Introduction

Large-scale organizational scandal is often preceded by an escalation of unchecked, low-level conduct violations that are significantly vaguer in their unethicality (Welsh et al., 2015). For example, the Wells Fargo fake-accounts scandal involved thousands of employees engaged in opening millions of potentially unauthorized or fraudulent accounts for customers (Kouchaki, 2016). In this case, employees engaged in unethical practices not only to benefit themselves but also to benefit the organization. Thus, the wrongness of such behaviors can be vague given the conflicting motives that may vary in salience or ascribed importance – a single incidence of this behavior could be overlooked as an over-zealous, proactive employee. Following Umphress and Bingham (2011: 621), such unethical behaviors are known as unethical pro-organizational behaviors (UPB), defined as "actions that are intended to promote the effective functioning of the organization or its members (e.g., leaders) and violate core societal values, mores, laws, or standards of proper conduct."

Prior research has documented that employees are more inclined to cross the ethical boundaries to help their organizations in cases where, they are more attached to their organization (Naseer et al., 2020; Umphress & Bingham, 2011) or the members in their organization (Johnson & Umphress, 2019), are more exposed to their supervisor’s unethical behavior (Lian et al., 2020), feel more obligated to reciprocate supportive leadership (Shaw & Liao, 2020), or are less morally sensitive (Matherne et al., 2018). Despite the breadth of scholarship interest in why UPB occurs, there remains relatively limited research on what happens afterwards, especially to UPB offenders themselves (see reviews by Mishra et al., 2021; Mo et al., 2023). The few existing studies on the aftermath of UPB tend to concentrate on affective mechanisms that are conducive to the organizational functioning (Tang et al., 2020; Wang et al., 2022), or the dual processes where affect as a positive enabler (i.e., guilt prompts prosocial responses), while cognitive processes act as a negative enabler (i.e., psychological entitlement prompts antisocial responses) (Chen et al., 2022). By contrast, in the current research, we attempt to further draw out the paradoxical nature of the cognitive processes associated with UPB – i.e., cognitive ambivalence – illustrating that employees may make sense of their own UPB in ways that simultaneously prompt prosocial and antisocial action.

UPB is inherently paradoxical; by definition, it is characterized by a prosocial motivation to benefit the organization, while also flouting broader ethical norms or standards (Umphress et al., 2010). This inherent paradox not only precedes the decision to engage in UPB but may also prompt cognitive ambivalence following acts of UPB. Relatedly, emerging empirical studies (cf. Jiang et al., 2022; Liao et al., 2024) has examined this cognitive ambivalence (i.e., moral deficits contrasted against psychological entitlement) following UPB, illustrating their paradoxical link with behaviors that are both prosocial (e.g., helping behaviors) and antisocial directed at the organization (i.e., deviance or reduced work effort). Our research extends their work by examining two distinct cognitive paths: perceived moral credits loss, which aids in course-correcting and moral disengagement, which promotes the escalation of UPB. The inclusion of this moral escalation path not only extend beyond examining "unethical pro-self behavior" (i.e., deviance) in previous research, but also address a critical theoretical inquiry: Why does UPB persist? Moreover, our study complements previous research by examining effects over a longer time frame, capturing cumulative impacts of UPB that may not be observable in shorter timeframes.

In particular, we draw on the theoretical arguments from moral self-regulation theory (Zhong et al., 2010) and moral disengagement theory (Bandura, 1990) to theorize a dual mechanism – recovery and recidivism (defined as the tendency of individuals to reoffend after previously being involved in ethically, socially, or legally problematic activities, Vugt et al., 2011). First, the unethical aspect of UPB can elicit a process of moral recovery: perpetrators compensate for their damaged moral self-concept, captured by the perceived loss of moral credits, by increasing constructive behaviors toward likely victims of UPB (operationalized herein as customer-directed citizenship behavior, or OCB-C, Bettencourt & Brown, 1997). Second, the pro-organizational aspect of UPB can prompt a process of moral escalation: perpetrators alleviate their threatened moral self-concept by rationalizing their unethical behavior as morally justifiable given the intent to benefit their organization, captured by moral disengagement, which leads to a "slippery slope" of repeated UPB.

This research advances our theoretical knowledge on UPB in several ways. First, this research extends this growing stream of research on the aftermath of UPB by juxtaposing two seemingly opposing moral-based cognitive paths – perceived loss of moral credits that promotes recovery and moral disengagement that promotes recidivism. Second, the extant research on employee reactions to UPB has primarily focused on positive, compensatory responses, but does not account for the fact that UPB might repeat itself or even escalate over time (as in the opening Wells Fargo example). This research is therefore particularly novel in its consideration of future UPB as a potential response. Notably, this moral escalation pattern enriching the ongoing scholarly discussion on UPB (e.g., Tang et al., 2020; Wang et al., 2022) by showing that reactions to one's own UPB are not always positive but can also result in future unethical choices. In doing so, this research also provides potential evidence of a slippery slope within the UPB context, offering a contrast to recent compensatory accounts. These innovations are theoretically and practically important because they provide a more comprehensive perspective of the downstream consequences of UPB, and help organizations develop policies and processes that both promote moral responsibility and discourage rationalization, contributing to a culture of accountability that prevents UPB escalation to the point of scandal.

