Hacker News Reader

Comment by Tossrock | Hacker News Reader

I agree that the Soyuz has been very reliable historically, but two major incidents involving it within a few weeks has the appearance of a trend.

View on HN

Soyuz is literally the most reliable launch vehicle and two datapoints are not a trend. Any two datapoints form a line, ie a trend.

"Once is happenstance, twice is coincidence, three times is enemy action." Maybe it's a coincidence, maybe it's not. I think it's very reasonable to want alternatives in any case.

Even with three failures I would not consider it a trend unless they are related in nature. For the current trend of "basically no failures" to go to "some failures" you'd need a couple dozen failures at minimum.

The alternative is to fix whatever broke Soyuz.

Russia at this point is largely deindustrialized and rapidly losing its tech know-how. The recent string of failures in space programme is only one (but certainly high profile) manifestation of it.

> Any two datapoints form a line, ie a trend.

Two datapoints are not a trend.

People who use run charts (control charts,time series charts) talk about trends being a set of five points all going the same way. http://www.qihub.scot.nhs.uk/media/529936/run%20chart%20rule...

While I generally agree, I'm being more generic.

A single data point doesn't establish any direction or trend at all.

With two data points you can establish a trend and make a prediction about the next data point. However, any two data points form a line, so any two results can be used to make a trend, however wrong it may be.

Only when more points are used can you confirm the trend and reduce the probability of being (un)lucky.

Two in a row is a coincidence, three in a row might be a rule

For what it’s worth, control chart sensitivity rules are selected based on economic considerations. So the decision to use five points is influenced by both statics and how affordable it is to investigate why the control chart alerted. Sensitivity rules can be adjusted, applied, or ignored based on the specific situation and the analysts tolerance for false positives/negatives.

If the worry is about the manufacturer rather than the design, then we have more than two data points. The Soyuz-2 launch vehicle, which is replacing Soyuz-FG, has about a 10% failure rate (including an ISS resupply mission in 2015). This was as far as I know the second-to-last -FG flight planned, as manufacturing has been discontinued and the -2 is the main production line now.

This reminds me of the fact that if you give most people the output of a true random number generator, they'll think it's rigged because there appears to be more clustering than is intuitive.

I recommend making a playlist and making a self-blind study which feels more random: your native music player shuffling it or a true random shuffling of the list.

You'll be surprised.

That's because most people don't really want a "random" playlist. They want a random shuffle: A set of songs played in random order with no repeats until all have been played. That's not the same as making independent random picks from a list, but that's what most people want when they "randomize" their playlist.

I was more thinking about taking a list and shuffling it instead of picking songs at random.

In both cases, people complain because, for example, the same artist might be picked twice in a row. Most shuffle functions aren't random but try to make distance between similar songs.

Reply on news.ycombinator.com