Crimes of Opportunity or Crimes of Emotion? Testing Two Explanations of Seasonal Change in Crime*

While past research has suggested possible seasonal trends in crime rates, this study employs a novel methodology that directly models these changes and predicts them with explanatory variables. Using a nonlinear latent curve model, seasonal fluctuations in crime rates are modeled for a large number of communities in the U.S. over a three-year period with a focus on testing the theoretical predictions of two key explanations for seasonal changes in crime rates: the temperature/aggression and routine activities theories. Using data from 8,460 police units in the U.S. over the 1990 to 1992 period, we found that property crime rates are primarily driven by pleasant weather, consistent with the routine activities theory. Violent crime exhibited evidence in support of both theories. Sociologists have long had an interest in how seasonal climatic changes may interact with social structures to influence the behavior patterns of individuals. Early work in this area includes Durkheim’s classic studies of seasonal differences in suicide rates (Durkheim 1952:107-18), a topic that has seen renewed attention in recent decades (Bollen 1983; Warren 1983). Seasonality in birth and death rates has also been investigated (Land & Cantor 1983), as has the linkage between seasonal changes in testosterone production and sexual activity, with mixed success (Smolensky et al. 1981; Udry & Morris 1967). This * We thank the members of the Carolina Structural Equation Modeling (CSEM) group for their comments and suggestions on earlier drafts of this article. The research on which this article is based was supported in part by a grant from NIDA (DA13148) and also a fellowship (F32 DA06062) awarded to the second author. Direct correspondence to John R. Hipp, UNC at Chapel Hill, Department of Sociology, Hamilton Hall, CB #3210, Chapel Hill, NC 27599. 1334 / Social Forces 82:4, June 2004 article focuses on one of the most robust and socially problematic seasonal trends in behavior, namely seasonal changes in crime rates. The notion that seasonal weather patterns affect crime rates was suggested at least as early as the nineteenth century, when Adolph Quetelet observed such a relationship with data from France (Quetelet [1842] 1969). More recent descriptive evidence from the U.S. also suggests that there are seasonal differences for at least some types of crime (Dodge 1980, 1988). This article addresses the question of why such a relationship should exist. While various explanations for seasonal changes in crime have been offered, rarely have these theories been empirically contrasted using methodological tools that directly and dynamically model seasonal changes in crime. Two dominant theories for explaining seasonal oscillations in crime rates are the temperature/aggression theory and the routine activities theory. While both theories suggest that temperature is related to crime rates, they propose different causal mechanisms for bringing about this relationship. In the more psychologically based temperature/aggression theory first proposed by Quetelet, uncomfortably hot temperatures increase the frustration of individuals, leading to aggressive behavior (Quetelet [1842] 1969). Thus one would expect violent crime to reach its highest levels during the hot days of summer, while the more calculating nature of property crime should be unaffected by heat and thus show no seasonal oscillations. In fact, in his own analyses, Quetelet noted that property crime in France in the late 1820s actually peaked during the winter, which he explained as a response by individuals to a shortage of basic needs. The more recent routine activities theory employs a social explanation, focusing explicitly on the changing activity patterns of individuals to explain seasonal oscillations in all types of crime (Cohen & Felson 1979). In this theory, pleasant temperatures encourage individuals to spend more time outside the home, increasing opportunities for criminal victimization. While much empirical work has looked at each of these theories separately, rarely have studies been conducted with the express purpose of comparing the two. As a result, advocates of both approaches often simply demonstrate a linear relationship between temperature and crime. Such a relationship is consistent with both theories and thus does not provide a basis for comparison. Our approach to comparing these two theories is both theoretical and methodological. By exploring the mechanisms proposed by each theory, we determine how they make subtly different seasonal crime predictions. This allows us to form hypotheses from each of the theories that differ in their implications. While we do not suggest that these theories are necessarily mutually exclusive — and indeed it is possible that both are at work in some instances — our approach allows us to evaluate the predictions of each theory with empirically observed seasonal crime patterns. Testing these hypotheses requires a methodological approach that will allow us to directly model seasonal fluctuations in crime rates, something that is Crimes of Opportunity or Crimes of Emotion? / 1335 notably lacking in previous research on this topic. The model we propose is a variant of the latent curve model (LCM) of Meredith and Tisak (1990) (see also McArdle 1988; McArdle & Epstein 1987; Muthen 1991). The LCM involves the estimation of trajectories of change that may vary over the units of study. While these trajectories are typically modeled with linear, quadratic, or higherorder polynomial functions, recent extensions of the LCM permit the estimation of trajectories that are nonlinear functions of time (Browne 1993; Browne & du Toit 1991; Cudeck 1996; du Toit & Cudeck 2001). While these extensions to the LCM allow the possibility of modeling oscillating functions over time, this strategy has rarely been exploited in applied research. Using the LCM framework allows us to explicitly model the nonlinear cycle in crime that takes place over the seasons. One important result is that we can also predict variation in these seasonal changes over communities, allowing us to test the predictions of these two theories. Further, while many past studies have focused on only one or two