Theoretical Background and Hypotheses Development

The Moral Self-Regulation Mechanism Following UPB

Complementing the above discussed emotion-driven compensatory process, our research focuses on the cognitive elements of moral self-regulation. Central to moral self-regulation theory is the notion that "threats to an individual's moral image induce compensatory behaviors to repair the moral self" (Zhong et al., 2009, p. 82). Being moral is an important aspect of a person's self-concept, and people strive to behave in alignment with their moral self-concept (Aquino & Reed, 2002). This desired moral self is also used as a yardstick to monitor the inconsistencies between the desired self and the actual behavior. When an actual behavior compares poorly, it will damage moral self-concept and, as a result, evoke reactions to rectify the inconsistencies between the desired moral self and the actual behavior through compensatory behaviors (Gromet & Okimoto, 2014; Zhong et al., 2009).

In the context of UPB, we posit that, by nature of UPB being immoral and norm-violating, engaging in UPB compromises one's moral self-concept, prompting them to engage in behaviors that help to restore their moral self-concept. In conceptualizing this process, moral self-regulation research suggests that an individual's moral self-concept is akin to a bank account (Liao et al., 2018). Individuals monitor fluctuations in this account and endeavor to maintain a favorable moral self-concept by balancing their moral and immoral actions (Zhong et al., 2009). When they engage in morally problematic actions, they will perceive a loss of moral credits, which prompts them to take more moral actions in the future. Supporting this notion, Zhong et al., (2010, p.2) posited that "after an unethical first choice, people acted significantly more ethically in their subsequent choice". That is, following ethical lapses, individuals tend to increase their moral behaviors toward the same targets to mend their damaged moral self-concepts (Zhong & Liljenquist, 2006; Zhong et al., 2009), rather than merely withdrawing the previously morally problematic behavior, as increasing moral behaviors is more likely to "reaffirm the values that have been undermined" (Gollwitzer & Melzer, 2012: 1356). Relatedly, empirical research has further illuminated this mechanism. For example, Liao et al.’s (2018) study showed that leaders tend to display more consideration to their subordinates to recover the deficit of moral credits following their abusive supervision.

This cognitive account is particularly relevant for service contexts, as UPB in this context usually involves ethically questionable tactics, such as lying and deception that are against service employees' ethical responsibilities to customer wellbeing (Kaptein, 2008). This may elevate concern for UPB victims and thus heighten the experience of incongruence between the desired moral self and the actual behavior. Therefore, we propose that the unethical aspect of UPB (i.e., incurring costs for inflicting harm on customers) will undercut individuals' moral self-concept, manifested in the perception of a shortfall of moral credits.

This, in turn, will prompt the perpetrator of UPB to increase helping behaviors toward customers as a way to restore their moral credits and thereby recover their ethical failings and damaged moral self-concept. In particular, aligned with previous research and our target population of frontline service employees, we focus on OCB-C, defined as the discretionary behaviors that employees initiate for the benefit of customers (Bettencourt & Brown, 1997; Dimitriades, 2007), as an avenue to recover the perceived loss of moral credits following their transgression. Examples of OCB-C include going the extra mile and providing extra attention when serving customers (Bienstock et al., 2003). Considerable research has documented that OCB-C is beneficial to customers as it is linked to greater customer satisfaction (Yen & Niehoff, 2004). We therefore expect that employees who engage in UPB are likely to engage in OCB-C subsequently as a result of the perceived loss of moral credits.

Hypothesis 1:

UPB is positively related to perceived loss of moral credits (a); and perceived loss of moral credits mediates the positive relationship between UPB and OCB-C (b).

The Moral Escalation Mechanism Following UPB

UPB is distinguished from other forms of unethical workplace behaviors as it is motivated by a prosocial desire to advance the interests of the organization (Umphress et al., 2010). As such, it can engender ambivalent cognitive states in individuals: UPB is both good and bad. In the moral recovery mechanism depicted earlier, the unethical and norm-violating aspect of UPB undercuts an individual's moral self-concept, promoting UPB actors to engage in OCB-C to restore moral credits. On the other hand, individuals can resolve the threat to their moral self by focusing their cognitive attention on the pro-organizational (positive) aspect of UPB, elevating their positive feelings in light of their willingness to do what is necessary to serve the greater good of the organization. Focusing on the favorable aspects of their past behavior can disrupt the connection between individuals' moral self-regulation systems and their thoughts and actions, facilitating moral disengagement (Bandura, 1999).

Grounded in social cognitive theory, moral disengagement refers to a set of cognitive justification mechanisms that allow people to circumvent internalized moral standards and commit unethical acts without self-condemnation (Bandura, 1990; 2001). In the literature, moral disengagement has been analyzed as an individual difference, which is a trait-like construct that predicates a pattern of unethical behavior, or a process, which is a state-like construct that follows as a consequence of unethical behavior (Schaefer & Bouwmeester, 2021). For example, Lee et al. (2016) theorized moral disengagement as an outcome state variable following social undermining; and Shu et al. (2011) investigated moral disengagement as a post-transgression cognitive process of self-serving rationalizations. Consistently, given the focus of this research is to unpack the underlying cognitive process following an unethical behavior, we conceptualize moral disengagement as a state variable that involves a set of cognitive rationalization mechanisms following morally problematic behaviors.

