Where Lines Blur: Exploring the Interconnections Between Workplace Sexual Harassment,Bullying,and Aggression Through Latent Class Analysis

Workplace Sexual Harassment, Bullying, and Aggression: A Comparative Perspective

Distinguishing workplace sexual harassment, bullying, and aggression is well justified in the literature, as each of the concepts captures unique manifestations of negative social behavior, reflecting meaningful differential experiences (Tepper & Henle, 2011). Workplace sexual harassment refers to unwelcome sex(gender)-related behaviors in the workplace that recipients perceive as offensive, overwhelming their coping resources, or threatening their well-being (Fitzgerald et al., 1997). Workplace bullying involves repeated and prolonged exposure to acts of harassment, insults, exclusion, or undermining, in a context where targets end up in an inferior position and have difficulties defending themselves (Einarsen et al., 2003; Nielsen & Einarsen, 2018). Finally, workplace aggression is often explicitly characterized as deliberate attempts to harm coworkers (Neuman & Baron, 1998).

Typically, the behavioral expressions of these three phenomena are argued to differ along dimensions of content, intent, power dynamics, and patterns (frequency, duration, escalation). Sexual harassment involves negative social behaviors of an explicit sexual or gender-related nature, whereas the bullying and aggression literature do not restrict the content of hostile acts. This also implies that discriminatory behaviors shaped by sociocultural characteristics – such as gender – are not necessarily excluded from aggression or bullying either (Cowie et al., 2002). Regarding intent, both sexual harassment and bullying do not require that perpetrators act with the explicit intent to cause harm, whereas intent is a key defining feature of aggression (Shewach & Sackett, 2024). Nevertheless, scholars note that perceived malintent may also become central in bullying, as targets frequently infer hostile intent due to the repeated nature of bullying behaviors (see below; Baillien et al., 2017). Of the three constructs, only bullying explicitly specifies and elaborates on the presence of a power imbalance between perpetrator and target as a defining criterion. In such situations, targets are perceived to end up in an inferior position compared to the perpetrator and therefore struggle to defend themselves. This imbalance can stem from various factors, including not only formal hierarchical status, informal social influence, or tenure, but also gender (Hershcovis, 2011), leaving room for overlap with sexual harassment. Although not a formal defining criterion in sexual harassment, its literature places significantly more emphasis on power relations in explaining the origins of the phenomenon than is typical for either bullying or aggression. Sexual harassment is often understood within a broader ‘gender hierarchy’ shaped by patriarchal societal structures and gender role socialization that carries over into workplace settings, reflected in the fact that the majority of reported cases involve women targeted by men (Cortina & Areguin, 2021). In this hierarchy, the masculine is associated with greater power than the feminine, and sexual harassment is viewed as a means, whether conscious or unconscious, of gaining or preserving that power (Burn, 2019; O’Donohue, 1998). The aggression literature, by contrast, gives relatively less explicit attention to power dynamics as a constituting element. Nevertheless, hierarchical power differences often create conditions under which aggression, particularly verbal aggression, can more easily be legitimized. Because supervisors hold authority over subordinates, the boundary between a firm management style and aggressive conduct can be thin, a dynamic that is central to the abusive supervision literature (Tepper et al., 2017). Finally, bullying distinguishes itself through its systematic nature. Its defining characteristic is persistence: involving repeated negative acts occurring over an extended period (Notelaers & Van der Heijden, 2019). A single incident cannot constitute bullying (Einarsen et al., 2003), whereas a single incident may suffice to classify sexual harassment and aggression, even though both often also occur repeatedly (Blindow et al., 2024; Rospenda et al., 2006). Bullying is therefore uniquely understood as a dynamic – an unfolding process often marked by escalation over time – rather than an isolated event (Notelaers & Van der Heijden, 2019).

