ArticlePDF Available

Overcoming the Inevitable Anchoring Effect: Considering the Opposite Compensates for Selective Accessibility

Authors:

Abstract

Anchoring effects—the assimilation of a numeric estimate to a previously considered standard—have proved to be remarkably robust. Results of two studies, however, demonstrate that anchoring can be reduced by applying a consider-the-opposite strategy. Based on the Selective Accessibility Model, which assumes that anchoring is mediated by the selectively increased accessibility of anchor-consistent knowledge, the authors hypothesized that increasing the accessibility of anchor-inconsistent knowledge mitigates the effect. Considering the opposite (i.e., generating reasons why an anchor is inappropriate) fulfills this objective and consequently proves to be a successful corrective strategy. In a real-world setting using experts as participants, Study 1 dem-onstrated that listing arguments that speak against a provided anchor value reduces the effect. Study 2 further revealed that the effects of anchoring and considering the opposite are additive.
PER SON AL ITY AND SO CIAL PSY CHOL OGY BUL LE TIN Mussweiler et al. / RE DUCING THE AN CHORING EF FECT
Over coming the In ev i ta ble An choring
Ef fect: Con sidering the Op po site
Com pen sates for Se lec tive Ac ces si bil ity
Thomas Mussweiler
Fritz Strack
Tim Pfeif fer
Universität Würzburg
Anchoring effects—the assim i la tion of a numeric esti mate to a
pre vi ously con sid ered stan dard—have proved to be remark ably
robust. Results of two stud ies, how ever, dem on strate that anchor -
ing can be reduced by apply ing a con sider-the-oppo site strat egy.
Based on the Selec tive Acces si bil ity Model, which assumes that
anchor ing is medi ated by the selec tively increased acces si bil ity of
anchor-con sis tent knowl edge, the authors hypoth e sized that
increas ing the acces si bil ity of anchor-incon sis tent knowl edge
mit i gates the effect. Con sidering the oppo site (i.e., gen er at ing
rea sons why an anchor is inap pro pri ate) ful fills this objec tive
and con se quently proves to be a suc cess ful cor rec tive strat egy. In
a real-world set ting using experts as par tic i pants, Study 1 dem
-
on strated that list ing argu ments that speak against a pro vided
anchor value reduces the effect. Study 2 fur ther revealed that the
effects of anchor ing and con sid er ing the oppo site are addi tive.
Human judg ment falls prey to a vari ety of sys tem atic
biases and dis tor tions (for an over view, see Kahneman,
Slovic, & Tversky, 1982). In many cases, these biases
result from the use of judg men tal heuristics (Tversky &
Kahneman, 1974) that are highly adap tive and ben e fi
-
cial under most cir cum stances but also may pro duce dis -
tor tions (e.g., Arkes, 1991). One typ i cal find ing—called
the anchor ing effect (Tversky & Kahneman, 1974)—is
that numeric esti mates are assim i lated to a pre vi ously
con sid ered stan dard of com par i son. In what is prob a bly
the best-known dem on stra tion of this effect, Tversky and
Kahneman (1974) first asked their research par tic i pants
whether the per cent age of Afri can nations in the United
Nations (UN) is higher or lower than an arbi trary num -
ber (the anchor) that had osten si bly been deter mined
by spin ning a wheel of for tune (e.g., 65% or 10%). Par -
tic i pants were then asked to give their best esti mate of
this per cent age. Abso lute judg ments were assim i lated to
the pro vided anchor value so that the mean esti mate of
par tic i pants who received the high anchor was 45%,
com pared to 25% for par tic i pants who received the low
anchor.
Anchoring effects such as these have proved to be a
truly ubiq ui tous phe nom e non that has been observed in
a broad array of dif fer ent judg men tal domains (see
Mussweiler, 1997; Mussweiler & Strack, 1999a). Despite
this ubiq uity of anchor ing and other judg men tal biases,
crit ics (e.g., Gigerenzer, 1991; Hogarth, 1981) have
argued that many of these biases are more appar ent than
real in that they dis ap pear in infor ma tion-rich, nat u ral
envi ron ments and are thus lim ited to the psy cho log i cal
lab o ra tory (but see Gilovich, 1991). This, how ever, is not
true for the anchor ing effect, which has clear prac ti cal
rel e vance for many deci sions in real-world set tings. For
exam ple, pric ing deci sions (Northcraft & Neale, 1987)
as well as esti mates for prime inter est rates (Russo &
Schoemaker, 1989) were found to be sus cep ti ble to the
anchor ing bias. More over, anchor ing appears to play a
sig nif i cant role in the nego ti a tion pro cess: It has been
dem on strated that the final agree ment of a nego ti a tion
is strongly influ enced by an ini tial offer (Chertkoff &
Conley, 1967; Liebert, Smith, Hill, & Keiffer, 1968). This
Au thors’ Note: The pres ent re search was sup ported by a grant from
the Deut sche Forschungsgemeinschaft. We thank Hans-Pe ter Erb,
Birte Englich, Jens Förster, Roland Neumann, and Lioba Werth for
dis cus sions of the is sue; Cornelia Florig, Iris Müller, Markus
Schuhmacher, and Anke Siebers for their as sis tance in data col lec
-
tion and ques tion naire con struc tion; and Karin Jae ger for her as sis -
tance in ed it ing the manu script. Cor re spon dence con cern ing this ar ti -
cle should be ad dressed to Thomas Mussweiler, Uni ver sity of
Würzburg, Psychologie II, Roentgenring 10, 97070 Würzburg, Ger -
many; e-mail: mussweiler@psychologie.uni-wuerzburg.de.
PSPB, Vol. 26 No. 9, September 2000 1142-1150
© 2000 by the So ci ety for Per son al ity and So cial Psy chol ogy, Inc.
1142
find ing has been con cep tu al ized as an anchor ing effect:
The ini tial offer serves as an anchor to which the agree -
ment is assim i lated (Neale & Bazerman, 1991; Ritov,
1996).
THE RO BUST NESS OF THE
AN CHORING PHE NOM E NON
Not only is the anchor ing effect influ en tial in a pleth
-
ora of judg men tal set tings, but this influ ence is also
remark ably robust. For one, anchor ing occurs even if the
anchor val ues are clearly unin for ma tive for the crit i cal
esti mate because—as in Tversky and Kahneman’s (1974)
clas sic study—anchors are ran domly selected (e.g.,
Cervone & Peake, 1986; Mussweiler & Strack, in press).
More over, anchor ing remains unin flu enced by the
extrem ity of the anchor (e.g., Chap man & John son,
1994; Mussweiler, Förster, & Strack, 1997; Strack &
Mussweiler, 1997) so that even implau si bly extreme val
-
ues yield an effect. For exam ple, in one of our own stud
-
ies (Strack & Mussweiler, 1997, Study 3), esti mates for
the age of Mahatma Gan dhi were assim i lated to an
unrea son ably high anchor value of 140 years.
Fur ther more, anchor ing effects appear to be inde
-
pend ent of par tic i pants’ moti va tion (Wil son, Hous ton,
Etling, & Brekke, 1996). Spe cifically, the attempt to
improve accu racy by award ing a prize for the best esti
-
mate proved unsuc cess ful.
1
In addi tion, it has been dem -
on strated that anchor ing occurs inde pend ently of par
-
tic i pants’ exper tise (Joyce & Biddle, 1981; Northcraft &
Neale, 1987; Wright & Ander son, 1989).
2
For exam ple,
Northcraft and Neale (1987) had a group of expe ri
-
enced real estate agents esti mate the value of a house.
Par tic i pants were given all the infor ma tion that is typ i
-
cally impor tant to make this esti mate (e.g., major char ac -
ter is tics of the prop erty, prices for neigh bor ing prop er -
ties) and had the oppor tu nity to inspect the house.
Although rel e vant infor ma tion was thus eas ily acces si -
ble, the experts were influ enced by the given list ing price
(i.e., the anchor).
Prob a bly the most strik ing dem on stra tion of the
robust ness of the phe nom e non, how ever, stems from
research dem on strat ing that explicit instruc tions to cor -
rect for a poten tial influ ence of an anchor do not mit i
-
gate the effect (Wil son et al., 1996). In fact, even explic -
itly fore warn ing judges about the poten tial dis tor tion
and inform ing them about its direc tion did not dimin ish
the effect. Taken together, these find ings indi cate that
anchor ing is an excep tion ally robust phe nom e non that
is dif fi cult to avoid.
