The Boston Hospital Workers Health Study (BHWHS) is a longitudinal occupational cohort study of over 27,000 hospital workers at two large academic medical centers in Boston (23). In periodic surveys of a random sample of workers, we measure working conditions, stressors, health, and wellbeing, with surveys directly linkable to administrative data (24–26). In 2023, we expanded the cohort from our historical population of patient care workers (nurses, low-wage nursing assistants) to include low-wage service workers at the hospitals (environmental services, patient and materials transport).

We conducted formative qualitative interviews with workers and managers from low-wage groups. Interviews informed survey content, built relationships and trust with managers, and identified potential barriers to participation (e.g. concerns with emailed surveys being mistaken for phishing).

The survey was distributed electronically to workers at their institutional email address from June 2023-January 2024. It was available in English, Spanish, Haitian Creole, and Portuguese and designed to take 30 minutes or less. Respondents received a $20 Amazon claim code after completion, sent to their institutional email address.

The database linkage component of BHWHS adds methodological strength, but presented barriers to survey response because a survey link unique to the workers’ study identification number had to be sent by email. Workers that did not regularly use work email struggled to navigate their email and the survey, or did not understand certain questions, even in their preferred language. To address this, we reserved a conference room with institutional laptops, mirroring processes for mandatory hospital trainings. Managers helped us determine optimal times for our team to sit in the room, explain the study, and help workers access their email. This strategy increased the response rate among environmental services workers by five percentage points, from 18% to 23%. In contrast, our initial approach of scheduled reminder emails to this group, without accompanying team visits or support, yielded an Increase of <1 percentage point per reminder.

For unit-based workers (nurses, nursing assistants), study team members visited inpatient units (n=149 units across the two hospitals) with flyers and study-branded candy to leave in break rooms, asking unit nurse directors to promote the survey to workers. During informal interactions with nursing assistants at these visits, the study team identified several barriers to response, even among those who reported receiving the survey: discomfort with navigating an online survey, literacy challenges, concern about permissibility of taking the survey on “work time,” and the belief that as (self-perceived) peripheral workers, their responses were not valued.

To address these challenges, the study team transitioned away from our initial approach of leaving materials in break rooms and speaking with nurse directors, and to a high-touch approach. During visits, we asked the charge nurse on each unit to pull the on-duty nursing assistant off the floor so a study team member could help them access the survey; this provided available workers with implicit manager permission to take the survey at work. After each visit, we sent emails to unit directors inviting them to tell nursing assistants about the importance of taking the survey so their voice could be heard. These approaches boosted response rate among nursing assistants from 33% to 46%, compared with increases of 1–2 percentage points with email reminders but no accompanying visits. Overall, the study had responses from 2,023 workers out of 4,618 sampled (44%), with response rates of 49% among nurses, 46% among nursing assistants, 23% among environmental services workers, and 50% among patient/materials transport workers.

- Erika L. Sabbath, Meg C. Lovejoy, Susan E. Peters

Community health workers (CHWs) improve care quality for socially and medically complex communities (27). The goal of this study was to understand how worker voice (ways that workers attempt to influence their work to meet individual and collective interests) may be associated with CHW health and wellbeing (28,29). The study took place in Massachusetts from March-November 2021.

We initially hoped to partner with organizations employing CHWs, but although organizations expressed enthusiasm, none opted to formally partner. Instead, we proactively used community organizing strategies for recruitment (30), including engaging with CHWs prior to the launch of the study to co-develop research questions, and building partnerships with CHWs, CHW professional associations, and employing organizations.

We initially sent recruitment emails to hospitals, health centers, and community partner organizations, using publicly available contact information; we asked organizations to forward the email to their CHWs. We quickly found that we could not rely on these organizations, instead developing more grassroots recruitment approach. The PI built trust with CHWs by volunteering to give invited talks at community meetings at which she shared findings from prior research on CHWs and discussed wellbeing for social service professionals.

At survey launch, many CHWs were exhausted from navigating the COVID-19 pandemic, and many felt over-surveyed. We established multiple systems to engage and support them. We sent recruited participants personalized emails (versus mass email): first to send them a unique survey link, and then to offer support for survey completion. These strategies resulted in minimal missing data (94% had complete responses). We offered respondents a wide range of $40 incentive options, including a Visa gift card or gift card from any store they wanted (postal mail or electronic), or electronic payment. Personalized communication helped build relationships between worker and researcher, increasing trust and willingness to participate.

