Employee voice as a behavioural response to psychological contract breach: The moderating effect of leadership style

Abstract Empirical evidence shows that psychological contract breach (PCB) leads to negative work behaviours of employees, including withholding of discretionary activities such as employee voice (EV). This research aims to determine empirically how PCBs are linked to different types of EV, and how different leadership styles affect these relationships. The paucity of literature on the relationship between all three variables necessitated this research. The study targeted medium to large South African organisations with more than 60 employees. The population sample was representative of a broad range of South African employees. This research adopted a crosssectional survey design, whereby the respondents were asked to answer a questionnaire about PCB, leadership styles and EV. Correlation analyses were used to test the direct links between variables and regression analyses to test for the moderation effect of leadership styles on the PCB–EV link. The data were collected from 620 respondents from 11 organisations. All the instruments showed acceptable psychometric properties. Three findings were dominant: PCB correlated negatively with promotive types of EV and positively with prohibitive types of EV; leadership styles were a weaker predictor of EV than PCB; and the PCB–EV relationship was, in most cases, partially moderated by leadership styles. PCB and leadership styles influence EV; however, leadership styles only partially influence the PCB–EV relationship. Applying a specific leadership style to influence EV under conditions of PCB is partially effective. Managers should circumvent PCB and focus on the fulfilment of PC, as this would elicit promotive EV and lessen prohibitive EV.

Psychological contract (PC), as a driver of employees' motivation and cooperation (Cullinane & Dundon, 2006), when honoured, is linked to helpful employee behaviours and constructive employment relations (Tekleab et al., 2020). However, unfulfilled or breached PCs lead to a multitude of negative work outcomes (see meta-analysis by Zhao et al., 2007) and compromise employee wellbeing (Gulzar et al., 2021). Extant research provides convincing empirical evidence that PCB relates negatively to employees' work engagement (Agarwal, 2014), affective commitment (Rigotti, 2009), in-role performance (Hartmann & Rutherford, 2015) and organisational citizenship behaviour (Lu et al., 2015).
Despite the fact that PCBs seem to be omnipresent (Jiang et al., 2017) and are perceived as the norm rather than the exception , managers are under pressure to foster positive relationships with employees (Guest, 2004) because these are salient to organisational success and even survival (W. Liu et al., 2010). One strategy to alleviate the effects of PCB is through leadership style, which can be adjusted and applied to achieve desired outcomes such as EV. Research by Ng et al. (2014) has shown how changes in social relations, which could result from changes in leadership styles, effect EV under conditions of PCB. The objective of this study is to assess the extent to which leadership style influences the relationship between PCB and EV.
The current research was conducted at a level of complexity that, to the authors' knowledge, no previous study had matched. Firstly, the independent variable (PCB) included two types of breach, namely transactional psychological contract breach (TPCB) and relational psychological contract breach (RPCB). Secondly, the dependent variable (EV) was studied as a four-dimensional construct with subconstructs related to supportive voice (SV), constructive voice (CV), defensive voice (DfV) and destructive voice (DsV). Finally, the four leadership styles-namely, transactional leadership (TsL), transformational leadership (TfL), directive leadership (DL) and empowering leadership (EL)were studied in the model as moderators in the PCB-EV link. Adding this complexity was necessary for several reasons. Firstly, EV evolved from a single construct (Hirschman, 1970;LePine & Van Dyne, 1998;Van Dyne & LePine, 1998) into a multidimensional concept, inferring dichotomous characteristics of different types of voice, which implies that the opposite to the promotive nature of voice also exists (Liang et al., 2012). Secondly, although research provides extensive evidence that PCBs negatively correlate with employees' discretionary behaviours (Zhao et al., 2007), including voice (Zagenczyk et al., 2015), it is still unknown in what way different types of PCB affect different types of EV. Finally, because employees' motivation to give voice is largely dependent on whether they believe their contributions are valued by their leaders (Farndale et al., 2011), it is necessary to establish how different leadership styles affect different types of EV under different conditions of TPCB and RPCB. Although some research on this matter was done previously (Ng et al., 2014), the relevant studies did not include this diversity of leadership style.

Literature review
The literature review briefly describes the three groups of variables used in this study and then reports on empirical research linking these variables.

