An American National Standard

Designation: E 141 – 91 (Reapproved 2003)

Standard Practice for

Acceptance of Evidence Based on the Results of Probability
Sampling1
This standard is issued under the fixed designation E 141; the number immediately following the designation indicates the year of
original adoption or, in the case of revision, the year of last revision. A number in parentheses indicates the year of last reapproval. A
superscript epsilon (e) indicates an editorial change since the last revision or reapproval.

1. Scope E 178 Practice for Dealing with Outlying Observations2

1.1 This practice presents rules for accepting or rejecting E 456 Terminology for Statistical Methods2
evidence based on a sample. Statistical evidence for this NOTE 1—Practice E 105 provides a statement of principles for guidance
practice is in the form of an estimate of a proportion, an of ASTM technical committees and others in the preparation of a sampling
average, a total, or other numerical characteristic of a finite plan for a specific material. Practice E 122 aids in deciding on the required
population or lot. It is an estimate of the result which would sample size. Practice E 178 helps insure better behaved estimates.
have been obtained by investigating the entire lot or population Terminology E 456 provides definitions of statistical terms used in this
standard.
under the same rules and with the same care as was used for the
sample. 3. Terminology
1.2 One purpose of this practice is to describe straightfor- 3.1 Definitions:
ward sample selection and data calculation procedures so that

--`,````,``,,,,,,````,``,`````,-`-`,,`,,`,`,,`---
3.1.1 Equal Complete Coverage Result, n—the numerical
courts, commissions, etc. will be able to verify whether such characteristic (u) of interest calculated from observations made
procedures have been applied. The methods may not give least by drawing randomly from the frame, all of the sampling units
uncertainty at least cost, they should however furnish a covered by the frame.
reasonable estimate with calculable uncertainty. 3.1.1.1 Discussion—Locating the units and evaluating them
1.3 This practice is primarily intended for one-of-a-kind are supposed to be done in exactly the same way and at the
studies. Repetitive surveys allow estimates of sampling uncer- same time as was done for the sample. The quantity itself is
tainties to be pooled; the emphasis of this practice is on denoted u. The equal complete coverage result is never actually
estimation of sampling uncertainty from the sample itself. The calculated. Its purpose is to serve as the objectively defined
parameter of interest for this practice is effectively a constant. concrete goal of the investigation. The quantity u may be the
Thus, the principal inference is a simple point estimate to be population mean, ( Ȳ), total (Y), median (M), the proportion
used as if it were the unknown constant, rather than, for (P), or any other such quantity.
example, a forecast or prediction interval or distribution 3.1.2 frame, n—a list, compiled for sampling purposes,
devised to match a random quantity of interest. which designates all of the sampling units (items or groups) of
1.4 This standard does not purport to address all of the a population or universe to be considered in a specific study.
safety concerns, if any, associated with its use. It is the 3.1.2.1 Discussion—The list may cover a specific shipment
responsibility of the user of this standard to establish appro- or lot, all households in a county, a state or country; for
priate safety and health practices and determine the applica- example, any population of interest. Every sampling unit in the
bility of regulatory limitations prior to use. frame (1) has a unique serial number, which may be preas-
2. Referenced Documents signed or determined by some definite rule, (2) has an
address—a complete and clear instruction (or rules for its
2.1 ASTM Standards:
formulation) as to where and when to make the observation or
E 105 Practice for Probability Sampling of Materials2
evaluation, (3) is based on physically concrete clerical mate-
E 122 Practice for Choice of Sample Size to Estimate a
rials such as directories, dials of clocks or of meters, ledgers,
Measure of Quality for a Lot or Process2
maps, aerial photographs, etc., referred to in the addresses.
3.1.3 sample, n—a group of items, observations, test results,
1
This practice is under the jurisdiction of ASTM Committee E11 on Quality and or portions of material, taken from a larger collection of such
Statistical Methods and is the direct responsibility of Subcommittee E11.10 on items; it provides information for decisions concerning the
Sampling and Data Analysis. larger collection.
Current edition approved August 15, 1991. Published November 1991. Origi-
nally published as E 141 – 59 T. Last previous edition E 141 – 69 (1975).
2
Annual Book of ASTM Standards, Vol 14.02.

