Computational Communication Research - Current Issue

Digital trace data retrieved via data donations holds significant potential for the study of individual behavior. However, data donation studies may be subject to bias. Researchers therefore need to quantify and address systematic error in digital trace data. To complement a-posteriori error correction methods like statistical modeling, we tested how ex-ante approaches, in particular survey design strategies, may help address bias in data donation studies. We conducted two preregistered experiments, one with a convenience sample of students (N_I = 345) and one with a convenience sample from an online access panel (N_II = 2,039). In both experiments, we analyzed the effects of survey design strategies – technical support during data donation, personalized incentives, and highlighting the societal relevance of participants’ data – on nonresponse rates and nonresponse bias. Our results indicate that while data donation studies are prone to both, our ex-ante strategies could not effectively decrease nonresponse rates or nonresponse bias. Overall, our study underlines the need to (a) make bias in digital trace data more transparent and (b) advance research on error correction methods.

The use of online media has led to an increase in digital footprints of human behavior, which has resulted in a growing interest in the collection and analysis of such data in non-intrusive ways. One such approach is digital data donation, which involves requesting participants to share data that they have requested from digital platforms with researchers. While this approach promises to provide an unprecedented level of detail for computational communication research, it also raises concerns about the representativeness and validity of the data. This study investigates the issue of potential selection bias and non-response bias in data donation samples. It aims to identify subparts of the population that might be underrepresented (or even absent) in data donation sampling methods, which might lead to inaccurate and biased research conclusions. Utilizing a survey with a sample frame of 1178 and 289 participants from the Netherlands, the study investigates the relation between demographics, knowledge, privacy and trust factors, and donation behavior. The results show biases in donation behavior in terms of age and digital and algorithmic efficacy, while privacy and trust factors are not related to it. It suggests that individuals decide not to donate to academic research due to skills and not due to concerns or lack of trust. This offers possibilities for improvement in study design so that all willing individuals are equally able to participate.

Currently, few tools are openly available that researchers can use to collect data donations. The Data Donation Module (DDM) presented in this paper addresses this gap and provides a new solution for collecting data donations from citizens. The DDM is a web application that provides researchers with the means to guide participants through the data donation process, define rules on how the personal data provided by the participants are preprocessed and filtered before being transmitted to the researchers, and optionally enrich the donated data with the participants’ self-reports, collected through an integrated questionnaire. The design of the DDM builds on a particular set of principles that distinguishes it from other data donation tools. First, it can be deployed at an institutional level and can support multiple data donation projects simultaneously. Second, it emphasizes the usability of the application for both participants and researchers. Third, it offers additional built-in data enrichment features, such as the possibility to include information contained in a data donation in the integrated questionnaire. In this paper, we further elaborate on these principles, present the DDM’s core functionalities from both the participants’ and the researchers’ perspectives, and provide an outlook on the application’s accessibility and sustainability.

Data donation methods have shown great potential as a means to measure a person’s media consumption behavior and exposure at an unprecedented level of detail. Yet what hampers this potential is that studies often suffer from high drop-out rates, and the accuracy of the digital trace data cannot be taken for granted. To improve the potency of this method, we need to systematically investigate how different recruitment strategies and design choices affect drop-out and accuracy. We used a novel open-source data donation application, and reflect on both a survey and field study where participants were asked to donate their browsing and YouTube history data from Google. Our results confirm that drop-out is high and non-random in the survey study, but adds the positive note that a field lab settings might help alleviate primary barriers of participation. We reflect on opportunities and challenges for data donation research and tools based on log data from our application, questions to participants, and our experience of building the application and guiding users through it.

Advertising has historically played a major cultural role in media and society; targeted advertising based on the detailed profiling of consumers has become central to the business models of commercial digital and social media platforms. Despite their immense cultural and economic significance, the operations of these advertising systems are difficult to observe, preventing the independent, systematic study of both advertising content and how it is delivered to audiences. This paper details a novel data donation infrastructure aimed at improving the observability of both of these aspects of platform-based advertising. The infrastructure includes custom plugins developed for major web browsers to capture web-based Facebook advertising content and targeting information from individual Facebook users for large- scale offline analysis. The paper describes the technical components of the infrastructure, including the plugin(s), data storage, data processing and enrichment, and researcher dashboard. We also discuss preliminary findings from case studies conducted using the tool, the ethical considerations associated with the research, and insights for future data donation projects.

The collection of data donations holds great potential for making digital trace data accessible to communication research. However, a major challenge is the recruitment of participants who are willing to donate their data for research. Hence, this research report presents our experiences with different strategies used in one study to recruit citizens to donate their personal media usage data. Based on our experiences during a data donation study in which Google Search and YouTube usage histories were collected, we compare three different online recruitment channels—(a) an access panel, (b) social media advertisements, and (c) a data marketplace—in terms of five dimensions: (1) administrative effort, (2) setup effort, (3) participation rate, (4) costs, and (5) analytical considerations. Overall, participation rates were rather low across all recruitment channels. Nevertheless, access panels remain a viable option for recruiting data donors even if significantly lower response rates must be expected compared to survey studies. Furthermore, we see potential in recruiting data donors through online advertisements, especially if they can be targeted through mobile ads. By contrast, our experience with the data marketplace was discouraging.