Misperceived Workplace Norms in Japan

We will use the average responses from Wave 1 to do a randomized information intervention in Wave 2. The details of the intervention are explained in Experimental Design section below.

We will use the average responses from Wave 1 to do a randomized information intervention in Wave 2. The details of the intervention are explained in Experimental Design section below.

2025-09-11

2025-09-29

In Wave 1, participants will be asked whether they agree that workplace norms in Japan are toxic. They will also be asked whether they agree that actions urgently need to be taken to improve workplace norms (see the Experimental Design section). In Wave 2, different participants guess the percentage of workers in Wave 1 who agreed with each statement. Their guessed agreement rates are the key outcome variables.

The primary test is whether these perceived agreement rates differ significantly from the actual agreement rates seen in Wave 1, i.e., thereby testing whether pluralistic ignorance exists.

Analysis method:
(a) Descriptive analysis and statistical tests (including both non-parametric and parametric tests, as appropriate).
(b) Regression analysis to test whether misperception differs by gender while also including other controls.

The primary outcomes are direct survey responses in Wave 2 (perceived agreement rates for each statement).

We will measure the following outcomes in Wave 2 (binary or four-point scales) and compare them between the treatment and control groups to investigate effects of the information intervention:
1. Actions and intentions related to addressing toxic workplace norms and job search.
2. Attitudes toward family/childcare initiatives (“Kurumin certification” and “Eruboshi certification”).
3. Knowledge/attitudes regarding Japan’s “Work Style Reform Related Laws.”
4. Support for actions to address toxic workplace norms.
5. Perceived social appropriateness of specific workplace behaviors (following Krupka & Weber 2013).

Analysis Method:
(a) Descriptive analysis and statistical tests: We will perform both non-parametric and parametric tests, as appropriate, to summarize and compare the key features of the data.
(b) Regression analysis: We regress each outcome variable on the treatment indicator and other covariates, including interaction terms between the treatment and selected variables in certain specifications.
(c) Robustness check: We will assess the robustness of our findings by including additional control variables.

The outcomes are measured as workers’ responses to the listed survey questions in Wave 2 (binary or four-point scales).

This study involves a two-wave survey targeting workers aged 18 to 69 who work as regular employees at large-sized firms in Japan. In Wave 1, workplace norms are elicited from them, and in Wave 2, another set of workers guesses the norms. The details will be summarized as follows:

A. Conducting Survey:
In Wave 1, we will survey 1,000 gender-representative regular employees at large-sized firms in Japan. Participants will be asked whether they agree that workplace norms in Japan are toxic. They will also be asked whether they agree that actions urgently need to be taken to improve workplace norms.
In Wave 2, we will recruit a different set of 4,000 regular employees at large-sized firms and ask them the same two questions. In addition, we ask them to guess the percentage of workers who agreed with each statement in Wave 1. These guesses are incentivized. Upon answering the questions, half of the employees (Treatment) will receive the actual results from Wave 1. The other half of the employees (Control) will not receive this information.
The remainder of the experiment in Wave 2 is identical to both the treatment and control groups: they will be asked about their beliefs, intentions, and actions related to workplace norms, including their willingness to take action to change the workplace practices.

B. Exclusion criteria for fraud answer/outliers:
Participants will be excluded from the analysis based on survey completion times, variability in responses to matrix-form questions, and implausible responses (e.g., reported age outside the eligible range).

C. Hypotheses:
H1: Pluralistic ignorance exists regarding (a) workers’ perceptions about toxic workplace norms and (b) demand for changes in Japan.
H1’: The degree of pluralistic ignorance differs by gender.
H2: Providing information about others’ actual opinions breaks the pluralistic ignorance, i.e., increases (i) their tendency to evaluate toxic workplace behaviors as inappropriate, and/or (ii) their willingness to take action to change such norms.
H2’: The effects of information intervention differ by gender.

This study involves a two-wave survey targeting workers aged 18 to 69 who work as regular employees at large-sized firms in Japan. In Wave 1, workplace norms are elicited from them, and in Wave 2, another set of workers guesses the norms. The details will be summarized as follows:

A. Conducting Survey:

Wave 1:
We will survey 1,000 regular employees at large-sized firms in Japan. The gender distribution of the recruiting sample will reflect that of regular employees at large-sized firms, based on the Employment Status Survey in Japan. Participants will be asked whether they agree or disagree with the following statements:
(i) “Workplace norms in Japan are toxic for employees”
(ii) “Some actions urgently need to be taken to improve workplace norms in Japan for the sake of improving worker welfare and well-being.”

