2.1 Formality and procedural fairness

This study is based on the theoretical proposition that formality satisfies fairness principles. The concept of formality reflects the extent to which superiors provide subordinates quantitative and written targets, quantitative and objective performance measures, and formulaic determination of reward. Hartmann and Slapničar (2009) explained that formality is a three-dimensional concept, which can be applied in target-setting, performance metrics and reward determination. In target setting, a high level of formality will see superiors set performance targets that are in quantitative and written terms (high level of objectivity) whereas a low level of formality will have superiors apply performance targets as qualitative and not formally communicate to subordinates (high level of subjectivity). In high formality, superiors will use performance metrics that are more objective and in quantitative form while in low formality, more subjective measures expressed in qualitative terms will be deployed. Superiors will more likely determine rewards based on formulae when formality is high, and they will use more untraceable personal judgement when formality is low. In summary, formality manifests in different degrees on a continuum where the highest level is reflected by superiors relying strictly on explicit procedures (Gibbs et al., 2004; Locke & Latham, 1990), quantitative measures and objective information (Moers, 2005) whereas the lowest level shows superiors total relying on implicit procedures, qualitative measures and subjective judgement.

Procedural fairness is the judgement on the fairness of the means or of the rules and processes that superiors use to make decisions in performance evaluation processes (Lau & Moser, 2008). It is fostered by adherence to six principles: (1) consistency (procedures should be consistent across persons and across time); (2) correctability (procedures should include mechanisms for correcting poor decisions); (3) accuracy (procedures should be based on valid facts, information and opinions); (4) bias suppression (procedures should be neutral and impartial); (5) ethicality (procedures should uphold standards of ethics and morality); and (6) representativeness (procedures should be representative of all groups' concerns) (Colquitt & Jackson, 2006; Leventhal, 1980). With performance evaluation systems, procedural fairness is determined by assessing employee experiences with performance ratings, bonus payments, or other rewards and benefits against the six criteria (Voußem et al., 2016). If employees perceive that the criteria were upheld, performance evaluation systems will be considered fair.

Formality can have a positive impact on subordinates' procedural fairness. Sholihin et al. (2011) point out that procedural fairness reflects all aspects of the organisation's procedures that are used by superiors to evaluate the subordinates' performance to communicate performance feedback and, to determine the subordinates' rewards such as promotion and pay increases. High formality is likely to be perceived by subordinates as superiors try to provide consistent and traceable criteria, which then act as a system-reference. Subordinates are more likely to perceive a greater likelihood of consistent evaluation process across team members who are in similar situations as they are than when there is not a system-reference in place (Principle 1). Lau and Buckland (2000) revealed that budget-related performance criteria (which is high in formality) are perceived by subordinates as superiors' attempt to provide ‘precise’ and ‘honest’ performance feedback. Thus, this study suggests that high formality assists employees in correcting errors and improving performance (Principle 2). In addition, standard procedures and evaluation criteria enhance objectivity in decision-making and suppress bias pertaining to subjective judgement (Merchant & Van der Stede, 2007) (Principle 4). Such objectivity illustrates superiors' desire to provide accurate direction and guidance to subordinates, who will perceive fairness in the process (Principle 3).

However, high formality may be rigid and not possible for highly uncertain tasks (Hartmann & Slapničar, 2012). With highly uncertain tasks, subordinates face a high level of uncertainty about the extent by which their efforts are recognised as outcomes in performance measures. Informed by conditional controllability, highly uncertain tasks could compromise sensitivity of performance measures, that is, whether employees can affect the statistical pattern of performance measures. Woods (2012) states that performance measures that are low in sensitivity are not completely informative about subordinate performance and need subjective adjustment by superiors. These discretionary adjustments will account for qualitative and uncertain factors, in order to provide a fair performance evaluation. However, both Woods (2012) and Moers (2005) found that superiors also made adjustments for reasons other than to improve objective measures. For instance, downward adjustments to unexpectedly high performances that are deficiently measured were used to encourage the departure of certain subordinates. Likewise, downward adjustments by supervisors appear to be avoided to preclude negative consequences for both subordinates and themselves. This sort of adjustments suffers from judgmental biases and result in evaluations that reflect superiors' personal favouritism and are not optimally aligned with organisational objectives (Bol & Smith, 2011; Ghosh & Lusch, 2000; Hartmann & Slapničar, 2012). It adversely impacts subordinate motivation (Prendergast & Topel, 1993) and redirect subordinates' effort toward influencing superiors' evaluation of their performance (Milgrom, 1988; Prendergast & Topel, 1993). It is at odds with the fairness principles of consistency, accuracy and ethicality. As such, Moers (2005) advises that subjective adjustments to outcomes of performance measures be relied upon only when high formality in performance measures result in deficient formal measures. Thus overall, it is the study's supposition that higher formality will have a stronger positive impact on procedural fairness than lower formality. However, as the studies by Hartmann and Slapničar (2009, 2012) demonstrated, the link between formality and procedural fairness is complex and not straight-forward and based on the discussions here, precision and sensitivity of performance measures could be two variables in this relationship (Hartmann & Slapničar, 2012).

