by contingentweight theory (Tversky et al., 1988) or expression theory (Goldstein & Einhorn,1987). Change-of-process theory states that preference reversals occur becausedecision makers use different strategies for combining information dependingon the task. Attractiveness ratings (“how attractive is this gamble (0–80)?”)were consistent with an additive combination strategy:

A (x, p; 0) 5 JA[k ? s(p) 1 u(x)], (1)

where A(x, p; 0) is the attractiveness rating for a gamble with probability pof winning amount x, otherwise nothing, s(p) is the subjective probability ofprobability p, u(x) is the utility of amount x, k is a scaling constant thatcalibrates subjective probabilities and utilities on the same scale for a givenexperiment, and JA is a monotonic judgment function which translates re-sponses onto the attractiveness rating scale. However, maximum buying priceswere better described by a multiplicative combination process.

PB(x,p; 0) 5 JB[s(p)?u(x)], (2)

where PB(x, p; 0) is the buying price for gamble (x, p; 0) and JB is a monotonicjudgment function for buying prices, placing responses on a dollar scale.Change-of-process theory assumes scale convergence (Birnbaum, 1974; Birn-baum & Sutton, 1992): that is, the psychological perceptions of probability andamount (i.e., subjective probabilities and utilities) remain constant across thetwo tasks.

In the Mellers et al. (1992b) study, the decision strategies used by subjectswere investigated by having them judge both buying price and attractivenessratings for a set of gambles constructed from a factorial combination of amountsto win ($3.00 to $56.70) and probability of winning (.05 to .94). Figure 1 presentsmean responses. Under certain assumptions about the judgment functions,change-of-process theory predicts that the pattern of these plots should differ


FIG. 1. Mean responses from Mellers et al. (1992a). Panel A presents mean attractivenessratings of simple gambles. Ratings plotted as a function of amount to win, with a separate curvefor each probability of winning. Panel B plots mean buying price responses by amount to win,with a separate curve for probability of winning. Responses were from the “positive skew” contextin Mellers et al. (1992a).

systematically. An additive combination strategy would result in parallelcurves, as shown in Figure 1A. A multiplicative strategy leads to a bilinearfan, where the curves meet at a common point and fan outward as in Fig. 1B.Since we cannot be certain that subjects are actually adding and multiplyingin these two tasks, we will use the form of the curves to describe the strategythat they use: parallel and bilinear.

Note that the inference that subjects are using different information pro-cessing strategies in the different tasks requires (a) evidence of preferencereversals, (b) monotonic judgment functions, and (c) the scale convergenceassumption. Considering each task separately, subjects could be using a varietyof different combination rules and judgment functions and produce curves thatare either parallel or bilinear. However, Birnbaum (1982) showed that theassumption of scale convergence between two tasks and monotonic judgmentfunctions imply that the two tasks cannot produce rank order differences ifsubjects are using the same combination process. Thus, given these assump-tions, preference reversals imply that subjects are using different strategies.

Risky Decision Making under Time Constraint

None of the current models of preference makes explicit predictions abouthow preferences will change under time pressure. This paper extends thechange-of-process theory for this purpose, seeking insights into the decisionstrategies employed by the subjects under different time constraints. Plots likeFig. 1 will be used to determine the information processing strategies used intime-constrained tasks. We will make the same assumptions as Mellers etal. (1992a, 1992b): scale convergence and monotonic judgment functions. InExperiment 1, we will apply a time constraint with the same set of gamblesused by Mellers et al. We expect that subjects will accelerate informationprocessing and possibly switch strategies to cope with the time constraint.


This switching of strategies to cope with time constraints may be moderatedby individual differences. There has been a great deal of research investigatingindividual differences in a general propensity to engage in cognitive activities,as measured by the Need for Cognition scale (NFC; Cacioppo, Petty, & Kao,1984). However, only one study has investigated the connection between NFCand reactions to time pressure (Verplanken, 1993). In this study, Verplankenfound that, under time constraint, subjects scoring low in need for cognitionappear to use more heuristic (i.e., simpler) information search strategies thando high-NFC subjects. We expect NFC scores to be related to strategies usedin our judgment tasks.

EXPERIMENT 1: TIME CONSTRAINT ON ATTRACTIVENESS RATINGSAND BUYING PRICES

Method

Stimuli and design. Subjects served in one of two conditions created by aTime Constraint by Task by Task Order factorial design (2 3 2 3 2). The TaskOrder was the only between subjects factor; it indicates whether subjects firstcompleted the attractiveness rating task or the buying price task.

Gambles were displayed as pie charts on a computer monitor. The circle wassaid to represent a hypothetical spinner device in which the outcome dependson where the spinner lands. The yellow section of the pie represented theprobability of winning, and the blue region indicated the probability of a zerooutcome. Amounts to win were indicated in a legend located near the pie.

