New findings on unconscious versus conscious thought in decision making: additional empirical data and meta-analysis

Judgment and Decision Making, Vol. 3, No. 4, April 2008, pp. 292–303
New ?ndings on unconscious versus conscious thought in decision
making: additional empirical data and meta-analysis.
Felix Acker?
School of Psychological Science
La Trobe University
Abstract
Ninety-eight Australian students participated in a functional replication of a study published by Dijksterhuis et al.
(2006). The results indicated that unconscious thought does not necessarily lead to better normative decision making
performance than conscious thought, which is contrary to the results found in Dijksterhuis et al. Since other studies
showed a positive, though statistically not signi?cant, effect for unconscious thought, a meta-analysis comprising a
total of 17 experiments was conducted. It suggests that there is little evidence for an advantage to normative decision
making using unconscious thought. However, a discussion of potential moderators shows that further study would help
to identify situations in which unconscious thought is truly helpful and those in which it is not.
Keywords: unconscious thought, meta-analysis, normative decision making.
1 Introduction
as lists (e.g., Newell, Wong, Cheung, & Rakow, submit-
ted), individually and randomized (e.g., this study), or in-
In a series of studies, Dijksterhuis and colleagues (e.g.,
dividually in a ?xed order (e.g., Phillips et al., 2007) for
Dijksterhuis, 2004b; Dijksterhuis, Bos, Nordgren, & van
a ?xed amount of time. Prior to the information presen-
Baaren, 2006; Dijksterhuis & Meurs, 2006) established
tation, participants are informed that they will have to a)
the surprising and counterintuitive ?nding that uncon-
choose one of the options (e.g., Dijksterhuis, 2004b, Ex-
scious thought leads to better decision making perfor-
periment 1) or b) rate each option (e.g., Lerouge, submit-
mance for complex problems than conscious thought.
ted). Then, following the presentation of the information,
Unconscious thought is de?ned as “cognitive and/ or
participants either immediately make a choice between
affective task-relevant processes that take place outside
options or think about their choice for a ?xed amount of
conscious awareness” (Dijksterhuis, 2004b, p. 586). A
time (except in Payne, Samper, Bettman, & Luce, 2007),
second postulate is that the reverse is true for simple de-
or are distracted for the same amount of time before mak-
cision: consciously thinking about them is better. This
ing their decision.
paper is concerned only with the ?rst.
The principal claim is that, when the number of fac-
A common denominator of all experiments described
tors that ought to be considered is high (as indicated by
by Dijksterhuis and colleagues is the experimental tech-
the number of attributes), and the decision is therefore
nique. Participants are split in two or three groups and
complex, unconscious thought will lead to better deci-
provided with a number of pieces of information about a
sion performance than conscious deliberation. Better de-
number of options to choose from, for instance four apart-
cision making (which I call “normative” here) is de?ned
ments or ?atmates. Each option is described by the same
as choosing, or ranking higher, options with more pos-
number of attributes, and usually these are attributes of
itive features. However, although most of Dijksterhuis
the choice option. For example, an attribute of all choice
and colleagues’ experiments found an effect in the hy-
options for “apartments” might be the size. Typically
pothesized direction, such effects often failed to reach
these attributes are conceptualized dichotomously — ei-
statistical signi?cance when comparing the unconscious
ther an apartment is spacious or it is not. All pieces of
and conscious group directly. For example, in the ?rst
information about all choice options are presented either
reported experiment in Dijksterhuis’ 2004 study the crit-
?I am indebted to Mary Omodei and Geoff Cumming (both La Trobe
ical comparison did not reach statistical signi?cance (p <
University), Ben Newell (UNSW) as well as one other anonymous
.08).
reviewer for helpful comments and critique. Address: Felix Acker,
School of Psychological Science, La Trobe University, 1 Kingsbury
In this context it is important to point out that in the se-
Drive, Melbourne, 3078. Email: f.acker@latrobe.edu.au
quence of experiments carried out by Dijksterhuis, only
292

