Servicios personalizados
Inglés (pdf)
Articulo en XML
Referencias del artículo
Traducción automática
Enviar articulo por email
Citado por SciELO 
Citado por Google 
Similares en
SciELO 
Similares en Google 
Permalink
Politics are present in every organization (Cheong & Kim, 2018; Ferris, Perrewé, Daniels, Lawong, & Holmes, 2017; Jordan, Hochwarter, Ferris, & Ejaz, 2018). Based on Lewin’s (1936) thoughts, Brouer, Harris, and Kacmar (2011) suggested that perception of organizational politics (POP hereafter), rather than de facto manifestation of politics, may affect individual and organizational outcomes. POP “involves an individual’s attribution of behaviors of self-serving intent and is defined as an individual’s subjective evaluation about the extent to which the work environment is characterized by co-workers and supervisors who demonstrate such self-serving behavior” (Ferris, Harrell-Cook, & Dulebohn, 2000, p. 90). The concept is generally perceived negative, and requires an understanding of its effect on organizational and employee outcomes (Brouer et al., 2011; De Clercq & Belausteguigoitia, 2017).
POP results in a wide array of adverse consequences such as alienation (Kumar & Ghadially, 1989), stress (Ferris et al., 1996), negligent behavior (Vigoda, 2000), organizational cynicism (Buenger, Forte, Boozer, & Maddox, 2014), decrease in organizational commitment (Chang, Rosen, & Levy, 2009), counterproductive behavior (Zettler & Hilbig, 2010; Wiltshire, Bourdage, & Lee, 2014) etc. Employee turnover intention is the most frequently studied consequence of POP (Chang et al., 2009; Landells & Albrecht, 2017; Miller, Rutherford, & Kolodinsky, 2008; Rosen, Harris, & Kacmar, 2009). Turnover intention refers to an employee’s purpose to leave her or his current organization (Bougheas, Davidson, Upward, & Wright, 2015; Saraih, Aris, Sakdan, & Ahmad, 2016). In other words, it is an individual’s perceived probability of staying or leaving the employing organization (Cotton & Tuttle, 1986). Turnover intentions lead to actual turnover (Allen, Shore, & Griffeth, 2003), which “is a serious issue for both employees and organizations” (Biron & Boon, 2013, p. 511). Therefore, any organizational phenomena that result in employee turnover intentions should be considered important.
Existing research has explained a robust relationship between POP and turnover intentions (Chang et al., 2009; Ferris et al., 2017; Rosen et al., 2009). However, relatively less attention has been paid to examine the mechanisms through which POP affect turnover intentions (Chang et al., 2009). Previously, some mediators of this relationship—such as perceived organizational support (García-Chas, Neira-Fontela, Varela-Neira, & Curto-Rodríguez, 2018; Harris, Harris, & Harvey, 2007), work engagement (Karatepe, 2013), job satisfaction (Poon, 2003; Rosen et al., 2009), exhaustion and cynicism (Huang, Chuang, & Lin, 2003), strain and morale (Chang et al., 2009)—have been examined in a variety of studies.
Job strain is a well-recognized mediator of POP-turnover intentions relationship (Chang et al., 2009). Among various forms of work strain (i.e., exhaustion, tension etc.), job anxiety is a widely discussed consequence of POP (Cropanzano, Howes, Grandey, & Toth, 1997; Ferris, Adams, Kolodinsky, Hochwarter, & Ammeter, 2002), and antecedent of turnover intentions (Balfour & Neff, 1993; Jensen, Patel, & Messersmith, 2013). Job anxiety is an employee’s subjective “awareness or feeling of personal dysfunction as a result of perceived conditions or happenings in the work setting” (Parker & DeCotiis, 1983, p. 161). It can be conceptualized as ‘job related feelings of anxiety’ (Parker & DeCotiis, 1983). Since job anxiety is an organization’s serious concern due to its costliness and increasing trends (McCarthy, Trougakos, & Cheng, 2016), a deeper understanding of its mediating role between POP-turnover intentions relationship is quite relevant.
Though previous research has suggested the mediating role of job anxiety in POP-turnover intentions relationship, empirical evidence is scarce. For example, Kacmar, Bozeman, Carlson, and Anthony (1999) examined ‘perceptions of politics model’ where job anxiety was shown as an intervening variable between POP and turnover intentions. Unfortunately, they neither developed argument for indirect effect nor statistically tested the mediating effect of job anxiety. Similarly, Vigoda’s (2000) model showed job anxiety as a mediator of the aforementioned relationship but did not provide theoretical foundations and empirical evidence of this mediating effect. Chang et al.’s (2009) meta-analytic work provided empirical evidence about the mediating effect of psychological strain between POP and turnover intentions. But they emphasized the need for additional evidence from the studies where POP and its outcomes are measured at different points of time. It entails to extend and replicate this mediating relationship by using more robust research designs such as longitudinal/time-lagged models.
In organizational sciences, it is well recognized that time plays a vital role in understanding the effects of organizational variables (Mitchell & James, 2001). Specifically, the passing of time is important to determine the effect of workplace stressors (such as organizational politics) on employee outcomes (Ford et al., 2014). Hence, the effect of POP on job anxiety, and subsequent effect on turnover intentions, can be better explained if these phenomena are measured at different time points. Moreover, longitudinal designs are recommended for mediational studies because cross-sectional data are not well-suited to test causal effects in mediation models (Boon, Eckardt, Lepak, & Boselie, 2018; Cole & Maxwell, 2003). On a more general level, cross-sectional tests of mediation cannot yield the same results as longitudinal tests until some specific conditions are satisfied (Cole & Maxwell, 2003; Maxwell & Cole, 2007; Maxwell, Cole, & Mitchell, 2011). For example; expecting that the coefficients of any pair of variables will remain invariant over time, and the correlations among predictor, mediator and criterion remain same at every time point (Maxwell et al., 2011). These conditions, however, seem highly unlikely to hold in most organizational phenomena. So, a longitudinal analysis of mediation processes needs attention.
