Content uploaded by Craig A Anderson
Author content
All content in this area was uploaded by Craig A Anderson on Nov 16, 2017
Content may be subject to copyright.
10
TEMPERATURE AND
AGGRESSION: PARADOX,
CONTROVERSY, AND A
(FAIRLY) CLEAR PICTURE
CRAIG A. ANDERSON
University of Missouri
KATHRYN B. ANDERSON
Our Lady of the Lake University
I pray thee, good Mercutio, let's retire;
The day is hot, the Capulets abroad,
And, if we meet, we shall not 'scape a brawl,
For now, these hot days, is the mad blood stirring.
Romeo and Juliet, Shakespeare
The day drags by like a wounded animal
The approaching disease, 92°
The blood in our veins and the brains in our head
The approaching unease, 92°
92° by Siouxsie and the Banshees, Tinderbox,
David Geffen Company, 1986
Imagine the following scenario. You are in your overpriced hotel room, re-
viewing your notes for an important presentation that you will be giving later in
the day. The people next door have their television cranked to maximum volume.
It is so loud that you can hear the dialogue as well as the gunfire and explosions;
it is from Sylvester Stallone's movie Judge Dredd. Despite an earlier request to
turn it down, the volume remains at an intolerable level. You pound on their door
Human Aggression: Theories, Research. and
Implications for Social Policy 247 Copyright © 1998 by Academic Press.
All rights of reproduction in any form reserved.
--------
248 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
to explain your situation and make one more appeal. The appeal is met by laugh-
ter and a comment on the legitimacy of your birth. How do you respond? We
would guess that most readers of this chapter would feel angry, but would not emit
any aggressive behavior.
Now, imagine the same scenario, but with one additional stressful factor. The
hotel air conditioning is off, and the rooms and hallways are at least 92°F. Now
what is the likelihood that you would aggress? Most people would agree that the
hot temperature would indeed increase the likelihood of an aggressive response of
some kind. Even if it is merely a verbal insult, the probability of a violent en-
counter occurring also increases, perhaps as the end result of a series of escalating
verbally abusive exchanges.
This chapter is structured to address theoretical, empirical, and practical issues
surrounding the temperature-aggression hypothesis. First, a brief history of the
temperature-aggression hypothesis is presented. Second, a paradox involving vio-
lence and lethargy is described and resolved. Third, the major issues and theories
surrounding heat effects are outlined, and an integrated model of aggression is pro-
vided. Fourth, several epistemological issues concerning empirical tests of various
theories are discussed. Fifth, modem empirical studies are reviewed, and new re-
sults that bear on issues of current concern are presented. Finally, the empirical
data base is more specifically compared to the major theories, noting convergences
and contradictions, and pointing out fruitfullines of inquiry for future research.
A BRIEF HISTORY OF THE
TEMPERATURE-AGGRESSION HYPOTHESIS
Social commentators have noted weather effects on human behavior and have
used heat-related imagery for thousands of years. Cicero (106-43 B.c.) noted that
"The minds of men do in the weather share, dark or serene as the day's foul or
fair." Shakespeare noted (in The Merchant of Venice) that "the brain may devise
laws for the blood, but a hot temper leaps o'er a cold decree." Shakespeare ex-
plicitly referenced what presumedly was a commonly held belief in the society of
his day in Romeo and Juliet, as in the opening quote linking hot temperatures to
violent behavior.
Social philosophers, social geographers, and other students of behavior began
to apply empirical methods to this theory in the late 1800s. Even earlier, Mon-
tesquieu traveled extensively and drew upon his observations in various writings.
Heat effects were included in his observations. In The Spirit of the Laws he stated
that, "You will find in the northern climates peoples who have few vices, enough
virtues, and much sincerity and frankness. As you move toward the countries of
the south, you will believe you have moved away from morality itself: the liveli-
est passions will increase crime. . . ." (Montesquieu, 1748/1989, p. 234). In this
work, Montesquieu even espoused a physiological theory for why hot tempera-
tures produced such extreme emotionality in the peoples who lived in hot south-
ern climates.
TEMPERATURE AND AGGRESSION 249
It was some time before this type of informal observation of and speculation
about heat effects was supplemented with more objective empirical methods. The
earliest such study located was by Leffingwell (1892), who examined quarter of
the year effects on two broad categories of violent crime in England and Wales in
1878-1887. Other early studies of the heat effect include those by Lombroso in
Italy (and elsewhere) (1899/1911), Guerry in France (as cited in Brearley, 1932),
Dexter (1899) in the United States, and Aschaffenburg (1903/1913) in Germany
and France. Although the empirical methods were somewhat crude by modern
standards, these early studies supported the prevailing theory that uncomfortably
hot temperatures produce increases in violent behavior (for a review of this work,
see Anderson, 1989).
THE TEMPERATURE-AGGRESSION PARADOX
Some of the early writings concerning heat effects contained a paradox that to
this day remains unresolved. Specifically, hot temperatures are seen as having two
effects that seem opposite and contradictory. On the one hand, hot temperatures
are seen as increasing aggression. On the other hand, hot temperatures are also
seen as robbing people of motivation, alertness, and energy. Aggression usually
requires considerable energy or effort, especially the aggression examined in stud-
ies focusing on violent crimes such as assault and murder. Thus, the paradox: How
can lethargic people behave violently? If heat makes people reluctant to engage in
energetic activities, how can it produce increases in aggression?
Before attempting to resolve this paradox, it is important to ask whether it truly
exists. There are two parts to consider, the "increases aggressive behavior" part
and the "decreases effort/energy" part. Much of this chapter is devoted to discov-
ering whether hot temperatures directly increase aggressive behavior. As most
readers will have guessed by now, the evidence of such direct effects is over-
whelmingly positive. Assuming that we are right (and that the later sections of this
chapter are convincing to you, the reader), then we are left with the second part,
whether hot temperatures also make people lethargic. There has been surprisingly
little research on this question, perhaps because it is somewhat vague.
There are several ways of thinking about lethargy, including at a simple affect
level (e.g., How do you feel?), at a physiological level (e.g., Is heart rate de-
creased?), and at a task level (e.g., Is cognitive performance lowered?). In addi-
tion, we can ask whether modern citizens of western society (e.g., U.S. college
students) believe there is a true heat effect. We begin by examining this "Do you
believe?" question.
STUDY 1: SOCIAL THEORIES ABOUT HEAT EFFECTS
The goal of Study 1 was to assess the social theories of our subject population
concerning the relation of temperature to the several variables of interest in this
domain: affect, arousal, and aggression. A questionnaire was developed to measure
250 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
subjects' social theories concerning the relation of both hot and cold temperatures
to these variables.
Method
Procedures
Fifteen female and 7 male undergraduates at a large midwestern university par-
ticipated in this experiment. Participants were given a two-page questionnaire,
which was titled: "Beliefs about temperature, emotions, and behavior." They were
instructed to "indicate your beliefs by circling a number for each item below."
After completing the questionnaire, participants were thoroughly debriefed and
thanked for their assistance.
Participants were asked six questions using the frame, "Compared to normal
temperatures, what do you think the effect of (hot/cold) temperatures would be on
(alertness and energy level, feelings of hostility and anger, aggression and violent
behaviors)." Each question was answered on a five-point rating scale, with the
lowest rating indicating a belief that the temperature (hot or cold) would decrease
the target variable (i.e., alertness, hostility, or aggression), the midpoint indicating
a belief in no heat effect, and the highest rating indicating a belief that the tem-
perature would increase the target variable. The three questions concerning effects
of hot temperatures and the corresponding three cold temperature questions were
presented on separate pages. Participants were randomly assigned to completing
the hot or the cold page first. This order manipulation allowed examination of the
possibility that thinking about one type of heat effect (e.g., hot) would influence
subjects' responses on the other (e.g., cold).
Results and Discussion
The six items and results are presented in Figure 10.1, in terms of average de-
viations from the scale midpoint of 3, which corresponded to a belief in no effect
of temperature. There were no reliable effects of task order or of sex (p >5)5), so
subsequent tests ignored these factors. A ttest was performed on each item mean
to see whether it differed reliably from "no effect." As can be seen, hot tempera-
tures were expected to have a very large impact on all three target variables. Com-
pared to comfortable temperatures, participants believed that hot temperatures
would produce a significant decrease in alertness and energy level [M =1.50,
t(21) =777777 p<0101], a significant increase in anger and hostility [M =4.64,
t(21) =13.24, p< 0101], and a significant increase in aggressive and violent be-
havior [M =4.45, t(21) =10.14, p< 0101].
Participants also expected that cold temperatures (relative to comfortable ones)
would have systematic effects on the target variables, but in every case the direc-
tion of the expected cold effect was opposite of the expected hot effect. Partici-
pants expected cold temperatures to produce a significant increase in alertness and
energy level [M =3.64, t(21) =2.32, P<5]5], a significant decrease in anger and
hostility [M = 2.45, t(21) =3,3,3, p<1],1], and a significant decrease in aggres-
sive and violent behavior [M =2.14, t(21) =4,4,4, p< 0101].
251
TEMPERATURE AND AGGRESSION
2
'U
~1
w
0
~
E0
.g
c:
0
.~
III .1
.~
0~Hot Effect
0Cold Effect
-2
Alertness/Energy
Level Feelings of
Anger/Hostility
Aggressive/Violent
Behaviors
FIGURE 10.1
Rating Scale
Social theories relating temperature to three dimensions.
Overall, these results confirm that people do have social theories relating tem-
perature to a host of aggression-related variables. IThe hot temperature paradox is
also illustrated by these beliefs. Our participants believed that heat reduces alert-
ness and energy levels, while simultaneously increasing aggressive behavior. The
finding of opposite social theories for the effects of uncomfortably cold tempera-
tures may prove useful in future research on the effects of temperature on aggres-
sion, especially in ruling out various alternative explanations that rely on
suspicion or demand characteristics.
TEMPERATURE AND AROUSAL
A second part of the paradox question involves the actual effect of hot temper-
atures on arousal. Study 1 indicated that people believe that heat reduces alertness
and energy, but what are its actual effects? This simple question turns out to have
a complex answer. It depends on what one means by arousal.
In several experiments in our laboratories, we have shown that subjective per-
ceptions of arousal, as measured by the self-report perceived arousal scale (An-
derson, Anderson, & Deuser, 1996; Anderson, Deuser, & DeNeve, 1995),
decrease at hot temperatures. Specifically, people assigned to play video games,
perform a cognitive reaction time task, or do brief aerobic exercise in hot temper-
atures report feeling less aroused than people performing the same tasks in normal
temperatures. Conversely, cold temperatures produce an increase in perceived
arousal (Anderson et aI., 1996).
IA larger study (n =55) was also conducted using these same procedures and items. However, the
order of hot versus cold questions was not varied. The result was a practically identical set of means,
each of which differed significantly from the scale midpoint of "no effect."
252 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
Physiological arousal though, as assessed by changes in heart rate, complicates
the picture. Hot temperatures systematically increase heart rate, relative to normal
temperatures (Anderson et aI., 1995, 1996). Interestingly, cold temperatures seem
to produce decreases in heart rate under laboratory conditions (Anderson et aI.,
1996). Other research from a variety of laboratories produces heart rate results
similar to ours (Bazett, 1927; Hardy, 1961; LeBlanc, 1975; Tromp, 1980).
TEMPERATURE AND PERFORMANCE
Performance on cognitive tasks seems to parallel the subjective arousal results.
Hot temperatures reduce performance on various types of tasks. Tedious and
repetitive tasks that involve a low level of physiological arousal are particularly
susceptible to performance decrements in heat. Visual vigilance tasks are im-
paired when temperatures exceed 90°F (Mortagy & Ramsey, 1973; Pepler, 1958),
sometimes as a result of perceptual distortions due to heat glare and shimmer (Ko-
brick & Johnson, 1991). Auditory vigilance decreases in temperatures above
10O°F (Poulton, Edwards, & Colquhoun, 1974). Heat has been shown to impair
performance in rifle marksmanship (Johnson & Kobrick, 1988, as cited in Ko-
brick & Johnson, 1991), flight simulations (Iampietro, Melton, Higgins, Vaughan,
Hoffman, Funkhouser, & Saldivar, 1972), arithmetic tasks (Ramsey, Dayal, &
Ghahramani, 1975), and short-term memory tasks (Wing & Touchstone, 1965, as
cited in Kobrick & Johnson, 1991). Across tasks, those that are more interesting
and arousing are less affected by increases in temperature as well as those that do
not involve use of materials that become uncomfortable in heat such as metal sur-
faces and bulky protective clothing (Kobrick & Johnson, 1991). Few studies have
been conducted on cold effects and performance, and the results do not warrant
generalized summaries at this point in time. (For reviews of this literature, see Ko-
brick & Fine, 1983; Kobrick & Johnson, 1991).
PARADOX RESOLUTION
Although the effects of hot (and cold) temperatures on various kinds of vari-
ables are undoubtedly complex, no real paradox is seen in the position that hot
temperatures can simultaneously decrease energy levels and increase aggressive
behaviors. Although uncomfortable heat can decrease one's willingness to do a
variety of things, it also increases one's irritability or state hostility (Anderson et
aI., 1995, 1996). Thus, any given provocation is received more negatively by a hot
person than by a comfortable one. In other words, those kinds of aggression that
are based on impulsive or affective reactions to provocation are likely to be in-
creased by hot temperatures.
Although the focus of this chapter is on heat effects, it is important to consider
another paradox. As will become apparent in our integrated model of aggression,
discomfort appears to be the underlying factor in heat-induced aggression, sug-
gesting that uncomfortably cold temperatures should also increase aggression.
TEMPERATURE AND AGGRESSION 253
However, real world violence does not appear to increase in cold temperatures.
Doesn't this contradict the discomfort theory? Actually, it does not. The simple
reason is that throughout most of human history and even in most modern soci-
eties, relief from cold discomfort is more available (via clothing, fire, heating sys-
tems) than relief is from heat discomfort. Thus, real world studies are not good
sources of tests of cold effects. As seen in subsequent sections, cold discomfort
does increase aggression in laboratory studies in much the same way as heat does.
MAJOR ISSUES IN THE STUDY
OF HEAT EFFECTS
There are three major issues in the study of heat effects on aggression. The first
one concerns whether temperature has a direct impact on aggressive tendencies.
The second concerns theoretical explanations for the various heat effects observed
in a wide variety of contexts. The third concerns the practical significance of find-
ings on the temperature-aggression hypothesis.
EXISTENCE OF THE HEAT EFFECT
It is clear from dozens of studies (e.g., Anderson, 1989) that hot temperatures
are associated with increased violence. Of course, it is also clear that race in the
United States is strongly associated with performance on standardized intelli-
gence tests, but mere association is not the same as causation. The race/IQ asso-
ciation is hotly contested for both political and methodological reasons. The most
obvious scientific reason for doubting that race is causally linked to these large
test score effects is that there is a well-developed body of evidence linking a host
of theoretically relevant variables to test performance and race. In other words,
race in 20th century u.S. society is confounded with a number of truly causal
variables such as poverty. Thus, there is good reason to doubt that race has a di-
rect impact on standardized test performance.
Similarly, in correlational studies of the temperature-aggression hypothesis,
there may be complexities that artificially give rise to strong heat effects. The ex-
istence question in this chapter refers to whether there is a true direct causal effect
of hot temperatures on aggression. By direct impact we mean one that occurs at a
psychologicallevel, having its impact on aggression via the individual's affective
state, way of thinking, or arousal level.