Voice
Initially, the concept of voice was associated with the customer's choice to express dissatisfaction with the company's offerings with the "intention to force a change in management" (Hirschman, 1970, p. 30). Since then, a few seminal authors have led the research on voice (see, Liang et al., 2012;Maynes & Podsakoff, 2014;Morrison, 2011;Van Dyne & LePine, 1998), propelling it into broad fields of interest related to voice, namely customer voice (Griffin & Hauser, 1993;Shillito, 2000), voice as a channel of employees' inputs into managerial decision making (Budd, 2004), voice as a means of communication with the management (Freeman & Medoff, 1984) and voice as employee extra-role improvement-orientated behaviour (Morrison, 2011). EV is defined as "discretionary communication of ideas, suggestions, concerns, or opinions about work-related issues with the intent to improve organizational or unit functioning" (Morrison, 2011, p. 375). It has developed into a universal organisational behavioural concept (Van Dyne et al., 2003) essential to organisational commitment (Farndale et al., 2011), employee engagement (Rees et al., 2013), individual performance (Ng & Feldman, 2012) and innovation (Veenendaal, 2015).
The organising framework for EV provided by Maynes and Podsakoff (2014) has become prevalent in recent conceptualisations. Building on the seminal work of previous scholars (Liang et al., 2012;Morrison, 2011;Van Dyne et al., 2003;Van Dyne & LePine, 1998), Maynes and Podsakoff distinguish between four types of EV, describing them as supportive, constructive, defensive and destructive. The definitions for each of the voice types are: 1) SV is defined as "the voluntary expression of support for worthwhile work-related policies, programmes, objectives, procedures, etc., or speaking out in defence of these same things when they are being unfairly criticised", 2) CV is defined as "the voluntary expression of ideas, information or opinions focused on effecting organisationally functional change to the work context", 3) DfV is defined as "the voluntary expression of opposition to changing an organisation's policies, procedures, programmes, practices, etc., even when proposed changes have merit or making changes is necessary", and 4) DsV is defined as "voluntary expression of hurtful, critical, or debasing opinions regarding work policies, practices, procedures, etc." (Maynes & Podsakoff, 2014, p. 91). Of the four, two types (SV and CV) relate to the promotive nature of voice, while the other two (DfV and DsV) relate to the prohibitive nature of voice.

Psychological contracts and psychological contract breaches
Without a doubt, Rousseau (Rousseau, 1989(Rousseau, , 1990(Rousseau, , 1998(Rousseau, , 2001(Rousseau, , 2004(Rousseau, , 2011 is the leading scholar of research on PC, and is more referenced in this field than any of her counterparts. Rousseau (1995, p. 9) defines PC as "individual beliefs, shaped by the organization, regarding terms of an exchange agreement between individuals and their organization". PCs are most often differentiated as transactional (TPC) and relational (RPC), based on the nature and tangibility of the employer's perceived obligations (Rousseau, 1995). TPCs inculcate specified, transparent, shortterm obligations and are described as the "monetisable" exchange relationships in which employees value instant rewards such as training, fair remuneration, compensation leave, etc. In contrast, RPCs emphasise broad, long-term, mutual obligations for both parties, such as support and development from employers and loyalty and commitment from employees. RPCs are characterised by a high degree of mutual interdependence, with an emphasis on socio-emotional (loyalty and commitment) elements of exchange (Rousseau & McLean Parks, 1993). Aligned with social exchange theory (SET; Blau, 1964), and specifically its notion of reciprocity, it is argued that employees will likely reciprocate the treatment that they receive from their employers, based on their perception of whether they receive less or more of what has been promised (Cropanzano & Mitchell, 2005). Employees' perceptions of a degree of fulfilment of individual PCs translate into distinct employee behaviours and reactions, either positive if employees perceive that their employers have fulfilled their obligations or negative if they perceive that their employers have failed to deliver on their promises (Morrison & Robinson, 1997). The perceived failure of employers to deliver on their promises refers to PCB, the variable that is the focus of this study.
A large number of empirical investigations is dedicated to evaluations of the role of PCBs in affecting employees' behaviours (Flood et al., 2001). These studies provide convincing evidence that employees reciprocate PCBs by lowering in-role performance (Haryanto et al., 2022), work engagement (Agarwal, 2014), affective commitment (Rigotti, 2009), organisational citizenship behaviour (Lu et al., 2015) and innovation (Botha & Steyn, 2022). Furthermore, extant literature provides evidence that employees react to breaches differently under different PC conditions. Studies that compared the effect of TPCB with the effect of RPCB report that because of their socio-emotional and symbolic nature, RPCBs tend to elicit stronger negative employee reactions associated with decreasing levels of trust and organisational citizenship behaviour (Restubog et al., 2008) and perceived organisational support (Zagenczyk et al., 2009).
Often in empirical investigations, PCB is presented as the independent variable (see, Akhtar et al., 2016;Hui et al., 2004;Kakarika et al., 2017;Thomas et al., 2016;Vander Elst et al., 2016). This research adopted a similar approach, with TPCB and RPCB both studied as independent variables.

