Unfortunately, if you want to answer the question: "how accurate are my gathered stats from the true distribution of this player's tendencies?", you'll need to look into some graduate level statistics. That said, we can make at least 2 observations that help us without getting a math degree.
The accuracy of your stats on a player depends on the amount of situations you were able to record them, not the number of hands recorded. Therefore, a basic PFR stat is a usable base after a couple of orbits, since PFR is a possibility in every hand. Compare this with the more specific PFR from UTG stat. This is available every n hands (assuming n-handed play), which means it should take ~n times more hands to have the same accuracy as your basic PFR stat.
So always compare the stat with the number of possibilities to create said stat, not the number of hands played. This will make very specific stats like triple barreling from BB after 3-bet from BB inaccurate possibly even after thousands of recorded hands.
The second thing to take into account is the percentage you actually expect from the stat. The closer your expectation is to 50%, the less samples you need to confirm this statistic. To construct an example, let's say you're playing 3-handed and expect that a player will raise from BU 50% of the time. If you have around 40-60% raise percentage after say 20 samples (exactly 20 orbits in 3-handed play), you know that at least you're in the ballpark with your 50% estimate. If, on the other hand, you expect - for whatever reason - that he'll open shove from the BU 1% of the time, you won't have enough information after 20 orbits. Anything between 0% and 5% (and even higher) for his open shove statistic could very well be a possibility, after all, they only had 20 chances so far. It's quite likely they just didn't get any high pairs / broadways with which they would use this move.
So always assign lower accuracy for stats the farther you expect that stat to be from 50%.
Those are the basics. Happy HUDding!