Choices and affective reactions to negative life events: An averaging/summation analysis

Judgment and Decision Making, Vol. 3, No. 5, June 2008, pp. 425–434
Choices and affective reactions to negative life events: An
averaging/summation analysis
John J. Seta?
Ashleigh Haire and Catherine E. Seta
University of North Carolina at Greensboro
Wake Forest University
Abstract
Three experiments investigated individuals’ preferences and affective reactions to negative life experiences. Par-
ticipants had a more intense negative affective reaction when they were exposed to a highly negative life experience
than when they were exposed to two negative events: a highly negative and a mildly negative life event. Participants
also chose the situation containing two versus one negative event. Thus, “more negative events were better” when the
events had different affective intensities. When participants were exposed to events having similar affective intensities,
however, two negative events produced a more intense negative affective reaction. In addition, participants chose the
situation having one versus two negative life experiences. Thus, “more negative events were worse” when the events
had similar affective intensities. These results are consistent with an averaging/summation (A/S) model and delineate
situations when “more” negative life events are “better” and when “more” negative life events are “worse.” Results also
ruled out several alternative interpretations including the peak-end rule and mental accounting interpretations.
Keywords: decisions, averaging, summation, affect, negative events, peak-end rule, mental accounting.
1 Introduction
From the A/S model, averaging effects are not rele-
gated to the judgment domain and summation effects are
“More is worse” when it comes to negative life events.
not relegated to the social in?uence domain. Rather, in-
This is a conclusion that follows from behavioral ap-
dividuals’ responses re?ect sensitivity to the sum of the
proaches. From behavioral accounts, the addition of a
values associated with each event as well as to central
negatively valenced stimulus to an already negatively va-
seeking tendencies — the average value of events. (See
lenced context should reduce approach and preference
the Appendix for a quantitative expression of this model).
tendencies (e.g., Hull, 1943; Young, 1936). The aver-
This model was initially developed to provide a theoreti-
aging/summation (A/S) model (e.g., Seta, Crisson, Seta
cal account of how individuals respond to the simultane-
& Wang, 1989; Seta, Seta & Wang, 1991; Seta & Seta,
ous presence of one or multiple audience members having
1992) demonstrated that individuals’ feelings and pref-
heterogeneous or homogeneous status levels (e.g., Seta et
erences are sometimes, but not always, the sum of the
al., 1989). It was extended to homogeneous and hetero-
affective values associated with each event; for example,
geneous life events more generally (e.g., Seta et al., 1991;
individuals may not always feel more negative after ex-
Seta, Seta, & McElroy, 2002).
periencing a compound event containing both a highly
negative plus mildly negative event than they feel after
experiencing a singular highly negative one.
1.1 The Averaging/Summation (A/S) model
The model incorporates ?ndings in the judgment and
Averaging is a process that establishes a central tendency
social-in?uence literatures (e.g., Anderson, 1974, 1981;
among stimuli. Establishing a central tendency among
Latane, 1981; Lichtenstein, Earle & Slovic, 1975). Aver-
stimuli provides individuals with perhaps the best single
aging, for example, was found in the judgment literature
index for predicting future events and a way of under-
when individuals were given traits or cues that had dis-
standing a diverse set of events. If individuals are sensi-
crepant values (see, however, our discussion of set size
tive to the average level of negativity in their lives, adding
effects in section 1.3), whereas summation of each indi-
a negative event of relatively low valence to a context in
vidual’s impact was seen in the social in?uence and au-
which a highly valenced negative event is present would
dience literatures (e.g., Latane, 1981), where participants
produce a decrease in resultant negative affect. Summa-
typically performed a task in front of spectators having
tive information is also functional and makes individuals
similar impact ratings such as similar status levels.
sensitive to the total quantity of resources that has been
?Email: jjseta@uncg.edu
depleted or accumulated and can, for example, provide
425