We posit that the pro-organizational aspect of UPB is likely to make it easier to morally disengage, as individuals can reconstrue their understanding of the offence by focusing on their good intention and its perceived immediate benefits to the organization. Moral disengagement, in turn, enables the perpetrator to decouple their internal moral standards from how they construe their past behaviors, rendering them more acceptable (Lawrence & Kacmar, 2017; Moore, 2015). Further, such cognitive rationalization process can even enable the degradation of individual behavior into a further downward spiral, producing a 'slippery slope' pattern of ethical failings (Welsh et al., 2015). So, not only does past behavior serve as a guide for future ethical choices, but moral disengagement makes previous UPB less problematic in retrospect, facilitating future unethical acts when similar scenarios arise. Therefore, in contrast to a recovery path (moral self-regulation), the moral disengagement perspective suggests a recidivism path where UPB can also lead to an escalating trend in future UPB. Thus, while the moral self-regulation process suggesting an upward constructive behavioral trend, moral disengagement explains why people might also be prone to a downward spiral of UPB. Therefore, we hypothesize the following:

Hypothesis 2:

UPB is positively related to moral disengagement (a); and moral disengagement mediates the positive relationship between UPB (Time 1) and UPB (Time 2) (b).

Overview of Studies

This research investigated frontline service employees' responses to their own UPB, a context in which perpetrators' relationship with the potential victims (i.e., customers) are particularly salient, and where there is ample opportunity for UPB. Importantly, as previously discussed, the service context may heighten the experience of ethical dissonance, making this an ideal context for the study. We conducted two studies to test our hypotheses. In Study 1, following the approach from Wang et al. (2021), we designed an experimental scenario for participants (N = 196) that presented an opportunity to engage in UPB. We then examined whether those who chose to conduct UPB experienced a greater perceived loss of moral credits and were more likely to morally disengage. This study allowed us to establish causal relationship between UPB and these cognitive responses. In Study 2, to enhance external validity and test the entire model in a real-world setting, we conducted a time-lagged critical incident survey with full-time customer-facing employees (N = 253), where they were to recall and describe a specific instance of UPB they had engaged in.

Study 1

Procedures and Participants

Prior to this main study, we conducted an initial pilot study (N = 50) to explore the prevalence and nature of UPB in the service context. This pilot study confirmed the frequency of UPB among customer-facing employees and typical types of UPB. Building on these findings, we recruited 200 full-time adults working in customer-facing roles through Prolific, an online research platform widely used in organizational research (Peer et al., 2017; for recent examples, see Loi et al., 2023; Park et al., 2023). After removing four participants who failed the attention checks, a total of 196 participant were included in the final sample. This sample included participants from the UK (89.47%) and the US (10.53%). Participants had an average age of 40 years (SD = 10.92) and an average tenure of 7 years (SD = 6.49), 67% were men, and 73% have an education level of university or above. Regarding job roles, 23.42% worked in information technology, 14.56% in software development, and 8.86% in marketing, with the remainder distributed across various customer-facing roles in other sectors.

Following the approach of Wang et al. (2022), we designed a scenario where participants were instructed to imagine themselves as a sales manager in a software development company. Participants were informed that “In this study, you will be asked to imagine you are a sales manager in a software company. You will complete a task related to selling antivirus software and then answer a list of questions related to the task”. We adapted Wang et al.'s (2022) study, which involved participants helping a face mask manufacturer design an advertisement for different types of masks accidentally mixed up during a warehouse change. We modified it to a software sales context to create a similar ethical dilemma for engaging in UPB in a non-pandemic scenario. The scenario read as follows:

"The company has two versions of the antivirus software: a "Premium" version that provides high-level protection against a wide range of cyber threats, and a "Basic" version that provides standard protection against common viruses. The "Premium" version is sold at a higher price than the "Basic" version, reflecting its superior protection capabilities. The market price range for similar "Premium version" product is typically between $60 and $80 for a 1-year subscription. The market price range for similar “Basic version” product is typically between $30 and $50 for a 1-year subscription. The two versions have the same user interface and look identical to the user, but the "Premium" version has an advanced codebase and enhanced features. However, due to an issue in the company's inventory system, the "Premium" and "Basic" software codes have been mixed up, and it is difficult to distinguish between them."

The ambiguity created by the mixed-up software codes mirrors the indistinguishability of the mixed-up face masks in Wang et al.’s (2022) study. It also provides the necessary moral wiggle room for this decision, mimicking real-world ethical challenges ins sales context (Dana et al., 2007). Participants were then given two options to sell this software: Option A, to sell all the software as the "Premium" version to potentially maximize profit at the risk of misleading consumers, or Option B, to sell all as the "Basic" version, safeguarding consumer interests. This design creates an ethical dilemma for participants, which allows us to distinguish UPB (i.e., Option A) from non-UPB (i.e., Option B).

Measures

UPB Acting

Participants were tasked with making sales decisions around three choices. First, participants were asked to select a label strategy for the software with two options presented: label the software as the "Premium" or label the software as the "Basic". Second, participants were asked to choose between two pricing strategies for a 1-year subscription: set the price at $70, reflective of the "Premium" version's market value or set the price at $40, reflective of the "basic" version's market value. The final decision involved selecting an advertisement slogan that would best represent the product under their chosen strategy. Option A was "Maximum Protection Against All Threats," suggesting the comprehensive coverage of the "Premium" version. Option B, "Standard Protection for Everyday Use," communicated the adequate but limited protection offered by the "Basic" version.

For each of the above three decisions regarding the sale of antivirus software, participants who selected options indicative of prioritizing company profits over consumer interests (e.g., labelling all software as the "Premium", choosing the higher pricing option of $70 for a 1-year subscription, and opting for the "Maximum Protection Against All Threats" slogan) were assigned 1 point on UPB. Conversely, participants who chose options that favored consumer interests (e.g., labelling all software as the "Basic", pricing the subscription at $40, and choosing the "Standard Protection for Everyday Use" slogan) received 0 points on UPB. This resulted in a total score ranging from 0 to 3.

