Human resource safety practices and employee injuries.

Compensation. Tying rewards (e.g., bonuses, pay, incentives, company perks) to safety indicates to employees the importance an organization places on safety. However, when tying compensation to safety, organizations must make sure that increased safe behavior, not reduced injury reporting, is reinforced. Collinson (1999) found that safety bonuses encouraged employees not to report injuries. Other studies, contrary to Collinson's, found that incentives actually improved safety performance (e.g., Harshberger and Rose, 1991). Perhaps the best support for the use of incentives is McAfee and Winn's (1989) review of twenty-four different studies that used incentives and/or feedback on safety for employees. Positive results (higher usage of protective equipment or lower numbers of injuries) were found in each of these studies.

Hypothesis 4a: Individual safety rewards have a positive association with reduced injuries.

Organizations and researchers continue to debate whether to reward groups or individuals for safety. Hoffman and Stetzer (1996) found both individual-level and group-level variables were associated with unsafe behaviors. Group compensation is most often used to improve the link between compensation and teamwork (Bartol and Hagmann, 1992). Lawler (1971) indicates that group compensation is most effective when employees believe cooperation will pay off. Many groups operate interdependently and perform tasks that affect others within the organization (Guzzo and Dickson, 1996). This interdependence occurs in safety when individuals "actively care" about each other (Hoffman and Stetzer, 1998), reiterating safety values. Employees may behave more safely with group safety rewards, since their actions affect several individuals.

Hypothesis 4b: Group safety rewards have a positive association with reduced injuries.

METHOD

Survey Distribution/Sample

Sampling Procedure/Respondents. A sample was obtained from the Association of Business and Industry and Safety Councils in a Midwest state. Because the use of HR practices across organizations in general was of interest, the study was open to organizations in all industries (type of industry was controlled for in later analyses). Fifty-four organizations completed the survey with 48 providing complete, useable data. Instructions asked organizations to have their top safety officer, CEO, HR Manager, or most knowledgeable individual of the organization's safety practices to complete the survey. Of the organizations responding, over half (51.9%) had multiple respondents. Organizations were classified into eight categories identified in OSHA's yearly reports. The majority (89%) of the responding organizations were from manufacturing (61.1%; including organizations manufacturing both durable and nondurable goods), services (16.7%; including personal, business, amusement and recreation, and health services), and transportation and public utilities (11.1%; including trucking and warehousing, communications, electric, gas, and sanitary services) industries. Agriculture (production of crops and livestock), mining (metal and nonmetallic minerals), finance (depository institutions, insurance carriers, agents, and brokers), wholesale and retail trade (trade of both durable and nondurable goods), and construction (building contractors and heavy construction) each had two or less organizations responding. Responding organizations employed anywhere from 4 to 6,000 employees (analyses controlled for industry size).

Measures

Independent Variables. The survey asked about selection, training, performance evaluation, and compensation practices associated with safety in the organization. New measures were developed for each of the independent variables in order to directly test the hypotheses. The survey requested that respondents answer Likert items based on a 1-5 scale (not at all = 1, great extent = 5). The items developed to assess each practice were examined using confirmatory factor analysis. Due to the low sample size, the Comparative Fit Index (CFI) and Root Mean Squared Error of Approximation (RMSEA) (Fan et al., 1999; Hu and Bentler, 1999) were examined. Fit indices ranged from .7 to .97. Although fit indices for performance evaluation and training were low, the scales were still included as fit indices may be negatively biased in small samples (Fan et al., 1999). Cronbach's alpha was estimated to determine internal consistency of each scale. These reliability coefficients were then used to correct for measurement error in the observed correlations. The resulting disattenuated correlations were used in the regression models (Schmidt and Hunter, 1996).

Selection. The five-item scale asked respondents to what extent employees are hired based on such criteria as their knowledge of safe work procedures and past safety performance, and if safety and the organization's safety policies are discussed in the interview process. Cronbach's alpha was .86, indicating a fairly reliable scale. Additionally, six questions about general selection practices were included. Respondents were asked to indicate to what extent (on a five-point scale) each of the following selection requirements/tests are used within the organization: prior work experience, personality-testing, intelligence-testing, drug-testing, alcohol-testing, and pre-employment physicals.

Training. The eight-item scale asked respondents to what extent safety training existed within their organization. Respondents were asked questions such as to what extent employees are trained on safe work procedures, if employees receive training on why safety is important, and if safety training is kept up-to-date. Cronbach's alpha (.97) indicated a reliable scale.

Performance Evaluation. Respondents were asked to answer six items about performance evaluation, including questions such as to what extent is safety emphasized and incorporated into employee evaluations, as well as to what extent employees receive formal and/or informal feedback about safety. Cronbach's alpha (.89) indicated the scale to be reliable.

Compensation. Eight items were developed for compensation (four asking about individual and four about group compensation). The scale asked questions such as to what extent individual employees were rewarded for safe behaviors and/or did individual employees receive compensation for not getting injured, as well as were employees rewarded for safety behaviors as a group and/or did employees receive compensation as a group for not getting injured. Confirmatory factor analyses demonstrated individual and group compensation were two separate factors. The reliabilities of the scales were fairly high (individual compensation = .87, group compensation = .91), indicating both of the scales to be reliable.

Dependent Variables. The organizational safety measure was obtained from the past five years of OSHA 300 logs (or an accident reporting sheet) provided by the safety officers. Organizational safety outcomes were measured as the average number of injuries, weighted for seriousness (based on type of injury and medical treatment required). The weighting was completed based on findings from Vredenburgh's (1998) study. In this study, Vredenburgh had twelve individuals in the medical field rate the seriousness of injuries. These ratings were then converted into an interval scale using Thurstone's discriminate model, the law of comparative judgment. From this, z-scores reflecting the interval separating the injuries in their severity were determined. The z-scores were then summed and averaged to express the severity value for each injury type, with the lowest injury type value being added to each of the other values to eliminate negative values. The scale values were from 0.0 to 3.92. This study then took the average number of injuries, weighted for seriousness, over a five-year period. Random fluctuation in reporting injuries was controlled for by figuring the reliability (.78) of injuries reported for one year, and then determining the reliability of the average injuries across five years (.95). This was then used to correct the correlations to help account for random fluctuation of responses over time (Schmidt and Hunter, 1996). The dependent variable was figured in this manner to provide a continuous safety measure for each of the organizations, while taking into account as many factors of injury variability as possible.

Control Variables. Size of the organization was measured with a single, open-ended item, asking "Approximately how many people are in your organization?" (Minimum = 4, Maximum = 6,000, Mean = 489). Type of industry was measured with a single open-ended item, asking "Type of industry." The organizations were then classified using OSHA's categorizations: agriculture, forestry and fishing; construction; manufacturing; transportation and public utilities; wholesale and retail trade; finance, insurance, and real estate; and services. These industries were then broken into high risk and low risk based on incidence rates per industry in OSHA records.

Analyses

Examining Respondents' Agreements. Because there were multiple survey respondents from organizations, the intraclass correlations (ICC) of the responses were determined (Bliese, 2000). The mean ICC was found to be .83 (Minimum = .64, Maximum = .96). Because the ICCs were fairly high, the means from each organization's responses were used for the analyses.