Time zone lines and suicides: west side story

A recent study (1) specifically finds high serotonin loadings in the brain of patients who attempt violent suicide. This is important because it connects to a previous area of research (2) showing that the human brain produces more serotonin when exposed to sunlight for longer periods of time. These two studies jointly suggest that the longer the exposure to sunlight, the higher the concentration of serotonin, and thus the more likely a patient is to attempt suicide. The extent to which sunlight exposure contributes to suicide has long been debated. On the one hand, lack of sunlight during the winter season is linked to seasonal affective disorder, and perhaps suicidal depressive moods (3). On the other hand, suicides peak in the spring and summer months (2) and midday rather than nighttime (4). Here, I argue that increased sunlight reduces suicide rates, thereby calling into question this logical chain. I offer a controlled test that identifies the independent role of sunlight on suicide. Observe that individuals living in counties just west of a time zone line receive one less hour of light at night compensated by an extra hour of light in the morning. But there is an asymmetry: Most people typically wake soon after sunrise, but go to sleep long after sunset. This suggests that a shift of one hour of light from morning to night is by no means neutral, but intuitively should increase the total amount of time spent in daylight. Indeed, this is the very motivation for daylight savings time. In principle, adjacent counties on opposite sides of the three major north-south time zone lines in the USA should see a discontinuity in the hours of waking sunlight. So motivated, the paper now pursues a controlled study of the variation in suicide rates along this ribbon of counties. I find that their suicide rate is higher by 7.7% – for a total 740 suicides in the years 1979– 2007 in just this ribbon of counties alone. This implies that holding all other factors constant, an hour more of sunlight at night, by itself, significantly reduces suicides. This has implications both for time zone policy and personal location decisions for those with at-risk family members. I have gathered county-level data from the compressed mortality files (CMF) accessed via the CDC WONDER web site (5). This consists of suicide rates for the years 1979–2007, contained in two distinct compressed mortality files 1979– 1998 and 1999–2007, organized by age group, gender, International Classification of Diseases (ICD) chapter, cause of death, injury, race, state, county, and year. Critically, these rates are also adjusted for age, because it is known that adult suicide rates generally rise with age. Also, males commit suicide more often than females and whites more than other races. I restrict focus to this largest class of suicides, namely, white males. I concentrate on the 205 counties adjacent to the time zone boundaries strictly inside the contiguous 48 states – namely, the Pacific-Mountain time zone boundary (denoted PT ⁄MN), the Mountain-Central time zone boundary (MN ⁄CT), and the Central-Eastern time zone boundary (CT ⁄ET). If a county s population falls below 100 000 in the last decadal census, then CDC WONDER suppressed death counts of five or less after 1989. So, unless two or more consecutive years of data are combined, or two or more adjacent counties are merged, the data are unavailable. After some county mergers necessitated by the data limitations, in 2005, there was an average of approximately 2.3 million white males in the 29 years of the data set, spread across 107 county units, for an average population of 20 796 per county merge. This amounts to a little over 3% of all American counties, with more than 2% of the American white male population. From 1979 to 2007, across all the included county units, there was a total average of 16 132 suicides committed by white males. This yields an average suicide rate of 25 per 100 000. I express the data as a panel. This accounts for the fact that each county group recurs over the entire time span, and its observations ought not to be viewed as independent. Specifically, I run the following random effects Maximum Likelihood Estimation panel regression with observations from the 107 county units for the age-adjusted suicide rates: