Effects of negative mood states on risk in everyday decision making

COGNITION AND EMOTION, 2000, 14 (6), 823–855
Effects of negative mood states on risk in everyday
decision making
G. Robert J. Hockey
University of Leeds, UK
A. John Maule
University of Leeds, UK
Peter J. Clough
University of Hull, UK
Larissa Bdzola
University of Leeds, UK
How does negative mood affect risk taking? A brief questionnaire was used to
measure state anxiety, depression, and fatigue, and a daily mood diary allowed
state and trait (average level) mood to be separated. Studies 1 and 2 used natural
moods and Study 3 a mood induction procedure. Risk was assessed using hypothetical everyday choice scenarios. Study 1 showed that riskiness was affected by
state fatigue, but not by anxiety and depression. Study 2 showed that increased
riskiness over a two-week period was predicted by fatigue changes, after controlling for riskiness and trait and state mood at time 1. Fatigue effects were
stronger for more important scenarios, and when state anxiety was also high. In
Study 3, covariance analyses showed that the observed increased in riskiness was
related to induced fatigue, rather than to anxiety or depression. The effects are
discussed in relation to the literature on fatigue effects, and models of mood and
cognition.
Please send correspondenc e and requests for reprints to to Professor G. Robert J. Hockey,
Department of Psychology, University of Leeds, LS2 9JT;
Email: [email protected] k
This research was supported by the UK Economic and Social Research Council’s Risk and
Human Behaviour Programme, Grant No 414206. We thank Christine Watson and Abi Hebbard for
their help in collecting data.
Ó 2000 Psychology Press Ltd
http://www.tandf.co.uk/journals/pp/02699931.html
824
HOCKEY ET AL.
INTRODUCTION
The impact of strong negative emotions on decision making and risk taking is
recognised both in anecdotal accounts (Janis & Mann, 1977; Kogan & Wallach,
1964; Mann, 1992), and in the analysis of major risk situations (Orasanu, 1997).
However, there has been little research on either the effects of more subtle
emotional changes, such as negative moods, or of decision making in more
typical everyday contexts. Moods may have greater relevance for behaviour in
everyday decision making, where extreme emotional states are relatively
uncommon. In addition, temporary mood may be expected to play a smaller part
in major life decisions, because these are normally taken over a protracted
period (buying a house, choosing a partner, deciding on a career). The present
study examines the effect of negative mood states on risk-taking behaviour in
everyday situations.
The most extensive programme of work on mood and risk has been carried
out by Isen and her colleagues (e.g., Isen & Geva, 1987; Isen & Patrick, 1983).
Their main findings are that positive moods (typically induced by small gifts)
produces risk-averse behaviour in gambling and lottery tasks. However, in ‘‘low
risk’’ tasks, where success is more likely, positive mood usually gives rise to
increased riskiness. Such results have been interpreted in terms of a mood
regulation model, which assumes a desire to maintain positive moods and to
repair negative moods. Risky decisions are thought to be rejected under positive
moods because the likely loss will upset the good mood state, whereas the likely
gain from a low risk decision would serve to enhance or maintain it. The mood
regulation interpretation has considerable appeal, though it is unable to account
for all effects of mood on cognition. For example, the maintenance of negative
states may be necessary in order to deal with the problems they signal (Martin,
Ward, Achee, & Wyer, 1993). Different mood regulation strategies (such as
rumination or distraction) may be used to either enhance or reduce negative
moods, such as depression, anxiety, and anger (Rusting & Nolen-Hoeksema,
1998; Thayer, Newman, & McClain, 1994).
Within the extensive social judgement literature, Forgas (1995) identifies
four different processing strategies in which mood effects (‘‘affect infusion’’)
might occur—direct access, motivated processing, heuristic processing, and
substantive processing. On this model, mood regulation is but one example of
motivated processing, in which decision outcomes are directed towards attaining
pre-existing goals, such as feeling good, or staying angry (Martin et al., 1993),
rather than on achieving optimal task solutions. Forgas argues that affect infusion occurs more readily in tasks that encourage heuristic and substantive processing. Mood congruent effects occur when current mood primes access to
affect-specific memory-based information (e.g., Bower, 1981; Mayer, Gashke,
Braverman, & Evans, 1992), biasing perceptions and values used in judgements.
For example, positive events are judged more frequently and negative events
NEGATIVE MOOD STATES AND EVERYDAY RISKS
825
less frequently in positive and negative moods, respectively (Johnson &
Tversky, 1983).
A second effect of mood concerns changes in processing strategies. Theorists,
such as Frijda (1986), have argued that emotion has a primary motivational
function, helping to regulate actions by signalling the presence or absence of
threat (and other) states. Models based on this perspective (e.g., Schwarz &
Bless, 1991) argue that negative moods (signalling the presence of a problem)
are more likely to promote analytic processing (the use of rational, elaborative
strategies) directed towards the source of the problem. By contrast, positive
states (signalling that all is well) give rise to simpler heuristic strategies. For
example, Forgas (1998) found that effects of induced mood on attribution errors
were strongly related to changes in information-processing style. Subjects in sad
moods performed better, making more effective use of memory for task information than controls, whereas those in happy moods made more errors and
recalled less. However, such effects of negative moods are not always found. In
one of their studies, Leith and Baumeister (1996) found that increased riskiness
in lottery choices under induced anger was effectively counteracted by
instructions to use rational decision strategies, implying that anger reduced
analytic processing.
Although most of this evidence comes from social judgements, it may also
have relevance for decision making and risk, although specific predictions are
difficult to make, particularly for the effects of negative moods. From a mood
repair perspective, people in negative moods may be expected to choose risky
options, in order to give themselves a chance of obtaining the positive outcome
that might improve their state. If negative affect acts to increase analytic processing, the choice of the safe option may be more likely, but only where this
results in resolution of the problem that initially gave rise to the state. Otherwise,
it may be directed towards a detailed assessment of the benefits and costs of the
risky outcomes. Conversely, if analytic processing is reduced under some
negative moods, risk choices may be more likely, as information will be processed less completely.
In fact, there is little systematic research on the effects of negative mood on
risk, and the findings are inconclusive. Broadly in line with the mood regulation
hypothesis, Leith and Baumeister (1996) found that a range of induced negative
states (such as anxiety and anger) increased the choice of risky options, in terms
of preferences for long-shot gambles over safe ones. However, another negative
mood, sadness, did not affect riskiness, and Pietromonaco and Rook (1987)
found that mild depression reduced the selection of risky options in everyday
decision scenarios. Taken together, these last two findings may indicate that
increased risk is found only with ‘‘high arousal’’ negative states (as suggested
by Leith & Baumeister, 1996). This may not be entirely satisfactory as an
explanation, as Mano (1992) found contradictory effects of aroused negative
moods within the same study. Individuals required to present a course report
826
HOCKEY ET AL.
(making them more tense and distressed) were more risky on one measure
(being willing to pay more for a lottery), but less risky on another (preferring a
‘‘sure thing’’ over a gamble). Overall, it seems likely that different negative
moods may have distinctive effects on risk behaviour, although the nature of
these differences has not been systematically explored to date.
A rather different source of evidence comes from research on stress and
human performance. Studies of decisional conflict (Baradell & Klein, 1993;
Janie & Mann,1977; Keinan, 1987) suggest a range of strategy changes under
stress, all associated with reductions in the amount of information used in
reaching decisions. In our own work (Hockey, 1997; Maule & Hockey, 1993),
decision making under time pressure and heavy workload is characterised by the
use of short-cuts in information processing, and reduced levels of mental effort.
In stress contexts such effects occur most strongly when fatigue is a part of the
strain response (Hockey, 1997). Shingledecker and Holding (1974) found that
subjects fatigued by up to 32 hours of continuous work gambled on low probability/low effort solutions on a post-work circuit testing task, instead of
working systematically through all possible options. Similarly, Webster, Richter
and Kruglanski (1996) showed that fatigued subjects gave a more stereotyped
response in a social judgement task, and made less use of available information
in reaching their decision.
Interestingly, there have been no direct studies of the effects of fatigue on risk
in decision making. Clearly, changes such as the foregoing are inconsistent with
the social judgement literature on negative moods (e.g., Forgas, 1998; Schwarz
& Bless, 1991), which generally predict an increased use of analytic processing.
One problem is the tendency to treat all negative states as the same, despite their
different motivational dynamics. Whereas anxiety is normally considered a
response to threat, and depression the result of loss (e.g., Frijda, 1986), fatigue is
more likely to signal a threat from overcommitment of processing resources to a
particular activity. On this view, although an analytic processing style may be
considered appropriate for anxiety and depression, it may not be for fatigue.
Feeling tired is a signal to disengage from current activity—to stop, or switch to
something else (Hockey & Meijman, 1998). At the psychological level, this is
most readily measured as a reduction in effort expenditure, persistence, and task
involvement (Hockey, 1997). Whether fatigue will elicit safe or risky actions
will then depend on the processing requirements of available alternatives. What
is needed at this stage is a study that compares the effect of fatigue and other
negative moods directly in the risky decision situation.
Although previous research has provided a valuable empirical and theoretical
base for the study of mood effects in risk behaviour, it has a number of
limitations for research on everyday decision making. For example, the widespread use of gambling and lottery tasks provide an effective way of defining
rational behaviour, but may have only limited relevance to everyday choices,
which normally have to be made in the face of uncertainty and ambiguity. In the
NEGATIVE MOOD STATES AND EVERYDAY RISKS
827
present study, we adopt an approach based on Kogan and Wallach’s (1964) use
of real life scenarios, although emphasising everyday situations rather than
major life events. Pietromonaco and Rook’s (1987) depression study is the only
one of this kind that we have encountered in the current literature.
