decision making :conscious versus unconscious_2009 (1)

8/10/2019 Decision Making :Conscious Versus Unconscious_2009 (1)

1/27


2/27

Think, blink or sleep on it? The impact of modesof thought on complex decision making

Ben R. Newell, Kwan Yao Wong, and Jeremy C. H. CheungUniversity of New South Wales, Sydney, NSW, Australia

Tim RakowUniversity of Essex, Colchester, Essex, UK

This paper examines controversial claims about the merit of unconscious thought for makingcomplex decisions. In four experiments, participants were presented with complex decisions and

were asked to choose the best option immediately, after a period of conscious deliberation, or aftera period of distraction (said to encourage unconscious thought processes). In all experiments themajority of participants chose the option predicted by their own subjective attribute weightingscores, regardless of the mode of thought employed. There was little evidence for the superiority ofchoices made unconsciously, but some evidence that conscious deliberation can lead to betterchoices. The final experiment suggested that the task is best conceptualized as one involvingonline judgement rather than one in which decisions are made after periods of deliberation or dis-traction. The results suggest that we should be cautious in accepting the advice to stop thinkingabout complex decisions.

Keywords: Decision making; Unconscious thought; Online judgement; Deliberation.

In order to get over the uncertainty thatperplexes us when we are faced with important,complex decisions, Benjamin Franklin advisedthe British scientist Joseph Priestly thus:

My way is to divide half a sheet of paper by a line into twocolumns; writing over the one Pro, and over the other Con.

Then, during the three or four days consideration, I putdown under the different heads short hints of the differentmotives, that at different times occur to me, for or against themeasure.. . . I find at length where the balance lies; and if,

after a day or two of further consideration, nothing new thatis of importance occurs on either side, I come to a determi-nation accordingly.. . . And, though the weight of reasonscannot be taken with the precision of algebraic quantities, yet

when each is thus considered, separately and comparatively,and the whole lies before me, I think I can judge better, andam less liable to make a rash step. (Franklin, 1772, cited inGoodman, 1931)

Wilson and Schooler (1991) noted that theessence of what Franklin called his moralalgebra is found in many of the formal descriptive

Correspondence should be addressed to Ben Newell, School of Psychology, University of New South Wales, Sydney, 2052,New South Wales, Australia. E-mail: [email protected]

The support of the Australian Research Council (Grant DP0558181) is gratefully acknowledged. The first author would like tothank John Payne for an enlightening discussion on the topic of unconscious thought and Brooke Hahn for assistance withprogramming.

# 2008 The Experimental Psychology Society 707http://www.psypress.com/qjep DOI:10.1080/17470210802215202

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY

2009, 62 (4), 707732


3/27

and prescriptive approaches to decision making.For example, techniques such as decision analysisadvocate careful consideration of all options andtheir attributes prior to choice (e.g., Edwards &Fasolo, 2001; Keeney & Raiffa, 1976; vonWinterfeldt & Edwards, 1986; see Newell,

Lagnado, & Shanks, 2007a, for an overview).In contrast to this intuitively sensible advice, a

recent high-profile book (Gladwell, 2005) andseveral articles (Dijsksterhuis, 2004; Dijksterhuis,Bos, Nordgren, & van Baaren, 2006;Dijksterhuis & Nordgren, 2006) have questionedthe use of conscious analytic thought for complexdecisions. Decision makers have been encouragedto make snap decisions (blink, Gladwell,2005) or to leave complex choices to the powersof unconscious thought (sleep on it,

Dijksterhuis et al., 2006). Such claims are seduc-tive and appealing and have received a great dealof attention in the media, giving rise to such head-lines as Want to make a complicated decision?Just stop thinking ( Jha, 2006) and Sleep on it,decision makers told (BBC News Online, 2006).This paper examines these important and rathersurprising claims and attempts to map out someof the boundary conditions of the power of uncon-scious thinking.

Not thinking versus thinking too much

The idea that thinking too much can sometimeslead to poorer choices is not new. Wilson andSchooler (1991) demonstrated that students whoanalysed their reasons for preferring particularbrands of strawberry jam revealed preferencesthat were less consistent with those of expertsthan participants who did not analyse theirreasons. A similar effect was found for preferencesfor college courses. Wilson and Schooler explainedthese effects by arguing that introspection causes adeparture from an initial adaptive preferencewhen that initial preference is inaccessible to con-scious report. Wilson and Schooler suggest thatalthough decision makers are in general able toweight relevant information appropriately, whenthey are asked to think about their reasonsduring the decision process it leads to a selective

focus on a subset of reasons that are accessible orplausible. As a result, this subset of reasons mayreceive greater (inappropriate) weight than otherpossible unarticulated reasons, diminishing thequality of the final preference or choice.

The data from the Wilson and Schooler (1991;

see also Wilson et al., 1993) studies suggest thatat least at times, the unexamined choice isworth making (p. 192). This general conclusionis echoed in work by Levine, Halberstadt, andGoldstone (1996) in evaluations of the likabilityof cartoon faces, Simonson and Nowlis (2000) inconsumer choices, and Tordesillas and Chaiken(1999) in a follow-up of Wilson and Schoolers(1991) study on university course selection.However, more recent work, in particular that ofDijksterhuis and colleagues, has made stronger

and more controversial claims on the merits orotherwise of conscious thought. While agreeingthat conscious thought may be detrimental forcomplex decisions, this recent work argues that aperiod of unconscious thought prior to making adecision confers benefits over decisions madeimmediately (Dijksterhuis, 2004; Dijksterhuiset al., 2006; Dijksterhuis & Nordgren, 2006).These researchers suggest that when decisionmakers are forced not to think about a decisionproblem (by being made to solve anagrams for

example), simultaneous unconscious processingoccurs during the distraction period that improvessubsequent choices.

Unconsciousthought is definedas object-relevantor task-relevant cognitive or affective thought pro-cesses that occur while conscious attention isdirected elsewhere (Dijksterhuis & Nordgren,2006, p. 96). It is explicitly distinguished fromideas popular in research on incubation in whicha period of distraction is said to allow problemsolvers to return to a problem with fresh eyesor to forget inappropriate strategies (e.g., Smith& Blankenship, 1989). Unconscious thought issaid to be an active process during whichinformation is organized, weighted, and integratedin an optimal fashion. The benefits of thisprocess are argued to be strongest when a decisionproblem is complexthose with multiple optionsand attributesbecause unconscious thought

708 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2009, 62 (4)

NEWELL ET AL.


4/27

does not suffer from the capacity limitationsthat hobble conscious thought (Dijksterhuis,2004).

This notion of the intelligent unconsciousstems partly from work claiming to demonstratethe ability to learn complex information in the

absence of awareness (e.g., D. C. Berry &Broadbent, 1984; Knowlton, Squire, & Gluck,1994; Lewicki, Hill, & Czyzewska, 1992; Reber,1992). This literature on implicit learningphenomena attributes considerable analytic skill tothe unconscious: an idea that meshes neatly withthe claim that complex decisions are best left tothese capacity-unlimited powerful processes. Suchconclusions are, however, not without strongcritics. In many demonstrations of implicit learning,cogent and compelling alternative explanations that

do not require recourse to intelligent unconsciousprocessing have been proposed (e.g., Dienes &Fahey, 1998; Lagnado, Newell, Kahan, & Shanks,2006; Newell & Bright, 2002; Newell, Lagnado, &Shanks, 2007b; see Shanks & St. John, 1994, for areview of earlier work). In spite of these alternativeand often far less startling interpretations,proponents have drawn strong and very generalconclusions about the benefits of unconsciousthought:

The current work demonstrates one thing the unconscious is

good at: making complex decisions. When faced withcomplex decisions such as where to work or where to live, donot think too much consciously. Instead after a little consciousinformation acquisition, avoid thinking about it consciously.

Take your time and let the unconscious deal with it.(Dijksterhuis, 2004, p. 596)

Evidence for the superiority of unconsciousthought in complex decision making

What is the evidence for these bold claims? Theexperimental paradigm used by Dijksterhuis andcolleagues is one in which participants are pre-sented with information about three or fourobjects (cars, apartments, potential room-mates)described by 10 or more attributes (e.g., mileage,building security, tidiness) and are asked tochoose the best object. In most cases best isdetermined normatively by the experimenter

assigning different numbers of positive and nega-tive attributes to each option. For example, thebest apartment might have 8 (normativelydefined) positive attributes and 4 negative attri-butes, while the worst apartment has the oppo-site arrangement (e.g., Dijksterhuis, 2004).

