M09_FOST7982_05_SE_Appendix.

indd Page 9A-1 15/06/12 5:40 PM user-s163

CHAPTER 9 APPENDIX
Acceptance Sampling Fundamentals

Our problem is not just the outgoing; it starts with the incoming.
—S. T. Foster

In Chapter 9, we introduced acceptance sampling. In this appendix, we start by discussing supply

chain fundamentals. We then move into creating sampling plans. To understand when you should
use these tools, refer to the discussion in Chapter 9.

ACCEPTANCE SAMPLING FUNDAMENTALS

We define acceptance sampling as a statistical quality control technique used in deciding to
accept or reject a shipment of input or output. When compared with statistical quality control,
acceptance sampling is defined by its occurrence after production has been completed. Accep-
tance sampling inspection can occur at the beginning of the process, such as when receiving
components, parts, or raw materials from a supplier. Or it can occur at the end of production, as
in the case of final inspection. Again, we focus here on inspection of incoming materials. One
interesting application of acceptance sampling occurs with seed producers. Because of the bio-
logical nature of seed, a supplier can place the seed on a truck in perfect condition, and the seed
can arrive in an unsuitable state. Therefore, large bulk purchasers of seed and other agricultural
products are avid users of acceptance sampling techniques.
Note that there are different methods for developing sampling plans. We discuss how to
develop sampling plans using Dodge and Romig tables and OC curves.

PRODUCER’S AND CONSUMER’S RISK

Producer’s risk is the risk associated with rejecting a shipment of materials that has good qual-
ity. Think of it this way. You are the producer of a product that has high quality. However, your
customer has concluded that your product has poor quality and returns the product to you. In this
case, you have been judged inaccurately. Consumer’s risk is the exact opposite. As a consumer,
you receive a shipment of poor-quality product and believe it has good quality. Therefore, you
pay for the product, use it in your production process, and suffer the consequences.
Producer’s risk is denoted by alpha () and is called a type I error. Consumer’s risk is
denoted by beta () and is referred to as a type II error. Table 9A-1 contrasts alpha and beta risk.
The goal of acceptance sampling is to reduce producer’s risk to low levels while maintaining
consumer’s risk at acceptable levels.

TABLE 9A-1 Alpha and Beta Risk

State of Nature

Product Is Good Product Is Defective

Consumer accepts product OK Consumer’s risk 
Outcome
Consumer rejects product Producer’s risk  OK

9A-1
M09_FOST7982_05_SE_Appendix.indd Page 9A-2 15/06/12 5:40 PM user-s163

9A-2

ACCEPTABLE QUALITY LEVEL

The acceptable quality level (AQL) is the maximum percentage or proportion of nonconform-
ing items or number of nonconformities in a lot or batch that can be considered satisfactory as a
process average. The AQL concept has been troublesome to many who consider it to condone an
acceptance of less-than-perfect quality. To statisticians, AQL simply denotes an economic deci-
sion that is associated with producer’s risk.

LOT TOLERANCE PERCENT DEFECTIVE (LTPD)

Lot tolerance percent defective (LTPD) is the level of poor quality included in a lot of goods.
The difference between AQL and LTPD is sometimes confusing to students. Lots of AQL or bet-
ter usually should have an alpha (that is, for example, a 5%) or less chance of rejection. AQL
relates to type I error or producer’s risk. Producer’s risk is the probability that good products will
be rejected by the consumer. For example, if a consumer concludes that an automobile is a lemon
even though it is a great vehicle, a type I error has occurred. Lots of LTPD or worse should have
a beta (that is, say, a 10%) or less chance of acceptance. LTPD relates to type II error or con-
sumer’s risk. Conversely, consumer’s risk is the chance that an automobile that seems perfect is
really a lemon. There is, theoretically, only one combination of sample size (n) and acceptance
number (c) that meets both conditions simultaneously. In practice, we will be unable to meet
both conditions precisely and must choose a combination of n and c that approximates both con-
ditions simultaneously.

n AND c
For the most part, the assignment of AQL, LTPD, alpha (), and beta () is a management deci-
sion. Once these values have been determined, n and c can be determined; these values in turn
define the sampling plan. The bottom line in acceptance sampling is that acceptance sampling
plans are designed to give us two things: n and c, where
n  the sample size of a particular sampling plan
c  a number that, if exceeded by the number of defectives in the sample, causes rejection
of the lot (acceptance number)
The average sampling plan can be stated in simple terms: n  20 and c  2. This clearly
communicates the bounds of the sampling plan. That is, take a sample of 20 items, and if more
than 2 are defective, reject the lot of materials. Remember that you should always randomize
when selecting products from a supplier to be inspected.

