-
Journal of Personality and Social Psychology2000, Vol. 79, No.
4, 644-655
Copyright 2000 by the American Psychological Association,
Inc.0022-351
-
EMOTION AND AGE 645
tual performance occurs in middle age. According to this
theory,children must suppress idiosyncratic affective judgments
aboutcollectively shared, symbol systems so that they acquire
uniform,culturally consistent representations of the world
(Labouvie-Vief& DeVoe, 1991; Labouvie-Vief, Hakim-Larson, et
al. 1989). Adultintellectual development, in contrast, involves the
reintegration ofsubjective information into existing knowledge
structures. Theo-retically, this increased complexity in cognitive
operations is as-sociated with increasingly more complex and
adaptive emotionalresponses and perhaps with greater flexibility in
coping with newlife events (Diehl, Coyle, & Labouvie-Vief,
1996).
A second theory relevant to emotional functioning in adulthoodis
socioemotional selectivity theory (Carstensen, 1993,
1995;Carstensen, Gross, & Fung, 1997; Carstensen, Isaacowitz,
&Charles, 1999). Although this theory also acknowledges
experi-ence as an important factor in emotional development, it
focuseson perceived time left in life rather than past experience.
Thetheory contends that the distinctly human ability to
consciouslyand subconsciously monitor time plays a fundamental role
inmotivation and emotion, providing the structure within whichgoals
are set, pursued, and evaluated. Because mortality places
theultimate constraint on time, chronological age is associated
withchanges in goals. Essentially, the theory contends that two
primarytrajectories of social motives operate throughout life: the
emotiontrajectory and the knowledge trajectory. The former is
character-ized by motives to achieve emotional satisfaction and
meaning, thelatter by motives to acquire new information and to
achieve indomains that are relevant to successful adaptation in the
future(e.g., educational and occupational domains). The central
changein adulthood is a shift in the salience of social goals.
Youngeradults, having much to learn and relatively long futures for
whichto prepare, are motivated by the pursuit of knowledgeeven
whenthis requires that emotional well-being be suppressed. For
olderadults, the reverse trend appears. Facing relatively shorter
futuresand having already accrued considerable knowledge about
others,older adults prioritize emotional goals because they are
realized inthe moment of contact rather than banked for some
nebulous futuretime.
The theory stresses that age does not entail the relentless
pursuitof happiness but rather the satisfaction of emotionally
meaningfulgoals, which entails far more than simply feeling good.
Findingmeaning in existing relationships, even conflictual ones,
emergesas a central task in later life. Emotional experience is
subsequentlyexpected to be more complex, and the experience of
mixed emo-tions, more frequent. In short, socioemotional
selectivity theorysuggests that constraints on time directly
influence emotionalexperience such that emotional states are
increasingly mixed.Whether pleasure or joy, sadness or pain,
knowledge that anexperience will soon end changes the emotional
experience itself.Rather than simply prompting negative emotions
related to antic-ipated loss, moments are savored, appreciated both
for what theyare and for their temporal fleetingness.
Despite these relatively optimistic empirical and
theoreticalpictures, well-documented declines in later life leave
an uneasinessthat positive portrayals of emotion in later life may
be overblown.First, much of the literature on emotion, and to the
best of ourknowledge all of the literature on emotion regulation,
has beenbased on global self-reports, a practice that may be
particularlyprecarious when questions elicit older adults* implicit
theories
about their own maturation (McFarland, Ross, & Giltrow,
1992).That is, if research participants believe that people
"should" con-trol their emotions better as they get older, they are
likely to saythat they do so. Moreover, global evaluations of life
are highlycognitive and involve comparisons with the past and the
present, aswell as with idiosyncratic standards (Schwarz, Park,
Knaueper, &Sudman, 1999). Subsequently, such evaluations are
susceptible toinfluence by cohort-specific experiences and mores
(Elder, Odum,& Hareven, 1994), as well as by memory of past
events (Levine &Bluck, 1997). A second important concern is
that even thougholder people may be able to respond well to
emotional tasks inlaboratory settings, emotional dysregulation may
be evident in lessstructured settings.
The purpose of the present study was to assess the
frequency,intensity, and the complexity of emotional experiences as
theyoccur in everyday life. Our general hypothesis was that
relative toyounger people, older people would show evidence of
improvedemotion functioning, including more differentiated
emotional ex-periences and better regulation of their emotional
states. By re-peatedly sampling the same participants over time, we
were able toexamine the frequency and intensity with which positive
andnegative emotions are experienced, as well as the stability
ofnegative and positive states over time and the complexity
ofemotional experience. Below we elaborate four experimental
hy-potheses about age differences in emotional experience.
