## wpiea2019261-print-pdf

## Source details

**Canonical URL:** [wpiea2019261-print-pdf](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019261-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019261-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019261-print-pdf.pdf.json)

---

### I. Introduction — context and motivation
- Choice experiment methodology used to estimate willingness to pay (WTP) of workers to enjoy better work-life balance (avoid extreme overtime, compulsory relocation and transfers, and job insecurity).
- Rationale for choice experiments over hedonic wage models:
  - Choice experiments address identification issues when the experimenter chooses options and permit analysis of non-marginal changes and people outside the labor market (paragraph 1–3).
  - Japan lacks publicly available, large-scale reliable data on pay and job characteristics (paragraph 3).
- Japan-specific context:
  - Long-hours culture; regular workers often required to work overtime and to follow employers’ orders to relocate or transfer regardless of preferences (paragraph 4–5).
  - Since 1997 the number of dual earner households has exceeded single earner households (paragraph 4).
  - Japan ranks 35th/40 in OECD Better Life Index work-life balance indicators (paragraph 5).
- Policy relevance:
  - Dissemination of limited-regular (gentei seishain) contracts suggested as a main component of work-style reform (paragraph 6).
  - Colacelli and Fernandez C. (2018) estimate that gradual introduction of limited-regular contracts would boost labor productivity in Japan by over 7 percent in the long-run (paragraph 6).
- Behavioral angle:
  - Study investigates the extent to which guilt drives differences in mean WTP between men and women (paragraph 7).

### II. Design, attributes, and experimental implementation
- Focus groups:
  - Three focus groups in summer of 2017 with working participants in their 20s-40s; key qualitative findings: wage level important, tolerance to wage cut varied, strong interest in work day/time policies, many prefer to avoid excess overtime, some raised work location concerns.
- Selected attributes for choice experiment:
  - wage level; required overtime; job security; mandatory relocation; mandatory transfer.
  - Clarifications: mandatory relocation = company-mandated move to another branch in country; mandatory transfer = move to different department same/nearby location.
- Attribute levels (preserved exactly):
  - Wage1 (Baseline) 3 million yen
  - Wage 2 3.5 million yen
  - Wage 3 4 million yen
  - Wage 4 6 million yen
  - Wage 5 7 million yen
  - Wage 6 8 million yen
  - Overtime 1 (Baseline) 0 hours per month
  - Overtime 2 0-15 hours per month
  - Overtime 3 15-45 hours per month
  - Overtime 4 45- hours per month
  - Relocation 1 (Baseline) Zero possibility of relocation
  - Relocation 2 Some possibility of relocation
  - Transfer 1 (Baseline) Zero possibility of transfer
  - Transfer 2 Some possibility of transfer
  - Security 1 (Baseline) As secure as regular contracts
  - Security 2  Less secure than regular contracts
- Rationale and reference points:
  - 15 hours/month cut-off = median reported by Ministry of Health Labor and Welfare (2018 Karoshi white paper) for full-time regular workers.
  - 45 hours/month cut-off = legal limit for overtime at normal time (with allowed exceptions).
  - “Medium” security chosen to be between regular and non-regular worker security; tenure lengths used as proxies.
  - About half of current wage earners in sample earn between 3 million yen and 8 million yen a year.
- Experimental design and pilot:
  - Theoretical combinations = 192; fractional D-efficient design used (DCREATE in STATA).
  - Each subject offered eight choice sets of three alternative jobs; respondents randomly assigned into four blocks; total of 32 different choice sets.
  - For each choice set respondents choose the best and the worst job among three alternatives.
  - Pilot in November 2017 with 107 subjects; added question in main survey asking which attributes respondents ignored (Q29 in Appendix 1).

### III. Estimation framework and WTP computation
- Random utility model (Train 2003); logit and rank-ordered logit used.
- Deterministic utility (preserved exactly):
  - V = β0 + β1 wage + β2 wage2 + β3 Overtime2 + β4 Overtime3 + β5 Overtime4 + β6  Security2 + β7 Transfer2 + β8 Relocation2
  - Except wage variables, other variables are indicator variables.
- WTP with quadratic wage term:
  - Simplified example (preserved formulas):
    - V = β1 Wage + β2 Wage2 + β3 Overtime
    - WTP = d Wage / d Overtime = -β3 / [β1 + 2β2 Wage]
  - Discrete treatment leads to quadratic equation for WTP:
    - β1 Wage + β2 Wage2 + β3 Overtime = β1 (Wage − WTP) + β2 (Wage − WTP)2
  - Note: β not identified but scale cancels; β2 < 0 implies WTP < WTA.

### IV. Survey administration, sample, and descriptive statistics
- Survey period: late December 2017 through early January 2018; online administration.
- Eligibility: Japanese, aged between 20 and 59, with at least one job experience.
- Effective sample: 1,046 respondents.
- Key sample vs national comparators (Table 3; preserved exactly):
  - Fraction male: 0.49 vs 0.5
  - Age: 41.67*** vs 40.46
  - Married: 0.42*** vs 0.37
  - Education: 3.35*** vs 2.74
  - No children: 0.58*** vs 0.24
  - Annual income (Yen m): 4.26*** vs 4.32
  - Notes: *** = p<0.01; education coded 1 = high school; 2 = 2-year college; 3 = 4-year college and 5 = graduate school; annual income for the sample is interval data and highest open interval assumed equal to 10m yen for calculation.
- Additional descriptive points:
  - 48 percent are four-year university graduates vs national figure 26 percent (2012).
  - Geographic concentration: Kanto-region, Aichi and Osaka; responses from all 47 prefectures.
  - Sample has slightly higher share of regular employees than non-regular; national ratio in 2012 was 7:3.
  - Ratio of total working people to non-working people in sample approximately 8:2.

### V. Data quality checks and respondent behavior
- Drop-out counts by choice-set group: 35, 29, 36, and 46.
- Drop-out ratios (drop-out to initial respondents, in percentage): 10, 9, 10, and 13 respectively.
- Ignored-attribute question:
  - 45 percent reported they did not ignore any variables.
  - Among those who ignored at least one variable, "security" and "transfer" were more often ignored; "wage", "overtime" and "relocation" less frequently ignored.
- Willingness to accept 45+ hours overtime:
  - 46 percent said they need to get at least 3 million yen more annual wage than current job to accept 45 hours or more overtime per month.
  - 35 percent say they would not accept it with pay increase of 3 million yen or less.
- Respondents judged hypothetical job choices as realistic and applicable.

### VI. Main econometric results (Benchmark Model 1) — coefficients and interpretation
- Rank-ordered logit estimates (Model 1; robust SEs clustered at individual level):
  - Wage: 1.350*** (0.0670)
  - Wage squared: -0.0773*** (0.00510)
  - Overtime2 (less than 15 hours/month): -0.0665** (0.0323)
  - Overtime3: -0.514*** (0.0351)
  - Overtime4: -1.327*** (0.0500)
  - Security2: -0.324*** (0.0313)
  - Transfer2: -0.905*** (0.0325)
  - Relocation2: -0.307*** (0.0221)
- Sample size:
  - Observations: 25,104
  - Number of groups: 8,368
- Interpretation:
  - Negative wage squared indicates diminishing marginal utility of income and implies WTP increases with wage.
  - All non-wage coefficients have expected negative signs and are significant at 1 percent except Overtime2 (significant at 5 percent).