Subjects were presented with 25 gambles created from a 5 (Probability) 3

5 (Amount) factorial design. Levels of probabilities were .09, .17, .29, .52, and.94. Levels of amounts were $5.40, $9.70, $17.50, $31.50, and $56.70. Theselevels were chosen to create several gambles with the same expected value,but differing levels of probabilities and amount; these stimuli were a subsetof those used in Mellers et al. (1992a, 1992b).

Instructions and procedures. All instructions were presented on the com-puter. The best and worst gambles were presented to indicate the range ofgambles before beginning the tasks. Gambles were presented one at a timein random order. Before each task, subjects completed five practice trials tofamiliarize themselves with the task and procedure. One half of the subjectsfirst rated the attractiveness of each of the 25 gambles on a scale from 0 to100 (0 5 neither attractive nor unattractive, and 100 5 extremely attractive).They then performed the buying price task in which they stated the maximumamount they would be willing to pay to play each of the 25 gambles. The otherhalf of the subjects performed the two tasks in the opposite order.

In the first block of trials, all subjects performed rating and pricing taskson the 25 gambles, taking as much time per trial as they needed. This allowedfor baseline measures of the subject’s normal time to complete the trials. (Pilottesting indicated that simply mentioning that there would be a time constraintin later trials caused subjects to significantly reduce their response times in


the unconstrained trials. Time constraints were therefore not mentioned inthese initial trials).

After completing both tasks without time constraint, the subjects were toldthat they would complete the remaining trails with limited amount of time tocomplete each trial. A bar at the upper left of the computer screen indicatedthe amount of time remaining: when the bar was filled, the allowed time wasover. Subjects were told that at the end of the time limit on any trial, thescreen would go blank and a loud beep would sound until they made a response.As an additional discouragement to exceed the time limit, subjects were toldthat every time they heard the warning beep (i.e., gone over the limit), theywould have to complete two extra “make-up” trials. In each task they completedthe 25 required trials, then gambles were randomly sampled from the set of25 to constitute the “make-up” trials. Before beginning the time constrainttrials, subjects completed 5 warm-up trials to familiarize themselves with thetime limit and procedure.

Subjects answered questions about their mood at the beginning of the ques-tionnaire and after the time constraint trials. At the end of the experiment,they completed the Need for Cognition (NFC) short form questionnaire andanswered a few additional demographic questions.

Time constraint method. As in Benson and Svenson (1993) constraints werebased on a pilot study. In the rating task, the average response time (RT) was5.82 s per trial with a standard deviation of 2.74 s. In the pricing task, themean was 9.28 s per trial with a standard deviation of 4.5 s. Thus, the timeconstraints imposed in the experiment were 3.08 s (5.82–2.74) for ratings and4.78 s (9.28–4.5) for prices.

Participants. Fifty undergraduate business students at the University ofArizona received extra credit for their participation. Each student worked alonein a sound-insulated room at his or her own pace. The experiment took from25 to 40 min to complete.

Manipulation checks. Several indicators suggested that subjects felt timepressure when under time constraint. First, subjects under time constraintsignificantly reduced the average time they took to complete each trial. Forratings, the average RT was 6.0 s without time constraint vs 2.9 s with timeconstraint (t49 5 8.87, p , .05). For prices, the mean RT without time constraintwas 8.6 s, 3.5s with time constraint (t49 5 8.05, p , .01).

Second, mood-related questions (all reported on a 0 to 100 “thermometerscale”) indicate that the subjects experienced stress due to the time constraint.On average, subjects reported feeling more “panicky” (58.0 vs 45.4 on a 0 5

tranquil to 100 5 panicky scale, t49 5 4.65, p , .05), more “vicious” (47.8 vs38.7 on a 0 5 loving to 100 5 vicious scale, t49 5 4.14, p , .05), and more“vigorous” (0 5 exhausted to 100 5 vigorous) after the time constraint trialsthan before (60.7 vs 54.4, t49 5 2.59, p , .05).

Finally, subjects’ responses indicated that, while they felt time pressure inthe time constraint condition, the pressure was not too extreme. After complet-ing the time constraint trials, subjects judged that they had a moderate amount


of time to think about their responses (4.5 on a scale from 0 5 little time to10 5 a great deal of time).