Wave 2:
We will conduct a randomized controlled survey experiment with two conditions (treatment condition and control condition). A total of 4,000 regular employees at large-sized firms will be recruited, with the following breakdowns:
# Treatment group: 2,000 workers consisting of 1,000 female and 1,000 male workers.
# Control group: 2,000 workers consisting of 1,000 female 1,000 male workers.
To allow for an analysis of gender effects, we recruit workers so that each group will be gender balanced. No participants from Wave 1 will be included in Wave 2.
Participants in both conditions will be asked about whether they agree with the same two statements from Wave 1. This question is identical to those in Wave 1. Immediately after this, they will be asked, in a free-form format, what they think are the most important actions that should be taken. Also, Unlike in Wave 1, they will be asked to guess the percentage of workers agreeing with each statement in Wave 1. These guesses are incentivized based on whether their answers are closest to the actual percentage.
The key difference between the treatment and control conditions lies in the information intervention. After providing their opinions and guesses, participants in the treatment group will be informed about the actual results from Wave 1. Control group participants will not receive this information. The rest of the experiment is identical for the treatment and control groups: They will have the following sections in the survey:
1. Questions about their actions or intentions regarding toxic workplace norms and job search
2. Questions about government initiatives supporting family and childcare (i.e., “Kurumin certification” and “Eruboshi certification”)
3. Questions about Japan’s “Work Style Reform Related Laws”
4. Questions to guess workers’ perceptions towards workplace norms and demand for change by gender
5. Questions about perceived social appropriateness on specific workplace behavior (using the method by Krupka and Weber 2013 in JEEA).

# Related to Point 4, we will ask respondents whether they are willing to support the research group in submitting a report—based on the survey results—that summarizes problems in the Japanese workplace environment and encourages improvements, to a newspaper or magazine. This will be done on the condition that the group proceeds only if at least 50% of respondents agree. We will also ask whether they support submitting a petition to the Japanese government, under the same condition.

Note: In both Waves, we will also collect information on demographics (e.g., age, education level, marital status), work experience, tenure, general well-being, and levels of stress.

B. Exclusion Criteria for Fraud Answer/Outliers
To ensure data quality, we will exclude participants from analysis based on the following information:
- Too short or too long survey completion times.
- Too small or too long average LongString index (i.e., the maximum number of consecutive identical responses in matrix-style questions).
- Reported age outside the eligible range (18–69 years old).
- Reported overtime exceeding 168 hours per week (the total number of hours in a week).
In addition, we will perform robustness checks by analyzing the data both including and excluding respondents who fail attention check questions.

C. Hypotheses
H1: Pluralistic ignorance exists regarding (a) workers’ perceptions about toxic workplace norms and (b) demand for changes in Japan.
H1’: The degree of pluralistic ignorance differs by gender.
H2: Providing information about others’ actual opinions breaks the pluralistic ignorance, i.e., increases (i) their tendency to evaluate toxic workplace behaviors as inappropriate, and/or (ii) their willingness to take action to change such norms.
H2’: The effects of the information intervention differ by gender.

Three types of randomization are employed.
(1) Treatment vs. control assignment: Participants are randomly allocated by the survey company through separate survey URLs, with the assignment intended to achieve a 50/50 split.
(2) Question order randomization: To control for order effects, the order of certain survey questions is randomized using SurveyMonkey’s built-in randomization function, which assigns question order probabilistically with equal chance.
(3) Bonus allocation: As an incentive, a certain number of workers will be randomly given money (points). For questions involving the bonus eligibility, assignment to “bonus winner” vs. “non-winner” is conducted in two ways: (a) by distributing separate URLs to participants so that a pre-specified number of respondents are designated as winners, and (b) for specific subgroups (e.g., respondents providing certain answers), winners are selected using random number generation in R.

The unit of randomization in all cases is the individual respondent.

No

4,000 individuals (survey respondents)
This is the number of participants in Wave 2 of the survey.

4,000 individuals (survey respondents)

Control group: 2,000 individuals (1,000 male and 1,000 female)
Treatment group: 2,000 individuals (1,000 male and 1,000 female)

With a balanced assignment across the treatment and the control groups and across male and female workers (creating four cells of size N/4), the standard error of the treatment-gender interaction is 4σ/√N, exactly twice that of the main treatment effect (2σ/√N). This implies that detecting an interaction of a certain standardized magnitude requires approximately four times the total sample size needed to detect a main effect. Therefore, we plan to calculate the sample size based on the ability to detect a treatment-gender interaction effect. At a significance level of α = 0.05 (two-sided) and power of 0.80, detecting an interaction effect (treatment × gender) of 0.20σ requires approximately N≈3,148. To allow for modest imbalance, possible variance underestimation, and multiple testing, we target a rounded total of N=4,000. Minimum detectable effects under N=4,000, α=0.05, and power=0.80: One-sample t-test (pluralistic ignorance): MDE≈0.044σ Two-sample t-test (treatment vs. control): MDE≈0.089σ Treatment-gender interaction: MDE≈0.177σ