A second issue relates to the usual practice of inducing mood states experimentally. Mood manipulation methods have the major advantage that they
permit the inference of cause-effect relationships. Induction procedures are,
however, rarely precise, typically giving rise to concomitant changes in several
emotional states (Polivy, 1981). By definition, such changes cannot be adequately measured without the use of multidimensional mood analyses. In
addition, because induction procedures are, in effect, stressors that disturb the
state of equilibrium, they may actually create an additional regulatory burden for
task management, especially when negative moods are induced. An alternative
to mood induction is to sample naturally occurring mood states, which may be
assumed to be already randomly assigned to participants (Mayer et al., 1992)
This has the advantage of avoiding transient stress effects of the induction
process. On the negative side, the lack of control with the use of natural moods
reduces the internal validity of the design, and the possibility for making causal
interpretations. We include both methods in the present approach. Studies 1 and
2 explore the broad relationships between mood and risk using naturally
occurring states. Study 3 uses a mood manipulation procedure to test our
inferred interpretation of the correlational findings through direct experimental
manipulation.
Finally, there is a possibility that mood effects are confounded with more
stable emotion-related traits. Rusting (1998) suggests that effects of mood on
cognitive processing are likely to be either moderated or mediated by personality factors, such as extraversion, neuroticism, or stable levels of positive and
negative affectivity. This has rarely been tested formally, although it has major
implications for theorising in this area. For example, models based on the
hypothesised informational function of affect (e.g., Schwarz & Bless, 1991)
assume that effects are attributable to current mood, rather than to stable traits. It
is clear that such information can be useful only if it is carried by dynamic
changes in affect, rather than by average levels.1 Experimentally induced moods
may be less vulnerable to problems of this kind, although effects may still be
moderated by trait differences, or constrained by predispositional tendencies. In
the case of natural moods, a single mood rating is likely to be strongly related to
stable mood traits, although the two influences may be teased apart by relating
measured moods to established baseline patterns. In the present study, this is
accomplished by assessing individual baseline mood over an extended period,
and using this to standardise current mood measures.
1
We are grateful to an anonymou s referee for pointing out the relevance of this.
828
HOCKEY ET AL.
Conceptual framework for mood analysis
As we have observed, most research on mood effects in risk and decision
making has adopted a unidimensional valence merit. Circumplex models indicate that measurement of differences in affect requires at least two dimensions.
One type of model (Russell, 1980) describes affect in terms of valence (pleasant–unpleasant) and arousal (degree of intensity or activation). An alternative
interpretation (Watson & Tellegen, 1985) is based on a rotation of the dimensions, to give positive affect (PA: enthusiasm and energy for life goals) and
negative affect (NA: active distress, anxiety, anger). A modification of the
rotated model (Thayer, 1989) expresses moods in terms of differentiated arousal:
‘‘energetic arousal’’ and ‘‘tense arousal’’, broadly equivalent to PA and NA.
There is still considerable debate about the relative merits of different models
(e.g., Feldman-Barrett & Russell, 1998), and detailed consideration of the issues
would be out of place here.
The rotated structure is preferred as the basis for our own research, as it
provides a more suitable framework for measuring changes in well-being under
stress or task demands (e.g., Hockey, Payne, & Rick, 1996; Hockey, Wastell, &
Sauer, 1998; Warr, 1990). In particular, it allows us to make distinctions
between different negatively valenced mood states associated with the response
to stress. In the present analysis we include three measures of negative mood:
anxiety, depression, and fatigue. Conceptually, anxiety is closely related to NA
and also to Thayer’s (1989) ‘‘tense arousal’’. Depression and fatigue are similarly related to (low) PA and Thayer’s (low) ‘‘energetic arousal’’.
The present research
The three studies presented here are concerned with the analysis of changes in
everyday risk behaviour under negative mood states. In particular, we address
the following questions. (1) Can we identify separate effects of the three
negatively valanced mood states: anxiety, depression, and fatigue? (2) To what
extent are effects of current mood attributable to stable or trait differences? We
adopt a diary-based methodology to identity individual mood patterns over an
extended period before administering the risk task. This provides a baseline for
standardising raw mood ratings, and allows us to separate the effects of state and
trait effects on risk behaviour. (3) Are there distinctive effects on risk of complex affective states, identified by conjunctions of fatigue or depression with
anxiety? (4) How are the effects of mood on risk moderated by the personal
context of decision making, as defined by within-person differences in the
familiarity, importance, and emotional impact of decision making problems?
Finally, we ask (5) whether the effects of natural moods can be replicated by
inducing a negative mood state experimentally. This provides better control and
allows causal interpretations may be made more confidently.
NEGATIVE MOOD STATES AND EVERYDAY RISKS
829
STUDY 1
Study 1 was an exploratory analysis of the mood-risk relationship. It also
allowed us to compare different assumptions about appropriate state and trait
indices of mood differences. By measuring moods at three times a day over a 14day period, we can obtain reliable individual differences at trait level (in terms
of the average levels of each mood dimension). Predictions from these stable (or
trait) measures can then be compared with those from state measures (obtained
at the time of testing). A further comparison may also be made between raw and
standardised mood scores. As the time of testing on the decision task was fixed,
we can also ask whether diary reports made at the same time of day as the task
provided a better baseline for the standardising procedure than either those made
at other times or the whole day aggregate.
METHOD
Participants
A total of 34 students (17 men and 17 women, mean age 20.2 years, range 18–
31) were recruited for the study through advertisements placed around the
university campus. They were not paid for their participation.
Materials
Risk behaviour (PRI). Risk behaviour was assessed through an instrument
specially developed for this study, the Personal Risk Inventory (PRI).2
Participants were presented with a set of everyday scenarios (in the form of
common choice dilemmas), chosen to be representative of a wide range of
situations (e.g., legal, health, social, moral, financial). These were designed to be
typical of choice situations frequently confronted by individuals in their normal
lives, based on a pilot study, in which respondents were asked to keep diaries of
frequently occurring decisions involving an element of risk. They were
instructed to imagine how they would feel in each situation, and to choose
which of two actions (A or B) they would take. For scoring purposes, one of
these was identified as ‘‘risky’’ and one ‘‘safe’’ (randomly designated as A or
B). In view of the demonstrated link between fatigue and the choice of low effort
strategies (Hockey, 1997) it was initially proposed to include two kinds of item,
differing in the implied effort required in order to carry out the safe option.
However, this goal was not pursued systematically, as it proved difficult to
match items satisfactorily for other features. The eventual set of 20 items used in
the pilot study attempted to balance effort differences across risky and safe
choice options.
2
Details of the full 13-item set of PRI scenarios may be obtained from the authors.
830
HOCKEY ET AL.
Because of the need for the PRI to measure changes in risk behaviour as well
as consistency, items were eliminated from the original set of 20 on the basis of
a preliminary study if their test-retest reliability was too high (>.8) as well as too
low (<.5). The final set of 13 scenarios had a compromise test-retest reliability
of r = .63, and an adequate level of internal consistency (Cronbach a = .74) for
this kind of study. In order to test our assumptions about effort and risk differences between options, a separate group of n = 22 participants were asked to
read through the scenarios and rate each option for ‘‘how much effort it would
involve’’ and ‘‘how much risk it would involve’’, using a 1–5 scale (1 = ‘‘hardly
any’’, 5 = ‘‘a great deal’’). Separate paired t-tests showed that, in every case,
options designated ‘‘risky’’ were rated as significantly more risky (at p < .05 or
better) than those designated ‘‘safe’’, the difference between overall means
(3.88 vs. 1.77) being highly significant, t(22) = 18.59, p < .001. For the effort
comparison, in five scenarios the safe option was rated as involving significantly
more effort than the risky option, in five others effort was higher for the risky
option, and in three others there was no significant difference. More importantly,
there was no difference between the overall (mean) perceived effort requirements of risky and safe options (2.65 vs. 2.60) across the full PRI set of 13
scenarios, t(21) = 0.49, p > .05. Thus, risky and safe choice options were
effectively balanced for effort demands. Two examples of scenarios are shown
later. In the first, choosing the safe option has a greater effort or cost, in terms of
both time and physical work. In the second example, perceived effort is rated as
higher for the risky option. The risky choice is shown by an asterisk in both
cases.
Hospital parking
You have to visit a close relation in hospital, and you manage to get away
from work for an hour at a busy time. As usual, the small visitors’ car park
opposite the hospital is full, and you know from experience that you will
probably have to wait 15 minutes or so at this time for a space. You could drive
into the staff car park, but this is occasionally patrolled by hospital security staff,
and you know that cars have been clamped.
You wonder where you should park:
(A) Use staff car park*
(B) Use visitors’ car park
Pub visit
You have been in a new job for a week and enjoy it. On the Friday, you
overhear people talking about visiting a pub together at the end of work. You
would like to get to know your colleagues better, but you have not received an
invitation to go along with them. You are unsure whether this is just an oversight
or a deliberate snub. On your way home you pass the pub where everyone is
NEGATIVE MOOD STATES AND EVERYDAY RISKS
831
meeting, and consider whether you should go straight home or call in. They may
be very pleased to see you, but it may be embarrassing, and make future work
less enjoyable.