Hypothesis 1. Older people experience negative emotions less
fre-quently than younger adults and experience positive emotions
just asfrequently as younger adults. According to socioemotional
selectivitytheory, increasing age is associated with greater
appreciation of lifeand greater investment in emotionally
meaningful social relationships.The theory predicts that this
emphasis on emotionally meaningfulgoals improves emotional
experience in everyday life.
Hypothesis 2. The intensity of positive and negative emotional
expe-rience is comparable across age groups. Socioemotional
selectivitytheory predicts that goal-directed behavior aimed at
obtaining emo-tionally meaningful goals results in less frequent
negative emotions.However, once negative emotions are elicited, the
theory makes noclaim about the intensity of the experience.
Hypothesis 3. Older as compared with younger adults show
differen-tial stability of emotional experience such that positive
states aremaintained longer and negative states are terminated more
quickly.Surveys that ask people how well they regulate their
emotions suggestthat where there are differences, older people
report greater control. Inthis study, we obviate global judgments
about emotion control byexamining whether positive states last
longer and negative statespersist for shorter periods in older
adults.
Hypothesis 4. Emotional experience is more complex in older
ascompared with younger adults. Because the pursuit of
emotionallymeaningful goals often entails mixed emotions, we
anticipate a morecomplex dimensional structure to the emotional
experience of olderadults.
Method
Sample
One hundred eighty-four African American and European
Americanresearch participants, ranging in age from 18 to 94 years
of age, (M 55,SD = 20.4), were recruited by a survey research firm
from the San
-
646 CARSTENSEN, PASUPATHI, MAYR, AND NESSELROADE
Francisco Bay area to participate in an experience-sampling
study ofemotional experience. Ethnic composition of the sample was
restricted tothese two ethnic groups rather than sampling the
ethnic diversity of the Bayarea so that sufficient numbers of
participants in subsamples would allowfor statistically meaningful
analyses. Thirty-one percent of the sample wereAfrican American;
the remaining 69% were European American. Forty-one percent of the
sample comprised blue-collar workers, and 59% werewhite-collar
workers; 54% of the participants were women, and 46% weremen.
Education ranged from 5 to 22 years (M = 15.0, SD = 2.7).
Gender,blue- or white-collar status, and race were distributed
evenly across age. Asshown in Table 1, the sample was diverse along
many dimensions.
Measures
Although our principal aim was to sample emotional experiences
ineveryday life, we also assessed health, personality, and verbal
fluencybecause each of these factors may influence at least some
features ofemotional experience or performance on the sampling task
(see, e.g.,McCrae & Costa, 1991; Watson & Pennebaker,
1989).
Emotion sampling booklet. On a 7-point scale that ranged from 1
(notat all) to 7 (extremely), participants indicated the degree to
which they werefeeling each of 19 emotions or feeling states.
Ratings greater than 1, thus,indicated that the emotion was present
and, consequently, both frequencyand intensity are captured in a
single rating. The list of emotions includedanger, guilt, pride,
sadness, happiness, fear, accomplishment, shame,amusement,
anxiety/worry, joy, contentment, irritation, frustration,
disgust,interest, embarrassment, boredom, and excitement. An other
blank wasalso provided on the response sheet to allow participants
to record addi-tional emotions not included on the sampler. A
week's supply of emotionresponse sheets were bound in a 5 in. by 5
in. pad for easy transport duringthe week of data collection.
Cornell Medical Index Health Questionnaire (CMI). The CMI
(Brod-man, Erdmann, & Wolff, 1949) is a widely used 195-item
index of physicaland mental health problems that allows the
computation of a general healthindex as well as subscales that
represent functioning in specific organsubsystems and symptoms
associated with specific psychological syn-dromes. Participants
report whether they experience each of the 195symptoms. We computed
two broad indexes from the CMI, one represent-ing the total number
of recent symptoms of physical illness and the otherrepresenting
the total number of recent symptoms of mental illness. Anexample of
a physical illness symptom item is "Are you troubled byconstant
coughing?" A sample mental illness symptom item is "Do youhave to
be on your guard even with your friends?"
Category instance fluency (Undenberger, Mayr, & Kliegl,
1993). As ameasure of verbal fluency, participants were asked to
name as manydifferent kinds of animals as possible in 90 s. This
test shows a strongrelationship to general intellectual ability and
has been extensively usedwith older adults.
