國立臺灣師範大學普通物理實驗講義編輯群

Processing of experimental data

Introduction to Uncertainty
1. uncertainty (of measurement)

The Definition of Uncertainty：

A parameter related to the measurement result, used to describe the dispersion
characteristics of the measured quantity. (Note1)
Uncertainty means doubt. Widely speaking, it is the degree of doubt about the
applicability of the measurement result. "Uncertainty" is often used in two ways, one is the
uncertainty concept, and the other is a quantitative measurement of the uncertainty
concept. Adjectives describe other special considerations. The article does not use the word
"error", mainly to avoid confusion with concepts such as errors or deviations [2],
“Inaccuracy (Error)” is defined as the difference between an observed value and the true
value. And it is approximately equal to the difference between observed value and
expected value (average value).
The formal definition of "measurement uncertainty" is: Usually, we do not know the
“true value”. We may get approximately “true value”, from earlier experiments data and
application of uncertainty analysis. Sometimes the accepted value can be regarded as the
true value.

2. Classification of Uncertainty：
According to the properties of uncertainty, it is generally divided into two types:
Systematic uncertainty and Random uncertainty.
A. Systematic uncertainty:
After a physical quantity is repeatedly measured many times, the result is
consistent, but there is still a certain deviation from the true value. We call there
is a systematic uncertainty in this experiment. Uncertainty of this type may result
from faulty calibration of equipment, experimental environment changes (ex:
temperature changes cause the original calibration value to be unsuitable), or from
bias on the part of the observer, etc. The characteristic of the system uncertainty is
“one-way certainty”, i.e. the measured value is either high or low, and the
deviation is the same in substance. Otherwise, there is no probability distribution.
Therefore, it cannot be reduced by increasing the number of measurements. In

1
(註 ): 陸健榮，量測不確定度表示法之國際標準概要，(The essentials of the international standards of the
expression of uncertainty in measurement)
U-1
 國立臺灣師範大學普通物理實驗講義編輯群

theory, we can completely eliminate the system uncertainty. An experiment

should be devoted to understanding and eliminating sources of systematic
uncertainty. It is a test of experimental ability.
B. Random uncertainty:
This type of uncertainty is caused by uncertain factors. The measured value
will fluctuate up and down, and the deviation from the true value may be positive
or negative, large or small, and there is no certain rule for the timing of the
uncertainty. The characteristic of random uncertainty is its probability. Under the
same experimental conditions, if the number of measurements is large enough,
the uncertainty distribution will generally conform to certain statistical
distribution functions. The most common function is the Normal Distribution
Function.
The purpose of uncertainty analysis is to extract useful data from a limited
number of measurements and to estimate the reliability and range of reliability of
the results.

3. Accuracy vs Precision
Accuracy: Indicates the degree of deviation between the measured value and the true
value (or accepted value). If the deviation is small, the accuracy is high. This is very
important in the design of the instrument, but here it is directly determined by the
instrument.
Precision: Represents the results of multiple repeated measurements, and the degree
of deviation between each other. That is a measure of the reproducibility of the result in a
given experiment. If the results of repeated measurements are similar, the precision of the
experimental measurement is high.
The high precision of an experimental measurement does not mean that the result of
the experiment is accurate. Although the random uncertainty is reduced, it is still possible
to hide the system uncertainty. Therefore, to evaluate the pros and cons of an experiment,
its accuracy and precision must be considered at the same time.

4. Significant Digits
An experimental result is presented by the Significant Digits faithfully. It consists of a
set of reliable numbers and an estimated number (or dubious number). Taking the
measurement of the length of a wooden pole as an example, the measured pole length is
123.0 cm, of which 1, 2, and 3 are reliable numbers, and the last number 0 is an estimated
number, so there are four significant figures. The four arithmetic operations of two
significant figures are illustrated as follows:

U-2
 國立臺灣師範大學普通物理實驗講義編輯群

A. Addition and subtraction

123.0 + 12 = 135 123.0 − 121.5 = 1.5

B. Multiply and divide

123.0 × 12 = 1500 or 1.5 × 103 123.0 ÷ 12 = 10

The above is the rigorous method of handling significant digits based on the results of
the elementary arithmetic. When multiplying or dividing two numbers, sometimes for a
quick calculate, the significant digits of the product or quotient obtained can be equal to the
smaller of the two numbers.

