Jaime Marques-Pereira

I study how social media messages are amplified by television news and how this amplification shapes public opinion. Using high-frequency data on President Trump's tweets and cable transcripts, I show that Trump's tweets caused near-immediate shifts in cable coverage, providing causal evidence of agenda-setting power. Linking text-on-screen data to a large public opinion survey, I find that broadcasts of Trump's tweets causally shifted viewers' approval of Trump and their 2020 voting intentions, widening gaps across TV audiences. Additional evidence shows that this mechanism is not specific to Trump, with cable news actively amplifying online statements by other prominent U.S. politicians.

U.S. drone strikes are electorally popular and authorized by the President. We test whether drone strike activity is shifted for political gain around U.S. federal elections. Using data on drone strikes in Afghanistan, Pakistan, Somalia, and Yemen, we document a pronounced pre-election increase in strike incidence. To separate political timing from military necessity, we exploit country-level cloud cover as plausibly exogenous variation in strike feasibility and study how strikes are reallocated across war theaters during electoral and non-electoral periods. Consistent with strategic behavior, cloud cover in one country induces more strikes in other clear-sky countries -- but only during elections. Finally, using a comprehensive corpus of cable-TV closed captions, we show that drone strikes are more likely in weeks when presidential coverage is negative, suggesting that drones may also serve as a diversionary tool. This relationship vanishes when news pressure is high, consistent with drones being used as a diversion only when there is scope for coverage.

This project studies how the emotional tone of televised political speech shapes polarization. Using a novel dataset of U.S. cable news audio spanning fifteen years, we quantify emotional expression in speech using transformer-based models. We document substantial changes in the emotional composition of primetime coverage over time. We complement this analysis with a survey experiment, in which participants are exposed to political statements delivered with different emotional tones, to assess how emotional delivery affects partisan attitudes and engagement.

This project studies gender disparities in Free and Open Source Software (FOSS) using the universe of public contributions archived by Software Heritage. We document large gender gaps in participation, contribution types, and visibility across tens of millions of developer activities. We investigate how institutional and family-related constraints contribute to these gaps, exploiting variation in school closures and parental leave policies across countries to show that child-related responsibilities significantly reduce women's participation and widen gender disparities in software collaboration.