Trajectories of Suicide Mortality in US Counties from 2000-2017 and the Influence of Selective Migration

Emma Gause | 2020

Advisor: Ali Rowhani-Rahbar


Objectives: To identify five distinct latent trajectory classes of suicide mortality among US Counties from 2000 to 2017 and to compare these trajectories to each county’s cumulative net migration over the study period to assess the influence of population change due to migration as a potential measure of economic opportunity. Methods: The annual county suicide mortality rates were prepped using a spatial Bayesian smoothing approach using Integrated Nested Laplace Approximations to reduce the influence of extreme high rates due to small numbers. Latent growth mixture modelling was employed to identify five linear mortality trajectories. The model allowed for random slopes and intercepts to allow individual county trajectories to vary around the class means. Each class was compared descriptively against three categories of cumulative net migration over the study period: persistent population loss due to migration, net even migration, and persistent population gain due to migration. Results: All five classes of latent trajectories displayed a positive slope indicating rising suicide mortality rates in all classes of counties from 2000 to 2017. While two classes exhibited greater slopes, the classes seemed to be more differentiated by their intercepts – the suicide mortality rate in 2000. The proportion of counties in each of the three migration categories differed between the five classes. However, the results of the Lo-Mendell-Rubin test failed to reject the null hypothesis that five classes were more appropriate than four classes, calling into question the validity of the individual county classifications. Conclusions: As the five-class trajectory model was not found to be statistically significant, further research should be done to explore different numbers of classes as well as higher order polynomial slopes to identify the best fit for the data. However, given the similarity in the five class slopes it may be possible that county suicide mortality rates exist on a continuum rather than clustered within distinct classes. Lastly, while the latent trajectories seemed to be marginally associated with cumulative net migration, it is possible that aggregating migration counts across the study period masked local temporal variation that should be explored in a more nuanced manner.