Content uploaded by Arthur A. Stone
Author content
All content in this area was uploaded by Arthur A. Stone
Content may be subject to copyright.
A Population Approach to the Study of Emotion: Diurnal Rhythms of a
Working Day Examined With the Day Reconstruction Method
Arthur A. Stone and Joseph E. Schwartz
Stony Brook University
Norbert Schwarz
University of Michigan
David Schkade
University of California at San Diego
Alan Krueger and Daniel Kahneman
Princeton University
To date, diurnal rhythms of emotions have been studied with real-time data collection methods mostly
in relatively small samples. The Day Reconstruction Method (DRM), a new survey instrument that
reconstructs the emotions of a day, is examined as a method for enabling large-scale investigations of
rhythms. Diurnal cycles were observed for 12 emotion adjectives in 909 women over a working day.
Bimodal patterns with peaks at noon and evenings were detected for positive emotions; peaks in negative
emotions were found at mid-morning and mid-afternoon. A V-shaped pattern was found for tired and an
inverted U-shaped pattern for competent. Several diurnal patterns from prior studies were replicated. The
DRM appears to be a useful tool for the study of emotions.
Keywords: affect, emotion, diurnal rhythm, DRM
Emotions are fundamental qualities of the human experience
and they play a key role in understanding human health and
behavior. Although there are many ways of measuring emotions,
most studies have individuals self-report on how they are or were
feeling over some specified period of time. A common assessment
is the adjective checklist, wherein a set of emotion words (e.g.,
happy, sad) are presented with intensity response options such as
not at all” through very much. Although some assessments target
current emotions, others specify reporting periods ranging from the
previous day to the previous month. Other assessments are in-
tended to assess trait components of emotion; they ask about the
respondent’s usual emotions, without a specific recall period.
Regardless of the assessment method, most research has focused
on between-person differences in the levels of emotion, either in
general or at particular points in time. The focus of this article is
on a quality of affect that is often overlooked: its rhythmicity over
the course of the day.
Conceptually, diurnal rhythms of affect are likely to be the
product of both environmental influences and physiological pro-
cesses. Both of these broad classes of determinants are themselves,
in part, a function of the rhythm of daily life. For most people,
activities of daily living are clearly entrained to the 24-hr clock and
obvious effects on affect might be expected at different hours of
the day, for example, comparisons between work and nonwork
hours. Regular daily events such as commuting and consuming
meals are also likely to be reliably associated with increased
negative and positive affect, respectively (Kahneman, Krueger,
Schkade, Schwarz & Stone, 2004). As for physiological influ-
ences, it is acknowledged that almost all biological processes have
some diurnal cycle component (Smolensky & Dalonzo, 1993); for
example, cortisol, a stress hormone, typically exhibits a steep
decline throughout the day and is regenerated during sleep
(Kirschbaum & Hellhammer, 1989). It may be the case that some
emotions are particularly sensitive to the psychosocial environ-
ment or to particular components therein, whereas others are more
influenced by physiological processes (Scherer, Wranik, Sangsue,
Atran & Scherer, 2004; Stone, Smyth, Pickering & Schwartz,
1996). In addition to being caused by environmental and biological
processes, diurnal cycles of affect may themselves affect behavior.
The diurnal cycle of some negative affects (e.g., depression, anx-
iety, tiredness), for example, may be predictive of suicidal behav-
ior or of the incidence of occupational accidents (Fortson, 2004).
Conversely, the cycle of positive affects may predict optimal
concentration and efficiency at work. As such, emotion cycles are
an essential part of daily experience and may provide insights into
daily behavior and symptoms.
From a methodological perspective, multiple assessments of
immediate affect across the day (or days) are used to construct
diurnal patterns, and greater numbers of assessments per day allow
for the detection of finer-grained patterns. Recently, several stud-
Arthur A. Stone and Joseph E. Schwartz, Department of Psychiatry and
Behavioral Science, Stony Brook University; Norbert Schwarz, Depart-
ment of Psychology, University of Michigan; David Schkade, School of
Business, University of California at San Diego; Alan Krueger, Woodrow
Wilson School and Department of Economics, Princeton University; and
Daniel Kahneman, Woodrow Wilson School and Department of Psychol-
ogy, Princeton University.
This work was supported by the William and Flora Hewlett Foundation,
the Woodrow Wilson School of Public and International Affairs at Prince-
ton, the National Science Foundation, and the National Institute on Aging.
Arthur A. Stone is the Associate Chair of the Scientific Advisory Board of
invivodata, inc, (Pittsburgh, PA), a company that provides electronic diary
services to the pharmaceutical industry.
Correspondence concerning this article should be addressed to Arthur A.
Stone, Department of Psychiatry, Putnam Hall, Stony Brook University,
Stony Brook, NY 11794-8790. E-mail: arthur.stone@sunysb.edu
Emotion Copyright 2006 by the American Psychological Association
2006, Vol. 6, No. 1, 139 –149 1528-3542/06/$12.00 DOI: 10.1037/1528-3542.6.1.139
139
ies of affect have been based on experience sampling or ecological
momentary assessment methods (Stone & Shiffman, 1994; Stone,
et al., 2000) wherein participants carry a paper or electronic diary
with them throughout the day and complete it according to preset
schedules. These methods have enabled the study of momentary
variables like affect in the environments people typically inhabit,
increasing the ecological validity of the measurements. Studies of
immediate emotion are, however, highly burdensome for partici-
pants and costly for investigators, which partially explains why
most studies of diurnal cycles are based on relatively small num-
bers of participants. Many studies have been conducted with fewer
than 20 participants and many with just a few participants although
often for longer periods of time.