Copyright © ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959, United States.

Copyright ASTM International 1

Provided by IHS under license with ASTM Licensee=Aramco HQ/9980755100
No reproduction or networking permitted without license from IHS Not for Resale, 10/08/2006 02:13:45 MDT
 E 141 – 91 (2003)
3.1.3.1 Discussion—A particular sample is identified by the 5.1.4 Such plans are defined and their relative advantages
set of serial numbers from the randomization device and by the discussed in [1], [2] and [6].
addresses on the frame generated by those serial numbers. 5.2 Replicate Subsamples—a number of disjoint samples,
3.1.4 sampling unit, n—an item, test specimen or portion of each one separately drawn from the frame in accord with the
material that is to be subjected to evaluation as part of the same probability sampling plan. When appropriate, separate
sampling plan. laboratories should each work on separate replicate subsamples
3.1.4.1 Discussion—If it is not feasible to select test speci- and teams of investigators should be assigned to separate
mens or laboratory samples individually, the sampling unit replicate subsamples. This approach insures that the calculated
may be a group of items, for example, a row, an entire case of standard error will not be a systematic underestimate. Such
items, or a prescribed area (as in the examination of a finishing subsamples were called interpenetrating in [5] where many of
process). their basic properties were described. See [2] for further theory
3.1.4.2 By a more expensive method of measurement (fu- and applications.
ture time, more elaborate frame) it may be possible to define a
5.2.1 Discussion—For some types of material a sample
quantity, u8, as a target parameter or ideal goal of an investi-
gation. Criticism that holds u to be an inappropriate goal selected with uniform spacing along the frame (systematic
should demonstrate that the numerical difference between u sample) has increased precision over a selection made with
and u8 is substantial. Measurements may be imprecise but so randomly varying spacings (simple random sample). Examples
long as measurement errors are not too biased, a large size of include sampling mineral ore or grain from a conveyor belt or
the lot or population, N, insures that u and u8 are essentially sampling from a list of households along a street. If the
equal. systematic sample is obtained by a single random start the plan
is then a probability sampling plan, but it does not permit
4. Significance and Use calculating the standard error as required by this practice. After
4.1 This practice is designed to permit users of sample dividing the sample size by an integer k (such as k = 4 or k
= 10) and using a random start for each of k replicate
--`,````,``,,,,,,````,``,`````,-`-`,,`,,`,`,,`---

survey data to judge the trustworthiness of results from such

surveys. Section 5 gives extended definitions of the concepts subsamples, some of the increased precision of systematic
basic to survey sampling and the user should verify that such sampling (and a standard error on k − 1 degrees of freedom)
concepts were indeed used and understood by those who can be achieved.
conducted the survey. What was the frame? How large (ex- 5.2.2 Audit Subsample—a small subsample of the survey
actly) was the quantity N? How was the parameter u estimated sample (as few as 10 observations may be adequate) for review
and its standard error calculated? If replicate subsamples were of all procedures from use of the random numbers through
not used, why not? locating and measurement, to editing, coding, data entry and
4.2 Adequate answers should be given for all questions. tabulation. Selection of the audit subsample may be done by
There are many acceptable answers to the last question. If the putting the n sample observations in order as they are collected,
sample design was relatively simple, such as simple random or calculating the nearest integer to =n , or some other conve-
stratified, then good estimates of sampling variance are easily nient integer, and taking this number to be the spacing for
available. If a more complex design was used then methods systematic selection of the audit subsample. The review should
such as discussed in [1] may be acceptable. Replicate sub- uncover any gross departures from prescribed practices or any
samples is the most straightforward way to estimate sampling conceptual misunderstandings in the definitions. If the audit
variances when the survey design is complex. subsample is large enough (say 30 observations or more) the
4.3 Once the survey procedures that were used satisfy regression of audited values on initial observations may be
Section 5, consult Section 4 to see if any increase in sample used to calibrate the estimate. This technique is the method of
size is needed. The calculations for making it are objectively two-phase sampling as discussed in [1]. Helpful discussion of
described in Section 4. an audit appears in [2].
4.4 Refer to Section 6 to guide in the interpretation of the
5.2.3 Estimate—a quantity calculated on the n sample
uncertainty in the reported value of the parameter estimate, u,
observations in the same way as the equal complete coverage
i.e. the value of its standard error, se(u). The quantity se(u)
result u would have been calculated from the entire set of N
should be reviewed to verify that the risks it entails are
possible observations of the population; the symbol u denotes
commensurate with the size of the sample.
the estimate. (In calculating u, replicate subsample member-
5. Descriptive Terms and Procedures ship is ignored.)
5.1 Probability Sampling Plans—include instructions for 5.2.3.1 Discussion—An estimate has a sampling distribu-
using either: tion induced from the randomness in sample selection. The
5.1.1 carefully prepared tables of random number, equal complete coverage result is effectively a constant while
5.1.2 computer algorithms, carefully programmed and run any estimate is only the value from one particular sample.
on a large computer, to generate pseudo-random numbers or, Thus, there is a mean value of the sampling distribution and
5.1.3 certifiably honest physical devices, such as coin flips, there is also a standard deviation of the sampling distribution.
to select the sample units so that inferences may be drawn from 5.2.4 Standard Error—the quantity computed from the
the test results and decisions may be made with risks correctly observations as an estimate of the sampling standard deviation
calculated by probability theory. of the estimate; se(u) denotes the standard error.

