News for hackers. Stuff that matters.
Re:Roche CEO's says some Covid-19 antibody tests are a disaster by jxramos on Wednesday April 29, 2020 @07:37PM (#23025875)
Proportionality frequently seems to escape journalists, why be satisfied with reports being confined to the qualitative and non-quantified realm? Give us some background please. In this molecular diagnostics domain what are typical figures for sensitivity and specificity? FDA lays down requirements for such things right? Anyone know off-hand what percentages are required for a diagnostic testing product to be released? For these tests in question how far off from those marks are we exactly?
Re:Roche CEO's says some Covid-19 antibody tests are a disaster by DenisM on Wednesday April 29, 2020 @07:58PM (#23026040)
There is no "typical" I'm afraid. Whether a test is useful or not depends on error rates, priors, and application of the result.

For example, consider this scenario:

- Virus has R0 of 2.5

- the test for the current infection (i.e. not antibody) is cheap and abundant

- the test has 20% false positives

- the test has 20% false negatives

- Application: test every single person and confine everyone who tested positive to their homes, releasing the rest. Repeat weekly. Take no other action.

Result #1: 20% of the population are in isolation. 20% of your economy is stalled, rather than 100%.

Result #2: 1 in 5 infected will go undetected. Still 4 out of 5 are isolated, reducing the number of infected out there by a factor of 5, therefore your R0 drops from 2.5 to 0.5 The number of infected people on the street gets cut in half every week. Epidemic fizzles out.

So even low-precision tests are useful. What is useless is tests that do not have a reliable metric for false positive / negative. Which is what many of the AB tests are.

Re:Roche CEO's says some Covid-19 antibody tests are a disaster by perl4ever on Wednesday April 29, 2020 @09:36PM (#23026598)
"Result #1: 20% of the population are in isolation. 20% of your economy is stalled, rather than 100%."

This bugs me, because in this hypothetical, you're implying something as a point of comparison which is nothing like reality. 100% of the population is not in isolation - the only people who are in isolation are those who have had such severe symptoms they've been tested and it came back positive, or they're really obviously linked to such a person. And obviously 100% of the economy is not stalled; it may not be precisely clear what the impact is so far, but people are talking about single digit GDP percentage point declines. If you frame that as the worst economic news since the great recession, or put it in context with the job losses, it's pretty bad, but it clearly could be ten times worse and it's not.

I get that maybe you meant it as just an example and specifics aren't important to your point about statistics, but I feel like I've read similar posts to yours that kind of get the nonsensical scenario out there, that we've shut down "100%" to where people kind of get used to it even though it's not reality.

Re:Roche CEO's says some Covid-19 antibody tests are a disaster by Retric on Wednesday April 29, 2020 @09:55PM (#23026707)
US’s isolation started very late in Q1. It requires a very large decline to average out to ‘just’ 1%.
Re:Roche CEO's says some Covid-19 antibody tests are a disaster by perl4ever on Wednesday April 29, 2020 @10:19PM (#23026826)
I don't feel that makes a difference to what I just wrote. But I can do some very rough arithmetic along those lines to see what the decline would have to be.

The GDP figure just reported is -4.8%. Based on what I recall, the inflection point was about halfway through March. Friday the 13th is when the state government in NY finally said anyone who can, work from home. That's 0.5/3 months or ~17% of the quarter I'd assign to "shutdown".

0.83 * 1 + .17 * x = .952

...I don't trust myself to do high school algebra, so I asked Wolfram Alpha and it said x = 72%. Ok, so the reported figure of nearly -5% probably equates to a little over a quarter of the economy missing.

I don't think I need to concede my point about a 100% stoppage not happening.

Re:Roche CEO's says some Covid-19 antibody tests are a disaster by Retric on Thursday April 30, 2020 @04:20AM (#23028664)
The larger point stands it’s just ‘but people are talking about single digit GDP percentage point declines‘ that I was correcting. That said, Pennsylvania and several other states waited until April to order a shutdown, but the observed GDP decline was annualizing a ~1% though there was growth in the ready of Q1.

Long story short, that 4.8% is absolutely worthless.

Re:Roche CEO's says some Covid-19 antibody tests are a disaster by perl4ever on Thursday April 30, 2020 @10:06AM (#23030916)
The bit you quote from me seems strictly accurate; by single digit I mean < 10%. That is what the news is reporting. It is factual, and my inference that the economy is not shut down hasn't been questioned.