You wonder what you should do:
(A) Go straight home
(B) Call in to the pub*
In order to obtain a more sensitive measure of choice, respondent s were asked
to indicate their strength of commitment to the selected option on a 10-point
scale (from ‘‘definitely A’’ to ‘‘definitely B’’). This provided a graded measure
of riskiness, rather than the dichotomous index of risk choices. Riskiness was
scored by averaging across the items in the set (after reverse-scoring in cases
where choice A was the risky one), so that higher values of riskiness refer to
increased endorsement of the risky alternative. Respondents were also asked to
rate how risky they perceived their decision to be in each case, using a 1–5 scale.
The average of these ratings was used as a measure of perceived risk. Finally,
they were asked to rate each scenario for three other attributes: (a) familiarity—
how familiar they were with this kind of problem; (b) importance—how
important it would be in real life for them to obtain a favourable outcome; and
(c) emotional impact—how much they were affected emotionally by the problem. A post-hoc classification of scenarios (high or low on each of the three
attributes) was carried out separately for individuals, to test for their possible
moderating role in the mood/risk relationship in Study 2.
Mood measurement. To establish a reliable basis for differentiating
individual traits and states, moods were measured over a large number of
occasions (Epstein, 1984). Participants completed a mood diary three times a
day for the 14 days preceding the PRI, and during the PRI test session itself.
They were asked: ‘‘How you have felt over the past few hours’’, with respect to
a set of 12 mood adjectives, four items (two reverse-scored) representing each of
the three negative mood measures included in the study: anxiety (Anx),
depression (Dep), and fatigue (Fat). The relatively small number of items in the
mood checklist was adopted in order to minimise the daily demands on
respondents, and encourage completion of the 14-day diary phase of the study.
Items were presented in alphabetical order in a single column (dimension in
brackets; R = reverse-scored): alert (Fat, R) anxious (Anx), calm (Anx, R),
cheerful (Dep, R), depressed (Dep), energetic (Fat, R), enthusiastic (Dep, R),
fatigued (Fat), miserable (Dep), relaxed (Anx, R), tense (Anx), tired (Fat).
Participants rated their feelings by putting a mark on a 100 mm line (labelled
‘‘not at all like this’’, at one end, to ‘‘very much like this’’ at the other). Scale
reliabilities from accumulated use of these measures over a number of current
studies are acceptable (Cronbach a for anxiety = .78, depression = .88, fatigue =
.83). Scale scores were obtained for the three dimensions, for each diary day, by
averaging over the four items, to give a value on a 0–100 scale.
832
HOCKEY ET AL.
Procedure
Participants were instructed to complete the mood diary over a 14-day period,
three times on each day (around 08.00 h, 13.00 h and 18.00 h). This allowed us
to obtain reliable estimates of trait differences in affect (based on averaged
reports) and to compare different indices of state changes, using the diary
sample to calibrate reports at the time of the PRI. Approximately one week
following completion of the diary, participants were invited to the laboratory to
complete the PRI. Testing took place in small groups of 2–4 persons during the
early afternoon (between 14.00 h and 16.00 h). This involved an orientation
briefing, in which they were introduced to the PRI and reminded that they
should imagine themselves in the various situations represented in the 13 scenarios. They were taken through three practice items, emphasising the roleplaying aspect of the task, and any queries answered. Mood state was assessed
by administration of the standard mood questionnaire at the end of the practice
period, and before the test scenarios were encountered.
Data analysis
Effects of mood on risk behaviour were assessed by carrying out three separate
analyses, designed to separate the contributions of trait and state effects. First,
we tested for possible effects of stable (trait) mood level by relating PRI risk
scores to the average level on each of the three mood dimensions for each
participant, over the whole 14-day period of the mood diary (mean Anx, Dep,
and Fat). Second, as the primary focus of the study was on effects of state
changes, we assessed the effects of between-person variations in mood at the
time of PRI testing. A conventional analysis of raw scores related individual risk
behaviour to the observed mood scores at the time of completion of the PRI (raw
Anx, raw Dep, and raw Fat). Such scores are, however, likely to be influenced
by individual differences in baseline mood and reporting biases. Consequently, a
third analysis was carried out in which effects of stable between-person differences in mood level and variability were removed. For this, raw scores for
each dimension were adjusted separately for each person by standardising with
respect to their 14-day baseline of diary reports (i.e., subtracting the mean and
dividing by the standard deviation). This procedure removes all between-person
effects, leaving only within-person variability, in the form of standardised (z)
state mood scores (zAnx, zDep, zFat).
For both average mood and standardised scores, four different sets of diary
data were available: the entire set of 42 observations (mean Anx-42, etc.) and
the samples of 14 observations for each of the three times of day (mean Anx-14,
etc.). There are various possible baselines for the standardising procedure.
Means for the 42-occasion sample may be best, since they are more representative of overall trait differences, and more reliable. However, means for
different times of day were assumed more sensitive to individual differences in
NEGATIVE MOOD STATES AND EVERYDAY RISKS
833
diurnal mood patterns, based on personal lifestyle (times of meals, sleep habits,
structure of work, and social activities). In particular, because the PRI was
administered in the early afternoon, the sample of lunchtime reports was predicted to provide a more appropriate baseline. For the computation of average
mood, both the full set (based on 42 reports) and the midday sample (based on
14 reports) were expected to be the most meaningful indices of stable (trait)
differences in mood.
RESULTS AND DISCUSSION
Comparison of mood baselines
As expected, the average of the midday diary reports proved to be the best
predictor of mood scores at the time of testing for all three mood dimensions: for
Anx, r = .56; Dep, r = .60; Fat, r = .45 (all ps < .01), and it was used as the basis
of the standardisation procedure. Despite having less face validity, the greater
reliability of the whole day average meant that it was also a reasonable predictor
of test mood, for all three dimensions (Anx, r = .37, p < .05; Dep, r = .65, p <
.01; Fat, r = .42, p < .05). For the analysis of trait effects, both midday (Anx-14,
etc.) and whole day (Anx-42 , etc.) averages were included.
Descriptive and correlational analysis
As described earlier, correlational analyses were carried out between risk
measures and three different measures of mood: stable (trait) differences, and
two measures of state changes—raw scores and standardised scores. These are
reported separately later.
Effects of trait mood on risk. The trait mood analysis was based on the
average levels of reported Anx, Dep, and Fat over the 14-day period of the mood
diary. An estimate of the reliability of these measures for each of the mood
dimensions may be obtained by considering each rating as an item in a 14- or
42-item scale. This procedure gives Cronbach a values of between .71 and .94
(slightly higher for the 42 occasion samples), indicating an acceptable degree of
stability in the day-to-day differences between respondents. Table 1 shows the
outcome of the correlation analysis for the two average mood measures from the
diary phase of the study and risk behaviour during the laboratory test.
Table 1 indicates that risk behaviour on the PRI does not depend to any great
extent on stable or trait differences in mood, at least as measured by average
midday mood over the 14-day diary period. Possibly because of their greater
reliability (Epstein, 1984), there is a stronger relationship with the trait measures
based on 42 reports, particularly for Dep and Fat, which correlate positively with
riskiness. The correlation for mean Fat-42 is significant (r = .36, p < .05). Thus,
participants who are generally more fatigued (less energetic) over the two-week
834
HOCKEY ET AL.
TABLE 1
Summary statistics for stable mood measures and
correlations between mood and risk variables
(Study 1)
Correlation with:
Stable mood measure
M
SD
Risk
P-Risk
Mean Anx-14
Mean Dep-14
Mean Fat-14
34.2
48.6
41.1
11.4
12.5
10.0
.10
.18
.03
.11
7.04
7.07
Mean Anx-42
Mean Dep-42
Mean Fat-42
35.3
54.2
46.6
10.5
9.2
0.3
.17
.24
.36*
.13
.17
.21
Note: The table shows the product-momen t correlation of
mood variables with Risk (riskiness rating) and with P-Risk
(perceived riskiness rating).
*p < .05.
period show a slightly stronger preference for risky options in the PRI. There
were no correlations of stable mood with perceived risk. There is a problem in
interpretation, however, because of intercorrelations between the three mood
averages: anxiety and depression, r = .50; anxiety and fatigue, r = .70, fatigue
and depression, r = .65, (all ps < .01). Controlling for the effects of anxiety and
depression has the effect of slightly reducing the correlation between Fat-42 and
riskiness, to a value which is no longer significant (partial r = .33, p > .05).
Effects of state mood. The main focus of the study is the influence of state
mood differences (i.e., how people felt at the time of PRI testing). As described
earlier, separate analyses were carried out using both raw scores and
standardised z-scores (based on the midday baseline data).
The overall pattern for state mood is similar to that for trait measures
(Table 2). Again, state depression and fatigue show positive correlations with
riskiness, significant for both raw Fat (r = .56, p < .001) and zFat (r = .50, p <
.01). There is no effect of state anxiety, and no correlations of any mood
measure with perceived risk. Thus, riskiness is associated with both a high stable
level of fatigue, and a relatively high state of fatigue at the time of testing.
Again, it is necessary to control for the small correlations between the three state
measures (r-values between 0 and .4), for both raw and standardised scores.
Partial correlation analysis confirms that the precedence of fatigue as a predictor
of riskiness is not dependent on these relationships. Correlations of mood with
riskiness are little changed from those reported in Table 2 when effects of other
measures are partialled out (partial r-values for zAnx, r = 2.10; zDep, r = .16;
zFat, r = .45).
