Psychological Capital and Perceived Stress in Nurses: The Mediating Role of Need for Recovery and Recovery Experiences Authors: Marie Charlotte Mollet, Oulmann Zerhouni, Arnaud Villieux, Corinne Isnard Bagnis, Lucia Romo Published: 2026 ABSTRACT ABSTRACTAimTo examine whether recovery experiences and need for recovery mediate the association between psychological capital and perceived stress in nurses.DesignCross‐sectional online survey.MethodsA total of 184 nurses currently practicing in France completed an anonymous online questionnaire administered in January 2023. Participants completed the French versions of the Psychological Capital Questionnaire, the Recovery Experience Scale, the Need for Recovery Scale, and the Perceived Stress Scale. Two mediation models were estimated using ordinary least squares regression with percentile bootstrap inference for the indirect effects.ResultsPsychological capital was positively associated with recovery experiences and negatively associated with need for recovery. Need for recovery was strongly and positively associated with perceived stress, while psychological capital showed no direct association with perceived stress. The indirect effect of psychological capital on perceived stress through need for recovery was statistically detectable, whereas a complementary indirect effect through recovery experiences was in the expected direction but did not meet the bootstrap criterion for statistical significance.ConclusionIn this cross‐sectional sample, psychological capital was associated with lower perceived stress primarily through its association with reduced need for recovery rather than through a direct association with stress. Interventions intended to protect nurse well‐being should therefore be conceived as combining individual resource‐building with organizational architectures that enable effective recovery.Implication for Nursing PracticePsychological capital should be seen as one element in integrated interventions rather than a stand‐alone solution to nursing stress. Brief recovery‐focused interventions for nurses need to be combined with scheduling practices and organizational policies that protect rest and make recovery feasible, a configuration that appears more promising than psychological capital training alone.Reporting MethodThe study followed the STROBE Statement for the reporting of cross‐sectional studies.No Patient or Public ContributionThis study focused on nurses as study participants. No patient or public stakeholder was involved in the design, conduct, or interpretation of the research, since the research question concerns occupational psychological resources rather than clinical practice or care delivery. Introduction Nurses constitute the largest segment of the global health workforce, and the sustainability of nursing staffing is a recognized public health priority. The World Health Organization projected a shortage of nearly six million nurses by 2020 (WHO 2020). The situation has continued to deteriorate in France, where surveys indicate that approximately 43% of nurses are considering leaving the profession within 5 years (Harris Interactive 2021; OECD 2021). Turnover intentions and early departure from the profession are strongly associated with psychosocial risks and demanding working conditions, including burnout, role conflict, and impaired psychological well‐being (INRS 2023; Liu et al. 2017; Machado et al. 2016). In this context, the identification of psychological resources that may protect nurses from the cumulative effects of occupational stressors has become a central research question. According to the conservation of resources theory (Hobfoll 1989), stress arises when perceived resources are insufficient to meet external or internal demands. These resources may be individual, such as self‐efficacy or resilience, social, such as colleague support, or material, such as time and financial means. The probability of maladaptive outcomes, such as exhaustion or disengagement, increases when these resources are depleted. In recent years, the construct of psychological capital has emerged as a higher‐order positive psychological state composed of hope, self‐efficacy, resilience, and optimism (Luthans et al. 2007). Psychological capital is theorized to support coping by providing motivational and cognitive resources that can be mobilized in the face of work demands. A meta‐analytic body of evidence has linked higher psychological capital to lower burnout (Spence Laschinger and Fida 2014) and lower anxiety (Zhou et al. 2018) in nursing samples, suggesting that psychological capital may attenuate the impact of occupational stressors. The effort‐recovery model (Meijman and Mulder 1998) provides a complementary mechanistic account. According to this model, work demands generate short‐term load reactions that need to be reversed during off‐work time. When recovery is incomplete, residual load accumulates and contributes to chronic strain. Recovery is supported by four behavioural and experiential processes, namely psychological detachment from work, relaxation, mastery experiences, and control over non‐work time (Sonnentag and Fritz 2007). A related construct, need for recovery, captures the subjective state of residual fatigue and the urgency of recuperation between work episodes (van Veldhoven and Broersen 2003). Recent research has emphasized the so‐called recovery paradox, that is, the observation that individuals who need recovery the most are often the least able to engage in restorative experiences (Sonnentag 2018). For nurses exposed to chronic stressors, this paradox underscores the importance of internal resources that may sustain or motivate recovery. Bringing together the conservation of resources framework and the effort‐recovery model, we propose that psychological capital may relate to perceived stress not only through a direct buffering effect, but also, and perhaps primarily, through its association with recovery‐related processes. More specifically, psychological capital may foster richer recovery experiences during off‐work time and may reduce the accumulated need for recovery, which in turn translates into lower perceived stress. Recovery‐related variables are therefore positioned as the intermediate processes through which an internal resource such as psychological capital influences a state‐level outcome such as perceived stress. The two recovery‐related constructs considered here, namely recovery experiences and need for recovery, are conceptually distinct and operate at different levels. Recovery experiences refer to discrete behavioural and experiential episodes that occur during off‐work time, while need for recovery reflects an aggregated state across recent work periods. We