Postponing a Date with the Grim Reaper: Ceremonial Events and Mortality

Abstract

Past research suggests that people are more likely to die after rather than before important ceremonial events (a death deferral effect). We replicated this finding in a sample based on more than 30 million decedents. In analyses in which we tracked deaths day by day, we analyzed four major events: Thanksgiving, Christmas, New Year's Day, and participants' birthdays. People were more likely to die just after rather than just before all four events. In addition, whereas people were less likely than usual to die on the exact days of Thanksgiving and Christmas, people were more likely than usual to die on the exact days of New Year's Day and their birthdays. Moderator analyses suggested that these effects reflected a will to live. For instance, effects for both Christmas and the birthday were much stronger for children than for adults.

ACKNOWLEDGMENTS

We thank Shira Gabriel and Sandra Murray for their insightful comments on this research.

Notes

¹As observed by an astute reviewer, O. Henry was more gifted as a writer than as a botanist. Ivy plants are evergreens and thus do not lose their leaves in the winter.

²It is plausible that people are less likely to die on the exact day of Thanksgiving because people are less likely to travel, and thus die in accidents, on this day. To assess this possibility, we looked at California mortality data from 1989 to 1998 (http://www.vitalsearch-ca.com) that contained detailed information about cause of deaths. The distribution of natural deaths (i.e., all deaths minus the number of accidental deaths) showed the same pattern that we found in the SSDI. This same point applies to Christmas.

³We also conducted a supplemental analysis to see if the mortality effects for Christmas would disappear for people who are unlikely to celebrate Christmas. Because the SSDI does not contain ethnicity data, we simply sampled the 3,227 participants who had one of the 106 typically Jewish surnames identified by Phillips and King (1988; e.g., Cohen, Goldberg). Although there was a weak trend for deaths to increase over this 4-day period, the death deferral effect was not significant, χ ²(1) = 1.82, p = .177. In addition, the decreased number of deaths on Christmas day was completely eliminated, χ²(1) = 0.08, p > .95. Given the fact that surnames are a reasonable but imperfect marker of Jewish identity, we consider these results supportive of our basic premise.

⁴Because of New Year's Eve parties, it is possible that there are more accidental deaths such as acute alcohol poisoning and traffic accidents. The California mortality data suggested that such deaths cannot account for our basic findings. National Highway Traffic Safety Administration's traffic fatality records (http://www.nhtsa.gov) also confirmed that car accidents constituted only about 2% of total deaths on New Year's Day. To be more concrete, only about 200 people in the United States die in motor vehicle accidents on a typical New Year's Day. In contrast, we observed an average of about 8,000 extra deaths on a typical New Year's Day.

⁵Birthdays are also celebrated with parties. However, the California mortality data showed that the distribution of natural deaths was similar to the pattern observed in the SSDI.

⁶The fact that our findings for both Christmas and birthdays were stronger among children could also be taken as evidence that all ceremonial events are more important to children than to adults. Because of the extreme difficulty of testing this hypothesis for Thanksgiving, we focused on a 5-year window (people who died in 1995–2000) comparing children (those ages 1–20, n = 1,425) to adults. Unlike the Christmas and birthday analyses, the difference between adults and children did not approach significance, χ²(4) = 6.92, p = .140. Because searches by age were much easier to conduct for New Year's Day, we conducted an exhaustive analysis for this holiday. Although sample sizes were still very small for some of these analyses (i.e., for the millennium), there was no evidence that our findings for New Year's Day were stronger for the 2,823 children who died in this 5-day window, χ²(4) = 4.76, p = .313. More specifically, for the entire window we searched, there was no death deferral effect at all for children (those ages 1–20) and virtually no difference between children and adults for an increase in deaths on New Year's Day. Moreover, for the much anticipated new millennium, there was no effect at all for children, except for an unexpected reduction in deaths on January 3. In short, the dramatic differences that existed between children and adults for Christmas and for birthdays did not appear to emerge for the two holidays that children do not consider unusually important.

⁷To see if seasonal variation in death rates could explain our day-by-day findings for Thanksgiving, Christmas, and New Year's Day, we sampled the immediate periods surrounding these three major holidays. Specifically, we analyzed death rates from November 2 to 19, December 2 to 22, and January 4 to 23 (excluding the 15th of the month, for reasons noted in Study 4a). We observed only a trivial increase in death rates in November. We actually observed a decrease in death rates in January, which is consistent with Phillips et al.'s (2006) findings. For December, we did observe a clear seasonal effect. On average each day in December yielded 158 more deaths than the day that preceded it. However, our finding of a death deferral effect was highly significant even after controlling for daily increase, χ ²(1) = 24.72, p  < .001. In fact, the magnitude of the death deferral effect in December was nearly five times larger than the seasonal effect. Finally, it is worth noting that seasonal effects cannot explain our findings for birthdays because we randomly sampled birthdays throughout the calendar year.

⁸In line with these arguments, gender might moderate some of the effects we observed, because women are more likely than men to have rich social networks (House et al., 1988). However, it is difficult to know what kind of gender difference to predict. For example, one might argue that if women care more about close relationships, women should show a larger decrease in the number of deaths on Thanksgiving and Christmas than men (see also Phillips et al., 1992). Alternately, if these holidays represent rare opportunities for men to fill their need for connectedness (Baumeister & Leary, 1995), one might predict a larger decrease in deaths on these holidays for men. At any rate, data on gender data are not provided in the SSDI records. Moreover, it is extremely difficult to code for gender in these data, because common male and female names differ dramatically in the degree to which they are associated with age. For instance, with the exception of Mary, there are few extremely common female names. Moreover, some of the most common female names are strongly confounded with age. For example, based on 1990 census records, Jennifer is a more common name than Dorothy. However, among decedents in the SSDI women named Dorothy outnumbered women named Jennifer 64:1.