STP 91-A-1963

F ft O F E R T Y O F
The
university of
Michigan
Liabraries
1 01 7
A ft T E S 8C1ENTIA VB RIT AS
A GUIDE FOR
FATIGUE TESTING AND THE
STATISTICAL ANALYSIS OF
FATIGUE DATA
Prepar'ed by
COMMITTEE £-9 ON FATIGUE
AMERICAN SOCIETY FOR TESTING AND MATERIALS
1 963
Reg. U. S. Pat. Off.
ASTM Special Technical No. 91-A* (Second Edition}
Price: $5.00; to Members: $4.00
Published by the
AMERICAN SOCIETY FOR TESTING AND MATERIALS
1916 Race Street, Philadelphia 3, Pa.
©BY AMERICAN SOCIETY FOR TESTING AND MATERIALS 1963
Library of Congress Catalog Card Number: 63-16331
Printed in Baltimore, Md.

February, 1964
F O REWO RD
The First Edition of this Guide was the composite work of many
people who contributed a great deal of time to the discussion and writ-
ing of the text under the guidance of Task Group Leader, F. B. Stulen.
A major portion of the statistical section was written by Miss Mary N.
Torrey. George R. Gohn not only contributed to the discussion and plan-
ning, but also edited and arranged for the printing of the advance copies
of the text. The coordination of contributions and discussions was done by
H. N. Cummings. Appreciable contributions to the statistical parts of the
Guide were also made by D. H. Shaffer. In addition to the above, R. E.
Peterson, H. F. Dodge, D. P. Gaver, R. Hooke, W. T. Lankford, R. B.
Murphy, W. C. Schulte, P. R. Toolin, and M. B. Wilk contributed to the
discussions at various conferences.
The original Task Group was organized under the leadership of J. T.
Ransom, and a first rough draft was prepared in 1954 and revised in 1955.
Other contributors to these drafts were E. W. Ellis, W. T. Lankford, F. A.
McClintock, R. E. Peterson, E. H. Schuette, F. B. Stulen, and E. J. Ward.
In 1956, F. B. Stulen became Leader of the Task Group and the Guide was
completed under his direction.
Upon the formation of Subcommittee VI on the Statistical Aspects of
Fatigue, this subcommittee was asked to review the First Edition and to
make any revisions necessary to bring the Guide up to date. As a result
of this study, extensive revisions have been made in various sections as
printed in this Second Edition. They include: (1) revisions in the definitions
(Section II) and their separate publication as ASTM Tentative Definitions
E 206,1 (2) an expansion of Section IV on the number of test specimens,
(3) changes in Section V on tests of significance, and (4) the preparation
of a new section, Appendix IV, on the use of the Weibull distribution func-
tion for fatigue Me.
This work was carried out by four Task Groups headed by S. M. Marco,
H. E. Frankel, Miss M. N. Torrey, and C. A. Moyer, respectively. Others
who assisted in the preparation of the Second Edition were W. N. Findley,
R. A. Heller, J. H. K. Kao, H. N. Cummings, W. S. Hyler, B. Ruley, and
G. R. Gohn, Chairman of Subcommittee VI.
1
Definitions of Terms Relating to Fatigue g and the Statistical Analysis of Fatigue Data
(E 206), 1962 Supplement to Book of ASTM Standards, Part 3.
iii
NOTE.—The Society is not responsible, as a body, for the statements
and opinions advanced in this publication.
CONTENTS
PAGE
I. Purposes of Fatigue Testing 1
II. Definitions, Symbols, and Abbreviations 2
III. Test Procedures 8
IV. Minimum Number of Test Specimens and Their Selection 16
V. Analysis of Fatigue Data 22
Appendices
Miscellaneous Reference Tables 55
Additional Technique for Distribution Shape Not Assumed 68
Analysis of Correlation Between Two Variables 69
The Weibull Distribution Function for Fatigue Life 71
References 78
Index 81
LIS T O F T A B L E S
TABLE PAGE
1.—Allocation of Test Specimens for "Probit" Method of Test 11
2.—Minimum Number of Specimens Needed for Determining 95 Per Cent Con-
fidence Intervals of Stated Width for a Population Mean, p 19
3.—Minimum Number of Specimens Needed for Determining 95 Per Cent Con-
fidence Intervals of Stated Width for a Population Standard Deviation, a 19
4.—Minimum Number of Specimens Needed to Detect if the Standard "Deviation
of a Population Is a Stated Percentage of a Fixed Value 20
5.—Minimum Number of Specimens Needed in Each Sample to Detect if a Stand-
ard Deviation of One Population Is a Stated Multiple of the Standard Devia-
tion of Another Population 20
6.—Minimum Number of Specimens Needed to Detect a Stated Difference Between
a Mean and a Fixed Value 21
7.—Minimum Number of Specimens Needed to Detect a Stated Difference Between
the Means of Two Populations 21
8.—Median Percentage of Survivors for the Population 24
9.—Confidence Intervals for the Median 26
10.—Approximate Confidence Intervals for the Mean 27
11.—Confidence Intervals for Percentages 28
12.—Fatigue Test Data 29
13.—Fatigue Test Data 31
14.—Percentages Surviving 108 Cycles 32
15.—-"Probit" Test Data 33
16.—Computations for Fitting a Response Curve by Method of Least Squares 35
17.—Computation of Standard Deviation, s 36
18.—Method of Computing 95 Per Cent Confidence Limits for Per Cent Survival
Values 37
19.—Method of Computing 95 Per Cent Confidence Limits for Fatigue Strength
Values 38
20.—Computations for Significance Tests 46
21.—R. R. Moore Rotating Beam; Step Tests of 42 Specimens 50
22.—Analysis of Data in Table 21 51
23.—Prot Test Computations 52
v
vi CONTENTS
TABLE PAGE
24.—Prot Test Computations 53
25.—Minimum Per Cent of Population Exceeding Median of Low Ranking Points.. 56
26.—Unpaired Rank Test2 58
27.—Percentiles of the x Distribution 60
28.—Areas of the "Normal" Curve 61
29.—Values of t 62
30.—Percentiles of the x2 /d-f- Distribution 63
31.—Mo.25 and uo.yis for Runs Among Elements in Samples of Sizes Ni and Nz 64
32.—F Distribution 65
33.—k Factors for S-N Curves (Normal Distribution Assumed) 67
34.—Working Significance Levels for Quadrant Sum 69
35.—Ordinate Locations Corresponding to Per Cent Failed Values 72
36.—Mean-Rank Estimates of the Per Cent Population Failed Corresponding to
Failure Order in Sample 74
37.—Typical Fatigue Test Data 75
38.—Typical Fatigue Test Data, Without Runouts 75
39.—Typical Fatigue Test Data, with Runouts 77
LIS T O F F I G U R E S
FIGURE
1.—Probability-Stress-Cycle (P-S-N) Curve for Phosphor-Bronze Strip 10
2.—Response or Survival Tests 11
3.—Illustration of Staircase Method 12
4.—Representation of "Step" Testing of Single Specimen 14
5.—Graphical Illustration of Prot Data 15
6.—"Normal" or Gaussian Distribution Curve 22
7.—Response Curves for a Particular Type of Steel 34
8.—Per Cent of Specimens Having at Least the Indicated Fatigue Strength at 107
Cycles 51
9.—Prot Test: Stress as Linear Function of Stress Cycles 53
10.—Log-Log Plot of Prot Data 54
11.—Scatter Diagram 70
12.—Typical Weibull Distribution Curves 72
13.—Construction of Weibull Probability Paper from Log-Log Paper 73
14.—Estimation of Weibull Distribution Function Parameters for Data in Table 38. 75
15.—Per Cent Failed at Weibull Mean 76
16.—Estimation of Weibull Distribution Function Parameters for Data in Table 39.. 77
RELATED ASTM PUBLICATIONS
Abstracts of Articles on Fatigue (STP 9)

