More than mirror images: Health associations of transitions between formal and informal employment Authors: Liwen Chen, Kun Liu Published: 2026 ABSTRACT Informal employment is an important social determinant of health, yet little is known about whether transitions into and out of informal employment are associated with health in similar ways. Using nationally representative panel data from the China Family Panel Studies from 2014 to 2020, this study examines whether transitions from formal to informal employment and from informal to formal employment are differently associated with self-rated health among non-agricultural workers in China. Asymmetric fixed-effects models show that transitioning from formal to informal employment is associated with a 0.078-point decline in self-rated health on a five-point scale, equivalent to approximately 0.07 standard deviations. The corresponding estimate for exit from informal employment is smaller and not statistically significant. Event-study analyses indicate that the entry-related health decline becomes apparent around the observed transition interval and is not preceded by differential health deterioration. The entry-related health decline amounts to approximately 0.16 standard deviations among workers over age 45 and 0.08 standard deviations among workers with a high school education or less, whereas the corresponding entry estimates are not statistically significant among younger and more-educated workers. These findings suggest that the direction of labor-market transition matters for health and highlight the importance of reducing avoidable movement into informal employment and extending social protection across employment forms. Introduction Over recent decades, labor markets worldwide have become increasingly flexible, accompanied by rapid growth in informal work arrangements linked to technological change, platform-mediated work, and gig economies (Günther & Launov, 2012; Kalleberg, 2011; Standing, 2011; Todolí-Signes, 2017). Informal employment refers here to non-standard work arrangements that lack a direct and stable employment relationship with the organization for which the work is performed (Aronsson et al., 2023; Lee & Di Ruggiero, 2022; Ling et al., 2024; Zhou et al., 2024). It now involves nearly 2 billion workers worldwide and accounts for 61.2% of total employment, with particularly high prevalence in the Asia-Pacific region, where it represents roughly 66% of employment and close to two-thirds of the global informal workforce (Bonnet et al., 2019; ILO, 2025). Against this backdrop, informal employment has drawn increasing academic attention as a central social determinant of health and health equity. Its significance lies in the fact that it systematically structures workers’ exposure to income insecurity, weak employment and social protection, and chronic psychosocial stress, thereby contributing to the unequal distribution of health risks and health outcomes across populations (Benach et al., 2014; Hajat et al., 2024). Despite growing interest in this issue, the health consequences of informal employment remain insufficiently understood (Lee & Di Ruggiero, 2022; Zhou et al., 2024). Much of the existing literature relies on cross-sectional comparisons between formal and informal workers at a single point in time (Aronsson et al., 2023; Julià et al., 2017, 2019; Koranyi et al., 2018; López-Ruiz et al., 2015, 2017; McNamara et al., 2021). Such studies are valuable, but they reveal less about how health changes when individuals move between employment statuses and how those changes unfold over the course of a labor-market trajectory (Amick et al., 2016; Benach et al., 2014). This omission matters because the health consequences of employment may depend not only on the type of job a person holds, but also on the direction of transition and the resources lost or gained in the move (Amick et al., 2016; Benach et al., 2014). Recent longitudinal research has made this point more explicit by showing that employment transitions are not simply mirror-image processes in their psychological well-being implications. Gebel and Voβemer (2014) report that unemployment and reemployment have similarly sized but opposite associations with psychological health, whereas Park et al. (2024) show that transitions into and out of informal employment are asymmetric in their associations with life satisfaction. This perspective raises the possibility that entering and exiting informal employment may have different health implications. Such asymmetry is theoretically plausible. The stress-process framework suggests that adverse events and chronic strains affect health not only through direct exposure to stressors, but also through the loss of coping resources and the spread of strain across work, family, and economic domains (Pearlin et al., 1981, 2005; Thoits, 2010). Entry into informal employment may constitute such a stressor because it often entails reduced fringe benefits, weaker regulatory protection, and more limited access to employment-based social protection (Benach et al., 2000, 2014; Fujishiro, 2019; Lee & Di Ruggiero, 2022; Naicker et al., 2021). It may also heighten uncertainty about job continuity and future prospects, thereby increasing psychosocial stress and undermining well-being and health (Allan et al., 2021; Connell & Burgess, 2006; Kalleberg, 2011; Llosa Fernández, Menéndez Espina, Agulló Tomás, & Rodríguez Suárez, 2018; Scherer, 2009). Recent evidence further shows that deteriorating employment conditions can operate as health-relevant stressors (Hajat et al., 2024; Oddo et al., 2024). Although informal employment may in some cases provide autonomy or a better fit with other non-work responsibilities (Kim et al., 2020; Temkin, 2016; Zhou et al., 2024), these potential benefits may depend on whether the transition is chosen or constrained. By contrast, weaker employment protection, reduced income predictability, and more limited institutional coverage may be more directly associated with entry into informal employment and may affect a broader group of workers. These considerations suggest that entering informal employment may be associated with a deterioration in health, particularly when the transition involves a decline in security, protection, and employment-related resources. By contrast, exiting informal employment may not yield a commensurate improvement in health. Although movement into formal employment can restore employment protection and improve access to health care and social protection (Lee & Di Ruggiero, 2022; Naicker et al., 2021), the health benefits of this transition are unlikely to be immediate or complete. A key reason is that the adverse effects of precariousness are often cumulative and persistent, with chronic insecurity and hazardous working conditions generating physiological and psychosocial wear that does not quickly dissipate when employment conditions improve (Amick et al., 2016; Benach et al., 2014). Research from the life-course perspective further suggests that exposure to poor-quality employment can become embedded over time through cumulative disadvantage, repeated stress exposure, and reduced opportunities for recovery (Dannefer, 2003, 2020; Kuh et al., 2003). Consistent with this view, recent research links adverse employment histories across the working life to allostatic load, indicating that prolonged exposure to insecure employment conditions may generate physiological wear that persists beyond the immediate employment episode (McEwen, 1998; Wahrendorf et al., 2022) and continues to leave “scarring effects” that are not fully offset by later improvements in job quality (Amick et al., 2016; Benach et al., 2014; Chen & Wang, 2024; Moscone et al., 2016). Existing research indicates that transitions from unemployment to employment tend to produce clearer improvements in psychological health than in physical health, while adaptation to employment-related life events varies substantially in both magnitude and duration (Gebel & Voβemer, 2014; Luhmann et al., 2012). Recent work on aging-related health outcomes further extends this life-course argument by showing that employment transitions in later life may have asymmetric associations with frailty, suggesting that labor-market entry and exit can have distinct implications for health over the life course (Nam & Kim, 2026). Existing research also suggests that workers may adjust to deteriorating employment conditions, but such adjustment does not necessarily imply a return to prior health. Recent evidence on life-event adaptation suggests that some individuals adapt to changed circumstances while others do not (Chilver et al., 2023), and qualitative research on precarious work shows that frequent adjustment to unstable workplaces can itself be stressful and exhausting (Irvine & Rose, 2024). Similarly, recovery after moving back into more secure employment may be incomplete if prior insecurity has already generated material or psychological costs (Wahrendorf et al., 2022). Thus, asymmetry may arise because deteriorating employment conditions have larger or more persistent health effects than subsequent improvements, consistent with recent evidence that different directions of employment transitions have unequal implications for mental health and subjective well-being (Jin et al., 2025; Nam & Kim, 2026; Park et al., 2024). At the same time, formal employment may itself introduce new pressures, including tighter supervision, lower discretion, and higher job demands, which can temper the expected benefits of improved employment status (Hessels et al., 2017; Nikolova, 2019). Consistent with this perspective, the health improvements of moving from informal to formal employment may be incomplete, pointing to the possibility that transitions into and out of informal employment need not be associated with health in symmetric ways. Given these considerations, the key question is not only how informal employment is associated with health outcomes, but whether health changes following employment transitions are symmetric and unfold over similar temporal patterns. If entering informal employment is associated with a health decline, whereas exiting informal employment is not associated with a comparably clear improvement, then reducing avoidable or unwanted