2.2 Conditional controllability

Accounting scholars acknowledge that employees' perceived controllability is critical in the use of performance evaluation systems (Burkert et al., 2011; Merchant & Otley, 2006; Merchant & Van der Stede, 2007). When managers perceive that their performance is evaluated free from the impacts of uncontrollable events and that they have control over their performance output, they are likely to invest effort in working to achieve their performance targets.2 In contrast, when managers believe they have little control over the performance results, they are likely to engage in dysfunctional behaviours, such as gaming the systems.

As stated, this study adopts the notion of conditional controllability, which is when subordinates ‘are evaluated on the basis of all information that provides insights into their actions’ (Antle & Demski, 1988, p. 716). There are two recognised principles of controllability – traditional controllability principle and conditional controllability principle. Traditional controllability is evident when subordinates ‘are held accountable only for what they can control’ (Merchant & Van der Stede, 2007, p. 533). Conditional controllability is different from traditional controllability because the latter ‘stipulates that managers should be able to literally control a measure instead of merely influencing its statistical distribution’, whereas the former focuses on whether performance measures provide useful information content, rather than whether managers can literally control the measures (Fischer, 2010, p. 20). Subordinates perceive greater controllability: (1) when performance measures are distorted by uncertain, uncontrollable events to a lesser extent (Merchant, 1985; Merchant & Van der Stede, 2007); and (2) when they believe their decisions and actions influence their performance measures to a larger extent (Merchant, 1989; Simons, 2005). In other words, perceived conditional controllability is determined by precision (lack of noise in performance measures) and sensitivity (being able to influence performance measures by one's own actions) (Banker & Datar, 1989; Bisbe et al., 2007; Burkert et al., 2011).

Woods (2012) observed the impact of a low level of formality on precision and sensitivity. Woods noted that supervisors made adjustments to the measurement outputs when they perceived deficiencies in the objective measures. ‘For example, one manager received values of 4, 3, 3, and 4 on the four objective measures. That manager's supervisor then adjusted the measures +1, 0, 0, and 0, respectively, yielding adjusted scores of 5, 3, 3, and 4’ (Woods, 2012, p. 406). He then concluded that performance measures that are low in precision are less informative about performance and superiors can make adjustments to correct this deficiency. Measures that are high in sensitivity are informative about employee performance and therefore do not need subjective adjustment from supervisors. Thus, subjective adjustments to performance measures depend on the levels of precision and sensitivity of measures. These discretionary adjustments supplement formality in performance measures because they help rectify the deficiency of objective measures (Bol & Smith, 2011; Du et al., 2012), although subjective evaluations as discussed in subsection 2.1 are not to be relied upon by themselves. The implication for our study is that formality and its links to perceptions of precision and sensitivity continue to be worthy of research.

2.3 The mediating role of precision of performance measures

2.4 The mediating role of sensitivity of performance measures

2.5 The mediating role of role clarity

2.6 Precision and sensitivity of performance measures and role clarity

This study predicts that precision and sensitivity of performance measures will enhance role clarity. An important role of strategic performance measurement systems is to translate organisations' strategies into actionable performance measures, and in so doing communicate and clarify the strategies across different levels within the organisations (e.g., Franco-Santos et al., 2012). Recall that a high level of precision means that employees think that there are clear and specific performance measures. It helps communicate to subordinates what they need to do in order to achieve their assigned objectives and contribute to overall organisational objectives. It also helps clarify their work's expected outcomes and the ways to achieve them (Lau, 2015; Long et al., 2011). In contrast, a low level of precision would lead to subordinates not being clear on what their superiors expect them to do (Burkert et al., 2011).

When there is a strong degree of sensitivity, subordinates believe that they are able to influence performance outcomes. It is proposed that as a result, subordinates will then think that they have a better knowledge of their roles and a better sense of how to plan and perform their tasks. For instance, when a production manager is told that his/her performance will be measured and evaluated based on the levels of production volume and product cost per unit, he/she will perceive ability to influence volume and cost. This, in turn, will help articulate his/her responsibility to manage production volume and product costs.