Copyright ASTM International 2

Provided by IHS under license with ASTM Licensee=Aramco HQ/9980755100
No reproduction or networking permitted without license from IHS Not for Resale, 10/08/2006 02:13:45 MDT
 E 141 – 91 (2003)
5.2.4.1 Example 1—When u is the population average of the where u is the average of the ui. The standard error is based
N quantities and a simple random sample of size n was drawn, on k − 1 degrees of freedom.
then the sample average ȳ becomes the usual estimate u, where The following estimates of percent “drug-in-suit” sales of
n prescription drugs were based on 20 replicate subsamples; each
u 5 ȳ 5 ( yi /n.
i51
(1) followed a stratified cluster sampling design. The separate
estimates were: 6.8, 7.1, 8.4, 9.5, 8.6, 4.1, 3.7, 3.2, 3.8, 5.8, 8.8,
The quantities y1, y2, ..., yn denote the observations. The 5.0, 7.9, 8.8, 8.4, 8.1, 6.0, 6.3, 4.5, 5.8. The value of u was
standard error is calculated as: 6.74 % and se(u) = 0.43 % on 19 degrees of freedom. Notice
se ~u! 5 se ~ȳ! 5 Œ( n

i51
~yi 2 ȳ!2/n~n 2 1!. (2)
that u = 6.58 does not equal u = 6.74. This is because u is a
ratio of two overall averages while u is the average of 20
There are n − 1 degrees of freedom in this standard error. ratios. For an example with k = 2, average 1⁄3 and 3⁄5 and
When the observations are: compare to (1 + 3)/(3 + 5).
81.6, 78.7, 79.7, 78.3, 80.9, 79.5, 79.8, 80.3, 79.5, 80.7 5.2.5 Procedures—must be described in written form and
then ȳ = 79.90 and se(ȳ) = .32. As this example illustrates, should cover the following matters; (1) parties interested in
formula (2) is correct when k replaces n and subsample collecting data should agree on the importance of knowing u
estimates are used in place of observations. and its definition including measurement methods, (2) the
5.2.4.2 Example 2 on the Finite Population Correction frame shall be carefully and explicitly constructed; N shall be
(fpc)—Multiplying se ( ȳ) by =1 2 n/N is always correct well established, (3) random numbers (or a certifiably honest
when the goal of the survey is to estimate the finite population physical random device) shall dictate selection of the sample.
mean (u = Ȳ). Using the previous data and if N = 50, then There will be no substitution of one sampling unit for another.
se(ȳ) becomes se(ȳ) = .28 after applying the fpc. If random The method of sample selection shall permit calculation of a
measurement error exists in the observations, then u8 based on standard error of the estimate (4) the use of replicate sub-
a reference measurement method may be a more appropriate samples is recommended (see section 5.24.2.2); an audit
survey goal than u (see section 4.1.4.1). If so, then se(ȳ) would subsample should be selected and processed and any depar-
be further adjusted upward by an amount somewhat less than tures from prescribed measurement methods and location
the downward adjustment of the fpc. Both of these adjustments instructions noted (see 5.2.2). A report should list u and its
are often numerically so small that these adjustments may be standard error with the degrees of freedom in the se( u).
omitted—leaving se(ȳ) of (2) as a slight overestimate.
5.2.4.3 Example 3—If the quantity of interest is (a) a 6. Adequacy of Sample Size
proportion or (b) a total and the sample is simple random then 6.1 Deciding on Increasing Sample Size: Choice of sample
the above formulas are still applicable. A proportion is the size should be made carefully in accordance with Practice
mean of zeroes and ones, while the total is a constant times the E 122 or on a comparable basis. Since procedures for setting
mean. Thus: sample size are based on judgements of the variability to be
(a) when u is taken to be the population proportion (u = P) encountered, there is a possibility that the standard error as
then; calculated from the data could greatly exceed that anticipated.
u 5 p 5 (yi/n 5 a/n (3) It may happen that the time period of interest for the population
has passed or for some other reasons it is not possible to take
where:
more observations and thus the following discussion should be
a is the number of units in the sample with the attribute, and ignored. Otherwise, a decision may be made to increase the
se~p! 5 =p~1 2 p!/~n 2 1! (4) number of replicate subsamples or even to carry out a census
of the universe. Such decisions must be made strictly indepen-
(b) when u = the population total (u = Y) then dent of knowledge of u, for example, in adversarial settings one
party may feel the size of u is inappropriate and will seek to
u 5 Nȳ and se ~u! 5 Nse ~ȳ! (5)
have it changed. Therefore experimental protocols along with
the standard error should be reviewed prior to announcement of
If a simple random sample of size n = 200 has a = 25 items the estimate u. Once all parties are satisfied (methods are
with the attribute then the conclusion is u = .125 and sound, standard error adequate) then the estimate can be
se(u) = .023 on 199 degrees of freedom. furnished.
5.2.4.4 Example 4. If u is a parameter other than a mean or 6.2 Increasing Sample Size by Calculating Costs and
if the sample design is complex, then replicate subsamples Losses: To assume that u = u is to make a judgement that the
should be used in the sample design. Denote the k separate cost of decreasing se(u) by increasing sample size is greater
estimates as ui, i = 1, 2, ..., k and denote by u the estimate based than the risks stemming from u not equal to u. If n is to be
on the whole sample. The average of the ui will be close to, but increased it is necessary to understand the survey costs as well
in general not equal to u. The standard error of u is calculated as the costs of inaccuracies in u. Survey costs are determined
as: through ordinary cost accounting procedures. In judging the
se ~u! 5 Œ( k

i51
~ui 2 u!2/k~k 2 1! (6)
seriousness of inaccuracy in u one needs to imagine losses
entailed if u were one standard error below u and above u.
Calculate these two losses and divide the average by two. This
--`,````,``,,,,,,````,``,`````,-`-`,,`,,`,`,,`---