Furthermore, empirical evidence on the boundary conditions of the above mentioned mediation process is scarce. A deeper understanding of mediating processes remains incomplete without knowing boundary conditions to decide when the indirect effect exists and when it does not (Hayes, 2018). Extant literature “has treated political skills as a boundary condition in stressor–strain relationships” (Munyon, Summers, Thompsoh, & Ferris, 2015, p. 167). Political skill refers to “the ability to effectively understand others at work and to use such knowledge to influence others to act in ways that enhance one’s personal and/or organizational objectives” (Ferris et al. 2005, p. 127). Studies examining the moderating effect of political skills on stressor-strain relationships are abundant (Cullen, Gerbasi, & Chrobot-Mason, 2018; Perrewé et al., 2005; Perrewé et al., 2004). However, a recent study calls for further inquiry into the buffering role of political skills in stressor-strain relationships as it did not find a significant direct effect of political skills on psychological strain (Munyon et al., 2015). So, there is a need to better understand how political skills affect the relationship between organizational stressors (i.e., POP) and employee strain or anxiety. Moreover, political skills have received relatively less attention as moderators of indirect effects. A recent research has suggested that employees’ political skills provide a boundary condition to affect the indirect effect of organizational phenomena on employee outcomes (Zhang, Song, Wang, & Liu, 2018). Given that political skills buffer the negative effects of stressors (García-Chas et al., 2018; Hochwarter, Perrewé, Meurs, & Kacmar, 2007; Munyon et al., 2015; Zellars, Perrewé, Rossi, Tepper, & Ferris, 2008), it is quite possible that the effect of POP on turnover intentions through job anxiety is affected by the level of political skills. However, this phenomenon has not been examined in previous research. Specifically, a time variant analysis of the aforementioned boundary condition is missing. So, there is a need to examine longitudinal moderated mediation model of political skills.
The purpose of this study was threefold. First, this study sought to examine the mediating role of job anxiety in POP-turnover intentions relationship. Based on the conservation of resources theory (COR theory) (Hobfoll, 1989), this study posited that POP, as an organizational stressor, tends to deplete employees’ valued resources such as time, energy, and focus (Cheng & McCarthy, 2018; Harris, Harvey, & Kacmar, 2009; Madison, Allen, Porter, Renwick, & Mayes, 1980). This resource depletion leads to emotional exhaustion and anxiety (Freedy & Hobfoll, 1994), which result in adverse employee outcomes such as job dissatisfaction and turnover intentions (Harris et al., 2009; McCarthy et al., 2016; Wright & Cropanzano, 1998). Second, this research aimed to examine how the indirect effect of POP on turnover intentions through job anxiety is moderated by political skills. Besides recognizing the resource depletion effect of organizational phenomena on employee outcomes, COR theory posits that people strive to mitigate the negative effects of resource depletion by calling on buffer-type resources available to them (Hobfoll, Halbesleben, Neve, & Westman, 2018; McCarthy et al., 2016). This idea makes a case for understanding the mitigating effect of buffer resources when people are faced with resources depletion derived from POP. Finally, given the need for determining true causal relationships in mediational models, this study strived to examine the above mentioned effects in a three wave time-lagged model.
This study contributes to existing literature in three ways. First, this study examined the mediating effect of job anxiety which has received relatively less attention as a mediator of POP-employee outcomes (Cho & Yang, 2018). Second, this study has extended POP and political skills research by investigating the buffer effect of political skills on a mediation process. Finally, this study is an addition to the scarce longitudinal moderated mediation research examining the mechanisms between POP and turnover intentions.
Theoretical Framework
Figure 1 shows our research model. The objectives of this study were achieved by developing mediation and moderated mediation hypotheses.
Job Anxiety as Mediator of POP-Turnover Intentions Relationship
It is generally believed that the most generic construction of a mediation hypothesis is based on Woodworth’s (1928) stimulus-organism-response (S-O-R) model, which describes that “an active organism intervenes between stimulus and response (Baron & Kenny, 1986, p. 1176)”. Consistent with S-O-R model, the psychological view of stress (Caplan, 1964; Lindemann, 1944) describes that psychological distress, such as anxiety (Hamer, Batty, Stamatakis, & Kivimaki, 2010), “could occur as the product of confrontation with especially stressful events” (Hobfoll, 1989, p. 514), which are likely to lead to stress responding (Hobfoll, 1989; Holmes, & Rahe, 1967).
The COR theory (Hobfoll, 1989) has emerged as a leading framework to understand responses to stressful events in work settings (Hobfoll, 2011). According to this theory, resources are those objects, personal characteristics, conditions, time, or energies that are valued by an individual (Hobfoll, 1989). The basic tenet of this theory is that “people strive to retain, protect, and build resources and that what is threatening to them is the potential or actual loss of these valued resources” (Hobfoll, 1989, pp. 513). This theory argues that people experience stress “when resources are perceived as unstable, threatened, or lost” (Tziner, Rabenu, Radomski, & Belkin, 2015, p. 208).
POP is a well-recognized organizational stressor because it creates uncertainty and ambiguity regarding organizational processes and outcomes (Chang et al., 2009; Ferris, Frink, Gilmore, & Kacmar, 1994; Vigoda, 2002), and depletes employees’ valued resources such as time, energy, and focus (Cheng & McCarthy, 2018; Harris et al., 2009; Madison et al., 1980). The COR theory implies that organizational politics is a process during which ongoing perceptions of uncertainty and ambiguity deplete employee resources. Over time, resource depletion reduces employees’ coping capacity and leads to increased job anxiety (Freedy & Hobfoll, 1994), which produces negative responses, such as job dissatisfaction and turnover intentions (Harris et al., 2009; Wright & Cropanzano, 1998).
Resource depletion, induced by POP, reduces not only the coping capacity of employees but also lowers their self-control (Gabriel, Koopman, Rosen, & Johnson, 2018). Lower self-control increases employee anxiety (Bertrams, Baumeister, & Englert, 2016), which provokes turnover intentions. In other words, the link between politics and withdrawal intentions is due to the experience of anxiety evoked by an employee’s lower self-control induced by high POP.
