intensity and frequency: dimensions underlying positive

Journal of Personality and Social Psychology1985, Vol. 48, No. 5. 1253-1265

Copyright 1985 by the Ai Psychological Association, [tic.0022-3514/85/J00.75

Intensity and Frequency: Dimensions UnderlyingPositive and Negative Affect

Ed Diener, Randy J. Larsen, Steven Levine, and Robert A. EmmonsUniversity of Illinois at Urbana-Champaign

Research on emotions and several happiness scales suggest that positive andnegative affect are strongly inversely correlated. However, work on subjective well-being indicates that over time, positive and negative affect are independent acrosspersons. In order to reconcile this inconsistency, two dimensions are proposed forpersonal affective structure: the frequency of positive versus negative affect andthe intensity of affect. Subjects in three studies completed daily and momentaryreports on their moods. In support of the intensity dimension, the correlationsbetween positive and negative intensity were strong and positive in all threestudies. The intensities of specific emotions across persons were also highlycorrelated. Across the three studies the frequency and intensity of affect variedindependently. Although average levels of positive and negative affect showed lowcorrelations, this relation became strongly inverse when intensity was partialedout. Thus the intensity dimension helps explain the relative independence ofpositive and negative affect. In addition, emotional intensity is offered as a newpersonality dimension that manifests interesting characteristics.

Scientific interest in states of subjectivewell-being has increased substantially in recentyears (for a comprehensive bibliography, seeDiener & Griffin, 1984). Although psycholo-gists have paid considerable attention to bothsingle session surveys of global well-being andthe structure of momentary affective states,fewer authors of empirical work have exam-ined the structure of affect within personsover time. It is important to note the differ-ence between research that focuses on mo-mentary affect and research in which thelonger-term structure of affect persons is ex-amined. If one wants to understand the on-going aspects of subjective well-being, affectmust be sampled within the lives of peopleover a relatively long period of time.

In a seminal work, Wessman and Ricks(1966) examined the fluctuations of dailyaffect in a small group of students. They

Support for this research was provided, in part, byNational Institute of Mental Health Grant MH15140 forresearch training in personality and social ecology. Weexpress sincere gratitude to Sharon Griffin, who was incharge of data collection for the first two studies.

Requests for reprints should be sent to Ed Diener,Department of Psychology, University of Illinois at Ur-bana-Champaign, 603 E. Daniel, Champaign, Illinois

61820.

concluded that persons differed in terms oftheir day-to-day affective states along twobasic dimensions that were independent ofeach other, (a) the amount of positive asopposed to negative affect a person experi-enced, which they called average hedoniclevel, and (b) the amount of variability aperson exhibited in his or her affect.

Bradburn (1969) made the next importantcontribution to this area. He collected datain several national samples and found thatpositive and negative affect, when measuredseparately, varied independently; that is, theamount of positive affect a person felt didnot substantially correlate with the amountof negative affect they experienced. Note thatBradburn was not referring to an ability ofpeople to experience positive and negativeaffect simultaneously in time. Rather, he wasinterested in investigating how people expe-rienced both types of affect over time. UntilBradburn's discovery, most researchers hadconceptualized affect as a single hedonic di-mension in which positive and negative affectwere defined and measured as bipolar oppo-sites. The advantages of allowing positive andnegative affect to vary independently havebeen demonstrated by various researchers(Cherlin & Reeder, 1975; Harding, 1982;Warr, 1978; Warr, Barter, & Brownbridge,

1253

1254 DIENER, LARSEN, LEVINE, AND EMMONS

1983). These investigators have shown thatpositive and negative affect, when measuredseparately, often correlate differently withother variables (e.g., personality measures).

The implications of Bradburn's (1969)work are important from a theoretical andmethodological point of view. However, hisconclusions have not gone unchallenged, thechief criticism being directed at the scales heused to measure positive and negative affect(Brenner, 1975; Kozma & Stones, 1980). Forexample, critics have argued that (a) thepositive items contain a confounding elementof arousal not found in the negative items,(b) ceiling effects for positive items in con-junction with floor effects for the negativeitems may attenuate the correlation betweenpositive and negative affect, and (c) specificfeelings receive the same weight as generalones and feelings experienced once are giventhe same weight as those experienced manytimes.

Although criticisms of Bradburn's (1969)work must be taken into account, other re-searchers have also found empirical supportfor the independence of positive and negativeaffect. Zevon and Tellegen (1982) factor an-alyzed mood reports of 23 subjects, collectedover a 3-month period. They compared in-dividual with group factor solutions and founda two-dimensional structure of positive andnegative affect. They also concluded thatthese dimensions are unipolar, thus support-ing Bradburn's findings. More recently, Dienerand Emmons (1984) have replicated Brad-burn's findings, using a diversity of measuresand techniques. Diener and Emmons's resultswere based on an ecological sampling of dailyaffect and the use of broad affective terms.Although negative and positive emotions wereinversely correlated for brief time periods,they found that when longer periods wereconsidered, the two were independent acrossindividuals. Thus one must take seriously theidea that the overall amounts of positive andnegative affect a person experiences are un-related to each other.

Although empirical evidence appears toindicate with increasing clarity that positiveand negative affect vary relatively indepen-dently within persons, other contradictoryconsiderations make this finding difficult tounderstand. Brenner (1975) reviewed a num-

ber of theoretical reasons why positive andnegative affect should vary inversely. In ad-dition, several researchers have constructedsubjective well-being scales in which negativeand positive affect show a strongly inverserelation (e.g., Kammann, Barter, Irwin, &Dixon, 1979). Warr et al. (1983) even foundthat when Bradburn's items are answered interms of frequency of occurrence (e.g., "of-ten" or "occasionally") rather than dichoto-mously, the two classes of affect becomenegatively correlated.