Fatigue Manual (STP 91) (1949)
Statistical Aspects of Fatigue (STP 121) )
Fatigue, with Emphasis on Statistical h (STP 137) (1952)
Papers on Metals (STP 196) (1956)
Fatigue of Aircraft Structures (STP 203) (1956)
Large Fatigue Testing Machines and Their Results (STP 216) (1957)
Basic Mechanisms of Fatigue (STP 237) (1958)
Acoustical Fatigue (STP 284) (1960)
STP91 A-EB/Feb. 1 964
GUIDE FOR FATIGUE TESTING AND

STATISTICAL ANALYSIS
About 15 years ago, ASTM Committee E-9 on Fatigue prepared a Manual
on Fatigue Testing. 1 That Manual attempted to standardize the symbols
and nomenclature used in fatigue testing, described the principal types of
testing machines then in use, presented detailed instructions for the prepara-
tion of test specimens, outlined test procedures and techniques, and gave
some suggestions for the presentation and interpretation of fatigue data.
Since the Manual was first prepared, a number of new techniques have been
developed for evaluating the fatigue properties of materials. Furthermore,
the application of statistical methods to the analysis of the test results of
samples offers a means for estimating the characteristics of the population
from which the samples were taken. To take cognizance of these develop-
ments, this guide has been prepared.
I. PURPOSES OF FATIGUE TESTING
The purposes of fatigue testing are (1) to estimate the relationship be-
tween stress- (load-, strain-, deflection-) amplitude and cycle life-to-failure
for a given material or component, and (2) to compare the fatigue properties
of two or more materials or components. In order to specify the reliability
of these estimates, they must be based on the results of testing a sample of
fatigue specimens which have been drawn at random from a population of
possible fatigue specimens and tested in accordance with acceptable testing
procedures. The principal acceptable procedures discussed in this guide are:
A. "Standard" tests (constant amplitude or classical Wohler method). 2
1. Single test specimen at each stress level.
2. A group of test specimens at each stress level.
B. Response tests (constant amplitude).
1. "Probit" method.
2. Staircase method.
3. Modified staircase method.
C. Increasing amplitude tests.
1. Step method.
2. Prot method.
The primary purposes of the statistical analysis of fatigue data are: (1) to
estimate certain fatigue properties of material or a component (together
with measures of the reliability) from a given set of fatigue data, obtained
by testing a sample of fatigue specimens in accordance with one of the previ-
ous test procedures, and (2) to provide objective procedures for comparing
two or more sets of fatigue data to determine whether or not the data come
from similar populations. Statistical theory also provides information on (a)
1
2
Fatigue Manual, ASTM STP 91, Am. Soc. Testing Mats., 1949.
The term "standard" test, as used here, does not imply an ASTM standard.
1
Copyright© 1 964 by ASTM International www.astm.org

2 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA
the most efficient use of a limited number of test specimens and (b) the num-
ber of test specimens required to give a specified degree of confidence in the
test results.
Even with some basic training, it is difficult to locate the techniques par-
ticularly useful in fatigue testing in the statistical literature. The purpose of
this guide is to describe some statistical treatments that are suitable for the
analysis of fatigue data obtained in any one of the foregoing test methods
and to present these statistical treatments in a form useful to the test en-
gineer. Definitions of certain statistical terms are included, but only enough
of the basic concepts of statistics are included to make the methods under-
standable; theory is left to the references.
Test procedures are discussed hi Section III while techniques for analyz-
ing the data obtained in these tests are given in Section V and the Appen-
dices.
II. DEFINITIONS, SYMBOLS, AND ABBREVIATIONS
Relating to Fatigue Tests and Test Methods:
To encourage uniformity of terminology, the terms dealing primarily with
fatigue testing and test methods are also published in ASTM Definitions
E 6.3 The symbols used are, hi general, those recommended in the Ameri-
can Standard Letter Symbols for Mechanics of Solid Bodies. 4
1. Fatigue (Note 1).—The process of progressive localized permanent struc-
tural change occurring in a material subjected to conditions which produce fluctu-
ating stresses and strains at some point or points and which may culminate in
cracks or complete fracture after a sufficient number of fluctuations (Note 2).
NOTE 1.—The term fatigue in the materials testing field, has—in at least one case
glass technology—been used for static tests of considerable duration, a type of test gener-
ally designated as stress-rupture.
NOTE 2.—Fluctuations may occur both in stress and with time (frequency), as in the
case of "random vibration."
2. Fatigue Life, N. —The number of cycles of stress or strain of a specified char-
acter that a given specimen sustains before failure of a specified nature occurs.
Definitions 3 to 19, inclusive, apply to those cases where the con-
ditions imposed upon a specimen result or are assumed to result in
uniaxial principal stresses or strains which fluctuate in magnitude.
Multiaxial stress, sequential loading, and random loading require more
rigorous definitions which are, at present, beyond the scope of this
section.
3. Nominal Stress, S.—The stress at a t calculated on the net cross-section
by simple elastic theory, without taking into account the effect on the stress pro-
duced by geometric discontinuities such as holes, grooves,'fillets, etc.
4. Stress Cycle.-^-The smallest segment of the stress-time function which is re-
peated periodically.
3
Definitions of Terms Relating to Methods of Mechanical Testing, 1962 Supplement
to 1961
4
Book of ASTM Standards (E 6), Part 3.
ASA No. Z10. 3—1948, Am. Standards Assn., 1948.
DEFINITIONS, SYMBOLS, AND ABBREVIATIONS 3
5/ Maximum Stress, Smax- —The stress having the highest algebraic value in the
stress cycle, tensile stress being considered positive and compressive stress negative.
In this definition, as well as in others that follow, the nominal stress is used most
commonly.
6./ Minimum Stress, Sm , n. —The stress having the lowest algebraic value in the
cycle, tensile stress being considered positive and compressive stress negative.
?. Mean Stress (or Steady Component of Stress), Sm. —The algebraic average
of the maximum and minimum stresses in one cycle, that is,
8. Range of Stress, Sr. —The algebraic difference between the maximum and
minimum stresses in one cycle, that is
9. Stress Amplitude (or Variable Component of Stress), Sa- —One half the
range of stress, that is
10. Stress Ratio, A or R. —The algebraic ratio of two specified stress values in a
stress cycle. Two commonly used stress ratios are:
The ratio of the stress amplitude to the mean stress, that is,
and the ratio of the minimum stress to the maximum stress, that is,
11. S-N Diagram. —A plot of stress against the number of cycles to failure. The
stress can be Smax, Smin, or Sa- The diagram indicates the S-N relationship for a
specified value of Sm, A, or R and a specified probability of survival. For N a log
scale is almost always used. For S a linear scale is used most often, but a log scale
is sometimes used.
12. Stress Cycles Endured, N. —The number of cycles of a specified character
(that produce fluctuating stress and strain) which a specimen has endured at any
time in its stress history.
13. Fatigue Strength at TV Cycles, SN. —A hypothetical value of stress for
failure at exactly N cycles as determined from an S-N diagram. The value of
SN thus determined is subject to the same conditions as those which apply to the
S-N diagram.
NOTE. —The value of SN which is commonly found in the literature is the hypothetical
value of Smai , -Smt^ror Sa , at which 50 per cent of the specimens of a given sample could
survive N stress cycles in which Sm = 0. This is also known as the median fatigue strength
at N cycles (see definition 47).
14. Fatigue Limit, S/. —The limiting value of the median fatigue strength as N
becomes very large.
NpTE.->-Certain materials and" environments preclude the attainment of a fatigue limit.
Values tabulated as fatigue limitsun the literature are frequently (but not always) values
6f SN for 50~per cent survival at N cycles of stress in which Sm = 0.
15. Cycle Ratio, C—The ratio of the number of stress cycles, n, of a specified
character to the hypothetical fatigue life, N, obtained from the S-N diagram, for
stress cycles of the same character, that is,
Theoretical Stress Concentration Factor (or Stress Concentration Factor),

