I won't blame anyone for forgetting this post about hockey and statistics, because, well, it's about hockey and statistics. It could make my eyes glaze over and I love both of those topics. But hey presto, now look at what I spied with my little glazed eyes: Fixing Plus/Minus by Tom Awad at Hockey Prospectus.
He had two simple improvements to suggest: first, adjust the individual's +/- to counter his team's influence over the overall number. Then, factor out the goalie (a person playing in front of Martin Brodeur is better off than one playing in front of Peter Budaj). The remaining number (he calls it RPM, for (i guess) Revised Plus/Minus) is a decent attempt to measure individual contributions to what is inherently a team-oriented activity.
One wishes he'd included more of the actual math, so one could duplicate his results. However, the chart at the bottom of the article is excellent, and at first glance it jives well with "I trust my eyes and guts, dagnabit" impressions of defensive play. The Islanders' Mark Streit, for example, is leading his team in scoring (as a defenseman!) and is still a plus player on a dreadful team. His RPM recognizes this. The list does seem to skew towards forwards, however: 23 of the top 30 instead of 18 of the top 30 (if it was an even distribution).** I don't know if that's a problem or if the forwards have just been a little better than the defenders this year. Perhaps there should be an adjustment for icetime - the average defender plays more, and should thus have a greater share in the numbers. (I'd check it myself but the lack of a formula or shown calculations is a hindrance.)
** EDIT - the 18 of 30 number is based on the ratio of three forwards to two defenders, which is the typical lineup; but an NHL team usually has four forward lines and three defensive pairs. Based on that ratio (2-to-1) a list of 30 would have 20 forwards instead of 18. I may wind up dusting off some of the old math I had been struggling with and try an adjustment based on ice-time, which seems the fairest measurement.