The above discussions on precision and sensitivity and their impact on role clarity, are consistent with the extant literature that state that applying a controllability principle enhances role clarity due to a decrease in uncertainty (Birnberg et al., 2007; Burkert et al., 2011). Thus, our discussion shows a positive association between precision and role clarity and a positive association between role clarity and procedural fairness. It also shows a positive association between sensitivity and role clarity and a positive association between role clarity and procedural fairness. Thus, the final set of hypotheses follows:

Figure 1 summarises the proposed influences of the intervening variables of precision, sensitivity and role clarity in the link between formality and procedural fairness.

3.1 Sample selection and data collection

To test our hypotheses, data were collected from 119 functional managers of Australian manufacturing firms using a survey questionnaire. This method allowed access to a larger sample and ensured anonymity of potential participants, which were selected from large Australian manufacturing firms with 100 or more employees and annual turnover of more than AU$10 million. Functional managers provide the most knowledgeable sample to respond to the survey (Marshall, 1996). They are at the middle rung of the organisational hierarchy and in their roles, they receive from superiors and send to subordinates, prescriptions and proscriptions on role expectations. The manufacturing industry was chosen because prior management accounting studies on procedural fairness (e.g., Lau, 2015; Lau & Tan, 2005) of the Australian manufacturing industry did not provide conclusive evidence on the factors influencing procedural fairness. This study responds to inconclusive insights on the Australian manufacturing industry by investigating the role of formality, precision, sensitivity and role clarity in enhancing procedural fairness. Finally, large firms are likely to put in place performance evaluation procedures because ‘it is probable that only large organizations with significant managerial expertise and resources will have the motivation and the means to develop them’ (Lau, 2015, p. 150).

We identified 668 functional managers based on information from Dun & Bradstreet database. One person in each function within the same firm was targeted and up to six managers in each firm were selected. The participants were required to have a minimum one-year experience in their current position so that they are familiar with their performance evaluation systems and able to provide reliable information. Following Dillman's (2000) tailored survey method, we called the target participants to verify their job titles and mail addresses. Managers who declined participation were removed from the mailing list. The final sample contained 651 managers from 223 firms. Three weeks after the first mail-out, reminder letters and supplementary questionnaires were sent. Of the 651 questionnaires sent, 120 were completed and returned. One participant was removed from the sample as s/he had less than one-year experience in the current position. This results in a total of 119 usable responses and a 19.29% response rate, which is at par with response rates in prior management accounting studies, such as Burkert et al. (2011) (12.6%) and Hall (2008) (22.5%).

Table 1 presents the demographic information of the participants who, on average, were 47 years old, with 11.7-year experience in their current positions and had about 76 employees in his/her functional area. The numbers of participants in manufacturing/production, human resources and other categories were similar, while the number of participants in the sales and marketing function category was slightly higher (30.30%).

To test for non-response bias, an independent two samples t-test was performed through SPSS v25. Of the 119 useable responses, 64 early-responses and 55 late-responses were identified based on the cut-off date. No significant differences were found between the two samples for all variables of interest.

3.2 Measurement of constructs

The construct of formality is adopted from Hartmann and Slapničar (2009),3 which contains three aspects of a performance evaluation system: target setting, performance measurement and rewards (Hartmann & Slapničar, 2009) (see Appendix for all measurement items). For target-setting, two items (FORM1 and FORM2) measured the extent of formality that superiors set subordinates' work objectives. Items are anchored by a 5-point4 Likert scale, from ‘1 = express work objectives implicitly and in qualitative terms’ to ‘5 = express work objectives explicitly and in quantitative terms’. For performance measurement, two items (FORM3 and FORM4) measured the extent of formality in using information when evaluating their subordinates' achievement, from ‘1 = evaluated performance based on superior's judgment and in qualitative terms’ to ‘5 = based on information systems and in quantitative terms’. Two items (FORM5 and FORM6) measured the extent of formality that superiors determine subordinates' rewards (fixed pay and bonus), from ‘1 = determine pay based on superior's judgment and in qualitative terms’ to ‘5 = determine pay based on systematic information and in quantitative terms’. All six items have factor loadings above 0.5 and explained approximately 51.43% of construct variance. The construct has a satisfaction level of composite reliability of 0.809 and Cronbach's α of 0.796. The results are presented in Tables 2–4.