Leadership style
The traditional two-way TsL-TfL typology has been criticised by a number of researchers. Yukl (1989, p. 212) points out that limiting the leadership paradigm to transactional-transformational aspects of leadership is to oversimplify the complex phenomenon. Following Yukl (1989), and even some of the full range leadership theories designers, Bass and Avolio (1993, p. 76) encourage researchers "to shape a leadership theory and model" to include a broader spectrum of leadership behaviours and attributes.
This study adopts the four-way leadership typology developed by Pearce et al. (2003), which is based on the historical analysis of various leadership models and theories, as well as traditional leader behaviours and attributes. Pearce et al. (2003) differentiate between four distinct leadership styles, namely TsL, TfL, DL and EL. TsL refers to the behaviours that establish the parameters of the exchange relationship between the leader and the follower; TfL refers to those leader behaviours that encourage vision, produce inspiration from their followers and motivate change; DL refers to behaviours where the leader gives orders on how the work needs to be done; and EL develops the followers so that they become effective and capable self-leaders.
Leadership styles are often presented in the conceptual models as traits (Walumbwa & Schaubroeck, 2009) that are relatively stable but also situational (Ridlwan et al., 2021), where leaders adjust their set styles to the situation. In this study, the leadership style variable was used as a moderator between TPCB and RPCB as independent variables and EV as a dependent variable. As a means of managing the workplace, this research advocates a situational perspective on leadership, allowing leaders the ability to behave differently should the circumstances dictate that changes are necessary.

Empirical links between psychological contracts, psychological contract breaches and employee voice
The extant research provides strong empirical evidence for the notion that PCB is associated with negative individual and organisational outcomes (see, Zhao et al., 2007). However, the research specifically focusing on the effects of PC and PCB on EV seems limited. Furthermore, in those limited sources, some researchers approach voice as a unitary construct while others report only on one or two broad subcomponents of EV. For example, Rees et al. (2013), in studying EV as a unitary construct, report that employees who experience positive relationships with their senior and line managers (which is generally associated with fulfilled PC) expressed voice more often as their reciprocal response to the positive treatment from the management. In the meta-analysis on the relationship between workplace strains and stressors (specifically strained relationships with supervisors, breaches of promises and expectations-all associated with PCB) and work behaviours, Ng and Feldman (2012) found that under conditions of major work strain and stressors (such as PCB), employees were unlikely to exercise voice behaviour. Interestingly, H. Liu et al. (2020), who studied the direct effects of relational and transactional PCBs on voice, reported that neither relational nor transactional breach had a direct effect on EV. In the broader approach, EV is perceived as a multidimensional construct, but these studies focus on one or two facets of a comprehensive EV conceptualisation. A number of models tested distinct types of voice, such as proactive voice (LePine & Van Dyne, 1998), promotive and prohibitive voice (Guo, 2017;Liang et al., 2012), and constructive and aggressive voice (Hagedoorn et al., 1999;Ng et al., 2014). When Ng et al. (2014) investigated the link between PCB and constructive and aggressive voice, they found that PCB was negatively related to CV but that the prediction of PCB having a direct relationship with aggressive voice was not supported. The findings on the absence of the PCBaggressive voice link by Ng et al. (2014) contradict the earlier study by Turnley and Feldman (1999), where it was concluded that PCB was significantly and positively related to aggressive voice. Employees' responses to dissatisfying work circumstances with aggressive voice are in line with the findings of Rusbult et al. (1988), who stated that employees' reactions to dissatisfying jobs may range from constructive to very destructive. This echoes the conclusions of Ng et al. (2014) that, under conditions of PCB, employees will not only reduce positive work behaviours but will also increase negative work behaviours. In other words, when employees experience PCB, they will likely withhold CV and engage in counter-productive, negative behaviours (Ng et al., 2014). In a recent study investigating the effect of PCB on employees' promotive and prohibitive voices, Guo (2017) reported that PCB was negatively related to both promotive and prohibitive EV. Thus, breaches resulted in withholding all discretionary behaviour. When organisational trust was added to the model as a possible mediator, the results revealed that organisational trust fully mediated the relationship between PCB and EV. Interesting findings were reported in the research conducted by Balabanova et al. (2019) on the effects of different types of PCB on employee exit and CV. The link between variables was tested among a sample of employees from Russia and Finland. The study reported that the Russian employees reacted only to breaches of transactional contracts, whereas employees in Finland were responsive to both transactional and relational breaches. It was also found that, compared to the negative association between RPCB and CV among Finnish employees, the TPCB and RPCB links with CV among the Russian employees were found to be statistically non-significant. Moreover, this study revealed that EV was not only organisational context dependent but also between-culture sensitive.
From this review, it is evident that although the interest in antecedents of voice is growing, the research is limited. The possibility that different types of PCB (namely TPCB and RPCB) may serve as predictors of voice operationalised as a complex multidimensional construct seemed unresearched, or at least, under-researched. To bridge the gap, this study aimed to answer the question "What will the distinct effects of transactional and relational psychological contract breaches be on supportive, constructive, defensive and destructive voice?"

