**2. Materials and Methods**

Three point-in-time surveys were conducted among worksites in Nebraska to test our central hypothesis that the prevalence of reported health promotion and safety strategies did not vary across years or industry sectors. The sample frames for all survey years (2010, 2013, and 2016) were generated from an employer database of establishments provided by the Nebraska Department of Labor. The sample frame data included worksite name, number of employees per worksite, worksite address, and industry code. Establishments were coded and grouped using the 2-digit industry sector according to the North American Industry Classification System (NAICS) ((Health Care and Social Assistance (62); Wholesale and Retail Trade (42, 44–45); Information, Finance, and Management services (51–55); Other Services (56, 71, 72, 81); Education Services (61); Construction (23); Manufacturing (31–33); Public Administration (92); Transportation and Warehousing (48–49); All Other Sectors (11, 21, 22)).

Worksites in the sample frame were defined as establishments with a Nebraska worksite address and 10 or more employees. In order to ensure worksites of all sizes were represented in the survey data, each sample was stratified by business size: Small (10 to 49 employees), medium (50 to 199 employees), and large (200 or more employees). Disproportionate stratification was used to allow for oversampling. All large businesses in the State were included in samples (*N* = 503 in 2010; *N* = 523 in 2013; and *N* = 525 in 2016). For small and medium-sized businesses, random samples were included (*N* = 1500 in 2010 and 2013, and *N* = 2010 in 2016 for both sizes). In 2016, two priority industry sectors with low responses in the 2013 survey, 'Construction' and 'Transportation and Warehousing', were oversampled. The sampling design allowed some businesses being surveyed across all three study periods, but none of the businesses were repeated within the same time period.

The survey questions were developed by the Division of Public Health, Nebraska Department of Health and Human Services (NDHHS) in consultation with the Bureau of Sociological Research (BOSR), University of Nebraska—Lincoln. Questions were adapted from a variety of sources, and a small pilot of the survey instrument was conducted with businesses randomly selected from the sample. Because these were point-in-time surveys, worksites were asked to report current workplace practices (i.e., if they have a specific health promotion policy or program in place) and perceived barriers in each survey year. The majority of questions remained unchanged across survey years to ensure comparability over time. Survey questions pertaining to this study are in Supplementary Materials (Table S1).

The 2010, 2013, and 2016 surveys were mailed to worksites, which included a cover letter, the survey, and a postage-paid envelope. The small and medium size surveys were addressed to the business owner or manager, while the surveys for large businesses were addressed to the human resource representative. For the 2016 survey, we provided an option in 2016 for businesses to visit a website and complete the survey via a web-based questionnaire.

Results were weighted to adjust for the business size differences found between the overall sample frame and the final compilation of businesses who are represented in the completed survey data. A weighting variable was calculated by applying the appropriate sampling weights and then also adjusting for nonresponse by strata [22].

A bivariable analysis was conducted to examine the prevalence of workplace governance and safety strategies among worksites over time, by industry sector and by worksite size. Weighted percentages and confidence limits were calculated. The Rao–Scott χ<sup>2</sup> statistic was used to assess differences according to year and industry sector. Effect modification was assessed by refitting the model multiple times, once for each of the main effects which was generated from the stepwise selection process. Point estimates and 95% confidence limits were calculated for multivariable analysis. The Wald χ<sup>2</sup> statistic was used to compare multivariable models fit to sectors with and without a workplace governance and safety strategies. Significance levels were set at α < 0.05. All data analyses were conducted using PROC SURVEYFREQ and PROC LOGISTIC commands in SAS version 9.4 (SAS Institute Inc., Cary, NC, USA).