---

Judgment and Decision Making, Vol. 3, No. 4, April 2008
Conscious versus unconscious thought
293
the ?rst two to be published (Dijksterhuis, 2004, Experi-
cally more robust estimate of the population effect size
ments 1 and 2) were exclusively directed at establishing
for the unconscious thought effect by meta-analytically
differences between decision success under unconscious
synthesizing data from a subset of studies on unconscious
versus conscious thought. The other experiments primar-
thought.
ily investigated further aspects of what was to become
the theory of unconscious thought (Dijksterhuis & Nord-
gren, 2006). For example, Experiments 1 and 2 of Di-
2 Empirical Study
jksterhuis, Bos, Nordgren, and van Baaren’s 2006 study
dealt with the deliberation without attention effect, which
2.1 Method
required two conditions: a complex and a simple one,
which were differentiated by the number of information
2.1.1 Participants
bits participants received. However, even though these
studies did not primarily contrast conscious with uncon-
All participants were third year psychology students that
scious thought, they frequently provided data supporting
took part in this experiment in the context of a tutorial se-
the notion that unconscious thought is a superior form of
ries on decision making. Participation in the experiment
integrating a large amount of information.
was voluntary and anonymous. The participation rate was
Conscious thought, according to Dijksterhuis, is very
90%, resulting in a total N = 98. There was a notable gen-
useful for monitoring information with respect to a partic-
der imbalance with females outnumbering males by 4:1.
ular criterion, for example the minimal amount of space
All students were naïve to the experimental hypothesis
that must be available in an apartment. It can engage in
and unfamiliar with the theory of unconscious thought.
logical operations and work with quantities in a precise
fashion, as is the case in mathematical operations. On the
2.1.2 Materials and procedure
other hand, it has limited capacity and is thus unsuitable
to integrate large amounts of information. Unconscious
The participants were tested in three groups of about 32
thought, on the other hand, is claimed to possess virtually
individuals. They were seated about 1.5 metres apart
in?nite processing capacity, to process information diver-
and were not permitted to talk during the experiment. In
gently and to employ a natural weighing mechanism for
line with the procedure described in Dijksterhuis, Bos,
acquired information. The alleged differences between
Nordgren, and van Baaren (2006) participants were pre-
the two forms of thought have been described by six prin-
sented with 48 sentences describing four ?ctitious cars
ciples and some additional characteristics in Dijksterhuis
(the Hatsdun, Kaiwa, Dasuka and Nabusi) on 12 at-
and Nordgren (2006), who re-stated the basic tenet that
tributes (e.g., milage, handling, service) in either negative
unconscious thought works well for complex decision
or positive terms. The sentences were randomized and
situations with multiple sources of relevant information.
each sentence was presented for ?ve seconds using Mi-
Despite its empirical successes and its advanced concep-
crosoft Powerpoint. The sentences were taken from the
tual development, the theory of unconscious thought has
supplementary material published in the Dijkersterhuis et
not been without criticism.
al. (2006) study. All sentences were examined by two
Shanks (2006), for example claimed that the uncon-
individuals to ensure that each sentence was intelligible
scious thought effects were an artefact of differential rates
to Australian subjects, and all were. Before the presen-
of forgetting rather than the product of different forms
tation, students were introduced to the names of the four
of processing. The theory was also criticised for lim-
cars and were told that they would have to pay attention
ited sample size in supporting studies, and concern was
to the material as they would later have to pass judgment
voiced about the applicability of the theory to the clini-
on each of the cars. Response sheets were handed face-
cal context (Bekker, 2006) for which, as Dijksterhuis and
down to all participants which sequentially assigned par-
colleagues pointed out, the theory was not intended. In
ticipants to groups c, u or i. After the presentation, each
addition to these criticisms, the currently published data
individual ?lled in the response form, following the di-
are fairly limited in that they almost exclusively include
rections printed on the form.
studies from one laboratory.
In the conscious thought group c participants were in-
The purpose of the present article is two-fold. The
structed to think for four minutes about the different cars
?rst aim is to provide replication data for the unconscious
and their attributes before rating them. Participants in
decision making effect with an English-speaking sample
group u (unconscious thought) had to work on a word
and English stimulus material. The study was also de-
search task for four minutes after which they ?lled in
signed to gather additional information to rule out poten-
their response sheet. Participants in group i (immediate)
tial alternative explanations for the unconscious thought
rated the cars right after the presentation ?nished. The
effect. Secondly, this article aims to provide a statisti-
ratings of the cars for all three groups had to be made on

---

Judgment and Decision Making, Vol. 3, No. 4, April 2008
Conscious versus unconscious thought
294
Conscious
overlap = .508
gap = ?.856
7
Unconscious
7
Immediate
6
6
gap = ?2
Average rating
5
5
4
4
H
N
H
N
H
N
Conscious
Unconscious
Immediate
Hatsdun
Kaiwa
Dasuka
Nabusi
Figure 1: Mean ratings of the four cars by group
Figure 2: 95% con?dence intervals for the difference
scores between Hatsdun and Nabusi for groups Con-
scious, Unconscious, and Immediate, with proportions
a 10-point rating scale, where 10 indicated the best possi-
overlapping or showing a difference. Proportion overlap
ble rating. The sequence within which the cars appeared
or gap is expressed in terms of the average of the half-
on the response sheet was also randomized separately for
widths of the two con?dence intervals.
each subject. Thus only a few individuals within each
group received the same rating sequence. Following their
to the predictions of the unconscious thought theory and
rating of the cars, participants were asked to indicate
the ?ndings in the original experiment by Dijksterhuis et
which two of the 12 presented attributes they considered
al. (2006). In fact, for group u the car with the second
to be the most important to themselves. The entire ex-
most positive attributes, the Kaiwa, was rated higher than
periment, from giving instructions to the collection of the
the supposedly better car, the Hatsdun. Looking at the
response sheets, took about 25 minutes per group.
group differences for the Hatsdun only, it becomes clear
that conscious thought was set apart from the other two
2.2 Results
groups. However, the overlap between c and u was still
substantial, mostly due to the con?dence interval width
Out of the 96 participants, 32 were assigned to group
for u (see Figure 3a). A similar result ensued when com-
c, 34 to group u and 30 to group i. The results clearly
paring the mean difference scores between the Hatsdun
demonstrated the main effect for cars. In all three groups,
and the Nabusi for each group. Again, the degree of vari-
the Hatsdun (best car based on number of positive fea-
ation was smallest for group c and largest for group u,
tures) was rated higher (mc = 6.93, sdc = 1.71; mu = 5.88,
but the overlap was still fairly substantial. Figure 3b il-
sdc = 2.66; mi = 6.16, sdi = 1.73) than the Nabusi (mc
lustrates this point.
= 4.35, sdc = 1.85; mu = 4.64, sdc = 2.08; mi = 4.19,
One potential explanation for the surprising effect
sdi = 1.95), which was the worst car based on number of
could be that the values for the Hatsdun were more ex-
positive features. Figure 1 illustrates this point. This dif-
treme for group c, whereas participants in group u gen-
ference was very pronounced for group c, somewhat less
erally were more careful in their ratings. For example,
clear for group i and least evident for group u (see Figure
participants in group c may have been more inclined to
2).
give a rating of 10 to their favourite car than participants
The biggest surprise in these results was that the differ-
in group u, who might have avoided the scale ends alto-
ences among the cars was smaller and the within group
gether. This hypothetical difference in scale usage would
variance was larger for the unconscious decision makers
have resulted in a clear distinction of the normatively
than for the conscious ones.1 This was exactly opposite
best car for group c and a very marginal distinction for
1The within-group variance is of interest because, according to the
That was precisely not the case as the scores for this group were highly
theory, we would expect a homogeneous improvement for all par-
variable. Some rated the Hatsdun very low and the Nabusi very high
ticipants following unconscious thought, and, especially, participants
and vice versa. It is this variability in the ratings for the different cars
should be better equipped to differentiate clearly between different
within each group that determines the width of the con?dence interval
choice levels (distinguish best from second best, from third best, etc.).
in Figure 3.

