3.1 Data
To test our hypotheses, we used individual-level data from the European Social Survey (ESS) round 5, conducted in 2010/2011. Data were collected through face-to-face interviews held with individuals aged 15 and over, residing within private households, regardless of their nationality, citizenship, language or legal status. Samples are representative at a country level and in general response percentages are high; the overall response rate is above 60 percent (European Social Survey
2012). For more information about the data, see
http://www.europeansocialsurvey.org.
The original dataset contains information on 50,781 respondents across 27 countries: Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Israel, Lithuania, the Netherlands, Norway, Poland, Portugal, Russian Federation, Slovakia, Slovenia, Spain, Sweden, Switzerland, the United Kingdom and Ukraine. Cyprus and Israel were left out of the analyses, for they do not fall into the typology of media systems, and there are no theoretical reasons to assume they do belong to one of the above mentioned types of media systems. Furthermore, only people between 25 and 75 years of age are included in the analyses, for there is possible health selection above the age of 75, and people younger than 25 have often yet to finish their education, possibly resulting in bias. After this age selection and listwise deletion of respondents with missing values on the individual characteristics, we estimated our models with 36,692 respondents across 25 European countries.
3.2 Measurements
To measure
general health, covering both physical and psychological health, respondents were asked: “how is your health in general?” The answer categories were: (1) “very bad”, (2) “bad”, (3) “fair”, (4) “good” and (5) “very good”. The categories are considered to be metric. Self-assessed health has been shown to be a reliable and valid measurement of health (Chandola and Jenkinson
2000; Lundberg and Manderbacka
1996). By using this measurement of self-assessed general health, we followed a prominent tradition in epidemiological research (Huijts and Kraaykamp
2011a). For the descriptive statistics (of all variables) we refer to Table
2 in the appendix.
To measure media exposure, four types of mass media exposure were assessed: television, radio, newspapers and internet. The first three items were measured with the questions “on an average weekday, how much time, in total, do you spend [watching television/listening to the radio/reading the newspapers]?” The answer categories range from (0) “no time at all” to (7) “more than 3 h”. These answer categories were considered to be continuous. Because a negative, yet nonlinear relation was found between radio and newspaper exposure with self-assessed health, we use dichotomous categories for both radio and newspapers exposure: (0) “no time at all”, versus (1) the other categories (see also the paragraph on linearity further on). For the use of internet, people were asked “how often do you use the internet, the World Wide Web or e-mail—whether at home or at work—for your personal use?” The answer categories ranged from (0) “no access at home or work” to (7) “every day”. These answer options constitute a metric scale, as will be explained at the end of this section.
Social isolation was measured with the single question: “How often do you meet socially with friends, relatives or work colleagues?” The answer categories being (1) “every day”, (2) “several times a week”, (3) “once a week”, (4) “several times a month”, (5) “once a month”, (6) “less than once a month” and (7) “never”. Previous studies used this variable to refer to the informal aspect of social capital (e.g. Savelkoul et al.
2011). We assessed the relation between these answer categories and the dependent variable to be linear, as will be shown at the end of this section.
Fear of crime was measured with the questions: “How often, if at all, do you worry about [becoming a victim of violent crime/your home being burgled]”. The answer categories are (1) “never”, (2) “just occasionally”, (3) “some of the time” and (4) “all or most of the time”. Although these items are often used as a part of a larger scale (Eitle and Taylor
2008; Taylor et al.
2009), several previous studies also relied on this selection of two items (Visser et al.
2013). The scale has been composed using Categorical Principle Component Analysis (CATPCA) in the 22nd SPSS edition.
The scale
perceived ethnic threat was measured with three questions: “Would you say it is generally bad or good for [country]’s economy that people come to live here from other countries?”; “Would you say that [country]’s cultural life is generally undermined or enriched by people coming to live here from other countries?”; and “Is [country] made a worse or a better place to live by people coming to live here from other countries?” The answer categories ranged from 0 to 10. In line with previous studies (Coenders et al.
2004; Visser et al.
2013), we constructed a scale where we used the mean sum score (Cronbach’s α = 0.865). The scale was reversed, so a higher score meant people perceived more ethnic threat. This scale is equivalent across all countries of the ESS (Coenders et al.
2004); our principal factor analyses also pointed in this direction (lowest communality = 0.457, lowest factor loading = 0.676). The relation between these answer categories and the dependent variable were assessed as linear.