As men tioned before, anchor ing plays an impor tant
role in many real-world sit u a tions, in which fall ing prey
to the bias may entail remark able costs for the deci sion
maker. For exam ple, in Northcraft and Neale’s (1987)
study, the experts’ esti mates of the appraisal value of the
house dif fered by more than $7,000 in the dif fer ent
anchor ing con di tions, which is equiv a lent to almost 10%
of the actual value. This points to the fact that anchor ing
can prove expen sive. In light of the ubiq uity of the
anchor ing phe nom e non, being able to pre vent this bias
seems impor tant to improve human judg ment. To
develop an appro pri ate strat egy, how ever, one has to
take into account the cog ni tive mech a nisms that under -
lie the anchor ing effect. Many judg men tal biases result
from inad e quate cog ni tive strat e gies rather than insuf fi -
cient moti va tion (Lord, Lepper, & Pres ton, 1984) so that
a cor rec tive strat egy may be best designed by com pen sat -
ing for the mech a nism that pro duces the dis tor tion in
the first place (Arkes, 1991; Fischhoff, 1982). Recently,
we (Mussweiler, 1997; Mussweiler et al., 1997;
Mussweiler & Strack, 1999a, 1999b, 2000 Strack &
Mussweiler, 1997) have pro posed a Selec tive Acces si bil -
ity Model that spec i fies the cog ni tive mech a nisms that
under lie judg men tal anchor ing. This model may help
iden tify a strat egy to mit i gate the effect.
THE SE LEC TIVE
AC CES SI BIL ITY MODEL
The Selec tive Acces si bil ity Model (for a more elab o
-
rate account, see Mussweiler & Strack, 1999a) pos tu lates
that anchor ing effects are medi ated by a selec tive
increase in the acces si bil ity of anchor-con sis tent seman -
tic knowl edge about the tar get (for related notions, see
Chap man & John son, 1999; Pohl, 1996). We assume that
judges com pare the tar get with the anchor by test ing the
pos si bil ity that the tar get’s value is equal to the anchor
value. For exam ple, judges who are asked whether the
aver age price for a Ger man car is higher or lower than
40,000 Ger man Marks are assumed to test the pos si bil ity
that the aver age price actu ally is 40,000 Marks. To do so,
they selec tively retrieve knowl edge from mem ory that is
con sis tent with this assump tion (e.g., “A Mercedes or a
BMW is even more expen sive,” etc.) (Trope & Liberman,
1996). As a con se quence, the acces si bil ity of anchor-con -
sis tent knowl edge is increased. To gen er ate the final
numeric esti mate, judges then rely pri mar ily on eas ily
acces si ble knowl edge (Hig gins, 1996; Wyer & Srull,
1989) so that their esti mate is heavily influ enced by the
anchor-con sis tent knowl edge gen er ated before. On the
sur face, this is appar ent in an assim i la tion of the final
esti mate to the anchor value (for empir i cal sup port of
these assump tions, see Mussweiler et al., 1997;
Mussweiler & Strack, 1999b, 2000; Strack & Mussweiler,
1997; for an over view, see Mussweiler & Strack, 1999a).
From this selec tive acces si bil ity per spec tive, anchor -
ing results because the knowl edge base that is used to
make the numeric judg ment is dis torted in that
anchor-con sis tent knowl edge is more acces si ble than
anchor-incon sis tent knowl edge. Con se quently, reduc -
Mussweiler et al. / RE DUCING THE AN CHORING EF FECT 1143
ing this selec tiv ity may mit i gate the effect. That is, if the
acces si bil ity of anchor-incon sis tent knowl edge is
increased, this knowl edge should be equally used to
make the final esti mate and may thus com pen sate for the
effects of eas ily acces si ble anchor-con sis tent knowl edge.
Hence, a pro ce dure that increases the acces si bil ity of
anchor-incon sis tent knowl edge con sti tutes a prom is ing
can di date for a suc cess ful cor rec tive strat egy.
CON SIDER THE OP PO SITE
AS A COR REC TIVE STRAT EGY
Inducing judges to con sider rea sons why the impli ca -
tions of the anchor value may be wrong may con sti tute
such a strat egy. In line with this assump tion, it has been
dem on strated that con sid er ing the oppo site (Lord,
Lepper, & Pres ton, 1984), that is, tak ing into account evi
-
dence that is incon sis tent with one’s ini tial beliefs, is an
effec tive strat egy to improve human judg ment in a vari
-
ety of domains. For exam ple, Koriat, Lichtenstein, and
Fischhoff (1980) found that apply ing a con -
sider-the-oppo site strat egy reduces over con fi dence in
the cor rect ness of a cho sen answer: Inducing par tic i -
pants to list argu ments that speak against the valid ity of
their response reduces their con fi dence in its cor rect
-
ness (see also Grif fin, Dunning, & Ross, 1990; Hoch,
1985). This may be the case because over con fi dence
results from a neglect of evi dence that con tra dicts the
cho sen alter na tive so that mak ing this evi dence more
salient reduces the effect (Koriat et al., 1980). Sim i lar
strat e gies were found to mit i gate other judg men tal
biases, such as the hind sight bias (Arkes, Faust,
Guilmette, & Hart, 1988; Davies, 1992), and biased pro -
cess ing of new infor ma tion (Lord et al., 1984).
The psy cho log i cal pro cesses that medi ate these phe -
nom ena (Koehler, 1991) appear to be sim i lar to those
that under lie judg men tal anchor ing. Hence, a con sider-
the-oppo site strat egy also may reduce the ubiq ui tous
anchor ing effect. In fact, some recent data (Chap man &
John son, 1999) sup port this assump tion. In one study,
par tic i pants were asked to esti mate the like li hood that a
repub li can would win the next pres i den tial elec tions
after indi cat ing whether this prob a bil ity is higher or
lower than the last two dig its of their social secu rity num
-
ber. Before giv ing their final esti mate, some of the par tic -
i pants were instructed to list one rea son why a repub li -
can would win, some why a Repub li can would not win,
and some were not instructed to list any rea sons. A sig nif -
i cant anchor ing effect was only obtained for those par tic
-
i pants who listed rea sons that were con sis tent with the
impli ca tions of the anchor value (e.g., pro argu ments for
a prob a bil ity of more than 50%) or no rea sons at all.
Con sidering rea sons that were incon sis tent with the
anchor (e.g., con argu ments for a prob a bil ity of more
than 50%), how ever, elim i nated the bias. From the cur -
rent per spec tive, this may have been the case because
list ing rea sons that oppose the impli ca tions of the
anchor value increased the acces si bil ity of anchor-incon -
sis tent seman tic knowl edge. Thus, anchor-con sis tent
and anchor-incon sis tent knowl edge was sim i larly acces si -
ble so that the knowl edge base used to make the final
esti mate was unbi ased. Con se quently, judg ments were
unbi ased as well.
In the pres ent research, we explore the role a con
-
sider-the-oppo site strat egy may play in reduc ing the
anchor ing effect. Spe cifically, Study 1 tested whether this
strat egy may be fruit fully applied to a real-world set ting.
Just as judg men tal heuristics may have stron ger effects in
the psy cho log i cal lab o ra tory where judg ment-rel e vant
knowl edge is scarce (e.g., Gigerenzer, 1991; Hogarth,
1981), sim ple cor rec tive devices may be inef fec tive if par
-
tic i pants are suf fi ciently informed about the judg ment
domain. An effec tive cor rec tive strat egy, how ever, is
espe cially needed for such real-world deci sions because
it is here that a biased judg ment has poten tially high
costs. Thus, we tested the effec tive ness of con sid er ing
the oppo site in a real-world set ting, in which expe ri -
enced judges had all the infor ma tion avail able that is
needed to make an accu rate judg ment. Study 2 fur ther
exam ined the effec tive ness of con sid er ing the oppo site
using a more con trolled lab o ra tory set ting.
STUDY 1
To inves ti gate the effects of anchor ing and con sid er
-
ing the oppo site in a real-world set ting, we chose a car-
sell ing sce nario. Spe cifically, experts in the car busi ness
were approached and asked to esti mate the value of a
10-year-old car. They were given all the infor ma tion that
is typ i cally deemed impor tant to make this esti mate
(e.g., mile age, year), had the car right in front of them
through out the whole exper i ment, and were given the
oppor tu nity to inspect it. In addi tion, their moti va tion to
make a seri ous esti mate was increased by hold ing out the
pros pect of a repair job to them. Taken together, these
mea sures cre ated a sit u a tion that is very close to a real-
world inter ac tion.
