All time references are in CEST
Social Media Ads as a tool for targeted survey sampling 2
|Session Organisers|| Dr Michael Zoorob (Meta Platforms)
Dr Steffen Pötschke (GESIS - Leibniz Institute for the Social Sciences)
Dr Bernd Weiß (GESIS - Leibniz Institute for the Social Sciences)
|Time||Thursday 20 July, 14:00 - 15:30|
The sampling of hard-to-reach populations remains a challenge in survey research. For example, conventional survey methods are often incapable of reaching very mobile groups, such as international migrants, or employees of specific industries, who are defined by an attribute that is usually not included in readily available sampling frames in most countries. Additional challenges occur in the field of comparative social research when a researcher must grapple with multiple national sampling systems that prevent consistency in sampling design.
Though not without its problems and challenges, the wide reach and detailed targeting options made available by social media advertising provide a potential solution to some of these core-problems in survey research. A nascent but growing suite of research tools, including automated Chatbots that provide customized messaging and automated survey response collection, has developed to enhance this approach. Recent research has underlined that social media constitute a useful recruitment tool allowing timely and cross-national data collections on relevant issues such as (forced) migration and the COVID-19 pandemic. This panel will discuss recent empirical research that samples via targeted social media advertising sampling in the absence of feasible alternatives. Preference will be given to contributions that, besides demonstrating the added value of this sampling approach, address additional methodological issues in terms of survey quality, e.g., sample composition in comparison to appropriate benchmarks, or nonresponse bias. We also welcome contributions that discuss issues of research ethics when applying this sampling approach.
Keywords: Sampling, sampling frames, social media, advertisements, hard-to-reach populations
Dr Bernd Weiß (GESIS - Leibniz Institute for the Social Sciences )
Ms Anna Hebel (GESIS - Leibniz Institute for the Social Sciences ) - Presenting Author
Dr Steffen Pötzschke (GESIS - Leibniz Institute for the Social Sciences )
In recent years, survey recruitment via social networking sites (SNS) such as Meta's Facebook or Instagram gained popularity. Particularly in the context of hard-to-reach populations, such as Ukrainian refugees or emigrants. Since there exists no (easily accessible) sampling frame of, for instance, Ukrainian refugees, researchers place advertisements on Facebook or Instagram. By clicking on the ad, potential survey respondents are redirected to the survey's landing page. These ads' central and most salient features are visual stimuli such as images (or videos). To the best of our knowledge, no survey methodological research systematically investigates the effect of ad images on ad performance, sampling composition, or measurement quality across multiple surveys using this sampling strategy.
In our contribution, we will utilize data based on six SNS-recruited surveys that address various research questions and populations (German emigrants or refugees from Syria, Iran, or Ukraine) and employ different ad images (and ad strategies). Utilizing these data, we will cover above mentioned research gap by addressing the following three research questions: (R1) Does ad performance differ by image over and within all six surveys? (R2) How is sample composition for sociodemographic characteristics affected by the choice of ad images? (R3) How is measurement quality affected by the selection of ad images?
Our research contributes to research on survey recruitment via SNS in multiple ways. First, we can answer whether (and, ideally, how) ad images contribute to the performance of ads on Facebook or Instagram. Second, by studying the sampling composition or measurement quality issues for different ad images, we can (a) identify whether there are any differences and (b) whether we find similar patterns among the six studies, which then could be generalized to a larger set of SNS-based studies and help to design better SNS-based surveys.
Dr Molly Offer-Westort (University of Chicago) - Presenting Author
Dr Leah Rosenzweig (University of Chicago)
During mass vaccination campaigns, social media platforms can facilitate broader access to public health information, but they may also engender vaccine hesitancy through the spread of false or misleading information. We design and deploy a Facebook Messenger chatbot to test interventions targeting sources of vaccine hesitancy among social media users in Kenya and Nigeria. The main goal of the study is to evaluate the extent to which a concerns-addressing chatbot conversation is effective at increasing vaccination acceptance and self-reported vaccine intentions. We compare the interactive concerns-addressing chatbot to a chatbot that delivers a non-interactive PSA treatment, as well as a pure control. Using Facebook advertisements we recruit social media users 18 years and older in Kenya and Nigeria, and deliver our interventions with a Messenger chatbot, facilitating interactions in a realistic setting. Using an adaptive experimental design, we are able to target the most effective interventions and learn what works best quickly during an ongoing global health crisis.
Professor Dante Donati (Columbia University)
Mr Nandan Rao (Universitat Autònoma de Barcelona) - Presenting Author
By reaching billions of people everyday, digital Ad platforms such as Google, Facebook and Instagram offer researchers and advertisers unprecedented opportunities to gather survey responses and conduct experiments entirely online. Researchers have successfully gathered “convenience samples” by creating ads for targetable populations. However, by manually creating these ads, researchers are limited to the explicit targeting criteria available on the platform as well as limited in the complexity of stratification which is feasible.
We formulate the problem of dynamically setting ad budgets for survey sample collection as one of minimizing the variance of a post-stratification weighted expectation subject to budget constraints. We then show that this problem can be continuously optimized, in real-time, by building software that connects to both the APIs of the ad platforms and to the API of a survey platform. We share data from surveys conducted this way with 100,000+ people across 20 countries, optimized for post-stratification weights matching country-level demographics for age, gender, and location.
In addition, we show that the same framework enables the use of advanced targeting and audience creation techniques (i.e. Lookalike Audiences on Meta) to stratify by variables not explicitly available on the platform. We share data from a survey and experiment related to malaria prevention in India where the sample was stratified by a known malaria risk-factor not explicitly available: permanent vs. non-permanent dwelling type. Using this technique, we successfully recruited a sufficient number of individuals who are at greater risk for this disease.
Finally, we present the software as an open-source platform that can be deployed as a web application on public or private clouds.