FSN/ABCT4423 Food Quality Assurance

Basic Statistics for Quality

Control/Assurance
Dr. GENG Ping (Sandy)
Office: TU321
Email: [email protected]
Content
1 Introduction of Statistics
2 Fundamentals of Probability
3 Probability Models for QA/QC
4 Central Limit Theorem
5 Hypothesis Testing

Reference:
Review of Basic Statistics (Quality Improvement, Chapter 5 & 8, Besterfield, 9th)
2
1. Introduction of Statistics
What is statistics?
The science that deals with the collection, tabulation, analysis, interpretation, and
presentation of quantitative data.

How are data classified?

▪ Variable: A quality characteristic that is measurable, such as weight, length, etc.

▪ Attribute: A quality characteristic that is classified as either conforming or not conforming

to specifications.

3
 1.1 Data collection: Accuracy vs Precision
Accuracy
How well a measurement agrees with an accepted value. It is the
deviation of measured or observed average value from the true
mean value.
Precision
How well a series of measurements agree with each other. It is a
measure of consistency of the system.

Quality Control/Assurance is used to monitor

both the precision and the accuracy of the
samples to provide reliable results.

4
 1.2 Measures of Central Tendency
A numerical value which describes the central position of the data.
n

x i
• Average: x= i =1
n
• Mode: the value that occurs with the highest frequency in a set of
numbers
e.g. 2, 4, 4, 5, 6, 6, 8, 8, 8,11

• Median: value which divides a series of ordered observations so that the

no. of items above it is equal to the no. below it.
e.g. 3, 4, 5, 6, 8, 8, 10
e.g. 3, 4, 5, 6, 8, 8, 8, 10 median = (6 + 8)/2 = 7
5
 1.3 Measures of Dispersion

Describes how the data are spread out or scattered on each

side of the central value.
• Range: highest value – lowest value
• Sample standard deviation:

 i
( x − x ) 2

s= i =1
n −1

The smaller the s, the better the quality.

n -1: degree of freedom. It equals to independent numbers (n) minus the number of
parameters estimated as intermediate steps (e.g., the sample mean)
6
 1.4 Comparison of Sample vs. Population
• Population: whole collection of measurements
• Sample: part or portion of the population, i.e., select some to represent the
whole collection
Population
distribution
Sample Population
Sample x : average µ: mean
Narrower spread

Note: when averages, standard deviations, and

spread other measures are computed from samples, they
are referred to statistics.
Sample Distribution vs. Population Distribution

7
Content
1 Introduction of Statistics
2 Fundamentals of Probability
3 Probability Models for QA/QC
4 Central Limit Theorem
5 Hypothesis Testing

Reference:
Review of Basic Statistics (Chapter 4 &7, Besterfield, 7th)

8
2. Fundamentals of Probability
What is probability?
It describes the chance that a specific event will occur, synonyms as
likelihood, tendency, and trend.
NA
Probability is defined by this equation: P( A) =
N
where P(A) = probability of an event A occurring to 3 decimal places
NA = number of successful outcomes of event A
N = total number of possible outcomes

• All probability values will lie between 0 and 1.

• The sum of the probability of all events is 1.
9
 2.1 Theorems of Probability (1)

Theorems which govern the computation of probability are:

• Theorem 1:
Probability is expressed as a number between 1.000 and 0, where a
value of 1.000 is a certainty that an event will occur, and 0 is a certainty
that an event will not occur.

• Theorem 2
If P(A) is the probability that an event will occur, then the probability
that A will not occur is 1.000 - P(A).

• Theorem 5
The sum of the probability of the events of a situation is equal to 1.000.

10
 2.1 Theorems of Probability (2)

One event Two events

‘Or’ case, P(A or B) ‘And’ case, P(A and B)

Mutually exclusive NOT Mutually exclusive Independent Dependent

P(A or B) = P(A) + P(B) P(A or B) = P(A) + P(B) – P(both) P(A and B) = P(A) * P(B) P(A and B) = P(A) * P(B|A)

Theorem 3 Theorem 4 Theorem 6 Theorem 7

Note:
o Mutually exclusive means that the occurrence of one event makes the other event impossible.
o P(B|A) is the (P) of B calculated knowing that A has occurred.
11
 Example 4
Data: Food inspection results by supplier
Supplier Conforming Nonconforming Total

X 50 3 53
Y 125 6 131
Z 75 2 77
Total 250 11 261

If 261 food products are contained in a box and we need to pick 1 to 2 samples from it, what
is the probability of selecting
• An item produced by supplier X or Z?
• A conforming item produced by X or Z?
• An item produced by X or being nonconforming?
• 2 items with 1st from X and 2nd from Y (with replacement)?
• 2 items with 1st from X and 2nd from Y (without replacement)?

