Youth, Politics and Discrimination

Participants take part in a two-wave online survey experiment. In the first wave, they are randomly assigned to one of two types of informational treatments.
The first treatment concerns youth political participation. Respondents are randomly assigned to one of four conditions: a control group receiving no additional content, a control group watching a short neutral video about Italian regions, a treatment group watching a short video citing statistics indicating low youth political participation (approximately one in three young people discuss politics at least once per week, ISTAT 2025), or a treatment group watching a short video citing statistics indicating high youth political participation (approximately 80% of young people discuss politics with family or colleagues, Istituto Toniolo 2024). Both treatment videos include an injunctive statement highlighting the importance of youth voting.
The second treatment concerns gender-related misperceptions. Respondents are asked to estimate one of four quantities: the share of sexual violence accusations that are false, the share of physical and sexual violence that is reported to the police, the share of young peers who believe feminism has gone too far, or the share of young people who identify as LGBTQ+. Respondents in the treatment group then receive corrective information comparing their estimate to the true value.
A follow-up survey is administered after the March 2026 constitutional referendum on the separation of judicial careers to measure persistence of effects and actual voting behaviour.

2026-03-10

2026-04-22

Research Question 1 — Youth Political Participation
1. Political Engagement Index (PEI): a composite index measuring the likelihood of engaging in political and civic activities, including signing a petition, attending a lawful demonstration, running as a city councillor, volunteering as a poll worker or scrutineer, and convincing a friend to vote. We will also report the separate components. An additional item measuring the likelihood of avoiding political participation may be included as a robustness check in the index construction.
2. Civic engagement: likelihood of helping plan or organise a neighbourhood or community event.
3. Referendum voting propensity: self-reported likelihood of voting in the March 2026 constitutional referendum on the separation of judicial careers.
Research Question 2 — Gender-Related Misperceptions
Outcomes 4–8 apply to all respondents. Outcomes 9–10 are administered to non-overlapping random halves of the sample. Outcome 11 is collected in Wave 2 only.
4. Confidence that the police reached the right conclusion in the rape vignette (Arms 1–2 treatment/control only)
5. Concern about being wrongly accused of sexual harassment
6. Belief that men are not protected against false accusations of gender-based violence
7. Perception that women exaggerate harassment claims
8. Support for fines for gender pay gap violations
9. Support for checking a partner's phone (random half of sample only)
10. Views about the family's situation when the woman works (random half of sample only)
11. Tolerance of sexist remarks (Wave 2 only)
Both Research Questions
12. Posterior beliefs: updated estimates of youth political participation (RQ1), measured immediately after treatment exposure, serving both as manipulation checks and as outcomes in their own right.

Political Engagement Index (PEI)
The PEI is constructed as the simple mean of standardised (z-scored) responses to five survey items measuring the likelihood of engaging in the following activities: signing a petition, attending a lawful demonstration, running as a city councillor, volunteering as a poll worker or scrutineer, and convincing a friend to vote. Each item is measured on a five-point Likert scale ranging from very unlikely to very likely. Prior to constructing the index, each item is standardised by subtracting the control group mean and dividing by the control group standard deviation. The final index is then re-standardised relative to the control group. The index is constructed separately for Wave 1 and Wave 2. Internal consistency will be assessed using Cronbach's alpha and reported in the paper. If alpha falls below 0.6, individual items will be reported separately alongside the index. All five items are equally weighted. We do not drop items on the basis of low item-total correlation, as the composition of the index is pre-specified and any ex-post adjustment would constitute a deviation from this plan. As a robustness check, we may add an item measuring the likelihood of not participating in political activities, recoded so that higher values correspond to greater engagement.
RQ2 Primary Outcomes
Outcomes 4–11 capture post-treatment attitudes and beliefs related to gender-based violence, false accusations, and gender norms. They are grouped into two clusters aligned with the experimental arms: outcomes 4–7 are most proximate to the corrective information provided in Arms 1 and 2 (false accusations and reporting rates); outcomes 8–10 capture broader gender norm attitudes relevant across all arms. Outcome 11 (tolerance of sexist remarks) is measured in Wave 2 to assess persistence of attitudinal change. The rape vignette (outcome 4) uses the same treatment/control assignment as Arm 1.
Posterior Beliefs
For RQ1, the posterior belief is measured by asking how common or uncommon it is for young people to discuss political issues, stay informed, or share opinions online, on a five-point scale. This qualitative measure of perceived norm serves primarily as a manipulation check but is also reported as an outcome in its own right.
Measurement and Scaling
The majority of outcome variables are measured on Likert scales and will be analysed as continuous variables in the main specification. Binary and standardised versions will also be constructed and reported. Standardisation uses the control group mean and standard deviation.

