Shop Talk

What's a Safe Distance?

Can you count the number of times in the past month that you’ve heard someone ask, “Did you get the Pfizer or the Moderna?” Before COVID, I was used to other dichotomous questions§. Red or white? Coffee or tea? Your place or mine? But COVID has unleashed a barrage of new questions and associated choices. One mask or two? In person or remote? And the latest, 6 feet or 3?

All kidding aside, the question of 6 feet or 3 is important. Local school boards want to know if it’s prudent to reopen classrooms when space allows only 3 feet of distance between pupils. Parents want to know if it’s wise to send their kids back under these conditions. Businesses are doing their own risk-benefit analyses as they decide when and how to bring employees back to the office.

Lucky for us, the literature has something to say on this matter. 

Last week, Polly van den Berg and her colleagues released a retrospective cohort study on the effectiveness of 3 versus 6 feet of distancing for controlling the spread of COVID-19 in schools. Unlike some other scientific reports in the age of COVID, I did not read about these data via press release. The manuscript underwent peer review and has been accepted for publication in Clinical Infectious Disease.

Below, we summarize the study, share the results, and discuss the value of a retrospective cohort study. 

van den Berg’s study measured the incidence of COVID-19 from September, 2020 to January, 2021 in a sample of 636,726 students and staff in Massachusetts schools, grades K – 12. All participants attended in-person school. Some were in a setting with a 6 foot distancing protocol; others were in schools with 3 foot distancing. Depending on the month, the COVID-19 case rates per 100,000 ranged from 5 to 55 for staff and 5 to 18 for students. (These are ballpark values that I pulled from one of their figures.) Case rates were not different between the 3 and 6 foot groups. So, we can conclude that a 3 foot distancing protocol in schools confers no additional risk over a 6 foot protocol for contracting COVID-19§.

To date, all our blogs about COVID-19 have involved randomized controlled trials, case reports, or mechanism-based reasoning. What’s exciting about today’s piece is that it involves a retrospective cohort study, which gives us an opportunity to discuss a new study design.

What’s a retrospective cohort study? 

As the name suggests, this study type is done retrospectively, which means that the data and the processes used to generate it existed before the study was conceived. Indeed, the Massachusetts schools did not select 3 or 6 foot protocols because a team of researchers asked them to. Likely, these selections were made for practical reasons. How big are our classrooms? How many kids must we accommodate, and how urgent is the need to get them back in school? Likewise, the COVID-19 incidence data were not collected for the purpose of this study. The Department of Elementary and Secondary Education was already collecting this information, and the authors plucked it from their database.

The main limitation of a retrospective study is lack of experimental control. van den Berg’s team couldn’t do anything to ensure that the 3 and 6 foot groups were matched in size or for demographic factors that might affect susceptibility to COVID-19. They could not specify who got tested or what counted as a case of COVID-19. They could not oversee distancing to make sure schools adhered to their protocol. Hence, this and all retrospective studies have to deal with a bunch of extraneous factors that may have influenced the result.

But, look at what we get in exchange for accepting some confound! A ton of data (>600,000 participants) sitting in a database and a natural experiment (3 versus 6 feet) that would have been impossible to orchestrate for the sake of research. Can you imagine randomizing your child to the 3 foot group just for the sake of learning whether it was riskier than 6 feet? 

Now, let’s talk about the word cohort. A cohort study is one in which you start with at least 2 groups of people who are at risk for developing a disease, but they have not yet contracted it. Each group, or cohort, is hypothesized to have a different level of risk. The groups are followed over time, and the case rate is tallied for each group. If it is true that one group was at higher risk than the other, you should see more cases in that group.

In the van der Berg study, there were 2 cohorts; 3 feet and 6 feet. The hypothesis was that 3 feet is riskier than 6 feet. Had the hypothesis been supported, we would have seen a higher COVID-19 case rate in the 3 foot group as compared to the 6 foot group. Instead, the data showed no difference in case rates between groups, which is why we can conclude that 3 feet is no riskier than 6.

Cohorts are a go-to design when the outcome of interest is something bad, like catching COVID-19. Smoking is another example. If you’re worried that it causes cancer, you cannot randomize participants to 2 groups; smoking and no smoking. But you can find people who have chosen to smoke and compare their cancer rates to non-smokers.

Thanks for reading. As always you can learn more from the materials below. If you are so inclined, share this piece with a friend and encourage them to subscribe. All they need to do is e-mail us at and say, “Add me to the list.”

Until next time, keep a safe distance. 


§ A dichotomous question is one with only 2 possible answers. I only have coffee and tea. Would you like coffee or tea? In research we often use dichotomous variables to split participants into 2 groups. Examples include dead/alive, pregnant/not pregnant, vaccinated/not vaccinated. You must pick one, and you cannot be both. 

§§ Distancing was not the only mitigation strategy in place. Masking, ventilation, symptom screening, and other protocols were also used.

Learn more: 

Read the full article from Polly van den Berg and her team at

Read the CDC’s guidelines for distancing in schools at

Mar 19, 2021

If you appreciate what we’re doing, please consider supporting our work -