Seasonal variation in deaths by suicide is an important study in terms of understanding the possible sociological and biological determinants of suicide and its prevention. Spring peaks for males and spring and autumn peaks for females were found in the UK (Reference Barraclough and WhiteBarraclough & White, 1978; Reference Meares, Mendelsohn and Milgrom-FriedmanMeares et al, 1981), Australia (Reference Eastwood and PeacockeEastwood & Peacocke, 1976; Reference Parker and WalterParker & Walter, 1982), Finland (Nayha, Reference Nayha1982, Reference Nayha1983) and Italy (Micciolo et al, Reference Micciolo, Zimmermann-Tansella and Williams1989, Reference Williams, Williams and Zimmermann-Tansella1991). A spring and late summer peak and a trough in winter months were found for suicides for both men and women in the USA (Reference LesterLester, 1971; Reference Lester and FrankLester & Frank, 1988). Massing & Angermeyer (Reference Massing and Angermeyer1985) give a comprehensive account of the effects of season on suicides in the 1960s and 1970s. Recently, a harmonic analysis was applied to examine suicides in Australia and New Zealand. Only one cycle for both men and women was observed (Reference Yip, Chao and HoYip et al, 1998). A similar finding was also observed in Hong Kong and Taiwan (Reference Ho, Chao and YipHo et al, 1997). The present study attempts to examine seasonal effects on suicides in England and Wales with respect to gender, age group and method of suicide. We have two objectives: to verify the findings in Meares et al (Reference Meares, Mendelsohn and Milgrom-Friedman1981) and Barraclough and White (Reference Barraclough and White1978) about the seasonal effect in England and Wales, i.e. men showed a single 12-month cycle, whereas women showed two cycles; and to investigate whether seasonal variation is found more strongly in some subgroups of the population (e.g. by age or method of suicide) than in others.

METHOD

The Office for National Statistics (England & Wales) provided data for all deaths by suicide for the period 1982-1996. Information on age, gender, method and month of suicide was also made available. The data included here relate to cases where judicial inquiries established that the cause of death was suicide. They were coded in the range E950-959 of ICD-10 (World Health Organization, 1992). The chosen period enabled comparison with results obtained in the 1960s and 1970s (Reference Barraclough and WhiteBarraclough & White, 1978; Reference Meares, Mendelsohn and Milgrom-FriedmanMeares et al, 1981).

The seasonal variations in suicides were examined in three ways. First, the number of suicides in England and Wales for each month of the period 1982-1996 was plotted separately for each gender. Second, a daily mean suicide incidence and cumulative number of suicides were calculated for each month of the study period. Third, a harmonic time series model was applied to different genders, age groups and methods of suicide. The method of analysis was similar to that employed by Barraclough & White (Reference Barraclough and White1978), Micciolo et al (Reference Williams, Williams and Zimmermann-Tansella1991), Ho et al (Reference Ho, Chao and Yip1997) and Yip et al (Reference Yip, Chao and Ho1998). In this model, the variation in suicides between the months is described as the sum of the sinusoidal curves. The seasonal variation consists of those components with cycles that repeat themselves an exact number of times per year. Let A_i be the number of suicides for month i; for a period of 15 years (180 months),

\batchmode \documentclass[fleqn,10pt,legalpaper]{article} \usepackage{amssymb} \usepackage{amsfonts} \usepackage{amsmath} \pagestyle{empty} \begin{document} \[\ \begin{array}{rl}A_{i}=a_{0}+{{\sum}_{j=1}^{90}}&\left(a_{j}\mathrm{cos}\frac{2{\pi}ij}{180}+b_{j}\mathrm{sin}\frac{2{\pi}ij}{180}\right)\\&{\ }{\ }i=1,{\ldots},180\end{array}\ \] \end{document}

where a_j and b_j are constant (j=1,…, 90), a _o is the mean suicide number and

\batchmode \documentclass[fleqn,10pt,legalpaper]{article} \usepackage{amssymb} \usepackage{amsfonts} \usepackage{amsmath} \pagestyle{empty} \begin{document} \[\ a_{j}\mathrm{cos}\frac{2{\pi}ij}{180}+b_{j}\mathrm{sin}\frac{2{\pi}ij}{180}\ \] \end{document}