Kt- —The ratio of the greatest stress in the region of a notch or other stress concen-
trator, as determined by the theory of elasticity (or by experimental procedures
that give equivalent values), to the corresponding nominal stress.
NOTE. —The theory ofplasticity should not be used to determine Kt .
17. Fatigue Notch Factor, K/. —The ratio of the fatigue strength of a specimen
with no stress concentration to the fatigue strength at the same number of cycles
with stress concentration for the same conditions.
NOTE. —In specifying Kj it is necessary to specify the geometry and the values of
Smax , Sm , and N for which it is computed.
18. Fatigue Notch Sensitivity, g. —A measure of the degree of agreement be-
tween Kf and Kt for a particular specimen of a given size and material containing
a stress concentrator of a given size and shape.
NOTE. —A common definition of fatigue notch sensitivity is q = (Kf — i)/(Kt — 1\ in
which q may vary between zero (where Kf = 1) and one (where Kf = t).K
19. Constant Life Fatigue Diagram. —A plot (usually on rectangular coordi-
nates) of a family of curves, each of which is for a single fatigue life, N, relating
Sa, Smax and/or Sm in to the mean stress Sm. The constant life fatigue diagram is
generally derived from a family of S-N curves, each of which represents a different
stress ratio, A or R, for a 50 per cent probability ofsurvival.
Relating to Statistical Analysis:
20. Population (or Universe). —The hypothetical collection of all possible test
specimens that could be prepared in the specified way from the material under
consideration.
21. Sample. —The specimens selected from the population for test purposes.
NOTE. —The method of selecting the sample determines the population about which
statistical inference or generalization can be made.
22. Group. —The specimens tested at one time, or consecutively, at one stress
level. A group may comprise one or more specimens.
23. Frequency Distribution. —The way n which the frequencies of occurrence of
members of a population or sample are distributed according to the values of the
variable under consideration.
24. Parameter. —A constant (usually unknown) denning some property of the
frequency distribution of a population, such as a population median or a population
standard deviation.
25. Statistic. —A summary value calculated from the observed values in a
sample.
26. Estimation. —A procedure for making a statistical inference about the
numerical values of one or more unknown population parameters from the ob-
served values in a sample.
27. Estimate.—The particular value, or values, of a parameter computed by an
estimation procedure for a given sample.
28. Point Estimate.—The estimate of a parameter given by a single statistic.
29. Sample Median.—-The middle value when all observed values in a sample
are arranged in order of magnitude if an odd number of specimens are tested. If
the sample size is even, it is the average of the two middlemost values. It is a point
estimate of the population median, or 50 per cent point.
30. Sample Average (Arithmetic Mean).—The sum of all the observed values in
a sample divided by the sample size. It is a point estimate of the population mean.
31. Sample Variance, s2.—The sum of the squares of the differences between
each observed value and the sample average divided by the sample size minus one.
It is a point estimate of the population variance.
NOTE.—This value of s* provides both an unbiased point estimate of the population
variance and a statistic that is used in computing interval estimates and several test sta-
tistics (see definitions 34 and 42). Some texts define s2 as "the sum of the squares of the
differences between each observed value and the sample average divided by the sample
size," but this statistic is not as useful.
— 32. Sample Standard Deviation, s.—The square root of the sample variance.
It is a point estimate of the population standard deviation, a measure of the
"spread" of the frequency distribution of a population.
NOTE.—This value of 5 provides a statistic that is used in computing interval estimates
and several test statistics (see definitions 34 and 42). For small sample sizes, j underesti-
mates the5 population standard deviation. (See the ASTM Manual on Quality Control of
Materials or texts on statistics for an unbiased estimate of the standard deviation of a
Normal population.)
33. Sample Percentage.—The percentage of observed values between two stated
values of the variable under consideration. It is a point estimate of the percentage
of the population between the same two stated values. (One stated value may be
— °° or + °°.)
34. Interval Estimate.—The estimate of a parameter given by two statistics,
denning the end points of an interval.
35. Confidence Interval.—An interval estimate of a population parameter
computed so that the statement, "the population parameter lies in this interval,"
will be true, on the average, in a stated proportion of the times such statements
are made.
36. Confidence Limits.—The two statistics that define a confidence interval.
37. Confidence Level (or Coefficient).—The stated proportion of the times the
confidence interval is expected to include the population parameter.
38. Tolerance Interval.—An interval computed so that it will include at least a
stated percentage of the population with a stated probability.
39. Tolerance Limits.—The two s that define a tolerance interval. (One
value may be — « or + °°.)
40. Tolerance Level.—The stated probability that the tolerance interval in-
cludes at least the stated percentage of the population. It is not the same as a con-
fidence level but the term confidence level is frequently associated with tolerance
intervals.
41. Significant.—Statistically significant. An effect or difference between popu-
6
ASTM STP 15-C, Am. Soc. Testing Mats., 1957.
6 FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA
lations is said to be present if the value of a test-statistic is significant, that is, lies
outside of predetermined limits.
NOTE.—An effect which is statistically significant may or may not have engineering
significance.
42. Test-Statistic.—A function of the observed values in a sample that is used
in a test of significance.
43. Test of Significance.—A test which, by use of a test-statistic, pur-
ports to provide a test of the hypothesis that the effect is absent.
NOTE.—The rejection of the hypothesis indicates that the effect is present.
44. Significance Level.—The stated probability (risk) that a given test of sig-
nificance will reject the hypothesis that a specified effect is absent when the hy-
pothesis is true.
Relating to Statistical Analysis of Fatigue Data:
45. Median Fatigue Life.—The middlemost of the observed fatigue life values,
arranged in order of magnitude, of the individual specimens in a group tested under
identical conditions. In the case where an even number of specimens are tested it
is the average of the two middlemost values.
NOTE 1.—The use of the sample median, instead of the arithmetic mean (that is, the
average), is usually preferred.
NOTE 2.—In the literature, the abbreviated term "fatigue life" usually has meant the
median fatigue life df the group. However, when applied to a collection of data without
further qualification the term "fatigue life" is ambiguous.
46. Fatigue Life for p Per Cent Survival.—An estimate of the fatigue life that p
per cent of the population would attain or exceed at a given stress level. The ob-
served value of the median fatigue life estimates the fatigue life for 50 per cent
survival. Fatigue life for p per cent survival values, where p is any number, such as
95, 90, etc., may also be estimated from the individual fatigue life values.
47. Median Fatigue Strength at N Cycles.—An estimate of the stress level at
which 50 per cent of the population would survive N cycles.
NOTE 1.—The estimate of the median fatigue strength is derived from a particular
point of the fatigue life distribution, since there is no test procedure by which a frequency
distribution of fatigue strengths at N cycles can be directly observed.
NOTE 2.—This is a special case of the more general definition 48.
48. Fatigue Strength for p Per Cent Survival at N Cycles.—An estimate of the
stress level at which p per cent of the population would survive N cycles; p may
be any number, such as 95, 90, etc.
NOTE.—The estimates of the fatigue strengths for p per cent survival values are derived
from particular points of the fatigue life distribution, since there is no test procedure by
which a frequency distribution of fatigue s at N cycles can be directly observed.
49. Fatigue Limit for^ Per Cent Survival.—The limiting value of fatigue
strength for p per cent survival as N becomes very large; p may be any number,
such as 95, 90, etc. (See Note, definition 14.)
50. S-N Curve for 50 Per Cent Survival.—A curve fitted to the median values of
fatigue life at each of several stress levels. It is an estimate of the relationship be-
tween applied stress and the number of cycles-to-failure that 50 per cent of the
population would survive.
NOTE 1.—This is a special case of the more general definition 51.

