Perceived organizational support (POS) is defined as the employees' “beliefs concerning the extent to which the organization values their contribution and cares about their well-being” (Eisenberger et al., 1986, p. 501).

Organizational support literature theorizes that a factor of business success is the extent employees develop beliefs concerning organization orientation to employees' welfare such as the organization valuing employees' contributions and caring about their well-being (Eisenberger and Stinglhamber, 2011). On one hand, employees develop an organization's identification evaluating the received treatment in the workplace. This perception is important both for business success and employee health (Eisenberger et al., 2002). On the other hand, social exchange theorists argue that receiving increased welfare activities from their organizations might contribute more to success. Specifically, employees compensate their employer with higher work performance (Eisenberger et al., 1986).

In a systematic review of 70 studies, Rhoades and Eisenberger (2002) found that positive organizational support is related to fair organizational procedures, supervisor support, and favorable rewards and job conditions, which in turn lead to positive outcomes for both the individual and the organization, such as increased affective commitment to the organization, increased performance, and reduced withdrawal behaviors.

In particular, Riggle et al. (2009) conducted an important meta-analytic study including 167 works. Results indicated that POS had a strong association with job satisfaction, organizational commitment, and intention to leave; whereas only moderate, positive effect on employee performance.

Research has pointed out that work organizations can be regarded as “employees' perceptions” as they represent contexts where people tend to assign the organization humanlike characteristics (Eisenberger et al., 1986) considering the company having its own unique capability, just like people do. Indeed POS seems characterized by specific psychological processes: the reciprocity norm, —the employee's felt obligation to care about the organization's welfare and to help the organization reach its mission—, the fulfillment of socio-emotional needs—leading employees to identify positively in the organization—, the development of beliefs that the organization recognizes and rewards increased performance (i.e., performance-reward expectancies; Rhoades and Eisenberger, 2002; Eisenberger et al., 2014; Kurtessis et al., 2015).

In the first process, when employees trust their organization they develop a sense of obligation. They might work harder as well as increasing voluntary citizenship behaviors because the basis of social exchange is the norm of reciprocity (Eisenberger et al., 1986, 2001; Rhoades and Eisenberger, 2002).

In the second process, employees consider their organization a source of social and emotional resources and are dedicated to work because they feel valued, cared for, and esteemed. As a consequence emotional identification toward the organization (brand, costumers) may increase with a higher levels of teamwork satisfaction and an increased perception of the positive image of the company (Eisenberger et al., 1990; Edwards, 2009; Edwards and Peccei, 2010).

Finally, the belief that one is being rewarded fairly in the organization motivates employees to work well and engage in activities without tangible rewards (e.g., extra role behaviors) (Rhoades and Eisenberger, 2002).

An increasingly important aspect central to the psychological process of the perceived organizational support theory is the perceived consideration of employee well-being and welfare.

Accordingly, a systematic review on POS conducted by Baran et al. (2012) analyzed 249 studies and found that the primary theoretical theme in association with POS was employee well-being (n = 43).

In particular, employees who experience lower levels of welfare and wellbeing might deteriorate the reciprocation psychological process when they don't feel organizationally and emotionally supported (e.g., when employees experience organizational stress).

Accordingly, social exchange theory points out that POS is negatively associated with stress (Baran et al., 2012). For instance companies who care about their employees' well-being are more likely to improve working conditions and job design, such as reducing conflicting job requirements (Jawahar et al., 2007) or bullying and incivility in workplace (Miner et al., 2012).

Additionally, POS, fulfilling socio-emotional needs, increase employees organizational membership, and role status (Rhoades and Eisenberger, 2002) and this plays a central role in developing a healthy employee–employer relationship and in reducing stress. POS fulfills emotional needs, decreases strain and thus enhances well-being (Byrne and Hochwarter, 2006).

Furthermore, several researches found an association between POS with health-related variables such as sense of accomplishment (Jain and Sinha, 2005), organization-based self-esteem (Lee and Peccei, 2007), general health (Bradley and Cartwright, 2002), decreased burnout (Kang et al., 2010), and anger (O'Neill et al., 2009).

As far as POS antecedents is concerned, job conditions (such as aspects of training, job discretion, role stressors, relations in workplace), seem to have a impact. These job characteristics overlap with the concept of stress, especially if we consider recent expanded models that go beyond job demand and job control based models (see for instance Giorgi et al., 2013). Similarly, individual antecedent such as personality factors and socio-demographic variables seemed to have a weaker relationship with POS (Baran et al., 2012).

