Political Narratives and Economic Behavior: A Survey Experiment on South Korea’s 2024 Martial Law

This study examines how competing media narratives about South Korea’s December 3, 2024 martial law declaration affect political attitudes, economic expectations, and economic behavior. Adult respondents in South Korea are randomly assigned to one of three groups: a no-information control group or one of two treatment groups that read a short passage based on Korean news reports and opinion columns. One passage presents the declaration as an unconstitutional attack on democratic institutions, while the other presents it as an exceptional response to severe political obstruction intended to restore government functioning.

2026-09-21

2026-10-12

The primary endpoints are: (1) economic behavior and intentions, including the likelihood of purchasing durable goods, expected changes in savings and financial assets, desired financial-portfolio shares, intentions to shift assets abroad, and intentions to seek information or prepare for overseas migration; (2) risk and intertemporal preferences; (3) personal economic expectations, including expected labor-income growth and subjective job-loss probability; (4) preferences regarding income redistribution, social safety nets, fiscal stabilization, and government intervention in the economy; (5) macroeconomic expectations regarding changes in South Korea’s consumer prices, unemployment rate, and real GDP over the next 12 months; (6) trust in political and social institutions and evaluations of government performance; and (7) respondents’ evaluation of the December 3, 2024 martial law declaration.

Economic behavior and intention outcomes will be analyzed using their original response scales. These outcomes include the likelihood of purchasing durable goods over the next 12 months; the expected direction of change in savings or financial assets; desired percentage shares allocated to deposits, stocks, funds or ETFs, bonds, cryptocurrency, and other financial assets; the likelihood of shifting assets abroad; and the likelihood of seeking information or preparing for overseas migration.

Risk preferences will be measured using general self-assessments and incentiveless lottery-choice items. Intertemporal preferences will be measured using self-assessed patience, willingness to give up a present benefit for a future benefit, and choices between smaller immediate and larger delayed payments. Components within each preference domain will be oriented in the same direction and combined into an equally weighted standardized index. Component-level estimates will also be reported.

Personal economic expectations will include expected labor-income growth over the next 12 months and the respondent's subjective probability of losing their job, measured from 0 to 100 percent. Macroeconomic expectations will measure expected changes over the next 12 months in South Korea's consumer prices, unemployment rate, and real GDP. Each expectation will be elicited using a slider ranging from -20 to 20, where negative values indicate an expected decrease, zero indicates no expected change, and positive values indicate an expected increase.

The policy-support index will combine four five-point items measuring support for income redistribution, stronger social safety nets, countercyclical fiscal spending, and active government intervention during economic crises. Items will be coded so that higher values indicate greater support.

Institutional trust will be measured separately for each listed institution. An equally weighted standardized index of trust in governmental institutions will combine trust in the government, National Assembly, courts, Constitutional Court, National Election Commission, and police. Evaluations of the former and current governments will be analyzed separately.

Evaluation of the December 3, 2024 martial law declaration will be measured on the same 0-to-10 scale before treatment and near the end of the survey, after the main substantive outcome modules. Higher values indicate a more positive evaluation. The primary analysis will use the post-treatment evaluation as the outcome and control for the corresponding pretreatment evaluation. The within-respondent change in evaluation will also be reported.

For each multi-item index, component outcomes will be oriented so that higher values have a common interpretation, standardized using the control-group mean and standard deviation, and averaged with equal weights. Individual component outcomes will be reported alongside the indices.

The secondary endpoints are: (1) subjective well-being, including happiness, meaning in life, and overall quality of life; (2) economic stress, political and social stress, uncertainty about the future, expected personal stability, relationship-related stress, perceived control over stress, and peace of mind; (3) interpersonal trust; (4) the perceived importance of institutional independence and checks and balances; (5) ideal daily time allocation across work, job search, household production, caregiving, education, leisure, civic participation, religious activity, and social interaction; (6) expectations about South Korea's future child-rearing environment and fertility intentions; (7) stress attributed to reported problems during the recent local-election voting process; (8) preferred sources for obtaining additional political, social, and economic information; and (9) current presidential-election voting intention; and (10) attitudes toward a potential special pardon of former President Yoon Suk Yeol.

