Systematic review of suicide in economic recession

Table 1 Positive Association Studies

Ref.	Countries studied	Study design	Suicide data source	Recession period	Economic indicator(s) used	Findings
Yang[45]	United States	Time series study Multiple regression analysis	Hollinger P.C Violent Deaths in the United States New York: Guild 1987[63]	Not Specified	Gross National Product per capita Unemployment rate	Total suicide rates were significantly and positively associated with increases in gross national product per capita and unemployment rate (P < 0.05 for both) However, total suicide rates were significantly and negatively associated with increases in gross national product with a one year lag (P < 0.05)
Lester et al[47]	United States and Japan	Time series study Pearson correlation analyses and regression analyses	Annual volumes of Vital Statistics of the United States	Not Specified	Change in Gross National Product Unemployment Rate	Increase in unemployment was significantly and positively correlated with increase in suicide rates in both the United States and Japan (P < 0.05 for both). Regression analyses, however, only showed a positive and significant relationship between unemployment and suicide rates in Japan (P < 0.05)
Gavrilova[16]	Russia	Time series study Descriptive statistics	Gokomstat (Russian statistical committee)	1992-1993	Average real earning and consumer prices	Male suicide rate + 61% and Female Suicide Rate + 22% over 1991-1994 Male morality rate (per 100000) 1991: 47.7 1994: 76.9 Female Mortality Rate (per 100000) 1991: 11.6 1994: 13.2
Ruhm[19]	United States	Panel Study Regression analysis	US Census Bureau	Not Specified	Unemployment Rate	Suicide rate predicted to increase by 1.3% for every percentage point increase in unemployment rate (P = 0.05)
Brainerd[21]	22 former Soviet Bloc Countries	Panel study using regression analysis	World Health Organisation “Health For All” Database 2000	1990-1994	Gross national product per capita and employment to population ratio	A $100 increase in GNP per capita predicted a decrease in suicide rate by 0.14% (P > 0.05) to -0.20% (P < 0.01) in males A one percentage point increase in employment to population ratio predicted decreased suicide rates in males by 3% (P < 0.01)
Lester[42]	United States	Panel Study Regression analysis	Statistical Abstract of the United States	Not Specified	Gross state product per capita Female labour force participation	A one percent increase in gross state product per capita was associated with a 0.45% decrease in total state suicide rates and a 0.55% decrease for male suicide rates (P < 0.05 for both)
Kim et al[28]	South Korea	Time Series Study Descriptive statistics and regression analysis	1999 report on cause specific mortality by the South Korean Statistical office	1997-1999	Unemployment rates and GDP	Suicide cases increased for three months after the recession onset, but decreased after this point Estimated excess suicide mortality caused by recession = 5.1 in 1998 and 0.5 in 1999 (P < 0.01 for both estimates)
Gerdtham et al[12]	Sweden	Cohort Study using descriptive statistics and probit regression	Statistic Sweden’s Survey of living conditions	Not specified	Advanced notification of job loss, changes in GDP, deviation from GDP trends unemployment rates, industry capacity utilisation, and industry confidence indicators	A one standard deviation increase in GDP decreased the risk of suicide by 22.7% (P < 0.05) A one standard deviation increase in the confidence indicator reduces suicide risk by 22.3% (P < 0.05). A one standard deviation decrease in the advanced notification of job loss rate, decreased suicide risk by 21.5% (P < 0.05)
Khang et al[15]	South Korea	Time-series Study Descriptive statistics	Death certificates from South Korean Statistical office	1997-2002	Unemployment rate and GDP per capita	Substantial increase in suicide rates post 1997 in males and females, and then a decline in the rate in both sexes until 2000 Age specific male and female suicide rates (per 100000) in 1996 and 1998: 1-14: MF = 0.8 (1996) and 0.8 (1998) 15-34: MF = 13.5 (1996) and 16.5 (1998) 34-65: M = 26.4 (1996) and 44.3 (1998) F = 8.9 (1996) and 11.9 (1998) 65-79: M = 42.4 (1996) and 58.5 (1998) F = 15.8 (1996) and 22.3 (1998)
Tapia Granados[44]	United States	Time series study Regression analysis	Historical Statistics of the United States	Not Specified	Unemployment rates	Suicide rates significantly associated with increases in the rate of change of unemployment rates from 1920-1944 (P < 0.01) and 1920-1996 (P < 0.01)
Granados[20]	Spain	Panel study Regression analysis	Instituto Nacional de Estadística. (National Statistics Institute)	Not specified	National and province level unemployment rates	Suicide rate predicted to increase by 1.28% for every percentage point increase in the unemployment rate (P = 0.05)
Chang et al[27]	27 European Countries 18 American Countries 8 Asian Countries One South African Country	Time series study	World Health Organisation Mortality Database and The Center for Disease Control Online Database (for the United States)	2008-2010	Gross Domestic Product and Unemployment	Overall suicide rates in men rose 3.3% (95%CI: 2.7-3.9) with a rate ratio of 1.033 (95%CI: 1.027-1.039) and an estimated excess mortality of 5124 (95%CI: 4219-6029) that could be attributed to the recession
Garcy et al[13]	Sweden	Time series study Cox regression	Swedish work and mortality database	1993-1996	Unemployment rates	No excess hazard of suicide mortality attributable to unemployment during the recession in men or women, although an excess hazard post-recession (1997-2002) in men (Hazard ratio = 1.43, 95%CI: 1.31-1.56) and women (Hazard ratio = 1.26, 95%CI: 1.04-1.54)
Chan et al[30]	South Korea	Time Series Study Regression analysis	National Statistical office of Korea	January 2009 to December 2010	Unemployment Rates	Between 2009 and 2010, there was a statistically significant positive association (P < 0.001 for most associations) between suicide rates and national unemployment rates in both the employed and unemployed except in employed men and women aged 50-59 and unemployed men aged 30-39
Madianos et al[32]	Greece	Time series study Descriptive statistics correlation and regression analyses	Vital Statistics Bureau of the Hellenic Statistical Authority	2008-2011	Unemployment Rates Debt as a percentage of Gross Domestic Product	Unemployment and debt as percentage of GDP were significantly and positively correlated with suicide rates ρ = + 0.64, P < 0.001 for unemployment, and ρ = +0.47, P < 0.05 for debt as a percentage of GDP In regression models, increases in unemployment and debt as percentage of GDP were significantly associated with increases in suicide rate (P < 0.05 for debt as a percentage of GDP and P < 0.01 for unemployment)
Phillips et al[25]	United States	Panel study Regression analysis	National Center for Health Statistics	2007-2009/2010	Unemployment Rates	Unemployment significantly (P < 0.05) and positively associated with changes in suicide rates but not in the elderly (65+) or young (15-24)
Pompili et al[33]	Italy	Time series analysis Joinpoint regression	Italian Mortality Database	2007-2010	Gross Domestic Product Unemployment rate	2006-2010: Age specific suicide rates increased in working age men (25-64) by 12% (RR = 1.12; 95%CI: 1.04-1.19), but not in older (65-85+) or younger age groups (1-24)
Reeves et al[26]	Europe and North America	Time series study	World Health Organisation Mortality Database and The Center for Disease Control Database	2008-2010 (Europe) 2008-2009 (Canada)	Unemployment rates	Using 2007 as a baseline, the excess suicide mortality attributable to the recession across the United States, Canada and Europe was 10000