Results

Buying prices. Mean buying prices are plotted in Fig. 2. (Individual levelplots showed that all subjects’ responses followed the same pattern as in Fig.2.) The surprising result is that the pattern of means in the two time constraintconditions appears virtually identical: both panels displaying bilinear fansconsistent with a multiplicative combination process. In fact, an Order (2) 3

Time Constraint (2) 3 Probability (5) 3 Amount (5) analysis of variance ofbuying price responses revealed that there are no significant main effects orinteractions with the Time Constraint factor. That is, buying price judgmentswere not affected by the imposed time constraint. Buying prices were also notaffected by the order in which the two tasks were performed: buying pricesfollowed by attractiveness ratings or the reverse order. As indicated in Fig. 2,buying prices were affected by Amount (F4,192 5 176.2, p , .05), by Probability(F4,192 5 130.0, p , .05), and by the Amount by Probability interaction(F16,768 5 57.6, p , .05). Thus, the hypothesis that subjects would changedecision strategies under time constraint in buying prices is not supported. Itappears that, in the buying price task, subjects were able to accelerate theprocessing of information to cope with the time pressure without changingtheir decision strategies.

Attractiveness ratings. Individual subject attractiveness ratings showed in-dividual differences in the pattern of results. The subjects were thereforegrouped according to the form of their curves.

Fifteen subjects showed similar parallel response patterns (suggesting addi-tive combination rules) in both time constraint conditions. Mean attractivenessratings are shown in Fig. 3A. Five subjects showed similar bilinear fan patterns

FIG. 2. Mean buying prices from Experiment 1. Mean buying prices averaging over all 50subjects were plotted as in Fig. 1B. Panels A and B show responses from the no time constraintand time constraint conditions, respectively.


FIG. 3. Mean attractiveness ratings from “same process” subjects in Experiment 1. Meanattractiveness ratings were plotted as in Fig. 1A. The upper panels show responses from theparallel subjects, while the lower panels display the responses from the bilinear subject group.The left and right panels show results for the no time constraint and time constraint condi-tions, respectively.

(suggesting multiplicative combination rules) in both conditions. Mean ratingsare shown in Fig. 3B. This group is labeled “same process”, indicating thatthe subject responses were consistent using the same information processingstrategy with and without time constraint. Twenty-six subjects showed a paral-lel pattern under no time constraint and a fan pattern under constraint. Meanratings for this “different process” group are shown in Fig. 4. The remainingfour subjects showed no classifiable patterns.

We compared “same process” (Figs. 3A and 3B, n 5 20) vs “different process”(Fig. 4, n 5 26) individuals. The two groups were not significantly different inaverage mood (before and after the time constraint trials), average responsetime in attractiveness ratings with and without time constraint, amount oftime to think about their responses, gender ratio, and GPA. However, the twogroups showed a marginally significant difference in Need for Cognition (NFC).The “different process” individuals had lower average NFC scores than the“same process” individuals (17.1 vs 26.9 on a 272 to 72 scale, p 5 .068). LowNFC subjects appear to be more ready to shift cognitive strategy to compensatefor the demands of time constraints.


FIG. 4. Mean attractiveness ratings from “different process” subjects in Experiment 1. Meanattractiveness ratings are plotted as in Fig. 1A, separated by time constraint condition.

Discussion

Experiment 1 produced two surprises. First, the bilinear process is robustwhen subjects perform the pricing task. Even under extreme time constraint,subjects continue to use this strategy systematically. Second, attractivenessratings seem to be much more malleable than prices under time constraint.Approximately half of subjects use different strategies with and withouttime constraint.

A somewhat different interpretation of this last result is suggested by a closeexamination of task order in this experiment. Two counterbalanced task orderswere used: Attractiveness Ratings (AR), then Buying Price (BP), or vice versa,with the sequence first completed under no time constraint, then repeatedunder time constraint. This implied that the time constrained AR task wasalways preceded by a BP task—and all 50 subjects displayed bilinear strategiesin both BP tasks. The subjects we labeled as showing different processes be-tween the time constraint conditions may be simply those who maintainedtheir bilinear strategies into the AR task. This interpretation would be consis-tent with the lower NFC score of these apparent strategy changers. It may bethat low-NFC subjects are reluctant to incur the cognitive costs of shiftingdecision strategies between task blocks. Experiment 2 was designed to castfurther light on these two alternative interpretations.

EXPERIMENT 2: FURTHER INVESTIGATION OF TIME CONSTRAINT ONATTRACTIVENESS RATINGS

One interpretation of the results in Experiment 1, is that when time con-straint is imposed during an attractiveness rating task, some subjects (primar-ily those with low NFC) continued to use the bilinear process they were pre-viously using in a pricing task. Experiment 2 explores this explanation.


