Background A growing number of studies indicate that sensationalist reporting of suicide is associated with increases in suicide rates, but in the light of some negative findings, the issue has remained controversial. The aim of this study was to evaluate the best current evidence on the association between celebrity suicide stories and subsequent suicides.
Methods Literature searches of six data sources (Medline, Psychlit, Communication Abstracts, Education Resources Information Center, Dissertation Abstracts and Australian Public Affairs Database (APAIS)) were conducted. Studies were included if they (1) adopted an ecological design, (2) focused on celebrity suicide, (3) had completed suicide as outcome variable, (4) analysed suicide rates across all suicide methods, (5) used data from after World War II and (6) satisfied basic quality criteria.
Results 10 studies with totally 98 suicides by celebrities met the criteria. The pooled estimate indicated a change in suicide rates (suicides per 100 000 population) of 0.26 (95% CI 0.09 to 0.43) in the month after a celebrity suicide. There was substantial heterogeneity between studies, which was explained by the type of celebrity (entertainment elite vs others) and the region of study, as indicated by mixed-effects meta-regression. The region-of-study–specific effect of reporting a suicide by an entertainment celebrity was 0.64 (95% CI 0.55 to 0.73) in North America, 0.58 (95% CI 0.47 to 0.68) in Asia, 0.36 (95% CI −0.10 to 0.61) in Australia and 0.68 (95% CI 0.51 to 0.85) in Europe. There was no indication of publication bias.
Conclusions Reports on celebrity suicide are associated with increases in suicides. Study region and celebrity type appear to have an impact on the effect size.
- mass media
- famous persons
- mental health
- social epidemiology
- social science
Statistics from Altmetric.com
- mass media
- famous persons
- mental health
- social epidemiology
- social science
Suicide is a major public health problem worldwide, accounting for approximately 1 million deaths each year.1 In spite of major advances in recent years, gaps remain in knowledge about risk and protective factors related to suicidal behaviour.1 ,2 Suicide seems to be a phenomenon that is susceptible to copycat behaviour, and there is a considerable body of literature which suggests that media reporting of suicide may lead to copycat acts.3–10 Most frequently, copycat behaviour following media reports of suicide has been explained by social learning theory. Vulnerable individuals (eg, those in psychosocial crisis) may learn that suicide is an acceptable solution to personal problems from media broadcasts and consequently adopt suicidal behaviour.11
Reporting of celebrity suicides has been at the core of scientific interest and of preventive endeavours in recent years.7 ,9 ,12–17 These suicides are extensively covered in the mass media and are of special concern because the celebrities are revered and may therefore act as particularly strong role models.7 Numerous guidelines for reporting on suicide have been prepared by national public health institutions and international organisations (eg, WHO). All of them acknowledge that there will be considerable public interest in a report of a celebrity's suicide and recommend that reports of these deaths should be sensitive and non-sensationalist.17
Despite this, debate continues about the actual impact of suicide reporting on suicide rates.5 ,18 ,19 There are several narrative reviews available which indicate a substantial heterogeneity with regard to individual study findings, including some negative findings.4 ,5 ,18–21 To date, no systematic meta-analysis has been conducted in which estimates have been provided on the cumulative effect size related to copycat behaviour.
The current meta-analysis was conducted with a view to filling this gap. It systematically identified and brought together the findings from relevant studies. Studies dedicated to the effects of media reporting on suicide can generally be classified into (1) descriptive studies (ie, studies of individual cases or a group of individuals with no comparator) and (2) analytical studies (ie, studies involving a comparison with another group).4 ,20 The analytical studies can be further divided into (2a) individual-level and (2b) ecological studies. The vast majority of analytical studies in this area have been ecological studies and have analysed the possible effects of media reporting on aggregated suicide rates, using ecological before-and-after designs. These studies are generally of low specificity (ie, cannot demonstrate that those individuals who subsequently died by suicide were exposed to the media stimulus).4 ,20 Some individual-level studies have been conducted in this area.4 In these studies, the outcome variable is typically a self-reported attitude towards suicide or a likelihood of suicidal behaviour. Individual-level studies overcome the problem of low specificity but are not appropriate for examining the most important outcome of interest, which is completed suicide.4 For this reason, the current review is limited to ecological studies. The Oxford Centre of Evidence-Based Medicine classifies well-conducted ecological studies as only level II studies, indicating that evidence from these types of studies is generally lower than that from well-conducted randomised clinical trials and cohort studies.22 Practical and ethical considerations make it almost impossible to conduct randomised clinical trials and cohort studies in this area. For these reasons, we aimed to systematically review the available ecological studies in order to evaluate the best current evidence on the association between reporting of celebrity suicides and suicide rates.
