ROB-Cognitive Repairs.pdf

7/26/2019 ROB-Cognitive Repairs.pdf

1/19

COGNITIVEREPAIRS:HOWORGANIZATIONALPRACTICES CANCOMPENSATE FOR

INDIVIDUAL SHORTCOMINGS

Chip Heath,RichardP. Larrick,and Joshua Klayman

ABSTRACT

The literature in cognitive psychology has described a variety ofshortcomings that

preventindividuals from learning effectively. W e reviewthis literature and provide

examples ofa numberoforganizational practices that may effectively repair the cog-

nitive shortcomings ofindividuals. We call these practices cognitive repairs. We

then discuss six tradeoffs that affect the success of cognitive repairs. W e close b y

considering h ow a cognitive perspective might benefit those who study organiza-

tional learning and those who manage it .

Research in Organizational Behavior, Volume20 , pages 1-37.

Copyright 1998 hy JAI PressInc.

Allrights ofreproduction inany form reserved.

ISBN: 0-7623-0366-2

1


2/19

CHIP HEATH, RICHARDP. LARRICK, and JOSHUA KLAYMAN CognitiveRepairs 3

INTRODUCTION

In a famous speech, Hamletdeclares, Whata pieceofworkis man.How noble in

reason,how infinite in faculties (Hamlet, II,2). Anobserver who summarizedthe

psychology ofthe late twentieth century would probably choose very different

phrases to describe the human conditionperhaps, What fools these mortals be

(Midsummer NightsDream, III, 2).

Arepeopleinfinite in faculties and noble in reason? HerbertSimon won a

Nobel prize for arguing thatsocial science mustunderstandthe waysthat human

faculties are limited. Instead of being infinite in faculties, Simons humans could

be only boundedlyrational because their cognitive abilitiestheir ability to

perceive, remember,and process informationwere restricted. Well, then, ifpeo-

ple are not infinite in faculties, a.re they noble in reason?Cognitive psycholo-

gists have spent30 years examining the actual processes that people use when

theycollect information, combine i t, and drawinferences about their world (Nis-

bett & Ross, 1980; Kahneman, Slovic, &Tversky, 1982; Holland,Holyoak, Nis-

bett, & Thagard, 1986). Instead of depicting people as noble (or magnificent) in

reason, this researchhas argued that people reason in ways thatproduce system-

atic errors. A pessimistic modern Hamlet might combine the observations of these

two research streams and describe humans as equipped with primitive hardwnre

andbuggy software.

However, outsiders have not always accepted the pessimistic description of

human faculties and reason that is found in the research literature. As one skepticputit, If weare so stupid,how did we get to the moon? (Nisbett & Ross,1980).

How should we resolve the apparent discrepancy between the pessimistic liter-

atureon human shortcomings and the optimistic evidence of human accomplish-

ment? One way is todismiss the laboratory research. Someresearchers have

argued that the shortcomings that have been documentedin the lab are so minor

that they do not constitute mistakes of any real consequence (Funder, 1987;

Cohen, 1981).Others haveargued that individuals are less likely to make errors in

natural environments than in contrived laboratory experiments (Anderson, 1991;

Gigerenzer, 1996; Cheng& Holyoak, 1985; Hilton, 1995).We propose another way to resolve the discrepancy. Unlike some researchers,

we do not dismissthe examples of limitations, errors, and biases reported in the lit-

erature; we assume that individuals are limited, their decision processes are

biased, and that they often make costly mistakes on important decisions. We

resolve the apparent discrepancy between evidence of individualshortcomings

and the empirical fact ofmoonwalks by observing that individualsdid not make it

to themoon, NASA did.

Organizationslike NASA may have discovered ways to avoid~or repair theindi-vidual shortcomings that have been documented b y cognitive researchers. Orga-

nizations may develop such repairs through deliberate analysis, learn them

through trial and error, or discover them through serendipitous accident. In some

cases, repairsmay derive from formal academicdisciplines likeeconomics or sta-

tistics (e.g., Nisbett, 1992;Nisbett, Krantz, Jepson, &Kunda, 1983; Larrick, Mor-

gan, & Nisbett, 1990), but in most cases they will not: They will be ad hoc,

intuitive rules that emerge from day-to-day practice. Ourthesis, then, isthat indi-

viduals indeed face cognitive limitations and shortcomings, but that organizations

can provide individuals with normsand procedures that mitigate their limitations

and reduce their shortcomings.

In this paper we describe a variety of potentiallyserious shortcomings that have

been documented in researchon humanjudgment and reasoning. We focus in par-ticular on learning and hypothesis testing, that is, how people use information to

develop and revise their mental model ofthe world. Foreach cognitiveshortcom-

ing we discuss, we provide examples of organizational practices thatmay repair

this shortcoming. We call these practices cognitive repairsto emphasize the fact

that they correct some cognitive process that was initially flawed and in need of

repair.

Weidentify potential cognitive repairs to spur researchers to consider how such

repairs might look and function.Although at this point, we canmake only an anec-

dotal case forinterpreting certain practicesa s repairs, wehope that,by pointing

out some plausible examples of such repairs, we will prompt researchersin both

psychology and organizationsto consider more systematically how such repairs

might function.

More generally, the concept oforganizational cognitive repairs illustrates that

researchers may find interesting relationships betweenindividual cognition and

organizational practice. These relationships have not received the attention they

deserve. On the one side, research in cognitive psychology has largely treated

individual learners as rugged individualists who face a difficult environment

alone, equipped only with their own, flawed cognitive strategies. On the otherside, organizational research has largely ignored the literature on individual cog-

nition, focusing insteado nissues of motivation or incentives. By studyingorgani-

zational sources of cognitive repairs, we bring together two frequently disparate

literatures and demonstratehow knowledge at one level ofanalysis can informthe

other.By reviewing individual shortcomings and identifying potential cognitive

repairs, we also hopeto contribute to the academic and popular literature on orga-

nizational learning (Epple, Argote, & Devadas, 1991; Deming, 1982; Senge,

1990; Cohen, 1991; Miner & Mezias, 1996). One important means to facilitate

learning byorganizations is to develop processes that overcome theshortcomings

of individuals within the organization.

Below, we start with abriefexample ofthe kinds of repairs that weconsider in

this paper. Then we introduce a framework that describes different stages in the

learning process, and we use it to review the literature onindividual shortcomings

and to suggest potential repairs. As a preliminary reply to Hamlet,we say that

even ifrugged individualsare unlikely to b e infinite in faculties and noble in rea-


3/19

4 CHIP HEATH, RICHARD P.LARRICK, and JOSHUA KLAYMAN CognitiveRepairs

son, individualswho have access to organizational and cultural repairs maysome-

times appear so.

An ExampleofOur Approach

Consider one studythat might be regardeda s anominous indication of ignoble

reasoningby individual experts. Hynes and Vanmarcke(1976) asked seven inter-

nationally known civil engineers to predict the height of a structure that would

cause a foundation to fail; they alsoasked the engineers to set a50 percent confi-dence interval around their prediction so that their confidence interval was wide

enoughto have a 50 percentchance of enclosing thetrue failure height. Theresults

were quite sobering: not one engineer correctly predicted the true failure height

within his or her confidence interval.

Evidently, the civil engineers thought they knew more than they didifthey

had been aware ofthe limitations of their analysis,they would havesetwider con-fidence intervals and would have predicted the truefailure height morecorrectly.

In the psychological literature this kind of finding has been labeled overconfi-

dence, and it is not anaberration. Similar results have been observed with a num-

ber of individual professionals(e.g., Russo &Schoemaker, 1992).In summarizing

the evidence, Griffin and Tversky (1992) quipped that experts are often wrongbut rarely in doubt.

To illustrate wh y this study paints an ominous picture of individual reasoning,

consider that (unlessyou arereading this paper outside) you are sitting in a build-

ing that was constructed by civil engineers who were substantially less accom-

plishedthanthe internationally known experts in thestudy. Yourcivil engineers

made numerous decisions to ensure the stability and safety of your building; they

decidedhow strong to make its roofsupports and how stable to make its founda-

tton. Ifeven expert engineers are overconfident, shouldyou be concerned about

your safety?

The answer, we believe, is no. Fortunately, the engineering profession has

developeda particular repair, called safety factors, that mitigatethe overconfi-

dent reasoningof individual engineers. In an actual assignment civil engineerswould preciselycalculate the amount andstrength offoundation materialsneces-

sary to hold a structure of a particular height, then they wouldmultiply their pre-

cise answerby a safety factor (i.e., a numberbetween three and eight), and use the

larger figure to build the foundation. Were theconfidence intervals ofthe engi-

neers too narrow?Yes. Werethey too narrow by a factor of three? No.

