Information-Accessing Behavior during Zika Virus Outbreak, United States, 2016

We used latent class analysis to examine Zika virus–related information-accessing behavior of US residents during the 2016 international outbreak. We characterized 3 classes of information-accessing behavior patterns: universalists, media seekers, and passive recipients. Understanding these patterns is crucial to planning risk communication during an emerging health threat.


Model Selection
There is some debate in best practices for LCA model selection, especially when applying weighted population estimates when likelihood ratio tests may not be appropriately run since maximum likelihood estimates are not possible (12). In accordance with the best practices set out by Nylund, Asparaouhov, & Muthen (2007), several criteria were used to determine the optimal number of classes (13). The criteria applied here were: 1. Akaike and Bayesian information criteria (AIC and BIC) (14); 2. Lo-Mendell-Rubin adjusted likelihood ratio test; 3. Entropy; 4. the relative size of classes in each model; 5. substantive interpretability; 6. and replication of the LCA solution in all three samples.

Results
LCA results suggested a replicable three-class solution of information users in the population, with classes distinguished by the number of sources accessed. Appendix Tables 1-3 demonstrate the selection criteria used to compare 2-6 classes and reflect the three classes solution had the best goodness of fit at each time point. Results as to the proportion of the population in each class and accessing each source by time point are below in Appendix Tables 4-6.

Sample 1: Spring 2016
The proportion of the population in each of the three-classes is shown in Appendix Table   1. The average latent class probability, an indicator of membership within a latent class, measures how certain an individual is to be in one class compared to another, was high-0.944, 0.893, and 0.906 respectively. Within Class 1, the probability of getting information from print news was 0.845, broadcast news was 0.814, social media was 0.564, doctor was 0.667, government was 0.645, and family/friends was 0.729. Within Class 2, the probability of getting information from print news was 0.844, broadcast news was 0.675, social media was 1.00, doctor was 0.00, government was 0.035, and family/friends was 0.390. Within Class 3, the probability of getting information from print news was 0.597, broadcast news was 0.786, social media was 0.004, doctor was 0.047, government was 0.108, and family/friends was 0.164.

Sample 2: Summer 2016
The proportion of the population in each of the three-classes is shown in Appendix Table   2. From time 1 to time 2, the proportion of the population in each class shifted. Class 1, people who sought information from many sources, was 13.8% of the population, Class 2, those who primarily sought information from mass media and social media were 51.5% of the population, and Class 3, the least active information seekers, was 34.7% of the population. The average latent class probability, an indicator of membership within a latent class, measures how certain an individual is to be in one class compared to another, was still high-0.890, 0.792, and 0.947 respectively. For Class 1, the probability of getting information from print news was 0.881, broadcast news was 0.735, social media was 0.635, doctor was 0.605, government was 0.518, and family/friends was 0.806. Within Class 2, the probability of getting information from print news was 1.00, broadcast news was 0.818, social media was 0.376, doctor was 0.00, government was 0.090, and family/friends was 0.277. Within Class 3, the probability of getting information from print news was 0.193, broadcast news was 0.665, social media was 0.169, doctor was 0.105, government was 0.073, and family/friends was 0.227.

Sample 3: Fall 2016
The proportion of the population in each of the three-classes is shown in Appendix Table   3. The proportion of the population in each class was similar to Sample 2. The average latent class probability, an indicator of membership within a latent class, measures how certain an individual is to be in one class compared to another, was also high-0.852, 0.860, 0.872respectively. For Class 1, the probability of getting information from print news was 0.818, broadcast news was 0.834, social media was 0.764, doctor was 0.564, government was 0.482, and family/friends was 0.789. Within Class 2, the probability of getting information from print news was 0.871, broadcast news was 0.816, social media was 0.414, doctor was 0.024, government was 0.151, and family/friends was 0.171. Within Class 3, the probability of getting information from print news was 0.000, broadcast news was 0.463, social media was 0.131, doctor was 0.044, government was 0.014, and family/friends was 0.118.