OC CURVES
The operating characteristic (OC) curve provides an assessment of the probabilities of accep-
tance for a shipment, given the existing quality of the shipment. An OC curve is constructed to
show the probability of accepting individual lots when the percent defective of the various indi-
vidual lots is known or assumed to be at a given level. These curves also can be used to develop
sampling plans.
Figure 9A-1 shows an OC curve for an optimal sampling plan. This ideal sampling plan
shows that given a probability of acceptance of 98%, this sampling plan should be used. In this
example, the probability of accepting a lot of goods with less than 3% defective is 100%. This
means the customer has considered the trade-offs and has determined that a lot of materials with
only 3% defective is acceptable given the circumstances.
However, OC curves never appear like the ideal case. Figure 9A-2 shows OC curves for a
sample size of n  100 and c  [0, 1, 2, 3]. Notice that as c gets smaller, the OC curve gets
M09_FOST7982_05_SE_Appendix.indd Page 9A-3 15/06/12 5:40 PM user-s163

Chapter 9 • Appendix 9A-3

Pa FIGURE 9A-1 Perfect

Operating Characteristic
(OC) Curve
1.00
Probability of acceptance

P'
.01 .02 .03 .04
Lot fraction nonconforming

steeper. Generally speaking, this means that higher values of c lead to higher probabilities of
accepting bad shipments (consumer’s risk). Also, higher values of n affect the OC curves in such
a way that we have greater confidence we have accepted a good shipment.
Figure 9A-3 shows an OC curve for n  50 and c  2 (for a lot of size 200; N, the entire
population  200). This shows that the probability of accepting (Pa) a lot with 2% defective
product is 92%. Also, the chance of accepting a lot of 12% defective product is 6%. This means
that if a lot has 2% defective, and this sampling plan is used (i.e., n  50, c  2), then the pro-
ducer’s risk is 8%. Notice that if the lot really has much poorer quality—say, 8% defective—the
chance of acceptance is quite low, about 24%. This means that there is a consumer’s risk of 24%
with the poor-quality lot.

FIGURE 9A-2 Various

OC Curves
1.0

Sample size
.8 n = 100
Probability of accepting shipment

.6

.4
c=3

c=2
c=1
.2
c=0

0 .10 .20 .30 .40 .50 .60 .70 .80

Quality of shipment
M09_FOST7982_05_SE_Appendix.indd Page 9A-4 15/06/12 5:40 PM user-s163

9A-4

Producer’s 1.0
Risk = α = .08
0.9

0.8

Probability of acceptance, Pa
0.7

0.6

0.5

0.4

0.3

0.2
Consumer’s
Risk = β = 0.24 0.1

0.0
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
Good quality Poor quality
(AQL) (LTPD)
Fraction nonconforming
FIGURE 9A-3 OC Curve for the Sampling Plan N ⴝ 200, n ⴝ 50, c ⴝ 2

EXAMPLE 9A-1 OC Curves

Problem: Let’s suppose we use a sampling plan with N  200, n  50, and c  2. In the past,
good-quality shipments from suppliers had about 1% defective, and poor-quality shipments had
about 6% defective. We desire to know producer’s and consumer’s risk. Using Figure 9A-3,
compute these probabilities.
Solution: From Figure 9A-3, it appears that the probability of acceptance of a good lot is 99%.
This translates to a producer’s risk of 1%. It also appears that the probability of accepting a poor
lot is 42%. This is consumer’s risk.