5. Mean and Standard Deviation

A. Parent Population and Parent Distribution; Sample and Sample Distribution
In the experimental measurement of the same physical quantity, the
collection of an infinite number of measurements is called the Parent Population.
In the Parent Population, the probability distribution of each different
measurement value is called the Parent Distribution.
B. Parent Population parameters：
1
i. Average：𝜇 ≡ lim ( ∑𝑖 𝑥𝑖 ) The average value of the Parent Population is
𝑁→∞ 𝑁

the true value.

ii. Deviation ：𝑑𝑖 = 𝑥𝑖 − 𝜇 。

1
iii. Variance：𝜎 2 ≡ lim { ∑𝑖(𝑥𝑖 − 𝜇)2 }。
𝑁→∞ 𝑁

1
iv. Standard Deviation：σ = √𝜎 2 ≡ √ lim { ∑𝑖(𝑥𝑖 − 𝜇)2 }。
𝑁→∞ 𝑁

U-3
 國立臺灣師範大學普通物理實驗講義編輯群

C. Sample parameter
In actual experiments, we can only make a limited number of measurements,
so the Parent Population and Parent Distribution are hypothetical. The collection
and probability distribution of a finite number of measurements are called sample
group and sample distribution, respectively. The sample group is a subset of the
parent group.
1
i. Average：𝑥̅ ≡ ∑𝑖 𝑥𝑖 。
𝑁

∑𝑖(𝑥𝑖 −𝑥̅ )2
ii. Standard Deviation:s ≡ √ 。
𝑁−1

For a set of N measurement values, the sum of squared deviations of the

sample average from each measurement value is the smallest. Therefore, the sample
average can be considered as the best estimate of the population average μ, that is to say
∑𝑖 𝑥𝑖
μ ≈ 𝑥̅ = 。
𝑁

Proof：
Let D = ∑𝑖(𝑥𝑖 − 𝑥̅ )2 = ∑𝑖 𝑥𝑖2 − 2𝑥̅ ∑𝑖 𝑥𝑖 + 𝑁𝑥̅ 2

𝑇𝑎𝑘𝑒 𝑡ℎ𝑒 𝑚𝑖𝑛𝑖𝑚𝑢𝑚 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝐷 𝑑𝐷 𝑑

⇒ = [∑(𝑥𝑖 − 𝑥̅ )2 ]
𝑑𝑥̅ 𝑑𝑥̅
𝑖

= −2 ∑ 𝑥𝑖 + 2𝑁𝑥̅ = 0
𝑖
∑𝑖 𝑥𝑖
⇒ 𝑥̅ =
𝑁
Similarly, the sample standard deviation s can be considered as the best estimate of
the Parent Population standard deviation σ, that is
∑𝑖(𝑥𝑖 −𝑥̅ )2
σ≈s=√ 。
𝑁−1

U-4
 國立臺灣師範大學普通物理實驗講義編輯群

6. Normal or Gaussian Distribution

1 1 𝑥−𝜇 2
P= 𝑒𝑥𝑝 [− ( ) ]
𝜎√2𝜋 2 𝜎

The probability that the measured value appears in the range of μ＋σ > x > μ − σis
68.3%,
The probability that the measured value appears in the range of μ＋2σ > x > μ − 2σis
95.4 %,
The probability that the measured value appears in the range of μ＋3σ > x > μ − 3σis
99.7 %.

7. Representation of measurement results

The measurement result is usually written in the form of 𝑋̅ ± ∆X. X represents the
measured value, and ∆X represents the uncertainty. ∆X is a function of the statistical
uncertainty (average standard deviation 𝜎𝑋̅ ) and instrument uncertainty. The formula is
simply expressed as follows:
2
1
√ 2
ΔX = (𝜎𝑋̅ ) + ( 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 𝑠𝑐𝑎𝑙𝑒)
2 √3

8. Propagation of Uncertainty
Often a physical quantity is obtained by measuring several components and then
calculating. For example, the volume is obtained by multiplying the three components of
length, width, and height. Therefore, the uncertainty of each component will also be passed
into the final result. Assuming that X represents a certain physical quantity, it is a function
of components such as u, v...,
X = 𝑓 (𝑢, 𝑣, … )
Let 𝑢̅, 𝑣̅ , …represent the sample average of 𝑢, 𝑣,… components,then the
average value of X can be written as
𝑋̅ = ∑ 𝑓 (𝑢𝑖 , 𝑣𝑖 , … )⁄𝑁
𝑖
If f is a linear function, then the average value of X can also be written as
𝑋̅ = 𝑓 (𝑢̅, 𝑣̅ , … )
U-5
 國立臺灣師範大學普通物理實驗講義編輯群

