Thanks for spotting this! I looked into it, and it seems to be double-counting the most recent result when computing the median. It’s an order-of-initialization issue; it thinks it’s getting all the results except the new one, adding it, then taking the median, but it’s actually getting a list of all the results. The fix is straightforward; I’ll email the admins to apply it.
It won’t prevent trolling but it will minimize its effects. As it stands, you can input numbers like 1e+19 which will seriously throw off the mean. If trolls can only give the highest or lowest reasonable bound then they’re not going to have much of an effect individually and that makes going through the effort to troll less worthwhile.
While the standard e.g. wolfram alpha definition (which isn’t normative) of the median would be 43.5 as well, it is an accepted practice (in plenty of CS grad classes, at least) to have the median guaranteed to be an element of the sample, normally the upper median is then chosen simply as “median”. Hence the wiki definition having the qualifier “usual”.
In fact, I was surprised that the median is strictly speaking not guaranteed to be an element of the set, using the majority mathematical definition.
So, not so much an error as a lazy CS convention …
Neat. Thanks!
This deserves karma. For fun, enter how much you think this post will get. [pollid:6]
It correctly interpreted ۲ as 2. :)
After my vote:
″ Mean 43.5 Median 75.0 Total votes 2″
Well this is mathematically impossible… My guess is the median isn’t properly calculated for even numbers of votes.
Thanks for spotting this! I looked into it, and it seems to be double-counting the most recent result when computing the median. It’s an order-of-initialization issue; it thinks it’s getting all the results except the new one, adding it, then taking the median, but it’s actually getting a list of all the results. The fix is straightforward; I’ll email the admins to apply it.
Your fix for the incorrect median calculation has been deployed.
I got a mean of −9.91765890411e+16, so something is still wrong.
No, that’s unfortunate but correct (several people entered things like entered −3e+18 as their estimate).
...trolltastic.
“Mean 1.25e+18 Median 45.0 Total votes 8”
Specifying a lower and upper bound on the input should be required.
That doesn’t really prevent trolling, so i’m not sure that it would be helpful.
It won’t prevent trolling but it will minimize its effects. As it stands, you can input numbers like 1e+19 which will seriously throw off the mean. If trolls can only give the highest or lowest reasonable bound then they’re not going to have much of an effect individually and that makes going through the effort to troll less worthwhile.
Not necessarily:
Votes: 12 and 75
Mean: 43.5
Median (upper median): 75
While the standard e.g. wolfram alpha definition (which isn’t normative) of the median would be 43.5 as well, it is an accepted practice (in plenty of CS grad classes, at least) to have the median guaranteed to be an element of the sample, normally the upper median is then chosen simply as “median”. Hence the wiki definition having the qualifier “usual”.
In fact, I was surprised that the median is strictly speaking not guaranteed to be an element of the set, using the majority mathematical definition.
So, not so much an error as a lazy CS convention …