The subjective well-being index will combine measures of current happiness, meaning in life, and overall quality of life. The stress and uncertainty index will combine economic stress, stress arising from the political and social situation, uncertainty about the future, expected personal instability, relationship-related stress, perceived lack of control over stress, and lack of peace of mind. Items will be oriented so that higher values have a common interpretation before standardization and averaging. Individual components will also be reported.

Interpersonal trust will be measured using ratings of trust in neighbors and in people encountered in public. The perceived importance of institutional independence and checks and balances will be analyzed as a separate outcome.

For time allocation, respondents report actual hours spent during the previous 24 hours on 14 activities before treatment and their ideal daily allocation across the same activities after treatment. Each activity will be analyzed separately. The post-treatment ideal allocation will be examined both in levels, controlling for the corresponding pretreatment actual allocation, and as the difference between ideal and actual hours.

Expectations about the future child-rearing environment and intentions to have a child or an additional child will be analyzed separately using their original response categories. Non-substantive responses such as “not applicable” and “prefer not to answer” will be treated as missing in scalar analyses. Stress attributed to reported problems during the recent local-election voting process will be analyzed on its original five-point scale.

Information-source outcomes will be constructed from respondents' ranked selections of up to three sources. For each source, we will construct indicators for being selected and for being ranked first, and compare post-treatment preferred sources with pretreatment sources normally used.

Presidential-election voting intention will be analyzed as a categorical post-treatment outcome. We will also construct indicators for movement toward or away from the major parties by comparing current voting intention with the respondent's reported vote in the 2025 presidential election. Attitudes toward a potential special pardon of former President Yoon Suk Yeol will be measured on a five-point scale ranging from strong opposition to strong support and analyzed as a separate outcome.

This study is a single-wave online survey experiment with pretreatment and post-treatment measures. Approximately 5,000 adult respondents will be recruited from a nationwide online panel in South Korea. Before treatment, respondents will report demographic and socioeconomic characteristics, political attitudes and behavior, media use, economic circumstances, actual time use during the previous 24 hours, and their evaluation of the December 3, 2024 martial law declaration.

Respondents will then be randomly assigned at the individual level to one of three groups: a no-information control group, a crisis/accountability framing group, or a legitimizing/order-restoration framing group. Following treatment assignment, all respondents will complete the same modules measuring economic behavior and intentions, economic preferences and expectations, policy preferences, subjective well-being, institutional trust, political attitudes, and ideal time allocation. The evaluation of the martial law declaration will be measured again near the end of the survey, after the main substantive outcome modules.

Not available

Randomization will be implemented electronically by the survey software after respondents complete all pretreatment questions. Eligible respondents will be assigned with equal probability to one of the three experimental arms: the no-information control group, the crisis/accountability framing group, or the legitimizing/order-restoration framing group. Neither respondents nor researchers will select the assigned condition.

The unit of randomization is the individual survey respondent.

No

5,000 individual respondents (the design is not clustered)

5,000 adult respondents

Approximately 1,667 respondents in each arm: 1,667 in the no-information control group, 1,667 in the crisis/accountability framing group, and 1,666 in the legitimizing/order-restoration framing group, for a total planned sample of 5,000 respondents.

With a total sample of 5,000 respondents allocated approximately equally across the three experimental arms, a two-sided test with a 5 percent significance level and 80 percent power can detect a pairwise difference of approximately 0.10 standard deviations between any two arms. For a binary outcome with a control-group probability of 50 percent, this corresponds to a difference of approximately 4.9 percentage points. These calculations are conservative because they do not incorporate potential precision gains from pretreatment covariates.