Table 2 Negative Association Studies and studies with inconclusive results

Ref.	Country studied	Study design	Suicide data source	Recession period	Economic indicators	Findings
Hintikka et al[37]	Finland	Time series Study Regression analyses	Causes of death statistics reported by Statistics Finland in 1996	1990-1995	Unemployment Rate and Gross Domestic Product	Female and Male Suicide rates positively associated with increases in GDP (P < 0.05)
Rancans et al[35]	Latvia	Time series study Descriptive statistics	Statistical Yearbooks of the Medical Statistics Bureau of the Latvian Ministry of Health and the Demographic Yearbooks of the Latvian Central Statistical Bureau	1990-1994	Gross Domestic Product and Unemployment rates	Rapid increase in suicide rates after 1989 reaching a maximum in 1993 (42.5 per 100000) The drop in GDP after 1990/1991 and the subsequent rise in unemployment did not correspond strictly with suicide rates changes
Neumayer[40]	Germany	Panel study using regression analyses	The German Federal Statistical Office	Not specified	Unemployment rates for initial analysis and gross domestic product for sensitivity analysis	A decrease in suicides for every percentage point increase in unemployment rate in males (0.98% decrease P < 0.1) and females (1.9% decrease P < 0.05) A decrease in suicides for every percentage point increase in unemployment rate in males (1.22% decrease NS) and females (1.97% decrease in females P < 0.01) Sensitivity analysis found that decreases in suicide rates were not associated with recession
Gonzalez et al[36]	Mexico	Panel study using regression analysis	Administrative records of death certificates	Not Specified	State Gross Domestic Product Per Capita	Decreases in state GDP per capita associated with decreases in the overall suicide rate across all states (P < 0.01)
Mackenbach et al[39]	25 European countries in Western and Eastern Europe	Time series study with regression analysis	International Mortality Data Base of the National Center for Health Statistics of the Centers for Disease Control	Not Specified	Gross Domestic Product	Changes in gross domestic product were not significantly associated with overall suicide rates across all countries in men and women
Saurina et al[41]	England and its sub-regions	Panel study using conditional and hierarchical mixed regression modelling	United Kingdom Office for National Statistics	2008-2010	Unemployment Rates	The increase in suicides between 2008 and 2010 was not statistically significant No significant association between unemployment rates and suicide at the national level, although at the regional level there were statistically significant negative and positive associations
Laanani et al[38]	Austria, Finland, France, Germany, the Netherlands, Spain, Sweden and the United Kingdom	Panel Study using regression analysis	Eurostat mortality database	2008-2010	Unemployment Rates	Overall 0.3% increase in suicide rates for every 10% increase in unemployment overall (95%CI: 0.1-0.5) Associations between unemployment and suicide within countries were inconsistent A positive significant association between unemployment and suicide in three countries: Netherlands, United Kingdom and France was found

