A fine-grained analysis of the jumping-to-conclusions bias in schizophrenia: Data-gathering, response confidence, and information integration

Judgment and Decision Making, Vol. 4, No. 7, December 2009, pp. 587–600
A ?ne-grained analysis of the jumping-to-conclusions bias in
schizophrenia: Data-gathering, response con?dence, and
information integration
Andreas Glöckner?
Max Planck Institute for Research on Collective Goods, Bonn
Steffen Moritz
University Hospital Hamburg-Eppendorf
Abstract
Impaired decision behavior has been repeatedly observed in schizophrenia patients. We investigated several cognitive
mechanisms that might contribute to the jumping-to-conclusions bias (JTC) seen in schizophrenia patients: biases in
information-gathering, information weighting and integration, and overcon?dence, using the process tracing paradigm
Mouselab. Mouselab allows for an in-depth exploration of various decision-making processes in a structured information
environment. A total of 37 schizophrenia patients and 30 healthy controls participated in the experiment. Although
showing less focused and systematic information search, schizophrenia patients practically considered all pieces of
information and showed no JTC in the sense of collecting less pieces of evidence. Choices of patients and controls both
approximated a rational solution quite well, but patients showed more extreme con?dence ratings. Both groups mainly
used weighted additive decision strategies for information integration and only a small proportion relied on simple
heuristics. Under high stress induced by affective valence plus time pressure, however, schizophrenia patients switched
to equal weighting strategies: less valid cues and more valid ones were weighted equally.
Keywords: decision making, schizophrenia, jumping to conclusions, heuristics.
1 Introduction
noid schizophrenia patients (Moritz, Woodward, & Haus-
mann, 2006; Peters & Garety, 2006). Recently, JTC has
Hasty decision-making is a hallmark feature of presently
been found to correlate with delusion conviction (Garety
deluded schizophrenia patients. A look of a stranger,
et al., 2005). A number of researchers ascribe JTC a fun-
sounds in the telephone line and certain initials on num-
damental role in the pathogenesis of delusions, that is,
ber plates are mistaken as proof of a conspiracy or
?xed false beliefs (for reviews see Bell, Halligan, & El-
surveillance. Cognitive studies have asserted that this
lis, 2006; van der Gaag, 2006).
so-called jumping-to-conclusions bias (JTC) in para-
Traditionally, JTC has been investigated with the beads
noid schizophrenia is not con?ned to idiosyncratic and
or probabilistic reasoning task: The subject is consecu-
delusions-related scenarios but extends to neutral situa-
tively presented a sequence of beads drawn either from
tions (Garety, Hemsley, & Wessely, 1991; Huq, Garety,
a jar that predominantly contains beads, for example in
& Hemsley, 1988; Moritz & Woodward, 2005; Moritz,
green, or a jar that predominantly contains beads in red
Woodward, & Lambert, 2007). While JTC is somewhat
(Huq et al., 1988). The chain of events usually strongly
aggravated among schizophrenia (Moritz & Woodward,
favours one of the jars. Compared to both healthy and
2005; Startup, Freeman, & Garety, 2008; Van Dael et
psychiatric controls, schizophrenia patients make early,
al., 2006) and sometimes also non-schizophrenia patients
premature and incautious decisions in 40–70% of the
with acute persecutory delusions (Corcoran et al., 2008),
cases (i.e., they decide after only one bead has been
other studies have found this bias also in remitted para-
drawn). When presented with the entire available infor-
mation all at once, group differences are abolished. In
?Both authors have equally contributed to the manuscript and
addition, probability ratings are usually not discrepant,
split ?rst authorship.
We thank Tanja Ostermann, Jan Multmeier
indicating that patients have a data-gathering bias rather
and Christoph Engel for helpful comments on earlier drafts of this
than de?cits with probabilistic reasoning. (For an older
manuscript. Address: Andreas Glöckner, Max Planck Institute for Re-
search on Collective Goods, Kurt Schumacher Str. 10, D-53113 Bonn,
but still relevant review on this topic, see Garety & Free-
Germany. Email: gloeckner@coll.mpg.de.
man, 1999.)
587

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Analysis of the jumping-to-conclusions bias
588
Our group has con?rmed this bias, ruling out de?cits
is pre-determined. Moreover, the beads task estimates
in memory and poor motivation as confounding factors
JTC on the basis of a single item, reducing its reliability.
(Moritz & Woodward, 2005). Others have found that
A ?nal aim was to investigate the impact of stress ex-
this bias is not a result of impulsivity (Dudley, John,
erted by time-pressure and emotionally framed scenarios
Young, & Over, 1997). Using an experimental vari-
within-subjects. On the basis of the available literature
ant of the “Who wants to be a millionaire” quiz, pa-
we expected that patients with schizophrenia (Szs) col-
tients with schizophrenia, irrespective of current delu-
lect less (H1) and particularly less valid (H2) informa-
sional ideation, displayed a lowered decision-threshold,
tion, show a less systematic information search inspect-
that is, they over-interpreted the available amount of ev-
ing less valid information ?rst (H3), are over-con?dent in
idence (Moritz, Woodward, & Hausmann, 2006). The
their judgments (H4), and that these biases might be more
precise nature of JTC is not entirely understood and un-
pronounced under conditions of stress induced by time-
der some circumstances (enhanced ambiguity and multi-
pressure or affective framing of the task (H5) compared
ple response options) the bias may even be diminished or
to controls (CPs). Following an exploratory account, we
abolished (Moritz, Woodward, & Lambert, 2007).
investigated whether there are differences in choice accu-
JTC can be conceptualized in different ways. The
racy, whether subjects particularly rely on a take the best
core contribution of this paper is to investigate JTC as a
strategy and if there is a relation between schizophrenia
data-gathering bias (less information is taken into account
severity measures (i.e., PANSS; see below) and the dif-
for decision-making relative to controls) and/or over-
ferent aspects of JTC biases.
