What is a good way to compare similarity between datasets with little variance?

Question

Let's say I have a list of 100 MLB pitchers and 5 statistics for each of them. The difference between, for example, an ERA of 3.5 and 3.1 might not look like a lot to a naive similarity algorithm, but is a lot in baseball. Given that a lot of the player statistics that I'm looking at have this little variance, a lot of a little variance like this, what is the best way to calculate similarity between two players?

An example of the data might look like this:

Player | ERA | Wins | Strikeouts
--------------------------------
A      | 3.5 | 15   | 180
B      | 3.1 | 12   | 210
C      | 3.4 | 13   | 150

I've used cosine similarity, and the results are all too similar, down to a thousandth of a decimal place.

Answer 1

This is not the kind of data you'd use cosine similarity for. Have you considered just averaging z-scores? z-scores will be sensitive to this low variability you talk about.