One hundred seven native English-speaking undergraduate students (74 females, mean age = 19.3 years, SD = 1.69) from Michigan State University participated in the experiment for course credit. An additional 11 individuals completed the experiment but were excluded from all analyses either because they were not native English speakers (n = 3) or because of experimenter error during testing (n = 8). Seventy-three of the participants (68 %) reported having Twitter accounts, and 1 additional participant formerly had an account but had deleted it in recent months.
Materials and procedure
All participants completed three experimental phases: encoding, misinformation, and confidence test. During encoding, participants viewed a series of 50 images that depicted a story of a man robbing a car (adapted from Okado & Stark, 2005). Each picture was displayed for 5 s, with a 500-ms interstimulus interval. After encoding, participants completed the operation span (OSPAN) task (Unsworth, Heitz, Schrock & Engle, 2005) to reduce rehearsal of information. In the OSPAN task, participants were asked to maintain verbal information in memory while performing arithmetic problems. On each item, participants performed simple mathematical verifications (e.g., 6 * 3 + 1 = 19) and responded by clicking on “True” if the statement was correct or “False” if the statement was incorrect. After each problem, they were given a letter to remember (from a set of 12 consonants). After 3–7 problems and letters, they were asked to recall the letters in the order in which they had been presented. During the recall phase, all 12 possible letters appeared in a grid on the screen, and participants clicked on the letters in the order in which they had been presented. After the letter recall phase, they were given feedback on the number of letters that were correctly recalled and the number of math errors for that set of trials. Participants completed three trials of each set size (3, 4, 5, 6, and 7 letters) and, in total, were asked to remember 75 letters amid 75 math problems.
During the misinformation phase, all participants viewed an information feed that presented 40 lines of text that narrated the events depicted in the images. While most of the information in the feeds was accurate, each participant was exposed to six details that directly conflicted with the images. For example, one line of text read, "The car had a Harvard sticker in the back window," when the car seen in the images actually had a Johns Hopkins sticker in the back window. Participants were not warned that they would see contradictory information.
To assess the effect of Twitter on false memory formation, participants were pseudorandomly assigned to one of three conditions: Twitter (n = 37), control (n = 40), or Twitter-control (n = 30). In all conditions, the information feeds were designed to provide information in a format similar to a Twitter feed. Each feed was a two-panel ticker in which new text appeared at the top. After 5 s, that text scrolled down to the lower panel, where it remained for an additional 5 s. Thus, each line of text was present for a total of 10 s.
The information feeds were visually similar; they were the same size and had the same format, but they differed in background designs across the conditions. In the Twitter and Twitter-control conditions, the information feed was labeled “Tweet Ticker” and was designed to resemble an official Twitter ticker; it had a light blue background and an image of the Twitter bird logo in the bottom right corner. Furthermore, a spatially blurred image of a person appeared to the left of the text, and black bars were placed at the top of each panel and scrolled with each response, creating the illusion of user censorship. The control feed was labeled “Photo Recap” and had a red background. Blurred images also appeared to the left of the text, but the feed did not contain censorship bars or social media logos (Fig. 1).
In the Twitter and Twitter-control groups, we told participants that the information feed consisted of tweets written by previous participants who had viewed the same images and tweeted responses as the images were presented. We told participants that the names and faces of the participants were blocked to protect their identities. To further strengthen our cover story, the text in the feed for the Twitter group was designed to resemble text found online; it was written with informal language and syntax. Moreover, some lines incorporated hashtags (#) or at signs (@), which are frequently used in tweets. Finally, to access the Twitter feed, experimenters navigated to a laboratory website and then clicked on a link to the Twitter feed. The control group was also told that the information in the feed was written by participants who had seen the photos in a previous experiment. This allowed us to isolate the effect of the medium of presentation, because all groups believed that the text had been written by past participants. The text for the control group was written in more formal language and consisted of complete sentences. To control for possible effects of language, the Twitter-control group was given the same cover story and visual display as the Twitter group but had the more formal text from the control condition (see Table 1, Appendix A, and Appendix B in the Supplemental Online Material [SOM]).
During the test phase, we presented events or details that may have appeared in the images and asked participants to respond on the basis of their memory for the images. Each stimulus was individually presented, and participants rated their confidence that the information was present in the images, using an 8-point Likert scale ranging from definitely did not see in pictures to definitely saw in pictures. There were 36 items in the test: information that appeared in the images and information feeds (n = 10), information that appeared only in the images (n = 10), information that appeared only in the information feeds (i.e., false information; n = 6), and novel lures that appeared neither in the images nor in the feeds (n = 10). For example, one item that appeared in both the images and the feeds was, “A man was walking down the street while wearing a Yankees t-shirt.” The man’s t-shirt was clearly visible in the images, and it was also mentioned in the first line of the scrolling feeds (see Appendix C in the SOM for a full list of test items). An example of an item that appeared only in the information feeds was, “The car had a Harvard sticker in the back window.” No time limit was imposed on responses.
Directly after the confidence test, we asked participants to rate how much attention they had paid to the information feed and how much they trusted the information in the feed, both on 5-point Likert scales from not at all to very much.
Finally, participants completed a source test. Each item from the test (except for novel lures) was presented again, and we asked participants to report where they had obtained the information about that item. Five options were provided: “saw it in the pictures only,” “saw it in the information feed only,” “saw it in both [the pictures and information feed] and they were the same,” “saw it in both and they were different,” and “guessed.” We were most interested in how often participants would attribute our suggested information to the images (the first and third options). This would constitute a richer form of false memory, since it would suggest some memory of the information appearing in the photos.