Organizations may use the responses of managers or workers to assess safety climate or practices [1-12]. They can use such surveys to identify priority areas for health and safety improvement and then change relevant policies and procedures to reduce injury rates [3, 13, 14]. The extent to which surveys for this purpose can be useful depends in part on their ability to predict injury rates.

Employers and researchers may prefer to survey managers because surveying managers may be easier and less expensive than surveying a large worker group and may result in higher response rates. This may be particularly true in low-wage workforces in which workers do not regularly access work email or if literacy levels are low. However, if worker perceptions better reflect injury risk (that is, are more predictive of future injury rates), the value of surveying workers may outweigh the cost.

Studies have shown that there may be consistent differences between mean manager and worker perceptions of safety climate, policies, or practices in the same work setting, but there have been few studies addressing whether relying on both groups provides more useful information for research, practice, and policy than just relying on one or the other [5, 7-9].

We relied on a conceptual model of managers’ and workers’ responses to survey questions about workplace policies and practices, similar to a conceptual model applied to team organizational climate described by Perrigino et al. [15] and to the framework described in Marín et al. [7, 16]. The underlying theory is that workers and their managers may either observe different aspects of the same phenomena or may observe the same phenomena but interpret them differently (see Fig. 1).

We illustrate this framework with a hypothetical example, using the following survey item: “Managers make sure that workers are able to take their entitled breaks during work.” In a hypothetical workplace, for each manager’s unit, workers are entitled to take a certain number of breaks daily. This is the workplace policy. However, workers and managers may see different aspects of how this policy is enacted. Managers may be more aware of organizational policies and be responsible for their implementation, while workers may be more aware of specific aspects of their jobs and practices on the unit. In the case of work breaks, managers may respond positively to requests for coverage so workers can take their breaks [17]. However, workers may not request coverage if there are high workloads, if they know that it will disrupt the workday of other team members, or if they fear retribution for taking breaks. If workers cannot take all their entitled breaks, managers still might assume that they do. However, workers may think that they are not receiving all their allotted breaks.

Workers and their managers may also have different beliefs about how many breaks workers are entitled to. In addition, workers may recall the breaks they missed more vividly than managers, and managers may more vividly recall their efforts to provide time for breaks. Although the reality – the absolute number of breaks taken – is shared, the perception of that reality may not be the same.

Managers’ perceptions may also be influenced by “social desirability bias,” as noted by Huang et al. [5] and Marín et al. [7]. Managers may consider themselves responsible for safety conditions, and better safety conditions thereby reflect well on their performance. This may lead to manager’s rating safety related aspects on their units higher due to the desirability of performing well in enacting safety policies and practices.

For other types of survey questions, the concept being measured is inherently subjective. For example, “The unit maintains excellent housekeeping.” Here, excellence is not defined using a shared metric and is in the eye of the beholder, influenced by one’s perception of what excellent housekeeping would be.

Research has generally shown that manager perceptions of organizational climate, safety climate, and related concepts are more favorable than worker perceptions [7-12, 18]. Although manager and worker responses may differ, they may still be equally associated with an outcome of interest. Given that manager and worker responses to survey questions about their work environment may differ, in this paper we ask how well these responses predict worker injury rates.

There are few studies that have explored this question, and their findings are inconsistent. In a study of limited-service restaurant workers, Huang et al. found that individual worker perceptions of safety conditions were associated with injury rates, while manager perceptions were not [8]. Similar results were found, perhaps not surprisingly, for workers working with a distant supervisor [5]. However, a study based in construction found that neither managers’ nor workers’ perceptions of safety climate had a statistically significant relationship with injury rates, but differences between worker and manager perceptions did [7]. A study of blue-collar workers in a university setting also found that workermanager discrepancies were predictive of injury rates, although it did not examine worker and manager perceptions separately [9]. Also, a 2017 study found that workers’ safety climate perceptions were not associated with injury rates, but that safety officers’ reported safety management practices were [16]. We note that the substantial body of safety climate research, pioneered by Zohar and others, relies exclusively on worker responses [1, 2].

Assessing the extent to which worker and manager responses are informative is notably influenced by: (1) the responses themselves, to the extent that workers and managers have different perceptions; and (2) the number of responses. When worker responses are included in a model, the analytical sample size is larger, resulting in increased precision of the estimates (smaller standard errors).

