I think I understand now - so we eliminate shots from bad positions, and what we're left with is the more true measure of the strikers ability, then? In other words, get rid of the "noise"?
Not quite, I'm not being very clear. We don't eliminate any shots. Let me try again.
So ExpG (or Expected Goals) is a new-fangled stat that assigns every shot in a game a number based on how likely it is for an average player to score from that shot. By far the most important factor in determining the likelihood of scoring from a particular shot is the position it was taken from--a shot from the six yard box is much easier to score than from the halfway line. Other factors go into it as well, though--it's easier to score with your feet than your head, easier to score off a throughball than a cross, etc.
So if a shot from the penalty spot off a throughball with your feet is scored 50% of the time, it is assigned an ExpG of 0.50. And that's done for every shot.
Now of course a striker is unlikely to score exactly as often as ExpG predicts. So let's say Sturridge takes 10 shots over the course of 2 games, with a total ExpG of 1.5 goals. This implies that each shot had an average ExpG of 0.15 (that is, a 15% chance of scoring). That 0.15 number can be described as his Expected Goals per Shot, or ExpG/S.
In reality, however, he scores 2 goals. That means that he's scoring at a higher rate than an average finisher would have with the same chance. We can find the exact rate by dividing 2 goals by 1.5 expected goals to get 1.33, showing that his finishing for those 10 shots was 33% above average. Sounds good.
So we've got 4 stats.
One is his Expected Goals per 90 minutes, which in this case is .75.
(That is we dived his total ExpG (1.5) over two games by 2 (the number of games)).
Second is his ExpG/shot, which is .15.
Third is the total goals he scored, 2.
And fourth is the quality of his finishing (goals per expected goal), 33% above average.
What the post I linked to argues is this:
A) The quality of finishing (i.e. G/ExpG) is over a non-giant sample
all noise. The distribution of finishing quality, as defined in this way, is random. So we don't give Sturridge any credit for finishing 33% above average.
B) The really important stat is ExpG/90. That is, what we care about is how many shots (especially, but not only, good shots) the player takes. So Sturridge gets credit in this example for his ability to create so many shots that he's expected to get .75gp90 (a great rate), but doesn't get any credit for finishing those chances at a higher rate than we expected.
The question I ask is what is being defined as "skill"?
The ability to "create" (or take shots that have) high Expected Goals. In less statty language, the ability to create lots of shots and/or shots that are very likely to be converted.