In July 2025, we hired SWG—one of Italy’s leading survey firms— to field a nationally representative conjoint experiment with approximately 2,000 respondents. In a moment of profound debate over Italy’s economy and the resilience of its democratic institutions, we seek to quantify how much Italians value free elections.
Intervention
We run a Conjoint Framework. Each respondent compared seven pairs of hypothetical “Italy-like” policy bundles, randomized on: • Individual net monthly income: 0.8×, 0.9×, 1.0× (baseline), 1.1×, 1.25× the national median. • Average society income: 0.8×, 1.0× (baseline), 1.5× median. • Free elections: Yes (1) / No (0). • Public services quality: High (48 h wait) / Moderate (5 days). For each profile p in pair j, respondent i chose their preferred society (Y(c)ijp = 1) and rated its desirability Y(r)ijp ∈ [0, 10]. No data were examined prior to this registration.
Experimental design
In July 2025, we hired SWG—one of Italy’s leading survey firms— to field a nationally representative conjoint experiment with approximately 2,000 respondents. Each respondent was exposed to five pairs of societies. For each pair, the respondent was asked to (i) assess the most preferred society and (ii) to evaluate the desirability of each society in the pair on a scale from 0 (least desirable) to 10 (most desirable). A society was characterized by four items: (i) respondent's individual income; (ii) average income in the society; (iii) free elections or not; (iv) level of public services. We used 5 levels of respondent's individual income, 3 levels of average income in the society; 2 levels of democracy [free elections or not] and 2 levels of public services. This gave rise to 60 possible combinations and thus to 60 different societies.
Design details
Randomization method
Each respondent was exposed to five pairs of societies. These ten different societies were randomly drawn by a computer among the 60 possible society types. We restricted the randomization not to have replacement, so that a particular society could not be included in the five pairs more than one, and each respondent was exposed to 10 different societies.
Randomization unit
The randomization was done at individual-pair level. For each individual, five pairs of societies were randomly selected.
Clusters
The analysis plan specifies clustering standard errors at the respondent level. Since each of the ~2,000 survey participants serves as a cluster for inference, the planned number of clusters is 2,000 (one cluster per respondent)
Observations
Choice‐task observations (one binary choice per pair):
2,000 respondents×5 pairs = 10,000 choice observations
Profile‐level evaluations (when you count each profile rating or desirability score):
2,000×5×2 = 20,000 profile‐evaluations
Arms
Given 2,000 respondents × 5 choice-pairs each (so 10,000 binary choices, or 20,000 total profile-evaluations) under a fully balanced conjoint, we have:
| Attribute | Levels | # Profile-evaluations per level |
| -------------------------- | ---------------------- | ------------------------------: |
| **Individual income** | 0.8× median | 20 000 / 5 = 4 000 |
| | 0.9× median | 4 000 |
| | 1.0× median (baseline) | 4 000 |
| | 1.1× median | 4 000 |
| | 1.25× median | 4 000 |
| **Society average income** | 0.8× median | 20 000 / 3 ≈ 6 667 |
| | 1.0× median (baseline) | 6 667 |
| | 1.5× median | 6 667 |
| **Free elections** | Yes | 20 000 / 2 = 10 000 |
| | No | 10 000 |
| **Public services** | High (48 h wait) | 10 000 |
| | Moderate (5 days) | 10 000 |
Since we have 60 societies, under a fully balanced conjoint, each society receives 20,000/6= 3333 evaluations
Primary outcomes
For each pair of societies presented to the respondents, the respondents (i) choose their most preferred and (ii) provide a value -- from 0 to 10 -- of the desirability of living in each society. These two variables -- a dummy for the most preferred society in a pair and the value of desirability for each society -- represents our main outcomes.
We use these two variables to obtain the Average Marginal Component Effects (AMCE) and the Willingness to Pay (WTP).
Average Marginal Component Effects (AMCE). We begin with the standard AMCE specification:
Y(m)ijp = αij + βdem Dijp + SUM (k) βk Aijkp + γ Posijp + εijp, m ∈ {c, r},
where αij are respondent–pair fixed effects, Dijp indicates free elections, Aijkp are attribute dummies, Posijp controls position, and SEs are clustered by respondent. We will also estimate the same equation including individual pre-treatment covariates (age, gender, education, region, political leaning). We use two Income Specification: 1. Continuous income: add η Incomei; and 2. Categorical income: replace that term with four dummies 1{Incomei = L} for L ∈ {0.8, 0.9, 1.1, 1.25}× median
Willingness to Pay (WTP). All WTP calculations use the democracy coefficient βdem from the AMCE regressions above:
• Continuous income: WTP = βdem/βincome
• Categorical bins: WTPL = βdem/ βL , for L ∈ {0.8, 0.9, 1.1, 1.25}.
Cited by
No citation data available for this registration yet.