Subjects served in one of two conditions: time-unconstrained and time-con-strained. In both conditions, subjects rated the attractiveness of the 25 gamblesused in Experiment 1, taking as long as needed to complete each trial, thenstated the buying prices for these 25 gambles without time constraint. Thenext block of trials differed across groups. The constrained subjects completedthe attractiveness rating task with a time constraint imposed on each trial,whereas the unconstrained subjects had no time constraint. At this point,some of the constrained subjects (i.e., those with low-NFC scores) should usea bilinear process in the second attractiveness rating block since they werepreviously using a bilinear process in the buying price task. However, theunconstrained subjects are not expected to use a bilinear process for the secondattractiveness rating block since they were not under time constraint and havethe ability to change from the bilinear process in the previous buying pricetask to a parallel process in the second rating block. In the final block of trials,both groups completed the attractiveness rating task without time constraint.For the constrained subjects who continue using a bilinear process in ratingsunder time constraint, the release of time constraint in the final rating blockmay allow these subjects to switch back to a parallel process.

Method

Stimuli and design. Subjects served in either the time-unconstrained (noTC) or the time-constrained (TC) conditions. For the time-unconstrained condi-tion, the order of the task blocks was AR no TC, BP no TC, AR no TC, and ARno TC (where AR and BP represent the attractiveness rating and the buyingprice tasks, respectively). The time-constrained condition differed only in thatthe second rating task block was performed under time constraint: AR no TC,BP no TC, AR TC, AR no TC. In each task block, subjects evaluated the same25 gambles used in Experiment 1. Gambles were displayed as pie charts as inExperiment 1.

Instructions and procedures. All instructions and procedures were the sameas in Experiment 1 except for the order of task blocks. The 25 gambles werepresented in four blocks, and gambles were presented one at a time in randomorder within each block on a computer monitor.

Subjects reported their stress level at the beginning of the questionnaire,after the second block of attractiveness rating trials, and at the end of theexperimental trials. Additionally, subjects responded to questions about thequality of their responses after the second block of rating trials: how comfortablethey felt with their responses and how satisfied they were with their decisions.The constrained subjects judged if they had sufficient time to think and if theexposure time was too short. At the end of the experiment, they completedthe Need for Cognition (NFC) short form questionnaire and a few additionaldemographic questions.

Participants. Fifty-one undergraduate business students at the Universityof Arizona received course credit for their participation. Twenty-five of these


subjects were randomly assigned to the time-unconstrained condition, while theremaining 26 subjects served in the time-constrained condition. Each studentworked alone in a sound-insulated room at his or her own pace. Two additionalparticipants who did not follow instructions were excluded from the analyses.The experiment took from 25 to 40 min to complete.

Manipulation checks. Constrained and unconstrained groups reported ap-proximately the same level of stress before the beginning of the experimentaltrials (3.0 vs 2.2 on a 0 5 no stress to 10 5 a lot of stress scale, p 5 .294).However, after the second block of attractiveness rating trials, when the timeconstraint was imposed, the constrained subjects reported significantly higherlevels of stress than the unconstrained subjects (5.3 vs 2.4, t49 5 3.96, p , .05).At the end of the experiment, both groups reported comparable stress levelsto those at the beginning of the experiment (3.0 vs 2.3, for the constrained andunconstrained groups, respectively). Constrained subjects were significantlyless comfortable with their responses than were the unconstrained subjects(5.3 vs 7.0, t49 5 22.46, p , .05 on a 0 5 not comfortable to 10 5 very comfortablescale). Constrained subjects felt that the time constraint moderately limitedtheir time to think (mean 5 4.7 on a 0 5 little time to 10 5 a lot of time scale)and was moderately too severe (mean 5 5.7 on a 0 5 agree a little to 10 5

agree a lot scale).

Results

Attractiveness ratings. Figure 5 presents the mean responses in all fourtrial blocks for the time-unconstrained group. The orders of the Panels Athrough D correspond to the order in which the trial blocks were performed:Panel A presents mean attractiveness ratings from the first block of trials.Panel B displays the mean buying prices as in Fig. 1. Panels C and D presentmean responses for the last two blocks of rating trials. Individual subjectplots were examined, and three subjects with uncategorizable patterns wereexcluded from the means shown in Fig. 5. The general result is that the re-sponses from the unconstrained subjects are similar to the original findingsby Mellers et al. (1992a, 1992b): parallel curves in ratings and bilinear curvesin prices.

As in Experiment 1, there were individual differences in the pattern ofresponses in the time-constrained group. One of the 26 subjects in the con-strained group, 11 showed a pattern of responses similar to that displayed inFig. 5: parallel curves in all rating blocks and bilinear fans in the pricing block.Thus, these subjects appeared to accelerate their information processing in thetime constraint rating trials (Block 3) and used the same decision strategy forall rating blocks. We will refer to these subjects as the “same process” groupsince they appeared to use the same process in ratings with and withouttime constraint.

Another 11 of the constrained group used a bilinear process in the timeconstraint rating trials, replicating the result found in Experiment 1. Figure


FIG. 5. Mean responses from the time-unconstrained condition in Experiment 2. Mean re-sponses are plotted as in Fig. 1. Panels A to D indicate the task and the order in which the taskwas performed. AR denotes the attractiveness rating task and BP is the buying price task. Notime constraint (No TC) was imposed in any of the four blocks of trials.