Our theoretical framework is based on the proposition that POS is associated with job characteristics of the working environment such as stress and welfare related factors.

Recently, a further systematic study and meta-analysis were conducted on POS pointing out its antecedents. Findings of the Ishfaq and Muhammad' (2015) review work, which included 170 studies, revealed that POS was mainly linked to justice, growth opportunities, supervisor support, and coworkers support.

Kurtessis et al. (2015) performed a meta-analysis, which included 492 papers containing 558 studies, grouped antecedents of POS into three categories: (a) treatment by organization members, (b) employee–organization relationship quality, (c) HR practices and job conditions.

In particular, job conditions played a substantial role in establishing POS, with employees being inclined to perceive organizational support as a way to reciprocate adequate working conditions (Shore et al., 2006; Kurtessis et al., 2015).

In addition, Kurtessis et al.'s study posited that POS is connected to favorable social interactions, both with colleagues and supervisors, with the latter more significant because leaders are perceived as representative of the organizations (Stinglhamber and Vandenberghe, 2003). Finally, contextual factors that convey the organization's regard for employees were associated with POS: common values shared with employee, psychological contracts, fairness of treatment, and perceived organizational politics (e.g., Eisenberger et al., 2001) An organizational factor not extensively investigated by the literature in connection with POS is training.

Conversely, high organizational concern with developing employees is linked with health and business success and might stimulate in workers a general felt obligation to reciprocate toward the organization in positive manner (Eisenberger et al., 1990; Kurtessis et al., 2015).

Instead, research has more recently looked at the relationship between low or high level of POS with safety perception. Safety climate seemed to play an important role in POS (Wallace et al., 2006). Similarly, as noted by Mearns and Reader (2008), the reporting of errors of unsafe situations was associated positively POS. On the other hand, the association of POS with safety seemed to be influenced by further variables (job demand, leadership, etc.) as previously noted.

Kath et al. study (2010) suggested that safety perception, as other stress related working conditions, have likely important associations with POS but do not represent exclusive antecedents (Kurtessis et al., 2015). To sum up, various stressors and strains need to be investigated in order to delineate more clearly the combination of interrelated processes that may occur in POS.

Accordingly, although meta-analysis exists in the field of POS (Kurtessis et al., 2015), current research offers limited empirical evidence of the work-related mechanisms and psychological processes potentially interacting in relation to perceived organizational support.

Given that multiple dimensions and distinguished categories have been suggested in studying POS, we expect specific combinations of antecedents (e.g., working conditions, social support, and work pressure) that interact with one another to impact worker's perceptions. This is also in accordance with theorists who advocate that the investigations of concurrent interconnected organizational variables for the prediction of several outcomes (Scott et al., 2011).

In doing so, we use regression tree analysis to explore main perceived organizational support relationships with correlated organizational and stress factors.

Of the parametric methods used in occupational health, regression is the most common. Regression carries with it a laundry list of requirements including linearity, normality, independence of error terms, and constant variance of error terms that need to be met in order to perform well (i.e., have unbiased parameter estimates). However, it is often the case in health research and practice that these conditions are not met or is overly restrictive. A particular limitation associated with parametric models is that it is challenging to estimate and interpret interactions occurring amongst more than two variables or to model the impact of variables with non-linear functional forms (Strobl et al., 2009a).

Yet, when trying to examine outcomes such as well-being, research tends to focus on main effects, and take into account limited numbers of variables at a time.

Regression tree models are based on graphs in which each circle in the diagram reflects a binary splitting point (i.e., a splitting node) in the model. Specifically, the application of graphical models on work related stress is motivated by the possibility to describe graphically the multitude of relations and dependencies among different variables. Recursive partitioning has been used in a variety of different fields to explain drug treatment retention (Hellemann et al., 2009) survival of cancer patients (Grossman and Ram, 2014); treatment effects in clinical trials (Doove et al., 2014); clusters in genomics (Nilsen et al., 2013); intimate partner violence (Salis et al., 2014); and to help detect problem gamblers (Markham et al., 2013), among others.

The present research illustrates the value of recursive partitioning in the prediction of perceived organizational support in a sample of more than 6000 Italian bankers.

Data were collected by a team of researchers with the aim of measuring work-related stress and perceived organizational support. The survey was designed in accordance with privacy and anonymity regulations and was administered on-line through the company intranet. In particular, employees were informed of the purpose of the assessment and of the aggregate data analysis performed by qualified academic researchers. Information of data protection were presented in a precise manner, so the test-takers' security was enhanced.