In this report we present an alternative to real-time momentary
assessment for assessing diurnal cycles of affect. The new method,
called the Day Reconstruction Method (DRM; Kahneman et al.,
2004), is based upon having participants recreate yesterday’s ac-
tivities and the affect associated with those activities using a
questionnaire. Because the questionnaire takes about an hour to
complete, it is feasible to design studies to examine activities and
affect with large numbers of participants. This feature of the DRM
opens the possibility of epidemiological studies of diurnal rhythms
of affect and other daily experiences. In many ways, the DRM
methods are similar to those used in time-budgeting surveys such
as the American Time Use Survey (Horrigan & Herz, 2004) and
food frequency surveys, because recall bias concerns led the de-
velopers of the instruments to limit recall to a single day. Data on
the instrument were recently published (Kahneman et al., 2004),
and the analyses presented here use the same data set to address a
different research question.
The two goals of this article are (a) to demonstrate the feasibility
of studying diurnal rhythms of specific affect adjectives with the
DRM in a large sample and (b) to test several small-sample
findings about diurnal rhythms of affect with a large sample of
women who used the DRM for a single working day. Relying
exclusively on work days has advantages and disadvantages. The
work day for a typical individual includes regularly scheduled
activities such as the morning commute, a morning segment of
work, lunch, an afternoon segment of work, the afternoon com-
mute, and a nonwork evening. Of course, these events do not occur
at exactly the same time for all working people, but in a large
sample of people there will certainly be a tendency for certain
activities to occur during certain parts of the day (e.g., lunch
around noon, commuting at the beginning and end of the day). If
such activities are responsible for diurnal cycles of affect, then an
advantage of studying work days is that a strong diurnal cycle
should emerge for the group. A disadvantage of sole reliance on
work days is that the hypothesized effects of activities will be
confounded with effects hypothesized to be due to physiological
processes. One strategy for teasing these effects apart would be
study both work days and nonwork days, where the effects of
activities on group cycles should be reduced. Unfortunately, only
a few nonwork days were available in our sample and they have
been eliminated from consideration in the analyses.
The feasibility goal of the article is addressed by determining if
there are significant time-of-day (ToD) effects on participants’
self-reports as assessed by the 12 emotion adjectives used, expand-
ing upon the information presented in our prior publication on the
DRM (Kahneman, et al., 2004). The article’s second goal is
addressed by exploring the following four previously documented
ToD effects:
1. Several researchers found that energy levels and positive
affect are highest in the morning (Wood, Magnello & Sharpe,
1992; Wood & Magnello, 1992). However, these studies sampled
a limited number of times throughout the day (6 fixed points),
yielding a relatively coarse degree of discrimination. In an attempt
to replicate these results, we examine self-reports of Tired, con-
ceptualized as the inverse of Vigor, and three positive emotions,
captured by adjectives (Happy, Calm/Relaxed, and Enjoy). Be-
cause the DRM characterizes affect through the entire day (by
breaking the day into discrete episodes), we were able to create
much finer depictions of diurnal cycles.
2. Several investigators reported that positive, but not negative,
emotions show diurnal cycles, although the evidence is mixed
(Monk, Fookson, Moline, & Pollack, 1985). In part, the absence of
diurnal effects in prior studies may be attributable to low statistical
power due to small sample sizes (although recording across many
days per individual, which some studies have done, increases
power). Our sample is relatively large and should have ample
power to detect diurnal cycles and moderators of cycles.
3. The impact of age on emotion has been a topic of consid-
erable interest in recent years and a striking finding is that the
elderly report higher levels of positive emotions (Carstensen,
Pasupathi, Mayr, & Nesselroade, 2000). Diurnal cycles of emo-
tions have been examined in one study and were found to be
independent of participants’ age (McNeil, Stones, Kozma &
Andres, 1994). The study of potential moderator variables such
as age is especially well-suited to survey methods such as the
DRM, given the large number of potential participants. In a
previous report (Kahneman et al., 2004), we examined the
diurnal cycles of Tired as a function of participants’ age and
found that age differences were limited to the morning hours.
Young adults (under 30 years old) were significantly more tired
in the morning than other participants, but this difference van-
ished over the course of the day. In this article we examine the
impact of age on all 12 emotion adjectives.
4. The fourth specific test of diurnal rhythms is based on Adan’s
(Adan & Sanchez-Turet, 2001) finding that women’s “optimal”
emotional point throughout the day is around 11 a.m. This finding
was based on a very limited sample (40 university student volun-
teers) aged between 18 and 23 years. We attempt to replicate the
finding with our sample of working women by creating an emo-
tional balance measure based on the difference between positive
and negative emotions. This is conceptually similar to the rationale
for the well-known Affect Balance Scale (Bradburn, 1969). The
hypothesis was that its level would peak at around 11 a.m.
