Article Text

Download PDFPDF

A conditional model for estimating the increase in suicides associated with the 2008–2010 economic recession in England
  1. Carme Saurina1,2,
  2. Basili Bragulat1,
  3. Marc Saez1,2,3,
  4. Guillem López-Casasnovas3,4,5
  1. 1Research Group on Statistics, Econometrics and Health (GRECS), University of Girona, Girona, Spain
  2. 2CIBER of Epidemiology and Public Health (CIBERESP), Madrid, Spain
  3. 3Center for Research in Health and Economics (CRES), Pompeu Fabra University, Barcelona, Spain
  4. 4Department of Economics and Business, Pompeu Fabra University, Barcelona, Spain
  5. 5Barcelona Graduate School (BSGE), Pompeu Fabra University, Barcelona, Spain
  1. Correspondence to Professor Marc Saez, Research Group on Statistics, Econometrics and Health (GRECS), CIBER of Epidemiology and Public Health (CIBERESP), University of Girona, Campus de Montilivi, Girona 17071, Spain; marc.saez{at}udg.edu

Abstract

Background Although evidence of the effects of the economic crisis on suicides is quite low, a recent article shows that the increase in suicides in England between 2008 and 2010 could be associated with the rise in unemployment. Our study analysed whether this effect was the same for all regions of England, using a conditional model which explicitly allows estimation of regional time trends and the effects of unemployment on suicides at the regional level.

Methods Hierarchical mixed models were used to assess both, suicides attributable to the financial crisis and the association between unemployment and suicides. The number and the (age-standardised) rate of suicides, for men and women separately, were the dependent variables. We considered the nine English regions based on the NUTS 2 level.

Results There was an (not statistically significant) increase in the number of suicides between 2008 and 2010. The variation in rates was not statistically significant in England as a whole but there were statistically significant increases and decreases in some regions. Statistically significant associations between unemployment and suicides were only found at regional level. For men, statistically significant unemployment rates were positively associated with age-standardised suicide rates in the South West (0.384), North West (0.260) and North East (0.136), and negatively associated in the East of England (−0.444), East Midlands (−0.236) and London (−0.168).

Conclusions The study provides evidence that, even with statistically significant associations, finding variability, but no clear pattern, between trends and associations and/or numbers and rates might in fact suggest relatively spurious relationships; this is a result of not controlling for confounders.

  • Public Health
  • Statistics
  • Suicide
View Full Text

Statistics from Altmetric.com

Introduction

The economic crisis, which for most European countries began in 2008, has caused an increase in suicides? Unlike general mortality, for which there is some controversy over whether there is an increase or decrease in periods of economic crisis,15 increases in unemployment have been associated with increased mortality from intentional violence, including suicides.1 In the USA, between 1928 and 2007, suicide rates increased during economic downturns, and decreased during economic expansions, but only for people between 25 and 64 years old.6 In Korea, during the period 1993–2008, fall in gross domestic product (GDP) was statistically significantly associated with increase in suicide rates, 1–4 months later.7 In the UK, for every 1% increase in unemployment, suicides in people under 65, increased by 0.79% and 0.79% of homicides from 1970 to 2007.1 Finally, according to Kentikelenis et al,8 suicides in Greece increased 17% from 2007 to 2009, and 25% from 2007 to 20108 ,9 (unofficial data).

However, evidence of the effects of the economic downturn on suicides is quite low. In this sense, we proceeded to conduct a literature review, searching MEDLINE, EMBASE and Scopus|SciVerse, in March 2013 using the term ‘suicides’ combined with the terms ‘economic recession’, ‘financial crisis’ and ‘economic downturn’, limiting the review to articles published in the last 5 years in English, French or Spanish. We found 44 citations. Most of them, however, contains narrative review, merely discussed hypothesis or proposed future research lines. Only seven articles provide quantitative evidence regarding the current recession.6–8 10–13

These include the paper of Barr et al.10 In this seminal paper, the authors conclude that the increase in suicides in England in the period 2008–2010 could be associated to the rise in unemployment, a consequence of the economic recession which began in 2008. The problem is that, although they use regional differences in suicide and unemployment rates between 2000 and 2010 to test the hypothesis that those regions with greater rise in unemployment have had corresponding increase in suicides, we believe that they have succeeded only partially in this. What we mean is that, through more observations, they manage to increase the efficiency of the estimates of both the time trend for suicides and the association between unemployment and suicides. However, they have not been able to estimate trends or associations at regional level, despite there being great heterogeneity between the two. In fact, they use a mixed design,14 specifically, a longitudinal one. That is to say, data were collected in order for geographical units (local areas in their case) to be observed over time.