Here, we offer three competing hypotheses about safety-related practices:

This study attempts to illustrate an approach to answering three questions: (1) On average, do managers and workers have similar perceptions about safety policies and procedures within the same work environment, (2) Are assessments of safety policies and practices predictive of injuries, and (3) Does the answer depend on whom you ask? To answer these questions for the specific measures in this study, we analyzed the relationship between worker and manager perceptions about workplace safety policies and practices (measured using a survey) in two large tertiary-care hospitals and administratively-recorded worker injuries in the year following the survey. The answers to these questions have implications for a wide range of workplace surveys to determine whether both worker and manager responses provide additional predictive power compared to surveying just one or the other.

Table 2 shows the scores of workers and managers for the three scales. For each, the managers’ score was higher. That is, managers viewed implementation of safety policies and practices as better than workers did. Using a 2-sample t-test, all three pairs of means differed at p < 0.05.

Table 3 shows odds ratios of injury for a one-unit increase in the ergonomics, safety, and safety leadership scales respectively. For two of the three measures, a higher worker score was associated with a significantly lower injury rate, with p < 0.05. For all three, the manager odds ratios were closer to 1.0 compared to the worker odds ratios. Moreover, none of the managers estimated odds ratios significantly differed from 1.0. Safety leadership responses were not significantly associated with injury odds for either.

To directly test our hypothesis, we computed the chi-square statistic of the difference in residual deviance between the model including both worker and manager scores and the models with only one of the two scores (likelihood ratio chi-square test). The results are shown in Table 4. For each of the three scales, this difference is small and not statistically significant for the worker-only model. On the other hand, the test statistic for the manager-only model was larger for all three and statistically significant for the first two. This implies that adding the manager scores does not appreciably improve predictions compared to the worker-only model.

As a further test, we computed the AIC for each model and scale. For all three scales, the AIC was lowest for the worker-only model, suggesting that this model was preferred to both the manager-only and the worker plus manager model.

In our study population, we found that worker responses to questions about safety policies and practices were associated with injury, but that there was no additional predictive value from including manager responses in these models. Indeed, the manager responses alone were at best weakly predictive of injury. While our study cannot determine the reasons for these differences, prior research suggests that workers are closer to hazards and thus may have more accurate perception of injury risk factors. Conversely, managers may want to see themselves as doing a good job and, thus, may overlook safety problems in their units [5, 7]. Alternatively, our null findings about the relationship between manager responses and injuries might be related to our sampling methods, question framing, or other aspects of study design.

Compared to using manager responses alone, using worker responses within the same model also increases the precision of the estimates (results in smaller standard errors) in this study. This is likely because there are more workers than managers and therefore a larger worker sample size.

Finally, we found that manager-reported perceptions of the safety policies and practices examined were more favorable than those of workers within the same units. This is consistent with results from other studies [7-10]. It may reflect several factors, including differential understanding of policies and procedures, different experiences of workplace hazards, different safety expectations, and managers’ self-perceived prioritization of worker health and safety, which could influence perceptions of actual working conditions on their units.

Earlier studies in different settings largely found that worker responses to safety questionnaires were predictive of future injury rates [5, 8], although this was not always the case [16]. These findings depend on the measures used and may also depend on the industry and occupations studied. The purpose of this study was not to generalize about the predictive value of worker versus manager perceptions. Rather, we present an example for how to determine, for a specific survey instrument, whether surveying one of these groups but not the other reduces researchers’ ability to predict injuries. Future studies examining the predictive value of worker and manager perceptions using existing safety climate instruments like the Nordic Occupational Safety Climate Questionnaire [22] could add to this understanding.

This study is based on employer-collected OSHA-recordable injuries. Past studies have shown that this data is substantially underreported compared to workers’ actual work-related injury rates [23-25]. If our concern is the relationship of safety scales to injury rates, this underreporting may lead to biased estimates, although we cannot infer the direction of this bias. The same is true of differential reporting by race/ethnicity [26], gender, or occupation. Also, there is evidence that better safety climate leads to more complete reporting [27]. This could lead to an underestimate of the relationship between both worker and manager responses and reported injury rates.

This study raises a cautionary flag. The safety climate concept has typically been considered as a measure of worker perceptions [1]. For ease of administration, increased response rates, or reduced cost, employers and researchers have, at times, administered workplace safety surveys only to supervisors or managers [4]. The results from the current analysis caution against such an approach, as we find that this practice may result in losing the ability to identify and address safety problems. The ability of a manager-only or worker-only approach to predict future injury rates should be piloted before relying on only one group’s survey responses to guide efforts to improve safety. The approach in this paper can be applied to specific safety climate instruments to determine whether, in a given worker population, surveying both workers and managers provides additional predictive power compared with surveying just one or the other. When surveying both workers and managers is impractical, our findings suggest that given a choice between administering a survey to only workers or only managers, researchers and practitioners should consider a worker-only survey.