View Single Post
Old 09-22-2015, 12:11 PM   #31
Galadriel55
Blossom of Dwimordene
 
Galadriel55's Avatar
 
Join Date: Oct 2010
Location: The realm of forgotten words
Posts: 10,299
Galadriel55 is lost in the dark paths of Moria.Galadriel55 is lost in the dark paths of Moria.Galadriel55 is lost in the dark paths of Moria.Galadriel55 is lost in the dark paths of Moria.
Quote:
Originally Posted by Inziladun View Post
My non-mathematical brain hurts.


Ok, to explain this, let's pretend that a person with their eyes closed has to decide if someone taps them very lightly on the shoulder or not (yes=he thinks he's tapped, no=he thinks he's not tapped). If the person is tapped, and he says yes, that's a true positive (or a hit). If he is NOT tapped but says yes, that's a false positive (or a false alarm). Likewise, a true negative is when he correctly identifies the lack of hit, and false negative is when he doesn't feel the tap.

If the person says "yes" all the time, he will identify all the taps - out of the potential hits, he would get all 100%. But he would also get the full 100% for false positives. It doesn't make him more accurate or more prone to feel the tap than a person who says a mix of yes and no. That's why scientists, rather than count just the number of correct "yeses", look at the proportion of true to false positives. The person in my example would have a 50/50 proportion, which in fact means that his choice is pretty much random and could be based on chance alone (which is in fact what happens in the story). If the person actually tries to identify and separate the tap vs no tap, he would probably have more true positives and less false positives.

The point of this exercise was to eliminate the effect of a person's individual tendency to answer yes or no when in doubt (so in this scenario, the person tends to answer yes more than no, which skews the data towards the positive side).

In relation to Werewolf, there are people who tend to be lynched early regardless of role. So logically, in absolute terms, they would probably have more fenrisses than average. Like a truckload of apples would probably contain more wormy apples than one basket. However, the percentage of wormy apples in both could be the same. We don't just want to count the true positives (how many wormy apples, how many fenrisses), but to find the proportion of true to false positives (how many wormy apples out of how many in total? How many fenrisses out of how many D1 lynchings in total?). This would actually tell you if you tend to be fenrissed more often than others, or if you just tend to be lynched more often than others.

Now that I think of it, if we were to do a full-out statistical analysis, we'd have to throw in a lot more factors in because of the small data sets. Most of us haven't played in that many games to make for a decent stats analysis. For example, I've never been fenrissed, but I've only been a wolf twice, and both times in my cub days. That's a great survival percentage, but it's not telling you much about be because of the small number of games.

Ok, I think I'll stop my rant and try to keep my math-science geekiness under control.
__________________
You passed from under darkened dome, you enter now the secret land. - Take me to Finrod's fabled home!... ~ Finrod: The Rock Opera
Galadriel55 is offline   Reply With Quote