As if it weren ’ t hard enough already: Breaking down hiring discrimination following burnout ☆

Hiring discrimination towards (former) burnout patients has been extensively documented in the literature. To tackle this problem, it is important to understand the underlying mechanisms of such unequal hiring opportunities. Therefore, we conducted a vignette experiment with 425 genuine recruiters and jointly tested the potential stigma against job candidates with a history of burnout that were mentioned earlier in the literature. We found candidates revealing a history of burnout elicit perceptions of requiring work adaptations, likely having more unpleasant collaborations with others as well as diminished health, autonomy, ability to work under pressure, leadership capacity, manageability, and learning ability, when compared to candidates with a comparable gap in working history due to physical injury. Led by perceptions of a reduced ability to work under pressure, the tested perceptions jointly explained over 90 % of the effect of revealing burnout on the probability of being invited to a job interview. In addition, the negative effect on interview probability of revealing burnout was stronger when the job vacancy required higher stress tolerance. In contrast, the negative impact of revealing burnout on interview probability appeared weaker when recruiters were women and when recruiters had previously had personal encounters with burnout.


Introduction
Across different regions and professions, researchers have discovered worrisome burnout numbers. 1 Besides compromising employee wellbeing, the relationship of burnout syndrome with turnover, absenteeism, and reduced job performance (Swider and Zimmerman, 2010) presents the 21st century's labour markets with tremendous challenges. 2  In response to this problem, many researchers (primarily in the field of psychology) have studied the symptomatology and determinants of burnout across a wide span of occupations (Bakker and Costa, 2014;Lesener et al., 2019;Maslach et al., 2001).Still, little is known about labour market re-integration following burnout (Kärkkäinen et al., ☆ Acknowledgements.We would like to thank Hannah Van Borm, Brecht Neyt and Ann Leen for their timely support throughout the research process.Research is funded by Ghent University's Special Research Fund.Because scientific research is a shared and continuous process rather than an isolated action, we welcome all sources of constructive criticism.We would also like to express our gratitude towards the two anonymous reviewers who improved our work by sharing their expertise.Data Processing is organised in line with Ghent University's code of conduct, therefore adhering to the General Data Protection (GDPR) standards.
* Corresponding author.E-mail address: Philippe.Sterkens@UGent.be(P.Sterkens). 1 In the scientific literature, population estimates of burnout have relied on various self-reporting scales, thus limiting comparability between countries.However, notwithstanding the existing methodological discrepancies, burnout is widely present.For instance, 78.4% (88.5%) of the physicians in the US (China) report themselves suffering burnout symptoms (Jha et al., 2019;Lo et al., 2018).The COVID-19-crisis further threatens the mental health of the working population, especially among health-care workers who appear at higher risk of burnout at the moment (e.g.Azoulay et al., 2020).Moreover, according to Schaufeli (2018), in European countries such as the United Kingdom or Sweden, respectively 13.5% and 7.3% of the working population always feels exhausted at the end of the working day (i.e.burnout's core symptom).Lastly, more comprehensive burnout measures in the Belgian and Dutch populations indicate that 16.9% of the Belgian and 17.3% of the Dutch workers are either at a high risk of developing clinical burnout or currently experience one (Hooftman et al., 2019;Schaufeli et al., 2020).
2 Indeed, recent studies have also addressed (other) severe health shocks on labour market participation (e.g.Bengtsson and Nilsson, 2018;Jones et al., 2020) 2017) -a gap requiring attention given (i) the health and financial benefits of returning to work (Kessler et al., 2008;Stuart, 2006) and (ii) the difficulties patients experience throughout their re-integration trajectories (Boštjančič and Koračin, 2014;Kärkkäinen et al., 2019). 3 One obstacle (former) burnout patients could encounter is hiring discrimination (Purvanova and Muros, 2010;Sterkens et al., 2021) which, we interpret as the phenomenon in which individual job candidates are evaluated based on perceived group characteristics, instead of their individual capacitiesand this regardless of the veracity of perceived group characteristics. 4For example, in a Belgian surveyconducted in the same region as our study population, infraapproximately 40 % of ex-burnout patients explicitly feared being discriminated against upon re-entering the labour market (Sterkens et al., 2021).This seems to be a realistic concern, given that hiring discrimination based on depression, another mental disorder, is well established in the literature (Baert et al., 2016a;Bianchi et al., 2015).
To tackle this obstacle, it is crucial to understand its driving forces.From an economists' theoretical point of view, both the seminal theories of taste-based (Becker, 1957) and statistical discrimination (Phelps, 1972) could explain differential chances in terms of hiring outcomes of burnout patients through stigmatisation.In a framework of taste-based discrimination, the applicants' burnout could be regarded as a cost in collaborations due to a distaste for the applicant.Indeed, burnout patients struggle with acceptance within organisations (Boštjančič and Galič, 2020;Boštjančič and Koračin, 2014).Alternatively, following statistical discrimination theory, employers could interpret applicant burnout as a negative signal (Spence, 1973) for candidate productivity, therein evaluating individual applicants based on their stigmatic beliefs regarding burnout patients in general (Brouwers, 2020;Mendel et al., 2015).
From an empirical point of view, survey and interview research (e.g.Bahlmann et al., 2013;Ozawa and Yaeda, 2007) suggests, in line with statistical discrimination theory, that employers perceive burnout patients as being less productive due to lingering symptoms (Boštjančič and Galič, 2020;Boštjančič and Koračin, 2014), reduced professional autonomy (Boštjančič and Galič, 2020;Ozawa and Yaeda, 2007), trainability (Boštjančič and Koračin, 2014;Diksa and Rogers, 1996) and manageability (Laberon, 2014;Stuart, 2006).However, traditional survey and interview studies both have their limitations (i.e.employers downplaying stigmatisationsocial desirability biasor patient-employees accepting their a priori convictions as general truths).In addition, the 'professional shape' of the burnout stigma (i.e.productivity perceptions patients are branded with by employers) has only been investigated in a single experimental study by Mendel et al. (2015).In their research, German managers rated experimental vignettes on the prospective job performance of an employee who had returned from sick leave due to burnout.They found that returning burnout patients are perceived as being unable to handle pressure and satisfactorily fulfil leadership roles, being more likely to be absent due to relapses and requiring adaptations to the job.
In the current study, we contribute to the literature by empirically testing a structurally assembled body of employers' candidate perceptions on burnout patients in relation to the likelihood of hiring.To this end, we conduct a state-of-the-art vignette experiment in which fictitious job candidates with and without a burnout experience are evaluated by subjects with genuine experience in recruitment.In doing so, we respond to prior researchers' calls to "bring firms back in" and to study the ground of employer behaviour (Bills et al., 2017;Rivera, 2020).We specifically extend the literature by (i) testing the empirical value of a broader spectrum of productivity-related stigma and (ii) determining candidate perceptions related to taste-based discrimination (e.g.how the employer or co-workers would appreciate their collaboration) as an alternative explanation for reduced hiring opportunities of former burnout patients.In view of the external validity of our study, we develop a more ecologically valid scenario to reveal a history of mental health problems than used previously in experimental hiring studies.In addition, hiring behaviours and candidate perceptions related to former burnout patients are studied across different job openings (instead of focussing on a single profession or sector as done in most of the previous research).Thereby, our experimental framework also allows us to investigate how unfavourable treatment of job candidates with a burnout experience varies with job characteristics.Finally, we also investigate moderators of this treatment at both the candidate level (gender, age, timing and duration of the inactivity due to burnout and extracurricular activities) and the recruiter level (among which core burnout knowledge and burnout encounters).