The target population was an Italian national bank with multiple locations widespread in Italy. All the employees were invited to fill in the questionnaire on the bank intranet portal.

The participation for the study was considerable for on-line survey, as around 30% of employees filled in the questionnaire. Questionnaires with more than four missing items were deleted from the database in order to increase the validity of the research. The final sample consisted of 6588 bankers (of which 57.5% were males). In addition, 4.3% of the respondents were <30 years old, 26.5% were 31–40 years old, 38.7% were 41–50 years, 28.7%, were between 51 and 60 years old; 1.7% of the respondents were over 60 years old. Regarding job seniority, 9.7% of the respondents had worked for up to 7 years in their current company; 25.1% for 8–15 years, 42.3% for, 16–30 years 22.8% for more than 30 years. Finally, 91% were office workers whereas 9% were managers or middle managers who co-ordinated a team.

The Stress Questionnaire (SQ) was chosen for this investigation because it is a validated Italian tool that measures not only work-related stress but also organizational perceptions. In brief, SQ has multiple scales which investigate classical stressors as well as emergent stressors and contextual variables. All the utilized scales are part of the SQ and detailed information can be found in the test validation study (Giorgi et al., 2013). The “Psychosocial risk scale of the Stress Questionnaire” (Giorgi et al., 2013) consists of 25 items and 5 subscales: job demand, lack of job control, role conflict, lack of supervisors' support, and lack of colleagues' support. A 5 point Likert scale ranging from 1 (absolutely agree) to 5 (absolutely disagree) was used: (a) role conflict, which measures the perception of lack of awareness in their roles and responsibilities (5 items: e.g., “I have a clear idea about what is expected of me at work,” reverse scored) Alpha.84; (b) lack of colleagues' support or collaboration and support among employees (5 items: e.g., “I get the support I need from colleagues,” reverse score it's difficult to assess whether my colleagues are competent”) Alpha.79; (c) lack of supervisors' support or the extent to which employees don't experience support and understanding from their supervisors/leaders (5 items: e.g., “My supervisor energizes me at work,” reverse scored “My supervisor is neither competent nor self-confident”) Alpha.82; (d) job demands, which refers to quantitative, demanding aspects of the job (6 items: e.g., “I have unrealistic deadlines,” “I'm under pressure at work”), Alpha.81; and e) lack job control or job resources that pertain to the task (5 items: e.g., “I can plan my work,” reverse scored “I'm fully autonomous in choosing my working tools” reverse scored) Alpha.77. The factorial structure and the reliability of this scale have been supported in previous studies (Giorgi et al., 2013).

The scale “Perceived organizational support” comprises 4 items in a 5-point Likert-scale (from 1: “strongly disagree” to 5: “strongly agree”) and measures the extent to which the organization values and cares for employees' welfare (e.g., “This organization pays little attention to the interests and well-being of its employees” reverse coded—; “This organization tries to take care of its employees' welfare”), Alpha.82. Consequently, after computing a total mean score, a higher score refers to higher perceived organizational support. This scale was included in the Stress Questionnaire (Giorgi et al., 2013).

The scale “Traning” consist of 3 items in a 5-points Likert-scale (from 1: “strongly disagree” to 5: “strongly agree”) and measures the perceived quality of training in workplace (e.g., “This organization trains adequately employees for performing job tasks” reverse scored; “employees are not well trained when they have to use a new tool/equipment”), Alpha.74.

Higher scores mean that employees perceived a lack of training or low quality training. This scale is part of the Stress Questionnaire (Giorgi et al., 2013).

The scale “Reporting” measures the employees' perception of an unsafe and unhealthy environment wherein procedures of safety and health are invisible or discouraged/under-reported (“workers are not trained neither informed about job related risks; “I can freely report stress and safety risks perceptions—for me and for others” (reverse coded) Alpha.52. It was measured using a four-item scale of the Stress Questionnaire (Giorgi et al., 2013).

Descriptives, Alpha of Conbrach, and correlation were calculated.

Regression Tree analysis was used beginning with a generation of a root node. After, a splitting rule based on algorithm, determined cutoff points that minimize the within-group variance on the outcomes. Progressively further binary splits were performed in order to divide the sample into subsamples, called nodes that ended finally in terminal nodes at the bottom of the figure.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.