suicide and unemployment in italy: description, analysis and interpretation of recent trends

Sm. SC;. Med. Vol. 34, No. II, pp. 1191-1201. 1992 Printed in Great Britain. All rights reserved

Q277-9536192 sS.00 + 0.00 Copyright C 1992 Pergamon Press Ltd

SUICIDE AND UNEMPLOYMENT IN ITALY: DESCRIPTION, ANALYSIS AND INTERPRETATION OF

RECENT TRENDS*

STEPHEN PLArr,‘.‘t Rocco MICCIOLO*~~ and MKHELE TANSELLA*

‘MRC Medical Sociology Unit, 6 Lilybank Gardens, Glasgow G12 SQQ, Scotland, ‘Servizio di Psicologia Medica, Istituto di Psichiatria, Universita di Verona, Verona, Italy and ‘Istituto di Statistica e Ricerca

Operativa, Universita di Trento, Trento, Italy

Abstract-This paper assesses the relationship between suicide and unemployment in Italy during the period 1977-1987, taking into account variations by gender and region. The first objective of the study is to provide descriptive longitudinal and cross-sectional aggregate-level analyses and also trends in individual-level and population risks for suicide in relation to unemployment. Our second objective is to use the Italian data to help discriminate between rival interpretations of the unemployment-suicide link, i.e. the operation of health selection or causal (susceptibility) mechanisms.

Evidence for an association between suicide and unemployment among women was not convincing. The annual rate of female unemployment was negatively correlated with the female suicide rate and unrelated to the suicide rate among the unemployed, the relative risk or population attributable risk. Individual-level analyses for each year confirmed that unemployed women were more likely to commit suicide than their employed counterparts, although the overall relative risk was low (1.5) and conference intervals for six of the eleven annual risk ratios included unity (1). Among men, the unemployment rate was positively correlated over time with the suicide rate. However, change in the suicide rate across 18 geographic regions of Italy was unrelated to change in the unemployment rate, a finding which did not appear to be invalidated by a regression to the mean effect. Unemployment was also positively related to the suicide rate among the employed and population attributable risk, but unrelated to the rate among the unemployed or the relative risk. Comparison of suicide rates among the employed and unemployed revealed an excess of suicide among the latter in each year, with an overall relative risk of 3.4. On the basis of this contradictory and inconsistent evidence, we are cautious about offering definitive interpretations concerning the nature of the unemployment-suicide link among men. We conclude by suggesting the need for further individual-level studies using retrospective case-control methods.

Key words-unemployment, suicide, health selection, causal mechanisms

INTRODUCTION

Over the past two decades the impact of adverse

economic conditions upon mortality and morbidity has been extensively investigated throughout Europe and in North America. The literature on unemployment and suicide has recently been reviewed by Platt [I]. He distinguished between four types of research studies: individual-cross-sectional (Type I), aggregate-cross-sectional (Type 2), individual-longitudinal (Type 3) and aggregate-longitudinal (Type 4).

Two (Type I) case control studies report significantly more unemployment among male psychiatric patient suicides than among matched psychiatric controls, but among women no such differences were found [2, 31. Five studies report higher rates of suicide among the unemployed than among other groups [4-81. However, in the two studies which deal separately with women, the findings are inconsistent.

*The publication of this paper has been inadvertently delayed by postal problems-Editor

tAddress correspondence to: Stephen Platt, MRC Medical Sociology Unit, 6 Lilybank Gardens, Glasgow G12 SQQ, Scotland.

Platt [I] lists seven independent aggregate-cross- sectional (Type 2) analyses of suicide and unemployment. Of these, five [4,9-121 reveal no relationship between the two variables, while one produces a positive correlation [13] and the other a negative correlation [14]. Platt [I] argued that the evidence failed to support the existence of an ecological association between suicide and unemployment rates; the findings of two more recent studies [l5, 161 confirm this conclusion.

Ten relevant individual-longitudinal (Type 3) studies have been identified. Three retrospective analyses [l7-191 reveal more occupational loss, job instability and unemployment among suicides than in other control groups. The evidence of an association between unemployment and suicide is also positive, albeit somewhat indirect, in several prospective analyses [2&23]. It is more impressively demonstrated in results obtained from four general population longitudinal studies.

The OPCS Longitudinal Study has traced all deaths among a 1% sample of the population of England and Wales selected at random from the 197 1 Census [24]. The standardised mortality ratio (SMR) for 1971-81 among men aged 15-64 years at death,

1191

1192 STEPHEN PLAIT et al

who were seeking work in 1971, was 169 for suicide and related causes of death, after standardising for social class [25]. The authors devise a novel test for the possibility of a health-related selection effect (rather than a causal effect) by positing that, if ill-health were a major influence on the selection of the unemployed, then their relative mortality should full (i.e. improve) over time. “Such achange would be expected because of the high initial mortality of those selected on the basis of ill-health; the proportion of sick men in this category would decline over time. This and other mechanisms would lead to a reduction in mortality over time. .” [25, pp. l326-13271. Trends in suicide mortality do not suggest health- selection, since the SMR in the second five years of the follow-up period is higher (186) than in the first five years (148). However, the confidence limits are extremely wide (the observed number of suicide deaths was small) and a statistically significant differ- ence in SMRs over time cannot be demonstrated. The authors conclude that “although these data provide no strong evidence for a selection effect, we cannot rule out the possibility that there is one operating” [25, p. 13271.

In a second test of selection effects mortality is compared across regions varying in the level of unemployment [26]. The finding that areas with comparatively high unemployment also had high excess mortality is interpreted by the authors as evidence for a causal effect of unemployment. More recently, Moser et al. [27] compared mortality among the unemployed in 1971-3 and 1981-3. Although the level of unemployment was two to three times higher in 1981 than IO years earlier, the same excess mortality (including suicide) was found among the unemployed. On this occasion, Moser et al. concluded that unemployment itself has a direct or indirect bearing on mortality and that the “high mortality . . . among men seeking work in 1981 . . is not explained by either the pre-existing health of these groups or their socio-economic status before unem-

ployment” [27, p. 901. A similar mortality study on the workforce in

Denmark at the 1970 Census has been reported [28]. A significantly increased death rate was found among the unemployed, after adjusting for occupation, housing category, geographical region and marital status. The estimated relative suicide rates for unemployed men and women over the period 1970-80 were 2.51 (95% CI 2.12-2.97) and 2.45 (95% CI l.72-3.49), respectively. The excess suicide was higher in the first half of the follow-up period than in the second half only among men. In both sexes relative mortality among the unemployed compared to the employed across 275 Danish municipalities tended to fall as the prevailing unemployment rate increased. The authors interpret the increased mortality among the unemployed as a consequence of health-related selection as well as increased susceptibility associated with the psychosocial stress of unemployment. How-

ever, it is noteworthy that on two of the tests suggested by Moser et al. [25, 26]-namely, regional variations in unemployment and excess mortality among the unemployed (a positive association supporting a causal effect of unemployment), and temporal variations in excess mortality among the unemployed (lower mortality in the first sub-period supporting a causal effect of unemployment)-the data (especially in relation to male suicide) give more support to the health-selection hypothesis.