It seems that the more a person experiencespleasant emotions, the less time is availableto experience negative ones. This straightfor-ward intuitive reasoning is supported by themomentary emotion literature, in which themajor dimension of overarching importanceis invariably a bipolar pleasantness-unpleas-antness dimension (Russell, 1980). Russellpresented a good deal of evidence to dem-onstrate that positive and negative emotionscorrelate highly negatively. Even those emo-tion researchers who place momentary neg-ative and positive moods on separate factorsfind that these two factors are not indepen-dent, but are highly inversely related. Thework of Diener and Emmons (1984), as wellas that of the emotion researchers, indicatesstrongly that the two types of affects areunlikely to occur within the same person atthe same time. Therefore, the more time aperson feels positive emotions, the less timehe or she should feel negative emotions. Thestructure of emotions felt at a particularpoint in time is not the same as the structureof affect within persons over time. However,despite this difference one would still expectthe two approaches to be compatible.

Thus the dilemma is the following: Brad-burn (1969) and other researchers have foundthat negative and positive affect vary inde-pendently across persons. Yet the two polesof affect seem unlikely to occur at the sametime. Therefore, it would seem as though themore frequently positive affect is experienced,the less frequently negative affect would beexperienced. This is supported by subjectivewell-being scales worded in terms of fre-quency, which show a strong inverse relationbetween positive and negative affect. Onemight argue that a person could simply feelpositive affect at certain times and negative

INTENSITY AND FREQUENCY OF AFFECT 1255

affect at other times. However, we know fromthe frequency scales such as that of Kammannet al. (1979) that the more often a personfeels one type of affect, the less frequently heor she will feel the other type. Why, then,doesn't the inverse relation in terms of fre-quency produce an inverse correlation foraverage levels of positive and negative affect?

In an attempt to resolve this issue, wepropose that it requires two underlying di-mensions to understand how affect is experi-enced over time. The first dimension is fre-quency of positive affect, or the amount oftime in which positive affect predominatesover negative affect. The second is intensity,or the strength with which one experiencesaffect. One can calculate frequency easily bysimply adding up separately the number ofinstances in which positive affect predomi-nates over negative affect and those instancesin which negative affect predominates overpositive affect. Over the total number ofobservations, positive and negative affect varyinversely as suggested by both common senseand the emotion literature. This method ofcalculating frequency is also strongly sup-ported by Diener and Emmons's (1984) find-ing that positive and negative emotions donot occur together at the same moment,especially when stronger emotions are expe-rienced. But how can positive and negativeaffect vary independently if the two mustvary inversely in terms of frequency? Theanswer we propose is that positive and nega-tive affect covary together on an intensitydimension; that is, a person who experiencesstrong positive emotions may also be a personwho feels strong negative emotions as well.In a recent study in which he assessed affectover time, Epstein (1983) had 30 collegestudents report their emotions for 28 consec-utive days. Averaging across 14 positive and14 negative emotions, he found that therewas a significant correlation of .58 betweenpositive and negative emotional reactivity, orintensity. Although the focus of the study wasnot on the structure of affect, Epstein's inci-dental finding of the covariation of positiveand negative intensity is of direct relevanceto our study.

If the intensity and frequency dimensionsare relatively independent across persons, agreat deal of confusion can be resolved. Spe-

cifically, both a strong positive correlationbetween the intensity of positive and negativeaffect and a strong negative correlation be-tween the frequency of positive and negativeaffect would tend to cancel each other outover time. In other words, the negative cor-relation of the two types of affect in terms offrequency is balanced by the positive corre-lation for intensity. The result is that overallmeans levels of the two types of affect willtend to be uncorrelated, because mean levelsof affect result from the independent contri-butions of frequency and intensity.

In summary, we propose two basic dimen-sions that describe the structure of subjectivewell-being within person over time: (a) fre-quency of positive versus negative affect and(b) intensity of affect. In order to address thisissue, we set up our study to examine severalquestions: (a) whether the intensity of positiveand negative affect are related and to whatdegree, (b) to examine the relation betweenfrequency of positive affect and intensity ofaffect, (c) to see whether average positive andnegative affect become inversely related whenthe influence of intensity is removed, and (d)if the two hypothesized dimensions of affectdo emerge, to examine their relation withother variables in order to determine if thesedimensions are theoretically meaningful. Anaffective intensity dimension would be aninteresting variable in its own right, indepen-dently of its ability to explain the positive-negative affect issue. Our proposal explainsthe confusion concerning past findings in twoother ways. First, it explains why some re-searchers find independence of the two classesof affect and others do not. We propose thatthe former are assessing mean levels of affect,whereas the latter use scales with questionsthat are worded in terms of frequency. Second,our proposal explains why frequency canvary inversely, as suggested by many re-searchers, and yet mean levels of affect canstill be independent of each other, as foundby Diener and Emmons (1984).

Study 1

Method

The subjects were I I male and 15 female undergrad-uates at the University of Illinois. They were enrolled ina semester-long independent study course and received a


grade of credit or no credit for their efforts. The coursewas publicized widely and there were few constraints onwho could enroll. Subjects completed a mood form everyday for a period of 10 weeks. Each evening beforeretiring, participants completed a form describing theiremotions for that day. Distortions due to memory were

minimized because subjects were required to turn intheir forms on a daily basis. Individual students wereallowed to make up any missed days at the end of theperiod. The procedure of sampling mood over a periodof 10 weeks, as opposed to a single measurement ofmood, allowed us to obtain a reliable measure of affectthat was not highly contaminated by short-term factors(Epstein, 1979). In addition, sampling how people felt ineveryday life allowed us to collect data that had ecologicalvalidity.

Mood form. The daily self-report measure consistedof 41 scales relevant to the subjective experience of affect.Twenty-three of the scales were monopolar and ranged

from 0 (he or she felt that way Not at all during thecourse of the day) to 6 (he or she felt that way Extremelymuch). These mood descriptions represented both positiveaffect ("happy," "pleased," "joyful," and "enjoyment/fun") and negative affect ("unhappy," "depressed/blue,"

"frustrated," "angry/hostile," and "worried, anxious andfearful"). On the basis of factor analytic work, we summed

four of these terms to provide a daily "positive affect"score, and we added five of the terms together to computea daily "negative affect" score (see Diener & Emmons,1984). Thus a person's total positive affect score couldrange from 0 to 24 and the negative affect score could

vary from 0 to 30. However, we placed the two types ofaffect on the same scale by dividing each total by thenumber of terms involved in each total. We then usedthe daily positive and negative affect scores to calculatethe frequency and intensity with which each subjectexperienced positive and negative affect on a daily basisover the course of the 10 weeks.