5. Finally, as we have implied above, any observed diurnal
patterns in emotion reports are likely to be associated with, if not
caused by, activities that occur on a regular basis during a working
day. In fact, our prior research has shown that most diurnal cycles
of mood are attenuated when the effects of concurrent activities are
statistically removed. An exception to this statement was the
observation that the daily pattern of tiredness, which did have a
distinct diurnal cycle, was not diminished by removal of activities.
We will test this activity hypothesis with a selection of mood
adjectives.
140
STONE ET AL.
Method
Participants
Participants were recruited by a professional survey firm in Texas, who
randomly contacted potential participants with telephone calls. The firm
selected 1,018 adult women by sampling from driver’s license lists; by
design, the recruitment strategy oversampled employed teachers, nurses,
and telemarketers. Selection was limited to women in this first application
of the DRM in order to reduce variability attributable to gender differences.
The average age was 38 years, and the sample consisted of 24% Black,
22% Hispanic, 49% White and 5% other, with an average household
income of $54,700 (slightly above average for the areas sampled). The
mean education level was “some college.” The 909 women who reported
working on the day sampled were retained for these analyses.
Materials
The DRM was developed to allow for the assessment of activities and
affect of the previous day with minimal recall bias. Respondents first
construct a diary of the previous day, with the following instructions:
Think of your day as a continuous series of scenes or episodes in a
film. Give each episode a brief name that will help you remember it
(e.g., “commuting to work” or “at lunch with B”). Write down the
approximate times at which each episode began and ended. The
episodes people identify usually last between 15 minutes and 2 hours.
Indications of the end of an episode might be going to a different
location, ending one activity and starting another, or a change in the
people you are interacting with.
This task is completed on a confidential questionnaire, which respon-
dents do not need to turn in. Next, respondents describe each of the specific
episodes identified, including the starting and ending times, what they were
doing (selected from a choice of 16 activities, arranged in order of rated
average positive emotion: engaging in intimate relations, socializing, re-
laxing, praying/worshipping/meditating, eating, exercising, watching TV,
shopping, preparing food, on the phone, napping, taking care of my
children, using a computer/e-mail/Internet, doing housework, working, and
commuting), with whom they were interacting (friends, relatives, spouse/
significant other, children, clients/customers, coworkers, boss, no one), and
how they felt during the episode.
The 12 adjectives comprising the emotion assessment are listed in Table
1. The adjectives are very similar to those used in other mood adjective
checklists such as the Nowlis (Nowlis, 1965), POMS (McNair, Lorr &
Droppelman, 1972), and PANAS (Clark, Watson & Leeka, 1989; Watson
& Tellegen, 1985), and, based on this previous research, 9 of the adjectives
were used to index the constructs of Positive Affect (Happy, Warm/
Friendly, and Enjoying Myself) and Negative Affect (Frustrated/Annoyed,
Worried/Anxious, Depressed/Blue, Hassled/Pushed Around, Criticized/Put
Down, and Angry/Hostile). The remaining three adjectives (Impatient,
Competent, and Tired) are considered individually. Response scales for the
adjectives ranged from 0 (not at all)to6(very much), with the remaining
points unlabeled. Other sections of the questionnaire asked respondents for
demographic information, the qualities of their job, and other personal
details; these questions are not used in this article.
1
Procedures
Respondents came at a prearranged time to a central location (hotel
ballrooms), where they filled out the questionnaire in large groups. Re-
search staff directed the groups in the completion of the DRM. Respon-
dents were paid $75 for their participation.
Analysis plan. Using traditional repeated measures analysis of variance
techniques to analyze real-time data (or near real-time data) can be prob-
lematic (Affleck, Zautra, Tennen & Armeli, 1999; Schwartz & Stone,
1998; Schwartz & Stone, in press). Hierarchical linear modeling or mul-
tilevel modeling is generally preferable in that it allows for unequal
numbers of repeated assessments, missing data, autocorrelation among
repeated measures, and various error structures. We test for diurnal
rhythms by including ToD variable as a categorical within-person predictor
in this type of model.
In preparing the data for analysis, the issue arises as to how to represent
the timing of DRM episodes, which are defined by beginning and end
times. For the multilevel analysis, we used each episode’s midpoint (the
average of starting and ending times) to represent when the episode
occurred. These midpoints were categorized into 15 one-hour blocks (e.g.,
9:00 –9:59 a.m.). Although this procedure ignores the duration of the
episode, it has the advantage that the degrees of freedom for the inferential
statistics are based, appropriately, on the number of episodes reported.
However, in order to graphically display the diurnal rhythms of affect at a
finer resolution, we constructed a data set in which each person’s mood at
7:00 a.m. and every quarter-hour thereafter until 9:00 p.m. was stored. In
this data set, the mood ratings of an individual episode were assigned to all
of the 15-min time points that occurred within that episode. Averaging over
all individuals for each time point creates a finer-grained profile of affect
over the day than collapsing the episode midpoints into one-hour catego-
ries. In this graphical analysis, every person contributes to the average
mood at a given time point, except if the person was asleep, either early in
the morning or later at night. We note that this second data set could not
be used to test for diurnal patterns (i.e., inferential analyses), because the
number of observations far exceeds the number of episodes on which mood
ratings were made.