Links between leadership and employee voice
The researchers' interest in how leadership influences EV is explicable, as eliciting employee extrarole or discretionary behaviours such as voice is salient for organisational survival (W. Liu et al., 2010). A large number of studies has been dedicated to understanding leadership behaviour as a predictor of voice. Leadership research provides broad evidence that the organisational context, partially created by leadership, influences employees' willingness to speak up (including providing creative ideas and constructive insights). Leadership behaviour is identified as instrumental in influencing employees' decisions to voluntarily offer their suggestions for organisational improvements (Gao et al., 2011). For leaders, it is fundamentally important to react to employees' expression of voice, as they are the targets thereof (Detert & Burris, 2007). Thus, in this research, the link between four leadership styles (namely TsL, TfL, DL and EL) and four types of EV (namely SV, CV, DfV and DsV) was investigated, with leadership style acting as a moderator between PCB and EV. From the literature search, it is apparent that more studies focus on the TfL-EV and EL-EV links than on the TsL-EV and DL-EV links. No studies that included all the mentioned leadership styles as well as types of EV was found, let alone research that included these eight variables as well as PCB.
It was deemed necessary, as background to the study, to provide information on the leadership style-EV association. In their investigation of the relationship between TfL and subordinates' improvement-orientated voice, Detert and Burris (2007) reported that transformational leader behaviour was positively related to voice. As TfL is indicative of an orientation towards organisational improvement, the authors concluded that TfL positively relates to employees' perceptions that it is safe to speak up and their willingness to engage in voice. In the more recent study on the links between TfL and EV conducted by Wang et al. (2018), it was found that TfL positively and significantly correlated with EV. The authors explain these findings as illustrating that transformational leaders build safe and supportive environments where employees are motivated to voice new ideas and opinions (Wang et al., 2018).
The empirical findings on the relationship between EL and EV are generally consistent in reporting on the positive correlation between the two variables. For example, Jada and Mukhopadhyay (2018) investigated the effect of EL on CV. Their study revealed that EL had both direct and indirect (through psychological safety) effects on CV. These findings are aligned with SET (Blau, 1964) in that when leaders exercise EL behaviour, they create a positive and psychologically safe environment where employees feel obligated to reciprocate to the organisation in a form of speaking up with suggestions and ideas for organisational improvement (Jada & Mukhopadhyay, 2018). Three types of EV (namely, pro-social, acquiescent and defensive) were studied as dependent variables in the investigation of Li et al. (2018). Their research concerned the indirect effect of EL on three types of voice, specifically a positive indirect effect of EL on pro-social voice and negative indirect effect on acquiescent voice and DfV. These indirect effects were moderated by employees' work stress. In the case of pro-social voice, the work stress lessened the positive link between EL and pro-social voice; in the case of acquiescent and DfVs, the work stress augmented the negative link between the variables.
Interest in studying the effect of ethical leadership on voice is understandable, as evidence exists that ethical leaders encourage employees to voice their ideas and opinions on improvements of work procedures and environments (Walumbwa & Schaubroeck, 2009). In line with the findings of Walumbwa and Schaubroeck (2009), the research on the links between ethical leadership, voice behaviour and creativity conducted by Chen and Hou (2016) showed a positive and significant relationship between ethical leadership and voice. Similar results were reported by Yousaf et al. (2019), who found that ethical leadership had a positive significant relationship with voice. The authors argue that ethical leaders are trusted by employees, they encourage employees to speak up and they strive to maintain better communication with their subordinates. The use of other leadership styles investigating links between the two variables was also found in the literature. For example, in a recent study of Soomro et al. (2021) on the relationships between authoritarian, benevolent and moral leadership styles and EV, it was argued that authoritarian leadership was associated with decreased levels of EV, the effect of benevolent leadership on voice was non-significant, and the correlation between moral leadership and voice was positive and significant.
The paucity of the research that focus specifically on the indirect effects of different leadership styles on EV, or multiple forms of EV, necessitated this study. Particularly, the possibility that TsL and DL may serve as indirect predictors of voice is unresearched or, at least under-researched. This study aimed to answer the second set of questions: "Which leadership style will likely activate the promotive types of EV (supportive and constructive) given the dominant type of the psychological contract breach?" "Which leadership style will likely weaken the prohibitive types of EV (defensive and destructive), given the dominant type of the psychological contract breach?"

Research framework and hypotheses
The research framework appears in Figure 1. It suggests a relationship between PCB and voice, with leadership style as a possible moderator.
SET (Blau, 1964) and psychological contract theory (PCT; Rousseau, 1995Rousseau, , 2011Rousseau & McLean Parks, 1993) form the basis of this framework. Both SET, being foundational in explaining workplace associations (Shore & Barksdale, 1998), and PCT which advocates that fulfilled PCs are linked to positive employee attitudes and behaviours focus on norms of reciprocity (Gouldner, 1960).
The following broad hypotheses were set: In view of the fact that PCBs are omnipresent and considered a norm of workplace relations , the senior and/or line manager's leadership style could be the mechanism buffering or exacerbating the negative effects of PCBs on employee behaviour, dependent on the type of leadership style applied by the manager. By definition, it is logical to assume that the TfL and EL styles will buffer or lessen the effect of breach on all four types of EV. Following the same line of thought, it can be assumed that TsL and DL styles will exacerbate the negative effects of breaches on all four types of EV. The third null hypothesis could then be stated as follows: H3 0 : All leadership styles affect the relationship between PCBs and EV.
The method for testing the aforementioned hypotheses is described below.

Method
This study was designed as a cross-sectional survey, which was appropriate for data collection and the proposed hypotheses testing.