### VII. Willingness to pay (Model 1) — by annual wage (millions of yen)
- For wage = 3M, 3.5M, 4M, 6M, 7M, 8M (values in million yen; significance shown):
  - Overtime2: 0.075**, 0.082**, 0.090**, 0.153**, 0.232***, 0.447***
  - Overtime3: 0.553***, 0.600***, 0.656***, 1.023***, 1.371***, 1.943***
  - Overtime4: 1.340***, 1.441***, 1.556***, 2.228***, 2.755***, 3.470***
  - Security2: 0.355***, 0.386***, 0.424***, 0.681***, 0.948***, 1.439***
  - Transfer2: 0.944***, 1.019***, 1.107***, 1.645***, 2.101***, 2.763***
  - Relocation2: 0.336***, 0.366***, 0.402***, 0.648***, 0.905***, 1.386***
- Examples:
  - People with annual 3-million-yen wage are willing to pay 1.3 million yen to avoid Overtime4 (45 hours or more overtime per month).
  - People with annual 8-million-yen wage are willing to pay 3.470 million yen to avoid Overtime4.
  - WTP to avoid transfer ranges from 0.944 million yen to 2.763 million yen across wage levels shown.
- Methodological notes:
  - Delta method used for standard errors.
  - Tests suggest rejection of IIA in general but WTP estimates robust to restricted models; alternative derivative method yields similar patterns though can diverge for some cases (example: derivative estimate for wage = 8m yen can be 11.7M yen).

### VIII. Heterogeneity: Model 2 interactions (WTP by socio-demographics; wage fixed at 3 million yen, age 30s)
- Model 2 interacts non-wage attributes with female, no_children, female*no_children, age cohorts, and 4-year college dummy.
- Reported WTPs (in million yen) for people in their 30s with 3 million yen wages:

  - Model 2 (4-Year College-Educated)
    - Overtime2: Base 0.092 | Female 0.216*** | No_children 0.149 | Female*no_children 0.050***
    - Overtime3: Base 0.330*** | Female 0.756*** | No_children 0.558*** | Female*no_children 0.753***
    - Overtime4: Base 0.955*** | Female 1.676*** | No_children 1.319*** | Female*no_children 1.616***
    - Security2: Base 0.095 | Female 0.421*** | No_children 0.327*** | Female*no_children 0.372***
    - Transfer2: Base 0.821** | Female 1.337*** | No_children 0.930*** | Female*no_children 1.313***
    - Relocation2: Base 0.208*** | Female 0.540*** | No_children 0.621*** | Female*no_children 0.630***

  - Model 2 (Less Than 4-Year College)
    - Overtime2: Base 0.049 | Female 0.174** | No_children 0.106 | Female*no_children 0.007
    - Overtime3: Base 0.236*** | Female 0.668*** | No_children 0.467*** | Female*no_children 0.665***
    - Overtime4: Base 0.777*** | Female 1.515*** | No_children 1.151*** | Female*no_children 1.454***
    - Security2: Base 0.059 | Female 0.386*** | No_children 0.292*** | Female*no_children 0.337***
    - Transfer2: Base 0.616*** | Female 1.148*** | No_children 0.719*** | Female*no_children 1.123***
    - Relocation2: Base -0.070 | Female 0.277*** | No_children 0.361*** | Female*no_children 0.370***

- Key observations:
  - Including interactions renders some WTPs insignificant for certain groups (e.g., men with children not WTP for improved security or to avoid relocation).
  - WTPs for transfer positive across groups and education levels.
  - Women tend to have higher WTPs than men to avoid relocation, transfer and extreme overtime.

### IX. Tests of differences by gender and children (4-Year College or Higher; wage = 3 million yen, 30s)
- Gender differences (Male - Female). Positive indicates higher WTP for male:
  - With children | No children
    - Overtime2: -0.12 | 0.10
    - Overtime3: -0.43*** | -0.20**
    - Overtime4: -0.72*** | -0.30***
    - Security2: -0.33*** | -0.04
    - Transfer2: -0.52*** | -0.39***
    - Relocation2: -0.33*** | -0.04
- Presence-of-children differences (With children - Without children), by gender:
  - Male | Female
    - Overtime2: -0.06 | 0.17
    - Overtime3: -0.23** | 0.002
    - Overtime4: -0.36*** | 0.06
    - Security2: -0.41*** | -0.09
    - Transfer2: -0.09 | 0.02
    - Relocation2: -0.23*** | 0.05
- Interpretation:
  - Women have higher absolute WTPs than men for most attributes.
  - Men without children have higher absolute WTPs than men with children.
  - For women, differences by presence of children are often not statistically significant.

### X. Emotional (guilt) measures — descriptive and WTP interactions
- Seven guilt items rated 1-5 (1 = very guilty, 5 = not at all guilty); lower = higher guilt.
- Mean guilt responses (selected groups: Male, Female, Male with children, Female with children) — preserved exactly:
  - q1: Male 3.074 | Female 2.760 | Male with children 2.82 | Female with children 2.535
  - q2: Male 3.166 | Female 2.867 | Male with children 3.08 | Female with children 2.668
  - q3: Male 3.003 | Female 2.548 | Male with children 2.865 | Female with children 1.979
  - q4: Male 2.595 | Female 2.304 | Male with children 2.235 | Female with children 1.838
  - q5: Male 2.788 | Female 2.427 | Male with children 2.695 | Female with children 2.274
  - q6: Male 2.706 | Female 2.591 | Male with children 2.535 | Female with children 2.278
  - q7: Male 2.798 | Female 2.573 | Male with children 2.600 | Female with children 2.203
- Findings:
  - Women report feeling guiltier than men across all seven items; differences statistically significant at 1 percent for all questions.
- Extended model (guilt interactions) — main patterns:
  - Persons who feel guilty tend to have higher WTPs.
  - For individuals with "No Guilt" (all guilt questions = 5), WTP estimates rarely significantly different from zero.
  - Women’s higher guilt partly explains higher female WTPs.
- Table 9 (selected WTP values in million yen; Male No Guilt, Male All Guilt, Female No Guilt, Female All Guilt, Difference All Guilt):
  - Overtime2: -0.015 | 0.051 | 0.122 | 0.187 | 0.136
  - Overtime3: -0.146 | 0.152 | 0.304 | 0.580*** | 0.429***
  - Overtime4: 0.018 | 0.902*** | 0.795* | 1.586*** | 0.684***
  - Security: -0.065 | -0.191 | 0.299 | 0.180 | 0.371***
  - Transfer: -0.194 | 0.631*** | 0.354 | 1.114*** | 0.483***
  - Relocation: 0.111 | 0.096 | 0.447* | 0.432*** | 0.337***
- Table 10 (Women with children, wage = 3 million yen, 30s) — WTP to avoid Overtime4 by single guilt item (millions of yen):
  - All guilt questions = 5 (No Guilt): 0.795*
  - q1=1, other guilt =5: 0.707
  - q2=1, other guilt =5: 0.394
  - q3=1, other guilt =5: 0.600*
  - q4=1, other guilt =5: 1.272**
  - q5=1, other guilt =5: 1.270**
  - q6=1, other guilt =5: 0.485
  - q7=1, other guilt =5: 1.394***
- Interpretation:
  - Female WTP for Overtime4 is highest when guilty about canceling children’s events (q4), not caring elderly parents (q5), and not caring kids while at home (q7).
  - Little or no significant WTP when guilty only about leaving office early (q2), taking paid leave (q1), or not earning enough (q6).