It is important to distinguish between two nondirect ways in which hot temper-
atures may be linked to aggressive behavior. The first way, which consists of indi-
rect links, causally links temperature to some external factor, which in turn is
causally linked to aggressive behavior. For example, hot temperatures may be as-
sociated with increases in violent crime rates because of the kinds of routine activ-
ities that people "do" in hot weather versus cooler weather. An analogous indirect
effect in the race/IQ domain might be that skin color influences the expectations of
254 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
teachers, whose behavior toward students of different races hinders the learning of
certain racial groups while facilitating the learning of other racial groups.
The second nondirect way that temperature may be linked to aggressive be-
havior is via totally spurious links. For example, the common finding of higher vi-
olent crime rates in the "southern" cities of the United States may be a function of
the higher proportion of impoverished and disadvantaged minorities in those
cities, who typically have higher violent crime rates than other groups throughout
the United States. Because temperature differences among cities do not "cause"
the obtained socioeconomic and racial compositions of the cities, this tempera-
ture/violent crime link could be totally spurious. (However, as seen in a later sec-
tion, controlling for socioeconomic and racial composition does not eliminate the
heat effect in such studies, so it is not spurious.)
It is important to realize that the epistemological status of spurious links is not
the same as that of indirect processes. Discovery that a particular heat effect dis-
appears when spurious links are controlled weakens the temperature-aggression
hypothesis by removing the spurious finding from the column of "successful pre-
dictions." However, discovery of indirect heat effects is largely irrelevant to the
questions of whether there are true, causal direct effects of hot temperatures on ag-
gressive behavior; surely there are many variables that influence the frequency
and severity of aggression in modem society, some of which may correlate with
temperature. The scholars' tasks in this area are to identify plausible causal influ-
ences on aggressive behavior, to develop testable hypotheses that allow for dis-
confirmation, to conduct tests of these hypotheses, and to refine the theories. Of
course, if all the observed heat effects can be explained parsimoniously via an in-
direct route or are found to be spurious, then the answer to this first major issue,
concerning the existence of direct heat effects, would be negative.
THEORETICAL POSITIONS
The second major issue asks, What are the major theories that might account
for hot heat effects? Five such theories have been identified, and an integration of
the most promising aspects of several of them has been provided in a broader
model of affective aggression.
Biological Theories
Physiological theories of heat effects should be viewed on a different level of
analysis from broader sociological, cognitive, and affective models. Biological
theoretical explanations can be seen as complementing higher-level theories by
suggesting the physiological mechanisms directly responsible for heat effects on
emotions. This section briefly describes both a physiological theory of heat effects
and the relation of thermoregulation to aggression.
Zajonc (1985, 1994; Zajonc, Murphy, & Inglhart, 1989) has proposed the
innovative vascular theory of emotional efference, which focuses on the role of
blood vessels in the cavernous sinus in cooling the blood that flows into the face
TEMPERATURE AND AGGRESSION 255
and brain. The degree to which the blood is cooled or heated influences the stim-
ulation of emotional centers in the brain. Zajonc's theory grew out of Wayn-
baum's (1907) work, which posited that the facial muscles (through constriction)
regulate the amount of blood that flows to the cerebrum, which in turn influences
subjective feelings. Zajonc clarifies that facial muscles are merely one of many
regulators of cerebral blood flow but maintains that constriction or relaxation of
certain facial muscles can affect the cooling of venous blood flow to the brain.
The cooling of the brain is suggested to release certain neurotransmitters that in-
crease the positive affect.
Zajonc et al. (1989) reported increases in forehead temperature in German and
American participants as they utter the German phoneme "ii", which constricts air
flow to veins in the cavernous sinus, while reading stories aloud or repeating a
tape-recorded voice, compared to no utterance of the "ii" or to utterance of
phonemes that open the sinuses (e.g., "ah" and "e"). Participants in these studies
reported less liking for the ii sound and stated that producing the non-ii sounds put
them in a better mood than did generation of the ii. Zajonc et al. (1989) further
found that cool air, when blown into the nostrils, both decreased forehead tem-
perature and increased positive subjective feelings. In line with the vascular the-
ory of emotional efference, he concluded that when cool air is introduced to the
nostrils, as occurs with the utterance of certain sounds, the blood flowing from the
nostrils up to the forehead and the brain is cooled, thereby cooling the brain and
increasing pleasant, positive feelings in the individual.
Another biological approach relates to thermoregulation, the process that the
body undergoes to heat or cool itself in response to uncomfortable temperatures.
Several of the physiological processes involved in thermoregulation have also
been linked to emotion. The amygdala, hypothalamus, and hippocampus are all
highly neuronally interconnected and are important brain centers for thermoregu-
lation and the release of both hormones and neurotransmitters related to aggres-
sion. For example, the hypothalamus releases acetylcholine in response to ambient
cold, which increases body temperature. Acetylcholine has been shown to in-
crease aggression (Reis, 1974). The male and female sex hormones of testosterone
and estrogen have also been associated with aggression. Increases in testosterone
have been linked to aggressive behavior in men and women (Blanchard & Blan-
chard, 1984), and decreases in estrogen and progesterone (as in the premenstrual
phase) have been associated with female aggression (Buchanan, Eccles, &
Becker, 1992). The production of testosterone is influenced by corticosteroids,
which are released from the adrenal cortex when the body sweats.
The amygdala acts as an emotional computer that assigns affective significance
to incoming stimuli (LeDoux, 1993). Its neurons are also responsive to changes in
heart rate and blood pressure, which can vary with ambient temperature. There-
fore, the amygdala can create aggressive interpretations and reactions as a func-
tion of sympathetic and parasympathetic autonomic responses to temperature.
The relation between thermoregulation and emotion is far from understood
due to the complex interrelation of relevant neuronal centers, hormones, and
256 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
neurotransmitters. However, the interconnectedness of these systems suggests a
relationship among ambient temperature, body temperature, and aggression.
Southern Culture of Violence
As evidenced in the earlier quote by Montesquieu (1748/1989), social theorists
have long noticed an increase in violence in southern regions, which are closer to
the equator. Theories of a u.S. southern culture of violence range from the soci-
ological (e.g., Gastil, 1971; Hackney, 1970) to the evolutionary and economic
(Nisbett, 1990, 1993).
Some sociological approaches focus on the relatively lengthy time period in
which the South was an unsettled wilderness frontier (Gastil, 1971; Hackney,
1970). Others attribute the development of a southern culture of violence (SCV)
to swashbuckling cavaliers who settled in the early South. The cavaliers held per-
sonal honor and virtue as ideals, which they combatively defended (Cash, 1941;
Nisbett, 1993).
Of particular interest is Nisbett's theory of a southern culture of honor (Cohen
& Nisbett, 1994; Nisbett, 1990, 1993). He posits that the livelihood of people who
primarily settled in the South depended on a herding economy. In order to thrive
in this economic system, male producers were required to be highly protective of
their livestock from poachers. Because of the relative isolation that these men ex-
perienced, they alone defended their herds, their families, and their honor. These
frontier people (adaptively) socialized their offspring to hold these aggressive de-
fensive attitudes toward potential intruders as well as taught them the behaviors
necessary to fight effectively (e.g., how to operate a gun).
Nisbett (1990, 1993) cites a variety of studies in support of this view and sug-
gests that the culture of honor explains the regional differences in U.S. homicide
rates. Of course this view also requires the assumption that once a culture of vio-
lence develops, it will persist even after the economic circumstances giving rise to
it have shifted. Otherwise, the culture of honor would be irrelevant to aggression
in urban environments.
Assumptions of some SCV theories have not been consistently supported by re-
search. Bailey (1976), for example, reanalyzed Gastil's (1971) study of "southern-
ness" effects on homicide rates and found that regional effects were greatly
diminished when appropriate socioeconomic factors were controlled. Some studies
have shown no differences in southern and nonsouthern samples in gun ownership
(O'Connor & Lizotte, 1978) or violent attitudes (Erlanger, 1975). Nisbett (1993)
provides the important caveat that southern violence is primarily linked to self-
protection, so only homicides that occur in the interests of self-defense (or of per-
sona1 honor) should (according to his theory) show regional differences. Similarly,
southerners should not hold more general violent attitudes than northerners, rather
southerners should endorse more violence for self-protection purposes.
In sum, the culture of honor view posits that the southern region of the United
States has a socioeconomic history that has created a more violent culture than in
northern regions. More specifically, this perspective predicts that the old south
--------
TEMPERATURE AND AGGRESSION 257
should have higher violent crime rates than other regions of the United States. Al-
though both the culture of honor and the temperature-aggression hypotheses at-
tempt to explain the high homicide rate often found in southern U.S. cities, they
need not be viewed as mutually exclusive. A southern culture of violence (or cul-
ture of honor) could (a) have an effect on violence that is independent of temper-
ature or (b) have partially (or wholly) evolved due to hot temperatures. Although
the latter supposition is impossible to test, critical tests of the former will examine
the relationships among SCV, temperature, and violence.
Routine Activity Theory
Cohen and Felson (1979) developed routine activity theory (RAT) to explain
the link between increases in crime and increases in temperature. This sociologi-
cal view states that opportunities to commit crimes increase in the summer be-
cause social behavior patterns change. In the summer, people (potential victims as
well as perpetrators) are more likely to leave their homes and their families. In-
creases in alcohol consumption and a reduction in guardianship have also been
posited as crime-related warm weather behaviors (Cohn, 1990; Landau & Frid-
man, 1993).
RAT has been supported by some archival studies of the temperature-aggres-
sion relation (Cohen & Felson, 1979; Field, 1992) and not by others (e.g., Michael
& Zumpe, 1986). Although heat and changes in social behavior patterns co-occur,
they can, and probably do, have independent effects on aggressive behavior. Some
proponents of RAT propose that the temperature-aggression effect is at least me-
diated by, if not an artifact of, changes in routine activities.
Negative Affect Escape Theory
Baron and Bell's negative affect escape theory (NAE; Anderson & DeNeve,
1992; Baron, 1972; Bell, 1992; Bell & Baron, 1976) focuses on the current state
of the individual and their behavioral motives. According to this theory, negative
affect increases as temperatures become uncomfortably hot or cold. Both aggres-
sive and escape motives are believed to increase as negative affect increases. At
high levels of negative affect, if escape from the situation is possible then escape
motives overcome aggressive motives and escape behavior is expressed (and ag-
gressive behavior is not). If escape is not perceived as an option (as in many lab-
oratory experiments), more aggressive behavior should result at uncomfortable
temperatures.
Most laboratory studies of this theory have manipulated temperature and anger.
Those that support the NAE have found a temperature x anger interaction in
which hot temperatures led to decreases in aggression in angry conditions and in-
creases in aggression in nonangry conditions (Anderson & DeNeve, 1992). This
interaction has not been found in some studies (Baron, 1972; Bell & Baron, 1977)
and has occurred in the reverse fashion in at least one (Bell, 1980). Overall, labo-
ratory data testing the NAE as applied to the temperature-aggression hypothesis
are inconsistent (Anderson, 1989). Further studies involving behavioral measures
258 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
of aggression as well as those that pit aggressive and escape motives against each
other are required to test the specific predictions of the theory.
Social/Cognitive Theories
Two theories that focus on the influence of environmental factors on aggressive
cognitions and behaviors are (a) Bandura's (e.g., 1973) groundbreaking social
learning theory (SLT) and (b) Berkowitz's (1984, 1993) contemporary cognitive
neoassociation theory (CNT) of emotion. According to SLT, witnessing aggres-
sion leads to the acquisition, performance, and maintenance of aggressive behav-
ior. An important aspect of this theory is that the aggressive behavior that is
viewed can be encoded into memory and retrieved when the witness is faced with
a similar situation (Geen, 1990). For example, a boy who grows up in an abusive
household may well learn to use physical force as a means of coping with un-
pleasant situations. He may learn both the "how to hit" aspects and the "why hit"
aspects, especially if the modeled aggression appears to work. As an adult he may
well recreate this abusive pattern in his own family. Hot temperatures contribute
to the unpleasantness of many situations and could therefore trigger the expression
of the learned aggressive behavior patterns.
CNA, a more recent cognitive theory, is based on network models of memory
that focus on the interconnectedness of related incidents or thoughts in memory.
According to CNA, thoughts, feelings, and behavioral programs are stored to-
gether in memory such that when a particular thought or feeling emerges into con-
sciousness, related thoughts, feelings, or behavioral propensities are activated
(may be experienced). Central to CNA is the role of negative affect as the initia-
tor of hostile thoughts, feelings, and behaviors. Berkowitz states that an increase
in negative affect (as occurs in uncomfortable temperatures) can automatically
bring aggressive thoughts and recollections to mind, as well as lead to the imme-
diate experience of angry feelings and aggressive behavioral inclinations.
Dodge and colleagues (e.g., Dodge & Crick, 1990) and Huesmann and col-
leagues (e.g., Huesmann, 1984; Huesmann, Eron, Lefkowitz & Walder, 1984)
have similarly shown that children learn aggressive scripts and apply them in
many normal situations in everyday life. These scripts also bias the interpretation
of ambiguous encounters in an aggression-related way. Presumedly, these scripts
are learned in a variety of ways from many sources (including television) and can
be linked to a variety of nodes in memory, such as negative affect.
A General Affective Aggression Model
A theory of affective aggression (Anderson et aI., 1995, 1996) incorporates as-
sumptions of some of the aforementioned theoretical perspectives. As the various
perspectives address different aspects of the generation and maintenance of ag-
gressive behavior, from the physiological to the social, the various theories are
best considered as addressing different levels of analysis. Our goal is to integrate
theoretical processes at different levels in order to more fully understand the
processes by which social/environmental variables (e.g, temperature) operate via
aggressive cognitions, feelings, and arousal to produce aggressive behaviors.
TEMPERATURE AND AGGRESSION 259
",,:/\
::.::.:::.:.:/\
.. ~
.
..
Arousal
hihysiological
ercerceived (subjective)
Reappraisal
eIneInterpretation of the situation
eteinterpretation of affect
onoping alternatives
LiLiely consequences
FIGURE 10.2 A general affective aggression model.
Our framework posits three main routes through which various input variables
can influence aggressive behavior. Certain environmental factors (e.g., frustra-
tion) or individual differences (e.g., aggressive personality) can increase the ac-
cessibility of hostile thoughts, hostile feelings, or physiological arousal (see
Figure 10.2). The automatic priming of aggressive thoughts and feelings may well
come about, in part, because of the biological processes at work. For instance, Za-
jonc's vascular theory may account (in part or full) for increases in hostile affect
in hot temperatures. Other thermoregulatory processes may also create specific
aggression-related emotional and cognitive states via hormonal and neural links.
Consistent with CNA and various cognitive models, these initial priming ef-
fects may well spread to a variety of associated thoughts, feelings, scripts, and
260 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
motor programs. All this sets the stage for subsequent aggressive behavior. When
in such a state, people may feel more angry and may interpret ambiguous remarks
as being more hostile than they normally would. In turn, this can influence how
they behave. Under the right (or wrong) circumstances, an aggressive behavior
may well result, which begins a reciprocal aggression cycle with another person
that can spiral out of control.