To exam ine the effects of con sid er ing the oppo site,
half of the par tic i pants were induced to list anchor-
incon sis tent argu ments (e.g., rea sons why a high price is
inad e quate) before giv ing their numeric esti mate. Con -
sis tent with the above anal y sis, we expected this con sider-
the-oppo site strat egy to com pen sate for the selec tive
acces si bil ity increase that results from the anchor ing
manip u la tion. Spe cifically, list ing anchor-incon sis tent
argu ments should ren der the knowl edge base that is
used for the judg ment less biased, which in turn should
lead to less biased final esti mates.
It is impor tant to note, how ever, that in prin ci ple the
expected debiasing effect of con sid er ing the oppo site
1144 PER SON AL ITY AND SO CIAL PSY CHOL OGY BUL LE TIN
also may be ascribed to con ver sa tional or prag matic
influ ences (e.g., Grice, 1975). For exam ple, it may be dif -
fi cult for par tic i pants to give an esti mate that devi ates too
much from the sug gested value (i.e., the anchor)
because doing so would vio late com mon norms of pro -
pri ety. Spe cifically, a mechanic who is asked about his or
her opin ion con cern ing a high value for the car may feel
obliged to give a fairly high esti mate because doing oth -
er wise may insult his or her poten tial cus tomer. Asking
for rea sons why this value may be inap pro pri ate, how -
ever, may release him or her from such pro pri ety con
-
straints so that he or she feels free to give a lower esti -
mate. From this per spec tive, con sid er ing the oppo site
would then reduce anchor ing because it alle vi ates prag
-
matic con straints and not because it debiases the infor -
ma tional basis for the judg ment.
To pro vide empir i cal evi dence against this alter na tive
expla na tion, Study 1 also exam ines whether the
debiasing effects of con sid er ing the oppo site depend on
the amount of anchor-incon sis tent knowl edge that par -
tic i pants gen er ated. If—as we assume—the effects of
con sid er ing the oppo site are medi ated by the impli ca -
tions of the gen er ated infor ma tion, then the more infor -
ma tion is gen er ated, the stron ger its debiasing effect
should be. Such a depend ency would be dif fi cult to
explain with prag matic influ ences.
Method
Par tic i pants. Par tic i pants were 60 male car experts; 44
of them were car mechan ics and 16 were car deal ers. All
of the par tic i pants had more than 5 years of expe ri ence
on the job. In fact, the major ity (N = 51) had worked in
the car busi ness for more than 10 years. Par tic i pants
were ran domly assigned to one of four exper i men tal
con di tions. Spe cifically, the con di tion they were
assigned to was ran domly deter mined before approach
-
ing them.
Mate rial. A 10-year-old car (1987 Opel Kadett E) was
used as the object to be eval u ated. An inde pend ent
expert esti mated an ade quate buy ing and sell ing price
for this car. The buy ing price (i.e., price for buy ing the
car from a dealer) was esti mated to be 4,500 Ger man
Marks (about U.S.$2,500 at the time); the sell ing price
(i.e., price for sell ing the car to a dealer) was 3,300 Ger -
man Marks (about U.S.$1,833 at the time). We used
these two esti mates to deter mine the anchor val ues. The
low anchor was set at 500 Ger man Marks below the sell -
ing price; the high anchor was set at 500 Marks above the
buy ing price. Thus, 2,800 Marks (about U.S.$1,556 at the
time) and 5,000 Marks (about U.S.$2,778 at the time)
served as anchor val ues.
Pro ce dure. Par tic i pants were approached indi vid u ally
at their place of employ ment. After arriv ing with his car,
the exper i menter requested to talk to an expert who
could tell him whether a lit tle bump he had in his car was
still worth fix ing given that the car was fairly old.
Typically, an expert was then sent to him and the exper i
-
menter explained that his girl friend had had a minor
col li sion with his car and that he was uncer tain whether
it would still be worth fix ing. To decide, he would like an
expert to esti mate what the actual value of this car was
and how much it would cost him to have it fixed. Then,
he pro vided the expert with the major facts about the
car, namely, its mile age (160,000 kilo me ters) and year
(1987), and gave his per sonal esti mate for its value (“I
thought that the car should sell for about 2,800/5,000
Marks”). For half of the par tic i pants, this esti mate was
equiv a lent to the low anchor value of 2,800 Ger man
Marks; for the other half of the par tic i pants, the esti mate
was equiv a lent to the high anchor of 5,000 Marks.
Con gru ent with the stan dard anchor ing pro ce dure,
the expert was first asked to indi cate whether the anchor
value was too high or too low (“Accord ing to your opin
-
ion, is this value too high or too low?”). He was then
asked to give his esti mate for the value (“Could you tell
me, what do you think is the approx i mate price for the
car as you see it?”). Before giv ing this esti mate, how ever,
half of the par tic i pants were asked for rea sons why the
anchor value might be inap pro pri ate (“A friend of mine
men tioned yes ter day that he thought this value is too
high/low. What would you say argues against this
price?”). For the other half of the par tic i pants, the abso -
lute ques tion imme di ately fol lowed their com par a tive
judg ment. Thus, the four exper i men tal con di tions
resulted from a com bi na tion of anchor (high vs. low)
and argu ment list ing (no argu ment vs. anchor-incon sis
-
tent argu ment). Both fac tors were manip u lated between
par tic i pants.
To main tain cred i bil ity, par tic i pants were then asked
to esti mate the costs for the required repair. Finally, the
exper i menter inquired how long the par tic i pant had
worked in the car busi ness already. The exper i menter
then thanked the expert, said that he would think about
whether he should have his car fixed, and left.
Re sults
Num ber of gen er ated argu ments. Par tic i pants who
received the low anchor gen er ated less anchor-incon sis -
tent argu ments (M = 1.33) than did par tic i pants who
received the high anchor (M = 2.67), t(28) = 3.89, p <
.001. This sug gests that find ing argu ments that indi cate
that the low anchor was too low was more dif fi cult than
find ing argu ments that indi cate that the high anchor
was too high. This may be the case because par tic i pants
saw the low anchor (which was close to the price a dealer
would typ i cally pay for the car) to be a more appro pri ate
esti mate than the high anchor. Con sis tent with this
Mussweiler et al. / RE DUCING THE AN CHORING EF FECT 1145
assump tion, the aver age esti mate over all four exper i
-
men tal con di tions (M = 2,999 Marks) is much closer to
the low anchor than to the high anchor.
Abso lute esti mates. Inspec tion of Table 1 reveals that the
typ i cal anchor ing effect is rep li cated. Over all, the high
anchor led to higher esti mates for the value of the car
(M = 3,347 Marks) than the did low anchor (M = 2,652
Marks), F(1, 56) = 16.92, p < .001. More impor tant, the
mag ni tude of the anchor ing effect depended on
whether par tic i pants were instructed to gen er ate
anchor-incon sis tent argu ments. Spe cifically, anchor ing
was weaker when par tic i pants were instructed to gen er -
ate anchor-incon sis tent argu ments, F(1, 56) = 4.25, p <
.04, for the inter ac tion of anchor and argu ment list ing.
Con trast anal y ses revealed that the dif fer ence between
the esti mates for the high and the low anchor was sig nif i
-
cant when no argu ments were listed, t(56) = 4.37, p <
.001, one-tailed. This anchor ing effect, how ever, was
only mar gin ally sig nif i cant if anchor-incon sis tent argu -
ments were listed, t(56) = 1.45, p < .08, one-tailed.
Correlational anal y sis. To deter mine whether the
debiasing effects of con sid er ing the oppo site depend on
the amount of gen er ated infor ma tion, we cor re lated the
num ber of gen er ated argu ments with the final esti mates.
For the low anchor con di tion, both quan ti ties were unre
-
lated (r = –.23, p > .4). For the high anchor con di tion,
how ever, both were neg a tively cor re lated (r = –.39, p <
.07, one-tailed). Thus, the more anchor-incon sis tent
argu ments (i.e., argu ments indi cat ing that the high
anchor is too high) were gen er ated the lower was the
final esti mate for the value of the car.
Dis cus sion
The impli ca tions of these find ings are man i fold. For
one, they dem on strate that anchor ing effects occur in
real-world set tings in which experts have all nec es sary
infor ma tion avail able to make the crit i cal judg ment.
