journal of Pttsoiuiity and Social Psychology Copynfbt 19*4 by the

1^5 Vd 47, No 5. 1105-1117 Amencan Psychological Anoculioji. I K

The Independence of Positive and Negative Affect

Ed Diener and Robert A. Emmons
University of Illinois at Urbana-Champaign

Five studies on the relation between positive and negative affect are reported. In
Studies 1 and 2 we found that positive feelings were remembered as being nearly
independent of negative feelings in the past year, but the two types of affect were
moderately negatively correlated for the past month. In Studies 3 and S, subjects
completed daily mood reports for 70 and 30 days, respectively. In Study 4,
subjects completed three-week, daily, and moment mood reports and also filled
out reports when they experienced strong emotions. The principal finding was
that the relation between positive and negative affect differed greatly depending
on the time frame. The strongest negative correlation between the two affects
occurred during emotional times. The correlation decreased in a linear fashion
as the time span covered increased logarithmically. It appears that positive and
negative affect are independent in terms of how much people feel in their lives
over longer time periods. Researchers need to focus on the processes that underlie
both positive and negative affect and that are responsible for producing their
relative independence.

Subjective well-being has become of wide- past few weeks. The major findings of Brad-
spread interest to psychologists in recent years. burn's studies were that (a) the correlation
Although much work has been carried out at between positive and negative affect items
the survey level (Andrews & Withey, 1976; was very low, (b) the correlation of items
Campbell, Converse, & Rodgers, 1976), there within positive and negative affect categories
remains an age-old debate on what happiness was much higher, and (c) the two dimensions
is and if it can be measured. Even so, re- of affect correlated differently with various
searchers have begun to empirically explore external variables. For example, worry and
the dimensions of subjective well-being anxiety correlated with NAS but not with
(Campbell et al., 1976). Bradburn and his PAS. Social participation and sociability cor-
associates have offered evidence that happi- related with PAS but not with NAS. Other
ness or subjective well-being is not a unitary researchers (Beiser, 1974; Cherlin & Reeder,
construct, but is composed of two separate 1975; Costa & McCrae, 1980; Moriwaki,
feelings: positive and negative affect. In a 1974; Warr, Barter, & Brownbridge, 1983)
series of studies, Bradburn (1969; Bradburn have also found that positive and negative
& Caplovitz, 1965) used a scale consisting of affect show different relations with external
ten items, five of which measured positive variables. Another important finding of
affect (the Positive Affect Scale, or PAS) and Bradburn's studies was that whereas scores
five of which measured negative affect (the on the two dimensions were independent,
Negative Affect Scale, or NAS). Respondents both were separate predictors of self-reported
were asked to indicate whether they had happiness.
experienced any of the ten feelings in the Bradburn's (1969; Bradburn & Caplovitz,
1965) findings have stirred considerable in-
Support for this research was provided, in part, by
terest. Within the subjective well-being area,
National Institute of Mental Health Grant MH1S140 for the results, if confirmed, would imply that
research training in personality and social ecology. We happiness must be measured along two di-
wish to thank Sharon Griffin and Randy J. Larsen for mensions that vary relatively independently
their assistance in data collection.
of each other. Thus Bradburn's findings are
Requests for reprints should be sent to Ed Diener,
Department of Psychology, University of Illinois at Ur- important to the debate about both how
bana-Champaign, 603 E. Daniel, Champaign, Illinois happiness should be conceptualized and how
61820. it should be measured. In addition, the finding
1105
1106 ED DIENER AND ROBERT A. EMMONS

of the independence of the two types of affect is not good domain or content sampling from
is not just of interest to those researchers the entire realm of positive and negative
working in the area of subjective well-being. emotions. Furthermore, most of the items
The finding has practical and theoretical im- are very specific (e.g., "proud because some-
plications to other areas. For example, clini- one complimented you"). It might be that
cians have often conceptualized distress and only these specific feelings are independent,
well-being as lying at opposite ends of a single and that this is not so for affect measured in
continuum. But if these two dimensions are a broader way.
indeed independent, then the distress of clients 2. There appears to be a ceiling effect on
could be alleviated without producing positive positive affect and a floor effect on negative
affect and vice-versa. Some clinical problems affect. This could attenuate correlations over-
might result from the absence of positive all, by reducing the variability of scores.
affect, whereas others may result from the 3. Questions are answered in terms of a
presence of excessive negative affect. yes-no format, thus raising the possibility of
Partly because of their importance and contamination of results that is due to ac-
partly because they are counterintuitive, quiescence.
Bradburn's (1969; Bradburn & Caplovitz, 4. Bradburn's (1969; Bradburn & Caplov-
1965) findings have ignited a controversy. itz, 1965) scale measures simple occurrence
Most people expect positive and negative of the feelings, and fails to take into consid-
affect to vary inversely and there are theoret- eration the frequency and intensity with which
ical reasons as well for believing they should the specific feelings are felt. Thus it is possible
do so (Brenner, 1975). However, a number of that a person may have felt both positive and
researchers have found additional empirical negative affect, but this does not mean that
support for the independence of the two both types of affect were felt in the same
dimensions (Andrews & Withey, 1976; Mor- proportion.
iwaki, 1974). Typically, the PAS and NAS 5. In the questions there is a great deal of
correlate inversely over a range of —.05 to specific nonaffective content that could influ-
-.30, which suggests that the two scales are ence responses (Brenner, 1975). For example,
not totally independent. However, the interi- in the case of "proud because someone com-
tem correlations reveal that the two scales plimented you," one could have been very
are nearly independent. Items within each happy and not have had this happen. Simi-
scale usually correlate around .25, whereas larly, one can be proud without having been
items across scales typically correlate only complimented. When Brenner (1975) used
about -.05 (Beiser, Feldman, & Egelhoff, more general items such as "depressed" and
1972; Warr, 1978). Several improvements on "unhappy," and changed the wording to "in
Bradburn's scale have not altered the inde- the past week," he obtained a significant
pendence of the dimensions (e.g., Beiser, inverse correlation between positive and neg-
1974). ative affect of -.52.
6. One of the most frequent criticisms is
Problems With the Bradburn Scale that the independence may be linked to the
time frame that the questions cover. Recall
Despite these ample supportive empirical that respondents are asked to indicate which
findings, it still remains to be established feelings they had experienced in the past few
whether positive and negative affect are truly weeks. As the time periods asked about be-
independent. Bradburn's (1969; Bradburn & come shorter, positive and negative affect may
Caplovitz, 1965) scales have been criticized become increasingly more inversely related.
on a number of grounds, and it is possible To summarize, despite the numerous find-
that the independence of the two affects may ings from Bradburn's (1969; Bradburn &
be due to the particular items used. A number Caplovitz, 1965) scale that support the inde-
of problems can be briefly described: pendence of positive and negative affect, there
1. Items on the PAS seem to reflect arousal are a number of arguments suggesting that
and activity, whereas items on the NAS do this independence could stem from the scale
not. Related to this is the criticism that there itself. The major difficulties with the scale
 POSITIVE AND NEGATIVE AFFECT 1107