Generalized social distrust was measured with the questions: “Generally speaking, would you say that most people can be trusted, or that you cannot be too careful in dealing with people?”; “Do you think that most people would try to take advantage of you if they got the chance, or would they try to be fair?”; and “Would you say that most of the time people try to be helpful or that they are mostly looking out for themselves?” The answer categories ranged from 0 to 10. These questions are derived from the Rosenberg Trust Scale, which has been shown to be reliable and valid for the ESS countries (Cronbach’s α = 0.789) (Reeskens and Hooghe
2008). Our analyses showed that the factor solutions were comparable across countries. (the lowest communality was 0.159, the lowest factor loading 0.398. The other communalities were all above 0.200, the other factor loadings above 0.400.) We assessed the relation between the answer categories and the dependent variable as linear.
Distrust in institutions was measured using the questions how much confidence people have in politicians, parliament, the legal system, the police, the European Parliament and the United Nations. The answer categories ranged from (0) “no trust at all” to (10) “complete trust”. The scale was constructed by calculating the mean of the sum, and is subsequently reversed. It was reliable (Cronbach’s α = 0.912), valid and equivalent (Zmerli and Newton
2008; Zmerli et al.
2007). Principal factor analyses also showed that the factor solutions were roughly comparable across countries, although in eight countries two factors were found. Restricting the factors to one in these countries still provided sufficiently high communalities and factor loadings (lowest communality = 0.195, lowest factor loading = 0.446). The relation between the answer categories and the dependent variable were assessed as linear.
Feelings of unsafety were measured using the question “how safe do you—or would you—feel walking alone in this area after dark?”, with the answer categories ranging from (1) “very safe” to (4) “very unsafe”. Even though single-item indicators might be less reliable and valid, this straightforward question makes it unlikely for interpretation issues to arise (Visser et al.
2013). We assessed the relation between the answer categories and the dependent variables to be linear, as explained at the end of this section.
In the analyses, we controlled for age (in years), age square, sex (0 = male, 1 = female), educational level,
cohabiting status, religiosity and main activity. The respondent’s highest completed level of education was assessed with multiple categories, ranging from (less than) lower secondary to higher tertiary. After testing for linearity, the categories were coded into three categories, namely: (less than) lower secondary, upper secondary and vocational and tertiary education. These categories constituted a linear relation with self-assessed health. The cohabiting status for all respondents was coded by the interviewers, stating if (1) “respondent lives with husband/wife/partner” or whether another situation is at place: (0) “all others”. Religiosity was measured by asking “[a]part from special occasions such as weddings and funerals, about how often do you attend religious services nowadays?” Answer categories ranged from (0) “never” to (7) “every day” and are considered continuous. The main activity of people was measured by asking “which of these descriptions best describes your situation (in the last 7 days)?” The answer categories were: “in paid work”, “in education”, “unemployed and actively looking for a job”, “unemployed, wanting a job but not actively looking for a job”, “permanently sick or disabled”, “retired”, “in community or military service”, “doing housework, looking after children or other persons” and “other”. The two unemployment categories were merged, for reasons of parsimony.
At the contextual level, we added one variable indicating the media system of one’s country. In line with Hallin and Mancini (
2004), we distinguished four types of media systems: the polarized pluralist model (France, Greece, Portugal and Spain), the democratic corporatist model (Belgium, Denmark, Finland, Germany, Netherlands, Norway, Sweden and Switzerland) and the liberal model (Ireland and United Kingdom). The former communist model consists of the countries not classified by Hallin and Mancini (
2004) (Bulgaria, Croatia, Czech Republic, Estonia, Hungary, Lithuania, Poland, Russia, Slovakia, Slovenia and Ukraine). Cyprus and Israel were left out of the analyses. We consider it important to note that Hallin and Mancini’s classification shows great similarities with the types of welfare state systems by Esping-Andersen (
1990).
Finally, following Huijts (
2011), we controlled for two macro-level control variables:
Gross domestic product (GDP) per capita (in US dollars divided by 1000, adjusted for purchasing power parities) and
welfare state expenditure, the percentage of the total health expenditure in a country that is covered by the government (World Health Organization,
2013). GDP per capita was retrieved from the United Nations Economic Commission for Europe (n.d.). Following Huijts (
2011), we used the log linear function to account for the influence of very low and high income countries. Incorporating these macro-level controls avoids spurious relationships, for these controls are linked to general health and differ over media systems.
ANOVA-tests for (deviance from) linearity are conducted for the variables which we originally planned to include as metric variables in the analyses. The results of these tests are presented in Table
3 in the appendix. These analyses indicated that most relations are predominantly linear, even though there are also deviations from linearity. The added value of categorization of independent variables is limited (F-value of deviation from linearity differs <10 % of linear F-value), and therefore these will be included as linear indicators. The exceptions are media exposure to radio and newspaper and religious attendance, leading us to include the media exposure indicators as dummies (0 = no exposure, 1 = exposure), and include all religious attendance categories as dummies (reference is never attending religious services). All metric variables are grand-mean centered.