Sean J. Taylor

Learning About At-Risk Veterans Using Online Network Surveys

03 Feb 2016

On December 12th last year, Carlos Diuk, Akos Lada, Alex Peysakhovich, and I hacked on Open Data and Innovations for Suicide Prevention, specifically working on addressing the issue for United States Veterans (of whom 22 commit suicide everyday). The hackathon was sponsored by the Whitehouse and the US Dept. of Veterans Affairs.

What We Did

We decided to treat at-risk veterans (those with depression, PTSD, or suicidal thoughts) as a “hidden” (i.e. hard to access and survey) population and designed a network-survey where we ask about friends who are veterans instead of the respondent, yielding information on network neighbors of the seed nodes we survey.  We’d like to thank Dennis Feehan, who’s an expert on this methodology, for his help in designing this strategy.

We first designed a survey, and then created a Facebook Page to create posts linking to the survey. We ran Facebook ads targeting those who were employees of the US Navy, Army, and Marine Corp, as well as those living near military bases and interested in those topics.

Within four hours, we had 400 brand new survey responses to analyze. We found patterns that seem to replicate existing understanding of incidences of risk factors in veterans. We also found some evidence for self-reporting bias in PTSD and suicidal thoughts, indicating that existing survey data may understate the problem. You can learn more from viewing our presentation slides:

What’s Cool About This

  1. Using Facebook ads to recruit survey respondents yielded excellent cost per respondent (about 40 cents).
  2. Survey instruments where you ask about the respondents friends instead of the respondent helps you find out about people who are hard to reach, such as people at-risk of mental health problems.
  3. Asking about friends has the added benefit of potentially yielding a less-biased signal when there may be response bias. 

Next Steps

We’re actively collaborating the San Francisco VA to get this method into production:

  • Iterate on survey design and recruitment procedures (optimizing ads targeting and copy).
  • Recruit for this survey regularly to detect longitudinal trends in the incidence of mental health risk factors for veterans.
  • Work on survey-weighting to adjust for biased survey population. Produce estimates of population levels instead of rates.
  • Potentially direct survey respondents to resources for themselves and the friends they identify as at-risk.