In our course, one of the rules that we stress for detecting misinformation is that if something seems too good or too bad to be true, it probably is and itโs time to dig deeper. Here @drjenndowd does that digging. slate.com/technology/202โฆ via @slate
A proper thread later, but it was a great experience to write about the challenges of credible data analysis in the time of COVID-19 for @FutureTenseNow@Slate #epitwitter @DearPandemic #Sturgis #infodemic #poptwitter
But @drjenndowd breaks the problems with this study down in much better detail.
TLDR: Sturgis has caused a bunch of cases -- but not this many. We probably won't ever know the exact number. But, again, it's not this many.
No, the Sturgis rally did not cause 19 percent of all the COVID cases in America. The study that claims that is not complete "fiction" (as the SD governor says), but it's also very, very unlikely to be accurate.
Excellent diagnosis and critique of the Sturgis Rally #COVID19 spread paper grabbing headlines. Clear description of some complicated elements of research design. Terrific job @drjenndowd! #pophealth #medsoc
. @drjenndowd nails many of the problems with that Sturgis paper. On @chrislhayes the other day, @FriedsonAndrew said "if you buy [our assumptions] then you buy the analysis." This is a strong & clear explanation of why I don't.
We need more fact-based journalism that punctures confirmation bias. I'm always proud when @Slate does it.
A careful takedown of the infection estimates from Sturgis, by @drjenndowd and @FutureTenseNow.
Great piece by @drjenndowd. The models in recent #Sturgis rally paper require many problematic assumptions & yield estimates that are widely variable to the point of being useless.
Researchers should perform a reality check before interpreting models.
The parallel trends assumption discussed on Slate. Hard to see how the authors of this study could possibly have justified it. Counties that sent a lot of bikers to Sturgis on similar pre-rally case count trajectories as those that didnโt?!
Another criticism of that unlikely Sturgis study. Not sure the "parallel trends" point really works though; it's a synthetic control study where the comparison counties are selected specifically *because* they had parallel trends before the rally.