---

Judgment and Decision Making, Vol. 3, No. 4, April 2008
Conscious versus unconscious thought
295
A
B
7.5
3.5
overlap = .77
3.0
7.0
2.5
overlap = .63
overlap = .71
6.5
2.0
6.0
Average rating
1.5
Mean difference
1.0
5.5
0.5
5.0
Conscious
Unconscious
Immediate
Conscious
Unconscious
Immediate
Figure 3: (A) 95% con?dence interval showing the mean comparison for Hatsdun ratings between groups; (B) 95%
con?dence intervals showing the comparison of mean differences between H and N by group. Proportion overlap or
gap is expressed in terms of the average of the half-widths of the two con?dence intervals.
group u. If this explanation were valid, then a conver-
handling, were restricted to the less important categories.
sion of scores into ranks should offset the effect, and the
The important ones had the same valence for either car,
Hatsdun should have been rated as the best car most fre-
and, in the case of handling, it was the Hatsdun that had
quently in all three groups. Table 2 shows that this was
a positive valence for this feature, not the Kaiwa. This
not the case. Even after the conversion into ranks, group c
result implies that, regardless of whether a decision about
showed much clearer results than group u. This suggests
the better car was made on the basis of counting the num-
that individuals in the unconscious decision group were
ber of positive attributes (Hatsdun = 9, Kaiwa = 7) or also
less sure about which car is best and, accordingly, score
by the value that a person assigns to the attribute into con-
preferences ?uctuated.
sideration, there should have been no difference between
A second explanation for the results is that the impor-
the two cars or, if anything, a preference for the Hatsdun.
tance individuals placed on particular features of cars dif-
Another transformation illustrates the discrepancy
fered across groups and therefore the Kaiwa was rated
even further. For each of the attributes, a proportion based
higher than the Hatsdun in group u. Since the feature
on the within group count was calculated that showed its
valence for the Kaiwa and the Hatsdun were not identi-
importance relative to the other eleven. These propor-
cal — the Kaiwa, for example, was described as having
tions were then multiplied by the valence score for each
more leg space than the Hatsdun — this may have in?u-
attribute for each car (1 if the attribute was positive for
enced the results if leg space was an important feature
the car and 0 if it was negative). These scores were then
to many participants. Note that this explanation is some-
summed and averaged, using the number of positive at-
what at odds with the unconscious thought theory, which
tributes for the car, which yielded the mean importance
posits that not so much individual features as the overall
score for each positive attribute for each car. Comparing
attractiveness of a choice option would be considered by
these scores for each of the cars and groups (Table 4), it is
unconscious thought in the decision making process.
clear that the differences across groups were minute and
The data that were collected on the two most important
that, in fact, the cars with more positive attributes also
features for each participant allowed an educated guess
were described in positive terms for those categories that
about the validity of this explanation. As Table 3 illus-
were regarded as important by the participants. Again,
trates, participants in groups u, c, and i had fairly similar
the data do not provide any hints for the unexpected group
feature preferences. Differences ensued only for those
differences between c and u.
features that were only seldom selected as one of the two
A third potential explanation is that the sequence of
most important. Linking the results of Table 3 to Table
statements during the presentation may have had a vari-
1, which shows the positive and negative attributes for
able effect on the groups. This is unlikely since all groups
each car, it is clear that differences in attribute valence be-
saw the same sequence. With the data collected, however,
tween the Kaiwa and the Hatsdun, with the exception of
it is also not possible to disprove this hypothesis. An in-