therefore position need for recovery as the more proximal mediator of perceived stress, while recovery experiences function as upstream contributors to the level of need for recovery, since the quality of recovery during off‐work time determines whether residual load is reversed or accumulates. The present study examines this configuration in a sample of nurses practicing in France. The French nursing context presents specific structural features that justify a dedicated empirical examination. The public hospital system is characterized by a centralized statutory framework, high administrative load, and chronic workforce understaffing, as documented in recent national reports on occupational risks (INRS 2023). These structural conditions plausibly intensify the role of recovery processes in shaping perceived stress. The choice of the French nursing context is therefore not merely descriptive but provides an informative setting in which the recovery‐related pathways can be examined under conditions of high baseline occupational demand. Research Aims and Hypotheses On the basis of the framework presented above, the present study tests two complementary hypotheses concerning the mediating role of recovery‐related variables in the association between psychological capital and stress‐related outcomes.Hypothesis 1 Psychological capital is associated with lower perceived stress indirectly through reduced need for recovery. Hypothesis 2 Psychological capital is associated with lower need for recovery indirectly through richer recovery experiences. Psychological capital is associated with lower perceived stress indirectly through reduced need for recovery. Psychological capital is associated with lower need for recovery indirectly through richer recovery experiences. Both hypotheses are tested using cross‐sectional mediation models. The resulting estimates are therefore interpreted as describing patterns of statistical association rather than causal relations, in line with current methodological recommendations for mediation analysis in non‐experimental designs (Hayes 2018). This interpretive constraint is maintained throughout the manuscript and is further emphasized in the discussion section. Methods Study Design and Setting This study employed a quantitative, cross‐sectional, online survey design. Data were collected between January 8 and January 29, 2023, that is, over a continuous 22‐day period, in the French care sector, with a focus on currently practicing nurses. The questionnaire was administered through the LimeSurvey platform, with full anonymity for all participants. The survey link was disseminated through personal and professional social networks of the first author, namely Facebook and Instagram, in posts shared in nurse‐related professional groups and on personal pages, with an explicit invitation to forward the link to colleagues. No paid promotion, sponsored content, or institutional mailing was used. The link circulated organically through reposts during the 3‐week recruitment window. The data collection window was deliberately kept short to limit external events, such as seasonal workload variation, from differentially affecting responses across participants. This restricted window and the social‐media distribution channel both constitute methodological constraints that are discussed in the limitations section. Participants Eligibility Criteria and Recruitment Participants were eligible for inclusion in the study if they met all of the following criteria, namely being currently employed as a nurse (registered nurse or specialized nurse), being aged at least 21 years, fluently reading written French, and providing electronic informed consent prior to questionnaire administration. Participants were excluded from the analytical sample if they failed to provide informed consent, did not complete at least one of the four focal scales (Psychological Capital Questionnaire, Recovery Experience Scale, Need for Recovery Scale, Perceived Stress Scale), or reported a profession other than nursing in the screening item. No additional exclusion criteria were applied. Recruitment was conducted through convenience sampling, with the survey link disseminated through personal and professional social networks as described in the Study Design subsection. No incentives, financial or otherwise, were provided for participation. Sample Description Out of 776 initial responses, 592 were excluded due to incompleteness on the focal scales, resulting in a final analytical sample of 184 nurses. This represents a completion rate of approximately 23.7%. The modest completion rate is acknowledged as a methodological constraint and is discussed in the limitations section. The sample was predominantly female. Of the 183 participants who reported their sex, 178 were women (97.3% of those who reported, 96.7% of the total sample) and 5 were men (2.7%). Age was collected in categorical bands. Specifically, 90 participants (48.9%) were in the 21 to 26 year band, 50 (27.2%) in the 27 to 35 year band, 27 (14.7%) in the 36 to 45 year band, and 17 (9.2%) in the 46 years and older band. Years of professional experience were also collected in categorical bands. Specifically, 49 participants (26.6%) had less than 1 year of experience, 56 (30.4%) had between 1 and 5 years, 28 (15.2%) had between 6 and 10 years, and 51 (27.7%) had more than 11 years. With respect to family situation, 133 of the 183 participants who reported their partnership status (72.7% of those who reported, 72.3% of the total sample) reported being in a relationship, and 75 of the 184 participants (40.8%) reported having at least one child. The full demographic profile is presented in Table 1. TABLE 1Sample characteristics (N = 184).VariableCategories n %SexFemale17896.7Male52.7Age band21–26 years9048.927–35 years5027.236–45 years2714.746 years and over179.2Years of professional experienceLess than 1 year4926.6Between 1 and 5 years5630.4Between 6 and 10 years2815.211 years or more5127.7Partnership statusIn a relationship13372.3Single5027.2ChildrenHas at least one child7540.8No children10959.2 Note: Percentages may not sum to 100 due to rounding. Sex was not reported by 1 participant; partnership status was not reported by 1 participant. Percentages are computed on the full sample (N = 184). Age and years of professional experience were collected as categorical bands and are not available as continuous variables. Sample characteristics (N = 184). Note: Percentages may not sum to 100 due to rounding. Sex was not reported by 1 participant; partnership status was not reported by 1 participant. Percentages are computed on the full sample (N = 184). Age and years of professional experience were collected as categorical bands and are not available