transitions into informal employment may be at least as important for population health as facilitating movement into formal employment. Yet few studies have directly examined these questions. Although recent research has considered whether employment transitions are asymmetrically associated with subjective well-being (Gebel & Voβemer, 2014; Park et al., 2024), evidence remains limited for broader health outcomes that predict subsequent morbidity and mortality. It is also unclear whether differences between entry and exit reflect pre-existing health trajectories, changes that become apparent around the transition interval, or changes that unfold across subsequent survey waves. Accordingly, this study examines whether transitions into and out of informal employment are differently associated with self-rated health and how these associations evolve around and after the transition interval. Employment transitions may affect health through several overlapping material and psychosocial pathways, including psychological distress, physical symptoms, functional capacity, and chronic health processes. Because these dimensions may respond differently and unfold over different periods, we examine their combined expression in self-rated health (Jylhä, 2009). Self-rated health is widely used and validated in population-health research (Adebisi et al., 2025) and predicts subsequent morbidity and mortality (Franks et al., 2003; Idler & Benyamini, 1997; Xiong et al., 2022). The analyses are based on nationally representative panel data from the China Family Panel Studies, covering 2014–2020. China provides an important setting for this analysis because informal employment is widespread and labor market stratification plays a central role in shaping workers’ exposure to insecurity and unequal protection. We estimate standard and asymmetric fixed-effects models, with the latter allowing the associations of entry into and exit from informal employment to differ (Allison, 2019). We then use event-study and segmented-regression analyses to assess whether health was already changing before the observed transition and whether subsequent changes are concentrated around the transition interval or continue across later survey waves. In doing so, the study treats informal employment not simply as a static employment category, but as a dynamic social exposure through which labor market precarity may shape health and contribute to health inequality over time. Data and measures Data This study uses data from the CFPS, a nationally representative panel survey conducted by the Institute of Social Science Survey at Peking University, covering the period from 2014 to 2020. The CFPS collects rich information on employment, health, and socioeconomic characteristics and is therefore well suited to examining the health implications of informal employment. Its panel structure enables within-individual analyses that reduce bias from time-invariant unobserved heterogeneity and capture labor-market transitions over time. The analytic sample is restricted to non-agricultural individuals aged 16 to 59 who were actively working at the time of the survey. Retirees, full-time students, and observations with missing values on key variables were excluded. The final analytic sample consists of 19,489 person-wave observations. Measures Health outcome. The outcome is respondents’ overall assessment of their health. We use self-rated health because the study focuses on changes in general perceived health rather than on any single mental or physical condition. Conceptually, self-rated health reflects how individuals synthesize information about both bodily and psychological states, and empirical studies confirm that physical and mental health both contribute to these assessments (Jylhä, 2009; Singh-Manoux et al., 2006). Although physical functioning may carry greater weight than mental or social functioning, respondents still draw on several dimensions of health when evaluating their overall condition (Mavaddat et al., 2011). The broad nature of the measure also captures aspects of health that are missed by more narrowly defined indicators (Au & Johnston, 2014). Its predictive validity is supported by extensive evidence (DeSalvo et al., 2006; Franks et al., 2003; Idler & Benyamini, 1997), including studies from China and other lower-income settings (Falk et al., 2017; Xiong et al., 2022; Yu et al., 1998). In the CFPS, respondents were asked to rate their overall health on a five-point scale originally coded from 1 (“very healthy”) to 5 (“unhealthy”). We reverse-coded this measure so that higher values indicate better perceived health. Informal employment. The main explanatory variable is informal employment. Following prior research on employment arrangements and health in China, we use a broad measure that includes workers without a formal labor contract and workers employed through a labor-dispatch arrangement (Ling et al., 2024). The indicator is coded 1 for informal employment and 0 otherwise. The analysis is restricted to the non-agricultural workforce because agricultural employment has distinct institutional and occupational characteristics. The absence of a written contract is commonly used as an operational indicator of informal employment because it may indicate that the employment relationship is not effectively covered by labor regulation, employment benefits, or social protection (Gallagher et al., 2015; ILO, 2013). Labor dispatch differs legally because dispatched workers may hold a contract with an intermediary agency. Nevertheless, it departs from the direct standard employment relationship by separating the legal employer from the organization that supervises and uses the worker's labor. Research in China consequently characterizes labor dispatch as a form of precarious work associated with fragmented employer responsibility and persistent employment insecurity (Feng, 2019; Zhang, 2021). We combine these arrangements in the main analysis because both can involve weaker attachment to a direct employer and reduced employment-related protection compared with direct formal employment. Recognizing their differences, we conduct sensitivity analyses using alternative definitions. Concurrent changes in employment conditions and protection. We also examine changes in employment conditions and social insurance protection associated with transitions into and out of informal employment. These measures are intended to characterize the employment changes accompanying the observed health associations. Employment conditions are captured by the natural logarithm of annual earnings and an indicator for receiving any cash benefits. Perceived job quality is measured using overall job satisfaction, assessed on a five-point scale, and a binary indicator of perceived career prospects based on whether respondents report promotion opportunities in their current job. Social insurance protection is measured by indicators for coverage under pension insurance, medical insurance, unemployment insurance, and work injury insurance. Control variables. All models control for age, gender, marital status, educational attainment, chronic disease status, and hukou status, which are commonly associated with both employment and health (Benach et al., 2014; Ling et al., 2024). Descriptive statistics for the analytic sample Table 1 reports descriptive statistics for the sample. The mean self-rated health score was 3.28 on a five-point scale, and 54.1% of observations were classified as informal employment. Respondents had a mean age of 38.3 years, 60.1% were men, and 43.3% held non-agricultural hukou. The descriptive statistics also show substantial variation in earnings, benefits, perceived job quality, and social insurance coverage. Fewer than half of respondents were covered by pension or medical insurance. The number of observations varies across variables due to item nonresponse. Additional balance checks indicate that item nonresponse was not associated with informal employment status.Table 1Characteristics of the analytic sample.Table 1VariableNMeanSDMinMaxSelf-rated health19,4893.2781.05115Informal employment19,4890.5410.49801Age19,48938.329.7321659Male19,4890.6010.49001Marital status (single)19,4890.1220.32701High school education or above19,4890.2670.44201Non-agricultural hukou19,4890.4330.49501Log annual earnings11,73010.171.510013.82Receipt of any cash benefits11,7220.3620.48101Job satisfaction19,4883.5450.80315Perceived career prospects17,9270.7140.45201Pension insurance13,9040.4500.49801Medical insurance14,0140.4550.49801Unemployment insurance12,5170.3890.48801Work injury insurance13,0490.4140.49301Notes: The number of observations varies across variables due to missing values. Table 1Characteristics of the analytic sample.Table 1VariableNMeanSDMinMaxSelf-rated health19,4893.2781.05115Informal employment19,4890.5410.49801Age19,48938.329.7321659Male19,4890.6010.49001Marital status (single)19,4890.1220.32701High school education or above19,4890.2670.44201Non-agricultural hukou19,4890.4330.49501Log annual earnings11,73010.171.510013.82Receipt of any cash benefits11,7220.3620.48101Job satisfaction19,4883.5450.80315Perceived career prospects17,9270.7140.45201Pension insurance13,9040.4500.49801Medical insurance14,0140.4550.49801Unemployment insurance12,5170.3890.48801Work injury insurance13,0490.4140.49301Notes: The number of observations varies across variables due to missing values. Characteristics of the analytic sample. Notes: The number of observations varies across variables due to missing values. Table 2 summarizes the employment status transitions that provide the within-person variation for the fixed-effects analyses. Of the 7,167 individuals in the analytic sample, 5,281 experienced no observed change in employment status, while 1,886, or 26.3%, transitioned between formal and informal employment at least once. Among those who transitioned, 1,500 experienced exactly one transition and 386 experienced two or more transitions. In total, the sample contains 2,302 transition episodes, including 1,165 transitions from formal to informal employment and 1,137 transitions from informal to formal employment. These episodes were contributed by 1,146 and 1,126 individuals, respectively. Because individuals with repeated transitions may contribute episodes