Researchers are facing a range of issues in recruiting participants, including decreasing response rates and increased difficulties in reaching participants through traditional modes such as phone and paper [1-3]. In turn, these challenges are associated with increased recruitment costs. As the presence of social media has increased and as our communication strategies have shifted, researchers are increasingly looking to these systems for potential recruitment to address current challenges.

Facebook is the largest social media platform with over 1.8 billion daily active users and 2.7 billion monthly active users [4]. Of these, over 200 million users are within the United States. This high proportion of users allows Facebook advertising to reach potential research participants in a short period of time. As of 2019, up to 90% of US adults use the internet, a number that has been steadily growing [5]. Additionally, although 24% of Americans do not have an internet connection at home, 17% of total US adults use a smartphone without home internet access, thereby increasing the feasibility of internet-based research [6].

One of the largest barriers in recruitment is cost. Mailed survey costs are relatively evenly distributed, ranging from US $6.51 to $30.24; however, this includes high personnel costs [7-9]. Telephone surveys have been observed to range from US $29 to $99 per participant, also requiring high personnel costs [10,11]. Some online health and behavioral studies reported costs as low as US $0.64 per participant, whereas others reported costs up to US $33 per participant; however, the median cost tends to be skewed toward the lower end of the scale [12-15].

Both telephone and mailed surveys require preexisting information to reach a targeted population, whereas Facebook’s integrated Ads Manager utilizes its internal user data to determine such demographics. However, this broad targeting introduces further concerns over multiple biases [13,16]. Although more traditional recruitment methods have been used and evaluated for agricultural populations, often with mixed results, our review of the literature indicates that there has been little data reported on the use of Facebook or other social media platforms for recruitment [17-21]. Therefore, we undertook this pilot study to investigate the effectiveness of targeted advertisements on Facebook as a tool to recruit participants in the agricultural sector, and to assess possible self-selection and geographic biases in this sampling method.

To determine the feasibility and cost-effectiveness of Facebook advertisements for an agricultural health and safety survey, we ran an advertising campaign consisting of three separate advertisements, each with a different agricultural image. The campaign was performed on Facebook from March 9, 2020 to March 16, 2020 at a total cost of US $294, distributed evenly among the three advertisements. Images were selected based on the apparent safety adoption displayed to determine if advertisement imagery affected performance. This included low (overturned tractor), neutral (child feeding a calf), and high (woman sweeping a barn in full personal protective equipment) safety adoption (see Multimedia Appendicies 1-3 for examples of advertising images). These advertisements were run concurrently through the National Children’s Center for Rural and Agricultural Health and Safety (NCCRAHS) Facebook page. We investigated the cost per click, cost per participant, and potential geographical biases that arose during this campaign.

We chose the following inclusion criteria to target our advertisements: individuals currently living in the United States, identified as female, within the ages of 21-50 years, and had a “role,” as defined by Facebook, in Farming, Fishing, or Forestry (FFF; delineated by Facebook as Demographics > Work > Industries > FFF). Facebook reported 170,000 eligible users according to these criteria. This campaign directed users who clicked on the advertisement to an online, REDCap-hosted consent document outlining a novel open 10-item survey assessing farm family demographics, views on child agricultural safety, and open-ended feedback on the Facebook advertisement they were shown [22,23]. This survey was devised in anticipation of a research project assessing the impacts of relevant news media on farm mothers’ knowledge, attitudes, and behavioral intentions toward childhood agricultural safety. No financial incentive was provided for participation. The survey instrument, consent form, and protocol for this study were deemed exempt from review by the Marshfield Clinic Research Institute Institutional Review Board.

We investigated the geographic distribution of advertisements by comparing Facebook’s advertisement targeting system performance to an expected distribution based on the number of female producers at the state level [24]. We then mapped the ratio of expected to observed reach of the advertisements.

Over this 1-week period, the total reach of the advertising campaign (measured by number of views) was 40,024 and 318 users clicked on one of the advertisements, representing a click rate of 0.79% (Table 1). Of these 318 visitors, 29 consented to the survey, representing a participation rate of 9.1%, with 24 fully completing the survey for a completion rate of 83%. This yields a completed response rate of 0.06% out of the total reach and 7.5% out of the total clicks. The total cost per click for this study was US $0.92, whereas the cost per participant was US $10.14. With the 17% dropout rate (completed surveys vs consenting participants) and a total personnel cost of US $31 per hour at 4 hours, the total cost per completed survey was US $17.42.

The neutral advertisement depicting a child feeding a calf generated the most clicks with 122 out of 318 total clicks (38.4%). However, the image of a young woman sweeping the barn had the highest click rate at 0.84% (Table 1).

Advertisements were posted more often than expected in eastern states, whereas western states (excluding California, Nevada, and Utah) tended to be overrepresented compared to the distribution of female producers (Figure 1). There was no observed difference in geographical distribution among the three advertising images.