SPH 3104: Experimental Physics
Lecture 5
• KyU Physics Department
1
Basics of statistics and data analysis
❑ In experiments physics, measurements are done for the purpose of
obtaining the value for a specific quantity.
➢ However, the value by itself is not enough. Determining the value is
half the experiment, and the other half is determining the uncertainty.
➢ Sometimes, the whole purpose of an experiment may be to determine
the uncertainty in the results.
➢ Error and uncertainty are synonymous in experimental physics, but
different concepts.
➢ Error is the deviation from the true value, while uncertainty is an
interval where the true value lies.
2
Sample and parent population
❑ Population is a collection of persons, things or objects under study.
Generally, there is a large number of items in this population. So to gain an
idea about the population, we select a sample (finite number of items)
from the population.
➢ Let’s try to measure something about the population. For example, the
population is all of the 6” hollow concrete blocks manufactured by Master
Industries (U) LTD Banda. This constitutes the parent population.
The quantity which we would like to measure is the weight of each
hollow concrete block. The sample population is the 30 blocks we have
bought.
3
Sample and parent distribution
❑ Imagine a process for manufacturing hollow concrete blocks. Although each
block is nominally the same, processing causes slight deviations in shape, size,
or other mechanical properties. If we measure the weight, say, of an infinite
number of such blocks, these weight measurements will spread into a
distribution around some mean or average value. This hypothetical, infinite
distribution is called the parent distribution. The parent distribution’s spread
depends, obviously, on the precision of the manufacturing process.
➢ We can never measure an infinite number of concrete blocks... Instead, we
measure a smaller subset of blocks, and from this sample, we again find that
our measurements spread into a distribution around the sample mean. This
finite distribution is called the sample distribution.
➢ In the limit of an infinite sample, this distribution should become the parent
distribution (assuming we have no systematic errors).
4
Examples of sample and parent distribution
# of Occurrences
Value of measurement
Figure 2: Sum of these 16
Figure 1: Sample distributions (Histogram) for a measurements (equivalent to 800
series of 16 sets of 50 measurements (also called measurements).
a frequency table).
❑ Note: 800 is close to infinity since the sample distribution now is
quite close to the parent distribution (red line).
5
Mean and standard deviation
❑ Measuring a quantity usually involves statistical fluctuations around
the “true” value.
❑ In fact, multiple measurements in a sample population may have
different values.
❑ Taking an average or mean value cancels the statistical fluctuations
to the first degree.
❑ Hence, the average value or the mean value of a quantity in a
sample population is a good estimate for that quantity.
𝑁
1
𝑥lj = 𝑥𝑖 where we use the notation 𝑥𝑖 = 𝑥𝑖
𝑁
𝑖=1
❑ The mean of the parent population is
1
= lim
N → N
i
x
6
Mean and standard deviation
❑ Although the average value from the sample population is the best estimate,
it is still an estimate of the true value.
❑ We should have another parameter that tells us how close we are to the true
value. This is the variance of the sample:
2
1 2
𝜎 = 𝑥𝑖 − 𝑥ҧ
𝑁
❑ Because the variance is the average of the squares, the square root of the
variance is a better quantity that shows the scatter around the mean value.
The square root of the variance is called the standard deviation:
1 2
𝜎𝑠 = 𝑥𝑖 − 𝑥ҧ
𝑁
7
Mean and standard deviation
❑ The best estimate for the standard deviation of the parent population is
given by
1 2
𝜎𝑝 = 𝑥𝑖 − 𝑥ҧ
𝑁−1
❑ As the number of measurements, N becomes large or as the sample
population approaches the parent population, the standard deviation of
the sample is almost equal to the standard deviation of the parent
population.
8
Mean and standard deviation: Example
You are given the following measurements, s = [110 105 93 125 98], which
represent counts of an event (for example cosmic rays) recorded from a
detector. The counts were recorded in 1-second intervals. Calculate the;
a) Mean of the sample of five measurements
1 1
𝑥lj = 𝑥𝑖 = 110 + 105 + 93 + 125 + 98 = 106.2
𝑁 5
(b) Standard deviation of the sample
1 2 1 2 2 2
𝜎𝑠 = σ 𝑥𝑖 − 𝑥ҧ = 110 − 106.2 + 105 − 106.2 + 93 − 106.2 +⋯
𝑁 5
𝜎𝑠 = 12.4
9
Weighted averages
❑ Sometimes we may measure the same quantity in different sessions. As a
result, we will have different sets of values and uncertainties.
