This proposal seeks to address a key challenge in public health : the ongoing and fine-grained
measurement of population mental health. Currently, in the largest population surveys, measurement of
mental health is limited in time to annual estimates, in space to predominantly metropolitan areas and in
scope to single questions about "mental health" or "depression". Through interdisciplinary work, we
propose to advance approaches to language-based analysis of social media to measure mental health,
sub-annually, ·at the county level and broaden the scope beyond depression to anxiety, stress, as well as to
protective mental health factors (such as healthy social relationships). This will provide the research
community with a much richer, timely and localized picture of population mental health.
The language of social media has been shown to be a flexible source of information about population
behaviors, thoughts and feelings It is available with high spatial and temporal resolution, suggesting great
potential for the study and monitoring of population mental health. However, approaches for tracking
psychological states across communities on social media were not developed with consideration for spatial
and temporal confounds or to fully leverage the multi-level structure (and sample sizes) of the data.
Proposed work will develop multi-level methods to control for spatial correlation and community
socioeconomic covariance to increase statistical power and the accuracy of measurement. The increased
power will also better enable quasi-experimental designs from epidemiology which will be combined with
the Twitter-based estimates to track the impact of policy and socioeconomic shocks on mental health.
The work in this proposal could significantly transform both research in population mental health and the
ability to apply and track the efficacy of policy to improve public health , It will allow researchers to observe
temporal changes in population mental health quarterly and for counties, which provides the measurement
infrastructure to observe changes in response to natural experiments such as economic shocks and policy
interventions. This will be possible in near real-time, without the reporting lag of a few years as in current
survey methodologies. The ongoing measurement will help identify areas of greatest need and may help
prioritize resource allocation. The improved quasi-experimental modeling of mental health determinants
may inform policy interventions, and the ongoing monitoring can establish evidence of their efficacy, In tum,
the burden of mental unhealth on society may be substantively reduced in the long term.