More over, the esti mates of the con trol group par tic i -
pants dem on strate that the size of the bias can be
remark able. Spe cifically, esti mates in the high and low
anchor con di tion devi ated by more than 1,000 Ger man
Marks, which is equiv a lent to more than 25% of the
actual value of the car. Thus, in line with pre vi ous
research (e.g., Northcraft & Neale, 1987; Russo & Shoe -
maker, 1989), our find ings dem on strate that anchor ing
is a potent judg men tal bias in every day judg ment and
deci sion mak ing.
More impor tant, these results also dem on strate that
the effects of anchor ing may be mit i gated by apply ing a
con sider-the-oppo site strat egy. Given the extraor di nary
robust ness of the anchor ing effect and pre vi ous fail ures
to find an ade quate means to reduce it (e.g., Wil son et al.,
1996), this find ing seems espe cially note wor thy. Fur ther -
more, our data dem on strate that anchor ing effects are
remark ably robust: Although con sid er ing the oppo site
reduced the effect, there is still a ten dency to give higher
esti mates after con sid er ing the high anchor than the low
anchor. Thus, con sid er ing the oppo site did not com
-
pletely remove the dis tor tion.
Finally, our results speak to the mech a nism that may
be respon si ble for the debiasing effect of con sid er ing
the oppo site. Spe cifically, our find ings seem more con -
sis tent with the assump tion that con sid er ing the oppo -
site mit i gates anchor ing because it debiases the infor ma -
tional basis for the judg ment. For one, con sid er ing the
oppo site reduced the effects of high as well as low
anchors. The debiasing effect for low anchors, how ever,
is dif fi cult to explain in terms of prag matic influ ences.
Because giv ing an esti mate that is higher than the sug -
gested low anchor is unlikely to offend the poten tial cus -
tomer, there exist no pro pri ety con straints that could be
lifted by ask ing for anchor-incon sis tent argu ments. Con
-
se quently, con sid er ing the oppo site should only influ -
ence the effects of high but not of low anchors. Our find
-
ings, how ever, dem on strate that this is not the case and
are thus dif fi cult to rec on cile with the prag matic
expla na tion.
More impor tant, the fact that the final esti mate was
cor re lated with the num ber of gen er ated argu ments sug
-
gests that the effect of con sid er ing the oppo site is medi -
ated by the knowl edge that is ren dered acces si ble. The
obtained cor re la tion seems dif fi cult to explain with
prag matic influ ences because the prag matic con straints
of the sit u a tion are the same for par tic i pants who gen er
-
ate few argu ments and for those who gen er ate many.
Nota bly, the cor re la tion of gen er ated argu ments and
final esti mate only held for the high anchor con di tion
but not for the low anchor con di tion. Although this
diver gence was unex pected and may shed some doubt
on the via bil ity of our account, our sup ple men tal anal y -
sis helps to rec on cile this find ing with the cur rent the o -
ret i cal per spec tive. Spe cifically, the fact that the over all
mean esti mate was very close to our low anchor value
sug gests that par tic i pants saw this value as rea son able.
Con se quently, they had dif fi cul ties gen er at ing argu -
ments that speak against it. This is appar ent in the
extremely low num ber of argu ments listed. As a result,
1146 PER SON AL ITY AND SO CIAL PSY CHOL OGY BUL LE TIN
TA BLE 1: Ab so lute Es ti mates for the Value of the Car by An chor
and Ar gu ment
Ar gu ment
An chor No An chor In con sis tent
High 3,563 3,130
Low 2,520 2,783
NOTE: Es ti mates are given in Ger man Marks. N = 15 in all cells.
the gen er ated argu ments only exerted a small effect so
that the final esti mates remain uncorrelated with the
num ber of argu ments. In sum, the pres ent find ings seem
more con sis tent with the assump tion that the effects of
con sid er ing the oppo site are due to its debiasing influ -
ence on the infor ma tional basis for the judg ment rather
than its pro pri ety lift ing qual i ties. Study 2 was in part
con ducted to fur ther rule out the prag matic account.
STUDY 2
To fur ther rule out the prag matic account, we exam
-
ined whether con sid er ing the oppo site also would mit i -
gate the effects of anchor val ues that were osten si bly
selected at ran dom. Because such arbi trary anchors are
not thought to have been delib er ately selected by the
exper i menter, par tic i pants are unlikely to feel obliged to
give an esti mate that is close to them. That is, eval u at ing
ran dom anchor val ues does not entail the kind of pro pri
-
ety con straints that may have medi ated the find ings of
Study 1. Con se quently, if the effects of con sid er ing the
oppo site were due to its pro pri ety-lift ing qual i ties, it
should not mit i gate the effects of ran dom anchor val ues.
Thus, one objec tive of Study 2 was to pro vide fur ther
evi dence that speaks against the prag matic account. In
addi tion, we attempted to fur ther explore the effec tive -
ness of the con sid er ing-the-oppo site strat egy. In Study 1,
the argu ment-list ing pro ce dure was explic itly designed
to com pen sate for the effects of anchor ing. That is, par
-
tic i pants were instructed to gen er ate argu ments that
speak against the pro vided anchor val ues so that the
impli ca tions of argu ments and anchors were opposed to
one another. One may well argue that this is a nec es sary
pre con di tion for the argu ment-list ing pro ce dure to be
effec tive. Spe cifically, list ing anchor-con sis tent argu -
ments may not have an effect on esti mates because doing
so may not change the knowl edge base that is used to
make the final esti mate. For exam ple, gen er at ing argu -
ments that speak for a high value of the car may not exert
an effect if a high anchor is pres ent because this anchor
value induces judges to think of these argu ments any way
so that no addi tional knowl edge is ren dered eas ily acces
-
si ble. Study 2 was designed to test whether argu ment list -
ing depends on this restric tion and is only influ en tial if
counter argu ments are gen er ated. To do so, we manip u -
lated the impli ca tions of the anchors and the gen er ated
argu ments inde pend ently of one another.
Method
Par tic i pants. We recruited 31 nonpsychology stu dents
at the Uni ver sity of Würzburg as par tic i pants and ran -
domly assigned them to one of the three argu ment-
list ing con di tions. They were asked to take part in a pre -
test for the con struc tion of a sur vey ques tion naire and
were offered a choc o late bar as com pen sa tion.
Mate rials. The ques tion naire con sisted of two pairs of
ques tions. The first ques tion pair per tained to the like li
-
hood that Ger man Chan cel lor Kohl would win the next
elec tion. Spe cifically, par tic i pants were first asked to
indi cate whether the like li hood that Chan cel lor Kohl
would win the next elec tion is higher or lower than
either 20% or 80%. In the sub se quent ques tion, they
were asked to give an esti mate of this prob a bil ity. The
sec ond ques tion per tained to the like li hood that the
Ger man oppo si tion leader Oskar Lafontaine would be
nom i nated as a can di date for the elec tion. Again, par tic i
-
pants were first asked to indi cate whether this per cent -
age is higher or lower than either 20% or 80% and then
asked to give a numeric esti mate.
The anchors were set at 20% and 80%. For half of the
par tic i pants, the first ques tion pair included the high
anchor and the sec ond ques tion pair included the low
anchor. For the other half, this assign ment was reversed.
The argu ment-list ing manip u la tion was inserted after
the com par a tive ques tion. Par tic i pants in the high con
-
di tion were instructed to list three argu ments that
implied that Chan cel lor Kohl would win the next elec -
tion and that oppo si tion leader Lafontaine would be
nom i nated, respec tively. Par tic i pants in the low con di -
tion were instructed to list three argu ments that spoke
against these pos si bil i ties. Par tic i pants in the no argu -
ment con di tion did not list any argu ments. For them, the
abso lute ques tions imme di ately fol lowed the com par a -
tive ones.
In sum, the six exper i men tal con di tions resulted from
a com bi na tion of anchor (high vs. low), which was var ied
within par tic i pants, and argu ment list ing (no vs. high vs.
low), which was manip u lated between par tic i pants.
Pro ce dure. Par tic i pants were run in groups of up to 10.
They were recruited in the uni ver sity caf e te ria, led to a
sep a rate room, and handed the ques tion naire. In the
instruc tions, par tic i pants were informed that they were
tak ing part in a pre test for the con struc tion of a sur vey
ques tion naire. It was empha sized that the pur pose of the
pre test was to find the best word ing for the opin ion sur -
vey. More over, they were told that some of the ques tions
would require a com par i son with a given numer i cal stan
-
dard and that these stan dards were ran domly selected by
using a mech a nism sim i lar to that of a wheel of for tune.