### XI. APPENDIX 2 — Extended Model (Model 2) coefficients (baseline and interactions)
- Baseline coefficients:
  - Wage: 1.375*** (0.0678)
  - Wage squared: -0.0781*** (0.00516)
  - Overtime2: -0.0720 (0.149)
  - Overtime3: -0.578*** (0.176)
  - Overtime4: -1.550*** (0.239)
  - Security2: 0.233* (0.131)
  - Transfer2: -0.543*** (0.135)
  - Relocation2: -0.0337 (0.0989)
- Interactions with Female dummy (coefficients; robust SEs):
  - Overtime2: -0.116 (0.100)
  - Overtime3: -0.422*** (0.0982)
  - Overtime4: -0.801*** (0.138)
  - Security2: 0.321*** (0.0919)
  - Transfer2: -0.555*** (0.0997)
  - Relocation2: -0.308*** (0.0701)
- Interactions with No-Children:
  - Overtime2: -0.0527 (0.0948)
  - Overtime3: -0.222** (0.0934)
  - Overtime4: -0.395*** (0.129)
  - Security2: 0.401*** (0.0885)
  - Transfer2: -0.104 (0.0838)
  - Relocation2: -0.217*** (0.0596)
- Interactions with Female * No_Children:
  - Overtime2: 0.20637 (0.132)
  - Overtime3: 0.224 (0.138)
  - Overtime4: 0.465** (0.190)
  - Security2: -0.312*** (0.117)
  - Transfer2: 0.130 (0.132)
  - Relocation2: 0.265*** (0.0916)
- Age cohort interactions (selected coefficients and SEs preserved exactly):
  - 30s # Overtime2: -0.0118 (0.127)
  - 40s # Overtime2: 0.0700 (0.137)
  - 50s # Overtime2: 0.0346 (0.133)
  - 30s # Overtime3: 0.270* (0.161)
  - 40s # Overtime3: 0.390** (0.169)
  - 50s # Overtime3: 0.284* (0.166)
  - 30s # Overtime4: 0.613*** (0.221)
  - 40s # Overtime4: 0.704*** (0.231)
  - 50s # Overtime4: 0.606*** (0.229)
  - 30s # Security2: -0.0415 (0.108)
  - 40s # Security2: -0.0279 (0.115)
  - 50s # Security2: -0.147 (0.116)
  - 30s # Transfer2: -0.253** (0.117)
  - 40s # Transfer2: -0.183 (0.124)
  - 50s # Transfer2: -0.237* (0.125)
  - 30s # Relocation2: -0.053438 (0.0845)
  - 40s # Relocation2: -0.146 (0.0907)
  - 50s # Relocation2: -0.131 (0.0900)
- Interaction with Educated dummy:
  - Overtime2: 0.0394 (0.0667)
  - Overtime3: 0.0895 (0.0706)
  - Overtime4: 0.185* (0.0950)
  - Security2: -0.255*** (0.0581)
  - Transfer2: 0.208*** (0.0671)
  - Relocation2: 0.0336 (0.0458)
- Sample and significance:
  - Observations: 25,104
  - Number of groups: 8,368
  - Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1
  - Note: Educated = 1 if graduated from 4-year universities or graduate schools and Educated = 0 otherwise.

### XII. Policy implications and conclusions (summary)
- Choice experiments are useful when revealed preference data limited; this experiment assesses preferences over overtime, compulsory relocations, transfers, and job security.
- Benchmark findings:
  - People significantly dislike overtime (especially extreme overtime), job insecurity, intra-firm job transfer, and relocation.
  - WTP calculations show people willing to forfeit substantial wage portions to avoid extreme overtime and job transfers.
  - Job choice preferences vary by gender, presence of children, age, and education.
  - Conditional on annual salary, women generally have higher WTPs than men to avoid overtime, relocation, transfer, and to have more secure jobs.
  - Men without children value flexibility more than men with children; for women, responses vary little by parental status but women are more prone to guilt which raises WTP.
- Policy recommendation on contracts:
  - Limited-regular contracts can appeal to workers (especially women with children) but risk creating glass ceilings or increased labor market segmentation.
  - For successful work-style reform, policymakers should improve usability of limited-regular contracts for both men and women by:
    - Restricting discrimination against limited regular contracts.
    - Ensuring limited contracts offer the same quality job (e.g., enough training and promotion opportunities) as regular contracts.
    - Ensuring mobility across different contracts so workers can switch work styles depending on life stage and changing work-life balance needs.

*Content derived from wpiea2019261-print-pdf.*

### REFERENCES ___________________________________________________________28

### REFERENCES

### Tables (listed)
- 1. Attributes and Levels _____________________________________________________11
- 2. Choice Set Example ______________________________________________________12
- 3. Mean Comparisons of Selected Socio-Demographic Variables _____________________16
- 4. A Benchmark Econometric Model (Model 1) __________________________________18
- 5. WTPs by Different Amount of Annual Wage (Including Overtime Pay) _____________19
- 6. WTP Estimation for Wage 3 Million Yen in their 30s ____________________________21
- 7. Testing Differences in WTP for 4-Year College or Higher Respondents _____________22
- 8. Mean Guilt _____________________________________________________________24
- 9. Guilt, WTP, and Gender ___________________________________________________25
- 10. Women, Guilt, and WTP for Reducing Overtime ______________________________25

### Appendices (listed)
- 1. Key Parts of the Survey Questionnaire ________________________________________33
- 2. Extended Model (Model 2) _________________________________________________36

### I. Introduction — context and motivation
- Choice experiment methodology used to estimate willingness to pay (WTP) of workers to enjoy better work-life balance (avoid extreme overtime, compulsory relocation and transfers, and job insecurity).
- Rationale for choice experiments over hedonic wage models:
  - Choice experiments address identification issues when the experimenter chooses options and permit analysis of non-marginal changes and people outside the labor market (paragraph 1–3).
  - Japan lacks publicly available, large-scale reliable data on pay and job characteristics (paragraph 3).
- Japan-specific context:
  - Long-hours culture; regular workers often required to work overtime and to follow employers’ orders to relocate or transfer regardless of preferences (paragraph 4–5).
  - Since 1997 the number of dual earner households has exceeded single earner households (paragraph 4).
  - Japan ranks 35th/40 in OECD Better Life Index work-life balance indicators (paragraph 5).
- Policy relevance:
  - Dissemination of limited-regular (gentei seishain) contracts suggested as a main component of work-style reform (paragraph 6).
  - Colacelli and Fernandez C. (2018) estimate that gradual introduction of limited-regular contracts would boost labor productivity in Japan by over 7 percent in the long-run (paragraph 6).
- Behavioral angle:
  - Study investigates the extent to which guilt drives differences in mean WTP between men and women (paragraph 7).
- Paper organization:
  - Background on choice experiments and prior evidence on value of work; experiment design; results reported in sections 6 and 7; conclusions in section 8 (paragraph 8).