NOTE 2.—In the literature, the abbreviated term "S-N Curve" usually has meant
either the S-N curve drawn through the means (averages) or the medians (50 per cent
values) for the fatigue life values. Since the term "S-N Curve" is ambiguous, it should be
used in technical papers only when adequately described.
51. S-N Curve for p Per Cent Survival.—A curve fitted to the fatigue life for p
per cent survival values at each of several stress levels. It is an estimate of the rela-
tionship between applied stress and the number of cycles-to-failure that p per cent
of the population would survive; p may be any number, such as 95, 90, etc.
NOTE.—Caution should be used in drawing conclusions from extrapolated portions of
the S-N curves. In general, the S-N curves should not be extrapolated beyond observed
life values.
52. Response Curve for N Cycles.—A curve fitted to observed values of per-
centage survival at N cycles for several stress levels, where N is a preassigned num-
ber such as 106, 107, etc. It is an estimate of the relationship between applied stress
and the percentage of the population that would survive N cycles.
NOTE 1.—Values of the median fatigue strength at N cycles and the fatigue strength
for p per cent survival at N cycles may be derived from the response curve for N cycles,
if p falls within the range of the per cent survival values actually observed.
NOTE 2.—Caution should be used in drawing conclusions from extrapolated portions
of the response curves. In general, the curves should not be extrapolated to other values
of p.
LIST OF SYMBOLS AND ABBREVIATIONS

The following terms are frequently used in lieu of or along with the terms
covered by the preceding definitions. In general the symbols are those
recommended in the American Standard Letter Symbols for Mechanics of
Solid Bodies (see footnote 4). For stress, the use of S with appropriate
lower case subscripts is preferred for general purposes; for mathematical
analysis the use of Greek symbols is generally preferred.
Symbol Term
A Area of cross-section, Stress ratio
C Cycle ratio
c Distance from centroid to outermost fiber
D or d Diameter
E Modulus of elasticity in tension or compression
e (epsilon) Strain
f t-lb Unit of work
G Modulus of elasticity in shear
7 (gamma) Shear strain
/ Moment ofinertia
i Subscript denoting ith term
in.-lb Unit of work
/ Polar moment ofinertia
ksi Thousands of pounds per square inch
or kips per square inch
Kf Fatigue notch factor
Symbol Term
Kt Theoretical stress concentration factor
Ib-ft Unit of torque
Ib-in. Unit of torque
M Bending moment
n (mu) pre- Poisson's ratio
ferred, or v
(nu) (gener-
ally used in
applied me-
chanics)
N Fatigue life; number of cycles
n JNumber of stress cycles endured; Sample size
P Load
psi Pounds per square inch
P Probability of failure; Per cent failure;
Probability of survival; Per cent survival
9 Fatigue notch sensitivity
R Stress ratio
s Sample standard deviation
s2 Sample variance
S Nominal stress
S or <r (sigma) Normal stress
Sa Stress amplitude
Sc Compressive stress
Sff Fatigue limit
»Jcj/ Compressive yield strength
»SJ>» Mean stress
S^ max Maximum stress
C .
LJrmn Minimum stress
SN Fatigue strength at N cycles
Sr Range of stress
Ss or T (tau) Shear stress
St Tensile stress
Stv Tensile yield strength
Su Tensile strength
(7 (sigma) Standard deviation; Stress
.A
(7 Estimate of standard deviation
2
(T Variance
<rx Standard deviation of x
<rx2 Variance of x
T Torque; Temperature
t Time
T (tau) or Ss Shear stress
III. TEST PROCEDURES
Until recently, there was only one accepted method of conducting labora-
tory fatigue tests on a material or component. This "standard" test, using
single specimens at several stress levels, is described in Section V of the
x TEST PROCEDURES 9
Manual on Fatigue Testing (STP91). Experience showed, however, that
this test method did not give adequate information for many of the pur-
poses for which fatigue data are needed. Therefore, within the last ten years,
a number of new methods for performing more meaningful fatigue tests
have been introduced, each method having certain advantages.
The choice of test method depends upon the objective of the test and the
number of available test specimens. When the objective is to determine an
S-N curve, the "standard" tests (Sections I Al and A2) are generally the
most suitable. To determine the long-life fatigue strength or the fatigue
limit, response tests (Sections I Bl, B2, and B3) or increasing amplitude
tests (Sections I Cl and C2) are recommended. The latter methods also are
used for comparing the long-life fatigue properties of different materials or
different methods of processing. All seven of these experimental fatigue
testing techniques are described in the following paragraphs. For analysis of
the data, see Section V.
A. "STANDARD" TESTS (CONSTANT AMPLITUDE)
1. Single Test Specimen at Each Stress Level:
In the "standard" test method described in STP 91, each fatigue speci-
men is cycled at a different constant stress (or strain) amplitude until frac-
ture occurs. The stress levels are usually selected to cover a series of stresses
ranging from high values, at which failure will occur within a limited number
of cycles, to low values at which no failure will occur (runouts) or at which
failure will occur only after an extremely large number of cycles. If the pri-
mary interest is in the longlife end of the S-N relationship (often called the
fatigue limit), the investigator usually has some preconception of this value
for the material or component to be tested. In this case, the first stress level
is selected somewhat above the estimated fatigue limit. Depending upon the
results of the first test, succeeding specimens are then tested at stress levels
either above or below this value, until a stress level is reached at which the
specimen does not fail within the prescribed number of cycles. Near the
fatigue limit, some specimens must be run at stress levels high enough to
produce failures in order to have data from which the fatigue limit may be
estimated.
This method of test is used when the investigator has available only a
relatively small number of specimens for test. Such is generally the situation
when (1) the fatigue specimens are expensive, (2) the supply of material is
limited, or (3) machine parts, full size sections, or assemblies are being tested.
2. Group of Specimens Tested at Each Level:
Since the "standard" test, using only one specimen at each stress level,
gives very little information concerning the variability of the material or
component and test procedure, it is more satisfactory to test several speci-
mens at each of a number of different stress levels. In this procedure, each
group should consist of at least four specimens in order to estimate the vari-
ability of the data. Ten or more specimens are preferable to obtain some
indication as to the shape of the distribution of fatigue life values. Three or
more different stress levels must be investigated for the determination of the
S-N curves for p per cent survival (see Fig. I). 6 Generally, at least four or
five stress levels are used in a test of this nature. To determine the fatigue
limit of the material, a number of groups also should be tested at stress
levels in the vicinity of the fatigue limit. This would increase the total num-
ber required to at least seven groups. Furthermore, to obtain approximately
an equal degree of precision throughout the range of the S-N curve, more
specimens should be tested in the long-life than in the short-life range.
FIG. 1.—Probability-Stress-Cycle (P-S-N) Curve for Phosphor-Bronze Strip.

