Statistical Quality Control: Basic Statistics
Statistical Quality Control: Basic Statistics
Basic statistics
Basic statistics is majorly divided into two types;
i. Descriptive Statistics: A straightforward presentation of facts. A survey or summary of
a population in which all data are known.
ii. Inferential Statistics: Drawing conclusions about a population from a random sample
Descriptive statistics
Descriptive Statistics includes;
The Mean- measure of central tendency
The Range- difference between largest/smallest observations in a set of data
Standard Deviation measures the amount of data dispersion around mean
x
n n
x
2
i i X
x i 1 σ i 1
n n 1
Inferential statistics
Inferential statistics is a valuable tool because it allows us to look at a small sample size and
make statements on the whole population.
Samples must be pulled RANDOMLY from a population so that the sample truly
represents the population. Every unit in a population must have an equal chance of being
selected for the sample to be truly random.
The distribution or shape of the data is important to know for analytical purposes. The
most common distribution is the bell shaped or normal distribution.
Parameters can be estimated from sample statistics. Two of the most common
parameters are the mean and standard deviation.
The mean (or average, denoted by μ) measures central tendency. This is
estimated by the sample mean or x-bar.
The standard deviation (σ) measures the spread of the data and is estimated by
the sample standard deviation
a) Descriptive statistics
It includes the mean, standard deviation, and range etc.
b) Statistical process control (SPC)
It involves inspecting the output from a process
The Quality characteristics are measured and charted
It is helpful in identifying in-process variation
c) Acceptance sampling
It is used to randomly inspect a batch of goods to determine acceptance/rejection
Does not help to catch in-process problems
Sources of variation
Variation exists in all processes. Variation can be categorized as either;
i) Common or Random causes of variation
Random causes are those that we cannot identify
They are somewhat unavoidable
Example: slight differences in process variables like diameter, weight, service time,
temperature
x
n n
xi
2
i X
x i 1 σ i 1
n n 1
2. Control chart for variables are used to monitor characteristics that can be
measured.
Example: length, weight, diameter, time
3. Control charts for attributes are used to monitor characteristics that have
discrete values and can be counted.
Example: percent defective, number of flaws in a shirt, number of broken eggs in a
box.
Percentage of values under normal curve Control limits balance risks like Type I error
Control charts for variables
In statistical quality control the x-bar and R chart is a type of control chart used to monitor variables
data when samples are collected at regular intervals from a business or industrial process.
It includes the use of x-bar and R-bar charts together. It is used to monitor different variables.
x-bar & R-bar Charts reveal different problems. These charts are called Shewhart Charts.
x-bar charts are used to monitor the changes in the mean of a process (central
tendencies).
R-bar charts are used to monitor the dispersion or variability of the process.
System can show acceptable central tendencies but unacceptable variability or system can show
acceptable variability but unacceptable central tendencies.
The ‘chart’ actually consist of a pair of chats; one to monitor the process standard deviation (as
approximated by sample moving range) and another to monitor the process mean, as is done with
x-bar and s-chart and individual control charts. The x-bar and R chart plots the mean value for
the quality characteristic across all units in the sample, x-bar plus the Range of the quality
characteristic across all units in the sample as follows;
𝑅 = 𝑋𝑚𝑎𝑥 – 𝑋𝑚𝑖𝑛
The quality characteristic to be mentioned in adequately modeled by a normally
distributed random variable.
The parameters µ and σ for the random variable are the same for each unit and each unit
is independent of its predecessors or successors.
The inspection procedure is same for each sample and is carried out consistently from
sample to sample.
Advantages
The chart is advantageous in following situations;
The sample size is relatively small.
The sample size is constant.
Humans must perform the calculations for the chart.
Graphical analysis
Graphical analysis is the first step in analyzing your data.