Results
Responses to the DRM
The average number of episodes reported was 14.1 (SD ⫽ 4.8) and the
median episode length was 61 min. There were no reported problems in the
administration of the DRM.
Levels of Affect and Diurnal Cycles
The levels and variability (across all respondents and episodes) of affect
as measured by the DRM are shown in Table 1, in the order in which the
adjectives were presented in the questionnaire. Positive moods (Happy,
Warm, Enjoy; the full adjective labels were presented above and we now
use shorter versions for brevity) clearly have the highest levels of endorse-
ment, whereas negative emotions (Frustrated, Depressed/Blue, Hassled,
Angry, Criticized, Worry) have considerably lower levels. Figures 1 and 2
present plots for each mood where the mean level of the affect was
computed at 15-min intervals. In order to show absolute levels of re-
sponses, while taking into account the higher levels of reporting for the
positive adjectives, different Y-axes were used for Figure 1 and Figure 2.
Tests of Diurnal Cycles for Each Emotion
We first tested whether each emotion adjective was significantly asso-
ciated with ToD. Results of the mixed model, where time was treated as a
categorical variable, are shown in the fifth column of Table 1. All of the F
tests were significant and the percentage of within-person variance that
hour of the day accounted for was computed for each adjective (shown in
column 6 of Table 1). The strongest diurnal patterns, accounting for more
than 4% of the variance, were observed for Tired, Enjoy, Impatient, and
Happy. However, Tired, was by far the adjective most strongly linked to
1
The full instrument is available at http://sitemaker.umich.edu/
norbert.schwarz/day_reconstruction_method.
141
DIURNAL RHYTHMS WITH THE DRM
ToD: It shared 18% of the variance with ToD. The weakest diurnal
patterns, accounting for less than 1% of the variance, were observed for
Criticized, Depressed/Blue, and Angry, which also happen to be those with
the lowest average levels.
On the basis of the multilevel analyses, standard errors were computed
for testing the difference between points on each plot, and they were
multiplied by 2 in order to approximate 95% confidence intervals. These
values are presented in the captions for Figures 1 and 2; a difference of at
least that amount between any two points in a plot is significant at the .05
alpha level. It is clear that many of the adjectives show a bimodal pattern
with midday as the major inflection point. Apparently, lunch is a high point
of the work day for many of these working women, associated with a drop
in negative affect and a rise in positive affect. On the other hand, two of the
adjectives have cycles that are not bimodal: Tired shows a V-shaped pattern
and Competent an inverted-U shape pattern.
To address whether there were overall increases or decreases across the
day, we examined the linear component of time for each adjective. ToD
was treated as a continuous variable and the results of these analyses (F
tests and regression coefficients) are shown in the last two columns of
Table 1. The adjectives that were most strongly associated with categori-
cally coded ToD also exhibited stronger linear associations with ToD.
However, the F tests for Impatient and Criticized were not significant
(though it was for Depressed/Blue). The sign of each regression coefficient
indicates the directionality of the linear component of the association and,
in general, positive feelings increase in magnitude over the day, whereas
the negative feelings decrease. These linear trends must be viewed with
some caution, given the nonlinear effects that were also evident in many of
the diurnal patterns.
Tests of Specific Patterns of Diurnal Affect
1. Positive affect and tiredness during the morning versus evening.
The hypothesis that affect and energy (viewed as the inverse of tired-
ness) were highest in the morning was tested by estimating the contrast
in positive emotions (Happy, Warm, and Enjoy) and Tired for the
morning hours of the day (7 a.m.–11 a.m.) versus the evening hours
(5 p.m.–9 p.m.). All four contrasts were significant [Happy: t(8670) ⫽
17.51, p ⬍ .001, difference
evening-morning
⫽ .56; Warm: t(8628) ⫽ 6.09, p ⬍
.001, difference
evening-morning
⫽ .20; Enjoy: t(8648) ⫽ 16.0, p ⬍ .001,
difference
evening-morning
⫽ .62; Tired: t(8643) ⫽ 29.03, p ⬍ .001,
difference
evening-morning
⫽ 1.02]. Contrary to the previously cited studies,
all tests were in the direction of evenings having higher levels of positive
affect than mornings and higher levels of tiredness. In terms of effect size,
the standard deviations of the momentary variables range from 1.7 to 2.2,
so the change is at about 40% of a standard deviation for Tired and at least
25% for Happy and Enjoy, whereas the effect for Warm is much smaller.
Inspection of Figure 2 shows that there is considerable variation within the
hours defining morning and evening, but the contrasts demonstrate that, on
average, the levels of positive affect and tiredness are greater in the evening
than in the morning.
2. Positive, but not negative, affects have a diurnal cycle. Several
earlier studies (Wood, et al., 1992) observed diurnal cycles for positive but
not for negative emotions; others observed diurnal cycles for both (Monk
et al., 1985). As shown in the first section of the Results section, significant
diurnal cycles were observed for all emotion adjectives in the present data.
