When you're dealing with tiny percentages you'd need a much bigger sample to be at all reliable.

What's not clear is whether these figures are based on what candidate people prefer, and how far on the likelihood of people actually voting for the candidate they prefer.