Examples:
Distribution (histogram, dotplot, boxplot)
Time Series plot for trending
I-chart (for Individual data points)
Normality
Dotplot of CONTROL Time Series Plot of CONTROL
80
75
70
CONTROL
65
60
50 55 60 65 70 75 80 55
CONTROL
50
1 7 14 21 28 35 42 49 56 63 70
Index
X=71.72 90
70 80
70
Percent
60
65 50
40
30
60 LCL=60.69 20
1 1 10
5
55
1
1
1 1
50 1
1 0.1
1 8 15 22 29 36 43 50 57 64 71 50 60 70 80 90
Observation CONTROL
Control charts for attributes
I. p-charts for proportion defective
It is a statistical control chart that plots movement in the sample proportion.
Procedure
1. Take a random sample and inspect each item.
2. Determine the sample proportion defective by dividing the number of defective items by
sample size.
3. Plot the sample proportion defective on the control chart and compare UCL and LCL to
determine if process is out of control.
The underlying statistical sampling distribution is the bionomial distribution but can be
approximated by the normal distribution with;
𝑀𝑒𝑎𝑛 = 𝜇 = 𝑛𝑝
𝑝(1 − 𝑝)
σ𝑃 = √[ ]
𝑛
Where,
p = historical population proportion defective
n = sample size
Control limits
𝑈𝐶𝐿 = µ + 𝑧𝜎𝑃
𝐿𝐶𝐿 = µ − 𝑧𝜎𝑃
z is the number of standard deviations from the mean. It is set based how certain you wish to be
that when a limit is exceeded it is due to a change in a process proportion defective rather than
due to sample variability.
Example:
If z =1, if p has not changed you will still exceed the limit in 32% of the samples (68% confident
that mean has changed if the limits are exceeded.
z = 2 – limits will be exceeded in 4.5 (95.5% confidence that mean has changed)
z = 3 – limits will be exceeded in 0.03 (99.7% confidence)
II. c-chart for number of defects per unit
c-chart a statistical control chart that plots movement in the number of defects per unit.
Procedure
1. Randomly select one item and count the number of defects in that item.
2. Plot the number of defects on a control chat.
3. Compare with UCL and LCL to determine if process is out of control.
The underlying sampling distribution is the poisson distribution but can be approximated by the
normal distribution with;
𝑀𝑒𝑎𝑛 = 𝑐
𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 = √𝑐
Where,
c = the historical average number of defect/ unit
Control limits
𝑈𝐶𝐿 = 𝑐 + 𝑧𝑐
𝐿𝐶𝐿 = 𝑐 – 𝑧𝑐
Confidence statement
In the case of attribute data sample size will determine the level that is reached with a confidence
statement. The higher the sample size used (with zero or minimal failures), the higher the
percent of population is when stating the confidence.
Below is a chart showing how sample sizes can affect the 95% confidence statements:
z 2s2
n 2
B
Where,
n = sample size to be calculated
z = table value for the specified confidence desired
s = estimated standard deviation
B = bound of the error estimation or ½ of the desired range of accuracy
The following simple formula may be used to estimate sample size to determine a proportion
(fraction) defective.
𝑛 = 𝑝 (1 − 𝑝) (𝑧 / 𝐵)
Where,
n = sample size to be calculated
z = table value for the specified confidence desired
s = estimated standard deviation
B = bound of the error estimation or ½ of the desired range of accuracy
p = estimate of the proportion fraction defective. If no estimate of p is available assume
worst case of p = 0.5
Example:
An engineer wants to estimate a sample size to determine the proportion of unacceptable attributes
that may be present in a manufacturing process, e.g., the number of molded components with flash
present on the parting line.
If a known history of scrap is already present in a similar product, then that proportion can
be used.
If the expected proportion is unknown, then you should use the worst case, or 0.5 as your
estimated proportion.
Let’s say the engineer does not know the proportion and uses 0.5 as the estimate.
He/she wants to know at 95% confidence what the sample size should be and is willing to
be accurate within ± 0.1.
1.96
𝑛 = 0.5 (0.5) ( )
0.1
96.04 𝑜𝑟 𝑟𝑜𝑢𝑛𝑑𝑒𝑑 𝑢𝑝, 97