Experimental design
A vignette experiment enables the analysis of human judgments and beliefs by integrating an experimental set-up in a survey (Auspurg and Hinz, 2014).It is, therefore, frequently used to study hiring decisions and discriminatory behaviour (e.g.Auspurg et al., 2017;Kuziemko et al., 2015;Van Belle et al., 2018;Van Borm and Baert, 2018).Specifically, in the context of hiring experiments participants evaluate candidate descriptions depicted in vignettes, for which the characteristics ('vignette dimensions', e.g.candidate gender) vary systematically or randomly over a number of categories ('vignette levels', e.g.male, female).
Because vignette experiments combine experimental and survey elements, they inherit favourable attributes of both: causal interpretations (experiment) and increased external validity (survey population).What makes vignette experiments particularly interesting in the context of hiring discrimination is the fact that, compared to traditional surveys, vignette experiments are more suitable for measuring sensitive issues (Auspurg et al., 2014) because the multidimensionality of the experiment forces participants to make trade-offs between dimensions and, thus, diminishes socially desirable answering.In the following subsections, the vignette design, data collection and experimental procedure are described.We discuss the limitations of our design in section 4. In the dimension 'timing of gap in working history', the gap duration was fixed to 5 ± 3 months.
3 Following Grossi et al. (2015), we define burnout patients as individuals suffering from 'clinically significant exhaustion and impaired performance, which motivates seeking professional help' (p.626). 4It should be noted that when a job candidate with a history of burnout is, indeed, less capable of performing in a job, we do not speak of discrimination (according to its strictest definition), but of 'a justified rejection in the selection procedure'.However, employers cannot estimate an individual candidate's productivity potential when she/he is denied the opportunity to participate in the selection procedure.Furthermore, in a labour market aiming to provide equal opportunities to all, a candidate's history of burnout should not a-priori limit her/his chances to prove oneself in selection procedures.Given the uncertainty employers have on (former) burnout patients' actual performance levels, we will also refer to this in our manuscript as 'differential hiring probabilities, chances or opportunities'.

Vignette design
In our vignette experiment, participants passed a series of judgments on four fictitious job candidates ('vignettes') demonstrating a gap in their working history.The deliberate choice to exclusively present job candidates with a gap in working history was motivated by methodological concerns.In particular, by having all candidates feature a gap in working historyalthough with different explanations -no substantial correlations arose between vignette dimensions ('gap in working history' and 'reason for gap in working history', infra), thus following Auspurg and Hinz's guidelines (2014).
Our job candidates varied systematically across five vignette dimensions on pre-determined vignette levels and were presented in a tabular design (Auspurg and Hinz, 2014).The employed vignette dimensions and their corresponding levels are displayed in Table 1.
Two vignette dimensions are related to the candidate's gap in working history.Our main dimension of interest was the provided reason for the gap in working historyits crucial level being burnoutas compared to 'personal reasons', 'physical injury' and 'unemployment' (control levels).The control level 'personal reasons' was based on the feedback of HR professionals (see subsection 2.3), who remarked that candidates hiding sensitive information during a job interview often produce vague explanations instead (e.g.not having worked for 'personal reasons').As a consequence, the level 'personal reasons' could additionally be interpretedand analysedas a more covert strategy for disclosing burnout.A second control reason was labelled 'physical injury' and was based on earlier comparisons of mental and physical health stigmas (e.g.Breen, 2018;Hipes et al., 2016;Lucas and Phelan, 2012;Stuart, 2006).Adding to the ecological validity of the experiment, our final control reason was labelled 'unemployment' (seeking a job).
The second dimension of interest reflected the timing of the gap in working history and allowed us to evaluate whether the effects of gaps (due to burnout) were more or less severe when situated further in the past.The length of the gaps varied between 2 and 8 months and, thus, aligned with sick leave estimates of Flemish burnout patients (Rooman, Sterkens, Schelfhout, Baert & Derous, mimeo;Sterkens et al., 2021).The gaps were placed at one of three time periods, at 0 (i.e. until the present), 2 or 5 years ago.These values were randomly adjusted ± 3 months to, again, maximise the ecological validity of the experiment.
The three remaining dimensions were common characteristics revealed during the selection procedure and were potential additional moderators in the relationship between revealing burnout and hiring chances.More concretely, the third dimension was sex (male, female).Because burnout could be perceived as a primarily female experience due to gender stereotypes (Eagly and Wood, 2012;Purvanova and Muros, 2010) and gender-incongruent behaviour is penalised by recruiters (Cohen and Bunker, 1975), hiring discrimination against burnout patients could be more common among male candidates (as is the case for depression (Baert et al., 2016a).
As a fourth dimension, we incorporated candidate age into vignettes via three levels: 30, 40 and 50 (randomly adjusted ± 3) years.Hiring discrimination against older burnout patients could be more outspoken compared to young workers due to age's signalling effect of worse health and reduced energy (Baert et al., 2016b).Adding a history of burnout to a vignette of older candidates could, therefore, strengthen these perceptions and elicit lower hiring chances.
A fifth dimension was extracurricular activity (sports, association, volunteering, none) because this dimension is commonly featured in other vignette studies mimicking real-life hiring decisions (e.g.Di Stasio, 2014;Van Belle et al., 2020).
Our selection of dimensions and levels resulted in a 2 (gender) × 3 (age) × 3 (timing of gap in working history) × 4 (reason for gap in working history) × 4 (extracurricular activities) design of 288 unique vignettes (the 'vignette universe').In a completely matched factorial design, participants would be exposed to all 288 stimuli, yet this was unachievable in terms of cognitive demands.The alternative, in which each participant would be presented with a single vignette, would have required an enormous sample to have each vignette rated by several recruiters.Therefore, following Auspurg and Hinz (2014), we randomly presented participants with a selection of four vignettes ('vignette decks').More specifically, we selected (40 × 4 =) 160 vignettes from the vignette universe and stacked them in decks of four using the D-efficiency algorithm (Auspurg and Hinz, 2014).A D-efficient design contains those selections of vignettes yielding the most precise parameter estimates.With a substantially high D-efficiency score of 98.347, a negligible covariance existed between vignette dimensions (Auspurg and Hinz, 2014).