Safetyfactors are an example of the kind of cognitive repair we consider inthis

paper.An organization (e.g., an engineering firmor the engineering profession at

large) provides its members with a repair that helps combat a systematic and

potentially serious bias in individual judgment. As a result, the organization

shields individualswithin th e organization from acting on their flawed decisions,

5

and it shields individuals inside and outside the organization from suffering the

consequences.

What is a Cognitive Repair?

Organizational repairscanroughly bedivided intotwo classes: (1) motivational

repairs increase the energy and enthusiasm with which individuals pursue a task

and (2) cognitive repairs improve the mental procedures individuals use todecidewhich task to pursue and how to pursueit . Organizational research onmotivation

and incentives can be regarded as the study of motivational repairs (Milgrom &

Roberts, 1992; Eisenhardt, 1989). Organizations may need to repair motivational

problems in order to encourage individuals to learn(e.g., see Heath, Knez, &

Camerer, 1993). For example, individuals may not be willing to experimentwith

newtasks because they have become endowed with the benefits associated with

the old task.

Although previous work hasrecognized the importance ofmotivational repairs,

ithas neglected cognitive repairs. Even when individuals have the right incentives

and resources, they may notlearnfrom their experience if they use the wrong

mental process togenerate hypotheses, collect information, anddrawconclusions.The civil engineers who misestimated the stability of the clay embankment.were

adequately motivatedto getthe right answer. However, they did not on their.own

invoke the kind of correctives (e.g., safety factors) that might have made their

guesses more appropriately cautious.

REVIEW OFINDIVIDUALLEARNING ANDEXAMPLES OF COGNITIVE REPAIRS

In thissection we organize the literature on learning and decision making around

three different stages of the learning process. Effective learners must (1) generate

hypotheses that explain the causal structureof the world, (2) collect information to

distinguish among their hypotheses, (3) draw conclusions that are appropriate and

~autious.The boundariesbetween these stages are fuzzytheyare interrelated

~nd interconnected (Klayman, 1995). However, we distinguish among them

ecause theyinvolvedifferent psychological processes.

Our strategy throughout the review is to consider first the individual then the

ganization. Foreach stageof learning, we describe how an ideal individual

irner might reason, and review psychological research showing how real mdi-

luals depart from this ideal. Then, we describe potential cognitive repairs b y

ich organizations might correct the individual shortcoming in question.


4/19

6 CHIPHEATH, RICHARDP . LARRICK, and JOSHUA KLAYMAN Cognitive Repairs 7

Generating Hypotheses

In thefirst stage of the learning process individuals must generate hypotheses

about the relationships amongevents. Subject to constraints of time and informa-

tion, individuals should generate hypotheses that are deep (i.e., by considering

causes that are more general or systemic) and broad (i.e., by considering a larger

number of potential causes). However,a greatdeal of psychological research sug-

gests that individuals develop hypotheses that are shallow and narrow.

Individuals Generate Hypotheses that areShallowRather than Deep

Individuals SearchforExplanations that Make Themselves Look Good

Individuals often conduct shallow searches when they try to explain success or

failure because they search in a self-serving way(i.e., in a w ay thatallows them

to feel good about themselves). I n ameta-analysis of9 1 tests of this self-serving

bias, Mullen and Riordan (1988) show that individuals typically conclude that

their successes resulted from stable, internal factors (e.g., ability), but that their

Jailures resulted from unstable, environmentalfactors (e.g., the difficulty of the

environment, insufficient effort, or bad luck) (see also Fiske & Taylor, 1991,

pp. 78-82).How might organizations repair self-serving biases? Some repairs may be quite

simple: Traderson Wall Street are warned, Dont confuse brains and a bull mar-

ket (Odean, 1996). This compact phrase prompts individual traders to consider

the base rate of success in the market, and it makes it more difficult for them to

indulge in self-serving explanations for their success.

AtFlorida Power andLight employeesdeveloped a new way to fight self-serv-

ing biases afteran incident that prominently featured a Japaneseinspector for the

Deming Prize w ho laterbecame a folk hero within th e company (Walton, 1990,

p. 61). To impress theinspector, FP&L managers tookhim to visit a new facility

that had been constructed faster and more economically than any facility in the

history of the industry. However, the Deming inspector did not simply accept theresults at face valueand congratulate themon their qualityproject management;instead, he asked anumber ofquestions todetermine why they were so successful.

The managers answerswere so inadequate that it soon became clear thattheydid

not understand enough about their success torecreateit inthe future. The inspec-

tor dismissed their success in his Japanese-accented Englishyou were

rucky. Lateron his phrase, complete with accent, became a common repair for

self-serving interpretations ofsuccess.

The Deming inspector deflated a self-serving bias by considering alternativehypotheses for success (e.g., luck rather thanskill or knowledge). Traditionally at

FP&L, managers were not questioned as long as they achievedgoodresults. After

this incident managers were much more likely to be asked to explain their sue-

cesses. Ifthey could not do so, the verdictwould be delivered: you wererucky

(Walton, 1990,p. 61). The strategic use of the accent was designed to remind man-

agers about the earlier incident where luckproduced dramatic results that were

unlikely tobe repeated.

IndividualsFocuson People Rather than Situations

Individuals also generatea shallow set of hypotheses because social settings

tendto highlight people as causes. I n Westernculture individuals typically chooseto explain events in terms of peoples actions and traits rather than situational fac-

tors (Gilbert & Malone, 1995; Ross, 1977; Ross & Nisbett, 1991). In a recent

study observers heard another student give a pro-life or pro-choice speech on

abortion. Afterward, observers assumed speakers held attitudes consistent with

their speeches even though the speeches were derived fromscripts written by the

experimenters and even though the observers themselves told the speakers which

position to speak for (Gilbert & Jones, 1986). Similarly, Deming (1982) describes

a company that used a variety offlammable products in their production process.

After analyzing the data on fires, Deming found that the fires were a stable and

predictable outcome of the production process. However, according to Deming,

the companypresident focusedhis attentionselsewhere. He sent a letter to everyone ofthe 10,500 employeesof the company to plead withthem to set fewer fires

(p. 325).

Peoples actions are frequently more obvious than their situations. Therefore,

when individuals generate hypotheses about why an event occurred, their first

hypothesis is likely to be that some person caused it (e.g., Ross & Nisbett, 1991).

Thistendency to focuson people rather than situations has been documented by-so

manyinvestigators in so many situations that it has been called thefundamental

attributionerror (Ross, 1977; for recent reviews see Ross & Nisbett, 1991; Gilbert

& Malone, 1995).

Organizations mightrepair thefundamental attributionerror by remindingiridi-

viduals toconsider causes other than people, especially thepeople who are likelyto be closest to any problem: front-line workers. Forexample, an old military

adage says, Thereare no such things as bad troops, only bad officers (Cohen &

Gooch, 1990, p. 228). Parallel repairs are found in total quality management

(TQM). Ishikawa says, whenever mistakes occur, two-thirds to four-fifths of

responsibility rests with management (Ishikawa, 1985, p. ix). Such maxims may

partially repair thefundamental attribution errorbecausethey encourage individ-

uals to lookbeyond th e front line. On the other hand, they may simply focus the

error on people at a higherlevel. Thus, abetter repair may be one from Deming,

who tells managers that ofthe problems he has seen, 94% belong to the system

(Deming, 1982, p. 315). Ishikawa and Deming both use vivid statistics to over-

come the fundamental attribution error even though it is unlikely thateither has


5/19

8 CHIPHEATH, RICHARDP . LARRICK,and JOSHUA KLAYMAN CognitiveRepairs 9

conducteda precise empirical analysis.Demings 94% is particularlynotewor-

thy because ofits apparent precision.

Individuals Stop Searching asSoon as TheyGenerate One Hypothesis

Self-serving biases andthe fundamentalattribution error are special cases ofa

much broader tendency:Individuals tend to stop searching for a cause as soon a sthey locate a plausible candidate hypothesis (Gregory, Cialdini, & Carpenter,

1982;Hoch, 1984).To counter this general tendency, organizations have developed some repairs

that are widely applicable across a number of domains.In one technique knowna s

the FiveWhys, workersa t Toyota learned to ask why?five times beforethey

stopped generating hypotheses. Whenthey did so, they were more likely to finda

rootcause rather than asuperficialone. Imai (1986) illustrates thetechnique with

the following example:

Question 1:

Answer I:

Question 2:

Answer2:

Question 3:Answer3 :

Question 4 :

Answer4 :

Question5:

AnswerS:

Why did the machine stop?

Because the fuseblew due to a n overload.

Whyw as therean overload?Because t he bearinglubrication w as inadequate.