It was pointed out that this was nec es sary to min i mize a
pos si ble influ ence that the stan dards may have on the
answers and to iden tify the impact of dif fer ent ques tion
for mats. The ran dom selec tion of the anchor val ues was
empha sized to reduce their ascribed infor ma tive ness
(Grice, 1975) and thus ensure that the obtained effects
were not medi ated by con ver sa tional infer ences (cf.
Jacowitz & Kahneman, 1995).
Mussweiler et al. / RE DUCING THE AN CHORING EF FECT 1147
Re sults
Pre lim i nary anal y sis. A pre lim i nary Anchor (high vs.
low) × Argu ment List ing (no vs. high vs. low) × Con tent
(Kohl vs. Lafontaine) ANOVA dem on strated that the
spe cific con tent exerted no effect (F < 1) for all effects,
includ ing con tent. Con se quently, this fac tor is not con
-
sid ered in the main anal y sis.
Abso lute esti mates. To allow for a com par i son of esti
-
mates given for dif fer ent con tent domains, responses
were trans formed into z scores for each ques tion. Thus,
the result ing cell means reflect par tic i pants’ aver age
devi a tions from the ques tion mean in units of the per ti -
nent stan dard devi a tion. As is appar ent in Table 2, the
typ i cal anchor ing effect was rep li cated. High anchors
led to higher esti mates (M = .33) than did low anchors
(M = –.32), F(1, 28) = 18.28, p < .001. More impor tant,
abso lute esti mates also depended on the impli ca tions of
the argu ments that were listed after the com par a tive
ques tion. Spe cifically, abso lute esti mates were low est for
low argu ments (M = –.46), high est for high argu ments
(M = .51), and inter me di ate if no argu ments were listed
at all (M = –.01), F(2, 28) = 3.95, p < .03. Finally, the mag -
ni tude of the anchor ing effect (i.e., the dif fer ence
between the esti mates given for high and low anchors)
remained unin flu enced by the argu ment list ing, F(2, 28) <
1, for the inter ac tion of anchor and argu ment list ing.
Dis cus sion
These results indi cate that the effects of argu ment list -
ing that we obtained in Study 1 also hold for ran domly
selected anchor val ues. Because eval u at ing such arbi -
trary anchors is unlikely to entail any pro pri ety con
-
straints, these find ings speak against a prag matic
account for the effects of con sid er ing the oppo site. At
the same time, they are con sis tent with the assump tion
that con sid er ing the oppo site mit i gates anchor ing
because it debiases the infor ma tional basis for the judg
-
ment. Fur ther more, the results of Study 2 sug gest that
the effects of argu ment list ing are not restricted to the
gen er a tion of anchor-incon sis tent argu ments. Although
inspec tion of the means depicted in Table 2 reveals a ten -
dency for anchor-incon sis tent argu ments to have a stron
-
ger effect than anchor-con sis tent argu ments, this ten -
dency does not yield a sig nif i cant effect. This sug gests
that regard less of the anchor value, instruct ing par tic i
-
pants to list argu ments leads them to access knowl edge
that they have not pre vi ously thought about. For exam
-
ple, although par tic i pants who received the high anchor
value of 80% for the prob a bil ity that Chan cel lor Kohl
would be reelected pre sum ably gen er ate evi dence that
favors this pos si bil ity already, they gen er ate addi tional
favor able evi dence when explic itly instructed to do so. As
a con se quence, gen er at ing argu ments makes an inde -
pend ent con tri bu tion to the acces si bil ity of judg -
ment-rel e vant knowl edge so that it also exerts an inde
-
pend ent influ ence on esti mates that are based on this
knowl edge. Thus, the effec tive ness of the argu ment-list -
ing pro ce dure does not appear to be restricted to the
gen er a tion of counter argu ments.
GEN ERAL DIS CUS SION
We have exam ined con sid er ing the oppo site as a cor
-
rec tive strat egy for the ubiq ui tous anchor ing effect. In a
real-world set ting using experts as par tic i pants, Study 1
dem on strated that judg men tal anchor ing can be mit i
-
gated by gen er at ing anchor-incon sis tent argu ments
before mak ing the numeric esti mate. This find ing seems
espe cially remark able because anchor ing has proved to
be an excep tion ally robust phe nom e non for which stan -
dard cor rec tive strat e gies, such as increas ing the moti va
-
tion to give an accu rate esti mate and inform ing judges
about the nature of the dis tort ing influ ence (Wil son et al.,
1996), remain uninfluential. Con sidering the oppo site
appears to be the first mech a nism that suc cess fully
reduces the anchor ing bias.
Cor rec tion by The ory-Based
Ad just ment Ver sus Con sidering
the Op po site
Recent con cep tu al iza tions of judg men tal cor rec tion
(e.g., Strack, 1992; Strack & Hannover, 1996; Wegener &
Petty, 1997; Wil son & Brekke, 1994) sug gest a gen eral
cor rec tive strat egy that—in prin ci ple—also may be
applied to the anchor ing phe nom e non. Spe cifically, it
has been sug gested that judg men tal cor rec tion often
takes the form of the ory-based adjust ment: To cor rect,
judges may con sult their naive the o ries about judg men
-
tal dis tor tion and deter mine the direc tion and mag ni -
tude of the bias. The ini tial judg ment is then adjusted in
the oppo site direc tion of the per ceived bias to a degree
that com pen sates for the assumed mag ni tude of the dis -
tor tion. Wil son et al.’s (1996) fail ure to reduce the
anchor ing bias by insti gat ing such the ory-based adjust -
ment, how ever, dem on strates that using this cor rec tive
device to elim i nate a pres ent bias is a dif fi cult task to mas
-
ter. This may be the case because—to cor rect suc cess -
fully—judges have to meet a mul ti tude of pre con di tions
(e.g., Strack, 1992; Strack & Hannover, 1996; Wil son &
1148 PER SON AL ITY AND SO CIAL PSY CHOL OGY BUL LE TIN
TA BLE 2: Ab so lute Es ti mates (z trans formed) by An chor and
Ar gu ment
Ar gu ment
An chor No High Low
High .37 (N = 11) .89 (N = 10) .26 (N = 10)
Low .39 (N = 11) .12 (N = 10) .67 (N = 10)
Brekke, 1994). Spe cifically, they have to be (a) moti vated
to give an accu rate judg ment, (b) aware of the poten -
tially dis tort ing influ ence, and (c) aware of the direc tion
and mag ni tude of this influ ence.
The con sider-the-oppo site strat egy, how ever, seems
less dif fi cult to mas ter. In con trast to the ory-based adjust
-
ment, judges merely have to be moti vated to give an
accu rate judg ment and to be aware of the dis tort ing
influ ence. Aware ness of the direc tion and the mag ni
-
tude of the dis tor tion is not nec es sary. Thus, con sid er ing
the oppo site seems an effec tive cor rec tive device able to
improve human judg ment even when other cor rec tive
strat e gies have failed. More over, its scope is not lim ited
to the anchor ing phe nom e non. As pointed out before, it
has proved to be suc cess ful for a vari ety of judg men tal
dis tor tions, such as the hind sight bias (e.g., Arkes et al.,
1988), over con fi dence (e.g., Koriat et al., 1980), biased
hypoth e sis-test ing (Lord et al., 1984), and biased pro
-
cess ing of novel infor ma tion (Lord et al., 1984).
3
CON CLU SION
Anchoring has been a long-stand ing enigma in psy -
cho log i cal research, with its remark able robust ness con -
sti tut ing one of the most enig matic char ac ter is tics. The
Selec tive Acces si bil ity Model, how ever, sug gests that
increas ing the acces si bil ity of anchor-incon sis tent
knowl edge reduces the mag ni tude of the anchor ing
bias. Con sidering the oppo site appears to be one way to
achieve this objec tive and con se quently proved to be an
effec tive cor rec tive strat egy. From the pres ent per spec
-
tive, con sid er ing the oppo site was suc cess ful because it
coun ter acted the very mech a nism that is respon si ble for
the bias, namely, selec tive acces si bil ity. In line with ear
-
lier con cep tu al iza tions of debiasing manip u la tions (e.g.,
Arkes, 1991; Fischhoff, 1982), this sug gests that first ana
-
lyz ing the psy cho log i cal mech a nisms that under lie a spe -
cific bias and then design ing a cor rec tive strat egy that
coun ter acts these mech a nisms is a fruit ful strat egy to
enhance human judg ment.