Figure 1 shows temporal trends from 2000 to 2010 for the number of suicides in the nine regions of Englanda and with the thicker broken line as the number of suicides in England.. As we can see, temporal trends in number of suicides display important regional differences. Likewise, in table 1 we can see that, although suicides (both in numbers and age-standardised rates) decreased on average between 2008 and 2010 compared with 2000–2007 for both sexes in England as a whole, there were notable regional differences.

Table 1

Descriptive of the variables of interest

Figure 1

Temporal trends for the number of suicides in the nine regions of England and in England as a whole (thicker broken line). (A) Men: North West, South East and South West with a thicker line. (B) Women: North West, South East and East Midlands with a thicker line. Access the article online to view this figure in colour.

The nine regions analysed have very different characteristics. The North East, West Midlands and Yorkshire and the Humber are the most deprived regions with high levels of unemployment and their economies are oriented mainly to the tourism and manufacturing. Although West Midlands was the industrial heartland of Britain, especially centred on Birmingham, the crisis has caused a great loss of workplaces. North West England as South West England and East Midlands regions are very diverse. North West population is historically linked to the textiles industry but it is also a rich farmland. In the South West, there are very large differences in prosperity between the eastern parts and the western parts of the region. Agriculture and tourism are important in many parts of the region but in other parts there are engineering, shipbuilding, electronics and food-processing industries. East Midlands is the fourth largest region in England and is known historically for its food. The region has the second lowest overall population density in England (after South West England). Economically the East Midlands bear similarity to South West England with some industrial counties and others primarily agricultural. Finally the East of England, historically called East Anglia, and the South East of England are the most prosperous areas together with Greater London. The East of England is a very flat area; a fact that makes it very suitable for agriculture and fisheries in this region is its most important industry, but we must also consider its high-technology industries. The South East of England is the most populous English region with the second largest regional economy in the UK (after London) making South East England the second richest region per capita after Greater London.

What the table and the figures show are the characteristic features of a mixed longitudinal design, that is, (1) units (regions in this case) do not behave in the same way over time; (2) the effect of the explanatory variables on the dependent variable may not be the same for the different units and (3) longitudinal observations within the same higher-level unit (regions) are not independent of one another.

These features require the application of appropriate statistical methods. In this sense, with respect to the first two features, rather than a pooled model either marginal (also called population-average) or conditional (also subject-specific) models must be used. Marginal model are constructed directly to describe marginal means and treat any covariance structure as nuisance parameters14 That is to say, in our case we are interested in the average time trend for the nine regions, or the average relationship, if any, between unemployment and suicides, but not in regional trends or relationships at regional level. In contrast, in conditional models (also known as conditionally specified models, or mixed models) the coefficients have region-specific interpretations, allowing us to observe regional differences in suicides and/or unemployment, again, if there are any. In addition, however, any conditional model leads to a specific marginal model.14

Our aim in this paper was to use a conditional model which explicitly allows estimation of regional time trends and the effects of unemployment on suicides at the regional level for England during the period 1993–2010. Our hypothesis is that neither the trends nor effects were the same for all regions of England, at least for the period of the economic recession, 2008–2010. In fact, regional differences were so significant that averaging them would mean the loss of valuable information.

Methods

Data setting

In this study, suicides were defined as deaths with an underlying cause of intentional self-harm or injury/poisoning of undetermined intent.15 ,16 For England, this corresponds to intentional self-harm (International Classification of Diseases (ICD)-9 codes: E950-E959; ICD-10: X60-X84); and to injury/poisoning of undetermined intent (ICD-9: E980-E989, excluding E988.8; ICD-10: Y10-Y34, excluding Y33.9 where the coroner's verdict was pending for the years 2001–2006; from 2007 onwards, deaths which were previously coded as Y33.9 are coded as U50.9). Note that we followed the definition provided by the Office of National Statistics (ONS)15, which led us, unlike Barr et al,10 to exclude the code Y33.9.

Annual suicide data for the nine regions of England for the period 1993–2010 (adults aged 15 years and over), including number of deaths and age-standardised suicide rates (rates/100 000 population standardised to the European Standard Population), and excluding deaths of non-residents, were obtained from the UK ONS.15

Although on average (ie, for England as a whole) there were no important differences between the number of suicides and the age standardised rates of suicides, some differences were notable for some regions, in particular the North East and East of England for men and the West Midlands (2008 onwards) for women (see figure 2). These differences could affect the results of the estimates, namely, it is not indifferent the use of the two measures to pick the number of suicides.