---

Judgment and Decision Making, Vol. 3, No. 5, June 2008
Averaging and summation
426
information about the demand or availability of resources
man, 1981) used prospect theory to account for responses
needed to successfully meet a goal. If individuals are sen-
to multiple events. Thaler’s (1985) mental accounting
sitive to the summative level of negativity in their lives,
analysis extended this analysis of multiple event judg-
adding a mildly negative event to a context in which a
ments. Thaler reasoned that the value attached to mul-
highly valenced negative event is present would result in
tiple events depends upon whether the events are orga-
an increase in negative affect. The A/S model assumes
nized as belonging to the same mental account or to sep-
that individuals are sensitive to both the summative and
arate accounts. If, for example, they belong to one ac-
average impact of life events. This assumption is con-
count, the events are mathematically combined before be-
sistent with suggestions that processes have evolved that
ing evaluated subjectively by using the value function of
result in the extraction of these types of statistical infor-
prospect theory. For example, Thaler and Johnson (1990)
mation (e.g., Chong & Triesman, 2003, 2005).
showed that, even when events have the same mathemat-
Several studies have supported predictions of the A/S
ical value, individuals can prefer two separate positively
model in audience settings using verbal, behavioral and
valued events (e.g., $50 and $50) to one (e.g., $100) (see
physiological measures (e.g., Seta et al., 1989; Seta &
also Linville & Fischer, 1991).
Seta, 1992), as well as in situations containing negative
Like mental accounting, the A/S model incorporates a
or positive life events (e.g., Seta et al., 1989; Seta, et al.,
negatively accelerating value function (Appendix). Indi-
2002; Seta, Haire & Seta, 2008). One prediction derived
viduals can assign more value to separate events of $50
from the A/S model is that, when two stimuli are rela-
& $50 than to a single event of $100. Because $100 is
tively discrepant in stimulus value (e.g., one highly neg-
further along the negatively accelerating part of the curve
ative and one mildly negative), the impact produced by
than is either of the $50 events. Contrary to the predic-
these events in combination can be less than the in?uence
tions derived from theorizing about the mental account-
of the highly negative event in isolation. One reason for
ing concept, however, the A/S model predicts that two
this is that the addition of a mildly negative event signif-
separate events may or may not be preferred to one sin-
icantly reduces the average value that characterizes the
gle event. These effects, for example, depend upon the
context while only marginally increasing the summative
discrepancy in the values of the events. The importance
total of negative consequences. When the same mildly
of discrepancy in predicting the impact of life events is
negative event, however, is added to a context containing
discussed in more detail in the General Discussion.
similar valenced events, the mildly negative stressor in-
creases the total amount of aggregated consequences but
has only a marginal in?uence on the average value char-
1.3 Overview of Present Research
acterizing the situation (e.g., Seta & Seta, 1992; Seta et
Research derived from the A/S model has examined in-
al., 1989, 1991). A possible reason why events discrepant
dividuals’ affective reactions when confronted with neg-
in affective intensity are more likely to produce an aver-
ative life experiences. It has not examined individuals’
aging effect may be that these events increase the weight
preferences for different negative life experiences. Would
and use of averaging (versus summation) whereas similar
individuals, for example, choose two negative events over
events are more likely to increase the weight of summa-
just one? According to the A/S model, when confronted
tion. Thus, individuals may average, sum or both. When
with a choice between option A (a highly negative event)
they are confronted, for example, with life events with
and option B (a highly negative plus mildly negative
similar affective levels, the central tendency is apparent
event) individuals are expected to choose Option B —
and thus attention does not need to be directed at estab-
the two negative event option. However, when all of
lishing a central tendency, allowing attention to be fo-
the events have a similar affective intensity level, such
cused on summation — each event builds upon the other.
as when option A has one mildly negative event and op-
When individuals are confronted with discrepant events
tion B has two mildly negative ones, individuals are ex-
the central tendency is not apparent and thus may draw a
pected to choose option A — the one negative event op-
disproportionately large amount of attention. Thus, indi-
tion. In Experiment 1, individuals were confronted with
viduals may average, sum, or engage in both types of life
an option containing a single highly negative event and an
event integration.
option containing a highly negative plus mildly negative
one. In addition to this condition, Experiment 2 included
1.2 Mental accounting
a second between-subjects condition — one in which in-
dividuals were exposed to a single mildly negative event
Prospect theory (e.g. Kahneman & Tversky, 1979, 1982)
and one in which they were exposed to two mildly nega-
was originally formulated to model responses to a sin-
tive ones. Support for averaging would be obtained if the
gle unitary outcome. Nevertheless, Kahneman and Tver-
one highly negative plus mildly negative event option was
sky (Kahneman & Tversky, 1984; Tversky & Kahne-
preferred over the option containing just a highly nega-

---

Judgment and Decision Making, Vol. 3, No. 5, June 2008
Averaging and summation
427
tive event; and support for summation would be obtained
by a highly negative plus mildly negative one. Further, it
if the one mildly negative option was preferred over the
should not produce a reaction that is equal or more neg-
option containing two mildly negative events.
ative than the one produced by a highly negative event.
One might argue that, if one negative event is chosen
To explore this issue, Experiment 3 included a six mildly
over two negative events, summation is supported at a
negative condition and a highly negative one. It also in-
descriptive level but not necessarily at a theoretical level.
cluded a highly negative plus mildly negative condition
Descriptively, summation occurs when the addition of a
and a single mildly negative one.
negative event increases negative affect or when the ad-
dition of a positive event increases positive affect. This
pattern of results, however, may not always represent the
2 Experiment 1
operation of a process involving summation. Rather, it
could represent a set size effect, which refers to a sit-
Experiment 1 presented participants with choices con-
uation in which numerically different sets of identically
taining a single highly negative event or containing a
valenced stimuli are perceived differently. The averaging
highly negative event plus a mildly negative event. When
model was extended to include this type of an effect and
perceivers seek a central tendency among events they
was explored within the impression formation literature
should choose a highly negative plus mildly negative
(e.g., Anderson, 1974, 1981; Levin & Kaplan, 1974). For
event over just a singular highly negative one, regardless
example, a set size effect exists when a target person de-
of the order of the events’ occurrences. We predicted av-
scribed as having four negative traits, each with a scale
eraging, rather than summation, because the events were
value of ?4, is not given the same evaluation as when the
associated with different levels of negativity. Further-
same person has only one or two traits with a scale value
more, the impact of each event on the accumulation of
of ?4. Without including perceivers’ initial impression
resources should not override averaging, as it would in a
of the person, an averaging model would predict that per-
situation when the implications of one event building on
ceivers would evaluate the two targets in the same way
the other is salient; in this situation, summation effects,
— both would be evaluated as a ?4. However, when per-
not averaging, would be expected.
ceivers’ initial impression of the person (based on dimen-
In a choice between events, averaging, for example,
sions such as physical appearance or race) is considered,
would not be likely when individuals who desire to max-
an averaging model would predict that the targets can be
imize make a choice between options containing out-
evaluated differently (see Anderson, 1974, 1981 for a fur-
comes having the same metric or function (e.g., option
ther discussion of set size effects). For example, if per-
A providing a single outcome of ?$100, and option B
ceivers’ initial impression of the target is a +2, then the
providing a ?$90 and a ?$20 loss). If there are no
average of the initial impression (+2) and one negative
other differences between the options, then individuals’
trait (?4) would be ?1 ([+2 ?4]/2) whereas the aver-
desire to maximize rewards and minimize losses would
age of the initial impression (+2) and four negative traits
lead them to choose the option containing ?$100 over
(?4) is ?2.8 ([+2, ?4, ?4, ?4, ?4]/5).
the one containing ?$90 and ?$20. In this example,
The same logic can be applied to situations involving
the difference in the total amount of loss associated with
negative life events. It may be the case that averaging,
each option is especially salient. Thus, it would be very
rather than summation, is the reason why one negative
easy for decision-makers to compare across options and
event can be preferred to two negative ones. This would
to see how option A has a lesser negative in?uence on
occur if perceivers consider their initial affective states
their wealth resources than option B. (See Hsee, 1996,
and if theses initial states were less negative than each of
for a discussion of how the characteristics of an item and
the two mildly negative ones, such as when perceivers’
the method of evaluation can alter the weight that deci-
initial affective state is ?1 and each of the mildly neg-
sion makers place on an item.) It is important to note that
ative events is a?4. In this example, one mildly nega-
we are not implying that averaging does not apply to ?-
tive event (e.g., ?4) would have a negative value of ?2.5
nancial outcomes. We simply are describing a situation
([?4, ?1]/2), ten mildly negative events a reaction of
in which it is easy for decision-makers to see how events
?3.7([?4 × 10, ?1]/11 and one hundred mildly nega-
would build upon one another — how ?$90 and ?$20
tive events a reaction of ?3.97 ([?4 × 100, ?1]/101). If
equal a loss of ?$110 and thus, a situation in which sum-
this is the case then it follows from an averaging analy-
mation is likely.
sis that the reaction to many mildly negative events will
Many everyday decisions involve outcomes that can-
approach but not exceed the value of a single mildly neg-
not be easily summed to determine movement toward (or
ative one (?4). Thus, when added to a highly negative
away) from a quantitative goal. Experiment 1 was de-
one, several mildly negative events should not produce a
signed to capture the characteristics of these situations.
reaction that is more negative than the reaction produced
Thus, we predicted that individuals would choose the