However, there was considerable variation in the strength of these effects,
with Tired showing the strongest ToD effect, followed by the positive
affect adjectives of Happy, Warm, and Enjoy. Three adjectives were
weakly associated with hour of the day: Depressed/Blue, Angry, and
Criticized. Nevertheless, some of the negative affects had robust associa-
tions with hour, including Frustrated, Hassled, and Worried. Thus, we
conclude that both positive and negative affects have diurnal cycles.
3. Age invariance of diurnal cycles. Earlier research (McNeil, et al.,
1994) suggested that diurnal cycles are invariant across age, although
Kahneman (Kahneman, et al., 2004) observed age differences in tiredness
over the course of the day. To examine the relationship between diurnal
cycles and age, the mixed model was expanded to include respondents’
age, both its main effect and its interaction with hour-of-day (see Table 2).
Age was coded as a continuous variable and hour remained a categorical
variable so that all potential interaction patterns, and not only linear trends,
would be tested. All three of the positive emotions were associated with
significant interactions between ToD and age, in contrast to earlier findings
(McNeil, et al., 1994). To examine the interaction patterns, least square
means were computed substituting age equals 30 and 60 to create two
profiles of affect (see Figure 3). For the positive affects, the interaction is
attributable to younger people having less positive affect in the morning
hours than older people. Fewer interactions were observed for the remain-
ing emotions. For Worried, younger people had higher levels in the
morning than older people. A similar pattern was observed for Tired where
Table 1
Descriptive Statistics and Results of Multilevel Analyses for Categorical and Linear Effects of
Time-of-Day on Emotional Adjectives (n ⫽ 909)
Average SD
N of
episodes
Mixed model
test of
categorical
hour-of-day
variable: F test
(df)
%of
Within-
person
variance
accounted
for by hour
Mixed model
test of linear
component of
time-of-day:
F test Beta
a. Impatient 1.99 2.20 11,656 37.5 (16, 8603) 5.2% 1.3ns ⫺.005
b. Happy 3.96 1.70 11,749 31.8 (16, 8670) 4.3% 404.4 .052
c. Frustrated 1.39 1.81 11,633 24.3 (16, 8583) 3.4% 220.7 ⫺.045
d. Depressed/Blue .65 1.37 11,615 3.1 (16, 8571) 0.3% 24.8 ⫺.009
e. Competent 4.22 1.96 11,683 11.4 (16, 8586) 1.5% 28.4 ⫺.013
f. Hassled .60 1.33 11,627 13.7 (16, 8579) 1.9% 167.9 ⫺.028
g. Warm 3.87 1.78 11,693 18.2 (16, 8628) 2.4% 52.6 .020
h. Angry .53 1.24 11,613 7.3 (16, 8569) 0.9% 71.2 ⫺.017
i. Worried 1.28 1.79 11,645 13.2 (16, 8594) 1.8% 164.7 ⫺.034
j. Enjoy 3.65 1.96 11,713 38.5 (16, 8648) 5.2% 400.5 .063
k. Criticized .24 .88 11,624 3.0 (16, 8580) 0.3% 4.2ns ⫺.003
l. Tired 2.75 2.17 11,709 154.6 (16, 8643) 18.5% 1473.4 .112
Note. F tests in Columns 5 and 7 are significant at the p ⬍ .01 level, except where noted with ns.
142
STONE ET AL.
Figure 1. Diurnal cycles of negative emotion adjectives and Impatient. The difference between two points on
a plot differ at the .05 alpha level when they differ by approximately .08 for Criticized, .22 for Impatient, .16
for Frustrated, .14 for Worried, .12 for Hassled, .10 for Depressed/Blue, and .12 for Angry.
Figure 2. Diurnal cycles of the positive emotions (Happy, Warm, and Enjoy), Tired, and Competent. The
difference between two points on a plot differ at the .05 alpha level when they differ by approximately .14 for
Warm, .15 for Happy, .18 for Enjoy, .16 for Competent, and .16 for Tired.
144
STONE ET AL.
the young reported higher levels of tiredness than older participants. Thus,
younger women clearly have less pleasant mornings than older women.
4. Optimal moment. Adan (Adan & Sanchez-Turet, 2001) suggested
that women experience their optimal emotional point around 11 a.m. Our
emotion data roughly replicate this pattern. During the day, the optimal
point is around noontime as all negative emotions are reduced and all
positive emotions are increased well above other points of the working day.
However, evenings are even more positive and less negative than noon.
Following the general approach behind the Affect Balance Checklist
(Bradburn, 1969), which is based on a balance between positive and
negative moods, we defined a single overall emotion score as the z score
of happiness minus the z score of frustration. There was a distinctive
association between this score and ToD, shown in Figure 4 (solid line),
with a bimodal pattern.
Another consideration for the definition of the optimal moment of the
day is tiredness. If tiredness is given a weighting equal to that of the mean
of positive and negative emotions and then subtracted, the pattern changes
as shown in Figure 4 (dotted line). High levels of tiredness, which are
viewed here as reducing overall mood, increase the magnitude of the noon
peak and greatly suppress the level of the evening hours.