12
 Example 4
Data: Inspection results by supplier
Supplier Conforming Nonconforming Total

X 50 3 53
Y 125 6 131
Z 75 2 77
Total 250 11 261

If 261 food products are contained in a box, what is the probability of selecting
• An item produced by supplier X or Z? P(X) + P(Z) = 53/261 + 77/261 = 0.498
• A conforming item produced by X or Z? P(co. X) + P(co. Z) = 50/261 + 75/261 = 0.479

13
 Example 4
Data: Inspection results by supplier
Supplier Conforming Nonconforming Total

X 50 3 53
Y 125 6 131
Z 75 2 77
Total 250 11 261

If 261 food products are contained in a box, what is the probability of selecting
• An item produced by X or being nonconforming?
P(X) + P(nc.) – P(X and nc.) = 53/261 + 11/261 – 3/261 = 0.234

14
 Example 4
Data: Inspection results by supplier
Supplier Conforming Nonconforming Total

X 50 3 53
Y 125 6 131
Z 75 2 77
Total 250 11 261

If 261 food products are contained in a box, what is the probability of selecting
• 2 items with 1st from X and 2nd from Y (with replacement)?
P (X and Y) = P(X) × P(Y) = (53/261) ×(131/261) = 0.102
• 2 items with 1st from X and 2nd from Y (without replacement)?
P (X and Y) = P(X) ×P(Y|X) = (53/261) × (131/260) = 0.102

15
 2.2 Counting of Events
NA
The calculation of probability is based on the number of possible outcomes, P( A) =
which sometimes are difficult to find. N

1) Simple Multiplication
If an event A can happen in any of a ways, and after it has
occurred, another event B can happen in b ways, the number of
ways that both events can happen is ab.
A witness to a hit-and-run accident remembered the first 3 digits of the
license plate out of 5 and noted the fact that the last 2 were numerals.
How many owners of automobiles would the police have to investigate?

Ans: ab = (10) × (10) = 100

16
 2.2 Counting of Events

2) Permutation

A Permutation is an ordered arrangement of a set of objects.

For example, what is the possible outcomes of forming a 2-letter code

from 3 cards, labeled with the letters of A, B and C respectively?

Ans: AB AC BC BA CB CA

17
 2.2 Counting of Events

2) Permutation
It is the number of outcomes of arranging n objects when taken r
at a time, noted as
Total no. of objects
n! n(n − 1)(n − 2)(n − 3)...
P = n Pr =
n
=
(n − r )! (n − r )(n − r − 1)(n − r − 2)(n − r − 3)...
r

no. of objects selected

Note: n! is called factorial, and 1! = 1, 0! = 1

18
 2.2 Counting of Events

2) Permutation

In the license plate example, suppose the witness further remembers that the
numerals were not the same, how many owners of automobiles would the
police have to investigate?

10! 10 ∙ 9 ∙ 8 ∙∙∙ 1
Ans: 𝑃210 = 10𝑃2 = = = 90
10 − 2 ! 8 ∙ 7 ∙∙∙ 1

19
 2.2 Counting of Events

3) Combination
It is the number of outcomes of arranging n objects when
taken r at a time, without regarding the order of selection,
noted as
n!
C = n Cr =
n

r!(n − r )!
r

Note: Prn  Crn (always)

20
 2.2 Counting of Events

3) Combination
An interior designer has five different colored chairs and will use
three in a living room arrangement. How many different combinations
are possible?