Copyright ASTM International 3

Provided by IHS under license with ASTM Licensee=Aramco HQ/9980755100
No reproduction or networking permitted without license from IHS Not for Resale, 10/08/2006 02:13:45 MDT
 E 141 – 91 (2003)
result represents roughly the gain to be expected by quadru- 7.4 Bounds on a Proportion in a Large Population When
pling sample size. If the cost of increasing sample size from n Zero is Observed in a Sample: It can happen, after observing a
to 4n is appreciably less than the above gain there is a basis for random sample of size n, that a = 0; that is, there are no
increasing sample size. observations showing the attribute among the n. In this case an
6.2.1 Example 5—The estimate of percentage“ drug-in- upper bound with level of significance a is computed as:
suit” sales (see 5.2.4.4, Example 4) was to be used in
u~U! 5 1 2 a1/n. (11)
determining how much drug companies might have to pay to
the state of North Carolina. Thus losses from inaccuracy in u in For example, with n = 18 and a = .05, u(U) = .15. Any
this example were relatively clear. The base sales of all values of the population proportion less than 15 % cannot be
prescription drugs in North Carolina was 700 million dollars. ruled out. See Example 8 below for a more exact treatment of
About 10 % of “drug-in-suit” sales could be judged as over- the case when N is of moderate size.
pricing. An initial sample of only four replicate subsamples 7.5 Nonnormality of the u Distribution:
was taken and the se( u) was found to be 0.7 %. Thus an
7.5.1 If any one of the observations is very much smaller or
overstatement by one standard error would represent a loss to
larger than the rest it should be investigated. If there is marked
the drug companies of .007 3 700 3 .1 = .49 million dollars,
asymmetry in the distribution of the observations (for example,
while an understatement of the same amount would be the
there are apparent outliers on one side of the average), be
same loss to the state of North Carolina. The average is
cautious in trusting the realism of a when calculating bounds
$490,000 and half of this is $245,000. Perhaps, from the
and confidence limits. If the following estimate of skewness in
court’s viewpoint, not all of this is loss since what one party
u is not larger in absolute value than .3, then the change in
overpays, the other gains. Still the survey would have cost
100a will likely be less than 1 % due to skewness.
approximately $50,000 to quadruple in size so it was decided
n
to take the total of 20 subsamples reported on in the above
g1 ~u! 5 @ ( ~ui 2 u!3/~n 2 1!~n 2 2!#/@se~u!#3n. (12)
Example 4. i51

For the example of the 20 replicate subsample estimates of

7. Reporting Results proportion“ drug-in-suit,” g1(u) = −.054 which is far from
7.1 Basic Technical Report: critical. The value .3 in the above rule is a relaxed form of that
7.1.1 The estimate of u should be reported as “u with a given on page 42 of [1].
standard error of se(u) on n degrees of freedom.” This form 7.5.1.1 Example 8 on Bounding Proportions Near Zero in