POP put job demands on employees’ valued resources such as time and energy (Crawford, LePine, & Rich, 2010), and the amounts of these resources used to confront politics are considered “as wasted resources, either in the effort used to engage in political behavior or through expenditure of effort on projects inappropriately given priority through political actions” (Madison et al., 1980, p. 93). Such organizational demands are likely to trigger employees’ intentions to leave due to higher levels of experienced anxiety derived from depletion or over-demand of resources (Astrachan, 2004; Chang et al., 2009; Cheng & McCarthy, 2018). This point is consistent with job demands-resource model of work stress which posits that imbalance between employees’ available resources and demands put on them creates anxiety (Bakker & Demerouti, 2007). Rapid resource depletion due to an organization’s political environment may create an imbalance in employee’s available resources and demands put on her. Based on this idea, Chang et al. (2009) described that employees are likely to look for less stressful jobs due to their experience of anxiety resulting from excessive job demands. Therefore, POP are related to employee turnover intentions because they produce feelings of anxiety which provokes intentions to leave.
In order to ascertain whether the relationship between POP and employee turnover intentions is mediated by job anxiety, the following hypothesis was examined.
Hypothesis 1: Job anxiety mediates the relationship between perceptions of organizational politics and turnover intentions.
Moderated Mediation of POP-Turnover Intentions Relationship
As already mentioned, a deeper understanding of mediating process requires to recognize the boundary conditions to decide when the indirect effect exists and when it does not (Hayes, 2018). So, considering boundary conditions would provide a more profound understanding of the mediation process explained above. Although “an active organism intervenes between stimulus and response” (Baron & Kenny, 1986, p. 1176), one should not ignore that “there are many contradictory and conflicting forces acting upon any organism” (Etkin, 1967, p. 116). Hence, the the intensity of an environmental stimulus’ effect on an organism will change when a contradictory force is in place. It may change the intensity of an organism’s response too. The COR theory suggests that under a stressful situation of resource loss, people tend to employ other resources to buffer or offset this loss (Hobfoll, 1989). The use of other resources may act as a contradictory force to buffer the effect of environmental stimuli on the organism. The resource loss neutralization resulting from the use of buffer resources, in fact, is an indirect conservation of the resources which might have been lost in the absence of buffer resources. The COR theory further describes that resources are not equally distributed (Hobfoll, 1989). So, the adverse effect of an environmental stimulus on an organism will be affected by the level of coping resources possessed by an individual. It suggests that the interaction of an environmental stimulus with a contradictory force acting upon an organism may change the intensity of stimulus-response relationship by the level of that contradictory force.
From the COR theory’s perspective, political skills are employees’ personal resources (Ferris et al., 2007; Hobfoll, 1989), which can be used to leverage other personal and social resources (Bentley, Treadway, Williams, Gazdag, & Yang, 2017; Hochwarter, Witt, Treadway, & Ferris, 2006). Since politics is highly embedded in an organization’s social environment (Cheong & Kim, 2018), politically skilled employees are expected to experience lesser anxiety in relation to those who are low in these skills because these skills allow them to develop social networks and exert interpersonal influence (Ferris et al., 2005). Politically skilled employees are able to effectively evaluate social cues, align their behavior with work environment, and succeed in becoming part of cooperative and resource rich social networks which are considered as employees’ social resources (Bentley et al., 2017). These resources help employees reduce feelings of job anxiety and its subsequent effect on turnover intentions (Argyle, 2013).
Political skills provide employees with greater coping capability and self-regulation which help restore lost resources or acquire new resources (Hobfoll, 1989; Perrewé et al., 2004; Perrewé et al., 2005; Ferris et al., 2007). Politically skilled individuals have greater control and understanding of the phenomena around them (Kimura, 2013) and are less likely to experience anxiety from their POP because they are able to “avoid aspects of work that represent potential stressors” and “perceive environmental stressors as less threatening” (Munyon et al., 2015, p. 8).
In addition, social astuteness and greater adaptability of politically skilled individuals provide them with social competence which helps them better meet the demands that emerge from an organization’s political environment (Cantor & Kihlstrom, 1987; Ferris et al., 2007). This competence allows them to avoid feelings of anxiety derived from the loss of resources due to political environment (Cheng & McCarthy, 2018; Ferris et al., 2007). So, employees’ political skills, when they interact with their POP, reduce their feelings of anxiety and subsequent outcomes (i.e., turnover intentions). Conversely, people low in political skills are vulnerable to experience resource loss and psychological distress when they perceive politics in their organizational environment (Perrewé et al., 2004).
This study argues that political skills reduce turnover intentions by weakening the effect of POP on feelings of job anxiety. In other words, differences in employees’ political skills matter for determining the effect of POP on turnover intentions through job anxiety. So, the anxiety enhancing effect of POP is reduced by higher political skills. Therefore, it can be expected that employees’ political skills will subvert the effect of POP on job anxiety and, in turn, reduce turnover intentions.
The above discussion suggests that the detrimental effect of POP on an employee’s job anxiety reduces when POP interact with that employee’s political skills. In other words, differences in political skills matter for establishing the effect of POP on job anxiety, and subsequent effect on turnover intentions. This leads us to formulate the following hypotheses.
Hypothesis 2a: Political skills moderate the relationship between perceptions of politics and job anxiety, such that the positive relationship between perceptions of politics and job anxiety is weaker when political skills are high. On the other hand, the positive relationship between perceptions of politics and job anxiety is stronger when political skills are low.
Hypothesis 2b: Political skills negatively moderate the indirect effect of perceptions of politics on turnover intentions through job anxiety.
Method
Research Design
This study used a longitudinal design for examining mediation and moderated mediation in three waves panel data. Mediation models involve causal relations where a predictor affects a criterion variable through a mediator (i.e., POP → Job anxiety → Turnover intentions). In such models, “a fundamental requirement for one variable to cause another is that the cause must precede the outcome in time” (Cole & Maxwell, 2003, p. 559). The basic question in longitudinal studies is “has there been any change over a period of time?” (Bouma & Atkinson, 1995, p. 114; cited in Saunders, 2011). In mediation models, this question can be answered by considering appropriate time lags in data collection even when actual manipulation of variables is impractical (Cole & Maxwell, 2003). The availability of three wave panel data allowed us to analyze how employees’ POP preceded turnover intentions over time through job anxiety. Similarly, we were able to examine the moderating effect of political skills on the aforementioned indirect effect.