Overall, the study had a sample of CHWs that demographically mirrored the known population of CHWs across Massachusetts (31). Of the 314 interested and eligible individuals, 220 responded (70% response rate), in contrast to a prior study of Massachusetts CHWs with a response rate of 36% (31).

- Yaminette Diaz-Linhart

The Shift Project (32) is a repeated cross-sectional survey of hourly workers employed at 150 of the largest service-sector firms (fast food, casual dining, grocery, pharmacy, big-box stores, hardware, other retail, delivery and fulfillment) in the United States. The survey is run twice per year, with 16 waves of data collected since 2016 accumulating surveys from 250,000 hourly workers.

Using surveys, the Shift Project measures workers’ self-reported wages, work schedule stability and predictability, access to paid leave, exposure to surveillance and automation, respect, harassment, job security, workers’ economic security, health, and wellbeing. These data are employer-employee linked, with workers nested within identifiable firms.

No accessible sampling frames of hourly workers at large U.S. service-sector firms exist. Given relational, legal, and logistical challenges associated with creating multiple employer-research partnerships to survey workers, there are formidable barriers to using employer partnerships to construct employer-employee matched survey data at scale. Thus, we developed a method to directly recruit workers at specific firms. We constructed a sampling frame using targeted advertising tools available through Meta to reach Facebook and Instagram users. Using the audience creation tool, we constructed employer-specific Meta “audiences” of workers at specific firms and then ran paid advertisements to each of these audiences, inviting users to complete the survey. Respondents were entered into a drawing to receive a $500 gift card. Over the last 14 waves of data collection, we averaged advertising/incentive costs of $13 per “sufficient” survey response, defined as progressing sufficiently through the survey prior to break-off to report on basic dimensions of job quality.

We have honed multiple recruitment strategies on Facebook and Instagram. First, recruitment ads perform best when the audience characteristic, advertising design, and advertisement message align (e.g. Facebook and Instagram advertisements targeting Walmart workers containing a picture of a Walmart worker and a recruitment message of “Survey for Walmart workers”). Advertisements missing any of these elements were less effective for recruiting target worker populations. For example, in April of 2021, a set of recruitment advertisements were erroneously run without including employer-specific targeting, but with employer-specific recruitment language in the ad text, so the ad was shown to the general population on Meta platforms. After discovering the error, targeting was fixed and data collection restarted. Using this “natural experiment” to compare efficacy of targeting, we found that untargeted advertisements produced 17 responses (0.3% of clicks), versus 473 responses (10% of clicks) when correctly targeted.

Second, fraudulent, low-quality data are concerns with online data collection (33). We assessed IP address and included “honeypot” questions to guard against bots; we have found no evidence of fraudulent responses across 16 waves of data collection. We also included basic attention checks; 91% of respondents pass checks, in excess of rates reported in online opt-in survey panels (33). We, like other researchers (34), find that guaranteed incentives (versus random prize draws) attract fraudulent responses.

Beyond imperfect coverage of the sampling frame and low response rates, Meta’s algorithm delivers recruitment advertisements non-randomly. Non-random selection is a common problem in modern survey research, often managed by post-stratification and weighting (35). Asking core demographics upfront allows for maximum preservation of sample size when weighting. While the sampling is non-probability, we benchmark Shift data to gold standard sources (e.g. National Longitudinal Survey of Youth-97, Current Population Survey (CPS)) to gauge representativeness. Before weighting, the Shift sample is biased on univariate measures of gender, education, and race/ethnicity, though not age, wages, or job tenure. However, the sample accurately represents bivariate associations (e.g. between job tenure and wages), as estimated in NLSY-97 and CPS (36).

- Daniel C. Schneider

The Fulfillment Center Intervention Study is a group-randomized controlled trial, comparing participants in worksites randomized to a participatory intervention (Health and Well-Being Committees, or HaWCs) with participants working for the same firm in control sites (37). HaWCs serve as a formal voice channel where a small group of frontline workers and supervisors solicit workers’ concerns and ideas about safety, the psychosocial environment, and work organization, then develop and implement improvement initiatives. The study utilized an established research partnership with the supply chain division of a mid-sized, non-unionized U.S.-based retailer. Sixteen fulfillment centers were randomized after matching on picking technology and building size.