emphasizes the quantity u which is to be taken in practice to be Finite Populations. If a simple random sample of size n = 200
the value of u. It also permits the user to rule out values of u has a = 3 items with the attribute, then the conclusion (see
as improbable under the evidence, by simple calculations based Example 3 above for comparison) is u = p = .0150 and
on widely available tables of the Student t distribution. se(p) = .0086. Treating the observations as 197 zeroes and 3
7.2 Upper and Lower Bounds on u: ones, allows the quantity g1(u) to be computed as .57 which
7.2.1 Values of u that can be ruled out because they are (a) exceeds .3 and warns of skewness in the distribution of u.
too large or (b) too small, can be calculated as follows: Further suppose that there are N = 800 items in a lot and let A
~a! Upper bound u~U! 5 u 1 ta ~n! se ~u!, or (7) be the unknown number of items with the attribute and that an
~b! Lower bound u~L! 5 u 2 ta ~n! se ~u!, (8) upper bound is needed for A. An upper bound with a 97.5 %
coefficient (a = .025) would be found from the formula in 4.2
where ta(n) is the value from the Student t distribution such as .0150 + 2(.0086) = .0322. This suggests that
that 100a percent of the distribution exceeds ta(n). A hypoth- 800 3 .0322 = 25.8 is an upper 97.5 % bound on A, but
esized value of u equal to or larger than u(U) would be rejected skewness may upset this. In fact setting A = 32 is required to
by the sample evidence at the a level of significance. For drive the chance of observing a = 3 or less to 2.3 %, while
values of t see, for example, [3]. For the theory of bounds see setting A = 31 only drives it to 2.8 %. The bound itself is set at
[4]. the half-integer A(U) = 31.5 since larger values can be ruled
7.2.2 Example 6. For the percent drug-in-suit data a lower out at the a = .025 level.
bound with 5 % level of significance is found as: 7.5.1.2 These probabilities (of .023 and of .028) are calcu-
u 5 6.74 2 1.729 3 .43 5 6.00, (9) lated from the hypergeometric distribution, see [1], which is
the sampling distribution of a when the sample is simple
random. The full formula is
where 1.729 is t.95 from [3] with n = 19.
7.3 Confidence Limits: The quantities u(U) and u(L) Pr ~a items with the attribute among n in the sample! (13)
calculated using ta/2(n) in place of ta(n), define an interval A!~N 2 A!!~N 2 n!! n!
(from u(L) to u(U)) called a 100a percent confidence interval. 5
a!~A 2 a!! ~n 2 a!! ~N 2 A 2 n 1 a!! N!
Again, see [4] or [1].
7.3.1 Example 7. For the percent condition estimate a 95 %
7.5.1.3 In practice, a hand held calculator may be used for
confidence interval would be from:
the needed probabilities to set bounds. The first step is to find
79.9 2 2.262 3 .32 to 79.9 1 2.262 3 .32, or (10) the probability for a = 0 as a sequence of A alternating
from 79.2 to 80.6, where 2.262 5 t.975 from @3# with n 5 9. multiplications and divisions where (from the above formula):
--`,````,``,,,,,,````,``,`````,-`-`,,`,,`,`,,`---