In the initial phase of the experiment attributeinformation is presented sequentially and typicallyin random order, about the four options. Thus fora choice set in which there are four options eachdescribed by 10 attributes, participants will see40 statements of the type Apartment A is spa-cious presented individually on a screen over aperiod of a few minutes. Participants are told atthe beginning of the presentation phase that at alater stage in the experiment they will be askedto choose their favoured option. Following presen-

tation of the attributes, participants are assigned toone of three (or sometimes only two) conditions.In the unconscious thought condition participantsare distracted from the task at hand (thinkingabout the best option) by being asked to solve asuccession of anagrams. In the conscious thoughtcondition participants are asked, in contrast, tothink carefully about their choice for a fewminutes, and in the immediate thought conditionparticipants are simply asked to decide as soon asthe presentation phase has finished. This last con-

dition is a necessary baseline (Dijksterhuis, 2004,p. 589) if one wants to draw any conclusions aboutthe relative merits of unconscious thought (asopposed to the possible deleterious effects ofconscious thought).

The take-home message from studies using thisparadigm is that unconscious thought improvedthe quality of decisions (Dijksterhuis, 2004,p. 596). Proponents argue that participants makebetter choices, show better differentiation ofgood and bad options and display more orga-nized memories for material following periodsof unconscious thought. Later we investigatesome of these claims in more detail, but firsttaking the claims at face value we examine theboundary conditions of such effects and attemptto discover why and indeed whether we reallyshould let the unconscious do the work whenfaced with a complex decision (cf. Bekker, 2006).

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2009, 62 (4) 709

UNCONSCIOUS THOUGHT AND COMPLEX DECISION MAKING


5/27

In the first two experiments we investigate howunconscious thought might work in two ways. InExperiment 1 we examine the claim that uncon-scious thought is beneficial because of its superiorability to weight multiple pieces of information. InExperiment 2, we ask whether benefits that have

been observed are due to thesuperiorityof uncon-scious thought or the inferiority of consciousthought. We argue that the standard paradigmpresents an unrepresentative analogue of the wayin which people typically deliberate aboutcomplex decisions, which may disadvantage con-scious thinkers unfairly.

Weighting information in complex decisions

When faced with a multi-option, multi-attribute

decision the acknowledged gold standardmethod is to use some form of weighted-additivemodel (WADD; Dawes & Corrigan, 1974;Gigerenzer & Goldstein, 1996; Payne, Bettman,& Johnson, 1993). Such models consider eachattribute, assign it some weight (degree ofimportance), and then sum the weights of theattributes for each option. The option with thehighest overall summed weighting is thenchosen. In such situations individuals may assignattributes different weights. Thus the idiosyncratic

nature of an individuals weighting scheme is animportant factor to consider when determiningthe quality of an individuals decision. Knowingthe subjective attribute weights assigned by anindividual allows the experimenter to comparethe congruency of objective choices (the optionchosen by an individual) with the option predictedby elicited subjective weights (the option with thehighest summed subjective weighting score forthat individual). Without this knowledge, onedoes not know whether a choice is congruentin the sense that it is the choice predicted by thatparticular individuals weighting schemeor arandom/unpredicted choice.

The unconscious thought theory (UTT;Dijksterhuis & Nordgren, 2006) states thatoptimal weighting of attributes occurs naturallyduring periods of unconscious thought and thatthis optimal weighting leads to a closer connection

between idiosyncratic preferences and objectivechoices for unconscious than for conscious orimmediate thinkers. Conscious processing isclaimed to be limited in capacity, leading peopleto focus only on a subset of available information,which may misrepresent true underlying prefer-

ences. To date, however, the evidence for thiscloser connection is rather limited. Dijksterhuis(2004) reported that the correlation between thedifference in attitude towards the best andworst options in a choice set and the differencein subjective importance ratings of the attributesof these options was higher for unconscious thin-kers than for conscious or immediate thinkers.But differences between conditions were not stat-istically significant.

A more direct way to investigate this optimal

weighting hypothesis is to examine the congruencyof objective choices and subjective weightings: Doparticipants choose the option predicted by theirown (idiosyncratic) subjective weighting schemes?A clear prediction of the theory is that thereshould be a higher proportion of congruent choicesfollowing unconscious thought than followingconscious or immediate thought.

We also tested a slightly more subtle predictionof UTT. The theory states that when faced withmultiple pieces of attribute information, conscious

thought is overwhelmed because of its capacity-limited nature. Unconscious thought on theother hand has unlimited capacity. This character-ization suggests that participants in a conscious(and presumably immediate) thought conditionmight more readily adopt a simpler heuristicway of integrating information (cf. Gigerenzer &Goldstein, 1996) than do those in the unconsciousthought conditions. To examine this possibility weconstructed a choice set in which the best optioncould be calculated in one of two ways. Thesimple method, following Dijsksterhuis, was tocount up the number of positive attributes andchoose the option with the highest number; themore complex method was to use a WADD algor-ithm in which not only the numberof attributeswas important, but also the weight assigned tothose attributes. Application of these two rules,TALLY and WADD, leads to different


NEWELL ET AL.


6/27

options being selected as the best (see Table 1and Method for more details). UTT predicts ahigher percentage of choices of the WADDoption following unconscious thought than fol-

lowing immediate or conscious thought(Dijksterhuis & Nordgren, 2006).

EXPERIMENT 1

Method

ParticipantsThe sample consisted of 44 female and 27 male(M 19.56 years, SD 1.73 years) undergradu-ates from the University of New South Wales.They either participated voluntarily or receivedcourse credits for an introductory psychologycourse.

DesignExperiment 1 was a 3 (mode of thought) 4(apartment) factorial design. The mode of

thought (immediate judgement, conscious delib-eration, unconscious thought) was betweensubjects, with random allocation of participantsinto conditions. The apartment variable wasmanipulated within subjects (all participantsviewed information about all four apartments).

MaterialsThe decision problem was to choose an apartmentto rent near the university. This is a real problemfaced by many undergraduate students, and wasmodelled closely on the materials used byDijksterhuis (2004). A set of pilot ratings on theimportance of 16 relevant attributes was obtainedfrom the same pool of Year 1 psychology students(N 44). For the purpose of Experiment 1, onlythe five dimensions with the highest ratings and

the five with the lowest were used. These dimen-sions were selected to allow for the constructionof options that could be differentiated both interms of the number of positive and negative attri-butes they possessed and in terms of their overallweighted sums. Table 1 shows the set of optionsconstructed. The overall weighting of each apart-ment was computed by adding the individualpositive attribute weightings and then subtractingthe negative attribute weightings. For example,Apartment B has five positive attributes (i.e.,

security of building, rent, crime rate, flatmate,and neighbours), which sum to 39.3 (i.e.,9.0 8.6 8.4 7.9 5.4 39.3); it alsohas five negative attributes, which sum to26.71, thus its overall weighted sum is (39.3 26.7) 12.6. In this choice set the TALLY rulepredicts an A . BC . D order of preference;in contrast the WADD rule predictsB . D . A . C.

Procedure

Participants sat at individual computer terminalsin a single testing cubicle. The experiment beganwith the following information:

Imagine you are planning to rent an apart-ment near university because you have beenliving in a faraway suburb. You are alsointerested in sharing the place with some

Table 1. Distribution of positive and negative attributes and totalweightings of the four apartments in Experiment 1

Weighting(out of 10)

Apartment

Attributes A B C D

Security of thebuilding 8.95 2 2

Rent 8.60 2 2 Crime rate of the area 8.36 2 2Flatmate (friend or

not)7.91 2 2

Size of the apartment 7.56 2 2 Kindness of

neighbours5.41 2 2

View 5.18 2 2Built-in wardrobe 4.70 2 2Direction 4.61 2 2Leisure facilities 4.59 2 2

Frequency of positiveattributes 6 5 5 4

Frequency ofnegative attributes

4 5 5 6

Total weightinga 2 0.17 12.59 2 12.59 0.17

aSum of positive minus sum of negative attributes.




7/27

other students. After some initial house-hunting effort, there are 4 apartmentsremaining in your list (Apartment A, B,C, & D). Since the semester will beginvery soon, youll have to make the decisionas soon as possible.

Following Dijksterhuis (2004), participants wereinformed that after the presentation of the infor-mation they would be asked to choose the apart-ment they perceived to be the best. Thepresentation phase followed in which the 40 attri-butes were shown individually in a random order,with each attribute appearing for 4 seconds.

After this information presentation stage, par-ticipants in the immediate judgement group wereasked directly to choose their preferred apartment.

After choosing, participants were given a 100-point scale (1 very unattractive to 100 very attrac-tive) to indicate their attitude toward each of thefour apartments in a fixed order (i.e., A, B, C, D).After making the ratings, participants were askedto recall the apartment attributes presented earlier.The memory test was included to examine thepossibility that poor performance in the consciousthought condition in previous demonstrations(e.g., Dijksterhuis, 2004; Dijksterhuis et al., 2006)was caused by self-generated memory interference

(Shanks, 2006). In the test, participants typed theattributes that they could remember for each apart-ment and stated whether the attribute was present/absent or good/bad (e.g., low securitybad). Thememory recall task was displayed on a singlescreen with Apartment A on the left, B and C inthe middle, and D on the right. Finally, participantsrated the subjective importance of each of the 10dimensions using a 10-point Likert scale, rangingfrom 1 very unimportant to 10 very important.