Nordstrom [29] has examined suicides occurring during the years 1961-5 among Swedish men aged 2&69 who had been unemployed for at least 4 months at the time of the 1960 census. The estimated age-standardised relative risk was 3.04 (95% CI 2.054.51), close to that reported by Iversen et al.

[28]. In the subsequent 5 years the relative risk was estimated to be 2.41 (95% CI 1.60-3.63). Nordstrom is sceptical, however, about the relevance of Moser et al. [25] temporal variation test for assessing selection effects, arguing that “the exposure, for example of the excess risk of unemployment which is assumed subsequent to the time of measurement, should level off with the passage of time”. He also makes the point that “sucidogenic characteristics (e.g. depression) that operate as selection mechanisms are probably more enduring than health selection mechanisms in general” [29, p. 3371.

The fourth and smallest population longitudinal study is based in Turin, Italy. In a preliminary

analysis of suicide mortality between I98 I and 1985, Costa et al. [S] report SMRs of 285 and 284 among males and remales (respectively) of working age. Analysis of trends over the period lead the authors to conclude that the higher suicide rate among the unemployed is unlikely to be due to health-selection effects.

The temporal association at an aggregate level between unemployment (or some related indicator of economic recession) and suicide rates has been examined in at least 36 separate (Type 4) studies. (Platt [I] reviews 30 such studies published prior to 1985. Since then, further relevant data have been provided by Stack and Haas [30], Crombie [3l], Cormier and Klerman [32], Charlton er al. [33], Pritchard [34], and Dooley et al. [35].) While analyses relating to the U.S.A. and Canada are almost unanimous in finding a positive association between the two variables, evidence relating to Great Britain and other European countries is conflicting. Very few Type 4 studies have specifically focused upon women, and their results have been inconsistent. Several analyses have been based on sophisticated multiple regression techniques, in order to overcome the problems of temporal precedence and autocorrelated error which bedevil the simple bivariate correlational approach. It should be noted, however, that a number of technical, methodological and conceptual criticisms have also been levelled at these time series analyses [3Wl]. Moreover, the aggegate longitudinal

Suicide and unemployment in Italy 1193

strategy in general is unable to resolve the competing claims of the causation or selection hypotheses. a view shared by exponents [42] and critics [36,38]

alike. Three broad conclusions can be drawn from these

findings. First, evidence of an association at the aggregate level between unemployment and suicide is found only in studies of longitudinal trends, and even these types of study produce inconsistent results or use questionable methodologies. Second, individual- level analyses tend to demonstrate an increased risk of suicide among the unemployed, but health- selection effects cannot entirely be ruled out. More generally, none of the research methodologies is capable of satisfactorily distinguishing between different interpretations of correlations between unemployment and suicide, although some attempts are considerably more sophisticated than others. Third, evidence of a link between suicide and unemployment is undoubtedly weaker for women than for men. This results partly from the actual empirical findings (which tend to differ by gender), and, more impor- tantly, from a relative paucity of studies specifically examining suicide and unemployment trends among women. (There is, in fact, a widespread tendency for researchers to neglect the study of the impact of female unemployment on female morbidity and mortality.)

One further general observation can be made. The four types of studies have almost without exception been carried out by different authors on different datasets. We are not aware of any research in which the association between unemployment and suicide has been explored at both aggregate and individual levels, using longitudinal and cross-sectional designs. In principle, statistics produced by the Italian Istituto Centrale di Statistica (ISTAT) should be amenable to Type 1, 2 and 4 analyses, but reports available at the present time have not fully explored all the possibil- ities. (The study by Costa et al. [S] does not consider regional data on suicide and unemployment.)

We have therefore undertaken a detailed examination of the relationship between suicide and unemployment in Italy, taking into account variations by gender, region and time period. This study is in part descriptive. We report the findings of longitudinal and cross-sectional aggregate-level analyses and also trends in individual-level and population risks for suicide in relation to unemployment. In addition we make use of the Italian evidence to attempt to discriminate between rival interpretations of the unemployment-suicide link, i.e. the operation of health selection or causal (susceptibility) mechanisms.

METHODS

Data sources

Suicide. In Italy, national data on suicide are collated and published by the Istituto Centrale di Statistica (ISTAT) in Rome. The data are obtained

from two sources. First, information based on final death certificates, completed by a doctor from the National Health Service who takes into account all available information, including the preliminary certificate issued by the examining doctor (see below), is published annually in the Annuario di Statistiche Sanitarie. However, publication is subject to considerable delay, usually some 5 years, and no data on economic position are provided. Secondly, information on suicide obtained from the police and carabinieri is published annually in the Annuario Statistic0 Italiano. On the same day that the death is notified to them, these officials compile a detailed case report based on a preliminary death certificate completed by the doctor who examined the body, and on extensive enquiries from key informants and relevant witnesses. At the same time they complete a form which is sent directly to ISTAT. One of the items on this form concerns economic position. Eight categories are used: employed, seeking a new job, seeking a first job, housewife, student, armed forces, retired and disabled. The present analyses were conducted using police/carabinieri data for the years 1977-1987, disaggregated into 18 regional units (16 of the 20 standard regions, together with a new unit combining Piemonte and Valle D’Aosta, and another combining Abruzzi and Molise). Cross-tabulations of data by economic position and age or social class were not available at the time this study was conducted.

