Assessing the Predictive Power of Online Social Media to Analyze COVID-19 Outbreaks in the 50 U.S. States

Round 1

Reviewer 1 Report

Very good article. Minor problems I see:
1. the real location of the tweets may be problematic - did the authors use any authentication mechanisms?
2. Pearson correlations did not come out very well - what could it mean?
3) As for the maps on p. 7 - wouldn't the basic correlation here be the number of inhabitants of a given state with the number of tweets and posts (the more numerous the state - the more?)

Author Response

please see attached response letter

Author Response File: Author Response.pdf

Reviewer 2 Report

For a paper of the presented kind, a non-trivial working hypothesis, its quantitative description and statistical evaluation is needed. I missed all these parts in the current manuscript. Thus, I would suggest that more work is invested into the proposed topic of research to publish more information per publication.

Comments for author File: Comments.pdf

Author Response

please see attached response letter

Author Response File: Author Response.pdf

Reviewer 3 Report

The topic addressed in this paper is appropriate to be published in this journal.

The work is well done and the presentation is good.

Only two recommendations:

• To add more background and relevant references (perhaps including a related work section).
• To include the organization of the paper at the end of the introduction section.

Author Response

please see attached response letter

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The improvement of the manuscript are very incremental. The introduction still contains a lot of unsupported statements.

The explicit statement of the working hypothesis is a substantial improvement, it makes following the authors' logic much easier. As formulated however it raises other questions. Authors have demonstrated *correlations* between tweets and COVID cases; however, their hypothesis is stated that "there will be statistically significant predictive power in the Google
search and tweeting behavior of a U.S. polity such as a state or a city", which is quite another issue. Correlation does not mean causation, and without knowing causation chains predictions are on a shaky ground, unless one has huge and various data (e.g. on *different* pandemics), which authors do not have. For COVID, correlation might be caused by high transmission rate of the virus, hence high spread rate, hence high attracted attention, hence increased tweets. And the same high transmission rate caused high incidence of COVID cases in the US. It is not clear if another virus, with different epidemiological characteristics (say different R0) will exhibit the same correlation, and attention (tweets) can be captured for reasons other than rapid spread. Incidentally, significant correlations (Table S1) are observed in Twitter, but not for Google Trends, which raises question what predictors are good for the pandemic prognosis and why. My overall conclusion is that the hypothesis about social media having "statistically significant predictive power in the Google search and tweeting behavior" is not supported by the provided evidence.

Author Response

Please see attached response letter

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

The 3rd revision of the manuscript now claims to have demonstrated correlation between the social media behavior and the COVID infection rates, which it indeed did. So the paper claims are now justified. Of course I find it quite expected that such correlations exist, so there is little surprise in the finding. It is quite regrettable that authors "make no claims about causality between actual infection rate and size of the lag between actual Covid-19 outbreak and twitter and Google search behavior"; if the research could demonstrate insights into such causality, it would have a good chance to become a predictive tool as the authors originally intended (and I would expect some surprises there, since human behavior is sometimes quite irrational at the first sight but often predictable), and would make the paper so much more interesting.

Author Response

please see attached response letter

Author Response File: Author Response.pdf