When representation shifts from indirect to direct election, politicians may change not only where they stand but what they choose to make visible. We study this distinction at the Seventeenth Amendment, which transferred the election of U.S. senators from state legislatures to voters. The reform created the strongest incentives for senators whose inherited partisan bases were weakest statewide. Using floor speeches, we show that these senators broadened their public agendas once direct election became credible, relying less on established specialties and speaking across a wider range of issues. Their agendas also became more similar to those of House members, chiefly because specialization declined rather than because senators adopted House priorities. Roll-call positions show no corresponding response, and we find no strong evidence of a response in amendment activity. Direct electoral accountability reshaped senators most on the margin they controlled most freely: the public agenda they chose to build.

Learned representations have transformed the measurement of unstructured data in economics. We extend it to relational data, showing that Temporal Graph Networks encode economically meaningful behavior in dynamic corporate ownership networks. Using a high-resolution Italian ownership graph anchored on 5,700 judicially confiscated firms, we train a TGN to produce time-varying firm embeddings and validate them along three dimensions: they forecast confiscation out of sample up to four years ahead, well beyond firm financials and conventional network statistics; their geometry places firms confiscated only later closer to firms already confiscated at the time of measurement; and they respond coherently to local changes in ownership. We summarize this geometry with an Infiltration Proximity Index and use it to date firms' transitions into a high-risk regime in a staggered difference-in-differences design. Around the dated transition, firms display sharp scale expansion, rising liabilities and receivables, cost reallocation, and persistent illiquidity, with only temporary profit gains, patterns consistent with firms operating as conduits for financial flows rather than as profit maximizers.

This paper studies peer effects in leisure consumption using a longitudinal dataset of daily museum visits in Northern Italy. Leveraging administrative records on museum cardholders, we construct a dynamic social network from repeated co-visits and estimate endogenous peer effects on museum attendance and membership renewal. Identification relies on quasi-experimental variation induced by age-based membership discounts, which shift peers' behavior through exogenous price changes. We document sizable social multipliers: increases in peers' visits and renewal decisions significantly raise individual attendance and repurchase. Mechanism tests indicate that the dominant channel is a preference for shared experiences, with a more limited role for informational spillovers. Overall, the results show how social networks amplify cultural consumption in settings where participation is intrinsically social.