Measuring Survey Error - there are four types of survey error that need to be considered:
Sampling Error relates to the likelihood that the sample is representative of the whole population being studied. Assuming a particular population can be randomly sampled, the error can be expressed as a margin of error of x% at a given confidence level. For example, "if the study was repeated, there is a 95% degree of confidence that the answers would vary by no more than plus or minus 3.5%". This is the type of error which is typically reported in survey findings.
Non-Coverage Error refers to the possibility that not all members of the population being studied have a chance of being sampled. For example, in the population of all US households, there is no complete, up-to-date list available. This is typically not a concern when the population being studied is a magazine's subscribers or an association's membership.
Non-Response Error stems from the fact that it is rare for all members of a sample to respond to a survey. Non-response error is an issue when differences exist between those who respond and those who don't respond. For this reason, researchers try to acheive as high a response rate as the budget for the project allows. Unlike sampling error, there is no statistical formula to quanitify non-response error.
Measurement Error can result from flaws in the survey design and mistakes made by respondents in completing the survey. Careful design with professional oversight, and thorough checking of responses for consistency and completeness can help minimize this problem. Some academic research suggests that self-administered surveys (mail and online) are more reliable than telephone and face-to-face surveys.