Web Survey Bibliography
As already known in 1924, the necessary and sufficient condition for a representative measurement is that the measurement should be independent of the sampling method. This independence condition allows a survey to provide representative values for some responses but not for others. This difference in representativeness explains why survey researchers us a telephone survey for representative responses in politics and health but not for responses to "Why don't you use the telephone?"
This paper proposes a metric called measurement dependency based on the independence condition for quantifying the extent to which different responses have different degrees of representativeness.
The metric can be computed using data from existing surveys. The dependency metric can quantify how much a question like email usage is biased by a method like an Internet survey. This paper further proposes that large numbers of dependencies can be collected by survey researchers routinely adding to surveys just one question on a sampling methodology.
The ease of measuring response dependencies will allow the survey community to work together with minimal effort to compile modern electronic databases for better choices of survey methods for desired responses.
Journal Homepage (abstract) / (full text)