- Email: [email protected]
Suicide and unemployment in Australia 1907–1990
Sm. Sri. Med. Vol.36,No. 6.pp.749.-756, 1993 Printed inGreatBritain
SUICIDE
AND
0277-9536193 $6.00+ 0.00 PergamonPressLid
UNEMPLOYMENT
IN AUSTRALIA
STEPHENMORRELL,RICHARDTAYLOR,SUSANQUINE Department
of Public
Health,
Faculty
of Medicine,
University
and
1907-1990
CHARLESKERR
of Sydney,
N.S.W.
2006. Australia
Abstract--Ever since Durkheim postulated a relationship between economic change and suicide there has been evidence of a general association between aggregate data on unemployment and the frequency of suicide. Quantitatively, however, the association has been variable and it is clear that due to differing cultural, social and individual determinants of suicide, the relationship is complex. Methodological difficulties abound with interpretation of aggregate data. Australian records for most of the present century are suitable for examining secular trends in suicide and unemployment by age group and gender to gain an indication of the extent to which both parameters may be causally related. An aggregate/ecological study was designed to incorporate quantitative and qualitative strategies. Annual age-adjusted male and female suicide rates and annual unemployment rates were derived for the period 1907-1990. Female suicide rates were generally stable throughout the period, whereas those for males demonstrated sharp fluctuations with the peaks coinciding with times of high unemployment. The association between suicide and unemployment for 15524 year old males was comparatively high for the recent period, 1966-1990. The increasingly youthful contribution to male suicide was demonstrated by a rise in the loss of life years during 1973- 1984. Despite the inability of any investigation based on aggregate data to establish an unequivocable causal relationship, no evidence was detected to suggest that relatively high population levels of unemployment were not related to the occurrence of suicide. Keya word.y-suicide,
unemployment,
youth,
Australia
INTRODUCTION Against and
high
countries
a background
of declining
unemployment since
the
levels early
industrial
in most
1970’s,
a
activity
industrialised generation
of
young people has been confronted with deteriorating employment prospects. Increasing rates of suicide among youth, particularly males, have been noted in many countries during the last 20 years [1, 21. While these increases have been recorded mainly for industrialised countries [3], similar increases have also been observed in some nonindustrialised nations [4]. Explanations for differentials and time trends associated with increasing suicide rates have centred around the impact of social stress brought about by rapid changes in culture and way of life through economic fluctuations, as postulated by Durkheim in 1897 [5]. Conversely, factors which promote social cohesion and sense of purpose, such as war, may lower suicide rates. Ready availability of the means to commit suicide (such as firearms [6, 71; coal gas [8]; barbiturates [9]; and herbicides [IO, 1l]), may contribute to the observed rates. Imitative behaviour in certain subgroups, has also been shown to be a factor [12, 131 and there has been extensive research into the complex psychodynamics of suicide, especially among the young [ 141. The precise relationship between aggregated data on unemployment and suicide remains unclear. A comparison of unemployment and suicide rates for 1973-1983 in 16 industrialised countries suggested that increases in unemployment rates accounted for only a small component of suicide rate increases, 749
although this varied among those countries studied [3]. Another comparison on data from 23 Western countries for 19641986 detected significant correlation between unemployment and suicide for both sexes from 1974 to 1986, when high unemployment rates were the norm in Western countries, but no statistically significant correlation for the 1964-73 period when unemployment rates were low [15]. There is evidence, however, that adaptation to long periods of unemployment and increased social acceptability of high unemployment levels are significant modifiers of any suicide-unemployment association, especially in longitudinal studies [ 161. In Australia, studies of suicide have emphasised a period of relatively high levels during the 1960’s, predominantly in young and early-middle-aged women, which was associated with the easier availability of barbiturate sedatives due to changes in prescription regulations [9, 171. A cross-sectional study revealed that a high proportion of suicides in the state of Victoria were unemployed at the time of death [ 181. Another study investigated the proportions of male to female suicide in Australia (without age adjustment) during the 20th century to test a ‘modernisation’ hypothesis. Taking into account several social indicators. including changing female work force participation and divorce rates, the investigators found positive correlation between the unemployment rate and the male/female suicide ratio using time, war and a modernisation index as covariates [19]. More recent studies during the 1980’s have detected increasing suicide rates amongst young people in Australia, especially males. Martina [20] compared
5 Sri
STEPHEN MORRELL
:nit\mai smc~de rates in males aged 15-34 years for *h< period 1966 1982 with corresponding unemploynlcnt rates. His most notable finding was a high coryrlation of suicide and unemployment rates among LO 24 year old males for the 196682 period. Martina ,>roposed an underlying cohort effect of post-war !taby-boomcrs resulting in an oversupply of labour from the late 1960’s. which in turn was associated ._I ith increased suicide rates over the same period. Suicide rates among young people in rural New .-,outh Wales have increased considerably over the past 25 years [21]. The rates have doubled in larger -ural ecntres, and increased h-fold in more isolated rural areas. compared to an increase of 50% in the cities. The rural increases were ascribed to jobless11~5s.bankruptcies. isolation, easy access to firearms :ind lack of mental health facilities. Several methodological difficulties have been noted during the investigation of aggregate data on suicide. Problems have existed with the diagnosis of suicide, and wjhether the registering of suicides represents an under-reporting of actual rates [22]. Hidden suicide, especially involving motor vehicle accidents [23] and illegal drug overdoses [24], is not possible to quantify. High iisk bchaviour, amongst youth in particular. :a11 seriously obscure suicide. Suicide rates are thus i..nlikely to be overestimated. Additionally, there are difficulties in accounting for changing attitudes to suicide over long periods, which may alter recognition and labelling of the event. Rigorous investigations of contemporaneous relationships between unemployment and health outcomes were initiated by Morris and Titmuss in 1944 [7_5].They demonstrated that death rates from rheumatic heart disease were significantly related to unemployment levels in England and Wales during the Great Depression of the 1930’s. Their approach was to relate the degree of change in rates for unemploy:nent with those for mortality from rheumatic fever .md to test a lag hypothesis for particular temporal relationships. Origins of rheumatic fever, however, are Xnown to be influenced by poverty and overcrowding. pi hese factors increased in centres of high unemployment nnd then subsequently decreased as conditions of full employment returned during the 1940’s. Comparable studies on the rise of unemployment during the 1970’s have been dominated by the work of M. H. Brenner. However, unlike Morris and Titmuss, Brenner investigated a range of morbidity and mortality outcomes (including suicide [26-301) whose origins were complex and necessitated the hypothesised involvement of economic stress. Accordingly, Brenner performed time-series analyses oil aggregate data using a ‘polynomial distributed lag method’ based on the assumption that there was probably no single discrete lag between economic change and mortality rates for a population. In addition, he attempted to account for different agerelated susceptibilities to disease and the influence of known risk factors to health.
rl al.
