Our new study confirms the tragic consequences of delaying the UK's first lockdown.
If it started 1 week earlier, there would have been 20k-35k fewer deaths. The required duration, for the same exit incidence, would also have halved from 69 to 35 days
1/6
https://t.co/yldNj2HBJS
Predictive #AI systems often fail because of “dataset shift” or lack of “domain adaptation”.
Without new data to retrain the algorithm, an explicitly causal approach is needed: Enter Counterfactual Prediction.
@barbradickerman and colleagues explain here
https://t.co/Wk0Og8ER1W
For no particular reason ...
A Sunday thread highlighting just a tiny sample of the brilliant women whose tireless work has shaped my understanding of Covid-19
I think the division of modeling tasks into “descriptive”, “predictive” and “causal (explanation)” is too narrow. Based on experience we also need “to finish thesis”, “uncontrolled chaos” and “want to make a point”
@PWGTennant One of these questions is 'disingenuous, snarky, and unconstructively provocative', while the other is (apparently) perfectly benign. Personally, I struggle to see the distinction.
YOU HAVE NO AUTHORITY HERE JACKIE WEAVER.
Please set aside 20mins today to watch the best bits of the Handforth Parish Council Planning and Environment Committee.
The best British comedy in decades.
https://t.co/gN9gEGE77n
'Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: review & recommendations'
Final version in @IJEeditorial!
#EpiTwitter#StatsTwitter#CausalTwitter#SER2020
https://t.co/2elISUpS0M