### II. Background — literature and prior evidence
- Choice experiment use in labor market research is limited; most studies focus on health/medical professionals and location choices in developing countries (paragraph 9).
- Examples of related studies:
  - Baum and Kabst (2013): choice experiment with graduates in Germany using ten highest ranked job choice factors; non-wage attributes had two levels (paragraph 9, footnote 3).
  - Mas and Pallais (2017) and Wiswall and Zafar (2018): found women’s WTP for flexible work arrangements higher than men’s; Mas and Pallais (2017) provides some external validity evidence (paragraph 9).
  - Yoo and Oh (2017): identified policy measures in South Korea including parental leave and workplace childcare center; found high WTP for workplace childcare (paragraph 9, footnote 5).
- Hedonic pricing model limitations:
  - Identification challenges when selection into market is part of equilibrium and when functional form may be unknown/non-linear (paragraph 10).
  - Lack of suitable official, large-scale reliable data for Japan (paragraph 10).
- Survey-based and hypothetical-wage studies in Japan:
  - Toda (2015): using 2012 Working Person Survey (~1,000 people aged 18-59 in Tokyo area) found 10 percent hourly wage discount to have a limit on work location and 10 percent hourly wage discount for having limit on work hours for female workers (paragraph 11).
  - Kuroda and Yamamoto (2013): found negative wage implications for having access to company’s work-life-balance policies, especially for male workers (paragraph 11).
  - Yasui et al. (2016): using RIETI 2015 online survey (~2000 people), found limited-regular workers have statistically lower “monthly” wages but about 80 percent and 90 percent of differences can be explained by observed characteristics; for “hourly wages” limited-regular workers receive statistically higher wages after controls (paragraph 12).
    - Controlled characteristics listed: gender, education, age, age squared, yeas of tenure, years of tenure squared, industry, occupation, marital status, number of children, prefecture of residence, and hours of work (footnote 6).
  - Morikawa (2010): respondents asked for minimum wage premium (ranging from +0 percent to +50 percent of a typical regular worker’s wage) to accept unstable employment and mandatory relocations; average asked-for premium ~10 percent (paragraph 13).
  - Kume et al. (2014): after excluding invalid responses, respondents on average asked for 21 percent premium for having unstable job and 19 percent for risk of mandatory relocations/intra-firm transfers; ~30 percent chose “I don’t know” (paragraph 14, footnote 7).
  - Tsuru et al. (2013): average asked-for premium 20 percent for switching to unstable job and 27 percent for switching to job with risk of mandatory relocations/intra-firm transfers (paragraph 14).
  - Kuroda and Yamamoto (2013, RIETI 2012): on average workers seem to accept a lower wage (about 20 percent lower) than companies would offer (about 10 percent lower) when valuing availability of work-life balance (WLB) policies (paragraph 14, footnote 8).
- Conclusion from prior studies:
  - Evidence varies across studies; wage cut extent for adopting worker-friendly working styles could vary by work location, company size, and whether job is white-collar (paragraph 12–14).
  - Existing studies often focus on one or two job features and have limited heterogeneity analysis; the present study aims to address these gaps (paragraph 15).

### III. Design and Method — experimental considerations
- Key design dimensions: number of choice sets, alternatives, attributes, and number and range of attribute levels (paragraph 16).
  - Definitions:
    - Choice set: group of alternative options respondents rank or choose from.
    - Alternatives: members of the choice set.
    - Attributes: features of the alternative (e.g., wage, overtime requirements).
    - Levels: particular values for attributes (e.g., a wage of 3 million yen) (paragraph 16).
- Trade-offs in experimental complexity:
  - Complex designs yield richer data but risk respondent cognitive overload, incomplete responses, and inconsistent choices (paragraph 16).
  - Two effects of complexity:
    - Respondent simplification of complex information can bias attribute weights/WTP.
    - Increased respondent error increases error variance (paragraph 16).
- Empirical findings on number of choice sets and attributes:
  - Chung, Boyer and Han (2011): variance of error first decreases then increases with number of choice sets; optimal number of choice sets per survey ~six (paragraph 17).
  - Caussade, et al. (2005): observe U-shaped relationship with error variance decreasing up to nine or ten choice situations (paragraph 17).
  - DeShazo and Fermo (2002) and Caussade, et al. (2005): increasing number of attributes results in increased error variance (paragraph 18).
  - Meyerhoff, Oehlmann and Weller (2015): probability of abandoning the survey significantly increases with number of attributes (paragraph 18).
  - As number of levels increases, experimental complexity increases due to more comparisons (paragraph 18).

*Source: wpiea2019261-print-pdf (REFERENCES ___________________________________________________________28).*

### 19. Given these trade-offs, we conducted a literature review and three focus group

### 19. Given these trade-offs, we conducted a literature review and three focus group

### Focus groups and attribute selection
- Three focus group interviews conducted during summer of 2017 with participants who:
  - were currently working, in their 20s-40s, and belonging to preformed groups (selection criteria per Rabiee 2004).
  - had similar socio-characteristics and would be comfortable talking to the interviewer and each other.
- Major qualitative findings from focus groups:
  - The majority of participants mentioned an appropriate level of wage as a factor they look for in a job.
  - Tolerance to wage cut varied among participants.
  - Work culture or policies related to work day and time received substantial interest.
  - Many agreed that excess overtime should be avoided:
    - Some prefer zero overtime.
    - Others accept up to certain level (example given: regularly working until 8 PM is “not a big problem”), but overtime above that level should be avoided.
  - No discussion about part-time or shortened work hours (likely because all participants were currently working full-time).
  - Some participants raised the issue of work location.
- Based on interviews, five attributes were selected for the choice experiment design:
  - wage level
  - required overtime
  - job security
  - mandatory relocation
  - mandatory transfer
- Clarifications:
  - Mandatory relocation: company-mandated relocation to another branch in the country (short or extended period), potentially highly disruptive to family life.
  - Mandatory transfer: moving the worker to a different department at the same or nearby location, potentially less disruptive but salient to some workers.