Population and sample
In line with recommendations for variability in responses (Zagenczyk et al., 2009), the targeted sample population consisted of employees of all races, genders and ages, and at different levels of responsibility; thus, several organisations were approached. Organisations with more than 60 employees were targeted, as it was assumed that the employment relations (PCs) would be formalised in these organisations and that a broad range of contracts would be in place. For assistance with the data collection and to gain access to various organisations, the researchers contracted a group of 11 students enrolled in the Master of Business Leadership programme at a major South African business leadership school. Once students had been granted access by the authorities at the 11 organisations concerned, they were assisted by these organisations' respective human resources departments with drawing a random selection of employees for participation in the study. As the result, each student obtained an average of 60 completed questionnaires from the related organisation.

Measures
A self-report survey was conducted across all three measures. This approach is acceptable, as perceptions of employees on the nature of their PC as well as the extent of its fulfilment or breach by the employer (PCB) are individual and subjective (Rousseau, 1995). With regard to voice, selfreported measures are widely used in the organisational research field (Axtell et al., 2000;Parker et al., 2006) due to employees (in comparison to their supervisors and peers) being more aware of the nuances of their suggestions and own beliefs, as well as whether these carry an instrumental value for the group or the organisation (Ng et al., 2014). Finally, self-reporting on discretionary behaviour such as voice (LePine & Van Dyne, 1998) provides better insight into individuals' assessment of intensity and frequency of own voice behaviour.
PC was measured using Millward and Hopkins's (1998) psychological contract scale (PCS). The original 33-item instrument (20-items for transactional contract and 13-items for relational contract scales) was reduced to five items for measuring transactional contracts and five items for measuring relational contracts, using the items with the highest average factor loadings of each construct, as recommended by Bateman and Crant (1993) and used by Strydom (2013). Each PC type was measured on a seven-point Likert scale ranging from 1 "Strongly disagree" to 7 "Strongly agree". The sample items were "I only carry out what is necessary to get the job done" for transactional contract and "To me working for this organisation is like being a member of a family" for relational contract. Millward and Hopkins (1998) reported a Cronbach's alpha of 0.86 for all relational items and 0.88 for all transactional items.
PCB was measured using Robinson and Morrison's (2000) nine-item scale, measuring the violation with four items and the breach with five items. The motivation for reporting on a total score for both violation and breach was based on the idea that, theoretically, employees' experiences of either violation or breach (or both) result in (negative) work outcomes (Bal et al., 2008;Raja et al., 2004;Tekleab & Taylor, 2003), and that both would affect discretionary behaviours such as voice in the same direction. Robinson and Morrison (2000) reported a Cronbach's alpha of 0.85 for violation and of 0.88 for breach. Reporting of breach and violation is done through reverse scoring of a fulfilment measure on a Likert scale ranging from 1 to 5 (disagree to agree). A sample item reads as follows: "Almost all the promises made by my employer during recruitment have been kept so far" (reversed).
Voice was measured using Maynes and Podsakoff's (2014), pp. 20-item voice scale, with five items related to each subscale, being the SV, CV, DfV and DsV subscales. The four types of voice were measured on a seven-point Likert scale, where 1 was "Strongly disagree" and 7 was "Strongly agree". The authors reported Cronbach's alphas for SV 0.89, for CV 0.95, for DfV 0.92 and for DsV 0.93. The sample items were: "I defend useful organisational policies when other employees unfairly criticise the policies" (supportive voice); "I frequently make suggestions about how to do things in new or more effective ways at work" (CV); "I vocally argue against changing work practices, even when making the changes is necessary" (DfV); and "I frequently make overly critical comments regarding how things are done in the organisation" (destructive voice).
Leadership styles: TsL and TfL were measured with the shortened scale of Podsakoff et al. (1990), with a total number of 11 items (comprising five items for TsL and six items for TfL) measured on a seven-point Likert scale ranging from 1 "Strongly disagree" to 7 "Strongly agree". The authors reported the Cronbach's alpha for each of the dimensions as ranging between 0.78 and 0.92. The sample items were "My manager always gives me positive feedback when I perform well" for TsL and "My manager is always seeking new opportunities for the organisation" for TfL.
DL was assessed using six items developed by Pearce and Sims (2002) and four items from Hwang et al. (2015). While Hinrichs (2011) indicated a Cronbach's alpha for the six-item scale of Pearce and Sims (2002) at an acceptable reliability level of 0.88, Hwang et al. (2015) reported a similar result for their four-item measure, which delivered a Cronbach's alpha coefficient of 0.85. In their recent study, Solomon and Steyn (2017) reported a Cronbach's alpha of 0.87 for the two measures combined. The sample items for DL were: "When it comes to my work, my team leader gives me instructions on how to carry it out" (Pearce & Sims, 2002) and "My manager identifies specific action steps and accountabilities" (Hwang et al., 2015). EL was measured using the 10-item scale adopted by Ahearne et al. (2005) from various sources. The leadership empowerment behaviour scale consists of four multi-item subscales, where each subscale focuses on a specific aspect of EL behaviour, namely enhancing the meaningfulness of work (Cronbach's alpha of 0.76), fostering participation in decision making (Cronbach's alpha of 0.92), expressing confidence in high performance (Cronbach's alpha of 0.90) and providing autonomy from bureaucratic constraints (Cronbach's alpha of 0.86). The sample items are: "My manager helps me understand how my objectives and goals relate to that of the company" and "My manager allows me to do my job my way". Solomon and Steyn (2017) reported an average Cronbach's alpha of 0.93.
Based on the structure of the questions, it is clear that the responses of the participants reflected the leadership styles of their immediate managers and not those of the organisations' leaders such as CEOs.

