For simplicity, was looking at 538 aggregate polls in 2016 vs. actual results for my examples on demonstrated uncertainties for a handful of selected states:
https://projects.fivethirtyeight.com/2016-election-forecast/
I'll take demonstrated actual historical error bands over theoretical statistical uncertainty bands (especially when the former is larger!), although my hypothesis on why there may have been such large errors is admittedly purely speculative. 20+ years removed from classwork in statistics, although currently work professionally on risks on energy investment projects including weather uncertainty, regulatory uncertainty, and future price uncertainty among other interesting risks in investment decisions (for wind and solar).
Walking through some potential sources of selection bias on even a "perfect" poll:
Take a hypothetically perfect poll that is a completely unbiased, completely random sample, with large random sample of all registered voters. Even so, this "perfect" sample poll will end up having multiple potential systemic biases:
Selection bias 1: This was from snapshot of registered voters as of a point in time, may be missing recent voter registration
Selection bias 2: This is limited to registered voters that they were able to get a phone number for (most voter registration lists do not typically include phone number - and won't be able to be contacted at all if unlisted phone or no phone)
Selection bias 3: The survey results are limited to subset of people that choose to answer phone call from unknown caller
Selection bias 4: And then further limited to subset that agree to answer poll
Selection bias 5: What time of day do they try to reach the registered voters - e.g. only calling in evening? (potentially under-sampling those that work night shifts)
Recognizing that polls can't obtain a random sample of actual voters is a key issue increasing uncertainty, and all phone polling faces the same hurdles and potential biases. Not that any of those individual particular items may in fact be biased to one party or the other, but they could be and we don't know. Are democrats or republicans more likely to have unlisted phone number or no phone? Are democrats or republicans more likely to refuse to answer phone from unknown caller or refuse to take time to answer poll?