Thursday, August 21, 2014
Judging Similarity (Part 2)
This post is by August/September GuestPrawf Irina Manta.
In my first post on this subject, I wrote about the background that motivated us to conduct our studies on copyright infringement for our paper “Judging Similarity”. Today, I would like to discuss our methodology and results in more depth.
In the first study, we tested the hypothesis that images seem more similar simply by virtue of being copies. We used an online format and recruited subjects on Amazon Turk. Subjects were told that they would be shown image pairs and would be asked to rate their similarity on a numerical scale. We specified that subjects had to base their answers entirely on a comparison of the works themselves. We also told subjects that while there would be both similarities and dissimilarities, they had to judge the works as a whole. Further, we told them to assume each time that the work shown on the left was created before the work shown on the right. Subjects were randomly assigned to either the independent group or the copy group. The former group was told that the creator of the work on the right did not know about the work on the left during the process of creation. The latter group received the instruction that the creator of the work on the right copied from the work on the left during the process of creation. Subjects then proceeded to view image pairs such as this one.
We conducted statistical analyses and found that the individuals who were told that copying occurred rated works as significantly more similar to one another than did individuals who believed that the creation process had been independent.
The results of the first study can mainly be explained in one of two ways. First, confirmation bias may lead individuals to find similarities more easily when told that a work was copied. Alternatively, motivated reasoning may influence people to punish copiers as part of the similarity analysis because they find copying morally distasteful. The first explanation struck us as more plausible in this case because subjects were not told about the consequences of the similarity judgment.
We then conducted a second study that would provide more legal context for the similarity judgment. Our goal was to test two hypotheses about how moral intuitions about unfairness of copying might affect judgments of similarity. We tested the effect of information about high versus low effort invested by the original creator. We separately tested the effect of information about negative versus no change in market demand for the original since the copy became available. This study used a single image pair in hopes of eliciting a clean response and approximating the conditions that a jury or judge would encounter in a trial. We changed our instructions to include explanations of copyright infringement and of the types of consequences that could result from a finding of liability. We also briefly explained “substantial similarity” and how it had to be present for a claim to succeed. The rest of our instructions resembled those from the first study, and subjects were told to rate similarity on a numerical scale comparing only the works themselves and viewing them as a whole. We told subjects across conditions that the work on the left is the original and the one on the right is the copy. We used this image pair.
We used four conditions. In our “high labor” condition, we told subjects that the creator of the original spent about two months designing and setting up the shot on the left. In the “low labor” condition, we said that this process had taken ten minutes. In the “market effect” condition, we told subjects that since the copy has become available, it has had a strong negative effect on the demand for prints and licenses of the original and that sales of the original decreased by over 60% since the copy came on the market. In the “no market effect” condition, we told subjects that the arrival on the market of the copy had no effect on sales of the original.
When we compared the two labor conditions, we found a significantly higher similarity rating for the group that believed that the original work had taken a high level of labor to create. When we analyzed the two market substitution conditions, however, we found no statistically significant difference between the groups. In part 3 of my post, I will discuss the implications of our findings.