PACs are most associated with legislators’ issue-attention in floor speeches compared with other legislator attributes, both in aggregate terms across the entire time period of our dataset as well as in individual congressional cycles.

We show how much various legislator attributes explain legislator’s issue-attention and how they compare with one another. Issue-attention is instantiated using topic distributions derived from floor speeches. A machine learning framework learns associations between legislator attribute values and their corresponding topic distribution on a training set, and then predicts the topic distributions for legislators on a held-out or validation set (i.e., these legislators are not used in the training of the model). A: shows the results in aggregate, with information about which individual PACs donated to legislators clearly explaining their issue-attention the most out of all the legislator attributes we consider; the cross-validation validation procedure was repeated 50 times, and the error bars (for results across the cross-validation procedure) are shown. B: shows the results when repeating our topic modeling and machine learning procedure separately for each of the twelve congressional cycles in our data, and demonstrates the same pattern as the aggregate picture (with the cross-validation procedure repeated 30 times; the error bars are provided in S5 Fig). (R) and (D) indicate which party was in the majority in the US House during that congressional cycle. Both of our findings are statistically significant (S7 Appendix).