❑ Combining all these sets may achieve a better result with a smaller
uncertainty.
❑ To calculate the overall average, we can assign weight to each value with
the corresponding variance and then calculate the weighted average.
𝑥
σ 2𝑖
𝜎𝑖
𝜇=
1
σ 2
𝜎𝑖
The overall standard deviation is given by;
1
𝜎𝜇 =
1
σ 2
𝜎𝑖
10
Uncertainty and significant figures
Typically, one uses the standard deviation to;
❑ Quantify the uncertainty in a measurement
❑ Determine the number of significant digits.
For experimental data the error in a quantity defines how many figures are
significant.
Assume that a measured value has a numerical value of 12.3456789.
Assume that the standard deviation is 0.99, then the data point has a
value of 12.35 ± 0.99, i.e. probably between 11.36 and 13.34. It
would be wrong to say 12.345 ± 0.99 since the '5' has no meaning. It
would also be wrong to say 12.35 ± 0.987, since the '7' similarly has
no meaning.
In simple cases, errors are specified to one or at most two digits.
Using significant figures is a crude but effective way of reporting errors.
11
Uncertainty and significant figures
What are significant figures?
❑ Significant figures (or digits) are the meaningful digits in a number that contribute
to its precision (Precision is how close repeated measurements are to each other).
Rules for determining significant figures:
❑ Non-zero digits: Always significant
➢ 453 has 3 significant figures
❑ Leading zeros: Not significant; only serve to position the decimal
➢ 0.0052 has 2 significant figures
❑ Captive zeros (between non-zero digits): Always significant
➢ 1003 has 4 significant figures
❑ Trailing zeros:
➢ Without Decimal: Not significant (e.g., 100 has 1 significant figure)
➢ With Decimal: Significant (e.g., 100.0 has 4 significant figures)
12
Uncertainty and significant figures
Calculations with significant figures:
❑ Addition/Subtraction: The result should have the same number of
decimal places as the number with the fewest decimal places
➢ 12.1 + 1.34 = 13.4 (rounded to 1 decimal place)
❑ Multiplication/Division: Result should have the same number of
significant figures as the number with the least significant figures
➢ 4.56 × 1.4 = 6.4 (rounded to 2 significant figures)
Express the following quantities to the correct number of significant
figures;
a) 4.122±3.74
b) 74±7.136
c) 4.782×3.05
d) 3.728/1.6781
13
Error analysis/ propagation of errors
❑ When evaluating the effect of uncertainties on a result, one has to be aware
of the underlying principles of the experiment.
❑ For example, consider an experiment involving a resistor with a 1%
tolerance (a percentage by which the actual resistance value of the resistor
may vary from its stated or nominal value). Say you connect such a resistor
to a battery and measure the current through it, I = V/R.
❑ The resistor’s value can be within 1% of the stated value, so say instead of
100 Ω it is 99 Ω. But repeated experiments with the same resistor will all be
with a 99 Ω resistor. The error is systematic.
❑ However, let’s say you are all experimenting, each with their own 1%-
tolerance resistor. Some will be 99 Ω, some will be 100 Ω, and some will be
101 Ω. The collection of results with that 1%-tolerance resistor now has a
random error of order 1%.
14
Error analysis/ propagation of errors
❑ Sticking with the resistor experiment, all resistors are affected by
temperature. Generally, a hotter resistor has greater resistance than a
colder resistor.
❑ This is yet another source of error. If you measure the resistance to be
99 ohm, and then make repeated measurements at different
temperatures (remember that the very act of passing a current through
the resistor affects its temperature), there is another source of
(essentially) random errors.
❑ Let’s say, as an example, that the temperature effect causes the
resistance to fluctuate with a standard deviation of about 0.5 Ω.