in both directions, these two individual counts are not mutually exclusive.Table 2Employment status transitions in the analytic sample.Table 2Transition patternPerson-wave observationsIndividualsTransition episodesPanel A. Employment status transitions in the full sample1. No employment-status transitionNo transition14,0675,2810 Remained in informal employment7,9023,0510 Remained in formal employment6,1652,23002. One employment-status transitionOne transition only4,0401,5001,500 Informal-to-formal employment1,999740740 Formal-to-informal employment2,0417607603. Two or more employment-status transitionsTwo or more transitions1,382386802 Informal-to-formal employment/386397 Formal-to-informal employment/3864054. Total analytic sample19,4897,1672,302 At least one formal to informal transition/1,1461,165 At least one informal to formal transition/1,1261,137Panel B. Employment status transitions in the analytic subgroupsAge≤45Age >45High school education or lessMore than high school educationPerson-wave observations15,4694,02014,4944,995Individuals5,6501,5175,3821,785Transition episodes19033991,792510Informal-to-formal employment953184871266Formal-to-informal employment950215921244Notes: Person-wave observations refer to survey records contributed by respondents. Transition episodes refer to observed changes in employment status between consecutive CFPS waves. Individuals may contribute more than one transition episode and may contribute transitions in both directions. The age and education groups correspond to the samples used in the subgroup regressions. Table 2Employment status transitions in the analytic sample.Table 2Transition patternPerson-wave observationsIndividualsTransition episodesPanel A. Employment status transitions in the full sample1. No employment-status transitionNo transition14,0675,2810 Remained in informal employment7,9023,0510 Remained in formal employment6,1652,23002. One employment-status transitionOne transition only4,0401,5001,500 Informal-to-formal employment1,999740740 Formal-to-informal employment2,0417607603. Two or more employment-status transitionsTwo or more transitions1,382386802 Informal-to-formal employment/386397 Formal-to-informal employment/3864054. Total analytic sample19,4897,1672,302 At least one formal to informal transition/1,1461,165 At least one informal to formal transition/1,1261,137Panel B. Employment status transitions in the analytic subgroupsAge≤45Age >45High school education or lessMore than high school educationPerson-wave observations15,4694,02014,4944,995Individuals5,6501,5175,3821,785Transition episodes19033991,792510Informal-to-formal employment953184871266Formal-to-informal employment950215921244Notes: Person-wave observations refer to survey records contributed by respondents. Transition episodes refer to observed changes in employment status between consecutive CFPS waves. Individuals may contribute more than one transition episode and may contribute transitions in both directions. The age and education groups correspond to the samples used in the subgroup regressions. Employment status transitions in the analytic sample. Notes: Person-wave observations refer to survey records contributed by respondents. Transition episodes refer to observed changes in employment status between consecutive CFPS waves. Individuals may contribute more than one transition episode and may contribute transitions in both directions. The age and education groups correspond to the samples used in the subgroup regressions. Table 2 also reports the transition support underlying the subgroup analyses. Respondents aged 45 or younger contributed 1,903 transition episodes, compared with 399 episodes among those older than 45. Respondents with a high school education or less contributed 1,792 episodes, while those with education above high school contributed 510 episodes. These counts provide information on the amount of within-person employment-status variation underlying the subgroup analyses. Empirical strategy Pooled OLS and fixed-effects models We begin by estimating pooled ordinary least squares (OLS) models to examine the association between informal employment and self-rated health:(1)Healthit=βXit+Witγ+εitwhere Healthit denotes the self-rated health of individual i in year t, Xit is a binary indicator of informal employment, Wit is a vector of time-varying control variables, and εit is an idiosyncratic error term. One concern with pooled OLS models is endogeneity, particularly due to reverse causality and omitted-variable bias. Individuals in poorer health may be more likely to enter or remain in informal employment because of reduced work capacity or barriers to stable jobs (Berkowitz & Johnson, 1974; Cai & Kalb, 2006; Olayiwola et al., 2018). In addition, unobserved individual characteristics such as preferences for flexible work, resilience, or family support may be associated with both employment status and health. As a result, pooled estimates may reflect both the association of informal employment with health and selection into informal work. To reduce bias from time-invariant unobserved heterogeneity, we next estimate individual fixed-effects models of the following form:(2)Healthit=βXit+Witγ+λi+μt+δj(i,t)+εit,where λi denotes individual fixed effects, μt denotes survey-year fixed effects, and δj(i,t) denotes industry fixed effects. The coefficient β captures the average within-individual association between informal employment and self-rated health after accounting for observed time-varying covariates and time-invariant individual characteristics. Standard errors are clustered at the individual level to account for serial correlation over time. Asymmetric fixed-effects model The standard fixed-effects model imposes a symmetry restriction by using a single within-individual coefficient for informal employment, implicitly assuming that entering and exiting informal employment have equal but opposite associations with health. As we expect that these transitions may have different health implications, we extend the baseline model using an asymmetric fixed-effects specification following Allison (2019). We first define the within-individual change in employment status as ΔXit=Xit−Xi,t−1, where Xit is a binary indicator equal to 1 if individual i is informally employed in year t and 0 otherwise. We then decompose this change into two non-overlapping transition indicators:(3)Xit+=max(Xit−Xi,t−1,0)(4)Xit−=max(Xi,t−1−Xit,0) Because Xit is binary, Xit+ equals 1 only when an individual moves from formal to informal employment between t−1 and t, and 0 otherwise. By contrast, Xit− equals 1 only when an individual moves from informal to formal employment. For the first observed wave, both transition indicators are coded as 0 because employment status in the previous wave is not observed. We next define the cumulative number of transitions during the observation period:(5)Zit+=∑s=1tXis+(6)Zit−=∑s=1tXis−Zit+ is the cumulative number of times an individual enters informal employment, and Zit− is the cumulative number of times an individual enters formal employment. We then employ the mean-deviation method to estimate the asymmetric effects:(7)Healthit=β+Zit++β−Zit−+Witγ+λi+μt+δj(i,t)+εit According to the data-generating model proposed by Allison (2019), the mean-deviation method is mathematically equivalent to the first-difference method, and it is more robust in handling unbalanced panel data and missing values. In Eq. (7), β+ captures the association between entering informal employment and changes in self-rated health, whereas β− captures the association between exiting informal employment and changes in self-rated health. To assess asymmetry, we use Wald tests to examine whether the adverse association of entering informal employment is larger in absolute value than the corresponding association of exiting it. Rejection of the null is interpreted as evidence that transitions into and out of informal employment are asymmetrically associated with changes in self-rated health. As in the baseline specification, we include survey-year effects, time-varying controls, and changes in industry affiliation. Standard errors are clustered at the individual level. Event-study and segmented-regression analyses To further examine the temporal pattern of the association between employment transitions and health, we complement the asymmetric fixed-effects model with two additional analyses. First, to test whether health changes occur prior to the transition into or out of informal employment, we use the following event-study specification:(8)Healthit=α+∑k≠−1πkDi×1[EventTimeit=k]+Witγ+λi+μt+δj(i,t)+εit Here, Healthit is self-rated health for individual i in wave t; Di identifies members of the transition group; and EventTimeit denotes time relative to the observed transition wave, with 0 indicating the wave in which the employment-status change is observed. Because CFPS interviews are conducted every two years, event time is measured in roughly two-year intervals. For instance, if an individual moves from formal to informal employment in 2016, the recentered event times for 2014, 2016, and 2018 survey waves are −1, 0, 1, respectively; similarly, for an individual who moves from informal to formal employment in 2018, the recentered times for 2016, 2018, and 2020 waves are −1, 0, 1, respectively. The survey wave immediately preceding the wave in which the employment transition is first observed, approximately two years earlier, is used as the reference category (k=−1). For entry into informal employment, we compare individuals who move from formal to informal employment with individuals who remain formally employed throughout the observation period. For exit from informal employment, we compare individuals who move from informal to formal employment with individuals who remain informally employed throughout. We then estimate models including event-time indicators, individual fixed effects, year fixed effects, industry fixed effects, and the same time-varying controls used in the main analysis. Standard errors are clustered at the individual level. Under this design, the coefficients πk represent differences in self-rated health between the transition group and the stable comparison group at each relative time point. Specifically, the coefficients on the lead indicators