We calculated the frequency with which a subjectexperienced positive affect by adding up the instanceswhen the daily positive affect score exceeded the dailynegative affect score, and dividing this number by thetotal number of days sampled. Therefore, the "frequency"

dimension refers here to the frequency of predominantlyhappy days. It is a percentage that can vary from 0 (mostunhappy persons) to 100 (most happy persons). In otherwords, "frequency" does not refer to the frequency ofdays on which any level, no matter how small, of positive

or negative affect is felt. Indeed, this type of frequencywould be very high for most persons for both positiveand negative affect because most people report at least asmall amount of both types of affect on most days. The

frequency estimate we used is based on the predominantemotion and thus ensures that more than a trivialamount of the emotion is reflected by the frequencyestimate. We refer to this measure as frequency becauseit is based on a frequency count of happy days. However,the measure enables one to estimate duration or howmuch of the time someone is happy, not frequency inthe sense of number of each individual occurrence ofpositive affect.

Intensity was the strength with which subjects experi-enced their dominant affect. We computed it by takingthe mean positive affect for each subject on happy days—that is, those days on which positive affect exceeded

negative affect. Similarly, we calculated the intensity ofnegative affect by taking the mean negative affect scoreon unhappy days for each subject. The strength of thedominant mood serves as a good indicator of intensitybecause it does not depend directly on frequency. Inother words, we can compute an index of how stronglyindividuals experience each type of affect regardless ofhow frequently their dominant mood is either positive ornegative. Our measure of intensity refers to how stronglyan emotion is felt when it is the dominant emotion. Ifan average value for an emotion were calculated overdays, it would not be a pure measure of intensity becauseit would be confounded by how often the emotion wasexperienced. For example, an average value for negativeaffect would not reflect the intensity of negative affectwhen it was experienced because such an average wouldbe highly influenced by the number of times whenabsolutely no negative affect was felt. We stress thatintensity and frequency calculated by these methods arefree to correlate very highly or not at all.

The daily mood report also contained 18 bipolar scalessuch as "Productive" and "Satisfied with my life." Therewere also a series of 23 bipolar terms such as "crabby-cheerful," "physically active-inactive," "feeling ill-feelingwell," "unaroused-aroused," and "low self-esteem-highself-esteem." Many of the scales constituting the moodform were included for the purposes of a larger study onsubjective well-being and are not presented in this report.

Artifact checks. When considering intensity scores itis important to examine the possibility that those scoresonly reflect differences in subjects' number use. To check

for that possibility, we had subjects rate 57 positive affectwords on the level of intensity implicit in their meaning.In this rating task any consistent tendency to use eitherhigh or low numbers had to be due to how people usedthe numbers of the scale and not to differences in theiraffective state because the same words served as stimulifor everyone. Because subjects were rating the words per

se in terms of intensity, a tendency to use high numberswould indicate a tendency to use high numbers whenresponding to emotion words on the mood rating forms.The word ratings were used later to covary out anyidiosyncratic effects that were due to how the subjectused the numbers of the mood scales.

In a further effort to detect whether the intensitydimension was nonartifactual, we had each participantwrite three descriptions of times when they felt veryhappy and three more about times when they felt veryunhappy. Each event was to have taken place in the last6 months. Subjects were required to take 30 min tocomplete each description and were asked to write ingreat detail. The stories were then randomly sorted andgiven to independent raters who were blind as to thestory condition (happy vs. unhappy), subject, and exper-imental hypothesis. Two judges working independentlyrated each description on the same adjectives that con-

stituted the positive and negative affect scales of themood forms. We could then compute an estimate ofintensity for each subject that was based on the judges'

ratings of the stories. Because the stories were qualitativeand were not based on the use of numbers by the subjects,a significant correlation between the judges' ratings ofintensity for each subject and the subjects' own self-reported intensity on the daily mood forms would rep-

resent compelling support that subjects' self-reported


intensities reflected substantive differences in how theyexperienced affect. One can seek additional evidence forthe validity of the intensity dimension by examining themagnitude of the correlation between the mean intensityof the subjects' three positive stories and the threenegative stories as rated by the judges.

Results and Discussion

As predicted, the relation between positiveintensity and negative intensity was strongand positive (r = .70, p < .01). In other words,the people who experienced intense positiveemotions tended to be the same ones whoexperienced intense negative emotions. Thisstrong positive correlation indicates that theintensity of one's positive emotions does notvary inversely with the intensity of one'snegative emotions, as common sense mightsuggest. The sum of positive and negativeintensity scores for each subject was used astheir overall intensity score.

The correlation between the key variablesis shown in Table 1. As might be expected,frequency and intensity tend to correlatesubstantially in most cases with mean positiveand negative affect. This should be expectedbecause the former two are components ofthe latter two. The intensity with which peopleexperienced affect was relatively independentof frequency, the correlation between the twobeing nonsignificant. The correlation betweenmean positive and mean negative affect wasalso nonsignificant and positive. However,when intensity was partialed out of the cor-relation between mean positive and meannegative affect, the relation was inverse (r =-.46, p < .05). This indicates that when in-tensity is removed, mean positive and negativeaffect are no longer independent, but areinversely correlated. The strong reversal ofthe correlation from .24 to -.46 when inten-sity was partialed out indicates that affectiveintensity is a key reason that average levelsof positive and negative affect are relativelyindependent.

To determine whether the intensity andfrequency with which subjects experiencedaffect was a consistent or stable phenomenon,we calculated split-half reliability coefficients.Specifically, because daily mood reports werefilled out for both the months of October andNovember, separate intensity and frequencyscores for each month were computed for

Table 1Correlations of Key Variables for Study 1

Variables 1

1. Mean positive affect —2. Mean negative affect .25 —3. Frequency .77 -.24 —4. Intensity of affect .90 .48 .24 —

each subject. The results indicated that boththe frequency and intensity with which sub-jects experienced affect were relatively stable.The reliabilities were .72, .60, and .65 forpositive intensity, negative intensity, and fre-quency, respectively. All correlations weresignificant beyond the .01 level.