Procedural fairness is measured by four items (PF1–PF4) developed by McFarlin and Sweeney (1992) on a 5-point Likert scale. Participants were asked to rate the extent to which they agreed with the fairness in target setting, performance, pay and the entire performance evaluation system, from ‘1 = completely disagree’ to ‘5 = completely agree’. Similar to prior studies (e.g., He & Lau, 2012; McFarlin & Sweeney, 1992), this construct shows high level of validity with a composite reliability of 0.897 and a Cronbach's α of 0.847, and all four items are loaded on a single factor with factor loadings above 0.7 and account for 69.55% of variance.

Precision and sensitivity constructs were adopted from Moers (2006) and Burkert et al. (2011).5 Precision (PRE1–PRE4) captured the extent to which managers perceive their performance measures are free of noise, such as ‘my performance measures are precise, that is, the influence of uncontrollable factors is minimal’. Participants were asked to rate the degree of agreement on statements, ranging from ‘1 = strongly disagree’ to ‘7 = strongly agree’. Sensitivity (SEN1–SEN4) captured the extent to which manager perceive they can influence the performance measures through their actions, such as ‘with my actions I can influence my performance measures’. The four items of sensitivity have factor loadings above 0.852, a composite reliability of 0.926 and a Cronbach's α of 0.893; and the four items of precision have factor loadings above 0.798, a composite reliability of 0.904 and Cronbach's α of 0.859. Total variances explained by these two factors are 71.96% and 72.03%, respectively.

The construct of role clarity is adopted from Rizzo et al. (1970). It consists of six items on a 7-point Likert scale. Participants were asked to rate their levels of agreement in relation to their planned goals and objectives, clear explanation, responsibilities, performance expectations, authorities and workloads. The scale is anchored from ‘1 = strongly disagree’ to ‘7 = strongly agree’. This construct has been well tested in prior research, including Chenhall and Brownell (1988), and He and Lau (2012). All six items loaded on to a single factor with 63.08% variance explained, have factor loadings from 0.523 to 0.874, and a composite reliability of 0.908 and a Cronbach's α of 0.877.

Hartmann and Slapničar (2012) found that low task uncertainty moderates the relationship between formality and procedural fairness, so we captured task uncertainty as one of control variables in our study. In particular, we used three items to capture the task analysability dimension of task uncertainty, which reflect employees' perception about their job requirements and information needed to perform their jobs (Hartmann & Slapničar, 2012; Kren, 1992).6 The participants were asked to indicate their level of agreement on three statements on a 7-point scale: (1) ‘I am always clear about what is necessary to perform well on my job’; (2) ‘I have adequate information to make optimal decisions to accomplish my performance objectives’; and (3) ‘I am able to obtain the strategic information necessary to evaluate important decision alternatives’ (Burney & Widener, 2007; Kren, 1992). The results reveal the mean scores of the three statements are 5.34, 4.88 and 4.87, respectively. Such high scores indicate that the managers face relatively low task uncertainty. These three statements load on to a single factor with satisfactory factor loadings from 0.870 to 0.926, a composite reliability of 0.927, Cronbach's α of 0.882.

We established the constructs' discriminant validity by examining Fornell–Larcker criterion and Heterotrait–Monotrait Ratio (HTMT) and present the results in Table 4. While Fornell–Larcker is commonly used in accounting studies, HTMT is said to be the most robust test for discriminant validity (Bedford & Spekle, 2018). An HTMT value below 0.9 indicates the discriminant validity between two constructs has been established (Hair et al., 2017, p. 119). Given that the HTMT values of the five key variables in our model are well below 0.9, we conclude that these constructs are distinctive. However, the HTMT value of task uncertainty and role clarity is 0.932 (i.e., the two constructs are highly correlated), we excluded task uncertainty from our main path analysis.

3.3 Examination of potential biases

Harman's one-factor test, and marker variable method indicated that common method bias is not a concern in this study. In Harman's (1967) single factor test through SPSS v25, all indicators of the constructs loaded on a single factor and the explained variance is 30.13%, less than the threshold of 50%. In the marker variable method (Lindell & Whitney, 2001) using SmartPLS 3, all constructs are linked to a marker variable. The results show no significant associations between the constructs of interest and the marker variable. Collectively, all constructs used in this study are of good quality.7

4.1 Main analysis

We conducted our analysis of the hypotheses through a step-wise fashion recommended by Hayes (2009), which is also consistent with Hartmann and Slapničar (2009) (Baron & Kenny, 1986; Hartmann & Slapničar, 2009).8 First, we tested the direct relationship between formality and procedural fairness (Stage 1), we then examined how this direct relationship altered with the introduction of mediating variables of perceived precision, perceived sensitivity and role clarity (H1–H4) (Stage 2). The results of direct relationship model indicate a significant positive direct relationship between formality and procedural fairness (β = 0.418, t = 5.502, p < 0.001, one-tailed) (see Table 5), which is line with our expectation.

