Workplace Harassment and Health: A Long- Term Follow Up

Abstract

Objective:

We examine relationships between workplace harassment (WH) and onset of health conditions over a 23-year period.

Methods:

Participants were surveyed at seven points between 1997-2006 and again in 2020. Regression analyses (n=921) assessed effects of chronic WH exposure on onset or recent health conditions by 2020.

Results:

Growth mixture modeling revealed infrequent and chronic classes of generalized workplace harassment (GWH; 33.39% chronic) and sexual harassment (SH; 32.32% chronic). Prevalence of health conditions ranged from 3.71% for myocardial infarction to 43.06% for hypertension. Analysis via propensity score matching showed chronic WH class membership increased odds of coronary heart disease (GWH, OR=3.42, p<.05), arthritic/rheumatic conditions (SH, OR 1.56, p<.05), and recent migraine (SH, OR=1.68, p<.05).

Conclusion:

WH is associated with CHD, arthritic/rheumatic conditions, and migraine. Worker health can be protected through strengthening and enforcing organizational and social anti-harassment policies and laws.

Introduction

Many adverse health outcomes have been linked to workplace harassment (WH) in previous studies. WH encompasses actions against a person that create a hostile work environment and can include generalized workplace harassment (GWH) or bullying that is not based on legally protected characteristics (e.g., gender and race), as well as sexual harassment (SH) [1]. Depending on how harassment is measured in national studies, past year prevalence has been estimated as 8.1% when using a single item measure that requires respondents to label their experiences as harassment or bullying [2]. When using a prevalence estimate based on multi-item scales that ask about experiences of one or more specific behavioral indicators of SH or GWH, past year prevalence was 47% for SH and 63% for GWH in a nationally representative sample [3]. Prevalence of, or risk for, harassment has been found to be elevated for women and individuals who are younger, multiracial, and unmarried [2; 3].

Notably, extensive evidence exists of associations between various forms of harassment and health or mental health disorders. Many of these studies, and related ones, use longitudinal survey data, cross-sectional studies, and cohorts to conduct their research. Longitudinal cohort and survey studies of US military veterans associate SH with onset of mental and psychotic disorders, such as PTSD, substance use disorders, anxiety, eating disorders, bipolar disorder, and schizophrenia [4; 5]. Markedly, most of the current literature on this subject explores the relationship between WH and bullying and mental health outcomes - with a variety of samples outlining the correlation with anxiety, depression, and suicidal ideation and behaviors [6; 7; 8]. A meta-analysis and systematic review explored the effects of sexual harassment and health outcomes, including all data from PubMed and PsychINFO through March 2021. This study found a correlation between harassment and numerous adverse outcomes, including posttraumatic stress disorder (PTSD), depression, suicidality, substance abuse, chronic pain, and gastrointestinal disorders [9]. Meta analytic research also supports a positive association between GWH and symptoms of psychological distress [10;11].

There are considerably fewer studies that investigate the effect of harassment on physical health outcomes – although the existing literature does suggest a negative impact. The strongest evidence involves risk for cardiovascular disease. In a community sample of US women, among those who had experienced workplace sexual harassment at some point in their lives, there was a significantly increased risk of midlife stage 1 or 2 hypertension [12]. In another study conducted via surveys over a 2-year period in a sample of 5432 Finnish hospital employees, there was an association between workplace bullying and incident cardiovascular disease (OR 2.3, 95% CI 1.2 – 4.6), defined as reported myocardial infarction, angina pectoris, cerebrovascular disease, or hypertension [13]. Although the mechanisms in which these outcomes are not understood, this study suggests that the adverse cardiovascular outcomes may be attributable to a high body mass index (BMI) secondary to poor health maintenance/nutrition that may be worsened by exposure to WH.