con?dence (the predictive value of information is over-
interpreted relative to controls) and/or suboptimal infor-
1.1 Methodological preliminaries
mation weighting and integration (the validity of cues
is not considered appropriately or heuristics are more
The aforementioned hypotheses for decision-making in
strongly preferred relative to controls). Our group has
schizophrenia were investigated using emotional and
recently investigated the second aspect, and we have re-
neutral probabilistic inference tasks that have been re-
peatedly found that patients with schizophrenia are over-
peatedly investigated in recent research on heuristics, that
con?dent in erroneous decisions (for a review see Moritz
is, simple short-cut decision strategies (Bröder & Gaiss-
& Woodward, 2006), which so far has been mainly inves-
maier, 2007; Bröder & Schiffer, 2006; Gigerenzer &
tigated in the context of memory tasks (for independent
Goldstein, 1996; Glöckner, 2006; Newell et al., 2003).1
replications see Kircher, Koch, Stottmeister, & Durst,
We used the standard process tracing paradigm of behav-
2007; Laws & Bhatt, 2005). We have also found that
ioral decision research: Mouselab (Payne, Bettman, &
patients tend to reach more incautious decisions when
Johnson, 1988).2 In Mouselab, different cues (i.e., pre-
asked to deduce the correct title of classical paintings, es-
dictors) and their varying validity (i.e., predictive accu-
pecially under stress (Moritz et al., 2009) which conforms
racy) are presented in a two-dimensional matrix (Figure
other studies assigning stress and arousal an aggravating
1). Information is (usually) hidden behind information
role for cognitive biases in the disorder (Lincoln, Lange,
cards and can be investigated by mouse click. Besides
Burau, Exner, & Moritz, in press).
choices, decision times and con?dence ratings, Mouse-
For the present study, we used a task that assesses the
lab allows recording and analyzing the amount, distribu-
data-gathering, con?dence, and the information integra-
tion and order of information search to infer individuals’
tion aspect of JTC in a single paradigm. We were espe-
decision strategies (for a discussion of the limitations of
cially interested to investigate cue selection in patients,
Mouselab, see Glöckner & Betsch, 2008c).
since a striking feature of schizophrenia is patients’ re-
liance on unreliable sources of information (e.g., Internet
2 Method
fora for conspiracy theories). JTC may not be a problem
if it is rested on the most valid pieces of information, and
indeed cognitive research has found that a subgroup of
2.1 Subjects
healthy subjects adopt a so-called take the best heuristic
Overall, 67 subjects took part in the experiment. They
(Ayal & Hochman, 2009; Bröder, 2000; Bröder & Schif-
belonged to a clinical condition of schizophrenic patients
fer, 2003; Gigerenzer & Goldstein, 1999; see also Hilbig,
(Szs) or were healthy persons (CPs). The sample con-
2008a; Hilbig, 2008b; Newell, Weston, & Shanks, 2003)
sisted of 37 Szs and 30 CPs. Patients were inpatients re-
and that the application of this heuristic in some environ-
ments leads to good decisions (Czerlinski, Gigerenzer, &
1In contrast to preference decisions (e.g., which car do you prefer),
Goldstein, 1999). Although the beads task remains the
probabilistic inference tasks have an objectively correct solution.
2For an early study using classic cognitive tasks to investigate the
gold standard to capture JTC, it does not shed light on
processes underlying formal thought disorders see also Persons and
this aspect of decision-making, as the sequence of events
Baron (1985).

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Analysis of the jumping-to-conclusions bias
589
cruited at the Department of Psychiatry and Psychother-
a validity of .70, .80, and .60 (i.e., predictive validity of
apy of the University Medical Center Hamburg Eppen-
70%, 80% and 60% accuracy). Subjects were explicitly
dorf. Healthy subjects were recruited via a subject pool,
informed about the validity of the cues which was equal
advertisement and word-of-mouth. No monetary or other
in all decision tasks. Half of the decisions used neutral
incentive was provided for any of the subjects. All pa-
materials (i.e., chose the better out of three brands of or-
tients gave written informed consent for participation.
anges; Figure 1, left), the other used more affective mate-
Diagnoses relied on DSM-IV criteria for schizophrenia
rial (i.e., chose one out of three persons who more likely
which were determined by experienced clinicians us-
committed a crime; Figure 1, right) constituting the factor
ing the neuropsychiatric MINI interview (Sheehan et al.,
Affective Valence.4 This factor was fully crossed with the
1998). Symptom severity was assessed by the same clin-
factor Presentation Format/Time Pressure. A third of the
icians with the Positive and Negative Syndrome Scale
decisions were presented in the classic Mouselab format
(PANSS; Kay, Opler, & Lindenmayer, 1989) following
with hidden information boxes and without time pres-
a semi-structured interview. The PANSS has 30 items.
sure, another third was presented in the same paradigm
The positive and negative syndrome scores were com-
but with explicit time pressure induction using a time-bar
posed following the standard algorithm (sum of all seven
(Figure 1, left). In the remaining decision tasks, informa-
positive and all seven negative items, respectively). In
tion was instantly available and subjects were instructed
addition, we computed a core delusion score which was
to decide as quickly as possible (Figure 1, right). Subjects
comprised of the items tapping delusions, suspiciousness
completed 18 decisions for the six combinations of con-
and unusual thought content. In keeping with factor an-
ditions each, constituting the factor Decisions. Decision
alytic studies which have reliably detected a syndrome
tasks were similar to the ones used in previous studies
called disorganization, we also computed a disorgani-
(Glöckner & Betsch, 2008c). Hence, we used a 2 (Szs
zation factor capturing formal thought disorder, postur-
vs. CPs) x 2 (Affective Valence) x 3 (Presentation For-
ing/mannerisms and disorientation.
