Study design

We used waves 10, 11, and 12 of the UKHLS. Data collection for each wave spans a two-year period and for our analyses covered 2018–2022. Many indicators are collected annually, via face-to face interview, including indices relating to mental health and labor market status, allowing for assessments of changes in mental health over time. However, the violence indicators have only been asked once of the entire sample, in wave 11, and it is for this reason that we limit our analyses to wave 11 and to those waves adjacent to it. It should also be noted that while a portion of data collection for wave 11 coincided with the COVID-19 pandemic, data collection did not stop as a result of it with interviews conducted online or by telephone from March 2020 to April 2022 to ensure continuity in the panel design, with 32.5% of the sample collected after the March 2020 UK national lockdown. We therefore conducted tests of a possible COVID-19 effect by de-selecting respondents who completed interviews during lockdown, which led to the same conclusions.

Measures of violence and fear of violence

We generated an amalgamate category of workplace violence, which included those who have been ‘physically attacked’ and/or ‘insulted, threatened or shouted at’ in the 12 months prior. The precise wording of the questionnaire asked: “In the last 12 months, have you been insulted, called names, threatened or shouted at, in any of these places? If so, which ones?”, and “In the last 12 months, have you been physically attacked in any of these places? If so, which ones?”. We combined both questions given low prevalence of being attacked. We determined fear of violence from positive responses to a question that asked: “In the last 12 months, have you felt unsafe in any of these places?”. For each of the above three questions, respondents were provided with a list of 11 locations and could select multiple, and we selected those who reported the workplace as the location of their exposure to violence or fear of violence. Our measure of fear of violence was unlike more widely used measures that measure fear of crime as feeling unsafe in one’s neighborhood at night-time (25). So, for our purposes, the data were better suited for an analysis of fear of violence at work. We could not, however, examine the perpetrators of violence using these data.

Measures of mental health

Associations between violence and CMD were examined using an aggregate dichotomous threshold variable based on the general health questionnaire (GHQ-12) (26). Construct validity of GHQ-12 against psychiatric diagnosis based on clinical interview of anxiety and depression have established strong overlap (27). The battery collects indicators on current mental health, with respondents asked to reflect on how they have felt over the past few weeks. The threshold variable we applied is widely used in the literature (28) and identifies those with clinically significant mental disorders. The variable scores 1 to those reporting positively to poor mental health indicators: for those with responses “rather more than usual” and “much more than usual”, while those who reported “not at all” and “no more than usual”, scored 0. The sum of the index ranges from 0–12, and the threshold variable groups those with an index of ≥4 as suffering from a CMD (26).

Measures of industrial sector

Industrial sector and occupational status, which we introduced as an important confounder of violence risk, were generated post-interview using respondents’ job description. We used the two-digit Standard Industrial Classification (SIC) 2007 to examine industrial sector and applied the simplified 10 categorical classification. Due to sample size, we excluded the armed forces and combined skilled agricultural occupations with craft and related trades resulting in 8 occupational categories (see table 1).

Confounders

Not all workplaces, or jobs held within them, are equal, and prior research has found certain features of the workplace to be confounders of workplace violence and of mental health (2) (21) (29) (9). We therefore introduced the following variables to improve estimate precision of our key covariates. Occupational status is an important predictor of violence, with those from lower status groups often found to be more exposed to violence risk (30). We therefore adjusted for occupational status and used the International Classification of Occupations 1988 (ISCO 88). We distinguished between those on permanent versus fixed-term or temporary contracts as job insecurity is known to be predictive of both poor mental health (31) and enhanced violence risk (32). We adjusted for working-time, distinguishing between full- and part-time workers, and adopted a cut-off of ≥30 hours per week. We adjusted for job autonomy and standard versus nonstandard working time, which were asked in wave 10 and fed forward. The job autonomy scale was constructed using responses relating to control over task performance, pace, manner, and order, as well as work hours, with answer categories ranging from (0) none to (3) a lot (Cronbach’s alpha=0.855). Non-standard working was defined as working most weekends, usually working nights, evenings, rotating shifts or with varying patterns. We further distinguished variance in risk by age (continuous), sex (men or women), and ethnicity (distinguishing between White (British), Asian (British), Black (British), and mixed/multiple or other ethnic backgrounds).

Statistical method

We examined two longitudinal samples, the first (N=11 366) examined the prevalence of workplace violence across industrial sectors as well as the association between workplace violence on CMD risk adjusting for CMD at t-1, with CMD one year prior correlated with current CMD risk. The second sample (N=9908) examined whether the effects of workplace violence are persistent, presenting tests of the association between workplace violence and CMD at t+1. All analyses select on employees as key controls were not asked of the self-employed, and we further excluded those with missing values.