Adjective checklist (John, Donahue, & Kentle, 1991). This is
a listof 54 adjectives presented in the form of self-descriptive
sentences. Ad-
Table 1Demographic Characteristics of the Sample
Characteristic Sample description
Age (in years)Education (in years)SexEthnicitySocioeconomic
statusMarital status
Number of children
M = 55.0, SD = 20.4, range = 18.0-94.0M = 15.0, SD = 2.7, range
- 5.0-22.054% women, 46% men31% African American, 69% European
American41% blue-collar, 59% white-collar26% single, 43% married,
19% widowed, 13%
divorcedM = 1.6, SD = 1.6, range = O-9
jectives representing all of the Big Five factors of personality
are repre-sented. Participants indicate whether a given statement
describes them byplacing a check next to it. Example items include
"I am talkative" and "Ican be somewhat careless." We computed
summary scores for each of theBig Five factors: Neuroticism,
Extraversion, Openness to Experience,Agreeableness, and
Conscientiousness.
ProcedureFollowing initial screening by the survey research firm
to ensure that
participants met recruitment criteria for the project,
participants werescheduled at their convenience for an initial
interview at Stanford Univer-sity or at the offices of the San
Francisco-based survey research firm thatdid the initial
recruiting. Participants were informed that the purpose of thestudy
was to examine feelings in everyday life. After obtaining
informedconsent and obtaining background information, such as
education level, thefollowing measures were administered: Category
Instance Fluency, CMI,and the Adjective Checklist.
At this point, participants were provided with detailed
instructions aboutthe experimental procedures, familiarized with
the operations of the elec-tronic pager (e.g., how to set it for
motion or sound, how to indicate thatthey received the page by
pushing a button, etc.), and instructed tocomplete the emotion
response sheets each time they were signaled. Next,two practice
trials were administered while participants were still in
thelaboratory so that responses could be reviewed with the
experimenter priorto beginning the study. The participant was left
alone; the intervieweractivated the pager from another room; the
participant completed thequestionnaire; and, on returning to die
room, the interviewer reviewed theparticipant's responses,
clarified any apparent mistakes, and answeredquestions.
During the ensuing week, participants were paged five times each
day.Paging times were determined by random selections from all
possible10-min intervals between 9 a.m. and 9 p.m. The only
constraint onsampling times was that participants were not sampled
more than oncewithin a single 20-min period. At the end of each
day, participants returnedthe day's completed response sheets by
mail in pre-addressed, stampedenvelopes, allowing us to monitor
responses during the data collectionperiod and assuring at least
rough adherence to the experimental protocol.Participants were
encouraged to telephone the laboratory if proceduralquestions or
problems arose and periodic calls were made to participants aswell
to ensure that the highest quality data were obtained.
After participants completed the week-long experience-sampling
datacollection, they returned to the laboratory for a follow-up
interview, atwhich time they returned the pagers and were
debriefed. Participants werepaid $125 for their participation.
Results
We organize our results into four sections. The first
sectiondescribes data reduction and preliminary analyses. The
secondsection of the results reports findings from analyses that
examineage differences in the frequency and intensity of emotional
expe-rience, controlling for individual differences that may
influenceemotional experience (Hypotheses 1 and 2). In this section
we alsoexamine the consistency of age differences across ethnic,
gender,and socioeconomic lines.
The third section of the results addresses emotion
regulation(Hypothesis 3), and the fourth section concerns the
complexity ofemotional experience (Hypothesis 4). In these latter
two sections,hypotheses are tested on the basis of
within-individual variabilitythat allowed us to examine emotional
complexity and the temporalexperience of emotional experiences.
Here, too, we examined theconsistency of effects across ethnicity,
gender, and class. Because
-
EMOTION AND AGE 647
our measure of differentiation was novel, we explored its
relation-ship to other measures, such as personality, mental
health, andintelligence.
Data Reduction and Preliminary AnalysesRatings of 19 emotions
were obtained on 35 separate occasions,
generating a total of 665 experience sampler data points
perparticipant. Data were reduced in the following way.
Frequencywas represented as the proportion of times across the 35
samplingpoints that a participant acknowledged that he or she
experiencedthe emotion to some degree, namely, ratings were greater
than 1.Intensity of emotional experience was calculated by
computing theaverage rating for each felt emotion (see Schimmack
& Diener,1997, for a discussion of this kind of decomposition).