as continuous variables. Measures and Scoring Procedures All study variables were assessed through previously validated self‐report scales. All scales were administered in their validated French versions. For each scale, the response format, the scoring procedure, the handling of reverse items, and the resulting score range are described below. The full scoring rules are reported in the main text in order to support replication, in addition to the demographic and item‐level information reported in the Supporting Information. Psychological Capital Psychological capital was measured using the French Psychological Capital Questionnaire (F‐PCQ‐24; Choisay et al. 2021), a 24‐item scale assessing the four facets of hope, self‐efficacy, resilience, and optimism through six items each. Responses were recorded on a 6‐point Likert scale ranging from 1 (strongly disagree) to 6 (strongly agree). The overall score was computed as the mean of the 24 items, with higher values indicating higher psychological capital. The scale contains three reverse‐coded items (items 13, 20, and 23) which were rescored as 7 minus the raw value before computing the mean. The possible range of the resulting score is 1 to 6. The internal consistency in the present study was good (Cronbach's alpha = 0.85). Recovery Experiences Recovery experiences were assessed using the 16‐item Recovery Experience Scale developed by Sonnentag and Fritz (2007). The instrument captures four subscales of four items each, namely psychological detachment, relaxation, mastery, and control over non‐work time. Responses were recorded on a 5‐point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). No reverse‐coded items are present in the original instrument. The overall score was computed as the mean of the 16 items, with higher values indicating richer recovery experiences. The possible range of the resulting score is 1 to 5. The French version administered in the present study has subsequently been validated by Le Moal et al. (2024) on samples of French entrepreneurs and agricultural employees. The internal consistency in the present study was good (Cronbach's alpha = 0.86). Need for Recovery Need for recovery was assessed using the 10‐item Need for Recovery Scale developed by van Veldhoven and Broersen (2003), in the French adaptation by Dupret et al. (2018). Each item is answered on a binary yes/no format coded as 1 and 0, respectively. The scale contains three items (items 4, 9, and 10) that are worded positively and require reverse coding before summation, that is, recoded as 1 minus the raw value. The total score was computed by summing the ten binary item scores after reverse coding and multiplying the result by 10, yielding a score ranging from 0 to 100 with higher values indicating greater need for recovery. The internal consistency in the present study was acceptable (Cronbach's alpha = 0.68). The implications of this modest reliability are addressed in the data analysis subsection (sensitivity analyses) and in the limitations section. Perceived Stress Perceived stress was assessed using the 10‐item Perceived Stress Scale (PSS‐10; Cohen et al. 1983), in its French version validated by Bellinghausen et al. (2009) and further examined in occupational samples by Dupret and Bocéréan (2013). The scale measures the frequency of stressful experiences during the past month on a 5‐point Likert scale ranging from 0 (never) to 4 (very often). The PSS‐10 contains four positively worded items (items 4, 5, 7, and 8) that were administered in their standard validated format and were reverse‐coded prior to computing the total score. Reverse‐coding was performed manually in the data preparation file (Excel) by recoding these items as 4 minus the raw value, so that higher recoded scores consistently reflected higher perceived stress. Item 10 was also inspected and treated according to the canonical French scoring rules. The total perceived stress score was then computed as the mean of the 10 recoded items. The possible range of the resulting score is 0 to 4. The internal consistency in the present study was good (Cronbach's alpha = 0.82). Ethical Considerations All procedures followed the relevant national and institutional ethical standards as well as the Declaration of Helsinki (1975, revised 2008). The study did not fall under the scope of article L. 1121‐1 of the French Public Health Code and therefore did not require formal review by a research ethics committee, in accordance with the Jardé law (loi n°2012‐300). The data processing procedures complied with Regulation (EU) 2016/679 (General Data Protection Regulation) and with the data protection framework of the University of Rouen Normandy. As a non‐interventional anonymous online survey that did not collect identifying information, the study did not require the issuance of a specific clearance number by the Data Protection Officer. The processing was nevertheless conducted under the institutional supervision of the Data Protection Officer of the University of Rouen Normandy. Electronic informed consent was obtained from all participants prior to questionnaire administration. The first page of the LimeSurvey questionnaire presented the purpose of the study, the identity and affiliation of the research team, the data processing procedures, the storage duration of the responses, the contact details of the Data Protection Officer, and a clear statement that participation was voluntary and that participants could withdraw at any time without justification and without consequence. To proceed to the questionnaire items, participants had to explicitly check a consent box stating their agreement to participate. No identifying information was collected at any point. Participants could exit the questionnaire by closing the browser window at any time, in which case the partial responses were not retained for analysis. After completion of the questionnaire, withdrawal of data was technically impossible because no identifier linked responses to participants, a procedure that was explicitly indicated in the consent page. Data Analysis All analyses were conducted using Jamovi version 2.5.5, based on R version 4.3. Descriptive statistics (means, standard deviations, frequencies, and percentages) were computed for all sociodemographic variables and focal measures. Bivariate Pearson correlations were computed between the four focal variables, and a simple linear regression was estimated to assess the association between psychological capital and need for recovery. Two mediation models were tested using ordinary least squares regression with 5000 percentile bootstrap resamples to derive confidence intervals around the indirect effects, in line with the recommendations of Hayes (2018) and Preacher