Reflections on My Research

A Critical Self-Review of Eight Major Articles

Published

May 14, 2026

Note

Language: 日本語版はこちら ・ This page is available in Japanese.

This page is neither a defense of the limitations of my own work nor an inflated celebration of its merits. By drawing on an independent critical review generated with AI assistance, it attempts to bring at least some measure of objectivity to looking back at what I have done and thinking about where to go next.

This document presents a candid self-review of my eight major English-language peer-reviewed articles, examining both their strengths and limitations. Social stratification research is rapidly evolving in terms of data, methods, and theoretical frameworks, and revisiting past work honestly is essential for setting research priorities going forward.

This review draws on independent critical assessments generated by two AI tools — Anthropic’s Claude Code and OpenAI’s Codex CLI — each of which read the full article PDFs and produced structured strengths-and-weaknesses analyses. The two sets of AI-generated reviews, together with my own reading, were integrated to construct the assessments on this page. The prompts used for the AI-assisted reviews are documented in the Appendix.

Articles Reviewed

# Article
[01] Fujihara, Sho and Hiroshi Ishida. 2016. “The Absolute and Relative Values of Education and the Inequality of Educational Opportunity: Trends in Access to Education in Postwar Japan.” Research in Social Stratification and Mobility 43:25–37.
[02] Fujihara, Sho, Toru Kikkawa, and Carmi Schooler. 2018. “Work Made Us What We Are: Complexity of Work, Self-directedness of Orientation, and Intellectual Flexibility of Older US and Japanese Men.” Research in Social Stratification and Mobility 54:36–45.
[03] Fujihara, Sho and Fumiya Uchikoshi. 2019. “Declining Association with Persistent Gender Asymmetric Structure: Patterns and Trends in Educational Assortative Marriage in Japan, 1950–1979.” Research in Social Stratification and Mobility 60:66–77.
[04] Fujihara, Sho. 2020. “Socio-Economic Standing and Social Status in Contemporary Japan: Scale Constructions and Their Applications.” European Sociological Review 36(4):548–561.
[05] Fujihara, Sho and Takahiro Tabuchi. 2022. “The Impact of COVID-19 on the Psychological Distress of Youths in Japan: A Latent Growth Curve Analysis.” Journal of Affective Disorders 305:19–27.
[06] Fujihara, Sho. 2023. “Explaining Class Differences in Educational Attainment in Japan: An Empirical Test of the Breen and Goldthorpe Model.” Research in Social Stratification and Mobility 83:100770.
[07] Fujihara, Sho and Hiroshi Ishida. 2024. “College Is Not the Great Equalizer in Japan.” Socius 10.
[08] Fujihara, Sho. 2024. “Identifying the Role of High School in Educational Inequality: A Causal Mediation Approach.” Social Science Research 124:103077.

Common Strengths Across the Body of Work

Reading the eight articles together, the following common strengths emerge:

  1. Systematic empirical testing of international theoretical frameworks in the Japanese context. Frameworks developed by Kohn-Schooler, Boudon, Breen-Goldthorpe, Chan-Goldthorpe, and Lundberg have been examined using Japanese data. This contributes to comparative stratification research, which has historically been dominated by Western European and North American studies.

  2. Methodological updating across articles. The articles employ generalized ordered logit, multi-group SEM with measurement invariance testing, log-multiplicative layer effect models, RC II models, latent growth curve modeling, conditional logit, entropy balancing, regression with residuals combined with the g-formula, and E-value sensitivity analysis. Each article adopts what was at the time a standard or state-of-the-art method.

  3. Open science orientation. Article [04] makes the constructed occupational scales publicly available on GitHub (ShoFujihara/OccupationalScales) to support reproducibility and community use.

  4. Use of large-scale and long-term data. The Social Stratification and Social Mobility surveys (SSM, 1955–2015), the Employment Status Survey, the Comprehensive Survey of Living Conditions, and long-term panel surveys are combined to provide internationally rare comparative data.

  5. Restrained and honest conclusions. Even where hypotheses are not fully supported, results are reported faithfully — for example, the limited explanatory power of the Breen-Goldthorpe model in Japan [06], the absence of college as an equalizer in Japan [07], and the relatively low E-values for indirect effects in causal mediation [08].