5! 5∙4∙3∙2∙1
Ans: 𝐶35 = 5𝐶3 = = = 10
3! 5 − 3 ! 3∙2∙1∙2∙1

Think: What is the possible outcomes of forming a 2-letter code from

the letters of A, B, C, without regarding the order of selection? AB, AC, BC

21
Content
1 Introduction of Statistics
2 Fundamentals of Probability
3 Probability Models for QA/QC
4 Central Limit Theorem
5 Hypothesis Testing

Reference:
Review of Basic Statistics (Chapter 4 &7, Besterfield, 7th)

22
3. Probability Models for QA/QC
• Continuous distribution:
➢ Normal distribution (Gaussian distribution)

• Discrete distribution:
➢ Binomial distribution
➢ Poisson distribution
➢ Hypergeometric distribution

23
 3.1 Normal Distribution

• Symmetrical, bell-shaped curve, Human BMI distribution

common shape
µ = 29
• Defined by population mean (µ) and σ=6
population standard deviation ()
• The mean, median, and mode are
the same value
• µ – locate at center of the curve
-3 -2 -1 +1 +2 +3
 – width of the curve 
- 0 +
Deviation from mean
Xi - µ
24
 3.1 Normal Distribution

• Determined by µ & , independently • The larger the , the flatter the curve.

Normal curve with different means but Normal curve with different standard
identical standard deviation deviations but identical means

25
 3.1 Normal Distribution

• All normal distributions of continuous e.g. Consider the value 35, Z = (35 – 29)/6 = +1
variables can be converted to the
standardized normal distribution by µ = 29
using the standardized normal value Z. σ=6

xi −  xi: measurement of interest

Z= Z: the distance from the centre
 measured in standard deviation unit

• Formula for the normal distribution µ=0

σ=1
curve.
1 −Z
2

f (Z ) = e 2

2 2
Standardized normal distribution

26
 3.1 Normal Distribution

• A relationship exists between the standard

deviation and the area under the normal curve.
• The total area under the curve is 1 (100%).
68.26%
• 68.26% of the items are included within ±1σ.
• 95.46% of the items are included within ±2σ.
95.46%
• 99.73% of the items are included within ±3σ.

99.73%

x scale: -3 -2 -1  +1 +2 +3

Basis for control charts
Z scale: -3 -2 -1 0 1 2 3

27
 3.1 Normal Distribution
Calculation of probability
• The area under the normal curve is the • Essential first step: determine Z value using
probability of occurrence. xi − 
Z=
• The area can be calculated by normal curve 
formula using calculus, but not necessary.
• Probability of X  xi can be made as
1 −Z
2

f (Z ) = e 2
xi −  xi − 
2 2 P( X  xi ) = P( Z  ) = ( )
 
• The % of data under the curve for various Z
values are given in the table of Normal
Distribution (Appendix Table A in Dale “Quality
Improvement”).
xi 
z 0
28
 3.1 Normal Distribution
Calculation of probability
• The table of “Areas Under the Normal
Curve” is a left-reading table, i.e., the given
area is from -∞ to a particular z value.
e.g., P (Z ≤ -0.72) = 0.2358, which means
23.58% of the data are less than -0.72.

Notes:
xi −  xi − 
P( X  xi ) = P( Z  ) = ( )
 
P(X  xi) = 1 - P(X  xi)
P(xi  X  xj) = P(X  xj) - P(X  xi)

29
xi xi xj
 3.1 Normal Distribution
Calculation of probability
The mean value of the weight of a particular brand of cereal for the past year is 0.297 kg with a
standard deviation of 0.024 kg. Assuming normal distribution, find the % of the data that falls
1) below the lower specification limit of 0.274 kg
2) above the higher limit of 0.350 kg, or
3) between the lower and higher limit.

30
 3.1 Normal Distribution
Calculation of probability
The mean value of the weight of a particular brand of cereal for the past year is 0.297 kg with a
standard deviation of 0.024 kg. Assuming normal distribution, find the % of the data that falls
1) below the lower specification limit of 0.274 kg
𝑋1 − 𝜇 0.274 − 0.297
Ans: 𝑍1 = = = −0.96
𝜎 0.024
P(X  0.274) = P(Z  -0.96) = 0.1685 = 16.85%

31
 3.1 Normal Distribution
Calculation of probability
The mean value of the weight of a particular brand of cereal for the past year is 0.297 kg with a standard
deviation of 0.024 kg. Assuming normal distribution, find the % of the data that falls
1) below the lower specification limit of 0.274 kg
2) above the higher limit of 0.350 kg, or
3) between the lower and higher limit.

