Methods note “There is space for everyone: Climate crisis messages across the political spectrum”
This brief note summarizes the methods used to collect the data for this project. The research was coordinated by d|part in collaboration with the Open Society European Policy Institute (OSEPI) and the data collection was undertaken by the survey company Bilendi.
The questionnaire used for the survey was developed following an extensive literature review of existing studies into climate change attitudes and perceptions. Some questions were adapted from existing projects and new questions developed, where required. Themes were agreed between d|part and OSEPI, an initial questionnaire drafted and feedback sought from several experts to enhance the questionnaire to address all the goals of the project. The master questionnaire was created in English and subsequently translated by professional interpreters. The translations were then checked by native speaker researchers from the respective countries to ensure the wordings reflected the original questions accurately, while being understandable for general audiences in each country.
Programming and piloting
The questionnaire was programmed by Bilendi using instructions by d|part to ensure the questionnaire could be accessed through a range of devices and browsers. Correctness and user friendliness were checked on desktop computers, laptops, mobile phones and tablets and for Windows, Apple and Android systems respectively (where appropriate). After preparation of the master programming, the approved translations were used to create country-specific versions, which were checked again before commencing data collection. Initially, a soft launch pilot with 50 participants in each country was conducted to check response times and any potential problems. After making small adjustments, the full data collection was carried out. The median completion time for the survey was 18 minutes and 10 seconds.
Sampling and data collection
All responses were collected online through Bilendi between 7 and 25 August 2020. In each country, just over 1000 respondents were surveyed, except for Germany, where over 2100 were surveyed. The age range of respondents was 18 to 74, except for Germany, where 16- and 17-year olds were also included for some additional analyses (for comparative analyses, only the 18–74-year old sample in Germany was included). The upper age limit was capped at 74, because in some of the countries included online data collection is not feasible in older age groups than those, if the goal is to achieve samples that are representative of the respective population. To ensure comparability across countries, the same upper age limit was used.
United Kingdom: Age range: 18–74; Sample size: 1039
USA: Age range: 18–74; Sample size: 1004
Spain: Age range: 18–74; Sample size: 1007
Sweden: Age range: 18–74; Sample size: 1031
Czech Republic: Age range: 18–74; Sample size: 1031
Poland: Age range: 18–74; Sample size: 1043
Italy: Age range: 18–74; Sample size: 1003
France: Age range: 18–74; Sample size: 1017
Germany: Age range: 16–74; Sample size: 2112
The sampling strategy aimed at achieving representativeness to overall population data for the 18- to 74- year olds in each country. To do this, quotas were set in each country for age groups, gender, region and education levels of the respondents, reflecting the distributions of those characteristics in the general population (based on official records). Additionally, to ensure distributions were balanced in the sample, cross-quotas were applied for education levels within each region, as well as age groups within each region. To make sure that respondents were recruited across age groups in as balanced a way as possible, quotas and cross quotas were applied comprehensively for as long as feasible and invitations to survey participants were staggered over a period of 19 days. Where specific cross-quotas could not be filled perfectly, those restrictions were only relaxed gradually in a targeted way towards the end of the data collection process.
Weighting and quality checks
After completing the data collection, deviations from population characteristics for each of the quotas and cross-quotas aimed for were assessed. Overall, the distributions were in many instances very close to population characteristics. Additionally, deviations in the distribution for a cross-quota not programmed (gender within each age group) were computed to account for potential unobserved distortions resulting from the unbalanced participation patterns. Two sets of weights were programmed: those adjusting only for main quotas and those that adjusted for the unprogrammed cross-quota (gender and age) in addition to the main quotas for region and education groups. The distribution of key variables used in the analyses were compared between the results for the unweighted sample, the sample weighted by main quota variables and the sample weighted by the additional cross-quota with remaining main quotas. This was done for each country. The differences between outcomes was very small. For most variables tested and in most countries, differences in estimates between the three weighting approach were less than one percentage point. In no instance exceeded the differences two percentage points. So even when adjusting for deviations from population characteristics, the weights only had a limited impact on the results. The sample overall reflected population characteristics well. In the analyses used in the reports, the simple weights, adjusting for main variables have therefore been applied.
Further details about the methodology, in terms of questionnaire design, sampling and analysis can be obtained upon request from d|part.