Two of the authors (TN and JP) conducted searches of Medline, Psychlit, Communication Abstracts, Education Resources Information Center, Dissertation Abstracts and Australian Public Affairs Database (APAIS) from their respective years of inception to January 2011 to identify studies on the association between media reporting and actual suicidal behaviour.4 ,20 We used the following search elements to identify relevant English-language articles: Suicide (suicid*), Imitation (Werther; copycat; imitat*; contagio*) or Media (media; newspaper*; radio*; televis*; film*; movie*; book*; play*).4 We screened titles and abstracts of retrieved articles and obtained full-text versions of articles reporting on studies that appeared to be relevant. We screened the reference lists of all retrieved articles for further potential articles. Efforts were also made to discover unpublished articles of relevance. We entered the above search terms into Google to identify any unpublished studies via the internet. In addition, we sought unpublished data through our informal networks, most notably the members of the International Association for Suicide Prevention's (IASP) Suicide and the Media Task Force. A protocol of the search strategy is available at the website http://www.mindframe-media.info/client_images/900016.pdf.20
Studies were eligible if (1) they adopted an ecological design comparing two or more time points before and after a suicide broadcast in the same population; (2) they focused on media reporting of celebrity suicide; (3) they had completed suicide as an outcome variable; (4) they analysed suicide rates across all suicide methods (deemed important because the evidence suggests that reports of suicide likely to have method-specific effects on suicide rates)7 ,9; (5) they used data from after World War II (because only a small proportion of the population would have been exposed to the media before World War II compared with the after-war period)23 and (6) met basic quality criteria, namely (i) the presence of a clear definition of the ‘celebrity term’ used in the study or a description of the celebrity analysed, (ii) the adequate reporting of statistical procedures which would allow for the repetition of the analyses made when the original data were available and (3) absence of any obvious flaws in the design or conduct of the study derivable from an in-depth reading of the relevant article. Editorials, comments and all other publications were excluded if they did not present primary or secondary data. Data on celebrity suicides that were reported in several studies were counted only once. This selection process yielded 10 studies. For an overview of the process of study selection, see the flowchart presented in figure 1.
The data extracted for each study were (1) the name of the country where the study took place; (2) the publication year; (3) the mean population size across the observation period of the respective study; (4) the number of celebrity suicides that were analysed; (5) the type of celebrity, based on previously developed taxonomy13; (6) the mean difference in suicide rate in the month before versus month after the media reporting; (7) the standard error (SE) of the difference and (8) a short description of the portrayed suicide(s). For four studies which did not report the difference in suicide rates and the SE of the difference,7 ,13 ,15 ,16 the data were obtained from authors. This was made easy by the fact that we shared coauthors who had access to all the original data of these studies (SS, TN and K-wF, respectively). Population sizes were obtained from relevant national statistical agencies if they were not stated in the particular study. All coders were familiar with the literature analysed. TN, K-wF, PSFY and JP extracted the data independently using a piloted Excel template and compared the extracted data afterwards, reaching an agreement rate of 100%.
Quantitative data synthesis
The effect estimates were the differences in monthly suicide rates between the month(s) when the respective celebrity suicide was reported and a reference period (preceding month(s)). The effect estimates for all 10 studies were pooled to obtain a combined effect size. Initial data inspection indicated potential heterogeneity between studies. Random-effects meta-analysis with restricted maximum likelihood estimation was used to estimate the summary statistics.25 I2 statistics and Cochran's Q test were used to estimate the amount of residual heterogeneity and to test for statistical significance, respectively.26 Publication bias was visually assessed using a funnel plot and quantitatively tested by rank correlation test27 and regression test for asymmetry.28
Sources of heterogeneity
We used mixed-effects meta-regression to determine which factors might be contributing to the observed between-study variation. We examined five variables: region of study (North America (reference group), Asia, Australia and Europe), decade of publication (2000–2011 (reference group) and 1980–1999), number of celebrities investigated (an integer), population size (taking the year average within the observation period of the respective study) and type of celebrity with regard to professional group and national identity (national entertainment elite, foreign or less known celebrity, mixed sample (reference group)).
The region-specific and time-specific variables were included because it was thought that the sociocultural circumstances associated with them might have an impact on the amount of copycat behaviour.5 ,29 ,30 Furthermore, we included the number of celebrities analysed in the respective study because the way the term ‘celebrity status’ is defined in a study may have an impact on any association with suicide rates. Definitions used to define celebrity status vary widely across studies on celebrity suicides and seem likely to be more inclusive of relatively less famous individuals in studies analysing the associations of several celebrity suicides with suicide rates compared with studies dedicated to the analysis of the effect of one specific celebrity suicide.