and Hayes (2008). The analyses were conducted using the medmod module in Jamovi (Selker et al. 2020), which implements both delta‐method standard errors and bootstrap‐based confidence intervals. This procedure is functionally equivalent to model 4 of the PROCESS macro for SPSS and SAS (Hayes 2018). For each model, we report the a path (predictor to mediator), the b path (mediator to outcome controlling for the predictor), the total effect c, the direct effect c′, and the indirect effect a·b with its 95% bootstrap percentile confidence interval. Standardized coefficients (β) are also reported. The significance level for all statistical tests was set at p < 0.05. Because the design is cross‐sectional, all estimates are interpreted as associations rather than as causal effects, and the term mediation refers to a statistical decomposition of variance rather than to a causal mechanism (Hayes 2018; MacKinnon et al. 2000). Because all variables were assessed through self‐report, a Harman single‐factor test was conducted as a preliminary check for common method variance (Podsakoff et al. 2003). The 60 items composing the four scales were submitted to an unrotated principal component/analysis. The first component accounted for 17.69% of the total variance, which is well below the conventional 50% threshold typically used to flag a substantial common method bias. Sixteen components had an eigenvalue greater than one, and the first ten components together accounted for 57.08% of the variance. Common method bias is therefore unlikely to substantially distort the observed associations, although we acknowledge that Harman's test is a minimal procedure and that more rigorous approaches require a study design that was not implemented here. A post hoc sensitivity power analysis was conducted to characterize the detection capacity of the achieved sample. With N = 184, alpha = 0.05 (two‐tailed), and a target power of 0.80, the minimum detectable bivariate Pearson correlation was |r| = 0.205, and the minimum detectable effect for a regression with one predictor was R 2 = 0.041 (Cohen's f 2 = 0.043). For the indirect effect of psychological capital on perceived stress through need for recovery, we relied on the bootstrap power recommendations of Fritz and MacKinnon (2007). The configuration observed in the present study, that is, a small a path and a large b path, corresponds to an estimated required sample size of approximately 120 for 80% power, which is below our achieved sample. We acknowledge that no a priori power analysis was conducted before data collection. The decision to retain the achieved sample as the analytical sample was driven by the available data collection window and by resource constraints typical of a Master's research project, which is acknowledged as a limitation. The post hoc sensitivity analysis presented above is therefore informative on what could be detected given the achieved sample, but it does not substitute for an a priori power calculation. Sensitivity Analyses Three additional sensitivity analyses were conducted to address potential threats to the validity of the main results. First, a correction for measurement unreliability was applied to the principal associations involving the Need for Recovery Scale, whose internal consistency was modest. The classical attenuation correction formula r_corrected = r_observed/square root of (alpha_X times alpha_Y) was used (Spearman 1904), as well as the analogous correction for regression coefficients based on the reliability of the predictor (Bound et al. 2001). Second, a comparison between pairwise deletion (the approach used in the main analyses) and listwise deletion (case‐complete analysis on all four focal variables) was conducted to assess the sensitivity of the results to the missing‐data treatment. Third, the two mediation models were re‐estimated after entering five demographic covariates (age band, sex, years of professional experience, parenthood, and partnership status) on both the mediator and the outcome equations. The results of all sensitivity analyses are reported in the corresponding subsection of the Results and in the Supporting Information. Missing values were below 2% on all key focal variables. The 184 case‐complete cases for the four focal variables were retained for the main analyses. Results Descriptive Statistics Descriptive statistics for the four focal variables are reported in Table 2. The mean psychological capital score was 3.54 (SD = 0.59) on the 1 to 6 scale, the mean recovery experiences score was 3.21 (SD = 0.63) on the 1 to 5 scale, the mean need for recovery score was 71.58 (SD = 21.83) on the 0 to 100 scale, and the mean perceived stress score was 2.08 (SD = 0.57) on the 0 to 4 scale. TABLE 2Descriptive statistics for the focal study variables (N = 184).Variable M SDMinMax α Possible rangePsychological capital3.540.591.715.130.851–6Recovery experiences3.210.631.314.690.861–5Need for recovery71.5821.8301000.680–100Perceived stress2.080.570.503.400.820–4 Note: For the Need for Recovery Scale, which uses binary items, α corresponds to the Kuder–Richardson 20 coefficient.Abbreviations: α, Cronbach's alpha; M, mean; SD, standard deviation. Descriptive statistics for the focal study variables (N = 184). Note: For the Need for Recovery Scale, which uses binary items, α corresponds to the Kuder–Richardson 20 coefficient. Abbreviations: α, Cronbach's alpha; M, mean; SD, standard deviation. These values can be situated in relation to published reference data. The mean perceived stress score of 2.08 is moderately higher than mean scores reported in general French employee samples assessed with the PSS‐10 (Lesage et al. 2012), which is consistent with the literature on occupational stress in nursing. The mean psychological capital score of 3.54 is at the lower end of the range reported in international nursing samples, where means typically fall between 3.7 and 4.5 on the 6‐point scale (Avey et al. 2011). The mean need for recovery score of 71.58 on the 0 to 100 scale is elevated and exceeds the threshold of 50 frequently used in occupational health practice to identify problematic recovery profiles (van Veldhoven and Broersen 2003). The sample therefore reports a substantial recovery debt at baseline, which is consistent with the structural conditions of nursing in the French public hospital sector documented in recent reports (INRS 2023). Correlations Pearson correlations between the focal variables are presented in Table 3. Psychological capital was positively associated with recovery experiences, r = 0.52, p < 0.001, and negatively associated with need for recovery, r = −0.18, p = 0.014. The association between psychological capital and perceived stress was weak and not statistically significant, r = −0.04, p = 0.622. Recovery experiences were negatively associated with need for recovery, r = −0.22, p = 0.002, and were not significantly associated with perceived stress, r = −0.01, p = 0.882. Need for recovery was strongly associated with perceived stress, r = 0.55, p < 0.001. The full item‐level Pearson correlation matrix is reported in the Supporting Information (item‐level correlation matrix file). TABLE 3Pearson correlations between the focal variables (N = 184).Variable12341. Psychological capital—2. Recovery experiences0.52*** —3. Need for recovery−0.18* −0.22** —4. Perceived stress−0.04−0.010.55*** —* p < 0.05.