provide evidence on whether health was already changing before the transition, whereas coefficients on the contemporaneous and lag indicators characterize the timing and persistence of health changes following the transition. If there are no significant health changes before the transition, πk (k<0) should be statistically insignificant and close to zero. Second, we estimate the dynamic associations of transition into or out of informal employment using the following segmented regression model:(9)Healthit=μt+θ1Postit+θ2Afterit+Witγ+λi+δj(i,t)+εit In this specification, Healthit denotes the self-rated health of individual i in year t. Postit is an indicator that equals 0 before the employment transition and 1 in the observed transition interval and all subsequent periods, so its coefficient θ1 captures the change in health during the observed transition interval. Afterit equals 0 before the transition and counts the number of periods since the transition afterward; accordingly, θ2 captures the post-transition slope in health. The segmented-regression model and the event-study model differ in the reference periods used. The segmented regression model utilizes all survey waves prior to employment transition as reference periods, while the event-study model uses the single wave prior to the employment transition as the reference period. All other terms are defined as in Equation (2). We estimate this model separately for transitions from formal to informal employment and for transitions from informal to formal employment. In this way, the segmented regression analysis provides a complementary assessment of asymmetry by showing whether one type of transition is associated with a larger change in health during the observed transition interval, a steeper post-transition trend, or both. Results Baseline results Table 3 reports the estimated associations between informal employment and self-rated health. In Column (1), the pooled OLS model shows that informal employment is negatively associated with self-rated health (β = −0.039, p < 0.05). In Column (2), after adjusting for individual fixed effects as well as industry and year fixed effects, the coefficient remains negative and is slightly larger in magnitude, but it is imprecisely estimated and does not reach statistical significance at the 5% level (β = −0.041, p > 0.05).Table 3Baseline results.Table 3(1)(2)(3)Self-rated healthSelf-rated healthSelf-rated healthPooled OLSStandard FEAsymmetric FEInformal employment−0.039*−0.041[0.019][0.024]Transitioning into informal employment−0.078*[0.032]Transitioning out of informal employment0.022[0.032]Age−0.015***--[0.001]--Gender0.166***--[0.019]--Single0.111***0.0260.025[0.029][0.050][0.051]Education: High school and above−0.056*−0.117*−0.117*[0.023][0.057][0.057]Chronic disease−0.868***−0.369***−0.369***[0.028][0.029][0.029]Non-agricultural hukou−0.068***−0.062−0.061[0.020][0.042][0.042]Observations19,48919,48919,489Overall R20.1030.6450.645Within R2-0.0150.015Individual FENoYesYesIndustry FENoYesYesYear FENoYesYesProvince by Year FENoNoNoNotes: The dependent variable in each column is the individual's self-rated health. Control variables include an individual's gender, age, marital status, educational level, presence of chronic disease, as well as non-agricultural hukou status. Age is collinear with the individual and year fixed effects, while gender is time invariant; both are therefore omitted from Columns (2) and (3). Overall R2 is reported for all models, and within R2 is additionally reported for the fixed-effects models. The standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *,p < 0.05. Table 3Baseline results.Table 3(1)(2)(3)Self-rated healthSelf-rated healthSelf-rated healthPooled OLSStandard FEAsymmetric FEInformal employment−0.039*−0.041[0.019][0.024]Transitioning into informal employment−0.078*[0.032]Transitioning out of informal employment0.022[0.032]Age−0.015***--[0.001]--Gender0.166***--[0.019]--Single0.111***0.0260.025[0.029][0.050][0.051]Education: High school and above−0.056*−0.117*−0.117*[0.023][0.057][0.057]Chronic disease−0.868***−0.369***−0.369***[0.028][0.029][0.029]Non-agricultural hukou−0.068***−0.062−0.061[0.020][0.042][0.042]Observations19,48919,48919,489Overall R20.1030.6450.645Within R2-0.0150.015Individual FENoYesYesIndustry FENoYesYesYear FENoYesYesProvince by Year FENoNoNoNotes: The dependent variable in each column is the individual's self-rated health. Control variables include an individual's gender, age, marital status, educational level, presence of chronic disease, as well as non-agricultural hukou status. Age is collinear with the individual and year fixed effects, while gender is time invariant; both are therefore omitted from Columns (2) and (3). Overall R2 is reported for all models, and within R2 is additionally reported for the fixed-effects models. The standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *,p < 0.05. Baseline results. Notes: The dependent variable in each column is the individual's self-rated health. Control variables include an individual's gender, age, marital status, educational level, presence of chronic disease, as well as non-agricultural hukou status. Age is collinear with the individual and year fixed effects, while gender is time invariant; both are therefore omitted from Columns (2) and (3). Overall R2 is reported for all models, and within R2 is additionally reported for the fixed-effects models. The standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *,p < 0.05. Column (3) presents the asymmetric fixed-effects model. Transitioning into informal employment is associated with a statistically significant decline in self-rated health (β = −0.078, p < 0.05). The coefficient of −0.078 corresponds to approximately 0.07 standard deviations based on the sample standard deviation of self-rated health. The standardized magnitude is comparable to that reported by Park et al. (2024) for the increase in life satisfaction associated with exiting precarious employment and falls within the range of cross-sectional associations between precarious employment and self-rated health reported by Ling et al. (2024). Transitioning out of informal employment is associated with a small and statistically insignificant increase in self-rated health (β = 0.022, p > 0.05). The within R2 is 0.015 in both fixed-effects specifications, indicating that the included time-varying regressors explain approximately 1.5% of the within-individual variation in self-rated health. Although the Wald test does not provide statistically significant evidence at the 5% level that the adverse association of entering informal employment is larger in absolute magnitude than the corresponding association of exiting it, the magnitude of this asymmetry is comparable to that reported in prior studies (Park et al., 2024). The estimated differences here are less precise, likely reflecting differences in sample size. These findings provide suggestive evidence that transitions into and out of informal employment are not symmetrically associated with self-rated health, with entry into informal employment associated with worse health and exit from informal employment not associated with a comparably sized improvement. Robustness checks Alternative samples and specifications. Table 4 examines whether the entry-related decline in self-rated health is robust to alternative sample restrictions and model specifications. Restricting the sample to respondents without chronic conditions at baseline or to those reporting fair or better self-rated health at baseline yields a similar pattern. Across these specifications, transitioning into informal employment remains negatively associated with self-rated health, whereas the estimated association for transitioning out of informal employment remains smaller and statistically indistinguishable from zero. The same pattern is observed after further adjusting for province-by-year fixed effects to account for time-varying contextual factors at the provincial level. Overall, these results suggest that the adverse association of entering informal employment with self-rated health is not driven solely by poor baseline health, pre-existing illness, or unobserved provincial shocks.Table 4Robustness checks with alternative sample restrictions and model specifications.Table 4(1)(2)(3)Self-rated healthSelf-rated healthSelf-rated healthNo chronic conditions at baselineSelf-rated health at baseline: fair or betterAccounting for time-varying contextual factorsTransitioning into informal employment−0.072*−0.066*−0.070*[0.034][0.033][0.032]Transitioning out of informal employment0.0290.0190.018[0.033][0.032][0.032]Single0.022−0.0220.032[0.053][0.051][0.050]Education: High school and above−0.131*−0.092−0.117*[0.061][0.061][0.057]Chronic disease−0.439***−0.336***−0.365***[0.038][0.033][0.029]Non-agricultural hukou−0.083+−0.063−0.061[0.044][0.044][0.042]Observations17,51416,23119,489Overall R20.6240.5940.649Within R20.0140.0120.015Individual FEYesYesYesIndustry FEYesYesYesYear FEYesYesYesProvince by Year FENoNoYesNotes: The dependent variable in each column is the individual's self-rated health. All specifications are estimated using the asymmetric fixed-effects model. Control variables include an individual's gender, age, marital status, educational level, presence of chronic disease, as well as non-agricultural hukou status. Standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *, p < 0.05; +, p < 0.1. Table 4Robustness checks with alternative sample restrictions and model specifications.Table 4(1)(2)(3)Self-rated healthSelf-rated healthSelf-rated healthNo chronic conditions at baselineSelf-rated health at baseline: fair or betterAccounting for time-varying contextual factorsTransitioning into informal employment−0.072*−0.066*−0.070*[0.034][0.033][0.032]Transitioning out of informal employment0.0290.0190.018[0.033][0.032][0.032]Single0.022−0.0220.032[0.053][0.051][0.050]Education: High school and above−0.131*−0.092−0.117*[0.061][0.061][0.057]Chronic disease−0.439***−0.336***−0.365***[0.038][0.033][0.029]Non-agricultural hukou−0.083+−0.063−0.061[0.044][0.044][0.042]Observations17,51416,23119,489Overall R20.6240.5940.649Within R20.0140.0120.015Individual FEYesYesYesIndustry FEYesYesYesYear FEYesYesYesProvince by Year FENoNoYesNotes: The dependent