NEGATIVE MOOD STATES AND EVERYDAY RISKS
835
TABLE 2
Summary statistics for state mood measures and
correlations between mood and risk variables
(Study 1)
Correlation with:
Stable mood measure
M
SD
Risk
P-Risk
raw Anx
raw Dep
raw Fat
51.0
45.8
46.1
27.3
21.1
26.7
7.07
.29
.56***
2.04
.09
.20
zAnx
Dep
Fat
0.42
0.33
0.17
1.14
0.95
1.24
7.01
.28
.50**
2.15z
2.03z
.20
Note: The table shows product-momen t correlations of
mood variables with Risk (riskiness rating) and with P-risk
(perceived riskiness rating).
**p < .01; ***p < .001.
The pattern of results in Tables 1 and 2 suggests a stronger effect of state
than of trait mood, and of state fatigue in particular. The standardised scores
offer strong evidence of the influence of a separate state effect, because they are
statistically independent of individual differences in overall level and variability
of reported mood. However, there may still be a functional link between trait
and state measures. Specifically, individuals who generally feel tired may be
more strongly predisposed to take a risky option when they are tired on a given
occasion. A simple test of this, based on a median split of mean Fat-42 ratings
(n = 17 per group), showed that the zFat/risk correlation was higher for participants who were habitually energetic (r = .68), and for those who were habitually tired (r = .44), although the two correlations did not differ significantly (z
= 0.95, p > .05). It is also possible that individuals who experience large mood
changes may be more susceptible to effects of state mood on risk.3 A similar
analysis to the foregoing based on standard deviation measures over the 42
occasions, again showed no differences in the zFat/risk correlation for subjects
with highly variable (r = .41) and less variable (r = .47) fatigue ratings (z =
0.18, p > .05).
Although Study 1 was largely exploratory, it suggests that risk taking in
everyday decision making may be affected by naturally occurring mood changes. In particular, fatigue at the time of testing is associated with higher levels of
riskiness in terms of choice of action, but not with perceptions of risk associated
with that choice. These effects were independent of stable differences in mood
3
We are grateful to an anonymou s reviewer for suggesting this analysis.
836
HOCKEY ET AL.
levels, which had little effect themselves on risk behaviour. In interpreting these
results in terms of naturally occurring mood changes, however, we need to
consider the possibility that testing under laboratory conditions may induce a
distinctive state of mild stress or test anxiety (e.g., Wine, 1971). Examination of
Table 2 shows that the standardised values observed in the laboratory testing
phase of Study 1 differed from the expected value (z = 0) for all mood
dimensions. Single sample t-tests showed that both zAnx: z = 0.42; t(33) = 2.16,
p < .05; and zDep: z = 20.32; t(33) = 1.98, p = .06, were higher than expected,
though the testing situation did not affect zFat (z = 0.17; t < 1). Thus, the effect
of state fatigue on risk may be partly accounted for by the induction of generally
higher levels of state anxiety and depression by the testing session.
STUDY 2
Study 2 was carried out to examine the robustness of these findings, and to
assess the role of possible moderator variables in the fatigue-risk relationship. In
addition, Study 2 includes a separate within-study replication, by administering
the PRI on two separate occasions over the diary period. A further procedural
difference is influenced by the observation of mildly increased stress in the
testing situation of Study 1. Study 2 investigates whether the same mood/risk
relationship effect is observed in a more naturalistic context. This was achieved
by arranging for subjects to complete the diary and the PRI in their own homes.
Finally, in addition to using larger samples of participants, the mood diary phase
was extended to 28 days. This provided a more reliable baseline and allowed the
PRI to be administered twice, two weeks apart.
There were two further goals of Study 2. The first was to examine the
possible moderating effects on the fatigue-risk relationship associated with other
aspects of the decisional context. Mood effects may be expected to depend on
the personal significance of the problem scenario. More important or more
familiar problems may be less vulnerable to the unpredictable effects of mood
than less important or familiar scenarios, for example because of a stronger
tendency to use direct access strategies (Forgas, 1995). To test this, we obtained
ratings of the familiarity and personal significance of scenarios, and used these
as moderators of the overall mood-risk effects. We also measured the overall
emotional impact of scenarios, in order to assess the possible interaction
between background and problem-generated effects. A final goal of Study 2 was
to examine the effects of more complex mood states. In particular, we were
surprised that state anxiety did not affect risk behaviour in Study 1, because it is
recognised as a major threat to decision making (Eysenck, 1982; Wine, 1971).
One possibility is that anxiety may interact with changes in fatigue or depression
to produce distinctive ‘‘high risk’’ mood states. We tested this by analysing risk
measures in relation to combinations of high and low levels of state fatigue and
anxiety.
NEGATIVE MOOD STATES AND EVERYDAY RISKS
837
METHOD
Participants
A total of 67 students (29 men, 38 women) were recruited by advertisements
placed around the campus. None had taken part in Study 1. They were not paid
for their participation, but received feedback at the end of the study on their
mood patterns. They were asked to complete a number of questionnaires (for the
purposes of other concurrent studies), so were not aware of any explicit link
between completion of the mood diary and responses to the PRI. Nine participants (4 men, 5 women) had to be dropped from the full analysis, because of
failure to complete at least 21 days of diary data, leaving a sample of n = 58 for
most analyses. Two of these did not complete the second PRI test.
Procedure
The mood diary and PRI were the same as those used in Study 1. Participants
completed mood diaries each day for 28 days, and the PRI at the end of weeks 1
and 3. They were instructed to complete diaries at the end of each afternoon
(17.00 h–19.00 h), followed by the PRI on day 7 (PR11) and day 21 (PR12).
PRIs were posted to participants on the Thursday of the preceding week (5 days
and 19) by first class post. (This normally gives delivery the next day, and
almost always within 2 days). A fixed day (Saturday) for completing the PRI
was adopted in order to control for possible day of the week effects in moods.
Saturday was selected because there is evidence of greater variability of moods
at weekends than during weekdays (Stone, Hedges, Neale, & Satin, 1985), thus
maximizing the range of possible mood changes that might be observed across
the sample. It also allowed for the possibility of completion on Sunday if
necessary. In fact, of the 114 PRIs returned (2 were missing from the second
set), 4 were completed on Fridays, 88 on Saturdays, 10 on Sundays, 9 on
Mondays (sometimes because of late postal deliveries), and 3 on other days.
Since there was no evidence of any differences between the data for Saturday
and other days, all data were used in the analyses.
Data analysis
Three separate analyses of the effects of mood on risk were again carried out, in
order to assess the separate effects of trait and state components of moods. Trait
mood was measured by taking the average levels of moods over the entire 28day diary period (mean Anx, etc.). For mean Anx, Dep and Fat, Cronbach a
values (treating days as items) were .87, .91, and .83, respectively. As in Study
1, we also included both raw mood scores at the time of PRI completion and
standardised mood, based on the 28 days of mood reports.
838
HOCKEY ET AL.
RESULTS AND DISCUSSION
Descriptive and correlational analysis
Effects of stable mood. Table 3 shows the outcome of the correlation
analysis for stable mood and risk behaviour for both PRI occasions. The pattern
of results is very similar. They show a small negative correlation of riskiness
with mean Anx, significant for PR1-2 (r = 7.34; p < .05) but not fir PRI-1 (r =
7.20, p > .05), and a positive correlation for mean Fat, significant on the first
occasion (r = .33, p < .05) but not the second (r = .26, p > .05). There is no
indication of an effect of mean Dep, and no correlations with perceived risk.
Unlike Study 1, these results hint at an effect of trait anxiety on risk behaviour. Being generally anxious appears to make individuals act more cautiously
in the risk scenarios. One problem of interpretation is the high correlation
between mean Anx and mean Dep (r = .82 in Study 2, compared to r = .50 in
Study 1). By contrast, in the present study, mean Fat is relatively independent of
both mean Anx (r = .20, p > .05) and mean Dep (r = .10, p > .05). The reason for
these differences is unclear, though the more frequent mood reporting of Study 1
may have made participants more aware of mood differences, and reduce their
focus on valence judgements (Feldman-Barrett & Russell, 1998). A partial
correlation analysis, controlling for the other moods, removes the apparent
correlation between riskiness and both mean Anx and mean Dep, for both PRI-1
(r = .01 and 7.09) and PRI-2 (r = 7.15 and .02). However, trait fatigue
maintains its small relationship with riskiness, significant at PRI-1 (r = .31, p <
.05), although still not at PRI-2 (r = .25, p > .05). Overall, the findings show that
trait mood has only small effects on risk behaviour.
TABLE 3
Summary statistics for stable mood measures and correlations between mood and risk
for each testing occasion (Study 2)
PRI-1 (n = 58)
Stable mood
measure
Mean Anx
Mean Dep
Mean Fat
PRI-2 (n = 56)
Correlation with:
Correlation with:
M
SD
Risk
P-Risk
M
SD
Risk
P-Risk
40.8
48.1
44.3
16.2
11.1
10.8
7.20
7.131
.33*
7.04
7.11
.06
40.4
48.3
54.1
15.7
11.6
10.7
7.34*
7.23
.26
7.15
.05
7.01
Note: The table shows the product-momen t correlations of mood variables with Risk (riskiness
rating) and P-Risk (perceived riskiness rating). The slight differences in n, M, and SD values between
PRI-1 and PRI-2 reflect the loss of data for two participants in the PRI-2 analysis.
*p < .05.
