Science & Industry

so you're trying to invent paladins rite?!

Most people struggle with standard deviation theory. 44 fails out of 45 attempts with a 27% chance of success would be well within expectations.

Nothing to see here. Move along.

This is normal, although pushing the envelope of bad luck. But as others have said the odds will even out over time.

look on the bright side. you just took 44 failures off the table. Statistically for every run of bad luck you should get a run of good luck to even it out.

So in theory if say it takes a batch of 1000 to consistently hit your 27% your remaining 955 runs will have a higher chance of success as you have 44 less failures in the 1000 run batch to worry about.

statistically at 27% you should get 12 successes in a 45 run batch. maybe you will hit a batch or two that will give you 15-20 successes out of 45 attempts. Just stick with it. in a large enough sample size the odds will always even out.

People win lottery jackpots with far higher odds against. You just got unlucky.

Overall, I'm getting around 60%-ish success rate. But of the ten jobs I ran yesterday, only one succeeded. 10%. Welcome to Probability Land.

"Invention is for people with poor math skills....."

The probability I obtain is 0.000012 for one or fewer successes out of 45, so one in ~100,000, which one can I think reasonably argue must happen to someone at some point. However...

The argument that with enough inventions things will average does not work, anyone who suggests it does does not know how to deal with cumulative probability or binomial distributions, but that is neither here nor there, this issue is worked directly into the OPs argument by using the binomial distribution. Maybe it will help if I say I was a math major in undergrad with continued studies at the graduate level (except I avoided stats cause its lame and unpleasant), point is, binomial takes this into account, whereas single event probability does not.

quick example of how binomial works:
Probability of single success: 0.27
Probability that one or fewer of 2 trials will succeed: 0.93 (9 in 10)
Probability that one or fewer of 5 trials will succeed: 0.59 (3 in 5)
" " 10 " " : 0.20 (1 in 5)
" " 50 " " : 0.0000028 (3 in 1,000,000)
" " 100 " " : 0.0000000000008 (~ 1 in a trillion)

See how it adjusts for sample size, so if you could kindly dispose of this line (numbers from http://stattrek.com/online-calculator/binomial.aspx)

I invent plenty, and likewise have very improbable strings of both successes and failures, and I see variation by ship type. I get the impression from this that things aren't governed by a random number generator, but also perhaps a weighting system, i.e. market quota and such to improve price stability or something. Given the volume, I feel this shouldn't be necessary, but I think anyone could comb through their results and find incredibly improbably streams of success/failures, of course the success streaks are rarely of note.

yup, the error people keep making here is thinking that long string of failures somehow affects the following trials. in reality each trial is completely independent.

sure you can say that some large sample (like 1000) will always produce exactly specified known percentage but the reality is, you don't KNOW which sample size will produce that percentage - it can be anything including 1e+3 and 1e+3000 with different probabilities.

I think you used inverse probability here instead of direct.

if probability of success is 0.27, probability that at least 1 of 2 attempts will succeed is the same as 1 minus probability of both attempts failing which is 1-(1-0.27)**2 = 0.467, 5 attempts is 1-(1-0.27)**5 = 0.793, etc.


keeping track of every invention attempt adds even more mindless clicking to already huge clickfest the invention is now, so I stopped doing that a while ago... but from what I have gathered, the overall chance is exactly as expected. I did not do per item stats but I don't think they are really item dependent. Overall score should be more trustworthy since by definition it has larger sample volume.

This is a very misleading statement as future events are independent of past events.

The only thing which is true is that you'll have a 27% chance of success on your next attempt. That's it. You do not get a run of good luck for a run of bad luck - ever. You just get a new result.

What people tend to call good luck and bad luck are simply deviations from the expected result. Thing is statistics tells us that deviations from the expected result is what is most likely to happen, not least. When you increase the number of results the deviation of the aggregated results is expected to diminish, but even over a million results you will not have a distribution which perfectly matches the expected result (though it probably would be pretty close).

Having said all that - ouch 44 failures out of 45 runs hurts.

Actually, according to statistics the rule of thumb is that you only need a sample size of about 30.

The OP experience a statistically unlikely event. This will eventually happen if you consider EVE's entire population of players. Sorry dude, you won the anti-lottery.