mat/Time Pressure) x 18 (Decisions) design with all fac-
None of the patients had substance dependency or neu-
tors except of the ?rst one being manipulated within sub-
rological disorders. Healthy subjects were screened for
jects. Subjects were assigned to one out of four balancing
absence of a psychiatric illness using the MINI inter-
conditions in which order of the relevant conditions was
view. Additionally, the premorbid intelligence of all sub-
varied (neutral vs. affective ?rst; hidden vs. open infor-
jects was tested using a vocabulary test, the Multiple
mation ?rst).5
Choice Intelligence Task (MWT-B; Lehrl, 1995). The
MWT-B requests the subject with 36 items each consist-
ing of 5 words, of which only one is a correctly spelled
2.3 Procedure
noun. Neuroleptic dosage was converted in % maximal
Subjects ?rst completed a mouse-ability pre-test in which
neuroleptic dosage following German prescription guide-
they opened the nine boxes of the information matrix
lines.
by mouse-click as quickly as possible. This test was
The groups were comparable in age (MSzs= 31.8,
later used to adapt trial duration to the subjects’ indi-
SDSzs=10.5 vs.
MCPs= 32.1, SDCPs=12.0 years), IQ
vidual speeds. This procedure was repeated ?ve times
(both M=106, SDSzs=12.8, SDCPs=38.8) and duration of
to determine the average time for information search.
school education (MSzs= 11.5, SDSzs=1.6 vs. MCPs= 11.9,
Subjects were introduced to the decision task using the
SDCPs=1.6 years). Szs were mainly male (28 male),
neutral material (i.e., select the best orange based on
whereas CPs were mainly female (20 female).3 Szs re-
testers). The three presentation formats of Presentation
ceived inpatient treatment on average 3.5 times (SD=3.7;
Format/Time Pressure were introduced (i.e., hidden infor-
including the current hospitalization).
mation Mouselab, hidden information Mouselab + time
4The effectivity of the manipulation of affective valence was shown
2.2 Materials and design
in a comprehensive pre-study (N=122; student population) using essen-
tially the same material and procedure. Subjects (among other mea-
All subjects completed a total of 108 probabilistic infer-
sures) indicated in how far they were emotionally affected by the differ-
ence decisions between three options based on 3 cues
ent types of decisions. A repeated measurement ANOVA with valence
which differed in validity (i.e., the percentage of correct
as within subjects factor indicated a strong and signi?cant effect on the
ratings, F (1, 121) = 64.8, p < .001, ?2 = .35. Subjects were much more
predictions given a certain criterion value). The cues had
emotionally affected by the criminal case decision tasks as compared to
the orange decision tasks.
3To control for gender effects, we also ran the core analyses correct-
5For pragmatic reasons, hidden information presentation under time
ing for gender effects using gender as covariate. Although we observed
pressure always followed after hidden information presentation with-
some gender effects, the results concerning our hypotheses were simi-
out time pressure. Due to some error in the randomization procedure,
lar to the results without covariate. In the following, only the simpler
subjects were not exactly equally distributed over the counterbalancing
analyses without the covariate are reported.
conditions.

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Analysis of the jumping-to-conclusions bias
590
Oranges 1
Oranges 2
Oranges 3
Suspect 1
Suspect 2
Suspect 3
Choose
Choose
Choose
Choose
Choose
Choose
Tester 1
Inspector 1
70% correct
+
?
?
70% correct
+
-
-
Tester 2
Inspector 2
80% correct
?
?
-
80% correct
-
+
-
Tester 3
Inspector 3
60% correct
+
?
?
60% correct
+
-
+
Please indicate how certain you are in making this decision!
Please indicate how certain you are in making this decision!
o absolutely certain o very certain o somewhat certain o guessing
o absolutely certain o very certain o somewhat certain o guessing
Continue
Continue
Figure 1: Decision screens for neutral orange decisions (left) and affective criminal case decisions (right). The left
picture shows an example for a Mouselab with time-pressure condition, the right picture shows an example for open
information presentation (cf. factor Presentation Format/Time Pressure).
pressure, open information Mouselab) and subjects com-
The aspects of JTC were tapped via the following vari-
pleted a test decision for each of them. Complete in-
ables:
struction can be found in the appendix. Each decision
trial started with the presentation of the information ma-
1. Biased gathering of information
trix with open or hidden information cards. In the two
• Amount of inspected information.
hidden-information conditions, information cards could
• Validity of inspected information.
be opened with the mouse and remained open for the rest
of the decision. Decision time was recorded from stim-
• Order of information search (i.e., is more valid
ulus onset (open information), or from the inspection of
information inspected ?rst).
the ?rst information card (hidden information). In the
2. Overcon?dence
explicit time pressure condition, only a limited time was
• Number of con?dence ratings “absolutely cer-
available for information search. The time was individ-
tain”. Given that even the best cue had a valid-
ually determined from the average time for information
search (see above) in the mouse-ability pre-test plus 2
ity of no more than 80%, any absolutely certain
seconds. Hence, there was just suf?cient time for in-
response was deemed incautious.
specting all information and applying a simple decision
3. Suboptimal weighting and information integration
strategy. A down-counting time-bar was used to induce
• Percentage of normatively correct answers ac-
time pressure (see Figure 1, left). Options were selected
cording to Bayes’ theorem.
by mouse-click. Afterwards individuals rated the con-
?dence in their decision on the following 4-point scale:
• Individuals’ decision strategies. The percent-
absolutely certain (1), very certain (2), somewhat cer-
age of subjects that take into account all in-
tain (3), guessing (4).6 Finally, subjects were informed
formation and weight them by their validity
that the next decision was about other oranges or other
(weighted additive strategy users), was com-
accused persons, and the next trial was started by mouse-
pared with the proportion of subjects that
click. Decision tasks were presented using a block design
ignore less valid information (take the best
of six blocks (2 Affective Valence x 3 Presentation For-
users), and the ones that ignore cue weights
mat/Time Pressure) with randomized order of decisions
(equal weight users).
within each block. After each block, a short break and a
new instruction for the following block was included.