B. RESPONSE TESTS (CONSTANT AMPLITUDE)
1. "Probit" Method:
In the "Probit" method, one or more groups of specimens are tested for a
fixed number of cycles at four or five different stress levels distributed about
the stress of interest. This test has been used primarily for estimating the-
fatigue limit of a material, that is, the stress at which 50 per cent of the test
specimens will fail prior to, and 50 per cent will survive, the preassigned
cycle life, N. The test is not limited to this application; it is just as valuable
for estimating the fatigue strength or the fatigue limit at any other per-
centages of survival, provided that the specimens are properly allocated to
the various stress levels. When used t estimate the fatigue limit at 50 per
cent survival, at least two stress levels should be selected so that the per-
centage of specimens surviving N cycles will be less than 50 and two more
stress levels selected at which the percentage of survivors will be more than
50. A fifth stress level producing approximately 50 per cent survivors is
desirable but not essential.
6
Such curves are sometimes referred to as probability-stress-cycle (P-S-N) curves.
TEST PROCEDURES 11
TABLE 1.—ALLOCATION OF TEST SPECIMENS FOR "PROBIT"
METHOD OF TEST.
Expected Per Cent Survival Relative Group Size"
25 to 75 1
15 to 20, 80 to 85 1.5
1 0,90 2
5,95 3
2, 98 5
a
The group size is the number of specimens included in a test at one stress level.
Thus, whatever group size is chosen for testing at stress levels for which the expected
per cent survival is between 25 and 75, the sizes of other groups must be increased by the
factor in the second column to obtain the number of test specimens required for testing
at stress levels for which the per cent survival is expected to be larger, or smaller, if
similar precision is to be obtained in the test results. If the stress levels are chosen suc-
cessively, starting with levels requiring the smallest group size, the group size required
for the other levels will be determined more easily. Previous data for the same material
or similar materials should be used as a guide for choosing the stress levels, whenever they
are available; otherwise a preliminary test such as that described under Sections IIIA1 or
B2 may be required. A properly designed "Probit" test will give more useful fatigue
data than any of the other response or increasing amplitude tests.
FIG. 2.—Response or Survival Tests.
In "Probit" tests a group should consist of not less than five specimens
and the total tested at all stress levels should be at least 50. The distribu-
tion of the total number of available t specimens will depend upon the
purpose of the test. The relative group sizes for different stress levels are
shown in Table 1. This allocation is suggested so that the observed percent-
age survival values will have approximately equal weight, a condition neces-
sary for fitting the response curve by the usual method of least squares. This
allocation also facilitates the computation of confidence limits on the re-
sponse curves. As an alternative to the use of the relative group sizes (Ta-
ble 1), groups of other sizes can be used at each stress level, provided that
weighting factors7 are employed and the analysis conducted as indicated in
references (1-3) .
Figure 2 presents data that might be obtained in a "Probit" test of the
type described if the preassigned number of cycles were 107. Although not
required for the "Probit" analysis, the actual number 7of cycles-to-failure
should be recorded for each specimen that fails before 10 cycles, so that the
data may be available for other types of analysis, such as the plotting of
P-S-N Curves.
2. The Staircase Method:
The staircase (or "up-and-down") method of testing is.a variation of the
"Probit" method. It may require fewer specimens than the latter but is
likely to be useful only when the primary interest is in the mean fatigue
strength corresponding to a preassigned cycle life, N* The advantage gained
FIG. 3.—Illustration of Staircase Method.

NOTE—Specimens numbered in chronological order. Number of cycles for each test is
constant unless failure occurs beforehand.
in reducing the number of specimens tested may be offset by an increase in
the time required to conduct the test.
In the staircase method the specimens are tested sequentially, one at a
time. The first specimen is tested at a stress level equal to the estimated
value of mean fatigue strength for the prescribed number of cycles or until
it fails, if it fails before that number of cycles. If the specimen fails, the next
specimen is tested at a stress level that is one increment below the first
stress level. If the first specimen does not fail, the second specimen is tested
at a stress level that is one increment above the first stress level, and so forth.
The data are recorded as shown in . 3. The specimens that did not fail
are designated by the o's and those that failed as #'s. The chart shows at a
glance the stress level that should be used for the next test.
The selection of the proper increment of stress level is very important.
7 The boldface numbers in parentheses refer to the list of references appended to this
guide.
8 The staircase method as described in the literature estimates the mean fatigue strength,
not the median fatigue strength which is used elsewhere in this guide.
TEST PROCEDURES 13
Ideally, most of the tests should be made at three stress levels, so chosen
that about 50 per cent of the test specimens survive at the middle stress
level, about 70 per cent survive at the lower stress level, and about 30 per
cent survive at the higher stress level. Previous data for the same or similar
materials are needed in order to choose the stress levels efficiently. If none
are available, some preliminary testing may be required. Such data are dis-
carded up to the first pair of data giving opposite results; for example, in
Fig. 3, data for tests 1, 2, and 3 should be discarded.
Since the testing is concentrated at stress levels near the mean fatigue
strength value, the number of specimens tested may be less than for the
"Probit" method, which gives results for a wider range of stress values. In
general, at least 30 specimens should be tested because, at most, only half
of the test results are actually used in the computation of the mean fatigue
strength. If data obtained by the staircase method are analyzed by response
curve methods, the results may be statistically biased because of the se-
quential nature of the staircase method. Further, if the main interest lies
in estimating the response curve—rather than the mean strength—at N
cycles, the staircase method is not an efficient experimental procedure.
3. Modified Staircase Method:
The time required to complete a test by the staircase method can be re-
duced by dividing the one long staircase program into several shorter, inde-
pendent staircases and conducting these several tests simultaneously. This
treatment is known as the modified staircase method. In the modified pro-
cedure, the total number of specimens, T, is divided into r groups of n each,
so that rn = T. Each group is tested as a separate staircase program, with a
separate chart for each group. Thus several machines may be used simul-
taneously. In the modified staircase method, as in any other test in which
specimens in a group are tested on more than one machine, a check should
be made to determine whether the machines give significantly different re-
sults. If the results are not significantly different, the data may be combined
for statistical analysis.
C. INCREASING AMPLITUDE TESTS
1. Step Method:
In many cases the "Probit" or staircase methods of test require more
specimens than are available. When only a few parts are available for de-
termining the fatigue limit, a natural desire is to test each part until it ac-
tually fails instead of just counting the number of runouts. When testing a
limited number of specimens sometime the practice is to run7 each specimen
at several stress levels for a large number of cycles, say 10 . If typical re-
sponse curves for the material are available, the test may be started at a
stress level corresponding to a percentage survival of approximately 90 per
cent. For each successfully completed run, the applied stress level is in-
creased by an amount corresponding to a decrease in the probability of sur-
vival of about 5 per cent and the test is repeated until failure of the speci-
men occurs. When response curves are not available, the test may be started
at a stress level equal to about 70 per cent of the estimated fatigue limit, and
the stress increments should be approximately 5 per cent of the estimated
fatigue limit.
In the past, this method has not been considered an acceptable technique
because the fatigue strengths of some materials will be increased or "coaxed"
by stressing them at stress levels below their fatigue limits, whereas the
fatigue strengths of other materials may be decreased by damage due to
"under-stressing." However, in recent years it has been observed that under-
stressing does not greatly affect the true fatigue limit of some alloys, such
as many of the alloy steels and a few of the nonferrous materials (4). For
those materials with which neither appreciable coaxing nor damage occurs,
it is possible to estimate the fatigue strength of each specimen or part by
stressing it at consecutively higher levels until the specimen fails.
FIG. 4.—Representation of "Step" Testing of Single Specimen.