Sample and Procedures
The respondents were full-time employees from private commercial banks in two districts of South Punjab (Pakistan). Banking has remained a sector of interest for organizational politics researchers (Sowmya & Panchanatham, 2017; Zhang & Yang, 2012). Owing to its very nature of using humans as the largest input, the banking sector is quite relevant for studying organizational politics and its effect on employee outcomes (Sowmya & Panchanatham, 2011). Pakistan’s banking sector is good enough to play a positive role in the country’s economic growth (Shahid Saeed, & Tirmizi, 2015). Specifically, banking has experienced sustainable growth with the emergence of China Pakistan Economic Corridor (CPEC) related activities (Pakistan Economic Survey, 2017). Pakistani banks have centralized recruitment systems and well developed organizational structures which are run by qualified bankers (Khan, Safwan, & Ahmad, 2011). Employees in these organizations have sufficient level of education to understand research surveys.
Using insights from previous research on organizational politics (Gandz & Murray, 1980; Hochwarter, Witt, & Kacmar, 2000; Zivnuska, Kacmar, Witt, Carlson, & Bratton, 2004), lower and middle level employees were chosen as the study population. The reason for targeting lower and middle level employees was that they are more susceptible to organizational politics as compared to their counterparts at higher level (Drory, 1993; Parker, Dipboye, & Jackson, 1995; Rahman, Hussain, & Haque, 2011). As this study sought to examine the negative consequences of POP (i.e., job anxiety and turnover intentions), selecting lower and middle level employees was quite appropriate.
The sample size was determined by performing a priori power analysis in G*Power software (Faul, Erdfelder, Buchner, & Lang, 2009). The use of G*Power is recommended in the context of PLS-SEM (Hair, Hult, Ringle, & Sarstedt, 2017). The results of power analysis suggested a sample size of 316 or greater for 0.90 statistical power and a large effect size of 0.025 (Aguinis, Beaty, Boik, & Pierce, 2005; Hair et al., 2017; Kenny, 2016).
Data were collected in 3 waves because a good mediation analysis in time lagged models requires at least 3 waves data from the same respondents in all waves (i.e. a three-wave panel) (Cole & Maxwell, 2003; Maxwell & Cole, 2007). For this purpose, a large sample size was targeted because most respondents “who complete the first wave of the survey fail to participate in subsequent waves” (Hillygus & Snell, 2015, p. 1). Moreover, a low response rate has been observed in existing research (Boon, Den Hartog, Boselie, & Paauwe, 2011). So, the target pool of respondents was three times larger than the size recommended in power analysis. At the time of survey, 1,452 lower and middle level employees were working in 158 bank branches of the target districts. A simple random sampling (SRS) strategy was used to select a sample of 950 employees. SRS is a probability sampling technique which is useful for reducing biases in sample selection (Haider, Fatima, Bakhsh, & Ahmed, 2019; Srinagesh, 2006), and enables researchers to assert “how confident he/she is that the research results reflect the situation in the underlying population” (Lohr, 1999; Reynolds, Simintiras, & Diamantopoulos, 2003, p. 88).
A six month time lag was used between each wave. The respondents were exactly the same in all 3 waves. A distinct code was assigned to each individual employee so that the responses of all three waves could be matched. All waves of the survey obtained employees’ self-ratings about all the study variables. The control variables (gender and education) were surveyed only in the first wave. After looking for missing values and matching the responses in preceding waves, 734, 496, and 347 responses were recorded from waves 1, 2, and 3, respectively. The overall response rate from the first survey to the final sample was 36.5%. The characteristics of the final sample are shown in Table 1. Gender was coded as 1 = male and 2 = female, and education was coded as 1 = Secondary School, 2 = Intermediate, 3 = Graduation, 4 = Masters, 5 = M.Phil.
One may note from Table 1 that respondents’ age reflects a relatively young sample. There are two major reasons for this young sample. First, respondents were lower and middle level employees, and typically employees at these levels are younger than their higher level counterparts. Second, Pakistan is a remarkably young country and has a relatively young labor force (Pakistan Bureau of Statistics, 2015; Pakistan Economic Survey, 2017). So, a young sample is quite normal.
Measures
Data were collected by using survey instruments already used in existing research (see Appendix). The instruments were used in their original form and language (i.e., English language). The target population had a sufficient level of education to understand the questionnaire. Any issues in understanding the questionnaire were solved by the researchers in follow-up visits. For all the questionnaires, a five point Likert scale was used which ranged from 1 = strongly disagree to 5 = strongly agree. Perceptions of organizational politics were measured by a 6-item scale used in Hochwarter, Kacmar, Perrewe, and Johnson (2003). A six-item scale was used to measure political skills (Ahearn, Ferris, Hochwarter, Douglas, & Ammeter, 2004). Dysvik and Kuvaas’ (2008) five item scale was used to measure employee turnover intentions. Job anxiety was measured by using four items from Parker and DeCotiis’ (1983) 13-item scale developed to measure two dimensions of job stress: ‘job-related feelings of anxiety’ (5 items) and ‘feelings of being under substantial time pressure’ (8 items). This study used four out of five anxiety related items. Insights from previous studies published in well reputed journals suggest that the four items selected for this study best capture the notion of job anxiety (Glazer & Beehr, 2005; Glazer & Kruse, 2008; Vanderpool & Way, 2013). Reliability results of all the questionnaires have been reported in Table 2.
Table 2. Evaluation of Measurement Model
1Factor/outer loadings;
2Cronbach’s alpha;
3CR = composite reliability;
4AVE = average variance extracted.