Regarding other health conditions that have received research attention, the findings are limited. Samples were taken mostly from cross-sectional studies and retrospective cohorts. There is a positive correlation between workplace bullying and fibromyalgia, a somatic syndrome. In a prospective cohort study of employees from 12 Finnish hospitals, those that experienced workplace bullying were much more likely to report physician-diagnosed fibromyalgia at the two year follow up (OR 4.1, 95% CI 2.0 – 9.6) [14]. Furthermore, in cohort studies sourced from Sweden and Denmark including 45,905 participants aged 40-65, the risk for developing type II diabetes has also been found to be increased in those experiencing WH [15]. In addition, there may be an effect of WH on the development and progression of asthma. In both European and American populations, there are mixed results supporting a relationship between various forms of harassment or disorders (violence, childhood, stress, PTSD) and asthma [16].

While there is a growing body of research on the effect of WH on physical health, the referenced studies and others in the current literature generally focus on a single particular physical health outcome, such as cardiovascular disease. Further, there are few studies that examine these associations in the US over a longer period. While a previous study considered the effects of bullying on risk of cardiovascular disease over a 2-year period, a sufficient breadth of understanding on this subject may require a decades-long timeframe due to the length of time it takes cardiovascular disease to develop [13]. Further, most works on this matter highlight cross-sectional studies on various populations globally, many of which may not be reflective of the US due to differences in sample diversity and culture, and variance in access to healthcare. This study aims to combat these limitations by examining the association between WH exposure and onset of health disorders – cardiovascular disease (CVD), coronary heart disease (CHD), myocardial infarction, cancer, type II diabetes, migraine, rheumatoid arthritis, and asthma – in a large, US-based diverse sample of workers over a 23-year study period.

Methods

Respondents were drawn from a sample of 4832 (2416 men and 2416 women) university employees from four occupational groups (faculty: 642 women, 819 men; graduate student workers/trainees: 664 women, 820 men; clerical/administrative workers: 846 women, 272 men; service/maintenance workers: 264 women, 505 men) at an urban university in the Midwestern United States. The goal of the original project was to enroll equal numbers of men and women within each occupational group. However, because of the smaller size of the population for women faculty, men clerical/administrative workers, and men and women service/maintenance workers, we sampled everyone in those groups. In other groups, random sampling was employed. Men were oversampled from faculty and RA/TA groups, the largest groups, because of historical challenges recruiting and retaining men in research. It was important to us to recruit a sample balanced by gender, in order to better study the phenomenon of sexual harassment among men.

Individuals were first surveyed in 1996-1997 and resurveyed about workplace harassment experiences at 6 additional time points through 2006 (T1-T7) (N_T1=2492, n_T1=1338 women, n_T1=1154 men; 52% response rate at T1). The T8 survey in 2007 did not assess workplace harassment. Past participants were again resurveyed by web or mail between July 2020 – February 2021 (T9), approximately 23 years following T1. Those who had previously requested to be removed from the study (n=62), those who had no available contact information from the prior study (n=3) and known deceased (n=40) were excluded from follow-up, leaving a possible sample for follow-up in 2020 of n=2387. All participants were provided consent information regarding risks and benefits of participation and consented to study participation. At T9, an additional 37 past participants were found to be deceased, 3 were incapacitated (e.g., very ill) and unable to participate, 30 refused, and 374 were not locatable. A total of n=921 valid responses were received at T9 (4 responses were disqualified due to demographics that did not match the original respondent; 39.2% reinterview rate excluding known deceased or 38.6% including known deceased [17]). The study was approved by the University of Illinois at Chicago Institutional Review Board (protocol 2019-0374).

Measures

Workplace harassment

Between 1997-2006 (T1-T7), ten time points of data were collected regarding experiences with workplace sexual harassment and non-sexual, generalized harassment (i.e., harassment not based on any demographic characteristic legally protected in the United States). At T3 and T6, respondents were asked to recall harassment experiences during each of the two prior years due to study funding gaps. Past 12-month sexual harassment (SH) at each measurement point was assessed with a modified version of the 19-item Sexual Experiences Questionnaire, the most-used, validated, and reliable measure of SH in research use [18; 19]. Questions were modified to be applicable to either women or men. Past 12 month generalized workplace harassment (GWH) was measured with the Generalized Workplace Harassment Questionnaire, a 29-item, reliable measure of GWH developed by our research group for the study [20]. Previously, we used growth mixture modeling on the T1-T7 data to identify individuals with different developmental growth trajectories for GWH and SH [21]. For both SH and GWH, two latent classes of harassment were extracted: chronic (a pattern of elevated mean levels of harassment across the study period: between 4.9-6.2 on a possible scale of 0-87 for GWH, and 2.0-3.0 on a possible scale of 0-57 for SH across 10 measurement points) and infrequent (a pattern of low mean levels of harassment across the study period: between 0.0 - 1.5 for GWH, and 0.0 - 0.5 for SH across 10 measurement points). 33.39% of our sample were in the chronic GWH class and 32.32% were in the chronic SH class. The latent classes for SH and GWH were used as predictors in our models (0=infrequent/no harassment, 1=chronic harassment over T1-T7).

