Stats require a sample from a smaller population that represents the entire population. Phone polls don’t include people a) without phones b) who ignore unknown calls c) who decline to reply to poll calls and d) who lie to poll calls (at least not accurately). I know I’ve been tempted to lie when someone reaches out on behalf of a politician I don’t like, though usually I just ignore the texts.

Exit polls can be similarly affected, though the dynamics are a bit different.