In our model, situational factors such as hot temperatures as well as individual
differences such as hot temperaments (aggressive personality) operate by making
aggression-related thoughts and emotions readily accessible for application to the
current situation. In other words, a variety of factors can increase the person's
"preparedness" to aggress. Other factors may mitigate or exacerbate the aggres-
sive impulse and its eventual expression in behavior. Such factors may influence
the initial appraisal or the reappraisal processes. For instance, learning about mit-
igating circumstances may decrease a person's anger at someone who has
thwarted their attempts at some task and may also decrease retaliative behaviors
toward that person (e.g., Dill & Anderson, 1995).
PRACTICAL SIGNIFICANCE
The third and final major issue concerns practical implications. Although many
implications follow directly from the work on the temperature-aggression hy-
pothesis, others follow from the broader theoretical context concerning affective
aggression. If the temperature-aggression hypothesis is correct, what can be done
to reduce unwarranted aggressive behavior in society? In what situations or con-
texts would an intervention be likely to work? What type of interventions are
likely to produce the desired effects? These issues will be examined at the end of
the chapter, after empirical and theoretical issues have been presented and dis-
cussed in detail.
EPISTEMOLOGICAL STRATEGIES
This section specifies the epistemological strategies and assumptions used in
examining the temperature-aggression literature. Definitions of key concepts used
throughout this chapter are also spelled out in some detail.
TRIANGULATION
Triangulation is best illustrated by the following quote by Richard Cardinal
Cushing (New York Times, 1964), commenting on the propriety of calling Fidel
Castro a communist, "When I see a bird that walks like a duck and swims like
duck and quacks like a duck, I call that bird a duck."
In other words, one extremely valuable way of examining a proposition is to
test it from several different perspectives. If the results of tests from several per-
------
TEMPERATURE AND AGGRESSION 261
spectives converge on the same answer, confidence grows. This approach has a
variety of names such as "multiple operationism" or "triangulation" (e.g., Ander-
son, 1987; Campbell & Fiske, 1959; Crano & Brewer, 1973; Feigl, 1958; Mc-
Grath, Martin, & Kukla, 1982).
In the temperature-aggression domain, three major perspectives have emerged:
geographic region effects, time period effects, and concomitant heat effects (An-
derson, 1989). Geographic region studies examine indices of aggression in re-
gions that differ in climate or concurrent temperature. Do hotter cities have higher
violent crime rates than cooler ones, for example? Time period studies examine
aggression rates in one location (e.g., in one city, or one country) but in a number
of different time periods that differ in temperature. Do hotter months produce
higher violent crime rates than cooler ones? Concomitant studies assess aggres-
sion and temperature at the same time (i.e., concomitantly) and at several differ-
ent times in which the temperature varies. For example, do subjects randomly
assigned to a hot condition behave more aggressively toward a provoking confed-
erate than subjects assigned to a comfortable temperature? In a sense, concomi-
tant studies are really a subset of time period studies. The advantage of assessing
temperature and aggression concomitantly is sufficiently important to warrant
treating it as a separate perspective. These studies are typically performed in lab-
oratory settings, although several impressive field studies of this type have also
been conducted.
The main advantage of the triangulation approach is that weaknesses of a par-
ticular type of study usually do not apply to other types. Therefore, consistency of
results across different types of studies allows us to triangulate or "home in" on a
true causal factor. Thus, if the temperature-aggression hypothesis is supported in
studies of geographic region effects, time period effects, and concomitant temper-
ature aggression effects, we can be fairly sure that hot temperatures do have a di-
rect effect on aggression.
META-ANALYSIS (IN SPIRIT, IF NOT IN METHOD)
The current popularity of meta-analytic techniques is often warranted and
sometimes misplaced. On the one hand, using statistical procedures to combine the
effect sizes of different studies examining the same hypothesis can be revealing, es-
pecially in literatures where there is considerable diversity of results and contro-
versy in their interpretation. On the other hand, there is an unwarranted tendency to
view the traditional narrative review approach as being vague and subjective. In ac-
tuality (at least, in our view) the traditional narrative review has many of the best
features of a good meta-analysis. Both collect as many of the relevant studies as can
be found. Both categorize the studies on the basis of potentially relevant features,
such as type of dependent variable measure used, or the environmental setting of
the research. Both test hypotheses about whether the underlying effect is consistent
and whether certain features tend to increase, decrease, eliminate, or reverse the
effect. Indeed, some "narrative" reviews of the past have statistically combined
262 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
results from different studies to get a clearer view of the overall pattern of evi-
dence (e.g., Anderson, 1989; Anderson, Miller, Riger, Dill, & Sedikides, 1994).
This chapter has features of traditional narrative reviews and of meta-analytic
approaches. The published literature was exhaustively searched for studies rele-
vant to the temperature-aggression hypothesis. Unpublished material was ex-
cluded both because we are not particularly interested in establishing the exact
effect size and because excluding unpublished work provides some quality con-
trol. In addition, a few of the published studies were excluded because their data
were reported elsewhere, data were reported in so poor a fashion that we could
not determine whether they really supported or contradicted a given position, or
the methods used were so poor that any conclusions based on them would be
misleading at best. Where possible, results of statistically combined sets of stud-
ies are presented. In most cases, officially sanctioned meta-analytic procedures
were not used, either because they were not needed or because they seemed in-
appropriate. In essence, a meta-analytic approach as adopted in spirit, if not al-
ways in method.
DESTRUCTIVE TESTING
"Destructive testing" is a term borrowed from structural engineering and ma-
terials science (Timoshenko, 1953; Wilson, 1984). It is best illustrated by analogy.
A new metal alloy has been developed. Initial tests have shown it to be fairly
strong. To find out just how strong it is, a series of destructive tests are conducted,
i.e., increasing stresses are applied to a sample of the new alloy until it breaks.
There is no question about whether it can be broken, only how much stress it can
take before it does break.
The initial test of strength is analogous to a zero-order correlation, perhaps be-
tween the hotness of U.S. cities and their violent crime rates. It is a simple test of
the basic theoretical hypothesis. Adding increasing stresses is analogous to adding
various statistical controls to the regression model, such as poverty rates in vari-
ous cities. The relevant question about the obtained relation between heat and
aggression is not, "Can it withstand all possible attempts to reduce it to non-
significance?" What is of interest in destructive testing is how much stress the tar-
get relation can withstand. The ultimate judgment concerning the strength of the
target relation (here, temperature and violent crime rate) is somewhat subjective
and will therefore differ from scholar to scholar. However, there are reasonable
rules of thumb that all scholars can apply. For instance, control variables with
good theoretical grounding are more appropriate and more informative than post
hoc variables of dubious relevance.
Destructive testing is a hybrid of the traditional "theory centered" approach to
science, in which hypotheses are derived from a formal theory and then tested,
and the "result centered" approach advocated by Greenwald, Pratkanis, Leippe,
and Baumgardner (1986), in which one asks, "Under what conditions does x lead
to y?" rather than "Does x lead to y?" This approach was applied in evaluating the
263
TEMPERATURE AND AGGRESSION
current temperature-aggression hypothesis literature (for a more detailed presen-
tation of destructive testing, see Anderson & Anderson, 1996).
DEFINITIONS
Some of the controversy in the temperature-aggression hypothesis literature
stems from ambiguous definitions of key concepts. Definitions for several con-
cepts are provided to reduce this problem.
Aggression, Affective Aggression, and Violence
"Aggression," "affective aggression," and "violence" are used interchangeably
in this chapter, with the only distinction being that violence is restricted to the
most extreme types of aggressive behavior. Our definition includes three compo-
nents. First, these terms refer to behavioral acts, not to thoughts or feelings. We
explicitly specify when we mean "aggressive cognitions" or "aggressive feel-
ings." Second, the primary intent of these acts is to harm the victim. Third, there
is a large anger/hostility component in the acts.
Some behaviors typically classified as violent do not meet this definition. For
instance, robbery is classified by the FBI as a violent crime. However, in our view
the primary intent in most robberies is not to harm the target, it is to gain some
economic benefit. Thus, because of the motive ambiguity, robbery is not included
as a measure of aggression or violence.
Another somewhat ambiguous case is rape. Although it is clear that rape is vi-
olent in its consequences to the victim, there is some debate about what portion of
rapes are perpetrated with harm as the primary intent (e.g., Felson, 1993). We be-
lieve that a significant portion of rapes are intended to harm the victim, and that
anger or hostility toward the victim is a part of many rapes. Thus, rape is included
as a measure of aggression/violence. However, because of the mixed motives in-
volved, one might expect weaker relationships between rape and aggression-
instigating variables than between more purely aggressive acts (e.g., assault) and
aggression-instigating variables. In sum, our definition of aggression includes
most acts that typically have been labeled as spontaneous aggression, pain-
induced aggression, and affective aggression.
Temperature Aggression Hypothesis vs Heat Effect
The temperature aggression hypothesis refers to the theoretical statement that
uncomfortable temperatures cause increases in aggressive motivation and (under
the right conditions) in aggressive behavior, and that they do so in a direct fash-
ion. In most cases the temperature-aggression hypothesis refers to the "hot" side
of this relation. On occasion, however, it also refers to the "cold" temperature side.
As noted in earlier works (e.g., Anderson & Anderson, 1996), any theoretical hy-
pothesis is protected from disconfirmation by multiple translation layers; the more
abstract the theory, the more translation layers are necessary to get to the specific
empirical realizations that can be subjected to testing.
264 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
The heat effect refers to an empirical observation that hot temperatures are pos-
itively associated with increased aggressive behavior. It is, essentially, a brief de-
scription of a specific type of empirical relation.
With these epistemological and definitional issues in hand, we now turn to the
empirical literature. When possible, we will summarize findings presented in An-
derson (1989). Newer results will be presented in greater detail.
GEOGRAPHIC REGION
Anderson's earlier reviews (1989; Anderson & DeNeve, 1992) revealed amaz-
ing levels of consistency in geographic region effects across countries and eras.
Hotter locations have higher violent crime rates than cooler locations within the
same country.
EARLY STUDIES
Some of the early studies of the temperature-aggression hypothesis do not pre-
sent data in sufficient detail to allow statistical analysis. Nonetheless, the results
are impressive in their consistency. For instance, Guerry (cited in Brearley, 1932)
reported that in the years 1826-1830 crimes against people (e.g., assault) were
twice as prevalent in southern France as in central or northern France, whereas
crimes against property (e.g., burglary) were twice as prevalent in the north. Sim-
ilarly, Lombroso (1911) reported that the homicide rate in the south of England
was almost 10 times that of northern England.
Other studies did allow some form of statistical analysis. Lombroso (1911) re-
ported several aggressive crime rates by degrees latitude of the region for both
Spain and Italy. In both countries, of course, latitude is essentially a proxy measure
of average temperature. Anderson (1989) found that in Lombroso's data violent
crime rates correlated significantly and positively with latitude in both countries.
Brearly (1932) reported the state-level homicide rates in the United States during
the 1918-1929 time period. Anderson (1989) showed that in Brearly's data "the
southern states had dramatically higher homicide rates (M =19.37 per 100,000)
than did the northern states [M =3.55, t(16) =7.93, p< 0101"] (p. 79).
MODERN ERA STUDIES
None of the early studies included controls for other variables such as poverty
rate, other than by restricting aggression rate comparisons to within country com-
parisons. Several modern era studies similarly focused primarily on the heat effect
with minimal attention to possible third-variable controls. Lester (1986) compared
the homicide rates of the 45 largest standard metropolitan statistical areas
(SMSAs, i.e., cities) in the United States in 1970. Predictor variables included av-
erage temperature and precipitation (30-year averages), latitude, and longitude. Of
-.----
TEMPERATURE AND AGGRESSION 265
most relevance here was the finding that temperature was strongly correlated with
the homicide rate (r =44, p< 0101).
Two studies compared aggression rates as a function of climate across coun-
tries and cultures. The one by Robbins, DeWalt, and Pelto (1972) included mea-
sures of behavior that would meet our definition of aggression (homicide) as well
as one that is less clearly an example of affective aggression (incidence of war-
fare). They found a significant positive relation between temperature and homi-
cide (p < 0101,but no relation with warfare. Schwartz (1968) similarly found no
relation among the temperatures of 51 countries and frequency of political vio-
lence. In other words, the heat effect is by and large restricted to what we have
termed affective aggression. More planful violence, such as wars and revolutions,
does not appear to be related to temperature.
Other studies in the modern era have included various types of socioeconomic
and social controls. deFronzo's (1984) study of crime rates in 142 SMSAs in the
United States in 1970 included such controls. Although the results provided sup-
port for the temperature-aggression hypothesis at the level of zero-order correla-
tions, statistical analysis shortcomings preclude any firm judgments about the
strength of the heat effect when socioeconomic factors are controlled (for more
details, see Anderson, 1989). Rosenfeld (1986) included some socioeconomic fac-
tors in two studies of crime rates in U.S. cities in 1970 and found significant re-
gion effects. Southern cities were especially higher than northern ones on murder
and assault. Other region studies (Rotton, 1986; Rotton, Barry, & Kimble, 1985)
have also found support for the temperature-aggression hypothesis, despite some
conceptual and statistical ambiguities (for more details, see Anderson, 1989). The
most extensive geographic region study is one on 1980 crime rates in 260 SMSAs
(Anderson, 1987). That study included 14 "social" variables for each city such as
unemployment, per capita income, education, age, and racial composition. The re-
sults were that hotter cities had higher violent crime rates even when other social
variables were statistically controlled.
More recently, Anderson and Anderson (1996) recreated this data set with sev-
eral modifications. The most important addition was a "southern culture of vio-
lence" index for each city. White arrest rate data were also gathered for these same
cities in 1980, in response to Nisbett's (1993) suggestions that his "culture of
honor" thesis can explain the higher incidence of violent crime in the south and
that only white perpetrator data are relevant. The specific procedures and results
are described in the following section.
Violent Crime Rates in U.S. Cities
Aggression
Two measures of violent crime were constructed. The first was a z-score com-
posite of overall murder, rape, and assault rates. The second was a z-score compos-
ite of murder, rape, and assault arrest rates for the white population, as suggested
by Nisbett's work.
266 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
Southernness
Each city's southernness was indexed by converting three indicators to zscores
and summing them. One indicator was a simple south-nonsouth distinction based
on U.S. Census Bureau classifications. Cities in southern states were assigned a
score of 1, whereas non south cities were given a score of O.The second indicator
was Gastil's (1971) southernness index, based on migration patterns from the old
south. The third indicator was the percentage of voters who voted for George Wal-
lace in the 1968 presidential election (Scammon, 1970). Preliminary analyses re-
vealed that this SCV index was linearly and curvilinearly related to violent crime.
We therefore converted the SCV index score to zscores and then created a qua-
dratic term (SCV2). This procedure allowed us to examine the linear SCV effect
on violent crime independently of the curvilinear component.
Temperature
The hotness of each city was assessed with a four-item summed index of z
scores. The items were (1) number of hot days (2::32.2°C, 90°F), (2) number of
cold days (~O°C, 32°F), (3) cooling degree days (amount of cooling needed to
maintain a comfortable base temperature of 18.3°C, 65°F), and (4) heating degree
days (amount of heating needed to maintain a comfortable base temperature of
18.3°C, 65°F). The number of cold days and heating degree days were multiplied
by -1 so that larger scores meant warmer temperatures on all indicators.
Social Variables
The 12 social variables used were unemployment rate, per capita income,
poverty rate, mobility (percentage living in a different home in 1975), high school
education (percentage of the 2::25-year-old population that had graduated), college
education (percentage of the 2::25-year-old population that had attended 4 or more
years), population size, percentage Black, percentage Spanish, percentage less
than 18 years old, percentage 18-64 years old, and median age.
