Data sources

Suicide rates were based on mortality and population data produced by the Office for National Statistics, 1950-1999 (Office for National Statistics, 2001). Suicide deaths were extracted on the basis of the codes used to identify suicides or probable suicides in the International Classification of Diseases (ICD, revisions 6 to 9, latterly World Health Organization, 1977). The years we have studied span four revisions of the ICD, and the codes used for suicide over these years are as follows: 6th revision ICD E970-E979; 7th ICD E970-E979; 8th ICD E950-E959; 9th ICD E950-E959. In the eighth and ninth revisions of the ICD, a new category of ‘undetermined deaths’ (E980-E989) — deaths given open verdicts by coroners — was introduced. As most such deaths are suicides, and in keeping with official reporting of national suicide rates (Reference Charlton, Kelly and DunnellCharlton et al, 1992; Reference Kelly and BuntingKelly & Bunting, 1998), we included these in our analysis (see below). We excluded deaths coded E988.8 (accelerated registrations) as most of these are homicides.

Age, period and cohort effects

Distinguishing age, period and cohort effects from each other is problematic because of the linear dependency between these three variables; this is known as the identification problem. A number of different graphical and modelling approaches to delineate these effects have been suggested in the literature (Robertson & Boyle, Reference Robertson and Boyle1998a ,Reference Robertson and Boyle b ). Previous authors have found that age—period—cohort modelling does not add greatly to the information obtained from graphical approaches (see, for example, Reference Allebeck, Brandt and NordstromAllebeck et al, 1996; Reference Newman and DyckNewman & Dyck, 1988) and for this reason we have mainly restricted our analysis to graphical assessments of trends.

Assessment of age and period effects

We assessed the extent to which suicide varies with age by plotting age- and gender-specific trends in suicide (1950-1999) within six age bands: 15-24; 25-34; 35-44; 45-54; 55-64; and 65+. Period effects refer to the extent to which living in a particular time period affects disease risk similarly in all groups in the population, regardless of age and birth cohort. We investigated period effects by plotting age-specific suicide rates in five different time (of death) periods: 1955-1959; 1965-1969; 1975-1979; 1985-1989; and 1995-1999. We omitted the intermediate periods (1950-1954, 1960-1964, etc.) to simplify the graphical presentation of the data. We also assessed period effects by: (a) looking for changes in age-specific trends common to all age—gender groups in the graphs of age-specific rates; and (b) by excluding some methods of suicide and undetermined deaths from our analysis of cohort effects (see below).

Assessment of cohort effects

We investigated the influence of cohort effects on suicide rates by plotting suicide rates at age 15-19, 20-24, 25-29, 30-34, 35-39, 40-44, 45-49, 50-54 and 55-59 for those born in nine successive 5-year birth cohorts — i.e. births around 1940, 1945, 1950 and so on up to 1980. Office for National Statistics mortality data are summarised as the number of deaths, by year, within 5-year age bands. Each of our estimates of the suicide rate for a particular birth cohort at a given age, therefore, includes the mortality experience of people born over a 9-year period centred around the specified year of birth of the birth cohort (see Table 1). For example, for the 1940 birth cohort (highlighted), we used the suicide rates for 15- to 19-year-olds between 1955 and 1959 as our estimate of their suicide rates at age 15-19. Likewise, for our estimate of their suicide rates at age 20-24, we used the suicide rates of 20- to 24-year-olds between 1960 and 1964, and so on. As can be seen in Table 1, deaths in any particular year within an age band never contribute to more than one birth cohort's estimated mortality rate. Deaths occurring among those born in the years around the mid-year of each birth cohort contribute to a diminishing extent as one moves towards the mid-year of the next birth cohort.

Table 1 Diagram showing data used to construct birth cohort specific suicide rates (focusing on 1940 birth cohort)

Assessment of cohort effects with adjustment for possible period effects

Changes in the toxicity of domestic gas, car exhaust gases and the drugs most commonly taken in overdose have all influenced patterns of suicide in England and Wales over the past 50 years (Reference KreitmanKreitman, 1976; Reference Gunnell, Middleton and FrankelGunnell et al, 2000; Reference Amos, Appleby and KiernanAmos et al, 2001). To assess the influence of these period effects on suicide rates, we first examined cohort effects for all suicides regardless of the method of suicide, and then for non-overdose, non-gas suicides only. The main methods of non-overdose, non-gas suicide in England and Wales are hanging, jumping and drowning (Reference Charlton, Kelly and DunnellCharlton et al, 1992) and these methods have not changed in lethality or availability in the post-war years.

Similarly, over the period covered by this analysis there was a change made in the official classification of suicide (see above). The category undetermined deaths (those given a coroner's open verdict) was introduced in 1968. As this change in the categorisation of deaths could have influenced the number of officially recorded suicides subsequent to 1968 (a period effect), we investigated the influence on any apparent birth cohort effects of restricting our analyses to those deaths given a suicide verdict alone (ICD—9, E950-959).

