| Seasonal variation of suicides and homicides in Finland: With special attention to statistical techniques used in seasonality studies. | ||
|---|---|---|
| Prev | Chapter 5. Results | Next |
A total of 44 studies on suicide seasonality were found trough database searches (V: Table 1). Of all articles, 21% were published in 1970-79, 39% in 1980-89 and 41% in 1990-97. The number of authors per article ranged from 1 to 6 (median 2 authors), and 33 different names of first author were found.
These reviewed articles were published in 15 different journals. Three journals published 52.5% (n=23) of all reviewed articles: British Journal of Psychiatry (9 articles, 20.5%), Acta Psychiatrica Scandinavica (8,18.2%), and Journal of Affective Disorders (6,13.6%). The rest of the articles were published in the following journals (in alphabetical order, number of articles in parenthesis): American Journal of Epidemiology (2 articles), American Journal of Psychiatry (1), Archives of General Psychiatry (1), Comprehensive Psychiatry (1), Medical Science and Law (2), Psychiatry (1), Psychological Medicine (1), Psychological Reports (2), Psychiatry Research (2), Social Science and Medicine (2), Social Psychiatry and Psychiatric Epidemiology (2), and Suicide and Life-Threatening Behaviour (4).
Of all reviewed articles, 72.7% (n=32) utilised the national and 27.3% (n=12) the regional suicide data. About half of the reviewed studies (24 articles, 54.5%) used data from European countries. Two articles originated from the Southern Hemisphere.
The length of the time series regarding suicides varied from 1 year to 59 years (median 9 years). The total number of suicides was not reported in 13 (29.5%) out of the 44 reviewed articles. Of those 31 articles, which actually reported the size of the data, the annual number of suicides ranged from 15 to 26918 (median 721).
The seasonal variation of suicides was examined with monthly values of suicides in 32 articles (72.7%) and with quarter-yearly values in 10 articles (22.7%). In addition, the seasonal pattern of suicides was analysed by days of a week (8 articles, 18.2%), time of day (4, 9.1%), phases of a lunar cycle (4, 9.1%), national holidays (4, 9.1%), daylight saving time changes (1, 2.3%), the time of week in month (1, 2.3%), and day number (1-365) in year (2, 4.5%).
The number of different time periods varied from one to four periods. Most frequently researchers studied simultaneously suicide seasonality by months and days of a week (8 articles, 18.2%) or by months and seasons (7, 15.9%). However, months and/or seasons were used in 35 articles (79.6%).The monthly values of suicides were reported as frequencies in only 9 articles. In 7 articles, they were presented in other units, such as monthly mean values or percentage deviation from chance expected numbers. When the monthly values were not reported in any way, they, however, could, roughly be estimated from graphical presentations in 9 articles. An adjustment for calendar effects, i.e. the effect due to the unequal number of days in a month, was reported in 10 articles.
The seasonal pattern of suicides was examined with some statistical significance test in 37 articles (84.1%), although the exact name of the test could not be identified in 2 articles. In 6 articles (13.6%), no statistical method for seasonality at all was used. One article investigated suicide seasonality only with relation to other phenomena, not the seasonal distribution of suicides itself.
A chi-square method was utilised in 14 (37.8%) out of those 37 articles, which actually used some significance test for seasonality (V: Table 2). Other methods were analysis of variance methods (7, 18.9%), non-parametric methods (2, 5.4%), and log-linear modelling (1, 2.7%). Of the time series analysis methods, a harmonic analysis was used in 7 articles (18.9%) and a spectral analysis in 4 articles (10.8%). Both the Edwards’ test for seasonality and Rayleigh’s test for seasonality were found in one article, each.
No apparent change was observed in the choice of a statistical significance test for seasonality during 1970-97 except regarding spectral analyses, which were all performed during the 1990s.