6 displays the mean responses for these “different process” subjects of Experi-ment 2. Again, without time constraint, we see parallel curves in ratings andbilinear fan in prices. However, in the critical Block 3, when the time constraintwas imposed, curves form a bilinear fan similar to the buying prices in PanelB. It appears that these subjects used a bilinear process under time constraintinstead of the parallel process they used without time constraint. The finalpanel (D) suggests that once the time constraint was released, these “differentProcess” subjects returned to an additive combination process. (One of these11 subjects displayed bilinear curves similar to Panel C). The remaining foursubjects in the time-constrained group showed uninterpretable responses andwere placed in the “other” group.

As in Experiment 1, the “same process” and “different process” groups weresimilar in many ways: they were not significantly different in GPA, gender,rated stress levels (before and after time constraint), and judgment of the timesufficiency, satisfaction, and comfortableness with the time-constrained task.However, “different process” subjects were significantly lower on average NFCscore than “same process” subjects (6.5 vs 24.1, t20 5 22.56, p , .05).


FIG. 6. Mean responses from the “different process” group in the time constrained conditionin Experiment 2. Mean responses are plotted in a similar format as Fig. 6. The only difference isthat Panel C presents mean attractiveness ratings when a time constraint (TC) was imposed inthese trials.

Preference reversals. Figure 7 presents mean preference reversal rates forthe “same process” and “different process” groups in the constrained conditionand for all subjects in the unconstrained condition. Two types of preferencereversals are possible: expected and unexpected. Expected preference reversalsare those that have been frequently found in the literature: the P bet receivesa higher rating than the $ bet, but the $ bet is priced higher than the P bet.These reversals are located in the upper-right corner in the boxes of Fig. 7.Unexpected reversals are the opposite ordering and are rare in previous studies:the $ bet receives a higher rating than the P bet, but the P bet is priced higherthan the $ bet. Unexpected reversals are displayed in the lower-left corner inthe boxes of Fig. 7.

In the unconstrained group, all subjects appeared to use a parallel processin each rating block and a bilinear process in the buying price task. Change-of-process theory predicts systematic preference reversals due to these differentstrategies. Figure 7A provides evidence for this prediction: the expected prefer-ence reversal rates are significantly different from the unexpected rates (all


FIG. 7. Mean individual preference reversals are presented separately for the time-uncon-strained, “same process”, and “different process” groups (upper, middle, and lower boxes). Threesets of preference reversals are computed for each group: buying price responses versus the first,second, and third blocks of attractiveness ratings. In each set, preference reversals for the 30equal expected value gamble pairs were obtained for each subject. The buying price vs. attrac-tiveness rating 2 for the “same process” and “different process” groups are the only two boxes inwhich time constraint was imposed for ratings (these are indicated by the asterisks). Percentagesof expected and unexpected preference reversals are presented in the upper-right and lower-leftof each box. The remaining two cells represent consistent preferences across tasks and are leftempty for simplicity.

x2 tests have p , .05).1 Since the “same process” subjects showed the samepattern of strategies in the two tasks as the unconstrained subjects, the change-of-process predictions are that there should be systematic preference reversalsacross all three rating blocks. Figure 7B is consistent with this prediction,showing a similar pattern to that of the unconstrained group rates (all x2 testshave p , .05). The interesting change-of-process predictions emerge with the“different process” group. This group used a bilinear process for the buyingprices and ratings under time constraint (attractiveness rating 2) but a parallelprocess for the rating conditions without time constraint (attractiveness ratings1 and 3). Thus, the prediction is that there should be preference reversals forprices vs ratings 1 and 3 (when strategies are dissimilar) but not for prices vs

1 Since both reversals are errors in judgment and there is no way of a priori determining thenumber of errors that should occur due to chance, previous studies compare the two reversal ratesfor systematic differences as evidence that expected reversals exist at a significant rate. Chi-square tests with 1 df were conducted comparing expected and unexpected reversal rates withineach box, where the expected frequency is the average of the two reversal frequencies.


ratings 2 (when strategies are similar). Figure 7C supports this prediction:there are significant differences between expected and unexpected reversalsfor ratings 1 and 3 (x2(1) 5 7.94, p , .05 and x2(1) 5 6.76, p , .05) but not forratings 2. Thus, these data support the change-of-process theory predictionsthat use of dissimilar (similar) strategies across tasks leads to dissimilar (simi-lar) preference orderings.