---

Judgment and Decision Making, Vol. 3, No. 5, June 2008
Averaging and summation
428
highly negative plus mildly negative option over just a
when “experiencing major car troubles” served as the
highly negative one.
highly negative event in the highly negative condition,
“having an argument with a signi?cant other” served as
the highly negative event in the mixed condition. The
2.1 Method
mildly negative event was “getting some food on your
Twenty female students from Introductory Psychology
shirt during your last class.”
classes participated in this experiment as an option that
partially ful?lled a course research requirement. In this
2.2 Results and discussion
and the other studies reported in this paper, experimental
sessions were conducted in small groups. We utilized a
Participants tended to choose the option containing one
within-subjects design that included two negative stimu-
highly negative outcome plus one mildly negative over
lus conditions: a highly negative event condition; and a
the option containing only a highly negative outcome: 15
mixed condition that contained two events, a highly neg-
vs. 5, respectively (?2
= 5.0, p < .05). The order of
N =20
ative and a mildly negative one. We measured partici-
the high and mild events was initially tested for and not
pants’ affective response each condition, and their choice
obtained, p = .39.
between the two.
A repeated-measures analysis was performed on the af-
Upon entering the experimental room, participants
fect scores of the two within-subjects conditions. When
were told that the experimenter was interested in their re-
participants were exposed to two discrepant negative
actions to different events. They then were given packets
events (M = 47.05) they reported less negative affect
that contained the experimental manipulations. We used
than when they were exposed to the one highly negative
a manipulation commonly employed to induce affective
event (M = 59.2, F1,19 = 14.04, p < .01). Again,
states (e.g., Schwarz & Clore, 1983), in which partici-
there were no signi?cant order effects, p’s>.35. The lack
pants were asked to think about, and write brie?y about
of order effects in this experiment demonstrates that av-
several negative events. This manipulation should in-
eraging effects can be obtained over and above sequence
crease the likelihood that participants are in contact with
effects, such as those predicted by the peak-end rule (e.g.,
the event’s implications. Thus, this type of manipulation
Kahneman, Fredrickson, Schreiber & Redelmeier, 1993).
is assumed to involve realistic affective consequences.
We will discuss this concept in more depth in the general
After considering the events, participants indicated the
discussion.
negativity of experiencing both a highly negative event
It could be argued that these data support summa-
alone and the negativity of experiencing both a highly
tion, rather than averaging if it is assumed that the per-
negative plus mildly negative one. Participants indicated
ceived value of the highly negative event was assimi-
how negative they felt after thinking about the events
lated (moved toward) the mildly negative one whereas the
happening to them on a 101-point scale where “0” in-
mildly negative one remained relatively constant. As long
dicated “low negative” and “100” indicated “super ex-
as the perceived value of the highly negative event was
tremely negative.” In addition, they also were asked
lowered suf?ciently (e.g., from ?8 to ?4) and the mildly
to choose between the two options (the highly negative
negative event remained relatively stable ( e.g., from ?3
event versus the highly negative plus mildly positive one)
to ?3), averaging would be explicable from an additive
by placing a circle around the letter (either A or B) that
(summative) account; the sum of the highly negative (?4)
appeared in front of each option. Order of options and
plus mildly negative (-3) would be less than that of the
event presentations were counterbalanced. The events
highly negative event (?8) in isolation.
used in this experiment, and in Experiments 2 and 3, were
To address this possibility, we employed a separate
events taken from a prior norming study. Participants ei-
group of participants who were given the same instruc-
ther read about the highly negative event ?rst or last; and,
tions as those provided to our participants in the highly
when asked to decide between options, the highly nega-
negative and in the mixed condition (highly negative plus
tive event option either preceded or followed the highly
mildly negative). However, instead of asking about par-
negative and mildly negative one.
ticipants’ overall negative affective reaction, we asked
Two different highly negative events were used in this
them to judge the negativity of each event separately in
study. They were as follows: “having an argument with
a between-subjects design. In the highly negative plus
a signi?cant other” and “experiencing major car trou-
mildly negative condition, participants were asked to re-
bles.” When “having an argument with a signi?cant
port on the negativity of the highly negative event and
other” served as the event for the highly negative out-
then on the negativity of the mildly negative one (order
come condition, “experiencing major car troubles” was
counterbalanced); in the highly negative condition, they
the highly negative event in the mixed condition — the
were asked to judge on the negativity of just the highly
highly negative plus mildly negative one. Conversely,
negative event. To obtain a summed score we added the