Figure 2

Number and age-standardised rates of suicides in selected regions of England, 2000–2010. (A) North-East, men. (B) East of England, men. (C) West Midlands, women.

Unemployment data for the period 2000–2010, both number of unemployed (persons aged 16–74) and unemployment rate (as a percentage of the economic active population), were obtained from the EUROSTAT (regions and cities, ‘regional labour market statistics’).17 All data were obtained in a stratified manner for men and women. According to EUROSTAT17 unemployed persons comprise persons aged 16–74 (in the UK, Spain, Iceland and Norway) who were (all three conditions must be fulfilled simultaneously): (1) without work during the reference week; (2) available for work at the time and (3) actively seeking work, or who found a job to start within a period of at most 3 months. Unemployment rate was defined as the ratio of unemployed on active population (employed and unemployed persons). EUROSTAT does not provide data on unemployment or GDP at regional level before 1999.

Statistical analysis

Hierarchical mixed (also conditional) models were used to assess both suicides attributable to the financial crisis (using a dummy variable for the crisis years, 2008–2010) and the association between unemployment and suicides. In both cases, we used the number and the (age-standardised) rate of suicides, for men and women separately, as the dependent variables.

On assessing the suicides attributable to the financial crisis we included, as explanatory variables, a time trend and the dummy variable. In this case, as in Barr et al,10 the dummy variable was designed to capture a break from past time trends.Embedded Imagewhere the subscript i denoted region; the subscript t time; Y was the dependent variable (suicides); D was the dummy variable (takes the value 0 for 2000–2007 and 1 from 2008); T was the time trend (t=1,2, …, 11); u a Gaussian random term and β denoted unknown parameters.

When assessing the association between unemployment and suicide we estimate two models: with the number of suicides the number of unemployed and a time trend were included as explanatory variables while with the age standardised rate of suicides we include the rate of unemployment and the time trend. In this case, a time trend was included in order to control for a possible spurious relationship, as suicides and unemployment evolve over time in the same way.Embedded Imagewhere U denoted unemployment; and for the rest the same meaning as above.

Both the intercept and the coefficients associated with all of the explanatory variables were considered random effects. That is to say, we allowed all coefficients to vary in the higher level unit considered, that is, the region. In the cases of the intercept and the coefficients associated with the dummy variable and unemployment, we assumed that the random effects were identical and independent Gaussian random variables with constant variance. In the case of the time trend, we assumed that the random effects vary by region and year, through a random walk of order 1 (ie, independent increments) for the Gaussian random effects vector18 (although we have also assumed a constant variance). Thus, we use a non-parametric approach to regional trends, which we assume to evolve non-linearly (see figure 1).

In all models we controlled for heteroskedasticity, the consequence of heterogeneity between regions, through the random intercept (at the region level), and for autocorrelation, that is, serial dependence of the longitudinal observations within the same region, through an autoregressive model of order 1 (within each region).

The inferences were performed using a Bayesian approach, with the Integrated Nested Laplace (INLA).19 All analyses were conducted using the free software R (V.2.15.2),20 through the INLA library.19 ,21

Results

We can see in figure 1 that since 2008 (the first year of economic recession in the UK), the regional trend has been similar to the national trend in only three regions (the North West, South East and South West) in the case of men, and other three (the North West, South East and East Midlands) in the case of women (marked with a thicker line in figure 1) and slightly similar (with a less pronounced increase between 2008 and 2009) in another two (the North East and East Midlands) in the case of men, and one (the South East) in the case of women.

In figure 3 we show a scatter plot of the age-standardised rate for suicides and the variation in unemployment rates from 2000 to 2010, with the (loess smooth) fitted lines representing each of the nine regions and the thicker broken line England as a whole. In this case, only the West Midlands (in the case of men; marked with a thicker line in figure 3) in some way resembles the line for England as a whole.

Figure 3

Age-standardised suicide rates and variation in unemployment rates, 2000–2010, and (loess smooth) fitted lines in nine regions of England and in England as a whole (thicker broken line). (A) Men: West Midlands with a thicker line. (B) Women. Access the article online to view this figure in colour.

