Simple. All his offensive stats are per 60 minutes. Karlsson averaged about 28 minutes per game the last 3 years. Even though he outscored all D in each year, per 60 minutes he didn't. Klingberg for instace has basically one year with limited ice time per game and is used in offensive situations far more often then not. Karlsson is used in all situations.
That is a huge flaw in his ratings. Offensively Karlsson is clearly the best over 3 seasons. Burns might be statistically if you solely looked at last year. But over 3 years it has to be no contest.
His ratings on D do not give higher rankings to D that get the most icetime, comparing 25-28 minute guys equal to 20-22 minute guys. I think icetime is a huge factor in how you should look at D-Men.
Doughty is ranked lower then he should be because he plays 27 mins a game for 3 years and so many of his ratings are based on /60 minutes.
Your post makes me realize something that I think could be very important.
Everywhere, you see people using /60 stats. But,
has anyone actually proved that most of the statistics are linear in terms of minutes played?
Because quite frankly, my theory would be that most of them are not and start to "plateau" the more a player plays per game. For instance, let's use Player A's pts/60 of 1.5, ie that player put up a point on average every 40 minutes. This assumes that if the player played 10 minutes per game, he would put up 20.5 points, then 41 points for 20 minutes per game, and 61.5 points for 30 minutes per game.
While it mathematically makes sense, I just don't think it realistically makes sense. Would be nice if someone found a way to confirm (or not) whether the use of /60 stats is justified, and if a non-linear model for stats adjustment in matter of time played is more adequate than a linear model.