Results: Heat Effect on Violence
Figure 10.3 presents the results from several regression analyses on the heat ef-
fect on violence. As can be seen, regardless of whether we examine overall crime
rates, or the white arrest rate for violent crime, temperature significantly predicts
violence rate differences in major U.S. cities. This is true even when all 12 social
variables and the southernness index are partialled out first.
Results: Southernness Effect on Violence
Figure lOA presents the parallel results when the southernness index is the pre-
dictor of interest. As can be seen in Figure lOA, the linear correlations between
the southernness index and both violence measures are positive and statistically
significant when the only other predictor in the model is the curvilinear term.
However, controlling for temperature and other predictor variables eliminates the
linear southernness effect on overall violent crime rate and appears to reverse its
effect on the white arrest rate for violent crime.
TEMPERATURE AND AGGRESSION 267
0.8
d
0
..=
ro
]
Q)
g
Q)
'0
>
~
B
S
8.
sp< 505
Q)
E-<
0.6
0 Overall Violent Crime Rate
ELI White Arrest Rate
0.4
0.2
0~v////
SCV & SCV2
~
~
None All Predictors
Variable(s) Partialled from the Temperature/Violence Relation
FIG U REI 0.3 Zero-order and partial correlations between temperature and violence.
0.6
§
.~
]
8
g0.2
~p< 505
0
>
~
'"
Q)
§p< 505
~.2.2
E
0
[fJ
0.4 0 Overall Violent Crime Rate
ELJ White Arrests
0.0
.4.4 SCV2 SCV2 & Temp. All Predictors
Variable(s) Partialled from the Linear SCV/Violence Relation
FIG U REI 0.4 Partial correlations between southernness (SCV) and violence.
Results: Latent Variable Analysis
In the preceding regression analyses, all 12 social variables were used. A look
at the list of variables reveals some redundancy in them. For instance, there are
several measures of average wealth or poverty. To get a better picture of the re-
lationship between temperature and the violent crime rate, a number of latent
268 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
)Direction Reversed
TLI... 494
GFI 9292
Chi-Sq 113
D.F. 28
tTVC 4.61
t SVC 1101
FIGURE 10.5 Latent variable model of temperature and southernness effects on violent crime.
variable models were tested. Getting a model with a good overall fit and without
serious statistical shortcomings proved to be a difficult task. Figure 10.5 presents
a simple model that includes only violence, temperature, and southernness vari-
ables.2 As can be seen from Figure 10.5, the overall fit was quite good according
to the Tucker-Lewis index, the goodness-of-fit index, and the ratio of X2 to de-
grees of freedom (Church & Burke, 1994; Marsh, Balla, & McDonald, 1988;
Marsh & Hocevar, 1985). Figure 10.5 also shows that the latent temperature fac-
tor is significantly related to the violent crime latent factor, whereas southernness
is not. Indeed, the southernness relation to violent crime is even in the wrong
direction.
Figure 10.6 adds the three social variables that produced the best fit, given the-
oretical constraints. The percentage of the city population that was classified as
Black or Spanish (Census Bureau designations) and the poverty rate combined to
produce a latent factor called low socioeconomic status (SES). Population size also
proved to be a valuable predictor of violent crime rate, but it did not form a mean-
ingfullatent factor. The results of this analysis were very similar to the model that
did not include any of the social variables. The various fit indices all yielded ac-
ceptable fits. The latent temperature factor was significantly related to the latent vi-
olent crime factor. Southernness was related to violent crime in the direction
predicted by the southern culture of violence model, but not significantly so (t < 1).
In addition, the low SES latent factor was positively related to violent crime. Cities
ZError terms and correlations among error terms in the final model are not displayed in Figures
10.5 and 10.6 to simplify the picture. We tested identical models with a quadratic southernness mea-
sure as well. The basic results shown in Figures 10.5 and 10.6 were replicated, but there were statisti-
cal shortcomings that led us to prefer the displayed models.
---..-..._---
TEMPERATURE AND AGGRESSION
269
SINDEX
South
Wallace Vote
IPopulation TLI. 9292
GFI. 989
Chi-Sq 216
D.F 54
t TVc. 2.35
t SVc. 9898
t SESVc.. 2.60
t PVC.. 9.27
ODirection Reversed
FIG U REI 0.6 Latent variable model of temperature and southernness effects on violent crime,
with three social variables.
with large, poor, minority populations had higher violent crime rates. Finally, pop-
ulation size was positively related to violent crime rates, even though this data set
(restricted to SMSAs) contains only fairly large metropolitan areas.
SUMMARY OF GEOGRAPHIC REGION STUDIES
Across a wide range of years and countries, the geographic region studies pro-
duce a highly consistent picture of heat effects. Hotter regions in the United States
and in several western European countries have higher violent crime rates than
cooler regions. The more recent studies that have included statistical controls for
various possible confounds and indirect effects show that the heat effect in region
studies cannot be easily dismissed as artifactual.
Of course, by themselves these studies do not rule out all possible indirect ef-
fects. For instance, it is possible that better measures of southern culture of violence
would yield somewhat different results. For example, it is possible that an attitude/
value measure of adherence to the culture of honor, administered to a representative
sample of residents of each city, would eliminate the heat effect when partialled out
statistically. In other words, it could be that hot locations have tended to produce
cultures of honor in which high violence is a part of being well socialized. If true,
this temperature/culture of honor relationship would then need explaining. As noted
elsewhere (Anderson & Anderson, 1996), the temperature-aggression hypothesis
could easily handle such a state of affairs. If hot temperatures do tend to increase
270 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
aggressive behaviors, cultures that develop and evolve in hot climates would tend to
develop rationales (or rationalizations) for violent acts committed by its high status
citizens. In other words, direct heat effects could, over time, lead to the development
of a set of attitudes and values that encourage violence, at least under certain "hon-
orable" conditions.
Several theories remain plausible as explanations for the obtained heat effect in
these region studies. Routine activity theory could be playing a role, although cer-
tain versions of it cannot account for the findings. Perhaps the typical "routine"
activities of people in warmer climates are those that promote aggressive behav-
iors. There may be more intermingling of people in general, increasing the poten-
tial for conflict and violence.
Although data argue against a strong version of southern culture of honor the-
ory-the version that says all region-based heat effects are due to SCV and none
are due to direct heat effects-a weaker version seems plausible and is not com-
pletely ruled out by extant findings. There may well be a southern culture of vio-
lence that has developed in the u.S. south, and it may continue to increase
violence independently of direct heat effects. Nisbett's preliminary studies on his
version of culture of honor (e.g., Nisbett, 1993) are quite interesting in this regard.
The general model of affective aggression is more strongly supported by these
region data than are other theories. The temperature-aggression hypothesis aspect
of the general model has survived a number of possible disconfirmations, and thus
gains strength.
TIME PERIOD
Time period studies examine aggression rates across time periods that differ
in temperature. Many such studies have been conducted using time periods of
various lengths, with the unit of analysis being as short as a day or as long as a
year. Because humans are skilled at carrying grudges and recreating earlier emo-
tional states, time periods of less than 1 day are probably inappropriate except for
studies in which the target aggressive behaviors can be carried out by the subject
almost immediately. (Most laboratory studies of aggression have this character-
istic, for example.) Anderson (1989) presents a detailed summary of many time
period studies. This section mentions those studies briefly and focuses on more
recent studies.
HOT DAYS
Anderson (1989) analyzed Dexter's (1899) New York City assault data and
found both linear and curvilinear effects of temperature on the relative frequency
of assaults (both ps < 0101).The specific form of this relation was that at cool tem-
peratures, slight increases in temperature had only a small impact on assault,
whereas at uncomfortably warm temperatures, further increases in temperature
TEMPERATURE AND AGGRESSION 271
yielded relatively large increases in assaults. Carlsmith and Anderson's (1979)
study of riots in the United States in 1967-1971, Harries and Stadler's (1988)
study of assault in Dallas, Rotton's (1982) study of rape in Dayton, Rotton and
Frey's (1985) study of assaults and family disturbances in Dayton, Cotton's (1986)
studies of violent crime in Des Moines and in Indianapolis, and Anderson and An-
derson's (1984) studies of violent crime in Chicago and in Houston all produced
similar findings of maximum aggression in the highest temperature ranges. Inter-
estingly, several of these studies also examined heat effects on less aggressive
crimes such as burglary. The heat effects were consistently weaker for these "less
violent" crimes (Anderson & Anderson, 1984; Cotton, 1986).
More recent studies add to the consistency of the heat effect in daily time pe-
riod studies. Three studies have used calls to police departments as criterion vari-
ables. LeBeau and Langworthy (1986) studied the frequency of "calls for service"
to the Chicago Police Department during 1976-1979. They found (among other
interesting effects) that temperature was the best predictor. As temperature went
up, so did the frequency of calls. However, because calls for service include calls
for relatively nonaggressive problems as well as aggression-related ones, the re-
sults must be interpreted with caution.
Walters (1991) examined temperature and pollen counts as predictors of fre-
quency of 911 calls to the Kansas City, Kansas, police department in the years
1986-1989 from March 1 through October 31 in each year. Data for each year
were broken down into three time periods: March I-May 24, May 25-August 14,
and August IS-October 31. Results were reported in the form of correlations in
each time period in each of the 4 years. The temperature/police call correlations
were averaged for each time period across years. For each time period the average
temperature/police call correlation was significant (all pvalues < 1)1),r=7,7, r=
8,8, and r=99, for the first, second, and third time periods, respectively. Even
though each was significant, it is obvious that the lowest correlations occurred in
the second time period. We suspect that this is due to a relatively smaller range (or
standard deviation) of temperatures in that late spring/early summer time period,
although we cannot be certain of this. It is also interesting to note that partialling
out the pollen count factor did not appreciably affect the temperature/police call
correlations. Once again, these data support the temperature-aggression hypothe-
sis, but caution in interpretation is warranted because not all 911 calls involve af-
fective aggression. (Walters reports that about 85% of these calls are for police
assistance of some kind.)
Another interesting study of the heat effect in daily time periods is Reifman,
Larrick, and Fein's (1991) study of aggression in major league baseball. The cri-
terion variable was the number of players hit by a pitch (HBP) in the 1986-1988
seasons. A regression analysis of the heat effect included statistical controls for
walks, wild pitches, passed balls, errors, home runs, and attendance. The heat ef-
fect was still significant (p < 0202), with relatively more HBPs in games played on
hotter days. Other control procedures were used to rule out the possibility that
some teams whose home games were played in hot climates were simply more
----
272 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
likely to have high HBPs regardless of temperature. Finally, in a replication study,
data from the 1962 season showed essentially the same heat effect correlation.
Cohn (1993) examined weather and temporal variables as predictors of police
calls for service for rape and for domestic violence in Minneapolis for the years
1985, 1987, and 1988. Although the results are complex and occasionally difficult
to interpret, the main findings of relevance to this chapter are (a) temperature was
strongly related to both types of violence (p values < 0101), (b) higher tempera-
tures yielded higher violence rates, and (c) rape was less predictable by the whole
set of variables than was domestic violence. Cohn (1993) discussed several pos-
sible reasons for the weaker results on the rape measure. Of particular interest is
her discussion of other work showing that a significant portion of rapes are
planned. This suggests, as noted earlier, that rape may well be a less pure exem-
plar of spontaneous affective aggression and thus should be examined separately
from purer acts of affective aggression such as assault and murder. One alterna-
tive explanation of the weaker results for rape concerns the incidence rate. Do-
mestic violence calls were about 40 times as frequent as rape calls, which would
tend to make the domestic violence measure less susceptible to random fluctua-
tions in frequency than rape.3
Cohn and Rotton (1997) conducted a time series analysis of the reported as-
saults in Minneapolis in 1987 and 1988 as a function of time of day, day of week,
month, and temperature. (Many other "control" variables were also included, but
are not particularly relevant to this chapter.) The study was performed to test
(a) the NAE-inverted U shape prediction of the temperature-aggression relation
and (b) the RAT assumption that time of day moderates the relation between tem-
perature and assault rates. The authors found significant positive relations be-
tween temperature and assault for every time period (in 3-hr intervals) except
between 6:00 and 11:59 a.m. This complex data set contains many other interest-
ing results. For instance, evidence shows that time of day and day of week mod-
erate the effects of temperature on assault. This is entirely consistent with any
model of aggression: opportunities are needed to aggress, and opportunities vary
by day of week and time of day.
One of the main conclusions of the article-that there was a significant down-
turn in assault as temperatures became hot-is simply not borne out by the re-
3Cohn's regression approach included linear, quadratic, and cubic terms, represented by the raw
temperature (T), raw temperature squared CT2), and raw temperature cubed (T3). In some analyses the
squared and cubed terms were kept without all of the lower order terms. This creates some difficult in-
terpretation problems because the higher order terms are necessarily highly confounded with the lower
order terms unless raw data are first converted to deviation score form. Thus, the finding that only the
T2 term yielded a significant effect in the 1985/1987 domestic violence model (Cohn's Table 7) could
be misinterpreted as meaning that only a curvilinear effect occurred. However, plotting a line using the
T2 beta and T2 scores (e.g., 40° becomes 1600,50° becomes 2500 . . .) and then converting the xaxis
back to linear temperatUre produces the heat effect curve frequently found in this literature. At low
temperatures, there were little increases in domestic violence rates with small increases in temperature,
but at uncomfortably warm temperatures, further increases produced large increases in domestic vio-
lence. This is actually both a linear and a quadratic effect. Similar interpretational ambiguities exist
with the 1985/1987 rape model, which included Tand T3, but not T2.
~-----
TEMPERATURE AND AGGRESSION 273
2
1
..
..............
~
~0.......................
......
,.................
......
"3
~-1
c;I)
<-2 --0-- Ignoring Time & Month
Using Time, Day of Week, & Month
-3 30 45 60 75
Temperature of
FIG U REI 0.7 Relationship between temperature and standardized assault rate in Minneapolis
(1987-1988). Slopes were derived from Cohn and Rotton (1997).
015 90 105
ported results. Conceptual, methodological, and statistical problems with this study
take us well beyond this chapter (for a more thorough critique, see Bushman, An-
derson, & Anderson, 1998). Two of these problems are particularly important.
The first problem concerns interpretation of the results reported in the article.
Most results actually showed a clear linear relationship between temperature and
assault rates, but little evidence of a downturn in assault at higher temperatures.
For example, in the 56 day of week (7) x time of day (eight 3-hr blocks) analyses
the linear temperature term was positively (p <5)5) related to assault 29 times.
However, the curvilinear temperature term yielded a statistically significant effect
in only 10 cases, and in 7 of those the slope was positive, indicating that at hotter
temperatures assault rates increased more rapidly with further temperature in-
creases. These findings clearly contradict Cohn and Rotton's (1997) paramount
conclusion that the shape of the temperature-assault relation is an inverted U; they
also exemplify how overaggregation of data is problematic when analyses reveal
interesting differences in effects at more microscopic levels of analysis. If the
analyses that purported to show important day of week and time of day effects are
accurate, then the best way to estimate the overall (or average) relationship be-
tween temperature and assault rate is to average the linear and curvilinear (qua-
dratic) slopes across the 56 day of week (7) x time of day (eight 3-hr blocks) sets
and plot the resulting equation. These averages are computed and the resulting re-
gression line (with the diamonds) is plotted in Figure 10.7. This line shows that
not only did assault rates continue to climb at the hottest of normal temperatures,
but they increased at a faster rate. This is because the average curvilinear slope
was positive. Even this regression line may underestimate the true heat effect be-
cause of other temperature-related "control" variables in the statistical model.
