The Hound of the Baskervilles effect:

A natural experiment on the Influence of psychological stress on the timing of death

David P. Phillips, Ph.D.,
Professor of Sociology

George C. Liu, B.A.

Kennon Kwok, B.A.

Jason R. Jarvinen

Wei Zhang, M.A.

Ian S. Abramson, Ph.D.,
Professor of Mathematics

October, 2001

To appear in the British Medical Journal, December, 2001

News Embargo Release Date (Pacific Time), Thurs, 4pm, December 20, 2001

Phillips, Liu, Kwok, Jarvinen, Zhang (Department of Sociology) and Abramson (Department of Mathematics) are from the University of California at San Diego. Address reprint requests to Professor Phillips at Sociology Department 0533, University of California at San Diego, La Jolla, CA 92093-0533 (Tel: 858-534-0482; Fax: 858-534-4753; Email:

What is already known on this topic:

Laboratory studies show cardiovascular changes following mild psychological stress. However, these studies do not reveal whether, in the world outside the laboratory, fatal heart attacks increase after psychological stress. Previous nonlaboratory studies were unable to control for physical and medical changes associated with most stressful occasions, and therefore were not able to indicate whether psychological stress per se could trigger fatal heart attacks.

What this study adds:

Unlike whites, Chinese and Japanese associate "four" and the fourth with death. Unlike whites, Chinese/Japanese cardiac deaths peak on the fourth day of the month. Because the fourth of the month is not objectively more hazardous than other days of the month, our data may provide the best evidence to date that cardiac mortality increases after psychological stress.

In The Hound of the Baskervilles, Charles Baskerville dies from a heart attack induced by stress. Because our data suggest that heart attacks increase on stressful occasions, our findings suggest that the "Baskerville effect" exists in fact as well as in fiction.


Objective: To determine whether cardiac mortality is abnormally high on days considered unlucky.

Design: The study group consisted of Chinese and Japanese (who consider the number "four" unlucky) and white controls (who do not). Cardiac and non-cardiac mortality on and around the fourth of each month was examined for each group.

Setting: Entire United States.

Subjects: All U.S. Chinese and Japanese (n=209,908) and whites (n=47,328,762) whose computerized death certificates were recorded between 1/1/1973 to 12/31/1998.

Main outcome measures: Ratio of observed number of deaths on day 4 to expected number of deaths on day 4; the expected number was estimated from mortality on other days of the month.

Results: Chinese and Japanese cardiac mortality peaks on the fourth of the month.    The fourth-day peak is particularly large for deaths from chronic heart disease (13% above expected; 95% confidence interval, 6%-21%) and still larger for chronic heart deaths in California (27% above expected; 15%-39%). Within this group, inpatients display a particularly large, 47%, fourth-day peak (19%-81%). The fourth-day peak is not followed by a compensatory drop in deaths below normal. White controls, matched on age, sex, marital status, hospital status, location, and cause of death, display no fourth-day peak in cardiac mortality.

Conclusions: Chinese and Japanese cardiac mortality is significantly higher on the fourth than on any other day of the month. The results are inconsistent with nine alternative explanations for the findings. For example, the fourth-day peak does not seem to occur because of changes in the patient's diet, alcohol intake, exercise, or medication regimes. Our findings are consistent with the hypothesis that cardiac mortality increases on psychologically stressful occasions.


In The Hound of the Baskervilles, by Conan Doyle, [1] Charles Baskerville has a fatal heart attack resulting from extreme psychological stress. Conan Doyle was a physician as well as an author; was his story based on medical intuition or literary license? Are fatal heart attacks and stress linked in fact as well as in fiction? Doyle's intuition is consistent with many laboratory studies, [2-4] which show cardiovascular changes following psychological stress. However, for ethical reasons, only non-fatal stressors can be studied in the laboratory, and one may not be able to generalize beyond these relatively mild stressors [5] to determine whether, in the real world, fatal heart attacks increase immediately after psychological stress.