Application of Brenner’s methodology has generated considerable controversy on grounds that include uncertainties over correct assignments of lag times for explanatory variables [31-331; the exclusion of non-economic influences on outcomes [34]; and general difficulties with time-series analyses involving the so-called ‘time series fallacy’ of construing causation from tortuous lagging of explanatory and outcome variables, measurement error, multicollinearity, simultaneity and different responses to given economic conditions [35]. It is clear that both quantitative and qualitative (historical and social) methods are needed because there is no simple relationship between quantitative economic indices and the suicide rate For instance, even if sufficiently detailed socio-economic information was available and spanned an appropriately long period, such data cannot adequately account for changes in cultural attitudes. patterns of living, individual adaptive mechanisms, access to lifeterminating technology and other factors that casestudies indicate as relevant to the origins of suicide. Accordingly, the investigation recorded here incorporates both methodological strategies in an aggregate/ecological examination of the possible contemporaneous relationship between suicide and unemployment rates. Utilizing Australian records from 1907 to 1990. secular trends in suicide. stratified by gender and age. were investigated. Emphasis is on suicide of young persons because of the consequent loss of relatively many more years of life by comparison with suicide in later years, In particular, secular trends in suicide and unemployment rates by age group are examined to determine the validity with which the two may be related in a causal hypothesis. METHODS
Data on suicides, population characteristics and unemployment were obtained from different sources. Annual numbers of suicides in Australia. stratified by gender and 5-year age group from 1920 to 1990, were obtained from the Australian Institute of Health which, in turn, obtained the data from the Australian Bureau of Statistics (ABS). Suicide data prior to 1920 were obtained from Commonwealth of Australia Yearbooks and ABS records. All mortality data are ultimately derived from state Registrars of Births, Deaths and Marriages. Annual population data stratified by gender and age were obtained from standard ABS sources, including census records. Intercensal interpolations were used by the ABS for yearly population estimates. Historical unemployment data were obtained from ABS records, Withers et al. [36] and Australians, Historical Statistics [37]. These data presented some problems because unemployment rates in Australia have not been uniformly collected, and therefore the overall historical trend had to be synthesised
Suicide and unemployment in Australia 1907-1990 from different sources. Currently, three separate sources of unemployment statistics exist in Australia [38]. The resulting historical unemployment curve appears to be reasonably accurate, although quantitatively imperfect, for the period before 1960. Annual unemployment rates by gender and age are available only from 1966 onwards. Male and female suicide rates were directly agestandardised using a standard population comprising averages of age-specific (Syear age brackets) male and female populations taken over the 1921-1987 period. Calculation of years of life lost from suicide was based on the preretirement (O-65 years) population, and, because suicide occurs very rarely before age 15, indicates years of actual or potential working life lost. Population increase and changing age structure of the O-65 male and female population was taken into account by using weightings based on the standard population. The following formula was used for each year: C [xi * (65~ma,) * Psi/Pi] * 100,000 z(<65
POP)
where xi is the number of deaths in the ith 5 year age group and mai is the mid-age for the ith 5 year age group. The weighting factor is Psi/P,, where P,, is the ratio of the ith age specific standard population to the total under-65 years standard population, and P,the ratio of the ith age-specific population to the total under-65 years population for the given year. The result represents the number of life years lost per 100,000 male or female population under 65 years of age, standardised for age. It should be noted that the estimates of pre-retirement years of life lost underestimate total years of life lost because of increasing life expectancies experienced by the Australian
751
population this century. Age-adjusted and age-specific suicide rates were regressed with corresponding unemployment rates using methods taking into account serial autocorrelation of time series data. Because the number of observations of age-specific unemployment rates in the present study is small (n = 25), the commonly used Cochrane-Orcutt iterative procedure was rejected because it omits the first (transformed) observation in the regression estimation. In addition, the processes underlying the residual terms from the initial ordinary least squares regression were first identified using autoregressive integrated moving average (ARIMA) techniques [39]. This technique was used because time series data should be tested for the order of serial autocorrelation before routinely applying estimated generalised least squares methods such as the Cochrane-Orcutt procedure. The DurbanWatson statistic, for example, does not indicate whether the autocorrelation is higher than first order. The main motivation for analysing residual terms in this way, however, was to shift statistically derived lagging of explanatory variables to the residual terms so as to maintain the explanatory function of the model. The method of significance estimation used was based on estimating the disturbance covariance matrix [40]. RESULTS
Annual age-adjusted male and female suicide rates and annual overall unemployment rates for 1907-90 are illustrated in Fig. 1. Female suicide rates demonstrated remarkable stability at about 4.5-5 deaths per 100,000 up to the rise attributable to the poisoning/ overdose barbiturate epidemic of the early-tomid-1960’s; since that time the rates have gradually declined, but are still above their former level, at
35
35 Great
30 World
Depression .t :;
-30
25
a, P
-20 i $
Year
Fig. 1. Male and female age standardised suicide rates and unemployment rates, 1907-1990. SSM 36,6--D
STEPHEN MORRELL et a[.
752
around 5.556 deaths per 100,000. The amount of variance in male and female suicide rates (equivalent to the correlation I?‘) explained by the overall unemployment rate for the 1907-1989 period is low: R' = 18% for males and R' = 11% for females. Male suicide rates during this century are characterised by sharp fluctuations compared to those for females. There were decreases across all age groups during both World Wars, with the lowest rate occurring during World War II (approx. 9.5 per 100,000). The century highpoint was the Great Depression (25.5). There was a suicide peak during the early 1960’s (2 1.O) which had similarities to the female suicide epidemic at around the same time. Since 1984 a sharp increase has occurred, reaching about 20.5 per 100,000 in 1987. and by 1990 stood at 19.0. The highest male suicide rate occurred during the Great Depression when unemployment levels were higher than at any time since the depression of the 1890’s. There also appear to be peaks in suicide rates in 1912~~1915 and 1920-21, coinciding with peaks of unemployment. These suicide peaks occurred chiefly among older males in the ranges of 50-54 years, 4549 years and 40-44 years, in decreasing order of magnitude. Suicide in males of 15-19 years and 2&24 years. and in females of all ages. remained relatively stable during these unemployment peaks. According to the 1933 Australian Census [41], unemployed 15-24 year olds comprised 29.4% of total unemployed males during the Great Depression, compared to 42.5% in 1989. Unemployed 45-54 year old males comprised 15.5% of the male total in 1933 compared to 9.0% in 1989. In terms of the 1933 male population age structure, these figures become 45.8% and 9.5% respectively. Historical studies of individ-
ual Australian states confirm this Great Depression phenomenon of unemployment affecting older age groups disproportionately compared to the current time [42]. Secular changes in age-specific suicide rates reveal the changing underlying age structure of contemporary suicide rates. As Figs 2 and 3 indicate, the decline in older age suicide rates since the mid-to-late-1960’s has been matched by increases in the 15-24 year age groups; more so among 20-24 year olds than 15-19 year olds, and more in males than females. Suicide rates among 20.-24 year old males reached their highest point this century in 1990 (approx. 36.1 deaths per 100,000). Additionally, by 1990 20-24 year old males were committing suicide at a higher rate than every other age group, with the exception of males 85 years and older. The relative stability in overall male age-adjusted suicide rates from 1974 to 1984 masks increasing rates amongst 15524 year olds. From 1985 to 1990 suicide increases occurred across all male age groups. There is no evidence of cohort effects in suicide rates for the twentieth century for any age group. male or female. Adjusted life years lost (O--65 years) from suicide in males and females is given in Fig. 4. Female life years lost from suicide were relatively stable until the early 1960’s when the total doubled and then slowly declined to former levels by the early-to-mid 1980’s. Annual male life years lost from suicide peaked in 1913-14, during the Great Depression, the early 1960’s, and with the highest peak occurring in 1987. The increasingly young component of male suicide rates is reflected in the rise in lost life years during 1973-84. Increased suicide rates across most male age groups from the mid-1980’s to 1988 are apparent in the 1987788 century maximum of lost life years.
