Graduation Year


Document Type




Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Public Health

Major Professor

Ismael Hoare, Ph.D., MPH

Committee Member

Getachew A. Dagne, Ph.D.

Committee Member

Richardo Izurieta, MD, DrPH, MPH, DTM&H

Committee Member

John Licato, Ph.D.


SARS-CoV-19, public health governance structure, social distancing measures, Twitter, language modeling, machine learning, model evaluation



Since the start of the COVID-19 outbreak, the responses to the pandemic varied greatly from nation to nation. In the US, state-level variance in responses has likely contributed to the disparate COVID-19 infection-related state-level outcomes such as incidence and case fatality rates. The large variances in national and sub-national responses to COVID-19 when combined with the abundance of factors that influence disease transmission rates has created a number of important research opportunities.


In this vain, this research study aims to describe the relationship that exists between public health governance structure and public sentiment and COVID-19 control measure effectiveness in the US. This overall goal was accomplished by focusing on three aims. Aim 1 Compare the effects that state-level and county-level COVID-19 control measure stringency had on disease transmission (basic reproductive rate R0) and human mobility, by using generalized linear modeling. Aim 2 Determine if there is a relationship between public sentiment and the state-level implementation of COVID-19 control measures, by utilizing sentiment analysis, Twitter data, and regression analysis. In the preliminary research conducted along these lines, it was found that at the national-level longer duration more stringent control measures have been found to delay the onset of peak COVID-19 morbidity rates. Aim 3 Evaluate the effect of the inclusion of emojis, addition of hashtags, and modifying the threshold for classifying sentiment to determine which of the text preprocessing steps are most effective when attempting to improve pre-trained language model accuracy.


The study found that state-level social distancing measures were more effective than count-level at reducing the COVID-19 R0 in Florida. Additionally, state-level social distancing measures were also associated with a larger increase in the amount of time that Florida residents spent at home. The study failed to detect a meaningful relationship between median monthly state-level sentiment and social distancing measure stringency levels. However, it should be noted that the language model used in the study was associated with an accuracy level of 42%. Finally, the inclusion of hashtags as text proved to be the most effective method of increasing model accuracy.


The observed relationship between state-level social distancing measures and the COVID-19 R0 indicates that during responses to complex public health emergencies too much decentralization is not ideal. However, it remains unclear which public health governance structure should be in these situations. This study did identify methods of improving language model accuracy when studying sentiment in COVID-19-related Twitter data. However, the keyword searching method commonly used to identify COVID-19-related tweets is not sufficient when attempting to identify tweets about COVID-19 disease control measures. Reliable and automated methods of identifying social media posts that include COVID-19 control measures as the subject should be further studied. Finally, future research should also be conducted to evaluate the effect that the inclusion of both emojis and hashtags has on language model accuracy.

Included in

Public Health Commons