3 Results
6The corresponding scale labels in German language were: “völ-
Checks for mouse handling. Overall, Szs and CPs
lig sicher”, “sehr sicher”, “wenig sicher”, and “geraten”. The original
question was: “Schätzen Sie bitte ein, wie sicher Sie sich bei dieser
were both well able to handle the pre-test as well as the
Entscheidung sind!”
main experiment. There were no drop outs during the

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Analysis of the jumping-to-conclusions bias
591
Table 1: Means of log-transformed decision times with SEs in parentheses. MOUS, MOUS+TP, OPEN refer to the
presentation formats mouselab, mouselab with time pressure, and open mouselab. Szs are schizophrenic patients, CPs
are controls.
Neutral
Affective
MOUS
MOUS+TP
OPEN
MOUS
MOUS+TP
OPEN
Szs
3.82 (0.04)
3.69 (0.04)
3.58 (0.03)
3.84 (0.04)
3.73 (0.03)
3.58 (0.03)
CPs
3.84 (0.05)
3.67 (0.04)
3.52 (0.04)
3.84 (0.04)
3.68 (0.03)
3.54 (0.04)
experiment. Since the program was computer directed,
formation inspection scores, which measure the percent-
no missing values occurred for our main dependent vari-
age of opened information boxes per cue for each of the 2
ables. To analyze whether the Szs differ from CPs in
(Affective Valence) x 2 (Presentation Format/Time Pres-
their mouse-handling skills, a t-test was conducted using
sure: hidden information Mouselab with vs. without time
mean reaction time in the mouse pre-test as dependent
pressure) blocks of decision tasks. Descriptive statistics
variable. The test revealed a marginally signi?cant dif-
are summarized in Table 2. A 2 (Szs vs. CPs) x 2 (Af-
ference, t(61.7) = 1.70, p = .09. As expected, reaction
fective Valence) x 2 (Presentation format/Time Pressure:
time was higher for the Szs (M=7.1 sec) compared to the
hidden information Mouselab with vs. without time pres-
CPs (M=6.1 sec) indicating lower mouse-handling skills
sure) x 3 (Cue) mixed model ANOVA was calculated us-
of the former. As outlined above, slowed subjects were
ing information inspection scores as dependent variable.
automatically allowed more time for the time-pressure
There was no main effect of clinical condition, F (1, 65)
tasks.
= .57, p = .45, ?2 = .009. In the structured information
environment Mouselab, Szs inspected 91% and CPs 88%
Decision times. Decision times were log-transformed
of the information. Hence, there was no reduced infor-
to base 10 before conducting the analysis to correct for
mation search of Szs and H1 was not supported by the
deviations from normal distribution and to reduce the in-
data. There was a main effect for Cue, indicating that
?uence of outliers. A 2 (Szs vs. CPs) x 2 (Affective
subjects focused more on more valid cues, F (1.2, 77.9)
Valence) x 3 (Presentation Format/Time Pressure) x 18
= 23.2, p < .001, ?2 = .26. For cue 2 (80% validity),
(Decisions) mixed model analysis of variance (ANOVA)
97% of the information was inspected, whereas for the
was computed to analyse log-transformed decision times.
cues 1 (70% validity) and 3 (60% validity) only 89% and
There was a highly signi?cant main effect for Presen-
83% were investigated. Descriptively, Szs focused less
tation Format/Time Pressure, F (1.4, 92.4) = 95.5, p <
strongly on this most valid cue and more on the less valid
.001, ?2 = .60 (as in all following analyses Greenhouse-
cues compared to CPs, indicating a less systematic in-
Geisser correction was used if the assumption of spheric-
formation search. The respective interaction effect, how-
ity was violated). The log-mean decision times for the
ever, did not reach conventional signi?cance levels in a
conditions hidden information Mouselab, hidden infor-
two-sided test, F (1.2, 77.9) = 1.86, p = .17, ?2 = .03.
mation Mouselab with time pressure, and open informa-
To explore H2 (less valid information) directly, we
tion Mouselab were 6.8 sec, 4.9 sec, and 3.6 sec. Thus,
compared information search differences of Szs and CPs
the time pressure manipulation worked, in that decision
for cue 2 (80%; most valid cue) and cue 3 (60%; least
times decreased if a time limit was enforced in Mouselab.
valid cue) using planned deviation contrasts. The interac-
Note, however, that decision time in the open Mouselab
tions between cue 2 (cue 3) vs. the grand mean and Szs
paradigm was even way below the decision time in the
vs. CPs both turned out to be marginally signi?cant, F
time pressure Mouselab condition. The main effect for
(1, 65) = 1.93, p = .08, ?2 = .03 (F (1, 65) = 2.15, p =
clinical condition and all interactions did not reach con-
.07, ?2 = .03; both one-sided).7 This provides support
ventional signi?cance levels (all Fs < 1.4; Table 1). Thus,
that SPs, indeed, look up the most valid cue less often
decision time did not differ between Szs and CPs. Hence,
than CPs and in contrast investigate the least valid cue
although Szs were slower in simple mouse-handling, they
more often. To test the robustness of these marginally
did not show longer decision times overall.
signi?cant results, we rerun the analysis using a multi-
level regression with random effects for subjects (Nezlek,
Schröder-Abe, & Schütz, 2006) and a clustered regres-
Information search.
To test the hypotheses that Szs
look up less information (H1) and concentrate more on
7Due to the fact that we had a directed hypothesis, we used one-sided
less valid information (H2) than CPs, we calculated in-
tests.

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Analysis of the jumping-to-conclusions bias
592
Table 2: Mean information inspection rate with SEs in parentheses by cue. MOUS, MOUS+TP, OPEN refer to the
presentation formats mouselab, mouselab with time pressure, and open mouselab. Szs are schizophrenic patients, CPs
are controls.
