Outside the margin of journalistic error
M. E. Sprengelmeyer and Alan Gathright of the Rocky Mountain News write:
Tancredo has been polling below the margin of error in polls around the country.
The "margin of error" we often hear quoted for polls is the range of sampling error for answer options that one expects about 50% of those surveyed to choose, e.g., as in a question on the general election preference between the nominees of the two major parties. In a poll with 500 respondents, at the 95% confidence level, a question where we expect a 50%-50% response split has a +/- 4.4% margin of error. However, if to the same question there is a third response we expect to be chosen 1% of the time, the margin of error on the frequency of that response would be +/- 0.9%. In other words, say Giuliani is polling at 50%; assuming a random sample, the "true" figure in the population is probably between 45% and 55%. If in the same question Tancredo receives 1%, the "true" figure is likely no higher than 2%.
It may not be worth it to a political reporter to gain the thorough understanding of survey research that would enable him or her to get such things correct; oddly, it is still his or her job to write about such things.
Point taken.
I think the "less than 2%" approach is right for a candidate who polls 1% -
that perfectly incorporates the lack of precision, i.e., sampling error.
If the latest poll result were 2%, then the range is essentially 1% to 3%;
of course, if you have twenty recent polls in which he never actually
registers the 3% but often registers 1%, you would probably want to say 1%
to 2%.
Thanks, Jason. Best of luck to you and all the number crunchers here.