Supplementary file 1_Dissonance between predicted and actual retirement statuses to address heterogeneous effects of retirement on mental health; evidence from JSTAR.docx

IntroductionMany studies have explored the relationship between retirement and health outcomes but findings are inconsistent, mainly owing to endogeneity in the relationship between retirement decisions and health, and the effect of heterogeneity across retiree attributes. Recent studies indicate that the mental health effects of retirement vary according to the volitionality of retirement choices taking an exogenous shock as an instrument. In this study, we proposed an alternative strategy to address retirement volitionality and effect heterogeneity using social comparison and cognitive dissonance theories, to treat the dissonance between retirement propensity and actual choice behavior.
MethodsA cross-sectional analysis was conducted using data for 1,544 Japanese men aged 60–75 years derived from the Japanese Study of Aging and Retirement. Drawing on social comparison and cognitive dissonance theories, we hypothesized that an individual’s preferred retirement status could be proxied by the predicted likelihood of retirement status determined in the reference population, and regarded the discrepancy between predicted and actual retirement status as the dissonance status of the retirement decision. The predicted retirement status was inferred from the retirement propensity estimated using a logistic regression model that included variables identified in previous studies as associated with retirement. By comparing predicted and actual retirement status, participants were categorized into four groups as follows: “predicted not-retired and actually not-retired” (PN-AN), “predicted retired and actually retired” (PR-AR), “predicted not-retired but actually retired” (PN-AR), and “predicted retired but actually not-retired” (PR-AN). We investigated between-group differences in the prevalence of depressive symptoms using logistic regression analysis.
ResultsCompared with PN and AN individuals, those who were actually retired regardless of their predicted status had higher odds ratios for depressive symptoms (1.91 [95% confidence interval: 1.16–3.12] for PR-AR and 1.84 [1.17–2.91] for PN-AR). The results were robust after adjusting for health conditions and social participation.
DiscussionOur findings indicate that retirement per se was related to depressive symptoms but dissonance between actual and predicted retirement statuses did not modify this association.

引言
既往诸多研究已探讨退休与健康结局间的关联，但所得结论并不一致，这主要源于退休决策与健康间的内生性问题，以及退休者属性异质性所带来的影响。近期研究表明，退休对心理健康的影响会因退休选择的自愿性而异，常以外生冲击作为工具变量。本研究提出了一种替代性策略，借助社会比较与认知失调理论来处理退休决策的自愿性问题及效应异质性，以化解退休倾向与实际选择行为间的失调矛盾。

方法
本研究基于日本老龄化与退休研究（Japanese Study of Aging and Retirement）的数据开展横断面分析，共纳入1544名年龄在60至75岁之间的日本男性。结合社会比较与认知失调理论，我们提出假设：个体的理想退休状态可通过参照人群中预测得到的退休状态概率来代理，并将预测退休状态与实际退休状态间的差异视为退休决策的失调状态。预测退休状态通过逻辑回归模型推导得出，该模型纳入了既往研究中已证实与退休相关的变量，以估计退休倾向。通过对比预测与实际退休状态，研究对象被分为四组：‘预测未退休且实际未退休（predicted not-retired and actually not-retired，PN-AN）’、‘预测退休且实际退休（predicted retired and actually retired，PR-AR）’、‘预测未退休但实际退休（predicted not-retired but actually retired，PN-AR）’以及‘预测退休但实际未退休（predicted retired but actually not-retired，PR-AN）’。我们采用逻辑回归分析，比较各组抑郁症状检出率的组间差异。

结果
与PN-AN组个体相比，无论预测状态如何，实际已退休者的抑郁症状比值比均更高：PR-AR组为1.91（95%置信区间：1.16~3.12），PN-AR组为1.84（1.17~2.91）。在调整健康状况与社会参与情况后，该结果仍保持稳健。

讨论
本研究结果表明，退休行为本身与抑郁症状存在关联，但实际退休状态与预测退休状态间的失调并未改变这一关联。