Twitter | Search | |
This is the legacy version of twitter.com. We will be shutting it down on 15 December 2020. Please switch to a supported browser or device. You can see a list of supported browsers in our Help Center.
John Burn-Murdoch
Stories, stats & scatterplots for | Daily updates of the coronavirus trajectory tracker | john.burn-murdoch@ft.com |
51,513
Tweets
4,191
Following
315,500
Followers
Tweets
John Burn-Murdoch 10h
Hi Paul, thanks for sharing my thread, though worth noting that I emphasised "in the initial outbreak phase". Log scales are useful when dealing with exponential growth or decay away from or towards zero. That's no longer the case when plotting cumulative numbers months in.
Reply Retweet Like
John Burn-Murdoch 10h
Replying to @PPudney
Yep, my daily run is as much about problem-solving as it is about exercise!
Reply Retweet Like
John Burn-Murdoch 10h
Replying to @TimBrock_DtD
Yup! • Jot down all current thoughts on the topic so I a) have them ready for tomorrow and b) no longer have them bouncing around my head • Worry-free sleep • Pick up where I left off, freshly energised
Reply Retweet Like
John Burn-Murdoch 10h
Extremely this. A couple of years ago I was struggling to work out how to build a particularly thorny predictive model. Went for a 13 mile hike to figure it out, and nailed it. Walking is amazing for problem-solving.
Reply Retweet Like
John Burn-Murdoch Nov 27
Replying to @joel_bkr @youyanggu
In a word, nope: Government also wrote up their decision-making for every area, and they quote observed case rates from the recent past (also 👋!)
Reply Retweet Like
John Burn-Murdoch Nov 27
You mean Worldometers? 😉
Reply Retweet Like
John Burn-Murdoch Nov 27
Totally agree. I was happy with it as an approx because a) government’s own written explanations mention each of the different metrics interchangeably, and b) what would get closest to gov decision-making is a different approach to the geography, more than to the metrics
Reply Retweet Like
John Burn-Murdoch Nov 27
Incidentally that’s a big shift I’ve made over the last few years: Pivoting from "what’s the graphic that I [and people like me] would pore over for hours?" (most of my work before, say, 2016) to "what’s the graphic that thousands of people will get use out of in one minute?"
Reply Retweet Like
John Burn-Murdoch Nov 27
I could be totally wrong here, but my instinct is that this is a classic case where the chart(s) that we (as data people) would want to explore, is not the chart(s) that a mass audience on social media wants.
Reply Retweet Like
John Burn-Murdoch Nov 27
This is a great discussion 😀 In this particular case, I only made the composite once I realised that making the six individual ones would a) be very visually repetitive in a way that subtracts value, and b) feel like information overload to a lot of the audience.
Reply Retweet Like
John Burn-Murdoch Nov 27
FT also in the midst of our very own transition period :-)
Reply Retweet Like
John Burn-Murdoch Nov 27
I haven’t, but I’ve shared all the raw data behind it so anyone can do that, or could do another thing that I think would get closer to the government’s actual process: Give each area a risk score that is the average of its own score and those of its neighbours.
Reply Retweet Like
John Burn-Murdoch Nov 27
Replying to @tombarton
(2/2) London more so than any other area in England can’t be split across tiers, because you walk across borough boundaries without even realising it. If London had been put wholly into T3, 3.3m live in boroughs where Covid risk is below any T3 place except Stratford...
Reply Retweet Like
John Burn-Murdoch Nov 27
Replying to @tombarton
There’s almost certainly an element of that, but: (1/2) Before national lockdown, London went into a *higher* tier than the data suggested, based on local officials being extra cautious, so it’s not clear that selfishness has always been the driver.
Reply Retweet Like
John Burn-Murdoch Nov 27
Replying to @GtGwentr @Grace_Kite
Hi Gwen, I first standardised each input variable to a z-score, then averaged the standardised scores in each area, then rescaled the result from 0 to 100.
Reply Retweet Like
John Burn-Murdoch Nov 27
See link to data at end of thread 🙂
Reply Retweet Like
John Burn-Murdoch Nov 27
Replying to @darcyrossiter
• Cases growth rate: cases divided by cases a week earlier (so < 1 = falling, > 1 = rising) • Positivity rate: cases last week divided by tests last week • Hospital occupancy: Covid patients in hospital per 100k people in area • Hosp admissions: as above but new admissions
Reply Retweet Like
John Burn-Murdoch Nov 27
Replying to @PeterBale
Tend to agree
Reply Retweet Like
John Burn-Murdoch Nov 27
Replying to @_PClough
Yes
Reply Retweet Like
John Burn-Murdoch Nov 27
Replying to @Grace_Kite
Ah sorry, I accidentally got rid of the "Circle size represents population" subhed in the process of updating it this morning. Y-axis is the combined score, x-axis is just sorting the data left-to-right based on that score.
Reply Retweet Like