### Attribute levels and reference points
- Rationale:
  - Levels chosen to cover a reasonable range of job features for typical regular jobs in Japan.
  - The cut off overtime hour of 15 hours per month is the median reported by the Ministry of Health Labor and Welfare in the 2018 Karoshi white paper for full-time regular workers in Japan.
  - The 45 hours per month cut-off is the legal limit for overtime at normal time (with allowed exceptions during busy seasons under certain conditions).
  - Employment security “high” reflects stability enjoyed by current regular workers (traditional notion of lifetime employment).
  - “Medium” employment security is somewhat lower than regular workers’ job security but higher than non-regular worker’s job security (“low” not included in the choice set).
  - Lengths of average tenure for regular, limited-regular, and non-regular workers used as proxy illustrations for high, medium, and low security and provided to respondents as reference information.
  - “Baseline” tag indicates omitted variables in regression equations (not presented to participants).
- Table of Attributes and Levels (preserved exactly as in source):
  - Wage1 (Baseline) 3 million yen
  - Wage 2 3.5 million yen
  - Wage 3 4 million yen
  - Wage 4 6 million yen
  - Wage 5 7 million yen
  - Wage 6 8 million yen
  - Overtime 1 (Baseline) 0 hours per month
  - Overtime 2 0-15 hours per month
  - Overtime 3 15-45 hours per month
  - Overtime 4 45- hours per month
  - Relocation 1 (Baseline) Zero possibility of relocation
  - Relocation 2 Some possibility of relocation
  - Transfer 1 (Baseline) Zero possibility of transfer
  - Transfer 2 Some possibility of transfer
  - Security 1 (Baseline) As secure as regular contracts
  - Security 2  Less secure than regular contracts
- Note on wage thresholds:
  - Purpose: estimate coefficient on wages and wage squared.
  - About half of current wage earners in sample earn between 3 million yen and 8 million yen a year.

### Experimental design and pilot
- Combinatorics:
  - With five attributes and several levels, theoretically 192 possible combinations.
  - Practical constraint: typical experiments involve less than 10 choice sets; cognitive burden must be limited.
- Design approach:
  - Fractional (less than 192) experimental design used.
  - Chosen efficiency measure: D-efficiency (minimizes determinant of covariance matrix).
  - Efficient designs rely on prior beliefs about sign and strength of coefficients.
  - Conjectures used for priors:
    - effect on utility of wages positive and increasing;
    - extreme overtime negative and worse than less extreme overtime;
    - no clear prior for within-company transfers (may start with coefficient zero).
  - Pilot design created using DCREATE command in STATA (Hole 2015).
  - Pilot results used to update design for the main survey; resulting design similar so initial design carried into main experiment.
- Choice set presentation:
  - Each subject offered eight choice sets consisting of three alternative jobs.
  - Respondents randomly assigned into four blocks; each block faces different set of choice sets.
  - Total of 32 different choice sets in experiment; each individual sees eight.
  - For each choice set respondents make two decisions: the best job and the worst job among the three alternatives.
- Example choice set (preserved exact values in example):
  - Job A: Annual Wage 6 million yen; Overtime 0 hours/month; Employment Security Medium; Transfer Possibility Some; Relocation Possibility None.
  - Job B: Annual Wage 7 million yen; Overtime 15-45 hours/month; Employment Security High; Transfer Possibility None; Relocation Possibility None.
  - Job C: Annual Wage 8 million yen; Overtime 0 hours/month; Employment Security Medium; Transfer Possibility None; Relocation Possibility Some.
- Pilot experiment:
  - Conducted in November 2017 with 107 subjects.
  - Pilot differed from main only in number of respondents (effective respondent size about 1/10 of main) and the absence of one question added after the pilot.
  - Added question in main experiment: asks what attributes respondents ignored when making stated choice decisions (Q29 in Appendix 1).
  - Coefficients from preliminary regression analysis on pilot data adopted as priors for D-efficient design.
  - Resulting efficiency level and choice sets were not very different from initial design; thus initial design used in actual experiment.

### Estimation models and WTP computation
- Underlying framework: random utility model (Train 2003).
  - Utility specification: Uij = V(si, xij) + εij = Vij + εij.
  - εij assumed to follow some distribution; logit model used with εj independently and identically distributed extreme value random variates.
- Logit and rank-ordered logit probabilities:
  - Pr(Uj>Uk, j≠k) = eVj/(eVj+eVk).
  - Rank-ordered extension: Pr(U1>U2 ... >UH, for H≤J) = ∏[eVh/∑eVm] from h=1 to H.
  - For this experiment J=H=3.
- Deterministic utility specification (preserved exactly as in source):
  - V = β0 + β1 wage + β2 wage2 + β3 Overtime2 + β4 Overtime3 + β5 Overtime4 + β6  Security2 + β7 Transfer2 + β8 Relocation2
  - With the exception of the wage variables, the other variables are indicator variables.
- Interpretation:
  - Significance and magnitude of β coefficients indicate relative importance of attributes influencing job preference.
  - Marginal rate of substitution between wage and another attribute provides estimates of willingness to pay (WTP).
- WTP considerations with quadratic wage term:
  - Because model includes wage squared, WTP depends on the level of wage.
  - Simplified example provided (preserved formulas exactly as in source):
    - Simplified model: V = β1 Wage + β2 Wage2 + β3 Overtime
    - WTP = d Wage / d Overtime = -β3 / [β1 + 2β2 Wage]
  - Discrete treatment for discrete attribute levels leads to quadratic equation in WTP to be solved:
    - β1 Wage + β2 Wage2 + β3 Overtime = β1 (Wage − WTP) + β2 (Wage − WTP)2
  - Note: though β is not identified (scale parameter), WTP measures can be identified because scale cancels out.
  - Income effects: β2 < 0 implies WTP < WTA; the paper uses WTP terminology.
- Interaction with socio-demographic characteristics:
  - Socio-demographic characteristics (gender, age, marital status, education, having children, income) may explain preference heterogeneity.
  - To identify differences, socio-demographic variables are interacted with non-wage variables; direct inclusion would drop out of equation (5).
  - Marginal utility parameterization (preserved exactly):
    - βk = βko + βk′ q for k>2, where q is vector of socio-demographic variables.
  - Example: q = (female, no_children); then βk = βko + βk1 female + βk2 no_children for k>2.