Table 2presents the average frequencies and intensities for
specific emo-tions across the entire sample. As can be seen, people
endorsednegative emotions relatively infrequently and positive
emotionsrelatively frequently, a finding consistent with earlier
experiencesampling studies (Diener & Diener, 1996). For each
specificnegative emotion, some people indicated that they did not
expe-rience the emotion on any of the sampled occasions;
however,only 2 individuals failed to endorse any negative emotions
at allduring the experience-sampling period.1
The above procedure resulted in each participant having 38scores
indicating the frequency and intensity with which he or shereported
19 emotions over the sampling period. To reduce thenumber of
statistical tests in our analyses and to increase thereliability of
our measures, we collapsed these 38 scores into four
Table 2Means and Standard Deviations of Frequency and
intensityof Experiencing Specific Emotions for the Entire
Sample
Emotion
AngerSadnessFearDisgustGuiltEmbarrassmentShameAnxietyIrritationFrustrationBoredom
HappinessJoyContentmentExcitementPrideAccomplishInterestAmusement
Frequency"
M
.20
.28
.20
.21
.17
.17
.14
.44
.39
.39
.30
.89
.78
.90
.69
.70
.74
.88
.73
SD
Negative
.24
.30
.28
.25
.27
.27
.25
.32
.28
.28
.32
Positive
.17
.27
.16
.32
.34
.27
.19
.27
M
3.313.313.133.462.993.232.933.283.393.403.38
4.354.104.403.984.094.404.553.95
Intensity
SD
0.911.051.041.090.901.090.930.930.900.860.95
0.971.021.000.941.101.000.990.86
n
147159138157128131112176177176149
1841811S4181180184184182
indicators of emotional experience: average frequency of
negativeemotions, average frequency of positive emotions, average
inten-sity of negative emotions, and average intensity of positive
emo-tions. Thus, the average frequency of experiencing positive
emo-tions reflects the average of the proportion of times a
personexperiences each of the positive emotions. Because there is
somedebate about the dimensionality of emotional experience and
be-cause the 38 scores were derived from only 19 ratings,
thisaggregation was verified by factor analysis.
We conducted a descriptive factor analysis using varimax
rota-tion with mean substitution for missing values. This revealed
thatfrequency and intensity scores for the emotions were
reasonablywell-characterized by a four-factor solution (69% of the
totalvariance was accounted for by this solution). Although six
factorswith eigenvalues above 1 could be extracted, Factor 5 had
highloadings only for the intensity of Fear (.60) and of Shame
(.80),whereas only the intensity of Boredom loaded significantly
onFactor 6 (.80). Together Factors 5 and 6 accounted for only
anadditional 5% of variance in the data. Inspection of the scree
plotshowed a clear dissociation of the latter two from the first
fourfactors. Thus, a four-factor solution was used (as per
Tabachnik &Fidell, 1989). This solution clearly reflects the
frequency andintensity with which positive and negative emotions
were experi-enced and was reliable both when cases with missing
values weredeleted and when oblique rotation was used. Table 3
presents thevarimax rotated factor loadings of frequency and
intensity vari-ables on the four factors as well as eigenvalues and
varianceaccounted for by each factor. As shown in Table 3, Factors
1 and 2reflect the frequency and intensity of negative affect,
respectively.Factors 3 and 4, respectively, represent the frequency
and intensityof positive affect.2
Hypotheses 1 and 2: Age Is Related to the Frequency butnot the
Intensity of Emotional Experience
We hypothesized that the frequency of negative, but not
posi-tive, emotional experience decreases across age cohorts, and
wehypothesized that intensity of emotion would not distinguish
age,
*n = 184.
1 Findings remain essentially unchanged when very high and very
low
scorers are eliminated.2 Oblique rotations suggested some
relationships between the factors,
with the factors for intensity of positive and intensity of
negative emotionscorrelated at .34, and the factors for intensity
and frequency of positiveemotions also correlated (r = .30).
Frequency of experiencing negativeemotions and frequency of
experiencing positive emotions were alsocorrelated, though less
strongly (r = .21). No other interfactor correlationswere above
.17. Relationships between the aggregate scores (not factorscores)
used in our analyses mirrored these oblique factor correlations,
andwere somewhat stronger. Intensity of negative emotion and
intensity ofpositive emotion showed a moderate relationship (r =
.40, p < .01).Frequency and intensity of positive emotion were
also correlated (r .42,p < .01), as were frequency of positive
and frequency of negative emotions(r .29, p < .01). All other
correlations were much lower (maximumabsolute value r = .16). Both
interfactor correlations and correlationsbetween our variables
imply the existence of individual differences inemotional intensity
and in the general frequency with which emotions areexperienced.
