7.3. Use of statistical techniques in studies of suicide seasonality

The use of statistical techniques for seasonality as well as some other important characteristics of the data was investigated from articles primarily examining the seasonal variation of suicides. The articles, which were located through database searches, had to fulfil the following inclusion criteria: (a) suicide seasonality was one aim of the study, (b) article was published during 1970-97 in a major psychiatric journals, (c) suicide data covered a national or regional population, and (d) the article was written in the English language.

A total of 44 articles from 15 different journal were reviewed. Three journals contained 53% of all articles: the British Journal of Psychiatry (9 articles), Acta Psychiatrica Scandinavica (8 articles) and the Journal of Affective Disorders (6 articles). The number of authors per article varied from 1 to 6 and 33 different names of the first author were discerned.

About half of the reviewed articles used data from European countries. Only two studies came from the Southern Hemisphere. National suicide data were used in 73% of the publications. The lengths of the time series for suicides varied from 1 to 59 years. The total number of suicides was given in 71% of articles. In these articles, the number of suicides ranged from 31 to 26918 per year.

The monthly variation of suicides was studied in 32 articles and the quarter-yearly variation in 10 articles. In addition, the time periods defined by days of a week, lunar phases, and national holidays were frequently used. The monthly values of suicides were reported in 9 articles as frequencies and in 7 articles with units such as monthly mean values or percentage deviation from chance expected values. The monthly values were estimable roughly from graphical presentations in 9 articles. Only 10 articles reported that the suicide rates were adjusted for the calendar effect before statistical analyses were undertaken.

The reviewed studies examined the seasonal distribution of the suicides themselves, compared seasonal patterns between subgroups of a population or related the seasonal suicide patterns to seasonality of a phenomenon other than suicides. Although the use of statistical significance tests was found in most of these articles, their frequency of utilisation varied depending on the type of study that was carried out.

When the purpose of the study was to investigate purely the seasonal distribution of the suicides themselves, a statistical significance test was used in the majority of these articles. The most frequently used of these tests was the chi-square one. Other popular methods were the analysis of variance tests or harmonic and spectral analyses. No substantial change in the use of the statistical tests occurred during the study period, except regarding spectral analyses, which began to be more widely applied in the 1990s.

Subgroup differences in the seasonal pattern of suicides were considered in 30 studies. The most common subgroup variables were the gender and the age group of a suicide victim. Subgroup differences were assessed with some statistical test in only 7 articles. The test was most commonly the chi-square test or a correlation analysis. The rest of the articles based their conclusions concerning subgroup differences either on graphical presentations or seasonal analyses, which were made separately in each category of a grouping variable.

The seasonal pattern of suicides was related to the seasonal pattern of a phenomenon other than suicides in 19 studies. Most often the seasonal trends in suicides were compared with the seasonality of meteorological factors or deaths due to causes other than suicides. A statistical significance test was used in 13 articles. Most frequently researchers were utilising the chi-square test or a correlation analysis. In 6 articles the interpretation of the findings was based on graphs or the seasonality analyses, which were made separately for each time series to which the suicide seasonality was related.

The adequate description of statistical techniques (mathematical formulae, brief presentation of the method, statistical references or appendix) was found in 20 articles. In 17 articles, such standard and well-known methods were used that the name of the test was considered to be satisfactorily provided for and the exact name of the test as well as that of relevant statistics were clearly apparent from the text. A total of 25 articles contained at least one drawn graphical presentation. The name of the statistical software was reported in only 11% of the articles.

Based on the results of this study it is suggested that statistical and methodological approaches must be standardised in suicide seasonality studies so that comparisons between different studies are possible and, moreover, useful. Furthermore, the total number of suicides must always be given and definitions, for example, the timeframe “winter” or “month” (not all have the same number of days) should be clearly stated. Enough details on the types of statistical techniques used in a study and an explanation why the latter was chosen should be provided.

A review study like the present one, which examined the statistical and methodological issues of published articles, is one way to increase the awareness among researchers on how important it is to pay attention to the statistical quality of a study. When the basic elements, such as the study design, characteristics of the data, and statistical methods, are clearly and unambiguously presented in an article, a study gets the attention that it deserves. Thus, it is very important to continue studying and addressing the statistical and methodological issues, even if they focus on only a limited aspect of a scientific problem as has happened, for instance, in this study.