The only differences in procedure for theremaining two groups were the following: Theconscious deliberation group was given 4 minutesto think consciously about which apartment todecide upon in between the presentation phaseand the choice task. In this period, a whitescreen appeared, and participants were remindedof the elapsed time after every minute thatpassed. Theunconscious thoughtgroup was given a

distraction task (i.e., solving simple 4 6-letteranagrams) for the same 4-minute period.

Results

Frequency of choiceThe percentage of participants choosing eachapartment in the three conditions is shown inFigure 1. The figure shows that Apartment B,the apartment with the highest weighted sum,was preferred by approximately equal numbers ofparticipants in each condition. Collapsed acrossall conditions, 15% of participants choseApartment A, 65% chose Apartment B, 3% choseApartment C, and 17% chose Apartment D. Achi-square test demonstrated that there was a sig-

nificant difference in the percentage of participantspicking each apartment, x2(3, N 71) 63.37,p, .001. This difference was also found in eachof the experimental conditions: immediate con-dition, x2(3,N 24) 23.00,p, .001; consciouscondition: x2(3, N 24) 19.33, p, .001;unconscious condition: x2(3, N 23) 25.87,p, .001. Follow up tests indicated thatApartment B was chosen by significantly moreparticipants than the next most preferred apart-ment in all conditions: immediate, x2(1) 7.20,p, .008; conscious, x2(1) 5.00, p, .025;unconscious, x2(1) 7.20, p, .008. Thus in allconditions the preferred choice was the apartment

Figure 1. Percentage of participants choosing each apartment inExperiment 1.


NEWELL ET AL.


8/27

predicted by the WADD strategy. However, incontrast to the predictions of UTT there was nodifference in the choice of Apartment B, or anyof the other apartments, as a function of condition:x2s , 1,ps . .607.

Attitude ratingsFigure 2 displays the attitude ratings toward eachapartment on a 1100 scale (with highernumbers indicating a more positive attitude).The attitude ratings mirror the choice data:Apartment B is clearly favoured in all conditions,and the order of preference is very similar acrossconditions (B . A . D C). This ordering iscloser to that predicted by the WADD rule thanthat predicted by the TALLY rule. Collapsedacross conditions, Apartment B was rated as themost attractive (M 73.4, SD 18.8). In theimmediate and the unconscious conditionsApartment B was rated as significantly moreattractive than each of the other three apartments(individual contrasts, Fs . 10.15, p, .01).1

However, for the conscious condition the com-parison of the attitude rating for B (M 69.3)did not differ significantly from that given for A(M 65.2), F, 1. This result provides somesupport for the notion that effortful deliberationcan lead to poorer differentiation between

alternatives (Wilson & Schooler, 1991).

Recall of attributesTable 2 displays the percentages of correctly recalledattributes collapsed across options. Only thedifference between the immediate and unconsciousconditions was significant, F(1, 46) 11.26,p, .01. This pattern contrasts with the predictionthat conscious deliberation would induce self-generated interference and impair memory. If

anything, it appears that distracting participantswith anagrams led them to forget some of theattributes.

Attribute weightings and congruency with choicesThe correlation between the average weighting foreach attribute collected in pilot testing and thoseelicited in Experiment 1 was .94, allowing us tobe confident that our experimental participantshad very similar views to our pilot participantsabout the attributes that are important whenrenting an apartment. For each participant, prefer-ences for each apartment were computed using theWADD rule. Our measure of congruency in

choice is the proportion of participants whochose the apartment predicted by the WADDrule. According to UTT, unconscious thought isbetter than conscious or immediate thought atweighting attribute information optimally. If thisis the case the highest proportion of congruentchoices should be in the unconscious thought con-dition. Table 3 displays the percentages of choicescongruent with the first and second highestweighted alternatives. The difference in summedtotals between the two highest weighted alterna-

tives was often very small (e.g., one or twopoints) so we present congruency with both. Thepattern provides no support for the prediction of

Figure 2. Attitude rating (0100; higher numbers indicate morepositive attitude) towards each apartment in Experiment 1.

1 In the analysis of attitude ratings and memory recall data a set of planned contrasts were conducted using PSY (Bird, 2004) forexamining main effects, simple effects, and interactions. They were done by controlling the per-contrast error rate (PCER), thusconstructing adjusted 99.74% confidence intervals. The Fcritical(1, 68) for this analysis was 9.75 with significance set at the alphalevel of .05.




9/27

UTT. There are small differences between con-ditions (in the opposite direction to the predictionof the theory) but the overwhelming finding is thatthe majority of participants in all conditions chosethe apartment predicted by their idiosyncratic sub-jective weighting profiles.

Discussion

Experiment 1 demonstrated that, regardless of themode of thought that participants engaged in, themajority chose the apartment predicted by aWADD rule. Moreover, the clear majority of par-

ticipants chose the apartment predicted by theiridiosyncratic attribute weightings. In previousexperiments it has not been possible to investigatethis issue because typically only the choice of theobjectively best alternative has been reported,and the congruency of actual choices and thosepredicted by attribute weightings has not beenexamined. There was very little evidence tosuggest that the opportunity to engage in uncon-scious thought improved choices or led to moreoptimal weighting of attributes. There was someevidence from the attitude ratings that consciousdeliberation led to poorer differentiation betweenApartment A (the apartment with the highestnumber of positive attributes) and Apartment B(the apartment with the highest objectiveweighted sum; cf. Levine et al., 1996; Wilson &Schooler, 1991).

An important consideration, given the choicepattern, is whether the choice set made it tooeasy to identify the best option and that thisease masked possible differences between con-ditions. We suggest that this is not the case.

Although Apartment B was the preferred optionit did not completely dominate choicesitaccounted for 65% of choices on average (afigure that is consistent with previous research,e.g., Dijksterhuis, 2004; Dijksterhuis et al.,2006). Moreover, according to UTT if the bestoption had been obvious due, for example, tothe inclusion of one attribute that overshadowedall others, then those in the conscious thoughtcondition should have performed best(Dijksterhuis & Nordgren, 2006).

A second consideration is whether the failure tofind a difference between conditions is due toinsufficient statistical power. Our sample sizes of23 or 24 participants per condition are consistentwith previous research (e.g., Dijksterhuis, 2004;Dijksterhuis et al., 2006), and thus we had noreason to suspect a priori that our samples were

Table 2.Percentage of correctly recalled attributes in Experiments1, 2, and 4

Mean SD

Experiment 1Immediate 50.2 17.4Conscious 43.6 15.4Unconscious 34.7 14.5

Experiment 2Immediate 60.0 18.4Conscious 63.7 18.2Unconscious 62.7 17.7Conscious & information 72.8 16.8

Experiment 4Immediate 47.3 19.6Conscious 42.3 18.2Unconscious 36.3 15.7

Table 3.Percentage of participants choosing the apartment or carcongruent with their subjective weighting schemes

Highest scorea Second highest scoreb Totalc

Experiment 1Immediate 83 17 100Conscious 75 12.5 87.5Unconscious 73 9 82

Experiment 2Immediate 74 17 91Conscious 74 26 100Unconscious 61 39 100Conscious &

information78 18 96

Experiment 3

Immediate 57 23 80Conscious 60 13 73Unconscious 47 30 77

Note: Apartment: Experiments 1 and 2. Car: Experiment 3.aChose option with highest summed subjective weighting

score. bChose option with second highest summed subjectiveweighting score. cTotal choices congruent with first andsecond highest scores.


NEWELL ET AL.


10/27

too small. Post hoc power calculations using G-Power 3 (Faul, Erdfelder, Lang, & Buchner,2007) indicated achieved power to detect signifi-cant differences between conditions (df1) of .92for a large effect, .53 for a medium effect (.30),and .10 for a small effect (using the definitions

of Cohen, 1992). The difference between con-scious and unconscious thought for complexchoices that Dijksterhuis et al. (2006) report corre-sponds to an effect size of f 0.35 (estimatedfrom their Figure 1), which corresponds to aneffect that is slightly greater than medium in size(Cohen, 1992). Similarly, in Dijksterhuis (2004,Exp. 2) the difference in processing style(global vs. specific) between conscious and uncon-scious thought was also moderate (f 0.33),though the differences in choice between con-

ditions were smaller in this study. Therefore,given our power analysis reported above, wemight reasonably have expected to have power inexcess of .6 to detect the kinds of effects thatDijksterhuis data suggest are there.

