Uncertainty in Measurement and Calculation
Consider a sponge cut into a block of length, width and height. If you were to make several measurements of the length with a precise electronic vernier caliper, you would have a distribution of values that represented the roughness and unevenness of the surfaces at each end. The results would also depend upon the pressure that you applied to the sponge and may vary with other variables such as mass of absorbed  water. The length itself is not perfectly defined and has an uncertainty which you determine by repeated measurements at different positions and pressures etc.

The mean of a set of two or more measurements is their sum divided by the number of measurements: 

Lave = Sum(Li)/N
We define the standard deviation or uncertainty of a set of two or more measurements to be the square root of the average square deviation from the mean
sL = (Sum(L-Lave)2/N)1/2
or root mean square deviation from the mean. In student labs we often take two independent measurements and call the uncertainty half the difference. If the number of measurements, N, were much larger than two, the standard deviation of the mean would be smaller than than that of individual measurements by the square root of N-1. Thus the mean of one hundred measurements is about ten times more accurate than any one of them and a single measurement has infinite uncertainy. With this many measurements you might keep two significant figures in the uncertainty rather than one.

If the block above were made of precision machined steel and polished square within a wavelength of light and measured with a meter stick of smallest increment equal to one millimeter, then you might expect repeated measments to all be the same. But the uncertainty is not zero as predicted by the the equation above since the instrument is limited in its accuracy by the smallest division, or least count. The human eye can easily resolve a tenth of a millimeter and we usually estimate the next decimal place. The total uncertainty is the sum of the uncertainties due to the object, the instrument and the observer. With all these sources of uncertainty, a good measurement requires many observations which vary over the object, the instrument and different observers.

What if we wanted to know the volume of the sponge above after carefully measuring each dimension? Since uncertainties are usually small, we can differentiate the equation for volume

V = LWH
to tell us what its uncertainty is due to the three measurements. The total differential change in V is the sum of contributions from all measured variables:
dV = Sum( (dV/dx) dx)
and for the sponge above
dV = WH dL + LH dW + LW dH
by the product rule. You can envision these little volumes as extensions to the original volume in each of the three axes. In this case we can divide the differential volume by the volume itself and find that
dV/V = dL/L + dW/W + dH/H.
The total percentage uncertainty in the volume is the sum of the percentage uncertainties in each measurement. This equation should be modified due to the fact that uncertainties are random and not in the same direction at the same time. The correct expression for total percentage uncertainty is the hypotenuse of the three percents, or square root of the sum of the squares. Since we usually keep only one significant figure in uncertainty, corresponding to the last significant figure of the mean, we usually just keep only the largest percentage uncertainty and drop the others as negligible. You should check your understanding of the propogation of uncertainties by showing that the percentage uncertainty in the volume of a sphere is three times the percentage uncertainty of its measured diameter.