Population. Yearly estimates of the breakdown of the Italian population by economic position are provided by ISTAT [43]. Five categories are used: employed, seeking new job, seeking first job, other economically active (14 years and over), and economically inactive (all ages). Regional data were available for the years 1977-1987. However, it was possible to subdivide the economically inactive by age group (O-13/14+ years) and region for the years 1977-8 I only.

Analysis

For the purposes of the study economic positions among suicides and in the general population were amalgamated into three categories: employed, unemployed and economically inactive. The unemployed consisted of those seeking a new job and those seeking a first job; both these categories were avilable for suicides and for the general population, and similarly defined. The economically inactive in the general population were here defined to include the category ‘other economically active’. There were two reasons for this decision. In the first place, despite the terminology it is apparent from the official description [43] that the primary economic position of individuals so classified is outside the labour force (although they may have expressed an intention to seek work in the future). Second, and more impor- tantly, there is no corresponding category in the classification of economic position among suicides.

1194 STEPHEN PLATT et al.

Females

070 s 77 78 79 80 61 02 03 84 85 06 87

Year

Males

10

9

8 8

8 7

? 6

% z 5

t 4

.; 3

‘5 2 VJ

1 n

0 Unemployment A Suicide

”

77 70 79 80 81 82 83 04 85 06 87

Year

Fig. I. Trends in suicide rate/lOO,OOO and unemployment rate percentage, 1977-1987. (A) Females. (B) Males.

Suicides who were outside the labour force at death would almost certainly be classified as retired, housewife, student, etc., even if they intended to seek work at some future date.

For each year, region and gender it was possible to calculate the suicide rates (per 100,000) among the employed and unemployed populations (aged 14+ years), both separately and combined. For the years 1977-81 only, we were also able to calculate a suicide rate among the economically inactive aged 14+ years, for each region and gender. Due to the absence of published data on age, social class and economic position, it was not possible to produce age- or social class-standardised suicide rates. The unemployment rate was the number of unemployed divided by the number of employed and unemployed, expressed as a percentage.

Statistical methods

Spearman’s rank correlation coefficient (rho) was used to assess the relationship between economic position and suicide. The consistency of the rank- ordering of regional suicide and unemployment rates across years was tested by means of the Kendall coefficient of concordance (w) (see Siegel [44] for descriptions of both statistics). Although it is argued [45] that significance testing should not be applied to ecological correlations due to the problem of ‘spatial

autocorrelation’ (i.e. the non-independence of con- tiguous geographical areas which tend to be like each other in social and economic characteristics), significance levels are commonly quoted as guidelines to the theoretical import of research findings. This conven- tion is followed in the present paper, although all tests are 2-tailed in order to minimize (as far as possible) artificially inflated ‘P’ values.

RESULTS

Over the 11 year period under review there were 13.555 suicides among economically active individuals aged 14+ years throughout Italy. Of these, 1512 were employed females, 246 unemployed females, lo,07 1 employed males and 1726 unemployed males.

Longitudinal analyses

The trends in unemployment and suicide rates for the whole of Italy during the years 1977-87 are shown in Figs l(A) and (B). Unemployment increased at a steady rate between 1977-8 and 1986-7 (by 64% among women and 87% among men); over the same period the female suicide rate fell by 22%, while the male suicide rate rose by 38%. Conse- quently, there was a negative (nonsignificant) correlation between unemployment and suicide among

Females

2 !! z

l.z-(*) 0O

i op 0 0

0.8 - 0 0

E” 0 0,” $ 0.4 - a 0 O0

2

.G - k?l 0.0 0

: Jz -0.4 - ’ ’ ’ ’ ’ ’ ’ ’ ’ I 0 -1.0 -0.6 -0.2 0.2 0.6 1 .o

Change in suicide rate

Males

E o-l1 ’ ’ 1 I I I I ,

-0.2 0.2 0.6 1 .o 1.4

Change in suicide rate

Fig. 2. Relation between change in mean annual suicide rate/lOO,OOO and change in mean annual unemployment rate percentage, 1986-87 compared to 1977-78 (IV = 18 regions).

(A) Females. (B) Males.

Suicide and unemployment in Italy

Table I. Suicide ~dtes/f~,~, 1977-78 and 1986-87. and changes in suicide rates (198&87 compared to 1977-X). by differing levels of regional unemployment

Regional unemploymeo~ rate’

1195

High Medium fN = 6 reeionsl IN = 6 reaionsl

Low (iv = 6 reaions)

FtWdFS Mean (SE) suicide rate. 1977-S 2.6 (0.3) 3.8 (0.3) 2.3 (0.2) Mean (SE) suicide rate. 1986-7 1.4 (0.2) 2.4 (0.2) 2.6 (0.2) Absolute change. 19867 compared to 197.7-S - 1.2 - 1.3 +0.2 Relative change. f986-7 compared to 1977-S -46.6% - 35.4% + 10.3%

W&s Mean (SE) suicide rate, 1977-S 5.0 (0.2) 7. I (0.3) 7.0 (0.3) Mean (SE) suicide rate, 1986-7 6.6 (0.3) 8.9 (0.3) 10.7 (0.3) Absolute change, 19867 compared to 1977-S +I.6 i- 1.9 +3.6 Relative change. 198&7 compared to 1977-S +32.1% + 26.8% +51x%

*Based on average rate prevailing over the whole study period. The regions were allocated as follows: High u~~~z~~o~~e~~ (females and males): Campania, Pug&a, Basilicata, Calabria, Sicilia, Sardegna. ,Wedium ~~~~~~o~~e~f (females): Liguria. Emilia Romagna. Toscana, Umbria, Lazio, Abruzzi-Molise; (males):

Piemonte-Valle d’Aosta, Liguria, Toscana, Umbria, Lazio, Abrussi-Molise. Low ene~p~o~~re~~ (females): Piemonte-Valle d’Aosta. Lombard& Trentino-Alto Adige, Veneto. Fruifi-Veneda

Giulia. Marche; (males): Lombardia, Trentino-Alto Adige, Veneto, Fruili-Venetia Giufia, Emifia Romagna, Marche.

women (r, = -0.53, NS) and a positive correlation between male unempIoyment and suicide rates (r, = 0.80, P < 0.01).