One way to circumvent the ethical problems of the laboratory experiment, and yet retain some of its rigor, is to use a natural experiment [6], one that seeks psychological effects of a real-world event occurring simultaneously to a study group (which is stressed by the event) and to a control group (which is not). To separate the effects of psychological factors from physical changes in the environment, the event under study should not coincide with destruction of the physical environment (as in disasters) or with disruption of medical services. [7] In addition, the event should occur several times, because it is easier to correct for accidental, confounding factors when studying recurring events. We found no previous studies meeting all these criteria. [3-5,7,8]

We approached the problem by identifying a cultural phenomenon that has unpleasant associations for one group (Chinese and Japanese) and neutral associations for another (whites). In Mandarin, Cantonese, and Japanese, "death" and "four" are pronounced nearly identically. [9,10] Extensive participant observation by three of our authors (Liu, Kwok, and Zhang) indicates that the number "four" evokes discomfort and apprehension in some Chinese/Japanese. Consequently, some Chinese/Japanese hospitals do not list a fourth floor or rooms numbered "four". [10,11] The Chinese omit the number "four" in designating military aircraft-an omission said to result from the link between "four" and death. [12] Some Japanese have reported avoiding travel on the fourth and some Chinese patients have reported apprehension about this date (these reports are based on participant observation).    Aversion to the number "four" is also evident from examination of Chinese/Japanese restaurants, which avoid this number (as shown below).

This paper shows that Chinese/Japanese cardiac deaths peak on the fourth; whites do not display this pattern.


We examined computerized daily U.S. mortality for Chinese/Japanese (n=209,908) and whites (n=47,328,762) from 1973 (when daily mortality was first recorded) to 1998 (the latest available data). [13,14] Chinese, Japanese, and whites were identified from separate racial codes on the death certificate.

For everyday life, American Asians use the Gregorian calendar.15 Therefore, we examined the Gregorian, rather than the ceremonial lunar calendar. There is no correlation between day of the month in one calendar and in the other; thus, the effects of day of the month in the two calendars are not confounded.

Because U.S. mortality is slightly higher in the first week of each month, [16] we used average daily mortality in the first week (days 1-3,5-7) to estimate the expected mortality level on day 4, given the null hypothesis (H0.)    Subsidiary analyses used days 1-3,5-28 to estimate expected mortality on day 4. The two approaches yield nearly identical estimates of the number of deaths expected on day 4.

We measured the size of the fourth-day peak by R=X/(Y/6), where X is the number of deaths on the fourth of the month, and Y is the total number of deaths in the rest of the first week of the month. The variance of R was determined from Table 3 in Gardner and Altman [17] As in previous work [16,18,19,20] we followed recommendations in Vital Statistics of the United States, [21] which justifies calculation of confidence limits for complete population data.


Statistical evidence for avoidance of the number "four"

New telephone subscribers in California have some choice of the last four digits in their telephone numbers. In this four-digit section, California restaurants listed as Chinese or Japanese [22] display significantly (P=.00000003) fewer "fours" than expected (observed=366; expected=4748/10=474.8; binomial test). California restaurants listed as "American" [22] display no such pattern (P=.879; observed=219; expected=2036/10=203.6; binomial test).

Asian and white cardiac mortality on and around the fourth

If the number "four" evokes superstitious stress in some Chinese/Japanese, and if Doyle's medical intuitions were correct, Chinese/Japanese cardiac mortality should peak on the fourth of each month. Doyle suggests that Charles Baskerville was susceptible to a stress-induced heart attack because he had a chronic heart condition. If so, chronic heart disease should display a particularly large fourth-day peak. Sir Charles' superstitious fear of an avenging, spectral hound was shared and reinforced by his neighbors; similarly, Chinese/Japanese superstitious fears are likely to be stronger where they are reinforced by large Chinese/Japanese populations. Hence, the fourth-day peak is likely to be stronger in California, which accounts for 42.6% of the Chinese/Japanese deaths under study.

These expectations are supported by our data: On the fourth, cardiac deaths are significantly more frequent than on any other day of the month, and are 7.3% higher than the 6-day average (Ratio=1.07, 95% confidence interval, 1.03-1.12). This percentage increase (7.3%) is bigger for chronic heart deaths (13%; Ratio=1.13, 1.06-1.21) and still bigger (27%; Ratio=1.27, 1.15-1.39) for chronic heart deaths in California. The fourth-day mortality peak is henceforth termed the "Baskerville effect".