BUILDING AN OC CURVE
There are two ways to construct OC curves. The first uses the binomial distribution, and the sec-
ond, the Poisson distribution. For simplicity, we use the Poisson distribution. Using a sample size
n and average percent defective p, we can develop an OC curve using a Poisson approximation
of a binomial distribution.
M09_FOST7982_05_SE_Appendix.indd Page 9A-5 15/06/12 5:40 PM user-s163

Chapter 9 • Appendix 9A-5

c=3 4 c=5 c=6 7 8 9 10 15 20 30 40 c = 50

0.99999 0.99999
5
0.9999 c=2 4 50 0.9999
10
3 9
0.999 8 0.999
7
c=1 2 6
5
0.99 4 0.99
1 40
3
Probability of accepting shipment with

2
0.9 30 0.9
c or fewer defective units.

c=0 1
0.8 20 0.8
0.7 0.7
0.6 15 30 0.6
0
0.5
Pa

0.5
0.4 10 0.4
0.3 89 0.3
0.2 7 0.2
6
5
0.1 4 0.1
3
0.01 2 20 0.01
1
0
15
0.001 0.001

0.0001 0.0001
10
0.00001 0.00001
c=0 c=5
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 2 3 4 5 6 7 8 9 10 20 30
Value of np

FIGURE 9A-4 Probability Curves for Poisson Distribution Source: H. F. Dodge and H. G. Romig, Sampling Inspection Tables (New
York: John Wiley & Sons, 1959). Reprinted by permission of John Wiley & Sons, Inc.

Rather than computing the distributions from formulas, we rely on tables that have been
developed for simplicity in computing single-sampling plans. Single sampling means that we
use only one value of c to make a decision. Multiple-sampling plans are discussed later. The
Dodge–Romig table1 in Figure 9A-4 shows a table that was developed to simplify such calcula-
tions. While holding c and n constant, we can vary the percent defective p to calculate various
points on the OC curve.
Steps in using Figure 9A-4 to develop an OC curve:
1. Select values of p for percent defective in a shipment.
2. Multiply these values of np (where n is your sample size).
3. Using these values of np, go to Figure 9A-4 and find Pa (probability of acceptance).
4. Draw the OC curve.

EXAMPLE 9A-2 Developing Single-Sampling Plans Using OC Curves

and the Dodge-Romig Table

Problem: We would like to develop an OC curve for a good single-sampling plan using the
Dodge-Romig table provided in Figure 9A-4. Our sample size n is 60 and maximum c is 2.
Solution: Table 9A-2 shows the resulting values needed to construct the OC curve. The values of Pa
are drawn from the Dodge-Romig graph (Figure 9A-4). Figure 9A-5 shows the resulting OC curve.

1
Dodge, H., and Romig, H., Sampling Inspection Tables (New York: Wiley, 1959).
M09_FOST7982_05_SE_Appendix.indd Page 9A-6 15/06/12 5:40 PM user-s163

9A-6

TABLE 9A-2 OC Curve Probabilities (c  2)

p n np Pa p n np Pa
.005 60 0.30 .996 .055 60 3.30 .460
.015 60 0.90 .940 .065 60 3.90 .340
.025 60 1.50 .840 .075 60 4.50 .200
.035 60 2.10 .680 .085 60 5.10 .160
.045 60 2.70 .550 .095 60 5.70 .120
.050 60 3.00 .500 .105 60 6.30 .080

1.00 .996
.940
.90
.840
.80
Probability of acceptance of shipment

.70
.680

.60
.550
.50 .500
.460
.40
.340
.30

.20 .200
.160
.120
.10
.080

0 .01 .02 .03 .04 .05 .06 .07 .08 .09 .10 .11 .12
Quality of shipment (in fraction defective)
FIGURE 9A-5 Resulting OC Curve for n = 60, c = 2

ESTIMATING AQL AND LTPD

OC curves can be used to estimate both AQLs and LTPDs. MIL STD 105 (also known as ANSI/
ASQC Z1.4) is a military standard for sampling procedures and tables for attributes that state
that AQLs and LTPDs of sampling plans should always be documented. Figure 9A-6 shows an
OC curve for a single-sampling plan with n  50 and c  1. The AQL is generally stated with a
95% chance of acceptance (  0.05). In Figure 9A-6, the AQL value is 72%. The LTPD is gen-
erally stated at a 10% chance of acceptance. In this case, the LTPD value is 7.6%. At the LTPD,
90% of the lots are rejected.2

