| First, the margin of error applies to the change, not the spread. In other words, on 50- 50, a three percentage points change yields a six percentage point spread, i.e. 53- 47. Second, a lot of polls exceed the plus or minus three percent (it is a function of sample size), sometimes reaching as much as six percent, or yielding a twelve point spread. Third, in any given number of polls, there will be "outliers" which exceed the designated margin of error as mere statistical anomalies. The outliers can be distinguished from true readings ordinarily by their infrequency. Fourth, there are different ways that the various pollsters do things that might or might not be a result of bias, for example, reweighting results for demographic balance, particularly party affiliation, if they think that the sampling was skewed; deciding how to determine "likely voters"; and ways of wording questions that may elicit different responses even though the questions are practically identical. |
