You could also try throwing in team defense records as a factor. Something like giving a certain amount of points to the top icetime defenceman on the team that gave up the least goals might give, say Langway in '84, weight that he wouldn't get in scoring stats.
Though I doubt there could be a model that would correlate with the voting results as accurately as a Hart Trophy model would. The Norris, much like the Selke, has a lot to do with a players reputation based on past seasons.
I have a model (covering 1999-2010) which says that Norris voting is a product of four statistics: goals, assists, plus/minus, and ice time. All four correlate positively with Norris votes (i.e. the higher the number is in each stat category, the more votes a player gets). I've broken out goals and assists (rather than treating points as one category) because goals seem to be valued nearly twice as much as assists.
The model isn't incredibly accurate, but it's not terrible either (correlation is 0.47). Correctly predicts 7 of the past 11 Norris winners.
I've considered other variables (age, whether they played on a playoff team, penalty minutes, shots on goal) but these don't seem to have a significant impact in either direction.
The model seems to suggest that voters do in fact take defensive play into account, above and beyond the plus/minus statistic. (I'd argue that plus/minus is a poor proxy for defensive play, but a lot of people think it's an accurate measure).
In many cases players who had "the right numbers" to win the Norris trophy, but didn't, have a reputation of being poor defensively (Sergei Gonchar 2002; Bryan McCabe 2004; Mike Green 2009; Mike Green 2010). On the other hand, excellent defensive play likely explains why some players are finalists despite the fact that they don't have "the right numbers" that voters are looking for (i.e. Ray Bourque 1999; Scott Stevens 2001; Derian Hatcher 2003).
Will try to see if I have penalty killing data in a usable format (maybe that would serve as a reasonable proxy for defensive ability) and make the model more accurate. If anyone has an excel spreadsheet with PK ice time from 1999-2010 please let me know. Don't have the patience to manually copy all of it from NHL.com.
Going to see if I can fine-tune this a little more then will show the results.
Last edited by Hockey Outsider: 12-22-2010 at 12:47 AM.