---

Judgment and Decision Making, Vol. 3, No. 4, April 2008
Conscious versus unconscious thought
296
Table 1: Valence for the 12 attributes for each choice op-
Table 2: Rank averages for the four cars by group.
tion. A value of “1” implies positive, “0” implies negative
valence.
Hatsdun
Kaiwa
Dasuka
Nabusi
Attributes
Hatsdun
Kawai
Dasuka
Nabusi
Conscious
1.68
1.97
2.71
3.29
Unconscious
2.31
1.97
2.54
2.86
Environment
1
1
0
0
Immediate
2.00
1.93
2.60
3.10
Cupholders
1
0
1
1
Many colours
1
1
0
1
Sound system
0
0
1
0
high methodological homogeneity and largely re?ect the
Service
1
1
0
0
method used in the empirical study described here. The
Handling
1
0
1
0
focus of the meta-analysis was restricted to those stud-
ies that directly compared unconscious and conscious
Milage
1
1
0
0
thought with respect to a decision making task after the
Leg Space
0
1
0
1
presentation of a large amount of stimulus material for
Trunk size
1
1
0
0
different choice alternatives.
Sunroof
1
0
1
1
Studies on incubation, which arguably also deal with
Gear shifting
0
1
1
0
unconscious thought, were not included. Incubation stud-
ies usually present the participant with a problem solv-
Age
1
0
1
0
ing task (e.g., Vul & Pashler, 2007) or judge the creative
Sum
9
7
6
4
output after a period of incubation (e.g., Dijksterhuis &
Meurs, 2006). Both of these approaches are clearly dif-
ferent from the methodological selection criterion out-
spection of the items showed only one permutation that
lined and thus do not qualify for the meta-analysis.
could have lead to a primacy or recency effect. Out of the
Other studies that deal more directly with unconscious
last eight items, four were negative statements about the
thought but also were not included are those that oper-
Nabusi, yet only one of these four related to an important
ationalise good decision making as post-choice satisfac-
attribute.
tion (e.g., Dijksterhuis & van Olden, 2006). Judging the
In summary, the experimental manipulation for the
degree of content with an item after a period of time does
cars was successful, in that the cars with more features
not appear to be functionally equivalent to making a rela-
were rated as the better ones, and some of these cars’
tively instantaneous decision based on comparative judg-
features were also considered the most important, such
ment. Thus, including studies which use post-choice sat-
as their environmental performance. The two cars with
isfaction as the dependent variable would have introduced
less positive features overall additionally had less desir-
further error variance into the analysis beyond the normal
able features, such as cupholders or trunk space. The
sampling variability and ultimately biased the overall ef-
clearest difference between the cars was obtained for the
fect size estimate.
conscious thinkers, the least distinction was achieved by
In order to be included a study had to focus on nor-
the unconscious thinkers. This result was not statisti-
mative decision making, compared unconscious and con-
cally conclusive due to the high variability within group
scious thought conditions, operationalise unconscious
u. Artefactual explanations such as primacy and recency
thought as a distraction period following a standardized
effects or different preferences for features and hence cars
encoding period, present each piece of information for
by the various groups were rejected on the basis of the
the same amount of time or at least make provisions for
data.
the participants to do so, make participants choose from
or evaluate multiple, speci?ed choice options, and ?nally
operationalise choice options in terms of degree of good
3 Meta analysis
decision making rather than choosing the correct out of a
number of incorrect options.
3.1 Method
Studies were sourced using the psychinfo database
(1806-present) and Google Scholar with the keywords
3.1.1 Study selection
“unconscious thought”, “decision making”, and “incu-
The purpose of the meta-analysis was to compile all
bation” and results were narrowed down with combina-
relevant study data that bear on the bene?t of uncon-
tions of these. Additionally, references cited in Dijkster-
scious thought for normative decision making.
The
huis (2004b), Dijksterhuis et al. (2006), and Dijksterhuis
studies carried out by Dijksterhuis and colleagues show
et al. (in press) were checked for compliance with the