The tendency to rate emotion words asintense (the mean rating for the 57 positiveaffect words) was partialed out of the corre-lation between positive and negative intensity.When this was done the correlation actuallywent up, from .70 to .74. This result indicatesthat a rating bias in the way in which subjectsused numbers did not account for the way inwhich subjects rated emotion words on theirdaily mood forms.

Finally, the independent ratings of intensityfor each subject that were based on judges'ratings of qualitative story data and the sub-jects' own self-reported intensity on the dailymood forms were correlated. The correlationswere as follows: For positive intensity basedon the story data with self-reported positiveintensity; r = .43, p < .05; for negative inten-sity based on story data with self-reportednegative intensity, r — .54, p < .01. Becausethe judges ratings were based on qualitativedata and not on the use of numbers by thesubjects, these correlations represent strongevidence that the variance between subjectson the intensity dimension in their self-reportsreflect meaningful differences in how theyexperience affect. Additional suppon can begleaned from the correlation between thepositive intensity and negative intensity basedon the data for each subject (r = .42, p <.05). The correlations based on the story datahave to be interpreted in light of the lowreliabilities of the judges ratings: For happystories, r= .55, and for unhappy stories, r =.37. Also, the amount of variance in thejudges' ratings of the subjects' stories was


small. Specifically, there were ceiling effectsfor the rating, of happy stories and flooreffects for the unhappy stories. Consequently,all intensity correlations based on the storydata would probably be much higher if thejudges ratings would have been more reliableand allowed for more variation between sub-jects.

Study 2

Study 1 confirmed our initial hypothesisand provided some evidence that the intensitydimension was not a self-report artifact. InStudy 2 we wanted to replicate our findingswith a larger sample. In addition, we wantedto sample peoples' affective experience forrandom moments. The moment ratings areeven less dependent on memory and thereforethe possibility of distortion in reporting shouldbe further minimized.

Method

The sample consisted of 42 University of Illinoisundergraduates (21 male and 21 female) recruited in thesame manner as in Study 1. The main measurementinstrument was the same mood form used in Study 1.In this study, however, participants rated their affectivestates within the context of different time frames. Specif-ically, subjects rated their mood (a) daily, summarizingthe entire day, (b) at two random moments each day, and

(c) when they felt a very strong emotion. The summaryrating was a replication of the procedures used inStudy I. The moment ratings were filled out at tworandom intervals every day for 6 weeks. Each subjectwore a watch with an alarm that would go off once earlyin the day and then again late in the afternoon or evening.The subject set his or her own alarm according to arandomized list of times that covered every 5-min wakingperiod during the 6-week course of the study. In orderto enhance spontaneity, the alarms were set hours inadvance. Most subjects reported thinking about the alarmfor 2 or 3 days, but thereafter it caught them unexpectedly.

When the alarm went off, subjects were to focus on theirfeelings at that moment and complete a mood question-naire immediately. They were also instructed to fill outa mood form whenever they felt a particularly strongemotion (positive or negative), but not more than onceper day. These emotion forms were included becauserandom moments were likely to miss these emotionallycharged times. We computed scores for frequency andintensity on all mood forms for the different time per-spectives, using the same formulas that were used for thedaily mood forms in Study 1.

In this study we included two different measures toindicate how each subject used the number system of thescales. As in Study 1, these artifact checks were intendedto assess response style bias. On one measure, subjectsdescribed in detail how they felt when they marked a 2

Table 2Study 2: Intensity Correlations for DifferentTime Periods

MomentTime

Daily Emotion

period Pos Neg Pos Neg Pos Neg

MomentPositiveNegative

DailyPositiveNegative

EmotionPositiveNegative

—

.88

.92

.71

.79

.70

—

.77

.78

.79

.80

—.70 —

.73 .67 —

.58 .61 .80 —

Note. N = 42. All coefficients are significant beyond p <

.01. Pos = positive; Neg = negative.

and a 4 on both positive and negative affect. These fourdescriptions were then rated on a scale from —100 to100 by two raters who showed high interrater agreementfor both the positive and negative descriptions (rs = .90and .92, respectively). These ratings could then be partialedout of the intensity scores for subjects. In this way wecould minimize the idiosyncratic ways in which thesubjects interpreted the meaning of a number whenrating their affective states. On the other artifact measure,subjects indicated where their happiness scale numbersfrom 0 through 6 would be on a line. The line wasmarked continuously with positive affect words that hadbeen prescaled normatively for intensity and were placedat the appropriate position on the line. This providedsubjects another opportunity to indicate the intensity of

feeling they meant by their number responses. Thus wecould check on the possibility that intensity ratings onthe mood forms might be due to response style ratherthan affect per se.


Because mood reports were completed bysubjects for three different contexts (days,random moments, and high emotion times),it was possible to compute the correlationbetween positive intensity and negative inten-sity as well as between frequency and summedintensity for each time period. The correla-tions between frequency and summed inten-sity were —.11, .05, and .08 for the moment,daily, and strong-emotion time periods, re-spectively. The correlation between positiveintensity and negative intensity mirrored thoseof Study 1. The correlations for each timeperiod are presented in Table 2. The corre-lation between mean positive and mean neg-ative affect was r = .13, ns. When intensity


was partialed out, this relation becamestrongly inverse (r = -.76, p < .01). Intensityand frequency correlated with positive affect(rs = .83 and .52, respectively) and with neg-ative affect (rs = .59 and -.66, respectively).

Interestingly, the correlations between pos-itive intensity and negative intensity for themoment and emotional times were evenstronger than those for the daily reports. Thisresult is probably related to the extent thatsubjects had to rely on their memories whenfilling out the mood forms. Specifically, inthe daily reports it is likely that more errordue to memory bias was present in the databecause subjects had to wait until the end ofthe day to complete the mood form. Thiswas not the case for moment and emotiontimes as subjects were instructed to fill out amood form immediately after their beepersignaled or immediately after they experienceda strong emotion. In addition, they wereinstructed to complete the mood forms sothat they would reflect how they felt at thatspecific time.

Overall, this consistent pattern of resultsindicates that the intensity dimension playsan important role in how people subjectivelyexperience affect when reviewing their entireday, considering their affective state at anygiven moment, and when they are interpretingthe experience of a strong emotion. In otherwords, intensity is such a pervasive dimensionin the organization of affective experiencethat there are some people who experienceaffect more intensely than do others no matterwhat the time period. Furthermore, peoplewho experience positive emotions more in-tensely are also the same people who experi-ence negative emotions more intensely.