NOTES
1. At first sight, these find ings appear to be incon sis tent with other
data dem on strat ing that anchor ing is atten u ated by increas ing par tic i -
pants’ eval u a tion appre hen sion (Kruglanski & Freund, 1983). This dis -
crep ancy, how ever, may be due to the dif fer ent judg men tal par a digms
used in both stud ies. Wil son, Hous ton, Etling, and Brekke (1996) used
the stan dard par a digm (Tversky & Kahneman, 1974), whereas
Kruglanski and Freund (1983) exam ined anchor ing in the con text of
choices between con junc tive and dis junc tive events (Bar-Hillel, 1973).
Both par a digms dif fer with respect to par tic i pants’ abil ity to find the
cor rect answer. In the stan dard par a digm, par tic i pants can not pro vide
a cor rect answer regard less of their effort. In the lat ter par a digm, how -
ever, they can do so by ana lyz ing the indi vid ual prob a bil i ties that form
the crit i cal con junc tive and dis junc tive events.
2. To be sure, anchor ing as a bias in judg ments under uncer tainty
(Tversky & Kahneman, 1974) is unlikely to occur if judges know the
exact value for the esti mate. Thus, exper tise will only remain
uninfluential if it does not involve knowl edge of the true value.
3. Nat u rally, con sid er ing the oppo site does not inev i ta bly improve
judg ment. To the extent that judges’ aware ness of a dis tort ing influ
-
ence is false, con sid er ing the oppo site also may dis tort a pre vi ously
accu rate judg ment. Spe cifically, judges may cor rect for an influ ence
that does not exist.
REF ER ENCES
Arkes, A. R. (1991). Costs and ben e fits of judg ment errors: Impli ca
-
tions for debiasing. Psy cho log i cal Bul le tin, 110, 486-498.
Arkes, H. R., Faust, D., Guilmette, T. J., & Hart, K. (1988). Elim i nating
the hind sight bias. Jour nal of Applied Psy chol ogy, 73, 305-307.
Bar-Hillel, M. (1973). On the sub jec tive prob a bil ity of com pound
events. Orga ni za tional Behav ior and Human Per for mance, 9, 396-406.
Cervone, D., & Peake, P. K. (1986). Anchoring, effi cacy, and action:
The influ ence of judg men tal heuristics on self-effi cacy judg ment
and behav ior. Jour nal of Per son al ity and Social Psy chol ogy, 50, 492-501.
Chap man, G. B., & John son, E. J. (1999). Anchoring, acti va tion, and
the con struc tion of values. Orga ni za tional Behav ior and Human Deci -
sion Pro cesses, 79, 115-153.
Chertkoff, J. M., & Conley, M. (1967). Open ing offer and fre quency of
con ces sion as bar gain ing strat e gies. Jour nal of Per son al ity and Social
Psy chol ogy, 7, 181-185.
Davies, M. F. (1992). Field depend ence and hind sight bias: Cog ni tive
restruc tur ing and the gen er a tion of rea sons. Jour nal of Research in
Per son al ity, 26, 58-74.
Fischhoff, B. (1982). Debiasing. In D. Kahneman, P. Slovic, & A. Tversky
(Eds.), Judg ment under uncer tainty: Heuristics and biases. Cam bridge,
UK: Cam bridge Uni ver sity Press.
Gigerenzer, G. (1991). How to make cog ni tive illu sions dis ap pear:
Beyond “heuristics and biases.” In W. Stroebe & M. Hewstone
(Eds.), Euro pean review of social psy chol ogy (Vol. 2, pp. 83-115).
Chichester, UK: Wiley.
Gilovich, T. (1991). How we know what isn’t so: The fal li bil ity of human rea -
son in every day life. New York: Free Press.
Grice, H. P. (1975). Logic and con ver sa tion. In P. Cole & J. L. Mor gan
(Eds.), Syn tax and seman tics 3: Speech acts (pp. 41-58). New York: Aca -
demic Press.
Grif fin, D. W., Dunning, D., & Ross, L. (1990). The role of construal
pro cesses in over con fi dent pre dic tion about the self and oth ers.
Jour nal of Per son al ity and Social Psy chol ogy, 59, 1128-1139.
Hig gins, E. T. (1996). Knowl edge acti va tion: Acces si bil ity, appli ca bil ity,
and salience. In E. T. Hig gins & A. W. Kruglanski (Eds.), Social psy -
chol ogy: Hand book of basic prin ci ples (pp. 133-168). New York:
Guilford.
Hoch, S. J. (1985). Counterfactual rea son ing and accu racy in pre dict -
ing per sonal events. Jour nal of Exper i men tal Psy chol ogy: Learning,
Mem ory, and Cog ni tion, 11, 719-731.
Hogarth, R. (1981). Beyond dis crete biases: Func tional and dys func
-
tional aspects of judg men tal heuristics. Psy cho log i cal Bul le tin, 90,
197-217.
Jacowitz, K. E., & Kahneman, D. (1995). Mea sures of anchor ing in esti -
ma tion tasks. Per son al ity and Social Psy chol ogy Bul le tin, 21, 1161-1166.
Joyce, E. J., & Biddle, G. C. (1981). Anchoring and adjust ment in
proba bil is tic infer ences in audit ing. Jour nal of Account ing Research,
19, 120-145.
Kahneman, D., Slovic, P., & Tversky, A. (1982). Judg ment under uncer
-
tainty: Heuristics and biases. Cam bridge, UK: Cam bridge Uni ver sity
Press.
Koehler, D. J. (1991). Expla na tion, imag i na tion, and con fi dence in
judg ment. Psy cho log i cal Bul le tin, 110, 499-519.
Koriat, A., Lichtenstein, S., & Fischhoff, B. (1980). Rea sons for con fi
-
dence. Jour nal of Exper i men tal Psy chol ogy: Human Learning and Mem -
ory, 6, 107-118.
Kruglanski, A. W., & Freund, T. (1983). The freez ing and unfreez ing of
lay-infer ences: Effects on impressional pri macy, eth nic ste reo typ -
Mussweiler et al. / RE DUCING THE AN CHORING EF FECT 1149
ing, and numer i cal anchor ing. Jour nal of Exper i men tal Social Psy chol
-
ogy, 19, 448-468.
Liebert, R. M., Smith, W. P., Hill, J. H., & Keiffer, M. (1968). The effects
of infor ma tion and mag ni tude of ini tial offer on inter per sonal
nego ti a tion. Jour nal of Exper i men tal Social Psy chol ogy, 4, 431-441.
Lord, C. G., Lepper, M. R., & Pres ton, E. (1984). Con sidering the oppo
-
site: A cor rec tive strat egy for social judg ment. Jour nal of Per son al ity
and Social Psy chol ogy, 47, 1231-1243.
Mussweiler, T. (1997). A selec tive acces si bil ity model of anchor ing: Linking
the anchor ing heu ris tic to hypoth e sis-con sis tent test ing and seman tic prim -
ing. Lengerich, Ger many: Pabst.
Mussweiler, T., Förster, J., & Strack, F. (1997). Der Ankereffekt in
Abhängigkeit ankerkonsistenter Infor ma tion: Ein Modell
selektiver Zugänglichkeit [Anchoring effects and the appli ca bil ity
of anchor-con sis tent infor ma tion: A selec tive acces si bil ity model].
Zeitschrift für experimentelle Psychologie, 44, 589-615.
Mussweiler, T., & Strack, F. (1999a). Com paring is believ ing: A selec tive
acces si bil ity model of judg men tal anchor ing. In W. Stroebe &
M. Hewstone (Eds.), Euro pean review of social psy chol ogy (Vol. 10,
pp. 135-168). Chichester, UK: Wiley.
Mussweiler, T., & Strack, F. (1999b). Hypoth e sis-con sis tent test ing and
seman tic prim ing in the anchor ing par a digm: A selec tive acces si bil
-
ity model. Jour nal of Exper i men tal Social Psy chol ogy, 35, 136-164.
Mussweiler, T., & Strack, F. (2000). The use of cat e gory and exem plar
knowl edge in the solu tion of anchor ing tasks. Jour nal of Per son al ity
and Social Psy chol ogy, 78, 1038-1052.