---

Judgment and Decision Making, Vol. 3, No. 5, June 2008
Averaging and summation
429
scores of the highly negative and mildly negative event
tive ratings (present or absent).1 In the discrepant va-
in the highly negative plus mildly negative condition and
lence condition, participants were given choices between
compared these scores to the scores obtained in the highly
a highly negative event and a highly negative plus mildly
negative condition. The combined negative implications
negative one. In the equal valence or similar condi-
of each of the two events (highly plus mildly negative)
tion, they were given choices between options containing
was signi?cantly higher (M = 77.68) than the negative
one mildly negative event and an option containing two
implications of just the highly negative event in isolation
mildly negative ones.
(M = 52.63, F1,33 = 6.98, p = .01). This effect is
Similar procedures as used in Experiment 1 were used
inconsistent with a summation account. From this view,
in this study, including counterbalancing and order con-
the combined implications of two events (highly negative
trols. We also used the same highly negative events. In
plus mildly negative) should have been lower, not higher,
this study, however, three mildly negative events were
than that of those associated with the singular highly neg-
used for the mildly negative options: “getting a bit of
ative event.
food on your shirt during your last class,” “?nding that
the drink machine is out of order,” and “disliking the mu-
sic on the radio.” “Getting a bit of food on your shirt dur-
3 Experiment 2
ing your last class” served as the mildly negative outcome
in the highly negative plus mildly negative event condi-
tion and one of the two events in the two mildly negative
Additive models predict that “more is worse” when it
event condition. The other two mildly negative events
comes to negative events. However, from an A/S perspec-
were counterbalanced across the one and two mildly neg-
tive this is not always the case; adding a negative event to
ative outcome conditions. Choice and affective responses
an already negative context can function to reduce nega-
were collected in the manner described in the previous
tive affect (averaging). One critical factor that determines
study.
the likelihood of obtaining averaging or summation ef-
fects is the discrepancy between the stimulus events. Ex-
periment 2 was designed to determine whether more is
3.2 Results and discussion
“worse” when participants are exposed to events having
3.2.1 Choice analyses
similar affective intensity levels — a summation effect
— and whether adding negative events reduces negative
A chi-square analysis of the choice data revealed a sig-
affect — an averaging effect. This experiment also was
ni?cant interaction between Negative Event Discrepancy
designed to retest the A/S position that averaging effects
and the Number of Negative Events conditions. Partici-
can be obtained when a highly negative event precedes or
pants chose the highly negative event plus mildly negative
follows a mildly negative one.
event option over the highly negative one: 28 (76%) vs.
In Experiment 1, the affective rating scale always pre-
9 (24%), respectively (?2
= 9.76, p < .01). But
N =37
ceded participants’ choices. It might be argued that par-
they chose the two mildly negative option over the option
ticipants’ choice ratings were a consequence of their de-
containing a single mildly negative event: 29 (91%) vs.
sire to be consistent with their affective ratings. Thus,
3 (9%) (?2
= 21.13, p < .001). These two results
N =32
when they demonstrated averaging in their affective rat-
produced an interaction between negative event discrep-
ings, they demonstrated a similar and consistent effect
ancy and the number of negative events (?2
= 30.49,
N =69
in their choice ratings. Would choice ratings be similar
p < .001). There were no effects of the order of events
when participants are not given affective ratings? To ad-
on participants’ responses, all p’s>.26.2
dress this question, Experiment 2 included a condition in
1The condition in which the affective rating scale was absent was
which participants were given choice ratings without af-
run in the same semester but at a somewhat later date than the affective
fective ratings.
rating present condition. We included this as a part of Experiment 2 be-
cause it was conducted by the same experimenter in the same academic
institution and with the same population of participants. It included a
smaller number of participants than the affective rating present condi-
3.1 Method
tion because of the availability of participants.
2The Chi-Square analysis did not reveal an interaction involving the
presence or absence of the affective rating scale. Nevertheless, we con-
Sixty-nine students participated in this experiment. The
ducted two additional Chi-Square analyses to be sure that similar results
design included 2 levels of the within-subject variable,
were found regardless of whether affect ratings were or were not made.
manipulating the number of negative events presented in
One analysis was conducted on participants’ choice ratings in the affec-
the options (one or two negative outcomes), 2 levels of
tive present score condition and another on participants’ choices in the
absent condition. When the affective rating scale was present, the anal-
a between-subjects manipulation of negative event dis-
ysis revealed a signi?cant interaction between Negative Event Discrep-
crepancy (discrepant and equal) and 2 levels of affec-
ancy and the Number of Negative Events condition (?2
= 16.29,
N =42