To investigate the correlation between pos-itive intensity and negative intensity from adifferent vantage point, we examined theintercorrelations for intensity between specificemotion words. Three positive affect words(pleased, joyful, and enjoyment) and threenegative affect words (angry, fearful, and de-pressed) were used. The analysis involvedexamining the daily rating for all six emotionwords for each subject each day. The scorefor the emotion word rated the highest eachday was recorded. We calculated the intensitymeans for each emotion word by adding upthe rating for each word, using the days on

which that word was rated higher than theother five emotion words. We then dividedthis sum by the number of days that thatemotion word was endorsed as the predomi-nate emotion of the day. In this way a separatemean intensity score was calculated for eachof the six emotion words. The separate inten-sities for each of the six specific emotionwords were intercorrelated across persons.The average correlation among the three pos-itive emotions was .42, the average correlationamong the three negative emotions was .43,and the average correlation between the threepositive and three negative emotion wordswas .41. All correlations were significant be-yond the .01 level, and none of these corre-lations differed from one another. These find-ings indicate that the tendency for somepersons to experience affect more intenselythan others is not simply limited to globalmeasures of positive and negative affect, butalso occurs at the level of more specificfeelings.

We made an additional attempt in Study2 to consider the degree to which the findingsconcerning intensity were biased by self-reportartifacts. When the scores from the measuresused to detect response styles were partialedout of the moment ratings, the correlationbetween positive and negative intensity re-mained virtually unchanged at .88. When thescores on the Marlowe-Crowne scale (Crowne& Marlowe, 1964) were partialed out in orderto control for social desirability, the correla-

tion between positive intensity and negativeincreased slightly. In retrospect, this makessense because those individuals who have atendency to present themselves in a positive

light would have strongly endorsed positiveaffect items and endorsed low levels on thenegative affect scale. Consequently, a socialdesirability response set would have deflatedthe correlation between positive intensity andnegative intensity.

In both Studies 1 and 2 the mean scoresfor the 18 bipolar self-rating scales includedon the daily mood forms were correlatedwith frequency, positive intensity, and negativeintensity. The relations that were significantacross both studies are shown in Table 3.

Scales that did not produce a significantpattern of correlations that replicated across


Table 3

Correlations Among Intensity, Frequency, and Daily Mood Variables

Daily mood scales

Frequency Positive intensity

Fall Spring Fall Spring

Negative intensity

Fall Spring

Self-EsteemInner HarmonyCheerfulSelf-ConftdentArousedPhysically ActiveProductiveDomineeringCaringSatisfied With DayLife Close to Ideal

Note. Fall n — 26, one-tj

,56,75.53.50.30.22.22.37.40.51.41

lilcd p < .05, r =

.55

.69

.68

.67

.12

.11

.09

.05

.28

.59

.43

.34, n < .01,

.29

.32

.31

.13

.47

.37

.58

.56

.31

.74

.46

r = .47: two-tailed p

.00

.05

.13-.04

.34

.41

.40

.31

.29

.42

.38

< .05. r =

.09

.04

.09

.09

.38

.51

.39

.45

.08

.36

.11

.40, n < .01,

-.21-.14

-.12.30.26.30.30.30.23.22.22

r = .51. Spring

both studies were Energetic, Well, Serious,Passive, Sociable, Lonely, and Melancholic.

Four self-rating scales relating to the psy-chological experience of global well-being (i.e.,High Self-Esteem, Inner Harmony, FeelingCheerful, and Feeling Self-Confident) covariedsignificantly with the frequency of positiveaffect but did not relate to either positive ornegative intensity. In contrast, self-rating ofhigh arousal, high levels of physical activity,high productivity, and a sense of dominationin personal relationships were all significantlycorrelated with positive and negative intensitybut were insignificantly related to frequency.This pattern of findings lends additional sup-port to the hypothesis that frequency andintensity represent different dimensions ofaffective experience. Not only do they varyrelatively independently of one another, butthey also correlate meaningfully with othervariables in different directions. In addition,the intensity of both positive and negativeaffect correlates in the same direction withthe daily variables.

Another finding common to both studieswas the pattern of correlations between bothintensities and personality variables. A largebattery of personality tests were administeredto subjects in both studies. This battery in-cluded tests such as the Personality ResearchForm (PRF; Jackson, 1974), Rosenberg's Self-Esteem Scale (Rosenberg, 1965), and theEysenck Personality Inventory (EPI; Eysenck& Eysenck, 1964). In Study 1, of the 57

personality scales completed by the subjects,47 (82%) correlated with positive intensityand negative intensity in the same direction.Of the remaining correlations, all of themwere close to zero. In Study 2, of 51 person-ality scales, 46 (90%) correlated with positiveintensity and negative intensity in the samedirection. These personality tests were givenfor the purposes of another study and do nothave a relation with positive and negativeintensity that was predicted beforehand.However, the fact that they correlated withpositive intensity and negative intensity inthe same direction in the overwhelming ma-jority of cases strongly suggests that intensityis a unitary construct.

Study 3

The results of both Studies 1 and 2 arestrong evidence regarding the nature of therelation between intensity and frequency, aswell as the relation between positive intensityand negative intensity. However, because theseresults are based on similar samples of collegestudents, it is important to determine whetherthe results could be replicated with older,more mature individuals, as well as withindividuals who live in different environmen-tal circumstances.

Method

Subjects were 34 adults living in the community. Theyranged in age from 33 to 85, with a mean age of 65.


They were recruited through both a senior citizen centerand a volunteer organization. The majority of participantswere retired, though still active in the community. Allwere living independently and were in reasonably goodhealth. Mood reports were completed daily for 30 con-secutive days. The forms used were similar to those usedin Studies I and 2, except that the print was enlarged inorder to facilitate readability. The daily mood forms werereturned weekly via self-addressed stamped envelopesprovided to each subject. If any days were missed, theywere made up at the end of the 30 days so that eachsubject completed an equal number of reports. For theirparticipation, the senior citizen center and the volunteerorganization received $25 for each participant who com-pleted the study. Again, scoring for positive and negativeaffect, we computed frequency and intensity of affect,using the same formulas used in Studies 1 and 2.


