Local newspapers are a vital but financially distressed pillar of U.S. civic life. To pursue economies of scale while avoiding antitrust challenges, the industry has consolidated through non-overlapping mergers, combining papers across different local markets. Yet whether such consolidation helps newspapers raise advertising revenue, deliver cost synergies, and how it affects consumer welfare remains unclear. This paper examines these questions using the largest consolidation event in U.S. newspaper history: the merger of Gannett and GateHouse. I assemble a novel dataset combining advertising, readership, and 4.5 million news articles analyzed with large language models and machine learning tools. Using a generalized synthetic control approach, I find that after the merger, advertising revenue rises by 40%, driven by an increase in advertising load rather than higher advertising rates. The share of syndicated content increases, while the core content characteristics readers care about remain relatively stable. I then estimate a structural demand model in which readers choose among newspapers and an outside option over price, content characteristics, and advertising load. Readers are ad-averse, and the merger reduces consumer welfare by 2.5%, driven almost entirely by the increase in advertising load, with a negligible contribution from price and content changes. This is roughly half the consumer-welfare loss from a within-market, duopoly- to-monopoly merger documented in the literature. Together, the evidence shows that consolidation can rescue struggling newspapers through higher advertising revenue and lower content-production costs, yet still harms readers even when it combines only non-overlapping local markets.

A growing number of professionals, such as physicians, use live streaming to attract clients and promote their services. Yet whether live streaming generates demand in professional service settings, and which features predict its effectiveness, remains unclear. We examine these questions using data from one of China’s largest online healthcare platforms, combining consultation records for more than 7,000 physicians with detailed information on their live streaming sessions. Using a generalized synthetic control approach, we show that live streaming adoption causally increases service demand. The effect persists for several months and is heterogeneous, with larger gains for more established physicians, including chief and associate chief physicians and those with higher baseline consultation volumes. To analyze live streaming effectiveness, we conduct a multimodal analysis of live streaming videos, extracting visual, auditory, and textual features using machine learning and large language models. Audience interaction, content, linguistic features, and emotional expression emerge as the strongest predictors of effectiveness. Interactive sessions, clear and concrete language, and a neutral emotional display are associated with stronger demand responses, whereas greater emotional variability is associated with weaker responses. Extending the analysis to investment fund managers reveals similar patterns, suggesting that live streaming is an effective demand-generation tool across professional service domains.

We examine how unanticipated income shocks shape taxi drivers’ labor supply within a shift, focusing jointly on break-taking and shift-ending decisions. Using high-frequency GPS data from 3.4 million shifts and more than 21,000 drivers in Singapore, we distinguish active work from non-work spells while drivers remain on shift, providing a direct measure of within-shift effort adjustment. Consistent with a dynamic model of reference-dependent labor supply, income shocks generate substantial responses along both margins, but the operative margin depends systematically on a driver’s position in the shift. Shocks primarily affect break-taking over a broad middle stretch of the shift and affect quitting only as drivers approach their typical end-of-shift time. Break adjustments account for a considerable portion of the total change in labor supply. Focusing on stopping behavior alone, therefore, understates the magnitude of reference-dependent labor supply responses and makes them appear more concentrated near the end of the shift than they are. The findings show that workers partially absorb income shocks through within-shift effort adjustments well before quitting becomes the relevant margin, with implications for measuring labor supply and designing incentives in goal-directed work settings.

Analyzing the transaction data of a retail chain selling storable products and targeting upper-middle-class customers, we find that, besides making larger expenditures on a payday, even on a non-payday, customers make larger expenditures as long as it is their first trip to the retail chain since a payday. Thereafter, the per-trip expenditures decrease over trips within the monthly paycheck cycle until an upward jump on the first trip since the next payday. This pattern suggests that consumers without facing monthly liquidity constraints may self-impose a monthly mental budget. Their daily expenditure decisions follow a rule of thumb trying not to overspend beyond the mental budget. They renew the mental budget on paydays and the salience of paydays also causes overshoots in expenditures