Mussweiler, T., & Strack, F. (in press). Numeric judge ment under
uncer tainty: The role of knowl edge in anchor ing. Jour nal of Exper i -
men tal Social Psy chol ogy.
Neale, M. A., & Bazerman, M. H. (1991). Cog ni tion and ratio nal ity in
nego ti a tion. New York: Free Press.
Northcraft, G. B., & Neale, M. A. (1987). Experts, ama teurs, and real
estate: An anchor ing-and-adjust ment per spec tive on prop erty pric -
ing deci sions. Orga ni za tional Behav ior and Human Deci sion Pro cesses,
39, 84-97.
Pohl, R. F. (1996). Der Rückschaufehler: Eine systematische
Verfälschung der Erinnerung [The hind sight bias: A sys tem atic
mem ory dis tor tion]. Report Psychologie, 21, 596-609.
Ritov, I. (1996). Anchoring in sim u lated com pet i tive mar ket nego ti a -
tion. Orga ni za tional Behav ior and Human Deci sion Pro cesses, 67(1),
16-25.
Russo, J. E., & Schoemaker, P.J.H. (1989). Deci sion traps: The ten bar ri ers
to bril liant deci sion-mak ing and how to over come them. New York: Simon &
Schuster.
Strack, F. (1992). The dif fer ent routes to social judg ments: Expe ri en -
tial ver sus infor ma tional strat e gies. In L. L. Mar tin & A. Tesser
(Eds.), The con struc tion of social judg ment (pp. 249-275). Hillsdale,
NJ: Law rence Erlbaum.
Strack, F., & Hannover, B. (1996). Aware ness of influ ence as a pre con -
di tion for imple ment ing cor rec tional goals. In P. Gollwitzer & J. A.
Bargh (Eds.), The psy chol ogy of action: Linking cog ni tion and moti va tion
to behav ior (pp. 579-596). New York: Guilford.
Strack, F., & Mussweiler, T. (1997). Explaining the enig matic anchor -
ing effect: Mech a nisms of selec tive acces si bil ity. Jour nal of Per son al ity
and Social Psy chol ogy, 73, 437-446.
Trope, Y., & Liberman, A. (1996). Social hypoth e sis test ing: Cog ni tive
and moti va tional fac tors. In E. T. Hig gins & A. W. Kruglanski (Eds.),
Social psy chol ogy: Hand book of basic prin ci ples (pp. 239-270). New York:
Guilford.
Tversky, A., & Kahneman, D. (1974). Judg ment under uncer tainty:
Heuristics and biases. Sci ence, 185, 1124-1130.
Wegener, D. T., & Petty, R. E. (1997). The flex i ble cor rec tion model:
The role of naive the o ries of bias in bias cor rec tion. In M. Zanna
(Ed.), Advances in exper i men tal social psy chol ogy (Vol. 29). San Diego,
CA: Aca demic Press.
Wil son, T. D., & Brekke, N. (1994). Men tal con tam i na tion and men tal
cor rec tion: Unwanted influ ences on judg ments and eval u a tions.
Psy cho log i cal Bul le tin, 116, 117-142.
Wil son, T. D., Hous ton, C., Etling, K. M., & Brekke, N. (1996). A new
look at anchor ing effects: Basic anchor ing and its ante ced ents. Jour -
nal of Exper i men tal Psy chol ogy: Gen eral, 4, 387-402.
Wright, W. F., & Ander son, U. (1989). Effects of sit u a tion famil iar ity
and finan cial incen tives on use of the anchor ing and adjust ment
heuristics for prob a bil ity assess ment. Orga ni za tional Behav ior and
Human Deci sion Pro cesses, 44, 68-82.
Wyer, R. S., & Srull, T. K. (1989). Mem ory and cog ni tion in its social con text.
Hillsdale, NJ: Law rence Erlbaum.
Re ceived March 4, 1999
Re vi sion ac cepted June 17, 1999
1150 PER SON AL ITY AND SO CIAL PSY CHOL OGY BUL LE TIN
... When participants had low con dence in their response, it might be even di cult for them to estimate public opinion, resulting in producing estimates that was off the mark (e.g., the answer was 10 and an estimate was 90). In contrast, when they had a higher degree of con dence, it seemed to be relatively easy to produce a plausible estimate of public opinion 49,50 . For future study, we plan to examine this speculation directly: put simply, when participants estimate public opinions, we will ask them how con dent they feel in taking different perspectives. ...
Preprint
Full-text available
Researchers have shown that even an individual can produce the wisdom of the crowds, called ‘the wisdom of the inner crowd’. However, the previous methods leave room for improvements in terms of efficacy and convenience. This paper proposes a more efficient method with low cognitive cost, based on findings from cognitive and social psychology. The procedure is to ask participants to give two answers to the same question: first, their own estimate and, second, their estimate of public opinion. Experiments using this method showed that the averages of the two estimates were more accurate than the participants’ first estimates. That is, the wisdom of the inner crowd emerged. In addition, we found that the method could be superior to other methods in terms of efficacy and convenience. Moreover, we identified the conditions where our method worked better. We further clarify the limitations of using the wisdom of the inner crowd, including people’s tendency to fall into overconfidence. Overall, this paper proposes an effective and convenient method for harvesting the wisdom of the inner crowd.
... Although Neal et al. (2022) systematic review of forensic psychological literature revealed a scarcity of research on debiasing techniques, one such technique showed promise: the 'consider-the-opposite' strategy (see also Mussweiler et al., 2000). This technique involves experts asking themselves why their initial judgment might be wrong and considering possible alternatives, just like the alternative-scenario method. ...
Article
Full-text available
Legal psychologists’ assessments can have a major impact on the fact finder’s evaluation of evidence and, consequently, perceptions of guilt. Yet, in the few studies about legal psychologists’ assessments and reports, great variability was found. As is the case with other forensic expert domains, legal psychologists are prone to cognitive biases, such as being adversely affected by irrelevant contextual information, confirmation bias, and allegiance bias. Based on the scientific literature, we propose several ways in which legal psychologists can minimize cognitive biases in their assessments, most notably the alternative scenario method. Furthermore, we propose guidelines for expert witnesses in the legal psychological domain, designed to make reports as scientifically grounded, applicable, readable, transparent, and bias-free as possible. We hope that the guidelines will enhance the quality of expert witness testimony provided by legal psychologists around the world.
... The 'consider-the-opposite' strategy (COS) is a strategy that corrects one's tendency for confirmation bias by considering possibilities at odds with one's beliefs and perceptions (Lord, Lepper, & Preston, 1984). COS instruction has been shown to be effective in reducing confirmation bias (Adame, 2016;Hirt & Markman, 1995;Lord et al., 1984;Mussweiler, Strack, & Pfeiffer, 2000). In general, a powerful learning tool to enhance the learning effect of instruction is feedback (Hattie & Timperley, 2007). ...
... Rezaei argued that both methods produce greater (than actual) weights for the less important attributes and lower (than actual) weights for the more important attributes. Generalization of the findings of his study should be made carefully due to two important facts: (i) in his study, (2016) Consider-the-opposite Consider-the-opposite strategy mitigates the effect of anchoring bias (especially for external anchors) because of the multiple and inconsistent anchors being provided Lord et al. (1984) and Mussweiler et al. (2000) Group decision-making Group decision-making leads to a reduction in anchoring bias because of the different anchors considered by different DMs involved de Wilde et al. (2018), Meub and Proeger (2018), and Sniezek (1992) graphical representation information is used to check the anchoring bias in subjects' estimation. Existing literature shows that different graphical representation information could lead to different levels of precision in estimation, something that has not been considered in that study and has been extensively studied in the literature (Korhonen and Wallenius, 2008;Gettinger et al., 2013;Liu et al., 2014;Miettinen, 2014;Wachowicz et al., 2019). ...
Article
Full-text available
In this study, the existence of anchoring bias-people's tendency to rely on, evaluate, and decide based on the first piece of information they receive-is examined in two multi-attribute decision-making (MADM) methods, simple multi-attribute rating technique (SMART), and Swing. Data were collected from university students for a transportation mode selection. Data analysis revealed that the two methods, which have different starting points, display different degrees of anchoring bias. Statistical analyses of the weights obtained from the two methods show that, compared to Swing (with a high anchor), SMART (with a low anchor) produces lower weights for the least important attributes, while for the most important attributes, the opposite is true. Despite their differences in anchoring bias, analytical approaches supported by empirical studies suggest that both methods (SMART and Swing) overweigh the less important attributes and underweigh the more important attributes. As such, we examined whether the best-worst method (BWM), which has two opposite anchors in its procedure (a possible promising anchoring debiasing strategy), could produce results that are less prone to anchoring bias. Our findings show that the BWM is indeed able to produce lower weights (com-pared to SMART and Swing) for the less important attributes and higher weights for the more important attributes. This study shows the vulnerability of MADM methods with a single anchor and supports the idea that MADM methods with multiple (opposite) anchors, like BWM, are less prone to anchoring bias.