Perceived Loss of Moral Credits

We measured perceived loss of moral credits using a five-item scale developed by Lin et al. (2016) and adapted to the ethical behavior context by following Liao et al. (2018). Participants were asked to reflect on the choice they just made and then indicated to what extent they were experiencing the following thoughts; e.g., "I lost moral credits for performing this behavior."

Moral Disengagement

We measured post-transgression moral disengagement by using five items from a scale developed by Shu et al. (2011). We excluded one item "Rules should be flexible enough to be adapted to different situations", as it is a more generalized item that does not suit the specific context of UPB in this study. Participants were asked to reflect on the choice they just made and then indicated to what extent they were experiencing the following thoughts; e.g., "Such behavior is appropriate because no one gets really hurt."

UPB Perception

To validate the measurement of UPB, following the approach of Wang et al. (2022), we measured "UPB perception" as a manipulation check to assess the extent to which participants agree they had conducted UPB in the task. We used a four-item short version of the original 6-item UPB scale (Umphress et al., 2010), removing two items that were not applicable to our service context: “To help our company, I had given a good recommendation on the behalf of an incompetent employee in the hope that the person will become another organization’s problem; and To help our company, I had withheld issuing a refund to a customer or client accidentally overcharged”. This adaptation is consistent with recent UPB research involving service employees (e.g., Liao et al., 2024; Tang et al., 2020, 2021; Wang et al., 2022; Yan et al., 2023). We also examined the correlation between the measure of UPB acting (the choice regarding labeling, pricing, and promotion) and participant’s UPB perception (r = 0.48, p < . 001), suggesting adequate validity of the UPB acting measure.

Control Variables

Aligning with previous UPB research (e.g., Chen et al., 2016), we measured social desirability to account for participant’s tendency to respond in a socially desirable way when answering sensitive self-report items in this study (i.e., our moral disengagement and perceived moral credit loss measure), we measure participants' impression management bias using a subscale from Steenkamp et al.'s (2010) social desirability bias scale. We also controlled for moral identity using the five-item internalization subscale of moral identity (Reed & Aquino, 2003), because research suggests that this aspect of moral identity can influence an employee's moral behavior (Matherne et al., 2018;) as well as how people interpret moral-related issues (Lin & Loi, 2019).

Results and Discussion

Means, standard deviations, correlations of all variables are presented in Table 1. Coefficient alphas are shown in parentheses on the diagonal. Results show that 87 participants did not conduct UPB (with a score of 0), 66 participants scored 1, 23 participants scored 2, and 20 participants scored 3. ANOVA results indicate that these four groups varied in their perception of UPB (F[3, 192] = 20.24, p < 0.001, η2 = 0.240 (95% CI [0.135, 0.328]).

Table 1 Means, Standard Deviations, Correlations, and Internal Consistency Estimates in Study 1
Full size table

A multivariate analysis of variance (MANOVA) was conducted to examine the effect of UPB on perceived loss of moral credits and moral disengagement while controlling for moral identity and social desirability. The results showed significant main effects of UPB on both perceived loss of moral credits (F[3, 192] = 15.393, p < 0.001, partial η2 = 0.196) and moral disengagement (F[3, 192] = 7.760, p < 0.001, partial η2 = 0.109).

Given the potentially sensitive nature of our constructs and the count nature of our UPB measure, we tested the effect of UPB on perceived loss of moral credits and moral disengagement using Mplus 8.9 with WLSMV (weighted least squares mean and variance adjusted) estimator. This estimator does not assume normality and is appropriate for models with categorical or ordered variables (Muthén & Muthén, 2017). The effect of UPB on perceived loss of moral credits is (b = 0.25, p < 0.001) and on moral disengagement is (b = 0.20, p = 0.002), thus supporting Hypotheses 1a and 2a. Through designing a context and a task for participants to engage in UPB, this study tested the causal relationship from UPB to two cognitive responses, revealing that the enactment of UPB can simultaneously evoke a heightened sense of moral credits loss and moral engagement.

Study 2

Procedures and Participants

To test the entire hypothesized model, we employed a two-wave survey design where working adults with customer-facing roles were recruited from Prolific to participate a two-part study with one month apart. Participants who were invited to the pilot and Study 1 were excluded from this study. Consistent with prior UPB research (e.g., Graham et al., 2019), at Time 1, we integrated a critical incident technique (Flanagan, 1954) in the survey design where participants were asked to think back over the past month to try to recall a time when they were involved in UPB (i.e., helping the organization at expense of a customer). Research has demonstrated that this approach is effective to examine an individual's emotional, cognitive, or behavioral reactions to a recalled situation (e.g., Aknin et al., 2013). Further, asking participants to recall contextual details enhances the accuracy and vividness of retrospection (Robinson & Clore, 2001). After writing about their UPB incident in detail, participants completed measures of the Time 1 UPB, perceived loss of moral credits, moral disengagement, demographic variables, and control variables. The mediating variables were collected immediately after the recall task because research shows that people's psychological reactions to their own behavior likely occur quickly (Salancik & Conway, 1975), but recognizing that this limits the causal interpretation of the relationship between the Time 1 UPB measure and Time 1 process measures. A total of 337 participants completed the Time 1 survey.