Statistical analyses
Demographic statistics were first calculated and then interpreted in terms of the representation of the population. The basis for comparison was information reported by Statistics South Africa (2020).
Next, the normality of the collected data was analysed in terms of skewness and kurtosis. The guidelines of Field (2009) were used to interpret skewness and kurtosis scores. If the observed SPSS value divided by the standard error of that value was to be larger than 1.96, or smaller than −1.96, the data would be interpreted as displaying a serious deviation from normality.
The reliability was calculated using the Cronbach's alpha coefficient. The recommendations of Tavakol and Dennick (2011) regarding Cronbach's alpha ranges were considered: larger than 0.90 (excellent), 0.89 to 0.80 (good), 0.79 to 0.70 (acceptable), 0.69 to 0.60 (questionable), 0.59 to 0.50 (poor), and smaller than 0.50 (unacceptable). In this study, the more lenient guidelines of Pallant (2013) were accepted, with the Cronbach's alpha coefficient considered satisfactory where scores exceeded 0.70 and with scores above 0.80 accepted as desirable.
The adequacy of the data was analysed through factorial validity testing, particularly by calculating Kaiser-Meyer-Olkin's score of sampling adequacy (KMO) and Bartlett's score of sphericity. The results were considered acceptable when the KMO score was excellent (0.90; Field, 2009) and when the Bartlett's test value was significant (p. 0.001; Pallant, 2013). When analysing factor loadings for fit, the dominant loading of items in the theorised manner, along with the absence of significant cross-loadings, was interpreted as indicative of factorial validity.
Regression analyses were also performed. In this study, the total size of the regression coefficient was of less concern, with the focus primarily on the significance of the beta values of the different predictors in the regression models. Statistically significant predictors (p < 0.01) were deemed unique and substantial contributors to the variance in the dependent variable (Fairchild & MacKinnon, 2009).
Moderation was tested based on the procedures recommended by Fairchild and MacKinnon (2009). Their method entails performing a regression without including the moderator as a variable in the regression (Model 1), thereafter adding the moderator (leadership styles; Model 2), and finally adding the moderator and the interaction effect (PCB × leadership style; Model 3). In general, the interest is in ∆R 2 , using Model 1 as a baseline model. If ∆R 2 is positive and significant across three models, this suggests improved models and the specific importance of adding the additional variable. Should leadership style directly predict voice (Model 2, with leadership style having a significant beta value), it is representative of a direct effect, making it an antecedent of voice. Should the interaction between leadership style and any predictor subcomponent be significant (Model 3, for example, PCB × TsL having significant beta values), this is representative of leadership style moderating the relationship between PCB and EV.

Demographics
A total of 620 respondents provided complete data on the variables of interest. There were 313 men (50.5%) and 301 women (48.5%); data from six respondents were missing. Most of the respondents (440) were black (71%), 103 respondents were white (16.6%), 42 were coloured (6.8%) and 28 were Asian (4.5%). In the sample, 254 respondents (41%) had obtained a higher degree or diploma, 203 respondents (32.7%) had their first degree or diploma, 138 respondents (22.3%) had matric (senior certificate), and 19 (3.1%) had less than 12 years of education. The average age was 37.8 years and the sample consisted of respondents from a well-distributed age group (standard deviation of 8.841), varying between 21 and 64 years. The average tenure was 6.59 years, with the range between 1 and 42 years (standard deviation of 5.848). The data implied that most respondents were well qualified for reporting on perceptions of employment relations as well as on observed organisational practices.

Reliability
The reliability of the different instruments used in the study is as follows.
The reliability of all the instruments used was acceptable, with the lowest value 0.764 (which was above the cut-off score of 0.70) and the majority of alpha values exceeding 0.80, placing these reliability scores within the desirable range (Pallant, 2013).

Validity
The structural validity of the instruments used was tested and found to be satisfactory. The results are not presented here due to their comprehensiveness, but they are available from the first author upon request.

Descriptive statistics
Reliability of research instruments is presented in Table 1. Mean scores and statistics related to the distribution of the data are presented in Table 2.
In accordance with the conceptual framework, two new variables were created. These were TPC × PCB and RPC × PCB, which were the interaction of the two psychological contracts and psychological contract breach.

