Dealing with variance in measurement

Here's a recent Q&A about the first assignment:

On 9/9/2015 1:26 PM, A Eager Student wrote:

First, as partners on this paper, do we submit the same paper, co-written?

Just one paper with both names on it.

Second, if you have several numbers, say 100 measurements of duration, 
what is a good statistical measure of how close these numbers are to each 
other? We thought to use standard deviation but thought that that might be
better suited to looking at how much a single value differed from the mean, 
whereas we want a total measure of variance.

If you think that the underlying population distribution is normal, then you can calculate a standard deviation.

However, the question is whether that is the right way to summarize the information.  If you are looking for the underlying cost of an operation, with as much of the noise removed at possible, then a minimum might be better.

And your first measurement might be much higher, do to caching effects.

Often the best way to display this information is to graph with a box-whisker plot.

--bart