These relationships (between the frequency and intensity ofpositive
and negative emotion) did not vary as a function of age at this,
thebetween-subjects, level.
-
648 CARSTENSEN, PASUPATHI, MAYR, AND NESSELROADE
Table 3Varimax Rotated Factor Loadings for Frequency and
Intensityof Positive and Negative Affect
Emotion
AngerSadnessFearDisgustGuiltEmbarassmentShameIrritationFrustrationAnxietyBoredomHappinessJoyContentmentExcitementPrideAccomplishmentAmusementInterest
Eigenvalues% of Variance
AngerSadnessFearDisgustGuiltEmbarassmentShameAnxietyIrritationFrustrationBoredomHappinessJoyContentmentExcitementPrideAccomplishmentAmusementInterest
Eigenvalues% of Variance
NEfrequency
PEintensity
Frequencies
.86
.85
.87
.88
.90
.88
.92
.86
.85
.78
.65
.06
.13
.02
.27
.24
.14
.19
.09
9.3425
- .04- .04- .01
.02- .04
.14
.06- .09- .14- .14-.17
.16
.32
.02
.30
.22
.16
.26
.07
8.3522
Intensities
- .09.03
-.05- .15-.01- .11-.07
.17
.06
.14- .04-.07
.03- .24-.01- .02- .07- .05- .07
9.3425
.18
.20
.07
.08
.12
.12
.07
.19
.10
.03
.15
.84
.92
.76
.84
.75
.78
.81
.72
8,3522
PEfrequency
.09
.11
.08
.01
.11
.09
.10
.06
.09
.13
.22
.85
.79
.79
.77
.76,85.80.79
5.6015
-.05- .02- .13-.08- .15- .02
.01-.07- .07- .03-.08
.34
.04
.21
.07
.18
.33
.14
.32
5.6015
NEintensity
.03- .03
.03
.03
.00- .08
.01- .04- .03- .05-.17-.12- .02
.00- .11- .13-.11- .18- .14
2.005
.66
.67
.66
.64
.56
.64
.53
.72
.83
.84
.45
.18
.16
.15
.22
.16
.27
.30
.22
2.005
Note. NE = negative emotions; PE - positive emotions. Boldface
indi-cates factor loadings above .40.
with the exception of excitement. Four regression analyses
werecomputed to explore relationships between the frequency
andintensity of experiencing positive and negative affect and
age.Both linear and quadratic age trends were explored. No
significantage effects were obtained for the frequency or the
intensity ofpositive affect, nor were there age effects for the
intensity withwhich negative affect was experienced, all Fs(l, 181)
< 1.0, ps >3. However, as predicted, age was associated with
the frequencyof experiencing negative affect F(2, 182) - 6.0, p
< .01. This
effect has both a linear and a nonlinear component, as shown
bythe joint effects of linear age, B = - .02, 0 = -1.47, z(157)
=-3.42, p < .01, and age squared, B = .0001 = 1.41,r(157) =
3.30, p < .01. Just to illustrate these effects in a
moreintuitive way, simple correlations between age and frequency
ofnegative emotion before and after age 60 were computed.
Thesecorrelations reveal a decrease in the frequency of negative
emo-tions from 18 to 60 years (simple r = .29, p < .01). After
60years, the decrease ceases (r = .14, ns) and characterizes
thepattern from age 60 onward.3 This effect is shown in Figure
1.
Our next step was to explore the robustness of age as a
predictorof emotional experience in conjunction with factors known
toinfluence emotions. We added the following predictors to
theregression equations described above: personality
(Neuroticism,Extraversion, Openness to Experience, Agreeableness,
and Con-scientiousness), health (self-reported mental health and
self-reported physical health), and demographic variables
(ethnicity,gender, and socioeconomic status). In addition, because
intellec-tual functioning reliably declines with age, we included
the mea-sure of general intellectual ability (i.e.. Category
Instance Flu-ency). All variables were entered simultaneously.
The results, displayed in Table 4, show that age findings
weremaintained even with these additional variables included.
Con-cerning the frequency of negative emotions, more mental
healthsymptoms and higher neuroticism were additionally
associatedwith higher frequencies of negative emotion. For the
frequency ofpositive emotions, African Americans and those who were
moreopen and agreeable reported experiencing positive emotions
morefrequently.