Copyright ASTM International 4

Provided by IHS under license with ASTM Licensee=Aramco HQ/9980755100
No reproduction or networking permitted without license from IHS Not for Resale, 10/08/2006 02:13:45 MDT
 E 141 – 91 (2003)
~N 2 n!~N 2 n 2 1!—~N 2 n 2 A 1 1! TABLE 1 Student t Values Required for Use with a Standard
Pr ~a 5 0! 5 . (14) Error on n Degrees of Freedom to Attain a = .0027
N~N 2 1!—~N 2 A 1 1!
n ta/2(n) n ta/2(n) n ta/2(n)

7.5.1.4 Alternating multiplications with divisions will help 1 235.78A 11 3.85 21 3.40
2 19.21 12 3.76 22 3.38
avoid rounding errors. Probabilities for a = 1, a = 2, etc. can 3 9.22 13 3.69 23 3.36
now be obtained recursively by at most two multiplications and 4 6.62 14 3.64 24 3.34
two divisions. For the case of A = 31, n = 200 and N = 800, 5 5.51 15 3.59 25 3.33
6 4.90 16 3.54 26 3.32
Pr(a = 0) = .0001096913978. Multiplying by (31/570) and 7 4.53 17 3.51 27 3.30
(200/1) brings us to Pr(a = 1); further multiplying by (30/571) 8 4.28 18 3.48 28 3.29
and (199/2) gets us to Pr(a = 2); and, finally, multiplying by 9 4.09 19 3.45 29 3.28
10 3.96 20 3.42 30 3.27
(29/572) and (198/3) produces Pr(a = 3) = .02087. Adding 40 3.20
these four (4) results gives .02841 which is above .025 and 50 3.16
requires us to go on to the case of A = 32. ` 3.00
A
7.5.1.5 A lower bound on A could also be found by trial and Note—When used to calculate an exact three sigma interval this value is
ta/2(n) = 235.80 for an exact a/2 = .001349898 and n = 1.
error after setting A = 3, A = 4, and so forth until the probabil-
ity of 3 or more, first exceeds a. When we set A = 3 the
(15)
probability of a = 3 becomes .0154 so that A = 3 can be ruled
out by the evidence at the a = .0154 level of significance. 8. Review of Important Points
However, when we set A = 4 the probability of getting a = 3 or
8.1 Probability sampling as practiced in accord with 5.2 is a
a = 4 is found to be .0503 and so A(L) = 3.5 becomes the lower
procedure by which one obtains a result from a selected set of
bound.
sampling units that will agree, within calculable limits of
7.5.1.6 Using the above formula for Pr(a = 0) it is now variation, with the equal complete coverage result.
possible to furnish an upper bound on the population propor- 8.2 The equal complete coverage result may or may not be
tion when zero proportion is observed, that takes account of the acceptable evidence. Whether it is acceptable depends on many
size of the finite population. We take the example of n = 20 considerations such as definitions, method of test, care exer-
with a = 0 and suppose N = 100. For A = 14 Pr(a = 0) is found cised in the testing, completeness of the frame, and on other
to be (80/100)(79/99)—(67/87) = .03413, while points not to be settled by statistical theory since these points
Pr(a = 0) = .0443 for A = 13, but Pr(a = 0) = .0574 when belong to the subject matter, and are the same whether one uses
A = 12. Thus an upper 5 % bound on A is set at A(U) = 12.5 sampling or not. Mistakes, whether in testing, counting, or
when N = 100 or the upper bound on the finite population weighing will affect the result of a complete coverage just as
proportion becomes .125. such mistakes will affect the sample result.
7.6 Extreme Security Limits: The extreme variation of an 8.3 When the audit subsample shows that there was reason-
--`,````,``,,,,,,````,``,`````,-`-`,,`,,`,`,,`---