In table 1, we can see that, for men, in the East of England and South West there were notable increases in the number of suicides (326 in 2000–2007 vs 346 in 2008–2010; 340 vs 363, respectively). However, age-standardised rates were only a little higher in the South West (17.09 vs 17.43). For women, there was a slight increase in the number of suicides (87 vs 90) in the East Midlands, although age-standardised rates remained constant.

Table 2 shows the results of estimating the variation in suicides during the period 2008–2010 of the economic recession compared with 1993–2007. Note that, although there was an increase in the number of suicides in 2008–2010 compared with 1993–2007, higher in men than in women (13.043 vs 4.530, respectively), no cases were statistically significant. In addition, variations in suicides at the regional level were virtually identical (all coefficients were very close to zero) to those occurring in England as a whole. It was a very different picture with age-standardised rates for suicide, however. Although the variation in such rates was not statistically significant in England as a whole, there were important and statistically significant increases in suicide rates in the North West (2.001 for men, and 0.460 for women) and North East (men, 1.828), and decreases in the East of England (−1.597 for men and −0.532 for women) and West Midlands (women, −0.333).

Table 2

Results of estimating the variation in suicides during the economic recession 2008–2010 compared with 1993–2007

The results of estimating the association between unemployment and suicide in the period 2000–2010 are shown in table 3. In no case was a statistically significant association found for England as a whole. However, statistically significant associations were again estimated for regions (with the exception of the relationship between unemployment rates and age-standardised suicide rates in the case of women). With respect to the association between the number of unemployed and the number of suicides, statistically significant positive associations were found for the South West (1.099 men and 0.584 women), North West (0.968 men and 0.309 women) and London (0.362 men and 0.210 women), and negative associations for the North East (−3.051 men and −1.868 women), East Midlands (−1.470 men and −0.798 women), East of England (−0.642 men and −0.391 women) and Yorkshire and the Humber (−0.505 men and −0.479 women); in all cases for both men and women. For men, statistically significant unemployment rates were positively associated with age-standardised suicide rates in the South West (0.384), North West (0.260) and North East (0.136), and negatively associated in the East of England (−0.444), East Midlands (−0.236) and London (−0.168).

Table 3

Results of estimating the association between unemployment and suicide in nine regions of England, 2000–2010

Discussion

We found that, compared with 1993–2007, during the period of economic recession from 2008 to 2010, there were statistically significant variations only in suicide rates, not in the number of suicides, and only at the regional level, not at the level of England as a whole.

In view of our results, it seems clear, therefore, that first, it could not be possible to interpret the temporal evolution of the number of suicides independently of the temporal evolution of the age-sex distribution of such number. Second, when there is a mixed design, with variation in at least two dimensions (between the regions of England-spatial dimension along the time-temporal dimension) appropriate statistical models should be used to represent such variability.

The only two regions with statistically significant increases in the suicide rate were North West and North East. The economy of the North West region, based mainly on manufacturing, was not able to thrive during the economic recession in the face of global competition, causing large-scale migration from the area. The North East is not known as an industrialised region, its economy being primarily oriented towards tourism.

In the East of England and West Midlands, on the other hand, suicide rates (statistically significant) decreased during the economic recession. The East of England is one of the fastest growing regions, with important high-tech industries such as electronics, aerospace and telecommunications. The West Midlands region is the UK's industrial heartland. The unemployment rate increased (by 17.26% for men and 10.25% for women) between the two periods of study and although the number of jobs in the service sector has grown, this has not been enough to offset job losses in the manufacturing sector.

During the study period (2000–2010), we have not found an association between suicide and unemployment for England as a whole. The estimated associations at regional level, however, varied both in sign and magnitude.

For men, statistically significant unemployment rates were positively associated with age-standardised suicide rates in the South West, North West and North East. For the regions of the North West and North East, these results are not surprising as the (real) annual percentage change in the number of suicides is positive and unemployment rates increased in both regions (see table 1). In the case of the South West, the figures in table 1 clarify our understanding of the estimate, as it is the only region with a positive increase in the suicide rate between the two study periods.

