Roots of Courage: Persistence of Past Repression and Anti-War Dissent (Survey Experiment)

The experiment is embedded at the end of a survey after question Q56. Respondents are randomly assigned with equal probability (p=1/3) to one of three groups:
- Group 0 (Control): proceeds directly to outcome questions (Q60-Q64), while questions about Soviet repressions (Q57-Q58) are asked after the outcome questions.
- Group 1 (Priming): before proceeding to outcome questions (Q60-Q64), asks two questions about Soviet repressions: 1) family experience of soviet repressions (two items: politically repressed, forcibly resettled) (question Q57), 2) a prompt on the scale of Stalinist repression and a request to estimate the percentage of those rehabilitated (question Q58).
- Group 2 (Priming + Information): before proceeding to outcome questions (Q60-Q64), asks the same questions as Group 1 (Q57-Q58), and then provides factual historical information about the number (and the percent) of repressed individuals officially rehabilitated (information prompt is Q59).

2026-02-02

2026-02-23

We use several binary and scale primary outcome variables based on the answers to the following outcome questions Q60, Q61:
1) Q60: Please tell me, to what extent do you approve or disapprove of the (military operation | military actions) of Russia on the territory of Ukraine? (1-Fully approve, 2-Partially approve, 3-Partially disapprove, 4-Fully disapprove, 99-Difficult to answer)
2) Q61: In your opinion, currently, should Russia continue the (military operation | military actions) on the territory of Ukraine, or move to peace negotiations? (1-Definitely, they should continue until the stated goals are achieved, 2-Rather, they should continue until the stated goals are achieved, 3-Rather, they should stop and move to peace negotiations, 4-Definitely, they should stop and move to peace negotiations, 99-Difficult to answer)

For each question Q60 and Q61, we construct four binary and one scale outcome variables. The binary variables are:
- 1 if the variable equals 1, and 0 otherwise,
- 1 if the variable equals 1 or 2, and 0 otherwise,
- 1 if the variable equals 4, and 0 otherwise,
- 1 if the variable equals 3 or 4, and 0 otherwise,
The scale variable substitutes “99-Difficult to answer” with missing and takes the rest of the answer as is.

We use several scale secondary outcome variables based on the answers to the following outcome questions Q62 - Q64, including sub-questions of Q63:
1) Q62: In your opinion, how much can most people be trusted in general? (from 0 - One must be very cautious in dealing with people, to 10 – most people can be completely trusted, 99- Difficult to answer).
2) Q63: How much do you trust the following groups of people? (0 – Do not trust at all, 10 – Fully trust, 99-Difficult to answer)
– People you personally know
– People you meet for the first time
– People of another nationality
– People who have religious views different from yours
– People who have political views opposite to yours
– People whose attitude toward the (military operation | military actions) in Ukraine is different from yours
3) Q64: How safe do you think it is to participate in public opinion surveys on the topic of the military operation in Ukraine? (0 – Completely safe, 10 – Very dangerous, 99-Difficult to say)

For questions Q62-Q64 (including sub-questions of Q63), we construct scale outcome variables by substituting “99-Difficult to answer” with missing.

The experiment design is a three-arm Randomized Controlled Trial (RCT) embedded within an online survey. Its goal is to see if priming people about past state violence (Soviet repressions) affects their attitude toward the war in Ukraine. The experiment concludes the survey. After a series of general questions, every respondent is randomly assigned to one of three groups with an equal probability (1/3):
- Group 0 (Control): These participants proceed directly to the outcome questions on war attitudes (primary outcomes), trust in different groups, and the risk of survey participation (secondary outcomes). This group serves as a control group, as they are not primed by the reminder of the Soviet repressions. This group answers the set of Soviet repression questions after the outcome variables.
- Group 1 (Priming): These participants are asked questions designed to make Soviet repression "top of mind." They are asked about their family experience of Soviet repressions and are told that over 24 million people were affected by the Soviet repressive system. They are asked to estimate the percentage of people later declared innocent (rehabilitated). After that, the participants answer outcome questions like Group 0.
- Group 2 (Priming + Information): These participants receive the same memory prime as Group 1, but right after this prime, they are given a specific historical fact (Q59) that the state eventually admitted 4 million of those people (17%) were innocent and rehabilitated them. After this information, the participants answer outcome questions like Group 0.

Sampling and Randomization
Survey sample: The survey covers people aged 18+ from all internationally recognized Russian regions.
Data Collection: An online survey is conducted by Intra Research. The expected start date is February 2, 2026. The expected survey period is two weeks.
Randomization Unit: Individual respondent.
Sample Size: Approximately 2,500 respondents.

Estimation methods
For primary and secondary outcomes, we are planning to estimate
1) difference-in-means for Group 1 vs. Group 0 and Group 2 vs. Group 0. Then we compare the sizes of the treatment effects.
2) OLS regression, comparing outcome variables for Group 1 vs. Group 0 and Group 2 vs. Group 0, controlling for the respondents' demographics and geographical characteristics. Then we compare the sizes of the treatment effects.
3) Heterogeneity of the treatment effects (Group 1 vs. Group 0 and Group 2 vs. Group 0) estimated via OLS regressions by the following moderators: gender, age, household welfare, risk attitudes, patience, marriage status and kids, occupation, education, nationality, religiosity, media consumption, VPN usage, voting patterns in parliament and presidential elections, family experience of soviet honors and WW2 participation, family experience of soviet repressions, distance of answer in Q58 from the factual 17%, geographical distance to the closest Gulag camp, anti-war detentions in a region\settlement of residence.

Estimation samples
For primary and secondary outcomes, we plan to use the full sample and a subsample excluding Moscow and Saint Petersburg.

Errors clustering
We use both robust non-clustered errors and errors clustered at the settlement level.

Hypotheses
H1 (Priming Effect): The priming of past state violence (Group 1) will decrease support for the current war compared to the Control Group (Group 0).
H2 (Information Effect): Providing factual information about mass rehabilitations (Group 2) will further decrease war support by highlighting state fallibility and historical lawlessness.
H3 (Heterogeneity): Treatment effects will be more pronounced among younger cohorts (ages 18–44), consistent with patterns of intergenerational memory transmission.
H4 (Geographical Amplification): We hypothesize that both treatment effects will be significantly amplified for respondents living in close geographical proximity to former Gulag camps where the repressed population was imprisoned. This is expected because local physical traces (graveyards) and memorial sites act as "memory anchors," making the historical information provided in the survey more cognitively available and personally resonant.

Not available

The survey script implements randomization during the survey.

Randomization is implemented at the individual level.

No

Randomization is not clustered. In the estimation, errors may be clustered at the settlement level. The number of different settlements is not limited and is not controlled by the researchers. The survey company can hypothetically reach respondents in all settlements in the internationally recognized Russian regions.

2500 individual respondents

We expect to have 2500/3 observations in each arm

With a total sample of 2,500 respondents randomized into three equal groups (with 833 per arm), our study is powered to detect a Minimum Detectable Effect (MDE) of 0.15 standard deviations for the primary outcome at the 5% significance level with 80% power. For binary specifications of war support, this corresponds to a shift of approximately 5 percentage points from a baseline of 15%.