2
Taylor, W., “Selecting Statistically Valid Sampling Plans,” Quality Engineering 10, 2 (1997): 365–370.
M09_FOST7982_05_SE_Appendix.indd Page 9A-7 15/06/12 5:40 PM user-s163

Chapter 9 • Appendix 9A-7

1.0 FIGURE 9A-6

OC Curve of Single-
0.9 Sampling Plan n ⴝ 50
and c ⴝ 1 Source:
0.8 Adapted from W. Taylor,

Probability of acceptance
“Selecting Statistically
0.7 Valid Sampling Plans,”
Quality Engineering 10,
0.6 2 (1997): 365–370.
0.56
0.5
0.4
0.3
0.2
0.1
0.0
0 1 2 3 4 5 6 7 8 9 10
Percent defective
AQL = 0.72% 3% LTPD = 7.6%

MORE COMPLEX SAMPLING PLANS

So far we have focused on sampling plans for single samples. This is not the limit of sampling
plans. More complex sampling plans are referred to as multiple-sampling plans or sequential-
sampling plans. With these sampling plans, the acceptance sampling rules might occur as follows:
n1  sample size for sample 1
n2  sample size for sample 2
nn  sample size for sample n
c1  acceptance number for sample 1
c2  acceptance number for sample 2
cn  acceptance number for sample n
r1  rejection number for sample 1
r2  rejection number for sample 2
rn  rejection number for sample n
Multiple-sampling plans have advantages over single-sampling plans. The sample size
used in multiple-sampling plans will be smaller, on average, with the same amount of protection
as a single-sampling plan. The decision to use multiple-sampling plans usually will be made in
the first phase of the sample. Example 9A-3 provides an illustration of a double-sampling plan.

EXAMPLE 9A-3 Demonstrating a Double-Sampling Plan

Problem: An engineer adopted a double-sampling plan with the following values: n1  40, c1  1,
n2  60, c2  5. r1 and r2 are 5 and 6, respectively. Explain to management how this double-
sampling plan works.
Solution: Forty items from the incoming lot are drawn at random. If one or zero defective pieces
are found, accept the lot. If five or more defective pieces are found, reject the lot. If between two
M09_FOST7982_05_SE_Appendix.indd Page 9A-8 15/06/12 5:40 PM user-s163

9A-8

and four pieces inclusive are found defective, perform a second sample with 60 pieces. If the
combined number of defective pieces in both samples is less than or equal to five, the lot can be
accepted.

DEVELOPING DOUBLE-SAMPLING PLANS

Although OC curves can be used to develop double-sampling plans, the calculations to develop
the OC curves are much more complex than for single-sample plans. For this reason, we use a
standard approach that is used by many practitioners. The main limitation of this plan is that
sample sizes must be specified, as well as AQL, LTPD, producer’s risk, and consumer’s risk.
The double-sampling plan we develop has the following parameters:
AQL  0.020
LTPD  0.040
Producer’s risk  0.05
Consumer’s risk  0.10
n1  .5n2
First, we start by computing the ratio (R in Table 9A-3) of LTPD to AQL:
LTPD/AQL  0.040/0.020  2 (9A.1)
The parameters for the appropriate double-sampling plan are found in Tables 9A-3 and 9A-4.
Table 9A-3 is used when n1  n2, and Table 9A-4 is used when n1  0.5n2. For this example, we
use Table 9A-4 because n2 is twice the size of n1.

TABLE 9A-3 Values for Constructing a Double-Sampling Plan Having a Specified p1
and p2 (n1  n2,   0.05,   0.10)
Acceptance Approximate
Numbers Values of pⴕn1 Approximate
Plan (ASN)/n1 for
Number R ⴝ pⴕ2/pⴕ1 c1 c2 0.95 0.95 Pointa
1 11.90 0 1 0.21 1.170
2 7.54 1 2 0.52 1.081
3 6.79 0 2 0.43 1.340
4 5.39 1 3 0.76 1.169
5 4.65 2 4 1.16 1.105
6 4.25 1 4 1.04 1.274
7 3.88 2 5 1.43 1.170
8 3.63 3 6 1.87 1.117
9 3.38 2 6 1.72 1.248
10 3.21 3 7 2.15 1.173
11 3.09 4 8 2.62 1.124
12 2.85 4 9 2.90 1.167
13 2.60 5 11 3.68 1.166
14 2.44 5 12 4.00 1.215
15 2.32 5 13 4.35 1.271
16 2.22 5 14 4.70 1.331
17 2.12 5 16 5.39 1.452
a
ASN is without curtailment on the second sample.
Source: Chemical Corps Engineering Agency, Manual No. 2: Master Sampling Plans for Single, Duplicate, Double
and Multiple Sampling (Edgewood Arsenal, MD: Army Chemical Center, 1953).
M09_FOST7982_05_SE_Appendix.indd Page 9A-9 15/06/12 5:40 PM user-s163