We screened out careless responses (i.e., participants who failed to pass any of the three attention-checking questions embedded throughout the survey) (N = 25). We also removed cases with significant missing data (> 50%) (N = 34), and cases of recalled incidents being inconsistent with the definition of UPB (N = 9). This resulted in 269 valid responses. One month later, these participants were invited to participate in the Time 2 survey, where they completed measures on OCB-C and Time 2 UPB. The choice of one-month temporal separation was informed by the pilot study findings, where most participants indicated that they engaged in UPB between a few times per year and a few times per month. Research has also shown that a one-month time lag helps to reduce common method variance (Ostroff et al., 2002). Further, this time lag offered the benefits of capturing significant predicted effects while avoiding erosion effects (Rindfleisch et al., 2008). This is also consistent with O’Laughlin et al.’s (2018) work, which underlines the importance of timing of measurement to capture mediation effects accurately as some effects may occur rapidly. Thus, the two-wave design allowed us to capture these immediate cognitive responses while still establishing temporal precedence for the behavioral outcomes.

A total of 263 submitted a complete survey at Time 2. After removing careless responses (N = 7), those with employment changes (N = 2), 253 matched samples were included for final analysis.Footnote 1 In the final sample of 253 participants, 58% were female, from various service sectors, primarily retail (30%), healthcare (16%), and hospitality and tourism (9%). The average age was 35 years (SD = 9.36), with an average tenure of 6.23 years (SD = 5.39). We divided valid participants at Time 1 (N = 269) into two groups: one group with valid responses at Time 2 (N = 253) and the other without (N = 16). Independent t-tests showed no significant differences between these two groups regarding Time 1 UPB (t = -0.43, p = 0.668), organizational tenure (t = 1.02, p = 0.307) and education (t = 1.50, p = 0.136), suggesting that our results were not likely to be influenced by selective attrition.

Measures

Responses for all items were made on a 7-point Likert-type response scale, from 1 = strongly disagree to 7 = strongly agree, unless otherwise noted.

UPB

We measured UPB using the same 4-item version as in Study 1. Moreover, following prior research (e.g., Yang et al., 2021), rather than probing behavioral intentions, we asked participants to think about their interaction with customers in the past month and rate their frequency on each item using a 7-point scale (1 = never to 7 = always); e.g., "Exaggerated the truth about my organization's products or service to customers to help my organization."

Perceived Loss of Moral Credits

We measured moral disengagement by using the same five items as in Study 1.

Moral Disengagement

We measured moral disengagement by using the same five items as in Study 1.

OCB-C

Early research on OCB-C posits that OCB-C extends the notion of work performance in the service domain beyond in-role service behaviors, capturing the discretionary elements of service behaviors that exceed formal role expectations (Van Dyne et al., 1995). Consistent with this conceptualization, we measured OCB-C using Bettencourt and Brown’s (1997) 5-item measure of extra-role customer service behavior. Participants reported the frequency of their enactment of OCB-C toward customers in general since the last survey (1 month ago) e.g., "Helped customers with problems beyond what is expected or required" (1 = never to 7 = always).

Control Variables

Consistent with Study 1, we controlled for moral identity internalization and social desirability. We also controlled whether employees’ salary is commission-based or not to account for the effect that high commission may motivate employees to engage in more UPB and OCB-C. Furthermore, we examined the relevance of demographic characteristics (i.e., gender and organizational tenure) as control variables, since research has shown that they can influence employees' UPB (Miao et al., 2013) and prosocial behaviors (e.g., Van Dyne & Pierce, 2004).

Preliminary Analysis

Means, standard deviations, and zero-order correlations of the study variables are presented in Table 2. Coefficient alphas are shown in parentheses on the diagonal.

Table 2 Means, Standard Deviations, Correlations, and Internal Consistency Estimates in Study 2
Full size table

Measurement Invariance over Time

Prior to testing our hypotheses, we examined whether measurement invariance existed across time for the latent variable of UPB. This is considered a pre-condition for adequately testing models with the same items collected at multiple time points (Little et al., 2007). Following Vandenberg and Lance's (2000) suggestions, we examined an unconstrained measurement model in which the factor loadings and indictor intercepts were allowed to be different across two time points. Then, we added additional constraints of time-invariant factor variances by setting the factor loadings equal across time (metric invariance). The chi-square change between these two models was not significant (Δχ2[3] = 4.557, p = 0.207). Thus, the relation between the latent variable UPB and its items was verified as consistent over time.

Confirmatory Factor Analysis (CFA)

We conducted a series of confirmatory factor analyses (CFA) using Mplus 8.7 (Muthén & Muthén, 2017) to examine the distinctiveness of the measured variables. We modelled the stability of UPB across time by allowing the items of UPB (Time 1) and UPB (Time 2) to covary (Little et al., 2007). The four-factor model provided a good fit to the data (χ2[216] = 428.864, CFI = 0.952, TLI = 0.944, RMSEA = 0.063, SRMR = 0.052; Hu & Bentler, 1999), with all factor loadings being statistically significant at the p < 0.001 level. We compared this four-factor model with alternative models by collapsing measures with the most conceptual overlap (see Zipay et al., 2021). As shown in Table 3, this model fits the data significantly better than several alternative models. The results supported the discriminant validity of the specified measurement model.

