Hacker News Reader

Comment by dotancohen | Hacker News Reader

If this incident happened 5 months after getting to C3, can we expect that on average we should see a similar collision event every few months, on the order of one to three (detected) collisions per year?

View on HN

A single incident will not give you any sort of sense for how often this happens. It could be luck either way. Might be this was a relatively calm period and it happens way more often in the future or incredibly unlucky and we never see another event for the rest of the mission.

Fwiw, the article mentions this was the fifth time it's been hit.

This being the most serious of the incidents.

I think this is too pessimistic. One always gains some information, even from a single event. If you're catching the bus one and shows up literally the minute that you arrive at the bus stop, yes, you may have been very fortunate and in fact the bus runs only once per day, but you really ought to assign a lower probability to that possibility now, given your recent single new piece of information. There are ways to express this quantitatively too - but your own intuition ought to be enough of a guide here.

This is classical Bayesian inference. You update your priors when new information comes in. If something happens that you thought was unlikely, you update your credence about this event.

yes and perhaps the difficulty of using a frequentist approach on a sample size of 1 is why the parent post thought that you cannot learn anything from a sample size of 1...

Hacker News is a very peculiar place. You make a statement that is fundamental to understanding statistics and you get downvoted till your post dies. On the other hand it's the place where the smartest people get together to discuss things. Smart people don't like statistics?

It helps to remember that the downvote button is so close to the upvote button that it’s very easy to hit it by mistake on mobile. And I reckon most people don’t take the time to double check and undo the downvote.

I feel you, and it’s a natural worry. I deal with it too. But remember, the point of HN is for the discussion to be for the readers. And reading about downvotes is boring.

One hint that helps me, is when I write things like “Smart people don’t like statistics?” —- it sort of helps me realize that I’m in the frame of mind where I should step back and close orange website for awhile. And I mean all of this genuinely; I’m not patronizing or just saying it. It happens a lot.

Oh I did. I got 99 problems and a downvote on HN isn't one.

To be fair, the statement rests on the assumption that it's a singular strike that TFA doesn't support.

Basically there is now a probability distribution for how frequently the JWST will be struck by meteorites. The most probable amount is once per five months. However it’s a distribution with pretty fat tails, almost equally likely is every four or six months, slightly less is every three or seven, etc.

It's a question of sample size. There might be frequent events out there that kill the entire mission but we've been lucky so far. This might be an anomalously "noisy" period. We just need more data to make the determination.

New data will adjust the probability distribution and flatten the tails, but even 1 data point is enough to get an initial probability distribution.

Reply on news.ycombinator.com