Experimental Design
1. Sampling & Randomization
This study utilizes a randomized controlled trial (RCT) embedded within an online survey designed for Japanese residents.
As described, this study conducts an RCT with three arms (Control, Treatment 1, Treatment 2). To reduce respondent burden and prevent fatigue bias, the full set of 36 profiles is divided into three blocks (Patterns A, B, and C). Therefore, the experiment consists of 9 distinct groups in total (3 intervention arms × 3 question patterns).
The survey will target approximately 1,200 monitors recruited via a research company. To ensure demographic balance across all 9 groups, participants are recruited and assigned using a stratified sampling method. In each of the 9 groups, the sample composition will be equal across the following 12 segments:
・Age: 3 categories (10-29, 30-49, 50-69 years old).
・Gender: 2 categories (Male, Female).
・Region: 2 categories (Three major metropolitan areas [Tokyo, Osaka, Nagoya] vs. Other regions).
2. Scenario & Attributes
In each arm, a binary CBC analysis comprising four attributes and three levels is conducted. Respondents select the preferred option between two tourist destinations characterized by varying attributes.
As previously stated, participants in the intervention groups are exposed to a nudge message designed to encourage congestion avoidance along with each choice task.
Respondents in Treatment 1 views a message emphasizing self-interest or loss aversion, while respondents in Treatment 2 receives a message appealing to social norms (describing the behavior of the majority).
The content of the choice tasks is as follows:
Respondents are asked to imagine a scenario where they are visiting Kyoto during the autumn foliage season and consulting an online guide map to decide which temple to visit. Under this setting, they are shown two distinct options, "Temple A" and "Temple B," which feature differing attribute levels, and are asked to select the one they would prefer to visit.
The two tourist destinations displayed in the comparison consist of the following four attributes, each with three levels:
・Crowding Level: Level 1 / 2 / 3 (Visualized with images and specific descriptions of waiting times and spacing to standardize perception).
・Admission Fee: 500 / 1,000 / 2,000 JPY.
・Travel Time (from current location): 10 / 30 / 60 minutes.
・Review Score: 2 / 3 / 4 stars.
3. Supplementary Questions
To capture the heterogeneity of treatment effects and facilitate a multifaceted analysis, this study collects data on respondents' socio-demographic attributes (e.g., marital status), past visitation history to Kyoto, and hypothetical travel companions within the experimental scenario.
Furthermore, we will measure behavioral economic traits, including conformity, altruism, and loss aversion. Additionally, participants in the intervention groups are asked about their impressions of the presented nudge messages. These responses will be used to deepen the discussion regarding practical challenges associated with the social implementation of such interventions.
4. Hypothesis
H1: Nudge messages emphasizing self-interest increase the importance of "crowding level" in tourist destination selection compared to the control group.
H2: Nudge messages emphasizing social norms increase the importance of "crowding level" in tourist destination selection compared to the control group.
5. Analysis Plan
5.1. Logistic Regression Analysis
We employ a Conditional Logit Model to analyze the choice data. By estimating the coefficients for each attribute (e.g., crowding level, admission fee), we will identify the relative importance respondents assign to these factors when selecting a tourist destination.
Crucially, we examine the interaction terms between the crowding attribute and the intervention group dummies. This allows us to verify the extent to which the nudge messages alter respondents' sensitivity (i.e., importance weight) toward the crowding attribute. The hypothesis is tested by calculating the WTP for congestion avoidance and comparing the differences between the control and intervention groups.
5.2. Heterogeneity Analysis
We will also conduct subgroup analyses using interaction terms between the intervention and respondent characteristics (demographics and behavioral traits) to explore potential heterogeneous effects.
5.3. Post-stratification weights (IPW)
While the sample is recruited with equal quotas across 12 strata defined by age, gender, and region., we plan to estimate the model using Inverse Probability Weighting (IPW) based on national census data to assess the generalizability of the findings to the Japanese population.