are that the items do not adequately assess mension being pleasantness-unpleasantness
positive and negative affect and that only a and the other being activity or arousal. Thus
single time period is sampled. the emotion literature does not support Brad-
The first of these problems was addressed burn. It suggests either only one dimension
in a recent study by Zevon and Tellegen on which positive and negative affect are
(1982). These investigators had 23 subjects bipolar ends or else separate, highly correlated
complete daily mood checklists for a period factors. In either case, it does not appear that
of 90 consecutive days. They circumvented positive and negative affect are independent.
the problem of specificity of item content by Can the findings of the emotion literature
using affect words selected to ensure adequate be reconciled with those of Bradbum (1969;
sampling of the domain of positive and neg- Bradburn & Caplovitz, 1965) and other sub-
ative affect. They accomplished this by se- jective well-being researchers such as Zevon
lecting three affect words from each of 20 and Tellegen (1982)? We think so, but first
mood categories. The principal finding was an important distinction needs to be made.
that the structure of mood could best be Bradburn and the other subjective well-being
interpreted as two unipolar dimensions, pos- researchers are interested in how much of
itive and negative affect. However, these in- each emotion a person feels over a long time
vestigators also sampled only a single time period. Thus it is quite possible that positive
period. and negative affect could be strongly inversely
related at particular moments in time—even
perfectly negatively correlated—and still be
Relevance of the Emotion Literature
independent in terms of how much people
There is an extensive literature of studies feel in their lives over a long period of time.
in which researchers have used affect adjective In other words, the positive and negative
checklists and have explored the relation be- affect states could be inversely correlated in
tween various emotion words. These studies a person's life. However, the longer term
have focused on momentary moods as op- mean levels of affect that people experience
posed to the long-term structure of affect. A (possibly more trait related) may be indepen-
number of researchers have found that hap- dent.
piness and sadness load on separate factors Our purpose, then, is to determine whether
(Clyde, 1963; Lorr & Shea, 1979). However, positive and negative affect are independent.
the happiness and sadness factors are not We incorporated a number of improvements
independent and are usually highly inversely over previous studies in order to determine
correlated. For instance, Lorr and Shea (1979) this. We used broad affective words such as
found that elation and sadness loaded on happy and unhappy to avoid the sampling
separate factors, but that these oblique factors and specific content problems encountered
were correlated —.55. Thus the emotion lit- by Bradburn (1969; Bradburn & Caplovitz,
erature, in which a number of emotion factors 1965). We also measured affect over different
are uncovered, cannot be taken as ready time periods (moments, days, weeks) to see
support of Bradburn's (1969; Bradburn & if a conceptual bridge could be built between
Caplovitz, 1965) position because the factors the findings of the emotion researchers and
are usually moderately or strongly correlated. the subjective well-being researchers. We
Recently, Russell (1979, 1980; Russell & measured affect ecologically, at random mo-
Mehrabian, 1977) argued in support of ments and across days in people's everyday
the idea that emotion falls along two bi- lives, so that the findings would show less
polar dimensions: pleasantness-unpleasant- influence of memory bias of the subjects.
ness (positive-negative) and arousal. Sjoberg, Perhaps Bradburn's findings are due to mem-
Svensson, and Persson (1979) factor analyzed ory biases because emotions are highly salient
responses to an 86-item mood adjective and memorable and people may remember
checklist, using a Guttman simplex analysis them disproportionately. Indeed, there is a
(Guttman, 1954) instead of a common factor great deal of evidence to suggest this is the
analysis. These investigators also report that case (Diener, Larsen, & Emmons, 1984a;
mood is basically two dimensional, one di- Erdelyi & Goldberg, 1979; Matlin & Stang,
1108 ED DIENER AND ROBERT A. EMMONS