Health Conditions

Health conditions were assessed at T9 in 2020 with items from the 2017 National Health Interview Survey [22] regarding lifetime diagnosis of each health condition (yes/no) and age at first diagnosis for the following conditions: hypertension, coronary heart disease, stroke, diabetes, cancer, arthritis, and asthma. We also added a question to assess experience of migraine in the past 3 months prior to the T9 survey (yes/no).

Statistical Analysis

As it was neither possible nor ethical to randomly assign respondents to the chronic harassment classes in the study, propensity score matching (PSM) with nearest neighbor matching was performed to help address possible confounding and selection bias. Building upon prior research in the study sample predicting the likelihood of class membership in WH [21], matched samples were created based on age, race and gender to ensure balance. Analysis was performed with and without a standard caliper (¼ of a standard deviation in the propensity scores) as established in the literature, but as there was a negligible difference in results, the results without the caliper were presented to reduce population bias [23; 24; 25]. After creating the matched samples, logistic regression analyses were conducted to examine the relationship between harassment classes (for SH and GWH separately) and the onset of various health outcomes over the study period (T1-T9), including cancer, arthritis or rheumatic conditions, asthma, diabetes, coronary heart disease (CHD), hypertension, myocardial infarction, any cardiovascular disease (any stroke, CHD, hypertension and myocardial infarction). Stroke was omitted as a direct outcome due to a very small number of observations. Respondents whose diagnosis of the above conditions occurred prior to the T1 survey were excluded from the analyses. In addition, the occurrence of migraine in the past 3 months was measured, but age of diagnosis was not asked in the survey. Although race, age and gender were used to match the samples, covariates included age, gender, Hispanic, Black and Asian and income at T9 to make the analyses more robust. [26]

While the harassment class variables were extracted from the T1-T7 data using growth mixture modeling to reflect different developmental growth trajectories across groups, as opposed to measuring exposure at specific time points, additional sensitivity analysis was conducted to examine the association between those exposed to harassment in the chronic harassment classes at T1 with outcomes occurring after exposure. 75.6% of respondents in the chronic sexual harassment class and 91.43% in the chronic generalized harassment class were exposed to harassment at T1, and US national data on patterns of harassment and international data on duration of harassment indicates that once harassment begins it tends to be chronic and enduring [27; 28]. The sensitivity analysis also used propensity score matching and logistic regression as described above but was limited to those exposed to harassment in the chronic harassment classes at Wave 1 and looking at outcomes occurring after Wave 1.

Results

Descriptive statistics

Table 1 depicts summary statistics and the percentage of respondents with each health condition. The sample was majority female (53.69%) and White (51.75%; Black 21.82%; Asian 16.52%; Hispanic 7.72%; Native American/Other/Multiracial 2.21%) with a mean age of 40.31 (SD =11.64) at T1. The percentage of respondents in the chronic GWH class was 33.39% while 32.32% were in the chronic SH class.