Concerning the intensity of emotions, negative affect was
pre-dicted by gender and socioeconomic status as well as by
mentalhealth symptoms, with women, blue-collar workers, and
individ-uals who were more extraverted endorsing more intense
negativeaffect. Intensity of positive affect was associated with
ethnicity andextraversion, with African Americans and more
extraverted indi-viduals endorsing more intense positive affect. We
also exploredthe consistency of the pattern of age effects and
noneffects inpredicting affective experience across gender, race,
and socioeco-nomic status by looking for interactions between age
and thesedemographic factors in predicting our emotional experience
vari-ables. There were no interactions for age and gender,
socioeco-nomic status, or race in predicting any of these facets of
emotionalexperience.
3 We chose 60 as an illustrative point because it represents
approxi-
mately the bottom of the line depicted in Figure 1. Choosing
other agecutoffs, such as 40,50, or 65, results in differences in
the magnitude but notthe pattern of relationships. To be concrete,
the correlations for the youngerportion of the sample are .46, .24,
.29, and .24, respectively, forages 40, 50, 60, and 65 years. The
same correlations for the older portionof the sample are .04, .17,
.14, and .19. In no case is the age relationshipnear zero for the
younger portion of the samplea decline is always seen.What happens
after the selected cutoff is somewhat more variable, rangingfrom
zero relationships to small positive correlations with negative
emo-tion. However, again, these are meant only to illustrate the
pattern detectedin the regression analyses.
-
EMOTION AND AGE 649
Table 4Emotional Experience, Age, and Other Variables
Predictors
Demographic variablesRaceSexSocio-economic status
HealthPhysicalMental
PersonalityNeuroticismExtraversionOpennessAgreeablenessConscientiousness
Verbal intelligenceLinear ageQuadratic ageOverall K2Regression,
F(13, 129)
r
.07- .04
.03
.11
.28**
.20**
.01
.03- .11- .18*- .02-.08- .04
.202.50**
NA Frequency
P
.11-.01-.07
.00
.17t
-20t- .01- .07
.05- .09- .03
-1.30*1.25*
B
.06-.01- .03
.0001
.01
.17- .01
.09
.05- .10
.0004- .01
.0001
r
.17*- .13t
.03
.01
.06
-.1122**.12.16*.17*
- .04.07.06.14
1.60t
PA Frequency
.21*- .13
.02
- .06.06
.08
.1115t.17f.05.04.39
- .38
B
.10- .05
.00
-.001.002
.70
.90
.18
.19
.05
.001
.004
.00
r
.16*
.18*
.25**
.16*
.22**
.03
.11
.00
.01- .02- .02- .06- .06
.172.00*
NA Intensity
IS
.12
.18*
.15
.02
.17f
.04
.17*- .06 -
.04
.02- .02 -
.37- .47 -
8
.19
.27
.23
.002
.02
.10
.50
.24
.15
.06
.001
.01
.0001
r
.37**
.07
.15*
.04
.03
-.24**.29**.14t.24**27**
- .10.06.05.28
4.00**
PA Intensity
.32**
.05
.12
- .04.04
- .09.20*.11.11.09
- .02.17
-.21
B
.62
.09
.22
-.003.005
- .29.69.51.53.38
-.002.007.00
Note. NA = negative affect; PA = positive affect. Being female,
being blue-collar, and being African American are represented by
larger values.tp < .10 (marginally significant). *p < .05. **
p < .01.
Finally, we looked specifically at age differences in the
intensityof excitement. Contrary to our hypothesis of reduced
intensity ofexcitement given the literature reviewed above, once
again wefailed to reveal a significant relationship between age and
theintensity with which excitement was experienced, B = .001 J3
=.08,1(182) = 1.1, p > .25.
Hypothesis 3: Compared With Younger People, OlderPeople Better
Regulate Emotional Experience
We operationalized emotion regulation as the maintenance
ofdesirable emotional states (defined as those where the
individual
.341
.32
ftM .30
S-28"8g .26
u.
feels more positive or less negative affect than usual) and
thecessation of undesirable emotional states (defined as states
wherethe person is feeling less positive or more negative than
usual). Theformer reflect adaptive aspects of emotional stability,
and the latterreflect adaptive aspects of emotional lability. To
test Hypothesis 3,we computed four scores that reflect these four
aspects of emotionregulation.
For each sampling occasion, participants were classified as
highon positive affect relative to their own idiosyncratic ally
calculatedmean across all sampled situations or as below or
equivalent totheir own idiosyncratically calculated mean across
situations. A
I18-34 35-64
Age in Years65-94
Figure 1, Frequency of negative affect across the life span.
