How excluding [some] luck helps for better statistics?

[indianaV8]

pzhon, thanks for the informative post. However, one of your (main) points I either did not understand, or it somehow deviated in different direction than my original goal.

I don't want to assume things based on larger sample than the very limited one I have from the current session.
The motivation is the following:
1) Now, as a poker player, you need to play several months of poker in order to get some statistically significant clue about do you play a winning poker at all. In several months, you change already your strategy several times, and the filed of player changes on top, and it's somehow digging in the dark. Basically once you have played X hands to have statistically significant data about your play, X is so big, that you are again in uncertainty because too many things will change during the timeframe of the next X hands.
2) So I'm looking for how you can get feedback much earlier, in cases when:
a) you change your strategy
b) your opponents strategy changes
and things that gives you better estimates than the standard one (take your results / divide by number of hands).
So it is about reducing the luck, but not by assuming profit.

In your example, if I win 10$ when I hit and 1$ when I miss, this is not indicative for if I or my opponents change strategy. I.e. if I change my strategy (to a dummy one that always fold no matter if I hit or not on the turn), then my EV will be 0 - and taking the average - that I as unlucky whatsoever - will lead me to wrong conclusions.
Another practical example is, if I have a winning strategy and I want to change limits. On the higher limits, now I only win 8$ when I hit, and 0.5$ when I miss (but I don't know this ... - actually I want to learn this as fast as possible, not needing to wait for another X hands).

For the Jam/Fold case let's say your actual EV is 2BB/100 hands. To find out (with 95% confidence) if you have A winning srategy AT ALL (i.e. BB/100h > 0) you have to play 220000 hands. If you reduce the STD from 45 to 27, you need only 70000 hands for the same conclusion.

For the same scenario, if you add knowledge about with what hands you will go all-in in this same situation, you may get and even better estimate. I will try this out.

So situation like the one you say (e.g. flush draw) will be very useful to look at, but not if you have estimate of previous results, but if somehow:
1) there is a way that you describe apriori your strategy in such situations
2) what the opponent do should not matter so much (maybe here your point about the bias is relevant).