In order to assess the consistency of these associ- ations (and provide further evidence for their interpretation) we examined changes in unemployment and suicide between 1977-8 and 1986-7 across IS geographic regions of Italy. Figures 2(A) and (B) show that changes in the two rates were not correlated in women (rS = -0.06, NS) or men (rr = -0.17, NS), respectively. (Further analyses using absolute

rather than relative measures of change produced virtually identicai resuhs.)

Another approach to the elaboration of the temporal reIationship between the suicide and unemployment rates is provided in Table 1. Were the 18 regions are divided into three equal groups, according to whether the ievel o/ ~~e~pl5y~ent over the whole study period was high, moderate or low. (Although the unemployment rate increased between 1977 and 1987, the ranking of regions by unemployment was aImost perfectly stable; w = 0.93, P c 0.0001 for

Table 2. Suicide rates (per lOO.000) by gender and economic position (economically active only), and measures of relative and popufation attributabfe risk, 1977-87

Year

Suicide among the employed

Rate (SE)

(A)

Suicide among the unemployed

Rate (SE)

(B1

Suicide among employed and

unemployed Rate (SE)

(Cl

Relative risk

Ratio’(SE)

(D) 95% CI

Population attributable

riskb

(E)

Population attributable

risk percentage’

(f=)

Femates 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1977-87

2.9 (0.2) 6.0 (I .2) 2.7 (0.2) 2.0 (0.7) 2.2 (0.2) 2.6 (0.7) 2.5 (0.2) 3.0 (0.7) I .7 (0.2) 2.4 (0.6) I .9 (0.2) 2.5 (0.6) I.7 (0.2) 3.0 (0.6) I.8 (0.2) 3.4 (0.6) I.9 (0.2) 2.3 (0.5) 2. I (0.2) 3.4 (0.6) f .9 (0.2) 4.7 (0.7) 2. I (0. I) 3.2 (0.2)

3. I (0.2) 2.1 (0.4) 1.41-3.19 2.6 (0.2) 0.8 (0.3) 1.43-1.50 2.3 (0.2) I .2 (0.3) 0.69-2.04 2.6 (0.2) I.2 (0.3) 0.75-2.02 I.8 (0.2) I .4 (0.4) O.Ss2.45 2.0 (0.2) I .3 (0.3) 0.81-2.15 I .S (0.2) 1.8 (0.4) 1.19-2.81 2.0 (0.2) 1.9 (0.4) f.26-2.85 I .9 (0.2) 1.3 (0.3) 0.79-2.00 2.2 (0.2) I .6 (0.3) l.lO-2.40 2.2 (0.2) 2.5 (0.4) 1.75-3.4s 2.2 (0. I) f.S(O.1) I .33-l .73

0.2 6.5

0 co 0 0 0. I 3.8 0.1 5.6 0.1 5.0 0.1 5.6 0.2 10.0 0 0 0. I 4.5 0.3 13.6 0.1 4.5

Mates 1977 5.6 (0.2) 27.7 (2.4) 6.3 (0.2) 5.0 (0.5) 4. I6-6.03 0.7 11.1 1978 6.2 (0.2) 16.8 (1.8) 6.6 (0.2) 2.7 (0.3) 2.17-3.38 0.4 6.1 I979 6.4 (0.2) IS.7 (1.9) 6.9 (0.2) 2.9 (0.3) 2.37-3.58 0.5 7.2 1980 6.2 (0.2) 22.1 (2.0) 6.8 (0.2) 3.6 (0.4) 2.95-4.33 0.6 8.8 1981 6.5 (0.2) 18.9 (1.8) 7.0 (0.2) 2.9 (0.3) 2.39-3.54 0.5 7.1 1982 6.8 (0.2) IS.1 (1.6) 7.3 (0.2) 2.7 (0.3) 2.23-3.22 0.5 6.8 1983 5.7 (0.2) 16.1 (1.4) 6.2 (0.2) 2.9 (0.3) 2.37-3.44 0.5 8.1 1984 6.5 (0.2) 20.1 (1.6) 7.3 (0.2) 3. I (0.3) 2.61-3.64 0.8 II.0 1985 7.2 (0.2) 26.6 (1.7) 8.3 (0.2) 3.7 (0.3) 3.224.28 I.1 13.3 1986 7.2 (0.2) 25.6 (1.7) 8.3 (0.2) 3.6 (0.3) 3.09-4.1 I I.1 13.3 1987 7.9 (0.2) 30.4 (1.8) 9.4 (0.3) 3.9 (0.3) 3.4w.39 I.5 16.0 1977-87 6.6 (0.1) 22.4 (0.5) 7.3 (0. I) 3.4(0.1) 3.24-3.59 0.7 9.6

‘Ratio of suicide rate among the unemployed to suicide rate among the employed (D = B:A). ?Suicide rate among the economically active (employed and unemployed) minus suicide rate among the employed (E = C - A). ‘Population attributable risk divided by suicide rate among the economically active, expressed as a percentage (F = E x 100/C).

1196 STEPHEN PLAlT Cl al.

Table 3. Correlawx~ between unemployment rate percentage and suicide ra1e,‘l00.000 by

eender and economic oosition (economicallv active onlv). 1977 to 1987 (N = I I vears)

Annual unemployment

rate correlated with:

Spearman correlation coeffiaent (rho)

Females Males

-Annual suiade rate among employed rs = -0.62 I, = 0.83

P < 0.05 P < 0.01 -Annual suiade rate amqng unemployed r5 = 0.21 Is = 0.32

NS NS -Annurtl relative risk ratio rs= 0.41 rs = 0.23

NS NS -Annual population attributable risk percentage rs = 0.35 I$ = 0.62

NS P con5

Note: two-tailed test.

women; II’ = 0.92, P < 0.0001 for men.) Among females the areas with high and moderate levels of unemployment showed a marked decrease in the rate of suicide over time, while the regions with low unemployment rates showed a modest increase. Among men the areas with low unemployment rates experienced the greatest increase in suicide. Overall,

therefore, there was a negative association between the regional unemployment level and increase in suicide. An alternative analysis of change in suicide rates by the extent of change in unemployment rates over the period showed similar results: areas with the lowest increases in unemployment tended to show less reduction in suicide (among women) or the greatest increase in suicide (among men).