To further investigate the presence of birth cohort effects, we assessed the extent to which the ranking of successive birth cohort's suicide rates at age 15-19 were sustained at later ages by calculating Kendall's rank correlations of their rankings using Stata (version 7.0; StataCorp, 2001).

Age and period effects

Trends in suicide (excluding undetermined deaths) between 1950 and 1999 differed markedly by age and gender (Fig. 1). Suicide rates doubled in men aged 15-44 years although, with the exception of 25- to 34-year-olds, they began to decline in the 1990s. In older men and women in all age groups, except those aged 15-24, rates have declined since the 1960s. The varying rates by age group provide evidence of an age-by-period interaction; the relative rates of different age groups vary in different time periods. The relatively consistent ranking of age-specific suicide rates in females provides evidence of an age effect; regardless of time period, suicide rates are highest in women aged 55+ and lowest in those aged <34. Possible period effects are suggested by the fall in suicide rates occurring in all age—gender groups in 1965-70 and the rise in most groups in the late 1970s/early 1980s.

Fig. 1 Age-standardised suicide rates: 1950-1999 England and Wales (3-year moving averages) in (a) males and (b) females.

Figure 2 further investigates period effects. Suicide rates in the six age groups are shown for five time (of death) periods. Period effects are present if the suicide rate at all ages in a particular time period is higher or lower than that in other time periods and if there is evidence of parallelism between the lines. There is no strong evidence of period effects in males; the profile of age-specific suicide rates varied markedly between 1955 and 1995. In females, Fig. 2 indicates that there have been year-on-year decreases in suicide rates across all age groups except 15- to 24-year-olds over the time studied, indicative of period effects.

Fig. 2 Suicide and undetermined death rates by time (of death) period in (a) males and (b) females.

Cohort effects

Figure 3 shows that in every 5-year age band up to age 30-34, suicide rates were higher in each successive birth cohort. This indicates a possible birth cohort effect on youth suicide. This observation is supported by the finding of significant correlations (generally >0.75) between the ranking of each birth cohort's suicide rates at age 15-19 with the ranking at subsequent ages (Table 2). There is evidence, however, that suicide rates for the 1975 and 1980 birth cohorts at age 15-19 and 20-24 (1975 birth cohort only) are similar to those of the 1970 birth cohort. For the earlier born (pre-1960) birth cohorts there is some evidence that suicide rates peak at successively younger ages: 45-49 years for 1940 and 1945 birth cohorts, and 40-44 for the 1950 birth cohort, indicating a possible period effect as peaks occur in the same time period (around 1990).

Fig. 3 Male rates of suicide and undetermined death in successive 5-year birth cohorts at different ages. (a) All suicides and undetermined deaths; (b) suicides and undetermined deaths except those by overdose and gassing; (c) overdose, gassing and undetermined deaths excluded.

Table 2 Kendall's rank correlations (and P value of test of independence) between ranks of each cohort's suicide rate at age 15-19 and the rank of their rates at subsequent ages

Gender (age)	n of rank estimates	Suicide and undetermined mortality rates (all methods) (15-19)		Suicide and undetermined mortality rates (all methods excluding overdose and gassing) (15-19)		Suicide mortality rates (excluding undetermined deaths and deaths by overdose and gassing) (15-19)
		Kendall's rank correlation	P value for test of independence1	Kendall's rank correlation	P value for test of independence1	Kendall's rank correlation	P value for test of independence1
Males
20-24	8	0.786	0.009	1.000	0.001	0.857	0.004
25-29	7	0.810	0.016	1.000	0.003	0.905	0.007
30-34	6	0.867	0.024	1.000	0.009	0.867	0.024
35-39	5	0.400	0.46	1.000	0.028	0.800	0.086
40-44	4	1.000	0.089	0.333	0.73	0.000	1.00
45-49	3	-0.333	1.00	0.333	1.00	0.333	1.00
Females
20-24	8	0.429	0.17	0.357	0.30	0.643	0.035
25-29	7	-0.524	0.13	0.429	0.23	0.238	0.54
30-34	6	-0.467	0.26	0.067	1.00	-0.200	0.71
35-39	5	-0.800	0.086	-0.400	0.46	-0.400	0.46
40-44	4	-0.667	0.31	-1.000	0.089	-1.000	0.089
45-49	3	-1.000	0.30	-1.000	0.30	-1.000	0.30

In Fig. 3b suicides by overdose and gassing are excluded. Similar birth cohort effects were observed, but in contrast to those shown in the full analysis, there was no evidence that rates peaked earlier in later born cohorts, supporting the suggestion that this was a period effect because of the popularity and lethality of a particular method of suicide. As this occurred towards the end of each cohorts' period of follow-up, the effect could reflect the recent decrease in suicides because of the reduced toxicity of car exhaust gases (Reference Amos, Appleby and KiernanAmos et al, 2001). Exclusion of undetermined deaths in Fig. 3c does not alter these conclusions.