274 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
The second problem concerns the use of temperature-related "control" vari-
ables. Statistically controlling for the month of the year may have artificially re-
duced the effect of the linear temperature term on assault. Because month is
highly correlated with temperature, removing the variance (in the assault rate)
that is associated with month effects may remove much of the variance that is
truly part of the temperature effect. Although Cohn and Rotton (1997) claim that
removing month from the statistical model produced little change, it did in fact
significantly change the shape of the temperature-assault curve. A model with
month statistically controlled yielded linear and curvilinear slopes (+.0337 and
006006, respectively) that produce a major decline in assault rates beginning at
about 77°F. However, a model that differed only in the removal of month as a
control variable yielded slopes that produce continued increases in assault rates
through 100°F. The line with the circles in Figure 10.7 displays this latter tem-
perature effect on assault. Other "control" variables in various models (including
both models in Figure 10.7) may also have artificially deflated the statistical esti-
mate of the linear temperature term as well. Furthermore, these two models (with
and without month as a control variable) further suffer from the over-aggregation
problem outlined in the preceding paragraph.
In sum, although both of the lines in Figure 10.7 already contradict the main
conclusions of Cohn and Rotton (1997), there is reason to believe that statistical
models that produced these lines inappropriately reduce the estimated increases in
assault rates at high temperatures. The "diamond" line in Figure 10.7 is based on
a statistical model that uses inappropriate temperature-related control variables.
The "circle" line is based on a model that overaggregates across time and day of
the week.
Other problems also warrant attention. Minneapolis is an inappropriate city to
sample when testing linear and curvilinear temperature effects because the NAE
downturn is posited to begin in the mid-eighties. Minneapolis reaches truly hot
temperatures far less often than do southern cities, and there may well have been
too few time periods with truly hot temperatures to accurately test for a hot down-
turn in assault. Furthermore, use of assault reporting rates is problematic when
studying time of day effects because assault reports may frequently occur well
after the crime has been committed. Also, heat stress may have a cumulative ef-
fect over time that is expressed later, sometimes during somewhat cooler periods
of time. The point is that although the Cohn and Rotton data are useful for some
purposes (e.g., testing whether there are general day of week or time of day ef-
fects, or testing the general heat hypothesis that hotter temperatures are generally
related to violence), such field data are not very useful for testing specific hy-
potheses involving the exact shape of the temperature-aggression relationship.
Thus, they cannot precisely test the NAE predicted downturn in aggression at
about 85° versus a simple negative affect model predicting continued increases in
violence through the normal temperature range (i.e., about 98°F).
In sum, the Cohn and Rotton (1997) results appear largely inconsistent with the
NAE model and consistent with a model in which aggressive tendencies continue
to increase throughout the normal temperature range. However, as noted earlier,
275
TEMPERATURE AND AGGRESSION
studies of reported crime rates do not appropriately test the exact shape of the tem-
perature-aggression relationship. More accurate tests of shape may be conducted
in laboratory settings where the time and temperature of the instigation to aggress
are known or in field studies that similarly assess instigation, temperature, and ag-
gression concomitantly (e.g., Kenrick & MacFarlane, 1984; Reifman et al., 1991).
As noted in other works and later in this chapter, we believe that under certain
conditions a downturn at moderately uncomfortable temperatures may well occur.
HOT MONTHS
Murder
Anderson (1989) combined monthly murder rate percentages in the United
States across studies by Brearley (1932), Cohen (1941), Iskrant and Joliet (1968),
Lester (1979), and Michael and Zumpe (1983). An analysis of variance on these
monthly percentages yielded a significant month effect (p < 0101). The peak mur-
der months were July and August, followed closely by December. The high De-
cember rate appears to be due to routine activities involved with Christmas, which
often involve excessive alcohol consumption.
Assault
A combined analysis of monthly assault rate patterns in the United States was
presented in Anderson (1989). The seven data sets came from Aschaffenburg
(1903/1913), Cohen (1941), Dexter (1899), Dodge and Lentzner (1980), Michael
and Zumpe (1983), and Perry and Simpson (1987). The result, displayed in Figure
10.8, was a significant month effect (p < 0101) with a peak in the hot summer
months.
10
..-
t\S
.......
0
Eo-<
;>-.
~
CI)
>-
4-<
0
~
9
8
....
]00
00
<
7
6Jan. Feb. Mar. Apr. May June July Aug. Sep. Oct. Nov. Dec.
FIGURE 10.8 Month
Monthly distribution of assaults. Adapted from Anderson (1989).
276 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
Rape
Monthly rape patterns were also examined in detail by Anderson (1989). The
data sets were taken from Arnir (1971), Aschaffenburg (1903/1913), Hayman,
Lanza, Fuentes, and Algor (1972), Lombroso (1899/1911), Michael and Zumpe
(1983), and Perry and Simpson (1987). There were eight data sets from four coun-
tries (England, France, Germany, and the United States) in these sources. An
ANOVA yielded a significant month effect that paralleled the assault results, with
the most rapes occurring in the hot summer months (p < 0101).
Spouse Abuse
Several other monthly studies are reviewed in Anderson (1989), with the same
general finding that aggression peaks in summer months. One of these studies de-
serves special mention because its findings relate to several alternative hypotheses.
Some critics have suggested that assaults and rapes occur more frequently in the
warmer months because of more frequent contact among potential victims and per-
petrators (i.e., people get out more in the summer) and because of "women's scant-
ier clothing" being provocative. Michael and Zumpe (1986) reasoned that if these
alternative explanations were true, then wife battering should not show the typical
summer increase. One could even argue that wife battering should go down, as the
increase in "getting out" should decrease the time and opportunity for wife batter-
ing. If aggression is directly temperature related, though, the same summer in-
crease observed for other violent crimes should be obtained with spouse abuse.
Michael and Zumpe (1986) examined crisis calls to 23 different women's shel-
ter organizations in five locations during 1981-1984 (at least 2 consecutive years
of data from each location). In each location the annual rhythm maximum oc-
curred in either July or August (p < 2525). In each case the pattern of monthly
abuse means corresponded very closely to monthly temperature means. Michael
and Zumpe (1986) further noted that ". . . the maxima for wife abuse in Atlanta
and Texas occurred about 40 days earlier in the year than those in Oregon and Cal-
ifornia; this difference in timing corresponded (within a few days) to the differ-
ences in the rape maxima in these states, which correlated with the times of the
local temperature maxima" (p. 640). Finally, they noted that the photoperiod max-
ima (the maximum amount of daylight, another alternative to the temperature in-
terpretation) in these locations did not show this 40-day difference.
These data do not rule out the possibility that some type of routine activity ef-
fect also occurs independently of temperature, but they do show that the routine ac-
tivity theory (or women's scantier clothing, or photoperiod shifts) cannot account
for the summer increase in wife battering. Michael and Zumpe imply that it is more
parsimonious to ascribe monthly effects on a variety of aggressive behaviors to the
same causal factor, i.e., the direct heat effect, than to adopt different (and largely
untested) explanations for different forms of affective aggression. We agree.
New Monthly Studies
A number of studies using month as the target time period have provided addi-
tional support. This section begins, however, with a study that at first appears to
TEMPERATURE AND AGGRESSION 277
support the temperature-aggression hypothesis, but is actually irrelevant to it.
Field (1992) reported a sophisticated time series analysis of 40 years of crime data
in England and Wales, primarily as a test of routine activity theory. Field reasoned
that many types of crime, including nonviolent property crimes, should be more
prevalent when people are out of their homes. He further proposed that people
should be out of their homes more often when the weather is nice, relative to nor-
mal weather patterns. He thus adjusted all the raw data for seasonal effects, i.e.,
monthly crime data as well as weather data were seasonally adjusted prior to the
main regression analyses. The main finding was that seasonally adjusted temper-
ature was a significant predictor of seasonally adjusted crime rates for violent
crimes, sexual offenses, burglary, theft, and criminal damage, but not for robbery.
On the whole, these data support routine activity theory, although the lack of an
effect on robbery is problematic, as is the failure of rainfall to predict crime rates.
However, these analyses are irrelevant to the temperature-aggression hypothesis,
as noted by Field. The seasonal adjustment procedure essentially "adjusts out"
most (or all) of the direct heat effects. Monthly crime patterns may well be influ-
enced by both routine activities and more direct heat effects. "It is entirely possi-
ble that temperature affects the level of crime both through a direct [effect] on
aggression and through the mediation of social behaviour" (Field, 1992, p. 348).
Field (1992) also noted that many temperature findings, such as the Michael and
Zumpe (1986) wife-battering study, ". . . (are) obviously not easily explicable in
terms of routine activities" (p. 349).
A similar study, which involved time series analysis (Landau & Fridman,
1993), tested seasonal fluctuations in monthly robbery and homicide rates in Israel
between January 1977 and February 1985. Landau and Fridman found an increase
in robbery rates during winter months (November-February) that they explained
with the routine activity explanation that cost of living increases in the winter
months, which leads to stealing. A seasonal effect was not found for homicide,
however. Homicides were highest in August, with March, May, and December
following in frequency. Landau and Fridman state that the lack of a seasonality ef-
fect on the homicide findings contradicts the temperature-aggression hypothesis
and that the strong August homicide effect is due to an increase in social interac-
tion in that month. As Landau and Fridman note, the lack of an effect of the other
summer months (June and July) on homicide does not support this routine activ-
ity theory. One additional problem with this study is that the total number of
homicides in such a relatively small population is so small that monthly rates are
likely to be quite unstable. Thus, it is probably wise to draw no conclusions from
this one study.
Several studies, including one by Haertzen, Buxton, Covi, and Richards
(1993), have examined monthly variations in aggressive behaviors by prison in-
mates. Haertzen et al. (1993) examined the frequency of "rule infractions" among
prisoners in a Maryland correctional institution from July 1987 to March 1991.
Seasonal as well as month-based analyses were reported. They found a weak cor-
relation between average monthly temperature and relative frequency of rule in-
fractions (r =5,5, N=45, P< 1010. However, because rule infractions include
278 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
many nonaggressive violations (about 50% according to the authors), these data
do not provide a clean test of the temperature-aggression hypothesis. The authors
correctly noted that to study more specific and purely aggressive incidents, a
larger sample of prisons over more years would be needed.
Other (earlier) studies of prisoner aggression suffer similar problems of ambi-
guity of aggression measures, relative infrequency of the target behaviors, and
short time spans. Ganjavi, Schell, and Cachon (1985) studied the effects of several
weather and geomagnetic factors on monthly rates of major violence (e.g., mur-
der, assault) and minor violence (minor assault, suicide, self-injury) in six Cana-
dian prisons from January 1980 to December 1983. The major violence rate was
too rare to be of value. One of the prisons was for psychiatric inmates; its rate of
minor violence was too low to be useful. The biggest problem from the standpoint
of testing the temperature-aggression hypothesis was the inclusion of suicide and
self-injury. Apparently, minor assaults were also quite rare. For these reasons, this
study was not included in Anderson (1989) and is uninformative for this chapter
as well.
Pettigrew (1985) examined monthly rates of simple fighting, aggravated fight-
ing (with a weapon or accomplice), and self-mutilation in five Louisiana prisons
from 1972 to 1982. Aggravated fighting is too rare to provide appropriate tests of
the temperature-aggression hypothesis, and self-mutilation is not an affective ag-
gression act as we have defined it. The simple fighting rate measure did correlate
significantly with monthly average temperature (r = 1414,N=180, p<5)5). Sim-
ilarly, an analysis of variance on the monthly rates of simple fighting showed a
significant month effect, with the July rate being significantly higher (p <5)5) than
every other month except June and August. However, the author pointed out that
many prisoners seemed to get into minor scrapes intentionally during the hot sum-
mer months in order to be "punished" in a way that gets them out of summer field
labor. Self-mutilation rates were similarly affected by temperature and a desire to
avoid field labor. As a result, these data cannot be seen as adequate tests of the
temperature-aggression hypothesis.
Finally, Linkowski, Martin, and DeMaertelaer (1992) reported a study of
monthly rates of violent and nonviolent causes of death in Belgium over a 5-year
period. Although there appeared to be some support for the temperature-aggres-
sion hypothesis-accidental violent death rates correlated positively and signifi-
cantly with temperature for both men and women-ambiguities in the meaning of
the different death rate categories and in the reported data analyses make firm con-
clusions very risky.
SEASONS/QUARTERS
Anderson (1989) presented considerable evidence (via reanalyses, in many
cases) that aggressive behaviors occur more frequently in the summer (or third
quarter) than in other times of the year. Data include a variety of types of violence:
simple and aggravated assaults, uprisings, family disturbances, rape, and murder.
TEMPERATURE AND AGGRESSION 279
40
35
-e- Uprisings
-+- Family Disturbances
--- Assault
+Assault
-e- Rape
-e- Murder
-+- Rape
-e- ViolentCrime
<ii
;2
>- 30
~
~
'0 25
c!<
20
15 Winter Spring Summer Fall
Season
234
Quarter
FIGURE 10.9
derson (1989).
Quarterly and seasonal distribution of aggressive behavior. Adapted from An-
Data come from a variety of researchers (Anderson, 1987; Chang, 1972; Leffing-
well, 1982; Lombroso, 1899/1911; Rotton & Frey, 1985), a variety of countries
(e.g., England, Wales, the United States, Spain, others), and a variety of centuries
(18th, 19th, and 20th). Figure 10.9 graphically summarizes these results.
Two additional studies of the temperature-aggression hypothesis with a sea-
sonal methodology were intentionally left out of Anderson's (1989) review
because of methodological problems. Both provide some support for the temper-
ature-aggression hypothesis. Atlas (1984) studied assault rates in four Florida
prisons. As noted by Anderson and DeNeve (1992), interpretational problems
arise in this study because "some of the prisons had air conditioning in some
places. The frequency of aggressive behaviors varied greatly from institution to
institution. The number of hot, moderate, and cool days at each institution is not
known" (p. 349). In addition, there did not appear to be any attempt to control for
differing numbers of inmates in different months. Nonetheless, a simple reanaly-
sis in which assaults were (a) totaled across the prisons, (b) converted to assault
rates (per day), and (c) subjected to an ANOVA by quarter of the year (with three
replications per quarter) produced a significant quarter effect [F(3, 8) = 9.35, p <
1].1].The summer quarter (July, August, September) had the highest assault rate
(M =3.02, 3.20, 3.76, and 2.84 assaults per day for the first, second, third and
fourth quarters, respectively). The methodological problems are severe, however,
so we advise extreme caution in interpreting these particular results.
The second problematic seasonal study is the Pettigrew (1985) study of Louis-
iana prisons described earlier. The same interpretational problems that existed
with monthly analyses also existed with seasonal ones. For the record though,
280 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
Pettigrew did find a significant increase in simple fighting among inmates in the
summer (p <5)5).4
YEARS
The first published study of the effect of hotter versus cooler years is a small-
scale study by Anderson (1987) of violent crime rates in the United States over a
lO-year period. This study found that hotter years were associated with higher vi-
olent crime rates.