EXPERIMENT 2

Experiment 1 found little evidence for the sup-eriority of unconscious thought using a choice

paradigm closely modelled on the work ofDijksterhuis (2004). However, the results alsoappear to provide little solace to advocates of adecision analytic approach. Why, when given theopportunity to deliberate about a decision, do par-ticipants not perform better than those who makean instant blink decision or than those who aredistracted prior to making a choice? One possi-bility is that the type of deliberation that consciousthinkers are able to engage in is unrepresentativeof the methods advocated by decision analystsand indeed the nature of conscious thinking thatdecision makers often engage in when faced witha complex choice (Payne, 2007; Shanks, 2006).How often in a consumer choice situation are wefaced with the problems of (a) having all infor-mation about objects in the choice set presentedbriefly and randomly, and (b) having to rely onmemory when considering the attributes of the

alternatives under consideration? More typicallywe would have the objects in front of us, or atthe very least we would consult websites or maga-zines that displayed the information we sought.Recall Franklins advice to Priestly: when each isthus considered, separately and comparatively,

and the whole lies before me, I think I can judgebetter, and am less liable to make a rash step(Franklin, 1772, cited in Goodman, 1931)inother words, have all the information presentwhen making a decision.

Experiment 2 examined these issues by varyingthe way in which information was presented to par-ticipants and whether they had access to it duringthe deliberation process. Specifically, in three con-ditions the presentation phase was changed so thatnow all information about the four options was

presented on an information board (Payne,1976; see Figure 3) rather than being presented dis-cretely and randomly. The aim of this manipulationwas to ensure that participants in all conditions hadsufficient time to encode the necessary informationbefore entering into the different modes of thought(conscious, unconscious, immediate). Interestingly,Dijksterhuis and Nordgren (2006) argue that,Complex decisions are best when the informationis encoded thoroughly and consciously, and thelater thought process is delegated to the

unconscious.. . .

One should look at the list, stopconscious thought for a while, and then wait forthe unconscious to deliver the decision in theform of an intuitive feeling (p. 107). Thus UTTpredicts that the unconscious thought conditionhas even greater potential to shine in Experiment2 than it did in Experiment 1 because now partici-pants have a much better chance of thoroughlyand consciously encoding information. In thefourth condition of Experiment 2, labelledconscious & information, participants werepresented with the information in the standardwayrandomly and discretelybut during thedeliberation period the information board was pro-vided. The experiment thus tests whether (a)unconscious thought improves choices whenencoding is facilitated; and (b) unconsciousthought is better than conscious thought wheninformation is provided (see Dijksterhuis, 2004).




11/27

A second change in the procedure ofExperiment 2 was the time allowed for delibera-tion or unconscious thought. Dijksterhuis andNordgren (2006) state that, longer unconsciousthought should lead to even better decisions thanbrief unconscious thought (p. 99). In unpublisheddata (cited in Dijksterhuis & Nordgren, 2006),Dijksterhuis found that unconscious thinkersmade better decisions when given 7 minutes ofunconscious thought instead of 2 minutes. Totest this prediction, and to provide potentially

more favourable conditions for unconsciousthought, the deliberation/distraction period wasextended from 4 to 8 minutes.

Method

ParticipantsThe sample consisted of 92 undergraduate stu-dents (65 female and 27 male; M 19.42 years,

SD

4.34 years) from the University of NewSouth Wales. They either received course creditsor participated voluntarily.

DesignThe simplest way to describe the design ofExperiment 2 is a 3 (mode of thought) 4 (apart-ment) between by within experiment, but with oneadditional condition: conscious & information.Due to the nature of the unconscious and immedi-ate conditions it was not possible to run these twoas & information conditions. (In the uncon-scious condition participants cannot be given theinformation whilst also being distracted andencouraged notto think about the choice; those

in the immediate condition do not have the oppor-tunity to consult the material prior to choice.)Participants were randomly allocated to the stan-dard mode of thought conditions; the conscious& information condition used new participantsfrom the same student population but was runafter the completion of the first three conditions.

MaterialsAn information board was prepared to display

all the attributes used in Experiment 1(seeFigure 3).2 The order of attributes, which wasalways shown on the left column, was randomized

Figure 3.Example of the information board used in Experiment 2.

2 One of the attributes (i.e., direction) used in Experiment 1 was changed to whether the balcony is facing the sun; this did notchange positivenegative weighting pattern for the four apartments. This change was made because some participants thoughtnorth-facing as a better direction while some thought south-facing as betterpresumably reflecting a mixture of northern andsouthern hemisphere natives in our sample.


NEWELL ET AL.


12/27

for each participant. The information board wasdisplayed on the computer screen either duringthe presentation phase (conscious, immediate,and unconscious conditions) or during thedeliberation period (conscious & informationcondition).

ProcedureParticipants in the immediate, conscious, andunconscious conditions were given a compulsory3 minutes to read the information on the infor-mation board, and then they could decide toproceed directly to the next phase or obtain moretime to encode the information; 3 minutes wereallocated because the information presentationstage in Experiment 1 lasted for 3 minutes.Participants in the conscious condition were

asked to think thoroughly about the apartmentsthroughout an 8-minute period (with remindersat the 2nd, 4th, 6th, & 7th minutes). Those inthe unconscious condition solved anagrams for 8minutes, and those in the immediate conditionmade their choice as soon as they had decided toproceed to the next phase. Participants in the con-scious & information condition were presentedwith the attributes one by one (as in Experiment1) and were then shown an information board tostudy for the 8-minute deliberation period.

Other measures (memory, attitudes, and attributeweightings) were elicited in the same manner asin Experiment 1.

Results

Frequency of choiceThe percentage of participants choosing each apart-ment in the four conditions is shown in Figure 4.The figure shows that Apartment B, the apartmentwith the highest weighted sum, was still preferredby the majority of participants in each condition,but there was more variability in choice in compari-son to Experiment 1. Collapsed across all con-ditions, 9% of participants chose Apartment A,71% chose Apartment B, 2% chose Apartment C,and 18% chose Apartment D. A chi-square testdemonstrated that there was a significant differencein the percentage of participants picking each

apartment, x2(3, N 92) 101.83, p, .001.This difference was also found in each of thefour conditions: immediate condition, x2(3,N 23) 15.09, p, .01; conscious condition,x

2(3, N 23) 41.52, p, .001; unconsciouscondition, x2(3, N 23) 22.74, p, .001;conscious & information condition, x2(3,N 23) 35.25, p, .001. Follow-up tests indi-cated that Apartment B was chosen by significantlymore participants than the next most preferred

apartment in all conditions, though the effect wasmarginal for the immediate and unconscious con-ditions: immediate, x2(1) 3.55, p, .059; con-scious, x2(1) 11.63, p, .001; unconscious,x

2(1) 3.86, p .05; conscious & information,x

2(1) 10.71,p, .001. Consistent with the find-ings of Experiment 1, the differences in theproportion of choices of Apartment B, or any ofthe other apartments, across conditions, failed toreach significance: x2 , 6.01, ps . .11. (Thelargest x2 value was for the comparison ofApartment A choices across conditions; thismarginal effect was driven by the relatively highproportion of choices of A in the immediate con-dition compared to that in the other threeconditions.)

Figure 4 shows that the highest proportions ofchoices for Apartment B are in the two consciousthought conditions. To explore the choice pattern

Figure 4. Percentage of participants choosing each apartment inExperiment 2.




13/27

further a chi-square analysis compared the percen-tage of participants choosing Apartment B in thetwo conscious conditions combined with the twoother conditions. The comparison allowed us toexamine whether some form of consciousdeliberationeither after information has been

effectively encoded or when the information is pro-videdcan improve choices relative to immediateor unconscious thought conditions. The analysisrevealed a significant difference between thesecombined conditions when compared to immediatethought: x2(1, N 92) 4.39, p, .05, but notwhen compared to unconscious thought: x2(1,N 92) 3.05, p. .05. The results suggest thatconscious deliberation can sometimes be beneficialfor decisions made in this choice task.

Attitude ratingsThe attituderatings for each apartment are showninFigure 5. Thegeneral patternof attitudes wassimilaracross the four groups, with all showing an orderingof B . A . D . C (closest to that predicted by theWADD rule). In every condition the ratings ofApartment B were significantly higher than theaverage of other apartments, Fs(1, 91) . 23.0,ps , .001, and in every condition Apartment Cwas rated significantly lower than the average of

the other apartments,Fs(1, 91) . 29.5,ps , .001.The failure to differentiate Apartments A and Bfound in the conscious group of Experiment 1 wasnot apparent in either of the conscious conditionsof Experiment 2.

Recall of attributesThe percentages of correctly recalled attributes aredisplayed in Table 2. In the three conditions inwhich the attributes were present on an infor-mation board throughout the study phase, per-formance was very similar and above the averageexhibited in Experiment 1 (62.2% compared to42.9%). In the conscious & information conditionperformance was better again, presumably reflect-ing the greater amount of time for examining theattribute information in this group. However, no

significant differences in recall were foundamong the four conditions (largest F 6.05 forimmediate versus conscious & information com-parison,Fcritical 9.75see Footnote 1).

Attribute weightings and congruency with choicesFor each participant, preferences for each apart-ment were computed according to the WADDrule. Table 3 displays the percentages of partici-pants whose choices were congruent with thefirst and second highest weighted alternatives.

Two aspects of the results are of interest: (a) Atleast descriptively, more participants choose con-gruently with their highest weighted option inthe three conditions that do not involve uncon-scious thought; (b) when both first and secondhighest weighted options are taken into account,the clear majority of participants across all con-ditions chose the apartment predicted by theirsubjective weighting profile.