Survey measurement occurred at baseline, 6 months, and 12 months. At each timepoint, all current workers in each fulfillment center received a paper invitation to complete a web-based survey. Survey packets also included a survey ID number to track repeat participation and a treat (granola bars, fruit snacks). Letters contained a QR code and link to the survey, allowing workers to take the survey electronically. Hard copies were also sent to buildings if workers preferred. Invitations were distributed at fulfillment centers during working hours; buildings were encouraged to provide workers with time off the floor to complete the survey. Upon survey completion, participants received a guaranteed incentive ($10–15, either a physical gift card sent to their home or an electronic card sent to their personal email). Weekly raffles were run for each building during the three-week-long survey periods, with prizes ranging from $25-$150. During three waves of data collection, 4,444 workers were eligible to participate; 63% participated in at least one survey (N=2813). Response rates were 48% (Wave 1), 47% (Wave 2), and 55% (Wave 3).

Data collection coincided with pandemic-related restrictions (baseline recruitment started July 2021) which limited the research team’s ability to be on-site to establish rapport and promote the survey. Instead, supervisors and managers announced the survey, which was a challenge as they were simultaneously managing COVID-related challenges and regular work duties. The research team designated a “survey point person” within each building to be an on-site contact to distribute and promote the survey. However, this reliance on management may have made workers skeptical, dampening participation. Once social distancing rules were lifted, we encouraged town halls where everyone in the building was introduced to the study and given time to take surveys.

The organization of work and incentive structure of the firm posed unique challenges. Time off the floor is frowned upon within fulfillment centers, as managers perceive it as reducing productivity. Additionally, some sites use performance-based pay, which may have disincentivized workers from taking the survey while “on the clock” for fear of losing pay. Furthermore, the company launched their own engagement survey the same year as the study survey. The overlap created survey fatigue and difficulty for managers, who were recruiting workers for both surveys.

We adopted several strategies in response. First, regular updates on response rates were shared with our building contacts to create friendly competition for high participation; buildings that reached over 50% participation received an additional three $100 raffle prizes. To encourage participation and combat survey fatigue in the final survey wave, we increased guaranteed incentive amounts from $10 to $15. There were also four $25–150 raffles for participants within each site per wave. Finally, the research team extended survey timelines while in the field to maintain positive relationships with management, stagger with company-wide surveys, and avoid peak periods where time off the floor was especially challenging. We implemented all presented strategies between Waves 1 and 3; the response rate increased by seven percentage points between waves, from 48% (Wave 1) to 54% (Wave 3).

- Grace DeHorn, Meg Lovejoy

The Work Organizational Health Study (WOHS) was a proof-of-concept cluster randomized control trial among food service workers, designed to test whether a multilevel participatory intervention targeting work organization and environment could be feasibly implemented and improve workers’ safety, health, and well-being (38,39). Collaborating with a large multinational food service organization, we worked with ten cafeteria sites providing food service to contracted corporate clients in greater Boston. Cafeterias were randomized to either intervention or control conditions. All food service workers were eligible to complete surveys. Baseline survey data were collected July-August 2018, with follow-up in early 2020; follow-up data collection was incomplete due to COVID-19.

During intervention development, we conducted focus groups with workers, and interviews with managers, to learn about the worker population and work environment (40). These data informed the intervention and alerted us to potential data collection barriers (e.g. literacy and language needs).

Due to reported low literacy of workers by management, surveys were conducted using an interviewer-administered protocol by two trained research assistants, one of whom was English-Spanish bilingual (16% of surveys were completed in Spanish). We hired experienced survey administrators who understood the importance of quickly establishing trust and rapport. The survey was designed to take 30 minutes or less. Survey respondents received a $25 Amazon gift card after completion. There was a 92% response rate (120/130 eligible workers).

Some workers had low to no reading or spoken English literacy, so in addition to hiring the bilingual survey administrator, we also created Spanish versions of all materials (e.g. recruitment posters). We were invited to attend standing morning huddles with employees; we delivered prepared huddle scripts about the survey in both languages, emphasizing privacy of responses and voluntary participation. We learned from our project champion that communication and trust between managers and staff were strained. Accordingly, pre-survey recruitment and study launch were driven by the research team, rather than managers, to communicate our independence.

Company management allowed surveys to be completed on work time; many workers had second jobs, making completion otherwise unfeasible. Because of already high workloads, we had difficulty accessing workers at times they were able to complete the survey. The research team needed to be flexible to work around peak times and unexpected catering jobs; we spent many hours waiting for employees to be available to survey. We also returned for a “cleanup” day to offer the survey to staff who had been unable to complete the survey initially.

- Susan E. Peters