More recently, Anderson, Bushman, and Groom (1997) have improved upon
that earlier study in several ways in two new studies. Study 1 used time series re-
gression procedures to test the effects of yearly average temperature and of age
distribution on violent crime in the United States from 1950 to 1995. As expected,
a significant (p <5)5) positive relation between temperature and violent crime rate
was observed, even after time series, age, and linear year effects were statistically
controlled. On average, each 1°F increase in average temperature produced 3.68
more murders and assaults per lOO,OOOpopulation. Nonviolent crimes were unaf-
fected by average temperature.
Study 2 examined the effects of number of hot days (~ 90°F) on the usual sum-
mer increase in violence. As expected, years with more hot days produced a big-
ger summer increase in violent crime than years with fewer hot days. Nonviolent
crime was unaffected by the number of hot days.
SUMMARY OF TIME PERIOD EFFECTS
All told, time period studies produce an impressive array of support for the
simple hypothesis that hot temperatures directly increase aggressive tendencies.
Although routine activity theory can account for some of the results, it cannot ac-
count for all of them, as its supporters sometimes acknowledge (Field, 1992). The
southern culture of violence position is totally silent on time period effects and
thus cannot account for them.
CONCOMITANT STUDIES
As noted earlier, concomitant studies of the temperature-aggression hypothesis
are actually a specific subcategory of time period studies. In concomitant studies,
temperature and aggression are assessed at the same time. This allows stronger
conclusions to be drawn in most cases because alternative explanations that rely on
"The reader should note that our criticisms of methodology should not be taken as criticisms of the
researchers. We know how very difficult it is to get archival data that meet the methodological re-
quirements of complex research questions. The first author, for instance, has tried unsuccessfully to get
acceptable data on prison aggression rates. We understand, and hope that our readers will too, that re-
searchers in this area must make do with whatever data are available.
TEMPERATURE AND AGGRESSION 281
timing differences are automatically ruled out. An additional methodological ad-
vantage is that concomitant studies can be conducted in the experimental labora-
tory. As noted by Anderson (1989), however, laboratory studies of the heat effect
have, on the whole, produced a very mixed set of results. This section first reviews
the few concomitant studies of the temperature-aggression hypothesis in naturalis-
tic settings, then reviews the laboratory studies via meta-analytic techniques, and
finally presents evidence from our laboratory that resolves many of the questions
raised by prior research.
NATURALISTIC SETTINGS
Two early studies of the temperature-aggression hypothesis in naturalistic set-
tings used horn honking as a dependent measure of aggression. Baron (1976) de-
layed motorists by a confederate whose car sat through a green light. Aggression
was measured by latency to horn honking. The study was conducted when the
temperatures were in the mid-80°F range. Subjects were classified as having air-
conditioned or unair-conditioned cars. Those without air-conditioning presumably
would be uncomfortably warm, and therefore should be more irritated by the con-
federate's blocking of the intersection and should, on average, honk sooner.
Among other things, Baron (1976) found that subjects without air-conditioning
honked their horns sooner than those with air-conditioning. Although this appears
to support the temperature-aggression hypothesis, we advise caution in interpret-
ing the study this way because latency to horn honking may be especially instru-
mental for those without air-conditioning. Thus, it is not clear that the latency
measure assessed affective aggression or a more instrumental intent.
The second study of this type also investigated horn honking in response to a
confederate blocking an intersection (Kenrick & MacFarlane, 1984). These re-
searchers, though, assessed latency to honk, number of honks, and total time spent
honking. The last two measures are not differentially instrumental as a function of
temperature or as a function of air-conditioning and therefore may be seen as mea-
sures of affective aggression. In both cases, once one has honked the horn the in-
strumental role of further honking is negligible. Because all three measures were
highly intercorrelated and yielded the same results, a composite of them was cre-
ated and reported by Kenrick and MacFarlane (1984). This study was conducted
in Phoenix with temperatures ranging from 84 to 108°F. As expected, there was a
significant linear effect of temperature on horn honking (p < 1)1). Furthermore,
this effect was significantly stronger for subjects without air-conditioned cars (r =
5757) than for subjects in cars with air conditioning (r =2,2, Z = 2.54, p<2)2).
Vrij, van der Steen, and Kippellaar (1994) reported a field experiment on the
heat hypothesis conducted in a police station. During a regular shooting exercise
using the Fire Arms Training System, 38 Dutch police officers were randomly as-
signed to the hot (80°F) or comfortable (70°F) condition. The Fire Arms Training
System uses video and laser discs and audio systems to provide officers with real-
istic scenarios to which they respond as police officers. The officers use laser
282 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
"guns" in these scenarios. The scenario used in this study involved the officer re-
sponding to a burglary call at a shed and being confronted by a man holding a
crowbar. The officers were videotaped during this session. After completing the
scenario, each participant went to a comfortable room and completed several
questionnaires measuring negative affect (annoyed and irritated), impression of
the suspect (aggressive or not), impression of how threatening the suspect had
been, and the officer's "tendency to shoot" the suspect. Finally, two measures of
aggressive behavior were obtained based on the video tapes (percentage with
firearm in hand, percentage who "shot" the suspect).
On each dependent variable, police officers in the hot condition displayed more
negative responses than those in the comfortable condition. Hot participants re-
ported more negative affect, a more aggressive impression of the suspect, a more
threatening impression of the suspect, and a greater tendency to shoot the suspect;
all these effects except the last one were statistically reliable. In addition, hot offi-
cers were significantly more likely to draw their weapon (41% vs 15%). Hot par-
ticipants also "shot" more suspects (62%) than did cool participants (45%), but
this difference was not statistically reliable.
On the whole, this work strongly supports the heat hypothesis. In all three nat-
uralistic studies, hot participants behaved more aggressively than comfortable
participants.
LABORATORY EXPERIMENTS
In many laboratory experiments, both temperature and anger have been ma-
nipulated to see if the heat effect is positive under low anger conditions and neg-
ative under high anger conditions. Anderson (1989) provides a detailed narrative
review of most of the laboratory experiments on the temperature-aggression hy-
pothesis. That article summarized the studies as follows: "On the whole, these lab-
oratory studies. . . yield more confusion than understanding. Sometimes hotter
conditions led to increases in aggression; at other times the opposite occurred. . . .
The anger by temperature interaction sometimes occurred and sometimes did not.
When it did occur, it usually took the form of a positive heat effect (increased
heat-increased aggression) in nonangry conditions and a negative heat effect in
angry conditions. In at least one instance the form was opposite" (p. 91).
Two studies were not included in that review. Baron and Lawton (1972) varied
both temperature (hot versus comfortable) and whether a confederate modeled ag-
gressive behavior. There appeared to be a weak negative heat effect in the no
model conditions and a slightly positive effect in the aggressive model conditions,
but neither effect approached significance. More recently, van Goozen, Frijda,
Kindt, and van de Poll (1994) found no heat effect for either high dispositional
anger or low dispositional anger subjects.
There has, nonetheless, been some controversy concerning whether the labora-
tory studies find effects consistent with the NAE model. [See Bell's (1992) com-
TEMPERATURE AND AGGRESSION 283
ment on the Anderson (1989) review article and Anderson and DeNeve's (1992)
response.] A meta-analysis was performed on the entire set of published labora-
tory studies in order to more objectively examine this question (Anderson & An-
derson, 1996). In all, 28 comparisons of comfortable and hot conditions were
found. Overall, there was no hint of consistency (d+ =6060, 95% confidence in-
terval =[-.114, 3434]).
The NAE posits that at low levels of negative affect, further increases in nega-
tive affect will increase aggression, but at high levels of negative affect, further in-
creases in negative affect will produce decreases in aggression because people
will be focused on trying to escape rather than to aggress. Obviously, this is rele-
vant only when escape and aggression motives lead to incompatible behaviors.
The most common way of testing this interaction prediction in the temperature do-
main is to factorially manipulate temperature and anger. Thus, the prediction is
that in high anger conditions hot temperatures should decrease aggression,
whereas in low (or no) anger conditions the opposite should occur. Other context
variables have also been manipulated, variables which either increase or decrease
negative affect or willingness to aggress. For example, attitudinal similarity of the
subject and the eventual target of aggression would decrease the negative affect
whereas attitudinal dissimilarity and insult increase the negative affect. Thus, it is
possible to categorize the experimental context on the basis of whether the non-
temperature factors (e.g., insult, similar attitudes, cooling drink) produce a net in-
crease in the negative affect. In the absence of hard data on the relative
effectiveness of these different factors, we adopted the simple rule of assigning a
+1 to positive factors (such as having similar attitudes) and -1 to negative factors
(such as receiving an insult or having dissimilar attitudes). Conditions in which
the net value of these nontemperature contextual factors was either positive or
zero were placed in a "neutral context" category. Conditions in which the net was
negative were placed in an "extra-negative context" category.
The 13 neutral context effects did yield a positive relation between temperature
(hot vs comfortable), but it was just barely significantly different from zero (d+ =
6464, 95% CI =[.00l, 2626]). That is, hot temperatures appeared to increase ag-
gression in these neutral context conditions, but not with great reliability. In con-
trast, the 15 extra-negative context effects yielded a negative heat effect, but this
negative effect did not approach significance (d+=10,101, 95% CI =[-.333, 1313]).
These results confirm Anderson's (1989) conclusions about the inconsistency
of laboratory results in this domain. However, they also tend to fall in the direc-
tion predicted by NAE. Before leaving this issue, one additional caveat is needed.
One of the studies that supports the NAE model at a behavioral level actually con-
tradicts it in other ways (Palamarek & Rule, 1979). Specifically, these researchers
included measures of escape motives and attributions for their affective state. Re-
sults of these measures contradict the NAE proposed mediating processes in two
ways. First, the desire-to-escape measure yielded no significant effects; thuS moti-
vation to escape was not supported as a valid mediating variable. Second, subjects'
284 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
ratings of the extent to which their mood was caused by the situation paralleled
the aggression choices. Those in the hot angry and the cool nonangry conditions
attributed their mood more to the situation than did the other subjects. Therefore,
using the aggressive behavior data as supportive of the NAE model is problematic
at best and misleading at worst. Furthermore, if these effects are removed from the
meta-analysis, the positive heat effect in the neutral context and the negative heat
effect in the extra-negative context both get even weaker (d+ =5050, 95% CI =
[-.028, 2828], d+ =89,89,95% CI = [-.332, 5454]).
Normally, we put high value on the external validity of laboratory studies (e.g.,
Anderson & Bushman, 1997). However, as noted in earlier works (e.g., Anderson,
1989; Rule & Nesdale, 1976) the laboratory studies of the temperature-aggression
hypothesis may be especially vulnerable to artifactual processes or may not
include all of the interpersonal dynamic processes that normally operate in
temperature-induced aggression in naturalistic settings. In either case, the result
could well be inconsistent findings. Still, we firmly believe that a better under-
standing of heat effects as well as of affective aggression in general requires the
precision and control available only in the laboratory. For this reason, we have
(along with several others in our laboratory) been conducting laboratory research
on heat and cold effects on aggressive behavior and on other aggression-related
affects and cognitions. All of this work has been done in the context of a broad
model of affective aggression, outlined earlier in this chapter. We turn now to a
summary of what we have learned so far.
Temperature and Affect
State Hostility
In several studies we have investigated heat effects on a variety of types of af-
fect. The results are quite consistent across study and paradigm. The most impor-
tant affect from our perspective is what we have labeled "state hostility" (which is
sometimes labeled as "anger"). The state hostility scale (Anderson et aI., 1995)
presents 35 statements (e.g., I feel furious) rated on five-point scales anchored at
"strongly disagree" (1), "disagree" (2), "neither agree nor disagree" (3), "agree"
(4), and "strongly agree" (5). Twelve items represent a lack of hostility (e.g., I feel
polite); these are reverse scored. Across a number of studies we have shown that
uncomfortably hot and uncomfortably cold temperatures increase state hostility
(Anderson et aI., 1996; Anderson et aI., 1995; Anderson, Dorr, Anderson, &
DeNeve, 1997). In general, participants in our studies report the least hostility at
about 75- 78°F. Cold temperatures in the 57-60°F range yield higher state hostil-
ity ratings than comfortable temperatures, and about the same as 93-96°F.
General Negative Affect
Similar increases have been found in general negative affect in hot and cold
conditions, compared to comfortable ones (Anderson et aI., 1996). Specifically,
self-ratings on the general descriptors "upset" and "distressed" were significantly
higher in the hot and cold conditions in that study.
TEMPERATURE AND AGGRESSION 285
Hostile Attitudes
In one study (Anderson et aI., 1996), participants completed the Caprara irri-
tability scale (30 items; Caprara, Cinanni, D'Imperio, Passerini, Renzi, & Trav-
ablia, 1985) and the Velicer attitudes toward violence scale (46 items; Velicer,
Huckel, & Hansen, 1989) in a comfortable (i.e., 75°F) room after doing some cog-
nitive tasks under varying temperature conditions. All items were combined to
form an overall hostile attitudes scale. The Caprara items focus on beliefs about
how one has typically behaved in the past (e.g., when I am irritated, I need to vent
my feelings immediately). The Velicer items focus on beliefs about various ag-
gressive ways of behaving (e.g., university police should beat students if they are
obscene). Despite the fact that all people were in a comfortable room at the time
they completed these items, those who had earlier been in the uncomfortably hot
or cold condition reported significantly more hostile beliefs.
Temperature and Arousal
The effects of hot and cold temperatures on physiological and subjective mea-
sures of arousal were examined (Anderson et aI., 1995, 1996). It was found that
hot temperatures increase heart rate but decrease perceived arousal relative to
comfortable temperatures. Conversely, cold temperatures decrease heart rate but
increase perceived arousal.
Temperature and Primed Cognitions
In one study, Anderson et al. (1996) used a modified Stroop procedure to in-
vestigate the possibility that uncomfortable temperatures might prime aggressive
thoughts. In the modified Stroop procedure, words were flashed on a computer
screen in one of five colors. On each trial the subject's task was to name the color,
not the word. Some of the words were aggression related (e.g., shoot) whereas
others were not (e.g., chant). If uncomfortable temperatures directly prime ag-
gressive thoughts, then naming the color of aggressive words should be relatively
harder for subjects in uncomfortable temperatures. This modified Stroop task was
sensitive to a photo prime manipulation; subjects who had seen and rated gun
photos took relatively longer to name the colors of aggressive words (relative to
control words) than subjects who had seen and rated photos of nature scenes.
However, there was no effect of temperature on the color-naming task.
Temperature and Aggressive Behavior
Two studies relating temperature to aggressive behavior in modified versions
of the Taylor competitive reaction time (CRT) paradigm (Taylor, 1967) have been
completed. In the standard version of this paradigm, subjects believe that they are
competing with another subject on a reaction time task. On each trial the "loser"
receives an electrical shock punishment. The shock intensity and duration are sup-
posedly set by the opponent. Thus, before each trial the subject sets the intensity
and duration of shock to be used against his or her opponent on that trial should
the opponent lose. The actual wins and losses, as well as the trial by trial shock
286 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
settings of the "opponent," are actually controlled by the experimenter. Intensity
and duration as set by the subject are measures of aggression.
Experiment 1
A simple modification was made in the first study. White noise delivered
through headphones was used as the punishment rather than shock. Subjects com-
peted in 25 trials with their opponent. On each trial, participants responded to a
tone by pushing a computer mouse button as soon as possible. Temperatures
ranged from uncomfortably cold (56°F) to uncomfortably hot (96°F). Participants
set the punishment level for their opponent prior to each trial.