Whyw as th e lubrication inadequate?Because the lubrication pumpw as notfunctioningright.W hy wasntthe lubricating pumpworking r igh t ?

Because t he pump axle woreout.Whyw as it worn out?Because s ludge gotin .

Imai argues that by asking why five times,workers identified the real cause

and therefore the real solution: attaching a strainer to the lubricating pump. If

workers had not gone through such repetitive questions, they might have settled

with an intermediate countermeasure, such as replacing the fuse (Imai, 1986, p.

50). Another illustration ofthe Five Whys deals directly with the fundamental

attributionerror: Problem: He doesnt manage well. (1)Why?Hes not on thefloor.(2) Why? Hes inthe equipment room. (3) Why?The newest equipment isnt

working. (4) Why? Purchasing gave the supplier a short lead time.(5) Why? Poor

planning system(Forum, 1992, p.54). In general, when individuals ask why

the first time, they are likely to develop answers that invokesome salient, recent,

or proximal event (e.g., some persons actions). Subsequent whys are likely to

cause individuals to think more broadly and situationally.

Althoughthe Five Whys i s an admirable cognitive repair because ofits power

and simplicity, individuals may find it difficult to execute b y themselves. Whenindividuals have one good hypothesis in mind, that hypothesis often blocks their

ability to see alternatives (Gregory, Cialdini, & Carpenter, 1982; Gnepp & Klay-

man,1992; Mynatt, Doherty, & Dragan, 1993).For example, Hoch (1984) found

that subjects who generated pro reasons for buying a producthad more difficulty

generating conreasonsimmediately afterward.

Ifindividuals find it difficult to generate alternate hypotheses on their own,then

organizations may repairshallow search b y confronting individuals with others

who are expertin asking questions that reveal deep causes. AtMicrosoft, Bill

Gateshas by personal example, encouraged a culture that relieson relentlessques-

tioning. Says oneWindows manager, you go intothe meetings and you come out

justsweating because, if there is any flaw, he will land on it immediately and pick

ittobits (Cusumano &Selby, 1995, p. 25). Employees overuse terms borrowedfromGates,likedrill downa s a euphemism for going into more detail (What

BillGates Really Wants, 1995).

A similar cognitive repair isfoundin the organization that administers the Dem-

ing quality prize. Here, official Deming inspectors examine managers using a

techniquecalled single-case borequestions.They begin with broad exploratory

queries and then relentlessly delve down into weaknesses and omissions intheanswers they receive. Single-case bore questionssometimes identify causes that

arequitedeep. Forexample,Florida Power andLight often had to deal with power

outagesthat occurredwhen a tree fell on apower line and severed i t. To improve

thereliability of its service, FP&L organized a unit to trim all the trees in sites

where damage had occurred, and thus prevent future outages. Managers at FP&L

congratulated themselves for creating a procedure that preventedfuture problems.However, the Deming inspectors were not satisfied withthe procedure since~it pre-

vented problems only in areas that had already experienced a crisis. Theysearched

for a solution at a deeperlevel, and asked managers anumberofquestions about

what might be considered forestry! What kind of trees grow in the region?Dopalms grow faster or slower than oaks? Managers a t FP&L realized they did not

know the answers to these questions, and that they had not searched deeply

enough tosolve their problems. After their experience with single-caseboreques-

tions, FP&L managers consulted with foresters and developed a regular mainte-

nance procedure to trim trees based on their growth ratesand acrossthe entire

region, notjustin areas where trees had previously severed lines Afterparticipat-

ing in sessions of this kind with the Deming inspectors, managers at the firmlearned to ask single-case bore questions in their o wn internal discussions, thus

institutionalizing this cognitive repair (Walton, 1990,pp. 57-63).

Individuals Generate Hypotheses that are Narrow Rather Than Broad

Inan ideal world individual learners would not only generate deeperhypothe-

ses; they would also consider a broad rather thannarrow set of potential hypothe-

ses. However, even when individuals generate alternative hypotheses, their

alternatives often differ only slightly from one another, and all lie within the

same general frame. Forexample, participants in one experimentwere asked to

considerthe serious parking problem facedb y their university, and they were


6/19

10 CHIP HEATH,RICHARDP. LARRICK, and JOSHUA KLAYMAN Cognitive Repairs 11

given time togenerate as many solutions astheycould (Gettys et al., 1987). Com-

bined, participants generated about 30 0 solutions thatresearcherswere later able

to classify into about seven major categories. One category, for example, sug-gested ways to reduce demand for parking (e.g., by increasing parking fees) and

another suggested waysto use parking moreefficiently (e.g., by segregating park-

ing slots according to size). The average participantproposed about 11 solutions

but these 11 solutions represented only about three of the seven possible catego-

ries. The authors asked an independentpanel ofexperts to compile a complete list

ofhigh-quality solutions, and they used this complete list to assess how manysolutionswere missedby each individual.The typical participant missed from 70

to 80 percent of the high-quality solutions.However, when asked, individuals

believed theyhad missed only 25 percent.

Evenexperts failto consider a hrQadrange ofalternative hypotheses.For exam-ple, one group ofresearchers showed professional auto mechanics a fault tree

that listed a number of hypotheses about why a car might not start (e.g., battery,starting system, fuel system, ignition). Some mechanics were presented with a

full tree that contained seven specific hypotheses, others were given a~pruned

tree that omitted some important hypotheses (e.g. the ignition system). The

resultsindicated that when hypotheses were pruned offthetree, mechanics did not

adequately consider them (Fischhoff, Slovic, &Lichtenstein, 1978).

How might organizations repair narrow search by individuals? Individualsmight search more broadly ifthey are cued tothink about a problem from different

perspectives. At Sharp, employees are told to be dragonflies but not flatfish.

Dragonflies have compound eyes and see things from multiple perspectives at

once, butflatfish have largeeyes that onlylookin one direction (Nonaka & Takeu-

chi, 1995).

The dragonfly repair exhorts individuals to consider different perspectives,

but thismay bedifficult for individuals to do by themselves. Organizations might

repairnarrow search more effectivelyby encouraging individuals to- recruitothers

who have different perspectives. A good example of this is provided b y Bridge-

stone Tire, which conducts kokai watches to generatealternative hypotheses for

improving workpractices. During a kokai watch a group ofup to a dozenpeople,from different areas of a factory, gather for a few hours to watch others work. I n

one four-hour watch a dozen people identified 63 potential dangers with a new

machine (Walton, 1990,pp. 200-201).

The kokai watch has a numberof features that ensure that watchersgenerateabroadarray of hypotheses. First, it mandates alargenumber of watchers (up to

12). Second, it selects watchers from a variety of different areasin one kokai

watch that examineddie and material changes, watchers included a plant visitor,

amemberof the human resources staff, a chemist, and a project manager. The

idea was that people could observe a process, even thosewh o were strangers toi t,with fresh eyes, seeing things that closely involved workers might not (Walton,

1990, p. 200). Third, it ensures that watchers do not discard hypothesesprema-

turely. The watchers are instructed to write down anything, Hey, looks like the

guy is walking too much, or Lookslike hesnot handling the knife right (Wal-

ton, 1990,p . 201). Only after watchers generate hypotheses independently arethe

results combined and filtered.

Other organizational procedures also repair narrow individual searchby ensur-ing that individuals generate hypotheses independently. For example, when

Motorola forms cross-functional teams to evaluate new products, they do not

allow employees who have participated in one product team to participate in

another team with a similar product. This prohibition limits the pool of potential

team members in a costly way; evaluation teams involve six to nine people and

spend two to three months to develop a business plan for the new product. How-

ever, by consciously disregarding previous experience, Motorola allows new

teams to develop recommendations independently from previous teams. At the

same time, Motorola avoids losing the knowledge ofprevious veteranstheyserve as a review team that evaluatesthe recommendations ofthenewest team.

2

Other repairs ensure that a broad range ofalternatives are considered simulta-

neously. Some companiesdividea productdevelopment team into competing sub-groups which develop separate project proposals, and only later recombine to

debatethe advantages and disadvantagesof the independent proposals. Again, this

strategy is costly because it is redundant. However, it may have advantages

becausethe built-in independence ensures that different subgroups will approacha problem from differentperspectives (Nonaka & Takeuchi, 1995,p. 14).

Collecting Information

Inthe second stage of the learning process ideal learners collect information to

test and revise their hypotheses. There are two main sources of such information:

the information that individuals already have in their memory and the information

that they collect from theenvironment. Both kinds of information have potential

flaws, but individuals might minimize these flawsif they collected information in

a rigorous way. However, learnersdo not always actas though they are aware of

the potential flaws in their informationthey frequently collect only a small,biased sample.