The results of Study 3 indicate that thefindings from the first two studies do gener-alize to a different sample. The correlationbetween positive intensity and negative inten-sity was .70, p < .01, and the correlationbetween intensity and frequency was slightlypositive, but not significant (r = .25). Thecorrelation between mean positive and meannegative affect was -.23 in this sample. Withintensity partialed out, this correlation be-came -.86, p < .01. Intensity correlated withpositive affect (r = .45) and with negativeaffect (r = .57). Frequency correlated withpositive affect (r = .59) and with negativeaffect (r = -.79). In addition to rating theiraffective experience, participants in Study 3also rated their level of physical activity on adaily basis. The correlations between averagephysical activity level and the intensity andfrequency scores for subjects mirrored similarfindings from Studies 1 and 2. Specifically,mean level of physical activity was signifi-cantly related to both positive intensity (r =.50, p < .01) and negative intensity (r = .37,p < .05), but was unrelated to the frequencyof positive affect (r = -.08).

General Discussion

In summary, our major findings were asfollows:

1. The correlations between positive andnegative affect intensities were strong andpositive for all studies, with a value of .70 inthe three sets of daily ratings and with highervalues in the momentary and emotion ratings.

2. This relation was remarkably consistentacross heterogenous samples that differed interms of both age and general life circum-stances.

3. We found even stronger, positive rela-tions between average positive and negativeintensities when we analyzed the mood reportsthat were completed (a) at random momentsand (b) just after particularly strong emotionshad been experienced.

4. The intensities of separate specific emo-tions (e.g., joyful, depressed, angry) also co-varied across persons.

5. The magnitude of the intensity corre-lations did not change when several potentialself-report artifacts were partialed out. Inaddition, when subject-generated qualitativedata was rated by independent judges, thosesubjects who described their positive emotionsas intense tended to be the same individualswho described their negative emotions inmore intense ways.

6. Both positive and negative affect inten-sities were found to be reliable over time. Allof the forgoing suggest that emotional inten-sity is an important personality variable.However, the .70 correlations between positiveand negative intensities indicate that onlyabout one half of the variance in emotionalintensity is accounted for by this underlyingpersonality factor. There must be other un-identified factors that influence the intensityof either positive or negative affect.

7. Mean positive and negative affect cor-related at very low levels in all three studies.However, when intensity was partialed out,the correlation between the two affects becamestrongly inverse in all studies. In the threedaily studies, the correlations between positiveand negative affect were —.46 (Study 1), —.76(Study 2), and -.86 (Study 3) when intensitywas partialed out. These values indicate thatthe personality dimension of affective intensityplays a large role in producing the typicallylow correlation across persons between posi-tive and negative affect. Clearly, they are notindependent when the influence of emotionalintensity is taken out.

Overall, the data provide compelling evi-dence that intensity is a meaningful andreliable component of affective experience.Our second basic question concerned therelation between intensity and frequency. Here


we found, as predicted, statistical indepen-dence. Consequently, it appears that frequencyand intensity represent separate processesthat contribute independently to affective ex-perience. Subsequent analyses (Diener, Lar-sen, & Emmons, 1983) suggest that intensityand frequency scores, when standardized,combine in an additive way to form meanlevels of affect.

There are a number of reasons to believethat our intensity dimension is not a mea-surement artifact. Intensity did not correlatewith the Marlowe-Crowne social desirabilityscale. We used a number of checks on artifactsand partialing these out from the intensitycorrelations did not diminish their magnitude.When subjects described their emotional ex-periences in paragraph form, an intensitydimension still emerged. Another reason torule out acquiescence resides in the AffectIntensity Measure (AIM), which Larsen(1985) developed. This questionnaire measurecontains 40 items, approximately one thirdof which are reverse keyed. Larsen found thatrespondents who answer the affirmativelyphrased AIM items in the keyed directionalso tend to respond to the reversed keyeditems in the intense direction as well. In amore recent daily study, Larsen (1985) foundthat the AIM correlated .61 with averagedaily affect intensity. This finding also tendsto rule out number use artifacts as an expla-nation of the intensity dimension because theAIM is answered in a true-false format.

It is important to understand that ourintensity dimension is not a mathematicalnecessity based on the method by which thescores were computed. For example, if aperson scores high on positive intensity, it isconceivable that when unhappy, he or shecould score very low on negative affect (butlower or zero on positive affect). There areother arguments that suggest that our com-putational procedures do not artifactuallycreate this intensity dimension. The storyintensities, for example, were based on astraightforward single affect score for eachstory and still the intensity of the happystories was positively correlated with the in-tensity of unhappy stories. Also, Larsen's(1985) Affect Intensity Measure is based ona straightforward single-score assessment ofintensity. The more complex computational

procedures used in the daily studies werenecessitated by the need to separate frequencyfrom intensity, but the results were totallycongruent with other methods of assessingintensity and show concurrent validity withexternal variables. Thus it appears that arti-facts had little influence on the findings ofthis study.

A question can still be raised as to whetherthe intensity dimension merely reflects theway subjects describe their emotional states(whether with rating scales or with words),and not the way they actually experiencethem. If so, this cognitive way of labelingtheir emotions would be quite interesting initself. However, data that we have more re-cently collected (including measures of phys-iological arousal, daily activities, and the wayevents are perceived) suggest that the intensitydimension represents a broadly pervasive styleof affective life that has cognitive, physiolog-ical, and behavioral consequences.

The importance of the frequency and in-tensity dimensions has been demonstrated inother areas of psychology as well. In psycho-physics, for example, the two basic physicaldimensions of a stimulus are its frequencyand intensity, and these dimensions combineto produce the psychological experience ofperception. In personality psychology, ac-cording to Allport (1961), there are two de-nning characteristics of a trait response: fre-quency and intensity. In other words, howfrequently and with what intensity does anindividual exhibit behaviors presumed to bereflective of a given trait? Murray (1938)discusses the periodicity (frequency) andstrength (intensity) of physiological and psy-chological needs. More recently, Ozer (1982)applied the dimensions of frequency andintensity to the scaling of behavioral responseclasses (e.g., how often and how much aggres-sion is exhibited in a particular situation).

