The Research Collaboration Program
To date, Crisis Text Line has supported more than 5 million conversations. We share data to support smarter research, policy, and community organizing. Research Collaborators are granted access to a scrubbed and anonymized version of this dataset.
In addition to the message content, Crisis Text Line has four main sources of data: (1) metadata, including over 200 variables, and content automatically collected from conversations, (2) Crisis Counselor surveys, (3) texter surveys, and (4) algorithm-derived data.
What are the goals of the Research Collaborators program?
- Access. To open the door for world-class machine learning / artificial intelligence researchers to work with one of the country’s most important data sets.
- Dissemination. To create a consistent flow of innovative and insightful media stories to share with community groups and policy makers.
- Implementation. To create data science products that are implemented by Crisis Text Line and community groups, and improve the crisis space as a whole.
What outputs are expected from Fellows?
- Research papers and long-form journalism. We expect these will come after the research is complete. Because Crisis Text Line’s data is so sensitive, developing the paper will be a collaborative effort between the Research Fellow and Crisis Text Line staff.
- Augmentation products. Examples of products include text classifiers that can automatically identify certain issues, like suicidality, or a recommendation engine that could automatically recommend the best resource for a high school student in need of a methadone clinic in Sacramento. These will improve Crisis Text Line’s service. We’ll also share products with other non-profits for free, to raise the crisis space as a whole. (Crisis Text Line has already proven this model twice, having incorporated products built by visiting researchers from Palantir and Pivotal Labs.)
- Blog posts. One Crisis Text Line Communications Team staff member will dedicate time to working with Research Collaborators to identify story ideas. These posts won’t require full accuracy; they’ll be written from the perspective of imperfect results in motion. The goal is to create a more consistent flow of meaningful stories around crisis in the media.
To be a Research Fellow, you must meet the following key requirements:
- Be affiliated with an academic or research institution. This includes obtaining the approval of an IRB, having a Principal Investigator that’s a full-time employee of the institution, and the written approval of your institution’s Office of Research or equivalent office.
- Work from our NYC offices. We provide our Research Collaborators (one researcher per team) with stipends to cover room, board, and travel to NYC. (Note: We offer remote, VPN access on a case by case basis due to COVID-19)
- Expect to spend 3-6 months working on their project. Expect to spend half or more of that time on-site in NYC. Time spent on-site does not need to be continuous, but it will be frequent!
- Have the technical expertise to work with massive data sets.
- Be willing to complete our 12 hour Crisis Counselor training to better understand our data.
- Be a U.S. or Canadian citizen. We need this for background checks.
See the full list of requirements.
Since 2013, more than a dozen papers have been published based on our data, and there is still more to come. We’re proud to fuel this kind of learning! To get a sense of the range of projects possible, see a list of published papers from our research collaborations below.
- Researchers studied changes in help-seeking behavior for adolescents and young adults in North and South Carolina before and after Hurricane Florence. They found an immediate and sustained increase in crisis texts for stress/anxiety and suicidal thoughts in the six weeks following Florence. These findings highlight the need to address mental health consequences in youth following a weather-related disaster. Researchers: Jennifer D. Runkle, Kurt D. Michael, Scott S.Stevens, Margaret M. Sugg; Science of The Total Environment, 2021. Click Here to read more!
- Researchers sought to understand how text message-based crisis counseling facilitates child maltreatment disclosures. The findings of this study have important implications for the use of technology to support disclosure of child maltreatment. Researchers: Laura Schwab-Reese, Nitya Kanuri, Scottye Cash; JMIR Mhealth Uhealth, 2019. Click here to read more!
- Researchers studied the dynamic between forwards and backwards oriented phrasing in crisis counseling. They found that focusing too much on discovering next steps happens at the expense of building rapport, while focusing too much on exploring can stall progress towards a goal and establishing next steps. The highest quality conversations tend to have “backwards” and “forwards” oriented statements that are congruent with the stage of the conversation outlined in Crisis Text Line training. Researchers: Justine Zhang, Cristian Danescu-Niculescu-Mizil; Proceedings of ACL 2020. Click here to read more!
- Researchers analyzed the engagement of subgroups of texters who reported suicidality. They found that most texters who reported suicidal ideation had similar issues, yet there appeared to be distinct subgroups of texters with different issues, which may inform how crisis counselors tailor strategies for both low- and high-volume texters. Researchers: Hannah Selene Szlyk, Kimberly Beth Roth, Víctor García-Perdomo; Psychiatric Services, 2019. Click here to read more!