Taken together, the discussion above suggests that although experiences of workplace sexual harassment, bullying, and aggression vary along several dimensions, their defining characteristics may be less differentiating as they initially appear (Aquino & Thau, 2009). This complicates drawing sharp boundaries and makes their communalities difficult to overlook. Historically, early scholars approached negative workplace behaviors in a far less segmented way than modern conceptualization. Even Leymann (1990a) composed his foundational Leymann (1990b) Inventory of Psychological Terror (LIPT) to consider a broad and comprehensive range of hostile acts – such as exclusion, verbal hostility, physical aggression, and unwanted sexual behaviors – examining them as systematic victimization altogether as opposed to differentiating subforms. Certain scholars even argue that defining features such as sexualized content, behavioral repetition, or deliberate harm could be rather trivial or peripheral in the lived experience of negative social behavior, implying that these concepts might just be more of the same (Aquino & Thau, 2009). Indeed, at their core, all three forms ultimately entail negative, unwanted, unsolicited, social interactions, often manifested through similar behavioral expressions, including psychologically and physically abusing acts as well as material and sociocultural forms of maltreatment, many of which recur over time (Aquino & Thau, 2009; Verschuren et al. 2021). Experiencing acts such as insults, negative comments, shouting, physical contact/attacks, vandalism, or sexualized remarks may, depending on their connotation or social context, constitute sexual harassment as well as bullying or instances of aggression. Additionally, each experience inflicts similar types of harm across physical, material, psychological, and social domains and involves comparable role dynamics among targets, perpetrators, and bystanders (Verschuren et al., 2021).

Given this growing acknowledgment of the conceptual similarities (and differences) between sexual harassment, bullying, and aggression at work, a substantial number of researchers argue that these phenomena are highly interrelated (Aquino & Thau, 2009; Bowling & Beehr, 2006; Dhanani & Bogart, 2025; Hershcovis, 2011). So far, scholars have highlighted different ways in which they may conceptually connect to one another. For instance, some scholars have suggested a nested structure in which one experience constitutes a specific manifestation of another, as depicted in Figure 1. Namely, workplace sexual harassment as well as bullying have been regarded as forms of a broader category of aggression (or, violence) at work (Hershcovis, 2011; Van De Griend & Messias, 2014). Sexual harassment is regarded as a sexually oriented type of aggression (Piquero et al., 2013; Schat & Kelloway, 2005) and bullying as a subset of aggression manifested in multiple incidents (Gururaj & Schat, 2025; Samnani et al., 2013). Since physical attacks can be a prominent component of aggression (Bartlett & Bartlett, 2011; Rayner & Hoel, 1997), behaviors such as sexual assault or physical manifestations of bullying are easily classified under aggression due to this shared physical feature (Bartlett & Bartlett, 2011; Piquero et al., 2013). Alternatively, when physical acts are excluded from definitions of bullying, some conceptualize bullying it as the ‘psychological’ segment of aggression (Schat et al., 2006). Conversely, albeit to a lesser extent, others have argued the reverse; that physical violence and verbal aggression could equally be components of a broader bullying conceptualization (Einarsen, 1999). Additionally, several researchers have followed the possibility of sexual harassment being a form of bullying that uses sexuality or gender as a means of oppression (Cowie et al., 2002; Einarsen, 2000).

OPEN IN VIEWER

Figure 1.

Visualization of suggested nested relationships between workplace sexual harassment, bullying, and aggression.

Another perspective views the behaviors as interconnected elements of the same overarching phenomenon, as represented in Figure 2. Recent literature has aimed to formulate a more comprehensive concept of negative social behaviors at work, often exemplified through operational definitions encompassing a wide range of experiences, including sexual harassment, bullying, and aggression (Verschuren et al., 2021). In this, all three concepts are seen as interrelated because they are similar components of one overarching superstructure or macro-construct. Accordingly, researchers have introduced umbrella-concepts like workplace harassment (Bowling & Beehr, 2006; Neall & Tuckey, 2014), workplace victimization (Aquino & Thau, 2009), workplace (interpersonal) mistreatment (Dhanani & LaPalme, 2019), and workplace emotional or generalized abuse (Keashly & Harvey, 2005) aiming to embody these experiences collectively. Yet, given the diversity of labels introduced, this has furthered conceptual fragmentation in a similar ‘semantic jungle’, transforming the initial promise for clarity of such superstructures into an additional source of confusion.

OPEN IN VIEWER

Figure 2.

Visualization of suggested overarching structures for workplace sexual harassment, bullying, and aggression.