1978; Teasdale & Fogarty, 1979). We sought Table 1

to determine if positive and negative affect Selected Mood Adjective Affect Loadings
are independent in people's everyday lives, Factor 1 Factor 2
not just among stronger and more memorable
emotions. Sample Sample Sample Sample
Affect words A B A B
Study 1
Pleasant
Method Happy 74 74 -08 -15
Joy 85 64 08 -06
Subjects were University of Illinois undergraduates Pleased 64 57 -12 -15
participating to fulfill an introductory psychology re- Enjoyment/Fun 78 60 -07 00
quirement. Although testing the independence of positive Glad 72 70 01 -09
and negative affect, we were simultaneously trying to Delighted 62 69 08 -10
devise a scale with broadly based items to measure Contented 45 57 -24 -25
happiness. Two subject samples were administered a last Unpleasant
of positive and negative affect words. The first, Sample Angry 00 03 52 60
A (n = 112), rated the extent to which they had felt each Fear/anxiety 04 -15 73 59
of 24 emotions over the past year The second. Sample Frustrated -21 -19 63 62
B (n = 346), rated the same words plus two additional Depressed -29 -20 74 70
emotions. Fourteen of these emotion words are shown Annoyed -01 -11 52 51
in Table 1. The emotion words were selected from lists Sad -15 -14 80 60
compiled by Russell (1979, 1980; Russell & Mehrabian, -11 -24 60
Gloomy 82
1977) and Plutchik (1980) so as to represent the emotions
that might most frequently be related to happiness and Note- Decimals are omitted. Sample A: N = 112, Sample
unhappiness. Potential items such as surprised or sleepy B: N = 346
were excluded either because of their questionable nature
as emotions or because they did not appear to be
substantial sources of affect in everyday life. A number
of additional items were included for other purposes (i.e. the satisfaction factor accounted for 7%. In
construction of a subjective well-being scale). In Sample Sample B the two affect factors accounted
A these items were sentences related to life satisfaction for 53% of the variance in responses and the
and in Sample B there were a large number of subjective satisfaction factor 5%. The factor loadings of
well-being items in sentence form. Subjects were asked
to indicate the degree to which they had experienced
specific affect words are presented in Table 1
each of the moods during the past year on a scale of I for Samples A and B.
(not at all) to 7 (extremely much). All subjects were One can see from the relatively simple
tested in a group setting. A different sample of 167 structure that the positive and negative affect
undergraduates was administered the Bradbum PAS and
NAS scales. Following completion of the questionnaires,
words load on separate factors, thereby sup-
subjects were thanked and given a written debriefing. porting Bradburn's position. The factor
structures for the two samples were very
similar. However, the lower loadings for Sam-
Results and Discussion ple B and the lower interitem correlations
Responses to the happiness questionnaires (which are presented in Table 2) may be
were subjected to factor analyses with varimax attributable to the mass setting in which these
rotation and squared multiple correlations subjects completed the questionnaire, thus
used as communality estimates. The number possibly leading to carelessness by some sub-
of factors to be extracted was determined by jects. One can see that the interitem corre-
a joint combination of the scree test (Cattell, lations within each category of affect are high
1966) and variance accounted for by each and the correlations between the two classes
factor. After we inspected the items and their of affect are much lower. These results provide
respective loadings, the factors were labeled evidence for the relative independence of
positive affect, negative affect, and satisfaction. positive and negative affect. In addition, we
The satisfaction factor included items such found that scores on the PAS and NAS were
as "The conditions of my life are excellent." independent of each other (r = —.11, ns).
This third factor is not germane to this study Incidentally, there were no significant sex
and is not included in our further presenta- differences found in this study, nor were there
tion. In Sample A the two affect factors any in the subsequent studies to be reported,
accounted for 47% of the variance, whereas so we do not discuss these further. In Study
 POSITIVE AND NEGATIVE AFFECT 1109
Table 2
Average Interitem Correlations for the Affect-Adjective Scale
Sample Time frame Positive-negative Positive-positive Negative-negative

Study T
n = 112 past year -.15 .57 .54
n = 346 past year -.10 .42 .40
Study 2"
n = 63 past month -.42 .71 .59
past year -.10 .58 .56
b
' University of Illinois College of the Virgin Islands

2, we sought to analyze affect structure with and negative affect words are relatively inde-
a cross-cultural sample and we also examined pendent, but over a shorter time period such
an additional time period. as a month, the two types of affect are
moderately inversely related (r = - . 4 2 , p <
Study 2 .01). When the items were summed into
overall positive and negative affect scales, we
Method found that for the month period the scales
The subjects were 63 (45 female, 18 male) undergrad- correlated - . 6 3 , whereas for the year period
uates enrolled in various psychology courses at the College they correlated only —.18 (ns). The correlation
of the Virgin Islands. Subjects were asked to indicate the between Bradburn's (1969; Bradburn & Ca-
degree to which they experienced 6 positive feelings
{delighted, happy, glad, content, satisfied, and pleased)
plovitz, 1965) PAS and NAS was low but
and 6 negative feelings (annoyed, frustrated, miserable, significant (r = - . 2 3 , p < .05).
sad, depressed, and gloomy). Because the factor analytic So, with a cross-cultural sample consisting
studies had not been completed at this time, the items largely of black Caribbean students, positive
used in the Virgin Island sample were not based on the
factor analyses. These terms were selected from Russell's and negative affect were relatively independent
(1980) to represent very pleasant and unpleasant emotions only over a longer time period. However, the
with roughly comparable arousal values. The words results obtained may be partly due to memory
miserable and satisfied used in this study were not distortion. Some people may remember more
included in the factor analysis. In addition to subjects' of all affective experiences, regardless of how
rating for the past year, participants reported the extent
to which they felt each of the 12 moods in the past much they actually experience. Similarly,
month, using the same 7-point scale. Bradburn's (1969; some people may recall more positive expe-
Bradbum & Caplovitz, 1965) positive and negative affect riences, others more negative. In addition,
scales were also administered. the longer the time period that is reported,
the more likely selective recall and other
Results and Discussion biases are to influence the reports. To elimi-
nate the possible alternative explanation that
Interitem correlations within and between
the independence of the two types of affect
the positive and negative affect items were
is due to memory distortion, in the next
computed for both the past year and the past
three studies we assessed actual affect on a
month, and the averages of these interitem
regular basis over an extended period of time.
correlations are presented in Table 2. Also,
the 6 positive and 6 negative affect words
were summed, the mean positive and negative Study 3
affect scores were computed for each subject
for both time periods, and the two means Method
correlated with each other across subjects for Subjects were 26 (15 female, 11 male) University of
each time period. Illinois undergraduates participating in order to obtain
The results for the rating of the past year research credit in a project on happiness and life satis-
faction. Participants completed mood reports daily for
are very similar to those obtained in Study 10 consecutive weeks. Mood reports were filled out at
1. One can see that for a long time period the end of each day just before subjects went to sleep. If
such as a year, it appears that specific positive it was impossible to complete the form at this time,
1110 ED DIENER AND ROBERT A. EMMONS

Table 3
Average Between-Subject and Within-Subject Interitem Correlations for Daily Studies
Positive-negative item Positive-positive item Negative-negative item
correlations correlations correlations