Table 1 presents unweighted descriptive statistics for both samples. Variance in the probabilities of workplace violence and fear of violence by industrial sector are presented in table 2 for sample 1, with adjusted and unadjusted odds ratios (OR) derived from weighted bivariate and logistic regression analyses. Figure 1 provides a graphic representation of violence prevalence by sector using average marginal effects based on the models presented in table 2. Pair-wise comparisons of these average marginal effects which show statistically significant differences in workplace violence by sector are presented in supplementary material, www.sjweh.fi/article/4230, table S1.

Table 3 applies lagged-dependent variable regression models (33), which include estimates of our dependent variable at t-1. These models allow us to control for CMD one year prior to approximate a causal measure of the effects of violence on well-being, with prior mental health conditions likely to be predictive of current mental health. Table 4 examines whether the effects of workplace violence persist over time through an assessment of workplace violence at t on CMD risk at t+1. We provide sensitivity tests to determine whether a linear specification of CMD GHQ-12 led to similar conclusions in appendix table 6. All analyses are weighted to account for the complex sample design, unequal selection probabilities, and non-response/attrition as is advised for these data (34).

Figure 1

Prevalence of workplace violence and fear of violence at work by industrial sector. Note: Weighted estimates derived from average marginal effects based on adjusted models shown in table 2 (N=11 366), which adjust for occupational status, contract-type, work hours, working time, work autonomy, sex, age, and ethnicity. Predicted probabilities and pair-wise comparisons are presented in supplementary table S1.

Lived experience engagement

Our statistical analyses are supplemented with the views of those with lived experience of violence and abuse in accordance with the aims of our research program. Participants’ views were obtained from an online panel discussions of our findings. Participants included men and women at different life stages and from a diverse range of ethnic backgrounds, and a more detailed account of the process is provided in the supplementary material. This study obtained ethics approval at City, University of London (ETH21220–299).

Effects and associations of workplace violence on common mental disorders

Prevalence of CMD in our sample was 20.3% at time t. Table 3 shows that of those who experienced violence in the workplace in the past 12 months, 34.1% reported CMD, and this was 39.4% among those who feared violence at work. The unadjusted and adjusted associations between violence and fear of violence are presented in table 3. Even when adjusting for CMD at t-1 and other confounders, we confirm that CMD were more common among people who had experienced violence at work compared to those who had not (OR_adj 1.400, 95% CI 1.182–1.658), as well as those who feared violence at work (OR_adj 2.103, 95% CI 1.779–2.487).

We also found the effect of violence to be persistent over time when we investigate CMD one year later at t+1 in table 4 (OR_adj 1.461, 95% CI 1.226–1.742 for violence at work; OR_adj 1.731, 95% CI 1.450–2.068 for fear of violence at work). Moreover, in these models, there was little evidence of strong difference in CMD risk by industrial sector or occupational status, suggesting a shared tendency by work type. Finally, we conducted sensitivity analyses of the functional form of our dependent variable, and found that a linear specification of CMD GHQ-12 led to the same conclusions (supplementary table S2).

Lived experience engagement

Our statistical analyses are supplemented with the views of those with lived experience of violent victimization in support of trauma-informed practice, with additional details of our research protocol provided in the supplementary material. There was a unanimous sense from those with lived experience that our statistics on prevalence were likely to be the ‘tip of the iceberg’, given disclosure issues. It was noted how hard it was for someone who may be traumatized from their experience of violence at work to also be responsible for raising the issue with management, with the process of describing the incident frequently traumatizing in and of itself for the victim. People with lived experience reported that their managers were often annoyed or upset when presented with their accounts of what they experienced at work, and indeed managers were accused of minimizing or dismissing incidents reported, especially if they were between co-workers. Victim-survivors noted the significant mental health impacts of violence exposure and some noted that they feared being declared unfit for work should they disclose these to managers. Others stated that it was easier to report physical or ‘routine’ violent interactions, but that insidious bullying behaviors were much harder to report and prove. They stated a need for improved policies on how to deal with abusive customers or co-workers, yet others noted that in some sectors employees were told to expect violent incidents at work.

Concluding remarks

We found exposure to workplace violence and fear of violence to increase the risk of CMD at baseline and one year after violence was reported. We also found high prevalence of workplace violence and fear of violence in many industrial sectors examined. We need better recognition of the extent to which workplace violence is experienced across multiple sectors and call for better systems wide interventions to mitigate the associated harms.