70 age groups (yrs)
- .-. 40-64
60
----
I”‘~~~~~‘~‘~~‘~l~“~““‘l”“‘~“““~I’~”””’I”~”””I”’~“‘~‘I 1910 1920 1930 1940 1950 1960 Year
1970
Fig. 2. Male suicide, age specific rates, 1907-1990
20-24 15-19
1980
1990
Suicide and unemployment in Australia 1907-1990
,Oi 25 -
age groups(yrs) -...-.. 40-64 20-24
----
1 ;..+ ‘\ _
15-19
i!+j : i Ii
0 ’
753
i I 1 1.h i! j
~~~,~~~~~~~~~,~~l~ll~ll~llllllll’~~~~~~~~~’~”””“‘~““““‘~“”’~~~~~~‘~~~
1910
1920
1930
1940
1950
1960
1970
1980
1990
Year Fig. 3. Female suicide, age specific rates, 1907-1990. Unemployment and suicide rates among male 20-24 year olds for 196690 with accompanying trend lines are shown in Fig. 5. The values obtained from regressing 20-24 year old male suicide rates on the corresponding unemployment rate were 15.38 for the intercept and 1.13 for the 20-24 year male unemployment rate. While the adjusted-R’ of 2&24 male suicide and unemployment rates is 70%, the Durbin-Watson statistic (D = 0.86 for 25 observations) indicates significant auto-correlation. A first order autoregressive and zero order moving average process (with zero differencing) [ARIMA (l,O,O)] gives a reasonable fit to the residual term. The
estimated significance of the coefficient of the original regression remains at P < 0.01, df = 2,23, after taking autocorrelation into account. The linear increase in male youth unemployment (20-24) from 196690 matches reasonably closely the linear increase in the male suicide rate in this age group. If 20-24 year male unemployment rates are taken as a ratio of the overall unemployment rate, and 2G24 year male suicide rates as a proportion of the male age standardised suicide rate, the R2 value increases to 80%. Figure 6 shows the closeness of these two proportions over the 19661990 period.
___
$ 500% Z $ 5 E g $ a ;j = ? Z x 2
400-
300.
200-
loo-
0’
“‘,““““‘,““““‘~““““‘~““““‘l”~“l””””’l””~~~’~~~~~~~
1910
1920
1930
1940
1950 Year
1960
1970
1980
1990
Fig. 4. Life-years lost from male and female suicide, 1907-1990 age standardised.
754
STEPHENMORRELL etal.
-
suicide rate
. unemployment rate
~
trend
~
1966
1970
1975
1960
1965
1990
year Fig.
5. Male
suicide
and unemployment,
DISCUSSION
2&24
year olds,
196C1990.
(R2= 60%, for 1624 year olds) and Canada (R2= 58%) follow Australia. In contrast, Sweden, U.S.A.
The association between suicide and unemployment rates in Australia has been shown to vary, depending on period, age, gender and geography [ 18-211. Findings from the present study reinforce the temporal and age variation in the relationship. Comparison with other OECD countries indicates that the suicide-unemployment association among 15-24 year old males in Australia (for the 196687 period) is comparatively high [42]. The strong link (R2= 68%) for this age grouping in Australia is exceeded only by France (R2= 77%, 1968-87). The
Japan and Germany exhibit correspondingly low associations, with R2 = 0.03, 0.01, and 0.005 respectively. Low unemployment levels among youth in these three economically successful countries during the last two decades, coupled with strong familial networks (Japan) or adequate social welfare systems (Sweden and Germany), may contribute to this disparity. The responsiveness of male suicide rates to unemployment levels compared to females is also an international phenomenon [43]. The highest percentage of variance in suicide explained by unemploy-
2.0
1 1.5
1.0 20-24 year male suicide rate as proportion of male age standardised suicide rate 20-24 year male unemployment rate as proportion of overall unemployment rate
0.5
I
1966
8
I
b
I
1970
t
I
I
I
I
I
1
I
I
I
1
I
1960
1975
I
I
I
1965
I
I
I
I
I
1990
year
Fig. 6. Male suicide and unemployment rate proportions,
20-24 year olds, 1966-1990
Suicide and unemployment in Australia 19074990 ment in Australia has occurred in 20-24 year old males for the period 19661990 (Rz = 70%). This occurs without taking into account cohort effects [20], modernisation indices [19], or other social variables. The R2 value actually increased since Martina’s study [20] to 78% for 1966-1987. Thus for 2&24 year old Australian males the association between suicide and unemployment has varied slightly, but remains strong. The other feature to note in these trends is a similarly strong association of suicide among males aged 25-34 with their corresponding unemployment levels for the 19661990 period (R2 = 53%). Unemployment amongst 15-19 year old male youth is 50% higher than in 20-24 year old males, yet their suicide levels remain lower. However, larger proportions of 15-19 than 2&24 year olds still live with their parents, and therefore have the benefit of a stronger network of familial social support. Furthermore, frustrated employment/career expectations, the need for financial independence and other societal pressures would conceivably be less significant in the younger group compared to 20-24 year olds. Using Durkheim’s hypothesis, it could be argued that sharp changes in unemployment rates influence suicide rates, rather than high or low unemployment levels as such., Very low correlations, between agespecific and age-adjusted suicide rates, or changes in these, and changes in corresponding unemployment rates, do not support this argument. Nevertheless, as indicated in Fig. 6, the link between suicide and unemployment may be more significant when considered as age-specific proportional deviations from respective overall rates. This would suggest that cross-sectional proportional differences from the contemporaneous social norm have more long term bearing on rates of suicide than diachronous changes in social conditions. It should be noted, moreover, that Durkheim formed his views under the belief that suicide rates would be consistently higher among older males. The present study demonstrates that this is no longer the case. The fact that 20-24 year old male suicide rates exceeded all other age groups (excepting over 85 year olds) in 1990, and that this is an accrual of a quarter century trend, clearly contradicts Durkheim’s age thesis. As can be seen from Fig. 2, increased suicide rates during the Great Depression years were experienced more by middle-aged men, 45-54 year olds in particular, whereas male youth suicide rates were hardly affected. While age-specific unemployment rates are not available for this period, historical accounts of unemployment in the Great Depression suggest that older Australians were affected more by unemployment in the 1930’s than younger people [44]. By contrast, since the early 1970’s, levels of suicide among older male age groups have declined, despite historically high overall unemployment rates. However, middle-aged males have experienced unemployment levels at approximately half the overall rate during this period [42].
755
Although the present study revealed no cohort effects for twentieth century suicides, Martina’s birthrate cohort hypothesis [20] is given some credence because the postwar baby boom in Australia had peaked by the early 1960’s and unemployment rates for 20-24 year old males had peaked by the early 1980’s. However, this hypothesis negates the influence of economic fluctuations, as exemplified by present (late 1991) levels of unemployment in Australia being the highest since the Great Depression. Since time series data can exhibit autoregressive processes amongst both explanatory and outcome variables, parameter estimates in classical regression analyses are often both biased and unreliable. Simple regression of time trend data is not a valid method for estimating precise relationships between different time series because, in general, the assumption of non-autocorrelation of residual terms from classical regression models is violated. However, if the residual terms from these regressions display first order autoregressive behaviour then the regression parameter estimates are in fact unbiased and reliable, but the significance levels are exaggerated. Most socioeconomic time series exhibit first order autoregressive behaviour [45], but this feature cannot be assumed in advance and should be investigated. Time series techniques, such as ARIMA (BoxJenkins) modelling and construction and analysis of transfer functions, may serve a useful purpose in forecasting, especially with processes under experimental control, but they offer no assistance in modelling historical data with a view to achieving a priori explanatory power. The current study explored the plausibility of contemporaneous relationship between unemployment levels and suicide rates, particularly in 20-24 year old males. This was undertaken by modelling the autoregressive behaviour of residual terms, without resorting to lag times in the explanatory variables themselves. Such an approach avoids the pitfalls of post hoc explanations of statistically derived lag times in explanatory and outcome variables, in effect recognising that a myriad of unaccountable social and personal factors act as chance influences on aggregate suicide rates away from the regression line. Correlation of time trends of variables as a method of causal analysis indicates whether it is plausible to hypothesise that the factors may be linked in a cause and effect relationship. This method is particularly useful if both factors increase and decrease in unison during the same period, thus providing stronger evidence of direct causation than do progressive trend increases or decreases occurring in both parameters. While the aggregate data show that the suicide/unemployment relationship is not a simple year-by-year correlation, the present study strongly supports the hypothesis that unemployment is significant as a predisposing factor for increasing the risk of suicide, especially in males. While no aggregate
156
STEPHEN MORRELL
study can unequivocally establish causation, there was no evidence from Australian records to suggest that unemployment was not related to suicide. The issue of suicide and unemployment is worthy of further detailed analyses so that one set of destructive consequences associated with unemployment can be emphasised to maintain a political agenda committed to the protection of vulnerable groups in times of economic recession.