Let {𝑢𝑖 , 𝑣𝑖 , … } be the sample data of the i-th component, then the corresponding 𝑋𝑖 is
𝑋𝑖 = 𝑓(𝑢𝑖 , 𝑣𝑖 , … )
ThenDeviation 𝑋𝑖 − 𝑋̅ = 𝑓(𝑢𝑖 , 𝑣𝑖 , … ) − 𝑓 (𝑢̅, 𝑣̅ , … )
𝜕𝑓 𝜕𝑓
= [𝑓 (𝑢̅, 𝑣̅ , … ) + ( ) (𝑢𝑖 − 𝑢̅) + ( ) (𝑣𝑖 − 𝑣̅ ) + ⋯ ] − 𝑓(𝑢̅, 𝑣̅ , … )
𝜕𝑢 𝑢̅ 𝜕𝑣 𝑣̅
𝜕𝑓 𝜕𝑓
= ( ) (𝑢𝑖 − 𝑢̅) + ( ) (𝑣𝑖 − 𝑣̅ ) + ⋯
𝜕𝑢 𝑢̅ 𝜕𝑣 𝑣̅
𝜕𝑋 𝜕𝑋
= ( ) (𝑢𝑖 − 𝑢̅) + ( ) (𝑣𝑖 − 𝑣̅ ) + ⋯
𝜕𝑢 𝑢̅ 𝜕𝑣 𝑣̅

⇒ 𝜎𝑥2 = lim ∑ (𝑋𝑖 − 𝑋̅ )2 ⁄𝑁

𝑁→∞ 𝑖
2
1 𝜕𝑋 𝜕𝑋
( ) ( )
= lim ∑ [( ) 𝑢𝑖 − 𝑢̅ + ( ) 𝑣𝑖 − 𝑣̅ + ⋯ ]
𝑁→∞ 𝑁 𝜕𝑢 𝑢̅ 𝜕𝑣 𝑣̅
𝑖

1 2
𝜕𝑋 2 2
𝜕𝑋 2
= lim ∑ [(𝑢𝑖 − 𝑢̅) ( ) + (𝑣𝑖 − 𝑣̅ ) ( )
𝑁→∞ 𝑁 𝜕𝑢 𝑢̅ 𝜕𝑣 𝑣̅
𝑖

𝜕𝑋 𝜕𝑋
+ 2(𝑢𝑖 − 𝑢̅)(𝑣𝑖 − 𝑣̅ ) ( ) ( ) + ⋯ ]
𝜕𝑢 𝑢̅ 𝜕𝑣 𝑣̅
let
1
𝜎𝑢2 = lim ∑𝑖(𝑢𝑖 − 𝑢̅)2
𝑁→∞ 𝑁

1
𝜎𝑣2 = lim ∑(𝑣𝑖 − 𝑣̅ )2
𝑁→∞ 𝑁
𝑖

2
1
𝜎𝑢𝑣 = lim ∑(𝑢𝑖 − 𝑢̅)(𝑣𝑖 − 𝑣̅ )
𝑁→∞ 𝑁
𝑖
2
In equation 𝜎𝑢𝑣 called Covariance
2 2
𝜕𝑋 2 2
𝜕𝑋 2 2
𝜕𝑋 𝜕𝑋
⇒ 𝜎𝑥 ≈ 𝜎𝑢 ( ) + 𝜎𝑣 ( ) + 2𝜎𝑢𝑣 ( ) ( ) +⋯
𝜕𝑢 𝑢̅ 𝜕𝑣 𝑣̅ 𝜕𝑢 𝑢̅ 𝜕𝑣 𝑣̅
If u and v is independent ,Then above formula becomes：
𝜕𝑋 2 𝜕𝑋 2
⇒ 𝜎𝑥2 ≈ 𝜎𝑢2 2
( ) + 𝜎𝑣 ( ) + ⋯ ____________(1)
𝜕𝑢 𝑢̅ 𝜕𝑣 𝑣̅

9. Uncertainty Propagation Formula

Using equation (1), we can derive the commonly used calculations, and Uncertainty

U-6
 國立臺灣師範大學普通物理實驗講義編輯群

Propagation Formula is as follows

A. Addition and subtraction
If 𝑋 = a𝑢 ± b𝑣，where a and b are constant，then
∂𝑋 𝜕𝑋
=a , = ±b
∂𝑢 𝜕𝑣

⇒ 𝜎𝑥2 = 𝑎2 𝜎𝑢2 + 𝑏2 𝜎𝑣2

[例][Ex](3.126 ± 0.014) − (1.892 ± 0.0095)