NS: Non-significant; RR: Rate ratio; GDP: Gross domestic product.

Table 3 Critique of Positive Association Studies

Ref.
Yang[45]	This study uses aggregated population data and only investigates contemporaneous and very short term lagged effects of economic variables on suicide rates. This study supports the evidence of other studies based in the United States that economic recession can exert a short term impact on suicide rates; although the utility of this study in identifying causal associations is questionable
Lester et al[47]	No individual level evidence included in this study. Suicide rates by age category were not computed
Gavrilova[16]	Mortality data quality is questionable as is cause of death coding. Suicide data could have been inaccurately coded leading to underestimates of suicide mortality. No individual level evidence was included and only descriptive statistical analysis was performed
Ruhm[19]	No individual level data was included in the study. No information on suicide rates by gender was provided. The inclusion of 4 yr lags for unemployment lags, whilst an improvement on similar published studies only provides a short term insight into the interaction between fluctuations in unemployment and suicide rates
Brainerd[21]	Gross national product per capita and employment rates used in the study are likely to be underestimates. This means that regression estimates are more likely to identity regressors as not exerting an effect on suicide rates. The panel used in this study is biased by missing or incomplete data for a number of included countries. No individual-level data was included in the study
Lester[42]	Spatial analyses of suicide rates within states were not performed, thus providing no insight into any clustering of suicide mortality within states. Age specific suicide rates were not included although a breakdown of suicide mortality by ethnicity was. No individual level data was included in the study
Kim et al[28]	Mortality data was derived from death certificates which may not be fully accurate. The use of aggregated population data limits the ability to make causal inferences about the interaction between economic recession and suicide. An age breakdown of suicide rates was not provided
Gerdtham[12]	Unlike most other included studies, a large individual level dataset was collected using a prospective cohort study design. Numerous sensitivity analyses were performed to address potential flaws with the initial study design. However, the decrease in the male mortality rates is concentrated in the first few years of the study period meaning there may be an variable that has not been accounted for in the study that is causing this effect
Khang et al[15]	Concerns have been raised by this study about the quality of cause of death coding in South Korea which may have affected the accuracy of data in this study. Total suicides may have been underreported due to a miscoding of such deaths as being unintentional - as a result of stigma. Only descriptive analyses of suicide data were performed, meaning any omitted variable bias has not been corrected for
Tapia Granados[44]	This study benefits from analysing the impact of a number of economic cycles on suicide mortality over a long period of time. Data on suicide mortality and economic indicators were taken from a range of official statistical sources, to provide data for the century studied in the study. This study does not explore the effect of mediating factors which relate economic recession to suicide. By studying aggregated population data rather than individual level data, this study only provides evidence for an association between recession and suicide
Granados[20]	The study design used is very similar to the design used in Ruhm (2000)[19] and the finding that suicide rates increase during recessions is consistent with Ruhm’s study. This study also expands on Ruhm’s work by providing sex specific mortality rates. However like Ruhm’s study, only short term effects of recession on suicide rates are explored; and the use of panel data can exacerbate attenuation bias resulting from underestimates of suicide mortality in official data. As with all ecological studies, the ability to make causal inferences between recession and suicide is impaired by a lack of individual level data
Berk et al[50]	Seasonality in suicide rates was not adjusted for in this study. No individual level data was included in this study either. Lagged effects of changes in included economic indicator were not computed
Granados[29]	Correlation and regression analyses were both performed in this study to elucidate the relationship between recession and suicide. The time series data in this study has also been detrended to make it more suitable for regression analysis. As the author suggests, the study would have been strengthened if a panel study design had been used as the association between economic recession and suicide could be tested across many sub-regions in Japan
Altinanahtar et al[49]	This study is notable for using auto regressive distributed lags which allow for the exploration of long run relationships between variables. This study’s focus on Turkey is also notable as it is a middle income rather than a high income country and high income countries form the majority of countries studied by the literature. This study however does not study suicides according to age and gender and no individual level data has been included
Chang et al[14]	This study provides one of the first analyses of the effects of the 1998 Asian financial crisis on the main countries in the region who were affected. Of concern is the potential difference in the accuracy of suicide coding practices across studied countries as well as the completeness of suicide data. State specific effects such as political unrest (e.g., in Hong Kong) were not fully controlled for in this study
Tapia Granados et al[18]	Regression analyses of the interaction between the depression and overall mortality were performed but not for suicide rates specifically. Only descriptive statistics were used to describe the interaction between suicide rates and the depression. A breakdown of suicide rates by age or gender was not provided
Stuckler et al[22]	Of note in this study is the analysis of the effect of social protection spending on the interaction between unemployment and suicide that this study performed. Using unemployment as a proxy for economic conditions, the effects of both smaller and larger increases in unemployment on suicide rates was tested as well. Other included studies tended to not perform the aforementioned analyses. This study only analysed the short term effects of recession and as with many of the included studies, this study was analysing the impact of recession on populations rather than individuals
