I'm not sure I fully understand your question, but I'll take a stab at what I think you are asking. Selection bias is not necessarily being introduced -- at least not in the usual statistical sense, but it depends on how you apply your analysis. My undergraduate and graduate degrees are in statistics and what we statisticians call "selection bias" is any sampling (random on not) mechanism which results in a set of data that is not representative of the population of which you would like to draw inferences. So if you want to apply your data to a specific market, you'll mostly likely need to train your data on that same market or at least take a random subsets of instruments that trade on that market. For example, if you intend to only trade small cap NASDAQ stocks, you should exclude large cap NYSE stocks when you are training your trading strategy. If you trained your strategy on DOW stocks for example because there are only 30 of them, and then applied them to all small caps, you might get completely different results than you were expecting because the DOW stocks inherently trade and perform different than the small cap stocks. In this example, you'd be better off taking a sample of all small cap stocks. Training your algorithms on a subset of them, then evaluating their performance by applying them to another subset of the data that was not used during the building/development of your algorithm. Does that help?