REFERENCES I. Boor M. The relationship between unemployment rates and suicide rates in eight countries, 1962-1976. Psycho/ Rep. 47, 1095-l 101, 1980. 2. Diekstra R. Suicidal behaviour in adolescents and young adults: the international picture. Crisis 10, 16-35, 1989. 3. Crombie I. K. Can changes in the unemployment rates explain the recent changes in suicide rates in developed countries? ht. J. Epidemiol. 19, 412416, 1990. 4. Rubinstein D. Epidemic suicide among Micronesian adolescents, Sot. Sci. Med. 17, 657-665, 1983. 5. Durkheim E. Suicide. A Stud?) in Sociologic, 1x9’97. Routledge & Kegan Paul, London, 1970. 6. Lester D. and Murrell M. The influence of gun control laws on suicidal behavior. Am. J. Psychiat. 137, 121-122, 1980. 7. Lester D. and Murrell M. The preventative effect of strict gun control laws on suicide and homicide. Suicide Life Threat Eehav. 12, 131-140, 1982. 8. Kreitman N. and Plan S. Suicide, unemployment and domestic gas detoxification in Britain. J. Epidemiol. Community Hlth 38, 14, 1984. 9. Oliver R. G. and Hetzel B. S. The rise and fall of suicide rates in Australia: Relation to sedative availability. Med. J. Aust. 2. 919-923, 1972. 10. Parshu R. A. M. and Uma R. A. 0. Paraquat poisoning in Fiji. Fui Med. J. Jan/Feb. 12-16, 1983. 11. Mowbray D. L. Pesticide poisoning in Papua New Guinea and the South Pacific. PNG. Med. J. 29, 131-141, 1986. 12. Phillips D. L. The influence of suggestion on suicide: subs&tive and theoretical implications of the Werther effect. Am. Social. Rec. 39, 340-354, 1974. 13. Phillips D. L. Suicide, motor vehicle fatalities and the mass media: Evidence toward a theory of suggestion. Am. J. Social. 84, 1150-1174, 1979. 14. Diekstra R. F. W. and Hawton K. (Eds) Suicide in Adolescence. Martinus Nijhoff, Dordrecht, 1987. 15. Pritchard C. Suicide, unemployment and gender variations in the Western world 19661986. Are women in Anglo-phone countries protected from suicide? Sot. Psychiat. Psychiatr. Epidemiol. 25, 73-80. 1990. 16. Warr P. and Jackson P. Adapting to the unemployed role: A longitudinal investigation. Sot. Sci. Med. 25, 1219-1224, 1987. 17. Burvill P. W. Changing _ __patterns of suicide in Australia, 191@1977. Acta psychiatr. stand. 62, 258-268, 1980. 18. Kessel A.. Nicholson A.. Graves G. and Krupinski J. Suicidal attempts in an ‘outer region of metropolitan Melbourne and in a provincial region of Victoria. ANZ J. Psychiatr. 9, 255, 1975. 19. Hassan R. and Tan G. Changing patterns of suicide in Australia, 1901-1985: An analysis of sex differentials. Suicide Life Threat Behao. 19, 362-380, 1989. 20. Martina A. Suicide and unemployment amongst young Australian males 1966 to 1982. Working Papers in Economic History. Australian National University, No. 56, pp. l-17, 1985.
et al
21. Dudley M., Waters B., Kelk N. and Howard J. Youth suicide in NSW: urban-rural trends 19641988. Med. J. Aust. 156, 83-88, 1992. of suicide. Epidem. Rev. 9, 22. Monk M. Epidemiology 51-69, 1987. 23. Olsen J. and Lajer M. Violent death and unemployment in two trade unions in Denmark. Sot. Psychiat. 14, 139-145, 1979. and illegal 24. Peck D. F. and Plant M. A. Unemployment drug use: concordant evidence from a prospective study and national trends. Er. Med. J. 293, 929-932. 1986. 25. Morris J. N. and Titmuss R. Health and social change: recent history of rheumatic heart disease. Med. Officer 2, 69-71, 77-79, 85-87, 1944. M. H. Mental Illness and the Econom?,. 26. Brenner Harvard University Press, Cambridge, MA, 1973. 27. Brenner M. H. Health costs and benefits of economic policy. ht. J. Hlth Sert?. 7, 581 -622, 1977. and 28. Brenner M. H. and Mooney A. Unemployment health in the context of economic change. Ser. Sci. Med. 17, 1125-1138, 1983. M. H. Mortality and economic stability: 29. Brenner Detailed analysis for Britain and comparative analysts for selected industrial countries. Int. J. Hlth Sew. 13, 563 420, 1983. 30. Brenner M. H. Relation of economic change to Swedish health and social well-being, 195(&1980. Sot. &i. Med. 25, 183. 195, 1987. 31. Eyer J. Review of mental illness and the economy. Iit. J. Hlth Serv. 1, 139-155, 1976. 32. Ever J. Reioinder to Dr Brenner. Inr. J. Hlth Sew. 1. 1;7-168, 1$76. G. and Stern J. Mortality 33. Gravelle H. S. E., Hutchinson and unemployment: A critique of Brenner’s time-series analysis. ,k&et 2, 675-679; 1981. J. R. and Funch D. P. Mental illness and 34. Marshall the economy: A critique and partial replication. J. HI/h Sot. Behar. 20, 282-289, 1979. 35. Gravelle H. S. E. Time series analysis of mortality and unemployment (Editorial). J. H//h Econ. 3, 297-305, 1984. 36. Withers G., Endres T. and Perry L. Australiun Historical Statistics.. L&our Statistics, Source Papers in Economic History No. 7. Australian National University, 1985. W. (Ed.) Austruliuns, Historical Stuti.strc.~. 37. Vamplew Fairfax, Syme & Weld, Broadway, NSW. 1987. 38. Smith D. J. and Callaghan A. H. Definitional aspects of measurement of unemployment. A comparative study of the Australian longitudinal survey and other sources of youth unemployment data. Aust. J. Stati.~/. 31A, 9%24, 1989. 39. Box G. E. P. and Jenkins G. M. Time Serie.s Ana/y.vi.s, Forecasting and Control. Holden Day. San Fransisco. 1976. J. and Weber N. C. Esti40. Keener R. W., Kmenta mation of the covariance matrix of the least squares regression coefficients when the disturbance covariance matrix is of unknown form. Econometric Theory 7. 2245. 1991. 41. Census of the Austrohcrn Population and Housing, p. Il. Australian Bureau of Statistics, 1933. Cooperation and Development. 42. Office for Economic Labour Force Statistics 1964-1984, and 1967- 1987. OECD, Washington DC, 1986, 1989. World Health Statistics 43. World Health Organisation. Annual, 1966-1989. WHO, Geneva. workers. A .Socicrl Hisfor), 44. Broomhill R. Unemployed of‘ the Great Depression in Adelaide. University of Queensland Press, St Lucia, 1978. 45. Ostrom C. W. Time Series Analysis Regression Tech n&es, 2nd edn. Sage, Newbury Park, 1990.