Common Weaknesses Across the Body of Work

At the same time, certain structural weaknesses recur across the corpus:

  1. Structural fragility of causal identification. All articles rely on observational data, and unmeasured confounding (cognitive ability, non-cognitive skills, local context, educational expectations, network ties) is not fully addressed.

  2. Bias toward men, married couples, and the employed. The early SSM was male-only, Kohn-Schooler studies focus on employed men, and the couple-level analyses are restricted to married, co-residing pairs.

  3. Coarse operationalization of educational and occupational categories. Some articles treat university as a binary category or combine four-year colleges and junior colleges, missing important fine distinctions in the Japanese educational hierarchy.

  4. Thin policy implications. The articles tend to stop short of concrete prescriptions about which interventions could improve outcomes for which groups.

  5. Limited direct testing of mechanisms. Patterns are often interpreted in light of theory, but the underlying mechanisms themselves are not always directly measured or tested.

Article-by-Article Assessment

[01] Fujihara and Ishida (2016): Absolute and Relative Values of Education

Strengths

  1. Explicit distinction between absolute, labor-market-relative, and distribution-relative measures of education. Disentangling the conflation between “more education” and “competitive advantage” in stratification research is a distinct contribution.
  2. Linking institutional context to measurement strategy. The institutional history of postwar Japanese educational expansion — high school expansion, higher education expansion, and university stratification — is reflected in the measurement design, including the University I/II distinction.
  3. Comparing multiple measures on the same data. This design shows that previously conflicting findings about Japanese educational inequality reflect different operationalizations of education, not only substantive change.
  4. Connection to international debates. The article engages with Mare, Thurow, and Bukodi and Goldthorpe, positioning the Japanese case within the broader literature on education as a positional good.
  5. Dual conclusions. The article reports a narrowing of inequality measured in years of schooling but a persistent or widening gap at the top of the distribution, offering a more nuanced picture of educational expansion than single-measure analyses.

Weaknesses

  1. The theoretical distinction between labor-market-relative and distribution-relative measures is not sharply drawn.
  2. Social origin is reduced to father’s education alone.
  3. Analysis is restricted to men, omitting the historically important expansion of women’s education.
  4. The University I/II distinction collapses considerable heterogeneity in selectivity, field, and region.
  5. The EPS measure (based on occupational prestige) introduces endogeneity and time-alignment issues.
  6. Identification is essentially descriptive trend analysis.
  7. Theoretical interpretations (relative risk aversion, market-share defense) are not directly tested.

Overall Assessment

The article persuasively shows, using postwar Japan as a case, that trends in educational inequality look different depending on how education is measured. Viewing educational expansion as a reconfiguration of relative status competition and institutional stratification — rather than merely an increase in years of schooling — has provided an important basis for subsequent work.

Implications for Future Research

  • Retain the three-axis design of absolute, labor-market-relative, and distribution-relative measures.
  • Expand origin measurement beyond father’s education to include father’s occupation, mother’s education, household income, region, and number of siblings.
  • Connect educational attainment to occupational, income, and employment-stability outcomes.

[02] Fujihara, Kikkawa, and Schooler (2018): Work and the Self

Strengths

  1. Engaging directly with a classic and important question. The reciprocal formation of personality and cognition by social structure, and the selection of work conditions by these psychological traits, is a topic of theoretical importance.
  2. Combining long-term panel data with multi-group SEM. Long panels (US 1964–1994/95; Japan 1979–2006) are analyzed with multi-group SEM and equality-of-coefficient constraints to assess the isomorphism of the reciprocal structure.
  3. Rarity of the data. A Kohn-Schooler-comparable panel that follows individuals for nearly three decades in Japan is virtually irreplaceable as a research resource.
  4. International comparison. The findings support the view that the Kohn-Schooler hypothesis is not US-specific but holds in stable industrial societies more generally.
  5. Co-authorship with Schooler. The collaboration with one of the originators of the framework ensures theoretical continuity.

