To avoid errors; for example in a research where it might be impractical or inaccessible to sample an entire population, we need to use a representative sample. Additionally, an experimenter/researcher must do whatever it takes to ensure that the sample is as representative as possible, and that statistical tests consist of checks and balances to avoid errors.
In the end, of course, conclusions are drawn about how an entire population would respond based on the responses from randomly selected group of people.
My question is this: But how do we ensure that the probabilities remain the same in one selection to the next?