Our study sample comprised articles pertaining to the topic of psychiatric disorders chosen from the six most widely read printed newspapers and magazines in both countries. This represents more than 75% of all readers of all newspapers and more than 50% of all magazines [22
]. Articles were retrieved by a media agency and were taken from five one-week periods throughout the year 2007. Articles from both countries were analyzed as a single pool, as the Czech Republic and Slovakia share similar socio-economical, political, and mental health policies. All parts of the newspapers were searched for key words (neutral terms, e.g., depression, dementia, as well as labeling terms, e.g., schizophrenic, alcoholic) covering all the major psychiatric disorders, including news, interviews, columns, and editorials.
We used content analysis [23
] to study articles in which a person with a psychiatric disorder represented the relevant content obtained after establishing the keywords.
Out of all articles obtained after setting the keywords, we preformed relevance sampling of the articles in which the subject of psychiatric disorder represented the relevant content. The initial numbers of articles identified were 1424 in the Czech Republic and 900 in Slovakia. The initial articles' search revealed numerous articles that did not use keywords in association with persons with psychiatric disorders, for example: "Depression on the stadium after losing the game; New anti-corruption law caused anxiety in the parliament, Alcoholic beverages cannot be advertised"; etc. The selection of articles with psychiatric disorder as the relevant content defined our final sample. A total of 375 articles were identified for the further analysis; 203 in the Czech Republic and 172 in the Slovak Republic.
This sampling is a combination of purposive sampling
] focusing on key media, using only the six most read print media, stratified composite sampling
, randomly selecting units over a time period (stratification by weeks or days) which has been identified as the most accurate sampling method for analysing media publications [25
], and then after stratification by weeks or days the third step included relevance sampling
], sampling of relevant content from those media based on a keywords search and the association of the article with psychiatric disorders, either as the main subject of the article or a sideline to another story.
These articles were further analyzed according to the PICMIN instrument's initial version developed for the purpose of this study [27
]. Its development was based on the theoretical framework of content analysis [23
The PICMIN instrument's initial version is composed of descriptive and analytical categories. Descriptive categories were used for easy identification of separate items and for finding the links with analytical categories. Within the analytical ones, aggressive behavior was assessed by two separate subcategories. In the first, the role of the mentally ill person in the violent act was recorded (focusing on whether the person with psychiatric disorder was depicted as a perpetrator or a victim of violent acts). In the second, we identified the particular type of aggressive act (homicide, physical assault, aggression against objects, completed suicide, attempted suicide and self-harm). In assessing the global impression of the article, the items were evaluated according to the presence of stigmatizing/de-stigmatizing statements and coded as either negative, positive, mixed (both statements present), or neutral (none of the statements present). Each analytical category included a paragraph-long definition to facilitate coding [21
Articles were coded as positive if the article: a) supported a positive picture of the mentally ill or psychiatric service by portraying it in a way that a mentally ill person is included in society, able to socially participate; b) presented examples of mental illness professionals, institutions or NGO's providing help to the mentally ill, their families and society; c) articles avoided reinforcing stereotypes of mental illnesses and were respectful of people's rights. Articles were coded as neutral if the article stated the facts in an objective way and did not give information which might sway the reader's perspective on mental illness. Articles were coded as negative if: a) mentally ill persons were portrayed as violent or dangerous; b) mentally ill persons were connected with criminal behaviour, endangering society; c) pejorative and colloquial terms were used and d) media presentations of mental illness promoted negative images and stereotypes. Articles were coded as mixed if both positive and negative impressions were found in their content.
To reflect the overall tone or global impression of the article related to stigma, six positive or negative themes pertaining to the topic of mental health/illness were defined "a priori". A list of themes was generated based on previous research [10
], and through a consultation process with the project's mentors. Each coder was supplied with this list with themes expressed in a list of statements, such as: "Treatment is beneficial"; "People with mental illness can socially function in the community", "People with mental illness are usually violent/aggressive", etc. Coders could choose one or more, if appropriate, of the six "a priori" defined themes, or write their own conclusion with the main message of the article within the open-ended box in the on-line version of the instrument that was used for data entries. These positive/negative themes served as a basis for the coding of the "global impression of the article" [27
Disorders that were included generally corresponded to classifications from the International Classification of Diseases [33
] and were grouped in the following clusters: Organic disorders (F00 - F09), Substance abuse disorders (F10 - F19), Psychotic disorders (F20 - F29), Affective disorders (F30 - F39), Neurotic disorders (F40 - F48), Eating disorders (F50), "Other psychiatric disorders" (F51 - F99) and "Not related to any specific psychiatric disorder". "Other psychiatric disorders" stands for personality disorders, including antisocial personality disorder; child and adolescent disorders, including conduct disorder; mental retardation and sexual disorders.
All categories were defined a priori by the research team during several workshops and were used as a basis for consensual coding. Reliability of the coding among raters in both countries was assured by their uniform training and regular international meetings, in which they discussed possible differences in interpretation. Inter-rater reliability (IRR) was determined for the descriptive and analytical categories using the indices: Average Pair-wise Percent Agreement
(APPA) and Krippendorff's α
(alpha). APPA is a more liberal index, comprising a single comparison of the level of agreement among coders and ratings, whereas and Krippendorff's α
(alpha), is a more conservative index of co-variation applicable to nominal and categorical data, which accounts for agreement expected by chance. Krippendorff's α
(alpha) values ≥.60 were considered reliable; values ≥.75 indicated high reliability [34
]. IRR was calculated with the ReCal ("Reliability Calculator"), an online utility that computes inter-coder reliability coefficients [37
Descriptive statistics were used to present all obtained data. The differences of various frequencies of variables were determined using χ2
tests. For cases in which the χ2
test is not appropriate, the p
value vas calculated using robust non-parametrical Monte Carlo test for independence of rows from columns. For each genuine table 99999999 random tables were made. Each random table has the same marginal totals as its genuine table. The random tables come from a population having independent rows and columns. A p value < .05 was considered statistically significant [38
]. All statistical analyses except the Monte Carlo test were carried out with SAS 9.1 statistical software package