I would agree if the sample size would be much smaller. However at 900 voters and looking at previous polls results AND the significant majority result (we are not in a 45%-55% bracket), the results are absolutely relevant.
You cannot use the "biased population" argument in this case as no vote manipulation seemed to have occurred (multiple account created for purpose of answering, call for "ballot stuffing" from other platform/site...).
The time frame and the limited time of poll being open were random, no significant event (Private Group invasion for example) could have influenced as well the voters.
I can guarantee that the population who voted was Random in statistical term. If you have 900 random persons as a sample size, participating in such poll, the results will be indeed correct within a 3% Error Margin considering the relatively small player base.
If it's Random, it is representative by nature. Remember, I assumed a 95% Confidence Level, which is a safe/conservative assumption considering previous similar polls.
You cannot guarantee that people that use the forum are random though. if the breakdown ingame for pvp only/pvp and pve/pve only was 20/30/50 and the breakdown of forum users was 10/15/75 then it is not a representative sample and the 95% confidence rules break down. The simple fact is we don't know what the breakdown is and therefore cant make assumptions that the poll is not biassed. To be valid it is not enough for a poll to be random it has to also be a representative sample of the demographic that you are inferring results for