While these perspectives offer valuable starting points for understanding the relationships between workplace sexual harassment, bullying, and aggression, they remain largely theoretical. To date, discussions have been grounded primarily in conceptual debate, often relying on literature reviews and definitional critiques (Shapiro et al., 2008). Empirical evidence on whether, and to what extent, employees perceive these experiences as connected is still lacking. Without empirical validation, our knowledge of the actual nature of these potential relationships is tentative at best. Despite the absence of joint empirical investigations into sexual harassment, bullying, and aggression, a handful of studies have explored how other, related forms of negative interpersonal conduct at work might interrelate. Unfortunately, the methods and methodological rigor employed vary widely and the results have produced disparate and inconclusive findings. For instance, researchers examining the psychometric properties of large and comprehensive instruments such as the LIPT (Leymann, 1990b) more in-depth, which initially include items that may reflect all three concepts, have identified multiple distinct underlying factors. For example, using principal component analysis, the French validation of the LIPT revealed separate components for, among others, verbal violence and threats, physical violence, and sexual harassment (Niedhammer et al., 2006). Similarly, the Turkish validation study employed confirmatory factor analysis and distinguished sexual and physical harassment from bullying behaviors (Korukcu et al., 2014). When comparing more narrowly defined scales, Larsen et al. (2019) conducted bifactor analyses and propose that sexual harassment, aggression at work, and gender discrimination share a considerable degree of overlap, to the extent that suggests a single underlying general factor, although individual constructs continued to account for a portion of the variance. Conversely, Lim and Cortina (2005) used confirmatory factor analyses to highlight that general incivility, sexualized harassment, and gender harassment do not fall under one overall negative behavioral experience but are better represented by three standalone factors. Yet, they did find strong correlations and notable co-occurrences among the three concepts. Similarly, through confirmatory factor analysis, Fendrich et al. (2002) observed that more generalized workplace abuse and sexual harassment ‘are two distinct but strongly related constructs’ (Fendrich et al., 2002, p. 501). Using LCA, Notelaers et al. (2018) found that the measurement items for bullying, conflict, and aggression were best represented by two distinct latent factors – one corresponding to bullying, and another combining conflict and aggression. In conclusion, our understanding of how the different negative social experiences might be entwined remains elusive.

Therefore, we aim to address this research gap by conducting a comprehensive empirical examination of workplace sexual harassment, bullying, aggression, and their potential interconnections. Based on the recognized communalities in the existing literature, we consider that employees may perceive sexual harassment, bullying, and aggression as more intertwined than traditionally assumed in research. This could imply they are seen as facets of a broader construct, or they remain distinct experiences but still tend to co-occur or relate in other meaningful ways. Accordingly, we aim to explore and untangle if and how sexual harassment, bullying, and aggression are connected in employees’ lived experiences. To this end, we employ LCA, an exploratory technique proven valuable in bullying research (Notelaers et al., 2018).

Sample and Procedure

For this study, we use data collected during June to July 2017 among employees of a university in the Low Lands. The data were collected by an independent work environment assessment company through submitting electronic questionnaires to 4,836 employees, of which the response rate was 26%. Participation in the survey was voluntary, and anonymity was ensured by all parties. For the present analyses, a subsample of 871 employees was selected, comprising only respondents who had completed all relevant behavioral items on sexual harassment, bullying, and aggression. In this sample, most employees, 61%, identified as female. The average respondent was 43 years old (SD = 12.06) and 57.4% of respondents worked full-time. Among the part-time workers, 20% worked more than 80%, 16.7% of employees worked between 50% and 80% of the time, and 6.3% occupied positions less than half-time. The sample included a nearly equal distribution of employees in administrative (51%) and academic (49%) positions.

Measures

Workplace Sexual Harassment was measured with five items of the 11-item Bergen Sexual Harassment Scale (Nielsen et al., 2010). The scale encompasses common verbal, nonverbal, and physical harassing behaviors such as ‘unwanted verbal comments with sexual content, such as jokes or sexist remarks’, ‘sexually charged staring or glances which felt uncomfortable’, or ‘unwanted physical contact with sexual undertones, like patting, pinching, touching or embracing’. Respondents indicated whether they experienced these behaviors (1) zero, (2) one, (3) two to five, or (4) more than five times during the last 6 months.

Workplace bullying was measured using the nine-item Short Negative Acts Questionnaire (SNAQ; Notelaers et al., 2019). Example items include both person- and work-oriented behaviors such as experiencing ‘gossip or rumors about you’ or ‘someone withholding information which affects your performance’. Respondents were asked to evaluate how often (from [1] never to [5] daily) they experienced these bullying behaviors during the last 6 months.

Workplace aggression was measured using one item from the original Negative Acts Questionnaire (Einarsen & Raknes, 1997) and two from the Negative Acts Questionnaire-Revised (Einarsen et al., 2009). Respectively, these items asked respondents whether they experienced ‘verbal violence’, ‘being shouted at or being the target of spontaneous anger outbursts’, and, more physically, ‘intimidating behaviors, such as finger-pointing, invasion of personal space, shoving, blocking your way’ during the last 6 months. The five response categories were used that ranged from (1) 0 times to (5) weekly or more often.