---

Judgment and Decision Making, Vol. 3, No. 4, April 2008
Conscious versus unconscious thought
297
Table 3: Distribution of feature importance rating across groups. Distributions are fairly similar for all three groups
and clearly distinguishes between important and unimportant features.
Group
EN
CU
CO
SOU
SE
HA
MI
LS
TR
SF
GE
A
C
12
1
4
3
9
11
12
2
0
3
3
3
U
10
0
4
2
12
11
9
6
1
3
1
7
I
15
2
3
1
8
11
8
2
0
1
5
3
Note. EN=Environment, CU = Cupholders, CO = Colours, SOU = Sound system,
SE = Service, HA = Handling, MI = Milage, LS = Leg space, TR = Trunk space,
SF = Sunroof, GE = Gears, A = Age. The numbers show how often each feature
was named as one of the two most important ones.
by Lerouge (submitted).
Table 4: Mean feature importance for each car by groups
and overall.
Experiment 2 of the Dijksterhuis et al. (2006) was the
most comparable to the present empirical study. The
Overall
C
U
I
main points of distinction were the difference in depen-
Hatsdun
0.096
0.097
0.096
0.096
dent variable scaling (50 point visual analogue versus 10
point rating) and the presentation time per item (8 versus
Kaiwa
0.096
0.095
0.093
0.099
5 seconds). Experiment 1 (Dijksterhuis, Bos, Nordgren,
Dasuka
0.063
0.063
0.061
0.065
& van Baaren, 2006) looked at proportional differences
Nabusi
0.041
0.040
0.049
0.034
between participants in the unconscious and conscious
thinking groups with respect to selecting the best choice
alternative. In Dijsterhuis’ 2004 study, Experiments 1
selection criterion. Finally, studies citing either Dijkster-
and 2 were fairly similar to the present experiment, albeit
huis (2004b) or Dijksterhuis et al. (2006) were checked.
the stimulus material was different. Experiment 3 (Dijk-
Two sets of data were identi?ed to be of potential use-
sterhuis, 2004b) was functionally similar to Experiment 2
fulness to this analysis but could not be obtained. These
but with yet again other stimulus materials. It also looked
were one data set pertaining to decision quality after
at the importance an individual places on the different in-
varying intervals of unconscious thought (Dijksterhuis,
formation attributes that are presented and correlated this
2004a, as cited in Dijksterhuis and Nordgren, 2006) and a
with the decision score. The present experiment had a
study that used the same stimulus material as the present
similar index, however, in a more rudimentary form.
study (Dijksterhuis, Bos, van Baaren, & van der Leij, in
Experiment 4 (Dijksterhuis, Bos, Nordgren, & van
prep., as cited in Dijksterhuis et al., in press). However,
Baaren, 2006) was not directly concerned with alternative
instead of a contrast between unconscious thought and
selection but rather with the re-attribution of item content
conscious thought, it focuses on the comparison between
to its source. In comparison to the other experiments, the
immediate decision making and unconscious thought.
encoded items did not have to be integrated into global
Overall, only two published studies comprising six ex-
judgments but rather had to be remembered individually.
periments were deemed to suf?ciently ful?ll the original
The experiment also fostered response speed as a second
selection criteria. These experiments were number 1 to
dependent variable, but these results were not included
4 in Dijksterhuis (2004b), Experiments 1 and 2 in Di-
here. The decision to include this experiment is debat-
jksterhuis et al. (2006). All other studies cited in this
able, but it seemed suf?ciently suitable as the attribution
meta-analysis have not yet been published and had ei-
still involved some decision making, although on an item-
ther been cited in work by Dijksterhuis or came to my
by-item level, after the same encoding procedure as in the
attention through word of mouth. This class of unpub-
other experiments.
lished works included ?ve additional studies incorporat-
Ham et al.’s experiments (Ham, Bos, & Doorn, submit-
ing 10 sets of relevant data. These data were extracted
ted) were again functionally quite similar to the present
from Experiments 1 and 2 of the Ham, Bos and Doorn
experiment but focused on justice judgments rather than
study (Ham, Bos, & Doorn, submitted), Experiments 1, 2
consumer choice. Newell et al. (submitted) presented
and 3 of Newell et al.’s (submitted) study, Experiments 1
four experiments. The ?rst three experiments were in-
and 2 reported by Payne et al. (2007), aggregate ?ndings
cluded in the meta-analysis, the fourth one was not. In
reported by Phillips et al. (2007), as well as from a study
all experiments Newell et al. collected data on a variety