Employees’ self-ratings were obtained for measuring all study variables. Single-source data may create the issue of common method variance (CMV; Podsakoff, MacKenzie, Lee, & Podsakoff, 2003). CMV refers to the “variance that is attributable to the measurement method rather than to the constructs the measures represent” (Podsakoff et al., 2003, p. 879). Although it is less likely in time-lagged data (Podsakoff et al., 2003), CMV was assessed by employing a PLS-SEM based full collinearity test proposed by Kock and Lynn (2012) and Kock (2015). This test is used to generate variance inflation factor (VIF) for all variables in a model acting as predictors of a ‘dummy’ criterion variable (Kock & Lynn, 2012). VIF values lower than 3.3 indicate the absence of CMV (Moqbel & Kock, 2018). Using procedures explained in Kock and Lynn (2012), full collinearity VIF (FVIF) values were obtained for all latent constructs in the model. Table 6 shows that VIF values for all latent construct are lower than 3.3. So, CMV was not an issue in this study.
Control variables. The effect of gender and education was controlled as these variables are likely to influence turnover intentions (Blomme, Van Rheede, & Tromp, 2010; Gardner, Wright, & Moynihan, 2011; Kim & Chang, 2014; Lee, Allen, Meyer, & Rhee, 2001.
Analytical Approach
Partial least squares structural equation modeling (PLS-SEM) was used to perform mediation and moderated mediation in three-wave longitudinal autoregressive models (Cole & Maxwell, 2003; Maxwell et al. (2011). PLS-SEM estimates were obtained from the latest version of SmartPLS 3 (3.2.7) software (Ringle, Wende, & Becker, 2015). The PLS approach to SEM is a component based procedure of estimation with iterative algorithms of least squares regressions (Haider, Fernandez-Ortiz, & de Pablos Heredero, 2017; Haider, Jabeen, & Ahmad, 2018; Hair, Hult, Ringle, & Sarstedt, 2014). Hypotheses are tested based on resampling method—bootstrapping. This variance based SEM is suitable for theory development and prediction of constructs (Hair et al., 2017). It is considered as an efficient tool for data analysis because it allows simultaneous estimation of item loadings and path coefficients, minimizes biases, and reduces measurement error (Hair et al., 2017; Mishra, Sharma, & Swami, 2016). PLS-SEM, when compared with covariance based SEM (CB-SEM), is recommended as a preferred technique in social sciences research because it is capable to handle small sample sizes and exempts the assumption of normality in data (Astrachan, Patel, & Wanzenried, 2014).
A PLS path model consists of two elements: measurement model and structural model (Hair et al., 2017). The measurement model validates data and the structural model tests the significance and relevance of hypothesized relationships.
Results
Evaluation of Measurement Model
Factor loadings of individual items and average variance extracted (AVE), composite reliability, and Cronbach’s α for the latent variables have been shown in Table 2. Overall, these values indicate that our data are valid and reliable at item and construct level. One item of turnover intentions (TOI-5) was deleted from analysis because of its factor loading being below .40 in all waves. Items T1-TOI4, T2POP1, and T2-POP2 were retained with their respective constructs despite their factor loadings being below .708. According to Hair et al. (2014, p. 107), “indicators with outer loadings between .40 and .70 should be considered for removal only if the deletion leads to an increase in composite reliability and AVE above the suggested threshold value”. We performed this operation, and deleting T1-TOI4, T2POP1, and T2-POP2 did not cause any increase in composite reliability and AVE. So, these items were retained as indicators of their respective constructs.
In addition to above validity measures, discriminant validity was established to assess that the measures of one construct do not correlate with other constructs (Ringle, Sarstedt, & Mooi, 2010). As a tradition, discriminant validity is evaluated by using cross-loadings and Fornell and Larker’s (1981) criterion. But these methods are insufficiently sensitive to detect discriminant validity. Henseler, Ringle, and Sarstedt (2015) introduced a more sensitive new criterion, heterotrait-monotrait ratio of correlations (HTMT), for measuring discriminant validity. We used this new criterion for establishing discriminant validity between constructs. Using more conservative approach (considered as the strictest criterion), HTMT value between two constructs must be less than .85 (HTMT.85). Table 3 shows that all HTMT values between constructs are below .85, except for constructs T3-TOI and T1-JA. However, HTMT values above .85 but below .90 is acceptable in a more liberal criterion (HTMT.90) (Henseler et al., 2015). So, in our model discriminant validity has been established based on HTMT.85 and HTMT.90 criteria.
Testing for Measurement Invariance
Longitudinal data are not considered valid for obtaining unbiased results over time until they are tested for measurement invariance (Cole & Maxwell, 2003). Measurement invariance determines “whether or not, under different conditions of observing and studying phenomena, measurement models yield measures of the same attribute” (Horn, & McArdle, 1992, p. 117). So, measurement invariance was tested by using a latest PLS based approach named as measurement invariance of composite models (MICOM) (Henseler, Ringle, & Sarsted, 2016). According to Henseler et al. (2016), “all variance-based SEM techniques model latent variables as composites” (p. 408). So, MICOM is quite suitable for testing measurement invariance in our reflective measurement model validated above.
Based on Henseler et al. (2016) and Matthews (2017), a three step MICOM process was used to test measurement invariance in our model. First, the configural invariance was followed, i.e., indicators, data treatment, and algorithm settings were identical in all waves of data collection and measurement. Second, compositional invariance was tested, which means that original correlation should be greater than or equal to 5% quantile. This test is performed by using permutation process in PLS-SEM. Table 4 shows satisfactory results obtained from 5,000 permutations.
Third, composite equality was established; if the mean original difference and the variance original difference (in parenthesis) fall between 2.5% and 97.5%, full invariance is established, and if only one of these variances falls between 2.5% and 97.5%, partial invariance is established. Table 5 shows that full invariance has been established in our longitudinal data.
Evaluation of Structural Model
Hair et al. (2014) suggested that collinearity between each set of predictor variables must be checked before hypotheses testing1. A frequently used measure of collinearity is variance inflation factor (VIF). Its value should be 5 or lower. The SmartPls results in Table 6 indicate that all VIF values are below 5, indicating the absence of collinearity among the predictors.