Sample Between- Within- Between- Within- Between- Within-

time frame subject subject subject subject subject subject

Fall 1981" Daily .17 -.35 .67 .62 .60 .52

Spring 1982b Emotion -.17 -.65 .83 .83 .70 .60
Moment .14 -.37 .74 .57 .74 .50
Daily .10 -.37 .78 .64 .71 .51
Three-week .04 -.10 .74 .63 .56 .46
Summer I982C Daily -.20 -.22 .89 .43 .74 .25

• n = 26. " n = 4 2 . c n = 34.

subjects were allowed to fill it out no later than the Questionnaire (Cattell, Eber, & Tatsuoka, 1970), the
following morning. Each subject was required to turn in Rosenberg Self-Esteem Scale (Rosenberg, 1965), a symp-
their form every day in order to ensure daily completion. tom checklist similar to the Hopkins inventory (Derogatis,
The mood report consisted of 43 self-report scales, 25 Lipman, Rickels, Uhlenhuth, & Covi, 1974), the Bern
monopolar and 18 bipolar. Eight of the monopolar scales Sex Role Inventory (Bern, 1974), and selected others.
were mood words selected from Study 1, and were
answered on a scale ranging from 0 (not at all) to 6
(extremely much). Several criteria were used to select the Results and Discussion
affect words used in this study. First, the selected affect
words had to have substantial factor loadings on the
We computed between-subject correlations
appropriate factor. Second, we tried to select words so across the means that were based on subject's
that positive and negative ones would overall have ap- daily positive and negative affect scores. In
proximately the same arousal value. Third, we attempted addition, we calculated within-subject corre-
to include several discrete emotions (Plutchik, 1980) that lations by converting each subject's scores to
appear to have substantial impact on everyday happiness
(e.g., anger) but we excluded emotions that appear to be standard scores around the individual's mean,
less frequent (e.g., disgust). We combined into one rating so that each individual had a mean of zero
category some emotions because of their apparent simi- and a standard deviation of one on both
larity (e.g., angry and hostile; worried and anxious). Thus positive and negative affect. This eliminated
four category ratings were chosen to represent positive
affect (happy, joyful, pleased, and enjoyment/fun) and
between-subject variability so that these cor-
five were chosen to represent negative affect (depressed/ relations were based on within-subject varia-
blue, unhappy, frustrated, angry/hostile, and worried/ tion across all subjects for all days. Table 3
anxious). An additional broad negative affect item, un- shows the average interitem correlations for
happy, was added to the list because we believed that it positive and negative affect, computed both
represents the most global level of negative affect It had
been inadvertently omitted from the factor analytic studies.
between and within subjects.
This list of positive and negative affect items covered The correlations between the positive and
many or most of the major feelings of happiness and negative affect sums were also computed and
unhappiness of our subjects. These items loaded distinctly are presented in Table 4. The most striking
on separate factors in Study 1 and were roughly compa-
rable in arousal value. Subjects were instructed to recall
finding is that the between-subject and within-
what their day had been like and to rate their day as a subject correlations between positive and
whole when filling out the form. For each daily report negative affect are in different directions for
for each subject, a composite positive affect score was both the interitem correlations and the cor-
obtained by summing the four positive affect words and relations based on the summed positive and
a composite negative affect score was obtained by sum-
ming the five negative affect words. A number of other negative affect scores. These within-subject
self-rating scales were included on the daily report, but correlations indicate that positive and negative
are not relevant to the present study. A total of 1,835 affect do vary inversely within days for people.
daily mood reports were collected. However, how much people feel overall—that
To examine the correlates of positive and negative is, their mean levels of positive and negative
affect, we administered a number of personality tests,
which included the Eysenck Personality Inventory affect—are not inversely related. Indeed, in
(Eysenck & Eysenck, 1968), the Sixteen Personality Factor this sample the two are positively related,
 POSITIVE AND NEGATIVE AFFECT 1111
Table 4 during which the mood report was to be filled out was
product-Moment Correlations Between Summed determined by a randomly generated sample of times
Positive and Negative Affect for Daily Studies between the hours of 9:00 a.m. and 11:00 p.m. Every
ten-minute period during the waking day was sampled
Sample Between- Within- at least once over the six weeks. Each subject was
Time frame subject subject provided with an alarm watch that was preset each day
so that the first time in a day would come in the morning
Fall 1981* Daily .26 -.54 and the second in late afternoon or evening. Each subject
Spring 1982" Emotion -.15 -.85 was provided with a list of times to which to set their
Moment .19 -.57 watch each day. When the alarm went off they were to
Daily .09 -.31 focus on their feelings at that moment and to complete
Three-Week -.08 -.10 the mood report immediately. If it was impossible to do
Summer 1982C Daily -.23 -.45 so immediately, it could be postponed up until one hour
later. Because of the possibility of memory distortion,
26. 42. cn = 34. under no circumstances could the form be filled out later
than one hour after the alarm.
In addition to moment and daily reports, one other
time period was also sampled. Subjects completed mood
though not significantly so. It is important to reports at three-week intervals, a total of four times
point out that the between-subject correlations throughout the semester. Hence these reports included
represent the entire 10-week period because time when daily reports were not being filled out. They
they are based on the summed positive and were instructed to rate their past three weeks as a whole
and to try to remember their overall feelings, and not
negative affect scales. Thus it does seem that how they had responded to the forms. In addition, mood
within days there is a negative correlation reports were also filled out when the subject felt any
between positive and negative affect, a finding strong emotion, either positive or negative, no more often
that supports emotion theorists. When a per- than once a day. Subjects were instructed to fill out a
son is experiencing either positive or negative report when they were experiencing what they thought
for themselves was a strong emotion. These emotion
affect, he or she is less likely to experience times were included because the random moments were
the other type of affect. The positive corre- usually likely to miss these, and these times were also
lation across subjects, although nonsignificant, rather unique in that most emotion research is based on
was unexpected. The nonsignificant positive rather low levels of affect Over the course of the semester,
correlation actually suggests that in this group 1,416 emotion, 3,528 moment, 1,748 daily, and 168
three-week mood reports were collected from the 42
the higher an individual was on one mean subjects.
(e.g., positive affect), the higher he or she The mood report was very similar to the one used in
would be on the other as well. The picture is Study 3. It consisted of a number of unipolar and bipolar
beginning to clear up, but unanswered ques- affect scales. The same affect words were summed to
tions remain. Two problems with this study form the composite positive and negative affect scores.
In addition, we included a number of artifact measures
are the small sample (n = 26) and that only to check on the possibility of response style bias. We
a single time period was sampled in the daily administered the Crowne-Marlowe (Crowne & Marlowe,
rating. In the next study, a larger sample is 1964) scale of social desirability as well as a number of
obtained and additional time periods are measures designed to assess how subjects used the scale
numbers. A response style of using high or low numbers
sampled in order to more clearly examine could potentially affect the correlation between positive
the relation between positive and negative and negative affect. On one measure, subjects rated 57
affect in regard to the time frame covered. positive affect words on the level of intensity implicit in
their meaning. In this rating task any consistent tendency
to use either high or low numbers had to be due to how
Study 4 people used the numbers of the scale and not to differences
in their affective states because the same words served as
Method stimuli for everyone. Because subjects were rating the
words per se in terms of intensity, a tendency to use high
The subjects were 42 undergraduates at the University numbers would indicate a tendency to use high numbers
of Illinois participating in order to obtain research credit when responding to emotion words on the mood rating
in a semester-long project on the structure of happiness forms. On another measure, subjects wrote descriptions
and life satisfaction. Subjects received three hours of in detail of how they felt when they had marked both a
credit for their participation. 2 and a 4 on both positive and negative affect. These
Participants completed mood reports daily for six four descriptions were then rated on a —100 to 100 scale
consecutive weeks. They followed the same procedure as for intensity of affect by two raters who showed high
in Study 3. The participants also completed mood reports mterrater agreement for both the positive and negative
at two random times every day for the six-week period, descriptions (n = .90 and .92). These ratings indicated
a total of three mood reports each day. The time of day the degree to which subjects used the number scale in a
1112 ED DIENER AND ROBERT A. EMMONS