### Survey administration, sample composition, and descriptive statistics
- Survey administration:
  - Conducted by an established online survey company in late December 2017 through early January 2018.
  - Invitation sent to targeted subjects: Japanese, aged between 20 and 59, with at least one job experience.
  - Online survey company collected responses separately from 20s-30s and 40s-50s groups to balance response numbers; extra measures taken to secure responses from younger cohorts.
- Sample size and representativeness:
  - Effective answers from 1,046 respondents in total.
  - Panel diverse by age, gender, etc., but not necessarily representative.
  - To match young and old group numbers, survey company collected relatively more people in the 35-39 age group compared to the 20-24 group.
  - Sample still covers current and potential labor force (focus of study).
- Key sample mean comparisons (preserved exactly as in source Table 3):
  - Panel vs Japan (national comparator):
    - Fraction male: 0.49 vs 0.5
    - Age: 41.67*** vs 40.46
    - Married: 0.42*** vs 0.37
    - Education: 3.35*** vs 2.74
    - No children: 0.58*** vs 0.24
    - Annual income (Yen m): 4.26*** vs 4.32
  - Notes from table: *** = p<0.01 for test of equality between sample mean and population; education coded 1 = high school; 2 = 2-year college; 3 = 4-year college and 5 = graduate school; annual income for the sample is interval data and highest open interval assumed equal to 10m yen for calculation.
- Additional descriptive points:
  - Sample is notably more educated: 48 percent are four-year university graduates vs national figure 26 percent (2012).
  - Geographic distribution: higher frequency in Kanto-region (Tokyo, Kanagawa, and Saitama), as well as Aichi and Osaka prefectures; responses from all 47 prefectures present.
  - Income comparison: national equivalent measurements unavailable; National Tax Agency survey used for comparison.
    - Share of high earners in panel slightly higher than national comparison.
    - Because of people reporting zero income, panel has a lower mean income than national mean reported in the table (possible artefact of upper-limit choice).
  - Employment type: panel has slightly higher share of regular employees than non-regular employees; national ratio in 2012 was 7:3.
  - Household roles: sample has more housewives/husbands and fewer students than national distribution.
  - Ratio of total working people to non-working people (including housewives/husbands and students) in sample approximately 8:2, similar to national distribution.

*Italic: Content derived from wpiea2019261-print-pdf (selected section).*

### 37. Before running the econometric models, we conducted various checks to see if the

### wpiea2019261-print-pdf - 37. Before running the econometric models, we conducted various checks to see if the

### Data quality checks and respondent behavior
- Drop-out counts by choice-set group: 35, 29, 36, and 46.
- Drop-out ratios (drop-out to initial respondents, in percentage): 10, 9, 10, and 13 respectively.
- Survey question on ignoring variables:
  - 45 percent of respondents reported they did not ignore any variables when making decisions and considered all variables in all choice sets.
  - Among respondents who ignored at least one variable, "security" and "transfer" were more often ignored; "wage", "overtime" and "relocation" were much less frequently ignored.
- Question on willingness to accept 45 hours or more overtime per month:
  - 46 percent respondents said they need to get at least 3 million yen more annual wage than current job to accept 45 hours or more overtime per month.
  - 35 percent say they would not accept it with pay increase of 3 million yen or less.
- Respondents on average judged the hypothetical job choice situations to be realistic and applicable to them.
- Conclusion: Checks support coherence and usefulness of the stated-choice data.

### Main econometric results (Benchmark Model 1) and WTPs
- Dependent variable: ranking of job in choice set (3 = best, 2 = second-best, 1 = worst).
- Independent variables: five attributes (wage, overtime levels, security, transfer, relocation) and socio-demographic characteristics (Section 4).
- Benchmark Model (Model 1) coefficient estimates (rank-ordered logit regression; robust standard errors clustered at individual level in parentheses):
  - Wage: 1.350*** (0.0670)
  - Wage squared: -0.0773*** (0.00510)
  - Overtime2 (less than 15 hours/month): -0.0665** (0.0323)
  - Overtime3: -0.514*** (0.0351)
  - Overtime4: -1.327*** (0.0500)
  - Security2: -0.324*** (0.0313)
  - Transfer2: -0.905*** (0.0325)
  - Relocation2: -0.307*** (0.0221)
- Sample size and groups:
  - Observations: 25,104
  - Number of groups: 8,368
- Significance notation: *** p<0.01, ** p<0.05, * p<0.1.
- Interpretation:
  - Negative coefficient on wage squared indicates diminishing marginal utility of income and implies WTP increases with wage.
  - All coefficients have expected signs and are significant at 1 percent except Overtime2 significant at 5 percent.

- Willingness to pay (WTP) by different annual wage levels (Model 1; values in million yen; significance shown):
  - For wage = 3M yen, 3.5M yen, 4M yen, 6M yen, 7M yen, 8M yen:
    - Overtime2: 0.075**, 0.082**, 0.090**, 0.153**, 0.232***, 0.447***
    - Overtime3: 0.553***, 0.600***, 0.656***, 1.023***, 1.371***, 1.943***
    - Overtime4: 1.340***, 1.441***, 1.556***, 2.228***, 2.755***, 3.470***
    - Security2: 0.355***, 0.386***, 0.424***, 0.681***, 0.948***, 1.439***
    - Transfer2: 0.944***, 1.019***, 1.107***, 1.645***, 2.101***, 2.763***
    - Relocation2: 0.336***, 0.366***, 0.402***, 0.648***, 0.905***, 1.386***
  - Example interpretations:
    - People with annual 3-million-yen wage are willing to pay 1.3 million yen to avoid Overtime4 (45 hours or more overtime per month).
    - People with annual 8-million-yen wage are willing to pay 3.470 million yen to avoid Overtime4.
    - WTP to avoid transfer ranges from 0.944 million yen to 2.763 million yen across wage levels shown.
- Notes on methods and robustness:
  - Because Model 1 includes wage squared, WTPs depend on wage level.
  - Delta method used to compute standard errors.
  - Tests suggest rejection of the IIA assumption in general, but WTP estimates are robust: restricted models (dropping worst option) do not produce significantly different WTP figures compared to unrestricted models.
  - Alternative derivative method yields similar patterns; largest divergence for Overtime4 in some cases (example: derivative estimate for wage = 8m yen can be 11.7M yen).

### WTPs by socio-demographic characteristics (Model 2 interactions)
- Model 2 includes interactions: non-wage attributes interacted with:
  - female dummy (female = 1 if respondent is female)
  - no_children dummy (no_children = 1 if respondent has no children)
  - female * no_children
  - age_cohort dummies (30s, 40s, 50s)
  - 4-year college degree dummy
- Baseline in model: male in their 20s with children and no degree.
- For simplicity, reported WTPs fix wage at 3 million yen and age cohort at 30s. All cases refer to people in their 30s with 3 million yen wages.
- Table 6: WTP estimation for wage 3 million yen in their 30s (in million yen)

  - Model 2 (4-Year College-Educated)
    - WTP (Base | Female | No_children | Female*no_children)
      - Overtime2: 0.092 | 0.216*** | 0.149 | 0.050***
      - Overtime3: 0.330*** | 0.756*** | 0.558*** | 0.753***
      - Overtime4: 0.955*** | 1.676*** | 1.319*** | 1.616***
      - Security2: 0.095 | 0.421*** | 0.327*** | 0.372***
      - Transfer2: 0.821** | 1.337*** | 0.930*** | 1.313***
      - Relocation2: 0.208*** | 0.540*** | 0.621*** | 0.630***
  - Model 2 (Less Than 4-Year College)
    - WTP (Base | Female | No_children | Female*no_children)
      - Overtime2: 0.049 | 0.174** | 0.106 | 0.007
      - Overtime3: 0.236*** | 0.668*** | 0.467*** | 0.665***
      - Overtime4: 0.777*** | 1.515*** | 1.151*** | 1.454***
      - Security2: 0.059 | 0.386*** | 0.292*** | 0.337***
      - Transfer2: 0.616*** | 1.148*** | 0.719*** | 1.123***
      - Relocation2: -0.070 | 0.277*** | 0.361*** | 0.370***
    - Significance notation: ***p<0.01 , **p<0.05, *p<0.10
- Key observations from Model 2:
  - Including interactions renders some WTP values statistically insignificant for certain groups (e.g., men with children are not WTP for improved security or to avoid relocation).
  - WTPs for transfer are positive across all specified individual characteristics and education levels, indicating respondents generally dislike transfer.
  - Gender and presence of children influence WTP levels; women tend to have higher WTP than men to avoid relocation, transfer and extreme overtime.