Discussion

The majority choice across conditions was again theoption favoured by the WADD rule (Apartment B)although there was more variability in choicepatternsespecially in the immediate condition.This variability lends weight to the argument thatthe choice set is not simply dominated by anobvious best alternative that masks potential

Figure 5.Attitude rating (0100; higher numbers indicate morepositive attitude) towards each apartment in Experiment 2.


NEWELL ET AL.


14/27

differences in choice across conditions. The con-gruency data again emphasize that the variabilityin choice is not random noisethe clear majorityof participants in all conditions chose the optionpredicted by their subjective weighting profiles (ortheir second highest option). The data from the

two conscious conditions provided an importantinsight: When participants either have the opportu-nity to encode information thoroughly or have theinformation in front of them, some period of con-scious deliberation can benefit choice (at least asmeasured by the objective best option). Despiteattempting to improve the conditions for uncon-scious thought (thorough encoding and extendedperiod of distraction) there was again very little tosuggest that unconscious thought improved choiceor led to better differentiation between options.

EXPERIMENT 3

In Experiments 1 and 2 we took the claims aboutunconscious thought at face value and attemptedto explore the boundary conditions of the effects.The findings thus far present an unexplained con-tradiction: Although we tried to follow the pro-cedures used in previous studies and modelledour stimuli closely on those used before, we were

unable to replicate the finding of unconsciousthought leading to superior choices. However,there is always a possibility that idiosyncrasies inprocedures or the particular choice set that weused in Experiments 1 and 2 masked potentialdifferences between the thought conditions. Thusin Experiment 3 we attempted a direct replicationof the result that provides the most compelling evi-dence to date of the benefits of unconsciousthought: Dijksterhuis et al. (2006, Exp. 1). In thestudy participants (N 40) were presented witha choice between four fictional cars described by12 attributes each. The approximate percentagesof participants choosing the best car were 60% fol-lowing unconscious thought but only 25% (i.e.,chance level) following conscious thoughta sig-nificant difference (an immediate group was notincluded). Experiment 3 attempted to replicatethis result, and, following the principle that a

replication should be more likely if sample sizesare increased, we used 50% more participants percondition. We also included an immediatethought group as a baseline.

MethodParticipantsThe sample consisted of 90 undergraduate stu-dents (56 female and 34 male; M 19.3 years,SD 2.02) from the University of New SouthWales. They received course credit for partici-pation. None of the participants had taken partin Experiments 1 or 2.

DesignExperiment 3 was a 3 (mode of thought) 4 (car)

factorial design. The mode of thought (immediatejudgement, conscious deliberation, unconsciousthought) was between subjects, with random allo-cation of participants into conditions (ns 30).The car variable was manipulated within subjects(all participants viewed information about allfour fictitious cars).

MaterialsThe materials were obtained from the supplemen-tary information document downloaded from the

Science website (Dijksterhuis et al., 2006). Thechoice set consisted of four fictitious cars eachwith 12 attributes. The attributes were either posi-tive (for example, The Hatsdun has goodmileage) or negative (for example, The Dasukahas poor mileage). Hatsdun was depicted with75% positive attributes,Kaiwa was depicted with58% positive attributes, Dasuka was depictedwith 50% positive attributes, and Nabusi wasdepicted with 25% positive attributes (seeAppendix A for the attribute list).

ProcedureThe procedure was identical to that of Experiment1, with the exception that participants were told tochoose the best car from the set rather than thebest apartment. Up to the choice phase the pro-cedure matched that of Dijksterhuis et al. (2006)as closely as possible; following the choice, attitude




15/27

ratings, recall measures, and attribute ratings werecompleted.

Results

Frequency and congruency of choiceFigure 6 shows the percentage of participantschoosing each car in the three conditions. Thefigure shows that the Hatsdun, the car with 75%positive attributes, was chosen by the majority ofparticipants in each condition. Chi-square ana-lyses showed there was a significant difference inthe percentage of participants picking each car,in each of the three conditions: immediate con-dition, x2(3,N 30) 19.87,p, .01; consciouscondition, x2(3, N 30) 22.00, p, .01;unconscious condition, x2(3, N 30) 9.73,

p, .05. In the immediate and conscious con-ditionsHatsdunwas chosen by significantly moreparticipants than the next most popular option:immediate, x2(1) 4.17, p, .05; conscious,x

2(1) 4.84, p, .05; in the unconscious con-dition only the comparison with the least popularoption was significant, x2(1) 10.28, p, .001.Importantly, there was no significant differencein the percentage of participants choosing theHatsdun in the three conditions, x2(2,N 90) 1.67,p . .05.

The congruency data are displayed in Table 3.Two aspects of the data are noteworthy: First, in

line with the findings of Experiments 1 and 2, thelowest percentage of choices congruent with thehighest weighted option is in the unconsciousthought condition; second the overall levels of con-gruence are slightly lower than those in the first twoexperiments. This latter pattern might be due to our

student participants having somewhat less clearlydefined preferences about car attributes than apart-ment attributes.

Our primary interest in Experiment 3 was thechoice data; thus in the interests of space we donot report a full analysis of the attitude andmemory data. To summarize: The attitude datashowed a clear preference for the best car in allconditions but no suggestion that unconsciousthought improved differentiation of options rela-tive to the other two modes of thought (consistent

with Experiments 1 and 2). There were no signifi-cant differences in the amount of informationrecalled across conditions, with all groups recallingapproximately the same number of attributes(average of 4 collapsed across options), a patternthat is comparable to the recall data ofExperiment 1 (see Table 2).

Discussion

Experiment 3 attempted to replicate the clearestfinding in the literature of the superiority ofunconscious thought for complex decisions.Despite using the same stimulus materials, follow-ing procedures as closely as possible, and increas-ing the sample size we were unable to find anadvantage for the unconscious thought condition.It is not clear why we were unable to replicate thepattern reported by Dijksterhuis et al. (2006). Atthis stage, perhaps the best that can be said isthat the stark contradiction in the data remainsto be solved through further theoretical under-standing of how and why the unconsciousthought effect arises. What is clear fromExperiment 3 is that adopting the same materialsand procedures but using a different sample popu-lation does not guarantee that the unconsciousthought effect will be observed (see also Acker,2008).

Figure 6. Percentage of participants choosing each car inExperiment 3.


NEWELL ET AL.


16/27

EXPERIMENT 4

Our initial aim in this set of experiments was toreplicate the unconscious thought effect and thento examine boundary conditions in an attempt toelucidate possible underlying mechanisms. Our

inability to replicate the effect is to our mindsimportant to document in its own right; however,it would be more satisfying to provide some expla-nation for the similarity in choice followingimmediate, conscious, or unconscious thought thatwe observed in all three experiments. Althoughsuch an approach may not directly resolve thecontradiction between our findings and those ofDijksterhuis and colleagues it has the potential toshed light on why our participants showed consist-ent choice behaviour across thought conditions.

Rethinking the choice task

Participants in the standard choice task are toldprior to the presentation of attribute informationthat at a later stage in the experiment they willbe expected to choose the best option.Attribute information is then presented discretelyand sequentially about each option. Setting upthe task in this way invites participants to treat it

as an online judgement task (Hastie & Park,1986)one in which a participant forms, andpossibly updates, a judgement as attribute infor-mation is encountered. This contrasts with end-of-sequence judgements (Hogarth & Einhorn,1992), or judgements or decisions from memory(see Broder & Schiffer, 2003) where a judgementis made after the relevant information has beenencoded. Given that a choice is expected, it islikely that a participant sequentially updatesimpressions of each option as more evidence isencountered (Edwards, 1968; Hogarth &Einhorn, 1992). The judgement reached at theend of the presentation phase is then the onerelied upon when the decision is asked for (e.g.,after the period of deliberation or distraction).Hastie and Park (1986) argue that so many judge-ments are made online (spontaneously) becausewhen a new judgement must be made in the

absence of perceptually available evidence, subjectsrely on previous judgements rather than remem-bered evidence (p. 263, emphasis added).Self-report data obtained in Experiment 2 are con-sistent with this interpretation: In all but the con-scious & information condition (in which evidence

was available during deliberation) the majority ofparticipants indicated that they had arrived attheir decision prior to engaging in deliberationor distraction.

If participants approach the experiment as anonline judgement task, then two predictionsfollow: First, there should be little difference inchoice between modes of thoughtthis predictionflies in the face of unconscious thought theory(Dijksterhuis & Nordgren, 2006) but is consistentwith the pattern in Experiments 1 3, and second,

that choices should be affected by the order inwhich attribute information is encountered(Hogarth & Einhorn, 1992). This latter predictionis yet to be investigated because in previous exper-iments care has been taken to randomize the orderof attribute presentation. In Experiment 4 attri-bute order was manipulated to investigate thisprediction.