5. Activities and diurnal cycles. Finally, we tested to which extent
diurnal patterns remain when the influence of specific activities is re-
moved. To do so, we selected three adjectives to represent the diurnal
cycles observed in this study. They are Enjoy (representing positive ad-
jectives), Frustrated (representing negative adjectives), and Tired (which
has a diurnal rhythm that is different than Enjoy and Frustrated). Three
multilevel models were computed, one for each adjective, with the activ-
ities recorded for each episode and ToD as predictor variables. Table 3
presents these the regression coefficients for each of the three models and
significance levels
It is evident from examination of Table 3 that activities that are associ-
ated positively with Enjoy are negatively associated with Frustrated. The
two exceptions are the activities, “On the phone” and “Taking care of
children,” which were not related to Enjoy, but were associated (in oppo-
site ways) with Frustrated. It is also notable that several activities have no
unique contribution to observed levels of enjoyment or frustration, includ-
ing shopping and preparing food. Tiredness is associated with many fewer
activities; those activities that were associated with tiredness were the same
in sign as for the associations with frustration.
Our hypothesis was that the diurnal cycle of emotion would be reduced
or flattened when the effect of activities were taken into account. Figure 5
presents diurnal plots for each adjective in two ways: without the effects of
activities removed from the least square means for ToD and with the
effects of activities removed from the means. Inspection of Figure 5 makes
clear that a substantial flattening of the lines has occurred for Enjoy and
Frustrated, but that there is almost no effect on the hourly means for Tired.
Thus, for Positive Affect and Negative Affect, how time is spent also
matters as well as it diurnal position.
Discussion
The two primary goals of this report were to demonstrate the
feasibility of studying diurnal rhythms of emotions with the
newly available DRM and to reexamine and extend several
diurnal rhythm findings that emerged from earlier small-scale
studies. With regard to the feasibility of using the DRM, our
observation that individuals had little or no difficulty with the
questionnaire suggests that the materials were understood and
completed properly. The average duration of completion is in
line with our expectations of the task, and obtaining compre-
hensive information about the activities and affect of a single
day in well under an hour also shows the efficiency of the
assessment procedure, at least compared with experience sam-
pling methods where the training alone often requires an hour.
It is also notable that the DRM can be administered in group
settings, which is more efficient than individualized adminis-
tration. The diurnal patterns of emotion adjectives appeared
sensible and the association between daily activities and emo-
tions provided further support for the validity of the assessment
method. Overall, it is our opinion that the DRM has operational
characteristics that make it suitable for studying diurnal cycles
of daily experiences in large samples. Although we limited the
experiences to activities and affect, it is notable that other
content, such as symptom experiences or health behaviors,
could easily be incorporated into the procedure by replacing or
extending the current emotion adjectives. Similarly, other affect
adjectives or characteristics of episodes could be used to meet
particular research goals.
With regard to the diurnal cycles observed in this sample of
Texas women, not only were several findings based on smaller-
scale studies replicated, we detected diurnal rhythms that to our
knowledge have not previously been reported. A consistent and
strong bimodal pattern was found for positive and negative emo-
tions. For the three positive adjectives, emotion levels during the
work day had a peak at noon and a second peak starting at about
7 p.m. and the higher level lasted the rest of the evening. Con-
versely, peaks for the six negative adjectives were at about
10 a.m. and then at 4 or 5 p.m., although this pattern was
relatively weak for some of the adjectives. One interpretation of
this bipolarity is that the elevation of negative emotions was
due to work and that lunchtime provided a respite from the
demands of the work environment, reducing negative emotions
(and increasing positive emotions); we discuss the association
between activities and affect below.
Two emotion adjectives had cycles that did not conform to
the bimodal pattern of the positive or negative adjectives:
Competent and Tired. Competent was lowest in the morning,
but quickly rose to the highest levels by midmorning, followed
by a gradual decline throughout the day with an ultimate return
to morning levels. We offer two speculative accounts. On the
Table 2
Results of Multilevel Analyses of Age, Time of Day (ToD), and
Their Interaction on Emotion Adjectives (n ⫽ 909)
F tests for age, hour-of-day, and interaction
between age and time-of-day
Age ToD Age ⫻ ToD
a. Impatient 0.4 5.8*** 1.8
b. Happy 6.1 13.9*** 5.8***
c. Frustrated 3.4 5.8*** 2.3
d. Depressed/Blue 0.8 2.4 1.8
e. Competent 21.1*** 1.9 1.1
f. Hassled 1.9 3.8*** 2.3
g. Warm 25.0*** 7.2*** 3.7***
h. Angry 2.3 3.8*** 2.0
i. Worried 0.4 6.1*** 2.9***
j. Enjoy 16.5*** 12.8*** 5.0***
k. Criticized 0.0 1.9 1.2
l. Tired 12.3*** 15.8*** 10.8***
Note. * p ⬍ .01. *** p ⬍ .001.
145
DIURNAL RHYTHMS WITH THE DRM
one hand, the observed decline in feeling competent may reflect
the accumulation of problems and disappointments during the
work day. On the other hand, feelings of competence may
require that one meets a challenge. Challenges may be more
likely to be encountered during the work day than during the
periods that precede or follow the work day, and one’s per-
ceived likelihood of meeting work related challenges may de-
cline as the work day nears its end.
Figure 3. Diurnal cycles of Enjoy, Happy, Warm, Worried, and Tired. Solid lines represent levels for those 30
and under and dashed lines for those 50 and older.
146
STONE ET AL.