When precision (PRE) is introduced into the model, the path between formality (FORM) and precision (PRE) is significant (β = 0.304, t = 3.644, p < 0.001, one-tailed) (see Table 5 and Figure 2), however, the path between precision (PRE) and procedural fairness (PF) is insignificant (β = 0.046, t = 0.514, p = 0.299, one-tailed), which suggest the relationship between formality and procedural fairness is not mediated by precision (β = 0.014, t = 0.466, p = 0.314, one-tailed). In addition, we examined the confidence interval to determine the existence of significant mediation (Hayes, 2009; Preacher et al., 2007). The confident interval for precision (H1) contains zero (LLCL = −0.038 and ULCL = 0.061), suggesting a non-significant mediation. Thus, H1 is not supported.

H2 predicts that the relationship between formality and procedural fairness is positively mediated by sensitivity. As reported in Table 5 and Figure 2, the results indicate that the path between formality (FORM) and sensitivity (SEN) is significant (β = 0.178, t = 2.007, p = 0.019, one-tailed), and the path between sensitivity (SEN) and procedural fairness (PF) is also significant (β = 0.259, t = 3.157, p = 0.001, one-tailed). After incorporating sensitivity, the direct association between formality and procedural fairness remains significant (β = 0.231, t = 2.734, p = 0.003, one-tailed). According to Baron and Kenny (1986), an indirect association is established when three conditions are met: (1) the independent variable significantly accounts for the variations in the presumed mediator; (2) the variations in the mediator significantly account for the variations in the dependent variable; and (3) when the mediator is incorporated, the previously significant association between the independent variable and dependent variable is reduced (Baron & Kenny, 1986). Therefore, a partial mediation exists when the initial direct relationship between the independent variable (FORM) and the dependent variable (PF) significantly decreases in strength but still exists (Burkert et al., 2011, p. 152). The indirect effect of formality on procedural fairness through sensitivity is 0.046, which is significant (t = 1.847, p = 0.033, one-tailed) and close to the meaningful threshold of an absolute amount of 0.05 (Lau & Moser, 2008, p. 67). Furthermore, the confident interval for sensitivity does not contain zero (LLCL = 0.013 and ULCL = 0.092), suggesting a significant mediation. Therefore, the results suggest that sensitivity partially mediates the relationship between formality and procedural fairness. Thus, H2 is supported.

H3 predicts the relationship between formality and procedural fairness is positively mediated by role clarity. The path between formality (FROM) and role clarity (RC) is positive and significant (β = 0.262, t = 3.769, p < 0.001, one-tailed), and role clarity (RC) is also positively associated with procedural fairness (PF) (β = 0.315, t = 4.013, p < 0.001, one-tailed). The indirect effect is 0.082 and significant (t = 2.572, p = 0.007, one-tailed). The confident interval does not contain zero (LLCL = 0.033 and ULCL = 0.134). Thus, H3 is supported.

H4a and H4b predict the relationships between precision and procedural fairness and, sensitivity and procedural fairness are positively mediated by role clarity. As reported in Table 5 and Figure 2, the results show precision (PRE) is not significantly associated with role clarity (RC) (β = 0.013, t = 0.144, p = 0.445, one-tailed) whereas sensitivity (SEN) is significantly associated with role clarity (RC) (β = 0.497, t = 6.06, p < 0.001, one-tailed), and the path of role clarity (RC) to procedural fairness (PF) is also significant (β = 0.315, t = 4.013, p < 0.001, one-tailed). Hence, the path of precision to procedural fairness through role clarity is not significant, H4a is rejected. The indirect effect of sensitivity on procedural fairness through role clarity is 0.156 and significant (t = 3.148, p = 0.001). Further, the confident interval does not contain zero (LLCL = 0.085 and ULCL = 0.248). Thus, the mediating effect of role clarity on the path of sensitivity to procedural fairness is significant. H4b is supported.