Table 6. Collinearity Assessment (inner VIF values)
Note.The values in this table have been provided from the last step in multiple linear regression analyses for testing moderated mediation. T1 = time 1; T2 = time 2; T3 = time 3; TOI = turnover intentions; POP = perceptions of politics; JA = job anxiety.
Mediation test (hypothesis 1). Figure 2 shows the estimated 3 wave autoregressive mediation model. This model has been constructed based on the procedures for longitudinal mediation explained in Cole & Maxwell (2003) and is analogous to a two unit time lag model explained in Maxwell et al. (2011, p. 822). In this model, the indirect effect of POP on TOI takes two units of time, one for POP to affect JA and another for JA to affect TOI. According to Maxwell et al. (2011), “if the indirect effect takes two units of time, an argument could be made for expecting the direct effect to take two units as well”. This argument justifies a 3 wave time-lagged model where the direct effect of independent variable on dependent variable takes two units of time (Maxwell et al., 2011).
Note. T1 = time 1; T2 = time 2; T3 = time 3; TOI = turnover intentions; POP = perceptions of politics; JA = job anxiety. *p lt; .05, **p lt; .01, ns = non-significant.
Figure 2. Estimated PLS longitudinal path model for mediation.
Insights from Cole and Maxwell (2003) and Maxwell et al. (2011) suggest that in a three wave time-lagged mediation model, the effect of time-1 predictor on time-2 mediator and subsequent effect on time 3 outcome variable (controlling for time 1 mediator, and time 1 and time 2 outcome variable) provide the best measures to avoid the issue of half longitudinal design. So, the effect of time 1 perceptions of politics (T1-POP) on time 3 turnover intentions (T3-TOI) through time 2 job anxiety (T2-JA) was tested, controlling for prior levels of job anxiety (T1-JA) and turnover intentions (T1-TOI and T2-TOI).
The mediation analysis for our autoregressive model was performed by using the latest mediation procedures explained in Hair et al. (2017). Based on Zhao, Lynch, and Chen (2010), Nitzl, Roldan, and Cepeda (2016), and Hair et al. (2017) suggested a two-step process for testing mediation in a model. The first step involves testing for the indirect effect of independent variable on dependent variable through the mediator. The significant indirect effect provides support for mediation hypothesis. The second step addresses the significance of the direct effect of independent variable on dependent variable. The insignificant direct effect, along with the significant indirect effect, suggests that the “mediator fully complies with the hypothesized theoretical framework” (Hair et al., 2017, p. 234).
Following the above mentioned procedures, the indirect effect and its significance were tested by using product of coefficients and bootstrapping procedures (Hair et al., 2015; Preacher, Rucker, & Hayes, 2007). These procedures are consistent with Cole and Maxwell (2003) who suggested that, for mediation to exist, the necessary (as well as sufficient) condition is that the indirect effect—the product of the coefficients of paths from time 1 predictor (T1-POP) to time 2 mediator (T2-JA) to time 3 dependent variable (T3-TOI)—must be non zero. In our model, the indirect effect was .056 [0.12*0.47]. Bootstrapping with 5,000 samples yielded a t-value = 2.36, which represents a significant indirect effect at p lt; .05.
The second step concentrates on testing the significance of direct effect of T1-POP on T3-TOI. As shown in Figure 2, the direct relationship between T1-POP and T3-TOI is weak (.008). The t-value from bootstrapping procedures shows that this relationship is nonsignificant (t = 0.22, p-value = .83). Insights from existing research suggest that “the absence of a direct effect after controlling for an initial mediator should not lead to conclusions of ‘full’ mediation… [The] exploration of mediation should be guided by theory” (Rucker, Preacher, Tormala, & Petty, 2011, p. 368). Therefore, it cannot be concluded that mediation is ‘full’ in our model. Previous research has found various mediators of POP-turnover intentions relationship (Chang et al., 2009; Huang et al., 2003). Based on Rucker et al. (2011), this study states that a mediator (job anxiety) has been documented between POP and turnover intentions, and the possibility of any additional mediators cannot be omitted. Accordingly, the significant indirect effect indicates that job anxiety mediates the POP-turnover intentions relationship (Hypothesis 1 supported).
Moderated mediation test (hypothesis 2). Moderated mediation was tested by extending Hayes’ (2015) first stage cross-sectional model to a 3 wave autoregressive mediation model (Figure 3). In Figure 3, dotted lines represent the optional moderating effect which can be included without disrupting the main moderated mediation effect (Hayes, 2015).
T1 = time 1; T2 = time 2; T3 = time 3; TOI = turnover intentions; POP = perceptions of politics; JA = job anxiety. *p lt; .05, **p lt; .01, ns = non-significant.
Figure 3. Estimated PLS Longitudinal Path Model for Moderated Mediation.
In a moderated mediation model, it is a tradition among researchers to test the significance of the moderating effect on the predictor-mediator relationship (Hair et al., 2017; Muller, Judd, & Yzerbyt, 2005; Preacher et al., 2007). In Figure 3, path, a3 shows a significant negative moderating effect on the relationship between time 1 perceptions of politics and time 2 job anxiety (β = -.13, t-value = 2.94, p lt; 0.01); t-value was obtained by using bias-corrected bootstrap confidence interval method with 5,000 samples in SmartPLS.
This significant moderating effect can be interpreted by using medium level of moderator (political skills) as a reference point (Hair et al., 2014). At a medium level of employees’ political skills, the relationship between time 1 perceptions of politics and time 2 job anxiety is .087 (simple effect). If employees’ political skills increase by one standard deviation point, the relationship between perceptions of politics and job anxiety would decrease by the size of interaction term’s coefficient, and obtains a value of .087 + (-.13) = -.043. Accordingly, with an increase in employees’ political skills, perceptions of organizational politics become less important for explaining job anxiety. These results support our Hypothesis 2a.
Figure 4 shows the interaction chart for low and high (±1 standard deviation) levels of political skills. The central (blue) line shows political skills at mean, which reflects no effect of political skills. At this level of political skills, the relationship between T1-POP and T2-JA is positive. This relationship becomes stronger with one standard deviation decrease in political skills (red line). However, one standard deviation increase in political skills changes the direction of relationship from positive to negative, as the green line shows.