Weaknesses

  1. Small samples (US 154; Japan 142) with strong selection on continued employment in older age.
  2. Women, non-employed individuals, and retirees are excluded.
  3. The Japanese sample is regionally restricted, limiting representativeness.
  4. The measurement invariance assumption across a roughly 30-year span and across two countries is strong.
  5. Two-wave SEM has structural limits in disentangling simultaneous and lagged effects.
  6. The article largely confirms an existing framework rather than developing Japan-specific theory.
  7. Policy and practical implications remain abstract.

Overall Assessment

The article rigorously tests the international applicability of the Kohn-Schooler hypothesis using a rare long-term panel. It is best read as establishing a robust association in older employed men in stable industrial societies, rather than as a universally generalizable finding.

Implications for Future Research

  • Maintain the international comparison, long panel, and latent-variable modeling approach while incorporating Japan-specific employment institutions.
  • Extend the analytic population to women, retirees, and non-standard workers.
  • Move toward identification strategies based on concrete institutional changes (job redesign, retirement-age policy, reassignment).

[03] Fujihara and Uchikoshi (2019): Educational Assortative Marriage

Strengths

  1. Conceptual contribution. Examining not only whether educational homogamy has declined but also whether the asymmetric gender structure has persisted is a clear analytical advance.
  2. Use of large-scale data. The analysis of 481,144 couples substantially improves on the small-sample limitations of prior Japanese studies, and the six-category education classification distinguishes professional training colleges, junior colleges, and graduate schools.
  3. Persuasive methodological design. Log-linear, log-multiplicative layer-effect, and regression-type models are compared in stages, avoiding reliance on any single specification.
  4. Clear empirical findings. The log-odds-ratio of the husband-wife association declined by roughly 25% between the 1950–54 and 1975–79 wife birth cohorts, while the gender-asymmetric structure of hypergamy persisted.
  5. International positioning. Results are placed in dialogue with the literature on Western and East Asian patterns of assortative mating.

Weaknesses

  1. Restriction to married, co-residing couples introduces selection bias related to divorce and widowhood.
  2. Marriage year and age at marriage cannot be directly observed; birth cohort serves as a proxy.
  3. Mechanisms behind the decline are inferred from existing theory rather than directly tested.
  4. The institutional meaning of educational categories has shifted over cohorts (e.g., professional training colleges).
  5. Sensitivity analysis regarding the integration of two surveys is limited.
  6. The relation between relative association and absolute marriage distributions is not made fully explicit.
  7. The analysis stops at the 1975–79 wife birth cohort.

Overall Assessment

This is a solid empirical study that can serve as a standard reference point for educational assortative marriage research in Japan. Large data and sophisticated association models jointly establish the simultaneous decline of homogamy and persistence of gender-asymmetric structure.

Implications for Future Research

  • Use data that directly capture the marriage formation process, including age at first marriage and meeting venues.
  • Connect educational assortative marriage to its consequences for household income, maternal employment, and children’s educational opportunities.
  • Extend the cohorts analyzed to those born in the 1980s and later.

[04] Fujihara (2020): Socio-Economic Standing and Social Status

Strengths

  1. Conceptual contribution. A Weberian distinction between class and status is reconstructed as an empirical framework applicable to Japan, with occupational prestige, socio-economic status, and social status compared directly.
  2. Methodological contribution. The JSEI is constructed from years of education, earnings, and existing prestige scores, while the JSSI is constructed from husband-wife occupational combinations using an RC II model. The construction logic of each is relatively transparent.
  3. Data strength. Scales are built on the large-scale Employment Status Survey (949,911 individuals) using a detailed 231-category occupational classification, with external validation drawn from SSM 2005 and 2015.
  4. Clear empirical results. The JSEI and JSSI capture intergenerational occupational correlations better than the existing prestige scale, and the JSSI is more strongly associated with cultural participation, consistent with theoretical expectations.
  5. Significance for international comparison. The European framework of class-status distinction is examined in a non-Western context, with comparisons to the SIOPS, ISEI, and ICAMS.
  6. Open science. The constructed scales are publicly available on GitHub.

Weaknesses

  1. The JSSI rests on circularity concerns, as marital combinations may reflect educational homogamy and gender role specialization rather than a pure status order.
  2. Gender differences are not adequately handled.
  3. The JSSI is highly correlated with education, raising questions about its distinctiveness.
  4. The analyses are validation-based and not causally identified.
  5. The set of cultural activities measured is somewhat narrow.
  6. The scales rely on occupational titles and do not capture employment type, firm size, or workplace authority.
  7. There is some misalignment between the time points of scale construction and application.