NEGATIVE MOOD STATES AND EVERYDAY RISKS
839
TABLE 4
Summary statistics for state mood measures and correlations between mood and risk
variables for each testing occasion (Study 2)
PRI-1 (n = 58)
Stable mood
measure
raw Anx
raw Dep
raw Fat
zAnx
zDep
zFat
PRI-2 (n = 56)
Correlation with:
Correlation with:
M
SD
Risk
P-Risk
M
SD
Risk
P-Risk
37.7
45.3
53.6
24.0
17.3
21.8
7.26
7.12
.32*
.16
7.05
.25
38.8
45.9
52.4
20.6
17.0
18.1
7.24
7.08
.35*
7.17
7.14
7.01
70.26
70.26
0.06
0.99
0.87
0.93
7.06
.06
.33*
.18
.03
.22
70.20
70.28
70.17
0.74
0.74
0.70
.12
.16
.44
7.07
.01
7.03
Note: The table shows product-momen t correlations of mood variables with Risk (riskiness
ratings) and P-Risk (perceived riskiness ratings). The slight differences in n, M, and SD values
between PRI-1 and PRI-2 reflect the loss of data for two participants in the PRI-2 analysis.
*p < .05, **p < .01.
Effects of state mood. The analysis of state effects again considered both
raw scores and standardised z-scores. The results are summarised in Table 4.
The only significant correlations, on both PRI occasions, are between riskiness
and fatigue. This applies to both raw Fat (r = .32 and .35 for PRI-1 and PRI-2,
both p < .05), and zFat (r = .33, p < .05; r = .44, p < .01). As with Study 1, risky
options are more strongly endorsed when participants are tired. State anxiety
shows only a nonsignificant negative correlation with riskiness (r = 7.26 and
7.24), and there is no sign of an effect of state depression. Again, there were no
effects on perceived risk.
Correlations between the mood states again need to be considered. Unlike
average mood levels, raw Anx and raw Dep were only moderately correlated (r
= .58 for both PRI occasions), and raw Fat was relatively independent of both
raw Anx (r = .06), and .11) and raw Dep (r = .28 and .33). Standardised scores
showed an even smaller correlation between anxiety and depression (r = .46 and
.38). Again, zFat was independent of zAnx (r = .25 and 7.20), although there
was a higher correlation between zFat and zDep (r = .44 and .46). Partial
correlation analyses confirmed that the effect of state fatigue on riskiness is not
affected by its correlation with other mood measures (for raw Fat, r = .34 and .36
for the two testing occasions; for zFat, r = .34 and .43) They also confirmed the
lack of association between riskiness and state anxiety/depression (r < |.15| in all
cases).
Both raw and standardised scores thus show that effects of state mood on
riskiness are best captured by variations along the energy-fatigue dimension,
840
HOCKEY ET AL.
rather than by changes in anxiety or depression. As noted earlier, raw scores
inevitably include a contribution from stable mood differences. By controlling
for stable individual differences in reported mood, the standardised score analysis shows that there is a genuine effect of intra-individual changes in state
fatigue. The two separate testing occasions serve as a replication, showing the
same overall pattern of effect. Furthermore, although raw mood scores show
significant correlations between PRI-1 and PRI-2 (r = .4 to .5, all p < .05),
correlations between standardised mood scores on the two occasions were very
small (all r < .2). This strongly suggests that the effect of mood on risk behaviour is also independent of individual-specific responses to the self-testing
requirements of the PRI. Different participants varied in tiredness or energy on
the two occasions, and their level of riskiness changed accordingly.
It may be argued that the use of a specific day (Saturday) for the PRI testing
may compromise the external validity of the findings. Although there is little
direct evidence for ‘‘day of week’’ effects in previous mood diary research (e.g.,
Stone et al., 1985), our ongoing diary studies have consistently observed lower
levels of strain (decreased levels of all three negative moods) at weekends. For
the present data, Table 4 indicates that the overall level of mood is not greatly
changed on PRI days (all z-values close to the expected value of 0). Nevertheless, one-sample t-tests show that, with the exception of zAnx at time 1, all
standardised values of strain are significantly lower (one-tailed tests, p < .05 or
better). As a group, participants were slightly less anxious and tired, and clearly
less depressed at the weekend, when they carried out their PRI tests.
Taken in conjunction with Study 1, where the opposite effects on mood state
were induced by the testing situation, we may reasonably conclude that the
effects of state fatigue on risk behaviour, as observed on three separate testing
occasions, are relatively robust. The cross-sectional analyses show that these
effects are independent of both general mood levels and inter-correlations
between mood variables. However, a more stringent test of the effect of state
changes is afforded by the quasi-longitudinal design of the study. Hierarchical
multiple regression may be used to test whether changes in mood from time 1 to
time 2 are associated with corresponding changes in rise behaviour. In particular, we tested the hypothesis that an increase in zFat between the two sessions
gives risk to an increase in riskiness.
To do this, change scores were computed, for the 56 participants who
completed both PRI sessions, for each mood variable (z 2–1Anx, z2–1Dep,
z2–1Fat) and for the dependent variable, riskiness (risk2–1). This was done by
subtracting the values for time 1 from those for time 2. Although there was only
a weak correlation between riskiness on the two PRI tests (r = .31, p < .05), we
controlled for individual preferences for risky options by entering riskiness at
time 1 (risk1) as the first step. In the next, we controlled for the standardised
mood scores at time 1 (z1Anx, z1Dep, and z1Fat). These may be correlated with
NEGATIVE MOOD STATES AND EVERYDAY RISKS
841
time 2 moods, because of characteristic lifestyle influences on mood patterns.
Finally, in step 3, we included the change scores for the three standardised
negative mood variables. The effect of interest concerns the last step, which
predicts a significant increase in R 2 for the combined effects of mood change,
and, in particular, a significant positive value for the effect of change in fatigue
(z2–1Fat). The results are summarised in Table 5.
The results of step 3 confirm that an increase in state fatigue between the two
PRI sessions is associated with an increase in riskiness (b = .574, p < .01). It also
confirms the absence of effects associated with other mood variables. An overall
indication of the size of the effect associated with the change in mood state over
the two-week period is given by the increase of 13% in variance accounted for
by step 3 (p < .05). In the absence of an overall change in riskiness, or of ceiling/
floor effects, the highly significant negative coefficients for Risk1 may be
assumed to reflect the influence of regression to the mean, commonly observed
in longitudinal studies. Subjects who are highly risky at time 1 are generally less
risky at time 2, and vice versa. In addition, there is a significant effect of
entering time 1 state fatigue at step 2 (b = .336, p > .05). This is difficult to
interpret, but it implies an increase in riskiness for participants who were more
tired at time 1. Because the zero-order correlation between zFat1 and risk2–1 is
negligible (r = .136, p > .05), this implies the operation of suppressor effects at
steps 1 and 2 (Cohen & Cohen, 1983).
TABLE 5
Summary of hierarchical regression analysis of riskiness
(Study 2)
Predictors
b(Step 1)
b(Step 2)
b(Step 3)
Step 1: Riskiness at time 1
Risk1
7.554***
7.644***
7.557***
Step 2: Mood at time 1
z1Anx
z1Dept
z1Fat
.018
7.046
.336*
Step 3: Change in mood
z2–1Anx
z2–1Dept
z2–1Fat
R2
DR2
.307***
.402***
.095
.078
.046
.681*
.135
7.023
.574**
.534***
.132*
Note: The table shows standardised b-weights for each step of the analysis.
* p < .05, ** p < .01, *** p < .001.
842
HOCKEY ET AL.
Effects of scenario characteristics
In addition to making decisions about whether to take risky or safe options,
participants also rated each scenario on a 1–5 scale for degree of familiarity,
personal significance, and emotional impact. Each of these may have an effect
on the way in which risk is interpreted or acted on, and moderate the effect of
background states, such as fatigue and anxiety. In terms of main effects of these
factors, we may expect more familiar everyday problems to reduce uncertainty
about outcomes, so be considered less risky in absolute terms. This would
predict an increased preference for the choices designated ‘‘risky’’ in our scenarios, with a corresponding decrease in perceived risk. Scenarios rated as more
important are hypothesised to give rise to lower riskiness, as the outcome is
perceived as more critical, and losses more imaginable (Arkes, Herren & Isen,
1988). Finally, the rated emotional impact of the scenarios is taken to be an
index of the emotion generated by the problem itself. As such, it may be
interpreted as a signal of potential loss and give rise to more cautious decision
making (Damasio, 1993). The effects of these three variables are examined in
turn later, in relation to state fatigue.
In order to increase the power for these analyses, data from the two studies
were combined (considering only the first PRI session of Study 2), and treated as
a single sample of n = 92. Mean ratings on all three characteristics were in the
middle of the 1–5 range, with no marked asymmetry: emotional impact 2.93 (SD
= 0.61), familiarity 2.73 (0.62), importance 3.14 (0.51). For each analysis scenarios were sorted on a within-person basis into ‘‘high’’ (ratings of 4 or 5) and
‘‘low’’ (1 and 2), with scenarios rated 3 omitted. Riskiness scores for high and
low items were averaged for each person. For some participants very few items
were rated in a particular category (notably low familiarity, high importance,
high emotion), so that riskiness scores were sometimes based on very small
numbers. To avoid using measures of very low reliability, only averages based
on at least 4 items at both low and high levels were included, resulting in a loss
of 3–6 subjects for different analyses. Each of the three dichotomous scenario
variables was included in separate mixed model ANOVAs, with state fatigue
(defined as a three-level between-subjects factor by recoding zFat scores as low,
medium, or high). The pattern of results for riskiness is shown in Figure 1. In all
three cases, the analyses confirm the main effect of zFat shown by the correlational data. (For different analyses, degrees of freedom for the error term
varied between 80 and 83, and F between 7.02 and 12.10; all ps < .01.) The main
interest in these analyses was in the main effect of the scenario characteristic and
interactions with zFat.