Table 3 Comparison of Alternative Measurement Models in Study 2
Full size table

Common Method Variance (CMV)

Although we included temporal separation when collecting data as a procedural remedy for CMV, our data may still be subject to CMV biases because the variables were collected from the same source. Thus, we used the unmeasured method factor approach (Podsakoff et al., 2003) to further test the presence of CMV. Specifically, a common method factor (all items loading onto a latent method factor) is added to the model. Then, following Schermuly and Meyer's (2016) approach, we fixed all unstandardized factor loadings associated with this method factor to 1, defining it as orthogonal to the other latent variables. The resulting model with the method factor fits the data neatly (χ2[232] = 448.116, CFI = 0.952, TLI = 0.943, RMSEA = 0.064, SRMR = 0.052). Next, we followed Castanheira (2016) in using CFI difference to compare this model with the original five-factor model. The CFI difference (0.001) was smaller than 0.010, suggesting that the method factor did not significantly improve the overall fit of the model (Widaman, 1985); therefore, CMV is unlikely to be a major problem in this study.

Hypotheses Testing

We tested a fully latent model using structural equation modeling (Full-SEM) in Mplus 8.7 with maximum likelihood estimation. The Full-SEM approach allows simultaneous estimation of multiple indirect paths and provides model fit indices. To appropriately test for indirect effects, we also modelled the direct effects of UPB (Time 1) on OCB-C (Time 2) and UPB (Time 2) (Preacher & Hayes, 2004), and the stability of UPB across time (Little et al., 2007). In addition, following Preacher and Hayes's (2008) recommendation for multiple mediator models, we allowed the two mediators (moral disengagement and perceived loss of moral credits) to covary. Our model provides good fit to the data (χ2[437] = 762.541, CFI = 0.934, TLI = 0.926, RMSEA = 0.055, SRMR = 0.069). Figure 1 presents unstandardized coefficient estimates for the overall research model.

Fig. 1
figure 1

Hypothesized results with unstandardized path coefficients (Study 2) Notes: N = 253. *p < 0.05, **p < 0.01, ***p < 0.001. Significance levels are two-tailed

Full size image

Hypothesis 1 predicted a positive relationship between UPB (Time 1) and OCB-C (Time 2) via perceived loss of moral credits, and Hypothesis 2 predicted a positive relationship between UPB (Time 1) and UPB (Time 2) via moral disengagement. As shown in Fig. 1, UPB (Time 1) was significantly positively related to perceived loss of moral credits (b = 0.21, p = 0.023) and moral disengagement (b = 0.37, p < 0.001). Further, perceived loss of moral credits was positively related to OCB-C (Time 2) (b = 0.10, p = 0.035), and moral disengagement was positively related to UPB (Time 2) (b = 0.27, p = 0.043).

We further tested the hypothesized indirect effect with a bias-corrected bootstrapping approach (MacKinnon et al., 2002). Specifically, we generate 95% confidence intervals (CI) based on 1,000 bootstrapped samples. The indirect effect of UPB (Time 1) on OCB-C (Time 2) via perceived loss of moral credits was 0.02, with a 95% bias-corrected CI of [0.003, 0.071]. The indirect effect of UPB (Time 1) on UPB (Time 2) through moral disengagement was 0.10, with a 95% CI of [0.011, 0.214].

Robustness Check

We conducted two robustness checks to strengthen our findings. First, we used a latent change score (LCS) approach (see Li et al., 2014; Matusik et al., 2021) to further examine the moral escalation path, as this involves a construct (i.e., UPB) measured at two time points. The LCS approach allows to directly test how moral disengagement relates to increases in UPB over time, addressing potential concerns about the stability of this relationship. Specifically, we created a latent change score (ΔUPB) based on two adjacent time points (McArdle, 2009), which is modeled as a latent variable with estimated measurement errors (Li et al., 2014). This ΔUPB represents the change in UPB from Time 1 to Time 2. We then regressed ΔUPB on moral disengagement. Results show that moral disengagement is significantly positively related to changes in UPB from Time 1 to Time 2 (b = 0.24, p = 0.038). This analysis provides additional for our hypothesis that moral disengagement leads to the increase of UPB over time, enhancing the robustness of our findings regarding the moral escalation path.

As a second robustness check, we tested alternative mediation paths to address potential concerns about the distinctiveness of our proposed mechanisms and to rule out alternative explanations. Specifically, we modeled the indirect effect of UPB (Time 1) on UPB (Time 2) via perceived loss of moral credits on UPB, and the indirect effect of UPB (Time 1) on OCB-C (Time 2) via post-moral disengagement. Results showed that the relationship between perceived loss of moral credits and UPB (Time 2) was nonsignificant (b = 0.13, p = 0.054), as was the relationship between moral disengagement and OCB-C (Time 2) (b = 0.09, p = 0.273). Furthermore, the indirect effects of UPB (Time 1) on UPB (Time 2) via perceived loss of moral credits was not significant (b = 0.03, p = 0.149). Likewise, there is no significant indirect effect of UPB (Time 1) on OCB-C (Time 2) via moral disengagement (b = 0.03, p = 0.298). These findings demonstrate that perceived loss of moral credits and moral disengagement are indeed specifically linked to their hypothesized outcomes (OCB-C and UPB, respectively).