𝑋2 − 𝜇 0.350 − 0.297
Ans: 2) 𝑍2 = 𝜎
=
0.024
= +2.21

P(X  0.350) = 1 - P(X  0.350) = 1 – P(Z  2.21)

= 1 – 0.9864 = 1.36%

3) P(0.274  X  0.350) = P(X  0.350) - P(X  0.274)

= 98.64% - 16.85% = 81.79%
32
3. Probability Models for Quality Control
• Continuous distributions:
➢ Normal distribution (Gaussian distribution)

• Discrete distributions:
➢ Binomial distribution
➢ Poisson distribution
➢ Hypergeometric distribution

33
 3.2 Binomial Distribution

Think about the problem:

In a milk product manufacturing factory,

there is a steady stream of products.
Based on previous record, the probability
of product nonconforming is 0.10.
If a random sample of 5 milk products is
selected, what is the probability of 1
nonconforming unit in the sample?

34
 3.2 Binomial Distribution

The binomial probability distribution is applicable to the following conditions:

• There are an infinite number of items or a steady stream of items coming from a work center.
• The process consists of a sequence of n independent trials, where the outcome of each trial is
either a success or a failure.
• If the probability of failure is denoted by p, and that of success by 1-p. The probabilities p and
1-p remain constant for each trial.

Then the probability function of x failure in n trials is referred to as the Binomial distribution:

P( x) = C xn p x (1 − p) n − x 1 2 3 4 5
x = 2,
x = 0, 1, 2,…, n n=5
p p 1-p 1-p 1-p
𝑝 𝑥 = 𝐶25 𝑝2 (1 − 𝑝)3 35
 3.2 Binomial Distribution

In a milk product manufacturing factory, there is a steady stream of products.

Based on previous record, the probability of product nonconforming is 0.10.
If a random sample of 5 milk products is selected, what is the probability of 1
nonconforming unit in the sample?

Ans: p = 0.10, n = 5, x = 1

𝑃 𝑥 = 𝐶𝑥𝑛 𝑝 𝑥 1 − 𝑝 𝑛−𝑥

𝑃 1 = 𝐶15 (0.1)1 1 − 0.1 5−1

= 0.328

How about the probabilities of 0, 3, 4, 5 nonconforming unit in the sample?

36
 3.2 Binomial Distribution

The probability of x  a for n trials with p on any trial is

a
P( x  a) =  C xn p x (1 − p) n − x
x =0

Use the same example as the milk product manufacturing factory with a steady stream
of products and nonconforming probability of 0.10. What is the probability that no
more than 2 nonconforming milk products in a sample of 5?

Ans: P (x  2) = P(0) + P(1) + P(2) = 0.590 + 0.328 + 0.073 = 0.991

37
 3.3 Poisson Distribution

The probability of a Poisson distribution is:

x e −
P( x) = x = 0, 1, 2, ...
x!

 is the average count of a given classification in a sample.

x is the count of any single event of a given classification in a sample.

38
 3.3 Poisson Distribution
 as average count x e −
P( x) = x = 0, 1, 2, ...
x!
x is the count of any single event

Example:
 is the average non-conforming beers in all boxes of beer product
x is 1 non-conforming beer in one sample box of beer

 is the average defects in all cake products

x is 2 defects in one cake

Note: The opportunities for the occurrence of an event should be equal and independent.

39
 3.3 Poisson Distribution
The Poisson distribution can sometimes be used as approximation to the binomial distribution.

For example: In a sample of 100 bottles of beers (n =100), the nonconforming probability of beer is 0.02
(p = 0.02) based on previous record. The average number of nonconforming beers in a sample  = np =
2. What is the probability of 1 nonconforming beer in a sample? (Hint: put  = 2, x = 1 into the formula )
x e −
P( x) = x = 0, 1, 2, ...
x!

Note: The Poisson distribution can be used as approximation to the binomial distribution, when n is
very large and p approaches to 0, such that np =  is a constant and (1 - p) approaches 1. In QA & QC,
p is usually very small.