Wu et al[46]	The authors of this study opted to detrend time series data in order to make it more suitable for regression analysis. The risk of omitted variable bias was well mitigated with the inclusion a range of potential confounders. The authors state that the results should be approached with caution due to data limitations, but these limitations were not specified
Ceccherini-Nelli et al[48]	Variations in suicide documentation across the four countries studied may vary which complicates comparisons made between these countries. The co-integration analyses performed in this study allow for the analysis of the long term interactions between economic variables proxying economic conditions and suicide rates; which is a major strength of this study. These analyses however only test for associations rather than causality
Luo et al[43]	The nonparametric analyses used in these studies did not allow for the analysis of the effect of recession severity or length on suicide rates, as all recessions were presumed to be of the same length and severity. Multiple regression analysis was not performed thus increasing the likelihood that omitted variables have biased the results
Stuckler et al[17]	This study represents a very preliminary analysis of suicide data from 2009. This means that any effects of the 2008 recession would not have arisen fully and any conclusions made based on data analysis may be premature
Barr et al[23]	The unemployment rate in this study is based upon the number of claimants of unemployment benefits which may underestimate the true unemployment rate. At the national level, the level of fiscal consolidation within England, does not seem to have been estimated or controlled. This is significant as this variable could influence both the unemployment and suicide rates, thus acting as a confounder. The sub-regions of England investigated in this study have different levels of deprivation and the unemployment rate may have differing impacts on the suicide rate in these regions, which are not expressed in the study itself. Only the immediate effects of economic recession have been investigated as the lagged effects have not been calculated
Reeves et al[24]	United States suicide data may underestimate the true suicide rate. The excess mortality estimates rely on short term extrapolations of pre-recession suicide trends
Lopez Bernal et al[31]	The use of interrupted time series analysis allowed for the controlling of seasonality in time series data. Unlike many other included studies, the extent of residual autocorrelation in the time series data was estimated and a control analysis was conducted; using mortality from accidental falls as an outcome measure rather than suicide mortality. Because of the low number of monthly suicide rates in some regions of Spain, regions within the country had to be combined in order to increase statistical power. This means there were less units of observation for the interaction between economic recession and suicide
Chang et al[27]	Differences in coding and classification of suicide may lead to an introduction of bias between countries, which may be exacerbated by the large number of countries studied (54 countries in this study). The sample of studied countries is also skewed by the presence of a large number of high income countries and an underrepresentation of low income countries; particularly from Sub-Saharan Africa. This means that overall excess mortality estimates may be more of a representation of the relationship between economic recession and suicide in high income countries than in low income countries
Garcy et al[13]	As in [12], this study analyses individual level evidence and the results seem to suggest a causal association. The great strength of this study is its very large sample size of individuals (more than 3 million) and the inclusion of a range of socioeconomic variables within the cox regression analyses that were conducted. As this study provided further analyses where individuals who had previously attempted suicide or had a previous mental health diagnosis were excluded, the relationship between unemployment and suicide in this study is less likely to be spurious; especially considering the fact that this study has been conducted at the individual level. However this study could not test whether mental health deteriorated as a result of unemployment meaning the behaviour of an important mediating factor has not been elucidated in this study
Chan et al[30]	Suicide data derived from death certificates may be more vulnerable to misreporting bias. This study attempts to focus the study of the South Korean population by stratifying by occupational status as well as gender; although the flaws associated with using aggregated population data remain. The inclusion of students and housewives in the “unemployed” label used by the S. Korean national statistical office, may distort the relationship between economic recession and suicide rates as these groups may respond to the onset of recession quite differently compared to those who lost their jobs as a result of recession
Madianos et al[32]	The study only explores the short term effects of economic recession on suicide rates. Monthly and regional suicide rates which may have provided a greater insight into the timing and spatial distribution of suicides were not included
Phillips et al[25]	This study performed age and sex stratified analyses of the association of between included economic variables and suicide rates. The authors acknowledge the potential impact of omitted variables such as gun ownership which were not included in analyses because of a lack of comprehensive data
Phillips et al[25]	This study only provides evidence of changes in suicide mortality during an economic recession. Data is provided up until 2010 but negative GDP growth only stopped in 2013 with a return to recession in 2014. Therefore this study does not capture the full extent of the suicide mortality response to recession conditions
Reeves et al[26]	The confidence intervals for the estimated excess European suicide mortality attributable to the recession are quite wide. The unadjusted difference between baseline suicide rates and recession suicide rates are compared to estimate suicide rates; meaning that the impact of confounders has not been adjusted for, nor has any seasonality in the time series data been controlled

