The tribes are developed from data drawn from a 10-country Chatham House-Kantar Public survey, conducted in late 2016 and early 2017.
The survey was finalized in English and French in November 2016. The full translation process followed shortly after for all other languages used in the survey. A three-stage process of revision was applied.
The general public survey was conducted between December 2016 and January 2017 among a representative sample of the population in 10 European countries (Austria, Belgium, France, Germany, Greece, Hungary, Italy, Poland, Spain and the UK). At least 1,000 interviews per country were conducted online using Lightspeed Research panels. Quotas were applied on age, gender and region, and deviations were corrected with post-stratification weights. The total sample size was 10,195.
Developing the tribes
Our previous work suggests that there are distinctive groups that exist across Europe with competing sets of attitudes and that individuals within these groups behave similarly at elections. Our goal was to identify and understand these unobserved groups. Put simply, we sought to determine who is most likely to be in such a group and how each group’s characteristics contrast with the other groups.
Latent class analysis is a commonly used statistical tool to generate sub-types or classes from multivariate categorical data, such as attitudes from survey responses. It is mainly utilized when the indicators used to determine the measurement of responses are binary or polytomous (i.e. more than two options). In other words, this is a kind of cluster analysis that puts people into a small number of groups based on their responses to many questions.
Latent class models consist of two parts. One assigns a probability to a person of belonging to a particular class or type. The other describes the relationship between the types and responses to the survey questions.
We selected eight questions from the original survey in order to detect distinctive groups of opinion across Europe. The latent class analysis enables us to estimate the probabilities that an individual belongs to a certain type given their responses to the questions. These questions form the basis of the quiz on this website. Based on their responses to the quiz and the conditional probabilities generated by the model, users are assigned to the tribe to which their attitudes are closest. We use traditional socio-demographic characteristics such as age, education, gender, income and employment status, as well as other attitudinal questions from the survey, as explanatory variables i.e. to describe the profile of each tribe.
The majority of the items or indicator variables selected asked the respondent whether they agreed/were neutral/disagreed with a statement, or otherwise provided three alternative responses to a question. Hence, we use these three part responses (trichotomous ordinal variables) for each of these indicators. The initial aim is to determine the appropriate number of types or ‘political tribes’ that exist across Europe. It is conventional to measure the appropriate number of types in the data using ‘goodness of fit’ statistics. In other words, our initial aim was to seek the most parsimonious model that provides the best fit to the observed data. In our case, using the statistical tools we identified more types than the six-type solution we eventually opted for, but comprised of very small number of members (less than 2% and 1% in type size and numerically a low sample size). We chose to balance the statistical evidence, the substantive meaning of each type solution and theoretical expectations. Utilizing this more pragmatic approach, we identified a six-type solution for the data.
The research and associated website/materials were originally produced in English. Translations from English into other languages are provided as a service to interested parties. However, if any text of this website or associated materials in English is inconsistent with the text in translated versions, the original English text shall govern.
Supporters and Partners
Chatham House is very grateful to Stiftung Mercator, the Robert Bosch Stiftung, the King Baudouin Foundation and the ERSTE Stiftung for their generous support of this project. Chatham House is also grateful to Kantar Public for their help in designing and conducting the survey.