estimate (from a probability sample) can often be placed at an able conformity with prescribed procedures and when the
interval of three standard deviations above or below the sample known instances of departures from the survey plan can be
result. When the sample is of sufficient size, only 27 out of shown to have no appreciable effect on the estimate, then the
10,000 intervals so calculated would not be expected to cover value u should be used.
the universe value. Table 1 shows values for ta/2 where
a = .0027, and gives some idea of the effect of having to 9. Keywords
estimate the standard deviation rather than using previous 9.1 Probability sampling; sampling unit; sampling frame;
knowledge of it. equal complete coverage; replicate subsamples; audit sub-
7.6.1 Example 9. For the percent condition estimate, ex- sample; sampling distribution; estimate; standard error; popu-
treme security limits would be set at: lation parameter; finite population correction; sample size
79.90 2 ~4.09 3 .32! to 79.90 1 ~4.09 3 .32! adequacy; confidence limits; skewness; hypergeometric
or from 78.59 to 81.21. distribution.

Copyright ASTM International 5

Provided by IHS under license with ASTM Licensee=Aramco HQ/9980755100
No reproduction or networking permitted without license from IHS Not for Resale, 10/08/2006 02:13:45 MDT
 E 141 – 91 (2003)
REFERENCES

(1) Cochran, William G., Sampling Techniques, Third Edition, John Wiley Wiley and Sons, New York, NY, 1986.
and Sons, New York, NY, 1977. (5) Mahalanobis, P. C., “Recent Experiments in Statistical Sampling in the
(2) Deming, W. Edwards, Sample Design in Business Research, John Indian Statistical Institute,” Journal of the Royal Statistical Society,”
Wiley and Sons, Inc., New York, NY, 1960. 109:325–370, 1946.
(3) Dixon, W. J., and Massey, F. J, Introduction to Statistical Analysis,
Fourth Edition, McGraw-Hill, Inc., New York, NY, 1983. (6) Yates, Frank, Sampling Methods for Censuses and Surveys, Fourth
(4) Lehmann, E. L, Testing Statistical Hypotheses, Second Edition, John Edition, MacMillan Publishing Co., Inc., New York, NY, 1981.

ASTM International takes no position respecting the validity of any patent rights asserted in connection with any item mentioned
in this standard. Users of this standard are expressly advised that determination of the validity of any such patent rights, and the risk
of infringement of such rights, are entirely their own responsibility.

This standard is subject to revision at any time by the responsible technical committee and must be reviewed every five years and
if not revised, either reapproved or withdrawn. Your comments are invited either for revision of this standard or for additional standards
and should be addressed to ASTM International Headquarters. Your comments will receive careful consideration at a meeting of the
responsible technical committee, which you may attend. If you feel that your comments have not received a fair hearing you should
make your views known to the ASTM Committee on Standards, at the address shown below.

This standard is copyrighted by ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959,
United States. Individual reprints (single or multiple copies) of this standard may be obtained by contacting ASTM at the above
address or at 610-832-9585 (phone), 610-832-9555 (fax), or [email protected] (e-mail); or through the ASTM website
(www.astm.org).
--`,````,``,,,,,,````,``,`````,-`-`,,`,,`,`,,`---

Copyright ASTM International 6

Provided by IHS under license with ASTM Licensee=Aramco HQ/9980755100
No reproduction or networking permitted without license from IHS Not for Resale, 10/08/2006 02:13:45 MDT