It does not... You have 20 data points. W or L. Unless you believe the better football team ALWAYS wins. Which is far from reality. 20 data points is what gets you in "Fooled by randomness" territory. 7-10 points underdogs still win the game outright around 25% of the time. 10-14 points underdogs win outright 20% of the time. The better team doesnt win 100% of the time in the NFL. Implying that any statistic from a W(wich is filled by variance to begin with) is any kind of relevant over 20 trials is.... Not valid.
Nope, I actually had a streak of 9 heads in a row during my little experiment. Thats as consistent as you're going to get, still doesnt mean anything.
Now you're talking, but thats not what you did. You simply need more data points to account for the fact that a win or a loss involves alot of randomness. In other words, Your argument is built around the attributes of the winning team, which involves a decent amount of randomness.
If you had say 200 games, then you could draw some conclusions, just as if I flipped the coin 200 times, I'd be alot less likely to flip heads 70% of the time. Sample size gets you closer to the truth.
There are a couple of things you could do to adress this problem.
- Split your data: Instead of just getting the winning team, split the dataset by quarters and account for winning quarters instead of winning games. This gives you 4x data points but doesnt account for game state that was influenced by bad luck. Its still an improvement.
- Split your data further: Just flat out go to PBP data and account winning plays. Depending on what you view as luck, just filter out plays where luck is involved. This would give you a big *** sample size to work with. ie. fumbles lost is mostly about luck, I have never seen a team thats better than anyone at recovering fumbles.
- Do a final four analysis: Instead of just going with super bowl winners, go with divisionnal participants, this will also 4X your sample size while including much more different game states to work with.
I think the last option is the most reasonable one and most likely to yield interesting results.