Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Suspicious Discontinuities (2020)

danluu.com

Post reply on HN
Print view View original

61–70 of 109 posts

Re: Suspicious Discontinuities (2020)

#61

Post by TeMPOraL »

XRG wrote: post #37
PaulDavisThe1st wrote: post #8

[ EDIT: Since the undated article is actually from Feb 2020, predating the expansion of federal health insurance subsidies, I withdraw everything I said below. ] ---------------------------------- The opening story is fabricated and/or bullshit. Last year (2025) there was no limit on income for health insurance subsidies. That ended for this year, but last year there would have been no reason for anyone who knew what…

Goes to show that having a creation date placed somewhere near the top of an article or blog post is a good idea.

It's always a good idea, and it's rather an obvious point - not to mention, something schools try to drill into kids since they're single-digit years old.

At this point I treat lack of dating articles as being done intentionally, and as evidence of malicious intent - i.e. that the entity publishing it has a reason they don't want readers to know the creation date, and there's really no reason other than trying to get away with lies or otherwise screw the reader over.

(Above is a heuristic I apply by default; this post is an example of a case-by-case exception, because of the type of site, my impression of articles read in the past, and of the author themselves.)

Re: Suspicious Discontinuities (2020)

#62

Post by TeMPOraL »

fc417fc802 wrote: post #54
dadkins wrote: post #43

Jim Roskind had a really nice talk [ https://www.youtube.com/watch?v=_uaaCiyJCFA ] at AWS Reinvent 2022 which highlighted the fence post problem. Basically, AWS was targeting P50 and P90 latencies in many services, and sure enough there were big clumps of latencies right under those target fence posts as engineers were basically gaming the metrics and nerfing the >P90 latencies to make it happen.

I don't think that counts as gaming the metric. It's an example of a metric that even when targeted in good faith results in consequences management didn't intend.

Good faith starts with understanding that "90" in P90 is an arbitrary number, and that the exact sampling threshold is not important to why anyone cares about that metric.

Re: Suspicious Discontinuities (2020)

#63

Post by TeMPOraL »

blharr wrote: post #12
hsuduebc2 wrote: post #5

I never understood why taxes or similiar absolute points aren't gradients instead.

I think because the gradient is simply too confusing for laypeople to understand. Even for a simple system like US social security that has a gradient. For every $2 you make over the limit, you lose $1 in benefits. I've heard countless times misconceptions of people thinking they'd be losing money (as in literally having less money net) by working.

Assuming that normies cannot understand gradients or feedback loops is IMO the source of a large amount of "tactical-level" problems in society, particularly around healthcare, taxation, and politics.

The better alternative would be to assume that normies are, in fact, capable of understanding these tools, and in the process forcing them to understand them by setting an expectation. I mean, this is what people should be told as answer when they ask, "what will learning math be useful for me in my life?". No, multiplying polynomials will not be useful to 99% of the people. But having a feel for basic linear algebra and feedback loops would be, because they describe the behavior of the simple and interesting control systems in any and all areas of life - which includes the Social Security examples of yours, too.

The worst thing is, letting majority of people off the hook here doesn't just impact them individually; in democracies, it prevents systems require certain level of understanding from being created in the first place. As shown in plenty of examples brought up in the discussion thread here.

So I guess one answer to "how is any of that math going to be useful for me in the future?" is, "if you all learn it, it will allow you to stop keeping yourselves in poverty in stupid and entirely unintentional ways".

Re: Suspicious Discontinuities (2020)

#64

Post by keynha »

dadkins wrote: post #43

Jim Roskind had a really nice talk [ https://www.youtube.com/watch?v=_uaaCiyJCFA ] at AWS Reinvent 2022 which highlighted the fence post problem. Basically, AWS was targeting P50 and P90 latencies in many services, and sure enough there were big clumps of latencies right under those target fence posts as engineers were basically gaming the metrics and nerfing the >P90 latencies to make it happen.

[dead]

Re: Suspicious Discontinuities (2020)

#65

Post by toast0 »

TeMPOraL wrote: post #63
blharr wrote: post #12

Earlier quoted context omitted.

I think because the gradient is simply too confusing for laypeople to understand. Even for a simple system like US social security that has a gradient. For every $2 you make over the limit, you lose $1 in benefits. I've heard countless times misconceptions of people thinking they'd be losing money (as in literally having less money net) by working.