---

Judgment and Decision Making, Vol. 3, No. 4, April 2008
Conscious versus unconscious thought
298
of dependent measures, such as a recall test for attributes,
investigation. A variety of decision processes and infor-
but only the data on choice preferences were used for the
mation conditions is represented. Decision making suc-
meta-analysis. Experiment 3 is virtually identical with
cess is identi?ed as choosing the best option and compar-
the empirical study presented here in that it is also a di-
ing the proportions across groups or by rating each choice
rect replication of Dijksterhuis et al. (2006) using the
option and then deriving some form of mean difference
same materials. In contrast, Experiments 1 and 2 also
between groups. Ham et al.’s study presents an exception
used the same normative choice methodology but with
to these two patterns. Regarding the information presen-
original stimulus material. In Experiment 2 an additional
tation, most experiments presented the items one by one,
experimental group is included: conscious thinkers that
but few presented multiple pieces of information simulta-
have access to the relevant information during the deci-
neously. Most studies included three conditions: imme-
sion making period. Data from this group were not in-
diate, conscious and unconscious decision making, but
cluded in the meta-analysis but based on the results re-
some (Newell et al., submitted; Payne et al., 2007) had
ported by Newell et al., their inclusion would not have
an additional condition to test speci?c predictions. The
much difference. Experiment 4 investigated the impact
gender ratio also varied, although females predominated
of primacy and recency effect on conscious and uncon-
in most studies. There were minor variations with respect
scious choices. It was unsuitable for inclusion to the
to the number of attributes for each choice option or the
meta-analysis as it only contained two choice alternatives
interval length after the presentation of the stimulus ma-
but mainly because the two cars were both described by
terial (with the exception of Newell et al., Experiment 2).
10 positive attributes so that there was no normatively ‘ra-
tional’ choice.
3.1.2 Meta analytic procedure
Payne et al. (2007) carried out two experiments that
explored the boundary conditions of the unconscious
For each of the selected experiments, standardized ef-
thought effect. In addition to the contrast between usual
fect sizes (g) were calculated following the guidelines
conscious thought and unconscious thought, they in-
presented in Grissom and Kim (2005) for mean differ-
cluded a further condition where the decision interval for
ences and DeCoster (2004) for proportion differences. In
conscious thought was self-paced. Only the conscious
line with recommendations of Schmidt, Oh and Hayes (in
thought condition with a ?xed time interval between pre-
press) a random effects model was chosen for the analy-
sentation and decision was included in the meta-analysis.
sis. Weights and the mean effect size were calculated us-
Both of Payne et al.’s experiments used a design whereby
ing the procedure described in Borenstein, Hedges, Hig-
different numerical values were assigned to each piece of
gins, & Rothstein (2008). Gender ratio, presentation time
information and which resulted in different expected util-
per item, and decision (or better distraction) interval span
ities for each choice option. Contrary to other studies,
were de?ned as moderators. Since the moderator vari-
this method required successful and precise analytic inte-
able investigated in Lerouge (submitted) had an appre-
gration of the values for each option, instead of choices
ciable effect on the results, the experiment was treated as
based on the gist of all items.
two separate data sets for the meta-analysis in order to
The data set obtained from Phillips et al.’s study is part
preserve this noteworthy contrast. This decision did not
of a larger online study. Phillips et al. looked at, among
affect the ?nal estimates of the population effect size and
other things, the effect of item presentation order, but the
margin of error. The data as presented below overesti-
data presented here has been collapsed across all presen-
mate the parameters by about 0.006, in contrast to esti-
tation order positions.
mates that treat Lerouge’s study as a single data set.
Finally Lerouge’s study (submitted) dealt explicitly
with consumer choice but primarily investigated the im-
pact of con?gural versus featural processing as a mod-
3.2 Results
erator of unconscious thought. Only Experiment 2 was
3.2.1 Unconscious versus conscious thought
included, since Experiment 1 generated data only for im-
mediate and unconscious thought, but not for conscious
Altogether 17 experiments were included in the analysis
thought. All experiments listed here aside from Dijk-
with a combined participant number of 888. The mean
sterhuis et al. (2006) also included the immediate choice
effect size was g = .251, and the range from g = (-.483)
condition, which is of subordinate interest in the present
in Payne et al.’s study (Experiment 2) to 1.25 (Dijkster-
study.
huis, Bos, Nordgren, & van Baaren, 2006, Experiment 2).
The selection of studies was homogeneous in terms of
Figure 4 shows a forest plot of effect sizes with respec-
general methodological approach, albeit with much varia-
tive con?dence intervals by study and Table 6 provides
tion in the exact procedure. Table 5 provides an overview
the numerical effect size values, the standard error and
of some key aspects that lend themselves to meta-analytic
the relative weights.

---

Judgment and Decision Making, Vol. 3, No. 4, April 2008
Conscious versus unconscious thought
299
Table 5: Overview of key features of the experiments included in the meta-analysis.
Expt.1 Groups2 Material
Gender
n
Judg-
ES4
Number Presenta- Filler task
Presen- Inter-
type
ratio
ment
of at-
tion
tation
val
(M/F)
type3
tributes
time
(min)
(sec)5
1
c,u,i
Cars
0.250
32
1
1
12
Random
word search
5
4
2
c,u
Cars
?
20
2
2
12
Random
anagrams
8
4
3
c,u
Cars
?
13
1
1
12
Random
anagrams
8
4
4
c,u,i
Apartments 0.312
21
1
1
12
Random
n-back task
4
3
5
c,u,i
Apartments 0.175
31
2
2
15
Fixed
n-back task
12
3
6
c,u,i
Person
0.355
48
1
1
12
Random
anagrams
2
4
7
c,u,i
Person
0.295
38
3
3
N/A
Random
anagrams
2
4
8
c,u,i
Person
0.250
18
3
3
14
Fixed
n-back task
26
3
9
c,u,i
Person
0.500
17
3
3
14
Fixed
n-back task
26
3
10
c,u,i
Notebooks
?
21
1
1
12
Fixed
anagrams
20
4
11
c,u,i
Notebooks
?
21
1
1
12
Fixed
anagrams
20
4
12
c,u,i
Apartments 0.610
23
2
2
10
Random
anagrams
4
4
13
c,u,i,(c+) Apartments 0.415
23
2
2
10
Fixed
anagrams
180
8
14
c,u,i
Cars
0.607
30
2
2
12
Random
anagrams
4
4
15
c,u,(c+) Lottery
1.158
20
2
2
12
Random
anagrams
6
4
16
c,u,i,(c+) Lottery
1.947
28
2
2
12
Random
anagrams
6
4
17
c,u
Cars
0.428
40
1
1
12
Collapsed anagrams
8
4
1 The sequence for this table is the same as in Figure 4 and Table 6.
2 The combination (c+) indicates that there was a special condition involving conscious thought. Data from
these conditions were not used here.
3 Value “1” indicates use of rating scales for all options, value “2” stands for selection of a particular (usually
best) alternative, value “3” are other, generic measures.
4 ES stands for effect size estimator. Value “1” indicates difference between highest and lowest rated option, “2”
indicates difference between percentages of correct choice in groups, and “3” indicates other, generic indices.
5 Values >10 indicate that all information for a given option was presented simultaneously, the value >100
indicates that all information for all options were presented simultaneously.
The amount of variability between the effect sizes was
3.2.2 Moderator variables
substantial (Q[df =16] = 54.994, p ? .000; I2= 70.906).
Given the high level of study heterogeneity, several meta-
Only ?ve out of the included 17 experiments returned re-
regressions were carried out to investigate the effect of
sults that can be described as “statistically signi?cant” in
classical terms. Each of these ?ve provided evidence for
potential moderator variables. The statistical package
Comprehensive Meta AnalysisTM (Borenstein, Hedges,
the superiority of unconscious thought. They also had
Higgins, & Rothstein, 2008) was used for this purpose.
the largest effect sizes but at the same time the smallest
sample sizes. Newer data that still await publication pro-
vided evidence con?icting with the unconscious thought
Gender ratio: Dijksterhuis (2004), in his discussion of
theory. The aggregate estimate shows a modest bene?t
Experiment 1, found an interaction between gender and
for unconscious thought, although, from a signi?cance-
thought condition. Males were choosing particularly well
testing perspective, the con?dence interval includes “0”
following unconscious thought. Other studies have not
and can thus be interpreted as non-signi?cant support.
speci?cally investigated this point, but it was worthwhile
to follow up with a large set of data here. The vast ma-