is commonly referred to as the jth harmonic of A_i with period 15/j years and frequency per annum j/15. The harmonics with j=15, 30, 45, 60, 75 and 90 have periods of 1, 1/2, 1/3, 1/4, 1/5 and 1/6 years and frequencies per annum of 1, 2, 3, 4, 5 and 6, respectively. For example, one such sinusoidal curve is that with a period of 6 months (1/2 year), which has just two peaks and two troughs in each year. These harmonics have a cycle that repeats an exact number of times per year and follows the same pattern in years as any combinations of such harmonics. The quantities a_j and b_j are estimated so as to give the best fit to the data, and they describe the amplitudes of the separate sinusoidal components. The significance of a particular component j is ascertained by testing whether a_j and b_j are significantly different from zero; if they are not, it may be concluded that there is no seasonal variation. It is assumed that the total variance of the distribution of the monthly suicide data can be broken down into three components: random, seasonal and non-seasonal. In this way it is possible to calculate the percentage of total variance attributable to seasonal variation, as well as to random and non-seasonal variation. Under the alternative hypothesis, that the variation is purely random, the monthly totals may be considered as independent identically distributed Poisson random variables. The details of testing the significance of the different components of the variation can be found in Pocock (Reference Pocock1974), Barraclough & White (Reference Barraclough and White1978) and Yip et al (Reference Yip, Chao and Ho1998).

RESULTS

A total of 59 608 (43 229 males and 16 379 females) suicides were identified in England and Wales for the period 1982-1996. The age-specific suicide rates by gender for the study period were plotted and are shown in Fig. 1. A statistically significant decreasing trend was observed in the group (male and female) aged 60 years or over (P <0.05). On the other hand, an increase in the suicide rate among males of age 15-24 years was noted (P <0.05). Suicide rates increased with age among females, but not among males. Males aged 25-39 years have shown the highest rate since 1994. The monthly distribution of suicides, for males and females is shown in Fig. 2. The cumulative number of suicides for both males and females, and the monthly mean number of suicides per day are shown in Table 1 and Fig. 3, respectively. The assumption of even distribution of suicides was rejected (P <0.01): December had a trough for both males and females.

Fig. 1

Age-specific suicide rates (per 100 000) in England and Wales, 1982-1996: [UNK], male 60+; [UNK], male 40-59; [UNK], male 25-39; [UNK], male 15-24; [UNK], female 60+; [UNK], female 40-59; [UNK], female 25-39; [UNK], female 15-24.

Fig. 2

Monthly distribution of suicide deaths in England and Wales, 1982-1996.

Fig. 3

Monthly averages of suicide incidence over 15 years in England and Wales, 1982-1996.

Table 1

The observed and expected number of suicides by month and gender in England and Wales, 1982-1996

	January	February	March	April	May	June	July	August	September	October	November	Dececember
Male
Observed	3755	3251	3777	3706	3717	3660	3669	3626	3481	3590	3605	3392
Expected1	3669	3345	3669	3550	3669	3550	3669	3669	3550	3669	3550	3669
χ2=43.98, d.f.=11, P<0.0001
Female
Observed	1362	1244	1496	1452	1448	1376	1370	1301	1337	1351	1416	1226
Expected1	1390	1268	1390	1345	1390	1345	1390	1390	1345	1390	1345	1390
χ2=50.90, d.f.=11, P<0.0001

Table 2 shows the results of harmonic analysis by gender. Only 15 and 17% of the total variations can be explained by the seasonal component for males and females, respectively. The effects of all seasonal harmonics are marginally statistically significant, with P≅0.05. In order to examine the possible determinant factors of season by different subgroups, the figures for age and methods of suicide were also subjected to the same analysis. The results obtained on the seasonal components (not shown here) are not statistically significant.

Table 2

Harmonic analysis of monthly data on suicides in England and Wales, 1982-1996

Component of variance	Male		Female
All seasonal harmonics	49.54	15%	24.02	17%
1 cycle1	18.36	37%	7.21	30%
2 cycles	0.21	0%	6.55	27%
3 cycles	0.00	0%	0.88	4%
4 cycles	1.41	3%	0.00	0%
5 cycles	20.31	41%	2.95	12%
6 cycles	9.26	19%	6.42	27%
Non-seasonal harmonics	45.95	14%	23.58	17%
Random variations	240.16	72%	91.02	66%
Total variance	335.65	100%	138.62	100%

DISCUSSION

The present study suggests that the marked seasonal fluctuation in suicide rates found in England and Wales in the 1960s and 1970s appears to be diminishing, and in some cases it has vanished completely (Reference Barraclough and WhiteBarraclough & White, 1978; Reference Meares, Mendelsohn and Milgrom-FriedmanMeares et al, 1981). Barraclough & White (Reference Barraclough and White1978) used the same method as described here. They examined the seasons during the period between 1968 and 1972 and found that 49.3% of the variation could be explained by season, with the adjusted ratio 2.5:1 of the components of all seasonal harmonics to random. The effect of season on suicide during the period 1982-1996 accounts for only 15-17% of the variation. The results suggest, therefore, a significant reduction in the proportion of variance accounted for by season. We examined the possible confounding factors, such as age and method of suicide, and found nothing to be significant. Also, we included the undetermined causes of death (E980-989) in the analysis and similar results were obtained. Seasonal components were not significant.

Clinical Implications and Limitations