-
650 CARSTENSEN, PASUPATHI, MAYR, AND NESSELROADE
similar split was made for negative emotional states. We
thencomputed four conditional probabilities: (a) maintaining high
pos-itive states: the probability that, given participants were
morepositive than average on Occasion 1, they would be more
positivethan average on the subsequent sampling occasion; (b)
maintain-ing the absence of highly negative states: the probability
that,given participants were less negative than average on Occasion
1,they would be less negative than average on the subsequent
occa-sion; (c) moving from low positive states to high positive
states: theprobability that, given participants were less positive
than averageat Occasion 1, they would be more positive than average
atOccasion 2; and (d) moving from highly negative states to
lownegative states: the probability that, given participants were
morenegative than average at Occasion 1, they would be less
negativethan average on the following occasion. Note that these
scores arenot necessarily related to the overall frequencies of
experiencingpositive or negative emotions, because these scores
reflect some-thing about the temporal distribution of positive and
negativestates. Also, note that these scores do not simply reflect
stability ofpositive or negative states, but rather, adaptive
features of stabilityand adaptive features of lability.
Age-related patterns were reasonably positive. Older men
andwomen showed greater stability of highly positive states (r =
.17,p < .05), and this pattern was consistent across all gender,
eth-nicity, and socioeconoraic groups. Age was also correlated
withstability of low negative states (r .20, p < .001). Age
wasnegatively, though not significantly, correlated with moving
froma low positive to a high positive state (r = .13, p < .11).
Therewere no quadratic trends for age and these effects were
constantacross genders, ethnicities, and socioeconomic status
groups. Fi-nally, age was uncorrelated with the likelihood of
moving from ahighly negative state to a low negative state (r =
.04, ns). How-ever, here a quadratic trend was present. Together,
linear age (B =.014, 0 = 1.47, p < .002) and quadratic age (B -
- .0001, /3 =1.44, p < .002) accounted for 6% of the variance in
the likeli-hood of moving from high negative states to low negative
states.This pattern was consistent across both ethnicities and
socioeco-nomic status groups, but did appear to vary by gender
(LinearAge X Gender interaction: A*2 = .03, &F = 6.7, p <
.02).Examining the regression equation separately for men and
womenrevealed that the pattern of age effects (linear and
quadratic) wasidentiqal for both genders, but the magnitude was
lower for women(overall R2 = .0(5) than for men (overall J?2 =
.13). In both cases,however, the age regression attained
statistical significance. To geta feel for what these results
suggest, we again computed correla-tions between age and the
lability of highly negative states sepa-rately for those under 60
(r = .30, p < .002) and those over 60(r = .15, p > .19).
Thus, older adults in our sample, up to someage, were less likely
to remain in a highly negative state overoccasions than were
younger adults. At some point in adulthood,this trend reverses,
although perhaps not significantly,4
To summarize, then, older adults were more likely to
maintainhighly positive states and were more likely to maintain the
absenceof negative emotional states. Thus, the stability indicators
suggestgreater stability of emotional experience in older adults
for theadaptive portions of emotional stability (i.e., one could
also bevery stable by being continuously very angry). This raises
thequestion of whether older adults are simply more stable in
general.This can be examined by computing a phi correlation for
positive
emotion and for negative emotion. This is equivalent to a
cross-lagged correlation but applied to the categorical states
(above orbelow one's idiosyncratic mean) that we defined for this
analysis.These correlations arc instructive. For positive emotions,
smallage-associated increases in stability (whether of low or high
pos-itive states) are evident (r = .17, p < .02). This is not
the case fornegative emotions (r = .11, p > .15).
Finally, the curvilinear pattern of results obtained for
lability inhighly negative states with age raises the possibility
that this kindof lability accounts for the age differences observed
in the fre-quency of negative emotions. Note that this is not
necessarily thecase. Older and younger adults could have different
frequencies ofnegative emotions without those emotions occurring in
temporallylinked ways. However, if older adults are moving out of
negativestates more quickly than young adults (at least up to some
point inadulthood), this might mean that older adults' better
emotionregulation, as assessed here, accounts for age differences
in thefrequency of negative emotions. We examined this by
computinga hierarchical regression predicting the frequency of
negativeemotion (from Hypothesis 1 above). The probability of
movingfrom highly negative to low-negative states was entered as
the firstpredictor, after which we examined whether age (linear and
qua-dratic) made any additional significant contribution to the
equa-tion. This analysis showed that once lability of highly
negativestates was entered, age (both linear and quadratic)
contributed anadditional 2% of the variance to predicting the
frequency ofnegative emotion F(2, 179) = 2.3, p = .10. Thus,
changes in thefrequency of negative emotion with age may be
interpreted asstemming from changes in the lability of highly
negative states.