= (3.126 − 1.892) ± √(0.014)2 + (0.0095)2

= 1.234 ± 0.017
B. Multiply and divide
If 𝑋 = ±a𝑢𝑣，where a is constant，then
∂𝑋 𝜕𝑋
= ±a𝑣 , = ±a𝑢
∂𝑢 𝜕𝑣

⇒ 𝜎𝑋2 = 𝑎2 𝑣 2 𝜎𝑢2 + 𝑎2 𝑢2 𝜎𝑣2 + 2𝑎2 𝑢𝑣𝜎𝑢𝑣

2

𝜎𝑋2 𝜎𝑢2 𝜎𝑣2 2

𝜎𝑢𝑣
⇒ = + + 2
𝑋 2 𝑢2 𝑣 2 𝑢𝑣
[Ex]：
i. (0.862 ± 0.007) × (0.401 ± 0.006) = 𝑋 ± 𝜎𝑋
𝑋 = 0.862 × 0.401 = 0.346
𝜎𝑋 2 𝜎𝑢 2 𝜎𝑣 2
[ ] =[ ] +[ ]
𝑋 𝑢 𝑣

0.007 2 0.006 2
√
⇒ 𝜎𝑋 = 0.346 ( ) +( ) = 5.89 × 10−3
0.862 0.401

⇒ 𝑋 ± 𝜎𝑋 = 0.346 ± 0.006
7.1±0.1
ii. = 𝑋 ± 𝜎𝑋
0.0340±0.0014

7.1
𝑋= = 210
0.0340

0.1 2 0.0014 2
𝜎𝑋 = 210 × √( ) +( ) = 9.1
7.1 0.0340

⇒ 𝑋 ± 𝜎𝑋 = 210 ± 9

U-7
 國立臺灣師範大學普通物理實驗講義編輯群

C. Power
If 𝑋 = a𝑢±𝑏 ，where a and b are constant，then
∂X 𝑏𝑋 𝜎𝑋 𝜎𝑢
=± ⇒ = ±𝑏
∂u 𝑢 𝑋 𝑢
[Ex]： Circle areaA = π𝑟 2
If r = (10.0 ± 0.3)𝑐𝑚，then the area is
A ± 𝜎𝐴 = 𝜋(10.0 ± 0.3)2
𝐴 = 𝜋 × (10.0)2 = 314 𝑐𝑚2
0.3
𝜎𝐴 = 314 × (2 × ) = 18.8 𝑐𝑚2
10.0

⇒ 𝐴 ± 𝜎𝐴 = 314 ± 19 𝑐𝑚2
D. Exponential
If 𝑋 = a𝑒 ±𝑏𝑢 ，where a and b are constant，then
∂𝑋 𝜎𝑋
= ±𝑏𝑋 ⇒ = 𝑏𝜎𝑢
∂𝑢 𝑋
−𝑡
[Ex]： Radiation intensity attenuation function N = 𝑁0 𝑒 𝜏 ，
If 𝑁0 = 1000⁄𝑠，τ = 5 𝑑𝑎𝑦𝑠，t = 20 𝑑𝑎𝑦𝑠 ± 1 ℎ𝑟，Please calculate N?
𝜎𝑡 = 1 ℎ𝑟 = 0.042 𝑑𝑎𝑦𝑠
20±0.042
⇒ 𝑁 ± 𝜎𝑁 = 1000𝑒 5

20
⇒ 𝑁 = 1000𝑒 5 = 18.3⁄𝑠
0.042
⇒ 𝜎𝑁 = (18.3) [ ] = 0.15⁄𝑠
5

⇒ 𝑁 ± 𝜎𝑁 = (18.3 ± 0.15)⁄𝑠 𝑜𝑟 (18.3 ± 0.2)/𝑠

E. Logarithm
If X = 𝑎 𝑙𝑛(±𝑏𝑢 )，where a and b are constant，then
∂X 𝑎 𝜎𝑢
= ⇒ 𝜎𝑋 = 𝑎 [ ]
∂u 𝑢 𝑢
𝑁
[If]：Radiation time function t = −τ ln [ ]，
𝑁0

If τ＝5 𝑑𝑎𝑦𝑠，What is the time required for

the radiation intensity to decay from 𝑁0 = 1000⁄𝑠 to

N = (10 ± 1)⁄𝑠?
U-8
 國立臺灣師範大學普通物理實驗講義編輯群

10 ± 1
t ± 𝜎𝑡 = −5 𝑙𝑛 [ ]
1000
10
⇒ 𝑡 = −5 𝑙𝑛 [ ] = 23 𝑑𝑎𝑦𝑠
1000
1
⇒ 𝜎𝑡 = 5 × [ ] = 0.5 𝑑𝑎𝑦𝑠
10