Several investigations of impression formationhave documented recency effectsin which pieces ofevidence encountered later are given more weight

than those encountered earlier (Denrell, 2005;Dreben, Fiske, & Hastie, 1979; Hogarth &Einhorn, 1992). Typically recency effects are foundwhen judgements are required after each piece ofinformation is presented, although that is not thecase with the current choice paradigm (a judgementis only required at the end); because participantsneed to update their impressions of four separateoptions the processing required is perhaps akin todeciding which of the four is ahead after each attri-bute is presented. In standard impression formationtasks participants evaluate only a single option (e.g.,a person) for, for example, likeability (Dreben et al.,1979) or suitability for a job (Hastie & Park, 1986).

To examine whether recency effects could befound in a more complex choice task we adaptedthe task used in Experiment 3 to one involvingtwo cars each described by 20 attributes (10 positiveand 10 negative; the names Hatsdun and Nabusi




17/27

were used and were counterbalanced across con-ditions; for ease of explanation we use the genericterms Car 1 and Car 2). (Two cars were usedrather than four to enable a more straightforwardmanipulation of attribute orderingsee Method.)On a simple tally rule the two cars were balanced

in terms of their goodness as choices because theyhad equal numbers of positive and negative attri-butes. The key question of interest was whetherthis overall balance was affected by the order inwhich attribute information was encountered. Iforder does have an effect then we would have evi-dence to support the claim that participants treatthis task as an online judgement task. This in turncould explain why we found little difference inthe choices made across modes of thought inExperiments 1 to 3. If participants are making

their choice at the end of the presentation phaseand then retrieving that decision when asked, onewould expect choices to be similar regardless ofthe mode of thought subsequently engaged in.

A second question is whether the mode ofthought interacts with any biases induced by attri-bute ordering. Arguably, proponents of unconsciousthought theory would predict that because uncon-scious thought is capable of more optimal weightingof information than conscious thought then itshould be least susceptible to ordering biases

(Dijksterhuis & Nordgren, 2006). That is, after aperiod of distraction (unconscious thought) anybias to choose one option should be counteractedby the optimal restructuring of information inmemory. In direct contrast to this prediction, someexperiments on impression formation have foundevidence that recency effects are stronger when pres-entation of attributes and judgements are interp-olated with distraction tasks (Dreben et al., 1979).

The presentation order of attributes of the carswas manipulated in an ascending (predominantlynegative attributes presented first followed by pre-dominantly positive attributes) or descendingorder (predominantly positive attributes followedby predominantly negative attributes; see

McAndrew, 1981). The presentation order wascounterbalanced for the two cars. In the Car1 ve last condition attributes of Car 1 were pre-sented in ascending order, and Car 2 attributeswere presented in descending order. In the Car1 ve first condition attributes of Car 1 were

presented in descending order, and Car 2 attri-butes were presented in ascending order. Recencyeffects would be indicated by the tendency tochoose the car about which a participant hasrecently encountered positive information: Car 1in the Car 1 ve last condition and Car 2 inthe Car 1 ve first condition.3

Method

Participants

The sample consisted of 119 undergraduate stu-dents (86 female and 33 male; M 19.7 years,SD 3.4) from the University of New SouthWales. They received course credit for partici-pation. None of the participants had taken partin Experiments 1, 2, or 3.

DesignExperiment 4 was a 3 (mode of thought) 2(order) 2(car) with the first two factors manipu-lated between subjects and the third within subjects.

Participants were randomly assigned to the six con-ditions resulting from crossing mode of thought(immediate, conscious, unconscious) with order(Car 1 ve last, Car 1 ve first). There was annof 20 in each condition except the unconscious,Car 1 ve first condition, which had 19.

MaterialsAppendix B contains a complete list of the attri-butes used and the order in which they were pre-sented. The 12 attributes used in Experiment 3were supplemented with 8 new attributes (e.g.,has a parking sensor) to construct the lists. ListA in Appendix B shows the ascending patternused in the Car 1 ve last condition in which

3The alternative predictionthat information encountered first will have a greater influence (i.e., primacy)could also be made,but previous research led us to predict that recency effects would be more likely in this paradigm (e.g., Denrell, 2005, p. 962), henceour framing of the hypothesis with respect to recency.


NEWELL ET AL.


18/27

predominantly positive attributes about Car 1 areshown later in the list. List B shows the descend-ing pattern used in the Car 1 ve last conditionin which predominantly negative attributes aboutCar 1 are shown later in the list.

ProcedureThe procedure was identical to that of Experiment3. Memory recall, attitude ratings, and attributeratings were all elicited in the same manner.

Results

Frequency of choiceFigure 7 displays the percentage of participantschoosing Car 1 in the six conditions. It is import-ant to note that because there are only two alterna-

tives in this experiment the percentage ofparticipants choosing Car 2 in each condition issimply 100 minus each of the values displayed in

Figure 7. The overall pattern suggests an effectof recency. This is evidenced by the left-hand bar(Car 1 ve last) being higher than the right-hand bar (Car 1 ve first) in each mode ofthought conditionindicating that more peoplechose Car 1 when they had recently seen positive

information about it than when they had seenpositive information about it at the start of thepresentation phase. This pattern is clearly mostevident in the unconscious thought condition.

Hierarchical log-linear analysis was used toexamine the interaction between mode ofthought and sequence order for the choice data.Thought condition (immediate vs. conscious vs.unconscious), order (Car 1 ve last vs. Car1 ve first), and choice (Car 1 vs. Car 2) wereentered as factors. The three-way effect

(Thought Condition Order Choice) was notstatistically significant, x2(2) 3.37, p .186.The two-way order-by-choice effect was

Figure 7.Percentage of participants choosing Car 1 in Experiment 4.




19/27

significant, x2(2) 12.45, p, .001the pro-portion choosing Car 1 was significantly higherwhen participants encountered its positive attributeslater. As order effects were a key concern in thisexperiment, we used three separate chi-square teststo perform planned comparisons of this effect of

order upon choice separately for each thoughtcondition. There was a large and statisticallysignificant effect in the unconscious condition,x

2(1) 11.35,p .001, f 0.54, consistent witha greater influence of recently viewed material.This effect was small and nonsignificant in theimmediate condition, x2(1) 0.92, p .337,f 0.15. In the conscious thought condition theeffect was moderate and approached significance,x

2(1) 3.14,p .077, f 0.28.

Attitude ratingsThe attitude ratings revealed a similar picture to thechoice data. Analyses of variance (ANOVAs) foundno main effects of conditions but a significant inter-action between order and car, F(1, 113) 5.86,p, .02, arising because on average Car 1 was ratedhigher than Car 2 in the Car 1 ve last condition(Car 1 60.5, Car 2 57.7; 1100 scale), withthe reverse being true in the Car 1 ve first con-dition (Car 1 55.9, Car 2 63.3). The interactionbetween order andthought condition was also signifi-

cant, F(2, 113)

4.70, p,

.02, because the meandifference in attitude ratings for the two cars washigher in the unconscious (mean difference of 8.7)and the conscious (mean difference of 7.7) than inthe immediate condition (mean difference of 2.0).The three-way interaction between car, order, andthought condition was not significant (F, 1).

Recall of attributesThe primary interest in the recall data was toexamine whether mode of thought had an overalleffect on the proportion of correctly recalled attri-butes. Table 2 shows the percentage of correctlyrecalled attribute rates collapsed across order andthe two cars. A main effect for mode of thoughtwas found, F(2, 118) 3.77, p, .03, with thedifference between immediate and unconsciousbeing the only significant contrast (95% CI:.013 .21). This pattern is consistent with that

found in Experiment 1 and demonstrates that par-ticipants in the unconscious thought conditionexhibited the poorest memory. In a secondaryanalysis we examined the effects of order onrecall of information about the two alternativesseparately. This revealed a marginal but reliable

advantage for recall of attributes of Car 1 (42.9%vs. 41.2% for Car 2), F(1, 113) 4.96, p, .03,and an effect of order indicating that participantsin the Car 1 ve first conditions recalled morecorrect attributes overall (47.8%) than those inthe Car 1 ve last conditions (36.4%), F(1,113) 13.05, p, .001. The reason for thislatter effect is unclear, but its presence does notimpact the interpretation of the choice data.There were no significant interactions betweenorder, car, and condition.

Attribute weightings and congruency with choicesParticipants provided ratings for each of the 20attributes used to describe Cars 1 and 2. Fromthese ratings preferences for each car were com-puted according to a WADD rule. The percen-tages of participants whose choices werecongruent with that predicted by WADD were57.5%, 55%, and 53.8% in the immediate, con-scious, and unconscious groups, respectively.These percentages are lower than those seen in

the earlier experiments; however, it is importantto note that with only two alternatives both with20 attributes (balanced in terms of the numberof positive and negative) one would only predictsubtle differences in overall weightings. Indeedthe mean absolute difference in weightingsbetween Car 1 and Car 2 was only 5.85 pointson a possible range of 0100 (i.e., a difference of5.8%). This finding of overall ambivalencebetween the two cars (in terms of attribute weight-ings) makes the recency data more compelling as itdemonstrates that order of information rather thanidiosyncratic weighting schemes was primarilyresponsible for choices.