Examination of the mean buying price and attractiveness ratings suggeststhat these preference reversals are strong and even occur for gamble pairswith nonequal expected values. In all panels of Figs. 5 and 6, two gambles areindicated with asterisks and are labeled 1 or 2. Gamble 1 is a .94 chance ofwinning $5.40 (otherwise nothing) and Gamble 2 is a .52 chance of winning$56.70 (otherwise nothing), with expected values of $5.08 and $29.48, respec-tively. In the time unconstrained group (Fig. 5) rating panels (A, C, and D),Gamble 1 is rated more attractive than Gamble 2. However, Gamble 2 is givena higher buying price than Gamble 1 (Panel B), resulting in three preferencereversals for this buying price task and all rating tasks. However, the “differentprocess” group in the time constrained condition (Fig. 6) reveals only twopreference reversals. Panels A, B, and D (unconstrained ratings or prices) showthe same pattern as in Fig. 5: Gamble 1 is rated higher than Gamble 2 butGamble 2 is priced than Gamble 1, resulting in two preference reversals.However, the constrained ratings (Panel C) show a different pattern: Gamble2 is rated higher than Gamble 1, resulting in no preference reversal comparedto the prices.

Discussion

These results replicate and extend those found in Experiment 1. The mainresult was that the subjects who used different processes in the rating tasksdid so under specific conditions: they used a bilinear process in ratings onlyunder time constraint and when they previously used bilinear process in prices.These “different process” subjects had significantly lower NFC scores. Sincethe general propensity to engage less in cognitive activities is related to whichsubjects use different processes, we can speculate as to why these subjects usea different process under time constraint. As suggested before, these subjectsmay have been attempting to reduce their cognitive effort by continuing to usethe decision strategy they had been using in the previous task, assuming thatthere is a “cognitive cost” to changing strategies. Thus, subjects may not alwaysswitch to easier strategies (as indicated by previous research) if switching itselftakes too much effort. Many of us have had the experience of using a lessefficient method of data analysis (e.g., hand calculation) because we believethat switching to an easier method requires effort (e.g., learning a new com-puter program).

An interesting consequence is that the “different process” subjects showedmore consistent preferences under time constraint. Since they used the samebilinear process in both ratings and prices, their preferences became moreconsistent and earlier preference reversals were eliminated. Thus, one way to


eliminate preference reversals is to use similar information processing strate-gies in both tasks. However, the effect was short-lived, since almost all subjectsreturned to a parallel process once the time constraint was released.

EXPERIMENT 3: INVESTIGATION OF THE “PREVIOUS PROCESS”HYPOTHESIS

It can be argued that subjects who used a parallel process in time-uncon-strained ratings and a bilinear process in time-constrained ratings changeddecision strategy purely in response to the time constraint. The “previousprocess” hypothesis, in contrast, states that the bilinear process was usedbecause of both the time constraint and the prior use of the bilinear process.Experiment 3 provides additional evidence that time constraint alone does notcause the subjects to change their decision strategy. It uses the same generaldesign as Experiment 2, except that the time-constrained rating task is pre-ceded by an unconstrained rating task. The “previous process” hypothesis pre-dicts that subjects will use the same process under constrained rating as theydid in the preceding unconstrained rating task.

Method

Stimuli and design. The design is similar to that of Experiment 2, exceptthe order of first two tasks was reversed. In the unconstrained condition, theorder of the task blocks was: BP no TC, AR no TC, AR no TC, AR no TC. Theconstrained condition differed only in that the second rating task was performedunder time constraint: BP no TC, AR no TC, AR TC, AR no TC. Thus, the timeconstrained rating conidition is preceded by an unconstrained rating task,unlike Experiment 2. All instructions and procedures were otherwise the sameas in Experiment 2.

Participants. Fifty-six undergraduate business students at the Universityof Arizona received course credit for their participation. Twenty-nine of thesesubjects were randomly assigned to the time unconstrained condition, whilethe remaining 27 subjects served in the time constrained condition. A fewadditional participants who did not follow instructions were excluded from theanalyses. The experiment took from 25 to 40 min to complete.

Manipulation check. The time-constrained group reported significantlyhigher levels of stress during the time-constrained task (4.1 vs 2.6, t54 5 2.36,p , .05); however there was no significant difference between the groups atthe beginning (2.8 vs 2.6, p 5 .71, n.s.) or end of the experiment (2.4 vs 2.8,p 5 .65, n.s.). As in Experiments 1 and 2, the constrained subjects felt timepressure, but were not overwhelmed by the time constraint: they reported thatthey had a moderate amount of time to think (mean 5 4.8) and that the timewas moderately too short (mean 5 5.4).