Saying that the difference is larger when you consider only part of the quarter is correct too, but 4.8% is not "absolutely worthless", it just needs to be interpreted properly.

As in my previous comment, you can approximately derive the actual amount of decline, which is a fraction of a complete shutdown.

Re:Roche CEO's says some Covid-19 antibody tests are a disaster by Retric on Thursday April 30, 2020 @11:18AM (#23031763)
By absolutely worthless I mean it’s a meaningless calculation. They take the actual change averaged out over a quarter, and then assume 4 more quarters will have exactly the same thing happen.

It’s a common calculation used to make quarterly numbers seem more important, but it’s at best misleading if not actively deceptive.

Re:Roche CEO's says some Covid-19 antibody tests are a disaster by perl4ever on Friday May 01, 2020 @09:45AM (#23041914)
Wait, wait, wait. Nothing is "meaningless" if it contains information. And "misleading" depends entirely on context.

It sounds as though you are suggesting I made an error that increases the apparent size of the shutdown, meaning it is actually smaller and my claim is stronger than I thought.

If the figures are annualized (I checked* and I think you're right) then it improves my original point - the 4.8% decline becomes 1.2% un-annualized, and the normal trend of +2% becomes +0.5%.

The equation should then be something like .167 * x + .833 * 1.005 = .988, "x" being the percentage of the economy active in the last half month, or about 90%.

I'm confident that there is information in the reported figure and that no reasonable calculation can show the economy was shut down completely.

*https://www.bea.gov/data/gdp/gross-domestic-product

Re:Roche CEO's says some Covid-19 antibody tests are a disaster by Retric on Friday May 01, 2020 @10:11AM (#23042123)
You’re ignoring information that you disagree with. Yep, of the multiple errors in your calculation, one of them lowers the estimate. But, I never said anything about a 100% shutdown just that single digit comment.

Going in the other direction, picking the earliest possible starting date underestimates the impact as does ignoring panic buying. Layoffs where also not instantaneous, and people’s final paychecks are delayed etc. All of this is going to show up in Q2 numbers even if many states are scaling back their shutdowns.

As to being a meaningless calculation, you’re describing a reversible calculation not a useful one. I can type a message in ROT13 and the result contains the same content, that does not make me doing the ROT13 useful.

Re:Roche CEO's says some Covid-19 antibody tests are a disaster by perl4ever on Friday May 01, 2020 @11:05PM (#23048791)
"You’re ignoring information that you disagree with"

You're the one saying information is "meaningless", and at the same time I don't see any alternative data that you've provided. You're just making general statements that sound to me like the fallacy that imprecision means no information.

In your first reponse to me, you wrote: "It requires a very large decline to average out to ‘just’ 1%"

So, "very large" is subjective, but I don't think 10% is "very large" in the context of people talking about 100% of the economy being down.

Re:Roche CEO's says some Covid-19 antibody tests are a disaster by Retric on Saturday May 02, 2020 @06:46AM (#23050927)
> imprecision means no information

It’s imprecise enough that you’re better using other means to estimate the decline. I am between 1cm and 1 mile tall, is true but does not convey new information.

Similarly without knowing how fast was the economy was growing or shrinking before the decline, the decline hit, when peak decline hit, it’s impossible to say how bad things where at the end of the quarter beyond well it’s worse. I mean presumably when news of the outbreak hit but before the shutdown there was some effect, but I don’t know if it was a net positive or negative.

If you want numbers I would say 10+x larger is objectively a very large decline relative to 1%. Anyone talking about a 100% shutdown while still buying food is being using hyperbolic or simply does not understand what their words mean. It’s at best symbolically true in that net wealth is being destroyed so the engine of human progress has shut down.

Re:Roche CEO's says some Covid-19 antibody tests are a disaster by jxramos on Thursday April 30, 2020 @12:08AM (#23027332)
So interesting, so even with inaccurate tests so long as we're dealing with reliable metrics backing those tests up then the degree of actual accuracy that intersects with the nominal accuracy is enough to carry out the compounding with and effect targeted quarantines with. Probability products stack up quick. Hope that makes some bit of sense.