---

Judgment and Decision Making, Vol. 3, No. 5, June 2008
Averaging and summation
430
3.2.2 Affect rating analyses
signi?cantly more negatively about six versus one mildly
negative event. It also allowed us to determine if the
Participants rated the situations containing highly nega-
descriptive summation effects obtained in Experiment 2
tive events higher (M = 41.64) than situations contain-
were due to a theoretical summation process. Theoreti-
ing the lower valenced events (M = 16.2), F1,40 =
cal summation would be shown if several mildly negative
20.93, p < .001. More interestingly, and as predicted,
events produced a more negative reaction than one mildly
this main effect was quali?ed by a signi?cant Nega-
event, and if they also produced a negative reaction that
tive Event Discrepancy X Number of Event interaction,
was equal to or more negative than that produced by a
F1,40 = 4.24, p < .05. Although neither of the two
highly negative event.
contrasts were signi?cant, the interaction was due to par-
ticipants reporting a higher level of negative affect in the
one high (M = 46.64) versus one high and one mildly
4.1 Method
negative event condition (M = 36.64), F1,40 = 3.75,
Participants were 64 students from Introductory Psychol-
p < .10 whereas an opposite pattern was obtained when
ogy classes who participated in this study for partial ful-
1 (M = 13.5) and 2 (M = 18.9) mildly negative event
?llment of course credit. We used a between-subjects de-
conditions were compared, F < 1. We found no order
sign in which participants were assigned randomly to one
effects in this analysis, p’s > .34.
of four negative events conditions: they considered either
a highly negative event, a mildly negative event, a mildly
3.2.3 Results summary
negative plus highly negative event or six mildly negative
events.
Participants chose a single highly negative plus mildly
Similar to previous experiments, participants were
negative event over a single highly negative one but chose
asked to think and write about events and were asked
the option containing a single mildly negative event over
to indicate how bad they felt after thinking about the
the one containing two such events. These results were
events happening to them, on the 101-point scale de-
obtained when participants’ affective ratings of the op-
scribed earlier. The highly negative event used in this
tions did or did not precede their choice ratings. In addi-
experiment was “being put on academic probation.” Six
tion, participants’ perceptions of the options’ negativity
relatively mild negative events were used. They were:
were consistent with their preferences. In addition, these
“having some dif?culties with friends, “having some car
averaging and summation effects were independent of the
troubles,” “waking up and ?nding the shower won’t rise
way in which the events were ordered.
above lukewarm,” “owing someone some money,” and
“having too many responsibilities,” “misplacing or los-
4 Experiment 3
ing something.” All of these events were used in the
six mildly negative condition (order counterbalanced). In
conditions containing one mildly negative event, each of
In Experiment 2, participants chose the option containing
these events was assigned as the mildly negative event,
a single mildly negative condition over the option con-
counterbalanced across conditions. The presentation of
taining two mildly negative ones. The negative affect
events having different affective intensity levels was also
associated with two mildly negative events was greater
counterbalanced in this condition.
— but not signi?cantly so — than that associated with
a single mildly negative one. In Experiment 3, we in-
cluded a six event mildly negative condition. This con-
4.2 Results and discussion
dition allowed us to determine whether participants felt
The four conditions did not produce equivalent ratings
p < .001). Participants chose the highly negative plus mildly nega-
(F3,60 = 7.69, p < .001). The means are contained in
tive option over the highly negative one, (?2
= 6.55, p < .01),
N =22
Table 1. Planned contrasts showed that the one highly
whereas they chose the two mildly negative option over the option con-
negative condition produced more negative affect than
taining a single mildly negative event (?2
= 9.8, p < .01). No
N =20
other signi?cant effects were obtained.
the mixed highly negative plus mildly negative condition
Consistent with the previous analysis, an analysis of the affect rating
(F1,60 = 10.04, p < .01) — the expected averaging
absent data also revealed a signi?cant Negative Event Discrepancy X
effect. The lack of order effects (p > .8) in the mixed
Number of Negative Events Interaction, (?2
= 14.85, p < .001).
N =27
Participants chose the highly negative plus mildly negative option over
condition supports the A/S position that averaging can be
the highly negative one (?2
= 3.26, p < .07) whereas they chose
obtained over and above sequence effects.
N =15
the two mildly negative option over the option containing one mildly
In addition, this experiment demonstrated summation
negative event (?2
= 12.0, p < .001). No other signi?cant ef-
N =12
in that the six mildly negative condition produced more
fects were obtained. Thus, averaging and summation effects were ob-
tained when affective ratings did or did not precede participants’ choice
negative affect than the one mildly negative condition
ratings.
(F1,60 = 11.3, p < .01), and it produced a similar nega-