variable in each column is the individual's self-rated health. All specifications are estimated using the asymmetric fixed-effects model. Control variables include an individual's gender, age, marital status, educational level, presence of chronic disease, as well as non-agricultural hukou status. Standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *, p < 0.05; +, p < 0.1. Robustness checks with alternative sample restrictions and model specifications. Notes: The dependent variable in each column is the individual's self-rated health. All specifications are estimated using the asymmetric fixed-effects model. Control variables include an individual's gender, age, marital status, educational level, presence of chronic disease, as well as non-agricultural hukou status. Standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *, p < 0.05; +, p < 0.1. Event-study analyses. To examine whether health was already changing prior to the employment transition, Fig. 1 presents the event-study estimates for transitions into and out of informal employment. Because CFPS is conducted every two years, event time should be interpreted in biennial survey-wave intervals rather than as the exact calendar timing of the employment transition. For instance, event time 0 refers to the survey wave in which a change in employment status is observed relative to the previous wave. In Panel A, the treated group consists of individuals who move from formal to informal employment, and the comparison group consists of individuals who remain formally employed. The coefficients in the pre-transition periods are small and statistically indistinguishable from zero, providing little evidence that those who later enter informal employment were already experiencing worse health trajectories than those who remained formally employed. This absence of an apparent pre-transition decline strengthens the interpretation that the main results are not simply capturing health deterioration already in motion before the employment change. By contrast, the estimates shift downward around the observed transition wave, indicating a deterioration in self-rated health during the two-year interval in which individuals are observed to move into informal employment. The post-transition coefficients are imprecisely estimated and not statistically distinguishable from zero. This dynamic pattern supports the interpretation that the entry-related decline is not simply a continuation of an earlier differential health trend.Fig. 1Event-study estimates around transitions into and out of informal employment.Notes: Event time is measured in CFPS survey-wave intervals, with each one-unit increase corresponding to approximately two years. Thus, the interval from event time 0 to 1 represents the two years between consecutive survey waves. Event time 0 denotes the wave in which a change in employment status is first observed relative to the preceding wave, rather than the exact calendar timing of the transition. The coefficients should therefore be interpreted as changes observed around the two-year interval in which the transition occurred. Event-study estimates around transitions into and out of informal employment. Notes: Event time is measured in CFPS survey-wave intervals, with each one-unit increase corresponding to approximately two years. Thus, the interval from event time 0 to 1 represents the two years between consecutive survey waves. Event time 0 denotes the wave in which a change in employment status is first observed relative to the preceding wave, rather than the exact calendar timing of the transition. The coefficients should therefore be interpreted as changes observed around the two-year interval in which the transition occurred. Panel B presents the corresponding event-study estimates for transitions from informal to formal employment, using individuals who remain in informal employment as the comparison group. The pre-transition coefficients are close to zero and statistically insignificant, suggesting no clear evidence of divergent health trajectories prior to transition. After the move into formal employment, the coefficient at event time 0 is slightly positive but statistically insignificant, and the subsequent post-transition estimates also have confidence intervals that generally overlap zero. Thus, unlike the pattern observed for entry into informal employment, the transition out of informal employment is not followed by a comparably clear shift in health around the observed transition wave. Nor is there evidence of a more gradual post-transition recovery in the subsequent observed survey waves. This dynamic pattern is consistent with the asymmetric fixed-effects estimates, which likewise show no statistically significant health gain of similar magnitude for transitions out of informal employment. Overall, the event-study results provide two pieces of supporting evidence. First, the absence of pre-transition differences in either panel reduces concern that the baseline results are driven primarily by pre-existing health trajectories. Second, the dynamic estimates indicate that the adverse association of entering informal employment is concentrated around the observed transition interval, whereas any improvement associated with exiting informal employment appears weaker, less clearly observed around the transition wave, and less consistent over time. Assessing sensitivity to omitted-variable bias. The fixed-effects specifications help account for time-invariant unobserved heterogeneity, while the event-study analyses reduce concerns that the results are driven by differential pre-transition health trajectories. However, these analyses cannot fully rule out bias from unobserved time-varying factors. As an additional sensitivity check, we therefore apply Oster's (2019) coefficient-stability approach to assess how strong selection on unobserved variables would have to be, relative to selection on observed covariates, to explain away the estimated association between employment transitions and self-rated health. This approach uses the change in the focal coefficient and the change in model fit after adding observed controls to infer how much selection on unobserved factors would be needed to reduce the estimated coefficient to zero. Specifically, the method requires specifying an upper-bound value of model fit, denoted as Rmax , which represents the hypothetical R2 that would be obtained if both observed and unobserved determinants of the outcome were included in the model. Following common practice, we set Rmax to 1.3 times the R2 from the fully controlled model. We then calculate δ, which indicates how strong selection on unobserved factors would have to be, relative to selection on observed covariates, to explain away the estimated association. For the informal-employment indicator in the standard fixed-effects model, the implied δ is 14.062, suggesting that selection on unobserved factors would need to be substantially stronger (more than 14 times) than selection on observed covariates to explain away the estimate. More importantly, for transitions from formal to informal employment, δ is 2.074. This indicates that selection on unobserved factors would need to be more than twice as strong as selection on observed covariates to reduce the estimated association to zero. The negative association between entering informal employment and self-rated health therefore appears unlikely to be highly sensitive to omitted variable bias under the proportional-selection assumption. For transitions from informal to formal employment, δ is negative (−0.668), indicating that unobserved selection would need to operate in the opposite direction to observed selection to move the estimate to zero. Considering that the baseline estimate for exiting informal employment is already small and statistically indistinguishable from zero, this test is consistent with the main finding that exiting informal employment is not associated with a clear or comparably sized improvement in self-rated health. Sensitivity analyses using alternative definitions of informal employment. Our primary measure combines employment without a written contract and labor-dispatch employment, although these employment types may differ in legal status, employment stability, autonomy, earnings, and access to social protection. We therefore examined whether the results are sensitive to distinguishing the two arrangements and applying narrower sample definitions. Table 5 reports these analyses. In Column (1), employment without a written contract and labor-dispatch employment are distinguished within the same model. Transitioning from formal employment into employment without a written contract is associated with a decline in self-rated health (β = −0.092, p < 0.05). The estimate for transitioning from formal employment into labor-dispatch employment is also negative and similar in magnitude (β = −0.075), although it is less precisely estimated and does not reach the 5% significance level. A Wald test indicates that the two entry coefficients do not differ significantly, providing no clear evidence that their associations with self-rated health differ in magnitude. The corresponding estimates for transitions from either employment type back into formal employment are positive but statistically insignificant. The results remain similar in the restricted samples. After excluding individuals ever observed in labor-dispatch employment, entry from formal employment into employment without a written contract remains negatively associated with self-rated health (β = −0.098, p < 0.05). Among individuals never observed in employment without a written contract, the estimate for entry from formal employment into labor-dispatch employment remains negative (β = −0.076) but statistically insignificant at the 5% level. These results indicate that the adverse entry-related health association is clearest for employment without a written contract, while the labor-dispatch estimates point in the same direction and do not differ significantly in magnitude.Table 5Sensitivity analyses using alternative definitions of informal employment.Table 5(1)(2)(3)Self-rated healthTransition into employment without a written contract−0.092*−0.098*[0.040][0.040]Transition out of employment without a written