⇒ 𝑡 ± 𝜎𝑡 = (23 ± 0.5) 𝑑𝑎𝑦𝑠

𝑜𝑟 (23 ± 1) 𝑑𝑎𝑦𝑠

10. Estimation and Uncertainty of True Value

We assume that any physical quantity has a corresponding parent population distribution.
The most likely value of the distribution is the true value we want. If the Parent Population
belongs to the normal distribution, the probability that the measured value is equal to 𝑥𝑖 is:
1 1 𝑥𝑖 − 𝜇 2
𝑃𝑖 = 𝑒𝑥𝑝 [− ( ) ]
𝜎√2𝜋 2 𝜎
In the formula, 𝜇 is the average value of the Parent Population, and it is also the most probable
value (that is, the highest probability of occurrence). But in a limited number of actual
measurements, we can't get the distribution of the Parent Population, and we can't know the
true value. Therefore, we must try to estimate its value and uncertainty range from the
directly measured sample data group.
Suppose we measure a set of sample data {𝑥1 , 𝑥2 , … 𝑥𝑖 , … , 𝑥𝑁 } of a physical quantity,
suppose that the average value of the parent population distribution corresponding to this set of
data is 𝜇′ (still unknown at this time), and the standard deviation is 𝜎 ′ (= 𝜎), then
the probability of measuring 𝑥𝑖 is:
2
1 1 𝑥𝑖 − 𝜇′
𝑃𝑖′ = 𝑒𝑥𝑝 [− ( ) ]
𝜎√2𝜋 2 𝜎

The probability of measuring the whole set of data {𝑥1 , 𝑥2 , … 𝑥𝑖 , … , 𝑥𝑁 } is:

𝑁 𝑁 𝑁 2
1 1 𝑥𝑖 − 𝜇′
′
𝑃 = ∏ 𝑃𝑖′ =[ ] 𝑒𝑥𝑝 [− ∑ ( ) ]
𝜎√2𝜋 2 𝜎
𝑖=1 𝑖=1

It can be seen from the formula that 𝑃′ is the 𝜇′ function. What should be the value of 𝜇′
so that 𝑃′ can be maximized?
∂𝑃′
⇒ ′=0
∂𝜇
𝑁 2
𝜕 1 𝑥𝑖 − 𝜇′
⇒ ′ [− ∑ ( ) ]=0
𝜕𝜇 2 𝜎
𝑖=1

U-9
 國立臺灣師範大學普通物理實驗講義編輯群

⇒ ∑(𝑥𝑖 − 𝜇′ ) = 0
𝑖=1

′
∑𝑁
𝑖=1 𝑥𝑖
⇒𝜇 = = 𝑥̅
𝑁
 The average value of the sample data group is the best estimate of the true value.
We continue to consider the standard deviation of 𝜇′ , because 𝜇′ is
calculated from the sample data, so
𝜇′ = 𝜇′ (𝑥1 , 𝑥2 , … 𝑥𝑖 , … , 𝑥𝑁 )，
Using the “uncertainty propagation formula”, we can get:
𝑁 2
𝜕𝜇′
𝜎𝜇2′ = ∑ [𝜎𝑖2 ( ) ]
𝜕𝑥𝑖
𝑖=1

∂𝜇′ 𝜕 ∑𝑁
𝑖=1 𝑥𝑖 1
式中 = [ ]=
∂𝑥𝑖 𝜕𝑥𝑖 𝑁 𝑁

𝑁
1 2
⇒ 𝜎𝜇2′ = ∑ [𝜎𝑖2 ( ) ]
𝑁
𝑖=1

If the measurement environment and conditions of each sample data are the same, then
𝜎𝑖 = 𝜎 is substituted into the formula to get
2
∑𝑁𝑖=1 𝜎
2
𝑁𝜎 2 𝜎 2
𝜎𝜇′ = = 2 =
𝑁2 𝑁 𝑁
𝜎
⇒ 𝜎𝜇′ =
√𝑁
Note that 𝜎 in the formula is the standard deviation of the sample data group,
𝜎𝜇′ is the standard deviation of 𝜇′ (i.e. average)
[Ex1]The following data is the time required for a spring to fully vibrate 20 times.
What is the period T of the spring vibration?
20T (sec) = 10.83 , 10.99 , 10.91 , 10.94 , 10.95
<sol>:
𝜎𝑋
20T (sec) = 𝑢′ ± 𝜎𝑢′ = 𝑥̅ ±
√𝑁
∑𝑁
𝑖=1 𝑥𝑖 (10.83 + 10.99 + 10.91 + 10.94 + 10.95)
𝑥̅ = = = 10.92 𝑠𝑒𝑐
𝑁 5