### Tests of differences in WTP by gender and presence of children (Table 7; 4-Year College or Higher)
- Gender differences in WTP (Male - Female). Positive indicates higher WTP for male; negative indicates higher WTP for female.
  - With children | No children
    - Overtime2: -0.12 | 0.10
    - Overtime3: -0.43*** | -0.20**
    - Overtime4: -0.72*** | -0.30***
    - Security2: -0.33*** | -0.04
    - Transfer2: -0.52*** | -0.39***
    - Relocation2: -0.33*** | -0.04
- Differences in WTP by presence of children (With children - Without children), reported separately for Male and Female:
  - Male | Female
    - Overtime2: -0.06 | 0.17
    - Overtime3: -0.23** | 0.002
    - Overtime4: -0.36*** | 0.06
    - Security2: -0.41*** | -0.09
    - Transfer2: -0.09 | 0.02
    - Relocation2: -0.23*** | 0.05
  - Significance notation: ***p<0.01, **p<0.05, *p<0.1
- Interpretation:
  - Women tend to have higher absolute WTPs than men for most attributes, for both with-children and without-children groups.
  - Men without children have higher absolute WTPs than men with children.
  - For women, differences by presence of children are not statistically significant in many cases.
  - Possible explanation: having children raises both marginal utility of income and marginal utility of leisure; effects may offset for women but income effect may dominate for men.

### Additional checks on children variables and robustness
- Alternative specifications tested:
  - Interaction dummies for number of children and age of youngest child, interacted with non-wage attributes and gender.
  - Results omitted here but summary: no clear additional patterns across different number of children nor across different age of the youngest children for both genders; suggests Table 7 results are not driven by the binary specification.

### Emotional (guilt) effects and gender differences
- Survey included seven guilt-related items rated on a 1-5 scale (1 = very guilty, 5 = not at all guilty). Lower number means higher guilty feelings.
- Mean guilt responses (selected groups: Male, Female, Male with children, Female with children):
  - q1: "I took paid leave when my managers and colleagues are working a lot of overtime."
    - Male: 3.074 | Female: 2.760 | Male with children: 2.82 | Female with children: 2.535
  - q2: "I left the office on time for a family event when my managers and colleagues are working."
    - Male: 3.166 | Female: 2.867 | Male with children: 3.08 | Female with children: 2.668
  - q3: "I did not prepare healthy dinner for me and my family for the entire week"
    - Male: 3.003 | Female: 2.548 | Male with children: 2.865 | Female with children: 1.979
  - q4: "Because I was working I missed my child(ren)’s event which I had promised to go"
    - Male: 2.595 | Female: 2.304 | Male with children: 2.235 | Female with children: 1.838
  - q5: "I did not see my elderly parents or other relatives who need care for the last one month."
    - Male: 2.788 | Female: 2.427 | Male with children: 2.695 | Female with children: 2.274
  - q6: "Not earning enough income to satisfy the demands of my child (extra-academic activities, clothes, games...)"
    - Male: 2.706 | Female: 2.591 | Male with children: 2.535 | Female with children: 2.278
  - q7: "Not being able to spend time with my child when we are at home because I must perform tasks that do not concern the family."
    - Male: 2.798 | Female: 2.573 | Male with children: 2.600 | Female with children: 2.203
- Findings:
  - Women in the sample tend to report feeling guiltier than men across all seven items.
  - Differences between men and women are statistically significant at 1 percent for all questions.
- Note on interpretation:
  - Guilty feelings are self-reported subjective ratings; cannot disaggregate gender differences into true experiential, perceptual, or reporting differences.

*Source: wpiea2019261-print-pdf - 37. Before running the econometric models, we conducted various checks to see if the — IMF Working Paper content unit.*

### 50. To see if the level of guilt in each item affect WTP, an extended model is run with

### 50. To see if the level of guilt in each item affect WTP, an extended model is run with

### Findings on guilt and willingness to pay (WTP)
- An extended model includes interactions between the guilt variables and non-wage attribute variables, and interaction effects between gender, having children and guilt.
- Table 9 reports WTP measures for hypothetical people with children, with an annual wage of 3 million yen and in their 30s.
  - "No Guilt" = person rating all guilt questions as 5 (not at all guilty).
  - "All Guilt" = person rating all guilt questions as 1 (very guilty).
- Main empirical patterns:
  - Persons who feel guilty tend to have higher WTPs.
  - For individuals who feel no guilt, WTP estimates are rarely significantly different from zero.
  - Women in the sample tend to have higher guilt levels than men; this may partly explain higher female WTPs observed earlier.
  - The final column of Table 9 examines the difference in WTP between men and women for the All Guilt case; differences are significant at the 1 percent level for most variables except Overtime2.
  - WTP figures are millions of yen in annual salary; differences are meaningful relative to the posited 3- million wage.

### Table 9: Guilt, WTP, and Gender (selected WTP values—millions of yen in annual salary)
- Columns: Male No Guilt, Male All Guilt, Female No Guilt, Female All Guilt, Difference (All Guilt)
- Overtime2:
  - Male No Guilt: -0.015
  - Male All Guilt: 0.051
  - Female No Guilt: 0.122
  - Female All Guilt: 0.187
  - Difference (All Guilt): 0.136
- Overtime3:
  - Male No Guilt: -0.146
  - Male All Guilt: 0.152
  - Female No Guilt: 0.304
  - Female All Guilt: 0.580***
  - Difference (All Guilt): 0.429***
- Overtime4:
  - Male No Guilt: 0.018
  - Male All Guilt: 0.902***
  - Female No Guilt: 0.795*
  - Female All Guilt: 1.586***
  - Difference (All Guilt): 0.684***
- Security:
  - Male No Guilt: -0.065
  - Male All Guilt: -0.191
  - Female No Guilt: 0.299
  - Female All Guilt: 0.180
  - Difference (All Guilt): 0.371***
- Transfer:
  - Male No Guilt: -0.194
  - Male All Guilt: 0.631***
  - Female No Guilt: 0.354
  - Female All Guilt: 1.114***
  - Difference (All Guilt): 0.483***
- Relocation:
  - Male No Guilt: 0.111
  - Male All Guilt: 0.096
  - Female No Guilt: 0.447*
  - Female All Guilt: 0.432***
  - Difference (All Guilt): 0.337***
- Significance notation:
  - *** = p<0.01, ** = p<0.05; * = p < 0.10