Whites do not display this effect, nor is it evident for Chinese/Japanese who die from causes other than chronic heart disease (for non-heart diseases: Ratio=1.02, .99-1.05; for heart diseases other than chronic: Ratio=1.04, .98-1.09). Thus, Doyle's medical intuition was remarkably precise: in our dataset, the fourth-day peak occurs only in persons with pre-existing heart conditions.

Alternative Explanations for the Findings

Because U.S. mortality peaks in the first week of the month, we used a six-day comparison period (1-3,5-7). Perhaps the Baskerville effect is somehow an artifact of this comparison period. A regression analysis, using days 1-3,5-28, generated a substitute for the 6-day average. The 6-day and 27-day approaches always yielded very similar Ratios and all findings remain statistically significant whichever approach is used (see, e.g., Figure 1).

Perhaps the Baskerville effect occurs because superstitious relatives attribute the decedent's death to the fourth when it actually occurred on the third. Given this hypothesis, inpatients should display a small or nonexistent fourth-day peak, because the deathdates of inpatients are accurately recorded. However, among California Chinese/Japanese dying from chronic heart disease, inpatients display a larger fourth-day peak (Ratio=1.47, 1.19-1.81), than do others (Ratio=1.16, .95-1.39).

On the fourth, patients may 1) change diets; 2) increase alcohol consumption; 3) refuse medicines; or 4) overstrain themselves. These behavioral changes are much less likely for inpatients (who are closely controlled and monitored) than for others. The large inpatient effect undermines all these explanations.

Perhaps the fourth-day peak in inpatient deaths occurs because patients refuse to leave hospital on this unlucky day, thus dying as inpatients rather than outpatients. This "discharge hypothesis" implies a compensatory drop in outpatient deaths on the fourth, and no such drop is found.

If the fourth-day peak occurred because deaths were merely precipitated by a day or two, then this peak should be immediately followed by a marked, compensatory drop in deaths. No such drop is evident in mortality fluctuations throughout the month (Figure 1A).

Perhaps the Baskerville effect appears because the study and control groups differ demographically. To test this hypothesis, we constructed a control group (Figure 1B) of whites matched to each member of the study group (Figure 1A) on state and cause of death, inpatient status, age, sex, and marital status. This matched control group does not show the Baskerville effect.

It is unlikely that the fourth day is objectively more hazardous than the six days surrounding it because white controls display no increased mortality on the fourth.


To assess the effects of psychological stress on cardiac mortality, we identified a recurring, aversive occasion which 1) induces stress in study groups but not in controls, and 2) is not objectively hazardous.

This paper demonstrates that Chinese/Japanese cardiac deaths peak on the fourth of the month. White controls display no such peak. Presently, the only explanation consistent with the findings is that psychological stress linked to the number "four" elicits additional deaths among Chinese/Japanese.

There is no linguistic link in English between "thirteen" and "death" and this may help to explain why white mortality shows no peak on the thirteenth, despite the Chinese/Japanese peak on the fourth.

The debate on whether there are fatal psychosomatic processes is unresolved. The question deserves further investigation, provided that such investigation uses rigorous methods, careful controls, and large samples.    The natural experiment presented here appears to have met these criteria, and has provided a new technique for examining fatal psychosomatic effects.

Our findings are consistent with the scientific literature and with a famous, non-scientific story: The "Baskerville effect" exists both in fact and in fiction, and suggests that Conan Doyle was not only a great writer, but a remarkably intuitive physician as well.

We thank Peter H.T. Liu, A.M.D., Ph.D., Lien-Fen S. Chu, Rachel Phillips, Chen Ruzhen, Daniel Smith, M.B.A., Elvira R. Strasser, Ph.D., Li Xianghui for helpful comments.
Contributors: DPP coordinated and designed the study and wrote most of the computer programs. GCL, KK, and WZ conducted informal participant observation of Chinese and Japanese cultures and performed most of the literature review. ISA calculated the standard errors. All authors participated in the writing and data analysis. DPP is the guarantor.
Funding: This study was supported in part by a grant from the Marian E. Smith Foundation.
Competing Interests: None.