SUICIDE
AND
0277-9536193 $6.00+ 0.00 PergamonPressLid
UNEMPLOYMENT
IN AUSTRALIA
STEPHENMORRELL,RICHARDTAYLOR,SUSANQUINE Department
of Public
Health,
Faculty
of Medicine,
University
and
1907-1990
CHARLESKERR
of Sydney,
N.S.W.
2006. Australia
Abstract--Ever since Durkheim postulated a relationship between economic change and suicide there has been evidence of a general association between aggregate data on unemployment and the frequency of suicide. Quantitatively, however, the association has been variable and it is clear that due to differing cultural, social and individual determinants of suicide, the relationship is complex. Methodological difficulties abound with interpretation of aggregate data. Australian records for most of the present century are suitable for examining secular trends in suicide and unemployment by age group and gender to gain an indication of the extent to which both parameters may be causally related. An aggregate/ecological study was designed to incorporate quantitative and qualitative strategies. Annual age-adjusted male and female suicide rates and annual unemployment rates were derived for the period 1907-1990. Female suicide rates were generally stable throughout the period, whereas those for males demonstrated sharp fluctuations with the peaks coinciding with times of high unemployment. The association between suicide and unemployment for 15524 year old males was comparatively high for the recent period, 1966-1990. The increasingly youthful contribution to male suicide was demonstrated by a rise in the loss of life years during 1973- 1984. Despite the inability of any investigation based on aggregate data to establish an unequivocable causal relationship, no evidence was detected to suggest that relatively high population levels of unemployment were not related to the occurrence of suicide. Keya word.y-suicide,
unemployment,
youth,
Australia
INTRODUCTION Against and
high
countries
a background
of declining
unemployment since
the
levels early
industrial
in most
1970’s,
a
activity
industrialised generation
of
young people has been confronted with deteriorating employment prospects. Increasing rates of suicide among youth, particularly males, have been noted in many countries during the last 20 years [1, 21. While these increases have been recorded mainly for industrialised countries [3], similar increases have also been observed in some nonindustrialised nations [4]. Explanations for differentials and time trends associated with increasing suicide rates have centred around the impact of social stress brought about by rapid changes in culture and way of life through economic fluctuations, as postulated by Durkheim in 1897 [5]. Conversely, factors which promote social cohesion and sense of purpose, such as war, may lower suicide rates. Ready availability of the means to commit suicide (such as firearms [6, 71; coal gas [8]; barbiturates [9]; and herbicides [IO, 1l]), may contribute to the observed rates. Imitative behaviour in certain subgroups, has also been shown to be a factor [12, 131 and there has been extensive research into the complex psychodynamics of suicide, especially among the young [ 141. The precise relationship between aggregated data on unemployment and suicide remains unclear. A comparison of unemployment and suicide rates for 1973-1983 in 16 industrialised countries suggested that increases in unemployment rates accounted for only a small component of suicide rate increases, 749
although this varied among those countries studied [3]. Another comparison on data from 23 Western countries for 19641986 detected significant correlation between unemployment and suicide for both sexes from 1974 to 1986, when high unemployment rates were the norm in Western countries, but no statistically significant correlation for the 1964-73 period when unemployment rates were low [15]. There is evidence, however, that adaptation to long periods of unemployment and increased social acceptability of high unemployment levels are significant modifiers of any suicide-unemployment association, especially in longitudinal studies [ 161. In Australia, studies of suicide have emphasised a period of relatively high levels during the 1960’s, predominantly in young and early-middle-aged women, which was associated with the easier availability of barbiturate sedatives due to changes in prescription regulations [9, 171. A cross-sectional study revealed that a high proportion of suicides in the state of Victoria were unemployed at the time of death [ 181. Another study investigated the proportions of male to female suicide in Australia (without age adjustment) during the 20th century to test a ‘modernisation’ hypothesis. Taking into account several social indicators. including changing female work force participation and divorce rates, the investigators found positive correlation between the unemployment rate and the male/female suicide ratio using time, war and a modernisation index as covariates [19]. More recent studies during the 1980’s have detected increasing suicide rates amongst young people in Australia, especially males. Martina [20] compared
5 Sri
STEPHEN MORRELL
:nit\mai smc~de rates in males aged 15-34 years for *h< period 1966 1982 with corresponding unemploynlcnt rates. His most notable finding was a high coryrlation of suicide and unemployment rates among LO 24 year old males for the 196682 period. Martina ,>roposed an underlying cohort effect of post-war !taby-boomcrs resulting in an oversupply of labour from the late 1960’s. which in turn was associated ._I ith increased suicide rates over the same period. Suicide rates among young people in rural New .-,outh Wales have increased considerably over the past 25 years [21]. The rates have doubled in larger -ural ecntres, and increased h-fold in more isolated rural areas. compared to an increase of 50% in the cities. The rural increases were ascribed to jobless11~5s.bankruptcies. isolation, easy access to firearms :ind lack of mental health facilities. Several methodological difficulties have been noted during the investigation of aggregate data on suicide. Problems have existed with the diagnosis of suicide, and wjhether the registering of suicides represents an under-reporting of actual rates [22]. Hidden suicide, especially involving motor vehicle accidents [23] and illegal drug overdoses [24], is not possible to quantify. High iisk bchaviour, amongst youth in particular. :a11 seriously obscure suicide. Suicide rates are thus i..nlikely to be overestimated. Additionally, there are difficulties in accounting for changing attitudes to suicide over long periods, which may alter recognition and labelling of the event. Rigorous investigations of contemporaneous relationships between unemployment and health outcomes were initiated by Morris and Titmuss in 1944 [7_5].They demonstrated that death rates from rheumatic heart disease were significantly related to unemployment levels in England and Wales during the Great Depression of the 1930’s. Their approach was to relate the degree of change in rates for unemploy:nent with those for mortality from rheumatic fever .md to test a lag hypothesis for particular temporal relationships. Origins of rheumatic fever, however, are Xnown to be influenced by poverty and overcrowding. pi hese factors increased in centres of high unemployment nnd then subsequently decreased as conditions of full employment returned during the 1940’s. Comparable studies on the rise of unemployment during the 1970’s have been dominated by the work of M. H. Brenner. However, unlike Morris and Titmuss, Brenner investigated a range of morbidity and mortality outcomes (including suicide [26-301) whose origins were complex and necessitated the hypothesised involvement of economic stress. Accordingly, Brenner performed time-series analyses oil aggregate data using a ‘polynomial distributed lag method’ based on the assumption that there was probably no single discrete lag between economic change and mortality rates for a population. In addition, he attempted to account for different agerelated susceptibilities to disease and the influence of known risk factors to health.
rl al.