This method is illustrated graphically in Fig. 4. In this manner, the fatigue
strength corresponding to a preassigned value of N for each specimen or
part may be estimated. The main disadvantage of the procedure is that the-
specimens are run initially at a sufficiently low stress level so that failure
will not occur. As a result, a number of stress levels of runouts are usually
necessary before failure of the specimen occurs.
Before the step technique of fatigue testing can be safely used, the effect
of coaxing or under-stressing the material must be known. Certain steels,
sensitive to strain-aging, will have r fatigue limits artificially raised by
coaxing or under-stressing at low stress levels. In other cases it is thought
that coaxing or under-stressing may damage the material artificially and
cause premature failures.
Although step tests have been made with a single specimen, four or more
are needed to estimate the median fatigue strength. A larger sample gives
greater precision in the estimates of the median and the variability of the
fatigue strength.
TEST PROCEDURES 15
2. The Prot Method (5-8):
In 1945 Marcel Prot, in France, devised a rapid method for estimating
the fatigue limit of a material. By using the Prot method, a good estimate
of the fatigue limit may be obtained in a fraction of the time required by
other methods but at the expense of more uncertainty than is present in
most of the other test methods. The use of this technique is restricted not
only to those materials which are not sensitive to coaxing effects, as dis-
cussed in Section III Cl, but also to materials that apparently have a fatigue
limit. In contrast to the step method, it is suggested that at least 20
test specimens be used to obtain the data needed for the Prot analysis be-
cause of the wide scatter in fracture stress usually found in Prot fatigue
data. To date, it has been found that, by the use of the Prot procedure, the
fatigue limits of many alloy steels may be obtained within a few per cent
FIG. 5.—Graphical Illustration of Prot Data.

«!, «2 , «s, indicate different loading rates in psi per cycle.
of the estimate found from constant amplitude methods. It is not certain,

however, that the long-life fatigue strengths of nonferrous alloys can always
be evaluated by this method.
In the Prot method, the test on a specimen is first started at an alternating
stress of about 60 to 70 per cent of its estimated fatigue limit and the stress
is raised at a constant rate. A number of specimens is tested at the same rate
of loading until each specimen fails. At least three rates of loading are used
to establish and check the linear relationship between stress and the power
of the loading rate, which is required in the Prot analysis. The lowest rate
should be as small as practicable and e highest rate should be low enough
so that the specimen does not fail by yielding before fracture. The type of
data observed is shown in Fig. 5.
One of the simplest methods for obtaining a constant rate of loading in a
fatigue test is to use a stream of water flowing at a constant rate into the
loading container. Another way is to arrange for small weights, such as shot,
to be poured into a container at a constant rate. Fairly good results have
been obtained by adding small weights by hand, one at a time, at small and
equal cycle increments. Any device that increases the stress at practically a
continuous, constant rate can be used.
NOTE.—For constant rates of loading, all the points obtained at a given rate should
fall on the same straight line. Small variations in the rate of loading or variations in the
testing speed may cause scatter such as that shown in Fig. 5.
IV. MINIMUM NUMBER OF TEST SPECIMENS
AND THEIR SELECTION
The practicability of fatigue tests is based upon the assumption that estab-
lishing the fatigue characteristics of a given material by studying the per-
formance of a random sample selected from a larger body of possible speci-
mens (the population or universe) is possible. Implicit in these tests is the
assumption that the sample tested is "representative" of the population.
By random selection and allocation of the test specimens, using a table of
random numbers (9, pp. 366-370) the influence of all variability inherent in
the material and testing procedures is given a fair chance of being reflected
in the test data.
There are innumerable stages in the testing program in which any one
specimen or any one group of specimens may be affected differently from
others from the same population. For example, if one bar of a batch of bar
stock is tested, it is often tacitly assumed that the remaining bars are the
same as the one tested. Usually they are not because, for example, such
blanks are heat treated in batches. For each batch the furnace settings are
slightly different. Within each heat-treatment batch, those specimens near
the walls of the furnace are under slightly different conditions from those in
the center. Specimens prepared at the start of the day are machined with
sharper tools than those that succeed them. Specimens tested at the begin-
ning of a program may have the advantage of being tested on newer, more
perfect testing machines than those that are tested later when wear of the
machines has modified their characteristics. These are but a few examples of*
the many factors that may produce significant biases in the results unless
controlled by appropriate randomization.
The following are some of the factors for which randomization might be
considered:
Position of specimen within the whole batch of material
Heat-treatment batch
Position of specimen in heat-treating furnace
Order of quenching
Order of polishing
Assignment to testing s (stress level and so forth)
Order of testing
Assignment to testing machine
Machine operator
This list will suggest other variables of importance in particular programs.
Onp job of thp pnginppr and statistician is to decide how the randomization
sb_n^1H he rqm'eH rît A common misconception is that randomization can_
be accomplished by rough-and-ready procedures such as reaching blindly
NUMBER or SPECIMENS AND THEIR SELECTION 17
into the box of specimensjQr^theîextjto^b^tested, but it is summing how
chance procedures.
The sample can be "biased" by unconscious and unrecognized trends of
human behavior as well as by unknown patterns of arrangement. The best
procedureJs_tQ_ae±-iiD the program on the basis of random numbers as previ-
ously suggested (9").
To obtain fatigue data that can be used most efficiently, atrained^tatisti^
cian shoiild_b£_£OJiau]lej J_whenever possible, in jjlanning the experiments
and specimen selection. In most cases the statistician will be able to plan the
experiments to measure not only the effects of the main variables under
study but also the effects of the more important secondary variables as well,
and do this without requiring many, if any, additional specimens. Tests con-
ducted in accordance with such a plan can be analyzed to give an estimate
of the importance of each of the known variables that contribute to the
scatter in the test results. The techniques of experimental design are too in-
volveoL however, to be included in this Guide. '
"Some indication of the minimum number of specimens needed for a given
degree of confidence in the results obtained when using the different test
procedures has been given in Section III. For supplementary references on
this subject, see references (10-12). 9
The following sections discuss the minimum number of specimens needed
for each type of analysis given in Section V when the sample size is fixed
before testing. All samples are assumed to be randomly selected samples
from the population under consideration.
A. LITE DISTRIBUTION SHAPE NOT ASSUMED
1. S-N Curves:
The minimum number of fatigue test specimens needed at each stress
level depends on: (1) which per cent survival curve is desired and (2) what
confidence level is desired.
For a 50 per cent confidence level and one group tested at each stress level,
Table 8 in Section VA1 shows the number of specimens needed for several
values of per cent survival. For example, a 95 per cent survival curve re-
quires at least 13 specimens at each stress level.
Table 25 10 provides similar information for one or more groups tested at
each stress level and seven values of confidence level, including 50 per cent.
For example, from Table 25, one group of five specimens at each stress level
is needed for an 87 per cent survival curve corresponding to a 50 per cent
confidence level. For a 90 per cent confidence level, at least five groups of 10
specimens at each stress level are d for an approximately equivalent
S-N curve.
When several S-N curves are to be drawn from the same data, Table 25
should be studied carefully to find the best combination of number of groups
and group size.
9
See also Recommended Practice for Choice of Sample Size to Estimate the Average
Quality
10
of a Lot or Process (E 122), 1961 Book of ASTM Standards, Part 3.
For Tables 25 to 33, see Appendix I, pp. 55-67.
2. Estimates of Parameters, Single Stress Level:
The minimum number of specimens needed depends upon the desired
width of the confidence interval for each parameter. In general, as the sam-
ple size increases, the confidence interval for any given confidence level be-
comes narrower and the difference between the observed value and the uni-
verse value becomes smaller.
For the median at a confidence level of 0.95, confidence limits are equal
to the observed minimum and maximum values up to a sample size of nine,
when the width of the confidence interval becomes less than the observed
range. (See Table 9 on page 26.) If ranges for prior samples from the same
population are known, a sample size can be chosen so that the interval will
have approximately the desired width.
If the number of specimens are only 3, 4, or 5, Table 10 (see Section V A2)
gives procedures for computing confidence intervals for the mean.
For per cent survival values, Table 11 (Section V A2) gives values of 95
per cent confidence limits for four sample sizes. Comparing the widths of
the confidence intervals gives some idea of the size of sample needed. If a
good estimate of p = per cent survival/100 is available, the minimum
sample size is approximately:
where E = one half the desired width of a 95 per cent confidence interval
(see ASTM Recommended Practice E 122)9.
It is more difficult to determine the minimum number of specimens needed
for a confidence interval of a given width for fatigue life corresponding to a
stated value of per cent survival other than 50 per cent. See footnote 15,
page 29, for equations for setting up tables similar to Table 9 (in Section
V A2) for other percentage points.
In general, the sample sizes would be larger than for medians.
3. Tests of Significance:
The minimum number of specimens needed depends upon the desired
magnitude of the difference that should be detected and the size of the risks
that can be tolerated.
When the rank test is used to test the differences of group medians, it is
difficult to relate the desired values and the criteria for the significance test
given in Table 26 for two groups and in Table 27 for more than two groups.
At least five specimens should be included in each group.
For differences of two or more s (other than 50 per cent) no
precise estimate of the minimum number of specimens needed is possible
unless prior estimates of the percentages are available. At least 15 speci-
mens should be included in each group.
4. Response Curves:
A discussion of the minimum number of specimens and their allocation
to stress levels is given in Section III Bl.
B. LITE DISTRIBUTION SHAPE ASSUMED
1. Normal Distribution:
It is assumed here, as well as in Section V B that the fatigue data can be
transformed so that they will be approximately Normally distributed. A
Normal distribution is assumed in all cases. Each sample is assumed to be
drawn at random from its population.
2. S-N Curves:
Table 33 gives k factors for computing points on 75, 90, 95, 99, and 99.9
per cent survival curves for four values of confidence level, including 50 per
cent, and for n = 3 to 25. The minimum number of specimens should in-
crease as the per cent survival increases, but there is no definite criterion
for choosing a particular group size except for the relative magnitudes of
the k values. (Note that the rate of decrease is less as,« increases.) The num-
ber of specimens tested at each stress level can be smaller than the group
sizes needed when the life distribution is not assumed (Section V A).
TABLE 2.—MINIMUM NUMBER TABLE 3.—MINIMUM NUMBER
OF SPECIMENS 0 NEEDED FOR DE- OF SPECIMENS" NEEDED FOR DE-
TERMINING 95 PER CENT CONFI- TERMINING 95 PER CENT CONFI-
DENCE INTERVALS OF STATED DENCE INTERVALS OF STATED
WIDTH FOR A POPULATION WIDTH FOR A POPULATION STAND-
MEAN, /*. ARD DEVIATION, a.
Standard Deviation, a, Assumed Known. Some Estimate of <r Available.
Width of Confidence Number of
Specimens, n
Number of
Width of Interval
Interval Limits Specimens, »
0 2<r ... X± 0. 1 <7 384 0.14<7 385