conservative or liberal fashion in describing their feelings. are computed using the six-week averages of
In the last artifact check, subjects indicated where each positive and negative affect Thus the between-
of their scale numbers from zero through six would be
on a line marked continuously with positive affect words subject correlations are based on overall mean
that had been prescaled for intensity. This provided differences between subjects. Although differ-
subjects another opportunity to indicate the intensity of ent reports are being averaged for the various
feeling they meant by their number responses. These time frames, the between-subject correlations
artifact checks are also described by Diener and Larsen always reflect the means for each subject for
(1984).
Three types of reliability coefficients on the positive that type of report for the entire study.
and negative affect scales were computed. The alpha The average interitem correlations between
coefficients (Cronbach, 1951) for positive and negative and within each class of affect for each time
affect were .89 and .84, respectively. We also computed period were also computed and are presented
two types of temporal stability coefficients: We computed
an odd-even reliability by comparing odd versus even in Table 3. The pattern of the within-subject
days across subjects, and we computed three-week stability interitem correlations is very similar to the
coefficients by comparing the first three weeks of the within-subject correlations based on the
study with the second three weeks. The coefficients summed positive and negative affect scores,
obtained were .83 and .79 for positive affect and .87 and
.81 for negative affect, respectively. It thus appears that
though the pattern is not as clearly marked
the positive and negative affect scales have adequate because the latter, of course, are based on
reliability. summing across all of the items.
To assess the correlates of positive and negative affect, Because the correlation between positive
we administered a number of personality inventories, the and negative affect is negatively related to the
majority of which were given in Study 3. Following the
six-week period, participants were informed as to the
time frame covered, we decided to aggregate
objectives of the study, and were given feedback on their across moments and compute the correlation
personality test scores. between the mean of the two affects at differ-
ent points in time. What we did was to
aggregate over adjacent time periods—com-
Results and Discussion
puting the correlation between positive and
Both between-subject and within-subject negative affect first for 2 adjacent moments,
correlations between the summed positive then 4, 8, 16, 32, and 64 adjacent moments.
and the summed negative affect scores were This has the effect of enabling us to examine
computed for each of the four time periods. the correlation between positive and negative
These product-moment correlations are pre- affect as longer and longer time periods are
sented in Table 4. considered. The nonstandardized summed
The principal finding is that the correlations affect scores for moments were used in this
differ greatly depending on the time period. aggregation procedure. We used 64 moments
Regarding the within-subject correlations, the per subject and looked at aggregates across
strongest negative correlation between the all subjects for all moments.
two affects occurs during the emotion times In Figure 1 we present the correlation
(r = -.85). This correlation decreases as the coefficients obtained between positive and
time period covered increases. The smallest negative affect as a function of the number
correlation between positive and negative af- of moments over which we averaged. One
fect was for the three-week period (-.10). So can see that the correlations drop off rapidly
the shorter the time period is, the more as the number of moments in the aggregates
negatively correlated are the two types of increases. Although the same moments are
affect. In other words, it is highly unlikely being used, they are aggregated into progres-
that one can feel both positive and negative sively longer time periods. In addition, the
affect at the same time, especially at strong correlations form a linear pattern when num-
levels. The emotion times are the most po- ber of occasions is plotted on a logarithmic
larized in this respect. None of the between- scale.
subject correlations, on the other hand, are The product-moment correlations between
significantly different from zero. This finding selected personality tests and mean daily pos-
is not all that surprising, considering that itive and negative affect scores are presented
these between-subject correlations are based in Table 5. Also in Table 5 are the correlations
on the longest time span of all because they between selected items of the daily mood
 POSITIVE AND NEGATIVE AFFECT 1113

(e.g., self-esteem, self-confident) in roughly

inverse ways, others tend to correlate with
positive or negative affect and not with the
other (e.g., extroversion, neuroticism, intelli-
gence), and some tend to correlate in the
same direction with both affects (e.g., pro-
ductive). Self-confidence and self-esteem show
the inverse pattern: they apparently affect
both types of feelings in roughly opposite
ways. Perhaps they derive from the feelings.
Productive, active, and domineering people
seem to experience both more positive and
more negative affect. Thus positive and neg-
ative affect sometimes correlate in the same
0.40
0 1 2 4 8 16 32 64 128 direction with some variables. Most fre-
Number of Moments Summed Over quently, however, one type of affect will cor-
Figure I Relation of positive and negative affect as a
relate with an external variable, whereas the
function of time. correlation for the other type of affect is close
to zero. These differential correlations with
external variables also clearly suggest that
report and positive and negative affect. Al- positive and negative affect are independent
though some measures correlate positively whenever they are examined over long time
with one affect and negatively with the other periods.