Application of Brenner’s methodology has generated considerable controversy on grounds that include uncertainties over correct assignments of lag times for explanatory variables [31-331; the exclusion of non-economic influences on outcomes [34]; and general difficulties with time-series analyses involving the so-called ‘time series fallacy’ of construing causation from tortuous lagging of explanatory and outcome variables, measurement error, multicollinearity, simultaneity and different responses to given economic conditions [35]. It is clear that both quantitative and qualitative (historical and social) methods are needed because there is no simple relationship between quantitative economic indices and the suicide rate For instance, even if sufficiently detailed socio-economic information was available and spanned an appropriately long period, such data cannot adequately account for changes in cultural attitudes. patterns of living, individual adaptive mechanisms, access to lifeterminating technology and other factors that casestudies indicate as relevant to the origins of suicide. Accordingly, the investigation recorded here incorporates both methodological strategies in an aggregate/ecological examination of the possible contemporaneous relationship between suicide and unemployment rates. Utilizing Australian records from 1907 to 1990. secular trends in suicide. stratified by gender and age. were investigated. Emphasis is on suicide of young persons because of the consequent loss of relatively many more years of life by comparison with suicide in later years, In particular, secular trends in suicide and unemployment rates by age group are examined to determine the validity with which the two may be related in a causal hypothesis. METHODS
Data on suicides, population characteristics and unemployment were obtained from different sources. Annual numbers of suicides in Australia. stratified by gender and 5-year age group from 1920 to 1990, were obtained from the Australian Institute of Health which, in turn, obtained the data from the Australian Bureau of Statistics (ABS). Suicide data prior to 1920 were obtained from Commonwealth of Australia Yearbooks and ABS records. All mortality data are ultimately derived from state Registrars of Births, Deaths and Marriages. Annual population data stratified by gender and age were obtained from standard ABS sources, including census records. Intercensal interpolations were used by the ABS for yearly population estimates. Historical unemployment data were obtained from ABS records, Withers et al. [36] and Australians, Historical Statistics [37]. These data presented some problems because unemployment rates in Australia have not been uniformly collected, and therefore the overall historical trend had to be synthesised
Suicide and unemployment in Australia 1907-1990 from different sources. Currently, three separate sources of unemployment statistics exist in Australia [38]. The resulting historical unemployment curve appears to be reasonably accurate, although quantitatively imperfect, for the period before 1960. Annual unemployment rates by gender and age are available only from 1966 onwards. Male and female suicide rates were directly agestandardised using a standard population comprising averages of age-specific (Syear age brackets) male and female populations taken over the 1921-1987 period. Calculation of years of life lost from suicide was based on the preretirement (O-65 years) population, and, because suicide occurs very rarely before age 15, indicates years of actual or potential working life lost. Population increase and changing age structure of the O-65 male and female population was taken into account by using weightings based on the standard population. The following formula was used for each year: C [xi * (65~ma,) * Psi/Pi] * 100,000 z(<65
POP)
where xi is the number of deaths in the ith 5 year age group and mai is the mid-age for the ith 5 year age group. The weighting factor is Psi/P,, where P,, is the ratio of the ith age specific standard population to the total under-65 years standard population, and P,the ratio of the ith age-specific population to the total under-65 years population for the given year. The result represents the number of life years lost per 100,000 male or female population under 65 years of age, standardised for age. It should be noted that the estimates of pre-retirement years of life lost underestimate total years of life lost because of increasing life expectancies experienced by the Australian
751
population this century. Age-adjusted and age-specific suicide rates were regressed with corresponding unemployment rates using methods taking into account serial autocorrelation of time series data. Because the number of observations of age-specific unemployment rates in the present study is small (n = 25), the commonly used Cochrane-Orcutt iterative procedure was rejected because it omits the first (transformed) observation in the regression estimation. In addition, the processes underlying the residual terms from the initial ordinary least squares regression were first identified using autoregressive integrated moving average (ARIMA) techniques [39]. This technique was used because time series data should be tested for the order of serial autocorrelation before routinely applying estimated generalised least squares methods such as the Cochrane-Orcutt procedure. The DurbanWatson statistic, for example, does not indicate whether the autocorrelation is higher than first order. The main motivation for analysing residual terms in this way, however, was to shift statistically derived lagging of explanatory variables to the residual terms so as to maintain the explanatory function of the model. The method of significance estimation used was based on estimating the disturbance covariance matrix [40]. RESULTS
Annual age-adjusted male and female suicide rates and annual overall unemployment rates for 1907-90 are illustrated in Fig. 1. Female suicide rates demonstrated remarkable stability at about 4.5-5 deaths per 100,000 up to the rise attributable to the poisoning/ overdose barbiturate epidemic of the early-tomid-1960’s; since that time the rates have gradually declined, but are still above their former level, at
35
35 Great
30 World
Depression .t :;
-30
25
a, P
-20 i $
Year
Fig. 1. Male and female age standardised suicide rates and unemployment rates, 1907-1990. SSM 36,6--D
STEPHEN MORRELL et a[.
752
around 5.556 deaths per 100,000. The amount of variance in male and female suicide rates (equivalent to the correlation I?‘) explained by the overall unemployment rate for the 1907-1989 period is low: R' = 18% for males and R' = 11% for females. Male suicide rates during this century are characterised by sharp fluctuations compared to those for females. There were decreases across all age groups during both World Wars, with the lowest rate occurring during World War II (approx. 9.5 per 100,000). The century highpoint was the Great Depression (25.5). There was a suicide peak during the early 1960’s (2 1.O) which had similarities to the female suicide epidemic at around the same time. Since 1984 a sharp increase has occurred, reaching about 20.5 per 100,000 in 1987. and by 1990 stood at 19.0. The highest male suicide rate occurred during the Great Depression when unemployment levels were higher than at any time since the depression of the 1890’s. There also appear to be peaks in suicide rates in 1912~~1915 and 1920-21, coinciding with peaks of unemployment. These suicide peaks occurred chiefly among older males in the ranges of 50-54 years, 4549 years and 40-44 years, in decreasing order of magnitude. Suicide in males of 15-19 years and 2&24 years. and in females of all ages. remained relatively stable during these unemployment peaks. According to the 1933 Australian Census [41], unemployed 15-24 year olds comprised 29.4% of total unemployed males during the Great Depression, compared to 42.5% in 1989. Unemployed 45-54 year old males comprised 15.5% of the male total in 1933 compared to 9.0% in 1989. In terms of the 1933 male population age structure, these figures become 45.8% and 9.5% respectively. Historical studies of individ-
ual Australian states confirm this Great Depression phenomenon of unemployment affecting older age groups disproportionately compared to the current time [42]. Secular changes in age-specific suicide rates reveal the changing underlying age structure of contemporary suicide rates. As Figs 2 and 3 indicate, the decline in older age suicide rates since the mid-to-late-1960’s has been matched by increases in the 15-24 year age groups; more so among 20-24 year olds than 15-19 year olds, and more in males than females. Suicide rates among 20.-24 year old males reached their highest point this century in 1990 (approx. 36.1 deaths per 100,000). Additionally, by 1990 20-24 year old males were committing suicide at a higher rate than every other age group, with the exception of males 85 years and older. The relative stability in overall male age-adjusted suicide rates from 1974 to 1984 masks increasing rates amongst 15524 year olds. From 1985 to 1990 suicide increases occurred across all male age groups. There is no evidence of cohort effects in suicide rates for the twentieth century for any age group. male or female. Adjusted life years lost (O--65 years) from suicide in males and females is given in Fig. 4. Female life years lost from suicide were relatively stable until the early 1960’s when the total doubled and then slowly declined to former levels by the early-to-mid 1980’s. Annual male life years lost from suicide peaked in 1913-14, during the Great Depression, the early 1960’s, and with the highest peak occurring in 1987. The increasingly young component of male suicide rates is reflected in the rise in lost life years during 1973-84. Increased suicide rates across most male age groups from the mid-1980’s to 1988 are apparent in the 1987788 century maximum of lost life years.