Research Question 1 — Youth Political Participation
1. Prior beliefs about youth political participation (to identify determinants of misperception)
2. Willingness to sign the youth-quota petition
3. Demand for additional information on the youth-quota petition
4. Demand for the link to sign the youth-quota petition
5. Posterior belief about the share of young people who discuss politics weekly (quantitative point estimate)
6. Expected share of peers who will vote at the referendum
7. Self-reported voting behaviour in the March 2026 referendum (Wave 2 only)
8. Posterior belief about youth referendum turnout (Wave 2 only)
Research Question 2 — Gender-Related Misperceptions
9. Prior beliefs about the relevant misperception in each arm (to identify determinants)
10. Support for legal penalties for false claims by content creators
11. Perceived harassment prevalence among women
12. Demand for the report on the percentage of false accusations
13. Demand for the report on the frequency of unreported cases of sexual violence
Mechanisms — Research Question 1
Mechanism variables are grouped into three pre-specified indices, each constructed as the simple mean of standardised components and re-standardised relative to the control group:
Trust & Institutional Responsiveness Index 14. Trust in politicians (0–10 scale) 15. Perceived government attention to youth-related issues (0–10 scale)
Political Motivation Index 16. Political interest / apathy 17. Free-riding norm: agreement with the statement that citizens do not look after the common good 18. Shame of voting: discomfort associated with having voted when peers did not 19. Social prioritisation over voting: preference for social activities over voting during an election period
Political Efficacy Index 20. Internal political efficacy: self-assessed qualification to participate in politics 21. External political efficacy: perceived influence of ordinary citizens over government decisions

Prior Beliefs (RQ1 and RQ2)
Prior beliefs are elicited before treatment exposure using a dynamic slider-and-pie-chart interface. Respondents are shown a fully grey pie chart and a slider set to zero; as they move the slider, the chart updates in real time to reflect their input. No anchor value is provided prior to elicitation, to avoid anchoring bias. Respondents subsequently report their confidence in the estimate by indicating by how many percentage points they believe they may be mistaken. These prior belief measures serve two purposes: as secondary outcomes to identify the correlates and determinants of misperception, and as moderators in heterogeneous treatment effect analyses, where prior belief distance from the true value is the primary pre-specified moderator.
For RQ1, a misperception score is constructed as the difference between the respondent's prior belief and the ISTAT (2025) true value. A composite measure of prior beliefs about peer engagement and voting will also be constructed. For RQ2, prior beliefs are arm-specific: the estimated share of false accusations (Arm 1), the estimated share of violence cases reported to police (Arm 2), the estimated share of young males and females agreeing that feminism has gone too far (Arm 3), and the estimated share of young males identifying as gay or bisexual and young females as lesbian or bisexual (Arm 4).
Behavioural Demand Outcomes (RQ1)
Willingness to sign the youth-quota petition, demand for additional information, and demand for the signing link are elicited sequentially at the end of the survey. These outcomes capture low-cost behavioural intentions and revealed interest beyond self-reported engagement likelihood. Respondents who request additional information will receive an email from the survey company containing the relevant links; researchers will not personally contact respondents. These outcomes are analysed as binary variables.
Peer Norm Beliefs and Turnout Expectations (RQ1)
The posterior belief about the share of young people discussing politics weekly is elicited as a quantitative point estimate after treatment exposure and serves as both a manipulation check and a secondary outcome capturing descriptive norm updating. The expected share of peers voting at the referendum captures downstream social norm updating beyond the respondent's own engagement intentions, and will be compared to official turnout statistics where available.
Wave 2 Outcomes (RQ1)
Self-reported referendum voting behaviour is collected in Wave 2, administered shortly after the March 2026 constitutional referendum. This outcome is subject to social desirability bias, which will be acknowledged in interpretation. Because Italian law requires citizens to vote in their municipality of registration, respondents who have relocated — a substantial share among young adults — cannot vote without travelling back, which is expected to attenuate treatment effects on actual turnout for this subgroup. Wave 2 analyses for RQ1 are designated exploratory given that the recontacted subsample of approximately 1,000 respondents is underpowered to detect effects on actual voting reliably.
Demand for Reports (RQ2)
Respondents are offered the opportunity to receive the reports cited in the discrimination treatment arms. Demand is measured as a binary indicator of whether the respondent requests the relevant report. These outcomes capture revealed interest in gender-related information and serve as low-cost behavioural complements to the attitudinal primary outcomes.
Mechanism Indices (RQ1)
Three mechanism indices are constructed to capture the psychological and attitudinal pathways through which the information treatments may affect political engagement. All indices are computed as the simple mean of standardised components and re-standardised relative to the control group. All variables are recoded so that higher values correspond to greater political engagement, efficacy, or institutional trust.
The Trust and Institutional Responsiveness Index captures whether treatment shifts perceived legitimacy of political institutions and perceived government responsiveness to youth concerns. The Political Motivation Index captures changes in political interest, free-riding attitudes, and the social norms around voting, including shame associated with voting when peers abstain and the prioritisation of social activities over civic participation. The Political Efficacy Index captures both internal efficacy (the respondent's sense of personal competence to engage in politics) and external efficacy (the respondent's belief that participation can influence government decisions).
These indices are examined as secondary outcomes to distinguish between competing mechanisms — in particular, the descriptive norm channel (whereby observing high peer participation strengthens legitimacy and efficacy) and the backlash or civic duty channel (whereby observing low peer participation increases perceived pivotality and personal motivation). Predicted directions by treatment arm are reported in the pre-analysis plan and summarised in Table 3 of the associated paper.
Measurement and Scaling
All secondary and mechanism outcomes measured on Likert scales are analysed as continuous variables in the main specification, with binary and standardised versions reported alongside. Mechanism indices are standardised using the control group mean and standard deviation. Demand outcomes are binary. Wave 2 outcomes are analysed using the same OLS specification as Wave 1, with treatment assignment from Wave 1 as the key independent variable.

