In the derivation of unbiased sample variance, it is considered that $X_i$ are iid random variables while $X_i$ actually represents a sample from a population. So my question is that shouldn't we consider $X_i$ to be an instance of same random variable whose probability distribution is probability of selecting $X_i$ ?
For example, consider I have population of people with different heights and I select a sample of people from this population. This sampling of people can be thought of a repetitive procedure of sampling a value from a random variable i.e uniform r.v. Isn't it so?
Also I have generally seen it that when we consider samples of a stochastic process, we model each sample as i.i.d instead of repetitive sampling (taking multiple values) from the stochastic process. For example in the definition of strict sense stochastic process we say that the joint distribution of different samples of $X_t$ will be same whether we sample it at any time. Here my question will be rephrased as how can we have joint distribution of constant numbers? Like if I have a series of dice-face-numbers then there is no meaning of considering joint probability distribution of these numbers?