The best baseball teams have a smaller edge over the worst teams in an invidual game, but baseball also plays 162 games per season, twice as many as the NBA and ten times as many as the NFL. Let’s do a quick R calculation:

baseball_season = 162
baseball_p = 0.55
football_season = 16
football_p = 0.65

baseball = 1 -- pbinom(baseball_season / 2, baseball_season, baseball_p)
football = 1 -- pbinom(football_season / 2, football_season, football_p)

The calculation shows that a hypothetical baseball team that wins 55% of its games has an 88.5% chance of finishing its 162-game season with a winning record, while a football team that wins 65% of its games has an 84% chance of finishing a 16 game season with a winning record. So even though the football team has a higher win probability for each individual game, the baseball team is more likely to finish with a winning record by virtue of the longer season

Sometimes teams have clinched the playoffs so they rest their starters, other times teams might intentionally tank to lose games in order to finish with a worse record and earn a better position in the draft.

As a concrete example, in the 2014 World Cup, once it became apparent that Germany and the USA would both advance out of their group if they played to a draw, the gambling odds on a draw increased from 19% to 37%, as gamblers figured that the teams might take it easy and coast to a tie (in fact Germany won):

Here’s the FiveThirtyEight/Inpredictable graph:

The reason every game starts at 50/50 is that the Inpredictable model’s inputs are:

• time remaining
• score difference
• possession

At the beginning of the game, it’s always 0-0 and neither team has the ball, which means the teams have equal win probability. The model doesn’t account for team quality, though Inpredictable’s website does have a checkbox to adjust for pregame odds. As an example, when my hometown Philadelphia 76ers played at San Antonio in November 2014, gamblers thought the 76ers began the game with only a 3.6% chance of winning, which seems reasonable given that the 76ers were 0-8 heading into the game, people were already wondering if they were the worst NBA team ever, and the Spurs were the defending NBA champions. Sure enough, San Antonio won the game handily, 100-75.

I’ve written a bit before about win probability models and how they compare to in-game gambling odds, and how models have both advantages and disadvantages when compared to gambling odds. Models are a great tool to estimate win probability when there’s no gambling data available, and if you can build a better model than the best one that currently exists, you might be able to use it to make money. On the other hand, gambling odds will tend to be more accurate than any publicly available model since presumably gamblers are at least as good as the best publicly available model, and probably gamblers are better.