Chapter 9 • Appendix 9A-9

TABLE 9A-4 Values for Constructing a Double-Sampling Plan Having a Specified p1
and p2 (n1  .5n2,   0.05,   0.10)
Acceptance Approximate
Numbers Values of pⴕn1 Approximate
Plan (ASN)/n1 for
Number R ⴝ p2/p1 c1 c2 0.95 0.95 Pointa
1 14.50 0 1 0.16 1.273
2 8.07 0 2 0.30 1.511
3 6.48 1 3 0.60 1.238
4 5.39 0 3 0.49 1.771
5 5.09 1 4 0.77 1.359
6 4.31 0 4 0.68 1.985
7 4.19 1 5 0.96 1.498
8 3.60 1 6 1.16 1.646
9 3.26 2 8 1.68 1.476
10 2.96 3 10 2.27 1.388
11 2.77 3 11 2.46 1.468
12 2.62 4 13 3.07 1.394
13 2.46 4 14 3.29 1.472
14 2.21 3 15 3.41 1.888
15 1.97 4 20 4.75 2.029
16 1.74 6 30 7.45 2.230
a
ASN is without curtailment on the second sample.
Source: Chemical Corps Engineering Agency, Manual No. 2: Master Sampling Plans for Single, Duplicate, Double
and Multiple Sampling (Edgewood Arsenal, MD: Army Chemical Center, 1953).

Because LTPD/AQL was 2, in Table 9A-4 we find we will use double-sampling plan num-
ber 15 (because 1.97 is closest to our value of 2). We see immediately that c1 and c2 are 4 and 20,
respectively. Next, we compute the sample size by using either column 5 or 6 from the table. If
we want to use a probability of acceptance of 0.95, we use column 5. For sample plan 15, the
value of pn1 is 4.75. Then we compute n1 as
n1  p'n0.95/AQL  4.75/0.020  237.5 ⬵ 238 (9A.2)
As we can see, n1  238 and n2 = 475 (remember that n2 is twice the size of n1). These are
rather large sample plans. We may want to reconsider our parameters to reduce these sample
sizes.
We also can use this approach to determine a double-sampling plan if we want to maintain
consumer’s risk at a low level. To do this, we use column 7 of Table 9A-4. If we wish to hold that
value of consumer’s risk to 0.10, we can use the column 7 value of plan number 15 of 9.35.
Using Formula 9A.2:
n1  pn0.95/AQL
9.35/0.020  467.5 or 468 (rounding)
n2  935 (remember that n2 is twice the size of n1, before rounding)
c1 and c2 are 4 and 20, respectively
Again, because this sampling plan has large sample sizes, we may be wise to experiment
with other parameters to find alternative sampling plans that meet targets for acceptance.
M09_FOST7982_05_SE_Appendix.indd Page 9A-10 15/06/12 5:40 PM user-s163

9A-10

Summary
The single- and double-sampling plans we have discussed are called lot-by-lot sampling plans.
As we perform separate samples, we receive additional lots of materials. Sometimes it is not
feasible to collect products into lots because they are produced in a continuous manner. In these
cases, acceptance sampling procedures for continuous production are used. These procedures
typically involve alternating between 100% inspection and sampling inspection. Although we
did not develop a methodology for continuous acceptance sampling here, MIL STD 1235C
(1988), named “Single and Multiple-Level Continuous Sampling Procedures and Table for
Inspection by Attributes,” is available from the Department of Defense. Any of the military stan-
dards mentioned in this appendix can be ordered from the Department of Defense directly by
accessing the DOD SSP on the Internet (Department of Defense Single Stock Point for Specifi-
cations and Standards). The URL is http://dodssp.daps.dla.mil/.

