I am trying to analyze data from an experiment where my collaborators and I were interested in whether tweets about risk can influence people's risk perception (a continuous dependent variable). We manipulated whether the tweets came from a high or low credibility source, and we also manipulated the content of the tweet (text only, text with accompanying photo, text with accompanying data table). We also wanted to assess whether our results would replicate across different risk topics, so we also included a within-subjects factor so that each participant viewed three tweets, each one about a different risk topic. After viewing a tweet about a given topic, participants self-reported their risk perception (i.e., three risk perception measures per participant).
However, participants did not necessarily see the same type of tweet across the three topics. For example, as I try to visualize in the figure below, one participant may have viewed the low-credibility/text-only tweet for Topic A, the low-credibility/text+photo tweet for Topic B, and the high-credibility/text+table tweet for Topic C. Another participant may have viewed different combinations of the tweet formats across the topics. In other words, within each of the three topics, participants could be randomly assigned to any one of the six tweet variations.
What would be the best way to analyze these data? Would this be considered a nested design (I'm assuming not a crossed design)? My hunch is that this would require some kind of multilevel model to analyze. Any suggestions (or recommendations for similar threads) would be greatly appreciated!