Correlation analyses
The Pearson correlation coefficient was calculated for all the predictor variables and the independent variables on EV. The results presented below relate to H1 and H2.
In Table 3, several statistically significant correlations are observed. It can also be observed that in two instances, these correlations were significant at a practical level (the shaded cells in the table).

Regression analyses: the moderation effect of leadership style on the PCB-EV, TPCB-EV and RPCB-EV links
Stepwise hierarchical regression analyses were performed with the sole purpose of identifying and specifying the moderation effect of leadership style on the PCB-EV relationship (Table 4), and the moderation effect of leadership style on TPCB-EV and RPCB-EV links (Tables 5 and 6). The focus in these tables should be on the improvement (if any) in the Adjusted R 2 from Model 2 to Model 3, and the statistical significance of beta values for interactions in Model 3.
From the observed changes in Adjusted R 2 (ΔR 2 ) for DfV and DsV in the table above, it can be concluded that the moderation models added exploratory power: For DfV ΔR 2 = 0.01 and for DsV ΔR 2 = 0.003. The defining test of moderation (Fairchild & MacKinnon, 2009), where the interactions of the independent variables were statistically significant predictors of the dependent variables, revealed that TsL moderated the relationship between PCB and both DfV and DsV. As TsL (as an independent variable) remained a significant predictor in Model 3, for both DfV and DsV, the moderation was partial only.
When considering the moderation effects of leadership styles on TPCB predicting CV, DfV and DsV, the moderation model displayed an additional exploratory power: For CV ΔR 2 = 0.005, DfV ΔR 2 = 0.008 and for DsV ΔR 2 = 0.012. In the test for moderation, where the interactions of the independent variables were statistically significant predictors of the dependent variables, it was found that TsL moderated the relationship between TPCB and both CV and DfV. TfL and EL moderated the TPCB-DfV relationship. As TsL (as an independent variable) remained a significant predictor in Model 3, for both CV and DfV, the moderation here was only partial. The same applied to TfL as a moderator, where the moderation was also partial. However, for EL, moderation was full.
When considering the moderation effects of leadership styles on RPCB predicting CV, DfV and DsV, this moderation model also showed an improvement in exploratory power: For CV ΔR 2 = 0.004, DfV ΔR 2 = 0.002 and for DsV ΔR 2 = 0.008. In the test for moderation, it was found that TfL moderated the relationship between RPCB and DfV. As TfL (as an independent variable) still remained a significant predictor in Model 3, the moderation was partial.
The outcomes of the hypotheses are presented graphically in Figure 2 below.

