SAMPLING ERROR

It is important to remember that a survey is merely a sample of the U.S. population, and as such can provide only an estimate which is subject to sampling error. Generally, larger sample sizes result in smaller random errors.

If we were to conduct a census of every single person in the U.S., our sampling error, theoretically, would be zero. Leaving aside other types of non-random error and bias, we could say that there are in fact 36,152,000 runners in the U.S. because we actually counted them.

But since practical considerations limit us to a survey in which each respondent represents about 17,341 people in the population, any projected number (in thousands of people) is merely an estimate which is subject to varying degrees of sampling error.

In the case of running, our estimate of 36,152,000 runners has a sampling tolerance of +/-3.9% at the 95% Confidence Level. This means that if this same survey were replicated on the same day by drawing 100 other population samples of similar size, in 95 out of 100 cases, the estimate for the projected number of runners in the U.S. would vary by not more than 3.9% or +/-1,410,000. In other words, the odds are 19-1 that there are somewhere between 34,742,000 and 37,562,000 runners in the U.S.