contract0.0200.015[0.035][0.035]Transition into labor-dispatch employment−0.075+−0.076+[0.046][0.046]Transition out of labor-dispatch employment0.0620.061[0.050][0.051]Observations17,08415,41510,499Individuals5,9755,5263,621Overall R20.6400.6460.634Within R20.0180.0180.016Time-varying controlsYesYesYesIndividual FEYesYesYesIndustry FEYesYesYesYear FEYesYesYesSample restrictionExcludes individuals observed in both no-contract and labor-dispatch employmentExcludes individuals ever observed in labor-dispatch employmentExcludes individuals ever observed in no-contract employmentNotes: Column (1) distinguishes transitions between formal employment and employment without a written contract from transitions between direct formal employment and labor-dispatch employment. Individuals observed in both no-contract and labor-dispatch employment are excluded. Column (2) excludes individuals ever observed in labor-dispatch employment, while Column (3) excludes individuals ever observed in employment without a written contract. All models include individual, industry, and year fixed effects and the same time-varying controls as the baseline specification. Standard errors clustered at the individual level are reported in brackets. ***, p < 0.001; **, p < 0.01; *, p < 0.05; +, p < 0.1. Table 5Sensitivity analyses using alternative definitions of informal employment.Table 5(1)(2)(3)Self-rated healthTransition into employment without a written contract−0.092*−0.098*[0.040][0.040]Transition out of employment without a written contract0.0200.015[0.035][0.035]Transition into labor-dispatch employment−0.075+−0.076+[0.046][0.046]Transition out of labor-dispatch employment0.0620.061[0.050][0.051]Observations17,08415,41510,499Individuals5,9755,5263,621Overall R20.6400.6460.634Within R20.0180.0180.016Time-varying controlsYesYesYesIndividual FEYesYesYesIndustry FEYesYesYesYear FEYesYesYesSample restrictionExcludes individuals observed in both no-contract and labor-dispatch employmentExcludes individuals ever observed in labor-dispatch employmentExcludes individuals ever observed in no-contract employmentNotes: Column (1) distinguishes transitions between formal employment and employment without a written contract from transitions between direct formal employment and labor-dispatch employment. Individuals observed in both no-contract and labor-dispatch employment are excluded. Column (2) excludes individuals ever observed in labor-dispatch employment, while Column (3) excludes individuals ever observed in employment without a written contract. All models include individual, industry, and year fixed effects and the same time-varying controls as the baseline specification. Standard errors clustered at the individual level are reported in brackets. ***, p < 0.001; **, p < 0.01; *, p < 0.05; +, p < 0.1. Sensitivity analyses using alternative definitions of informal employment. Notes: Column (1) distinguishes transitions between formal employment and employment without a written contract from transitions between direct formal employment and labor-dispatch employment. Individuals observed in both no-contract and labor-dispatch employment are excluded. Column (2) excludes individuals ever observed in labor-dispatch employment, while Column (3) excludes individuals ever observed in employment without a written contract. All models include individual, industry, and year fixed effects and the same time-varying controls as the baseline specification. Standard errors clustered at the individual level are reported in brackets. ***, p < 0.001; **, p < 0.01; *, p < 0.05; +, p < 0.1. Dynamic evidence on health changes around employment transitions In this section, we examine whether health changes are concentrated around the observed transition interval or instead unfold across subsequent observed survey waves. Because CFPS is conducted biennially, the transition indicator captures the change in self-rated health during the observed two-year interval in which the transition is recorded, while the post-transition trend captures changes across subsequent observed waves. The segmented-regression results, reported in Table 6, show the temporal pattern of health change surrounding employment transitions. For individuals moving from formal to informal employment, the coefficient on the transition indicator is negative and statistically significant (β = −0.073, p < 0.05), indicating a decline in self-rated health during the observed transition interval. By contrast, the post-transition time trend is small and statistically insignificant (β = 0.017, p > 0.05), suggesting no clear evidence of continuing deterioration across subsequent observed waves. For transitions from informal to formal employment, neither the change around the observed transition interval (β = 0.028, p > 0.05) nor the post-transition trend reaches the 5% significance level (β = −0.027, p > 0.05). These results complement the event-study evidence by showing clearer evidence of health decline around entry into informal employment than of health recovery after exit.Table 6Segmented regression estimates of changes in self-rated health around transitions into and out of informal employment.Table 6(1)(2)Self-rated healthFrom formal to informal employmentFrom informal to formal employmentChange during the observed transition interval−0.073*0.028[0.037][0.039]Post-transition trend0.017−0.027[0.020][0.019]Single0.067−0.065[0.072][0.071]Education: High school and above−0.122−0.144[0.082][0.094]Chronic disease−0.345***−0.399***[0.040][0.042]Non-agricultural hukou−0.099−0.084[0.066][0.055]Observations9,30910,807Overall R20.6450.661Within R20.0160.017Individual FEYesYesIndustry FEYesYesYear FEYesYesNotes: The dependent variable in each column is the individual's self-rated health. All models include individual, industry, and year fixed effects and the same time-varying controls as the baseline specification. The standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *, p < 0.05. Table 6Segmented regression estimates of changes in self-rated health around transitions into and out of informal employment.Table 6(1)(2)Self-rated healthFrom formal to informal employmentFrom informal to formal employmentChange during the observed transition interval−0.073*0.028[0.037][0.039]Post-transition trend0.017−0.027[0.020][0.019]Single0.067−0.065[0.072][0.071]Education: High school and above−0.122−0.144[0.082][0.094]Chronic disease−0.345***−0.399***[0.040][0.042]Non-agricultural hukou−0.099−0.084[0.066][0.055]Observations9,30910,807Overall R20.6450.661Within R20.0160.017Individual FEYesYesIndustry FEYesYesYear FEYesYesNotes: The dependent variable in each column is the individual's self-rated health. All models include individual, industry, and year fixed effects and the same time-varying controls as the baseline specification. The standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *, p < 0.05. Segmented regression estimates of changes in self-rated health around transitions into and out of informal employment. Notes: The dependent variable in each column is the individual's self-rated health. All models include individual, industry, and year fixed effects and the same time-varying controls as the baseline specification. The standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *, p < 0.05. Changes in employment conditions and social insurance protection Table 7 complements the main health results by showing that transitions into and out of informal employment are accompanied by broad changes in job quality and social insurance protection. Entry into informal employment coincides with declines in earnings, cash benefits, perceived career prospects, and coverage by pension, medical, unemployment, and work injury insurance. Exit to formal employment is associated with improvements in these same domains, along with a modest increase in job satisfaction. The fact that these improvements upon exit are not matched by a similarly sized change in self-rated health underscores that changes in employment conditions and changes in health do not unfold in parallel.Table 7Changes in employment conditions and social insurance protection associated with transitions into and out of informal employment.Table 7Panel A: Employment conditions and perceived job qualityLog annual earningsReceipt of any cash benefitsPerceived career prospectsJob satisfactionTransitioning into informal employment−0.283***−0.097***−0.058***−0.014[0.077][0.022][0.015][0.027]Transitioning out of informal employment0.344***0.064**0.078***0.065*[0.076][0.020][0.016][0.028]Time-varying controlsYesYesYesYesIndividual FEYesYesYesYesIndustry FEYesYesYesYesYear FEYesYesYesYesObservations10,14110,12917,36219,487Overall R20.5600.6070.5900.575Within R20.0090.0060.0060.001Panel B: Social insurance protectionPension insuranceMedical insuranceUnemployment insuranceWork injury insuranceTransitioning into informal employment−0.159***−0.157***−0.162***−0.178***[0.018][0.019][0.019][0.020]Transitioning out of informal employment0.225***0.197***0.233***0.214***[0.018][0.019][0.019][0.019]Time-varying controlsYesYesYesYesIndividual FEYesYesYesYesIndustry FEYesYesYesYesYear FEYesYesYesYesObservations12,12412,22810,38111,101Overall R20.7830.7660.8420.777Within R20.0410.0310.0610.043Notes: All specifications are estimated using asymmetric FE, including individual, industry, and year fixed effects and the same time-varying control variables as the baseline specification. The standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *, p < 0.05. Table 7Changes in employment conditions and social insurance protection associated with transitions into and out of informal employment.Table 7Panel A: Employment conditions and perceived job qualityLog annual earningsReceipt of any cash benefitsPerceived career prospectsJob satisfactionTransitioning into informal employment−0.283***−0.097***−0.058***−0.014[0.077][0.022][0.015][0.027]Transitioning out of informal employment0.344***0.064**0.078***0.065*[0.076][0.020][0.016][0.028]Time-varying controlsYesYesYesYesIndividual FEYesYesYesYesIndustry FEYesYesYesYesYear FEYesYesYesYesObservations10,14110,12917,36219,487Overall R20.5600.6070.5900.575Within R20.0090.0060.0060.001Panel B: Social insurance protectionPension insuranceMedical insuranceUnemployment insuranceWork injury insuranceTransitioning into informal