Actually, the mere fact you are assuming the sample is random increases the error margin.
If you CONFIRM the sample is random, then the error margin is between 2 and 3 percent, which might be okay in an election sample, but wouldn't be acceptable in a scientific sample.