70 age groups (yrs)
- .-. 40-64
60
----
I”‘~~~~~‘~‘~~‘~l~“~““‘l”“‘~“““~I’~”””’I”~”””I”’~“‘~‘I 1910 1920 1930 1940 1950 1960 Year
1970
Fig. 2. Male suicide, age specific rates, 1907-1990
20-24 15-19
1980
1990
Suicide and unemployment in Australia 1907-1990
,Oi 25 -
age groups(yrs) -...-.. 40-64 20-24
----
1 ;..+ ‘\ _
15-19
i!+j : i Ii
0 ’
753
i I 1 1.h i! j
~~~,~~~~~~~~~,~~l~ll~ll~llllllll’~~~~~~~~~’~”””“‘~““““‘~“”’~~~~~~‘~~~
1910
1920
1930
1940
1950
1960
1970
1980
1990
Year Fig. 3. Female suicide, age specific rates, 1907-1990. Unemployment and suicide rates among male 20-24 year olds for 196690 with accompanying trend lines are shown in Fig. 5. The values obtained from regressing 20-24 year old male suicide rates on the corresponding unemployment rate were 15.38 for the intercept and 1.13 for the 20-24 year male unemployment rate. While the adjusted-R’ of 2&24 male suicide and unemployment rates is 70%, the Durbin-Watson statistic (D = 0.86 for 25 observations) indicates significant auto-correlation. A first order autoregressive and zero order moving average process (with zero differencing) [ARIMA (l,O,O)] gives a reasonable fit to the residual term. The
estimated significance of the coefficient of the original regression remains at P < 0.01, df = 2,23, after taking autocorrelation into account. The linear increase in male youth unemployment (20-24) from 196690 matches reasonably closely the linear increase in the male suicide rate in this age group. If 20-24 year male unemployment rates are taken as a ratio of the overall unemployment rate, and 2G24 year male suicide rates as a proportion of the male age standardised suicide rate, the R2 value increases to 80%. Figure 6 shows the closeness of these two proportions over the 19661990 period.
___
$ 500% Z $ 5 E g $ a ;j = ? Z x 2
400-
300.
200-
loo-
0’
“‘,““““‘,““““‘~““““‘~““““‘l”~“l””””’l””~~~’~~~~~~~
1910
1920
1930
1940
1950 Year
1960
1970
1980
1990
Fig. 4. Life-years lost from male and female suicide, 1907-1990 age standardised.
754
STEPHENMORRELL etal.
-
suicide rate
. unemployment rate
~
trend
~
1966
1970
1975
1960
1965
1990
year Fig.
5. Male
suicide
and unemployment,
DISCUSSION
2&24
year olds,
196C1990.
(R2= 60%, for 1624 year olds) and Canada (R2= 58%) follow Australia. In contrast, Sweden, U.S.A.
The association between suicide and unemployment rates in Australia has been shown to vary, depending on period, age, gender and geography [ 18-211. Findings from the present study reinforce the temporal and age variation in the relationship. Comparison with other OECD countries indicates that the suicide-unemployment association among 15-24 year old males in Australia (for the 196687 period) is comparatively high [42]. The strong link (R2= 68%) for this age grouping in Australia is exceeded only by France (R2= 77%, 1968-87). The
Japan and Germany exhibit correspondingly low associations, with R2 = 0.03, 0.01, and 0.005 respectively. Low unemployment levels among youth in these three economically successful countries during the last two decades, coupled with strong familial networks (Japan) or adequate social welfare systems (Sweden and Germany), may contribute to this disparity. The responsiveness of male suicide rates to unemployment levels compared to females is also an international phenomenon [43]. The highest percentage of variance in suicide explained by unemploy-
2.0
1 1.5
1.0 20-24 year male suicide rate as proportion of male age standardised suicide rate 20-24 year male unemployment rate as proportion of overall unemployment rate
0.5
I
1966
8
I
b
I
1970
t
I
I
I
I
I
1
I
I
I
1
I
1960
1975
I
I
I
1965
I
I
I
I
I
1990
year
Fig. 6. Male suicide and unemployment rate proportions,
20-24 year olds, 1966-1990
Suicide and unemployment in Australia 19074990 ment in Australia has occurred in 20-24 year old males for the period 19661990 (Rz = 70%). This occurs without taking into account cohort effects [20], modernisation indices [19], or other social variables. The R2 value actually increased since Martina’s study [20] to 78% for 1966-1987. Thus for 2&24 year old Australian males the association between suicide and unemployment has varied slightly, but remains strong. The other feature to note in these trends is a similarly strong association of suicide among males aged 25-34 with their corresponding unemployment levels for the 19661990 period (R2 = 53%). Unemployment amongst 15-19 year old male youth is 50% higher than in 20-24 year old males, yet their suicide levels remain lower. However, larger proportions of 15-19 than 2&24 year olds still live with their parents, and therefore have the benefit of a stronger network of familial social support. Furthermore, frustrated employment/career expectations, the need for financial independence and other societal pressures would conceivably be less significant in the younger group compared to 20-24 year olds. Using Durkheim’s hypothesis, it could be argued that sharp changes in unemployment rates influence suicide rates, rather than high or low unemployment levels as such., Very low correlations, between agespecific and age-adjusted suicide rates, or changes in these, and changes in corresponding unemployment rates, do not support this argument. Nevertheless, as indicated in Fig. 6, the link between suicide and unemployment may be more significant when considered as age-specific proportional deviations from respective overall rates. This would suggest that cross-sectional proportional differences from the contemporaneous social norm have more long term bearing on rates of suicide than diachronous changes in social conditions. It should be noted, moreover, that Durkheim formed his views under the belief that suicide rates would be consistently higher among older males. The present study demonstrates that this is no longer the case. The fact that 20-24 year old male suicide rates exceeded all other age groups (excepting over 85 year olds) in 1990, and that this is an accrual of a quarter century trend, clearly contradicts Durkheim’s age thesis. As can be seen from Fig. 2, increased suicide rates during the Great Depression years were experienced more by middle-aged men, 45-54 year olds in particular, whereas male youth suicide rates were hardly affected. While age-specific unemployment rates are not available for this period, historical accounts of unemployment in the Great Depression suggest that older Australians were affected more by unemployment in the 1930’s than younger people [44]. By contrast, since the early 1970’s, levels of suicide among older male age groups have declined, despite historically high overall unemployment rates. However, middle-aged males have experienced unemployment levels at approximately half the overall rate during this period [42].
755
Although the present study revealed no cohort effects for twentieth century suicides, Martina’s birthrate cohort hypothesis [20] is given some credence because the postwar baby boom in Australia had peaked by the early 1960’s and unemployment rates for 20-24 year old males had peaked by the early 1980’s. However, this hypothesis negates the influence of economic fluctuations, as exemplified by present (late 1991) levels of unemployment in Australia being the highest since the Great Depression. Since time series data can exhibit autoregressive processes amongst both explanatory and outcome variables, parameter estimates in classical regression analyses are often both biased and unreliable. Simple regression of time trend data is not a valid method for estimating precise relationships between different time series because, in general, the assumption of non-autocorrelation of residual terms from classical regression models is violated. However, if the residual terms from these regressions display first order autoregressive behaviour then the regression parameter estimates are in fact unbiased and reliable, but the significance levels are exaggerated. Most socioeconomic time series exhibit first order autoregressive behaviour [45], but this feature cannot be assumed in advance and should be investigated. Time series techniques, such as ARIMA (BoxJenkins) modelling and construction and analysis of transfer functions, may serve a useful purpose in forecasting, especially with processes under experimental control, but they offer no assistance in modelling historical data with a view to achieving a priori explanatory power. The current study explored the plausibility of contemporaneous relationship between unemployment levels and suicide rates, particularly in 20-24 year old males. This was undertaken by modelling the autoregressive behaviour of residual terms, without resorting to lag times in the explanatory variables themselves. Such an approach avoids the pitfalls of post hoc explanations of statistically derived lag times in explanatory and outcome variables, in effect recognising that a myriad of unaccountable social and personal factors act as chance influences on aggregate suicide rates away from the regression line. Correlation of time trends of variables as a method of causal analysis indicates whether it is plausible to hypothesise that the factors may be linked in a cause and effect relationship. This method is particularly useful if both factors increase and decrease in unison during the same period, thus providing stronger evidence of direct causation than do progressive trend increases or decreases occurring in both parameters. While the aggregate data show that the suicide/unemployment relationship is not a simple year-by-year correlation, the present study strongly supports the hypothesis that unemployment is significant as a predisposing factor for increasing the risk of suicide, especially in males. While no aggregate
156
STEPHEN MORRELL
study can unequivocally establish causation, there was no evidence from Australian records to suggest that unemployment was not related to suicide. The issue of suicide and unemployment is worthy of further detailed analyses so that one set of destructive consequences associated with unemployment can be emphasised to maintain a political agenda committed to the protection of vulnerable groups in times of economic recession.
