Correlations between weather and crime have been reported (Cohn 1990; Lab and Hirschel 1987) and generally crime occurs more frequently in good weather. This correlation seems to be strongest in rural areas, where people tend to be influenced by weather to a greater extent (Lemieux 2004). Temperature and humidity are also correlated with the occurrence of crime (Rotton and Cohn 2004), and since weather changes tend to be seasonal, a correlation between season and murder rate has also been found (Tennenbaum 1994).
Anecdotally, we have noticed that forensic autopsies are rare on days after rain, but we have not previously examined this relationship statistically. In Japan, murder, bodily injury resulting in death, and hit-and-run traffic deaths are the main cases that undergo forensic autopsy. Such crimes are categorized in three ways, based on the action of the offender, but we divided the three crimes into only two categories: murder and bodily injury resulting in death, and hit-and-run traffic death. We took this approach because murder and bodily injury are planned offences, whereas a traffic accident may be unintentional. Hence, in this study we investigated the relationship between weather and cases of murder or bodily injury resulting in death, and that between weather and hit-and-run traffic death in the eastern part of Tokyo from 1998 to 2002.
Our department is in charge of forensic autopsy for cases from the eastern part of Tokyo. We investigated 204 cases of murder or bodily injury resulting in death and 58 cases of hit-and-run traffic death that happened in the eastern part of Tokyo and for which the date and place of the incidents were known; these cases were among 471 forensic autopsies handled by our department from 1998 to 2002. The rest of the cases were excluded because the date of the incident was unknown or death was due to another cause.
The exact time of some of the incidents was unknown, and therefore the weather on the date of each incident was checked using the database of the Japanese Meteorological Agency (JMA). The monthly average temperature and humidity were obtained from the database, and the discomfort index (DI) was calculated as follows: DI (%) = 0.8t + 0.01H (0.99t - 14.3) + 46.3 (t: temperature [degrees]C, H: humidity %). At DI = 75 about half of those tested experienced discomfort (Fukuda and Samata 2005).
Weather types were divided into seven categories based on the classification of the JMA: (1) "sunny day," (2) "sunny and cloudy day" (this category included days that were sunny after being cloudy, cloudy after being sunny, sunny but sometimes cloudy, and cloudy but sometimes sunny), (3) "cloudy day," (4) "cloudy and rainy" (this category included days that were cloudy after being rainy, rainy after being cloudy, rainy but sometimes cloudy, and cloudy but sometimes rainy), (5) "rainy day," (6) "rainy day and sunny day" (this category included days that were sunny after being rainy, rainy after being sunny, rainy but sometimes sunny, and sunny but sometimes rainy), and (7) "snowy day" (this category included days that were snowy; snowy after being sunny, cloudy, or rainy; snowy but sometimes sunny, cloudy, or rainy; and sunny, cloudy, or rainy, but sometimes snowy).
The seven weather types did not occur equally throughout the year over the five-year period (Table 1). Therefore, the annual number of incidents occurring on a day with a particular type of weather was divided by the annual number of days on which this kind of weather occurred; the result of this calculation was defined as the autopsy index (AI) for each weather type. The annual rates of murder, bodily injury resulting in death, and hit-and-run traffic deaths were calculated for each weather type.
As a preliminary analysis, event rates of murder and bodily injury resulting in death during each type of weather in the five-year period were compared all possible combinations, following the Poisson regression model that had weather type as a factor and days as an offset term. Multiplicity was adjusted using the Bonferroni's correction method. After that, seven weather types were grouped using the likelihood test based on the Poisson regression model. Finally, on the basis of the grouped Poisson regression model all groups were compared (Agresti 2002). All comparisons were conducted with a significance level of 5% and a confidence coefficient of 95%. All statistical analyses were conducted by SAS 9.1.
Relationship between daily weather types and cases of murder and bodily injury resulting in death
The total number of cases of murder and bodily injury resulting in death...