---

Judgment and Decision Making, Vol. 3, No. 5, June 2008
Averaging and summation
431
hand in cold water for 60s whereas in the long sequence
Table 1: Mean negative affect ratings by condition.
they were exposed to the same cold water for the ?rst 60s,
Number of Negative Events
Negativity Rating
but also were exposed to a less aversive water tempera-
ture for the last 30s. The majority of participants chose to
1 highly negative event
63.25
experience the long sequence even though this sequence
1 mildly negative event
25.33
represented 30s of additional aversive stimulation. Thus,
1 highly and 1 mildly negative event
35.76
participants preferred the episode containing longer dura-
tions of aversive stimulation. Other studies have extended
6 mildly negative events
55.44
this “duration neglect” ?nding to other forms of aversive
stimulation, such as aversive sounds (e.g., Ariely & Za-
uberman, 2000), unpleasant movie clips (Fredrickson &
tive affective reaction as that produced by the one highly
Kahneman, 2003) and medical procedures and treatments
negative event condition (F < 1).
(e.g., Chapman, 2000; Redelmeier, Katz & Kahneman,
1997; Redelmeier & Kahneman, 1996).
5 General discussion
People also have been shown to neglect the duration
of positive episodes. For example, Fredrickson and Kah-
Three studies were conducted that tested assumptions of
neman (1993) showed that the duration of a movie clip
the A/S model in the context of perceivers’ choices and
had little impact on viewers’ evaluations of an unpleas-
affective reactions to negatively valenced events. Consis-
ant movie when the peak and end intensities were taken
tent with this perspective, “more” negative events were
into consideration. And Diener, Wirtz, and Oishi (2001)
“worse” when participants were confronted with stimuli
found that participants rated another person’s “wonder-
of similar affective intensity levels whereas “more” nega-
ful life” as more desirable when it ended abruptly than
tive events were “better” when they were confronted with
when it lasted longer but the additional years were only
stimuli of discrepant affective intensities.
relatively good ones.
These studies provide information on when predictions
In addition to an episode’s peak and end intensities,
from additive accounts, such as hedonic calculus, will
evaluations and memories for events have been shown
and will not occur. Additive accounts predict that “more
to depend upon an episode’s rate of improvement (e.g.,
negative events are worse” — a summation effect. As
Baumgartner, Sujan & Padgett, 1997; Hsee & Abelson,
seen in the present series of studies, “more” negative
1991; Hsee, et al., 1994; Loewenstein & Sicherman,
events can be “worse” as predicted by additive accounts
1991) and how good and bad aspects of the episode are
but more negative events can also be diminutive — an
distributed over time (Loewenstein & Prelec, 1993). The
averaging effect. The discrepancy between the negative
importance individuals place on the pattern of a sequence,
stimuli was the key variable in this study affecting the
such as duration, peak and end intensities is not static.
likelihood of obtaining averaging and summation effects.
Ariely and Zauberman (2000), for example, found that
When the affective intensity discrepancy between
the pattern of an episode — its peak and end intensities
events was large, an averaging effect was obtained; when
— had less of an impact on participants’ judgments when
the discrepancy was small, summation was obtained.
the hedonic experience was composed of multiple seg-
ments.
5.1 Peak-end rule
5.2 Comparison of models
Kahneman and colleagues (e.g., Kahneman et al., 1993;
Fredrickson & Kahneman, 1993; Redelmeier & Kahne-
The A/S and peak-end rule differ in important ways.
man, 1996) have shown that participants often neglect the
First, and in contrast to the peak-end rule, the A/S model
duration of an episode and concentrate on the episode’s
predicts that, even when the sequence of events is not
peak and end intensities and thus follow a peak-end rule
salient, a mildly negative plus highly negative event can
of judgment. This rule is now a part of a “judgment by
still produce less negative affect than the highly negative
prototype” model (e.g., Ariely, 1998; Ariely & Carmon,
one in isolation. Second, the A/S model takes the discrep-
2000; Hsee, Salovey & Abelson, 1994). Like the A/S
ancy between events into consideration in its predictions
model, the peak-end rule concludes that utility is not al-
concerning when averaging versus summation effects are
ways positively related to the sum of the values associated
likely to occur. Thus the A/S model makes predictions
with events.
about when individuals’ reactions to life events will re-
In one study, Kahneman et al. (1993), asked partici-
?ect summation and averaging effects.
pants to experience sequences of aversive sensations. In
Research on the peak-end rule has dealt with an unfold-
the short duration sequence, participants immersed their
ing continuous experience (e.g., Kahneman et al., 1993;