Hypothesis 4: Age Is Associated With the Complexity ofEmotional
Experience
To test Hypothesis 4, we computed the eigenvalues of
eachindividual's 19 X 19 emotion ratings correlation matrix on
thebasis of his or her 35 occasions of measurement. We took as
anindex of differentiation the number of eigenvalues greater than
1.Across the whole sample, the average number of
eigenvaluesexceeding 1.0 was 5.8 (SD = 1.1, range = 2 to 9). The
corre-sponding principal components accounted for, on average, 77%
ofthe total variance in emotional ratings across time (SD ~ 3.9).
Theamount of variance accounted for by the principal componentswith
eigenvalues larger than unity was uncorrelated with age (r =.09).
The correlation between age and the number of eigenvalueslarger
than 1.0 was, as predicted, positive and significant (r = .28,p
< .01) and is shown in Figure 2. This evidence of
age-relateddifferentiation held across all levels of ethnicity,
gender, andsocioeconomic class. There was no quadratic trend for
age. Therelationship between differentiation and age was not
accounted forby individual difference variables of personality,
health, or verbalfluency.
4 Again, if these correlations are computed for cutoff points of
40,
50, 60, and 65, the respective correlations in the younger
portion of thesample are .23, .31, .30, and .26. The respective
correlations in the oldersample are .14, .15, .15, and .17. Once
again, the pattern is clear:Highly negative states are increasingly
labile across adulthood, but at somepoint, this increase levels off
or becomes negative.
-
EMOTION AND AGE 651
18-34 35-64Age in Years
Figure 2. Differentiation of emotional experience over time.
65-94
Because our operational measure of differentiation represented
anovel way of thinking about qualitative aspects of
emotionalexperience, it was unclear how greater differentiation was
associ-ated with mental health. For this reason, we conducted
exploratorycorrelation analyses about the relation of our
complexity measureswith other variables. We focused on three
questions. First, isdifferentiation associated with better mental
health? In otherwords, is it meaningfully related to indicators
aside from theemotion sampling data on which complexity scores were
derived?Second, how does complexity relate to the emotional
experiencemeasures (frequency and intensity of negative and
positive affect)?Third, is complexity meaningfully related to other
conceptuallyrelevant measures, specifically, verbal fluency and
neuroticism?Verbal fluency especially may be related to
individuals' capacityto represent situations in complex ways.
Neuroticism, on the otherhand, is a diffuse tendency to perceive
life negatively.
Differentiation was uncorrelated with overall mental health (r
=-.07). Differentiation was negatively associated with the
fre-quency of experiencing both negative affect (r = .30, p <
.01)and positive affect (r .22, p < .01). Differentiation was
notassociated with the intensity of positive affect (r = .06) but
wasnegatively correlated with the intensity of negative affect (r
=- .16, p < .05). It was unrelated to verbal fluency (r = -.03)
butwas negatively associated with neuroticism (r = .19, p <
.05).In sum, differentiation appears to be a positive feature of
emotionalexperience, as it is associated with greater emotional
control andless intense negative affect and with less
neuroticism.
In addition to this factor-based indicator of emotional
complex-ity, we computed a second analysis to examine the degree to
whichindividuals experienced both positive and negative emotions
onthe same sampling occasion. We refer to this feature of
emotionalexperience as "poignancy." Poignancy was computed by
calculat-ing, for each participant, a correlation between positive
and neg-
ative affect across the 35 sampling occasions. On average,
thiscorrelation was - .35 (SD = .33), suggesting that positive
andnegative affect tended not to be present on the same occasion.
Alinear-age effect emerged, with older age associated with
thegreater potential for co-occurrence of positive and negative
emo-tions, (r = .26; p < .01). This effect is shown in Figure 3.
Forpeople under age 60, the average correlation between positive
andnegative emotion within occasions was - .42 (SD = .28),
whereasfor those above age 60, that correlation was - .25 (SD =
.36). Justas in the above analyses, these age differences remained
aftercontrolling for personality, health, and verbal fluency. Age
differ-ences in poignancy also held within race, gender, and
socioeco-nomic classes. Poignancy was unrelated to the frequency of
expe-riencing positive or negative affect, suggesting that
individualswho frequently endorse all emotions are not more likely
to expe-rience negative and positive emotions in the same moment.
Fi-nally, differentiation and poignancy were correlated, (r = .23,
p