Statistical Errors in Studies

When working with statistical sampling (e.g, public opinion surveys, studies of illness recovery due to medication,...) there is an error range caused by the fact that one is sampling at random. More details are provided in our Statistics Tutorial.

For example, in a public opinion survey of 600 people, you might encounter 320 Democrats and 280 Republicans. If you repeat the same survey with another group of 600 people, you might encounter 283 Democrats and 317 Republicans.

If we toss a coin 100 times, "most likely" (95% of the time!) we will get between 40 and 60 "heads". This range (40 to 60) is called "the 95% Confidence Level (CL) range" - it means that if we repeat the tossing of 100 coins again and again, 95% of the time we will get between 40 and 60 "heads". In medical studies, this is typically used as an estimate of the range in which the "true" result will be. We will also refer to this range (interval) as the "Margin of Error".

If we get the "true" result only 95% of the time, this means that 5% of the time (in 1 out of 20 experiments) the true result will be outside the quoted range - the quoted result of the experiment is wrong! This is an additional significant source of errors, in addition to the systematic error sources, listed by Dr. Shmerling.


If you benefitted from this site, please consider supporting us.


Back to previous page


Last Modification - November 20, 2005