Introduction to experimental

errors
J W Cockburn
Department of Physics and Astronomy
University of Sheffield
Physics & Astronomy Laboratories
Outline of talk
Why are errors important?
Types of error random and systematic
(precision and accuracy)
Estimating errors
Quoting results and errors
Treatment of errors in formulae
Combining random and systematic errors
The statistical nature of errors
Why are errors important?
Two measurements of body temperature before
and after a drug is administered

38.2C and 38.4C

Is temperature rise significant? It depends on the
associated errors

(38.20.01)C and (38.4 0.01)C - significant

(38.20.5)C and (38.4 0.5)C not significant


Random errors
An error that varies between successive
measurements
Equally likely to be positive or negative
Always present in an experiment
Presence obvious from distribution of values
obtained
Can be minimised by performing multiple
measurements of the same quantity or by
measuring one quantity as function of second
quantity and performing a straight line fit of the
data
Sometimes referred to as reading errors
Systematic errors
Constant throughout a set of readings.
May result from equipment which is
incorrectly calibrated or how
measurements are performed.
Cause average (mean) of measured
values to depart from correct value.
Difficult to spot presence of systematic
errors in an experiment.
Random vs systematic errors
True value
Random errors only
Random + systematic
A result is said to be accurate if it is relatively free
from systematic error
A result is said to be precise if the random error is
small
Quoting results and errors
Generally state error to one significant
figure (although if one or two then two
significant figures may be used).
Quote result to same significance as error
When using scientific notation, quote value
and error with the same exponent
Value 44, error 5 445
Value 128, error 32 13030
Value 4.8x10
-3
, error 7x10
-4

(4.80.7)x10
-3

Value 1092, error 56 109060
Value 1092, error 14 109214
Value 12.345, error 0.35 12.30.4


Dont over quote results to a level inconsistent
with the error 36.6789353720.5

Quoting results and errors

Estimating reading errors 1
Oscilloscope related to width of trace
3.8 divisions @ 1V/division = 3.8V
Trace width is ~0.1 division = 0.1V
(3.80.1)V
Digital meter error taken as 5 in
next significant figure
(3.3600.005)V
Estimating reading errors 2
Analogue meter error related to width of pointer
Value is 3.25V
Pointer has width 0.1V
(3.30.1)V
Estimating reading errors 3
Estimating reading errors 4
16 17
Linear scale (e.g. a ruler)
Need to estimate precision with which measurement
can be made
May be a subjective choice
16.770.02
Estimating reading errors 5
16 17
The reading error may be dependent on what is
being measured.
In this case the use of greater precision equipment
may not help reduce the error.
16 17
Treatment of errors in formulae
In general we will calculate a result using a
formula which has as an input one or more
measured values.
For example: volume of a cylinder
How do the errors in the measured values
feed through into the final result?

2
A r h t =
Treatment of errors in formulae
In the following A, B, C and Z are the absolute
values
AA, AB, AC and AZ are the absolute errors in A,
B, C and Z
Hence AA/A is the fractional error in A and
(AA/A)100 is the percentage error in A etc
AA and A will have the same units
Assume errors in numerical or physical
constants (e.g. t, e, c etc) are much smaller
than those in measured values hence can be
ignored.

( ) ( ) ( ) ( )
2 2 2 2
2 2 2 2
2 2 2 2
/
ln
exp
n m l
Z A B C Z A B C
Z A B C
Z ABC or AB Cetc
Z A B C
Z A B C
Z A B C n m l
Z A B C
A
Z A Z
A
Z
Z A A
Z
= A = A + A + A
A A A A
| | | | | | | |
= = + +
| | | |
\ . \ . \ . \ .
A A A A
| | | | | | | |
= = + +
| | | |
\ . \ . \ . \ .
A
= A =
A
= = A
Treatment of errors in formulae
Example of error manipulation 1
2
A r t =
Where r=(50.5)m
A=78.5398m
2

2 2
2
2 2
0.5
0.1
5
2 (2 0.1) 0.04
0.2 0.2 0.2 (78.5398 ) 16
r
r
A r
A r
A
hence A A A m m
A
A
= =
A A
| | | |
= = =
| |
\ . \ .
A
= A = A = =
Hence final result is A=(7916)m
2
P=2L+2W where L=(40.2)m and
W=(50.2)m
P=18m



P=(18.00.3)m
( )
2
2 2
(0.2) (0.2) 0.08
0.28
P
P m
A = + =
A =
Example of error manipulation 2


2
l
g
t t =
l=(2.50.1)m, g=(9.80.2)ms
-2
t=3.1735s
2
2 2
1 1
2 2
l g
l g
t
t
| | A A A
| | | |
= + =
| | |
\ . \ .
\ .
2 2
4
1 0.1 1 0.2
5.04 10
2 2.5 2 9.8

| | | |
+ =
| |
\ . \ .
At/t=0.022 hence At=0.022x3.1735=0.070
t=(3.170.07)s
Example of error manipulation 3
Random + systematic errors

Combine random error and systematic error (if known) by
adding the squares of the separate errors.

Example: A length is measured with a reading (random
error) given by (892) cm using a rule of calibration
accuracy 2%.
( )
2
2
2
2
2
( )
( )
2
(0.02) 0.000905
89
0.03
Total fractional error fractional reading error
systematicerror
Total fractional error
=
+
| |
= + =
|
\ .
=
Absolute error =
0.03x89=2.7cm

Value =(893)cm
The statistical nature of errors
Because of the way in which errors are
combined to generate the total error this
does not give the maximum possible
range of values.
Instead the total error associated with a
value provides information concerning the
probability that the value falls within
certain limits.
The statistical nature of errors
If a quantity o has an associated error Ao then
There is a 67% chance that the true value lies
within the range o-Ao to o+Ao
There is a 95% chance that the true value lies
within the range o-2Ao to o+2Ao
P
r
o
b
a
b
i
l
i
t
y


o
f

r
e
s
u
l
t



o o
2Ao
2Ao


Comparing values

Need to look at overlap of distributions
Case of two quantities o
A
and o
B
which differ by
sum of errors Ao
A
+Ao
B




Probability of agreement ~2x1/36 = 6%
Conclusions
Systematic and random (reading) errors
accuracy and precision
Quoting errors
Estimating reading errors
Manipulating and combining errors
The statistical nature of errors


Further reading: Document on website or any text
book on practical physics e.g. Experimental Methods
L Kirkup or Practical Physics G L Squires