If you have ever used probability or percentage to measure some value then you know that it is very hard to predict how often an event will occur and if it occur at all. In a completely deterministic universe if you somehow could know the initial starting measurement of any object or thing you could predict how the process will unfold and thus predict the outcome. But in real world it is almost impossible to measure every single aspect of any given situation so giving probabilities as a likelihood that one thing would happen is more useful. If you imagine a card game then you never know what are the cards in a shuffled deck but imagine if after playing a card game you see how the dealer takes those cards and shuffles, you see where each card goes and in the end you know what will the next card be and thus have an edge over other players. But imagine if you could see just half of the shuffling actions and partially the other half. In this situation you could assign probabilities to how likely is that a card will come before other cards as well as where are the strongest but you could never be sure because you do not have the complete information.
If we see how statistics is related to probability theory then we can see that in every statistical prediction there is a certain level of probability that the prediction is true but you can never say anything with complete confidence. This is because statistics takes a lot of data, organizes and finds the underlying model and analyses it but this method cannot tell you that the next event that will occur will be that and this method can just guess the likelihood. This means that statistical data and analysis as well as predictions are all just informed probabilities and they will never be 100% accurate.
And if we delve deeper into this subject then there are even more unpredictability as for example in quantum physics everything consists of likelihoods that something will happen but in reality the thing may or may not happen. And that means that any statistical measurements will not be completely accurate as even after a significant sample data there would be accourances when something goes out of control and basically is unpredictable. And this is also true for humans as we are unpredictable creatures and we can change the odds of something occurring in our favor but that is a completely different story.