Discussion

Experiment 4 found evidence of recency effects onthe choices made in a two-alternative choice task.


NEWELL ET AL.


20/27

The effect was present in all three modes ofthought condition but was strongest in the uncon-scious thought condition. The pattern of data sup-ports the claim that participants engage in onlineupdating of the options, a process that is affectedby the order in which pieces of information are

encountered. Finding that participants appear totreat the task as an online updating task providesa plausible explanation for why we found similarpatterns of choice across modes of thought inExperiments 1 to 3. If, as Hastie and Park(1986) suggested, participants retrieve judgementsmade online when subsequently asked for adecision in this type of task, we would expectlittle difference between conditions when theorder of information is randomized. Such anexplanation begs the question why Dijksterhuis

and colleagues find differences between modes ofthought when randomized lists are used, but atthis stage, as noted earlier, the theoretical basisfor the unconscious thought effect is not yet suffi-ciently understood to resolve such contradictions.4

An important contribution of Experiment 4 isthe finding that when the order of attributes ismanipulated participants in the unconsciousthought condition appear to be most affected(though the data are suggestive, rather than defini-tive, with respect to this). This effect is contrary to

the prediction of unconscious thought theory andsuggests that a period of distraction can enhancerecency effects (cf. Dreben et al., 1979) and, inthis case, lead to poorer choices. This conclusionis reinforced by the attribute rating data, whichshowed that participants were largely ambivalentbetween the two options and yet were influ-encedheavily in the case of the unconscious

thought conditionby the order in which attri-butes were encountered.

GENERAL DISCUSSION

In contrast to the advice of decision analysts,recent claims in the media have urged peopleto Leave big decisions to your unconscious(Munro, 2006). In four experiments we examinedthe techniques on which these claims were based.All four experiments showed that when thecongruency of choices was examinedby compar-ing the choice predicted by the sum of weightedattributes with the actual choicethe majority ofparticipants chose options congruent with theirhighest or second highest weighted option.

This effect held regardless of the mode ofthought engaged in but, if anything, was clearerin the conditions that did not involve unconsciousthought. This finding is both novel and importantas previous investigations of this choice paradigmhave not reported sufficiently detailed data toexamine congruency of choice. Experiment 4investigated the idea that the choice task involvesthe online updating of options throughout thepresentation phase. We found evidencein theform of recency effectsto support this claim

and argued that framing the task in this wayprovides a plausible explanation for why wefound little difference between the modes ofthought in Experiments 1 and 3. In starkcontrast to claims in the literature and themedia we found very little evidence of the super-iority of unconscious thought for complexdecisions.

4We believe that there is some reason to be cautious in interpreting some of the differences between thought conditions reportedin previous studies. For example in Dijksterhuis, 2004, Experiment 2 (one of the few studies to report choice proportions as opposed

to attitude differences) the percentage of participants choosing the best apartment was 59.3%, 47.1%, and 36.4% in the uncon-scious, conscious, and immediate groups respectively. The comparison between the unconscious and the immediate group was sig-nificant on a chi-square test but only using a one-tailed test with a pvalue of .04 (it is questionable whether one-tailed tests ofsignificance are suitable in this instance given that a number of alternative theories that would predict effects in the opposite direction:Kimell, 1957; Howell, 2002). The crucial comparison between the conscious and unconscious groups was not statistically significant.In the same paper in Study 1 the difference in attitude towards the best and worst options between the unconscious and immedi-ate conditions was significant only for males in the sample (15 out of a total of 63 participants); there was no significant differencebetween the conscious and the unconscious condition. In Study 3 the difference between the conscious and unconscious conditions

was significant but only for the males in the sample (38 out of 145).




21/27

Resolving a contradiction

Caution should always be taken when interpretingfailures to replicate. Although the experimentsreported here provide important insights into thecongruency of judgements, the role of online updat-

ing, and the effect of providing attribute infor-mation, it is the absence of significant differencesbetween the modes of thought that is perhapsmost contentious. One reaction to this null resultis to dismiss it (e.g., the experimenters must havedone something wrong). We used four differentchoice tasks, including a direct replication of theDijksterhuis et al. (2006) study, and even alteredsome conditions in an attempt to facilitate uncon-scious thought (e.g., extending the period of distrac-tion, providing information for thorough encoding)but could not find an unconscious thought effect. Ifwe did do something wrong it is not clear to uswhat that something was. This is not to deny out-right the existence of the effect, rather it is to suggestthat the appropriate reaction to these findings is toacknowledge that sometimes contradictory patternsappear in the literature and that their presence forcesfurther theoretical and empirical development (e.g.,see C. J. Berry, Shanks, & Hensons, 2006, reexami-nation of Merikle & Reingolds, 1991, unconsciousmemory effects). The direction that this develop-ment will take is not yet clear, but we speculate,

along with other recent investigations (e.g., Acker,2008; Payne, Samper, Bettman, & Luce, in press)that it might result in a need to temper some ofthe bold and general claims made about the benefitsof unconscious thought. For researchers familiarwith the trajectory of research on related topicssuch as implicit learning, in which initial claimsabout the sophistication of unconscious processeswere subsequently unsupported, such tempering ofconclusions and reconsideration of the evidencewill not come as a surprise (e.g., Lagnado et al.,

2006; Shanks, 2005).

Other indices of unconscious thought

Proponents of UTT couldargue that other indices ofbehaviour may have greater potential to reveal thebenefits of unconscious thought for decision

making. We suggest that the choice data from con-trolled, laboratory experiments are the most compel-ling and important (it is after all what the personultimately chooses to do that is of most conse-quence); nonetheless it is possible that the benefitscan be more clearly seen in other dependent

measures such as attitude ratings, memory for attri-butes, or postchoice regret (Dijksterhuis &Nordgren, 2006; Dijksterhuis & van Olden, 2006;Wilson et al., 1993)or indeed from studies per-formed outside the laboratory (e.g., Dijksterhuiset al., 2006, Studies 3 and 4). Our preference is toobtain clear demonstrations of the phenomena ofinterest in the laboratory before speculating onwhat might or might not be underlying effects seenin peoples recollections of consumer choices (forexample); nevertheless we can examine some of the

other dependent measures that we collected in ourexperiments to look for evidence of the effects ofunconscious thought.

Inspection of the attitude ratings collected inExperiment 1 did indeed provide insight intopossible deleterious effects of conscious delibera-tion, but again no evidence for improved differen-tiation between the best and worst optionsfollowing unconscious thought could be found.In terms of memory, there was no suggestionthat overall memory was improved following

unconscious thought (it was in fact poorer thanthat following immediate thought in Experiment1 and Experiment 4); however, the claims madeabout the benefits of unconscious thought formemory are slightly more subtle than a simpleoverall improvement. According to the bottom-up-versus-top-down principle, of UTT, a distinc-tion exists between conscious and unconsciousthought in terms of schematic structures.Conscious thought is regarded as being guidedby expectancies and schemas, which can lead todistortions in the representation of material (e.g.,an overreliance on a subset of plausible orexpected information, cf. Wilson & Schooler,1991). Unconscious thought, in contrast, worksfrom the bottom up and delivers an objectivesummary judgement via the slower process ofinformation integration (Dijksterhuis &Nordgren, 2006). This distinction is demonstrated


NEWELL ET AL.


22/27

in the polarization of memory whereby uncon-scious thinkers recognize more positive infor-mation than negative for desirable options andmore negative information than positive for unde-sirable options. Such polarization is not as appar-ent following conscious and immediate thought

(Dijksterhuis, 2004, Study 4).We examined the memory data of Experiment 1

(the experiment that most closely resembled that ofDijksterhuis, 2004, Study 4) for possible polariz-ation effects. Figure 8 displays the percentage ofaccurate recall of positive and negative attributes ofthe objectively best apartment (Apartment B) andworst apartment (Apartment C). The pattern isvery similar to that found by Dijksterhuis (2004).In all conditions there was a tendency to polarizeinformation: Memory for positive attributes of the

best apartment was better than memory for nega-tive attributes, with the reverse true for the worstapartment. A significant apartment by attributevalence interaction supports this interpretation,F(1, 61) 10.04, p, .003. This effect appearsstrongest in the unconscious thought condition butthe three-way interaction between mode ofthought and memory for positive and negativeattributes that UTT predicts was not significant(p. .60). Dijksterhuis (2004) also reported a non-significant three-way interaction.These data

suggest that a period of distraction can lead tosubtle changes in the memory for attributes;however, whether this change is due to active and

qualitatively different processes of consolidationand organization or a reflection of poorer overallmemory (see Table 2) is unclear.

