This dissertation examines how “Big Data” collected from Twitter can be used to understand how collective traumas (e.g., mass violence, campus shootings, and large-scale life-threatening warnings) impact individuals and communities. Prior research suggests that when collective traumas occur, people use social media channels (e.g., Twitter) to express psychological discomfort and seek critical updates, especially when updates from emergency management officials are lacking. Study 1 demonstrates the methodological value of mining locally-generated Twitter data after the 2015 San Bernardino terrorist attack. Results show that negative emotion increased 6.2% on the day of the attack and remained elevated for several days after. Studies 2a and 2b combine the methodological strengths of survey and Twitter data to understand distress responses, rumor exposure, generation, and transmission during a protracted lockdown event at a major university. Among about 3,400 students trapped in the lockdown, results indicate that a) rumor exposure and social media channel use during the lockdown were each associated with distress, b) students who trusted social media for critical updates reported increased acute stress, and c) exposure to rumors was associated with consulting Twitter for critical updates. In Study 2b, rumor generation and transmission occurred in the 90-minute gap in which no updates were transmitted to the campus during the lockdown. Study 3 exclusively uses Twitter data to explore the community- and individual-level impact of exposure to a life-threatening, but false, ballistic missile alert sent to smartphones of all residents and visitors in Hawaii in January 2015. Analyses at three time-scales reveal that a) anxiety expressed on Twitter increased 4.6% on the day of the false alert and remained elevated for at least two days, b) anxiety increased 3.4% every 15-minutes during the 38-minute alert period (incubation of threat), and c) users who expressed low pre-alert anxiety on Twitter exhibited the highest increase in anxiety expression at the onset of the alert (9.5%) and their anxiety remained elevated for 41 hours, compared to medium (5.6%; 23 hours) and high pre-alert anxiety users (-10.5%; 0 hours). The benefits of this big data approach for advancing knowledge and theory across all three studies are discussed.