He simply has no clue about the term or concept of "probability". He confuses the observed realization/outcome of a random variable (which for the case of a binary variable is either 0 or 1, win or lose, 100% or 0%, etc.) with the probability that this random variable will take on a particular value before it is realized/before the "random experiment is carried out".
Simple example: A flipped (nonrigged) coin will either be heads or tails (or land on the side of the coin), but before you flip the coin, you dont know which outcome you will get.
Same story with elections: at the booth, each voter will vote for precisely one of the candidates on the ballot (if he doesnt fuck up...), but we as the bystanders dont know beforehand which candidate will receive how many votes/win the particular race. This obviously doesnt mean that we are entirely clueless about the outcome - for example, even in absence of further information, we know that the GOP candidate is more likely to win a race in rural Alabama than the Democratic candidate.
Probabilities are, in a certain sense, the language with which these differences are formalized and expressed.
And finally, statistical election models try to produce estimated probabilities which are as close to the underlying true but unknown probability as possible. They do so by looking at historical experience about how elections play out coupled with recent polling data for the particular race.
This post was edited by Black XistenZ on Jul 25 2019 09:10am