Experimental Design
To identify the causal effects of participation in deliberative democracy, we rely on random treatment assignment. The treatment group consists of 50 people from the pool of registered individuals who serve on the Good Council. Registered individuals are selected into the treatment group based on a random process that is specifically designed to ensure representativeness of the population when selecting members for citizens’ assemblies (Flanigan et al., 2021). The control group comprises a subgroup of other registered individuals. Since individuals opt in for our study’s data collection, the control group sample size is smaller than the complete set of registered individuals. The balance table (Table 2) details the means in the characteristics used for stratification between treatment and control groups. Since the treatment group is selected to correspond to the characteristics of the total population but the entire pool of those interested in participating does not, the control group differs by definition in some observable characteristics from the treatment group. Yet, the detailed stratification yields balanced groups for most covariates. Statistically significant imbalances emerge by age (for those 60 years and older 2), educational attainment (specifically, compulsory schooling, apprenticeship, and university degrees), employment type (dependent employment, retirement), household income (first income quartile), and views on wealth equality. The differences reflect the disproportionate self-selection of individuals with 30-44 years, higher educational attainment, dependent employment, with incomes in the upper half of the distribution, and more pessimistic views about the fairness of the perceived wealth distribution. Further, Table 3 compares the characteristics of participants with replacement group members, and Table 4 with control group members and replacement group members. There are no significant differences, though in part due to the small sample size of 14 individuals in the replacement group. In addition, Table 5 compares the characteristics of participants with those of the general population. The comparison confirms that the treatment group is highly representative of the general population.
We use three contrasts to adjust for imbalances between our treatment and control groups: First, we compare individuals in the treatment with the control group. The groups are balanced in most, but not all, characteristics. They should further be identical in unobserved characteristics conditional on the imbalanced observable characteristics. Second, we compare our treatment group with a random sample of the Austrian population that is freshly drawn in June 2024. Both groups should balanced in terms of observed characteristics since the treatment group is representative in observable characteristics. However, the treatment group may differ in unobserved characteristics as treatment is based on opting-in by registering for the Good Council citizens’ assembly. Third, we compare our control group with the random sample of the general population. By reweighting respondents in the control group to correspond to the general population in terms of observable characteristics, any differences in outcomes can be inferred to stem from unobservable differences due to self-selection. This serves as a test to quantify the magnitude of the potential selection bias from registering for the Good Council. It allows us to assess how clean or biased the comparison between the treatment group and the random sample of the Austrian population is.