Neutral
Affective
Overall
MOUS
MOUS+TP
MOUS
MOUS+TP
Cue 1 (70% correct)
Szs
0.88 (0.03)
0.90 (0.04)
0.94 (0.02)
0.92 (0.03)
0.91 (0.02)
CPs
0.88 (0.04)
0.85 (0.04)
0.90 (0.03)
0.88 (0.04)
0.88 (0.02)
Cue 2 (80% correct)
Szs
0.95 (0.02)
0.95 (0.02)
0.96 (0.02)
0.98 (0.01)
0.96 (0.01)
CPs
0.97 (0.03)
0.98 (0.03)
0.98 (0.02)
0.98 (0.01)
0.98 (0.01)
Cue 3 (60% correct)
Szs
0.84 (0.05)
0.87 (0.05)
0.88 (0.04)
0.85 (0.05)
0.86 (0.02)
CPs
0.80 (0.05)
0.76 (0.05)
0.82 (0.04)
0.80 (0.05)
0.80 (0.02)
sion (Rogers, 1993) using robust standard errors (Hayes
ratings (i.e., feelings of absolute con?dence; Moritz &
& Cai, 2007). In the former, both effects reached the con-
Woodward, 2006). Therefore, in line with previous re-
ventional signi?cance level (both p < .05; one-sided), in
search (Moritz, Woodward, & Rodriguez-Raecke, 2006;
the latter the cue2-interaction-effect was marginally sig-
see also Moritz, Woodward, Whitman, & Cuttler, 2005)
ni?cant (p < .10) whereas the cue 3-interaction-effect was
we tested H4 by investigating the frequency of the rating
signi?cant (p < .05). Hence, results seem to be robust and
“absolutely certain”. In the probabilistic inference task
we conclude that there is support for H2 in that patients
used in the study, one could never be absolutely certain
are less guided by valid information.
that a chosen outcome would be realized, because pos-
To analyze H3 (that Szs use a less focused and
terior probabilities were all below 1. Hence, “absolutely
more unsystematic information search), we analyzed the
certain” ratings are deemed incautious. An extreme rating
amount of information search that focused on the most
score (i.e., frequency of absolutely certain ratings) was
important cue (i.e., cue 2) for the ?rst three information
analyzed using an independent t-test. In line with earlier
acquisitions in each decision. Assuming that all pieces of
?ndings, we observed a signi?cantly higher number of
information are looked up, this represents the ?rst third
extreme ratings from Szs (M = 28.7) as compared to CPs
of information inspections.8 In all three cases, CPs fo-
(M = 9.3), t(52.2) = 2.8, p = .007, supporting H4.
cused much more strongly on the most valid cue than Szs
We additionally conducted a 2 (Szs vs. CPs) x 2 (Af-
(total proportion of acquisitions on the most valid cue in
fective Valence) x 3 (Presentation Format/Time Pressure)
1st: 47% vs. 24%; 2nd: 65% vs. 40%; 3rd acquisition:
x 18 (Decisions) mixed model ANOVA to analyse mean
47% vs. 21%). We calculated initial-focus scores indi-
con?dence ratings. There was also a tendency that Szs
cating the average frequency of inspections of the most
(M = 2.17) were more con?dent in their decisions than
valid cue in the ?rst three acquisitions per person. The
CPs (M = 2.40) (low scores indicate high con?dence),
score was signi?cantly higher for CPs (M = 1.06, SE =
which, however, did not reach conventional signi?cance
0.13) as compared to Szs (M = 0.56, SE = 0.10) accord-
levels, F (1, 65) = 2.15, p = .15, ?2 = .03. There were
ing to an independent t-test, t(65) = 3.04, p < .01. Hence,
main effects for affective valence, F (1, 65) = 8.3, p =
there is support for H3 that Szs show a less systematic
.005, ?2 = .11 and Presentation Format/Time Pressure,
information search.
F (2.0, 129.8) = 4.8, p = .01, ?2 = .07. Subjects were
more con?dent in the neutral (M = 2.22) as compared to
Con?dence ratings. It has been shown that overcon-
the emotional (M = 2.34) decisions. There was, however,
?dence of Szs does not always result in a generally in-
no interaction with the clinical condition. Subjects under
creased con?dence level, but in more frequent extreme
time pressure were less con?dent than in the two other
conditions. Nevertheless, these results should be inter-
8Due to the fact that information boxes remain open after being in-
spected, this effect was naturally reversed for the remaining two-thirds
preted cautiously because it is not entirely clear whether
of information inspections.
our measure for con?dence ratings is interval-scaled.

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Analysis of the jumping-to-conclusions bias
593
Table 3: Mean con?dence scores with SEs in parentheses. MOUS, MOUS+TP, OPEN refer to the presentation for-
mats mouselab, mouselab with time pressure, and open mouselab. Szs are schizophrenic patients, CPs are controls.
Con?dence scale ranged from 1 (absolutely certain) to 4 (guessing).
Neutral
Affective
MOUS
MOUS+TP
OPEN
MOUS
MOUS+TP
OPEN
Szs
2.06 (0.11)
2.19 (0.11)
2.15 (0.11)
2.24 (0.11)
2.27 (0.11)
2.12 (0.10)
CPs
2.31 (0.13)
2.33 (0.12)
2.31 (0.13)
2.41 (0.13)
2.55 (0.12)
2.47 (0.11)
To investigate the effect further, we examined the num-
tion about cue validities and allowed information search
ber of extreme ratings separately for correct and wrong
in a limited space, Szs decided with similar accuracy to
decisions according to Bayes’ Theorem (see also below).
CPs.