Individuals Often CollectSmall SamplesofInformation

Individuals often collect only a limited sample of information becausethey are

constrainedby time or attention. I n a classic study, Payne (1976) asked hissub-

jectsto choose one apartment out of a number of alternatives, each of which was

describedon several different dimensions(e.g., rent, cleanliness, landlord quality,

noise level). When subjects chose among onlytwo apartments, they tended to con-

sider all of the information before they decided. However, individuals searcheda

smaller and smaller percentage of information as more information becameavail-

12 CHIP HEATH RICHARD P LARRICK and JOSHUA KLAYMAN C i i R i


7/19

12 CHIPHEATH, RICHARD P . LARRICK, andJOSHUAKLAYMAN Cognitive Repairs13

able. For example, one subject, who was deciding among 12 apartmentscharac-

terized on eight different dimensions, looked at only about 25 percent of the

information before makinga finalchoice.

It would be reasonable forindividual learners to collect only a small sample of

information ifthey performed a cost/benefitanalysisand decided thatcollecting a

large sample was too costly. However, there is evidence that individuals collect

onlya smallsample of information becausethey systematically underestimate the

benefits of larger samples. Tversky and Kahneman (1971) arguethat individuals

typically believe that small samples will be quite similar to the population fromwhichthey are drawn. They labeled thisbeliefthe law ofsmall numbers to high-

lightthat it contradictsthe statistical law of large numbers, which argues that

samples can yield an accurate picture of a population when they are sufficiently

large. When individuals believe in the law of small numbers,they assumethatany

sample willbe sufficient, no matterhow small.

At the extreme, individuals may not collect any information from the external

environment because they believe that they already have adequate information

stored in theirhead. Organizations may overcome thistendency by encouraging or

requiring individuals to collect larger samples. This kind of repair is pervasivein

writingson TQM. In promoting statistical quality control, we have used the slo-

gan, Let us talkwith data (Ishikawa, 1985,p . 200). At manyTQM companiesone of the main principles of the quality effort is Management by Fact(Walton,

1990, p. 37).

And TQM not only talks aboutdata, it provides individuals with tools that help

them collect andanalyzedata. For example, six of the SevenTools ofTQM pro-

vide ways tocollect data (e.g., checksheets) or to simplify and display large quan-

tities of data (e.g., histograms, scatter plots, Pareto diagrams, control charts)

(Deming, 1982;Imai, 1986;Ishikawa, 1982, 1985;Juran, 1992).

IndividualsCollect Biased SamplesofInformation

Individuallearners not only collect small samples ofinformation, they also tendto collect samples that are biased (i.e., that are unrepresentative of the larger

world). Consider the common claims that the other linealways moves faster or

it only rains after I wash mycar. Unless we want to believe that a malevolent

spirit is in charge of such harassment, these examples demonstrate that ourmem-

oriesdo not store a random sample ofall waiting times orall rainstormswe are

more likely to remember the rainstorms thatspoil the finishon-ourfreshly washed

car.Even whenindividualscollect information from theoutside work1(rath~vtha~i

from memory), theydo not always attend to the most relevant and importantinfor-

mation.Below,we discuss a number offactorsthat might lead individual learners

to collect biased samples.

Individuals OnlyConsider Available Information

As indicated b y the car wash example, individuals often collect biased samples

because they collect information that is easily available in memory, for example,

because it isespeciallyvivid or recent. The problem i s that individuals typically

assumethat the information that is available is also most frequent, probable, and

causally important (Tversky & Kahneman, 1973). This assumption is often

wrong. Individuals dramaticallyoverestimate the likelihood of vivid causes of

death like accidents or homicides, and they underestimate the likelihoodof less

vivid causeslike disease or strokes. Individuals estimate that accidents caused a s

many deaths as diseases and that homicides were as common as strokes. In fact,

diseasestake 16 times more lives than accidents and strokes take 11 times morelives than homicides. Individuals also overweight recent information. They

assume that the most recent floodprovides an upper boundon possibleflood dam-

age,and the purchase of earthquake insurance increases sharply after a quake and

then decreases steadily a s memories fade (Slovic, Fischhoff, & Lichtenstein,

1982,p . 465).

Many organizationsrepair individuals tendency to rely on biased, available

informationby institutinga process that collects information more systematically.

At a Motorola division that develops equipment for cellularphone systems, one

group realized thatan availability bias was causing it to overlook certain custom-ers when it evaluated new products. The unit assigned account managers only to

large accounts, so when managers evaluated newproducts, theyprimarilyconsid-

ered the needs and requirements of only large customers. However, the unit also

served anumber of smaller customers that did not have their o wn account man-

ager. Together, these small customers accountedfor a large percentage of reve-

nues. Motorola overcame the availability bias by developing a Feature

Prioritization Process; they surveyed customers up to four times a year and then

weightedall ofthe inputsbasedon customervolume andpriority.3

Hospitals also have a variety of procedures to force individuals tocollect infor-

mation more systematically. Trauma physicians are often confronted by vivid but

potentially misleading information. One doctor states that, contrary to what one

might expect, stabbings and bullet wounds are relatively straightforward affairs

because they leave clear tracks on the body. Other injuries are more difficult to

treat becausethey leave no visiblecues. It would be all too human to focus on a

lacerated scalpagory but basically insignificant injuryand miss a fractured

thighbone that hadinvisibly severed a major artery (Rosenthal, 1994,p. 48). The

medical profession has developed a series of strict protocolsfor trauma situations

that allow doctors to quickly collect all the relevant information, notjust that

which is salient. For example, when a patient first entersthe emergency room,

physiciansfollow the ABCs; they establish airway, then breathing, thencircu-

lation.4

For situations that are more critical, such ascardiac emergencies, proto-

cols are even more rigorous and specific.

14 CHIP HEATH RICHARD P LARRICK and JOSHUA KLAYMAN 15


8/19

14 CHIPHEATH, RICHARDP. LARRICK, and JOSHUA KLAYMAN 15Cognitive Repairs

Ifindividualstend to focuson information that is highly available, itis not ter-ribly surprising that they are frequently unaware of missing information. Even

when informationis present, learnersdo not pay as much attention to what doesnt

happen as what does (Agostinelli, Sherman, Fazio, & Hearst, 1986; Newman,

Wolff, & Hearst, 1980).

Certain professions and organizations have learned to repair the tendency to

ignore missing information. Homicide detectives learn to notice the absence of

itemsa t murder scenes, since many murdererstakebacksomething that belongs to

them after committing thecrime.Youlook at whats been taken and you findout

who itbelonged to originally (Fletcher, 1990, p . 75).

A particularlyimportant form of missing information is the absence of experi-

ence with highly unusual events. Bank examiners rarely see a bank fail, nuclear

technicians rarelysee a meltdown, airline personnel rarelywitness a crash (March,Sproull, & Tamuz, 1991; Perrow, 1984). Certain organizations institutionalize

procedures that encourage individualsto pay attention to such information:desjtito

the factthat such events are unlikely to be available in their own experience. For

example, at the Federal Reserve Bank, whichcertifies the security of banks, senior

bank examiners deliberately recount stories offailed banks to keepjunior exam-

iners aware that they should be vigilant.5

At one banks commercial lending

department, senior creditofficers would hold seminars and informal brown-bag

lunches to discuss past lending mistakes, particularly in areas characterized byunusual or rare events (e.g., problems with highly leveraged companies, real

6estate, environmental liability on contaminated property). By forcing individu-als to rehearse such information, organizations help individuals learn from vicar-

ious experiences that are rare but highly informative. Furthermore, organizations

remind individuals of potentially painful information that self-serving biases

would make them prefer to ignore.

IndividualsCollectBiased Information Basedon TheirPreexistingTheories

Research suggests that individuals tend to think of facts, experiences, and

arguments thatsupport a current hypothesis more readilythan those that refuteit(Klayman, 1995; see also, Baron, 1988; Kunda, 1990; Nisbett & Ross, 1980).Thus, when individuals collect information from memory, they may focus on

information that supports their preexisting theories. Individuals may also do this

whenthey collect information from the externalenvironment. For example, when

individuals collect information from others, they often askspecific, directive

questions that are likely to elicit the answer they expect (Hodgins & Zuckerman,1993; Zuckerman, Knee,Hodgins, & Miyake, 1995).

The ChicagoBoard of Trade has astaffof in-houseinvestigators who scrutinize

tradesthat may violate exchange rules. In these investigations, which are obvi-

ously quite sensitive,it is veryimportant thatinvestigatorsdo not collectinforma-tionthat isbiasedby their initial theories. To repair thistendency, the investigators

are trained to avoid questions thatcan be answered with a yes or no response.