Supplementary Analysis

Our theorizing proposed that UPB would lead to both recovery (OCB-C) and recidivism (subsequent UPB) via two mediating mechanisms—perceived loss of moral credits and moral disengagement. However, we did not initially explore which mediating mechanism played a more profound effect. Research has noted that different moral cognitive processes might function differently when predicting ethical conduct (Bandura et al., 1996). Therefore, we conducted a supplementary analysis to explore which effect—moral recovery or moral recidivism—was stronger in the context of post-UPB behavior. Specifically, following Hu et al. (2023) approach, we created a new variable that measured the difference between the two mediating paths. The results showed that this difference was not significant (b = -0.06, p = 0.251, 95%CI = [− 0.172, 0.043]), suggesting that both processes play important roles in shaping subsequent behaviors.

General Discussion

In addition to adversely affecting the functioning of the organization, UPB can also affect people's moral cognitions and subsequent behaviors. Consistent with the paradoxical nature of UPB, our research shows that the moral self-regulation mechanism and the moral escalation mechanism can both follow the enactment of UPB, promoting subsequent paradoxical behaviors that reflect recovery (i.e., OCB-C) and recidivism (i.e., more UPB). Below, we discuss the theoretical and practical implications of our findings.

Theoretical Implications

Overall, our research makes several contributions to theory by studying UPB from an offender-centric perspective. First, this study extends our understanding of how individuals react to their own unethical helping behaviors. Research to date has primarily presented remorse and recovery for consideration on how people respond to their own UPB. Indeed, our test of moral self-regulation provides converging evidence of moral recovery process where the unethical aspect of UPB dampens an individual's moral self-concept, prompting perpetrators of unethical behaviors rectify their wrongdoing and restore their moral credits through customer-focused constructive behaviors. However, in contrast to this recovery account, the moral disengagement literature indicates the prevalence of moral escalation, where wrongdoers will cognitively justify their behaviors as less problematic, resulting in repeat engagement in unethical acts (Shu et al., 2011). Our test of moral escalation provides empirical support for this process, where the pro-organizational aspect of UPB disrupts the connection between one's moral self-regulation systems and their actions, prompting perpetrators of UPB progress down a slippery slope with future UPB. Thus, taken together, our research integrates these contrasting views by evidencing that moral self-regulation aids in course-correcting the UPB of individual employees; while moral disengagement is a countervailing cognitive process that may also promote the escalation of UPB. In doing so, the current research corroborates the notion that "the dynamics of ethically questionable behavior such as UPB are not always so straightforward" (Fehr et al., 2019, p. 36).

Second, our focus on employees in a customer-facing service context offers another important complement to the UPB literature. Service scholars have long recognized that customer-facing employees tend to experience or behave differently from those who work in non-service contexts (e.g., accounts or IT analysts) as they face a more bifurcated social landscape – the organization on one side and customers on the other (Jiang et al., 2016). However, UPB research to date has paid relatively limited attention to the unique cognitive processes experienced by service employees (cf. Chen et al., 2022; Liao et al., 2024). From a moral self-regulation perspective, we found that following the use of ethically questionable tactics during service delivery, employees will remedy their cognitive feelings of immorality as their UPB can impose a cost on customers. As a result, these employees will exert more discretionary effort to remedy any harm caused to customers (i.e., OCB-C). In parallel, from a moral disengagement perspective, service workers can also cognitively rationalize harmful behaviors toward customers, for example, by reasoning that "I was able to sell more of our products over rival organizations". This rationalization, in turn, enables the likelihood of future similar behaviors. Thus, this research enhances the UPB literature by focusing on cognitive ambivalence as a potential trigger for paradoxical behaviors toward the UPB victim themselves: customers.

Third, our research addresses recent calls for more insight into the cognitive ramifications of UPB (e.g., Tang et al., 2020), an area with relatively limited empirical evidence (cf. Jiang et al., 2022; Liao et al., 2024). We shift focus from generic, unethical pro-self behaviors (i.e., reduced work effort, Jiang et al., 2022; deviance, Liao et al., 2024) to antisocial behaviors directed specifically at the victim group (i.e., customers). Doing so not only offers empirical support for the mechanism of moral escalation but also shed light on why UPB may become a chronic issue within organizations. Moreover, extending Jiang et al.’s (2022) work that examined the generic form of OCB as a morally laudable act to counteract moral credits loss, our research focused on recovering moral credits relevant to that particular victim group. Our study also offers a complementary perspective by focusing on a cognitive reaction centered on the self (i.e., moral disengagement), as opposed to reactions centered on the self in relation to others, (i.e., psychological entitlement) in their work. Additionally, from an empirical perspective, our study also complements Liao et al.’s (2024) focus on immediate or momentary effects of UPB by capturing cumulative impacts over a longer, one-month interval in Study 2. This extended timeframe allows us to observe how the behavioral consequences of UPB unfold and potentially accumulate over time, providing insights into the longer-term implications in this scenario.

Finally, regarding the findings on the moral escalation path, this research contributes to the UPB literature by including subsequent UPB as a potential response. Our findings shed light on the dark side of moral self-regulation, where individuals relax their self-regulation and rationalize their behaviors, absolving moral accountability and predicting repetition of problematic behavior. By providing evidence that moral disengagement also follows UPB, and that such cognitive justification predicts subsequent UPB recidivism, we proffer evidence of a slippery slope (Shu et al., 2011; Welsh et al., 2015) within the UPB context where prior engagements in UPB can lead to future UPB. In their seminal theoretical paper on UPB, Umphress and colleagues (2011) suggest that UPB can create a sense of cognitive dissonance where people will change their perception through neutralization processes to resolve the dissonance; however, empirical support for such an account has remained ambiguous. The current investigation provides novel empirical evidence that UPB can evoke moral escalation through the process of moral disengagement. This finding is particularly important given that existing research typically focuses on constructive, moral compensatory responses among UPB perpetrators (e.g., Tang et al., 2020; Wang et al., 2022).