employment−0.159***−0.157***−0.162***−0.178***[0.018][0.019][0.019][0.020]Transitioning out of informal employment0.225***0.197***0.233***0.214***[0.018][0.019][0.019][0.019]Time-varying controlsYesYesYesYesIndividual FEYesYesYesYesIndustry FEYesYesYesYesYear FEYesYesYesYesObservations12,12412,22810,38111,101Overall R20.7830.7660.8420.777Within R20.0410.0310.0610.043Notes: All specifications are estimated using asymmetric FE, including individual, industry, and year fixed effects and the same time-varying control variables as the baseline specification. The standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *, p < 0.05. Changes in employment conditions and social insurance protection associated with transitions into and out of informal employment. Notes: All specifications are estimated using asymmetric FE, including individual, industry, and year fixed effects and the same time-varying control variables as the baseline specification. The standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *, p < 0.05. Heterogeneity by age and educational attainment We next examined whether the entry and exit associations between employment transitions and self-rated health varied across groups differing in social and economic vulnerability. Results are presented in Table 8.Table 8Heterogeneity by age and educational attainment.Table 8(1)(2)(3)(4)AgeEducational attainment≤45>45High school education or lessMore than high school educationTransitioning into informal employment−0.057−0.169*−0.085*−0.064[0.035][0.077][0.036][0.065]Transitioning out of informal employment0.031−0.0140.0170.023[0.034][0.087][0.038][0.058]Time-varying controlsYesYesYesYesIndividual FEYesYesYesYesIndustry FEYesYesYesYesYear FEYesYesYesYesObservations15,4694,02014,4944,995Overall R20.6380.6620.6480.640Within R20.0160.0170.0160.016Notes: All models include individual, industry, and year fixed effects and the same time-varying controls as the baseline specification. The standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *, p < 0.05. Table 8Heterogeneity by age and educational attainment.Table 8(1)(2)(3)(4)AgeEducational attainment≤45>45High school education or lessMore than high school educationTransitioning into informal employment−0.057−0.169*−0.085*−0.064[0.035][0.077][0.036][0.065]Transitioning out of informal employment0.031−0.0140.0170.023[0.034][0.087][0.038][0.058]Time-varying controlsYesYesYesYesIndividual FEYesYesYesYesIndustry FEYesYesYesYesYear FEYesYesYesYesObservations15,4694,02014,4944,995Overall R20.6380.6620.6480.640Within R20.0160.0170.0160.016Notes: All models include individual, industry, and year fixed effects and the same time-varying controls as the baseline specification. The standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *, p < 0.05. Heterogeneity by age and educational attainment. Notes: All models include individual, industry, and year fixed effects and the same time-varying controls as the baseline specification. The standard errors in the brackets are clustered at the individual level. ***, p < 0.001; **, p < 0.01; *, p < 0.05. Although employment-status transitions were less common among individuals older than 45, transitioning into informal employment is associated with a significant decline in self-rated health (β = −0.169, p < 0.05), corresponding to approximately 0.16 standard deviations, indicating a more pronounced association than in the full sample. Transitioning out of informal employment is not associated with a statistically significant improvement (β = −0.014, p > 0.05). The Wald test supports asymmetry in this subgroup at the 5% level (p = 0.047). By contrast, among respondents aged 45 or younger, neither coefficient is statistically significant, and the asymmetry test is not significant (p = 0.279). A related, albeit weaker, pattern is observed by educational attainment. Among respondents with a high school education or less, transitioning into informal employment is associated with worse self-rated health (β = −0.085, p < 0.05), whereas the coefficient for transitioning out of informal employment is not statistically significant (β = 0.017, p > 0.05). Because the Wald test does not reach the 5% significance level (p = 0.085), these results show an entry-related decline but do not establish a statistically significant difference between entry and exit. No comparable pattern is observed among respondents with more than a high school education, for whom neither transition coefficient is statistically significant and the asymmetry test is also not significant (p = 0.300). These subgroup analyses indicate that the adverse association of entering informal employment is more pronounced among older and less educated workers. Among older workers, the Wald test provides statistically significant evidence that the entry and exit associations differ in magnitude. Among less educated workers, the entry coefficient is statistically significant, but the Wald test is not; the evidence therefore supports an entry-related health decline rather than asymmetry between entry and exit. More broadly, these patterns are consistent with the view that the health consequences of employment precarity are socially stratified, with larger adverse associations among workers who may have fewer resources to absorb labor-market shocks. Discussion and conclusion Using nationally representative panel data from China, this study shows that entering informal employment is associated with worse self-rated health, whereas we do not detect a comparably sized improvement following exit from informal employment. The event-study estimates further show no evidence of worse pre-transition health trajectories among those who later enter informal employment, alleviating the concern that the main findings are driven primarily by differential health trends preceding the employment change. The segmented-regression results further indicate that the adverse association of entering informal employment with self-rated health is concentrated around the observed transition interval rather than in a continuing decline across subsequent observed waves. One plausible interpretation is that movement into informal employment involves a decline in employment protection, income predictability, and institutional support, along with greater uncertainty about future work and access to benefits; because self-rated health reflects a broad assessment of physical and psychological well-being, it may be sensitive to these declines in security and working conditions (Bamberger et al., 2012; Benach et al., 2014; Falkenberg et al., 2013; Moscone et al., 2016; Vahtera et al., 2004). The absence of a statistically significant health improvement after exit from informal employment should be interpreted cautiously. It suggests that gains in earnings, benefits, career prospects, job satisfaction, and social insurance coverage do not translate into a clearly detectable improvement in self-rated health in the observed period (Bamberger et al., 2012; Falkenberg et al., 2013; Vahtera et al., 2004), but it does not rule out smaller, delayed, or heterogeneous health benefits of exit. The subgroup analyses further suggest that the adverse association of entering informal employment is socially stratified. The entry-related decline in self-rated health is more pronounced among older working-age respondents and those with lower educational attainment, whereas no statistically significant decline is observed among younger or more highly educated respondents. Evidence that the associations of entry and exit differ in magnitude, however, is statistically significant only among older workers. The findings for older working-age respondents are consistent with life-course research showing that employment trajectories and cumulative work histories are closely linked to later self-rated health and frailty (Lu et al., 2017; McDonough et al., 2017). Entry into informal employment may be especially consequential at older ages because accumulated work stress, declining health reserves, and shorter remaining career horizons can reduce workers’ capacity to absorb adverse employment changes. Poor working conditions have been documented to be associated with later-life frailty, while involuntary job loss among older workers is followed by poorer physical functioning and mental health (Gallo et al., 2000; Kalousova & de Leon, 2015). Recent evidence also indicates that older age can intensify the mental-health risks associated with precarious employment (Baek et al., 2024). This strand of literature suggests that older workers may have fewer physiological and psychological resources with which to offset the loss of employment protection and social insurance associated with entry into informal employment. Age-related labor-market vulnerability may further amplify this association. Older displaced workers face greater difficulties in reemployment and are more likely to experience persistent labor-market disadvantage after job loss (Chan & Huff Stevens, 2001; Heisig & Radl, 2017). The lower frequency of employment-status transitions among older respondents in our sample may similarly reflect reduced labor-market mobility and fewer opportunities to move between employment arrangements. In this context, entry into informal employment may be perceived as less temporary or reversible and more as a signal of declining security, weaker bargaining power, and limited prospects for regaining formal protection. Workplace age discrimination is also associated with poorer mental, self-rated, and occupational health (Marchiondo et al., 2019). Moreover, later-career workers may perceive employment instability differently because work-related goals, perceived time horizons, and opportunities for recovery change with age (Kanfer & Ackerman, 2004). Consistent with this interpretation, perceived employability appears less able to buffer the adverse consequences of job insecurity among older workers (Yeves et al., 2019). Recent asymmetric fixed-effects studies likewise show that employment entry and exit can have distinct implications for life satisfaction and frailty in later life (Nam & Kim, 2026; Park & Kim, 2025). Although these studies examine retirement-age employment transitions rather than informal employment specifically, they reinforce the broader argument that the