A V-shape was observed for Tired and this confirms prior
real-time data analyses of this emotion (Stone, et al., 1996).
Interestingly, on average, Tired reached its nadir at lunchtime and
was followed by a steep rise through the remainder of the day. We
and others previously speculated (Stone, et al., 1996; Wood, et al.,
1992) that tiredness is largely independent of activities and may
have a physiological basis: from midday onward, the longer we are
awake, the more tired we become, leading, ultimately, to a strong
desire for sleep. Following this argument, we suspect that whereas
positive and negative emotion cycles may be very different on
work versus nonwork days, the diurnal cycles of tiredness will be
very similar for work and nonwork days.
With regard to the prior findings on rhythmicity, we replicated
several findings from smaller scale studies. Diurnal cycles were
observed for all of the emotion adjectives included in the DRM,
although the strength of the associations varied widely. The form
of the rhythms was similar, although not identical, to those ob-
served in prior studies and that greatly increases our confidence
that the DRM is a feasible and valid method for studying diurnal
rhythms with large samples. Of particular interest were the anal-
yses that examined diurnal rhythms taking into account the effects
of the activities. For Enjoy and Frustrated, broadly representing
positive and negative affects, diurnal patterns flattened consider-
ably, indicating the important contribution of activities to diurnal
cycles. A good example of this is the apparent pleasantness of
lunch for this sample. All of the positive adjectives showed a
sizable jump around noon and the negative adjectives showed a
corresponding decline. When the influence of activities (including
lunch) was partialed from the means, the lunchtime peak in posi-
tive affect (and trough in negative affect) was also entirely elim-
inated. However, activities, though associated with the adjective,
had almost no effect on the diurnal cycle of tiredness, a finding that
replicates a previous result (Stone, et al., 1996).
Discrepancies between the present results and past results
may be due to the current sample being limited to working
women in a single geographic location, especially since males
and females may have different diurnal cycles of affect. Our
ability to detect patterns in all emotion adjectives may be a
function of the increased statistical power afforded by the large
number of participants examined, which we argue is facilitated
by use of the DRM.
As an illustration of the potential application of the diurnal
pattern of emotions to explain behavior, Figure 6 shows the diurnal
pattern of tiredness (solid circles, right scale) and the diurnal
pattern of time of death by suicides (gray bars, left scale) for a
sample of women in Italy from 1994 –1997. The latter series was
reported by Preti and Mioto (Preti & Miotto, 2001) in 3-hour time
intervals, the midpoint of which is displayed on the graph. The
pattern of suicides is almost the mirror image of the pattern of
tiredness. Suicides peak in the midmorning, near the time when the
DRM sample was most alert. Suicides are least common late at
night and very early in the morning, when the sample is most tired.
We further find that the diurnal cycle of sleep for those who are
depressed (as indicated by the response that keeping up enough
enthusiasm to get things done in the past month has been a “big
problem”) bears an even stronger inverse relationship to the diur-
nal pattern of suicide (data not shown). One interpretation of this
correspondence is that depression is more likely to lead to suicide
when individuals have a sufficient level of energy to carry through
with a plan to take their own lives. Another example of corre-
sponding diurnal cycles is that of work accidents and tiredness
(Fortson, 2004).
Table 3
Results of Multilevel Analysis of Time of Day and Activities on
Enjoyment, Frustration, and Tired (n ⫽ 909)
Activities
Frustration
Enjoyment Beta Tired
a. Commuting ⫺.35*** .30*** ⫺.04
b. Working ⫺.20*** .32*** ⫺.08
c. Shopping .15 .08 ⫺.14
d. Preparing food .00 .04 .04
e. Doing housework ⫺.60*** .22*** .17**
f. Taking care of children ⫺.02 .37*** .14*
g. Eating .45*** ⫺.39*** ⫺.16***
h. Praying/worship .38*** ⫺.26** .02
i. Socializing .71*** ⫺.41*** ⫺.44***
j. Watching TV .32*** ⫺.20*** .05
k. Nap/resting .30*** ⫺.23** .76***
l. Computer/internet ⫺.02 ⫺.03 ⫺.03
m. Relaxing .57*** ⫺.35*** .05
n. On phone ⫺.08 .25*** .05
o. Intimate relations .82*** ⫺.34* ⫺.55***
p. Exercising .73*** ⫺.55*** ⫺.39**
Note. * p ⬍ .05. ** p ⬍ .01. *** p ⬍ .001.
Figure 4. Diurnal cycle of optimal emotions. The solid line defines
optimal as the standardized happiness score minus the standardized frus-
tration score. The dotted line subtracts the standardized tiredness score
from the first definition.
147
DIURNAL RHYTHMS WITH THE DRM
The results of these analyses are limited in several ways. Our
sample was composed entirely of women and we limited the
analyses to a working day. Certainly, studies of diurnal cycles of
emotion should be extended to men, who may have different
patterns of emotions throughout the day. We also expect that large
differences in cycles will be observed on nonwork days compared
with working days; daily activities are likely to be more loosely
linked to time of day on nonwork days and may display less
distinctive diurnal rhythms. Regarding the DRM, it is a newly
developed technique, which should be kept in mind when consid-
ering these results.