The negative associations between unemployment rates and age-standardised suicide rates correspond to the East of England, East Midlands and London. In the case of the East of England, the result is consistent with the values observed in the two tables. Table 1 shows a decrease in the number of suicides of 1.19% and an increase in unemployment of 13.82% between the two periods of study and table 2 indicates a decrease in the estimated annual variation of the suicide rate by 1597 for men. For the East Midlands and Greater London, table 2 shows no significant results with regard to estimated annual change. However, table 1 indicates a decrease of 7.78% in the suicide rates and a 4.61% increase in the unemployment rate between the two periods for the London region and a decrease of 14.53% in the suicide rate along with an increase of 13.59% in the unemployment rate between the two periods of study for the East Midlands. One explanation for the negative association found in the case of the East Midlands might be that it is a region made up of a patchwork of counties, some more industrial and some more agricultural in nature.

We believe that the variability in sign and in magnitude, both in the temporal evolution of suicides and, above all, on the association between suicide and unemployment, could suggest that, in fact, we could have estimated relationships (relatively) spurious, probably as result of not having enough control for potential confounders. The fact of introducing a (non-linear) time trend in all estimated models, suggests us that these confounders could have only a spatial dimension (regional level).

Although with quite limited evidence, an important confounder could be the urban/rural location of suicide. Many studies (but not all) found that rates of suicide are often higher in rural than urban areas.22 In a systematic review, Hirsch23 found that there are characteristics specific to rural locales and individuals, such as geographic, psychological and sociocultural barriers to treatment of suicide, which may increase the risk for suicide. Although quite old, Saunderson et al,24 found that for England and Wales, whereas female suicides (standardised mortality rates) were highest in urban areas, male suicides presented an excess in rural districts. Saunderson et al24 pointed out three factors that could explain such differences, difference in suicide methods, the likelihood of communicating suicidal intent and, perhaps, variations in access to psychiatric services. Recently, it has been argued that loss of reputation as a result of unemployment could be highly prominent in rural areas.25 On the other hand, poverty, one possible cause of suicide, is an urban phenomenon, although according to Qin,26 suicide risk associated with urbanicity varies significantly by sex and across age. Living in a more urbanised area reduces the risk of suicide significantly among men whereas it increases the suicide risk among women. Moreover, he pointed out that there has been a greater decline in the risk of suicide among men living in urban areas, compared with those men living in rural ones.

The general problem is the lack of available data at the individual level and, with regard to our study, the geographical units are too heterogeneous to characterise them as rural or urban. This lack of availability of data at the individual level is also present in other possible confounders such as long-term unemployment and the proximity (of unemployment) to retirement age. In fact, what is observed at the aggregate level is not observed at the individual level. Considering at the individual level the reason for a suicide (professional, personal, mental health issues, etc) is very difficult and appears to be an important limitation for this studyb.

Another limitation of our paper could be the change of classification from ICD-9 to ICD-10 during the period under study, which can imply artefacts in the series. However, this same limitation could also be found in Barr et al.10

What is already known on this subject?

  • The evidence of the effects of the economic recession on suicides is quite low. Only a small number of articles provided quantitative evidence and a recent article concludes that the increase in suicides could be associated with the rise in unemployment.

  • Unlike general mortality, for which there is some controversy over whether there is an increase or decrease in periods of economic crisis, increases in unemployment have been associated with increased mortality from intentional violence, including suicides.

  • The evidence of the effects of the economic recession on suicides is quite low.

  • Only a small number of articles provided quantitative evidence and a recent article concludes that the increase in suicides could be associated with the rise in unemployment.

What this study adds?

  • Applying a conditional model which explicitly allows estimation of trends and the effects of unemployment on suicides at the regional level, we are able to show that neither the trends nor effects were the same for all the regions.

  • Compared to 1993–2007, during the period of economic recession from 2008–2010, there were statistically significant variations only in suicide rates, not in the number of suicides, and only at the regional level, not at the level of England as a whole.

  • The variability in sign and in magnitude, both in the temporal evolution of suicides and on the association between suicide and unemployment, could suggest that, in fact, we could have not enough control for potential confounders.

Acknowledgments

We appreciate the comments of three anonymous reviewers for improving the manuscript.

References

View Abstract

Footnotes

  • Contributors MS and CS had the initial idea of the paper and conducted the statistical analysis; BB and MS built the database; BB, MS and CS constructed the tables and the figures; MS, CS and GL-C redacted the text and discussed.

  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement Data and code will be available in the GRECS'Web http://www.udg.edu/grecs.

  • a The nine regions are: North East, North West, Yorkshire and the Humber, East Midlands, West Midlands, East of England, London, South East and North East (Sources: Office for National Statistics15 and NUTS2 EUROSTAT16).

  • b The limitations pointed out in this paragraph were indicated by one of the anonymous reviewers.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.