Data sources
Our analysis was based on publicly available Eurobarometer Survey data collected from 28 EU member states from June 2014 to May 2015. [30] Eurobarometer surveys are conducted in several waves every year and each wave includes a number of core questions and bespoke modules in a wide range of topics. In recent years, a question on life satisfaction has been added to the core Eurobarometer questionnaire. For the purpose of this analysis, we analysed data in waves 81.4 (data collected in May/June 2014; n = 28,004); 81.5 (June 2014; n = 27,910); 82.1 (September 2014; n = 28,050); 82.2 (October 2014; n = 27,868); 82.3 (November 2014; n = 27,901); 82.4 (November/ December 2014; n = 27,801); 83.1 (February/March 2015; n = 27,980); 83.2 (March 2015; n = 28,082); 83.3 (May 2015; n = 27,758); and 83.4 (May/June 2015; n = 27,718). These waves were selected in order to obtain data collected during a 12-month period.
Data was collected through personal interviews in the respondents’ language from a total of 279,092 individuals aged 15 years and older; however, some region-level data were not available for Croatia, hence it was excluded from the analysis; the final sample size was 268,696. A multi-stage random sampling and post stratification weighting were used to ensure samples are representative in terms of age, gender and area of residence. Sampling design and interview protocols were consistent across waves. Eurobarometer reports the country and Nomenclature of Units for Territorial Statistics (NUTS) region of residence. NUTS-2 level detail is available for all countries with the exceptions of Germany and the United Kingdom, where data at the NUTS-1 level was recorded. NUTS is a geocode standard to reference the subdivisions of EU countries, with NUTS-1 representing major socio-economic regions or government office regions and NUTS-2 representing basic regions for the application of regional policies, counties or groups of counties. [31]
The Climatemps database was used to obtain data on weather variables, [32] while macroeconomic and region-level aggregate data were downloaded from the Eurostat database. [33] All data were publicly available and deidentified, hence no ethical approval was required.
Measures
Life satisfaction
Life satisfaction was measured using a one-item measure which asked participants, “On the whole, are you very satisfied, fairly satisfied, not very satisfied or not at all satisfied with the life you lead?” We used the original responses for our main analysis. A sensitivity analysis with an alternative grouping of responses (“very satisfied” vs. “fairly satisfied”, “not very satisfied” and “not at all satisfied”) was also conducted.
Sociodemographic factors
The Eurobarometer survey also collected data on participants’ age (15–24, 35–34, 35–44, 45–54, 55–64, 65+ years), sex (male, female), occupation (employed, house person, student, unemployed, retired), difficulty paying bills (never, from time to time, most of the time) as a proxy of financial difficulties, marital status (single, married, divorced, widowed, other) and area of residence (urban, rural).
Environmental factors
Our analysis includes annual precipitation (presented per 100 litres per square meter/100 L/m2), sunshine (hours[h] per day), and average annual temperature (°Celsius) extracted from ClimaTemps. Weather data were matched to Eurobarometer respondents at the lowest regional level available (NUTS-1 in Germany and the United Kingdom and NUTS-2 in the remaining countries). For each NUTS region, the capital -or the biggest city if the capital was not available in Climatemps- was selected to represent the weather in the entire region. For 46 out of the 220 regions a direct match for climate data was not available in ClimaTemps, in which case the closest city to the capital (straight line distance) was used instead.
Data on land use with heavy environmental impact (% of total land use) was extracted from Eurostat at a regional level. Land use with heavy environmental impact includes mining and quarrying; energy production; industry and manufacturing; water and waste treatment; construction; and transport, communication networks, storage, protective works.
Macroeconomic factors
Data on gross domestic product (GDP) per capita (results presented per 1000 Euros) for the year 2015 was extracted from Eurostat at a regional level. To control for medium-term macroeconomic trends, we compared GDP per capita between 2008 and 2015; regions where GDP per capita decreased over this period were considered affected by the economic crisis.
Statistical analysis
A three-level (country/region/individual) mixed effects ordered logistic regression model, allowing for clustering of observations within region and country, was used to assess the associations of life satisfaction with sociodemographic (age, area of residence, sex, financial difficulties, marital status, occupation), environmental (temperature, sunshine, precipitation) and macroeconomic factors (GDP per capita, having been affected by the crisis). The model was further adjusted for the season when data collection was conducted (summer, autumn, winter or spring). The final specification of the model was decided following considerations of collinearity and comparing alternative models using the Bayesian Information Criterion and Akaike Information Criterion. Ordered regression models account for the ordered nature of the outcome variable. The results are presented as adjusted odds ratio (OR) with 95% Confidence Intervals (95% CI) and are interpreted as the OR of reporting a higher level of life satisfaction for a one-unit change in the independent variable (or compared to a reference category in categorical independent variables). A sensitivity analysis with a multi-level logistic regression model with similar specifications was conducted to compare those who reported being ‘very satisfied’ with all other categories combined. Observations with missing responses in any of the above variables (n = 6122, 2.3% of total observations) were excluded from the analysis. Individuals with complete and missing data were compared using chi-square tests. Descriptive results as presented as weighted % with 95% CI. Survey weights provided in the official Eurobarometer datasets were used as appropriate to control for the complex study design. All analyses were conducted using Stata 15.0 (StataCorp LP, College Station, TX).