CMU's COVIDcast Is Tracking Face Mask Use Across The Nation

Oct 16, 2020

New data from Carnegie Mellon University allows users to see the percentage of people who wear masks in their county. COVIDcast, which already localized collected data about coronavirus activity, began asking survey respondents about face coverings and test access last month. 

According to the data, which comes from surveys distributed by Facebook, 93 percent of responding Allegheny County residents claim they wear a mask most or all of the time while in public. In the first few days of October, between 83 and 87 percent of responders in Westmoreland, Beaver, Butler and Washington Counties reported wearing masks.

Delphi, a Carnegie Mellon University research group, has been collecting data about COVID-19 symptoms, test results and social activities since April. The data is updated daily and not currently available anywhere else.

The surveys appear in Facebook users’ newsfeeds. Respondents are asked a series of questions about how often respondents leave their home and for how long, whether they’ve had a coronavirus-related doctors’ visit and if respondents had participated in social activities like going to a bar or riding public transit.

The project is meant to forecast the spread of COVID-19. Delphi has been using a similar method to forecast influenza activity for several years. The Centers for Disease Control and Prevention designated the group as one of two National Centers of Excellence for Influenza Forecasting. The CDC requested Delphi adapt their forecasting efforts to cover COVID-19 last spring.

Alex Reinhart, an assistant teaching professor in the Department of Statistics and Data Science and a member of the Delphi group, hopes officials will use the new data about mask use to create better informed public health policies.

"Our survey doesn't replace official public health reporting on COVID testing and case counts, but it can provide insights not available any other way," Reinhart said in a statement announcing the new metrics. "By providing these signals to the public, we hope to provide researchers, public health officials and journalists the information they need to form a more complete picture of the pandemic."

The vast majority of health experts agree that wearing masks in public serves a vital role in preventing the spread of the coronavirus. Reinhart said as the presence of masks grows in the U.S., it’s valuable to see if a correlation exists between a county’s mask usage and rate of COVID-19 cases. Data from the first few weeks of October shows the states with a lower number of people reporting wearing masks are the same states that have higher numbers of COVID-19 cases.

But, Reinhart is quick to note that correlation doesn’t always mean causation. “But sometimes correlation can certainly suggest something that is worth further investigation,” he said. Health officials could look at the same data and see an opportunity to educate the public about the impact of wearing face masks, according to Reinhart.

“So that public health authorities could understand ‘how do we best get people to wear masks?’ Are there groups that are not that we should be doing outreach to?”

The data also showed the absence of an upward trend in the number of people wearing masks, something that surprised Reinhart.

Also new to the survey are questions about access to testing. COVIDcast’s data collection about testing provides a uniform source to compare testing data by state. Such a source didn’t exist before because the criteria for reporting testing varies widely among states.

After reviewing some statewide data about testing access, Reinhart found that some respondents wanted to get tested for COVID-19, but didn’t know where to go. “So it seemed like an information problem that perhaps one could solve with advertising or public service announcements or something like that,” he said.

Researchers will consider adding new questions to its survey as the pandemic continues. One example, according to Reinhart, is a new section of questions about a vaccine.

Delphi is working with Google and Facebook to manage the surveys. Last month, Google gave Delphi $1 million and a dozen fellows to help process and analyze the forecasting data. Facebook targets a cross section of users in an area and distributes the surveys, but cannot see a user’s answers. Similarly, Delphi can’t see a respondent’s Facebook page, just information provided in the survey. That relationship ensures privacy for the respondent, according to Reinhart.

So what if a respondent provides inaccurate answers or outright lies on the survey? Reinhart said while trolling is possible, the data shows most people who complete the 30 question survey do so earnestly. Reinhart said respondents often send clarifications and corrections to the email address attached to the survey.  

“We get so many people who follow up and say ‘I answered this, but I need to clarify…’ They want to share their story because it’s something that has happened to them and it’s such a major event nationwide,” he said.

Reinhart has also compared answers to questions about whether the respondent knows someone with COVID-19 against confirmed case counts from a state and found the data correlates, “startlingly well.”

More than 10 million respondents have submitted surveys since the spring.