0. 4 ± 0.2 96 0.2 190
06 =t 0.3 43 0.3 84
08 ± 0.4 24 0.4 47
10 ± 0.5 15 0.5 30
1.2 ± 0.6 11 0.6 21
1.4 ± 0.7 8 0.7 16
1.6 ±0 . 8 6 0.8 13
1.8 ± 0.9 5 0.9 10
2. 0 ± 1 .0 4 1 .0 8
Where:
._ n=
E= width of interval
2 0
Based on Fig. 1 of Greenwood and
x = sample mean. Sandomire (13).
"ASTM Designation E 122 (see foot- * The values of re given in
note 9). Table 30.
3. Confidence Intervals:
For the Mean. —If a good estimate of the population standard deviation,
0, is available, Table 2 gives the minimum number of specimens needed
for confidence intervals of stated width for the mean, p, of the population.
If the sample is used to estimate a as well as /*, the sample sizes should be
larger, since ^0.975 values from Table 29 should be used instead of the 1.96
in the equation for n (Table 2).
For the Standard Demotion. —In order to find the minimum number of
specimens needed for determining confidence intervals of stated width for
the standard deviation, <r, of a population, some estimate of a must be avail-
able, since the width of the interval is measured in units of a. However,
Table 3 can be used as a guide even if no good estimate of a is available.
For example, if n = 8, the sample-standard deviation, used to estimate the
population standard deviation, may be above or below a by 0.5 <r, whereas an
estimate based on n = 30 will not be expected to deviate from the true value
by more than 0.25<r.
TABLE 4.—MINIMUM NUMBER TABLE 5.—MINIMUM NUMBER
OF SPECIMENS" NEEDED TO DE- OF SPECIMENS 0 NEEDED IN EACH
TECT IF THE STANDARD DEVIA- SAMPLE TO DETECT IF A STAND-
TION OF A POPULATION IS A ARD DEVIATION OF ONE POPULA-
STATED PERCENTAGE OF A FIXED TION IS A STATED MULTIPLE OF
VALUE. THE STANDARD DEVIATION OF
ANOTHER POPULATION.
Number of Specimens, n
Percentage of Fixed Number of Specimens, «
Value 80% 90%
Chance Chance
of of Multiple 80% 90%
Detection Detection
Chance Chance
of of
40 7 8 Detection Detection
45 8 9
50 9 12 1 .5. 39 52
55 12 14 2.0 15 20
60 15 19 2.5 9 13
65 20 26 3.0 7 9
70 28 38 3.5 6 8
75 42 55 4.0 5 7
0
a
Based on Fig. 2 from Ferris et al (14). Based on Fig. 3 from Ferris et al »(14).
Difference Between Two Standard Deviations. —The sample sizes for testing
the difference between two means are given in Tables 6 and 7. In some cases,
the principal interest is in the difference between standard deviations.
1. One Standard Deviation a Fixed Value. —If one standard deviation is a
fixed value—for example, the long-time standard deviation of data based
upon an old procedure—and if the other standard deviation is to be com-
puted from data based upon a new procedure that may reduce the varia-
bility, Table 4 gives the minimum number of specimens needed to detect a
reduction of a stated amount. These e sizes apply when the observed
standard deviation, s, for the new procedure is indeed smaller than the fixed
value, and the ratio s2/(fixed value) 2 is compared with 1/F 0.95, corresponding
to °o and n — 1 degrees of freedom for numerator and denominator re-
spectively. (See Section V B4(a) and Table 32.)
2. Two Sample Standard Deviations. —If the problem is to test whether
the variability of procedure 1, say, is greater than the variability of pro-
cedure 2 (the numbers having been assigned prior to taking the data), Ta-
ble 5 gives the minimum number of specimens needed in each sample to
detect that si is a stated multiple of s%. If the observed value of si is indeed
larger than the observed value of sz, compare s?/s<? with ^0.95 corresponding
to (HI — 1) degrees of freedom for numerator and denominator (since
n\ = HZ). (See Section V B4(a) and Table 32.) In this case it is not correct
to make the test if s22 is greater than Si2.
Difference Between Two Means:
1. One Mean a Fixed Valise. —If one mean is a fixed value—for example,
the long-time mean of data based on an old procedure or a commonly used
material—and the other mean is to be computed from data based upon a new
TABLE 6.—MINIMUM NUMBER OF TABLE 7.—MINIMUM NUMBER OF
SPECIMENS 0 NEEDED TO DETECT SPECIMENS 0 NEEDED TO DETECT
A STATED DIFFERENCE BETWEEN A STATED DIFFERENCE BETWEEN
A MEAN AND A FIXED VALUE. THE MEANS OF TWO POPULATIONS.
a = Unknown Standard Deviation of the <7 = Unknown Standard Deviation of
Population Being Estimated. Each Population; <n = az •
Number of Specimens, » Number of Specimens, n
Difference Difference
80% 90% 80% Chance 90% Chance
Chance Chance of Detection of Detection
of of
Detection Detection
0 50<r 64 86
0 50<r 34 44 0 75 29 39
0 75 16 21 1 00 17 23
1 00 10 13 1 25 12 15
1 25 7 9 1 50 9 11
1 50 6 7 1 75 7 9
1 75 5 6 2 00 6 7
2 00 4 5 2 50 4 5
2 50 3 4 0
0
Taken from Table E.I of reference
Taken from Table E of reference (11). (11) •
procedure that may shift the mean, Table 6 gives the minimum number of
specimens needed to detect a shift in either direction, measured in terms
of the population standard deviation of the new procedure. These sample
sizes apply when the computed value of
is compared with £0.975 in Table 29. e Section V B4(6).) No F-ratio test