---

Judgment and Decision Making, Vol. 3, No. 4, April 2008
Conscious versus unconscious thought
300
This study
Table 6: Effect sizes (g), standard errors (SE) and rela-
Dijksterhuis (2004), Exp 1
tive weights (w) for the experiments included in the meta-
Dijksterhuis (2004), Exp 2
analysis. The abbreviation sfp means “submitted for pub-
Dijksterhuis (2004), Exp 3
Dijksterhuis (2004), Exp 4
lication.”
Dijksterhuis et al. (2006), Exp 1
Study name
g
SE
w
Dijksterhuis et al. (2006), Exp 2
Ham et al. (sfp), Exp 2
This study
0.471
0.243
6.743
Ham et al. (sfp), Exp1
Dijksterhuis (2004), Exp 1
0.434
0.306
5.998
Lerouge (sfp), Exp 2 configural
Dijksterhuis (2004), Exp 2
0.242
0.277
6.340
Lerouge (sfp), Exp 2, featural
Newell et al. (sfp), Exp 1
Dijksterhuis (2004), Exp 3
0.241
0.203
7.205
Newell et al. (sfp), Exp 2
Dijksterhuis (2004), Exp 4
0.065
0.267
6.459
Newell et al. (sfp), Exp 3
Dijksterhuis et al. (2006), Exp 1 0.968
0.390
5.054
Payne et al. (2007), Exp 1
Payne et al. (2007), Exp 2
Dijksterhuis et al. (2006), Exp 2 1.247
0.417
4.774
Phillips et al. (2007)
Ham et al. (sfp), Exp 2
0.883
0.352
5.469
Overall
Ham et al. (sfp), Exp 1
1.055
0.349
5.503
?1.0
?0.5
0.0
0.5
1.0
1.5
2.0
Lerouge (sfp), Exp 2 con?gural 1.116
0.326
5.765
Lerouge (sfp), Exp 2, featural ?0.064
0.303
6.033
Figure 4: Forest plot of studies displaying effect sizes and
Newell et al. (sfp), Exp 1
0.171
0.336
5.650
95% con?dence intervals.
Newell et al. (sfp), Exp 2
?0.504
0.381
5.150
Newell et al. (sfp), Exp 3
?0.367
0.285
6.245
jority of experiments exhibited a surplus of females, with
the exception of Payne et al. (2007). The experiments by
Payne et al. (2007), Exp 1
0.722
0.393
5.025
Lerouge (submitted) and Dijksterhuis et al. (2006) were
Payne et al. (2007), Exp 2
?0.483
0.340
5.604
not included as no data on gender were available. The
Phillips et al. (2007)
?0.251
0.222
6.984
regression analysis suggested that the gender ratio of a
study is a poor predictor of effect size (? = ?0.214, CI
Overall
0.251
0.137
95
[?0.786, 0.357], SE = 0.291).
Presentation format: A last moderator variable was
Item presentation duration: The analysis of item pre-
the effect of presenting all pieces of information either in-
sentation duration focused only on those studies that
dividually or in clusters (as lists for each choice option or
showed each piece of information individually. Stud-
all simultaneously). This information was not available
ies (Dijksterhuis, 2004, Experiment 4; Ham et al., sub-
for Phillips et al.’s data, which was therefore excluded
mitted, Experiments 1 and 2, Lerouge, submitted, both
from the analysis. The results showed that the aggregate
conditions; Newell et al., submitted, Experiment 2) that
effect size was lower (g = 0.147, CI
showed items list-wise or all simultaneously were ex-
95 [?0.037, 0.331],
SE = 0.094) for experiments that presented the items in-
cluded. The analysis showed a trend that longer pre-
dividually (n = 9) than for the overall estimate. On the
sentation times per item led to less advantage for uncon-
other hand, the results for studies that presented multiple
scious thought, but this relationship was slight and did not
pieces of information at the same time (n = 6) suggested
reach statistical signi?cance (? = ?0.095, CI95 [?0.232,
a higher and most likely positive effect (g = 0.369, CI
0.042], SE = 0.07).
95
[0.110, 0.627], SE = 0.132). This means that unconscious
thought may actually be helpful when much information
Thought interval: All studies were included in this
is presented simultaneously, but not when bits of infor-
meta-regression. Similar to the item presentation vari-
mation are presented individually.
able, the results suggested that a longer interval between
information presentation and decision is favourable for
3.2.3 Unconscious thought versus immediate deci-
conscious thought rather than unconscious thought.
sion making
Again, though, the result did not reach statistical signif-
icance (? = ?0.199, CI95 [?0.445, 0.048], SE = 0.126)
Altogether 13 out of the 17 data sets were included in
and was strongly in?uence by Newell et al.’s (submitted)
a meta-analysis comparing decision making after uncon-
second experiment, which had a substantially longer in-
scious thought with immediate decision making. The ?ve
terval than all other studies.
excluded data sets did not feature the immediate decision

