Do most election polls have a higher margin of error than commonly reported?
A 2016 study from the American Statistical Association examining 4,221 late-campaign polls found "the historical margin of error is plus or minus 6 to 7 percentage points" rather than the typically reported plus or minus 3 points that is usually cited as sampling error.
This higher margin of "total survey error" is affected by coverage error, i.e., when not all of a target population has an equal chance of being sampled. For example, phone surveys exclude people without phones. A Pew Research analysis found "politically and civically engaged individuals" were overrepresented in online surveys. Most of the online surveys Pew studied also had a disproportionate number of childless, low-income adults.
The higher margin of error also allows for nonresponse error—which can arise, for example, when supporters of a trailing candidate become less likely to respond to surveys.