Table 5
Selected Correlates of Positive and Negative Affect
Fall 1981 (n = 26) Spring 1982 (« = 42)

Positive Negative Positive Negative

Personality scales affect affect affect affect

Sixteen Personality Factor

Questionnaire
Warmth .03 -.43" .09 -.21
Intelligence .03 .36* -.10. Alt*
Dominance .30 .05 .26* .17
Guilt Proneness -.19. .34* .04 .34"
Suspicious -.11 .23 .01. .39f»
Symptom checklist
Obsessive-compulsive items .11 .39" -.17 .16
Interpersonal sensitivity items -.16. .53** .22 .11
Depression items -.02. .53** -.05 .10
Anxiety items .21 53" -.13. .31?*
Neuroticism -.21. .46** -.05 .13
Extroversion .52?* -.05b .28* .00
Rosenberg S-E .35: -.31b .13 -.08
Masculinity .39" -.16 .35" .04
Femininity -.29 -.04 -.20 -.26*
Daily Mood Report
Self-confident .36? -.34* .36." -.58**
Self-esteem .46." -.40** .34." -.46J*
Productive .56" .42" .56" .12
Domineering .57" .12 .37" .24
Active .41" .30 .47" .21

Note. Differing subscripts denote correlations significantly different from each other, two-tailed / test.
•p<.05. "p<01
1114 ED DIENER AND ROBERT A. EMMONS

The Crowne-Marlowe correlated with neg- (-.23), though nonsignificant, whereas in

ative affect (r = -.38), but not with positive Studies 3 and 4 the between-subject correla-
affect ( r = -.01). Our three number-use ar- tions were in the positive direction but also
tifact measures correlated with negative affect nonsignificant. One can see in Table 3 that
.08, .07, and -.02, and with positive affect the within-subject average interitem correla-
.02, - . 3 8 , and -.14. None of the correlation tions within each class of affect are consid-
patterns help explain why positive and nega- erably lower than in the previous two studies.
tive affect are independent over longer time We believe this resulted from the small
periods. When the measures of potential re- amount of variance in affect scores. Almost
sponse bias were partialed out of the positive- all subjects reported considerably more pos-
negative relations for each of the time periods, itive than negative affect each day, and several
there were virtually no changes in the size of reported no negative affect at all on any day.
the correlations. In other words, there was a restriction of
The results seem clear and convincing, yet range on many negative affect items that was
note that we performed each of these four due to a floor effect. This is revealed by the
studies using college students as subjects. Can low within-subject correlation between nega-
these results be replicated with another, more tive affect items (.25).
heterogeneous sample? In Study 5 we exam- It appears that these results confirm the
ined the issue of generalizability, using a earlier ones based on college student popu-
sample of adults in the local community. lations. Positive and negative feelings vary
inversely within the same day, but are more
Study 5 unrelated across subjects over longer time
Method periods.
Subjects were 34 (24 female, 10 male) adults from the
local community. They ranged in age from 33 to 85 General Discussion
years, with a mean age of 65 years. They were recruited
through both a senior citizen center and a volunteer It is now clear that if one uses adequate
organization. The majority of the participants were retired, measures of positive and negative affect and
though still active in the community. All were living samples several different time periods, the
independently and were in fairly good health.
Mood reports were completed daily for 30 consecutive
findings of the emotion researchers can be
days. The forms were returned weekly via self-addressed, reconciled with those of the subjective well-
stamped envelopes provided to each subject. If any days being researchers. The relation between pos-
were missed, they were made up at the end of the 30 itive and negative affect depends on the time
days so that each subject completed an equal number of period being considered. Positive and negative
reports. The nine mood items were identical to those
used in Studies 3 and 4, but the ancillary items were affect states do vary inversely, but only over
different (for purposes of other studies) The print of the short time spans; the two are unlikely to
form was enlarged in order to facilitate its readability. occur together within the same person at the
For their participation, the senior citizen center and the same moment. Our results clearly show that
volunteer organization received $25 for each participant the strongest inverse correlations are found
they recruited.
when one feels strongly emotional, and that
the correlation between positive and negative
Results and Discussion affect decreases in a linear fashion as the
As in Studies 3 and 4, between- and within- logarithm of the time span covered increases.
subject correlations were computed for both For long time periods of weeks or more, the
the interitem and the summed positive and two types of affect become relatively indepen-
negative affect scores. These correlations are dent: How much a person feels of one is
presented at the bottoms of Tables 3 and 4, unrelated to how much he or she feels of the
respectively. The within-subject correlations other. All of the across-subject correlations
between summed positive and negative affect between positive and negative affect were low,
for days (—.45) is similar to the correlations and most were in the positive direction. The
obtained in the two previous studies (—.54 within-subject correlations were highly nega-
and —.31). However, the between-subject cor- tive only when short time periods were con-
relation between the sums was negative sidered. The factor analytic results, the inter-
 POSITIVE AND NEGATIVE AFFECT 1115