---

Judgment and Decision Making, Vol. 3, No. 5, June 2008
Averaging and summation
432
Loewenstein & Prelec, 1993). The temporal sequencing
provement goals are salient, a positive (successful) end
of events was a salient feature of these procedures. In
event that follows a few negative (unsuccessful) events
contrast, research on the averaging/summation model has
may be perceived to be more positive than either the same
dealt with contexts, such as audiences, in which sequen-
positive end event that follows a single negative event or
tial information was not available or salient. For example,
the same positive event in isolation (e.g., Festinger, 1957;
when an audience was composed of multiple members,
Seta & Seta, 1982).
performers were exposed simultaneously to all members
The weight assigned to events also may be in?uenced
in this context. Thus, the impact of the arrangement of
by the discrepancy in the affective intensity levels of
stimuli cannot be interpreted via changes in the sequenc-
events. In some situations, a mildly intense event may
ing of the audience members over time.
have little or no impact (see Seta & Seta, 1996). For
In our prior research, and in the present study, the
example, when the discrepancy between events is very
events that were presented were discrete and unique from
large, as when a mildly negative event is only slightly
one another, and were not presented as having unfolded
negative and a second event is highly negative, the mildly
as a meaningful stream. Therefore, the sequencing of
negative event may be given little or no weight in per-
events was not a salient feature in this research testing
ceivers’ reactions. Thus, in this situation, perceivers
the A/S model. Consequently, neither the peak-end nor
would respond in a similar way to a context containing a
improvement rules were driving forces in the results of
highly negative event and a situation containing the same
these studies. Rather, we found that, because of the oper-
highly negative event plus a very mild negative one.
ation of central seeking tendencies, a mildly negative plus
highly negative event produced less negative affect than
just a highly negative one, regardless of the sequencing
5.3 Conclusion
of the stimuli.3 Thus, the results of these studies provide
The results of these experiments demonstrated the oper-
evidence that averaging effects can be obtained over and
ation of averaging and summation in situations in which
above sequence effects.
individuals are confronted with negative life events. In
It is important to note that sequencing effects like
doing so, they delineated situations when “more” neg-
those predicted by the peak-end rule can be accounted
ative events can be “better” and when “more” can be
for within the A/S model. For example, in situations
“worse.”
More negative events produced decrements
in which the peak and end is especially salient, partic-
in negative affective reactions and increments in prefer-
ipants’ judgment would be heavily in?uenced by these
ence level when individuals were confronted with events
two factors — these factors would have an especially high
having discrepant affective intensity levels; more, how-
weight. (See the Appendix for a mathematical presenta-
ever, produced increments in negative affective levels and
tion of weights). If participants average the peak and end
decrements in preference levels when individuals were
values and the end is less negative than the peak, then the
confronted with events having similar affective intensi-
average of the peak and end would be less negative than
ties.
a sequence without this end event. If the end is more neg-
ative, however, the average of the peak and end would be
more, not less, negative than a sequence without this end
References
event. Of course, from the A/S model, when the peak and
end are critical factors, the discrepancy between the peak
Anderson, N. H. (1981). Foundations of information in-
and end’s intensity levels would in?uence the likelihood
tegration theory. New York: Academic Press.
and strength of averaging.
Anderson, N. H. (1974). Cross-task validation of func-
The sequencing of events also may change the value of
tional measurement using judgments of total magni-
the end event. When multiple events comprise a mean-
tude. Journal of Experimental Psychology, 102, 226–
ingful stream, events that precede the end event can easily
233.
alter the value of the end event. For example, when im-
Ariely, D. (1998). Combining experiences over time: The
3It is not entirely clear from the peak-end rule whether a mildly neg-
effects of duration, intensity changes and on-line mea-
ative event plus a highly negative one will be preferred to a singular
surements on retrospective pain evaluations. Journal
highly negative one when the highly negative event is presented last.
of Behavioral Decision Making 11, 19–45.
For this effect to occur, the mildly negative plus highly negative event
would need to have less value when it appears in combination versus
Ariely, D., & Carmon, Z. (2000). Gestalt characteristics
when it appears by itself. This might be the case because, when a highly
of experiences: The de?ning features of summarized
negative event is preceded by a mildly negative one, it is the second ex-
events. Journal of Behavioral Decision Making, 13,
perienced negative event and thus it may have less value than when it
191–201.
the only experienced negative event. However, because the highly neg-
ative event is counter to improvement goals it may have more, not less,
Ariely, D., & Zauberman, G. (2000). On the making of
negative value in combination than in isolation.
an experience: The effects of breaking and combining