1    Doyle AC. The Hound of the Baskervilles. London: J. Murray, 1964.
2    Soufer R, Bremner JD, Arrighi JA, Cohen I, Zaret BL, Burg, MM et al.    Cerebral cortical hyperactivation in response to mental stress in patients with coronary artery disease. Proc. Natl. Acad. Sci. USA 1998;95:6454-9.
3    Krantz DS, Sheps DS, Carney RM, Natelson BH. Effects of mental stress in patients with coronary artery disease. JAMA 2000;283:1800-2.
4    Rozanski A, Blumenthal JA, Kaplan J. Impact of psychological factors on the pathogenesis of cardiovascular disease and implications for therapy. Circulation 1999;99:2192-217.
5    Steptoe A, Vogele C. Methodology of mental stress testing in cardiovascular research. Circulation Supplement 1991;83:II14-24.
6    Phillips DP, Glynn L. Field studies. Encyclopedia of Psychology. Oxford: Oxford University Press, 1999.
7    Phillips DP, Halebsky SC. The epidemiology of found experiments. JAMA 1995;273:1221.
8    Witte DR, Bots ML, Hoes AW, Grobbee DE. Cardiovascular mortality in Dutch men during 1996 European football championship: longitudinal population study. BMJ 2000;321:1552-4.
9    Min Su Shu Zi De Bing Tai Yu Lao Fa [Popular morbid beliefs in numbers and their alleviation]. (accessed April 19, 2000).
10    Schauwecker's Guide to Japan: Superstition. http://www.japan- (accessed December 7, 2000).
11    Visitors' brochure for Chungshan Hospital, Taipei; copy available on request.
12    rec.aviation.military FAQ, Part 5. Subject: H.11. Chinese aircraft designations.
Http:// (accessed October 14, 2001).
13 Mortality Detail File. Hyattsville, Md.: U.S. National Center for Health Statistics (computer data file).
14 Mortality File. Sacramento, Ca.: California Department of Health Services (computor data file).
15    The Western Calendar and Calendar Reforms.,5716,108734+5+106114,00.html (accessed December 8, 2000).
16    Phillips DP, Christenfeld N, Ryan NM. An increase in the number of deaths in the United States in the first week of the month. N. Engl. J. Med. 1999;341:93-8.
17 Gardner MJ, Altman DG, eds. Statistics with Confidence. London: British Medical Journal, 1989.
18.    Phillips DP, Ruth TE, Wagner LM. Psychology and survival. Lancet 1993;342:1142-5.
19    Phillips DP, Smith DG. Postponement of death until symbolically meaningful occasions. JAMA 1990;263:1947-51.
20    Phillips DP, King EW. Death takes a holiday: Mortality surrounding major social occasions. Lancet 1988;ii:728-32.
21    National Center for Health Statistics. Vital statistics of the United States, 1973-1994. Vol. 2. Mortality. Part A. Technical Appendix. Washington DC: Government Printing Office, 1976-1997.
22    AltaVista - Yellow Pages Home. (accessed August 13, 2000).
23    National Center for Health Statistics. Vital statistics of the United States, 1994. Vol. 2. Mortality. Part A. Washington DC: Government Printing Office, 1997.

Figure 1. Number of California inpatient deaths from chronic heart disease, by day of the month (1989-1998). The study period begins in 1989 because, prior to that date, inpatient status was seldom recorded on California death certificates, [14] resulting in no pre-1989 recorded Chinese/Japanese inpatient deaths from chronic heart disease. Deaths from chronic heart disease are those coded 410-414 in the International Classification of Diseases, 9th Revision. [23] The error bars represent 95% Poisson confidence intervals determined from Gardner and Altman, Table 3. [17] The dotted line represents a regression line fitted to days 1-3,5-28. Panel A: Chinese and Japanese deaths. Panel B: Deaths for white controls, matched to each Chinese and Japanese decedent on state and cause of death, inpatient status, age, sex, and marital status. For each Chinese/Japanese decedent we randomly selected twelve white controls who had the same age, sex, and marital status. Twelve matches were selected because it was possible to find at most twelve matches for some of the Chinese/Japanese decedents. Chinese, Japanese and whites were identified from race codes on the death certificate.

Click on the links below to see Figure 1a (for Chinese/Japanese mortality) and Figure 1b (for white mortality).

Figure 1a
Figure 1b
Baskerville Press Release

Converted by David Phillips and Andrew Scriven