Goodman and Kruskal's gamma statistic for likert scale data

#1
I have been told to use the Goodman and Kruskal's gamma statistic to look at Likert Scale data. I The data that I am looking at used a Likert Scale (1=very dissimilar to 7 very highly similar) asking one question 20 times for different images of maps. The question was"how similar or dissimilar are the images? The images of maps were presented in different sets: 2, 3, 4, 6, 9. For instance, the user would see two maps side by side and have to make their 1-7 rating on how similar or dissimilar they were. The question was the same every time, but the number of maps per trial varied.

Is the gamma statistic the right way to analyze this data? I want to make sure I am thinking about this correctly. Would I be calculating the gamma statistic to see what the association between each set combination would be with respect to the similar/dissimilar rating (e.g., sets of 2 and sets of 3, sets of 3 and sets of 4)? The examples of this that I am finding all compare different questions that were answered using the same likert scale, whereas this data has the same question for each. To me, the interpretation of a significant finding does not seem to make sense. Say the correlation was .36 (or 36%) with p<.05; which would mean there is a positive correlation -- as their similar/dissimilar score increases to being more similar, their reported similar/dissimilar score increases on another set. This makes me think there must be another way to look at this data.

Thank you for the assistance!