∑𝑁
𝑖=1(𝑥𝑖 − 𝑥̅ )
2
𝜎𝑖 = √
𝑁−1

U-10
 國立臺灣師範大學普通物理實驗講義編輯群

= {[(10.83 − 10.92)2 + (10.99 − 10.92)2 + (10.91 − 10.92)2

1
+ (10.94 − 10.92)2 + (10.95 − 10.92)2 ]/(5 − 1)}2

⇒ 𝜎𝑋 = 0.06 𝑠𝑒𝑐
0.06
⇒ 20𝑇 = 10.92 ± = 10.92 ± 0.03
√5

⇒ 𝑇 = (0.546 ± 0.002) 𝑠𝑒𝑐

From above, if the number of repeated measurements is more
(that is, the greater the value of N), the less uncertainty of the result is obtained.

11. Least Squares Mapping Method：

The “Least-Square Curve Mapping Method” is to find a curve (polynomial) so that
the sum of the squares of the vertical distances from all data points to the curve is the
smallest. That is if (𝑥𝑖 , 𝑦𝑖 ) is used to represent the data points, and 𝑓(𝑥) is their least
square difference curve, then 𝑓(𝑥) must satisfy the condition that the value of 𝐸 =
∑(𝑓(𝑥𝑖 ) − 𝑦𝑖 )2 is the smallest. 𝐸 is the sum of the squares of the longitudinal distance.
Here, when 𝐸 is the minimum, the sum of the squares of the vertical distances from all the
data points to the curve is also the minimum.
Here we only do linear least square difference approximation, with N experimental
data (𝑥𝑖 , 𝑦𝑖 ), the least square difference curve is:
𝑓 (𝑥) = 𝑎 + 𝑏𝑥
Where slope b and a are the intercepts, both of which are unknown and to be determined.
From the definition, it is necessary to minimize E,

E = ∑(𝑓 (𝑥𝑖 ) − 𝑦𝑖 )2 = ∑(𝑏𝑥𝑖 + 𝑎 − 𝑦𝑖 )2

and the conditions are:

∂E
= 0 ⇒ 𝑏 ∑ 𝑥𝑖2 + 𝑎 ∑ 𝑥𝑖 − ∑ 𝑥𝑖 𝑦𝑖 = 0 ____________(2)
∂b
∂E
= 0 ⇒ 𝑏 ∑ 𝑥𝑖 + 𝑁𝑎 − ∑ 𝑦𝑖 = 0 ____________(3)
∂𝑎
From (2) and (3), we can get:
(∑ 𝑥𝑖2 )(∑ 𝑦𝑖 ) − (∑ 𝑥𝑖 )(∑ 𝑥𝑖 𝑦𝑖 ) 𝑁(∑ 𝑥𝑖 𝑦𝑖 ) − (∑ 𝑥𝑖 )(∑ 𝑦𝑖 )
𝑎= 2 , 𝑏 =
𝑁 (∑ 𝑥𝑖 ) − (∑ 𝑥𝑖 )2 𝑁(∑ 𝑥𝑖2 ) − (∑ 𝑥𝑖 )2
The denominator can be represented by another symbol: ∆= 𝑁(∑ 𝑥𝑖2 ) − (∑ 𝑥𝑖 )2
Then draw this curve on the data graph according to the obtained slope and intercept.
In this way, you get the curve closest to your data and complete your drawing.
When using the least square method to calculate, the standard deviation σ in the y-
U-11
 國立臺灣師範大學普通物理實驗講義編輯群

direction is evaluated by the sum of squares of the vertical distance from each data point to
the fitted straight line. The formula is as follows:
1
Standard deviation formula：𝜎 2 = ∑(𝑦𝑖 − 𝑎 − 𝑏𝑥𝑖 )2
𝑁−2

To evaluate the slope and intercept uncertainty from 𝑦𝑖 , first treat the slope and intercept
as a function of 𝑦𝑖 . So the slope and intercept uncertainty can be calculated by the
following equation:
(Coefficient) standard deviation formula:
2
𝜕𝑎 𝜎2
𝜎𝑎2 2 2
= ∑ ( ) 𝜎 ⇒ 𝜎𝑎 = ∑ 𝑥𝑖2
𝜕𝑦𝑗 ∆
𝑗

2
𝜕𝑏 𝑁
𝜎𝑏2 = ∑ ( ) 𝜎 2 ⇒ 𝜎𝑏2 = 𝜎 2
𝜕𝑦𝑗 ∆
𝑗

There will be practical exercises later.

