Zee said:
Saw a weird stat on Andersen. His save % is actually worse the more days of rest he has. He hadn't played for 3 days. His best games are with 1 day of rest which lines up nicely for a bounceback game on Wednesday. (let's hope)
?P-hacking? is the name for the phenomenon that occurs when you ask many, many questions of a particular data set (said another way, when you search for correlations in a data set) and then advertise any correlation you find as if it is statistically significant, even when it probably isn?t. P-hacking goes something like this:
1. suppose a correlation (like the correlation between Andersen?s save percentage being low and having 3 days of rest) has a 1% chance of appearing in the data set even though it is just a fluke and there is no real causation between variables
2. Now check 100 different correlations/questions of that sort
It is almost certain if you do that, you will stumble across a correlation where where there is no real causation at work.
The Andersen save percentage/3-days rest thing is likely an instance of that. Other instances are things like:
* the leafs rarely score first
* the leafs rarely lead after 1
* the leafs are better/worse in the 1st, 2nd, 3rd period
* the leafs always lose on Tuesday?s, or when there is a full moon or etc
Basically, broadcasters look through all possible combinations of periods, home/road splits, days of the week looking for an outlier. If we look for all possible statistical anomalies, we are going to find some very unusual things in the data, but they are going to be flukes. They won?t be predictive/meaningful.