Although the above mentioned moderating effect is significant in our model, a nonsignificant moderation “does not imply the indirect effect is not moderated by” political skills because this effect “does not quantify the relationship between the moderator and the indirect effect” (Hayes, 2015, p. 9). Hayes recommends testing the significance of moderating effect on the indirect effect as a whole rather than testing an isolated moderating effect on the independent variable-mediator relationship.
According to Hayes (2015), the indirect effect in a moderated mediated model is the product of conditional effect on the mediator (path a3), and the mediator’s effect on the criterion (path b), controlling for predictor. In our model, this effect is significant (β = -.13 * .42 = -.055, t-value = 2.66, p lt; .01). ). Statistical significance of this effect (t-value) was tested by using bias-corrected bootstrap confidence interval method with 5,000 samples in SmartPLS, instead of using the traditional Sobel’s (1982) test which has low statistical power (Salgado, Blanco, & Moscoso, 2019). The above mentioned significant indirect effect (-0.055) is called Hayes’ (2015) index of moderated mediation, which “is a direct quantification of the linear association between the indirect effect and the putative moderator of that effect” (p. 3). Equation (1) shows Hayes’ linear function of moderator adapted for this study. In this equation, the intercept is the indirect effect of the predictor on the criterion through the mediator (the product of paths a1 and b), and the slope is the indirect effect of the interaction term on the criterion through the mediator (the product of path a3 and b).
A visual representation of this linear function can be obtained by using the values of intercept and slope in above equation, at different values of the moderator. In this study, the intercept is .037 (.087 * .42) and slope is -.055. The linear function in Figure 5 was obtained by using arbitrary values of moderator from – 5 to 5. The negative slope of the line (or the index of moderated mediation) indicates that the indirect effect of perceptions of politics on turnover intentions through job anxiety seems to decrease with increasing political skills.
Overall, the results indicate that political skills moderate the relationship between perceptions of politics and job anxiety (Hypothesis 2a). In addition, a moderated mediation pattern exists in our model, supporting our Hypothesis 2b, i.e., the indirect effect of perceptions of politics on turnover intentions via job anxiety is negatively moderated by political skills. The results confirm that political skills reduce employees’ turnover intentions by changing influence of perceptions of politics on job anxiety.
Determining the effect size of the moderator. The estimation of Cohen’s (1988) effect size f2 is highly recommended in moderation models (Hair et al., 2017). This effect is used “to assess the relative impact of the predictor construct on an endogenous construct” (Hair et al., 2017, p. 317). For a specific construct, the effect size f2 can be calculated by using the R2 values of endogenous variables in the following formula when that specific construct is included in and excluded from the model.
For example the effect size of the moderator (political skills) on mediator (job anxiety) and criterion (turnover intentions) can be obtained by using R2 values of these constructs from mediation and moderated mediation models by using the above formula, f2 values of political skills as .04 and .09 for job anxiety and turnover intentions, respectively. General guidelines for effect size assessment suggest that values .02, .15, and .35 represent low, medium and high effect size, respectively (Cohen, 1988). However, recent developments indicate that .005, .01, and .025 are more realistic values for small, medium, and large effects in moderation tests (Aguinis et al., 2005; Hair et al., 2017; Kenny, 2016). Based on this illustration, the moderator’s (political skills) effect on job anxiety (.04) and turnover intentions (.09) is large. It indicates that political skills contribute much to explain the effect of politics perceptions on job anxiety and turnover intentions.
Monte Carlo power analysis for indirect effects. The statistical power of indirect effects in mediation and moderated mediation models was assessed by performing a “Monte Carlo power analysis for indirect effects” (Schoemann, Boulton, & Short, 2017; Thoemmes, MacKinnon, & Reiser, 2010). This analysis is performed in a computer based application and requires mainly the following inputs to calculate the power of indirect effect; the sample size used in the study, standard deviation of latent constructs, and correlations between study variables. This information was put in appropriate boxes of the Monte Carlo application by using 2,000 replications, 20,000 Monte Carlo draws per replication, and 95% confidence level. The application generated a power of 0.92 and 0.84 for indirect effect in the mediation and moderated mediation models, respectively. This power is above the desired level of 0.80 (MacCallum, Browne, & Sugawara, 1996). Therefore, it can be stated that the indirect effect in the mediation and moderated mediations models is statistically powerful. For details on these calculations, the readers are referred to Schoemann et al. (2017).
Discussion
This study examined why and how perceptions of organizational politics affect turnover intentions. Data from 347 bank employees confirmed our research model. To our knowledge, this is the first study in which POP and political skills together predicted job anxiety and subsequent effect on turnover intentions. The results indicate that POP tend to increase job anxiety because a political “work environment is often less helpful and more threatening” (Cropanzano et al., 1997, p. 175). Conversely, political skills were negatively related to job anxiety because these skills provide employees with a greater self-regulation and coping capability against threats (Ferris et al., 2007; Hobfoll, 1989; Perrewé et al., 2004; Perrewé et al., 2005). The indirect effect of POP on turnover intentions through job anxiety was significant, and this effect was moderated by political skills. This study enhanced our understanding of the mechanisms that relate POP to turnover intentions. The findings suggest that job anxiety and the subsequent intentions to leave are minimized when employees’ perceptions of politics are coupled with their political skills.
This study contributes to existing literature in three ways. First, this study is an addition to the scarce research examining the mediating role of job anxiety between POP and employee outcomes (Cho & Yang, 2018). The results support not only the mediation hypothesis but also confirm the untested mediation of job anxiety between POP-turnover intentions relationship proposed by Kacmar et al. (1999) and Vigoda (2000). The findings of this study are consistent with previous research examining the psychological mechanisms (i.e., strain, burnout etc.) between POP and turnover intentions (Chang et al., 2009; Huang et al., 2003), and respond their call for testing this link at different time points (i.e., in a time lagged model).