4.2 Additional analysis

To gain a better understanding of the relationships among the variables of interest, we conducted additional analyses examining the impacts of several factors, including managerial functions, the level of task uncertainty, participants' experience in their current positions and their ages. Abernethy and Brownell (1997) and Moers (2005) suggested that functional areas, such as manufacturing and sales/marketing are likely to adopt more quantitative measures and objective information, while supporting functions, such as human resources, information technology, strategic development and research and development, are likely to adopt more qualitative measures and superiors often use subjective judgement. Hartmann and Slapničar (2009) labelled the difference in the ease of measuring functions' measurable outputs as contractability. Contractability was proposed to be different in their study of bank managers, who were either doing front-office or back-office functions. Front-office functions which include granting loans and hiring deposits, are different from back-office functions that fulfilled reporting requirements and provided legal counsel, because the former have direct financial consequences that are measurable and verifiable whereas the latter do not. Given its implications on managers' perceptions, we also examined whether participants' managerial functions have an impact on the hypothesised relationships among the variables of interest.

The 119 participants were split into two groups based on their responsibility areas: (1) quantitative functions (i.e., sales and marketing, manufacturing and production, logistics functions), which contain 72 responses; and (2) qualitative functions (i.e., human resources, engineering, strategic management, IT and other supporting functions), which contain 47 responses. Path models for each group were drawn and similar testing procedures were applied. We conducted SmartPLS Multi-group Analysis (MGA) comparing the results of the hypotheses between the two groups.

The results are presented in Table 6. The only significant difference between the quantitative and qualitative function groups is the association between formality (FORM) and precision (PRE) (β _difference = 0.337, p = 0.001, one-tailed). The association is not significant in the quantitative functions (β = 0.171, t = 1.324, p = 0.093, one-tailed) but significant in the qualitative functions (β = 0.508, t = 4.837, p < 0.001, one-tailed). This can be interpreted to mean that higher level of formality of performance evaluation systems in the qualitative functional areas has a stronger impact on managers' perception of precision of performance measures than in those quantitative functional areas.

Task uncertainty may also influence the hypothesised path relationships (e.g., Hartmann & Slapničar, 2012), though excluded from the main path analysis due to a lack of discriminant validity. We examine its potential effects as part of additional analysis. Like Hartmann and Slapničar (2012), we dichotomised the sample into higher and lower task uncertainty subgroups (higher TU vs. lower TU) using the mean score of 5.031. The higher task uncertainty group contains 56 cases, and the lower task uncertainty subgroup contains 63 cases. The PLS-MGA results are presented in Table 7.

Panel B shows that the association between formality and precision is significant in the higher-TU group (β = 0.428, t = 3.099, p < 0.001, one-tailed), and that this association is stronger than that in the lower-TU group (β = 0.221, t = 1.157, p = 0.124, one-tailed), despite the difference between two subgroups was not statistically significant. Likewise, the association between formality and sensitivity is significant in the higher-TU group (β = 0.289, t = 2.436, p = 0.008, one-tailed) but not significant in the lower-TU group (β = −0.060, t = 0.43, p = 0.334, one-tailed). The difference between the two subgroups is significant (β = −0.349, p = 0.03, one-tailed). These findings are in line with our prediction that formality enhances precision and sensitivity and does so differentially.

To gain more understanding of the effects of different degrees of formality, we also divided our sample into high-formality and low-formality groups using the median value of 3.9 The results presented in Table 8 show that formality is significantly associated with role clarity in the high-formality group (β = 0.415, t = 4.314, p = 0.000, one-tailed) but not significant in the low-formality group (β = 0.088, t = 0.521, p = 0.302, one-tailed), and this difference is statistically significant (β = 0.327, p = 0.039, one-tailed). As such, the mediating effect of role clarity is only significant in the high-formality group.10

We further analysed the potential effects of participants' age and work experience in both SmartPLS 3 and Amos 25 but no significant impact on the hypothesised relationships was detected. Furthermore, a separate moderation model was analysed to test the potential moderating effects of precision and sensitivity on the relationship between formality and procedural fairness. The results reveal that sensitivity positively moderates the relationship between formality and procedural fairness (β = 0.255, t = 3.196, p < 0.003) whereas there is no moderating affect found for perceived precision (β = 0.049, t = 0.607). This finding suggests formality leads to procedural fairness when there is a higher level of sensitivity. Sensitivity is a function of formality. When sensitivity is strong, managers will think that the use of formal performance evaluation is more supportive and favourable in reflecting subordinates' interest, and as such, formality has a stronger influence on procedural fairness. In other words, with a stronger sense of sensitivity, the association between formality and procedural fairness is stronger, and vice versa.