We conduct a two-wave online survey experiment on a nationally representative sample of approximately 3,000 Italian citizens aged 18–34, recruited in partnership with the market research company Bilendi Srl. The sample uses quota sampling based on gender, age group (18–24 vs. 25–34), and Nielsen macroregion (Northwest, Northeast, Centre, South). Wave 1 is fielded before the March 2026 constitutional referendum on the separation of judicial careers; Wave 2 is administered shortly after to approximately one-third of Wave 1 respondents.
The study addresses two independent research questions. RQ1 examines whether informational treatments about youth political participation affect civic and political engagement. RQ2 examines whether corrective information about misperceptions related to gender-based violence and sexism affects attitudes toward gender equality.
For RQ1, respondents are randomly assigned in a 1:1:2:2 ratio to one of four conditions: a pure control group (no video), an active control group (neutral video on Italian institutional facts), a Treatment Low group (video citing ISTAT data indicating that approximately 1 in 3 young people discuss politics at least once per week), or a Treatment High group (video citing Istituto Toniolo data indicating that more than 3 in 4 young people discuss politics with family, colleagues, or friends). Both treatment videos include an injunctive statement on the importance of youth voting.
For RQ2, respondents are exposed to corrective information about two or more of four misperceptions: the prevalence of false accusations of sexual violence (Arm 1), the share of gender-based violence cases reported to police (Arm 2), the share of peers holding anti-feminist attitudes (Arm 3), and the share of young people identifying as LGBTQ+ (Arm 4). Arms 1 and 2 use within-phase randomisation. Arms 3 and 4 exploit phase assignment as the source of identification, with Phase 1 respondents serving as the control group and Phase 2 respondents receiving the treatment.
RQ1 and RQ2 treatment assignments are crossed and independent: every respondent receives one RQ1 arm assignment and two separate RQ2 arm assignments.

Not available

Respondents are randomly assigned to treatment and control conditions by the survey platform's computerised randomisation algorithm just before the prior beliefs. RQ1 arm assignment follows a 1:1:2:2 ratio (Control 1 : Control 2 : Treatment Low : Treatment High) and is implemented as simple randomisation stratified by the quota variables (gender, age group, and four Nielsen macro-region). RQ2 Arms 1 and 2 are randomised independently of RQ1 assignment and of each other, with equal probability allocation to treatment and control. RQ2 Arms 3 and 4 use phase assignment as the randomisation device: Phase 1 respondents are assigned to the control condition and Phase 2 respondents to the treatment condition, with the 1:1:2:2 RQ1 randomisation running identically in both phases to support the identifying assumption of comparability across phases. All randomisation is performed automatically by computer with no manual intervention.