Assuming that normies cannot understand gradients or feedback loops is IMO the source of a large amount of "tactical-level" problems in society, particularly around healthcare, taxation, and politics. The better alternative would be to assume that normies are, in fact, capable of understanding these tools, and in the process forcing them to understand them by setting an expectation. I mean, this is what people should…

Given that normies don't geneally understand tax brackets, I can understand not wanting to do more complex stuff either.

I don't know how many people I've heard express that they didn't want to move into the next tax bracket because they didn't want to pay that rate on their whole income, but it's a lot.

A very small minority are worried about hitting a ceiling that removes eligibility for important services, and some are worried about the weird phaseouts where marginal rates are very high, but most just don't understand.

Maybe that means having a more complicated system is fine because it's already incomprehensible to most, but I dunno.

Re: Suspicious Discontinuities (2020)

#66

Post by subscribed »

mnahkies wrote: post #3

The UK tax system also has a bunch of unfortunate cliffs, and tapers that create >60% marginal tax rates and worse. There's a calculator here that illustrates it well https://tax-cliffs.britishprogress.org/calculator The childcare cliff edge is probably the worst, but the personal allowance taper isn't ideal either as it's compressed over a relatively short income range And of course all the thresholds remain frozen,…

What's most infuriating it's only the job income, work, that's taxed like this.

Passive income of any sort? Bribes from the lobbying groups? Absolutely not.

Re: Suspicious Discontinuities (2020)

#67

Post by jefftk »

cadamsdotcom wrote: post #53

The marathon one has a simple - and fun - explanation: it’s great to run with people! Many (most?) marathons have pace runners who run the course, hewing to each 30 minute and 15 minute finish. So there’ll be a 3:30:00 pacer, a 3:45:00 pacer and so on. Your local pacer might even have their pace on their shirt and may even have a flag so they’re super hard to miss as you’re running. One friend of mine runs with a spe…

I'm from Boston and I'd never heard of this, but reading a bit it sounds like Boston is unusual in not having pace runners.

Given how hilly the course is, this makes sense: running an even pace the whole distance is not a good plan.

Re: Suspicious Discontinuities (2020)

#68

Post by khuey »

TeMPOraL wrote: post #62
fc417fc802 wrote: post #54

Earlier quoted context omitted.

I don't think that counts as gaming the metric. It's an example of a metric that even when targeted in good faith results in consequences management didn't intend.

Good faith starts with understanding that "90" in P90 is an arbitrary number, and that the exact sampling threshold is not important to why anyone cares about that metric.

The 99 in his TM99 is also an arbitrary number.

The more interesting part of the talk is how using what he calls "fenceposts" naturally results in bunching up against the fenceposts because even small regressions that cross the fenceposts are blocked but regressions of any size that don't cross a fencepost go unnoticed.

Re: Suspicious Discontinuities (2020)

#69

Post by hsuduebc2 »

blharr wrote: post #12
hsuduebc2 wrote: post #5

I never understood why taxes or similiar absolute points aren't gradients instead.

I think because the gradient is simply too confusing for laypeople to understand. Even for a simple system like US social security that has a gradient. For every $2 you make over the limit, you lose $1 in benefits. I've heard countless times misconceptions of people thinking they'd be losing money (as in literally having less money net) by working.

I have a different take. I think the state wants the system to be this rigid because, paradoxically, that makes it more flexible for them.

For example, they can keep certain tax exemption thresholds low while overall incomes rise. That means more people gradually fall into higher tax brackets or lose exemptions, even though the money they earn is worth less over time.

The rigidity itself is not a bug but a feature. It is a sneaky way to raise the taxes without openly raising taxes. But it's just a theory.

Re: Suspicious Discontinuities (2020)

#70

Post by fc417fc802 »

TeMPOraL wrote: post #62
fc417fc802 wrote: post #54

Earlier quoted context omitted.

I don't think that counts as gaming the metric. It's an example of a metric that even when targeted in good faith results in consequences management didn't intend.

Good faith starts with understanding that "90" in P90 is an arbitrary number, and that the exact sampling threshold is not important to why anyone cares about that metric.

It's just a plain bad metric. Useful due to brevity though.

A decent metric for testing against might consist of a target mean, the sigma specifying a gaussian about that mean, a minimum sample size, and a hard upper cutoff.

Post reply on HN