Key Terms
Acceptable quality level Lot tolerance percent Operating characteristic Producer’s risk
(AQL) defective (LTPD) (OC) curve Sampling plan
Consumer’s risk

Discussion Questions
1. Define producer’s risk and consumer’s risk.
2. Define the concept of acceptable quality level. Why has this concept been troublesome to many
people?
3. What is the term used to designate the level of poor quality included in a lot of goods? Describe the
role of this term in the quality management process.
4. What is an operating characteristic curve? What is the function of this curve in the quality manage-
ment process?

Problems
1. Using Figure 9A-2, with a sample size of n  100 and an acceptance number c  1, if a good shipment
has no more than 0.05 defective, what is the probability of acceptance? What type of risk is this?
2. Using Figure 9A-2, with a sample size of n  100 and an acceptance number of c  2, if a bad ship-
ment has 40% defective, what is the probability of acceptance? What type of risk is this?
3. Using Figure 9A-2, with a sample size of n  100 and an acceptance number of c  3, if a good ship-
ment has no more than .02 defective, what is the probability of acceptance? What type of risk is this?
4. Using Figure 9A-2, with a sample size of n  100 and an acceptance number of c  1, if a good ship-
ment has no more than .30 defective, what is the probability of acceptance? What type of risk is this?
5. Develop an OC curve using Figure 9A-4, a sample size of 100, and the following p values: .01, .02,
.03, .05, .07, .09, .11, .13, .15, .17, .19, .21. The maximum acceptance number is c  2.
6. From the OC curve developed in Problem 5, if a good shipment is 0.02 defective, what is the probabil-
ity of a type I (producer’s) error?
7. From the OC curve developed in Problem 5, if a bad shipment is defined as having at least 10% defec-
tive, what is your estimate of type II (consumer’s) risk?
8. Develop an OC curve using Figure 9A-4, a sample size of 100, and the following p values: .01, .02,
.03, .05, .07, .09, .11, .13, .15. The maximum acceptance number is c  4.
M09_FOST7982_05_SE_Appendix.indd Page 9A-11 15/06/12 5:40 PM user-s163

Chapter 9 • Appendix 9A-11

9. From the OC curve developed in Problem 8, if a good shipment is .04 defective, what is the probability
of a type I (producer’s) error?
10. From the OC curve developed in Problem 8, if a bad shipment is defined as having at least 15% defec-
tive, what is your estimate of type II (consumer’s risk) error?
11. Develop an OC curve using Figure 9A-4, a sample size of 100, and the following p values: 0.01, 0.02,
0.03, 0.05, 0.07, 0.09, 0.11, 0.13, 0.15, 0.17, 0.19, 0.21. The maximum acceptance number is c  3.
12. From the OC curve developed in Problem 11, if a good shipment has 5% defective, what is the prob-
ability of type I error? How is type I error defined?
13. From the OC curve in Problem 11, if a bad shipment has more than 20% defective, what is the proba-
bility of accepting a bad shipment?
14. Interpret the following sampling plan in plain English:
n1  50
c1  2
n2  100
c2  5
r1  4
r2  6
15. Interpret the following sampling plan in plain English:
n1  125
c1  3
n2  150
c2  6
n3  200
c3  12
16. We wish to develop a double-sampling plan where n1  n2 (see Table 9A-3). Here are the needed
parameters:
AQL  .010
LTPD  .030
Producer’s risk  .05
Consumer’s risk  .10
17. Rework Problem 16 where n2 is twice the size of n1 (see Table 9A-4).
18. We wish to develop a double-sampling plan where n1  n2 (Table 9A-3). Here are the needed
parameters:
AQL  .020
LTPD  .080
Producer’s risk  .05
Consumer’s risk  .10
19. Rework Problem 18 where n2 is twice the size of n1 (Table 9A-4).
20. Your boss wants you to develop a double-sampling plan where n1  n2 (using Table 9A-3). Here are
some parameters for your use:
AQL  .015
LTPD  .040
Producer’s risk  .05
Consumer’s risk  .10
21. Rework Problem 20 where n2 is twice the size of n1 (Table 9A-4).