The unit of randomisation is the individual respondent. There is no cluster randomisation. All treatment assignments (RQ1 arm, RQ2 Arms 1 and 2, and RQ2 Arms 3 and 4) are assigned at the individual level. The sole exception is RQ2 Arms 3 and 4, where the source of identification is phase assignment (Phase 1 vs. Phase 2), which is also determined at the individual level upon entry into the survey.

No

This is not applicable as the randomisation is at the individual level. In the first wave, there are approximately 3,000 individual respondents.

There are approximately 3,000 individual respondents in Wave 1, while approximately 1,000 individual respondents will be recontacted in Wave 2. This is the same as the value stated in the planned number of clusters because it is not clustered.

RQ1:
• Control 1 (pure control, no video): 500 respondents
• Control 2 (active control, neutral video): 500 respondents
• Treatment Low (ISTAT video, low participation statistic): 1,000 respondents
• Treatment High (Istituto Toniolo video, high participation statistic): 1,000 respondents

RQ2 Arms 1 and 2 (within-phase randomisation, crossed with RQ1):
• Control (no corrective information on false accusations or reporting rates): approximately 1,500 respondents
• Treatment (corrective information on false accusations and reporting rates): approximately 1,500 respondents

RQ2 Arms 3 and 4 (phase-based identification):
• Control (Phase 1, no peer-norm correction): approximately 1,000 respondents
• Treatment (Phase 2, peer-norm correction on feminist attitudes and LGBTQ+ identification): approximately 2,000 respondents

Wave 2 (exploratory):
• Approximately 1,000 respondents recontacted across all RQ1 arms, proportional to Wave 1 allocation

All power calculations use two-sided tests at α = 0.05 and 80% power, benchmarked against Haaland et al. (2023). Outcomes are z-scored with respect to the control group mean and standard deviation (SD = 1.0), so effect sizes are expressed in standard deviation units. Calculations use the Stata 17 command power twomeans 0, sd(1) n(#). RQ1 — Treatment High and Treatment Low vs. pooled control: Each treatment arm (n ≈ 1,000) is compared against the pooled control group (n ≈ 1,000, assuming Control 1 and Control 2 are equivalent following the placebo check). The minimum detectable effect (MDE) is 0.113 SD at 80% power. Power at the central benchmark effect size of 0.10 SD is approximately 0.65; power at the optimistic benchmark of 0.15 SD is approximately 0.85. Once stratum fixed effects (macro-region × gender × age group) and pre-registered baseline controls are included, residual variance is expected to decrease by approximately 25-35%, reducing the effective MDE to between 0.085 and 0.098 SD. Under this adjusted specification, power at the central benchmark of 0.10 SD rises to approximately 0.70-0.75, and power at the optimistic benchmark of 0.15 SD exceeds 0.90. For Treatment Low specifically, where the expected effect size is more uncertain and plausibly smaller, power at a central benchmark of 0.08 SD is approximately 0.50, reaching 0.80 at an optimistic benchmark of 0.12 SD. RQ2 Arms 1 and 2 (balanced design, n ≈ 1,500 per arm): The MDE is 0.071 SD at 80% power. Power at the central benchmark of 0.10 SD is approximately 0.85. With covariate adjustment, the effective MDE falls to approximately 0.060–0.065 SD. RQ2 Arms 3 and 4 (unbalanced design, control n ≈ 1,000, treatment n ≈ 2,000): The unequal 1:2 allocation is less efficient than a balanced design. The MDE is 0.085 SD at 80% power for z-scored outcomes (SD = 1.0) and 0.102 SD for ordinal outcomes (SD = 1.2). Power at the central benchmark of 0.10 SD is approximately 0.78. With covariate adjustment, the effective MDE falls to approximately 0.072-0.078 SD. Null result protocol: A non-significant average treatment effect will be interpreted as evidence that the true effect is smaller than 0.113 SD, ruling out the optimistic benchmark. All primary outcomes are supplemented with a two one-sided equivalence test with pre-specified bounds of ±0.10 SD, and 95% confidence intervals are reported throughout.