... Stemming from the concept of disfluency (Mills, 2020), reflection increases the cognitive load of decisionmaking, forcing participants to confront the assumptions of their pre-existing position, and at the same time open the door to considering alternates. Experimental studies have shown the effectiveness of reflection across different contexts such as managerial decision-making (Nagtegaal et al., 2020), price estimation of cars and probabilities of election outcomes (Mussweiler et al., 2000). Moser (2020) found that reflection does significantly reduce purchase impulsivity and intent, but she only tested it against anchoring on discounted pricing. ...
Article
Full-text available
Dark patterns – design interfaces or features that subtly manipulate people in making suboptimal decisions – are ubiquitous especially in e-commerce websites. Yet, there is little research on the effectiveness of dark patterns, and even lesser studies on testing interventions that can help mitigate their influence on consumers. To that end, we conducted two experiments. The first experiment tests the effectiveness of different dark patterns within a hypothetical single product online shopping context. Results show that, indeed, dark patterns increase the purchase impulsivity across all dark patterns, relative to the control. The second experiment tests the effectiveness of three behaviorally informed interventions on four different dark patterns also in a hypothetical online shopping scenario, but this time offering multiple products instead of a single product. Between-subject analysis shows that not all interventions are equally effective, with uneven impact across dark patterns. However, within-subject results indicate that all interventions significantly reduce purchase impulsivity pre- versus post-intervention, indicating that any intervention is better than none when it comes to combating dark patterns. We then end by discussing the policy implications of our results.
Article
Full-text available
When making consumption choices, people often fail to meet their own standards of both ethics and frugality. People also generally tend to demand more of others than they do of themselves. But little is known about how these different types of hypocrisy interact, particularly in relation to attitudes toward ethical consumption. In three experiments, we integrate research methods using anchoring and hypocrisy within the context of ethical consumption. Across three experiments, we find a default expectation that people (particularly people other than ourselves) should spend less on consumer items than they actually do. This default position can be inverted by making the ethical context of consumption salient, whereby the expectation is then that people (particularly other people) should spend more on consumer items than they actually do. Experiments 2 and 3 show that a moderate price anchor for ethical consumption is sufficient to shift expected standards for other people, but a higher price anchor is required to shift expected standards in personal behaviour. We discuss the countervailing roles of frugality and ethical consumption in understanding hypocrisy and ethical decision-making.
Article
Full-text available
While many studies have highlighted the existence of the anchoring effect in a wide variety of domains, no study to date has investigated its impact on memory. The present study aims to test whether an irrelevant numerical anchor not only influences an estimate but also modifies the memory of the associated event. Two experiments (total N = 259) were conducted, combining the methodology used by Loftus & Palmer (1974) and a classic anchoring paradigm. The results show that an irrelevant numerical anchor can modify the estimate of a car's speed and produce false memories of the event. We discuss the link between the processes underlying the anchoring effect and the false memory phenomenon.
Article
A healthy, immunocompetent South Asian man in his mid-20s, with a medical history of gastric ulcer, presented to Accident & Emergency with pleuritic chest pain, shortness of breath, fever, night sweats, weight loss, dry cough and asymptomatic iron deficiency anaemia. Following his initial assessment and investigations (chest X-ray, CT and blood tests), a diagnosis of miliary tuberculosis (TB) was made and empirical antimicrobial treatment started. However, subsequent microbiological testing, including urine, blood, induced sputum and lymph node sampling, was negative. Being interpreted as non-diagnostic, the antimicrobial therapy was continued. Following a clinical deterioration while on treatment, the patient's case was re-evaluated and further investigations, including a repeat CT and a liver biopsy, confirmed a diagnosis of stage IV (T1aN3bM1) gastric carcinoma. Our case highlights the diagnostic challenges in differentiating metastatic cancer from miliary TB. We also focus on possible cognitive biases that may have influenced the initial management decisions.
Article
The purpose of this study is to verify the effect of challenge and hindrance job demands on two-dimensional view of work-related subjective well-being and the role of employee’s dominant promotion-focus and LMX based on Russell's circumplex model(1980). A sample of 361 employees were participated voluntarily in this study and data were analyzed. The results and implications of this study are as follows. First, challenge stressor had a positive effect on the positive factors of work-related subjective well-being, while hindrance stressor had a negative effect. The effect of challenge stressor on work engagement was relatively large than the effect on job satisfaction. Second, both challenge and hindrance stressor had a positive effect on the negative factors of work-related subjective well-being. In comparison of the effect size considering activation level, the effect of challenge stressor on workaholic and hindrance stressor on burnout was relatively large compare to each other. Third, employee’s dominant promotion-focus showed the role of enhancing effect of challenge demands on the positive factors of work-related subjective well-being. The LMX has been shown to mitigate the effects of hindrance stressor on workaholic. Finally, theoretical and practical implications of these findings, along with this study limitations and future research directions were discussed.
Article
Full-text available
Results of four studies support the notion that anchoring effects are mediated by mechanisms of hypothesis-consistent testing and semantic priming. According to the suggested Selective Accessibility Model, judges use a hypothesis-consistent test strategy to solve a comparative anchoring task. Applying this strategy selectively increases the accessibility of anchor-consistent knowledge which is then used to generate the subsequent absolute judgment. Studies 1 and 2 demonstrate that absolute estimates depend on the hypothesis implied in the comparative task, suggesting that a hypothesis-testing strategy is used to solve this task. Study 3 shows that limiting the amount of knowledge generated for the comparative task retards absolute judgments. This suggests that knowledge rendered easily accessible in the comparative judgment is used for the subsequent absolute judgment. Finally, Study 4 suggests that self-generation of knowledge contributes to the robustness of the effect, thus resolving the seeming inconsistency that anchoring effects are at the same time remarkably robust and mediated by typically fragile semantic priming mechanisms.
Article
Many decisions are based on beliefs concerning the likelihood of uncertain events such as the outcome of an election, the guilt of a defendant, or the future value of the dollar. Occasionally, beliefs concerning uncertain events are expressed in numerical form as odds or subjective probabilities. In general, the heuristics are quite useful, but sometimes they lead to severe and systematic errors. The subjective assessment of probability resembles the subjective assessment of physical quantities such as distance or size. These judgments are all based on data of limited validity, which are processed according to heuristic rules. However, the reliance on this rule leads to systematic errors in the estimation of distance. This chapter describes three heuristics that are employed in making judgments under uncertainty. The first is representativeness, which is usually employed when people are asked to judge the probability that an object or event belongs to a class or event. The second is the availability of instances or scenarios, which is often employed when people are asked to assess the frequency of a class or the plausibility of a particular development, and the third is adjustment from an anchor, which is usually employed in numerical prediction when a relevant value is available.
Chapter
This chapter describes the role of naïve theories of bias in bias correction in the flexible correction model. The notion of bias correction was reviewed across a variety of research domains. Corrections are often the result of people consulting their naive theories of the influence of potentially biasing factors on their perception of the target. This view differs from competing views of bias correction because a view of corrections based on perceivers' naive theories of bias allows for a more flexible set of corrections than those proposed by other current models of bias removal. The chapter illustrates that flexible correction model (FCM) principles demonstrate the relevance of the perspective to a variety of research areas (including persuasion, attribution, impression formation, stereotyping, and mood). Finally, this chapter hopes that research and theory based on flexible correction notions will help to build a unifying framework within which correction processes in many areas of psychology can be investigated and explained.
Article
Most so-called “errors” in probabilistic reasoning are in fact not violations of probability theory. Examples of such “errors” include overconfidence bias, conjunction fallacy, and base-rate neglect. Researchers have relied on a very narrow normative view, and have ignored conceptual distinctions—e.g. single case versus relative frequency—fundamental to probability theory. By recognizing and using these distinctions, however, we can make apparently stable “errors” disappear, reappear, or even invert. I suggest what a reformed understanding of judgments under uncertainty might look like.