JMIR Publications

ABSTRACT

Background:

Facebook’s advertising platform reaches most U.S. households and has been used for health-related research recruitment. The platform allows for advertising segmentation by age, gender, and location; however, does not explicitly allow for targeting by race/ethnicity to facilitate a diverse participant pool.

Objective:

We looked at the efficacy of zip code targeting in Facebook advertising to reach Blacks/African Americans and Hispanics/Latinos who smoke daily for a quit-smoking social media online study.

Methods:

We ran a General Market campaign for 61 weeks in all continental U.S. zip codes as a baseline. Concurrently, we ran two campaigns to reach Black/African American and Hispanic/Latino-identified adults, targeting zip codes ranked first by the percentage of households of the racial/ethnic group of interest, then by cigarette expenditure per household. We ran a Spanish Language campaign for 13 weeks, targeting all continental U.S. zip codes but utilizing Facebook’s Spanish language targeting. The ad images and language were common across campaigns. Costs were compared for ad clicks, queries, applications, and participants; and yields were compared for the final three outcomes. We examined outcomes before and after the Cambridge Analytica scandal that became public in March 2018. Finally, we examined two promoted Facebook features, Lookalike Audiences and Audience Network Placement.

Results:

Zip code targeting campaigns were effective for yielding the racial/ethnic groups of interest. The Black/African American-focused versus General Market campaign increased Black/African American weekly queries (mean=9.48 vs. General Market=2.83, p < .001) and applicants (mean=1.11 vs. General Market=0.54, p < .001). The Hispanic/Latino-focused versus General Market campaign increased Hispanic/Latino weekly queries (mean=3.10 vs. General Market=0.71, p < .001) and applicants (mean=0.36 vs. General Market=0.10, p=.001). Cost metrics did not differ between the campaigns at generating participants, overall p=.537. Costs increased post-vs. pre-scandal for the Black/African American-focused campaign for queries (mean=$8.51 vs. $5.86, p=.001) and applicants (mean=$59.64 vs. $38.96, p=.004) and for the Hispanic/Latino-focused campaign for queries (mean=$9.24 vs. $7.04 , p=.005) and applicants (mean=$61.19 vs. $38.19, p=.001). Lookalike Audience while cheaper than General Market for ad clicks (mean=$0.73 vs. $1.07, p < .001) was similar for queries (mean=$3.98 vs. $4.27, p=.637). Audience Network Placement on websites and apps versus ad placement on Facebook alone was cheaper for ad clicks (mean=0.11 vs. 0.69, p < .001), but substantially more expensive for queries (mean=13.40 vs. 5.62, p=.048).

Conclusions:

Zip code targeting in Facebook advertising is an effective way to recruit diverse populations for health-based interventions. Public concerns around data privacy should be considered regarding effects on recruitment and costs. Audience Network Placement should be avoided. The Facebook Lookalike Audience may not be necessary for recruitment with drawbacks including an unknown algorithm and unclear use of Facebook user data. Clinical Trial: ClinicalTrial.gov NCT02823028