In the standard paradigm as well as our first modification of it, the subject can
use any of several motives in setting punishment levels. For instance, if angered
the subject can set high noise levels to hurt the opponent. Because of the trial-by-
trial nature of this task, the subject may refrain from setting high levels in order to
prevent the opponent from responding in kind or the subject may adopt a tit-for-
tat strategy in order to bring down the punishment level set by the opponent. Al-
though the standard paradigm is well established, this ambiguity of subject motive
may make it somewhat less sensitive to subtle effects, such as the heat effect, than
is desired. Nonetheless, we kept this feature of the standard paradigm for our first
experiment.
The pattern of punishments set by the "opponent" was also manipulated. For
half of the subjects, their opponents gave consistently low punishments (selected
intensities of 1-3 on the 0-10 intensity level scale). The other half of the subjects
were initially given low punishment levels by their opponents, but across the 25
trials the intensity levels were increased.
As expected, subjects who received consistently low punishments by their op-
ponent gave significantly lower punishments to their opponents than did those
who received a pattern of increasing punishment. This was particularly true on the
later trials, as shown by the provocation x trial block interaction [F(2, 378) =49.5,
p< 00101] depicted in Figure 10.10.
As expected, provocation also increased state hostility (Ms =2.19 & 1.91),
F(1, 212) = 15.21, p< 00001. In addition, both hot and cold temperatures led to
higher levels of state hostility than did comfortable temperatures, as shown by the
significanteffect of a quadratic temperature term in the model [F(l, 184) =7.30,
p< 1].1].Similarly, escape motives, assessed via a 23 item self-report scale, were
also curvilinearly related to uncomfortable temperatures, with hot and cold sub-
jects reporting heightened motives for escape [F(1, 181) =5.22, p<5]5].
However, temperature had no significant impact on noise intensity settings.
This was true on the first setting, which was made prior to receiving any punish-
ment by the opponent, as well as on each block of eight trials that followed. There
are, of course, a host of reasons for an independent variable to fail to influence a
dependent variable. In the present case, we believed that the standard procedure of
having the subject and opponent set punishments for each other on a trial-by-trial
basis instigated attempts to control the opponent. These attempts may well have
overridden any increase in aggressive tendencies induced by hot or cold tempera-
287
TEMPERATURE AND AGGRESSION
7
BProvoked
Nonprovoked
e
~6
.~
~
-5
<3)
'"
....
0
Z4
31 2
Trial Blocks
3
FIGURE 10.10 Noise intensity means by trial block and provocation.
tures. Alternatively, the low and high provocation manipulations themselves may
have been so clear that they produced floor and ceiling effects, respectively, as
soon as the subject figured out what the opponent was doing. To address these
possibilities, we further modified the reaction time paradigm, and created what we
call the retaliation reaction time (RRT) paradigm.
Experiment 2
In the RRT paradigm there are two sets of trials. One competitor sets punish-
ment levels for the other on all trials in set one; the roles are reversed in set two.
The real subject is always "randomly" assigned to receive punishment on the
"lose" trials in set one, and therefore always sets punishment levels for the oppo-
nent in set two. This procedure allows subjects to retaliate for whatever transgres-
sions they perceive were committed by their opponents during the set one trials,
and to do so without fear of further retaliation by that opponent.
The temperature effect on aggression is likely a fairly subtle one. It may occur
primarily as a brief outburst in reaction to some provocation. If this is true, then
the RRT paradigm should be more sensitive to heat effects than the standard CRT,
especially on the first trial in which the subject gets to set the opponent's punish-
ment level.
This outburst by subjects in the hot and cold conditions should occur primarily
when they have experienced some unwarranted aggression by the opponent, but
not when the opponent's behavior during set one was either very nice (i.e., all low
punishments) or very clearly not nice (i.e., systematically increasing punishment).
To fully test this notion, a third pattern of opponent-set punishments was added.
This set contained exactly the same frequency of each intensity setting as those in
the original "high provocation" conditions, but instead of systematically increasing
288 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
7.50
---0-- Trial 1
6.50 --0-- Trials 2-25
5.50 0, , ~ a..
"0.., ~O~~ "
"~0"~ ~.(Y'~ /"0
'~~
-
Q)
:>
Q)
~
Q)
CZ!
'0
Z4.50
3.50
2.50 50 70 80
Temperature of
FIG U REI O. 1 1 Noise level set by participants in ambiguous provocation conditions as a func-
tion of temperature and trial.
60 90 100
across trials, the various punishment settings were in a random pattern. In other
words, there was no relation between trial number and intensity settings by the
"opponent," thus creating ambiguity for the subject: "What is my opponent trying
to do?" We expected these conditions-provocation with ambiguous intent-to be
maximally sensitive to temperature effects. Here, the hostility state induced by un-
comfortable temperatures can influence interpretation of the opponent's actions
and of one's own state of anger. Once again, this should be especially true on the
first trial for which the subject sets punishment for the opponent.
On subsequent trials the hot and cold subjects may well return to punishments
that are very similar to those set by comfortable subjects. After all, the opponent has
been justly punished. Indeed, for the last block of trials we expected that maximal
aggression would occur by subjects in the moderately uncomfortable conditions.
Hot and cold subjects may give relatively low punishments out of guilt over their
earlier outburst or because that outburst fully satisfied their thirst for revenge or their
desire to "set things right." Comfortable subjects may give relatively low punish-
ments on these last trials because temperature never increased their hostility or their
punishing behavior. Moderately hot and cold subjects, however, may well have in-
creased feelings of hostility that are not expressed in an initial outburst and so they
may continue to deliver moderately high punishments even in this last block.
Our results produced this very complex but meaningful pattern of aggression.
On trial one punishments, there was a significant quadratic temperature x provoca-
tion interaction [F(2, 191) =3.74, p<5]5].5 Further examination of the results
5Temperature was converted to deviation score form prior to analysis so that quadratic and linear
effects could be examined simultaneously and meaningfully.
-_.----
289TEMPERATURE AND AGGRESSION
showed that, as expected, hot and cold subjects gave the most intense punishments
to their opponent, but this curvilinear effect occurred only in the ambiguous provo-
cation condition [F(l, 63) =5.45, p<5]5]. Figure 10.11 displays these results.
Interestingly, this outburst in the ambiguous provocation condition ended quite
rapidly; by trial two it was no longer significant. Indeed, averaging over all subse-
quent trials yielded a downturn in aggression at the temperature extremes, al-
though the downturn did not quite reach the low levels of aggression displayed by
the most comfortable subjects. Interestingly, people in the moderately uncomfort-
able conditions gave the highest punishment levels on these subsequent trials, as
shown by the significant quartic temperature effect [F(1, 61) =3.99, p=5]5], also
displayed in Figure 10.11.
SUMMARY OF CONCOMITANT FINDINGS
Although much work remains to be done to cleanly establish the causal routes
through which various temperature-related effects exert their influence on aggres-
sive behavior, we now have a good start on drawing that road map. Consider first
the laboratory conditions that produced the increase in aggression at hot (and
cold) temperatures. It required a paradigm that could capture an outburst (RRT)
and a provocation level that was maximally ambiguous. In the real world, of
course, an aggressive outburst directed at a target will typically serve as a provo-
cation for that target. As is well known to all (e.g., Bettencourt & Miller, 1995),
provocation plays a huge role in instigating further aggression. Thus, the consis-
tent increases in aggression at hot temperatures, found in geographic region and
time period studies, could well result from this outburst phenomenon.
The quartic pattern found in subsequent trials of our RRT paradigm also pro-
vides some explanatory power for the violent crime data of various types. As many
people have noted, hot days also have somewhat cooler periods. The elevated ag-
gression displayed by our subjects in moderately warm conditions could well con-
tribute to the overall pattern of increased aggression on hot days and in hot regions.
These data do not, however, completely detail the underlying psychological
processes. For instance, even though we believe that escape motives will eventu-
ally prove to play an important role, there still is little evidence that directly sup-
ports the NAE.
KEY THEORETICAL MODELS REEXAMINED
We now examine each of the main theoretical models to see how they fit with
data from each triangulation perspective. We focus on southern culture of vio-
lence, routine activity theory, and our general affective aggression model.
SOUTHERN CULTURE OF VIOLENCE
Although the SCV idea is a fascinating one, temperature data provide little
support at best. The strong version of SCV claims (a) that heat effects are either
290 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
indirect or artifactual and (b) that regional differences in violence are due to the
relative differences in endorsement of values associated with the SCv. There are
several problems with this position. First, the SCV model implies that if regional
differences in "southernness" are controlled in u.S. data, then heat effects must
disappear. Our city violent crime data show, instead, that temperature remains a
significant predictor even after controlling for southernness in various ways. Fur-
thermore, when temperature is statistically controlled, the southernness effect
disappears.
Second, the SCV model is totally silent on time period and concomitant stud-
ies of the temperature-aggression hypothesis because SCV is essentially con-
trolled in both types of studies. The parsimonious explanation for regional
differences in violent crime rates is temperature, not an additional cultural factor.
The third problem is the weak empirical base of general SCV theories as well
as Nisbett's more specific culture of honor model. Note that we believe culture of
honor effects exist as well as heat effects. Nisbett's studies provide hints of this
(e.g., Cohen & Nisbett, 1994; Nisbett, 1993). What is needed to bolster the culture
of honor model are studies showing the following: (a) Representative samples
from high vs low culture of honor regions should differ in aggressive behavior, es-
pecially in public responses to public insults; (b) culture of honor scale scores
should mediate the culture of honor/aggressive behavior relation in "a"; and
(c) culture of honor scale scores should mediate aggressive behavior differences
between "southern" and nonsouthern participants in laboratory studies. To date,
none of the culture of honor laboratory studies has measured aggressive behavior.
Cohen, Nisbett, Bowdle, and Schwarz (1995) measured the extent to which
insulted and noninsulted southerners and northerners gave way to a large confed-
erate in a hallway "chicken game" as well as the perceived firmness of the partic-
ipants' handshakes. Insulted southerners tended to step aside later and to give
firmer handshakes than northerners, but these measures relate to domineeringness.
They do not meet standard definitions of aggression, and thus do not directly test
the theory.
Similarly, there has not yet been a clear statement of what values and attitudes
constitute the culture of honor, so there has not been an individual difference mea-
sure developed to assess culture of honor. Such a measure would be of immense
value to the study of culture of honor and to violence more generally. It is hoped
that future work will address these issues.
Finally, there are alternative explanations of the southern culture of violence
that differ from the sociological theories and Nisbett's culture of honor model.
One major historian (Wyatt-Brown, 1986) suggests that the violent aspects arose
from the particular circumstances surrounding the institution of slavery and the
need to maintain control. Other scholars (e.g., Pennebaker, Rime, & Blankenship,
1996) have shown that climate plays a major role in the development of emotional
expressiveness differences between cultures, a suggestion that fits well with our
position that hot temperatures may contribute to the development of cultures of vi-
olence (Anderson & Anderson, 1996). In sum, although both temperature and cul-
TEMPERATURE AND AGGRESSION 291
tural explanations are likely to playa role in the violence in modem society, the
southern culture of violence notion is not a viable alternative to the temperature-
aggression hypothesis. It may best be thought of as a "partner."
ROUTINE ACTIVITY THEORY
The strong version of routine activity theory also states that heat effects are ei-
ther indirect, working through the kinds of activities that people normally do in
various temperatures, or artifactual. If RAT is the only factor underlying observed
heat effects in time period studies, then the following predictions must hold:
(a) Routine activities during hotter periods of time must differ in ways that in-
crease aggressive behavior; (b) for different types of aggressive behavior, which
differ in the kinds of routines that increase or decrease them, the observed time
period heat effects should differ in corresponding ways (e.g., wife battering
should not increase in the summer in the same way that assault does); (c) any re-
gional heat effects should be related to regional differences in routine activities;
and (d) there should be no heat effect in laboratory experiments or in field studies
where routine activities are controlled. At this point in time, there is some limited
support for prediction (a). Prediction (b) is contradicted by the time period stud-
ies in that all measured violent behaviors appear to show much the same time pe-
riod heat effect. To our knowledge, prediction (c) has never been examined, i.e.,
no studies appear to link various high violent crime routine activities to geo-
graphic regions. Indeed, one of the major shortcomings of RAT is that too little is
known about what routine activities promote what kinds of aggressive behaviors,
about the temporal and spatial distribution of these routine activities, and about
the intra- and interpersonal dynamics that link these routine activities to aggres-
sive behaviors. It is an area ripe for more research. Finally, concerning prediction
(d), concomitant studies and even some of the day-unit time period studies cannot
be handled by RAT. For instance, the Reifman et al. (1991) study of batters hit by
pitched baseballs cannot be explained by RAT.
A different version of RAT, one which accepts that heat directly affects ag-
gressive tendencies, can reasonably posit that the relationship between heat and
aggression might differ as a function of routine activities. Specifically, some "rou-
tines" reduce the opportunity for aggression. In such cases we would expect the
link between heat and aggression to be attenuated. The Cohn and Rotton (1997)
study provides some evidence for this version of RAT.
GENERAL AFFECTIVE AGGRESSION MODEL
This theory states that heat effects are both direct and indirect. It predicts that,
all else being equal, hotter regions should have higher rates of aggressive behav-
ior for those aggressive behaviors with a large affective/impulsive component.
Region studies strongly support this prediction, even when all sorts of statistical
controls are imposed. The theory also predicts that hotter regions will, on average,
292 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
develop more violent cultures. That is, people who grow up in hotter regions
should have more positive attitudes and values toward aggressive behaviors, at
least in certain kinds of circumstances. There is some support for this from the
culture of honor literature (e.g., Cohen & Nisbett, 1994; Nisbett, 1993) and some
from cross-cultural studies (e.g., Pennebaker et aI., 1996; Robbins et al., 1972).
This theory also predicts direct heat effects in time period studies. Research
consistently shows higher rates of aggression during hotter time periods in daily,
monthly, quarterly, and yearly time period studies.
The general model makes a host of predictions concerning heat effects in con-
comitant studies, both in laboratory as well as in field settings. Research in the
laboratory has resolved many of the inconsistencies found in earlier laboratory
work and shows that hot temperatures do increase aggressive behavior in the lab-
oratory. The few concomitant field studies of aggressive behavior also support the
theory. In addition, the general model makes a host of predictions concerning pos-
sible routes through which hot temperatures (and cold) may influence aggressive
behavior. We have shown that both affective and arousal routes may carry the heat
effect from discomfort to (eventually) aggressive behavior. However, consider-
ably more work must be done to test this model.
PROMISING DIRECTIONS
There are many promising directions for new research in this domain. We elab-
orate on some that are particularly exciting to us.
Development of a culture of honor scale would enable the testing of many hy-
potheses, both about cultures of honor as well as about individual differences in
violence proneness. A culture of honor scale should be developed independently
of indicators of aggressiveness so that the hypothesis that a culture of honor pre-
dicts aggressiveness can be fairly tested. Once developed, such a scale would also
be useful in examining child development issues. How are these attitudes and val-
ues passed on? Are there alternative attitudes and values that could be substituted
and that would be more beneficial to the individual and society?
More work is needed on the biological aspects. How can we relate biological
factors (e.g., hormonal, neuronal) to heat effects on affective state, cognitions, at-
tention, and arousal? Are some people more susceptible to such heat effects, per-
haps because of their biological responsiveness to temperature?
As noted earlier, more work is needed on the specific ways that routine activi-
ties may be tied to heat effects on the one side and to violent crime on the other.
