MoE is a product of sample size: The bigger your sample, the smaller your Moe in general. But if your sample all came from a non-representational pool, then your results can be way off. You can sample every single member in that pool and get your MoE down to zero, and your results will still be wrong.
I am more concerned with pollers methodology: How are they reaching the people they poll? If they are still using random land-line calls then their methodology is wildly biased. I don't even answer my land-line any more and I think few people under 50 do or don't even have land-lines.
