Regarding polls. I wonder how many people do not answer their phone if they see a number they

[Silent3]

77. What's more important is if there's a systematic bias behind answering/not answering

If Biden voters and Trump voters are equally unlikely to answer the phone, and equally likely or unlikely to respond truthfully if they do answer, then it really doesn't matter how many people answer. Just keep calling and calling until you hit a reasonable sample size of people who answer, and you'll still be getting pretty good data.

It only gets tricky if, say, Biden voters hate answering phone calls from strangers more than Trump voters do. Or if Trump voters like fucking with pollsters more than Biden voters do (which is easy to imagine in one way, if you think about a general propensity for lying, but tougher to imagine if you think of a Trump supporter trying to choke out the words, "I'm for Biden!" ).

There might be some systemic bias, such as Trump voters being older, more of them being home more of the time, and therefore more likely to answer the phone. If you are aware of such a bias, however, you can try, with varying degrees of success, to factor that into your model.