Results

The main question to be addressed in this experiment is whether any subjectsused a parallel process in ratings without time constraint but a bilinear processwith time constraint: the answer is clearly negative. Individual plots similarto Fig. 5 showed that none of the 27 constrained subjects used different decisionstrategies in the three different rating blocks: 81% of these subjects used aparallel process, while the remaining 19% used a bilinear process in all blocks.The same approximate percentages were found with the unconstrained condi-tion: 90% used a parallel process and 10% used a bilinear process. (x2(1) 5 .76,p 5 .38, n.s.). Thus, Experiment 3 supports the “previous process” hypothesisof Experiment 2: in order for subjects to use a parallel process in unconstrainedratings, but a bilinear process in constrained ratings, it is apparently necessaryto have a completed a task using a bilinear process (e.g., pricing) in betweenthe two rating tasks.

GENERAL DISCUSSION

These experiments examined the effects of time constraint and comparedresults to change-of-process theory predictions of preferences for gambles with-out time constraints. We applied time constraint to both attractiveness ratingsand buying price tasks for a set of gambles. We found no evidence that subjectsswitched decision strategies with buying prices to cope with time constraint.

In addition, results suggested that many subjects may have changed thestrategy used in the rating task in response to time constraint. Approximatelyhalf changed from a parallel process to a bilinear process in the rating task.These “different process” subjects seemed to simply continue using the strategythey had used in the previous trials. Patterns of preference reversals suggestthat these subjects changed decision strategies in response to time pressure.Under no time constraint, they used a parallel process for ratings and a bilinearprocess for prices, consistent with using an additive and multiplicative decisionstrategy. Following the arguments of change-of-process theory (Mellers et al.,1992a, 1992b), the use of these two different strategies resulted in preferencereversals. However, the bilinear process in time-constrained ratings and pricesled to the elimination of preference reversals, indicating that subjects used thesame decision strategy in these two tasks.

The next important issue to address is why did only half of the subjects usedifferent strategies in ratings with and without time constraints. Our only cluelies in the fact that these “different process” subjects tended score low on athe Need for Cognition personality scale (Cacioppo, Petty, & Kao, 1984) thatmeasures people’s general propensity to engage in cognitive activities. Wepropose that these subjects were actually attempting to reduce their cognitiveload, but that the load was associated with strategy change, rather than withthe content of either strategy. Prior to the time-constrained rating trials, theyhad been using a bilinear process in the pricing task. When the time-con-strained rating task began, they did not expend the cognitive effort to change


from bilinear to a parallel process as they had previously done in the time-unconstrained rating task. Of course this line of reasoning is post hoc sincewe could not interrupt the subjects during the time-constrained trials to askthem why they used their current decision strategy. (Also, they probably couldnot have told us had we asked).

Why did the time constraints affect strategies in the rating task but not thepricing task? Rating responses appear to more malleable than prices. Not onlyis this evidenced in the current study by the effects of time constraint, but thisis also suggested by rating responses found in Mellers et al. (1992b). In thisstudy, subjects produced parallel rating curves to gamble stimuli similar tothe current study. However, when Mellers et al. presented subjects with gamblesthat had zero winning payoffs or zero probabilities of winning, most producedresponse curves with a bilinear pattern. Mellers et al. suggest that these un-usual gambles made subjects aware of the problems with using an additivemodel: varying one factor (amount to win or probability of winning) while theother factor is zero would produce varying attractiveness ratings. A multiplica-tive process leads to the more realistic conclusion that these gambles shouldreceive a zero rating. Although the current study finds similar results (i.e.,contextual manipulations resulted in bilinear rating responses), we do notsuggest that our subjects were necessarily aware of the implications of theirdecision strategies. Instead, the commonality appears to lie in the fact thatboth studies showed the flexibility of rating responses. Note, however, thatOrdonez, Mellers, Chang, and Roberts (1995) showed the preference orders forboth ratings and prices changed in response to an experimental treatmentwhich reduced preference reversals.

Finally, why do subjects appear to add subjective probabilities and utilitiesin ratings but multiply this information in prices? Mellers, Weber, Ordonez,and Cooke (1995) summarize several studies showing the parallel pattern inratings and bilinear fan in prices. They suggest that decision makers treatprobability as a modifier in prices (thus, multiply it with amount) but as anindependent variable that contributes to overall attractiveness in ratings (thus,it is added to amount). However, the inclination to use an additive process inratings appears to be quite robust since subjects who used a bilinear processwith time-constrained ratings, revealed ratings consistent with an additiveprocess as soon as the time restriction was lifted. This result seems to suggestthat an additive process is a more familiar or natural strategy to use withrating responses.

These current results suggest that more complex models of preference undertime constraint need to be developed. Models that suggest that subjects simplyaccelerate processing or switch to “easier” strategies (however defined) appearto be incomplete. Previous research has indicated that decision makers appearto tradeoff effort against accuracy when deciding which strategies to employ(Johnson, Payne, & Bettman, 1993). The present data suggest that we alsotradeoff the immediate effort of switching strategies against using strategiesthat are easier or more familiar.