Data collection
The vignette experiment was administered online through Qualtrics to a sample of Flemish HR representatives (hereafter referred to as recruiters) who submitted vacancies with the Public Employment Agency of Flanders (PEAF) -Belgium's largest job site (Delbeke, 2019).E-mail addresses from eligible recruiters were collected in January 2020 by screening (i) the 500 most recent vacancies posted (in general) and (ii) up to 2000 of the most recent PEAF vacancies for one out of eight job types studied in our experiment (as discussed in subsection 2.3).
Following this search strategy, we collected and subsequently contacted a total of 2,327 unique e-mail addresses through two points of contact (i.e. an initial invitation and one reminder).In view of avoiding non-response bias (Tourangeau et al., 2013) due to a specific interest in mental health, we deliberately concealed the actual topic of study in the invitation.More specifically, recruiters were invited to complete a questionnaire "on recruitment and selection… in which fictitious hiring decisions were to be made".Additionally, to increase response quality (Tourangeau et al., 2013), we organised a raffle, with prizes having a total value of 750 euro, among participants who had submitted complete and accurate responses (i.e.passing an attention check in the post-experimental questionnaire; 'This is a quality check, please indicate six for this item.').In total, 928 recruiters opened the web link, and of them fully completed the experiment.Next, we filtered out participants who failed the attention check (Liu and Wronski, 2018), yielding 425 suitable responses (18.3 %) and, thus, (425 × 4 vignettes =) unique vignette observations.

