3.1.2. Design

We manipulated News Source within subject, attributing headlines to one of three sources. In addition, subjects assigned themselves into one of three Political Affiliation categories.

#### 3.1.3. Materials and Procedure

The experiment was identical to Experiment 1a, except as follows.

**News Sources.** We chose different news sources for this experiment. To more formally identify and quantify experimentally useful news sources, we first asked a separate sample of 202 Mechanical Turk workers to provide familiarity, trustworthiness, and bias ratings for each of the original list of 42 news sources [14]. The preregistration for this norming study is available at FA2018 - Fake news - Sources norming (#13611). https://aspredicted.org/4ep7 p.pdf (accessed on September 2021). Subjects observed the names of these news sources, one at a time, in a randomized order. For each source, subjects provided a familiarity rating (1 = Not at all, 5 = Extremely), a trustworthiness rating (1 = Not at all, 5 = Extremely), and a bias rating (1 = Strong liberal bias, 5 = Strong conservative bias). We also asked subjects about their own political leaning and party affiliation. From these data, we identified the two news sources rated maximally different on trustworthiness across Democrats and Republicans: CNN (*M*Democrat = 3.73, *<sup>M</sup>*Republican = 2.52) and Fox News (*M*Democrat = 1.95, *<sup>M</sup>*Republican = 3.27).

Subjects in the current study were presented with 48 headlines, randomly selected from the original set of 50, so that subjects rated 16 headlines per source. Each headline was attributed to one of three news sources. Specifically, subjects read: "X reported that ... ," where X was replaced with either "CNN," "Fox News," or "It was" for the unspecified source. This time, subjects did not rate the familiarity of each source. The data are available at https://osf.io/h6qen/ (accessed on 27 September 2021).

#### *3.2. Results and Discussion*

We analyzed data only from subjects who gave complete responses, and we did not exclude subjects on any other basis, contrary to our preregistration. Most subjects responded correctly to each attention check item (97% and 98%, respectively) and did not look up any headlines (98%).

Of the 201 subjects, 44 identified as Republicans, 92 as Democrats, and 65 as Other (or none). Distributions of the political leaning variable were consistent with these data: The modal selections were "somewhat conservative" for Republicans, "somewhat liberal" for Democrats, and "Moderate" for Other.

Recall that our primary question, as in Experiment 1a, was: To what extent does political affiliation influence how source information affects people's interpretations of the news? To answer that question, we examined subjects' mean headline ratings as a function of their political affiliation and news source. Table 1 shows the mean rating for each condition. A RM-ANOVA on mean headline ratings revealed—as in Experiment 1a—a statistically significant interaction between political affiliation and news source, suggesting that the influence of political affiliation on headline ratings depends on source information, *F*(4, 396) = 2.52, *p* = 0.04, η2 p = 0.025. We also included age as a covariate in an additional exploratory RM-ANCOVA, but found that age had no meaningful influence (all age-related *p* values > 0.18).

To determine where any meaningful differences occurred, we then ran three one-way ANOVAs testing the influence of political affiliation on mean headline ratings for each news source (we did not explicitly specify these follow-up analyses in our preregistration). As in Experiment 1a, these analyses yielded mixed results: Subjects' political affiliation influenced ratings of headlines only from CNN and Fox News, *F*CNN(2, 198) = 3.84, *p* = 0.02, η2 p = 0.037; *F*Fox News(2, 198) = 7.78, *p* <0.01, η2 p = 0.073.

More specifically, Tukey-corrected post hoc comparisons for these two sources revealed that Democrats rated headlines from CNN as less real than Others ( *M*Diff = 0.25, 95% CI [0.03, 0.47], *p* = 0.02). Democrats also rated headlines from Fox News as less real than both Republicans ( *M*Diff = 0.26, 95% CI [0.04, 0.49], *p* = 0.02) and Others ( *M*Diff = 0.31, 95% CI [0.11, 0.51], *p* <0.01).

Taken together, this collection of results is consistent with our hypothesis, but only partially so. We predicted that people would rate headlines attributed to sources favoring their political affiliation as more real than headlines attributed to other sources. That prediction was correct, but in contrast to Experiment 1a, only for headlines attributed to a source favoring people who lean politically right: Fox News.

Overall, the results of Experiments 1a and 1b sugges<sup>t</sup> that source information contributes to people's interpretations of the news. However, there are two key limitations to this conclusion. First, the observed differences were small, and not entirely consistent across our two samples. Consider, however, that subjects were provided with only the mere name of a source. It is perhaps surprising that such limited information can have any influence at all. Second, the headlines were normed to be relatively unfamiliar. We chose to use unfamiliar headlines in an effort to control for pre-existing knowledge, but it is possible that unfamiliar headlines convey so little information that they are almost meaningless. Again, however, it may be surprising that source information can influence interpretations of almost meaningless headlines.

Having conducted these initial investigations, we were then presented with a unique opportunity. In November of 2018, a United States White House intern attempted to take a microphone away from CNN's Jim Acosta during a press conference. Acosta clung to the microphone, resulting in brief contact between the two. Shortly afterward, then-Press Secretary Sarah Huckabee Sanders posted video footage of the interaction to Twitter. Sanders used the video as justification for revoking Acosta's White House press pass, claiming his behavior was inappropriate. However, rather than posting the original CSPAN footage, Sanders posted a subtly altered video that appears to have originated from a conservative media site [31].

Several media agencies raised concerns about the potential suggestive influence of this manipulated footage. Consistent with these concerns, a partisan split emerged, with those on the left tending to claim Acosta's behavior was unremarkable, while those on the right tended to claim his behavior was problematic. One explanation for this split is that the version of the video people observed guided their interpretations of Acosta's behavior. However, we suspected that the explanation was more nuanced, hypothesizing that any influence of the video would depend on political affiliation. More specifically, we predicted that due to beliefs about media sources, Republicans would be more susceptible to any potential influence of the altered video than Democrats. In Experiments 2a and 2b, we therefore tested the extent to which altered video footage of a real-world event affected people's interpretations of that event. In contrast to Experiments 1a and 1b, video footage of a real-world event provides a richer context than a sparse headline and allows us to explore the role of familiarity with the news story.

## **4. Experiment 2a**

The preregistration for this experiment is available at https://aspredicted.org/da3hg. pdf (accessed on 27 September 2021). The data were collected on 2 May 2019.