Table 1:

Summary Statistics

Variable	n	M (SD) / %	Range
Gender
Women	1338	53.69%
Men	1154	46.31%
Race/ethnicity
Black	543	21.82%
White	1288	51.75%
Hispanic	192	7.71%
Asian/Pacific Islander	411	16.51%
Native American/Other/Multiracial	55	2.21%
Age T1	2441	40.31 (11.64)	20 - 86
Income T9
Under $25,000	36	4.11%
$25,000 to $49,999	107	12.22%
$50,000 to $74,999	88	10.05%
$75,000 to $99,999	77	8.79%
$100,000 to $149,999	147	16.78%
$150,000 to $199,999	126	14.38%
$200,000 to $249,999	95	10.84%
$250,000 or more	200	22.83%
Chronic Sexual Harassment Class	791	32.32%	0 - 1
Chronic Generalized Harassment Class	816	33.39%	0 - 1
Cancer T9	156	17.03%	0 - 1
Diabetes T9	97	10.61%	0 - 1
Asthma T9	113	12.35%	0 - 1
Arthritis/Rheumatic Disease T9	278	30.48%	0 - 1
Coronary Heart Disease T9	59	6.46%	0 - 1
Hypertension T9	394	43.06%	0 - 1
Myocardial Infarction T9	34	3.71%	0 - 1
Any Cardiovascular Disease T9	311	38.30%	0 - 1
Migraine T9	134	14.66%	0 - 1

Note.

T1 = Wave 1, n=2492; T9=Wave 9, n=921.

Aside from migraine which was assessed for experience in the past 3 months, diagnoses include lifetime diagnosis reported at T9.

Table 2a denotes the associations between harassment type (generalized harassment, sexual harassment) and health outcomes. Results were significant at the p ≤ 0.05 level for generalized harassment and coronary heart disease, as well as sexual harassment and migraine and rheumatoid arthritis. Compared to the infrequent class, being exposed to chronic generalized harassment increased the odds of being diagnosed with coronary heart disease during the study period by more than three times. In addition, the odds of experiencing migraines increased by 67.7% when subjected to chronic SH, compared to those subjected to no or low levels of SH. The odds of being diagnosed with arthritis/rheumatic disorder increased by 56.3% when subjected to chronic SH, compared to those subjected to no or low levels of SH (see Table 2a).

Table 2a:

Associations between Harassment Latent Class and Health Outcomes

	Generalized Harassment			Sexual Harassment
Outcome	OR	CI	p	OR	CI	p
Cancer (n=577, n=625)	1.21	[0.73, 2.00]	0.468	0.76	[0.47, 1.22]	0.249
Diabetes (n=565, n=613)	0.68	[0.36, 1.30]	0.243	0.69	[0.37, 1.28]	0.241
Asthma (n=492, n=585)	0.78	[0.36, 1.70]	0.529	0.95	[0.45, 2.04]	0.900
Arthritis/Rheumatic Disease (n=524, n=589)	1.04	[0.69, 1.57]	0.836	1.56	[1.02, 2.40]	0.041*
Coronary Heart Disease (n=571, n=581)	3.42	[1.09, 10.72]	0.035*	1.81	[0.77, 4.26]	0.173
Hypertension (n=524, n=569)	0.74	[0.51, 1.07]	0.107	0.78	[0.54, 1.11]	0.166
Myocardial Infarction (n=546, n=553)	2.53	[0.76, 8.44]	0.131	1.37	[0.46, 4.06]	0.572
Migraine (n=579, n=629)	1.49	[0.91, 2.44]	0.112	1.68	[1.06, 2.65]	0.027*
Any Cardiovascular Disease (n=510, n=562)	0.84	[0.58, 1.22]	0.360	0.94	[0.66, 1.35]	0.733

^*p<0.05;

^**p<0.01;

^***p<0.001.

Generalized and sexual harassment were modeled as latent classes, with 1=chronic harassment class and 0=infrequent/no harassment class.

OR=Odds Ratio; CI=Confidence Interval

Table 2b denotes the associations between harassment (GWH, SH) at T1 and health outcomes occurring after exposure, for those in the chronic harassment classes. Being exposed to SH at T1 increased the odds of being diagnosed with arthritis/rheumatic disorder by more than two times. Being exposed to GWH at T1 increased the odds of being diagnosed with coronary heart disease by more than four times, although the results were “borderline” significant at p=0.054 and should be interpreted with caution. In addition, being exposed to GWH at T1 was also associated with increased odds of cancer by 81% at p=0.055.