Procedure
The following paragraphs describe participants' trajectory from invitation to experiment completion.Study participation took approximately 15 min and consisted of four parts: (1) introduction, (2) job vacancy and instructions, (3) candidate evaluations and (4) a postexperimental questionnaire.
The first part of the experiment comprised two screens providing participants with an introduction to the study in terms of its duration, confidential data processing, participants' rights and raffle participation.After reading through the information and providing their consent, participants indicated whether they had a recent experience with one of the following vacancies: driver (and salesperson), welder, telemarketer, massage therapist, chemical engineer, researcher, ICT administrator or tutor (i.e.78.1 % of the sample).
In the second part of the experiment, participants were presented with one out of eight fictitious job vacanciesaccording to their indicated hiring experiences, otherwise displaying one out of eight vacancies randomly (infra) -and received detailed experimental instructions.Our selection of eight vacancies was based on four underlying job characteristics that served as potential job-side moderators in the relationship between revealing burnout and hiring outcomes.First, vacancies varied according to their required level of education because workers suffering from mental health problems are more likely to be employed in jobs requiring lower levels of education (Stuart, 2006).Second, vacancies varied in required leadership because the burnout P. Sterkens et al. stigma of reduced leader capacities (Mendel et al., 2015) could create for an additional disadvantage when applying for leadership positions (Brohan et al., 2012).A third dimension in which occupations differed was required level of stress tolerance.The relationship between stress and burnout (i.e.cumulating exhaustion due to work stressors; Bakker and Demerouti, 2017) could be reflected in lower suitability ratings for jobs requiring a high level of stress tolerance because those jobs could be perceived as being too demanding.Fourth, the presented vacancies varied in their required emotional labour.That is, the emotional impairments burnout patients experience (Schaufeli et al., 2020), could encourage employers to treat former patients differently for jobs with emotional demands out of fear of reduced performance.
To select jobs for our experiment differing in these four characteristics and write fictitious job descriptions, we examined the corresponding O*Net classifications per job characteristic and, as presented in Table A1, isolated each characteristic in our job choices by constructing an (8 × 4) matrix varying the levels of job characteristics. 5After reading one job description, participants rated the job's requirements (i.e.leadership abilities, stress tolerance and emotional labour), thus collecting participant estimates of job requirements for further analyses.
Consistent with (upcoming) scales throughout the experiment, statements were rated on response scales ranging from 0 'Fully disagree' to 10 'Fully agree' (Auspurg and Hinz, 2014).
After appraising the job vacancy on its underlying characteristics, participants received information on the experimental context and instructions on the simulated hiring assignment.Specifically, participants were told that another colleague had one prior interview with several applicants.Before sending out invites for a second round in the selection procedure, participants were asked for their advice on a couple of candidates, herein relying on extracts of the colleagues' interview notes summarised in the organisation's HR software package (i.e. the vignettes described in subsection 2.1).In our opinion, and that of real-life HR professionals,6 this description improved the ecological validity (i.e. the study context approximating the real-world situation that is being examined) compared to earlier correspondence and vignette experiments (e.g.Baert et al., 2016a;Hipes et al., 2016) in which personal information such as mental health problems was revealed on a candidate resume.The implications of our choice of experimental context are discussed in Section 4.
In the third part of the experiment, participants were randomly assigned to one deck consisting of four job candidates (vignettes) each and shared hiring advice on each candidate.That is, they assessed candidates employing two sets of evaluative statements (again, using 11-point response scales).In the first set of statements, they indicated the probability with which they advised (i) to invite the candidate for the second phase of the solicitation procedure (a 'proximal hiring outcome') and (ii) to hire them (a 'distal hiring outcome').
Next, a set of statements estimated potential signals emitted by the different candidate profiles.All statements employed for candidate evaluations are presented in Table 2 below.Ten of these 'signalling statements' reflected stigma as potential sources of statistical discrimination.Seven represented distinct productivity perceptions derived from prior research.More specifically, we asked whether the recruiters thought that applicants had sufficient (i) leadership abilities (Mendel et al., 2015), (ii) autonomy (Ozawa and Yaeda, 2007), (iii) ability to work under pressure (Mendel et al., 2015;Ozawa and Yaeda, 2007), were sufficiently (iv) manageable (Laberon, 2014;Stuart, 2006), had sufficient (v) learning abilities (Boštjančič and Koračin, 2014;Grossi et al., 2015;Öhman et al., 2007) and were perceived as sufficiently healthy in terms of both (vi) current health and (vii) the likelihood of future sick leave (Laberon, 2014;Mendel et al., 2015) to perform well in the job.The remaining three statements related to statistical discrimination gauged for estimations on adaptational requirements in terms of work context, conditions and job content for the candidate to perform well in the job (Brohan et al., 2012;Laberon, 2014).The last three statements stemmed from the theory of taste-based discrimination and gauged taste for collaboration with the candidate from three different perspectives (i.e. the employer, co-worker and client).
Based on principal component analyses (PCA) and items' theoretical underpinnings, the three statements measuring adaptational requirements (α = 0.865) and the items related to taste-based discrimination (α = 0.902) were combined into scales; they were used in item form in robustness checks discussed below, however.
The fourth and final part of the experiment comprised a postexperimental questionnaire investigating variables that served as (potential) participant-side moderators in the relationship between revealing burnout and hiring outcomes or were collected for robustness checks.
Three potential demographic moderators were gender ('male' or 'female'), age (20-67 years) and education level ('no tertiary education', 'Bachelor level tertiary education' or 'Master level tertiary education').Compared to their female colleagues, male recruiters were expected to Notes.Each item was rated on a scale from 0 (Completely disagree) to 10 (Completely agree).
show more discriminatory tendencies because men support stronger stigmatic perceptions when evaluating job applicants (Cole et al., 2004).Further, because older employers (aged over 60 years) were reported to have a more negative attitude on the expected productivity of applicants with mental health problems (Ozawa and Yaeda, 2007), discrimination against burnout patients could be more common among older recruiters.
A first potential psychographic moderator was core burnout knowledge. 7Although it was expected that (public) knowledge could decrease discriminatory behaviours (Brouwers, 2020), the literature remains indecisive on the impact of (burnout) knowledge on hiring discrimination (Brohan et al., 2012;Brouwers, 2020).Recruiter encounters with burnout syndrome ('none', in their 'professional' life, 'personal' life or as a former 'patient') was a second potential psychographic moderator we surveyed.Based on Allport's (1979) 'in-group contact hypothesis', recruiters with more personal burnout encounters would actually be increasingly likely to hire candidates with a history of burnout.A third potential psychographic moderator is risk-taking.When a history of burnout emits negative productivity signals, hiring a former burnout patient could represent an elevated risk for employers (Spence, 1973), which recruiters more prone to risk-taking might be willing to take.Following Baert's (2018a) six-item application in the study of hiring discrimination against gay men, we assessed risk-taking using the validated Domain-Specific Risk-Taking Scale (Blais and Weber, 2006).An example item of a professional risk rated on a scale from 1 'extremely unlikely' to 7 'extremely likely' is 'investing 10 % of your annual income in a new organisation'.
Next, the post-experimental questionnaire surveyed participant characteristics in view of robustness analyses.Specifically, we asked participants for their general hiring experience in terms of frequency ('daily', 'weekly', 'monthly', 'once per semester', 'once a year', 'less frequent') and tenure ('less than a year', 'one to five years', 'greater than five years').Moreover, we asked for prior experiences with hiring candidates suffering from mental health problems ('yes' or 'no').Lastly, social desirability was measured through the shortened Marlowe--Crowne Social Desirability Scale (Beretvas et al., 2002;Reynolds, 1982;Sârbescu et al., 2012;Van Borm and Baert, 2018).This scale contained 13 items expressing behaviours that are either socially sanctioned or approved (e.g.'I sometimes feel resentful when I don't get my way.') and participants indicated their applicability (score 1) or not (score 0).Afterwards, participants' total social desirability scores were calculated by summing up and standardising item scores.
Finally, participants were thanked for their time and were able to leave an e-mail address if they wanted to participate in the raffle or when interested in the research results. 7Because there did not appear to be a validated scale on burnout (mis)conceptions in the literature, we developed a scale ourselves indicative of participant's familiarity with the (scientific) concept.The scale comprised 10 statements rated from 0 'Fully disagree', 5 'Neutral' to 10 'Fully agree'.
Addressing shortcomings from earlier misconception scales (Bensley and Lilienfeld, 2017;Bensley et al., 2014;Gardner and Brown, 2013), we allowed participants to express uncertainty by introducing the central point of the scale as a neutral option.Participants' correct responses (i.e.scores between 0 and 4 for incorrect statements and between 6 and 10 for correct statements, here printed in italics) were added to form a single core knowledge score.Specifically, the statements surveyed burnout symptomatology (exhaustion, cognitive problems, psychological distance from work and reduced self-efficacy), differentiation from other conditions (overstrain and lacking motivation), the main determinant (sleep deprivation) and misperceptions on adequate treatment (taking a holiday) and occurence (contagiousness within organisations and the possibility of relapses).An example statement is: 'People with burnout often feel exhausted'.

