However, the best data is - as pointed out by ElstonGunn - in basketball and baseball (primarily because those sports lend themselves to advanced metrics much better than football).
Er, no. It's because the statistics used have high frequencies, like pass completion in football. High frequencies reduce the effect of experimental error, which is why it's better to have large sample sizes.
Where 'form' becomes relevant is where there are lower frequencies of specific events occurring which are, due to their low frequency, also affected more by psychology.
If you make a mistake with something you do 100 times in a game, you will probably put it right with only a statistically insignificant effect on your success rate. It might even make you take more care/time to ensure successful completion, but effectively slow the game down.
If you make a mistake for a low frequency task, you have less opportunity to make up for this, and also, it will be more likely to affect your confidence. Low frequncy events are less likely to be autonomous, making them more vulnerable to cognitive processes, affected by mistakes.
In any case, players who feel good about their game (confidence/efficacy), for whatever reason (including thinking they are in good form), are more likely to play well. There is no way that can even be disputed with the amount of emperical evidence there is that demonstrates that.
What is frustrating me about this discussion is that it is even taking place. Form has absolutely nothing to do with the hot-streak research. It is being completely misused to argue something completely different.