item correlations, the correlations between and negative affect would be independent
the summed affect means, and the correlations over time, and partialing out the artifacts
with external variables all suggest the major made absolutely no difference in our findings.
finding: that positive and negative affect are These artifact checks in general showed low
relatively independent in people's lives. This correlations with our affect measures. There-
finding held up with the use of different fore, response style cannot explain the relative
scales—Bradburn's (1969; Bradburn & Ca- independence of the two types of affect. Most
plovitz, 1965) and our own—as well as with important, we found a decreasing correlation
very different samples. Although mean levels between positive and negative affect over time
of affect over longer periods may be due to within subjects. This occurred both when we
personality traits, this conjecture remains un- aggregated daily measures and when we used
certain because those levels could also be due measures that enabled us to assess different
to longer term external factors in peoples' time periods. Because the within-subject cor-
lives. relations are not subject to individual differ-
Our studies demonstrate the importance ences in response bias, this finding is a con-
of studying mood within individuals over an vincing argument that our results are not
extended period of time (see also Diener, spurious. Also, the differences in correlations
Larsen, & Emmons, 1984b). If we had relied between the different time periods could not
solely on the memory of the subjects the be due to individual response biases because
results may have been different. That is, in the same subjects were completing the differ-
memory positive and negative affect could be ent assessments.
coded such that they were strongly inversely The present findings have implications for
related—even over longer time periods—be- the emotion literature. The high polarity of
cause of people's implicit theories about how responses during emotion times indicates that
positive and negative affect go together. In positive-negative is a fundamental dimension
fact, the room for such memory distortions of momentary emotion, as suggested by Rus-
could become larger the longer the time period sell (1980). Polarization usually occurs during
sampled is, and this could make positive and strong emotions in which a person feels either
negative affect appear more inversely corre- positive or negative. Nevertheless, it is possible
lated because memory distortion and judg- at times to feel "mixed emotions" that include
mental bias can be greater the longer the time both specific positive and negative elements
period is. (Plutchik, 1980). Positive and negative affect
There are a number of reasons to believe may load on separate factors in any factor
that the present results are valid and are not analytic study because within the same affect,
due to artifacts or unreliability of the mea- items correlate highly with each other, and
sures. Both the internal consistency and the because the factors can be oblique. However,
temporal stability of the positive and negative the two are normally not independent as
affect measures was quite high. Another ar- experienced in emotion.
gument against an unreliability explanation Although positive and negative affect seem
is that the daily results become less correlated to be experienced in a bipolar way during
when aggregated over time even though such emotion times, this does not mean that there
aggregates are more reliable than single mea- may not be a number of neurologically more
sures. Last, we found that correlations de- specific emotions. However, we have found
creased as the assessed time period increased, very high correlations between reports of
even though measures with which we assessed emotions of the same polarity. For example,
longer time periods were not less reliable. when people experience anger, they also tend
Thus the relative independence of positive to report other negative feelings. Unlike the
and negative affect over time could not be positive-negative relation, the correlations
due to low reliability of the scales. In addition, between emotions of the same polarity re-
we used a number of artifact checks because mained high no matter how long or how
response style had been a problem in some short the time period examined was. This
emotion research. The modest correlations may mean either that the conditions that lead
that did appear cannot explain why positive to a positive emotion also tend to lead to
1116 ED DIENER AND ROBERT A. EMMONS

other positive ones (and the same for negative) ing in the area of subjective well-being (Diener,
or it could mean that the positive emotions 1984) should be to uncover the processes that
tend to blend together in experience, as do are responsible for producing the indepen-
the negative. In any event, whether for mo- dence of the two types of affect.
ments or weeks, emotions of the same polarity
seem to covary strongly.
The strength of the negative correlation References
within subjects for emotion times suggests Andrews, F. M., & Withey, S. B. (1976). Social indicators
that when one affect is dominant there may of well-being. New York: Plenum Press.
be some mechanism that suppresses the other. Beiser, M. (1974). Components and correlates of mental
well-being. Journal of Health and Social Behavior, 15,
This suppressive mechanism occurs at the 320-327.
moment emotions are being experienced, but Beiser, M., Feldman, J. J., & Egelhoff, C. i. (1972). Assets
one type of affect does not seem to suppress and affects: A study of positive mental health. Archives
the likelihood of the other occurring later in of General Psychiatry, 27. 545-549.
time. In other words, the suppression of the Bern, S. L. (1974). The measurement of psychological
androgyny. Journal of Consulting and Clinical Psy-
opposite affect occurs only for a short period chology. 42. 155-162.
of time. Diener and Iran-Nejad (1984) argue Bradburn. N. M. (1969). The structure of psychological
that there is a mutual incompatibility between well-being Chicago: Aldine.
positive and negative affect, but not a linear Bradburn, N. M., & Caplovitz, D. (1965). Reports on
happiness A pilot study Chicago: Aldine.
inverse relationship. The suppressive mecha- Brenner, B. (1975). Quality of affect and self-evaluated
nism could be biological in origin or related happiness. Social Indicators Research. 2, 315-331.
to phenomena such as attentional focus. On Campbell, A., Converse, P. E., & Rodgers, W. L. (1976)
the other hand, it could be that extremely The quality of American life New York: Sage.
good and extremely bad events are rare and Catteil, R. B. (1966). The scree test for the number of
factors. Multivanate Behavioral Research. 1. 245-276
therefore unlikely to occur together. These Catteil, R. B., Eber, H. W., & Tatsuoka, M (1970)
various explanations of the suppressive Handbook for the Sixteen Personality Factor Question-
mechanism await further research. The po- naire. Champaign, IL: Institute for Personality and
larity implied by the suppressive mechanism Ability Testing.
is captured by circumplex models (e.g. Plut- Cherlin, A., & Reeder, L. G. (1975). The dimensions of
psychological well-being. Sociological Methods and Re-
chik, 1980) that postulate that certain emo- search, 4. 189-214.
tions are opposites. Clyde, D. J. (1963). Manual for the Clyde Mood Scale
Finally, we need to ask why might positive Coral Gables, FL: Biometric Laboratory, University of
Miami.
and negative affect be independent across Costa, P. T, & McCrae, R. R. (1980). Influence of
persons. Separate processes must influence extroversion and neuroticism on subjective well-being.
both, and we need to search for the processes Happy and unhappy people. Journal of Personality
that lead to both types of affect. We believe and Social Psychology, 38. 668-678.
that important components of person affect Cronbach, L. J. (1951). Coefficient alpha and the internal
structure of tests. Psychometnka, 16, 297-334.
are the frequency and intensity with which Crowne, D. P., & Marlowe, D. (1964). The approval
both affects are felt. Frequency may vary motive New York: Wiley.
inversely, whereas intensity of affect may Derogatis, L. R., Lipman, R. S., Rickels, J., Uhlenhuth,
covary across persons. In other words, those E. H., & Covi, L. (1974) The Hopkins symptom
Checklist: A measure of primary symptom dimensions
who feel one type of affect strongly may also In P. Pichot (Ed.), Psychological measurements in
experience more intense levels of the other psychopharmacology Modern problems in pharmaco-
affect (Diener, Larsen, Levine, & Emmons, psychiatry (pp. 79-110). Basel, Switzerland: S. Krager
in press). Thus mean levels of the two types Diener, E. (1984). Subjective well-being. Psychological
of affect do not correlate when longer time Bulletin, 95, 542-575.
Diener, E., & Iran-Nejad, A. (1984). The relationship
spans are considered because the influence of between positive and negative affect Independent, in-
frequency and intensity cancel each other verse, or mutually exclusive? Manuscript submitted for
out. This model suggests that positive and publication.
negative affect may actually be controlled by Diener, E., & Larsen, R. J. (1984). Temporal stability
the same processes but are structured in such and cross-situational consistency of positive and negative
affect. Journal of Personality and Social Psychology,
a way that they are independent in expression 47, 871-883.
across persons. A primary task of those work- Diener, E., Larsen, R. J., & Emmons, R. A. (1984a).
 POSITIVE AND NEGATIVE AFFECT 1117