---

Judgment and Decision Making, Vol. 3, No. 5, June 2008
Averaging and summation
433
experiences on their overall evaluation. Journal of Be-
Levin, I. P., & Kaplan, M. F. (1974). The set-size effect
havioral Decision Making, 13, 219–232.
in personality impression formation is not an artifact.
Baumgartner, H., Sujan, M., & Padgett, D. (1997). Pat-
Bulletin of the Psychonomic Society, 3, 187–188.
terns of affective reactions to advertisements: The in-
Lichtenstein, S., Earle, T. C., & Slovic, P. (1975). Cue
tegration of moment-to-moment responses into overall
utilization in a numerical prediction task. Journal
judgments. Journal of Marketing Research, 34, 219–
of Experimental Psychology: Human Perception and
232.
Performance, 1, 77–85.
Chapman, G. B. (2000). Preferences for improving and
Linville, P., & Fischer, G. (1991). Preferences for sepa-
declining sequences of health outcomes. Journal of
rating or combining events. Journal of Personality and
Behavioral Decision Making, 13, Special Issue: Time
Social Psychology, 60, 5–23.
and Decision, 203–218.
Loewenstein, G., & Prelec, D.(1993). Anomalies in in-
Chong, S. C., & Treisman, A. (2003). Representation of
terpersonal choice : Evidence and an interpretation. In
statistical properties. Visual Research, 43, 393–404.
J. Elster & Loewenstein, G. (Eds.) Choice over time.
Chong, S. C., & Treisman, A. (2005). Statistical process-
New York: NY: Russell Sage Foundation.
ing: Computing the average size in perceptual groups.
Loewenstein, G.,& Sicherman, N. (1991). Do workers
Vision Research 45, 891–900.
prefer increasing wage pro?les?
Journal of Labor
Diener, E., Wirtz, D., & Oishi, S. (2001). End effects of
Economics, 9, 67–84.
rated quality of life: The James Dean effect. Psycho-
Redelmeier, D. A., & Kahneman, D. (1996). Patients’
logical Science, 12, 124–128.
memories of painful medical treatments: Real-time
Festinger, L. (1957). A theory of cognitive dissonance.
and retrospective evaluations of two minimally inva-
Stanford, CA: Stanford University Press.
sive procedures. Pain, 66, 3–8.
Fredrickson, B. L., & Kahneman, D. (1993). Dura-
Redelmeier, D. A., Katz, J., & Kahneman, D. (2003).
tion neglect in retrospective evaluations of affective
Memories of colonoscopy: A randomized trial. Pain,
episodes. Journal of Personality and Social Psychol-
104, 187–194.
ogy, 65, 45–55.
Schwarz, N., & Clore, G. L. (1983). Mood, misattribu-
Hsee, C. K. (1996). The evaluability hypothesis: An ex-
tion, and judgments of well-being: Informative and di-
planation for preference reversals between joint and
rective functions of affective states. Journal of Person-
separate evaluations of alternatives. Organizational
ality and Social Psychology, 45, 513–523.
Behavior and
Human Decision Processes 67, 247–
257.
Seta, C. E., & Seta, J. J. (1992). Increments and decre-
Hsee, C. K., Salovey, P., & Abelson, R. P. (1994). The
ments in mean arterial pressure as a function of audi-
quasi-acceleration relation: Satisfaction as a function
ence composition: An averaging and summation anal-
of the change of velocity of outcome over time. Jour-
ysis. Personality and Social Psychology Bulletin, 18,
nal of Experimental Social Psychology, 30, 96–111.
173–181.
Hsee, C. K., & Abelson, R. P. (1991). Velocity relation:
Seta, C. E., & Seta, J. J. (1996). When more is less: An
Satisfaction as a function of the ?rst derivative over
averaging/summation analysis of social anxiety. Jour-
time. Journal of Personality and Social Psychology,
nal of Research in Personality. 30, 496–509.
60, 341–347.
Seta, J. J., Crisson, J. E., Seta, C. E., & Wang, M. A.
Hull, C. L. (1943). Principles of Behavior, an Introduc-
(1989). Task performance and perceptions of anxi-
tion to Behavior Theory. Oxford, England: Appleton-
ety: Averaging and summation in an evaluative set-
Century.
ting. Journal of Personality and Social Psychology,
Kahneman, D., Fredrickson, B., Schreiber, C., & Re-
56, 387–396.
delmeier, D. (1993). When more pain is preferred to
Seta, J. J., Haire, A.,& Seta, C. E. (2008). Averaging and
less: Adding a better end. Psychological Science, 4,
summation: Positivity and choice as a function of the
401–405.
number and affective intensity of life events. Journal
Kahneman, D., & Tversky, A. (1979). Prospect Theory :
of Experimental Social Psychology, 44, 173–186.
An analysis of decision under risk. Econometica, 47,
Seta, J. J., & Seta, C. E. (1982). Personal equity —
313–327.
An intrapersonal comparator system analysis of reward
Kahneman, D.,& Tversky, A. (1982). The psychology of
value. Journal of Personality and Social Psychology,
preferences. Scienti?c American, 246, 160–173.
43, 22- 235.
Kahneman, D., & Tversky, A. (1984). Choices, values
Seta, J. J., Seta, C. E., & McElroy, T. (2002). Strategies
and frames. American Psychologist, 39, 341–350.
for reducing the stress of negative life experiences: An
Latane, B. (1981). The psychology of social impact.
averaging/summation analysis. Personality and Social
American Psychologist, 36, 343- 356.
Psychology Bulletin, 28, 1574–1585.

---

Judgment and Decision Making, Vol. 3, No. 5, June 2008
Averaging and summation
434
Seta, J. J., Seta, C. E., & Wang, M. A. (1991). Feelings of
information that will be contained in the context; S1, S2
negativity and stress: An averaging-summation analy-
refers to the de?ning features of each stimulus such as its
sis of impressions of negative life experiences. Person-
properties and implications; W refers to the weight or in-
ality and Social Psychology Bulletin, 17, 376–384.
?uence of the stimulus that is determined by factors such
Thaler, R. H. (1985). Mental accounting and consumer
as importance or attention; V refers to the strength or in-
choice. Marketing Science, 4, 199- 214.
?uence of these two integration rules; V1 corresponds to
Thaler, R. H., & Johnson, E. J. (1990). Gambling with the
the strength of averaging; V2 refers to the strength of sum-
house money and trying to break even: The effects of
mation.
prior outcomes on risky choice. Management Science,
Essentially, this formula expresses a method of inte-
36, 643–660.
grating various components of a setting and includes in-
Tversky, A., & Kahneman, D. (1981). The framing of
dividuals’ sensitivities to both the average (or central ten-
decisions and the psychology of choice. Science, 211,
dency) impact of the various components (?rst bracketed
453–458.
section) and the summative (or accumulated), impact of
Young, P. T. (1936). Motivation of Behavior. New York:
every individual component (second bracketed section).
John Wiley & Sons, Inc.
Each stimulus within the context (S) contributes to the
determination of a central tendency (the average) and the
Appendix
summative impact of the components, and these stimuli
combine to produce an overall response to the setting (R).
The averaging/summation (AS) model can be expressed
Each stimulus has a value along some dimension of judg-
more precisely by the equation below.
ment. These values may be along dimensions such as
The ?rst bracketed section contains an averaging rule;
the magnitude of negativity, positivity, status level, con?-
the second bracket contains a summation rule. R refers
dence, or consequences. In addition, the summation pro-
to the individual’s overall response (e.g., stress level); S
cedure is raised to a power (t) with an exponent less than
refers to stimulus value; S0 represents the initial state of
1 to re?ect the commonly found marginally decreasing
the individual as he or she enters the context as well as
utility function for increments in the number of stimuli in
the individual’s initial expectation concerning the type of
the setting.
W0S0 + W1S1 + W2S2 + . . .
V1
+ V2 [W0S0 + W1S1 + W2S2 + . . . ]t <1
W0 + W1 + W2 + . . .
R =
V1 + V2

---