Familiarity of scenarios. Riskiness of decisions was greater for scenarios
rated as more familiar than those rated unfamiliar (5.59 vs. 4.89). There was also
a difference in perceived risk for familiar and unfamiliar items (2.22 vs. 2.41).
843
Figure 1. Riskiness ratings as a function of level of state fatigue and scenario characteristics: left panel, familiarity (fam); centre, importance (imp); right,
emotional impact (emot).
844
HOCKEY ET AL.
Thus, participants made more risky choices for familiar scenarios, but perceived
these actions to be less risky than those for unfamiliar events. Analysis of the
riskiness data (Figure 1, left panel) confirms the highly significant main effect of
familiarity, F(1, 84) = 20.77, p < .001. Although the effect of zFat appears
greater at high familiarity the interaction term was not significant, F(2, 84) =
1.89, p > .05.
Importance of scenarios. Riskiness was lower for scenarios rated as more
important (4.80 vs. 5.77), though there was little difference in perceived risk
(2.26 vs. 2.32). The centre panel of Figure 1 shows the highly significant main
effect of importance on riskiness, F(1, 81) = 66.50; p < .001, but also the
interaction between importance and fatigue, F(2, 81) = 5.59; p < .01. Contrary to
expectations, the overall effect of increased riskiness under state fatigue is
stronger for scenarios rated as more important to the individual. Fatigue may be
interpreted as reducing the normally higher level of caution with which more
important decisions are taken.
Emotionality of scenarios. There were no large differences between
scenarios which had a strong or a weak emotional impact, either for riskiness
(5.22 vs. 5.35) or perceived risk (2.32 vs. 2.36). Figure 1 (right panel) suggests
that riskiness is reduced under the highest level of fatigue for items which attract
a stronger emotional response. However, the analysis shows that there were no
overall effects of emotionality, and no interaction with zFat (both Fs < 1).
Effects of combined mood states
A final analysis was carried out to investigate the possible effects of different
combinations of mood dimensions. Although high levels of state fatigue have
shown the most consistent association with riskiness, there was evidence in both
studies of covarying changes in anxiety. Some theorists (e.g., Thayer, 1989;
Tomkins, 1963) have argued that emotional states are dynamically interdependent, so that the implications for a person’s behaviour of high levels of
depression or fatigue may depend on whether anxiety is also present. In addition,
high anxiety may provide a more distinctive and enduring emotional context,
augmenting effects of other changes (Taylor, 1991). Matthews and Westerman
(1994) found an interaction between energy and tension [equivalent to our (low)
fatigue and anxiety] on visual and memory search tasks. Expressed in the terms
used here, anxiety impaired search performance, but did this more when subjects
were less fatigued. In the context of decision making, it is not possible to make
strong predictions about such interactions. Anxiety may predispose participants
to avoiding the negative consequences of risky choices, and so reduce the riskpromotion effects of fatigue. Alternatively, we may expect it to further increase
distraction from task processing, and therefore add to effects already present.
NEGATIVE MOOD STATES AND EVERYDAY RISKS
845
In view of these possibilities, we decided to test the effect on the fatigue-risk
relationship of differences in the level of state anxiety, again using the combined
sample of 92 participants from Studies 1 and 2. Each of the state mood variables
(zFat and zAnx) was recoded to give subgroups differing in combinations of
energy and anxiety at the time of testing. Again, zFat was defined as low,
medium, or high, however, because of the low numbers in some cells, zAnx
could only be recoded as low or high. The two classifications of state mood were
crossed to give six subgroups, and the effects on risk behaviour assessed through
a 362 analysis of variance.
The results for riskiness are shown in Figure 2. Consistent with the correlational data, there was a strong overall effect of zFat, F(2, 84) = 10.41, p < .001,
and no main effect of zAnx F(1, 84) = 2.00, p > .05. There was, however, a small
but significant interaction, F(2, 84) = 3.30, p < .05. As can be seen from Figure
2, the increase in risk with fatigue has a different pattern for low and high
anxiety groups. For less anxious individuals, riskiness is increased at moderate
levels of fatigue. However, for individuals who are more anxious at the time of
testing, most of the effect on riskiness occurs in the transition from medium to
high levels of fatigue. Thus, although state anxiety has no overall effect on risk
Figure 2.
Riskiness ratings as a function of state fatigue and state anxiety.
846
HOCKEY ET AL.
behaviour in these studies, it appears to moderate the effects of fatigue. The
pattern of results is consistent with the view that state anxiety predisposes
individuals towards risk avoidance. Only at the highest level of fatigue is this
effect nullified.
STUDY 3
The results of the first two studies show a large measure of agreement. They
reveal stable effects of variations in state fatigue on risk behaviour. However,
because they are based on naturally occurring mood states, they leave open the
possibility that effects are due to other factors that covary with the observed
mood changes. In order to check our interpretation, it is necessary to carry out an
experimental study in which mood is manipulated. As we have already indicated, mood induction procedures may give rise to separate methodological
problems, notably that state changes are often more complex than intended
(Polivy, 1981). Nevertheless, they clearly provide much stronger internal
validity than is possible with natural moods. In Study 3 the effect of mood
changes on riskiness is investigated by using a manipulation designed to
increase fatigue. However, because the most effective fatigue manipulations
inevitably involve sustained periods of demanding work, it is also likely that
anxiety (and possibly depression) will be increased. In this case, such effects
may be partialled out by treating them as covariates.
METHOD
Participants and design
A total of 55 management trainees (32 men, 23 women), aged between 23 and
30 (mean 25: 4) years, were recruited for the study, representing all the participants in two successive five-day professional development courses for a large
chemical engineering company. Because it was not possible to assign subjects
randomly to treatment groups, the study effectively adopts a nonequivalent
control group design (Cooke & Campbell, 1979). The first course group (n = 33)
was designated the experimental group, receiving a mood-induction programme
designed to increase fatigue. The second group (n = 22) acted as a control group,
receiving no mood manipulation. A pre–post design was incorporated, in order
to control for baseline mood changes from day 1 to day 2. The analysis was
therefore carried out on post–pre change scores in PRI and mood scores.
Procedure
All participants, in both groups, completed the PRI at the end of day 1 and again
at the end of day 2, following either the mood manipulation or a period of free
activity. The mood induction procedure was given as part of the training course
NEGATIVE MOOD STATES AND EVERYDAY RISKS
847
for the experimental group. It was carried out on the afternoon of the second
day, with the PRI test and mood questionnaire immediately afterwards (as part
of a batch of questionnaires, unconnected with the study). Trainees were given a
set of demanding managerial planning exercises, lasting for about one hour, with
little feedback or opportunities to exercise control. The exercises required the
group to act as a team in carrying out a set of practical tasks, which could not be
completed successfully because of ‘‘unexpected’’ problems and constraints. In
previous work with management courses, this procedure has been found to
induce a high level of stress, particularly a large increase in fatigue. In contrast,
the control group spent the time in private study and unstructured group discussions. In order to obtain individual baselines for mood measures, all participants were asked to complete mood diaries for three weeks, as part of a
‘‘follow-up’’ to the training course. This took place about a month later. They
were not paid for their participation, but received feedback at the end of the
study (following the diary collection) on their mood pattern and its relation to
risk taking.
RESULTS AND DISCUSSION
Manipulation check
Table 6 summarises the pre- and post-manipulation data for mood and risk
variables, as well as the change scores. There was a strong overall effect of the
mood manipulation. Change scores showed a market effect of the manipulation
for the experimental group (shifts of around half a standard deviation for zAnx
and zFat, and slightly less for zDep). There were no strong effects for the control
group. To test for the effect of the manipulation, change scores for the three
mood variables were subjected to a multivariate analysis of variance
TABLE 6
Mean pre- and post-test levels of moods and risk measures for control (C) and
experimental (E) groups, with change (post± pre) scores (Study 3)
Control
Measure
zAnx
zDep
zFat
Risk
P-Risk
Experimental
Pre
Post
Change
Pre
Post
Change
E–C
(change)
.13
7.15
7.02
.24
.02
7.24
+.11
+.17
7.22
.17
7.15
7.04
.74
.25
.46
+.57
+.40
+.50
+.46
+.23
+.72
5.46
2.32
5.50
2.26
+.04
7.06
5.32
2.38
5.83
2.51
+.51
+.13
+.47
+.19
848
HOCKEY ET AL.
(MANOVA) with group as the between-subjects factor. The multivariate test
was highly significant on all criteria, F(3, 51) = 8.71, p < .001. Univariate tests
revealed significant differences between conditions for zFat: F(1, 53) = 18.97, p
< .001, and zAnx: F(1, 53) = 12.75, p < .001, but not for zDep: F(1, 53) < 1. (As
can be seen in Table 6, although there is an increase of .34 in zDep for the E
group, there is also an increase of .17 for group C.) Thus, following the mood
induction procedure, subjects in the experimental group were more tired and
anxious, but not more depressed, than the control group.