We determined for each decision whether people did or
did not chose the option with the highest posterior proba-
Decision strategy analysis. Individuals’ decision
bilities for being of good quality from the set of available
strategies were analyzed by within-subjects compar-
options.9 Interestingly, the effect was driven by more ex-
isons of the distribution of choices using ?2-tests (for
treme ratings in the cases in which subjects made correct
detailed description of the method see Glöckner &
decisions and the difference between CPs and Szs was not
Betsch, 2008c).10
The method allowed to determine
signi?cant for the wrong decisions. Hence, in the struc-
if individuals used a take the best heuristic (TTB, i.e.,
tured environment Mouselab, Szs used more extreme rat-
ignore less valid information and based their decision on
ings, but mainly in cases in which they made normatively
the most important information only), an equal weight
correct decisions. Thus, in line with the previous results,
heuristic (EQW, i.e., ignore the validity of cues and
Szs seem to show more extreme con?dence ratings, but
chose the option which has more positive predictions),
often these occur in cases in which their decisions are
or a weighted additive strategy (WADD, i.e., choose the
normatively correct.
option with the higher weighted sum of cue values and
cue validities) (Payne et al., 1988). The results indicate
Quality of choices. To analyze the quality of choices
that there was no increased usage of TTB for Szs (Table
we calculated whether choices were in line with the op-
4).11
timal solution according to Bayes’ Theorem. In both
In line with recent ?ndings, Szs and CPs both mainly
conditions, a high proportion of correct choices was ob-
used a WADD strategy (cf. Bröder, 2003) that might be
served. Szs showed 75% correct choices, CPs showed
based on automatic processing (Glöckner, 2008; Glöck-
77%. There was no signi?cant difference concerning
ner & Betsch, 2008a, 2008b, 2008c; Glöckner & Her-
quality of choices between Szs and CPs (see logistic
bold, in press; Glöckner & Hodges, 2009; Glöckner &
regression in additional analyses below). In a struc-
Witteman, in press). Interestingly, this is not the case for
tured information environment which provided informa-
Szs who have to make (affective) criminal decisions un-
9
der time pressure. Under this very speci?c condition Szs
The optimal solution to the problem is to choose the option with
the highest posterior probability being of good quality given the base-
show an increased usage of EQW. This result indicates
rate and all cue values. The cue validities provided in this experiment
that Szs ignore cue weights in high stress conditions in
can be interpreted as prior probabilities [p(cue+|O+) = 1 ? p(cue-|O+);
which time pressure and high affective valence coincide.
with O indicating options (A, B or C) and subscripts +/- indicating posi-
tive/negative criterion values or cue values, respectively]. Hence, under
According to rational standards of probability theory, this
the assumption that the cues make independent predictions the posterior
can be considered a bias because the predictions of the
probability for good quality of, for instance, option A, p(A | cues, base-
cues differ in their validity, which should be taken into
rate), according to Bayes’ theorem can be determined using the base
rate, the cue values and the prior probabilities of the cues according to:
10Per subject two ?2-tests were conducted, which tested against the
p(A|p
null hypotheses that individuals ignored less valid cues (i.e., used TTB)
A, pC1, pC2, pC3)
p
p
p
p
=
A
C1
C2
C3
and that they did not take into account cue weights (i.e., used EQW).
1? p(A|pA, pC1, pC2, pC3)
1? pA 1? pC1 1? pC2 1? pC3
Only if both hypotheses could be rejected, individuals were classi?ed
pA is the base-rate for good quality of option A. Because each decision
as WADD users. In the case that the error rate for the classi?ed strategy
is made between new options no informative base-rate information is
was above .50, individuals’ decision strategy was not classi?ed (but see
available (we set pA=.50 which can be ignored in calculations). pC1,
Glöckner, 2009, for an improved methodological approach).
p
11
C2, and pC3 are the prior probabilities of the respective cue value given
Because of a programming error, decision strategies could be reli-
the option has good quality (i.e., for positive cue values: .70, .80, .60;
ably determined only for the Presentation Format/Time Pressure condi-
for negative cue values: .30, .20, .40). The option with the highest
tions hidden information Mouselab with time pressure and open infor-
posterior probability for good quality should be chosen.
mation Mouselab.

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Analysis of the jumping-to-conclusions bias
594
signi?cant interactions of these factors with clinical con-
Table 4: Proportion of subjects using the respective de-
dition, indicating that there are no differential effects of
cision strategy by condition. TTB (take the best) strat-
stress induced by time pressure and affective valence on
egy indicates ignorance of less valid cues, EQW (equal
Szs as compared to CPs concerning decision time, con-
weight) strategy indicates inappropriate equal weighting
?dence and information search. As reported in the last
of cue information, WADD (weighted additive) strategy
section, we found a shift in decision strategies speci?-
indicates an integration of cue information according to
cally for Szs under high stress induced by time pressure
its validity. For the clinical condition, Szs stands for
and affective valence. Under this condition, many Szs
schizophrenia patients, CPs for controls.
used EQW which means that they seemed to ignore the
Decision strategy classi?cation (in %)
validity of cues. Hence, H5 that biases of Szs should be
more pronounced under stress was supported by the data
Clinical
TTB
EQW
WADD
not class.
for information integration strategies, but not for con?-
condition
dence and information search.
Time pressure Mouselab neutral (oranges)
Szs
0.16
0.16
0.59
0.08
Correlations between schizophrenia measures and de-
CPs
0.2
0.2
0.57
0.03
pendent variables. The observed mean PANSS sum-
score largely corresponds to a “mildly ill” clinical state
Open Mouselab neutral (oranges)
according to the criteria adopted by Leucht et al. (2005).
Szs
0.14
0
0.78
0.08
For Szs, the PANSS sum-score and the subscales of the
CPs
0.2
0.13
0.67
0
PANSS (positive, negative, disorganization, delusions)
were correlated with the total amount of information
Time pressure Mouselab affective (criminal case)
search, and the number of absolutely certain ratings (Ta-
Szs
0.19
0.35
0.38
0.08
ble 5). The amount of information search correlated
CPs
0.17
0.23
0.57
0.03
marginally signi?cantly and negatively with the PANSS
sum-score. The effect was mainly driven by correlations
Open Mouselab affective (criminal case)
with the PANSS positive and PANSS delusion subscale
Szs
0.14
0.08
0.73
0.05
and it did also hold in a regression when simultaneously
controlling for intelligence and gender differences (at p <
CPs
0.17
0.1
0.7
0.03
.05). This indicates that the amount of information search
decreases with the severity of schizophrenia symptoms.