REFERENCES I. Boor M. The relationship between unemployment rates and suicide rates in eight countries, 1962-1976. Psycho/ Rep. 47, 1095-l 101, 1980. 2. Diekstra R. Suicidal behaviour in adolescents and young adults: the international picture. Crisis 10, 16-35, 1989. 3. Crombie I. K. Can changes in the unemployment rates explain the recent changes in suicide rates in developed countries? ht. J. Epidemiol. 19, 412416, 1990. 4. Rubinstein D. Epidemic suicide among Micronesian adolescents, Sot. Sci. Med. 17, 657-665, 1983. 5. Durkheim E. Suicide. A Stud?) in Sociologic, 1x9’97. Routledge & Kegan Paul, London, 1970. 6. Lester D. and Murrell M. The influence of gun control laws on suicidal behavior. Am. J. Psychiat. 137, 121-122, 1980. 7. Lester D. and Murrell M. The preventative effect of strict gun control laws on suicide and homicide. Suicide Life Threat Eehav. 12, 131-140, 1982. 8. Kreitman N. and Plan S. Suicide, unemployment and domestic gas detoxification in Britain. J. Epidemiol. Community Hlth 38, 14, 1984. 9. Oliver R. G. and Hetzel B. S. The rise and fall of suicide rates in Australia: Relation to sedative availability. Med. J. Aust. 2. 919-923, 1972. 10. Parshu R. A. M. and Uma R. A. 0. Paraquat poisoning in Fiji. Fui Med. J. Jan/Feb. 12-16, 1983. 11. Mowbray D. L. Pesticide poisoning in Papua New Guinea and the South Pacific. PNG. Med. J. 29, 131-141, 1986. 12. Phillips D. L. The influence of suggestion on suicide: subs&tive and theoretical implications of the Werther effect. Am. Social. Rec. 39, 340-354, 1974. 13. Phillips D. L. Suicide, motor vehicle fatalities and the mass media: Evidence toward a theory of suggestion. Am. J. Social. 84, 1150-1174, 1979. 14. Diekstra R. F. W. and Hawton K. (Eds) Suicide in Adolescence. Martinus Nijhoff, Dordrecht, 1987. 15. Pritchard C. Suicide, unemployment and gender variations in the Western world 19661986. Are women in Anglo-phone countries protected from suicide? Sot. Psychiat. Psychiatr. Epidemiol. 25, 73-80. 1990. 16. Warr P. and Jackson P. Adapting to the unemployed role: A longitudinal investigation. Sot. Sci. Med. 25, 1219-1224, 1987. 17. Burvill P. W. Changing _ __patterns of suicide in Australia, 191@1977. Acta psychiatr. stand. 62, 258-268, 1980. 18. Kessel A.. Nicholson A.. Graves G. and Krupinski J. Suicidal attempts in an ‘outer region of metropolitan Melbourne and in a provincial region of Victoria. ANZ J. Psychiatr. 9, 255, 1975. 19. Hassan R. and Tan G. Changing patterns of suicide in Australia, 1901-1985: An analysis of sex differentials. Suicide Life Threat Behao. 19, 362-380, 1989. 20. Martina A. Suicide and unemployment amongst young Australian males 1966 to 1982. Working Papers in Economic History. Australian National University, No. 56, pp. l-17, 1985.
et al
21. Dudley M., Waters B., Kelk N. and Howard J. Youth suicide in NSW: urban-rural trends 19641988. Med. J. Aust. 156, 83-88, 1992. of suicide. Epidem. Rev. 9, 22. Monk M. Epidemiology 51-69, 1987. 23. Olsen J. and Lajer M. Violent death and unemployment in two trade unions in Denmark. Sot. Psychiat. 14, 139-145, 1979. and illegal 24. Peck D. F. and Plant M. A. Unemployment drug use: concordant evidence from a prospective study and national trends. Er. Med. J. 293, 929-932. 1986. 25. Morris J. N. and Titmuss R. Health and social change: recent history of rheumatic heart disease. Med. Officer 2, 69-71, 77-79, 85-87, 1944. M. H. Mental Illness and the Econom?,. 26. Brenner Harvard University Press, Cambridge, MA, 1973. 27. Brenner M. H. Health costs and benefits of economic policy. ht. J. Hlth Sert?. 7, 581 -622, 1977. and 28. Brenner M. H. and Mooney A. Unemployment health in the context of economic change. Ser. Sci. Med. 17, 1125-1138, 1983. M. H. Mortality and economic stability: 29. Brenner Detailed analysis for Britain and comparative analysts for selected industrial countries. Int. J. Hlth Sew. 13, 563 420, 1983. 30. Brenner M. H. Relation of economic change to Swedish health and social well-being, 195(&1980. Sot. &i. Med. 25, 183. 195, 1987. 31. Eyer J. Review of mental illness and the economy. Iit. J. Hlth Serv. 1, 139-155, 1976. 32. Ever J. Reioinder to Dr Brenner. Inr. J. Hlth Sew. 1. 1;7-168, 1$76. G. and Stern J. Mortality 33. Gravelle H. S. E., Hutchinson and unemployment: A critique of Brenner’s time-series analysis. ,k&et 2, 675-679; 1981. J. R. and Funch D. P. Mental illness and 34. Marshall the economy: A critique and partial replication. J. HI/h Sot. Behar. 20, 282-289, 1979. 35. Gravelle H. S. E. Time series analysis of mortality and unemployment (Editorial). J. H//h Econ. 3, 297-305, 1984. 36. Withers G., Endres T. and Perry L. Australiun Historical Statistics.. L&our Statistics, Source Papers in Economic History No. 7. Australian National University, 1985. W. (Ed.) Austruliuns, Historical Stuti.strc.~. 37. Vamplew Fairfax, Syme & Weld, Broadway, NSW. 1987. 38. Smith D. J. and Callaghan A. H. Definitional aspects of measurement of unemployment. A comparative study of the Australian longitudinal survey and other sources of youth unemployment data. Aust. J. Stati.~/. 31A, 9%24, 1989. 39. Box G. E. P. and Jenkins G. M. Time Serie.s Ana/y.vi.s, Forecasting and Control. Holden Day. San Fransisco. 1976. J. and Weber N. C. Esti40. Keener R. W., Kmenta mation of the covariance matrix of the least squares regression coefficients when the disturbance covariance matrix is of unknown form. Econometric Theory 7. 2245. 1991. 41. Census of the Austrohcrn Population and Housing, p. Il. Australian Bureau of Statistics, 1933. Cooperation and Development. 42. Office for Economic Labour Force Statistics 1964-1984, and 1967- 1987. OECD, Washington DC, 1986, 1989. World Health Statistics 43. World Health Organisation. Annual, 1966-1989. WHO, Geneva. workers. A .Socicrl Hisfor), 44. Broomhill R. Unemployed of‘ the Great Depression in Adelaide. University of Queensland Press, St Lucia, 1978. 45. Ostrom C. W. Time Series Analysis Regression Tech n&es, 2nd edn. Sage, Newbury Park, 1990.