The population size of the country where the study was conducted might help explain heterogeneity across study findings because population size constitutes an indicator of the complexity of media structures in countries.31
The type of celebrity was considered relevant based on a US study, which analysed data on a large number of celebrities and found that suicides by national elite entertainers were most strongly associated with increases in suicide rates in the population.13 It has been suggested that this finding may be explained by the perceived high social status of entertainers as compared with other famous persons (eg, politicians) and the more pronounced social identification with individuals from the own country as compared with foreigners.7 ,32 Stack suggested, using Dyer's theory of the star, that the suicide of the entertainment celebrity promotes mass audience identification by representing a duality. Stars represent the celebration of ordinary people, by playing the roles of ordinary people on stage, while being rich and famous offstage.13 ,33
The model was statistically adjusted by using Knapp and Hartung34 method to help control the risk of spurious findings. The Variance Inflation Factor was calculated to diagnose the model's multicollinearity. If its value is larger than 10, this is an indicator of severe multicollinearity in the model.35 The number of celebrities was strongly correlated with the year of publication (Pearson's r=0.65) and with population size (r=0.63). The other study characteristics did not show multicollinearity.
We used R meta-analysis package metaphor to conduct all statistical analyses.36
Review by a research ethics committee was not required for this work because of the use of published aggregate data. The study was conducted in accordance with the Declaration of Helsinki.
Identification of studies
Figure 1 shows how we arrived at the 10 studies that were included in the analysis. In total, we identified 102 studies on the association between media reporting of suicide and actual suicidal behaviour. Fifty-five of these were ecological in nature, 49 of which used completed suicide as the outcome variable. From this sample, 16 studies focused on one or several celebrity suicides.3 ,7 ,9 ,10 ,12–16 ,19 ,23 ,24 ,37–40 Of these, one study analysed a specific suicide method but not total suicides.38 One study reported on celebrity suicides that happened before 1930.23 Four studies were replications that used the same data or a smaller amount of the same data, which were also used in another identified study.3 ,19 ,39 ,40 The remaining 10 studies were checked for basic quality criteria as outlined in the methods section. All 10 studies met these criteria and were eligible for the meta-analysis.7 ,9 ,10 ,12–16 ,24 ,37
Table 1 shows the characteristics of the 10 eligible studies. Four studies were conducted in Asian countries—Hong Kong,16 Taiwan (two studies15 ,37) and South Korea9—and one in Australia.12 Two studies were conducted in North America (USA13 and Canada14). There were three European studies that were conducted in Austria,7 Germany10 and France,24 respectively. Four of the included studies investigated the effects of several celebrity suicides (ranging from six to 38 studied suicides)7 ,10 ,13 ,24 and six studies focused either on a single celebrity suicide12 ,14–16 ,37 or provided original data related to one specific celebrity suicide.9 Seven papers were published between 2000 and 2011,7 ,9 ,14–16 ,24 ,37 two were published in the 1990s10 ,12 and one was published in the1980s.13
Quantitative data synthesis
Figure 2 shows the forest plot. The random-effects summary estimate of the 10 eligible studies on 98 celebrity suicides indicated an average increase in the suicide rate of 0.26 (95% CI 0.09 to 0.43) in the month after a report on a celebrity suicide. The heterogeneity test indicated significant between-study variation (I2=91%, Q=66.8, p<0.0001).
Table 2 shows the results of the mixed-effects meta-regression model used to identify possible sources of heterogeneity. Type of celebrity (national entertainment elite) and region of study (Australia) were found to be statistically significant.
The coefficients of the other explanatory variables were not significant. The variables number of celebrities and population size were not used in the analysis because of strong multicollinearity in the resulting model, yielding inflated Variance Inflation Factors. The heterogeneity test for the meta-regression model indicated no remaining statistically significant between-study variation (QE (df=3)=0.40, p=0.94).
In summary, type of celebrity and region of study appeared to explain the observed heterogeneity. If a study of a national entertainment elite's suicide was published between 2000 and 2011, the effect was 0.64 (95% CI 0.55 to 0.73) if the study was undertaken in the North America. If it was an Asian study, the effect size was estimated at 0.58 (95% CI 0.47 to 0.68). If the study was conducted in Australia, the estimate was 0.36 (95% CI −0.10 to 0.61), and if it was a European study, the estimate was 0.68 (95% CI 0.51 to 0.85).