Statistical Considerations

Using Latent Gold 6.1, we conduct LCA because this statistical approach can address some of the methodological challenges faced by prior research on negative social behaviors, particularly those related to response scale types and distributional assumptions (Notelaers et al., 2018). Ordinarily, these measures all employ response categories reflecting exposure frequency (from ‘never’ to ‘always’). However, strictly speaking, the observed responses to such behavioral items are ordinal in nature rather than continuous, despite commonly treating them as interval-level data in traditional factor analysis or (linear) regression techniques (Hershcovis & Reich, 2013). Even when categorical accommodations are made by adopting a weighted least squares estimation method, thresholds are estimated for a pseudo-continuous latent variable for which normality still is assumed (Jöreskog, 2003). While moderate deviations from the normality assumption can be tolerated, severe departures can bias results and threaten statistical validity (Li, 2016; Magidson & Vermunt, 2001). Consequently, conventional techniques such as correlational or covariance modeling pose notable challenges when applied to negative social behaviors at work. LCA offers several advantages that address the aforementioned challenges. Importantly, it does not rely heavily on distributional assumptions and therefore accommodates non-normal data. Moreover, LCA can incorporate count, continuous, interval, nominal and ordinal measures variables, making it a versatile technique for capturing heterogenous response patterns. This flexibility gives LCA an advantage over traditional clustering methods, as it can more effectively model categorical response patterns (Magidson & Vermunt, 2004).

Ultimately, LCA is a statistical method that allows to uncover hidden structures within datasets. It classifies respondents into mutually exclusive classes or ‘profiles’ that represent the categories of an unobservable variable (McCutcheon, 1987; Notelaers et al., 2006). LCA begins with a baseline, single-class model. Sequentially, classes are added until an optimal model fit is achieved (Magidson & Vermunt, 2004; McCutcheon, 1987). Not only the number of classes can be increased, but also the number of underlying latent variables, factors, can be expanded as well. Given its conceptual resemblance to traditional factor analysis, a model with multiple latent variables is called a ‘Latent Class Factor analysis’ (LCF; Magidson & Vermunt, 2001).

Determining the fit of LC models requires an extensive and nuanced evaluation process. Model comparison primarily relies on the Bayesian Information Criterion (BIC), with the model showing the lowest BIC value considered the most suitable (McCutcheon, 1987, Vermunt & Magidson, 2002). In addition, several other statistics inform us about the quality of the retained model. First, there is the (mis)fit of the model itself. When a model in its own does not fit the data, it has a significant L². However, assessing significance in cases of heavily skewed data (and resulting sparse multiway tables) is not as straightforward. In such cases, Langeheine et al. (1996) proposed using a bootstrapping procedure. Beyond global fit measures such as the BIC and the L², local misfit and its origin can be inspected through the Bivariate Residuals (BVRs), which should ideally be below 3.84 (Statistical Innovations, 2013). However, given BVR’s sensitivity to large sample sizes, a more relative and pragmatic threshold suggests at least 85% reduction in BVR as criterion (Notelaers et al., 2006). This may also avoid adding spurious clusters arising from residual associations among only a few indicators (Uebersax, 2009). Finally, a model’s classification quality is evaluated using R², entropy R², and the adjacent classification error. This multifaceted assessment ensures a comprehensive evaluation of the model’s fit, significance, and classification accuracy.

Limitations and Future Research

This study entails several limitations. First, consistent with the dominant perspective in the literature, we focused exclusively on targets’ experiences of sexual harassment, bullying, and aggression. Accordingly, our study relied on questionnaire-based behavioral measures that capture the subjective evaluations of acts by targets themselves. Yet, incorporating the perspectives of other actors involved, such as perpetrators or bystanders, introduces different challenges. Prior research shows that these actors may interpret and experience negative workplace behaviors differently from targets (Nielsen & Einarsen, 2013; Samnani, 2013). Nevertheless, future studies could compare the co-occurrence and convergence of the behaviors across these different perspectives, validate exposure profiles accordingly, and/or complement self-reports with registered complaints or case reports. Moreover, we strictly focused on three popularly studied (yet more general) constructs of negative social behavior at work, aiming to examine the relationships between experiences of sexual harassment, bullying, and aggression. However, each of these literatures is itself marked by internal distinctions and ongoing debates on subtypes and finer-grained conceptual nuances. To maintain conceptual clarity and avoid unnecessary complexity, we did not further unpack variations within each construct. Instead, we focused on the interrelationships between the three broader constructs, that is, how the three general forms relate to each other, rather than exhaustively addressing all potential subconcepts. Consequently, important conceptual questions – such as, for example in the bullying literature, the differentiation between workplace bullying, mobbing, incivility, ostracism, psychological harassment and more – were not addressed in the present analysis. However, our study offers an important first demonstration of how questions of conceptual overlap can be rigorously empirically unraveled in case of highly non-normal behavioral data. In doing so, it may inspire future research to extend this approach in addressing more fine-grained conceptual distinctions within each domain, such as those in the bullying literature. At the same time, examining overlap among even more closely related constructs as bullying, incivility, or mobbing may prove challenging, given the resemblances and similarities among their most commonly used measurements (Nixon et al., 2021).