Data description
Exploring the experimental data collected we did not expect correlations between vignette dimensions and job characteristics (discussed supra), thus enabling us to unravel the effects of both candidate (vignettes) and job characteristics.Indeed, the insignificant t-tests (and chi-squared tests in case of discrete variables) shown in Panel A of Table 3 indicate that the randomisation of vignettes across vacancies was successful.Similarly, Panel B indicates that no correlations existed between candidate (vignette) and participant characteristics, thus adhering to the experimental logic proposed in Auspurg and Hinz (2014).
In addition, Panel B shed light on the sample characteristics.Both male and female (71.9 %) recruiters participated, with an average age of about 38 years.Adding to the representativeness of our sample, participants differed substantially in their level of education.Where 43.8 % of the sample did not enjoy any tertiary education, respectively 18.6 % and 37.6 % of the sample followed Bachelor or Master level tertiary education, respectively.Our sample was comparable in both age and gender distribution with HR professionals from the European Social Survey (ESS) 8 but was slightly more educated (i.e.12.7 percentage points more participants enjoyed Master level tertiary education).Participants' (i) considerable knowledge of burnoutas indicated by the average scores (7/10) on the core burnout knowledge scale discussed in subsection 2.3 -and (ii) encounters with the syndrome further support the sample's eligibility.More specifically, no less than 88.4 % of our participants have had an encounter with burnout in their private or professional lives.

Results
To investigate whether (3.1), how (3.2) and when (3.3) revealing burnout affects hiring likelihood, we conduct three consecutive series of linear regressions in Stata/MP (version 15) where the error terms are consistently corrected for the clustering of the observations ('vignette evaluations') at the participant level.In all cases, ordered logistic regressions yield equal results.

Effect of revealing burnout on hiring chances
We first set the stage by investigating the total effect of revealing burnout on hiring outcomes (displayed in Fig. A1 and referred to as path c).As such, Table 4 presents the results of hiring outcomes regressed on combinations of (i) candidate, (ii) job and (iii) participant characteristics.In eight regressions, we assess the stability of relationship between revealing burnout and hiring outcomes by comparing regression results from different control specifications ('reasons for gaps') and independent variables.Because the experimental set-up required each candidate to explain their gap in working history (subsection 2.1), our analyses pinpoint differences in unequal treatment among various stigmatized groups.Indeed, candidates' employment gaps are, to some degree, generally accompanied by stigmatisation (e.g.unemployment (Van Belle et al., 2018) and physical injury (Ameri et al., 2018).
In confirmation of burnout patients' fears (Sterkens et al., 2021), candidates revealing a history of burnout are at a severe disadvantage during the hiring process.As model (1) and model ( 5) demonstrate, when a gap in working history is explained by burnout, candidates receive lower ratings (i.e.β = − 0.598 for interview probability and β = − 0.623 for hiring probability when controlling for job and participant characteristics) compared to all other gap explanations together (i.e.personal reasons, a physical injury and unemployment).Moreover, out of the four explanations for gaps in working history, the coefficient estimates of model (2) show that burnout has the largest negative impact on interview probability (burnout: β = − 0.984, personal reasons β = − 0.705 (p < 0.001), unemployment β = − 0.439 (p < 0.001) and physical injury (p < 0.001, as the reference category)).Additional F-tests further confirmed the differences in hiring chances between burnout and the other control conditions i.e. personal reasons (F(1, 424) = 4.54; p = 0.034) and unemployment (F (1, 424) = 14.91; p < 0.001).Compared to candidates with a gap due to a physical history, the chance of a recommendation for a second job interview is 9.7 percentage points lower for (former) burnout patients. 9This equals a decrease of 0.428 (0.480) standard deviation in interview (hiring) probability.Estimation results are very equal when controlling for job and participant characteristics, which is not surprising given our experimental design (subsection 2.4).
These findings resonate well with earlier research on hiring discrimination by Hipes et al. (2016), which stated mental health patients have a stronger disadvantage in the labour market when compared to applicants with a history of physical illness.Still, in our opinion this makes for an interesting addition to the literature because to the best of our knowledge we are the first to empirically test hiring discrimination against former burnout patients.Although Bahlmann et al. (2013) suggest that burnout might be a label carrying little stigma perhaps due to patients' history of successful employmentour findings illustrate that patients are even at a more severe disadvantage compared to the aforementioned gap explanations, including physical injury.Moreover, in comparison to the category of reference, the negative impact of burnout on hiring chances is 2.283 times more harmful than experiencing an unemployment spellwith according to Baert et al. (2016), unemployment gaps already decreasing positive call-back ratios (in the Belgian labour market) with 33.8 percentage points compared to candidates without gap in their resumes.

