When is the earliest reliable opinion polls could be invented with a POD of 1900?
They didn't get the kinks worked out until the 1950s, most spectacularly shown in 1936 and 1948. (Gallup for one had a massive inbuilt GOP bias that put Ras to shame) My question is how to do that earlier.
It's possible to conduct scientific polling as soon as regular, cheap postal service becomes available, as long as you 1) understand sampling techniques and 2) have the resources to print the questionnaires, mail them, and study the responses.
The problem though with mail-in polling is that the sample which responds is smaller than that sought out. It's a biased sample because the only respondents are those who care enough to fill out the questionnaire and mail it back, possibly at their own expense.
Selection bias is inherent in opinion polling - and Gallup used postage-paid envelopes, for one thing. It was less of a problem in the first half of the 20th century, when mail was the primary form of long-distance communication for most Americans. Response rates to mail contacts were greater then than today.
I agree that mail is a horrible way to build a sample today. There is one newspaper which does a famous mail-only poll, but the name is escaping me.
The sampling bias would also reflect who it is that is mailed, and who at the address fills out the survey. The ANES has arguably the most reliable, but also the most costly, approach.
I'm not sure if ANES can be replicated in the OP's scenario. In polling or in GOTV, face to face is always the best, but it's also the most resource-intensive.