Hacker News Reader

Comment by phreeza | Hacker News Reader

In my experience, the obsession with correct significant figures is an american high school thing. Why should you deliberately have to choose a number that is not your best guess? Ideally you should include confidence bounds, but if you don't, it shouldn't imply that you are confident to that level.

View on HN

Because it gives an idea of what range is expected for the actual price. If someone gets an estimate of "about fifty bucks" versus "about $51.04", their reaction to seeing a final price of say $62.17 would be different. The first one implies that it's a rough estimate, while the second one doesn't.

"About" is doing a lot of work, too, no?

If someone said "about fifty bucks" and it turns out to be $62.17, I'd immediately understand them to be scamming me.

Really depends on what the context is. Maybe for a taxi ride when negotiating a price ahead of time, but pointing at an object, asking for its price, getting an estimate that's off, then getting the actual price before the transaction is completed doesn't feel like a scam.

Please price estimates for car repairs in dollars and cents and state it's only an estimate all the time.. with variations possible depending on what's discovered that might have been missed it that particular context.

Yep. The weird thing is that their sigfig rules of thumb aren't particularly accurate and after any non-trivial operations (or more than a few operations) just give wrong results.

Tracking precision through calculations requires interval arithmetic, which is a pretty big PITA and not frequently used.

I think the sig figs cult comes out of mistaking a simplified lesson on precision for a useful practice.

Not sure exactly what you mean. It is well known that you should keep additional digits throughout the calculation and then drop non-significant ones for final results.

The number of digits out is only a simple function of the digits in if you're doing fairly trivial calculations.

Checkout https://en.wikipedia.org/wiki/Interval_arithmetic for what it takes to monitor precision through complex calculations (in particular, see the bits about using the taylor expansions of non-trivial functions).

When I was getting my engineering degree too many significant figures would earn you -2 pts.

Measure with a micrometer.

Mark with a chalk.

Cut with an axe.

(Apparent practice at a workshop I used to frequent.)

Old crusty semi-retired engineer I knew said when he was young he liked designing things to close tolerances. But once he was old he found more pleasure in designing things that worked with really shitty tolerances.

-I had a colleague once, fresh out of mechanical engineering at the local univerity college, who as his first project did a locking pawl for a winch drum.

Tolerance? H7.

The fitters threw a fit trying to make that happen.

Clearly he wasn't working in any sort of mechanical capacity while getting that community college degree.

That's definitely where I learned that it was expected.

I look at it as one of many possible conventions for communicating precision. For what it is, it works well. But it's not universal, and people who do and don't use it should both keep in mind that the other group exists.

It hints that someone doesn’t understand these confidence values at all.

Why $47.10 and not $47.1032471?

Reply on news.ycombinator.com