** p < 0.01.*** p < 0.001. Pearson correlations between the focal variables (N = 184). p < 0.05. p < 0.01. p < 0.001. Linear Regression Predicting Need for Recovery A simple linear regression was estimated to assess the association between psychological capital and need for recovery. Results are reported in Table 4. The model was statistically significant, F(1, 182) = 6.23, p = 0.014, R 2 = 0.033. Psychological capital was negatively associated with need for recovery, b = −6.69, SE = 2.68, t = −2.50, p = 0.014, β = −0.18. The intercept was 95.27 (SE = 9.62, t = 9.90, p < 0.001), consistent with the 0 to 100 range of the need for recovery score. TABLE 4Linear regression of need for recovery on psychological capital (N = 184).Predictor b SE t p Intercept95.279.629.90< 0.001Psychological capital−6.692.68−2.500.014 Note: Dependent variable: need for recovery score (0–100). R 2 = 0.033, F(1, 182) = 6.23, p = 0.014. β = −0.18 for psychological capital. Linear regression of need for recovery on psychological capital (N = 184). Note: Dependent variable: need for recovery score (0–100). R 2 = 0.033, F(1, 182) = 6.23, p = 0.014. β = −0.18 for psychological capital. Mediation Analyses Model 1: Need for Recovery as Mediator of the Psychological Capital and Perceived Stress Association The first mediation model tested whether need for recovery accounted for the association between psychological capital and perceived stress (Hypothesis 1). Results are reported in Table 5. Psychological capital was negatively associated with need for recovery, b = −6.69, SE = 2.68, p = 0.014, β = −0.18. In turn, need for recovery was strongly and positively associated with perceived stress after controlling for psychological capital, b = 0.015, SE = 0.002, p < 0.001, β = 0.56. The total effect of psychological capital on perceived stress was not statistically significant, b = −0.04, SE = 0.07, p = 0.622, nor was the direct effect when need for recovery was included as a mediator, b = 0.06, SE = 0.06, p = 0.304 (Figure 1). TABLE 5Mediation Model 1: Need for recovery as mediator (N = 184).Effect b SE β p 95% bootstrap CI a (PsyCap → NfR)−6.692.68−0.180.014— b (NfR → PSS | PsyCap)0.0150.0020.56< 0.001— c (total: PsyCap → PSS)−0.040.07−0.040.622— c′ (direct: PsyCap → PSS | NfR)0.060.060.060.304— a·b (indirect: PsyCap → NfR → PSS)−0.0980.045−0.10—[−0.187, −0.012] Note: Bootstrap CI based on 5000 percentile resamples.Abbreviations: NfR, need for recovery; PSS, perceived stress; PsyCap, psychological capital. Mediation Model 1: Need for recovery as mediator (N = 184). Note: Bootstrap CI based on 5000 percentile resamples. Abbreviations: NfR, need for recovery; PSS, perceived stress; PsyCap, psychological capital. Mediation models testing the indirect associations between psychological capital and perceived stress via need for recovery (Model 1, H1) and via recovery experiences (Model 2, H2). Unstandardized coefficients (b) and standardized coefficients (β) are reported. Bootstrap 95% CI based on 5000 percentile resamples. *p < 0.05, **p < 0.01, ***p < 0.001. The slight sign reversal between the total effect (−0.04) and the direct effect (+0.06) reflects a small statistical suppression effect, that is, the inclusion of need for recovery in the regression partials out the negative indirect path and leaves a small positive residual association that fluctuates around zero. Such sign reversals are formally described as classical suppression configurations and have been extensively discussed in the methodological literature on mediation (MacKinnon et al. 2000; Tzelgov and Henik 1991). Importantly, neither the total nor the direct effect is statistically significant, and the apparent sign reversal reflects sampling fluctuation around zero rather than a substantive change in the association. The indirect effect was statistically significant under the bootstrap criterion, ab = −0.098, SE = 0.045, 95% bootstrap CI [−0.187, −0.012], β = −0.10. Hypothesis 1 was therefore supported. Because the design is cross‐sectional, this statistical mediation should be read as a pattern of association consistent with the proposed configuration rather than as evidence of a causal mechanism. Model 2: Recovery Experiences as Mediator of the Psychological Capital and Need for Recovery Association The second mediation model tested whether recovery experiences accounted for the association between psychological capital and need for recovery (Hypothesis 2). Results are reported in Table 6. Psychological capital was positively associated with recovery experiences, b = 0.55, SE = 0.07, p < 0.001, β = 0.52. Recovery experiences were in turn negatively associated with need for recovery after controlling for psychological capital, b = −6.23, SE = 2.91, p = 0.034, β = −0.18. The total effect of psychological capital on need for recovery was statistically significant, b = −6.69, SE = 2.68, p = 0.014, while the direct effect was no longer significant when recovery experiences were entered as a mediator, b = −3.28, SE = 3.10, p = 0.292. The point estimate of the indirect effect was of comparable magnitude to that observed in Model 1, ab = −3.42, SE = 1.75, β = −0.09 (Figure 1). However, the 95% bootstrap confidence interval marginally included zero [−6.75, 0.18]. Under the conservative bootstrap criterion, the indirect effect did not reach the conventional threshold of statistical significance. Hypothesis 2 was therefore not supported under the bootstrap criterion. The point estimate is in the predicted direction, but the inferential evidence is weaker than for Model 1 and should be interpreted with caution. We refrain from drawing strong theoretical inferences from this borderline result. The pattern is discussed below in light of the substantial shared variance between psychological capital and recovery experiences. TABLE 6Mediation Model 2: Recovery experiences as mediator (N = 184).Effect b SE β p 95% bootstrap CI a (PsyCap → RES)0.550.070.52< 0.001— b (RES → NfR | PsyCap)−6.232.91−0.180.034— c (total: PsyCap → NfR)−6.692.68−0.180.014— c′ (direct: PsyCap → NfR | RES)−3.283.10−0.090.292— a·b (indirect: PsyCap → RES → NfR)−3.421.75−0.09—[−6.75, 0.18] Note: Bootstrap CI based on 5000 percentile resamples. The 95% bootstrap CI for the indirect effect marginally includes zero, indicating that this indirect effect does not reach the conventional threshold of statistical significance under the bootstrap criterion.Abbreviations: NfR, need for recovery; PsyCap, psychological capital; RES, recovery experiences. Mediation Model 2: Recovery experiences as mediator (N = 184). Note: Bootstrap CI based on 5000 percentile resamples. The 95% bootstrap CI for the indirect effect marginally includes zero, indicating that this indirect effect does not reach the conventional threshold of statistical significance under the bootstrap criterion. Abbreviations: NfR, need for recovery; PsyCap, psychological capital; RES, recovery experiences. Sensitivity Analyses Correction for Measurement Unreliability Because the internal consistency of the Need for Recovery Scale was modest (alpha = 0.68), the principal associations involving this variable may be attenuated by measurement error. The classical attenuation correction formula was applied to estimate the magnitude of this attenuation. For bivariate correlations involving need for recovery, the correction increases the magnitude of the association. Specifically, the corrected correlation between psychological capital and need for recovery moves from r = −0.18 to r = −0.24, the corrected correlation between recovery experiences and need for recovery moves from r = −0.22 to r = −0.30, and the corrected correlation between need for recovery and perceived stress moves from r = 0.55 to r = 0.73. For the regression coefficients, the corrected estimates also increase in magnitude, with the a path moving from b = −6.69 to approximately −7.87 and the b path moving from b = 0.015 to approximately 0.022. Consequently, the corrected indirect effect of psychological capital on perceived stress through need for recovery is approximately −0.18, that is, about 1.80 times larger in magnitude than the observed estimate of −0.098. These results indicate that the modest reliability of the Need for Recovery Scale induces a systematic underestimation of the focal associations and that the observed effects are conservative. Sensitivity to Missing‐Data Treatment Missing values on the four focal variables were below 2% in the present study, and the four‐variable case‐complete sample comprised all 184 participants retained after the initial completion screening. Pairwise deletion (the approach used in the main analyses) and listwise deletion across the four focal variables therefore yield identical analytical samples for the focal correlations and for the mediation models. As expected, the corresponding estimates are strictly identical under the two missing‐data treatments, with the indirect effect of psychological capital on perceived stress through need for recovery estimated at ab = −0.098, 95% bootstrap CI [−0.187, −0.012] under both schemes. The main results are therefore not sensitive to the choice between pairwise and listwise deletion within the four focal variables. Robustness to Demographic Covariates To address concerns about potential confounding by demographic characteristics, both mediation models were re‐estimated after entering age band, sex, years of professional experience band, parenthood, and partnership status as covariates on both the mediator and the outcome equations. The full results are reported in the Supporting Information. The principal effects of Model 1 remained stable. Specifically, the a path was b = −7.27, SE = 2.71, p = 0.008, the b path was b = 0.014, SE = 0.002, p < 0.001, and the indirect effect was ab = −0.103, 95% bootstrap CI [−0.197, −0.020]. For Model 2, the inclusion of covariates yielded an indirect effect of ab = −3.60, 95% bootstrap CI [−6.79, −0.12], which is in the predicted direction and which, in this specification, reached the conventional threshold under the bootstrap criterion. The change should not be overstated, since the lower bound of the bootstrap confidence interval is still close to zero. Overall, the central findings of the study can be interpreted as robust to the demographic configuration of the sample. Discussion The present study examined the associations between psychological capital, recovery‐related variables, and perceived stress in a sample of nurses practicing in France. Three principal patterns emerge. First, psychological capital was not directly associated with perceived stress, although several reports in the international literature describe a direct association. Second, psychological capital was associated with lower need for recovery and with richer recovery experiences. Third, the association between psychological capital and perceived stress was statistically accounted for primarily by need for recovery rather than by a direct buffering effect. A complementary model involving recovery experiences as the mediator of the association between psychological capital and need for recovery yielded a borderline indirect effect under the unadjusted bootstrap criterion. We emphasize at the outset that all these results derive from a cross‐sectional design and therefore describe statistical patterns of association rather than causal relations between psychological capital, recovery processes, and perceived stress. Theoretical Implications These findings position psychological capital, in this sample, as a correlate of resource restoration within the frameworks of the conservation of resources theory (Hobfoll 1989) and recovery research (Sonnentag and Fritz 2007). Individuals with higher psychological capital reported richer recovery experiences, namely detachment, relaxation, mastery, and control, which are usually considered essential for restoring depleted resources and reducing cumulative strain. The strong association between need for recovery and perceived stress suggests that stress, as captured by the PSS‐10, functions in part as a recovery debt in this sample. Psychological capital therefore appears, in our data, to be associated with behaviours and circumstances that reduce this debt, rather than with a direct buffering of the perception of stress. This process‐oriented reading is consistent with the recovery paradox described by Sonnentag (2018), according to which higher job strain simultaneously increases the need for recovery and hinders it. In environments characterized by chronic occupational demands, psychological capital may not be sufficient to reduce perceived stress in the short term, but it may still be associated with better recovery dynamics across time. These interpretations remain at the level of plausible reading of the cross‐sectional data and would need to be examined through longitudinal