[Homework] The relationship between time and speed measured in the free fall experiment
is as follows:
t (s) 0 1 2 3 4 5
v (m/s) 15.0 24.8 34.1 44.4 55.0 64.0
Using the least square method (y = a + bx), calculate
(1) slope?
(2) intercept?
(3) What is the standard deviation of the speed? And explain its meaning?
(4) Mapping?

[Ex2]： Data analysis of the Simple pendulum period

A. Measure one cycle. 100 times in total:
1.656 1.661 1.656 1.653 1.656
1.657 1.659 1.656 1.657 1.655
1.657 1.656 1.656 1.656 1.655
1.657 1.656 1.655 1.655 1.655
1.652 1.655 1.661 1.657 1.656
1.657 1.658 1.657 1.659 1.665
1.657 1.656 1.657 1.656 1.666
1.655 1.663 1.656 1.656 1.657
1.656 1.657 1.662 1.649 1.656
1.650 1.657 1.655 1.655 1.651
1.657 1.660 1.656 1.656 1.661
U-12
 國立臺灣師範大學普通物理實驗講義編輯群

1.659 1.656 1.656 1.659 1.656

1.655 1.657 1.656 1.656 1.655
1.653 1.649 1.657 1.656 1.654
1.656 1.657 1.650 1.661 1.656
1.655 1.657 1.657 1.652 1.652
1.656 1.658 1.649 1.658 1.656
1.656 1.656 1.656 1.656 1.657
1.658 1.656 1.656 1.656 1.657
1.655 1.657 1.656 1.656 1.650

∑ 𝑇𝑖 1.656+1.657+⋯
Average period：𝑇̅ = = = 1.65614
𝑁 100

Standard deviation：
∑(𝑇𝑖 −𝑇̅)2 (1.656−1.65614)2 +(1.657−1.65614)2 +⋯
σ=√ =√ = 2.7 × 10−3
𝑁−1 100−1

𝜎 2.68185×10−3
Average Standard deviation：𝜎𝑇 = = = 2.7 × 10−4
√𝑁 √100

Analysis result: The cycle of a Simple pendulum： T = 1.65614 ± 0.00027 (𝑠)

Measure the time of the Simple pendulum for 25 cycles(25T). 8 times in total
41.416 41.416 41.412 41.422 41.412
41.415 41.408 41.407

41.416+41.416+⋯
(Average period)：25T = = 41.414
8

(41.416−41.414)2 +(41.416−41.414)2 +⋯
Standard deviation：σ = √ = 4.8 × 10−3
8−1

𝜎 4.840×10−3
Average Standard deviation：𝜎𝑇 = = = 1.7 × 10−3
√𝑁 √8

the time of the Simple pendulum for 25 cycles：

25T = 41.414 ± 0.0017
 The cycle of a Simple pendulum：
41.414±0.0017
T= = 1.65656 ± 0.00007 (𝑠)
25

[Ex3]：The relationship between the period and length of a Simple pendulum

A. Change the pendulum length seven times. Repeat the measurement of the time of
each swing 30 cycles (30T) five times
Length (cm)
82.14 85.45 88.59 95.39 105.79 119.85 132.15
Times(s)
U-13
 國立臺灣師範大學普通物理實驗講義編輯群

30T 54.556 55.706 56.703 58.867 61.941 65.967 69.225

30T 54.553 55.712 56.712 58.858 61.945 65.959 69.227
30T 54.542 55.727 56.705 58.871 61.945 65.967 69.242
30T 54.546 55.714 56.732 58.872 61.943 65.968 69.254
30T 54.541 55.720 56.720 58.880 61.944 65.963 69.278

The relationship between the period and length：

Length (cm) 82.14 85.45 88.59 95.39 105.79 119.85 132.15
Time(s) 1.818 1.857 1.891 1.962 2.065 2.119 2.308