REFERENCES

Benson, L. (1993). Studies in human decision making: On the effects of experimental instructions,framing, and time constraints. Unpublished doctoral dissertation, Lund University, Lund Sweden

Benson, L., III, & Svenson, O. (1993). Post-decision consolidation following the debriefing of subjectsabout experimental manipulations affecting their prior decisions. Psychological Research Bulle-tin, 32, 1–13.

Benson, L., III & Beach, L. (1996). The effects of time constraints on the pre-choice screening ofdecision options. Organizational Behavior and Human Decision Processes, 67, 222–228.

Birnbaum, M. H. (1974). The nonadditivity of personality impressions. Journal of ExperimentalPsychology Monograph, 102, 543–561.

Birnbaum, M. H. (1982). Controversies in psychological measurement. In B. Wegener (Ed.), Socialattitudes and Psychophysical measurement. Hillsdale, NJ: Erlbaum.

Birnbaum, M. H., & Stegner, S. E. (1979). Source credibility in social judgment: Bias, expertise,and the judge’s point of view. Journal of Personality and Social Psychology, 37, 48–74.

Birnbaum, M. H., & Sutton, S. E. (1992). Scale convergence and utility measurement. Organiza-tional Behavior and Human Decision Processes, 52, 331–346.

Birnbaum, M. H., & Veit, C. T. (1974). Scale convergence as a criterion for rescaling: Informationintegration with difference, ratio, and averaging tasks. Perception & Psychophysics, 15, 7–15.

Cacioppo, J. T., Petty, R. E., & Kao, C. F. (1984). The efficient assessment of need for cognition.Journal of Personality Assessment, 48, 306–307.

Edland, A., & Svenson, O. (1993). Judgment and decision making under time pressure: Studiesand findings. In O. Svenson & A. J. Maule (Eds.), Time pressure and stress in human judgmentand decision making. New York: Plenum.

Goldstein, W. M., and Einhorn, H. J. (1987). Expression theory and the preference reversal phenom-ena, Psychological Review, 94, 236–254.

Johnson, E. J., Payne, J. W., & Bettman, J. R. (1993). Adapting to time constraints. In O. Svensonand A. J. Maule (Eds.), Time pressure and stress in human judgment and decision making. NewYork: Plenum.

Lichtenstein, S. & Slovic, P. (1971). Reversals in preference between bids and choices in gamblingdecisions, Journal of Experimental Psychology, 89, 46–55.

Lindman, H. R. (1971). Inconsistent preferences among gambles. Journal of Experimental 89, 390–397.

Mellers, B. A., Chang, S., Birnbaum, M. H., & Ordonez, L. D. (1992a). Preferences, prices, andratings in risky decision making. Journal of Experimental Psychology: Human Processes andPerformance, 18, 347–361.

Mellers, B. A., Ordonez, L. D., & Birnbaum, M. H. (1992b). A change-of-process theory for contextualeffects and preference reversals in risky decision making. Organizational Behavior and HumanDecision Processes, 52, 319–330.

Mellers, B. A., Weber, E. U., Ordonez, L. D., & Cooke, A. D. J. (1995). Utility invariance despitelabile preferences. In J. R. Busemeyer, R. Hastie, & D. L. Medin (Eds.), Psychology of learningand motivation: Vol. 32. Decision making from the perspective of cognitive psychology.

Miller, J. G. (1960). Information input overload and psychopathology. American Journal of Psychia-try, 116, 695–704.

Ordonez, L. D., Mellers, B. A., Chang, S, & Roberts, J. (1995). Are preference reversals reducedwhen made explicit? Journal of Behavioral Decision Making, 8, 265–278.

Smith, J. F., Mitchell, T. R. & Beach, L. R. (1982). A cost benefit mechanism for selecting problem-solving strategies: Some extensions and empirical tests. Organizational Behavior and HumanPerformance, 29, 370–396.

Svenson, O. & Benson, L., III (1993a). Framing and time pressure in decision making. In O.


Svenson & A. J. Maule (Eds.), Time pressure and stress in human judgment and decision making.New York: Plenum.

Tversky, A., Sattath, S., & Slovic, P. (1988). Contingent weighting in judgment and choice. Psycho-logical Review, 80, 204–217.

Verplanken, B. (1993). Need for cognition and external information search: Responses to timepressure during decision-making. Journal of Research in Personality, 27, 238–252.

Wright, P. (1974). The harassed decision maker: Time pressure, distraction and the use of evidence.Journal of Applied Psychology, 59, 555–561.

Received: December 30, 1996

Decisions under Time Pressure: How Time Constraint Affects Risky Decision Making

Documents