Discussion
This study aimed to empirically evaluate how PCBs, as well as particular types of PCB (being TPCB and RPCB), are linked to different types of EV and how different leadership styles affect these  relationships. The literature on the interplay between variables investigated in the proposed conceptual model is scarce, which prompted this research.
The cross-sectional design of the study was adequately suited to meeting the objectives of the research. The respondents fairly represented the population sample in terms of their sex and race, aligned as they were to the numbers reported by Statistics South Africa (2020). Therefore, it was concluded that generalising the findings of this study across the broader South African context would be reasonable.
The reliability scores of all the instruments used in the study fell within the range between 0.764 and 0.992, which is acceptable (Pallant, 2013). The structural validity of the instruments was tested and found to be satisfactory. As stated above, the results are available from the first author upon request.
Three null hypotheses were set for the investigation: H3 0 : All leadership styles affect the relationship between PCBs and EV.
The first hypothesis was addressed in Table 3 by means of the Pearson correlation analyses. It was found that PCB related positively and significantly to the prohibitive forms of EV (DfV and DsV), but correlated negatively with the promotive forms of EV (SV and CV) even though the relationship with CV was not significant. CV is mentioned here to indicate the (negative) direction of the effect, and to illustrate that PCB positively relates to prohibitive forms of voice and negatively to promotive forms of voice. The only practically significant relationship between PCB and EV occurred in the instance of DsV. It could therefore be stated that PCB and DsV go hand in hand. It was found that the transactional type of PCB, compared to overall PCB which is also illustrated with the (negative) direction of the correlation with the prohibitive forms of voice, is a stronger facilitator of the prohibitive voice and to a lesser extent an inhibitor of the promotive voice because in all these cases, the relationships were statistically significant. It is important to note that TPCB and DfV had a practically significant relationship, thus implying that TPCB incites DfV. In general, RPCB presented lower correlations with EV. Interestingly, the direction of the associations was positive throughout, but it was not practically significant in any of the cases. These results affirm the earlier findings on the significant and positive correlation between PCB and aggressive voice (Turnley & Feldman, 1999), and the significant and negative correlation between PCB and CV/promotive voice (see, Guo, 2017;Ng et al., 2014). Furthermore, this study provides new insights into the effects of transactional and relational breaches of four distinct types of EV. The disparate results in previous investigations on TPCB/RPCB-EV links are understandable, as the findings were dependent on the choice of voice studied as a dependent variable. In addition, inconsistent results could also be explained by voice having been studied by some researchers as a unitary construct (LePine & Van Dyne, 1998;Rees et al., 2013) and by some as a particular voice dimension (Liang et al., 2012;Ng et al., 2014). The current study addressed this dilemma of disparity by providing empirical evidence on the interplay between TPCB/RPCB and four dimensions of EV.
The second hypothesis was also addressed with the correlation analyses (Table 3). It was found that none of the four leadership styles were significantly related to DfV. Although none of the correlations between leadership styles and all four types of EV were practically significant, EL had the highest correlation with SV and CV, and TfL impacted most negatively on DsV. Although the findings on the positive and significant EL-CV link are supported by previous studies (Jada &   Mukhopadhyay, 2018), the findings on the direct effects of leadership styles on other dimensions of voice are novel.
Considering the outcomes of Hypotheses 1 and 2, it can be stated that in a comparison between the effect of PCB and leadership style on EV, and at a practical level, PCB proved to have a higher direct impact on EV than any of the four leadership styles. Leadership styles were thus not a direct predictor of voice, which made testing for Hypothesis 3 necessary.
Hypothesis 3 was related to the likelihood of leadership style moderating the PCB-EV relationship. It was particularly important to find out whether some or any of the leadership styles had a propensity to buffer, or whether any of the leadership styles tended to exacerbate, the effect of PCB. The conclusions presented below are related to each type of leadership style given the type of PCB: • Under conditions of PCB (in general), TsL buffers PCB's effect on DfV and DsV. Regarding TPCB, TsL buffers the effect on DfV but lessens the intensity of CV. It thus seems that when psychological contracts are violated, dealing with consequences of breach explicitly, in a contractual manner (TsL), would lessen defensive and destructive but not necessarily increase promotive forms of voice. Linking TsL with DfV seems acceptable, as both parties respect the "rulebook". The link with CV also seems understandable, as parties would allow constructive negotiation, particularly when a breakdown occurs in a transactional environment (TPCB). However, the significant and negative link between PCB × TsL interaction and DsV is difficult to explain.
• TfL exacerbates the effect of TPCB on DsV. This can be explained using the example of a manager who acts as a transformational leader, but who breaks the "rulebook" (TPCB), making an employee likely to respond with DsV behaviour. TfL also aggravates the effect of RPCB on DfV. This can be explained by the fact that when a transformational leader breaks an employee's trust, that employee may become confrontational.
• DL has no impact on the PCB-EV relationship.
• Finally, the findings suggest that EL tends to have a negative effect on DsV and that this occurs particularly under conditions of TPCB. It could mean that in the environment typified by TPCB, when a leader acts in an empowering manner, DsV behaviour could be lessened. These results are in many respects novel, as the present body of research on this matter is-as far as our investigation goes-absent. However, questions could legitimately be asked about the interpretations presented immediately above, and many counterarguments could be provided. Limited to the present data, the authors will abide by these results and interpretations.

Conclusions
This research aimed to identify leadership styles that are particularly effective in managing EV, given conditions of PCBs. Based on the comprehensive empirical analyses, it was revealed that no single leadership style was omni-influential. Nevertheless, it was found that TsL had the most profound buffering effect on the PCB-EV links, for both defensive and destructive forms of voice, although the effect on CV within a transactional environment was the opposite. Similarly, EL had a buffering effect on the PCB-EV link. This showed that, where transactional breaches take place, the empowering leader will lessen the expression of employees' destructive voice.
In light of these implicit relational dynamics, this study aimed to further advance an understanding of social exchange relationships that reinforce the PCB-EV link by exploring the relative contribution of leadership style in moderating the way PCB is associated with EV. In this research, a significant contribution was made by investigating relationships between TPCB, RPCB, four leadership styles and four types of EV, studied together in one conceptual model-something that has not been done in previous literature.
From an academic perspective, complexity was added to the PCB, leadership and EV debate. Leadership does not seem to be as influential as theories suggest. Researchers are encouraged to explore different mechanisms that could possibly influence the PCB and EV relationship, as EV is a potent predictor of effective change. When dealing with leadership styles per se, it is suggested that the focus should be shifted away from TfL and that more attention should be given to transactional forms of leader behaviour as, in this study, TsL demonstrated its relative strength in influencing the PCB-EV link.
From a practical perspective, the findings of this study will assist managers to better understand the salience of maintaining positive social exchange relationships with their employees. In addition, it has become evident that some leadership styles are particularly useful in buffering the negative effects of psychological contract breaches on prohibitive forms of EV. The endeavours of leaders in engaging employees in discretionary behaviours-particularly when they are expected to voice creative ideas and suggestions about organisational improvements-will be more successful under an empowering form of leadership.
This research could benefit from including additional leadership styles such as the authentic leadership style, as it is a known predictor of creativity among employees (Alzghoul et al., 2018). This and other leadership styles could be included in future research as the critique against the older theories such as the full range theory of leadership (Bass & Avolio, 1993) and the four-way leadership typology of Pearce et al. (2003). Future researchers are also encouraged to look beyond immediate managers' leadership styles and to explore the impact of strategic leaders such as chief executive officers on EV behaviour.