In some sense our dimensions are reminis-cent of the ancient Greek Galen's four tem-peraments based on the four humors. In thisscheme, a choleric person feels strongly orintensely, whereas the phlegmatic is relativelyunemotional and therefore does not experi-ence emotions intensely. The sanguine personis one who is frequently happy, whereas themelancholic frequently experiences negativeaffective states. However, unlike Galen, we


conceptualize frequency and intensity as con-tinuous variables that tend to vary indepen-dently. Thus in our system a person couldbe, for example, both a choleric and a melan-cholic. In terms of experiencing happiness,those who are high in frequency of positiveaffect and high in intensity tend to feel exu-berance and joy, whereas those high in fre-quency but low in intensity experience con-tentment and serenity. Those who are low onfrequency and high in intensity often expe-rience depression or other strong negativeemotions. Low-frequency persons who arealso low in intensity experience affect char-acterized by mild unhappiness. Thus, in termsof subjective well-being, the intensity dimen-sion implies that low- and high-intensity per-sons typically experience happiness and un-happiness in very different ways.

Wessman and Ricks (1966) also proposeda two-dimensional model of affective experi-ence. They denned their dimensions as theamount of positive affect as opposed to neg-ative affect a person experiences, or averagehedonic level, and the emotional variabilitya person exhibits day-to-day. Note that theconstructs of variability and intensity are notsynonymous. Variability is the standard de-viation of mean affect measured on a singlebipolar hedonic scale, and as a result neces-sarily combines frequency and intensity. Onthe other hand, intensity refers to the strengthof a particular affective state, regardless ofhow frequently the state in question is expe-rienced. Although Wessman and Rick's (1966)measure of variability may have to someextent been contaminated with frequency,their descriptions of variable individualssound as if they are describing individualswho would score high on our intensity di-mension. Similarly, our frequency dimensionis congruent with their description of happypersons. Thus we believe that in our work weborrow from Wessman and Ricks, but clarifythe subjective well-being dimensions. Under-wood and Froming (1980) also suggestedaffective dimensions of level and variabilitythat are similar to frequency and intensity.Our intensity dimension may also appearsimilar to the trait of "emotionality" discussedby personality researchers (e.g., Buss &Plomin, 1975). However, it is unclear to whatextent measures and concepts of "emotion-

ality" combine affective intensity with nega-tive affect frequency. This is a topic for futureresearch.

Our suggestion that intensity and frequencyare the basic underlying dimensions of affectis congruent with the broad structure ofemotions proposed by Russell (1980). How-ever, whereas Russell's pleasant-unpleasantdimension appears to be one embodying themagnitude of positive versus negative mo-mentary affect, the frequency dimension pro-posed here refers to the frequency of positiveand negative affect when persons are assessedover time. Russell's arousal dimension maybe congruent with our intensity dimension.Although it is encouraging that the two sys-tems have similarities, the structure of affectexperienced by persons over time does notnecessarily have the same structure as mo-mentary emotional states.

One significant contribution our researchcan make is to help clear up the controversysurrounding the relation of positive and neg-ative affect. Specifically, one group of re-searchers finds that positive and negativeaffect are inversely related (Brenner, 1975;Kammann et al., 1979), whereas others reportindependence (Bradburn, 1969; Bryant &Veroff, 1982; Diener & Emmons, 1984; Zevon& Tellegen, 1982). Diener and Emmons's(1984) work represents an initial step forwardin clarifying this confusion. What they didwas to sample affect over varying periods oftime (i.e., from moments to weeks). Theirfindings indicate that positive and negativeaffect are inversely correlated at particularmoments in time, but that the correlationbetween the two decreased as the time intervalincreased. Thus when one considers a periodof a few weeks or longer in a person's life,the amount of positive and negative affectone experiences is independent, even thoughexperiencing the two emotions simultaneouslyis unlikely. These findings put some of thecontroversy into perspective.

When one examines longer time periods,however, it is still unclear why mean levels ofpositive and negative affect are independent.This is precisely the point at which thedimensions of frequency and intensity becomecritical. When one analyzes positive and neg-ative affect scores in terms of frequency andintensity, the data become readily interpret-


able. Consider the idea that the more fre-quently one feels positive affect, the lessfrequently one feels negative affect. Thisproposition is consistent with Diener andEmmons's (1984) finding that people rarelyexperience strong negative and positive affectat the same time, and it implies that the twovary inversely in frequency. Consequently, ifpositive and negative affect are a combinationof frequency and intensity, then they mustcovary in terms of intensity in order for meanlevels to be independent. This is exactly whatwe find across all of the studies we conductedfor this report. Because frequency and inten-sity appear to be uncorrelated, the resultinginfluence of the two, when persons are con-sidered over time, is to make positive andnegative affect independent.

We have presented evidence that frequencyand intensity are important dimensions ofaffective structure. Frequency and intensityexplain why some scales show that positiveand negative affect are independent and somescales suggest that they are inversely related,depending on whether their wording tapsmean levels or frequency. When mean levelsof affect are correlated, with intensity par-tialed out, a strong inverse correlationemerges. This finding suggests that the inten-sity dimension is at least partly responsiblefor the low correlation between average pos-itive and negative affect.

Should investigators measure positive andnegative affect or intensity and frequency?Depending on the purposes of the investigator,both mean levels of affect and frequency andintensity could be of interest. Note that meanpositive and negative affect are rotationalvariants of each other; that is, each can bederived from the other. If subjective well-being scales assess mean levels of affect overtime, one will find independence, whereas ifthey measure frequency, positive and negativeaffect will vary inversely. This pattern hasindeed held for all of the scales we haveencountered, the majority of which appearto measure the frequency dimension only. Ifresearchers want to assess emotional intensity,but do not want to use a longitudinal ap-proach, then Larsen's (1985) Affect IntensityMeasure is appropriate.