Table 2b:

Associations between Harassment Type and Health Outcomes for those with Exposure to Chronic Harassment at T1

	Generalized Harassment			Sexual Harassment
Outcome	OR	CI	p	OR	CI	p
Cancer (n=446, n=439)	1.81	[0.99, 3.31]	0.055	0.90	[0.51, 1.58]	0.720
Diabetes (n=431, n=436)	1.17	[0.55, 2.50]	0.689	0.97	[0.45, 2.09]	0.944
Asthma (n=358, n=385)	0.73	[0.31, 1.75]	0.485	0.92	[0.31, 2.76]	0.882
Arthritis/Rheumatic Disease (n=402, n=421)	1.02	[0.63, 1.65]	1.942	2.31	[1.34, 3.97]	0.002**
Coronary Heart Disease (n=405, n=384)	4.67	[0.97, 22.42]	0.054	2.60	[0.78, 8.61]	0.119
Hypertension (n=394, n=400)	0.69	[0.46, 1.05]	0.087	0.83	[0.54, 1.27]	0.388
Myocardial Infarction (n=388, n=381)	2.06	[0.55, 7.64]	2.282	1.17	[0.30, 4.51]	0.821
Migraine (n=446, n=444)	1.68	[0.95, 3.00]	0.077	1.39	[0.82, 2.36]	0.224
Any Cardiovascular Disease (n=381, n=398)	0.84	[0.55, 1.30]	0.442	1.07	[0.70, 1.65]	0.752

^*p<0.05;

^**p<0.01;

^***p<0.001.

Generalized and sexual harassment were modeled as latent classes, with 1=chronic harassment class and 0=infrequent/no harassment class.

T1=Wave 1; OR=Odds Ratio; CI=Confidence Interval

Discussion

This is the first study of long-term risk for morbidity associated with exposure to WH in the United States. Given the wealth of previous data characterizing the sample, this study presents a unique opportunity to inform viable strategies to prevent and address WH-related chronic health issues that may contribute to premature mortality, and to inform targeted treatment strategies to ameliorate negative health-related effects of WH. The results indicate that exposure to chronic generalized WH, which due to the multiple experiences of such behavior is consistent with the definition of workplace bullying, is associated with an increased risk of coronary heart disease. This was in concordance with the sensitivity analysis, which showcased a more than four-fold increase in the risk for coronary heart disease, although this was just above the threshold for statistical significance. In addition, experience of SH is associated with an increased risk of migraine and arthritic/rheumatic conditions via the original propensity score analysis. Interestingly, although the sensitivity analysis is consistent with the results for arthritic/rheumatic conditions with a greater than two-fold increase, it fails to establish a relationship between migraine and WH. Additionally, the sensitivity analysis returned increased odds of malignancy by 81%, albeit falling just above the threshold of statistical significance. Although some of these findings fell slightly out of the range for statistical significance, it does illustrate a degree of consistency in the study results. The findings align with and extend upon the established literature, as WH was previously connected to heart disease in European samples [29; 13]. As stated previously, workplace bullying was linked to a higher likelihood of physician-diagnosed fibromyalgia – indicating potential expectations for a relationship between WH and rheumatoid arthritis, which this study supports [14]. While other researchers have argued that childhood abuse is more consequential to the development of health-related issues in adulthood, specifically migraine [30], our findings suggest that abusive or harassing workplace experiences in adulthood can also contribute to the development of health issues such as migraine.

Existing literature makes various attempts to convey the pathophysiologic explanation of the associations between WH and the various health outcomes. Many studies illustrate an association between psychological distress resulting from childhood trauma (childhood abuse, physical violence, sexual violence) and migraine [31; 32]. However, the mechanism in which migraine could be caused by other stressors is not as clear. Leading research attributes migraine headaches to dilation and inflammation of cephalic vasculature [33]. This is mediated by release of vasoactive neuropeptides (substance P, calcitonin gene-related peptide) which perpetuate vasodilation. The surrounding cranial nerves may subsequently become irritated and magnify the symptoms of pain [34]. A viable explanation could be that stress and psychological trauma activate the sympathetic nervous system, causing corticotropin releasing factor (CRF) to regulate the activity of the hypothalamic-pituitary-adrenal (HPA) axis to trigger modulation of serotonin receptors which in turn contributes to vasoactive peptide release and vasodilation.