Driving signals of hiring discrimination against burnout patients
Next, we dive more deeply into the effect of revealing burnout on hiring outcomes by examining what proportion of the total 'burnout effect' (calculated in subsection 3.1) can be ascribed to the candidate perceptions operationalised in subsection 2.3.That is, the total burnout effect can be decomposed into indirect effects via signals and attitudes (paths a × b; see Fig. A1) and a remaining 'direct' effect of burnout (path c') via multiple mediation analyses (Hayes, 2017).To do so, we jointly estimate 10 regression models.Nine of these models regress our candidate perception scales on the same independent variables as adopted in Table 4 (i.e. the candidate characteristics, job and participant characteristics collected throughout the experiment).The tenth model regresses the interview probability on the same variables as well as the nine perception scales. 10 Table A2 presents the full estimation results and Fig. 1 contains a visual representation of burnout's signalling effects.Revealing a history of burnout clearly emits several negative signals to employers.More specifically, compared to candidates with a gap due to physical injury, burnout patients are perceived as having lower (i) leadership capabilities (β = − 0.887; p < 0.001), (ii) autonomy (β = − 0.518; p < 0.001), (iii) ability to work under pressure (β = − 1.806; p < 0.001), (iv) being less manageable (− 0.411; p < 0.001), (v) having less learning abilities (β = − 0.325; p < 0.001), (vi) worse current health (β = − 0.696; p < 0.001), (vii) an increased probability of future sick leave (β = 0.702; p < 0.001), (viii) requiring increased adaptational requirements (β = 0.319; p = 0.003) and (ix) collaboration with them is regarded more negatively (β = − 0.425; p < 0.001).Thus, these estimation results provide empirical evidence for all potential burnout signals we identified in the literature (subsection 2.3).In particular, the effect on perceived ability to work under pressure is substantial: compared to candidates with a gap due to physical injury, burnout patients score about 18 percentage points lower on the related scale.
Even though we provide evidence for multiple burnout signals, not all of these signals necessarily drive hiring discrimination against burnout patients, as recruiters might not take each signal into account when making hiring decisions.Therefore, to test the investigated signals' potential as mediators, we calculate, in line with (Hayes, 2017), the indirect effects of burnout on hiring outcomes (compared to candidates with a history of physical injury) via the signals over a bootstrapping 9 The likeliness of advicing to invite a candidate is rated on a scale going from 0 to 10.The average probability, across all participants, therefore also ranges from 0 to 10 (as a continuous scale).By dividing the values for the scale by 10, we arrive at a classical probability (between 0 and 1) and by multiplying by 10 at a probability in percentage terms.The difference between two percentages (i.e. the probability of being invited for two separate groups) then provides a value that can be interpreted as a difference in percentage points. 10Further analyses predominantly focus on the outcome variable 'interview probability' -instead of hiring probabilitybecause interview probability is the outcome closest to the experimental set-up described in Section 2 (i.e.before reaching the (final) hiring decision the selection procedure could require additional steps such as assessment centers and contacting references).The complete analyses with hiring probability, nonetheless, yield similar results and are available from the first author upon request.procedure with 500 reps. 11More concretely, these indirect effects are estimated by multiplying the regression coefficients from (i) burnout to the signal (path a in Fig. A1) with (ii) coefficients from the signal to interview probability (path b in Fig. A1) (Hayes, 2017).Table 5 below provides an overview of the percentages of the total burnout effect on interview probability as explained by each of the investigated signals.
Five signals emerge as significant mediators of the relationship between burnout and interview probability.Most importantly, the lion's share of the burnout effect can be explained by the candidate's perceived ability to work under pressure (i.e.no less than 45.1 %).By consequence, perceptions of reduced stress tolerance represent a major obstacle for job candidates with a history of burnout.The remaining four statistically significant mediatorsperceived probability of future sick leave (10.9 %), perceived autonomy (9.9 %), perceived current health (9.6 %) and perceived manageability (6.8 %) -all explain proportions of the effect that are lower but still substantial.
Our proposed mediation model fully mediates the effect of revealing burnout on interview probability because the (remaining) direct effect of revealing burnout is no longer significant when controlling for indirect effects via mediators (β = − 0.069, p = 0.517; path c' in Fig. A1).
Our data provide evidence for the role of burnout's productivityrelated stigma (as suggested based on the theory of statistical discrimination (Phelps, 1972)) in explaining burnout patients' reduced hiring chances (compared to candidates with a history of physical injury).In contrast, we do not find evidence for a role of attitudes towards collaboration (as suggested by the theory of taste-based discrimination (Becker, 1957)) in explaining burnout patients' hiring chances.We namely conclude that more than 90 % of the burnout effect on interview probability can be explained by productivity-related stigma, whereas a statistically insignificant 1.3 % is explained by the perceived employer, co-worker and customer attitudes on collaboration with former patients.
Because the present study is, to our knowledge, the first to delineate the unequal hiring chances of burnout patients into its underlying stigmatic perceptions, direct comparisons to other burnout studies are impossible (for now).Nonetheless, where in earlier experiments Van Borm et al., (2021) age discrimination's driving perceptions leave 35 % of the (direct) age effect unexplained, following a similar statistical framework, we find that discrimination against burnout is remarkably well captured by our proposed stigma framework.We return to the practical consequences of these findings in section 4.
Next, to investigate the robustness of our results, we conduct similar mediation analyses with (i) hiring probability as a dependent variable and (ii) single items for perceptions on adaptational requirements and collaboration as mediators (instead of the scales comprising three items).The main results of the additional mediation analyses are presented in Table A3.Again, indicating robustness, the following mediation effects are statistically significant across conceptualisations: perceived ability to work under pressure, perceived current health, perceived future sick leave and perceived autonomy, with the first perception being the most important driver.
As additional robustness checks, we redo our mediation analysis relying on more homogeneous subsamples of participants identified in subsection 2.3.More specifically, we developed subsamples of recruiters characterised by (i) experience with the vacancy, (ii) experience at hiring candidates with mental disorders, (iii) a hiring frequency of at least once per month, (iv) a hiring tenure greater than one year and (v) a low or average social desirability tendency.In Table A4, the decomposition of the burnout effect is once more presented, this time at the level  11 Following the mediation literature, we speak of mediation 'effects'.However, a causal effect from signals on hiring outcomes cannot be demonstrated from the current study design because the signals could correlate with unobserved causal mechanisms (path b in Fig. A1).Hence, mediation effects should be interpreted as associationsnonetheless, supported on a theoretical basis.In contrast, our experiment does allow for causal interpretations of the total effect of burnout on (i) hiring outcomes (path c) and (ii) on candidate perceptions (path a).We return to this point in section 4. of these five subsamples.Again, our main mediatorthe perceived ability to work under pressureconsistently remains the dominant explanation, thus indicating the robustness of our results and only slight deviations occur between subsamples.