designs with repeated measurements. The pattern observed for the second mediation model deserves a specific comment. The point estimate of the indirect effect of psychological capital on need for recovery through recovery experiences was in the predicted direction and of a magnitude comparable to that of the first mediation. However, the bootstrap confidence interval marginally included zero in the unadjusted model, and the indirect effect did not reach the conventional threshold. We therefore consider Hypothesis 2 as not statistically supported by the unadjusted analysis and we refrain from drawing strong theoretical inferences from this borderline result. Two elements may account for the observed pattern. First, psychological capital and recovery experiences were strongly correlated in the present sample, r = 0.52, with psychological capital alone accounting for 26.5% of the variance in recovery experiences. This substantial shared variance reduces the statistical separability of the two constructs in the partial‐effect estimation and inflates the standard error of the partial b path coefficient. The borderline indirect effect should therefore not be read as evidence against the conceptual link between psychological capital, recovery experiences, and need for recovery, but it does indicate that the present data do not provide robust support for this specific chain in the unadjusted analysis. Second, this result is also compatible with the possibility that recovery experiences and need for recovery capture partially overlapping processes, namely the behavioural and experiential side of off‐work recovery on the one hand, and the perceived residual fatigue requiring further recovery on the other hand. In contrast, the mediation by need for recovery (Model 1) yielded a more robust indirect effect, which is interpreted as reflecting the more proximal positioning of need for recovery relative to the experience of perceived stress in the proposed framework. Cross‐National Comparisons The pattern observed here, namely a statistically detectable indirect effect through need for recovery combined with a non‐significant direct effect, contrasts with several reports in the international literature in which psychological capital shows a direct association with perceived stress. Several potential explanations may be considered, although they remain speculative in the absence of direct between‐country comparisons within the same study. First, structural differences in the organization of nursing work may modulate the relative weight of direct versus indirect pathways linking psychological capital to perceived stress. The French public hospital sector is characterized by a centralized statutory framework and by chronic workforce understaffing, while other healthcare systems exhibit greater institutional variation. Such variations could plausibly influence the buffering capacity of individual psychological resources, although this conjecture would require direct empirical examination. Second, cross‐cultural differences in non‐work life arrangements may also play a role, since recovery dynamics depend not only on individual resources but also on the availability of supportive contexts outside work, and these contexts vary substantially across countries. Direct cross‐national comparisons within the same study would be needed to test these conjectures rigorously. In the meantime, the absence of a direct effect in our sample should not be interpreted as evidence against psychological capital as a stress‐relevant resource, but rather as an indication that, in our cross‐sectional French nursing context, its statistical association with perceived stress is more visible through recovery‐related variables. Practical Implications The findings suggest possible directions for interventions aimed at protecting nurse well‐being. We must underline that psychological capital should not be positioned as a stand‐alone solution to nursing stress, since such a framing would risk over‐individualizing a systemic problem. The present data, although cross‐sectional and therefore not directly informative on intervention efficacy, support the principle of integrated interventions combining three levels of action. At the individual level, brief structured interventions targeting recovery‐related behaviours during off‐work time may be considered, such as psychological detachment training and structured relaxation protocols, which have shown promising effects in occupational samples (Hahn et al. 2011). These interventions can include training in the deliberate planning of off‐work activities, in micro‐mastery experiences, and in time‐control strategies. Such interventions should be adapted to shift work and acute care constraints, and should specifically emphasize self‐efficacy for recovery behaviours. At the team level, schedule design should protect recovery time between shifts, with particular attention to the avoidance of quick returns of less than 11 h, in line with European occupational health guidelines. Team‐level interventions may also include the protection of breaks during shifts, the establishment of quiet spaces for short recovery episodes, and team autonomy over micro‐scheduling, all of which create the conditions for the individual‐level skills to be deployed effectively. At the organizational level, the integration of routine recovery monitoring through brief instruments such as the Need for Recovery Scale may be considered as part of occupational health surveillance. Such monitoring would allow early identification of nurses or of units at risk of chronic recovery debt and would provide a basis for the targeted allocation of intervention resources. Furthermore, organizational policies that limit off‐shift contact and that establish predictable scheduling can reduce the structural barriers to recovery. In summary, the configuration of “psychological capital plus recovery architecture” is likely to be more effective than psychological capital training alone. Occupational health interventions in nursing should therefore integrate individual resource‐building with the organizational conditions that make recovery feasible. These recommendations remain provisional in the absence of direct intervention evidence in this population and would benefit from rigorous evaluation in pragmatic trials. Limitations and Future Directions Several limitations should be acknowledged. First, the cross‐sectional design precludes causal inference. Although mediation analysis is sometimes interpreted in causal terms, the present results should be understood as describing statistical patterns of association rather than causal pathways. We have therefore systematically used associative language throughout the manuscript. Future