IMO this looks to by a symptom of lack of testing. Overall the ratio of tests coming back positive is unchanged:<p><a href="https://raw.githubusercontent.com/lettergram/covid19-analysis/master/graphs/us-test-ratios.png" rel="nofollow">https://raw.githubusercontent.com/lettergram/covid19-analysi...</a><p>Even though we are doing more tests. Meaning the reduction in speed-of-spread likely has more to do with lack of testing.<p>That being said, I’m sure spread is being reduced, we just don’t know to what extent because we don’t have enough effective testing.<p>It’s also possible (probably likely) our tests are not exceptionally accurate. It’s the best we have, but given all of this is <6 months old the false positive/negative rate can be high.<p>Generally, Testing isn’t an effective way to measure spread. Probably we should use “hospitalizations with flu symptoms“ as the best indicator.<p>You can compare testing per state and deaths here:<p>Tests: <a href="https://raw.githubusercontent.com/lettergram/covid19-analysis/master/graphs/us-tests-by-state.png" rel="nofollow">https://raw.githubusercontent.com/lettergram/covid19-analysi...</a><p>Deaths: <a href="https://raw.githubusercontent.com/lettergram/covid19-analysis/master/graphs/us-deaths-by-state.png" rel="nofollow">https://raw.githubusercontent.com/lettergram/covid19-analysi...</a><p>Overall repo: <a href="https://github.com/lettergram/covid19-analysis" rel="nofollow">https://github.com/lettergram/covid19-analysis</a>
Here's an alternative set of per-US-state Rt estimates from the London School of Hygiene & Tropical Medicine, with somewhat different results: <a href="https://epiforecasts.io/covid/posts/national/united-states/" rel="nofollow">https://epiforecasts.io/covid/posts/national/united-states/</a><p>They believe there's insufficient data to estimate Rt <i>today</i>, though, due to lags in symptoms/diagnosis/testing/reporting. So their estimates lag about 10 days, with the x-axis for the Rt-over-time graphs currently running February 17 through April 7.<p>They also have per-country estimates: <a href="https://epiforecasts.io/covid/posts/global/" rel="nofollow">https://epiforecasts.io/covid/posts/global/</a>
Great looking site. For those who don't know/realize - this is the work of Kevin Systrom, Mike Krieger (and team?) - co-founders of Instagram.<p>Statistically speaking - love that it has error bars and timeseries. Makes it infinitely more informative to track relative to news, events, and mitigation strategies and have a sense of the confidence of the estimate. The provided Jupyter notebook is a great resource.<p>Alphabetically speaking - I enjoyed the puzzle of figuring out how the states were sorted in the timeseries list (e.g. why was Alaska before Alabama, Iowa before Idaho, etc.). Turns out it's alphabetical by the two letter state abbreviation which is not shown - just a fun observation :)
This really illustrates that the parameter in question is not just a characteristic of the virus, but also of the environment.<p>I think a lot of people forget that if try to return to normal activity this is just going to shoot back up. I have yet to see an exit strategy other than "modulate social distancing such that the health care system is near capacity, we can gradually open up as more people are immune due to prior infection."
It's wild that California has been locked down for a full month and the reproduction number is still hovering around 1 and even went up a couple of days ago.<p>If that's actually true, we shouldn't expect to see a quick decrease in cases and we probably aren't on track to end the lockdown any time before at least June. I can't help but wondering if the unlimited outside time under the guise of "exercise", but which people are clearly taking advantage of by going to hang out with friends in the park, has anything to do with it. Or perhaps it's the lack of masks in indoor areas like grocery stores that has prevented a bigger drop (in SF, only starting today has that finally changed).<p>For all that California has been praised for acting quickly, it still feels like we've really been slow playing this lockdown by being so lenient, and without really having gained anything. And it doesn't really seem like there's any urgency anymore around continuing to increase testing, even though we know that's needed to open back up. Now we're going to be stuck inside for weeks or months longer than we would have needed to be if we had just copied what successful countries (Taiwan, South Korea) did from the beginning.
Great visualizations - I'd love to see this done with hospitalizations though as case count overestimates r given increasing testing. (Granted hospitalizations are also breaking down recently due to recent disproportionate nursing home infections)<p>E.g. CA is unlikely to have had an r above 2 even at beginning of March, but this is calculating close to 3. (Source: <a href="https://www.medrxiv.org/content/10.1101/2020.04.12.20062943v1.full.pdf" rel="nofollow">https://www.medrxiv.org/content/10.1101/2020.04.12.20062943v...</a>).
They list the following known issues:<p>1) Changes in testing will affect numbers.<p>2) The delay between infection and testing is ignored so actual Rt values are delayed by some amount.<p>Both are huge actual issues.
The ramp-up in testing has an effect on these observations (thinking specifically about the early spikes in many places):<p>> Absolute testing levels should not affect this algorithm much, but a fast ramp or decline in testing will affect numbers.
Does anybody know if the testing numbers are per patient or per test? I was trying to run some bayes simulations yesterday and while debating it we couldn't figure out anything useful, because the statistics we were able to find are already rolled up and the methodology is hidden.
Great job! I live in HI (top 4 states according to the website), it looks like they are gonna start loosening up restrictions in a couple of weeks. It will be interesting to see how Rt evolves after that. There's a couple of states where Rt started increasing again after some time (e.g. WA).
This could also be an interesting data point for anyone who needs to pick a location for establishing a new business.<p>Lesser Rt number <i>might</i> mean safer operating environment and healthier employees.