In females (Fig. 4a), unlike males, there is no clear evidence that successive birth cohorts have experienced monotonic increases in their suicide rates at all ages. The statistical analysis in Table 2 supports this finding. In the analysis excluding overdose and gassing (Fig. 4b) and undetermined deaths (Fig. 4c), which together account for over half of all female suicides, there is a suggestion that the 1975 and 1980 birth cohorts experienced higher suicide rates than previous birth cohorts, but because of the limited number of data points it is not possible to reach a firm conclusion about this.

Fig. 4 Female rates of suicide and undetermined death in successive 5-year birth cohorts at different ages. (a) All suicides and undetermined deaths; (b) suicides and undetermined deaths except those by overdose and gassing; (c) overdose, gassing and undetermined deaths excluded.

Main findings

We have found evidence for a cohort effect on suicide rates in males born after 1940. Over the period we examined, successive male birth cohorts appear to have experienced a higher risk of suicide throughout their lives than earlier-born cohorts. These effects do not seem to be because of changes in the official classification of suicide or in the lethality of commonly used methods. The cohort effects may be diminishing in the most recent two male birth cohorts (1975 and 1980). In females, apart from an indication in the assessment of non-overdose/non-gas suicides of a rise in the suicide rates of the 1975 and 1980 birth cohorts, no similar effects were seen.

Strengths and limitations

The main limitation of our analysis is the rather short period of follow-up for the cohorts that have experienced some of the highest levels of suicide before the age of 30. It is possible that as they age these birth cohorts will no longer continue to experience such high death rates. Indeed, there is evidence of such a diminution in the cohorts born before 1960.

We did not attempt to undertake age—period—cohort modelling in view of the recognised limitations of these approaches (Reference Newman and DyckNewman & Dyck, 1988; Reference Lee and LinLee & Lin, 1995). Our graphical displays allow us to investigate age and period effects, but because of their linear dependency assessment of their separate effects is problematic. The most consistent evidence of a period effect was the decrease in suicides in all age—gender-groups in the late 1960s (Fig. 1). The most likely explanation for this is the detoxification of the domestic gas supply and the subsequent reduction in both domestic gas and overall suicides (Reference KreitmanKreitman, 1976). The rise in suicides in most age— gender—groups in the late 1970s/early 1980s coincides with the rise in unemployment at that time (Reference Gunnell, Lopatatzidis and DorlingGunnell et al, 1999) and could reflect the effects of economic recession on suicide.

As far as we could, we controlled for the period effects of changes in the availability/lethality of common methods of suicide (domestic gas poisoning and overdose) by carrying out an analysis restricted to those suicides using other methods. Although this approach takes no account of possible ‘method transfer’ to methods other than overdoses and gassing, there is some epidemiological evidence that people who use another method of suicide rather than gassing tend to use overdose, and this method was also excluded from this restricted analysis (Reference Gunnell, Middleton and FrankelGunnell et al, 2000). Furthermore we investigated the effects of changes in the official coding of suicide deaths by inspecting figures based on suicide deaths alone, excluding undetermined deaths (ICD—8 and 9 E980—E989). The change in coding did not appear to influence the patterns we observed. By restricting our analysis to post-war deaths we ensured the well-recognised effects of war on national suicide rates (Reference Charlton, Kelly and DunnellCharlton et al, 1992) did not influence observed trends. We were, however, unable to control for other possible period effects, such as the economic recession in the 1980s and associated raised levels of unemployment (Reference Gunnell, Lopatatzidis and DorlingGunnell et al, 1999). Our analysis of period effects in males (Fig. 2) revealed no strong evidence of such effects. Thus, the year-on-year rises in young male suicide are more likely to be cohort effects or reflect the influence of year-on-year changes in risk factors for suicide, which have varying effects on different age— gender groups.