These findings appear to cast some doubt on a causal explanation for the temporal association between unemployment and suicide among males. However, this evidence may itself be untrustworthy, due to the effect known as regression to the mean, which is the tendency “for a variable that is extreme on its first measurement. . to be closer to the center of the distribution for a later measurement” [46, p. 4931. We therefore examined scatterplots showing the association between unemployment (or suicide) in 1977-8 and proportional change in unemployment (or suicide) over the period 1977-8 to 1986-7 across the 18 regions. The resulting patterns appeared to be entirely random, providing no evidence of a regression effect. Statistical analyses confirmed the absence of a relationship between initial values and change over time for both suicide (rr= -0.16, NS) and unemployment (rr = -0.12, NS).

Suicide rates among employed and unemployed, separately and combined, together with the relative and population attributable risks for each year of the study, are shown in Table 2. With only one exception (females, 1978), the rate among the unemployed was always higher than the rate among the employed. In 6 of the II years, however, the confidence limits for the female relative risk ratio included unity(l). Among males the minimum relative risk was 2.7. Over the whole period the mean relative risks were I .5 (95% CI 1.3-1.7) and 3.4 (95% CI 3.2-3.6) among females and males, respectively. The final column in Table 2 shows trends in population attributable risk percentage, i.e. the maximum percentage of the

suicide rate among the economically active which is ‘attributable’ (in a statistical sense) to unemployment. The mean value was 4.5% (minimum O%, maximum 13.6%) among women, and 9.6% (minimum 6.1%, maximum 16.0%) among men.

Table 3 presents findings on the temporal relationship between the annual unemployment rate and suicide rates by economic position. Among women unemployment was correlated negatively with the suicide rate among the employed; no other correlations were significant. Among men unemployment was correlated positively with the suicide rate among the employed and also with trends in population attributable risk percentage; all other correlations were non-significant.

Cross-sectional analyses

The ecological (geographical) relationship between the mean annual suicide rate (1977-87) and the mean annual unemployment rate (1977-87) across 18 Italian regions was significantly negative for both women (r; = -0.60, P < 0.01) and men (r, = -0.53, P < 0.05) (see Figs 3A and B). (An examination of correlations for each of the 11 study years shows that all values were negative; 4 reached statistical significance among women, 6 among men. The scatter plots shown in Figs 3(A) and (B) are therefore valid representations of the annual patterns of relation- ships between the two variables.)

Table 4 summarises the cross-sectional association between suicide rates by economic position and the unemployment rate, across the 18 regions, for the whole study period. Among both females and males, all correlations were significantly negative. Thus, the higher the unemployment rate, the lower the suicide rate among the employed and unemployed, and the lower the relative risk ratio. (Correlations were also calculated separately for each year. With only one exception (relative risk ratio, females, 1978) all were negative. In general significance levels tended to be higher for males, especially with respect to the correlation between the unemployment rate and the suicide rate among the unemployed.)

Comparison between economically actice and inactive

It was possible to calculate the rate of suicide among the economically inactive aged 14+ years for


Females

20

E 18 $ & 16

a 14

0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

Suicide ratellOO.OOO

Males

0

3 5 7 9 11 13 15 17

Suicide rate/lOO,OOO

Fig. 3. Relation between mean annual rates of suicide/100,000 and unemployment percentage, 1977-87

(N = 18 regions). (A) Females. (B) Males.

each year between 1977 and 1981 and for each region of Italy. The mean rate was 3.8 and 13.8 among females and males, respectively (Table 5). The inactive group had the highest rates overall (and in each year except 1977) among females, while the unemployed had the highest rates overall (in each year) among males. (It was reasonable to take mean values over the period since the ranking of regions by suicide rates among the economically active was almost identical from year to year; w = 0.86, P < 0.0001 for women; w = 0.85, P < 0.0001 for men.)

DISCUSSION

Before summarising and discussing our main findings, some comment is required about the adequacy

of both denominator and numerator data. The estimated annual breakdown of the population by economic position is derived from three-monthly stratified

random samples of the total population. ISTAT [43] has carried out a number of reliability and validity checks and concludes that these data provide a reasonably accurate picture of economic activity at the regional level. As reported above, the suicide dataset used in this study is derived from the rlnnu- ario StaGsfico Italiuno (ASI), which itself relies on reports made by the police and carabinieri. Cecere and Florenzano [47] have suggested that these data, based on a preliminary assessment of cause of death by the examining doctor, underestimate suicide as compared with data based on the final death certificate. This was confirmed by Williams et al. [48] who found, however, strong positive correlations between trends in suicide identified by the two different sources (r, = f0.75 for the years 1973-80; and rr = f0.87 across the regions). The problem of under-reporting of suicide is not confined to Italy [49] and there is no reason to suppose that this phenom- enon varies systematically in relation to economic position in such a way as to bias findings reported in this study. The advantage of the AS1 dataset is that it alone produces counts of suicide by economic position in addition to breakdowns by gender and region. The unusual opportunity to calculate suicide rates among individuals allocated to different economic positions is of considerable scientific import- ance. The relationship between economic conditions and suicide had mostly been examined by means of aggregate-level time-series analyses [I], with the attendant risk of committing the ecological fallacy [50, 511.

Our study encompasses both longitudinal and cross-sectional approaches and presents data at both aggregate and individual levels. In the interest of greater clarity we discuss separately findings relating to women and to men.

Among women, the annual rate of unemployment was negatively correlated with the overall female suicide rate and also the suicide rate among the female unemployed; there was no relationship with the rate among the unemployed, the relative risk or population attributable risk. Individual-level analyses for each year confirmed that unemployed women were more likely to commit suicide than their

Table 4. Correlation between unemployment rate percentage and suicide rate/lOO,OOO by gender and economic position (economically active only), 1977 to 1987, across geographical

regions (N = 18 regions)

Regional unemployment rate correlated with:

Spearman correlation coefficient (rho)

Females Males

-Regional suicide rate among employed

-Regional suicide rate among unemployed

-Regional relative risk ratio

Note: two-tailed test.

r,= -0.56 r, = -0.49 P < 0.02 P < 0.05

r,= -0.85 ‘, = - 0.88 P < 0.001 P < 0.001

I,= -0.82 I,= -0.91 P < 0.001 P < 0.001

SSM 34111-c

1198 STEPHEN PLAlT et a/.

Table 5. Suicide rates (per 100.000), by economic positmn and gender. 1977 10 1981 (all years

combined)

Females

Rate (SE)

Males

Rate (SE)

A. Suicide rate among employed 2.4 (0. I ) 6.2 (0. I)

B. Suicide rate among unemployed 3. I (0.3) 20.7 (0.9)

C. Suicide rate among economically active (A + B) 2.5(0.1) 6.7 (0. I)

D. Suicide rate among economically inactive 3.8 (0.1) 13.8(0.2)

.

employed counterparts, although the overall relative risk was low (1 S) and confidence intervals for 6 of the 1 I annual risk ratios included unity (1). Across geographical regions female unemployment was significantly negatively correlated with the overall female suicide rate, and also with rates among the employed and unemployed (and the relative risk ratio). A final comparison of suicide rates across all economic positions showed that among women the rate is higher among the economically inactive than among the unemployed. On the basis of these findings we may conclude that the association in recent years between unemployment and suicide among Italian women had been, at best, rather weak.