---

Judgment and Decision Making, Vol. 3, No. 4, April 2008
Conscious versus unconscious thought
301
conditions. These were Dijksterhuis et al. (2006, Exper-
experiment and the original study is the distractor task
iments 1 and 2), Payne et al. (2007, Experiment 1) and
used. Dijksterhuis et al. mainly used anagrams to keep
Phillips et al. (2007). The results were similar to the
participants in the unconscious thought condition busy; a
comparison of unconscious and conscious thought. There
word search task was used here. Anagrams have been
was a modest trend in favour of unconscious thought (g
widely used in the study of incubation (see for exam-
= 0.189, CI95 [?0.05, 0.428]) but a signi?cant amount
ple Vul & Pashler, 2007) and seem to be positively re-
of heterogeneity across studies (Q[df =12] = 26.691, p =
lated to the kind of processes activated during uncon-
.009; I2= 55.041). Further analyses to identify true mod-
scious thought. Hence, anagram solving might have had
erator variables were not carried out.
a positive mediator function. Word search, on the other
hand, does not have much in common with unconscious
3.2.4 Conscious thought versus immediate decision
thought, as most people scan the array of letters system-
making
atically for the correct combinations. Hence it is more
akin to conscious thought. However, in some studies Di-
The same 13 data sets were used for the con-
jksterhuis (2004b) and also Ham (Ham, Bos, & Doorn,
scious thought versus immediate decision making meta-
submitted) used the n-back task (Kane, Conway, Miura,
analysis. The results showed only a very slight advantage
& Col?esh, 2007). This task puts high demand on ex-
for conscious thought over immediate decision making
ecutive functioning and also can hardly be conceived as
(g = 0.084, CI95 [?0.72, 0.24]) with substantially more
supporting unconscious thought. Nevertheless, the pos-
agreement among experiments than for the other compar-
sibility that the type of distractor tasks affects decision
isons (Q[df =12] = 9.77, p = .636; I2= 0).
making ef?ciency under unconscious thought conditions
warrants further study and could be used as a moderator
4 Discussion
variable in future meta-analyses when more experiments
with distractor tasks other than anagrams are available. If
all these three alternative explanations can be discounted,
4.1 Empirical study
then the results of the present study provide strong evi-
The results of the present experiment are noticeably at
dence that the true effect size for unconscious thought is
odds with the theory of unconscious thought. According
much smaller than assumed so far or that this particular
to the obtained data set, it is a better idea to consciously
experimental approach is not very suitable to demonstrate
think about different choice alternatives in order to ar-
the unconscious thought effect reliably. Further support
rive at the best choice than letting the unconscious do
for either of these two conclusions comes from the meta-
the work. Not only, it seems, does conscious thinking
analytic ?ndings presented here.
lead to the better identi?cation of the top choice, but it
also allows to differentiate between gradual choice op-
4.2 Meta-analysis
tions. With unconscious thought the choice is a muddy
one; it did not consistently help individuals to differen-
The statistical synthesis of all available data provides
tiate between choice alternatives. Unconscious thought
at best suggestive evidence in favour of unconscious
failed to allow a clear distinction of the cars; contrary to
thought, but, on the basis of 888 subjects tested under
the expectations conscious thought did do just that.
similar conditions, there is no convincing statistical evi-
While some alternative explanations for the diver-
dence. The true effect in the population may be anything
gence from the expected results were investigated (re-
between a moderate bene?t after unconscious thought
cency effects, weighing differences, scale usage differ-
to a slight advantage following conscious thought. One
ences), other factors in which the present study differed
sign for caution is that the experiments with fewer par-
from Dijksterhuis et al. (2006) may have been responsible
ticipants consistently generated substantially larger effect
for the reversal of results. An obvious difference between
sizes than the larger studies.
this and the Dijksterhuis study was that this experiment
was carried out with English material and Australian stu-
4.3 Moderator variables
dents. It is dif?cult to imagine, however, how cultural
or linguistic variation could have had such a radical ef-
Four moderator variables were investigated in the present
fect on decision making quality. Similarly, delivering the
meta-analysis. Of these, only the presentation format
information to a group rather than on an individual ba-
as either single item or list-wise, did help to explain
sis is unlikely to have had any impact, especially since
the variance between studies, which is substantial. On
participants were engaged in the task and did not distract
the other hand, gender-ratio, presentation time per item
one another during the task presentation or the thought
and thought interval length were very weak predictors
interval. One other noteworthy difference between this
of effect. This weakness, however, may have partially

---