Binomial and Poisson probability distributions are the basis for attribute control charts and
acceptance sampling.

40
 3.3 Poisson Distribution
The probability for x  a in Poisson distribution is:
a x e −
P( x  a) = 
x =0 x!

The number of defects on a pack of PE food packaging material has been

found to follow the Poisson distribution with  = 3. What is the probability
that a randomly selected pack will have two or less defects?

Ans:
2 3 x e −3
P( x  2) =  = P(0) + P(1) + P(2) = 0.049 + 0.150 + 0.224 = 0.423
x =0 x!

41
 Binomial & Poisson Distribution
The proportion of fish cans that are nonconforming is 0.04, calculate the probability of 3 or
fewer nonconforming fish cans in a sample of 100.
3
Ans: Binomial Distribution P( x  3) =  C 100
x ( 0.04) x
(1 − 0.04)100 − x

x =0

= P( 0 ) + P(1) + P(2) + P(3)

= 0.017 + 0.070 + 0.145 + 0.197
= 0.429
3
4 x e −4
Poisson Distribution P( x  3) = 
x =0 x!
p = 0.04 < 0.1
= e − 4 (1 + 4 + 8 + 64 / 6)
λ = np = 100 × 0.04 = 4
= 0.433
 3.4 Hypergeometric Distribution
When the population N is small, then if one selects a nonconforming
sample without replacement, the probability of subsequent samples will be
affected. The probability distribution for such small population follows
Hypergeometric distribution.

Let’s consider two cases:

Case 1: A lot has 10,000 bottles with 1,000 nonconforming units. After taking 1
nonconforming unit from the lot, (p) is unchanged
p(before) = 1,000/10,000 = 0.1 Binomial or Poisson distribution
p(after) = 999/9,999  0.1

Case 2: A lot has 20 bottles with 2 nonconforming units. After taking 1

nonconforming unit from the lot, (p) is changed
p(before) = 2/20 = 0.1 Hypergeometric distribution
p(after) = 1/19  0.05
43
 3.4 Hypergeometric Distribution
When drawing samples from a lot of finite size N, containing a specified
number of nonconforming units, D, the P of a hypergeometric distribution that
a sample of size n contains x nonconforming units is:
Nonconforming Conforming
combinations combinations
Population: N, D
D N −D
C C
P( x) = x
N
n− x
x = 0, 1, 2, ...min(n, D)
Sample: n, x Total number: N
C n

Total combinations
N-D
The probability for x  a in a hypergeometric distribution is:
D
a C DC N − D
P( x  a) =  x n − x
x =0 CnN

The hypergeometric distribution is the appropriate probability model for sampling

without replacement.
44
Example
A lot of 9 chocolate balls located in a container has 3 expired balls.
What is the probability of drawing 1 expired chocolate ball in a
random sample of 4?

N = 9, D = 3
n = 4, x = 1

C13C49−−13
Ans: P (1) = 9
= 0.476
C4

45
Example
A company produces egg cartons with 24 eggs in each carton. One small bakery shop purchased
one carton and would like to test 4 eggs in it, and if all of them are fresh, the bakery shop will
purchase the egg brand in the future. If the carton purchased has 3 nonconforming eggs, what is
the probability of rejection by the bakery shop?
3 24 −3
Ans: N = 24, D = 3 Hypergeome C 0C 4 − 0
tric : P(0) =
Acceptance = 0.563
C24
4
n = 4, x = 0
Rejection:= 1 − P (0) = 1 − 0.563 = 0.437
rejection

1 1
Binomial:: P (0) = C xn p x (1 − p ) n − x = C04 ( )0 (1 − ) 4−0 = 0.586
Binomial
8 8
( 0.5) 0 e −0.5
Poisson:: x = 0,  = 4 * 18 = 0.5, so P(0) =
Poisson = 0.607
0!