Bias in mood recall in happy and unhappy persons Rosenberg, M. (1965). Society and the adolescent self-
paper presented at the Annual Convention of the image Princeton, NJ: Princeton University Press.
American Psychological Association, Toronto. Russell, J.-A. (1979). Affective space is bipolar. Journal
Diener, E., Larsen, R. J., & Emmons, R. A. (1984b). of Personality and Social Psychology, 37, 345-356.
Person X situation interactions: Choice of situations Russell, J. A. (1980). A circumplex model of affect.
and congruence response models. Journal of Personality Journal of Personality and Social Psychology, 39, 1161-
and Social Psychology, 47, 580-592. 1178.
Diener, E., Larsen, R. J., Levine, S., & Emmons, R. A. Russell, J. A., & Mehrabian, A. (1977). Evidence for a
(in press). Intensity and frequency: Dimensions under- three-factor theory of emotions. Journal of Research
lying positive and negative affect. Journal of Personality in Personality, II, 273-294
and Social Psychology.
Sjoberg, L., Svensson, E., & Pcrsson, L. (1979). The
Erdelyi, N. H., & Goldberg, B. (1979). Let's not sweep
measurement of mood. Scandinavian Journal of Psy-
repression under the rug: Toward a cognitive psychology
chology. 20. 1-18.
of repression. In J. Kihlstrom & F. Evans (Eds.),
Function disorders of memory (pp. 355-402). Hillsdale, Teasdale, J. D., & Fogarty, S. J. (1979). Differential
NJ- Erlbaum effects of induced mood on retrieval of pleasant and
Eysenck, H. J., & Eysenck, S. B. G. (1968). Manual of unpleasant events from episodic memory. Journal of
the Eysenck Personality Inventory San Diego, CA: Abnormal Psychology, 89, 248-257.
Educational and Industrial Testing Service. Warr, P. (1978). A study of psychological well-being.
Guttman, L. (1954). A new approach to factor analysis: British Journal of Psychology, 69, 111-121.
The Radex. In P. F. Lazarsfeld (Ed.), Mathematical Warr, P., Barter, J., & Brownbndge, G. (1983). On the
thinking in the social sciences (pp. 258-348). New independence of positive and negative affect. Journal
York: Columbia University Press of Personality and Social Psychology, 44, 644-651.
Lorr, M, & Shea, T. M (1979). Are mood states bipolar? Zevon, M. A., & Tellegen, A. (1982). The structure of
Jounal of Personality Assessment, 43. 468-472. mood change: An idiographic/nomothetic analysis.
Matlin, M. W., & Stang, D. J. (1978). The Pollyanna Journal of Personality and Social Psychology, 42, 1 1 1 -
principle Cambridge, MA: Schenkman. 122.
Moriwaki, S. Y. (1974). The affect balance scale: A
validity study with aged samples. Journal of Gerontology,
29. 73-78.
Plutchik, R. (1980). Emotion A psychoevolutionary syn- Received March 23, 1983
thesis New York: Harper & Row. Revision received June 27, 1983

Manuscripts Accepted for Publication in the Section

Personality Processes and Individual Differences

Compensatory Self-Inflation: A Response to the Threat to Self-Regard of Public Failure. Jeff Greenberg
(Department of Psychology, University of Arizona, Tucson, Arizona 85721) and Tom Pyszczynski.
Individual Differences in Information Integration During Moral Judgment Donelson R. Forsyth (Department
of Psychology, Virginia Commonwealth University, 810 W. Franklin Street, Richmond, Virginia 23284).
Effects of Sex Guilt, Repression, Sexual Arousability, and Sex Experience on Female Sexual Arousal during
Erotica and Fantasy. Patricia J. Morokoff (Uniformed Services University of the Health Sciences, School
of Medicine, Medical Psychology, 4301 Jones Bridge Road, Bethesda, Maryland 20814).
Validity of Self-Reports of Alcohol and Other Drug Use: A Multitrait-Multimethod Assessment. Alan W.
Stacy (Department of Psychology, University of California, Riverside, California 92521), Keith F. Widaman,
Ron Hays, and M. Robin DeMatteo.
Passive-Avoidance in Syndromes of Disinhibition° Psychopathy and Extraverson. Joseph P. Newman (De-
partment of Psychology, University of Wisconsin, 1202 West Johnson Street, Madison, Wisconsin 53706),
Cathy S. Widom, and Stuart Nathan.
Identifying and Predicting Sex Differences in Married Couples: An Evaluation With Two Data Sources.
Michael L. Barnes (Department of Psychology, Yale University, P.O. Box 11A Yale Station, New Haven,
Connecticut 06520) and David M. Buss.

(continued on p 1128)