communities, our approach facilitates comparisons over many communities — in our case a sample of 8,460 police units in the U.S. Thus, our article makes four contributions. First, while past work has only viewed seasonal crime patterns in a descriptive manner, using structural equation modeling allows us to statistically test for the presence of seasonal crime patterns. Second, we construct a unique data set that combines crime rates in cities with nearby climate data. Third, we explicitly compare the two theories, and by specifying the implications of the causal mechanisms for each are able to derive testable hypotheses. Finally, we meld these theoretical derivations with a methodology uniquely suited to testing the hypotheses, allowing us to compare crime rates between cities at the same time that we model seasonal crime patterns within cities. The article takes the following course. We first provide an overview of the two theories and then deduce a set of hypotheses on seasonal crime trends that differ between the two theories. Following that, we discuss the limitations of the methodological strategies used in past research on this topic and show how our approach addresses these limitations. We also note that, over any evaluation period, seasonal fluctuations in crime may be overlaid on both stable intercommunity differences and longer-term increases or decreases in crime rates, and we use the social disorganization perspective to help explain these differences. After describing our data and measures, we present our analytical model for capturing seasonal oscillations in crime rates and show how it allows us to evaluate the role of key predictors. In addition to the statistically powerful results obtained by analyzing a nationally representative sample of police units, we also explore specific case studies of crime rates for communities in particular states. We conclude with a summary of the results and their implications for future research. 1336 / Social Forces 82:4, June 2004 Temperature Aggression Theory The earliest explanation for the observed regularity of seasonal crime oscillations was the temperature/aggression (T/A) theory. As initially formulated by Quetelet ([1842] 1969], this theory suggests that hot temperatures lead to greater discomfort, which in turn gives rise to more aggressive behavior. Because the focus is on the psychological level of discomfort, some investigators have suggested that both hot and cold temperatures should lead to greater discomfort and hence aggression (for a nice review, see Anderson 1989). This has been generalized to other forms of discomfort, such as crowding (Calhoun 1962), and laboratory studies have even looked at the relationship between noxious smells and aggressive behavior (Berkowitz 2000). However, incontrovertible empirical evidence for the T/A theory has been hard to come by. For instance, laboratory experiments have not fared particularly well. Scholars have attributed these null results to the possibility that entering a laboratory with an inordinately warm temperature might alert participants to the focus of the study and lead them to alter their behavior (Anderson 1989; Anderson & Bushman 1997). These subjects might then attribute provocative behavior by a confederate “to the heat” and therefore show an even more restrained response than would otherwise be the case. As a result, much of the evidence for the temperature/aggression theory consists simply of studies showing correlations between temperature and crime rates (Anderson 1989, 2001). For instance, in support of the T/A theory, studies using daily data from Chicago and Houston (Anderson & Anderson 1984) and Des Moines and Indianapolis (Cotton 1986) found evidence of a linear trend between temperature and violent crime but no relationship between temperature and property crime. However, focusing on particular cities limits the generalizability of the results of such studies; in addition, Cohn (1990a) points to other studies that have found contradictory evidence regarding the relationship between temperature and homicide rates. Fewer studies have looked at a large number of communities at a given time, also showing inconsistent findings. DeFronzo (1984) looked at 142 standard metropolitan statistical areas (SMSAs) in the U.S. with populations greater than 200,000 in 1970. Most notable about this study was that it found that after adding demographic controls, the number of hot days (temperature greater than 90 degrees Fahrenheit) experienced by an SMSA had a positive effect only on homicide and burglary rates. While the finding for homicide is consistent with the T/A theory, the lack of findings for other types of violent crime, along with the finding for burglary, are at odds with the theory’s predictions. Proponents of the T/A approach have countered that the large number of control variables employed by Cohn’s study may have introduced collinearity problems, making the estimates unstable. Additionally, the focus on only large SMSAs limits the generalizability of the study and raises possible selection issues (Berk 1983; Crimes of Opportunity or Crimes of Emotion? / 1337 Heckman 1979). A second study looking at 260 SMSAs in the U.S. in 1980 also found that the number of hot days had a positive effect on homicide, even with other controls in the model (Rotton 1993). In sum, while the results of these studies are sometimes consistent with T/A theory, they are too often based on simple tests of a linear relationship between violent crime and temperature. Rountine Activities Theory In contrast to T/A theory, routine activities (RA) theory suggests that seasonal oscillations in crime rates are not due to increased aggression on the part of individuals, but rather to altered behavioral patterns (Cohen & Felson 1979). For a crime to occur in this model, there must be a concurrence in space and time of three elements: (1) an offender, (2) a suitable target, and (3) the absence of guardians (Cohen & Felson 1979). Temperature can play an important role in determining whether these conditions are met. For instance, when it is very cold, individuals are more likely to stay at home, reducing the number of suitable targets, and as a result burglary becomes much more difficult (since people are in the home) as do such crimes as assault and