Such work may be particularly important in its implications for crime control
efforts.
Finally, the integrated model itself presents many research questions that, to
date, do not have clear answers. Many of these questions involve temperature, of
course. How does an uncomfortably hot temperature influence the interpretation
---
293
TEMPERATURE AND AGGRESSION
of ambiguously aggressive stimuli? Does it influence attributions people make for
aversive events? How do individual differences interact with temperature and
other aggression-related factors such as provocation?
APPLICATIONS
Throughout this chapter we have concentrated on the more purely scientific as-
pects of the temperature-aggression hypothesis. What are the underlying causes?
Although the potential for application may seem obvious to many readers, we feel
it important to at least briefly mention a few.
The integrated model itself suggests a host of ways that society can intervene
to reduce unwarranted aggression. Some of these are not new ideas arising solely
from temperature research, but instead can be derived from many contemporary
research programs that are congruent with our own. For instance, by reducing
children's exposure to aggressive material, especially to aggression that appears to
be rewarded, we might be able to produce children (and later, adults) who are less
prone to making hostile attributions for observed events (e.g., Dodge & Crick,
1990; Huesmann, Eron, Lefkowitz, & Walder, 1984). The role of temperature in
the broad model also has implications. Many instances of violence in society
begin as small disputes that escalate. If people were more generally aware of how
hot temperatures can lead to attribution and interpretation biases, perhaps they
could counteract them. Public service announcements about the need to "cool"
one's temper as well as one's temperature during hot periods of time might pro-
duce a reduction in heat-induced aggression.
The various field study results themselves may be useful in some settings. For
instance, knowing when and where violence is likely to erupt (e.g., hot days and
nights, near bars) may be used to change police presence patterns in productive
ways.
Finally, most people spend much of their lives in "built" environments. By
making them more comfortable, we may be able to reduce unwarranted aggres-
sion. The most obvious places (to us, at least) are schools, prisons, homes, and the
workplace.
REFERENCES
Amir, M. (1971). Patterns in forcible rape. Chicago: University of Chicago Press.
Anderson, C. A. (1987). Temperature and aggression: Effects on quarterly, yearly, and city rates of vi-
olent and nonviolent crime. Journal of Personality and Social Psychology, 52, 1161-1173.
Anderson, C. A. (1989). Temperature and aggression: Ubiquitous effects of heat on occurrence of
human violence. Psychological Bulletin, 106, 74-96.
Anderson, C. A., & Anderson, D. C. (1984). Ambient temperature and violent crime: Tests of the lin-
ear and curvilinear hypotheses, Journal of Personality and Social Psychology, 46, 91-97.
294 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
Anderson, C. A., Anderson, K. B., & Deuser, W. E. (1996). Examining an affective aggression frame-
work: Weapon and temperature effects on aggressive thoughts, affect, and attitudes. Personality
and Social Psychology Bulletin, 22, 366-376.
Anderson, C. A., & Bushman, B. J. (1997). External validity of "trivial" experiments: The case of lab-
oratory aggression. Review of General Psychology, I, 19-4].
Anderson, C. A., & DeNeve, K. M. (1992). Temperature, aggression, and the negative affect escape
model. Psychological Bulletin, Ill, 347-351.
Anderson, C. A., Bushman, B. J., & Groom, R. W. (1997). Hot years and serious and deadly assault: Em-
pirical tests of the heat hypothesis. Journal of Personality and Social Psychology, 73, 1213-1223.
Anderson, C. A, Deuser, W. E., DeNeve, K. (1995). Hot temperatures, hostile affect, hostile cognition,
and arousal: Tests of a general model of affective aggression. Personality and Social Psychology
Bulletin, 21, 434-448.
Anderson, C. A., Dorr, N., Anderson, K. B., & DeNeve, K. M. (under review). Temperature-
aggression inconsistencies res01ved: Laboratory evidence of multip1e functiona1 re1ations between
uncomfortable temperatures and aggressive behavior.
Anderson, C. A., Miller, R. S., Riger, A. L., Dill, J. C., & Sedikides, C. (1994). Behavioral and char-
acterologica1 attributional styles as predictors of depression and 10neliness: Review, refinement,
and test. Journal of Personality and Social Psychology, 66, 549-558.
Anderson, K. B., & Anderson, C. A. (1996). Laboratory effects of hot temperatures on aggressive be-
havior: A meta-analysis. Unpublished manuscript.
Aschaffenburg, G. (1903/1913). Crime and its repression. Boston: Little, Brown.
Atlas, R. (1984). Vi01ence in prison: Environmental influences. Environment and Behavior, 16(3),
275-306.
Bailey, W. C. (1976). Some further evidence on homicide and a regiona1 culture of violence. Omega,
7, ]45-170.
Bandura, A. (1973). Aggression: A social learning theory analysis. Englewood Cliffs, NJ: Prentice Hall.
Baron, R. A. (1972). Aggression as a function of ambient temperatures and prior anger arousal. Jour-
nal of Personality and Social Psychology, 21, ]83-]89.
Baron, R. A. (1976). The reduction of human aggression: A fie1d study of the influence of incompati-
ble reactions. Journal of Applied Social Psychology, 6, 260-274.
Baron, R. A., & Lawton, S. F. (1972). Environmenta1 influences on aggression: The faci1itation of
modeling effects by high ambient temperatures. Psychonomic Science, 26, 80-82.
Bazett, H. C. (1927). Physi010gical responses to heat. Physiological Reviews, 7, 53]-599.
Bell, P. A. (1980). Effects of heat, noise, and provocation on retaliatory evaluative behavior. Journal of
Social Psychology, lID, 97-100.
Bell, P. A. (1992). In defense of the negative affect escape model of heat and aggression. Psychologi-
cal Bulletin, Ill, 342-346.
Bell, P. A., & Baron, R. A. (1976). Aggression and heat: The mediating role of negative affect. Jour-
nal of Applied Social Psychology, 6, 18-30.
Bell, P. A., & Baron, R. A. (1977). Aggression and ambient temperature: The faci1itating and inhibit-
ing effects of hot and c01d environments. Bulletin of the Psychonomic Society, 9, 443-445.
Berkowitz, L. (1984). Some effects of thoughts on anti- and prosocial influences of media events: A
cognitive-neoassociation analysis. Psychological Bulletin, 95,410-427.
Berkowitz, L. (1993). Aggression: Its causes, consequences, and control. New York: McGraw Hi11.
Bettencourt, B. A., & Miller, N. (1995). Sex differences in aggression as a function of provocation: A
meta-analysis. Manuscript submittted for publication.
Blanchard, D. c., & Blanchard, R. J. (1984). Affect and aggression: An animal mode] applied to
human behavior. In R. J. Blanchard & D. C. Blanchard (Eds.), Advances in the study of aggression
(Vol. 1, pp. ]-62). Orlando, FL: Academic Press.
Brearley, H. C. (1932). Homicide in the United States. Montclair, NJ: Patterson-Smith.
Buchanan, C. M., Eccles, J. S., & Becker, J. B. (1992). Are adolescents the victims of raging hor-
mones: Evidence for activational effects of hormones on moods and behavior at ad01escence. Psy-
chological Bulletin, Ill, 62-107.
295
TEMPERATURE AND AGGRESSION
Bushman, B. J., Anderson, K. B., & Anderson, C. A. (1998). Assaults and temperature in Minneapo-
lis reexamined. Manuscript in preparation.
Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the multitrait-
multimethod matrix. Psychological Bulletin, 56,81-105.
Caprara, G. V., Cinanni, V., D'Imperio, G., Passerini, S., Renzi, P., & Travablia, G. (1985). Indicators
of impulsive aggression: Present status of research on irritability and emotional susceptibility
scales. Personality and Individual Differences, 6, 665-674.
Carlsmith, J. M., & Anderson, C. A. (1979). Ambient temperature and the occurrence of collective vi-
olence: A new analysis. Journal of Personality and Social Psychology, 37, 337-344.
Cash, W. J. (1941). The mind of the South. New York: Knopf.
Chang, D. H. (1972). Environmental influences on criminal activity in Korea. Criminology, 10, 338-
352.
Church, A. T., & Burke, P. J. (1994). Exploratory and confirmatory tests of the Big Five and Tellegen's
three- and four-dimensional models. Journal of Personality and Social Psychology, 66, 93-114.
Cohen, D., & Nisbett, R. E. (1994). Self-protection and the culture of honor: Explaining southern vio-
lence. Personality and Social Psychology Bulletin, 20, 551-567.
Cohen, D., Nisbett, R. E., Bowdle, B., & Schwarz, N. (1995). Insult, aggression, and the southern cul-
ture of honor: An "experimental ethnography." Unpublished manuscript, University of Michigan.
Cohen, J. (1941). The geography of crime. Annals of the American Academy of Political and Social
Science, 217, 29-37.
Cohen, L. E., & Felson, M. (1979). Social change and crime rate trends: A routine activity approach.
American Sociological Review, 44, 588-608.
Cohn, E. G. (1990). Weather and crime. British Journal of Criminology, 30, 51-64.
Cohn, E. G. (1993). The prediction of police calls for service: The influence of weather and temporal
variables on rape and domestic violence. Environmental Psychology, 13,71-83.
Cohn, E. G., & Rotton, J. (1997). Assault as a function of time and temperature: A moderator-variable
time-series analysis. Journal of Personality and Social Psychology, 72, 1322-1334.
Cotton, J. L. (1986). Ambient temperature and violent crime. Journal of Applied Social Psychology,
16, 786-801.
Crano, w., & Brewer, M. (1973). Principles of research in social psychology. New York: McGraw-
Hill.
deFronzo, J. (1984). Climate and crime. Environment and Behavior, 16, 185-210.
Dexter, E. G. (1899). Conduct and the weather. Psychological Monographs, 11( 10), I-1m.
Dill, J., & Anderson, C. A. (1995). Effects of justified and unjustified frustration on aggression. Ag-
gressive Behavior, 21, 359-369.
Dodge, K. A., & Crick, N. R. (1990). Social information-processing bases of aggressive behavior in
children. Personality and Social Psychology Bulletin, 16, 8-22.
Dodge, R. W., & Lentzner, H. R. (1980). Crime and seasonality: National crime survey report. U.S.
Department of Justice (#SD-NCS-N-15).
Erlanger, H. (1975). Is there a 'subculture of violence' in the South? Journal of Criminal Law and
Criminology, 66, 483-490.
Feigl, H. (1958). The mental and the physical. In H. Feigl, M. Scriven, & G. Maxwell (Eds.), Min-
nesota studies in the philosophy of science (Vol. II). Minneapolis: University of Minnesota Press.
Felson, R. B. (1993). Motives for sexual coercion. In Felson & Tedeschi (Eds.), Aggression and vio-
lence: Social interactionist perspectives (pp. 233-253). Washington, DC: American Psychological
Association.
Field, S. (1992). The effect of temperature on crime. British Journal of Criminology, 32, 340-351.
Ganjavi, 0., Schell, B., & Cachon, J. (1985). Geophysical variables and behavior. XXIX. Impact of at-
mospheric conditions on occurrences of individual violence among Canadian penitentiary popula-
tions. Perceptual and Motor Skills, 61, 259-275.
Gastil, R. D. (1971). Homicide and regional culture of violence. American Sociological Review, 36,
412-427.
Geen, R. G. (1990). Human aggression. Pacific Grove, CA: Brooks Cole.
296 CRAIG A. ANDERSON AND KATHRYN B. ANDERSON
Greenwald, A G., Pratkanis, A. R., Leippe, M. R., & Baumgardner, M. H. (1986). Under what condi-
tions does theory obstruct research progress? Psychological Review, 93, 216-229.
Hackney, S. (1970). Southern violence. In H. Graham & T. Gurr (Eds.), Violence in America: Histori-
cal and comparative perspectives (pp. 505-527). New York: Bantam.
Haertzen, C., Buxton, K., Covi, L., & Richards, H. (1993). Seasonal changes in rule infractions among
prisoners: A preliminary test of the temperature-aggression hypothesis. Psychological Reports, 72,
195-200.
Hardy, J. D. (1961). Physiology of temperature regulation. Physiological Reviews, 41, 521-605.
Harries, K. D., & Stadler, S. J. (1988). Heat and violence: New findings from Dallas field data, 1980-
1981. Journal of Applied Social Psychology, 18, 129-138.
Hayman, C. R., Lanza, C., Fuentes, R., & Algor, K. (1972). Rape in the District of Columbia. Ameri-
can Journal of Obstetrics and Gynecology, 113,91-97.
Huesmann, L. R. (1986). Psychological processes promoting the relation between exposure to media
violence and aggressive behavior by the viewer. Journal of Social Issues, 42, 125-139.
Huesmann, L. R., Eron, L. D., Lefkowitz, M. M. & Walder, L. O. (1984). Stability of aggression over
time and generations. Developmental Psychology, 20, 746-775.
Iampietro, P. F., Melton, C. E., Higgins, E. A., Vaughan, J. A., Hoffman, S. M., Funkhouser, G. E., &
Saldivar, J. T. (1972). High temperature and performance in a flight task simulator. Aerospace Med-
icine, 43, 1215-1218.
Kenrick, D. T., & MacFarlane, S. W. (1984). Ambient temperature and horn-honking: A field study of
the heat/aggression relationship. Environment and Behavior, 18, 179-191.
Kobrick, J. L., & Fine, B. J. (1983). Climate and human performance. In D. J. Oborne & M. M.
Gruneberg (Eds.), The physical environment at work (pp. 69-107). Chichester, England: Wiley &
Sons.
Kobrick, J. L., & Johnson, R. F. (1991). Effects of hot and cold environments on military performance.
In R. Gal & A. D. Mangelsdorff (Eds.), Handbook of military psychology (pp. 215-232). Chich-
ester, England: Wiley & Sons.
Landau, S. F., & Fridman, D. (1993). The seasonality of violent crime: The case of robbery and homi-
cide in Israel. Journal of Research in Crime and Delinquency, 30, 163-191.
LeBeau, J. L., & Langworthy, R. H. (1986). The linkages between routine activities, weather, and calls
for police services. Journal of Police Science Administration, 14, 137-145.
LeBlanc, J. (1975). Man in the cold. Springfield, IL: Charles C. Thomas.
LeDoux, J. E. (1993). Emotional networks in the brain. In M. Lewis & J. M. Haviland (Eds.), Hand-
book of emotions (pp. 109-118). New York: Guilford Press.
Leffingwell, A. (1892). Illegitimacy and the influence of the seasons upon conduct. New York: Scribners.
Linkowski, P,. Martin, F., & DeMaertelaer, V. (1992). Effect of some climatic factors on violent and
non-violent suicides in Belgium. Journal of Affective Disorders, 25, 161-166.
Lombroso, C. (1899/1911). Crime: 1ts causes and remedies. Boston: Little, Brown.
Marsh, H. W., Balla, J. R., & McDonald, R. P. (1988). Goodness-of-fit indexes in confirmatory factor
analysis: The effect of sample size. Psychological Bulletin, 103, 391-410.
Marsh, H. W., & Hocevar, D. (1985). Application of confirmatory factor analysis to the study of self-
concept: First and higher order factor models and their invariance across groups. Psychological
Bulletin, 97, 562-582.
McGrath, J. E., Martin, J., & Kukla, R. A. (1982). Judgment calls in research. Beverly Hills, CA: Sage.
Michael, R. P., & Zumpe, D. (1983). Annual rhythms in human violence and sexual aggression in the
United <