direction of labor-market transition may be especially consequential in older adulthood. This interpretation is consistent with our finding of a more pronounced decline in self-rated health following entry into informal employment and a statistically significant difference between the entry and exit associations among older workers. The stronger entry-related association among less educated workers is consistent with research linking education to health through employment conditions, economic resources, psychosocial assets, and health-related behaviors (Ross & Wu, 1995). Fundamental-cause theory similarly suggests that education provides access to resources, such as knowledge, income, power, and social connections, that help individuals avoid or mitigate health risks (Phelan et al., 2010). Less educated workers may therefore have fewer alternative job opportunities, weaker bargaining power, and fewer resources with which to absorb movement into informal employment (Baek et al., 2023; Burgard & Lin, 2013; Rönnblad et al., 2019; Schram et al., 2021). These mechanisms may help explain why entry into informal employment is associated with poorer self-rated health among less educated workers. However, because the Wald test does not establish a statistically significant difference between the entry and exit coefficients, this result should be interpreted as evidence of an entry-related health decline rather than educational variation in asymmetry. This study makes four main contributions to the literature. First, the study adds to research on informal employment and health by showing that entry into informal employment is associated with a clear decline in self-rated health, whereas exit from informal employment is not associated with a statistically significant improvement in the full sample. Statistically significant evidence that the entry and exit associations differ in magnitude is clearest among older working-age respondents. Whereas prior research has focused primarily on static comparisons between formal and informal workers or on average associations between employment status and health (Aronsson & Huijts, 2025; Lee & Di Ruggiero, 2022), this study highlights the importance of transition direction. It extends prior longitudinal research by showing that, even within employment, movements between formal and informal work may have different health associations (Gebel & Voβemer, 2014). In doing so, it also extends recent asymmetric transition research on precarious employment and subjective well-being (Park et al., 2024) by showing that the form and timing of asymmetry may vary across outcomes. While prior work suggests that subjective well-being responds more to exit from precarious employment, our findings indicate that self-rated health is more closely tied to deterioration at entry and shows less evident improvement after exit in the observed period. Second, the paper adds a dynamic perspective by showing that the adverse health association of entering informal employment emerges around the transition and is not preceded by clear evidence of differential pre-transition health trajectories. This finding helps clarify when health change occurs in relation to labor-market transitions. Relative to recent asymmetric fixed-effects studies that identify differential associations by transition direction (Park et al., 2024), the present analysis further shows that health deterioration becomes apparent around the observed entry interval. Third, by examining concurrent changes in earnings, benefits, career prospects, job satisfaction, and social insurance coverage, the study clarifies the employment-condition changes that accompany transitions into and out of informal employment and situates the main health findings within broader changes in work quality and social protection. This responds to recent calls for greater attention to the material, psychosocial, and institutional factors linking poor-quality employment to health (Aronsson et al., 2023; Aronsson & Huijts, 2025; Hajat et al., 2024). The finding that improved employment conditions after transitions from informal to formal employment are not accompanied by statistically detectable health gains further highlights that the health costs of entering informal employment may be more apparent around the transition period than the potential health benefits of exiting it, and that recovery may be delayed or constrained even after employment conditions improve. Fourth, the study shows that the adverse association of entering informal employment is more apparent among older and less educated workers. In doing so, the paper further addresses the broader question of how labor-market transitions may contribute to the production and persistence of health inequality (Benach et al., 2014; Hajat et al., 2024; Lee & Di Ruggiero, 2022). These contributions are especially relevant in emerging economies, where informal employment remains widespread and systems of social protection are often fragmented or still evolving. China provides a particularly important setting in which to examine these issues because of the scale of its labor force, the prevalence of movement between formal and informal employment, and the central role of institutional stratification in shaping exposure to insecurity and unequal protection (Ling et al., 2024; Zhou et al., 2024). Evidence from this context therefore broadens current understanding of how labor-market precarity may translate into health inequality under conditions of rapid economic and institutional change. Several limitations should be noted. First, our main measure of informal employment combines employment without a written contract and labor-dispatch employment. The combined measure may conceal variation in the health implications of different employment arrangements. We address this concern through sensitivity analyses using alternative definitions, but more detailed longitudinal information on employment contracts and effective access to workplace protection would permit a more complete assessment. Second, because the analysis is restricted to individuals who are working, it does not capture transitions between employment and non-employment. Excluding such transitions may understate the broader health consequences of unstable labor-market trajectories. Third, although we conducted several robustness checks to reduce concerns about reverse causality and selection bias, unobserved time-varying factors may still affect both employment transitions and health. The estimates should therefore be interpreted as associations rather than causal effects. Moreover, we cannot directly observe whether transitions into or out of informal employment are voluntary or involuntary. Our estimates therefore represent average associations across workers with different reasons for changing employment status. Future research incorporating direct measures of workers’ reasons for transitioning is needed to distinguish these heterogeneous experiences more clearly. In addition, self-rated health captures overall health but does not identify the specific dimension underlying the observed association. Changes following entry or exit may reflect psychological, physical, functional, or chronic health processes, which cannot be distinguished in the present data. Future research using domain-specific health measures is needed to clarify these pathways. Finally, the biennial structure of CFPS limits the temporal precision of the event-study and segmented-regression analyses. Because employment transitions may occur at any point between two survey waves, we cannot identify the exact timing of employment changes or distinguish short-term responses from more gradual changes within the two-year interval. This limitation is especially relevant for interpreting the absence of clear recovery following transitions out of informal employment, as short-term improvements may be difficult to detect with biennial observations. The findings also carry several policy implications. They suggest that transitions into informal employment should be understood not only as labor-market events but also as health-relevant transitions. Because entry into informal employment is associated with worse self-rated health and coincides with declines in earnings, access to benefits, career prospects, and social insurance coverage, policies aimed at reducing avoidable or unwanted transitions into informal employment may have health-protective value. This points to the importance of stronger employment stability, contract enforcement, and protection against arbitrary dismissal. It also highlights the need to make social protection less dependent on employment form by expanding portable and inclusive access to medical insurance, work injury compensation, unemployment protection, and pensions for workers moving across formal and informal jobs. More broadly, the results suggest that reducing health inequalities requires attention not only to whether workers are informally employed, but also to the transition processes through which they enter informal employment and to the unequal health risks these transitions generate (Benach et al., 2014; Hajat et al., 2024; Lee & Di Ruggiero, 2022). Ethical statement The datasets analyzed in the current study, which were collected under the ethical approval (No. IRB00001052-14010) granted by the Peking University Institutional Review Board for the CFPS project, are available through the official CFPS website at https://cfpsdata.pku.edu.cn/#/home following a formal application process. Funding This work was supported by the 10.13039/501100001809National Natural Science Foundation of China (No. 72572059; 72102072) and the Humanity and Social Science Foundation, China Ministry of Education (No. 21YJC630005). CRediT authorship contribution statement Liwen Chen: Conceptualization, Funding acquisition, Methodology, Writing – original draft. Kun Liu: Data curation, Formal analysis, Methodology, Software, Writing – review & editing. Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:Liwen Chen reports financial support was provided by National Natural Science Foundation of China. Liwen Chen reports financial support was provided by Humanity and Social Science Foundation, China Ministry of Education. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.