There were, however, no signi?cant correlations between
account. To test this result statistically, we conducted a
the amount of absolute certain ratings and the schizophre-
logistic regression with usage of WADD (0=no, 1=yes)
nia measures. Due to the relatively small power in the
as categorical dependent variable, and clinical condition
analysis (power = .45; two-tailed test assuming a medium
(Sz=1 vs. CP=0), and a contrast comparing the high stress
effect r = .30; Faul, Erdfelder, Lang, & Buchner, 2007),
conditions (i.e., Mouselab with time pressure and affec-
further research must determine whether this null result
tive content; coded 1) against the remaining conditions
replicates (but see also the general discussion for similar
(coded ?1/3) as well as their interaction as predictors.12
?ndings in previous studies).
The interaction between clinical condition and the con-
Maximal neuroleptic dosage following German pre-
trast for high versus low stress condition turned out sig-
scription guidelines (in percent) did not differ between
ni?cant, Odds-ratio = .33, z = ?2.19, p = 0.029 indicating
users of different strategies and did not correlate signi?-
that (after correcting for the main effects) the probability
cantly with information search, and con?dence. The av-
for usage of WADD was reduced in the stress condition
erage dosage was M = 61% (SD = 44%).
to one third as compared to the other conditions.
Differential in?uence of time pressure and affective
Additional analyses. We observed training effects over
valence on Szs vs. CPs. As indicated by the previous
the 108 choices in the main experiment. Subjects de-
analyses, there was a main effect of time pressure on de-
cided more quicly over time. We regressed decision time
cision time and a main effect of affective valence and
(in milliseconds) on order (1 to 108), clinical condition
time pressure on con?dence. We did not, however, ?nd
(Sz=1, CP=0), and their interaction and found a signi?-
cant order effect, b = ?25.62, t = ?4.72, p < 0.001, but no
12As in all following regressions we corrected for clusters in the
interaction with clinical condition, b = ?9.80, t = ?1.19,
data due to repeated measurement (Rogers, 1993) and used robust stan-
dard errors (Hayes & Cai, 2007) relying on STATA standard commands
p = 0.24. Hence, training effects were not signi?cantly
“cluster” and “robust” (Gould, Pitblado, & Sribney, 2006).
different for Szs and CPs. Furthermore, we investigated

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Analysis of the jumping-to-conclusions bias
595
information. In a structured information environment,
Table 5: Descriptive statistics for schizophrenia measures
however, they do not inspect fewer pieces of information
and correlation with decision parameters (amount of in-
than CPs. This ?nding re?ects recent evidence that JTC,
formation search, amount of absolute certain ratings for
in the sense of a data-gathering bias, is not found with
con?dence). + p < .10; * p < .05.
all paradigms (Ziegler, Rief, Werner, Mehl, & Lincoln,
Amount
2008) and may under some conditions also be abolished
M
Info
Schizophrenia measures
absolute
in the beads task (Moritz, Woodward, & Lambert, 2007).
(S.D.)
search
certain
52.1
4.1 Con?dence Ratings
PANSS Sum-Score
?.30+
?.01
(13.8)
We further observed that Szs tend to be overcon?dent, in
PANSS positive (positive
11.6
that they use the extreme and inappropriate rating “abso-
items 1-7, conventional
?.38?
?.05
(6.4)
lutely certain” more often than controls, which accords
algorithm)
with ?ndings using memory paradigms (Moritz, Wood-
PANSS negative (negative
ward, & Hausmann, 2006; Moritz, Woodward, Jelinek,
11.3
items 1-7, conventional
?.27
.03
(3.9)
& Klinge, 2008). Interestingly, extreme ratings seem to
algorithm)
be more sensitive to capture group differences than sim-
PANSS desorg (positive item
3.4
ple means, on which we did not ?nd differences. Our
?.01
?.17
2, global items 5 and 10)
(0.9)
?ndings do not indicate a general overcon?dence but only
PANSS core delusional
an increased number of extreme ratings, which we view
7.5
items (positive items 1 and 6,
?.40?
?.10
as a speci?c kind of overcon?dence (see also Moritz &
(4.3)
global item 9)
Woodward, 2006). “Overcon?dence” implies that “abso-
lutely certain” ratings are considered irrational. We think
that this is justi?ed by the fact that all Bayes-posterior
whether the correctness of choices according to the nor-
probabilities were below 1 and hence ratings for absolute
mative Bayes’ standard changed over time by conduct-
certainty are incautious. Of course, some subjects might
ing a logistic regression with correct scores (1=correct,
have misinterpreted the scale and could have used the rat-
0=wrong) as dependent variable and order (1 to 108),
ing as an indicator for some high level of certainty. There
clinical condition (Sz=1, CP=0), and their interaction as
is, however, no good reason to assume that this should
predictors. None of these effects turned out signi?cant
appear more often for Szs compared to CPs. However,
(clinical condition: Odds?ratio=.97, z = ?0.12, p = 0.91;
considering the fact that the increased number of extreme
order: Odds-ratio=.997, z = ?1.25, p = 0.211; interac-
ratings in Szs was mainly found for correct choices we
tion: Odds-ratio=.998, z = ?0.36, p= 0.72). Hence, also
cannot completely rule out that this increase might also
for performance there were no differential learning ef-
be partially due to better discrimination.
fects for Szs as compared to CPs.