Second, this study has examined the effect of political skills and POP interaction (i.e., political skills*POP) on turnover intentions. Although the research on the moderating effect of political skills is growing, an examination of political skills’ interaction with different organizational phenomena and the effect of this interaction on employee outcomes is still a fertile area of research (Ferris et al., 2017). Studies examining the effect of political skills and POP interaction on employee outcomes are limited, and provide inconsistent findings. For example, the effect of this interaction on employee’s job satisfaction, job performance, manager-rated commitment, and affective commitment (Kimura, 2013) was nonsignificant in Brouer et al. (2011) and Kimura (2013). However, Kapoutsis, Papalexandris, Nikolopoulos, Hochwarter, and Ferris, (2011) found a significant negative effect on job performance. These contradictory findings required further research for achieving greater consistency. This study has found a significant negative effect of the political skills*POP interaction on job anxiety (Hypothesis 2a was supported), but a nonsignificant effect on turnover intentions (optional in this study). The results of this study including those of the previous research indicate contradictory findings and open up the issue for further discussion.
Third, this study extends existing research by examining the moderating effect of political skills on the indirect effect of politics perceptions on turnover intentions through job anxiety (Hypothesis 2b). Existing literature has recognized that understanding mediating mechanisms without considering boundary conditions provides limited insights into the scientific inquiry of a causal process (Hayes, 2018). A significant indirect effect was found in the moderated mediation model (Hypothesis 2b supported). The results indicate that, to a greater extent, political skills can protect employees from the adverse effect of POP on job anxiety, and subsequent effect on turnover intentions. This findings of this study suggest that mediational studies should consider boundary conditions (i.e., individual differences, context etc.) to explain whether an indirect effect “exists or not, or is strong versus weak, or positive versus negative” (Hayes, 2018, p. 4).
Finally, this study has examined POP’s positive effect (arising from the threat of losing valued resources) and political skills’ negative effect (emerging from the possession of coping or buffer resource) on job anxiety. Empirical examination of contradictory forces on an organism has value because contradiction represents “a process of change and development, and that the motion and change of things is the outcome of the contradictions which are inherent in them” (Sayers, 1982, p. 409). Testing the effect of conflicting forces on job anxiety provides useful insights into an under researched area of ‘resource utilization’ and ‘resource passageways’ in COR theory (Halbesleben, Neveu, Paustian-Underdahl, & Westman, 2014). People are motivated to conserve resources and utilize these resources when the situation demands their use (Hochwarter et al., 2006). This study describes a situation where the threat of resource loss resulting from POP provides a passageway to utilize conserved resources such as political skills. This idea is consistent with the notion of needs in motivation theory which describes that resources are valued for and invested in the goals where the need for those resources arises (Halbesleben et al., 2014; Locke & Latham, 1990). Future research may examine the amount of conserved resources utilized when demand for their use arises.
Although a non-political work environment is important for having a healthy and satisfied workforce (Cropanzano et al., 1997), politics is a fact which cannot be ignored (Cheong & Kim, 2018; Ferris et al., 2017). Previous research informs that emotional intelligence is positively related to increased political skills (Meisler, 2014) and these skills can be improved through training (Slaski & Cartwright, 2003). Organizations can support employees to improve their political skills by providing them with training on emotional intelligence. It may be important, especially in case of reducing job anxiety and turnover intentions of those employees who are important for organizational performance and competitiveness.
Despite a handsome contribution to human resource management literature, this study also has some limitations. First, the study sample was from the banking sector of Pakistan, and this setting may be distinct enough to limit the external validity of our results. Although the procedures adopted in this research allow us to generalize findings across the study population, future researchers may extend the results in different contexts and organizational settings, across employees at different organizational levels, and demographically diverse samples. Despite the potential concerns that may arise about its generalizability or external validity, this study provides the researchers with a causal explanation of the mechanisms through which, and the conditions under which, perceptions of organizational politics lead to employee turnover intentions. It is valuable for the reason that “causal explanation is an important route to the generalization of causal descriptions because it tells us which features of the causal relationship are essential to transfer to other situations” (Shadish, Cook, & Campbell, 2002, p. 10). If future research supports our findings, then this study would be able to provide a vigorous process to apprehend the dual impact of politics perceptions and political skills on employee outcomes through psychological mechanisms. So, we recommend future researchers to amplify the scope of this research by conducting replication-extension studies as they would provide “a more precise estimate of some effect size measure”, increase “the generalizability of statistical results”, and “may advance non-statistical argument to generalize findings from larger populations of interest” (Bonnet, 2012, p. 409).
Second, single-source data might have created common method bias (Podsakoff et al., 2003). However, a PLS-SEM based full collinearity test was performed (Kock, 2015; Kock & Lynn, 2012) and no such bias was found. Future researchers may extend this research by using multi source data, where possible.
Third, job anxiety was measured by using four out of five items from Parker and DeCotiis’ (1983) measure of ‘work related feelings of anxiety’. Although previous research has used these four items to measure job anxiety in business/organizational settings and got the standards of construct validity, the items do not measure anxiety in the clinical sense. Job anxiety could be measured by using more than one measure or Beck Anxiety Inventory (Beck, Epstein, Brown, & Steer, 1988) measure, which is well recognized in both clinical and other settings. Future researcher may extend and replicate this study by using more than one measure or Beck Anxiety Inventory as a more comprehensive measure of anxiety.
Fourth, though theory and evidence support our theoretical model, other possible illustrations may emerge. For example, employees with higher feelings of anxiety (or intentions to leave) may be more likely to perceive their organizations’ environment as political. One recommendation for future research is to initiate and examine a more comprehensive construction of the possible predictors of job anxiety and turnover intentions, and to also discover the level of mutual cause effect relationships. Finally, this study did not provide an analysis of the modeled relationships across different cultural contexts. Future studies may replicate and extend this work by considering how the consequences of perceived organizational politics in Pakistani context contrast with other cultural contexts.
Despite these limitations, our study provides a substantial contribution to existing body of knowledge as it developed a moderated mediation model for answering the question of how and why perceptions of organizational politics affect turnover intentions, and how employees’ political skills are contingent upon this relationship by reducing their job anxiety.
