This game is pretty screwy in terms of prediction/evaluation systems. 538's ELO has this game basically as a push (51%/49% Pats), ESPN's FPI has the Pats as more like 64%/36% favorites, while DVOA has KC ranked much higher and gives them almost double the Pat's chances to win the Super Bowl (which by necessity means winning this game), though only a 52%/48% to win this game (because while the odds of making the AFC championship game are close for both teams, their system sees KC as much more likely to win it).
Basically all three systems see Ari and Den as heavy favorites, and all three favor Car (though DVOA only very slightly). However FPI is running its predictions (and from what I recall that is that FPI is based on marginal expected points added on a per play basis to get the ranking, and then monte carlo simulator on those outputs), it seems to have significantly more lopsided projections than the others. I know that they add modifiers into the prediction engine after computing the raw rating scores to account for things like seasonality, home field, days of rest, etc., which would seem to double count those factors a bit (as they were already affecting the inputs for the rating), but it is interesting how confident they are in the Pats given the lack of agreement from the other systems.