Although intensity may explain much ofthe independence between positive and neg-

ative affect, this does not mean that otherfactors may not operate as well. For example,Warr et al. (1983) showed that good environ-mental events correlate with positive affectand bad events with negative affect, eventhough the number of good and bad eventsdo not correlate with each other. We havereplicated these findings, which suggest thataverage positive and negative affect may beindependent for reasons in addition to inten-sity. Although much of the independencebetween positive and negative affect can beaccounted for by their covariation on theintensity dimension, certain factors may in-fluence one or the other to a greater extent.The two do not covary perfectly in terms ofintensity, and therefore certain influencesprobably affect either positive or negativeintensity, but not both. This reasoning sug-gests that there may be some inherent inde-pendence between average positive and neg-ative affect in the sense that a causal variablecan influence one, but not the other. Thusthere is justification for still measuring averagepositive and negative affect, as well as intensityand frequency. However, the covariation ofpositive and negative affect intensity un-doubtedly explains a large portion of thestatistical independence of mean levels ofpositive and negative affect.

Taking all the findings into consideration,the two dimensions of frequency and intensityappear to have heuristic value. In addition,they seem to be stable and theoreticallymeaningful. However, as of yet, we do notunderstand the underlying processes that me-diate these variables. Preliminarily, we havesome evidence that emotional intensity issimply related to how subjectively good orbad the events happening to the person are.We have also found that persons who expe-rience more intense emotions show greatervariations of emotion across various situa-tional dimensions. Finally, we have foundpreliminary evidence for peripheral physio-logical differences between high- and low-intensity individuals. Researchers should tryto tease apart the mechanisms underlying theintensity dimension. It appears also to haveimplications for theories of subjective well-being, for the psychopathology of affectivedisorders, and for personality assessment, allof which should be explored. Finally, some


may question the general utility of self-reportsof affect or our daily reports in particular.However, there are many reasons to haveconfidence in at least moderate validity forsuch reports (see Diener, 1984, for a review).The measures show low correlations withsocial desirability, but theoretically meaningfulcorrelations with other variables. Weinstein(1982) found that self-reported happiness wasstrongly related to an unobtrusive measureof smiling and laughing in an interview.Correlations between self-ratings and expertratings of affect averaged an encouraging r =.52. Thus although some distortion or biasundoubtedly does enter these measures, theydo seem to contain substantial amounts ofvalid variance.

References

Allport, G. (1961). Pattern and growth in personality.New York: Holt, Rinehart & Winston.

Bradbura, N. M. (1969). The structure of psychologicalwell-being. Chicago: Aldine.

Brenner, B. (1975). Quality of affect and self-evaluatedhappiness. Social Indicators Research, 2, 315-331.

Bryant, F., & VerofT, J. (1982). The structure of psycho-logical well-being: A sociohistorical analysis. Journalof Personality and Social Psychology, 43, 653-673.

Buss, A. H., & Plomin, R. (1975). A temperament theoryof personality development New York: Wiley.

Cherlin, A., & Reeder, L. G. (1975). The dimensions ofpsychological well-being. Sociological Methods and Re-search, 4, 189-214.

Crowne, D. P., & Marlowe, D. (1964). The approvalmotive. New York: Wiley.

Diener, E. (1984). Subjective well-being. PsychologicalBulletin, 95, 542-575.

Diener, E., & Emmons, R. A. (1984). The independenceof positive and negative affect. Journal of Personalityand Social Psychology, 47, 1105-1117.

Diener, E., & Griffin, S. (1984). Happiness and lifesatisfaction: A bibliography. Psychological Documents,14, 11. (Ms. No. 2623)

Diener, E., Larsen, R. J., & Emmons, R. A. (1983).Person affect and subjective well-being: Intensity andfrequency as basic dimensions. Unpublished manu-script, University of Illinois at Urbana-Champaign.

Epstein, S. (1979). The stability of behavior: I. Onpredicting most of the people much of the time.Journal of Personality and Social Psychology, 37, 1097-1126.

Epstein, S. (1983). A research paradigm for the study ofpersonality and emotions. In H. Howe & R. Dienstbier(Eds.), Nebraska symposium on motivation: 1982 (pp.91-154). Lincoln: Nebraska University Press.

Eysenck, H. J., & Eysenck, S. B. G. (1964). Manual ofthe Eysenck Personality Inventory. San Diego, CA:Educational and Industrial Testing Service.

Harding, S. D. (1982). Psychological well-being in GreatBritain: An evaluation of the Bradbum affect balancescale. Personality and Individual Differences, 3, 167-175.

Jackson, D. N. (1974). Personality Research Form manual.Goshen, NY: Research Psychologists Press.

Kammann. R., Barter, J., Irwin, R., & Dixon, G. (1979).Properties of an inventory to measure happiness andpsychological health. New Zealand Psychologist, 8, 1-9.

Kozma, A., & Stones, M. J. (1980). the measurementof happiness: Development of the Memorial Universityof Newfoundland Scale of Happiness (MUNSH). Journalof Gerontology, 35, 906-912.

Larsen, R. J. (1985). Theory and measurement of affectintensity as an individual difference characteristic.Dissertation Abstracts International, 45, 2297. (Uni-versity Microfilms No. 84-22112)

Murray, H. A. (1938). Explorations in personality. NewYork: Oxford University Press.

Ozer, D. (1982). Consistency, coherence, and convergence:A framework for the study of personality. Unpublisheddoctoral dissertation, University of California at Berke-ley.

Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press.

Russell, J. A. (1980). A circumplex model of affect.Journal of Personality and Social Psychology, 39, 1161-1178.

Underwood, B., & Fronting, W. J. (1980). The moodsurvey: A personality measure of happy and sad moods.Journal of Personality Assessment, 44, 404-414.

Warr, P. (1978). A study of psychological well-being.British Journal of Psychology, 69, 111-121.

Warr, P., Barter, J., & Brownbridge, G. (1983). On theindependence of positive and negative affect. Journalof Personality and Social Psychology, 44, 644-651.

Weinstein, L. (1982). Positive contrast due to happiness.Bulletin of the Psychonomic Society, 19, 97-98.

Wessman, A. E., & Ricks, D. F. (1966). Mood andpersonality. New York: Holt.

Zevon, M. A., & Tellegen, A. (1982). The structure ofmood change: An idiographic/nomothetic analysis.Journal of Personalitv and Social Psychology, 42,111-122.

Received August 30, 1983

Revision received December 26, 1983