Overall Assessment

The article provides reusable scales for stratification research in Japan and links theory, scale construction, large-scale data, and international comparison. Interpretation of the JSSI as “social status” requires care, particularly with respect to its overlap with education, marriage market dynamics, and gender structure.

Implications for Future Research

  • Use the scales while explicitly examining their overlap with education, gender, and employment type.
  • Develop gender-specific occupational indices.
  • Examine the impact of non-standard employment, professionalization, and the growth of service occupations on the status order.

[05] Fujihara and Tabuchi (2022): COVID-19 and Youth Mental Health

Strengths

  1. Longitudinal data with a pre-pandemic baseline. Including December 2019 as a baseline before COVID-19 is relatively rare and allows analysis of changes in distress rather than only levels.
  2. Latent growth curve modeling to separate level and change. Modeling initial distress and rate of change separately helps disentangle “those who were already distressed” from “those who became more distressed during the pandemic.”
  3. Broad set of predictors. Demographic, socio-economic, subjective status, schooling and employment, social network, health, maternal K6, personality, and school adjustment variables are all examined.
  4. Internationally comparable measure (K6). Findings can be set alongside US and UK longitudinal studies.
  5. Treatment of missingness. Multiple imputation with auxiliary variables addresses the high attrition typical of online youth panels.

Weaknesses

  1. The initial sample is drawn from an online access panel, with a 45% initial response rate, limiting generalizability.
  2. High attrition makes the analysis heavily dependent on the missing-at-random assumption.
  3. The “impact of COVID-19” is identified only by temporal coincidence, not by a counterfactual design.
  4. Time-varying factors (economic anxiety, class delivery format, social contact) are not measured in detail.
  5. Effect sizes are small, and clinical significance is not strongly established.
  6. Exploratory tests of many predictors raise multiple testing concerns.
  7. Some short-form scales (anxiety, self-esteem) have low reliability.

Overall Assessment

The article uses a rare panel that begins before the pandemic and shows that young women’s psychological distress increased through July 2020, with the change largely unexplained by pre-pandemic characteristics. The design of separating level and change has applicability beyond pandemic studies.

Implications for Future Research

  • Measure time-varying factors (schooling format, employment changes, family roles, online communication) wave by wave.
  • Treat baseline vulnerability and acute deterioration during crises as analytically distinct problems.
  • Continue tracking young women’s mental health with greater attention to gendered stressors.

[06] Fujihara (2023): Testing the Breen and Goldthorpe Model

Strengths

  1. Theoretical clarity. The Breen and Goldthorpe model is decomposed into subjective cost, success probability, status-maintenance benefit, and status-maintenance motivation, and tested directly. Treating status-maintenance motivation as a moderator rather than a mediator is theoretically coherent.
  2. Non-European application. Direct empirical tests of the BG model have been concentrated in Europe; testing in Japan is comparatively valuable.
  3. Longitudinal design. Subjective evaluations are measured during high school and linked to subsequent educational outcomes, providing a clearer temporal ordering than cross-sectional designs.
  4. Appropriate analytic methods. Conditional logit models handle alternative-specific subjective evaluations, and the KHB method is used for mediation proportions in non-linear models.
  5. Restrained conclusions. The article reports partial and limited support for the BG model rather than overstating its applicability.

Weaknesses

  1. The sample is drawn from an access panel, raising representativeness concerns.
  2. Sample size is small for a multinomial choice model.
  3. Origin classes are dichotomized into service vs. non-service.
  4. Subjective cost, success probability, and benefit measures are general rather than precise.
  5. The “status-maintenance benefit” item may not fully capture the BG concept.
  6. The reasons for the small mediation effects are not fully explored.
  7. Controlling for self-reported GPA risks over-control of an intermediate variable.

Overall Assessment

This is an important direct empirical test of the BG model in Japan. The central conclusion — that BG mechanisms operate as predicted in some respects but explain only a limited portion of class differences — is itself a substantively important finding about the distinctive structure of educational inequality in Japan.