This point is worthy of some further consider-ation. Dijksterhuis (2004) concedes that his test ofmemory for attributes is actually an allocation

testparticipants were shown attributes and hadto identify the source (room-mate A, B, or C).Therefore, from his data on correct recognition(allocation) one can infer incorrect recognition/allocation. One can see (Dijsksterhuis, 2004,Fig. 1) that incorrect allocation is high in thecase of negative attributes associated with attrac-tive options and positive attributes associatedwith unattractive options. Therefore, the datathat Dijksterhuis presents as evidence of polariz-ation in memory, and discusses as a positive

feature of unconscious thought, may simplyreflect a rather poor memory for attributes in hisunconscious thought conditions. This poormemory is presumably allied to a tendency to mis-allocate attributes, of which one has little or nomemory, on the basis of preferences that one hasrecently acquired. For this reason, we feel thatfree recall tests of attributes are much more likelyto give genuine insight into the representation ofattribute information. Furthermore, there is noreason to suppose that memory polarization

precedes or explains patterns of choice. Forinstance, if, as seems to be the case, the judgementsthat drive choice are made online, memory forattributes may be influenced by the valence ofattitude (rather than attitude being influenced byattribute information stored in memory).

Reliability of ratings

The choice congruency data revealed interestinginsights into the relation between actual choicesand those predicted by a weighted additive model(WADD). However, their interpretation is notwithout potential problems. Ratings of attributeswere always taken after choices had been made,raising the possibility that there was a demand forparticipants to appear consistent (i.e., if they hadjust chosen the apartment without a wardrobe theyshould not rate having a wardrobe as an important

Figure 8. Percentage of accurate recall of positive and negativeattributes of the objectively best apartment (Apartment B) andworst apartment (Apartment C) in Experiment 1.




23/27

attribute). Wilson and Schooler (1991) attempted tocounter this problem by having participants believethat their reasons for making particular choices inthe jams study would not be required. There arepotential problems with both before and after elici-tation methods, but on balance, we felt that eliciting

ratings after the choice was less problematic. If theattributes had been prerated this may have increasedthe potential for all participants to approach theexperiment in a more analytic (conscious) frameof mind, thus (in the eyes of UTT proponents atleast) hampering unconscious thought.

Other decision models

One of our key aims was to examine the claim that

unconscious thought is better than consciousthought in weighting attribute information. Onetest of this prediction was to examine the pro-portion of choices consistent with a simpledecision strategy that requires no attribute weight-ing (TALLY) and a more complex one that does(WADD). It is important to note, however, thatthere are many more ways in which attribute infor-mation can be used to arrive at a decision. We didnot attempt to construct a choice set that wouldallow us to distinguish between other decision

models but clearly this is an important goal forfuture research. For example, the choice set usedin Experiments 1 and 2 does not distinguishbetween the predictions of WADD and asimpler lexicographic model such as take-the-best (Gigerenzer & Goldstein, 1996)bothwould choose Apartment B. (The latter does soby choosing on the basis of the single best attri-bute that discriminates options in the choice set,when attributes are searched in order of import-ance.) This is instructive because it means that aparticipant who chose Apartment B might havearrived at that choice using a weighted sum of allthe attributes, or she might have utilized asimpler mechanism like TTB. Given the interestin discovering the conditions under which peopleadopt such different strategies (see Broder, inpress; Newell, 2005; Newell & Shanks, 2007, forreviews) in future research it would be intriguing

to design choice sets in which a lexicographicstrategy and WADD make different predictions.

CONCLUSION

Claims about the powers of the unconscious arealways appealing and seductive, and the claimthat we do not need to think about complexdecisions is very tempting. Although proponentsof this idea may argue that their original con-clusions were not so bold (see, for example, thereplies by Dijksterhuis et al. to the letters ofBekker, 2006, and Shanks, 2006, in Science), theoverwhelming message from these studies wasthat choices in complex matters. . .should beleft to unconscious thought (Dijksterhuis et al.,

2006). We believe that existing empirical evidenceand theoretical understanding is not sufficientlyclear to warrant this conclusion. On the contrary,our data suggest that unconscious thought ismore susceptible to arbitrary ordering effects, andthat if conscious thinkers are given adequatetime to encode material, or are allowed toconsult material while they deliberateconditionsthat reflect Benjamin Franklins sage advicetheirchoices are at least as good as those madeunconsciously.

Original manuscript received 19 February 2008Accepted revision received 18 April 2008

First published online 23 August 2008

REFERENCES

Acker, F. (2008). New findings on unconscious versusconscious thought in decision making: Additionalempirical data and meta-analysis. Judgment andDecision Making, 3, 292303.

BBC News Online. (2006, February 17). Sleep on it,decision makers told. Retrieved May 10, 2007, fromhttp://news.bbc.co.uk/2/hi/health/4723216.stm

Bekker, H. L. (2006). Making choices without deliber-ating.Science,312, 1472.

Berry, C. J., Shanks, D. R., & Henson, R. N. A. (2006).On the status of unconscious memory: Merikleand Reingold (1991) re-examined. Journal of


NEWELL ET AL.


24/27

Experimental Psychology: Learning, Memory, andCognition,32, 925934.

Berry, D. C., & Broadbent, D. E. (1984). On therelationship between task performance and associ-ated verbalisable knowledge. Quarterly Journal of

Experimental Psychology,50A, 124.Bird, K. D. (2004).

Analysis of variance via confidenceintervals.London: Sage Publications.Broder, A. (in press). The quest for take the best

insights and outlooks from experimental research.In P. Todd, G. Gigerenzer, & the ABC ResearchGroup (Eds.), Ecological rationality: Intelligence inthe world.New York: Oxford University Press.

Broder, A., & Shiffer, S. (2003). Take the best versussimultaneous feature matching: Probabilistic infer-ences from memory and effects of representationformat. Journal of Experimental Psychology: General,132, 277293.

Cohen, J. (1992). A power primer. Psychological Bulletin,112, 115159.

Dawes, R. M., & Corrigan, B. (1974). Linear modelsin decision making. Psychological Bulletin, 81,95106.

Denrell, J. (2005). Why most people disapprove of me:Experience sampling in impression formation.Psychological Review,112, 951978.

Dienes, Z., & Fahey, R. (1998). The role of implicitmemory in controlling a dynamic system. Quarterly

Journal of Experimental Psychology,51A, 593614.Dijksterhuis, A. (2004). Think different: The merits of

unconscious thought in preference development

and decision making. Journal of Personality andSocial Psychology,87, 586598.

Dijksterhuis, A., Bos, M. W., Nordgren, L. F., &van Baaren, R. B. (2006). On making the rightchoice: The deliberation-without-attention effect.Science,311, 1005 1007.

Dijksterhuis, A., & Nordgren, L. F. (2006). A theory ofunconscious thought. Perspectives on PsychologicalScience,1, 95109.

Dijksterhuis, A., & van Olden, Z. (2006). On the benefitsof thinking unconsciously: Unconscious thought canincrease post-choice satisfaction. Journal of

Experimental Social Psychology,42, 627631.Dreben, E. K., Fiske, S. T., & Hastie, R. (1979). The

independence of evaluative and item information:Impression and recall order effects in behavior-based impression formation. Journal of Personalityand Social Psychology,37, 1758 1768.

Edwards, W. (1968). Conservatism in human infor-mation processing. In B. Kleinmuntz (Ed.),Formal

representation of human judgment (pp. 1752).New York: Wiley.

Edwards, W., & Fasolo, B. (2001). Decision technol-ogy.Annual Review of Psychology,52, 581606.

Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A.(2007). GPower 3: A flexible statistical power analy-sis program for the social, behavioral, and biomedicalsciences.Behavior Research Methods, 39, 175191.

Gigerenzer, G., & Goldstein, D. G. (1996). Reasoningthe fast and frugal way: Models of bounded ration-ality.Psychological Review,103, 650669.

Gladwell, M. (2005). Blink: The power of thinkingwithout thinking.London: Penguin.

Goodman, N. G. (1931). The ingenious Dr. Franklin:Selected scientific letters of Benjamin Franklin.Philadelphia: University of Pennsylvania Press.

Hastie, R., & Park, B. (1986). The relationship betweenmemory and judgment depends on whether the

judgment task is memory-based or on-line.Psychological Review,93, 258268.

Hogarth, R. M., & Einhorn, H. J. (1992). Order effectsin belief updating: The belief adjustment model.Cognitive Psychology,24, 1 55.

Howell, D. C. (2002). Statistical methods for psychology(5th ed.). Belmont, CA: Wadsworth.

Jha, A. (2006, February 17). Want to make a complicateddecision? Just stop thinking. Guardian Online.Retrieved May 10, 2007, from http://www.guar-dian.co.uk/science/story/0,,1711859,00.html

Keeney, R. L., & Raiffa, H. (1976).Decisions with mul-tiple objectives: Preference and value tradeoffs.

New York: Wiley.Kimmel, H. D. (1957). Three criteria for the use of one-

tailed tests.Psychological Bulletin,54, 351353.Knowlton, B. J., Squire, L. R., & Gluck, M. A. (1994).

Pro

decision making :conscious versus unconscious_2009 (1)

Documents