The results also note an association between WH and arthritic conditions. This was a particularly interesting finding, given that one rheumatic disorder, rheumatoid arthritis, is an autoimmune condition which primarily affects the joints, but also often has systemic effects. It is presumed that neuroendocrine hormone cascades activated during stress lead to immune dysregulation or altered cytokine production, resulting in autoimmune disease or impaired host defense [35]. In addition, chronic stress may induce an acute phase response, perpetuating a chronic inflammatory process such as atherosclerosis and metabolic diseases. Evidence nominates the liver, endothelium, and fat cell depots as the primary sources of cytokines such as interleukin-6 and C-reactive protein which are strongly associated with inflammatory processes which lead to insulin resistance, the primary vehicle for metabolic disease. Via the same mechanism, the relationship between WH and coronary heart disease can be explained, as atherosclerosis and elevated blood pressures generally cause this condition.

To postulate further, the correlation between stress or WH and psychological disorders (anxiety, depression, PTSD, insomnia, psychotic disorders, substance use, eating disorders) is well established. These disorders could lead to a decreased ability to achieve health-conscious outcomes regarding diet, lifestyle, sleep quality, exercise, and access to healthcare providers for routine screenings and health maintenance. In turn, this may incidentally lead to poor physical health outcomes as mentioned above, along with worsening of disease. It is notable that each of the psychological disorders could interfere with sleep quality and quantity - along with the development of insomnia altogether, because sleep loss is known to be associated with an array of health issues including hypertension, diabetes, obesity, depression, myocardial infarction, stroke, and all-cause mortality [36]. Remarkably, sleep loss also affects the immune system by stimulating the sympathetic nervous system, resulting in increased activation of the aforementioned inflammatory cytokines, HPA axis, catecholamines, cortisol, and vasoactive intermediates. Not only could this disrupt regular immune function and increase likelihood of autoimmune conditions, but it may increase risk of malignancy due to the decreased clearance of endogenous carcinogens and impaired immune ability to eliminate cells that develop oncogenic mutations. Thus, it can be inferred that sleep loss as a result of WH may act as another vehicle in disseminating the pathophysiological conditions found significant in the propensity score analysis.

In established literature, WH has been known to be associated with onset of a number of adverse health outcomes. Interestingly, many of the outcomes evaluated in this study demonstrated no statistically significant relationship; namely between WH and myocardial infarction, type II diabetes, or asthma, when there were expectations otherwise. Sexual violence was associated with CVD risk in the literature, which led us to expect an association between sexual harassment and myocardial infarction via the aforementioned inflammatory and metabolic mechanisms, due to similar medical pathophysiology [9; 37]. Along with this, there was a discrepancy within the analysis results, as the sensitivity analysis failed to establish a relationship between migraine and WH whereas an association was found in the original propensity score analysis. This can be explained by several factors, including a fault in the data set that does not report age information regarding timing of migraine diagnosis. Therefore, we cannot conclude from our data whether the outcome of migraine was a result of the exposure chronologically. Additionally, migraine is the only diagnosis among the health conditions assessed in this analysis exclusively made clinically - meaning the diagnosis is made through the evaluation of patient symptoms and lab values or imaging are not required. This may introduce potential inadequacies in assessing patients for this condition and misdiagnoses with other headache conditions, such as tension headache, caffeine withdrawal, and symptoms of sleep deprivation.

Moving forward, a possible explanation for why our analyses failed to find an association between WH and certain health conditions is also the main limitation of this study: the comparison group. In each analysis, the sexual harassment (SH) and generalized harassment groups are not referenced to a control group that had zero experience of harassment. Rather, each SH and GWH group is referenced to members within the same group who experienced little to no harassment over the study period, but not necessarily none. This could impact the results such that some may be returned without statistical significance due to the lack of a clear, non-harassed control group. Future research should attempt to incorporate a control group that has experienced no WH and develop thresholds for SH/GWH indicating the level or amount of harassment that is most likely to lead to clinically significant health impacts.

