Obviously if you use one number to characterize a distribution, it will “overstate” some samples and “understate” others.
If the distribution is reasonably close to a normal (Gaussian) distribution, reporting the mean and the standard deviation will be appropriate and sufficient.
If the data contains some outlier points that skew the mean, the median is a good alternative. If 10 people lost 2 pounds and 1 person gained 20 pounds, the mean is a weight less of zero, but the median is 2 pounds.
Well, we see this kind of thing happen 15 to 20 to 80% of the time. 
If the number of people/cases/examples isn’t huge, and if the ‘raw data’ isn’t massive, you can take a number of people who most closely resemble a given subject and do a basic average, while realizing averages are composed of a range… and if 30 is the high end and 10 is the low end, the most you can hope for is somewhere in the range… if we’re talking about weight loss.
.
[QUOTE=Blue Blistering Barnacle]
Well, we see this kind of thing happen 15 to 20 to 80% of the time.
[/QUOTE]
Huh? Who is seeing what happen?
Just a silly comment in response to the 5% bullshit comment. We would sometimes say this when we were pulling “statistics” out of thin air.