If a celebrity was foreign or less known to the public, the effect was estimated at 0.16 (95% CI 0.01 to 0.31) for studies conducted in the North America and 0.09 (95% CI 0.04 to 0.15) for Asian studies. For studies conducted in Australia and Europe, the coefficients were not statistically significant and estimated at −0.13 (95% CI −0.35 to 0.09) for Australian studies and 0.19 (95% CI −0.01 to 0.40) for European studies, respectively. Figure 3 shows the funnel plot. The distribution of effect estimates seemed to be rather symmetrical, suggesting no direct evidence of publication bias. Both rank correlation test (Kendall's τ=0.31, p=0.21) and regression test (z=0.98, p=0.33) were not significant and supported this interpretation.
This study provides novel cumulative insights into the association between news stories about celebrity suicides and suicide rates. It summarises the strength of the effects in the literature and contributes to the examination of variables that moderate the effect. The pooled estimate of the 10 eligible studies on 98 celebrity suicides indicated an average increase in the suicide rate of 0.26 (95% CI 0.09 to 0.43) in the month after a report on a celebrity suicide. There was substantial heterogeneity between studies, which could be explained by the continent where the study was conducted and by the type of celebrity whose suicide was reported. Effects were most pronounced after reports on suicides by national top entertainers. There appeared to be no publication bias as indicated by a funnel plot and related statistical tests.
The finding that study heterogeneity was partially explained by the study region may indicate differences in reporting practices and quality across regions. Previous research showed that reporting quality varied from country to country.30 This variation was considered to reflect differences in cultural attitudes and values towards suicide.30
The finding that effect sizes for studies on national top entertainers were higher than for other celebrity groups is consistent with earlier literature which identified suicides by national top entertainers to be consistently associated with post-report increases in suicide rates.5 ,13 The high social status of national top entertainers as compared with other celebrity types, and the greater potential of identification with celebrities from one's own country as compared with foreign celebrities may contribute to this finding.32 The lack of any direct indication of publication bias is also noteworthy. Concerns about selective publication of positive findings related to effects of celebrity suicide reporting have been raised in earlier related literature.21
Strengths and limitations
The main strength of the present analysis is the extensive and systematic literature review which used a wide set of search terms and databases. This is likely to have exhaustively identified studies in the topic area. A further strength is the independent extraction of data from the studies by several individuals, which resulted in complete agreement concerning the data used for the analysis. Furthermore, the selection of studies that met predefined quality criteria resulted in the inclusion of studies with acceptable quality.
This study also has limitations. Like the primary studies included in this review, this meta-analysis could only identify temporal associations and could not establish causality.41 Furthermore, our emphasis on celebrity suicides meant that only 10 studies were eligible for inclusion, which had implications for the statistical power of the analysis. We justified this restriction on the grounds that celebrity suicides are the most likely to be reported and to have the largest impact on suicide rates, but it influenced the level of precision of our results.
Implications for research and public health practice
Further research is needed to investigate the potential impact of study region on copycat behaviour. In particular, studies investigating international differences in the quality of suicide reporting and the association between regional reporting quality and suicide rates are needed. The present study also indicates that more uniform approaches to study design and definitional issues are necessary to facilitate a comparison of studies.
This meta-analysis has clear public health implications. It confirms that particular caution is necessary when reporting suicides by celebrities. Celebrity suicides will always be reported because they are particularly newsworthy, and we are not suggesting that they be censored. Instead, we would argue that they should be reported in a responsible manner in view of the existing results. There are vulnerable individuals in the community who would be negatively affected by sensationalist reporting. This recommendation is consistent with international guidelines on media reporting of suicide.17 Public health professionals (particularly those working in the areas of mental health promotion and suicide prevention) should collaborate with media professionals to make sure that reporting minimises the risk of copycat acts.
What is already known on this subject
Reporting of celebrity suicides has repeatedly been found to be followed by increases in suicides.
However, the topic has remained controversial because of some negative findings.
What this study adds
Bringing together the data from individual studies from different countries into a single meta-analysis allows us to conclude that celebrity suicide is associated with increases in subsequent suicides.
The effect size appears to depend on the region where the study is conducted and on the type of celebrity.
Cautious reporting on celebrity suicide is necessary.
The authors thank for the useful suggestions from the reviewers. This work occurred as a result of a collaboration between members of the International Association for Suicide Prevention's Suicide and the Media Task Force (http://www.iasp.info/suicide_and_the_media.php).
Funding This work was supported by the Faculty of Social Science of the University of Hong Kong and a GRF grant (HKU 784210M) and benefited from a working meeting held in Hong Kong in 2010. The funding bodies had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
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.