B. Least Squares Method：

i. linear relationship 𝑦 = 𝑎 + 𝑏𝑥，𝑥: (pendulum length)1/2 ，𝑦: period

𝑥𝑖 𝑦𝑖 𝑦𝑖 − 𝑎 − 𝑏𝑥𝑖 (𝑦𝑖 − 𝑎 − 𝑏𝑥𝑖 )2

9.063 1.818 −1.9 × 10−3 3.61 × 10−6
9.244 1.857 7.47 × 10−4 5.59 × 10−7
9.412 1.891 1.01 × 10−3 1.01 × 10−6
9.767 1.962 7.03 × 10−4 4.94 × 10−7
10.29 2.065 −1.34 × 10−3 1.80 × 10−6
10.94 2.199 2.11 × 10−3 4.44 × 10−6
11.50 2.308 −1.37 × 10−3 1.88 × 10−6

∑ 𝑥𝑖 = 70.216, ∑ 𝑦𝑖 = 14.100

∑ 𝑥𝑖2 = 709.39

∑ 𝑥𝑖 𝑦𝑖 = 142.45
2
∆= N ∑ 𝑥𝑖2 − (∑ 𝑥𝑖 ) = 7 × 709.39 − (70.216)2 = 35.42
coefficient：
(∑ 𝑥𝑖2 )(∑ 𝑦𝑖 ) − ∑ 𝑥𝑖 (∑ 𝑥𝑖 𝑦𝑖 )
𝑎=
∆
709.39 × 14.100 − 70.216 × 142.45
= = −4.045 × 10−4
35.42
𝑁 ∑ 𝑥𝑖 𝑦𝑖 − ∑ 𝑥𝑖 ∑ 𝑦𝑖
𝑏=
∆
U-14
 國立臺灣師範大學普通物理實驗講義編輯群

7 × 142.45 − 70.216 × 14.100

= = 2.0085 × 10−1
35.42
(Thus) y = −4.045 × 10−4 + 0.2009𝑥
Standard deviation：
1 1
𝜎2 = ∑(𝑦𝑖 − 𝑎 − 𝑏𝑥𝑖 )2 = × 1.38 × 10−5
𝑁−2 5

⇒ 𝜎 = 1.7 × 10−3

The Standard deviation of coefficient：

𝜎2 2
2.76 × 10−6
𝜎𝑎2 = ∑ 𝑥𝑖 = × 709.39 = 5.53 × 10−5
∆ 35.42
−3
𝜎𝑎 = 7.5 × 10
𝑁 7
𝜎𝑏2 = 𝜎 2 = × 2.76 × 10−6
∆ 35.42
𝜎𝑏 = 7.4 × 10−4
The complete equation is：
y = (−4.0 ± 75) × 10−4 + (2.0085 ± 0.0074) × 10−1 𝑥
The linear correlation coefficient changes the function relationship in the
same way:

y = 𝑎′ + 𝑏′ 𝑥，x： period，y：(pendulum length)1/2 ，

get 𝑎′ = 1.041 × 10−5 ，𝑏′ = 4.979

Correlation coefficient：
𝑟 = 𝑏𝑏′ = 2.0085 × 10−1 × 4.979 = 0.99994
After calculating the relationship between period and length of the simple
pendulum by the least square method, the g value is calculated by the
coefficient, and the uncertainty range of the g value is calculated by applying
the uncertainty transmission law:
Known from the calculation of the least square method：
y = 𝑎 + 𝑏x = −4.0 × 10−4 + 0.20085𝑥
In here x is
1
(pendulum length)2 = √𝑙 ，𝑦 𝑖𝑠 𝑝𝑒𝑟𝑖𝑜𝑑 𝑇，

Standard deviation of coefficient a：𝜎𝑎 = 7.5 × 10−3

Standard deviation of coefficient b：𝜎𝑏 = 7.4 × 10−4

U-15
 國立臺灣師範大學普通物理實驗講義編輯群

ii. Find the g value：

𝑙 2𝜋
∵ 𝑇 = 2π√ = √𝑙
𝑔 √𝑔
2𝜋
∴ = 0.20085 ⇒ 𝑔 = 978.62
√𝑔
iii. Find the standard deviation 𝜎𝑔 of 𝑔 value：
2𝜋 ∂𝑓 3
iv. Let 𝑓(𝑔) = ，then = −𝜋𝑔−2
√𝑔 ∂𝑔
2
𝜕𝑓
𝜎𝑏2 = ( ) 𝜎𝑔2
𝜕𝑔
1 𝑔3
⇒ 𝜎𝑔2 = 𝜎𝑏2 ∙ = (7.4 × 10 −4 )2
∙ 2 = 52
𝜕𝑓 2 𝜋
( )
𝜕𝑔

⇒ 𝜎𝑔 = 7.3
cm
So from ii. and iii. we know 𝑔 = 978.6 ± 7.3
𝑠2
7.3
Uncertainty ratio： ≈ 0.75%
978.6

U-16