Among men, the unemployment rate was positively correlated over time with the suicide rate and also with the rate among the employed and population attributable risk. There was no temporal association between unemployment and the suicide rate among the unemployed or the relative risk. Furthermore, change in the suicide rate was unrelated to change in the unemployment rate. In fact, the data presented in Table 1 suggest that the greatest increase in suicide over the period occurred in the regions with the lowest unemployment rates. On the other hand, comparison of suicide rates among the employed and unemployed reveal an excess of suicide among the latter in each year, with an overall relative risk of 3.4. The suicide rate among the unemployed was also shown to be higher than the rate among the economically inactive. Cross-sectional analyses consistently demonstrated a negative association between the regional unemployment rate and all regional suicide rates: the higher the level of unemployment, the lower the suicide rate among the employed and unemployed, and the lower the relative risk.

This evidence is contradictory and inconsistent. The strength of the temporal relationship between the movement of suicide and unemployment rates might be thought to constitute per se evidence of a causal impact of joblessness. However, there are three reasons for suggesting that such an assumption may be naive and unwarranted. In the first place, we should note that the trends in unemployment were in one direction only (upwards) over the 11 year period. Thus, there is no empirical evidence concerning the impact of declining unemployment on suicide rates. (By contrast, the high correlation between unemployment and parasuicide in Edinburgh, Scotland was based on a 15 year period in which there was downward, as well as upward, movement in unem-

ployment [52].) Second, it has to be remembered that, although there was a positive temporal association between unemployment and population attributable risk (especially from 1982 onwards), between 84% and 94% of suicide incidence could not be attributed statistically to unemployment. In contrast, more than 50% of the male parasuicide rate in Edinburgh in the early 1980s was attributable to unemployment [52]. Finally, we demonstrated that the association between unemployment and suicide disappeared when trends were analysed at regional level. This analysis was not invalidated by a regression to the mean effect. Moreover, identical discrepant findings were noted by Crombie in his analysis of male unemployment and suicide in Scotland [3 I] and across I6 developed countries [53] during a similar time period.

The demonstration of a nonsignificant (temporal) or negative (ecological) correlation between the prevailing unemployment rate and the rate of suicide among the unemployed tends to confirm the finding of the Danish (but not the British) Longitudinal Study. Using the test proposed by Moser et al. 1261 this finding can be construed as evidence against a causal explanation of the unemployment-suicide relationship. On the other hand, Platt and Kreitman [54] have suggested an interpretation of the negative ecological association between unemployment and parusuicide (non-fatal deliberate self-harm) among the unemployed which does not rule out the possibility of a causal mechanism. They speculate that the unemployed individual considers himself less socially stigmatized and personally deviant in areas of high unemployment than in areas of low unemployment. He may be able to count on a greater degree of support and companionship from other unemployed people during the working day. Consequently, the psychological impact of unemployment may be significantly less marked in high-rate than in low-rate unemployment areas. (Empirical evidence in support of this proposition was subsequently provided by Jackson and Warr [55].) This may be the reason why there was a comparatively low risk for parasuicide among the unemployed in those areas of Edinburgh where unemployment was more widespread, and, by analogy, the reason why the risk for suicide among the unemployed was lower in areas of Italy where there was a higher level of unemployment.

The interpretation of the finding of excess suicide mortality among unemployed men is hampered somewhat by the fact that suicide rates by economic position could not be adjusted for age or social class.


In Italy (as elsewhere in the industrialised West) the suicide rate increases markedly with age [47]. Since the unemployed seeking new jobs are of higher average age than the employed, the rate of suicide among the unemployed will tend to be higher than the rate among the employed, regardless of any effect of unemployment itself. Costa ef al. [8] have analysed separately suicide incidence among those seeking a first job (on average younger) and those seeking a new job (on average older). As predicted, suicide rates among the former are considerably lower then suicide rates among the latter. The effect of social class standardisation can only be conjectured, since certain crucial data are not available. However, it is known that unemployment is concentrated in the lower occupational groups [56] while evidence from Great Britain [57] would suggest that there is an inverse relationship between suicide and social class. Once again, therefore, we would expect the rate of suicide among the unemployed to be higher than the rate of suicide among the employed, due to the confounding effect of social class. In a future analysis we hope to be able to produce age-standardized suicide rates. In the meantime the quoted mean relative risk (3.4) should be treated with some caution.

There may indeed be a causal link at the individual level between unemployment and suicide, but the relationship is unlikely to be direct or to exclude entirely some element of self-selection. Any interpretation of the findings presented in this paper has to take into account the evidence from several well designed retrospective case studies [58-60] that mental illness is almost certainly a necessary ante- cedent condition, though not a sufficient cause, of suicide in industrial Western countries [61]. Further- more, detailed analyses of individual case histories [6,62-65] and evidence from a careful case-control study [19] suggests that unemployment in suicide victims is to a large extent a consequence or reflection of an underlying psychiatric disorder: rather than unemployment being a cause of psychiatric illness and suicide, it is more likely that the risks of both unemployment and suicide are elevated by the pres- ence of a major psychiatric illness (in particular, depression). However, we should bear in mind that the size of this group at risk of suicide may itself be related to prevailing macro-economic conditions.