❑ How do we combine the effects of these different types of errors, in the
case of a current measurement: I = V/R?
15
Relative Error (percent error)
❑ Relative error is just the inverse of the signal-to-noise ratio (SNR),
𝜇
= SNR
𝜎
𝜎𝑝
❑ So the relative error is = Relative Error parent distribution
𝜇
𝜎𝑠
= Relative Error (sample)
𝑥ҧ
❑ In our case of the 100 Ω resistor with ±1 Ω tolerance, there is a relative
error of 1
= 0.01 = 1%
100
❑ Which is why we said the resistor has a 1% tolerance.
❑ Our example ± 0.5 Ω temperature effect gives an additional 0.5% error.
16
Relative Error (percent error)
❑ What if we have a 1%-tolerance 10 k Ω resistor? Then the error in
resistance is larger—it is 100 Ω—but the relative error is still the
same: 100
= 0.01 = 1%
10000
❑ Likewise, the random temperature effect causes larger fluctuations
in resistance, ± 50 Ω, but the relative error is still 0.5%.
17
Propagation of errors
❑ What is propagation of errors? It is the effect on a measurement of errors
in one or more parameters affecting the measurement. In our current
measurement V
I= ?
R R
❑ How does the error in resistance affect the current?
❑ Say V = 1.5 volts. We can always calculate using the two extremes
V 1.5 volts
I= = = 14.85 mA
R + R 101 ohm
V 1.5 volts
I= = = 15.15 mA
R − R 99 ohm
❑ So we see that the error is about ± 0.15 mA. What is the relative error?
❑ This is the case for combining measurements from a lot of different 1%-
tolerance resistors, but if we make repeated measurements with a single
resistor, again we will systematically measure either a high or low current.
18
Propagation of errors
❑ What we have done is to examine a simple case of propagation of
error, but the actual case can be more complicated. For example, what
if different measurements are done by students in a lab, using different
resistors and different batteries?
❑ When the measurements of current are combined, there are errors
contributed by both the different voltages of the batteries and different
resistances of the resistors:
𝑉 ± 𝛿𝑉
𝐼=
𝑅 ± 𝛿𝑅
❑ We need a general way to treat such cases so that we do not have to
resort to trying different extremes.
❑ We can use differential calculus to help us. Let’s start with the original
equation, which tells us how the measurements of voltage and
resistance combine to give current: 𝑉
𝐼=
𝑅
19
Method of propagation of errors
V
❑ Start with the original relation, in this case: I = .
R
❑ Use the chain rule:
dV 1 dV dV V
dI = + Vd = + VdR −1 = − 2 dR.
R R R R R
❑ Here, dI represents the deviations of individual measurements:
dI = Ii − I = xi − x
❑ We now square both sides to give the square deviations:
2 2 2
dV V dV V V
( dI ) = ( Ii − I )2 = − 2 dR = + −
2
2 dR 2 dVdR.
3
R R R R R
❑ Finally, average over many measurements:
2 2
dV V V
( dI ) = ( Ii − I ) = + 2 dR − 2
2 2
dVdR
R R
3
R
By definition, 1 2 V2 2 V
the variance = 2 V + 4 R − 2 3 dVdR
2
Covariance term
I
R R R
20
Method of propagation of errors
❑ Notice that we can take:
1 2 V2 2 V
= 2 V + 4 R − 2 3 dVdR
2
I
R R R
❑ And divide through by 𝐼 2 = 𝑉 2 Τ𝑅2 :
Square of relative error in I
I2 V2 R2 1
= + −2 dVdR
I V2 R2 VR
❑ What about this term <dVdR>? This is the product of random
fluctuations in voltage and resistance. When we multiply these
random fluctuations and then average them—they should
average to zero!
❑ Thus, we have the final result:
I V2 R2
= +
I V2 R2
21
Propagation of errors: Exercise
x =u+v x2 = u2 + v2
Show that: x2 u2 v2
x = uv 2
= 2
+
x u v2
u x2 u2 v2
x= 2
= 2
+
v x v2 u
x2 u2 4 v2
x = uv 2 = + 2
x2 u 2 v
Use the CHAIN RULE to determine the
propagation of errors….
22
