Hacker News Reader

Comment by yreg | Hacker News Reader

This dashboard is not intended to visualize computed estimates from models. It tracks the official confirmed cases, which is a very different and useful piece of data.

You are unfair by stating it is garbage and misrepresenting what it shows (confirmed cases vs. estimated cases).

View on HN

The comment above is partially true. The data sources for each country are different, and thus have different error profiles.

It's statistically incorrect to assume they can be merged and treated as a single data set.

What is more useful is to examine well bounded and vetted subsets of the data.

Once the South Korean cases become more mature in a couple weeks, I think we will have very high confidence mortality data.

Including poor data quality sources will just introduce biases.

What's garbage is the US response to it.

Italy is running 2500 tests a day.

Oregon has the capacity to run a total of about 40.

Agreed. Seems like people in Oregon choose to mostly ignore the issue

It's not just Oregon; that's just a number I saw this morning in the local paper, but I suspect it's similar everywhere. The US did not use the advance warning we had to prepare.

Georgia ran 50 tests yesterday. And that's after only 3 positive results so far. All states not equally dysfunctional.

Let's do some numbers:

Georgia, with 50 tests/day and 10,000,000 people is at 0.005 per 1K people.

Oregon, with 40 tests/day and 4,200,000 people, is at 0.0095 per 1K people.

Italy, with about 2500 tests/day and about 60,000,000 people, is doing around 0.04 per thousand people. It's about an order of magnitude more.

The point isn't a pissing match between states, it's that the entire US has not responded well so far. Hopefully things ramp up, so that we stop seeing these situations where people get exposed and have no idea what's going on.

https://www.theatlantic.com/health/archive/2020/03/how-many-...

I do not know why this is, just that it's not going well so far.

Oh, and Italy went back to a higher number. Today they ran almost 4,000 tests.

https://lab24.ilsole24ore.com/coronavirus/

More testing != useful testing.

The fact that Georgia and Oregon are testing at virtually equivalent rates is what's wrong.

I feel like ~50/day is an appropriate number for GA. I feel like ~50/day is not an appropriate number for Oregon (sandwiched in-between Washington and California).

The federal goverent should be doing a stronger job of quarterbacking this to allocate scarce resources.

If Oregon needs surge capacity of testing labs across the Midwest, then by god make that happen.

Some amount of testing likely needs to happen in every state in order to prevent panic, but the most resources should flow towards the places with the most known infections.

This is a decent article explaining how things have gone wrong in various ways:

https://www.washingtonpost.com/health/what-went-wrong-with-t...

Yesterday, I read the US had 100 confirmed cases, and had tested 500 people.

Trump fired the a big part of the CDC that deals with pandemics back in 2018 and hasn’t replaced them. I don’t think the US government will make any serious effort to contain it, despite the recent theatrics from congress.

Not testing sick people will inflate the official mortality rate. If the disease hits 70% of the population (projection from yesterday), there will be more panic than there should be. Last I heard, WHO (poor testing on average) estimated 3.4%, but in areas with thorough testing, it looks closer to 0.6%.

I've seen comparisons like that doing the rounds. They're just plain wrong. The 500 number is the number of patients that had been tested by the CDC about a week ago, excluding state and local labs, whereas the 100 confirmed cases is a newer figure and a lot of those cases were diagnosed by the very state and local lab tests that aren't being counted. From looking at an old version of the CDC web page, at the point when diagnosed cases hit 100 the CDC alone had tested about 1,500 patients. At one point the CDC pulled the number of tests from their web page to try and stop people making this error, but everyone just started comparing the old and increasingly outdated number from before they did this to the number of confirmed cases in the latest news headlines.

The stuff blaming it on Trump firing the part of the CDC that deals with pandemics is untrue too. (Though sadly, the testing might have rolled out more smoothly if he had culled the CDC a bit... their in-house facilities seem to have really screwed things up.)

The CDC stopped posting the numbers of tests they'd run a few days back. I don't know if they started again. It's not confidence inspiring.

I don't know if this has anything at all to do with firing the pandemic team, but you have to figure that didn't help, either.

Any administration has a lot of competent career professionals working for them. Events like this are the time to let them do their thing without hindrance, and without random commentary from uninformed leaders.

If I fired my team of system security engineers, then a data breach was discovered, and great harm came to my company because there was no one there familiar enough with the system to know what to do, you'd probably think it was my fault. Ya know, because I did piss poor risk management. And did not conduct due diligence with risk analysis. Especially if my company had more than enough funding to have them there... Because data breaches don't happen often, but they are BAD NEWS when they do; there's a good reason that job exists.

The WHO fact-finding mission to China reported that they ran 350,000 tests in the province of Guangdong alone (though it must be noted that its population is 110 million) ...

You are right but unfortunately almost everyone calculates the mortality rate based on confirmed cases alone, and that leads to a gross over-estimation of the dangerosity of this virus. So these visualisations are still kind of misleading in a way.

Yeah but if every virus is modelled this way then the data is representative...

Isn't that true of all flu viruses, every year? Meaning this measure is still valid to compare with other flu epidemics.

So that leaves this one at 10X or more fatalities, however you calculate it.

I am not sure. Looking at the CDC website [1], it looks like they extrapolate cases for the flu. They seem to estimate that roughly half of the people contaminated with the flu do not see a doctor (I would have guessed the numbers would be even lower, I had the flu a couple of times, it didn't even cross my mind to see a doctor).

[1] https://www.cdc.gov/flu/about/burden/how-cdc-estimates.htm

I don't agree. The sample is not representative of the whole. The only useful inference is a lower bound on how many cases there are, but this is more likely to mislead that inform.

If you know that officially confirmed cases don't reflect reality, and you care about reality, then the dashboard is not useful.

Reply on news.ycombinator.com