This forces an investigator to askopen-ended questions and allows her to draw

out as much information about the situation as possible. By asking open-ended

questions, investigators avoid the possibility of directing the interview in a way

that elicits only information that is consistent with their preexisting theories.7

Some organizationshave developed maxims that seem designed to encourage

individuals tocollect datarather than relyingon their (potentially biased) theories.

At Bridgestone Tire employees use two Japaneseterms: genbutsu (actual product)

and genba (actual place) (Walton, 1990,p. 194). These terms remind employeesnot to rely on their o wn theories, but to actually go out andinvestigate the actual

product in the actual place where the problems arose. Another group (Forum,

1992)uses a similar cognitive repair theycall thethreeactualrule: (I) Go to the

actual place; (2)See the actual problem; (3) Talkto the actual people involved.

Individuals ConsiderOnly Partofthe RelevantInformation

Finally, individual learners may collect biased samples because theytend to col-

lect information from only one small corner of the universe of information. This

arises from basic cognitive processes. Memory is associativewhenindividuals

retrieve onepiece of information, they tend to thinkof other information that is

linked to it by strong associations, common features, or similar meaning. Even

when individuals collect information from the external environment, they are

likely to collect information based on the same kind of associative process.

Research in cognitivepsychology hasshown that individuals attend to and process

information more comprehensively when they have a mental schema that tells

them what information is needed in a given situation and where to find it(Ander-

son, 1995).

Accordingly, organizations can repair biased information collectionby provid-

ing individuals with aschema that reminds them of the full range ofrelevantinfor-

mation.Many schemas of this kindcanbe found in the financial services industry,

where individuals must assess a wide variety ofinformation todetermine whether

to buy,sell, or lend. At the Federal ReserveBank of NewYork, the BankExami-nations groupprotects the FDIC insurance fund by ensuring that individual banks

are in soundfinancial condition.When reviewing eachbank, examiners use a rat-

ing system known a s CAMEL:they review Capitaladequacy,Assetquality, Man-

agement, Earnings, and Liquidity.8

In another banks commercial loan

department,creditanalysts use the Five Cs of Credit:Collateral, Capacity,Cap-

ital, Conditions, and Character.9

Organizational schemaslike CAMEL and the Five Cs are likely to encourage

individuals to collect a broaderrange ofinformation than they would normally

collect. It would be very easyfor individuallearnersto collect informationon only

on the most salient factors (such as cash flow in a loan decision). Although cash

flowis certainly important, it can also be misleading or unreliable, particularly in

16 CHIP HEATH RICHARD P LARRICK and JOSHUA KLAYMAN C i i R i


9/19

16 CHIP HEATH, RICHARDP. LARRICK, andJOSHUA KLAYMAN Cognitive Repairs 17

an environment where conditions are changing. By emphasizing the Five Cs, a

bankcan repair th e tendency of individual analysts to neglect information about

important variablesthat are less obvious orare harder to assess. For example, the

Five Cs reminds loan officers to consider characterWhat are the management

skills ofth e owners? Do they have good personal credit records? Although the

answersto such questions are quite important, individualanalysts might forget to

ask them in a numbers-oriented environment like a bank, without a cognitive

repairlike theFive Cs.

Individuals Who CollectBiasedInformation Fail to CorrectforBias

We have discussed a number of factors that might lead individual learners to

collect biased information. However, even if learners collect biased information,

they mightstillbe ableto draw effective conclusion as long as theyrecognized the

bias and corrected forit. Forexample, suppose an individual made the statement,

the other linealwaysmoves faster, but then reminded herself that such situations

might be overly available in her memory. This kind of correction improves the

conclusions drawn from even a biased sample. On the other hand, even ifindivid-

uals are aware that they have collected biased information, they may not know

how tocorrect for biases after the fact. For example, after individuals ask biased

questions and therefore receive biased answers,theydo not take into account ho w

much theanswers were biased b y their initial questions (Zuckerman, Knee, Hod-

gins, & Miyake, 1995).

Because individuals do not always correct their information for biases,some

organizations attempt to ensure that individuals collect unbiased samplesfrom-the

start. Microsoft requires its software developers to use the same programsand

machinesthat areused b y their customers. For example, programmers who weredevelopingth e new Windows NT operating system ran the current days version

of theprogram as they programmed the nextdays version. At Microsoft this pro-

cess isknown as eating yourown dog food. It ensures that developers collect a

large, unbiased sample of information about the current state of the program. If

Windows NT crashed while a developer was designinga new printer driver, he

had tofixthe problem with NT before he could return to his driver (Cusumano &

Selby, 1995,p. 331). Microsoft also requires developers to use the same machine

used by customers, a requirement that has been controversial at times because

developers like to have the fastest, coolest machines on their desks. However,

when developers have better technology than the average customer they collect

biased information about how well their software programs perform. One man-

ager said, every time Ive had a project where the developers had hardware thatwas a generation beyond what customers had, the [software] always had perfor-

mance problems (Cusumano & Selby, 1995,p. 347). Byrequiring developersto

use the same machines as their customers, Microsoft forces them to collect an

unbiased sample of information about the operating speed and memory demands

of the software they a re developing.

DrawingConclusions

After generating hypotheses and collecting information, ideal learners should

evaluate the information they have collected and draw conclusions that are appro-

priate andcautious. Researchers have suggested three main classes ofproblems

that real individuals face when they interpret evidence. First, they often weighinformation in a way that is not statistically appropriatefor example, they

emphasizethe importance of extreme evidence butthey do not emphasize the rel-

ative amount of extreme versus non-extreme evidence. A second, even more

insidious problem is that individuals use their initial theories to interpret the evi-

dence. While individuals may readily accept information that is consistent withtheir initial hypothesis, they cast a critical eye on information thatcontradicts it .

Third, as a resultofthetwo previous processes and others, individuals frequently

draw conclusions that are overconfident and overly optimistic.

Individuals Weigh Vividand Extreme EvidenceMoreHeavily

Once individuals have collected information, h ow shouldthey combine it andweigh it? An ideal learner would weigh information basedon the quality of the

information.However, actual learnersdo not alwaysassign appropriateweights to

all aspects ofthe decision. For example, they tend to weigh morevivid, easily

imagined information more heavily(Keller & McGill, 1994). They also focus on

the extremity or strength of the available information (e.g., the warmth of a rec-

ommendation letter) without adequately attending to the amount or weight ofthe

evidence (e.g., thewriters amount ofcontact with the recommendee) (Griffin &

Tversky, 1992).

Ifindividuals tend to overemphasize vivid or extreme information, organiza-

tions might prevent this by requiring individuals toconsciously classifyinforma-

tion according to its appropriate weight. Many companies have internal auditgroups that examine the recordsof company divisions to ensure that they are using

proper accounting procedures and spending money on legitimate expenses. An

audit usually uncovers a variety of major and minor exceptions(i.e., situations

where correctprocedureswere not followed). One auditor says that auditors must

be careful not to place too much emphasis on memorableerrors, e.g., an error in

the presidents expense reportor th e misuse ofthe company car. One auditing

group repaired this temptation by first classifying each exception a s major orin

minor then consciously ignoring theminor issues.Consistent with the tendency to overweight the extremity of information and

ignore the amount, individuals frequently place higher weight on one vivid case

than on a much larger sample of information. Joseph Stalin is reported to have

18 CHIP HEATH RICHARD P LARRICK and JOSHUA KLAYMAN Cognitive Repairs 19


10/19

18 CHIPHEATH, RICHARD P. LARRICK,and JOSHUA KLAYMAN CognitiveRepairs 19

said, The death of a single Russian soldier isa tragedy. A million deaths i s a sta-

tistic (Nisbett& Ross,1980,p. 43). Ina study that supports this observation, Bor-gida and Nisbett (1977) showed some students a statistical summary of how

dozens ofstudents had rated variouscourses in the previous term. Other students

attended a panel discussion, during whichtwo or three upper-division students

rated eachcourse on a numerical scale and provided somegeneric, uninformative

comments. Despite the fact that the statistical summary provided students with a

larger amount of information, individuals who heard the small sample of vivid

information were more likely to change the courses they selected.

Microsoft also discovered that individualsdiscountlarge samples of statistical

information.At one point, Microsoft started surveying users to see how many of

them founditeasy tousea particular feature. Software developers often refused to

believe the statistics. The usability groupwould tell the development group Six

out often couldnt dothis. And thedevelopersreaction wouldbe, Whered you

findsix dumbpeople? (Cusumano &Selby, 1995, p . 379).In orderto repairthistendency to ignore base rate information, Microsoft made the information more

vivid. I t built a usability test lab where developers can watch real usersstruggle

with new products from behind a one-way mirror. Instead of presenting develop-

ers with pallid statistics, the test lab presents them with real people (albeit amxtch

smaller sample). The lab managersays that when developers see auser, twenty

ideasjustimmediately come tomind. First of all,you immediately empathize withthe person. Theusual nonsense answer Well, they canjust look in the manual if

they dont know how to useit, orMy idea is brilliant; you just found ten stupid

people...that kind of stuffjustgoes out thedoor... (Cusumano &Selby, 1995,p .