- Researchers studied crisis patterns in youth following the release of 13 Reasons Why (Season 2) and celebrity suicides. They found that both the release of 13RW2 and the celebrity suicides in the summer of 2018 were followed by an abrupt, but transient rise in crisis help-seeking among adolescents. The findings suggest media outlets should consider pairing suicide-themed content with crisis support services as a core best practice to reduce the risk of population-level adverse reactions to suicide portrayals or coverage. Researchers: Margaret M. Sugg, Kurt D. Michael, Scott S.Stevens, Robert Filbin, Jaclyn Weiser, Jennifer D. Runkle; Preventive Medicine Reports, 2019. Click here to read more!
- Researchers sought to quantify the extent to which Crisis Counselors change their linguistic behavior with experience. The study revealed that Crisis Counselors evolve in their choice of words as they gain experience. They also get faster at writing messages and write more linguistically diverse messages. Researchers: Justine Zhang, Robert Filbin, Christine Morrison, Jaclyn Weiser, Cristian Danescu-Niculescu-Mizil; Proceedings of ACL, 2019. Click here to read more!
- Researchers studied the relationship between crisis support-seeking behavior in young adults/adolescents and temperature in the United States (2019). This study suggests that there is a significant association between high minimum or maximum temperatures and crisis help-seeking behaviors in young adults and adolescents in urban areas in the United States. Researchers: Margaret M. Sugg, P. Grady Dixon, Jennifer D. Runkle; Science of The Total Environment, 2019. Click here to read more!
- Researchers developed a 3-step set of guidelines for organizations that wish to develop their own data-sharing program with academic researchers. The guidelines explain how to (1) determine the value and suitability of the data and organization for creating a data-sharing program; (2) decide on an appropriate data sharing and collaboration model; and (3) develop protocols and technical solutions for safe and ethical data sharing and the best organizational structure for implementing the program. They determined that with careful planning and appropriate resources, data sharing between big data companies and academic researchers can advance their shared mission to benefit society and improve lives. Researchers: Anthony R Pisani et al.; J Med Internet Res, 2019. Click here to read more!
- Researchers conducted an analysis of help seeking behavior in youth following the release of Netflix series 13 Reasons Why (Season 1). The found elevated search engine volume for terms indicating an increase in suicidal thoughts in the days following the release of the show as well as a significant but momentary rise in Crisis Text Line conversation volume. Researchers; Laura K. Thompson, Kurt Michael, Margaret M. Sugg; Prev Med Rep., 2019. Click here to read more!
- Researchers investigated the relationship between geospatial suicide clusters in the US and service use, and emergency responses initiated by Crisis Text Line. Researchers: Mark E. Larsen, Michelle Torok, Kit Huckvale, Bilal Reda, Sofian Berrouiguet, Helen Christensen; 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2019. Click here to read more!
- Researchers conducted a geographical exploration of help seeking behavior in adolescents using Crisis Text Line data. Findings indicated that increased rates of text-based help-seeking occurred in counties with higher mean household incomes, higher divorce rates, and lower residential stability. Rurality was the strongest predictor for low rates of help-seeking. This finding is particularly concerning in light of elevated rates of suicide among rural counties. The findings suggest that rural communities, particularly those with low support-seeking behavior and comparatively high suicide rates, should be the target of future research and outreach. Researchers: Laura K. Thompson, Margaret M. Sugg, Jennifer R. Runkle, Social Science & Medicine, 2018. Click here to read more!
- Researchers developed a set of novel computational discourse analysis methods to measure how various linguistic aspects of conversations are correlated with crisis conversation outcomes. Findings led to actionable crisis conversation strategies that are associated with better conservation outcomes. Researchers: Tim Althoff, Kevin Clark, Jure Leskovec; Association for Computational Linguistics, 2016. Click here to read more!
- Researchers developed “Fathom”, a system that provides topic modeling of conversations and graphical visualization of topic distributions, updated in real time. They developed a mixed-initiative paradigm to train coherent topic and word distributions and used them to power real-time visualizations aimed at reducing Crisis Counselor cognitive overload. Researchers: Karthik Dinakar, Jackie Chen, Henry Lieberman, Rosalind W. Picard; MIT Media Lab, 2015. Click here to read more!