Conversation

Pediatric COVID-19 Cases in Counties...

2) Take, e.g., cdc.gov/mmwr/volumes/7. There are two big problems with how this study is presented.

cdc.gov

This report describes pediatric COVID-19 cases in counties with and without school mask requirements.

168

3) First, the highlight is that mask mandates decreased COVID in schools, P<0.001. Which, sure, it would be pretty weird if they didn't decrease COVID *at all*. But--everything's a tradeoff. How much impact do they have?

215

4) Well, there's not much discussion of that! Because the core focus is that there's a *statistically significant* result, rather than how *important* that result is. That's stupid. The more important question is how large the effect is, not how large the study was!

373

5) In this case, it seems like the mandates decreased COVID in schools by ~20 cases per 100,000 children per week. That's ~0.02%/week, or ~1%/year. Would you rather: a) wear a mask for 100 years b) get COVID once I'll let you judge how the benefits compare to the costs.

572

11:41 PM · Feb 15, 2022Twitter Web App

6) But there's a second problem--it's not clear that those results are controlling for confounders. Now, the study did look at confounding variables! It's just, they first produced the chart, and *then* re-ran the numbers controlling. Or, at least, they imply so.

Replying to

7) After controlling for "age, race and ethnicity, pediatric COVID-19 vaccination rate, COVID-19 community transmission, population density, social vulnerability index score, COVID-19 community vulnerability index score, percentage uninsured, and percentage living in poverty":

142

8) "school mask requirements remained associated with lower daily case rates of pediatric COVID-19 (β = −1.31; 95% confidence interval = −1.51 to −1.11) (p<0.001)." Without controlling, it's ~1%/year. After controlling, it's... P<0.001.

160

9) As far as I can tell, they don't actually say how large the effect was after controlling for confounding variables. They just clarify that it's still statistically significant--it could be massive, or it could be essentially 0. They don't report it!

259