Cohort effects on suicide reported in other countries

Analyses of national mortality data up to the early 1990s for the USA (Reference Murphy and WetzelMurphy & Wetzel, 1980), Canada (Reference Solomon and HellonSolomon & Hellon, 1980), Italy (Reference LaVecchia, Bollini and ImaziaLa Vecchia et al, 1986), New Zealand (Reference Skegg and CoxSkegg & Cox, 1991), Spain (Reference Granizo, Guallar and Rodriguez-ArtalejoGranizo et al, 1996), Sweden (Reference Allebeck, Brandt and NordstromAllebeck et al, 1996) and Belgium (Reference Moens, van Oortmarssen and HonggokoesoemoMoens et al, 1987), using a range of different analytical approaches, have reported that suicide rates in successive male birth cohorts have increased in the post-war years. In contrast, an analysis of successive Danish birth cohorts, up to 1971 births, found no evidence of such an effect (Reference Bille-Brahe and JessenBille-Brahe & Jessen, 1994). Similarly in Australia, successive birth cohorts followed up to 1979 showed little evidence of cohort effects, except in the youngest (1955 and 1960) birth cohorts (Reference Goldney and KatsikitisGoldney & Katsikitis, 1983). A more recent analysis of Australian data up to 1999, however, reports post-war patterns of male suicide consistent with those seen in our analysis (Reference Snowdon and HuntSnowdon & Hunt, 2002).

Findings for female suicides are less consistent than those for males. In Britain declines in the suicide rates of successive post-war cohorts have been reported (Reference Surtees and DuffySurtees & Duffy, 1989), whereas there have been increases in successive female birth cohorts in the USA (Reference Murphy and WetzelMurphy & Wetzel, 1980), Canada (Reference Solomon and HellonSolomon & Hellon, 1980), Belgium (Reference Moens, van Oortmarssen and HonggokoesoemoMoens et al, 1987) and, to a lesser extent, Australia (Reference Goldney and KatsikitisGoldney & Katsikitis, 1983). No effects were seen in New Zealand (Reference Skegg and CoxSkegg & Cox, 1991), Denmark (Bille-Brahe & Jessen), Italy (Reference LaVecchia, Bollini and ImaziaLa Vecchia et al, 1986) or Sweden (Reference Allebeck, Brandt and NordstromAllebeck et al, 1996).

Differences between studies could reflect real variations in national patterns of suicide, differences in the period studied, the analytical approaches used, or study power. Analyses of suicide trends in small countries and in women, where the numbers of suicides are relatively low, are unable to give precise estimates of suicide rates within a particular age— period—cohort grouping. Detection of cohort effects in this context is difficult because of fluctuations in rates based on few events. The different findings could also reflect the varying influence of period effects in different countries and the capacity of the analytical approaches used to assess and control for such effects. Goldney & Katsikitis (Reference Goldney and Katsikitis1983), for example, suggested that differences in the findings in Australia compared with the USA and Canada could have arisen through the period effects of restrictions on the prescribing of barbiturates in Australia introduced in 1967. More recent analyses of Australian and Canadian suicide trends suggest similar birth cohort effects to those seen in this analysis of data in England and Wales (Reference Mao, Hasselback and DaviesMao et al, 1990; Reference Snowdon and HuntSnowdon & Hunt, 2002).

Possible explanations for birth cohort effects

Taken together, the evidence from a range of industrialised countries suggests that, at least up until the 1970 birth cohort, each successive post-war cohort of males has experienced a higher suicide rate than its predecessor. Findings for females are less clear-cut. There are a number of possible explanations for the cohort effects observed in males. First, cohort effects could result from the long-term impact of changing levels of exposure in childhood or early adulthood to some environmental influence on long-term mental health. Such factors might include increasing levels of parental separation (Reference Agid, Shapira and ZislinAgid et al, 1999), increased substance misuse (Reference Christophersen, Rooney and KellyChristophersen et al, 1998), or changes in the labour market such as increased female participation (Reference Platt, Hawton, Hawton and van HeeringenPlatt & Hawton, 2000). Second, it is possible that increased media portrayals of idealised lifestyles in televised drama and advertising materials results in young people having increased, but unrealistic, expectations from life. Such mismatches between expectation and reality could lead to increases in suicide risk. Similarly, changes in the media portrayal of suicide (as relatively common-place and acceptable) could permanently influence the attitudes of young viewers regarding its cultural normality, and hence their likelihood of committing suicide in times of crisis. Last, the effects could conceivably result from changes in the popularity of particular methods of suicide in successive generations. Such an effect could occur if an individual's preferred method of suicide is based on early life experiences (e.g. media portrayal) and remains the same thereafter. In the UK, for example, it is possible that the increased suicide rates in successive male birth cohorts could reflect a greater preference for hanging as a method of suicide; this method is more usually fatal than the other commonly used methods such as overdose and gassing (Reference PounderPounder, 1993).

Reasons why these effects should influence female suicide rates to a lesser extent are unclear, although there is a suggestion that similar patterns could be beginning to emerge in younger female generations in England, and in the analyses based on large populations such as the USA and Australia (see above). The main social change among women in the post-war years has been their increased engagement in the labour market and greater equality with men.

Public health implications

If younger birth cohorts carry their increased suicide rates through into later life then the recent falls in suicide rates in England and Wales will be reversed (Reference Kelly and BuntingKelly & Bunting, 1998). Possible explanations for the observed effects require clarification to inform suicide prevention activities.