4.2 Decision strategies and decision quality
4 Discussion
Szs and CPs mainly used complex WADD strategies to
make decisions (i.e., they took into account all pieces
In the current study, we used Mouselab to investi-
of information according to their importance of valid-
gate different aspects of jumping to conclusions (JTC)
ity).
Under most conditions, there was no tendency
in schizophrenia versus healthy subjects: information
gathering, overcon?dence and information integration in
for Szs to rely more on simple heuristics such as Take
probabilistic inference decisions. Overall, we observed
the Best (TTB) or Equal Weighting (EQW). Only under
group differences on all three dependent variables but the
high stress induced by affective valence and time pres-
magnitude of these differences was smaller than we ex-
sure did Szs rely more on EQW strategies, which, ac-
pected, perhaps owing the rather mild psychopathological
cording to probability theory, implies a less appropriate
status. We found that Szs show a less systematic informa-
weighting of information (i.e., all pieces of information
tion search compared to CPs. In line with the clinical ob-
are weighted equally although they differ in validity).
servation that patients give undue weight to less relevant
One might, however, argue that applying EQW under
and sometimes random aspects, schizophrenia subjects
time pressure is adaptive (or even rational) if only mi-
focused more strongly on less valid information and, in
nor effects on accuracy can be expected, because, with
particular, started the information search with less valid
less cognitive effort, EQW often leads to rather accu-

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Analysis of the jumping-to-conclusions bias
596
rate choices (Dawes & Corrigan, 1974). Note, however,
4.4 Severity of schizophrenia and decision
that this classic adaptive-strategy-selection explanation
behavior
cannot account for the observations a) that the effect is
not found for the time pressure condition without affec-
For Szs, we observed a correlation between schizophre-
tive stimuli, and b) that the effect is found only for Szs
nia measures (particularly driven by PANSS positive and
but not for CPs of similar intelligence. Hence, it seems
PANSS delusion scores) and the amount of information
more likely that the effect is caused by a speci?c reac-
search. Subjects who scored higher on these measures
tion of Szs to stress induced by time pressure and affect.
gathered fewer pieces of information. We did not, how-
ever, ?nd a respective effect for the amount of absolute
Considering recent ?ndings, it might even be questioned
con?dence ratings, which might be partially due to the
whether WADD strategies that are based on automatic-
low power of the analysis. Note, however, that a null-
intuitive processes are indeed cognitively more effort-
effect for the latter relation has also been observed in
ful than deliberate EQW strategies (Glöckner & Betsch,
prior studies (Moritz, Woodward, & Rodriguez-Raecke,
2008a, 2008c; Glöckner & Herbold, in press; Hilbig &
2006). Overall, there were no reliable differences in strat-
Pohl, 2009; see also Horstmann, Ahlgrimm, & Glöckner,
egy use related with increasing schizophrenia scores.
2009).
Interestingly, in spite of these differences concerning
aspects of decision making, we did not observe an overall
4.5 Partial task speci?city of ?ndings
difference in the quality of the choices compared with the
The present ?ndings may seem at odds with data ob-
normative standard provided by Bayes’ theorem between
tained with the beads task where stronger JTC in the
Szs and CPs. In Mouselab, the effects of JTC biases on
sense of less gathered information has been quite con-
decision quality seem to be low.
sistently found (Fine, Gardner, Craigie, & Gold, 2007).
A striking difference between our task and the beads task
could be that the subject in our task was confronted with
4.3 Structured information presentation
place holders for the other options, which could have
and decision quality of Szs
fostered curiosity or encouraged a search through all in-
formation — especially since it was not associated with
It should be noted that the Mouselab environment differs
great time loss. In contrast, in the draws-to-decision con-
from real world decisions in that it provides a clearly
dition of the beads task the subject knows that more ?sh
structured and rather simple information environment
can be drawn, usually no placeholders for other beads
with a ?nite number of data. All these pieces of infor-
are presented, and after each draw the subject is asked
mation are available and can be nicely compared and in-
whether or not to terminate data gathering, which could
tegrated within short time. Decisions in the real world,
also prompt hasty decision-making. Therefore, we regard
in contrast, are often characterized by incomplete in-
?ndings obtained from both tasks as complementary and
formation, provided in an unstructured way and with-
they allow for the conclusion that patients rest too much
out easily comparable probability information. Perhaps,
con?dence on scarce information.
most importantly, the amount of available information is
In sum, our results qualify ?ndings concerning JTC bi-
unknown, gathering of further information is associated
ases in structured information environments. Although
we found JTC biases for speci?c aspects of information
with high (time) costs, and the number of available cues
search (i.e., less focused, less ordered), con?dence rat-
is uncertain. The present ?nding that patients, especially
ings (i.e., more extreme ratings) and information integra-
in the ?rst phase of each trial, tend to use less valid cues
tion (i.e., inappropriate cue weighting under stress) other
might be more consequential in the real world, where the
aspects (i.e., amount of information search, average con-
search process might not be prolonged. To summarize,
?dence rating, integration strategy without stress, quality
the structured environment is likely to in?uence choice
of decisions) were not in?uenced. This more differenti-
behavior (cf. Glöckner & Betsch, 2008c) and it therefore
ated view was made possible by a paradigm tapping dif-
has to be tested whether our results generalize to other
ferent aspects of JTC in a single paradigm which could
realistic situations. Against the backdrop of earlier ?nd-
stimulate further research. Cognitive treatment programs
ings concerning stronger JTC biases in other situations,
such as the Metacognitive Training for Schizophrenia pa-
the structured information presentation, however, seems
tients (MCT; Moritz & Woodward, 2007; Moritz, Wood-
to have a positive in?uence on choice behavior. Hence,
ward, & Group, 2007) have began to train patients to
structured information presentation in a matrix format in-
search both for more and especially valid pieces of in-
cluding information on cue validities might be a means to
formation and to tone down con?dence in case of incon-
enhance decision quality of Szs.
sistent evidence or ambiguity.

---