Moderators in the relationship between revealing burnout and hiring outcomes
We conclude the analyses with an exploration of potential moderators (introduced in section 2.3) of burnout patients' hiring chances compared to candidates with a history of physical injury.Yet, the current experiment does not allow for a causal interpretation of moderators at the job and participant level and, thus, demands caution during interpretation.Table 6 presents the results of our moderation analysis with interview probability.The first three models test two-way interactions separately for candidate, job and participant characteristics.Then, a fourth regression jointly tests all possible interactions.The results of similar regressions with hiring probability as outcome variable are also presented between brackets.
While no evidence emerges for two-way interaction effects between revealing a history of burnout and other candidate characteristics in model ( 1), there are significant interactions with both job and participant characteristics in models (2) and (3).When conducting the joint test of interaction effects in model ( 4), no notable differences are found compared to the separate tests.In what follows, we refer to the coefficients of the latter model.
First, a significant interaction effect between revealing burnout and the job's required stress tolerance indicates that burnout patients are additionally disadvantaged, in terms of interview chances, when applying for jobs that require higher stress tolerance (β = − 0.133, p = 0.036, see Fig. 2; hiring probability: β = − 0.113, p = 0.065).The statistical significance of this interaction is in accordance with our expectations from subsection 2.3 because the stress-related nature of burnout syndrome (Bakker and Demerouti, 2017) is clearly comprehended by recruitersas is reflected in their familiarity with burnout (subsection 2.1) and the perceived burnout signal of a reduced ability to work under pressure (described in subsection 3.2).Second, and in line with the predictions from subsection 2.3, there is marginal evidence that female recruiters are more likely to invite former burnout patients for a job interview compared to their male counterparts (β = 0.427, p = 0.083, see Fig. 3; hiring probability: β = 0.535, p = 0.019).
Third, as suggested in subsection 2.3, statistical evidence is found for an interaction effect between revealing burnout and recruiters' encounters with burnout, where recruiters with more personal burnout encounters (i.e. in their personal lives (β = 0.971, p = 0.009; hiring probability: β = 0.384 p = 0.238), or as a former patient themselves (β = 1.150, p = 0.021; hiring probability: β = 1.092, p = 0.014) are more likely to invite candidates with a history of burnout for a job interview (see Fig. 4).
A final notable result is the absence of an interaction effect between participant's core burnout knowledge and candidates' history of burnout on hiring decisions (β = − 0.024, p = 0.740; hiring probability: β = − 0.087, p = 0.226).Whereas both Brohan et al. (2012) and Brouwers (2020) point out that the effect of employer's knowledge on subsequent hiring behaviours remains a topic of debate, our findings steer in the direction that there is no effect of core burnout knowledge on hiring evaluations of ex-burnout patients.However, given the scale's explorative nature, the same cannot be claimed for more specialised knowledge concerning e.g.treatment and prognosis.