Practical Implications

Our research offers important practical implications for managers. It reveals that to resolve the cognitive experience of immorality, employees may, quite paradoxically, change their behaviors by performing constructive behaviors to help customers or change their beliefs and engage in further harmful behaviors toward customers. Although there is a chance that UPB can lead to short-term gains or functional outcomes, such as customer-helping behaviors, this does not lend support to UPB perpetrators' illusory belief that UPB is legitimate, as they do not offset its long-term detrimental costs for the organization and its stakeholders (Tang et al., 2021). Our findings provide evidence about how UPB can also pave the way for future ethical transgressions, even within the same perpetrator; thus, underscore the importance of manager recognizing UPB’s long-term costs and suggest implementing training to increase employee awareness and curb UPB.

Our research also highlights the need for organizations to understand how employees' cognitive processing of their own unethical behavior (i.e., UPB) may shape their subsequent work behavior. That is, after the enactment of UPB, employees may justify their behaviors as less problematic or even believe it is legitimate as they were trying to help the organization. With this possibility, organizations should establish clear norms against any stakeholder-harming behaviors, regardless of pro-organizational intent, and promote interventions for moral recovery and prevention of moral disengagement. For example, the organization could set up a strong ethical infrastructure with both formal and informal systems that support and embed ethical expectations and clarify expectations in advance via training programs and mission statements to encourage stakeholder responsibility and empathy (Kaptein, 2008; Roy et al., 2023). However, it is important to recognize that the solution may not be so simple; as our additional exploratory analysis shows, alignment with stakeholder interests may prevent the behavior from occurring in the first place, but it may not prevent the motivated justification of UPB once it has occurred.

Limitations and Future Research Directions

We acknowledge several limitations of our study, highlighting opportunities for future research. In Study 1, the correlation between our measure of UPB acting and UPB perception (r = 0.48) was moderate, which is lower than in previous research (e.g., Wang et al., 2022). This highlights the need for future validation of UPB measurement in experimental settings. In Study 2, first, although our study followed previous best practices by including time lags between variables to alleviate common method bias (Podsakoff et al., 2003), this time separation cannot completely rule out this bias due to the use of single-source data. Future research may address this concern by assessing the behaviors using third-party behavioral reports. While UPB may not be easily assessed by others due to its typically private nature, it might be worthwhile to have others (e.g., supervisors, coworkers, or customers) rate focal employees' OCB-C to obtain a more objective assessment to complement the self-reported measure.

While our two-wave design allowed us to capture immediate cognitive response and establish temporal precedence for behavior outcomes, multi-wave design could provide stronger evidence for causal mechanism and temporal ordering in mediation models (Götz et al., 2021; O’Laughlin et al. 2018; Ployhart & MacKenzie, 2015). To strengthen the mediation mechanisms proposed in our study, future research may consider using an experimental design to test the causal relationship between UPB and subsequent cognitive and behavioral reactions (Pirlott & MacKinnon, 2016). Furthermore, we encourage future research to adopt a longitudinal design with more time points to provide more insights on how unethical behaviors, particularly UPB, might evolve over time.

Our unexpected finding of a negative relationship between perceived loss of moral credits and moral disengagement warrants further investigation. A potential explanation is that when individuals perceive a loss of moral credits, they are less likely to morally disengage, perhaps because they are more aware of the ethical implications of their actions; conversely, when individuals morally disengage, they may be less likely to perceive a loss of moral credits, as they have already rationalized their behavior as acceptable. Future studies could explicitly examine the relationship between perceived loss of moral credits and moral disengagement and how these moral cognitive processes might interact over time. Additionally, the marginally significant residual covariance (p = 0.049) between OCB-C and UPB (time 2) in Study 2 suggests a complex relationship between these constructs that warrants further investigation. This association might stem from the pro-organizational aspect inherently in both constructs. While UPB involves unethical actions intended to benefit the organization, OCB-C represents extra efforts in customer service that can also benefit the organization. Future research could examine if organizational contextual factors can foster a “whatever it takes” mentality that inadvertently encourage both positive extra-role behaviors (OCB-C) and potentially harmful actions (UPB).

The goal of the current research was to focus on evidencing countervailing cognitive processes (i.e., the why) within the individual as a way to understand potentially paradoxical outcomes; however, this does not undercut the critical importance of future research for better understanding the scope (i.e., the who and when) of these mechanisms. For example, future research may also examine That is, employees may be predisposed to a particular pathway (e.g., moral self-regulation or moral disengagement); prior research already suggests that people with high moral attentiveness or with high moral identity tend to reflect more on the moral implications of their behaviors and thus might be more likely to engage in moral self-correction followed by their transgressions (Jiang et al., 2022; Liao et al., 2018).

Conclusion

To date, UPB research has primarily focused on why people engage in UPB. We depart from this focus by adopting an offender-centric perspective of UPB in the service context to investigate how service employees think and behave after engaging in UPB. Our findings revealed that consistent with the paradoxical feature of UPB, the engagement of UPB can elicit complex cognitive experiences in the forms of perceived loss of moral credits and moral disengagement, which in turn lead to customer-helping behaviors or a downward spiral of UPB. By elaborating a dual-pathway model of the aftermath of UPB, we hope our work can encourage future research to extend scholarly understanding of this paradoxical organizational behavior.