We should also note that our chosen approach, PSM, is a widely used approach in medicine, epidemiology, social work, sociology and psychology; it has been used or referenced in over 141,000 articles as of 2019. Propensity score matching has long been a subject of debate, including recent literature by King and Nielson (2019) that PSM may “increase imbalance, inefficiency, model dependency, and bias” and thus should not be used for matching. These authors were concerned that PSM may increase imbalance via random pruning (a process of deleting observations in a dataset independent of treatment and covariates), when covariates were roughly balanced between groups and the worst balanced observations were pruned. They also point out that using an established caliper (¼ of a standard deviation in the propensity scores) can increase model imbalance and dependence and that including all available pre-treatment variables in the propensity score estimation may also generate pruning that is closer to random [24].

On the other hand, more recently scholars have noted that previous studies have shown that even if there was increased imbalance after heavy pruning, it was much less significant than alleged by King and Nielson [38; 39; 40; 25]. Wang (2021) argues that the right question should be when and how to use PSM rather than to use or not to use it. Wang specifies that PSM may not be a good approach for indirect comparisons between studies or data sources with small to moderate sample sizes. The article also argues that some issues of PSM may be due to inadequate caliper selection and that PSM has some desirable properties to equal percent bias reduction. Guo et al. (2020) also offered responses to the King and Nielson article by raising what they believe to be a misleading title and other incorrect points made in the article. However, they agree that model specification in propensity score analysis is challenging, and that researchers need substantive knowledge of model assumptions and plausible causal structure. They conclude that from prior research, sources of selection bias must be understood [39].

In the present study, careful consideration was given to the selection of variables to avoid suboptimal matching and random pruning. The selection of variables was informed by a prior empirical study where age, race and gender predicted the likelihood of class membership in our university employee sample. This provided an empirical basis to inform sources of selection bias and propensity score estimation, as opposed to including random pre-treatment covariates for matching and thus increasing random pruning and potential imbalance. In addition, the authors performed analysis with and without the standard 1/4 SD caliper as established in the literature with minimal differences in results – the caliper did not change any significance for the outcomes and minimally influenced the coefficients as very few observations were pruned [23]. Hence, the results without the caliper were presented. However, it is possible that unmeasured variables such as diet and exercise may have impacted our results. Future research should consider and control the possible impact of such factors. There are also different philosophies to standard error estimation for matched samples and the critique that PSM may affect variance and effect estimation. [41; 42] However, scholars have argued that it is still a more robust approach to perform PSM and regression analyses after matching of the samples, as opposed to conducting simple covariate adjustment in the unmatched sample only and therefore not correcting for confounding biases [39; 43]. According to Guo et al. (2020), OLS is a less than ideal analytic approach in the presence of selection bias based on bias-reduction criterion for both selection on observables and on unobservables in Monte Carlo stimulations [39].

Despite the limitations of this study, it is the only U.S. study to date to investigate the long-term health impacts of exposure to workplace harassment and indicates that workplace harassment represents a significant contributor to the development of disease over time. Thus, prevention of workplace harassment is very important for protecting the long-term health of workers. Employers and occupational safety and health professionals should renew efforts to prevent WH and enforce current SH law. These findings also support efforts in the U.S. to pass legislation such as the Healthy Workplace Bill that would help protect the health of workers by prohibiting workplace bullying. Additionally, this study suggests the need for surveillance efforts to assess and monitor levels of workplace harassment in the population and in specific workplaces. Regular surveys in the workplace, for example, would be useful to identify job types or work groups where harassment is occurring, allowing for more targeted intervention and prevention efforts rather than typical boilerplate harassment-prevention training distributed to all workers, which is not designed for specific workplaces, job types, or work groups. Our results also indicate the need for large scale representative longitudinal studies on the impacts of psychosocial workplace exposures, particularly those such as workplace harassment which can be traumatic in nature, on long-term morbidity and mortality.

SMART Learning Outcomes.

After completing this enduring educational activity, the learner will be better able to:

• Describe the variety of health-related outcomes that have been associated with exposure to workplace harassment.

• Describe the significant associations between exposure to a pattern of chronic workplace harassment and onset of specific health problems in a large sample of U.S. workers over a 23-year period.

• Discuss implications for workplace surveillance and training programs.