Experiments
When conducting experiments, it is almost always a good idea to take
replicate measurements at the same conditions. The exact number will
depend on the nature of the experiment, but you will usually want at least
three replicates.
Data taken in real systems will yield different results each time, even at the
same operating conditions.
How do we deal with these differences in the data?
By making use of appropriate statistics.
1
Useful statistics in uncertainty analysis
For repeated measurements, a useful statistical measure is the standard
deviation, it shows how much variation or "dispersion" exists from the
average (mean, or expected value).
The mean of the data values, µ, is given by
N
1
µ=
N
∑x
i =1
i , where N is the number of data points
The standard deviation, s, is then
1 N
s = s= ∑
N − 1 i =1
( xi − µ ) 2 (‘corrected’ sample standard deviation)
However, sometimes the standard deviation is define as follows:
N
1
s=
N s=
N
N
∑ (x − µ)
i =1
i
2
(‘uncorrected’ sample standard deviation)
2
Exercise
The following data points were taken on a system.
For either temperature or pressure, compute the mean and standard
deviation of the replicate data points.
Temperature (C) Pressure (kPa)
replicate 1 27 211
replicate 2 25 203
replicate 3 31 180
replicate 4 22 192
mean, µ ? ?
standard deviation, s ? ?
3
Exercise: Solution
Temperature (C) Pressure (kPa)
replicate 1 27 211
replicate 2 25 203
replicate 3 31 180
replicate 4 22 192
mean, µ 26.25 196.5
standard deviation, s 3.77 13.5
For the temperature:
N
1 1
µ=
N
∑ x=i
i =1 4
( 27 + 25 + 31 + 22=) 26.25 C
1 N
=σ ∑
N − 1 i =1
( xi − µ ) 2
1
= ( 27 − 26.25 ) + ( 25 − 26.25 ) + ( 31 − 26.25 ) + ( 22 − 26.25 )
2 2 2 2
4 −1
= 3.77
4
Uncertainties on plots
Error bars are a common way of representing uncertainty in repeated
measurements. The point on the plot is usually the mean value of the data
and the length of the bar from this point is typically the standard deviation in
the data (though it can represent the standard error of the mean).
In general, you will always want to show error bars in your plots of
measured data.
5
Here are some examples of plots with error bars from scientific papers:
The Effect of Fruit and Vegetable Intake on Risk for Coronary Heart
Disease (Ann Intern Med. 19 June 2001;134(12):1106-1114)
What conclusions can you draw from this plot?
6
Animal manure is applied to agricultural land as a means to provide crop nutrients.
However, animal manure often contains antibiotics as a result of extensive therapeutic
and sub-therapeutic use in livestock production.
What conclusions can you draw from this plot?
7
Without even knowing the specifics of the study, what conclusions can be
drawn from this plot?
How significant are the differences between the controls, and the BEB and
AB treatments?
8
Propagation of uncertainties
When you combine (add, subtract, multiple, divide) measurements with
uncertainties, the uncertainties ‘propagate’.
The uncertainty of the ‘combined’ or calculated quantity will always be
larger than that of any of the individual uncertainties.
9
Rule for the adding or subtracting quantities with uncertainties:
q =x + + z − (u + + w)
δq
= (δ x ) + + (δ z ) + (δ u ) + + (δ w )
2 2 2 2
example:
q = (3.2 ± 0.1) – (5.21 ± 0.03) – (17.2 ± 0.8)
q = (3.2 – 5.21 – 17.2) ± ?
q = -19.21 ± ?
δq = ( 0.1) + ( 0.03) + ( 0.8) = 0.8086
2 2 2
Therefore, the answer, using appropriate significant figures, is
q = -19.2 ± 0.8
10
Rule for the multiplying or dividing quantities with uncertainties:
x × × z
q=
u × × w
δq δx δ z δu δw
2 2 2 2
= + + + + +
q x z u w
example:
q = (3.2 ± 0.1) x (5.21 ± 0.03)/(17.2 ± 0.8)
q = 3.2 x 5.21 / 17.2 ± ?
q = 0.9693 ± ?
δq
2 2 2
0.1 0.03 0.8
= + + = 0.0563
q 3.2 5.21 17.2
δq
= 0.0563
0.9693
δ q = 0.0546
Therefore, the answer, using appropriate significant figures, is
q = 0.97 ± 0.05
11
Other rules
Other rules for powers, exponents, special functions, etc. of numbers with
uncertainties are contained in most books on uncertainty analysis.
12
Software used for error analysis: Demo
It is essential that you know the mathematics behind uncertainty
propagation before using software for these calculations.
There is a large list of software packages for uncertainty analysis on
Wikipedia:
https://en.wikipedia.org/wiki/List_of_uncertainty_propagation_software
In addition, other examples are as follows:
• MATLAB: ‘Propagation of Uncertainty’ package
https://www.mathworks.com/matlabcentral/fileexchange/17901-propagation-of-
uncertainty
• python: ‘uncertainties‘ module
http://packages.python.org/uncertainties/
• Abacus
http://www.eecs.berkeley.edu/~megens/abacus/abacus_description.html
13