Implications for Future Research

  • Move beyond replication of the BG model toward analyzing why its explanatory power is limited in Japan.
  • Refine the measurement of subjective cost, success probability, and benefit.
  • Incorporate institutional structures (high school rank, course tracks, exam preparation resources) that shape the opportunity structure before subjective evaluations are formed.

[07] Fujihara and Ishida (2024): College and the Great Equalizer

Strengths

  1. Clear question with international significance. The classic question of whether college offsets disadvantage by social origin is re-examined in the Japanese institutional context, rather than assumed in the Western form of “college as the great equalizer.”
  2. Extending the OED triangle. Rather than only examining origin-destination associations within educational levels, the article examines heterogeneous returns to college by social origin.
  3. Data quality. Multiple representative datasets are combined, with separate analyses for men and women, first and current jobs, and prestige and SEI outcomes.
  4. Attention to selection. Entropy balancing is applied across a rich set of pre-college covariates, including parental education and occupation, sibship, ninth-grade academic performance, high school rank, and high school track.
  5. Plausible institutional interpretation. The article links its findings to the difference between school-mediated job placement after high school and family-resource-mediated job search after college.

Weaknesses

  1. The reformulation of “great equalizer” as “heterogeneous returns” shifts the original question.
  2. The treatment definition combines four-year colleges and junior colleges asymmetrically for men and women.
  3. Professional training colleges, dropouts, and graduate-school graduates are excluded.
  4. Outcomes are limited to occupational status (prestige and SEI), excluding income, employment type, and firm size.
  5. Institutional mechanisms are not directly measured.
  6. Period heterogeneity (expansion, employment-ice-age cohorts, women’s labor market entry) is not central to the main analysis.
  7. The causal language is somewhat stronger than the design fully warrants.

Overall Assessment

The article carefully examines the equalizing function of college in Japan from theoretical, methodological, and institutional perspectives. It offers an important counter-finding to the Western-centered “college as equalizer” thesis.

Implications for Future Research

  • Treat the expansion of college access as insufficient on its own; consider gaps in college quality, fields of study, and labor-market support.
  • Re-evaluate the role of high-school-mediated job placement as a model of less family-dependent institutional support.
  • Connect within-college stratification to labor market outcomes including income, employment type, and career progression.

[08] Fujihara (2024): The Role of High School Through Causal Mediation

Strengths

  1. Explicit specification of estimands in causal mediation. Rather than relying on a simple comparison of regression coefficients, the article defines randomized analogs of the natural total, direct, and indirect effects, and the controlled direct effect, treating “how much high school mediates” and “how the income effect varies by high school rank” as distinct questions.
  2. Treatment of treatment-induced mediator-outcome confounding. Variables measured in ninth grade (academic performance, educational expectations, study time, cram school attendance) are explicitly handled as treatment-induced confounders, addressing a problem that standard regression approaches typically fail to manage.
  3. Use of regression with residuals. This approach accommodates continuous treatment, continuous mediator, intermediate confounding, and treatment-mediator interactions. Comparison with conventional regression shows that the latter overestimates the mediation share substantially.
  4. Use of randomized analogs in place of natural effects. Because natural effects are not identified in the presence of treatment-induced confounding, the article appropriately switches to randomized analogs and clearly states this in a footnote.
  5. Sensitivity analysis with E-values. The article reports that randomized direct effects are relatively robust while indirect effects have low E-values, candidly acknowledging that mediation conclusions are sensitive to unmeasured confounding.

Weaknesses

  1. Identification assumptions remain strong even with the RWR approach.
  2. E-values are useful but a limited form of sensitivity analysis.
  3. Comparing the lowest and highest income percentiles is conceptually transparent but politically and practically distant from realistic interventions.
  4. High school rank is reduced to a single continuous variable (standardized rank score).
  5. Linear and linear-probability specifications introduce specification dependence.
  6. The placement of intermediate confounders in time may not perfectly correspond to their developmental ordering.
  7. Generalizability is limited by Japan’s distinctive high school entrance examination system.