robbery (as individuals are not out and about providing potential targets). However, it is important to note that RA theory does not focus exclusively on temperature, viewing it as only one of many factors that change the normal behavior patterns of individuals in a community. In part for this reason, studies attempting to evaluate RA theory often do not explicitly address temperature effects. In their initial test of the theory, Cohen and Felson (1979) noted how structural changes in female labor force participation affected opportunities for crime, asserting that more women entering the labor force moved them outside the home and increased the risk of criminal victimization. Their model then used changes in the percentage of women in the labor force to explain changes in crime rates in the entire U.S. (Cohen, Felson & Land 1980). Nonetheless, RA theory has strong implications for the seasonal oscillations observed in crime rates due to the hypothesized change in social patterns. For instance, Cohn (1990a) points out that vacations occur more often during warmer weather, leaving homes exposed to burglary and putting individuals out and about in environments and hence at risk of criminal victimization. In general, a greater amount of time spent outside the home during nicer weather should lead to more opportunities for criminal activity. This implies the opposite effect for cold weather, and evidence of this comes from a study of SMSAs finding that the number of cold days in a month has a significant negative effect on various crime types (DeFronzo 1984). 1338 / Social Forces 82:4, June 2004 Contrasting the Theories It is notable that while some studies attribute seasonal changes in crime to increased aggression (e.g., Anderson & Anderson 1984), a study of monthly crime data for England and Wales attributed a similar finding to more time spent outside the home during nicer weather (Field 1992). These different conclusions suggest that the climatic patterns in a community may be important for distinguishing which of these two theories is at work: the fact that a seasonal effect is found in Britain where the summers are fairly mild lends support to RA theory, while the presence of a seasonal effect in an area with hot summers might suggest the T/A theory. This difference points out a possible way to compare these two theories, especially when evaluated using a large sample of communities with considerable variation in climate patterns. While each of these theories suggests a positive relationship between seasonal temperature changes and oscillations in crime rates, a close inspection of the two approaches reveals that they have subtle, but key, differences in their predictions. First, while the T/A approach suggests that hotter temperatures in the summer will lead to greater aggression and hence an increase in violent crime rates, this aggression is not hypothesized to affect rates of property crime. To the extent that property crime involves calculating behavior and not aggression, it should not be affected by seasonal temperature fluctuations. In contrast, routine activities theory suggests that altered behavior patterns will result in seasonal relationships for both property and violent crime. In particular, favorable weather makes individuals more likely to leave home. This may provide additional tempting targets that will particularly affect property crime rates. This yields a key distinction between these two theories: Hypothesis 1: The routine activities theory predicts that there will be a positive seasonal effect for the property crime rate, while the temperature/aggression theory predicts that there will not be a seasonal effect for property crime rates. And while each of these theories predicts a seasonal effect for violent crime rates, the mechanisms they propose for the effect of temperature on violent crime suggest subtle distinctions in this relationship. We illustrate these hypothesized relationships in Figure 1. Because the T/A theory focuses on higher temperature bringing about the psychological causal mechanism of greater frustration/ aggression, there is little reason to expect that increases at the low end — or even the midrange — of the temperature scale will increase violent crime rates. That is, a community that experiences temperatures around 40 degrees Fahrenheit in the winter and 75 degrees Fahrenheit in the summer should see no seasonal change in crime since there is little reason to expect that this temperature range leads to greater discomfort, and this is represented by line 4 in Figure 1. Arguably, the level of discomfort decreases for increasing middlerange temperatures and only starts becoming uncomfortable at higher temperatures. The precise point at which temperature becomes uncomfortable Crimes of Opportunity or Crimes of Emotion? / 1339 is not clear: while some studies have used 90 degrees Fahrenheit (Anderson, Bushman & Groom 1997; DeFronzo 1984; Rotton 1993), this has been criticized as arbitrary (Cohn 1990a). We sidestep this issue by focusing on the climatic patterns of communities and suggest that looking at the typical range of temperatures within a community can yield a clue to which of these theoretical mechanisms is at work. The crucial point is that the T/A theory predicts areas with hotter climates will experience the greatest seasonal crime oscillations, as shown by line 1 in Figure 1. On the other hand, the routine activities theory suggests that the relationship between temperature and crime rates will be most pronounced in the midrange of temperatures. That is, fewer crimes will be committed during colder temperatures as individuals spend more time inside their homes to avoid the inclement weather, thus reducing the risk of victimization. But as the temperature begins to warm, people venture outside their homes, increasing the possibility of criminal acts, as shown by line 3 in Figure 1. At some point increasing temperature ceases to become more pleasant and no longer induces increasing numbers outdoors (Rotton & Cohn 2000), and thus line 2 in Figure 1 shows that variations in already hot temperatures will have little effect on crime rates. Thus, this model implies that seasonal fluctuations in crime will be 3.8 3.9 4 4.1 4.2 4.3 4.4 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18