Relevance & Research Question:

Under GDPR Article 9(2) political opinions are classified as special category data (sometimes known as sensitive data) and explicit consent is required from data subjects to process this data. Guidance from the UK Market Research Society is that questions relating to voting can only be asked to individuals that have explicitly consented to responding on this topic.

This paper examines whether offering respondents the chance to opt-out from answering voting questions potentially reduces the accuracy of political research.

Methods & Data:

Kantar Public does regular political opinion polling in the UK. This analysis is based on five online opinion polls conducted between June and November 2018 (n=6,072).

All respondents were asked demographic questions and questions regarding the economy, Brexit, and policy issues. Questions relating specifically to voting behaviour – both in the past and future intentions – were only asked to the sub-set of respondents (92.6%) that consented to answer these.

The ‘ask all’ questions have been analysed to determine whether those that chose not to answer the voting questions are systematically different from those that consented.

Results:

The following analysis controlled for demographic differences (age, gender, region, education and working status) between respondents that consented to the voting questions (n=5,623) and those that opted out (n=449).

The response given to the consent question was found to be significantly associated with:

• Self-reported job security (chi-square p<0.01)

• Economic sentiment (rating of the economy now vs 12 months ago - chi-square p=0.05)

• The order in which respondents ranked eleven policy areas, in particular “reducing unemployment” and “increasing healthcare spending”

These attitudes are likely to be correlated with political opinions, and this suggests that the voting intentions produced from these polls may be biased.

Added Value:

This paper seeks to help inform the methods used for future political research. The findings indicate that complete case analysis is likely to be biased even with demographic weighting applied. Further research is required to determine the best way in which to deal with this.

Relevance & Research Question: Researchers in the quantitative social sciences have traditionally relied on survey methods for studying online behavior. In recent years, however, there has been increased interest in the use of so-called digital trace data, in particular data from social media platforms. Both types of data have specific limitations. Hence, combining these two data sources holds great promise for social scientific research on online behavior. Linking survey data and social media data at the individual level, however, requires explicit and informed consent from study participants. Especially in Europe with the recent introduction of the GDPR, researchers have to provide detailed information about what data they collect, how they collect it, and for what purpose.

Methods & Data: We conducted a web survey with participants of a German online access panel in which browsing behavior was tracked. Of 2042 invited panelists, N = 1347 completed our online questionnaire that focused on political attitudes and behavior and media use.

Results: 22.8% of the respondents (n = 307) reported having a Twitter account, and of those, 65.8% (n = 202) consented to the collection of their Twitter data. 196 supplied a Twitter handle, of which 68 were unusable due to typos, invalid strings or accounts that belonged to somebody else (e.g., celebrities). In a logistic regression model, we found that male, younger, and lower income Twitter users as well as those who have recently used Twitter are more likely to consent, whereas education and the incentive condition had no effect. Specifically, we compared a 5 Euro prepaid to a 5 Euro postpaid condition. Interestingly, only a very small minority of the respondents read the extended privacy information that was provided on the project website via a link embedded in the short version of the short informed consent in the web questionnaire.

Added Value: Our study highlights challenges that need to be considered when linking social media data and survey data. Moreover, we provide a critical perspective on obtaining consent and gathering information needed to link both sources for panelists of online access panels—populations that are frequently used to study consent decisions and generalize findings.

Relevance & Research Question:

Current scientific debates, such as on new technologies and their application, often involve emotional rhetoric styles. Those who read or listen to these kinds of scientific arguments have to decide whom they can trust and which information is credible. How do information seekers make such decisions?

Methods & Data:

Using a 2x2 between-subject online experiment, the current study investigates how the language style (neutral vs. enthusiastic) and the professional affiliation (scientist vs. lobbyist) of a forum post author influence his trustworthiness and the credibility of his information.

Results:

Results show that lobbyists were perceived as more manipulative than scientist. Furthermore, if the forum post author used an enthusiastic language style, he was perceived as more manipulative, less knowledgeable and his information was perceived as less credible. Moreover, both experimental factors interacted: When the forum post author was a scientist, enthusiastic language led to lower benevolence and integrity ratings. For lobbyists, this effect did not occur.

Added Value:

Since most of the Internet is not governed by editors, the validity of online information cannot be guaranteed. The current study demonstrates that information seekers use language characteristics to decide whether they should accept scientific arguments they encounter in online forums.