In addition, the sample originated from a single organization within the higher education sector. Accordingly, the findings could be highly dependent on the organization’s specific context, workforce composition, and its established social climate, and should therefore not be generalized. Nevertheless, given the study acts as a first, more thoroughly empirical demonstration of examining conceptual overlap, extending this methodological approach to more diverse samples will be important for assessing the generalizability of these findings across organizational contexts that differ in industry, job characteristics, workforce demographics, and cultural backgrounds. Cultural differences, in particular, may determine how these concepts are understood and interpreted, which in turn may influence the extent to which they appear to converge or diverge. For instance, research by Salin et al. (2019) demonstrates that HR professionals across cultures differ in what they consider acceptable conduct and what they regard as workplace bullying. Similarly, employees’ own perceptions of the acceptability of behaviors associated with sexual harassment, bullying, or aggression may influence the extent to which these experiences are perceived as overlapping or distinct. This is an important avenue for future research as we hope empirical work on these conceptual questions continues to develop. Another sample aspect concerns the size of the most extreme exposure profiles, despite the inclusion of almost 900 respondents. The ‘severe target’ groups of all constructs were small, ranging from 1.85% for sexual harassment to 1.18% for bullying and 0.42% for aggression, reflecting a generally low prevalence of severe social maltreatment in the sampled organization. Such low prevalence rates for severe victim classes are consistent with prior LCA findings in this domain (Nielsen et al., 2009). However, the (non)convergence patterns observed in the biplots at the highest exposure levels should be interpreted with considerable caution, as they are based on a limited number of respondents. While the results seem to point toward reconvergence, the small class sizes make it impossible to draw definitive conclusions on how the three concepts behave in their most extreme levels. In consequence, future replication studies are needed to re-examine the re-convergence of workplace sexual harassment, bullying, and aggression at its most severe levels. This means larger and more diverse samples are required to be able to pass a conclusive judgment on the severe exposure levels of these three experiences.

Next, our findings reveal escalating profiles of intensity across all three experiences and suggest that such escalation processes may help explain why the experiences reconverge. However, the cross-sectional nature of our data prevents us from directly testing or modeling these escalation patterns over time. Therefore, future research could examine how these negative experiences actually unfold and whether they follow identifiable escalatory trajectories. For example, beyond merely co-occurring behavior (e.g., verbal abuse and gossip experienced simultaneously), the phenomena might also follow a sequential pattern in which one type triggers another (Verschuren et al., 2021). Established accounts (Einarsen, 2000; Kaukiainen et al., 2001) argue how singular aggressive acts precede more intense victimization, afterwards mutating into bullying or harassment. To deepen our understanding of how these experiences evolve, and when they diverge or converge, future studies could adopt longitudinal designs that capture these processes over time (Cole et al., 2016). Provided sufficiently large and diverse samples, temporal approaches such as Latent Transition Analysis (Reknes et al., 2021) could model changes across the three behaviors, identify latent classes, and illuminate when and how people move between them.

Furthermore, given that the present study identified distinct exposure profiles for each construct, another useful next step may be to further examine how these profiles relate to health and well-being outcomes. Prior research suggests that higher exposure is associated with more adverse consequences (Leon-Perez et al., 2014; Notelaers & Van der Heijden, 2019; Notelaers et al., 2006), and future studies could extend the current findings by testing whether a similar gradient emerges across the identified classes. Particularly so for sexual harassment and aggression, where evidence into criterion validity remains most limited.

Finally, given the variety of operationalizations of sexual harassment, bullying, and aggression at work, it would be valuable to replicate the current study using alternative measures to examine whether our findings hold across different scales and instruments. Additionally, traditional measures often struggle with construct validity, not fully capturing the exact theoretical definitions or dimensions that are assumed to distinguish these behaviors (e.g., intent; Shewach & Sackett, 2024). Examining conceptual overlap using updated measures that address these limitations may therefore provide the most meaningful insights.