CONCLUSION

In this study we have attempted to describe the association between unemployment and suicide in Italy, using individual- and aggregate-level data in cross-sectional and longitudinal analyses. We suggested that there is little evidence of any such association among women, but were more circum- spect in our conclusions relating to men. On the one hand, both suicide rates and population risk of suicide attributable to unemployment tended to rise over time in line with trends in unemployment, and

the rate of suicide among the unemployed (compared to the employed) was significantly elevated over the whole period. On the other hand, most of the suicide incidence could not be ‘attributed’ to unemployment even in the ‘worst’ year; there was no relationship between change in unemployment and change in suicide at the regional level; and the impact of falling unemployment could not be assessed due to the ever-rising trend during the period under review.

Even where the descriptive evidence was reasonably straightforward, it was virtually impossible to achieve the study’s second objective, namely deciding between rival (causal vs self-selection) interpretations of the suicide-unemployment link. It might be thought that the individual-longitudinal design is the research study type most likely to be able to resolve this controversy. As we have seen, however, even large-scale prospective longitudinal studies are unable to produce unequivocal results. There are at least three reasons for this failure: first, the measure of unemployment is restricted to one observation at a fixed point in time, with no information being provided on duration of the current spell or on lifetime unemployment patterns; second, the individual’s psychiatric status and history is entirely missing; and, thirdly, data on employment status at death is (usually) unavailable. We suggest that the way forward should consist primarily of individual-level studies, but using the retrospective case-control method, as pioneered by Shepherd and Barraclough [19]. Careful consideration would have to be paid to the choice of a suitable control group, the identifi- cation and classification of episodes of psychiatric illness during the individual’s lifetime (including the dating of onsets and offsets), and the reconstruction of his/her employment history (including the dating of periods of unemployment). Such studies are long overdue.

Acknowledgements-We are grateful to the Istituto Centrale di Statistica (ISTAT, Roma) for providing the data. The study was supported by the Consiglio Nazionale delle Ricerche (CNR, Roma,) Progetto Finaliuato Fattori di Malattia (FATMA), Sottoprogetto Stress, contract to Professor M. Tansella

I.

2.

3.

4.

5.

6.

REFERENCES

Platt S. Unemployment and suicidal behaviour: a review of the literature. Sot. Sci. Med. 19, 93-l IS, 1984. Robin A. A., Brooke E. M. and Freeman-Browne D. L. Some aspects of suicide in psychiatric patients in Southend. Br. J. Psychiat. 114, 739-147, 1968. Roy A. Risk factors for suicide in psychiatric patients. Archs. fen. Psychiar. 39. 1089-1095. 1982. Sainsbury P. -Suicide in London. Chapman & Hall, London, 1955. Cumming E., Lazar C. and Chisholm L. Suicide as an index of role strain among employed and not employed married women in British Columbia. Can. Rev. Socioi. Anthrop. 12, 463-469, 1975. Yap P. M. Suicide in Hong Kong. J. Menr. Sri. 104, 266-301, 1958.

I200 STEptim PLArr et al

7. Kraft D. P. and Babigian H. M. Suicides by persons with and without psychiatric contacts. Archs gen. Psy- chiar. 33, 209-2 15, 1976.

8. Costa G.. Crepet P. and Florenzano F. Unemployment and mortality in Italy: The Turin longitudinal study. In Current Research on Suicide and Parasuicide (Edited by Kreitman N. and Platt S.), pp. 40-46. Edinburgh University Press, Edinburgh, 1989.

9. Wallis C. P. and Maliphant R. Delinquent areas in the county of London: ecological factors. Er. J. Critninol. 7, 250-284, 1967.

10. Farmer R. D. T., Preston Y. 0. and O’Brien S. E. M. Suicide mortality in Greater London: Changes during the past 25 years. Br. J. Prec. Sot. Med. 31. 171-177. 197;. ’

I I. Buglass D. and Duffy J. C. The ecological pattern of suicide and oarasuicide in Edinburgh. Sot. Sci. Med. 12. 241-253, 1678.

12. Sainsbury P., Jenkins J. and Levey A. The social correlates of suicide in Europe. In The Suicide Syndrome (Edited by Farmer R. and Hirsch S.), pp. 38-53. Croom Helm, London, 1980.

13. Bagley C., Jacobson S. and Palmer C. Social structure and the ecological distribution of mental illness, suicide and delinquency. Psychol. Med. 3, 177-187, 1973.

14. Lester D. Suicide and unemployment. A re-examination. Archs. enuirn. Hhh 20, 277-278, 1970.

15. Araki S. and Murata K. Social life factors affecting suicide in Japanese men and women. Suit. Life-Threat. Behau. 16, 458-468, 1986.

16. Zimmerman S. L. States’ public welfare expenditures as predictors of state suicide rates. Suit. Life-Threar. Behav. 17, 271-287.

17. Humphrey J. Social loss: a comparison of suicide victims, homicide offenders and non-violent individuals. Dis. Nerv. Sysf. 38, 157-160, 1977.

18. Olsen J. and Lajer M. Violent death and unemployment in two trade unions in Denmark. Sot. Psychiar. 14, 139-145, 1979.

19. Shepherd D. M. and Barraclough B. M. Work and suicide: an empirical investigation. Br. J. Psychiar. 136, 469-478, 1980.

20. Theorell T., Lind E. and Floderus B. The relationship of disturbing life-changes and emotions to the early development of myocardial infarction and other serious illnesses. Inr. J. Epid. 4, 281-293, 1975.

21. Cobb S. and Kasl S. V. Some medical aspects of unemployment. Indusr. Geronr. 8, S-15, 1972.

22. Hagnell 0. and Rorsman B. Suicide in the Lundby Study: a controlled prospective investigation of stressful life events. Neuropsychobiology 6, 319-332, 1980.

23. Borg S. E. and Stahl M. Prediction of suicide: a prospective study of suicides and controls among psychiatric patients. Acfa psychiaf. stand. 65, 221-232, 1982.

24. Fox A. J. and Goldblatt P. 0. Longitudinal Study: Socio-demographic and Mortality Differentials, 1971-1975. HMSO, London, 1982.

25. Moser K. A., Fox A. J. and Jones D. R. Unemployment and mortality in the OPCS Longitudinal Study. Lancet ii, 1324-1328, 1984.

26. Moser K. A., Fox A. J., Jones D. R. and Goldblatt, P. 0. Unemployment and mortality: further evidence from the OPCS Longitudinal Study 1971-81. Lancer i, 365-367, 1986.