379). This cognitive repair is interesting because it uses onekind of bias (over-

weighting of extreme, or vivid information) to fight another (underweighting of

statisticalinformation).

IndividualsUse TheirPreexistingTheories toInterpretthe Evidence

Individuals not only weigh information inappropriately,they also have diffi-

culty interpreting information independently of their preexisting theories. Instead

of using the information to test their theories, they use their theories to test theirinformation. This often leads them todiscount information that disagrees with

their preexistingbeliefs.

In a classic demonstration of such discounting, Lord, Ross, and Lepper (1979)

selected undergraduates who strongly supported or opposed capital punishment

and presented them with two purported academic studies that evaluated capital

punishments effectiveness using very different methods. A study using one

method found that capital punishment was effective and a study using the other

method found it w as ineffective (the researchers counterbalancedwhich method

was associated with which result). Participants applauded the positive aspects ofwhichevermethod supported their own preexistingtheory, andthey critiqued th e

design flaws in the other. I n fact, after receiving mixed results from th e two

studies, subjects becamemoreconvincedofthe validity of their original position.Seemingly, they regarded the evidence as one good study that supports my

beliefs, and one lousy study that draws the wrong conclusions. Individual sub-

jects thus failed to evaluate the incoming information separately from theirpreex-isting theories. Unfortunately, similar results have been noted withprofessional

scientists (Mahoney, 1976;Koehler, 1993).

One bank helped its loan officers repair the way they interpret evidence b y

encouraging them toconsider anonstandard theory of lending.In mortgage lend-

ing, loan officers oftenlookforreasons to deny loansbecause loans are difficult to

make (they are subject to a mountain of regulations) and potentially quite costly

(e.g., foreclosure on a bad loan may cost up to 20% of the property value). Thus,

the initial hypothesis in manyloan decisions isthat an applicant should be denied

a loanunless proven otherwise. One mortgage loan department grewa t an annual

rate of30 percent by forcingloanofficers toconsideran alternative tothe standard

hypothesis. Instead of asking whether an applicant should bedenieda mortgage

loan, it asked whether the applicantshould be approved. This reversal led the

department to develop special programs for qualified applicants who had low

incomes or other special circumstances.11

Individuals use their theoriesto develop expectations about what is normal, and

they frequently label unexpected events as problems or failures. These labelsmay be misleading, however, particularly in research and development where

unexpected events may point the way to important breakthroughs. One research

organization has developed a repair that discourages individuals from thinking

that unexpected events are failures(Sapolsky, 1997).JacksonLaboratoriesbreeds

mice that exhibit physiologicalor behavioral traits that are of interest to medical

researchers. For example, it sells mice that lack growth hormone to researchers

who are interested in understanding the biology of mammalian growth. It found

that the animal technicians (e.g., the people w ho cleanedthe cages) often noticed

unusual behavior that was scientifically important. The mice that lacked growthhormone were discovered by a technician who noticed a particular mouse that

didnt grow at a normal rate. Another technician noticed a mouse that didnt

respond normally to the loud noisesthat occurredwhen the cages were cleaned--

its offspring were found to be susceptible tohereditary deafness.After several

experienceslike this where unexpectedbehavior produced important discoveries,

the company started holding regular meetings with a nimal techniciansto inquire

whether they have spotted anything unusual. These forums for highlighting the

importance of unexpected eventsare called deviant searches.

CRSS,an architectural firm,developed a special position to repair the problemof theory-based interpretation ofevidence. Most designers loveto draw, to make

thumbnail sketches, says one manager, but this rush todraw conclusions is

often premature. CRSS created a uniquejob description,the programmer, to

ensure that some members of its design teams were not allowing their o wn theo-

20 CHIPHEATH, RICHARD P . LARRICK, andJOSHUA KLAYMAN Cognitive Repairs 21


11/19

g p 21

ries to dominate the way they evaluated information from clients. Programmers

arenot in charge ofdesigning or problem solving, they are in chargeofproblem

seeking. They are trained to use techniques that help them to resist premature

conclusions,and thus listen more carefully to clients. The experienced, creative

I programmerj withholdsjudgment, resists pre-conceived solutions and the pres-

sure to synthesize...he refuses to make sketches until he knows the clients prob-

lem (Peters, 1992, p . 402).

Often, organizations ensure that individuals weigh information effectively b y

forcingthem to interact with others who might weigh the information differently.

Oneresearcher has explored whether training as a scientistcures the problems thatother individuals have in evaluatingevidence(Dunbar, 1995). The answer is no.

For example, scientists, especially young ones, often believethat a single experi-

mental result hasjust resolved an important problem. However, when Dunbar

studied aset ofmicrobiology labs thathad been particularly successful, he found

that they placed more emphasis on group lab meetings. At these meetings an indi-

vidual scientist presented his or her results to a variety ofskeptical, uninvolved

peers. When the individual scientist presented a striking new piece of evidence

(e.g., I have detected Enzyme Z in a biological process where ithas never been

observed before), the individuals peers were typically quite willing to propose

alternate ways of interpreting the evidence (e.g., the sample w as contaminated

with residual Enzyme Z from a prior procedure). In successful labs, even when

individual scientists failed to weigh a particular piece of evidence appropriately,their peers did so for them. Moreover, the most successful labs were those that

includedmembers with different training and backgrounds. Such lab meetings

are not limited to successfulmolecular biology labs; similar meetings takeplaceat

venture capital firms wherefirms decide whether to allocate money to new ven-

tures (Kaplan, 1995).

IndividualsDraw Conclusions thatare Overconfidentand Overly Optimistic

Imagine that individuals have generateda set of hypotheses, collected some

new information, and interpreted the relevance ofthe information for the initial

hypotheses. How much confidence should they place in the conclusions theyhavedrawn? Ifindividual learners were adequately cautious, their conclusions would

reflect the degree of uncertaintyin the data on which they are based. Over the

years, research h as documented that individuals often express more certainty in

their conclusions than is warranted by the facts available to them (or by their

actual performance). This kind of problem has been documentedextensively in

laboratorystudies, but also in field studies of individualjudgment in a variety of

professions, likethe civil engineers in the introduction (Griffin &Tversky, 1992;

Lichtenstein, Fischhoff, &Phillips, 1982; Russo & Schoemaker, 1992).

Individuals often exhibit aparticular kind of overconfidencethatwe might label

a planningfallacy (Buehler, Griffin,& Ross, 1994)or an optimism bias. This opti-

mism bias is pervasive in work environments. Software developers at Microsoft

often experience burnout because they grossly underestimateh ow long it will

takethem to accomplish certaintasks (Cusumano & Selby, 1995, p. 94). Organi-

zationsdo not always successfully overcome this individual bias. A study of pio-

neerprocess plants revealed that the typicalplant experienced actual constructioncosts that were almost double the original projections; similarly, a study ofstart-

ups showed that more than 80 percent fell short of their projected market share

(Davis, 1985). These examples suggest that individuals drawconclusions that

underestimate the amount of uncertainty and error in their predictions, but they

tend to do itasymmetricallytheyrarely overestimate a projects cost or time tocompletion.

Ofcourse, individuals may display anoptimism bias because theyconfront mis-

aligned incentives. Perhaps if engineerscorrectlyestimated the true cost of a new

energy plant, decision makers might choose not to build it. However, the realcauses of the optimism bias seem to be cognitive, since individuals are overconfi-

dent by the same magnitude even in lab experimentsthat reward accuracy. Forexample,individuals typically assumethattheir predictions are more precise than

theyare.Whentheyare asked to setconfidence intervals around aquantity, so that

their confidenceinterval has a 98percentchanceofincluding the true answer,they

are typically surprised by the true answer not 2 percent of the time, but 20 to 50

percent (Lichtenstein, Fischhoff, & Phillips, 1982; Russo & Schoemaker, 1992).How might organizations repair individual tendencies toward optimism bias

and overconfidence? One strategy is to allow individuals to make overconfident

predictions, thenadjust them overtly. This was the strategy pursuedby the engi-

neering profession withits safety factors. Microsoft uses a similar strategy to cor-

rect the overly optimistic projections of individual software developers: It has

rules about the amount ofbuffer time that should beadded to projects. For reason-

ably well-understoodprogrammingchallenges, such as applicationsprograms like

Excel andWord, Microsoft typically adds buffer time that constitutes 30 percent

of the schedule.However, for operating systemslikeWindows, where developersmustcreate a system that has to mesh effectively with numerous piecesof hard-

ware and software, Microsoft may add buffer time that reaches 50 percent

(Cusumano & Selby, 1995). Similar repairs have evolved in otherindustries. Atone Big Six accounting firm, where teams must prepare formal plans for a con-

sulting engagement, project leaders develop their best estimates oftime, expense,t2

and contingency costs, then increase the final numberby 15 percent. This repairhas evolved despite the fact that this environment provides some incentives to

underestimatecostsbidsthatare too high may not be accepted.