Effects of induced fatigue on risk behaviour
Table 6 also shows that the mood manipulation produced a marked increase in
risk, as well as an apparent increase in perceived risk. A MANOVA carried
out on the change scores for riskiness and perceived risk showed an overall
multivariate effect of the manipulation: F(2, 52) = 8.45, p < .001, with significant univariate effects on both riskiness: F(1, 53) = 16.48, p < .001, and
perceived risk: F(1, 53) = 4.62, p < .05. The main goal of the manipulation
was to induce fatigue, but as reported earlier, there were inevitably concomitant changes in other mood variables, which may have contributed to the
observed effects. In order to test for this, MANCOVA examined the effect of
conditions on the risk changes, with zAnx and zDep change entered as covariates. Multivariate tests confirmed the overall effect of conditions: F(2, 50) =
5.36, p < .01, and univariate tests that the effect was significant for riskiness:
F(1, 51) = 10.32, p < .01. Although perceived risk also increased as a result of
the manipulation, the univariate effect was not significant: F(1, 51) = 3.14, p >
.05. In addition, there were no effects of either covariate on multivariate or
univariate tests (all ps > .05). As a final check, zFat change was entered as the
covariate instead. In this case, the effect of conditions was no longer significant on the multivariate test: F(2, 51) = 3.05, p > .05, but there was now
an effect of the covariate itself: F(2, 51) = 3.60, p < .05. Univariate tests again
showed that this was confined to riskiness: F(1, 52) = 7.02, p < .05, with no
effect on perceived risk (F < 1).
In addition, as in Studies 1 and 2, there was also a correlation between risk
and fatigue at time 1. For the combined group on the pre-test, riskiness correlated with zFat (r = .34, p < .05), but not with zAnx (r = .09) or zDep (r =
7.19). These results confirm the relatively strong effect of state fatigue on
riskiness, in comparison with other negative mood states. However, it should be
noted that the average post-induction level of anxiety in the experimental group
was high (z = 0.74, compared to 0.4 and 70.2 in Studies 1 and 2). Taken in
conjunction with the results illustrated in Figure 2, it remains an interesting
possibility that a high level of background anxiety facilitates increased risk
taking under fatigue.
NEGATIVE MOOD STATES AND EVERYDAY RISKS
849
GENERAL DISCUSSION
The findings from the three studies show that the degree of risk taken in
everyday decision making may be affected by variations in state mood. In
addressing the five research questions raised earlier, they go some way towards
understanding how mood affects risk, although they also raise a number of new
questions. In all studies, and for all comparisons, the observed risk effects apply
primarily to the degree of preference expressed for the risky or safe alternative.
Little effect of mood was observed on judgements of how risky the decision
was.
Our first research question asked whether there were differences between
the three negative moods. The findings are unequivocal in showing that the
strongest effects on risk behaviour occurred with changes in fatigue, with little effect of concomitant changes in anxiety or depression. This was surprising, given the large literature on effects of anxiety and depression on social
judgement and cognition, and also the less specific literature on effects of
‘‘negative mood’’ (which typically refers to anxiety or depression, rather than
fatigue).
The second question concerned differences between state- and trait-type
mood effects. We were able to distinguish between current mood and more
stable patterns of affect by obtaining two separate state measures. In addition to
measuring raw scores, the use of a standard score transformation enabled us to
correct for individual differences in average level and variability of reporting for
each mood variable. Although raw scores produced the same pattern of effects
as standardised scores, these are subject to bias from participant response style
and stable differences in day-to-day levels of mood reporting. Our use of
standardised scores allows us to conclude that the observed effects of fatigue on
risk taking are associated specifically with state changes, rather than with stable
mood differences.
The third and fourth research questions related to possible moderating effects
of state anxiety and the personal significance of scenarios on the fatigue-risk
relationship in Study 2. Although there was no overall effect of state anxiety,
fatigue effects were found to depend on whether individuals were anxious or not
at the time of testing. At low or moderate levels of fatigue more anxious individuals showed a preference for the safe option, consistent with the view that
anxiety promotes an increased concern for avoiding losses. However, when
fatigue was very high the effect disappeared, and choices were more risky at
both levels of anxiety. There was also a moderating effect of the personal
importance of scenarios. The increase in riskiness under fatigue was greater for
more important decisions, but there was no effect of the familiarity or emotional
impact of scenarios. These findings are considered more fully later, in relation to
our interpretation of the central effects of state fatigue.
850
HOCKEY ET AL.
Finally, we asked whether the effects of naturally occurring negative moods
could be replicated by inducing mood changes experimentally. It is possible that
the unusual pattern of mood effects obtained in Studies 1 and 2 may partly
reflect the difference between effects of natural moods and those of induced
moods in other studies. However, the results of Study 3 showed the same overall
pattern when fatigue was induced, rather than simply measured. In addition, the
stronger control of such a study provides a stronger basis for assuming a causal
interpretation of the effects of fatigue on risk.
The present findings do not fit easily into the current mood/risk literature.
Although they are superficially consistent with some findings of increased risk
under negative mood (Leith & Baumeister, 1996; Mano, 1992), the specific
link with fatigue, rather than with anxiety and depression, is at odds with the
more usual operational definition of negative mood effects. There is also an
apparent discrepancy between our results and those of Pietromonaco and Rook
(1987). Using broadly similar hypothetical everyday scenarios, they found a
reduction in riskiness in students with high levels of depression, whereas we
found no such effects. An important difference may be that Pietromonaco and
Rook defined depression in terms of BDI scores. These clearly represent a
more stable and motivationally distinctive affective state than the day-to-day
changes in background mood measured here. It is conceivable that the use of
hypothetical, rather than real, decisions is a relevant factor in the observed
pattern of effects, and it would be valuable to test effects of fatigue (and other
negative moods) in actual decision situations. However, we believe that participants’ responses to our scenarios represent at least a predisposition to act in
safe or risky ways in real situations. The summary data reinforce this view.
The scenarios were clearly not considered trivial by participants, as mean ratings of both importance and emotional impact were around the midpoint on the
1–5 scale.
The central interpretative problem is how to explain the effects of fatigue on
risk behaviour. As we have previously argued, the human performance literature
makes a strong link between risk and fatigue, but there is a conceptual discrepancy. What is meant by ‘‘risk’’ in this context is the adoption of a low effort
information-processing style—cutting corners and preferring long shots in solving problems (Hockey, 1997; Holding, 1983), or making judgements (Webster
et al., 1996). On the surface, this is more like the heuristic processing typically
associated with positive mood states, rather than the analytic style expected of
negative states (Schwarz & Bless, 1991). However, this may be misleading. As
we said earlier, fatigue is not a typical negative emotion. Rather than giving rise
to problem-focused processing, it appears to operate more as a signal for
reducing engagement in active information processing.
In what way is such a state likely to elicit risky choices in decision making? It would not be surprising for fatigued individuals to select risky options
NEGATIVE MOOD STATES AND EVERYDAY RISKS
851
if a safe course of action involved a higher effort commitment, because effort
conservation is a strong feature of the fatigued state. However, this is not the
case here. As mentioned earlier, PRI scenarios were balanced in terms of the
effort requirements of risky and safe options. Greater effort was implied in
some items by taking the safe option, and in others by taking the risky option;
still others showed no difference. A simple comparison between the safe/effort
and risky/effort types (both based on mean responses to five items) shows that
the correlation between riskiness and fatigue is slightly higher for safe/effort
items, both for Study 1 (r = .454 vs .351) and Study 2 (.422 vs. .343). However, the effect is present (and significant) for both types of scenario, and a
test for differences between non-independent correlations (Steiger, 1980)
shows that the r-values do not differ significantly, t(34) = 0.63, and t(53) =
0.65; both ps > .05. Given the fatigue/performance literature (see Hockey,
1997), it is likely that effort considerations do have a role to play in determining the choice of actions under fatigue. However, the present series of
studies shows that risky choices are made more often whether they involve
more effort or less.
A second possibility is that a natural bias exists in favour of risky options,
because of the possible gains that such choices offer. Under normal conditions,
such bias may be overridden by an inhibitory self-control process, allowing safe
choices to be selected where appropriate. However, such control may be less
effective under fatigue, because such a state is often the end result of extended
regulatory control activity (Hockey, 1997). Muraven, Tice, and Baumeister
(1998) have shown that this kind of manipulation strongly depletes the capacity
for further self-control, as inferred from impaired performance on a range of
tasks administered after the manipulation. This is an intriguing possibility,
although at this stage, it must be considered largely speculative. One testable
prediction is that fatigue will have a greater effect on increased riskiness
whenever a choice situation is characterised by a preference for processing
information about possible gains. Under these circumstances, less effective
control under fatigue will increase the likelihood that the greater gains associated with the risky outcome will result in the choice of the risky alternative.
There is some evidence that increased risk taking is associated with less effortful
processing strategies. Mann and Ball (1994) showed that more risky choices
followed briefer information searches, focusing more on gains than losses.
Unfortunately, there is little direct research on the dynamic aspects of information use in risky choice behaviour—the order in which information is used in
reaching decisions. This is a promising area for further work on mood effects in
decision making.
In addition to clarifying the basis of the main effect of fatigue in these
studies, any new research must consider the unexpected pattern of interactions
with importance and anxiety. The effect of importance is broadly consistent
852
HOCKEY ET AL.
with the above regulatory hypothesis. Important items are those where the
outcome matters to the individual, and show a much stronger endorsement of
the safe option (through self-control?). Under fatigue, the loss of control
capacity will then fail to provide protection from the attractiveness of the
risky option. We are testing this interpretation in current work, by examining
decision processes directly (in terms of the temporal pattern of information
use), and by measuring the attractiveness and subjective probabilities of gains
and losses. On the same reasoning, safe choices may be made more readily
under anxiety because of increased inhibitory control, which protects against
the possibility of loss. This is consistent with the view that analytic processing is increased under negative affect (Schwarz & Bless, 1991; Taylor,
1991). The hypothesised impaired control under high fatigue would then
reduce the effectiveness of this protective function, leading to an increase in
risky choices. If this interpretation is true, analytic processing under negative
affect should occur only under low levels of fatigue (i.e., when subjects are
alert and energetic).