Table 4 Critique of Negative Association and Inconclusive Studies

Ref.
Hintikka et al[37]	Potential confounding variables such as state health expenditure were not included. Finnish suicide data seems to be highly reliable so incomplete data is less likely to be a major source of bias. The finding that rapid increases in unemployment were not associated with increases in suicide rates is interesting, but the lack of individual level evidence means that underlying explanations for this association cannot be arrived at from this study alone
Rancans et al[35]	Only a descriptive analysis of economic and suicide data was performed here meaning the effect of any confounding variables was not controlled for. This study only demonstrates a simple temporal association (or strict lack of it) between unemployment rates, gross domestic product, and suicide rates
Neumayer[40]	The panel study design used here is very similar to Ruhm, 2000[19]. Interestingly on sensitivity analysis using gross domestic product as the main proxy for economic fluctuations, variations in suicide mortality no longer behave pro-cyclically. As with the other included panel studies, attenuation bias may be a problem if suicide statistics underestimate suicide mortality
Gonzalez et al[36]	Age breakdowns for suicide mortality were not provided. No individual level analyses were performed in this study. This study takes great care in ensuring the quality of its mortality data; going as far as to test whether quality of death classification varies with the economic cycle
Mackenbach et al[39]	Differences in the accuracy and quality of cause of death coding and classification between countries and within countries over time may have biased the results of this study. The investigation into the relationship between income and mortality in this study implies a causal relationship between the two, although this is contentious
Saurina et al[41]	Using a different regression model to Barr et al, this study contradicts the findings of Barr et al despite the fact that the study location (England) and the study period are very similar. The authors of this paper argue that Barr et al’s methodology is incomplete and has probably produced a spurious relationship between recession and suicide due to not controlling for confounding variables; particularly those with a regional dimension. Despite claiming to improve upon Barr et al’s methodology, this study is still probably biased by the change in classification from ICD-9 to ICD-10 (although Barr et al’s study would be similarly affected). Furthermore the location of suicide (e.g., rural or urban) may have a confounding effect; and this kind of spatial confounding may not have been adequately controlled for in this study
Laanani et al[38]	This study provides an interesting insight into the question of a causal link between recession and suicide. By performing a sensitivity analysis to check for confounding by a “crisis effect”, this study found that such an effect varied between countries and the association between unemployment and suicide in studied countries was statistically significant, but fairly weak and variable. This however may be a reflection of the quality of unemployment as a proxy for economic conditions in studies that investigate multiple countries at once. This study also emphasises the inherent problem in trying to investigate causal associations between economic recession and suicide using ecological study designs