Conclusion
To explain differential treatment of burnout patients in terms of underlying stigma, we conducted a vignette experiment in which genuine recruiters evaluated fictitious job applicants with different explanations for a gap in their working history, among which burnout.More concretely, these recruiters rated four applicants for one out of eight job vacancies on a total of 13 statements related to all dominant explanations for hiring discrimination against burnout patients derived from the scientific literature.Besides being the first to empirically test the theoretical body of burnout stigma and its role in explaining hiring discrimination, we contributed to the scientific literature in various aspects.More specifically, we explored potential candidate, job and employer-side moderators of burnout patients' reduced hiring chances (compared to candidates with a history of physical injury) and developed a more ecologically valid scenario to reveal a history of mental health problems in experimental hiring studies.
Revealing a history of burnout during a selection procedure reduces the candidate's subsequent hiring chances.More so, the negative effect of burnout on hiring likelihood is larger than the negative effects of physical injury, unemployment or personal reasons as reasons for a gap in working history.Perhaps most striking is that approximately half of this adverse effect of revealing burnout, compared to a history of physical injury, can be captured by employer perceptions of reduced stress tolerance.Moreover, our results indicate that the negative impact of revealing a burnout could be more prominent for jobs requiring higher levels of stress tolerance.Conversely, the disadvantageous interview likelihood of burnout patients might be less pronounced when recruiters previously encountered burnout in their personal lives or when the recruiter is female.The adverse hiring outcomes of former burnout patients are additionally explained by candidate perceptions of increased future sick leave, lower autonomy, worse current health and reduced manageability.Although not directly associated with hiring outcomes in our experiment, burnout patients are also perceived as having lower leadership abilities, learning abilities, requiring job adaptations to work productively, and the collaboration with them is regarded more negatively.
These results have practical implications for the three parties (in) directly involved in the employment relationship: former burnout patients, employers and policy-makers.12First, when applicants decide to   disadvantageous position of burnout patients is most likely worsening in the less tight labour markets resulting from the economic downturn (Lippens et al., 2021).This provides employers with more opportunities to take stigma against burnout patients into account when making hiring decisions, thus increasing the relevance of our study's implications.
P. Sterkens et al. disclose their burnout history to potential employers, they should be aware of the potential negative signals they 'transmit' in doing so.A decision to disclose shoulddespite employers' expectations to do so (Brouwers, 2020;Mendel et al., 2015) be well-considered and prepared.Following signalling theory, one potential route out of differential treatment could be that candidates signal their renewed stress tolerance and improved health to counteract negative perceptions based on their burnout history, although concrete implementations require further investigation.
Second, in their efforts to optimise hiring decisions, employers should ideally be aware of ungrounded negative perceptions they have about job candidates with a history of burnout.When employers incorrectly deduce candidate characteristics based on a history of burnout, sub-optimal hiring decisions could be made (Travis, 2002) and creates legal risks for employers.After all, burnout patients were productive employees before they fell victim to the syndrome (Schaufeli et al., 2020) and discrimination based on prior and current health is forbidden (FOD, 2021).In addition, burnout is not necessarily indicative of a candidate's unsuitability because the determinants of burnout are primarily situated in the (prior) workplace rather than with potentially unfavourable individual characteristics (Alarcon, 2011;Bakker and Demerouti, 2017).
Third, as we did not find main or interaction effects of core burnout knowledge on hiring decisions for candidates revealing burnout, when combating hiring discrimination against burnout patients, policymakers could consider to not solely base interventions on educating employers on the fundaments of burnout (prevention e.g.symptomatology and causes) (Gronholm et al., 2017).Instead, the results from moderation analyses suggest that in the first place research agendas, and subsequent interventions, could investigate opportunities for more personal encounters with the syndrome and its patients because of the association with less negative perceptions and are, therefore, a promising route to explore.Alternatively, training for selection personnel to counter prejudice and bias could be another promising route to explore (e.g.Derous et al., 2020;Isaac et al., 2009).More concretely, knowledge of the burnout stigma could be implemented in the development of such programmes to foster participant's understanding of burnout and stigmatisation and ultimately reduce hiring discrimination.
We conclude our article by acknowledging our study's limitations and formulating suggestions for future research.First, whereas the experimental set-up of our study allows for a clear causal interpretation of the effect of revealing burnout on (i) hiring outcomes and (ii) candidate perceptions, it cannot claim causality for the associations between candidate perceptions and hiring outcomes.
A second limitation, or rather caveat, of this study is in the decision to disclose mental health problems.We are aware that whether a job candidate reveals a history of mental health problems is often a choice when applying for a job.For one, applicants could hide their history of burnout to avoid a negative backlashor at least in the short term (Brouwers, 2020;Rüsch et al., 2018).Our findings which indicate that mentioning 'personal reasons' (i.e.remaining vague on the actual reasons behind an absence period) is perceived less negatively than disclosing a burnout history align with this idea.There are, however, many reasons why applicants reveal prior burnout: to obtain crucial work adjustments, serve as a role model for others, having positive experiences with disclosure, to obtain employer's support, build an authentic working relationship, as an explanation for their own behaviour and to avoid the stress accompanying concealment (Brohan et al., 2012;Brouwers, 2020).Furthermore, we carefully developed our experimental instructions to approximate a realistic hiring context in which burnout was revealed, thus enhancing ecological validity.While the complexity of the experimental design greatly increased the realism of the designand the D-efficient design enabled statistical investigation of two-way interactions (Auspurg and Hinz, 2014) -a third limitation of our study lies in the design's limited power to capture some higher-level interactions, like 'candidate gender × burnout × participant gender'.In our data we did not find evidence for such effect (β = − 0.100; p = 0.837).However, given this study's pioneering role in the investigation of discrimination against burnout patients, current analyses already paved the road for further investigation of more complex relationships between different moderators we identified.
Also, a final limitation is related to the laboratory setting in which the experiment was organised.Because participants were fully aware that they were partaking in an experiment, the set-up could have induced a certain degree of measurement bias.In acknowledgment of this limitation, our vignette experiment approximated the complexity of hiring decisionswithin experimental boundariesby simultaneously varying candidate characteristics besides revealing burnout (e.g.sex and extra-curricular activities).Indeed, research has demonstrated that decisions made in vignette experiments are highly correlated with actual behaviour (Baert and De Pauw, 2014;Hainmueller et al., 2015;Van Belle et al., 2018).Moreover, we performed several robustness checks to test for potential measurement errors.In particular, analyses on a sub-group of participants with low-to-average social desirability scores yield similar results.Furthermore, the estimates in the current study are probably lower-bounds of burnouts' true effects on hiring chances and candidate perceptions (relative to candidates without employment gap, a condition we did not implement) because of our reference category i.e. candidates who recovered from a physical injurya group that is already disadvantaged in the labour market (Ameri et al., 2018;Hipes et al., 2016) and because of our scenario depiction of 'advising another colleague'.More specifically, the presence of significant others (like colleagues) might externally motivate recruiters to respond with fewer prejudice (Derous et al., 2012) because of a heightened sense of accountability (compared to 'isolated' decision-making) and, hence, be a conservative test of our hypotheses (Derous and Ryan, 2019).
To conclude, future research could complement our findings by estimating the exact magnitude of burnout on hiring chances in a field setting, for instance through a traditional correspondence test (Baert, 2018b).Moreover, although we approached the hiring penalties ex-burnout patients face from a purely economic perspective in this paper (i.e.stigmatisation through taste-based and statistical discrimination theories), discrimination theories from related social fields can be a valuable addition to the literature.In fact, multidisciplinary research is becoming increasingly popular and advancing insights in areas like discrimination and also burnout.Concretely, we are convinced that psychological (Dovidio and Gaertner, 2010;Gilbert et al., 1998) and sociological (Bills et al., 2017;Rivera, 2020;Tilcsik, 2021) approaches to (workplace) discrimination could enrich the current economists' approach.Moreover, interdisciplinary approaches of discrimination are not limited to interwoven theory, but also allow for methodological 'cross-fertilisation'.Based on reviewers' suggestions, future research could reproduce our experiment with other constructs such as 'feeling thermometers' or 'implicit bias' measures such as the Implicit Association Test (Bertrand et al., 2005;Blommaert et al., 2012;Dovidio and Gaertner, 2010)

Fig. 1 .
Fig. 1.Average perception scores of job candidates with a history of burnout compared to physical injury.

P
.Sterkens et al.

Fig. 2 .
Fig. 2. Interview probability by candidate's history of burnout and the vacancy's required stress tolerance.

Fig. 4 .
Fig. 4. Interview probability by candidate's history of burnout and participant's encounters with burnout.

Table 1
Vignette dimensions and levels presented in experimental materials.

Table 2
Statements employed for candidate profile evaluations.

Table 3
Data description, sorted by fictitious candidate's reason for gap in working history.

Table 4
Regression results with interview and hiring probabilities as outcome variables.
8We consulted data from the 2018 wave to compare our sample distribution of recruiters to the European HR professionals in terms of gender and education level.Survey respondents were selected according to the ISCO-08 occupation codes of 1212 (Human resource managers), 2423 (Personnel and careers professionals), 3333 (Employment agents and contractors) and 4416 (Personnel clerks).P.Sterkens et al.

Table 5
Mediation analysis: Percentages of burnout effect on interview probability explained by each mediator.

Table 6
Regression results with interview probability as the outcome variable, two-way interactions included.