from Travis Yost of TSN,

Last May, a team executive asked where I thought statistical analysis was having the biggest impact at the NHL level. A few answers came to mind, but I immediately mentioned that I thought our valuation of defencemen – especially puck-moving defencemen – had rapidly improved.

It's not a coincidence that a few days later, I wrote about how the Tampa Bay Lightning's stealing of Anton Stralman out of free agency was paying huge dividends. Stralman is something of an idyllic case, but it does capture the essence of what paying attention to the underlying numbers can do for a team. He never had lofty counting numbers and wasn't the most physical presence on the ice, but his impact on the possession game was indisputably strong. With that, of course, comes the goal differentials that win games.

One other question: he asked where I thought statistical analysis was struggling. This was a much easier answer, and it's a topic that gets a lot of play around the blogosphere.

We haven't figured out goaltenders.

There are myriad problems with the position, especially as it pertains to predictive analysis. The biggest driver is that save percentage – perhaps the best metric we have for capturing a player's ability to stop shots – is swamped by random variance, team effects and (potentially) sample size issues that kill any hope at meaningful confidence intervals. It leads to very little repeatability in goaltender performance, which makes it a struggle to forecast what many players will do in upcoming years.

continued