When Microsoft adds buffer time to a schedule, it corrects the predictions of

overconfident individuals by overriding them. However, it has also developed

procedures that help individual developers decrease their initial level ofovercon-

fidence. For example, the company has improved itsschedules by requiring devel-

opers to create a detailed work plan that specifies which tasks they will perform

22 CHIPHEATH,RICHARD P. LARRICK, andJOSHUAKLAYMAN Cognitive Repairs


12/19

Cognitive Repairs

during specific windows of time. Says one manager, The classic example isyou

aska developerhow long it will take him to do something andhell say amonth,

becausea month equals an infinite amount of time.And you say, Okay,a month

has22 working days init. What are the 22 thingsyoure going to do during those

22 days? And the guy will say, 0h, well, maybe it will taketw o months. Even

by breaking it downinto22 taskshe realizes, Oh, itsa lot harder than I thought

(Cusumano &Selby, 1995, p. 254).

Some organizations repair overconfidence by forcing individuals to interact

with others who are trained to question their conclusions. For example, the Penta-

gon for many years had what they called the murder board, a group of experi-

enced officers that reviewed theplans for important missions, with the goal of

killing the mission. According to Pentagon lore, the failed Iranian hostage rescue

during the Carteryears wasnotvetted by this boardbecause highgovernment offi-

cials were too concerned about security leaks.13

Otherorganizations have developed norms of frank feedback to ensure that

individuals question others conclusions honestly andopenly.In its feature anima-

tion unit, Disney regularly holds Gong Shows where personnel (including

department secretaries) canpitch ideas to a group ofsenior executives. Gong

Shows may attract 40 people who present their idea to the executives and other

presenters for three to five minutes. The senior executives are careful to give

exceptionally frank feedback at the end of the session, highlighting both good and

bad aspects of the presentations. Somebody may have a great concept, but the

story may not be very good. [Wecantsay]Oh, thats fabulous. Great pitchguys!

and when they leave, mumble, That was awful!....We dont pull our punches.

[Eventually] people begin to understand that no matterhow good,bad, or indiffer-

enttheidea, itcan be expressed, accepted, andthought about (McGowan, 1996).

GENERAL DISCUSSION

In this paper we have reviewed the literature on individuallearning using a simple

frameworkthat considers three broad stages of the learning process. W e arguedthat ideal learners would generate a broad and deep set of hypotheses, test them

with a large, unbiasedset ofinformation, and draw conclusions in a cautious and

balancedway. The psychologicalliteratureindicates, however, that real individu-

als are not ideal learners; they think and act in ways that reduce their ability to

learneffectively.

Fortunately, individual learners do not have to go it alone. We have argued that

organizations frequently repair the shortcomingsof individual learners through

the use of sayings, informal routines, and formal procedures. We believe the

examples we have offered illustrate the tremendous promise of organizational

sources of cognitive repairs.

23

Nevertheless, we do not think that cognitive repairs will overcome every indi-

vidual problem. Cognitive repairs are heuristicslike the mental processes they

repair, they are pragmatic and often efficient, but also approximateand inexact.

For example, they may solve 75 percent of individual shortcomingswhileincur-

ring only one-third ofthe costs ofoptimalprocedures (e.g., fromeconomics or sta-

tistics). However, theyare unlikely to be perfect.

Consider the five whys. It undoubtedly prompts individuals to think more

deeply about causes, but it is only a roughheuristic.Why five questions and not

three or seven? And which questions? Problem: He doesnt manage well. (1)Why? He doesnt manage conflict well. (2)Why? He grewup in a dysfunctional

family. (3)Why? His parentswere alcoholics... In this example, the answers took

an unhelpful detour away from potential solutions sometime around answer 2.

Even when repairs are reasonably effective, they maystill leave room forfur-

therrepair. Consider,for example, the militarys partial repair forthe fundamental

attribution error: There are no bad troops,only bad officers. This adage may

repair tendencies to attributeblame to the people whoare closest to a problem (the

troops whoare on the battlefield);however,it merely focuses attentionon another

group of people. Thus, it may prevent individuals from fixing systems or proce-

duresthat havebasic flaws (Cohen &Gooch,1990). A more effective repair might

say, There are no bad people, only bad systems.

Other repairs may be imperfect becausethey fix one problem well, but exacer-

bate others. For example, the Five Cs may help individual loan officers collect

more kinds of information but they may create secondary problems. First, by

emphasizing character,the Five Cs may provokethefundamental attribution error.

Second, although they expand the set of factors loan officers will consider in a

loan decision, they may also institutionalize any tendency that they may have to

ignore other potentially relevant factors. Third, they may help loanofficers collect

information, butthey do not necessarily help them interpret it. They seem to indi-

cate that each C should be weighted equally, whereas an ideal statistical model

wouldweigh someCs more heavily than others.

As these caveats illustrate, cognitive repairs are unlikely to completely repair

the shortcomingsofindividual learners. Thus, when we assess whether agiven

cognitive repair is successful, we must consider the costs and benefits ofthe

repair. Below,we consider six dimensions that may affect the costs and benefitsof

repairs, and therefore their success.

Tradeoffs Associated with Successful Repairs

In order to be successful, a cognitive repair must be effectiveit must mend

some individual shortcoming and improve learning relative to the status quo. To

be truly successful, however,a cognitive repair must also be accepted in theorga-

nization and actively used. A repair that is not implemented is not a repair.

24 CHIPHEATH,RICHARDP . LARRICK,and JOSHUA KLAYMAN CognitiveRepairs 25


13/19

Cognitive repairsare a kind ofinnovation, and a s such, their use will undoubt-

edly be affected by many of the characteristics that have previously been men-

tioned in literatureson diffusion and adoption (Rogers, 1995; Scott, 1995). We

will focus on innovation characteristics thatare particularly relevant for cognitive

repairs. Cognitive shortcomingsnot only create the need for a repair, they also

limit what repairs may succeed.

Below, we consider six dimensions that affect whether a repairwill be success-

ful: simple versuscomplex,domain-specific versus domain-general,familiar ver-

sus novel, corrective versus preventative, social versus individual, and top-downversus bottom-up. (We will typically focus on the endpointsof these dimensions,

but they should be regardedas continuousrather than dichotomous.) Mostdimen-

sions involve tradeoffs.For example, qualities that make a repair more effective in

solving an individual shortcoming sometimes reduce the chances that it will be

acceptedand usedby individuals. I n the absenceofperfectly effective andaccept-

able repairs, we must recognizeand understand the tradeoffs that make repairs

more or less successful.

Simple versus Complex

One obvious dimension along which cognitiverepairs vary is whether they are

relatively simple or complex. Many of the repairs we have discussed in this paper

are strikingly simplethey require an individual to rememberand apply a proce-

durethat is only a fewsteps long(e.g., thefive whys or the physicians ABCs). In

contrast, many of the procedures that are taught as formal repairs in academic

environmentsar e quite complex, and involve many stages ofsorting, arranging,

and calculating(e.g., formal financial or statistical analysis).

Simple repairs have profound advantages over complex repairs. First, they are

more likely to be usedbecause the costs are small; individuals will find iteasierto

learnand implement shorter procedures. By contrast, complex repairs typically

require extensive background knowledge and tax basic cognitive resources like

attention and memory (Bell, Raiffa, & Tversky, 1988; Nisbett& Ross, 1980).

Thus, when individuals encounter a complex repair, they are likely to perceive thecosts of learning it as large and immediate, and the benefits of using it a s small,

uncertain, and delayed.

Second, simple repairs are easier to remember a nd reconstruct than complex

repairs, and this increases the probability that individuals will accurately apply

them and accurately transmit themto others. Because complex repairs require

individuals to remember a number of stages, they a re more likely to be distorted

whentheyare transmitted from individual toindividual.This problem will be par-

ticularly pronounced in situationsthat require learning by observation and imita-

tion (DiMaggio & Powell, 1983). Individuals who learn a repair through

observationmay find it difficultto infer thecomplete rules ofbehavior forcom-

25

plexrepairs becauseinformation aboutthe rules is incomplete, unavailable, or dis-

tributed across time in a way that makes learning difficult.

