You only go for it if you have better than average odds of getting it. Its not like random draws from a known distribution. Any included data on 4th down is going to be subject to some variation of survivorship bias. See, for example, also the consideration:
I used only data from the 1st and 3rd quarters to exclude situations hurried by an expiring clock and by desperate teams or teams with large leads playing differently late in games.
So, by this logic the 4th down data is similarly non-standard, no? It would clearly be a situation different than normal 1-3 downs in 1-3 quarters.
4th down data is not non-standard for 4th down data. Sure, you can't say for sure what would have happened if they had gone for it, but there's nothing particularly different about these downs that would lead to a higher conversion rate.
I think the point is that if you are using game data then the only times the offense is currently going for it on fourth down is because they have some reason to believe they can make it with a high percentage. Football is not played with random number generators, the coaches and players understand the matchups and the chances they have of making X yards with the players and plays they have available. The teams that didn't go for it might have a reason to believe that with the players on the field the probability of them getting X yards was small.
I used only data from the 1st and 3rd quarters to exclude situations hurried by an expiring clock and by desperate teams or teams with large leads playing differently late in games.
So, by this logic the 4th down data is similarly non-standard, no? It would clearly be a situation different than normal 1-3 downs in 1-3 quarters.