The regulation hypothesis needs to be tested formally, by manipulating
regulatory capacity independently of fatigue and anxiety, for example using
Muraven et al.’s (1998) regulatory loading tasks. In addition, we need to
compare the effects of the background emotional state associated with current
mood and the emotion generated by the decision problem itself. Natural state
anxiety may not be strongly tied to problem information, and may have different motivational implications from anxiety generated by negatively valenced
decision scenarios. For example, the increased use of analytic processing under
negative affect may be more pronounced when the decision problem is, itself,
the source of the state change. Such effects may also be less likely to occur
when regulatory control capacity is depleted by tiredness or by previous
activity. We are currently exploring these questions in experiments that
manipulate the emotional context of decision problems independently of background mood.
Manuscript received 4 November 1998
Revised manuscript received 5 January 2000
REFERENCES
Arkes, H.R., Herren,L.T., & Isen, A.M. (1988). The role of potential loss in the influence of affect on
decision making. Organizationa l Behavior and Human Decision Processes, 47, 181–193.
Baradell, J.G., & Klein, K. (1993). Relationship of life stress and body consciousnes s to hypervigilant decision making. Journal of Personality and Social Psychology, 64, 267–273.
Bower, G.H. (1981). Mood and memory. American Psychologist, 36, 129–148.
NEGATIVE MOOD STATES AND EVERYDAY RISKS
853
Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behaviora l
sciences (2nd ed.). New York: Erlbaum.
Cooke, T.D., & Campbell, D.T. (1979). Quasi-experimentation : Design and analysis issues for field
settings. Chicago, IL: Rand-McNally.
Damasio, A.R. (1994). Descartes’ error: Emotion, reason and the human brain. New York: Putnam.
Epstein, S. (1984). The stability of behavior across time and situations. In R.A. Zucker, J.A. Aronoff,
& A.T I. Rabin (Eds.), Personality and the prediction of behavior (pp. 209–268). San Diego, CA:
Academic Press.
Eysenck, M.W. (1982). Arousal and attention: Cognition and performance. New York: Springer.
Feldman-Barrett, L., & Russell, J.A. (1998). Independenc e and bipolarity in the structure of current
affect. Journal of Personality and Social Psychology, 74, 967–984.
Forgas, J.P. (1995). Mood and judgement : The Affect Infusion Model (AIM). Psychologica l Bulletin,
117, 39–66.
Forgas, J.P. (1998). On being happy and mistaken: Mood effects on the fundamenta l attribution error.
Journal of Personality and Social Psychology, 75, 318–331.
Frijda, N. (1986). The emotions. London: Cambridge University Press.
Hockey, G.R.J. (1997). Compensatory control in the regulation of human performance under stress
and high workload. A cognitive energetical framework. Biological Psychology, 45, 73–93.
Hockey, G.R.J, & Meijman, T.F. (1998). The construct of psychological fatigue: A theoretical and
methodologica l analysis. Paper presented to Third International Conference on Fatigue and
Transportation Fremantle, Australia.
Hockey, G.R.J., Wastell, D.G., & Sauer, J. (1998). Effects of sleep deprivation and user interface on
complex performance : A multilevel analysis of compensator y control. Human Factors, 40, 233–
253.
Hockey, G.R.J., Payne, R.L., & Rick, J.T. (1996). Intra-individual patterns of hormonal and affective
adaptation to work demands: An n = 2 study of junior doctors. Biological Psychology, 42, 393–
411.
Holding, D.H. (1983). Fatigue. In G.R.J. Hockey (Ed.), Stress and fatigue in human performance (pp
145–168). Chichester, UK: Wiley.
Isen, A.M., & Geva, N. (1987). The influence of positive affect on acceptabl e level of risk and
thoughts about losing: The person with a large canoe has a large worry. Organizational Behavior
and Human Decision Processes, 39, 145–154.
Isen, A.M., & Patrick, R. (1983). The effects of positive affect on risk-taking: When the chips are
down. Organizationa l Behavior and Human Decision Processes, 31, 194–202.
Janis, I.L., & Mann, L. (1977). Decision making: A psychologica l analysis of conflict, choice and
commitment. New York: Free Press.
Johnson, E.J., & Tversky, A. (1983). Affect, generalization, and the perception of risk. Journal of
Personality and Social Psychology, 45, 20–31.
Keinan, G. (1987). Decision making under stress: scanning of alternatives under controllable and
uncontrollable threats. Journal of Personality and Social Psychology, 52, 639–644.
Kogan, N., & Wallach, M.A. (1964). Risk-taking: A study in cognition and personality. New York:
Holt, Rinehart & Winston.
Leith, K.P., & Baumeister, R.F. (1996). Why do bad moods increase self-defeating behavior?
Emotion, risk and self-regulation. Journal of Personality and Social Psychology, 71, 1250–
1267.
Mann, L. (1992). Stress affect and risk-taking. In J.F. Yates (Ed.), Risk-taking behavior (pp. 201–
230). Wiley: New York.
Mann, L., & Ball, C. (1994). The relationship between search strategy and risky choice. Australian
Journal of Psychology, 46, 131–136.
854
HOCKEY ET AL.
Mano, H. (1992). Judgements under distress: Assessing the role of unpleasantnes s and arousal in
judgement formation. Organizational Behavior and Human Decision Processes, 52, 216–245.
Martin, L.L., Ward, J.C., Achee, J.W., & Wyers, R.S. (1993). Mood as input: People have to interpret
the motivational implications of their moods. Journal of Personality and Social Psychology, 64,
317–326.
Matthews, G., & Westerman, S.J. (1994). Energy and tension as predictors of controlled visual and
memory search. Personality and Individual Differences, 17, 617–525.
Maule, A.J., & Hockey, G.R.J. (1993). State, stress and time pressure. In O. Svenson & A.J Maule
(Eds.), Time pressure and stress in human judgement and decision making (pp. 83–102). New
York: Plenum.
Mayer, J.D., Gashke, Y.N., Braverman, D.L., & Evans, T.W. (1992). Mood-congruen t judgement is a
general effect. Journal of Personality and Social Psychology, 63, 119–132.
Muraven, M., Tice, D.M., & Baumeister, R.F. (1998). Self-control as limited resource: Regulatory
depletion patterns. Journal of Personality and Social Psychology, 74, 774–789.
Orasanu, J. (1997). Stress and naturalistic decision making: Strengthening the weak links. In R. Flin,
E. Salas, M. Strub, & L. Martin (Eds.), Decision making under stress (pp. 43–66) Aldershot, UK:
Ashgate.
Pietromonaco, P.R., & Rook, K.S. (1987). Decision style in depression: The contribution of perceived risks versus benefits. Journal of Personality and Social Psychology, 52, 399–408.
Polivy, J. (1981). On the induction of emotion in the laboratory: Discrete moods or multiple affect
states? Journal of Personality and Social Psychology, 43, 803–817.
Russell, J.A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology,
393, 1161–1178.
Rusting, C.L. (1998). Personality, mood and cognitive processing of emotional information: Three
conceptua l frameworks. Psychologica l Bulletin, 124, 165–196.
Rusting, C.L., & Nolen-Hoeksema , S. (1998). Regulating responses to anger: Effects of rumination
and distraction on angry mood. Journal of Personality and Social Psychology, 74, 790–803.
Schwarz, N., & Bless, H. (1991). Happy and mindless but sad and smart? The impact of affective
states on analytic reasoning. In J.P. Forgas (Ed.), Emotion and social judgements (pp. 55–71)
Oxford, UK: Pergamon.
Shingledecker , C.A, & Holding, D.H. (1974). Risk and effort measures of fatigue. Journal of Motor
Behavior, 6, 17–25.
Steiger, J.H. (1980). Tests for comparing elements of a correlation matrix. Psychologica l Bulletin,
87, 245–251.
Stone, A.A., Hedges, S.M., Neale, J.M., & Satin, M.S. (1985). Prospective and cross-sectional mood
reports offer no evidence of a ‘‘Blue Monday’’ phenomenon . Journal of Personality and Social
Psychology, 49, 129–134.
Taylor, S.E. (1991). The asymmetrical effects of positive and negative events: The mobilizationminimization hypothesis. Psychological Bulletin, 110, 67–85.
Thayer, R.E. (1989). The psychobiology of mood and arousal. New York: Oxford University
Press.
Thayer, R.E., Newman, J.R., & McClain, T.M. (1994). Self-regulation of mood: Strategies for
changing a bad mood, raising energy and reducing tension. Journal of Personality and Social
Psychology, 67, 910–925.
Tomkins, S.S. (1963). Affect, imagery, consciousness : Vol. 2. The negative affects. New York:
Springer.
Warr, P.B. (1990). The measurement of well-being and other aspects of mental health. Journal of
Occupational Psychology, 63, 193–210.
Watson, D., & Tellegen A. (1985). Toward a consensual structure of mood. Psychologica l Bulletin,
98, 219–235.
NEGATIVE MOOD STATES AND EVERYDAY RISKS
855
Webster, D.M., Richter, L., & Kruglanski, A.W. (1996). On leaping to conclusions when feeling
tired: Mental fatigue effects on impressional primacy. Journal of Experimental Social Psychology, 32, 181–195.
Wine, J. (1971). Test anxiety and the direction of attention. Psychological Bulletin, 76, 92–104.