A total of 30 studies (68.2%) have considered suicide seasonality in certain subgroups of a population. Of these investigations, gender differences with regard to the seasonality of suicide were investigated in 25 articles (83.3%). Subgroup differences were also reported according to age groups (9 articles, 30.0%), suicide methods (6, 20.0%), geographical region (3, 10.0%), ethnicity (2, 6.7%), marital status (1, 3.3%), social class (1, 3.3%) and occupational status (1, 3.3%). Two studies have examined time trends in the seasonal pattern of suicides.
The number of different grouping variables per article varied from one to four (median 1 variable). Most commonly the age and gender differences were analysed simultaneously (9 articles, 30.0%).
A statistical significance test for subgroup differences was found in only 7 out of 30 articles (23.3%). The chi-square method was used in 3 articles. A Spearman correlation analysis, a cross-correlation coefficient, a two-way analysis-of-variance test, and log-linear modelling were used in one article each. The exact name of the statistical test could not be identified in two articles.
Although in 21 articles no significance test at all was used for subgroup differences in seasonal distribution of suicides, such a difference was frequently reported to exist. In those cases, conclusions were based on either the interpretation of graphical presentations or “eyeball“ comparisons of results obtained from separate subgroup analyses.
The seasonal distribution of suicides was related to the seasonal pattern of a phenomenon other than suicides in 19 (43.1%) out of the 44 reviewed articles. Ten (52.6%) out of these 19 studies investigated an association between suicides and various meteorological factors, such as ambient temperature, length of day, duration of sunlight or humidity.
In 9 articles (47.4%), the seasonal distribution of suicides was related to that of deaths from causes other than suicides (e.g. homicides). In one study, the suicide seasonality was associated with seasonality in sociological (e.g. unemployment rate) and biological (e.g. serotonin) factors. Eight studies (42.1%) explored a seasonal association between suicides and hospital admissions due to mental disorders, traffic accidents, attempted suicides or electroconvulsive therapy.
A total of 13 articles (68.4%) out of those 19 articles, which examined relationships between suicides and other phenomena, used some significance test. The chi-square method was used in 3 articles (15.8%), Pearson’s correlation in 3 articles (15.8%), Spearman’s correlation in 4 articles (21.0%) and a cross-correlation in 4 articles (21.0%). In addition, a logistic regression analysis, multiple regression analysis, a spectral analysis and the Mann-Whitney U-test were utilised in two articles (10.5%) each. In 6 articles (31.6%), researchers based their conclusions only either on the graphical interpretation of the data or results obtained from separate analyses of the studied phenomena.
An adequate description of statistical methods – i.e. reporting of mathematical expressions, descriptions of the basic ideas of the statistical test, statistical references, or a statistical appendix - was found in 20 (45.4%) of all reviewed articles. In 17 articles (38.6%), however, the name of the statistical test without any detailed description was presented, but that was satisfactory, if a standard and well-known statistical method was used like, for instance, the ordinary chi-square goodness-of-fit test. The exact name of the statistical test could not be identified in two articles, although the p-values were reported in the text.
About half of the reviewed articles contained at least one reference to the statistical literature like a statistical textbook or a methodological article. The name of the statistical software was mentioned only in 11% of all articles.
At least one graph was found in 25 (56.8%) of all reviewed articles. The type of graphs were histograms (4 articles, 9.1%), “line diagrams“ or scatterplots (20, 45.5%) or a combination of these (1, 2.3%). In 12 articles (27.3%), the seasonal pattern of suicides was visualised by plotting the monthly values averaged over the study period (months on x-axis). In 6 articles (13.6%), the annual monthly values were plotted against time including (number of years) x 12 cut-off points on x-axis. For example, if the study period had been 10 years, the x-axis included 10 years x 12 months =120 monthly intervals.
In particular, gender differences in suicide seasonality studies were visualised with the help of graphs (12 articles, 27.3%). Of these 12 articles, the seasonal patterns of male and female suicides were plotted in the same figure in 7 articles (58.3%). In 5 articles (41.7%), there were separate graphs for males and females. Unequal scaling of the y-axis, which makes it difficult to compare the extent of seasonality between different graphs, was found in 3 articles.