studies should adopt longitudinal designs with repeated measurements that can model temporal precedence between psychological capital, recovery experiences, need for recovery, and perceived stress. Second, the use of social networks to recruit participants resulted in a sample that is younger and more strongly female than the French nursing workforce as a whole. Women represent the large majority of nurses in France, and our sample (96.7%) exceeds even that baseline, while the age distribution is skewed toward younger professionals (48.9% in the 21 to 26 age band). The findings should therefore not be extrapolated to older nurses or to male nurses without further replication. Third, the completion rate of 23.7% is modest, which raises the possibility of responder bias. Participants who completed the entire questionnaire may differ systematically from those who dropped out, particularly with respect to motivation, available time during work hours, and engagement with the topics of stress and recovery. Our results may therefore not be fully representative of the population of French nurses, and the absolute levels of psychological capital, need for recovery, and perceived stress reported here should be interpreted with caution. Fourth, key occupational confounders, namely clinical department, shift work configuration, workload intensity, and social support at work, were not measured in this study. The questionnaire battery prioritized the focal psychological constructs and was designed to minimize participant burden. Their absence is a limitation, which constitutes a priority for future research. The robustness analyses conducted with the available demographic covariates suggest that the principal findings are stable, but a complete adjustment for occupational confounders remains an objective for further investigation. Fifth, all variables were assessed through self‐report. Although the Harman single‐factor test did not indicate a substantial common method bias, more rigorous approaches such as the unmeasured latent method construct technique or marker‐variable designs would require a study design that was not implemented here. Multi‐method approaches combining self‐report with objective indicators of recovery, such as actigraphy and sleep monitoring, or with administrative data on shift schedules and off‐hours messaging, would provide a more complete picture. Sixth, the internal consistency of the Need for Recovery Scale was acceptable but modest in the present study (Cronbach's alpha = 0.68). For binary items, alpha is equivalent to the Kuder–Richardson 20 coefficient, which typically yields lower values than alpha computed on continuous items. The value obtained here is consistent with the original validation study by Dupret et al. (2018). The sensitivity analysis using the attenuation correction formula indicated that the modest reliability of the Need for Recovery Scale induces a systematic underestimation of the focal associations, with the corrected indirect effect being approximately 1.80 times larger than the observed estimate. The observed effects can therefore be considered conservative, but future studies could consider complementary instruments with higher internal consistency to assess recovery debt with greater precision. Seventh, no a priori power analysis was conducted before data collection. The post hoc sensitivity power analysis indicates that the achieved sample provided adequate power for the principal mediation pathway but limited power for detecting small effects, such as the direct association between psychological capital and perceived stress. The null direct effect should therefore not be interpreted as evidence of no relationship but rather as a constraint of the present design. A priori power calculations should be conducted in future replication studies. Priority next steps include a short longitudinal study with repeated measures of psychological capital state, recovery experiences, need for recovery, and perceived stress, alongside objective indicators of recovery such as actigraphy and sleep data. A pragmatic cluster‐randomized trial comparing psychological capital training combined with recovery architecture against psychological capital training alone or against usual practice would also provide informative evidence on the relative contributions of individual and structural levers. Conclusion In this sample of nurses practicing in France, psychological capital was not directly associated with lower perceived stress. Rather, it was associated with stress through its association with recovery processes that, in this cross‐sectional configuration, reduced recovery debt. Interventions intended to protect nurse well‐being should therefore align individual resource‐building with organizational conditions that make recovery feasible. The findings illustrate that, in the present sample, the protective association of psychological capital with perceived stress is visible primarily through recovery‐related variables and not through a direct buffering effect. These patterns are based on a cross‐sectional design and would benefit from confirmation in longitudinal and interventional studies. Author Contributions Corinne Isnard Bagnis: writing – review and editing, validation, resources. Arnaud Villieux: conceptualization, validation. Lucia Romo: validation, writing – review and editing, resources. Marie Charlotte Mollet: conceptualization, investigation, writing – original draft, methodology, visualization, writing – review and editing, formal analysis. Oulmann Zerhouni: methodology, validation, writing – review and editing, formal analysis, supervision, resources. Funding The authors have nothing to report. Ethics Statement All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. This psychological research complies with the ethical guidelines of the Helsinki Committee. Consent Informed consent was obtained from all participants included in the study. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Table S1: Recoding of demographic covariates for robustness analyses. Table S2: Model 1, mediator equation: Need for recovery regressed on psychological capital and demographic covariates (N = 183). Table S3: Model 1, outcome equation: Perceived stress regressed on psychological capital, need for recovery, and demographic covariates (N = 183). Table S4: Model 1, mediation summary with covariates (N = 183). Table S5: Model 2, mediator equation: Recovery experiences regressed on psychological capital and demographic covariates (N = 183). Table S6: Model 2, outcome equation: Need for recovery regressed on psychological capital, recovery experiences, and demographic covariates (N = 183). Table S7: Model 2, mediation summary with covariates (N = 183).