### Table 10: Women, guilt, and WTP to avoid Overtime 4 (female with children, wage = 3 million yen, in her 30s)
- All guilt questions = 5 (No Guilt): WTP to Avoid Overtime 4 = 0.795*
- q1=1, other guilt =5 (Guilty: Taking paid-leave): 0.707
- q2=1, other guilt =5 (Guilty: Leaving the office early): 0.394
- q3=1, other guilt =5 (Guilty: Not cooking): 0.600*
- q4=1, other guilt =5 (Guilty: Canceling Kids' Event): 1.272**
- q5=1, other guilt =5 (Guilty: Not caring elderly parents): 1.270**
- q6=1, other guilt =5 (Guilty: Not earning enough for kids): 0.485
- q7=1, other guilt =5 (Guilty: Not caring kids while at home): 1.394***

- Interpretation from Table 10:
  - Female WTP for Overtime 4 is highest when she feels guilty about canceling children’s events, not caring for elderly parents, and not caring for kids while at home.
  - No significant WTP when she only feels highly guilty about leaving the office early, taking paid leave, or not earning enough.

### Broader interpretation and behavioral patterns
- Two key implications:
  - Higher female WTPs relative to male WTPs may be driven by (a) a higher likelihood of women feeling guilty and (b) greater sensitivity of WTP to guilt for women.
  - Female WTPs are mostly constant across family structure or marital status, but women are more prone to feel guilty and thus have higher WTPs as working mothers.
- Specific observations on guilt:
  - Women felt guiltier about compromising home responsibilities beyond child-related events (e.g., cooking dinner).
  - In the context of Japan’s ageing population, women felt guiltier about not caring for parents; WTP was especially sensitive to this factor.
  - Guilt creates a multiplicative impact raising women’s WTP for flexibility relative to men.
- Policy implication: Policies to reduce gender gaps in labor market behavior may need to consider how feelings of guilt can be altered.

### Policy implications and conclusions (summary of policy-relevant findings)
- Choice experiments are useful to analyze labor market reforms when revealed preference data are limited; this study investigates preferences over overtime, compulsory relocations, transfers, and job security.
- Benchmark model findings:
  - People significantly dislike overtime (when it is more than the typical amount), job insecurity, intra-firm job transfer, and relocation.
  - WTP calculations suggest people are willing to forfeit a large portion of wage to avoid extreme overtime and job transfers.
  - Job choice preferences differ by gender, presence of children, and age.
  - Conditional on annual salary, women generally have higher WTPs than men to avoid overtime, relocation, transfer, and to have more secure jobs.
  - Men without children value flexibility more than men with children; for women, responses differ little by parental status.
- Policy recommendation on contracts:
  - Limited-regular contracts can be appealing for workers with various characteristics, especially women with children.
  - Policy challenge: avoid creating new glass ceilings or increasing job market segmentation.
  - For successful work style reform, policymakers should improve usability of limited-regular contracts for both men and women by:
    - Restricting discrimination against limited regular contracts.
    - Ensuring limited contracts offer the same quality job (e.g., enough training and promotion opportunities) as regular contracts.
    - Ensuring mobility across different contracts so workers can switch work styles depending on life stage and changing work-life balance needs.

*Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019261-print-pdf.pdf*

### APPENDIX 2: EXTENDED MODEL (MODEL 2)

### APPENDIX 2: EXTENDED MODEL (MODEL 2)

### Estimated coefficients (baseline variables)
- Wage: 1.375*** (0.0678)
- Wage squared: -0.0781*** (0.00516)
- Overtime2: -0.0720 (0.149)
- Overtime3: -0.578*** (0.176)
- Overtime4: -1.550*** (0.239)
- Security2: 0.233* (0.131)
- Transfer2: -0.543*** (0.135)
- Relocation2: -0.0337 (0.0989)

### Interactions with Female Dummy
- Overtime2: -0.116 (0.100)
- Overtime3: -0.422*** (0.0982)
- Overtime4: -0.801*** (0.138)
- Security2: 0.321*** (0.0919)
- Transfer2: -0.555*** (0.0997)
- Relocation2: -0.308*** (0.0701)

### Interactions with No-Children
- Overtime2: -0.0527 (0.0948)
- Overtime3: -0.222** (0.0934)
- Overtime4: -0.395*** (0.129)
- Security2: 0.401*** (0.0885)
- Transfer2: -0.104 (0.0838)
- Relocation2: -0.217*** (0.0596)

### Interactions with Female * No_Children
- Overtime2: 0.20637 (0.132)
- Overtime3: 0.224 (0.138)
- Overtime4: 0.465** (0.190)
- Security2: -0.312*** (0.117)
- Transfer2: 0.130 (0.132)
- Relocation2: 0.265*** (0.0916)

### Interactions with age cohorts (Overtime2, Overtime3, Overtime4, Security2, Transfer2, Relocation2)
- 30s # Overtime2: -0.0118 (0.127)
- 40s # Overtime2: 0.0700 (0.137)
- 50s # Overtime2: 0.0346 (0.133)

- 30s # Overtime3: 0.270* (0.161)
- 40s # Overtime3: 0.390** (0.169)
- 50s # Overtime3: 0.284* (0.166)

- 30s # Overtime4: 0.613*** (0.221)
- 40s # Overtime4: 0.704*** (0.231)
- 50s # Overtime4: 0.606*** (0.229)

- 30s # Security2: -0.0415 (0.108)
- 40s # Security2: -0.0279 (0.115)
- 50s # Security2: -0.147 (0.116)

- 30s # Transfer2: -0.253** (0.117)
- 40s # Transfer2: -0.183 (0.124)
- 50s # Transfer2: -0.237* (0.125)

- 30s # Relocation2: -0.053438 (0.0845)
- 40s # Relocation2: -0.146 (0.0907)
- 50s # Relocation2: -0.131 (0.0900)

(Note: the coefficient listed as "30s # Relocation2 -0.0534  38  (0.0845)" in source is rendered here as -0.053438 to preserve the numeric token sequence exactly as presented.)

### Interaction with Educated Dummy
- Overtime2: 0.0394 (0.0667)
- Overtime3: 0.0895 (0.0706)
- Overtime4: 0.185* (0.0950)
- Security2: -0.255*** (0.0581)
- Transfer2: 0.208*** (0.0671)
- Relocation2: 0.0336 (0.0458)

### Sample and significance
- Observations: 25,104
- Number of groups: 8,368
- Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1
- Note: Educated = 1 if graduated from 4-year universities or graduate schools and Educated = 0 otherwise.

*Source: APPENDIX 2: EXTENDED MODEL (MODEL 2) from the supplied PDF content.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019261-print-pdf.pdf_