To avoid look-ahead bias/confounding variables, here is the same analysis but using 2016 margin of victory as the predictor. Similar results.
This basically says that 2016 election results is a better predictor of the severity of the fall wave than intervention levels in 2020!
For completeness, here is the correlation between political leaning and intervention levels for all 50 states (R^2 = 0.30).
In the coming months/years, you will see lots of published research incorrectly attribute causal relationship between level of intervention and severity of the fall wave.
But the reality is that political leaning is a heavy confounding variable, and we must be cognizant of this.
Rather than judging a state's pandemic response in a vacuum based on raw cases/deaths, we should instead look at how states expect to do given the demographics.
If we were to use the plot above, then states like WV/ME/VT outperformed expectations, while SD/WI/CA underperformed.
(Of course, we probably need a more complex model with more variables to do a better "expectations" analysis)
For some more single variate correlation analysis, see my thread from last month:
Caveat: just because correlations may be "baked in" does not mean that we should just give up and do nothing.
Rather, it's important to tamper expectations. Interventions and restrictions will help, but unless we can change political beliefs, there seems to be a very real limit.
If anyone finds a stronger explanatory variable, please let me know.
I'll end here. Here's a final takeaway:
It's human nature to want to feel like we can control the future. This applies to everyone, including policy makers and public health experts.
But this pandemic has taught us that sometimes, there is a limit to what we have control over.
Update: A follower suggested incorporating county-level election results (from github.com/tonmcg/US_Coun…).
Here are the results (2016 left, 2020 right). The correlation is decent given 3000+ data points.
It'd be hard to find another explanatory variable with stronger correlation.
Update #2: Another suggestion was to look at deaths, so here it is.
Left plot is total deaths in the pandemic. Right plot is total deaths since Sep 1, 2020. No correlation on left, decent correlation on right.
Update #3: Had some good convos regarding this with @Afinetheorem, @WesPegden & others. Spatial correlation is another confounding factor. Obviously we need more than a Twitter thread to prove any causal relationships, but my main purpose was to show an interesting correlation.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
I've seen many news articles cite that "the UK variant could be the dominant strain by March". This is emphasized by @CDCDirector.
While this will likely to be the case, this should not be an automatic cause for concern. Cases could still remain contained.
Here's how: 🧵
One of @CDCgov's own models has tracked the true decline in cases quite accurately thus far.
Their projection shows that the B.1.1.7 variant will become the dominant variant in March. But interestingly... there's no fourth wave. Cases simply level out:
This conclusion is based on the following new developments over the past month:
- Remained high levels of vaccine hesitancy
- New variants that may lower vaccine efficacy
- Rollout of the J&J vaccine (efficacy ~70%)
- Delayed arrival of the children vaccine
That said, herd immunity does not have a hard threshold, and being close to herd immunity may be sufficient to prevent large outbreaks.
Our goal should not be to reach "herd immunity", but to reduce COVID-19 deaths & hospitalizations so that life can return to normal.
This seems to confirm analysis announced by White House Data Director @cyrusshahpar46 that shows that the majority of delivered first doses have already been administered.