Arrest Charges’ (Non)-Independence in Black and White Men An Extension and Application of SimRank to include Network Structure when Computing Black and White Men’s Arrest Charge Similarities

We extend the SimRank algorithm (Jeh and Widom 2002) to efficiently examine the similarity of charge pairs in a large dataset of men arrested in 2013. We compare computed SimRank scores of charge pairs within black and white arrestees and evaluate how these measures differ from the more traditional measure of similarity of Chi-squared residuals, which does not account for network structure. Our procedure identified certain charges involving drugs, vandalism, theft, and assault to have a higher similarity in terms of the network than would be assumed based on their simple co-occurrences, and found that these measures were similar across black and white men.