Overall Assessment

The article reframes the role of high school in educational attainment using causal mediation analysis rather than conventional mediation regression. The combination of regression with residuals, controlled direct effects, randomized analogs, and E-value sensitivity analysis represents a methodologically refined approach. The conclusion — that high school rank does mediate but to a smaller degree than conventional methods suggest, with much of the family income effect operating directly — is appropriately cautious.

Implications for Future Research

  • Make the choice of estimand explicit before mediation analysis (natural effects, randomized analogs, or controlled direct effects).
  • Design longitudinal studies that measure intermediate confounders at the right developmental stages.
  • Combine high school reform with interventions on pre-high-school learning opportunities and family resources, rather than expecting high school policy alone to close attainment gaps.

Future Research Directions: A Two-Axis View

Axis 1: Building on Strengths

The strengths that have accumulated across the eight articles can be summarized as:

Strength Articles
Multi-dimensional measurement of education and occupation [01] [04] [07]
Sociological application of causal inference [07] [08]
Internationally comparable empirical evidence on Japan [01] [02] [03] [04] [07]
Open science (publicly available scales and code) [04]
A culture of methodological updating [01] → [08]

Axis 2: Addressing Weaknesses

The structural weaknesses across the corpus can be summarized as:

Weakness Articles
Bias toward men, married couples, and the employed [01] [02] [03] [04]
Coarse educational and occupational categories [01] [04] [07]
Fragility of causal identification All articles
Limited direct testing of mechanisms [03] [05] [06] [07]
Thin policy implications All articles
Conflation of “average effects” and “distributional changes” [01] [05] [07] [08]

Synthesis

Viewed individually, each article has substantial limitations. Viewed as a program, however, three patterns become clear:

  1. The same set of questions (educational inequality, stratification and occupation, stratification and health) has been repeatedly revisited across different periods, methods, and datasets.
  2. The weakness of one article often becomes the motivation for the next. For example, the single-indicator measurement in [01] motivates the multi-dimensional scale construction of [04]; the descriptive analysis in [04] motivates the heterogeneity analysis of [07]; and the limitations of conventional causal inference in [07] motivate the causal mediation approach of [08].
  3. The weaknesses thus reflect the current location of an unfinished research program rather than fundamental flaws.

The direction for future research is therefore:

  • Sustain and extend the strengths — international frameworks applied to Japanese empirical evidence, methodological updating, and open science.
  • Address the structural weaknesses through next-generation research designs and data collection, rather than through retrofits to individual articles.
  • Bring a research-question- and estimand-first orientation to bear on the accumulated methodological tools, aiming to raise the standard of causal inference in social stratification research.

Closing Remarks

Research is always conducted within the constraints of the theoretical frameworks, data, and methods available at the time. As this review suggests, my own work has both genuine strengths and real limitations. What matters, in my view, is not treating past work as definitive but acknowledging its limits honestly and letting them motivate the next study. Going forward, I aim to advance research on stratification, educational opportunity, occupations, and health with more rigorous causal identification, richer measurement, broader analytic populations, and clearer policy implications.

Appendix: Prompts and Workflow

The independent critical reviews for each article were obtained from two AI tools: Anthropic’s Claude Code and OpenAI’s Codex CLI. Both tools were given the full PDF of each article and the same review template (translated from Japanese):

You are a strict and fair reviewer well-versed in social stratification, quantitative sociology, and causal inference. Examine the article critically and constructively for the purpose of a self-evaluation document. Use the following structure (in Japanese):

Strengths (3–5 points, 2–3 sentences each): Evaluate the article’s distinctive contributions in conceptual, methodological, empirical, data-related, and international dimensions. Avoid exaggeration.

Weaknesses (5–7 points, 2–3 sentences each): Be similarly strict.

Overall assessment (2–3 sentences): Net contribution, distinctiveness, and value for inheritance in the research program.

Implications for future direction (2–3 points): Both directions for building on strengths and directions for addressing weaknesses, with concrete suggestions.

The two sets of AI-generated reviews — independent of each other and produced by models from different research organizations — were read alongside my own reading and integrated into the assessments above.

I view this kind of AI-assisted critical self-review as a way to obtain additional, structured “second opinions” that complement but do not replace careful reading by the author or standard peer review. The text of this page reflects my own judgments; the AI-assisted reviews informed the structure and ensured that important critical points were not overlooked.