If the acceptance criterion allows no more than one defective egg in a carton, what is
the probability of acceptance? Ans: P(x ≤ 1) = P(0) + P(1) = 0.563 + 0.375 = 0.938
46
Distribution Inter-relationship

Hypergeometric

n/N ≤ 0.1 n: sample size

N: population size

Binomial

p ≤ 0.1, n large

Poisson

47
Content
1 Introduction of Statistics
2 Fundamentals of Probability
3 Probability Models for QA/QC
4 Central Limit Theorem
5 Hypothesis Testing

Reference:
Review of Basic Statistics (Chapter 4 &7, Besterfield, 7th)

48
 4. Central Limit Theorem
Data on the length of chocolate bars
Each individual measurements are coded from 6.00 mm,
so 6.35 is recorded as 35.

The individual and average data are summarized below:

49
4. Central Limit Theorem

Individual values compared to sample averages

• Individual values: Wider base, higher frequency
• Sample averages: Narrower base, lower frequency
• The mean of the distribution of sample averages equals to the
mean of the individual values from the population.
𝑥Ӗ = 𝜇𝑥
• The variance of the sample averages equals to the variance of
the individual values from population divided by the samples
size. 2
𝜎𝑥 𝜎𝑥
𝜎𝑥2ҧ = 𝜎𝑥ҧ =
𝑛 𝑛

50
4. Central Limit Theorem
Consider the frequency distribution of non-normal population and average distribution of
sample sizes 2, 3, 4, 5, 6.

51
4. Central Limit Theorem
If the population from which samples are taken is not normal, the
distribution of sample average will tend towards normality
provided that the sample size n is at least 4 (n  4).

• As the sample size n becomes larger, the

distribution of average becomes more normal
and narrower, irrespective of the distribution
shape of the individual terms Xi.

• The central limit theorem is one of the reasons

behind X chart in statistical process control.

52
Content
1 Introduction of Statistics
2 Fundamentals of Probability
3 Probability Models for QA/QC
4 Central Limit Theorem
5 Hypothesis Testing

Reference:
Review of Basic Statistics (Chapter 4 &7, Besterfield, 7th)

53
5. Hypothesis Testing
To get results from statistical analysis, we first make a statement or claim, i.e.,
Hypothesis, about the variable of interest, which we believe to be true.

Example A: we believe that the diameters of sausages have a true mean of 6 mm.
The following hypothesis could be formed before the samples were taken:
H0: x = 6 null hypothesis Currently accepted value for a parameter

H1 : x  6 alternative hypothesis The complement or opposite of null hypothesis

Example B: we wish to test whether the Melamine content of a dairy product

exceeds 2.5 ppm?
H0: melamine equal or larger than 2.5 ppm: x ≥ 2.5 ppm
H1: melamine less than 2.5 ppm: x < 2.5 ppm

Note: These two states of nature (H0 and H1) cannot be true simultaneously.
54
5. Hypothesis Testing
Example: Please find the H0 and H1

It is believed that a candy machine makes chocolate bars that are

on average 5 g. A worker claims that the machine after
maintenance no longer makes 5 g bars.

H0: x = 5 g null hypothesis

H1 : x  5 g alternative hypothesis

55
 Statistical Hypothesis - Errors

With a ‘YES/NO’ type of question, there are two ways of being wrong:

• Deciding ‘NO’ when ‘YES’ is correct → Type I error → We reject a null hypothesis when
the null hypothesis is in fact true.
e.g., The probability that a lot of conforming units is rejected or a process producing
acceptable units is stopped (). Producer's risk

• Deciding ‘YES’ when ‘NO’ is correct → Type II error → We accept a null hypothesis
when the null hypothesis is in fact false.
e.g., The probability that a lot of non-conforming units is accepted or a process producing
inferior units is allowed to continue (). Consumer's risk

56
 Statistical Hypothesis - Errors

Ho true Ho false
Accept Ho Right decision Type II error ()

Reject Ho Type I error () Right decision

Type I error is called α type error. α is also named as significance level.

α type error and β type error are important terms in acceptance sampling.

57
Summary

• Distinguish the important terms such as variable & attribute, sample &
population, permutation & combination;
• Master theorems of probability;
• Understand normal distribution for continuous data, mean and standard
deviation, standardization method, probability calculation;
• Understand probability models for discrete data (Binomial, Poisson and
Hypergeometric distribution), the prerequisite for use, probability calculation;
• Have basic understanding of central limit theorem;
• Distinguish null & alternative hypothesis, type I (α type) & type II (β type)
errors in hypothesis testing.

58