27. Moser K. A., Goldblatt P. 0.. Fox A. J. and Jones D. R. Unemployment and mortality: comparison of the 1971 and 1981 lonnitudinal study census samples. Br. Med. J. 294, 86-90, i987. -

28. Iversen L., Andersen O., Andersen P. K., Christoffersen K. and Keiding N. Unemployment and mortality in Denmark, 1970-80. Br. Med. J. 295, 879-884, 1987.

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

39.

40.

41.

42.

43.

44.

45.

46.

47.

48.

49.

50.

51.

52.

Nordstrom T. Deriving relative risks from aggregate data 2. An application to the relationship between unemployment and suicide. J. Epidemiol. Commun. Hlth 42, 33640, 1988. Stack S. and Haas A. The effect of unemployment duration on national suicide rates: a time series analysis, 194881982. Social. Focus 17, 17-29, 1984. Crombie I. K. Trends in suicide and unemployment in Scotland, 197686. Br. Med. J. 298, 782-784, 1989. Cormier H. J. and Klerman G. L. Unemployment and male-female labour force participation as determinants of changing suicide rates of males and females in Quebec. Sot. Psychiat. 20, 109-114, 1985. Charlton J. R. H., Bauer R., Thakhore A., Silver R. and Aristidou M. Unemployment and mortality: a small area analysis. J. Epidem. Comm. Hlfh 41, 107-I 13, 1987. Pritchard C. Suicide, unemployment and gender in the British Isles and European Economic Community (1974-1985). Sot. Psychiar. Psychiarr. Epidem 23, 85-89, 1988. Dooley D., Catalan0 R., Rook K. and Serxner S. Economic stress and suicide: multilevel analyses. Part 1: Aggregate time series analyses of economic stress and suicide. Suit. Life-Threat. Behav. 19, 321-336, 1989. Dooley D. and Catalan0 R. Economic change as a cause of behavioural disorder. Psychol. Bull. 87, 450-468, 1980. Gravelle H. S. E., Hutchison G. and Stern J. Mortality and unemployment: a critique of Brenner’s time-series analysis. Lancet 2, 675-679, 1981. Kasl S. V. Mortality and the business cycle. Some questions about research strategies when utilising macro-social and ecological data. Am. J. publ. Hlth 69, 784-788, 1979. Kasl S. V. Strategies of research on economic instability and health. Psychol. Med. 12, 637-649, 1982. Stern J. Unemployment and its Impact on Morbidity and Morraliry. Centre for Labour Economics, L.S.E. Discussion Paper No. 93, London, 1981 Wagstaff A. Time series of the relationship between unemployment and mortality: a survey of econometric critiaues and reolications of Brenner’s studies. Sot. Sci. Med: 21, 985996, 1985. Brenner M. H. Health costs and benefits of ecomomic policy. Inr. J. Hlth Serv. 7, 581623, 1977. Istituto Centrale di Statistica (ISTAT). Statistiche del Laooro. ISTAT, Rome, 1986. Siegel S. Nonparametric Statistics for rhe Behavioural Sciences. McGraw-Hill, New York, 1956. Bagley C., Jacobson S. and Palmer C. Social structure and the ecological distribution of mental illness, suicide and delinquency. Psychol. Med. 3, 177-187, 1973. David C. E. The effect of regression to the mean in epidemiologic and clinical studies. Am. J. Epidem. 104, 493-498, 1976. Cerere F. and Florenzano F. II suicide in Italia. ISIS41, 5-12, 1984. Williams P.. De Salvia D. and Tansella M. Suicide and the Italian psychiatric reform: an appraisal of two data collection systems. Eur. Arch. Psychiar. Neurol. Sci. 236, 237-240, 1987. World Health Organisation (WHO) Changing Patterns in Suicidal Behaoiour. Euro reports and studies No 74. WHO Regional Office for Europe, Copenhagen, 1982. Selvin H. C. Durkheim’s ‘Suicide’ and problems of empirical research. Am. J. Social. 63, 607-619, 1958. Catalan0 R. and Doolev D. Health effects of economic instability: a test of economic stress hypothesis. J. Hlth Sot. Behao. 24, 46-60, 1983. Platt S. and Kreitman N. Parasuicide and unemployment among men in Edinburgh 1968-82. Psychol. Med. 15, 113-123, 1985.

Suicide and unemployment in Italy I201

53. Crombie I. K. Can changes in unemployment rates explain the recent changes in suicide rates in developed countries? Inr. J. Epidem. 191, 412-416, 1990.

54. Platt S. and Kreitman N. Trends in parasuicide and unemployment among men in Edinburgh, 1968-82. L3r. Med. J. 289, 1029-1032, 1984.

55. Jackson P. R. and Warr P. Mental health of unemployed men in different part of England and Wales. Br. Med. J. 295, 525, 1987. *

56. Florenzano F. La salute dei disoccupati. Psichiufriu- Informa:ione 4, 22-27, 1986.

57. Kreitman N., Carstairs V. and Duffy J. The association of age and social class with suicide among men in Great Britain. J. Epidem. Corm. Hlth In press.

58. Barraclough B., Bunch J., Nelson B. and Sainsbury P. A hundred cases of suicide: clinical aspects. Br. J. Psychiat. 125, 355-373, 1974.

59. Robins E., Gassner S., Kayes J., Wilkinson R. H. and Murphy G. E. The communication of suicidal intent. A study of 134 consecutive cases of successful (completed) suicide. Am. J. Psychiuf. 115, 724-733, 1959.

60. Dorpat T. and Ripley H. S. A study of suicide in the Seattle area. Comp. Psychiur. 1, 349-359, 1960.

61. Brown J. H. Suicide in Britain. More attempts, fewer deaths, lessons for public policy. Archs. gen. Psychiat. 36, I 119-l 124, 1979.

62. Tuckman J. and Lavell M. Study of suicide in Philadel- phia. Publ. Hlrh Rep. 73, 547-553, 1958.

63. Ovestone I.M.K. An Epidemiological Study of Suicidal Behaviours in Edinburgh. M.D. thesis, University of Edinburgh, 1972.

64. Stearns A. W. Suicide in Massachusetts. Ment. Hyg. 5, 752-777, 1921.

65. Breed W. Occupational mobility and suicide among white males. Am. Social. Rev. 28, 179-188, 1963.

suicide and unemployment in italy: description, analysis and interpretation of recent trends

Documents