You people do realize that homogeneous statistics through entire season is impossible?
Homeogeneous statistics through pretty much anything in the real world is impossible. There will always be random or worse semi-random occurrences. E.g. the number of overtakes can be affected by the number of opportunities for overtaking, which in turn can be affected by other factors, e.g. how good the driver is at qualifying.
For the random factors, we would like to measure over a larger number of races, which would allow the random factors to cancel themselves out. That's where the concept of the standard error of the mean comes from and is why it decreases the more trials you have.
For other factors such as larger or smaller numbers of opportunities to overtake, then one thing that can be done is to acknowledge this problem, and therefore take any statistics obtained with an academic pinch of salt.
There is the possibility of deciding exactly what statistic to use, and to perhaps use a more complicated statistic. But this must be done before data collection, otherwise subconscious or conscious biases may affect the choice of statistic in light of the data. But, if a non-obvious statistic is going to be used, then it should be justified (again, before data is collected). If a really convincing argument for using a more complicated statistic cannot be created, then the default is to use the simpler one, giving caveats in analysis.
Designing these more complicated statistics is much much harder than some people on this thread seem to assume. E.g. you might want to count a pass on a leading car more than a pass on a tail end car. But a tail end car at Monaco might be much harder to pass than a car with similar power at a track with an excessively long DRS zone into a suitable corner. The weather might affect things. Trying to make the statistic "better" often falls into the trap of not considering enough ways in which one pass can be more valuable than another. The data just isn't homegeneous enough (yes Vlad, this hasn't wandered completely away from what you are saying). And once some passes are dropped, well then the number of trials goes down, and the standard error of the statistic rises. So even if the statistic does measure "something" more accurately, that additional accuracy (and more) can easily be lost due to greater variance due to random sampling.