In summary, researchers may wish to consider using the DRM
for examining diurnal rhythms of emotion. It is feasible to admin-
ister to large samples of people, it can be readily modified to study
other emotions, behaviors, or symptoms of interest, and it incor-
porates methodological features to reduce memory bias. A limita-
tion of the method is that it only captures the experiences of a
single day and some research questions—for example, whether
Figure 5. Diurnal cycles of Enjoyment, Frustration, and Tired without the effects of activities removed (solid
lines) and with the effects of activities removed (dotted lines).
Figure 6. Diurnal cycle of suicides of Italian women (1994 –1997) and the cycle of tiredness.
148
STONE ET AL.
work days differ from nonwork days or whether stable individual
differences in diurnal patterns exist—require the assessment of
multiple days.
References
Adan, A., & Sanchez-Turet, M. (2001). Gender differences in diurnal
variations of subjective activation and mood. Chronobiology Interna-
tional, 18, 491–502.
Affleck, G., Zautra, A., Tennen, H., & Armeli, S. (1999). Multilevel daily
process designs for consulting and clinical psychology: A preface for the
perplexed. Journal of Consulting and Clinical Psychology, 67, 746 –754.
Bradburn, N. (1969). The structure of psychological well-being. Chicago:
Aldine.
Carstensen, L., Pasupathi, M., Mayr, U., & Nesselroade, J. (2000). Emo-
tional experience in everyday life across the adult life span. Journal of
Personality and Social Psychology, 79, 644 – 655.
Clark, L., Watson, D., & Leeka, J. (1989). Diurnal variation in the positive
affects. Motivation and Emotion, 13, 205–234.
Fortson, K. (2004). The diurnal pattern of on-the-job injuries. Monthly
Labor Review, 127, 18 –25.
Horrigan, M., & Herz, D. (2004). Planning, designing, and executing the
BLS American Time-Use Survey. Monthly Labor Review, 127, 3–19.
Kahneman, D., Krueger, A., Schkade, D., Schwarz, N., & Stone, A. (2004,
December 3). A survey method for characterizing daily life experience.
Science, 306, 1176 –1780.
Kirschbaum, C., & Hellhammer, D. (1989). Salivary cortisol in psychobi-
ological research: An overview. 1. Neuropsychobiology, 22, 150–169.
McNair, D., Lorr, M., & Droppelman, L. (1972). Profile of Mood States
[Annals]. San Diego: EdITS.
McNeil, J., Stones, M., Kozma, A., & Andres, D. (1994). Age-differences
in mood: Structure, mean level, and diurnal-variation. Canadian Journal
on Aging, 12, 201–220.
Monk, T., Fookson, J., Moline, M., & Pollack, C. (1985). Diurnal variation
in mood and performance in a time-isolated environment. Chronobiol-
ogy International, 2, 185–193.
Nowlis, V. (1965). Research with the mood adjective checklist. In S. S.
Thompkins & C. E. Izard (Eds.), Affect, cognition, and personality (pp.
352–389). New York: Springer.
Preti, A., & Miotto, P. (2001). Diurnal variaitons in suicide by age and
gender in Italy. Journal of Affective Disorders, 65, 253–261.
Scherer, K., Wranik, T., Sangsue, J., Atran, V., & Scherer, U. (2004).
Emotions in everyday life: Probability of occurrence, risk factors, ap-
praisal, and reaction patterns. Social Science Information, 43, 499 –570.
Schwartz, J., & Stone, A. (1998). Data analysis for EMA studies. Health
Psychology, 17, 6 –16.
Schwartz, J., & Stone, A. (In press). Coping with daily work problems.
Contributions of problem content, appraisals, and person factors. Work
& Stress, pp. 1–16.
Smolensky, M., & Dalonzo, G. (1993). Medical chronobiology: Concepts
and applications. American Review of Respiratory Disease, 147(Suppl.),
S2–S19.
Stone, A., & Shiffman, S. (1994). Ecological Momentary Assessment
(EMA) in behavioral medicine. Annals of Behavioral Medicine, 16,
199 –202.
Stone, A., Smyth, J., Pickering, T., & Schwartz, J. (1996). Daily mood
variability: Form of diurnal patterns and determinants of diurnal pat-
terns. Journal of Applied Social Psychology, 26, 1286–1305.
Stone, A., Turkkan, J., Jobe, J., Bachrach, C., Kurtzman, H., & Cain, V.
(2000). The science of self report. Mahwah, NJ: Erlbaum.
Watson, D., & Tellegen, A. (1985). Towards a consensual structure of
mood. Psychological Bulletin, 98, 219 –235.
Wood, C., & Magnello, M. (1992). Diurnal changes in perceptions of
energy and mood. Journal of Research in Social Medicine, 85, 191–194.
Wood, C., Magnello, M., & Sharpe, M. (1992). Fluctuations in perceived
energy and mood among patients with chronic fatigue syndrome. Jour-
nal of the Royal Society of Medicine, 85, 195–198.
Received March 4, 2005
Revision received September 8, 2005
Accepted September 30, 2005 䡲
149
DIURNAL RHYTHMS WITH THE DRM
A preview of this full-text is provided by American Psychological Association.
Content available from Emotion
This content is subject to copyright. Terms and conditions apply.