Althoughsimple repairs have profound advantages overcomplex repairs, they

also have some disadvantages. Fundamentally, thetradeoffbetween simple and

complex repairs is a tradeoffbetween ease of use a nd accuracy. Complex proce-

dures are often complicatedbecause they attempt to be precise. Simple repairs

gain easeofuse by sacrificing accuracy. For example, a simple aphorism such a s

dont confuse brains and a bull market suggests the correct direction to adjust

onesjudgment, but providesno guidance about exactlyhow much one should dis-

credit the success of an individualtrader. To precisely estimate the amount ofcredit due to brains versus the market, an individualwould have to perform a more

complex procedure, such as calculating the overall market performance and mea-

suring an individuals performance relative to the dispersion and central tendency

ofthe market.

Domain-Speczflc versus Domain-General

Cognitiverepairs also vary in the rangeofcontexts to which theycanbe applied

(Nisbett, 1992), with some repairs being relatively more domain-specific and

some being more domain-general. Domain-specific repairs are tailored narrowly

for a specific context (e.g., the Feature PrioritizationProcess at Motorola or theFive Cs of Credit). Domain-general repairs are described so generally, and

abstractlythat they applyacross mostjudgment tasks(e.g., the Five Whys or most

economic or statistical principles).

Domain-specific rules havea t leasttwo advantages overdomain-generalrules.

First, individuals findit easier to recognize that a domain specific rule is relevant

because the situation itself remindsthem to usethe rule (e.g., a creditanalyst whohas learned to think aboutthe Five CsofCredit willfind it difficult to think about

lending decisions withoutconsidering all five categories of information). Second,

individualsmay find it easier to apply domain-specific than domain-general rules.

Consider, forexample a loan officer who istrying to apply a general rulelike cal-

culate the netpresent value (NPV) of makingthe loan.This domain-general rule

applies tomany more financial decisions than the FiveCs; but it contains no hints

about how it should be applied to a loan decision. In contrast, the Five Cs pointout

specific aspects of the loandecision that might affect the loans quality.Similarly,

securities traders might find it hard to benefit from a domain-general warning

against self-serving biases (e.g., pay attention tosituational determinants ofsuc-

cess, and dont over-attribute achievement to personalcharacteristics). In con-

trast,they are unlikely to missthe point of a more domain-specificwarning not to

confuse brains and a bullmarket.

Although domain-specific rules have advantages, they also have limits. Their

specific content will make them more likely tospread within their domain, but it

may also prevent them from spreading across domains. For example, engineers


14/19

28 CHIP HEATH, RICHARDP . LARRICK, and JOSHUAKLAYMAN Cognitive Repairs 29


15/19

than novel repairs. Anything that requires individuals to throw out old practices or

adopt new beliefsmay be technically and psychologically difficult.However, familiar repairs may sometimes be too familiar for their own good.

First,they may be less likely tocreate enthusiasm. Ifindividuals think thata new

repair differs only trivially from current practice, they may seeno advantage toit.

Because individuals often ignore thefamiliar, would-be change agents often strive

to create the perception that their programs are novel and unique (Abrahamson,1996).Second, familiar repairs may be subject todistortion. Ifa repair seems par-

tially familiar, individuals may neglect its unfamiliar aspects or force them to

mimic the more familiar aspects (a process that psychologists call assimilation).

Forexample, the proper techniquefor brainstorming requires a specificsequence

of steps: first, a creative, idea-generation stage which does not allow criticism,

then a stage where ideasare evaluated and selected. Although organizations fre-

quently report that they use brainstorming, c areful examination reveals that the

organizations are actually engaged in a more familiar activity: a basic business

meeting (Zbaracki, in press). Thenovel aspects of the brainstorming procedure,

sucha s the separation ofstages and the no criticism rule, are often lost as brain-storming isassimilated to the more familiar practice of the standard business

meeting. In the end, only the attractive label remains. In situations where assimt-

lation is a problem, repairs that are novel may be less likely to suffer distortion

than repairs that are more familiar because novel repairs do not evokethe preex-istingknowledge that leadsto assimilation.

Socialversus Individual

Many ofthe cognitiverepairs we have considered are social; theyworkbecause

individuals interact with others (e.g., single-case borequestions in the Deming

Prize organization, or th e murderhoard at the Pentagon). Otherrepairs are indi-vidual; individuals apply them to theirow n learning processes without theinter-

vention of others (e.g., individuals learn to use the Five Whys, and individual

investigators at the Board ofTrade learn to avoid yes or no questions).

In general, we suspectthat manysuccessful repairs will be social because indi-vidualsmay notrecognizethe need to repair themselves. The very cognitiveshort-

comings that organizations might hope to repair will make it difficult for

individuals to see their own flaws. As we have discussed,individuals tend to retain

current assumptions in the face of conflicting data (Klayman, 1995). Also, they

interpretevents in ways that protect their self-image; they avoidpotentially threat-

ening feedback(Frey, 1986; Larrick, 1993)and attribute their poor outcomes to

luck or forces outside theircontrol. Although individualsmay find it hard to rec-

ognize theirown biases, they may find it easier torecognize the biases ofothers.

Manyoftherepairswe document have thefeel of(friendly) social gamesmanship.

Forexample, learners at FP&L did notconsider thehypothesis thatthey had been

ruckytheir colleagues considered it for them. Similarly,duringweekly micro-

biologylab meetings,researchers did nothave to suggest alternative ways ofinter-

preting their evidence, their peers did so. Social competitionamong individuals

aids the spreadofrepairs even when individuals are overconfident and believe

they wouldhave donejustas well without therepair.

Social repairs do have to overcomesome disadvantages. For example,individ-

uals may not appreciate others who attempt to repair their biases, and they may

dismiss therepairattempts as the productof picky or abrasivepersonalities. Thus,

social repairs may be moresuccessful when an individual understands that his or

her tormentors are playing an important, formal role. Individuals may find it easier

to entertain an antagonists critiques when he or she is labeled as adevils advo-

cate, or whenthe individual isappearing before the murder board. Disney clar-

ified the role of the evaluators and theoccasion by establishingits norm of frank

feedback andby labelingits tryouts a s The Gong Show.

Eventually,social repairsmay be transformed into individual repairs as individ-

ualslearn to imitate the patterns of analysis forced on them b y others. In order for

individuals to learn, they need vivid, immediate feedback. Social encounters are

likely to provide a key source of such feedback. For example, when Deming

examiners ask single-case-bore questions, or when lab colleagues try to shoot

holes in a lab presentation, individual learnersmay eventually learn to engage in

preemptive self-criticism inorder tolookbetter during social encounters(Tetlock,

1992).(Many academic papers arebetterbecause authors learn to mentally simu-late potentialreviewerscomments.)Such repairsinvoke social forces at two dif-

ferentlevels: individuals who anticipate social interaction may be more aware of

someoftheirown shortcomings, andthenactual social interaction mayovercome

additionalshortcomings that individualscannot repair on their own.

Top-Downversus Bottom-Up

Some cognitive repairs originate fromtop-down withinan organization. Typ -ically these repairs are deliberately designed and implemented by managers or

outside experts.Others arise from bottom-upthrough informal observation orser-

endipitous discovery from the people who are doingthe work.

The source of the repair isimportantbecause it is likely to affect its form.In

general, bottom up repairs, such as organizational adages, willbe simpler and

more domain-specific than top-down repairs designed by technically sophisti-

cated engineers, statisticians, ormanagement gurus. The localoriginofbottom-uprepairs may also make them feel more familiar and acceptable than top-down

repairs. Thus, the originofa repair will be highly correlated withmany of the

tradeoffs we have already discussed.

More importantly, theorigin of the repair is also likely to affect h ow potential

adopters perceive it . Top-down repairs may beperceivedwith suspicion orhostil-

ity precisely because they originateoutside the organizationor are imposed from

above. Front-line workers may doubt that outsiders like consultants understand

30 CHIPHEATH, RICHARD P. LARRICK, andJOSHUAKLAYMAN Cognitive Repairs 31


16/19

their situationwell enough to make wise recommendations. When managers sug-

gest a repair, theyseem to imply that employees have been performing poorly or

cannot be trusted to perform theirjobcorrectly. Ifso, then individuals m ay resist

adopting a repair because of the same kind ofself-serving biases we discussed ear-

lier.

Top-downrepairs may also be resisted because they will be perceived a s driven

by politics or fashion and notby thedemands of the task. Some top-down repairs

may be resisted because they seem too political. Particularly when top-down

repairs rely on fixed procedures,they may provoke resistance because individuals

may think thatthey a redesigned tocentralize control or removeindividual- discre-t

ROB-Cognitive Repairs.pdf

Documents