is needed.
2. Two Sample Means. —The minimum number of specimens needed in
each sample to detect a difference in two population means, stated as a
multiple of their equal universe standard deviations, is given in Table 7.
These sample sizes apply when (1) the two sample standard deviations are
not significantly different and (2) the computed value of / (see Section
V B4(6)) is compared with /0.9?5 in Table 29.
V. ANALYSIS OF FATIGUE DATA
A bask concept of statistics is that a group of one or more specimens is a
sample taken from a larger body or population. Such a sample is considered
to be just one of a "number," often very large, of samples that could have
been taken. The sampling procedure used delimits the-population being
estimated. The results obtained from tests on a random sample from the
population can be used to estimate the characteristics of the whole popula-
tion and to measure the precision of the estimates.
In the case of fatigue tests the data observed are usually the lives of
specimens tested at a constant applied stress (strain or deflection) amplitude.
Since the cycle life varies from specimen to specimen, this measurable char-
FIG. 6.—"Normal" or Gaussian Distribution Curve.

acteristic is not a fixed value and is best described by a frequency distribu-
tion. The graphical presentation of the distribution of cycle lives for the
population of specimens that have lives between certain limits is known as a
frequency distribution curve. Such a distribution curve may be estimated
from the raw test data or from transformed test data, that is, either from
values of N or from values of log N, log log N, N1 '2, and so forth.
When the frequency distribution curve has a particular kind of bell shape,
as shown in Fig. 6, the data are said o have a "Normal" or Gaussian dis-
tribution. This Normal probability distribution curve, f(x), is represented
by the equation:
The constants in the formula are the population mean, /*, and <r, the popula-
ANALYSIS or FATIGUE DATA 23
tion standard deviation (a measure of the dispersion). 11 It should be empha-
sized that values of the parameters of the population can only be estimated
from tests on the specimens in the sample; to obtain exact values would re-
quire that the total population be tested.
While some fatigue tests, particularly those made in the finite life range
of an S-N curve, may yield approximately Normal distributions of cycle
life, generally a transformation to log cycle life is required. Others do not
yield Normal distributions, even after various transformations are performed
on the data. This is particularly true in the case of tests made at applied
stresses near the fatigue limit where runouts are observed. Hence, other
distributions, such as the Weibull distribution, 12 the "extreme value" dis-
tribution with and without lower limits, as used by Freudenthal and Gumbel
(IS), and other distributions, that are just as normal in the usual sense, as the
Normal or Gaussian distribution, have been applied to the analysis of fatigue
data. While references to some ,of these distributions are included in this
Guide, analysis of the fatigue data has been confined mostly to methods that
require no assumptions of distribution shape or to the methods based upon
the assumption that the raw data or the transformed data have a Normal
distribution.
As stated previously, however, any set of observations to which these
statistical methods are applied is assumed to come from a random sample
from the population of interest. If a series of samples is drawn, procedures
for testing for statistical control are given in the ASTM Manual on Quality
Control of Materials (see footnote 5). Lack of statistical control in data in-
dicates that the series of samples does not come from the same population.
A. LIFE DISTRIBUTION SHAPE NOT ASSUMED
1. S-N Curves:
These techniques should be used when the actual shape of the distribution
of fatigue life values for a given material is unknown or sketchy and the
number of specimens tested at each applied stress level is too small, say less
than 50, to estimate the shape of the distribution. In such cases, these tech-
niques give conservative results.
(a) One Group at Each Stress Level. —Usually the first step in the analysis
of fatigue data is to draw the S-N curve for 50 per cent survival; it is the
curve fitted to the medians of the groups at the several applied stress levels.
The median, an "order statistic," is the middlemost value when the observed
values are arranged in order of magnitude, or the average of the two middle-
most values if the group size is even.
Other S-N curves, those for p per t survival (where p is not 50), may
be fitted to other order statistics if the group size is greater than 1. If the
group values are arranged in order of magnitude, NI is the minimum cycle
life value, or the first order statistic, A7 2 is the second observed value, or the
second order statistic, and so forth.
The estimated percentage of survivors for the population at cycle life
11
In the Normal distribution, the median and the mean are equal.
12
See Appendix IV, p. 71 .
values of Ni, or 7V 2 , depends upon the group size.13Table 8 gives the median
percentages at Ni and Nz for several group sizes. Some of these percent-
ages also are given in Table 25 for one group, m = 1, and the 50 per cent
confidence level.
The median percentage of survivors at the maximum value of the sample,
Nn , is 100 — (per cent for N\), etc. Examples of how Table 8 may be used
follow:
1. The 50 per cent survival curve may be estimated from the median of
any sample size.
2. If three specimens are tested at each applied stress level, the 79, 50, and
the 21 per cent survival curves may be estimated from the entries in Table
8 and their complements. The value 79 per cent is found opposite sample
size 3 in the second column, the value 50 per cent is taken from the median
TABLE 8.—MEDIAN PERCENTAGE OF
SURVIVORS FOR THE POPULATION.
cSample
™~i«, Size,
c; n
M - At^the Lowest At the Next^Lowest
Vajue ValuCj
1 . 50
2 70 30
3 79 50
4 84 61
5 87 69
6 89 73
7 90 77
8.". 91 80
:' 9. 92 82
10 93 84
" :• 11.-. 94 85
12 94 86
13 95 87
14 95 88
15.' 95 89
16, 96 90
S-N curve, and the value 21 per cent is obtained by subtracting the value
in the second column from 100 per cent.
3. If 7 specimens are tested at each applied stress level, the 90, 77, 23, and
10 per cent survival curves may be estimated from the entries in Table 8 and
their complements. The 50 per cent survival curve may be estimated from
the median.
At least 13 specimens must be tested at each applied stress level to esti-
mate the 95 per cent survival curve.
In practice, values of per cent l less than? 50 usually are not wanted.
Hence, if all of the specimens in a sample are tested simultaneously, the tests
may be stopped as soon as the specimen having the median value of fatigue
life for the sample has failed, unless the data are required for other purposes.
13
These are called "median percentages" because, half of the time, the true*percentage
will be larger, and for the other half of the time, smaller. They are close to, but usually
not equal to, the "expected" percentage of survivors, which is equal to 1 — i/(n + 1), where
i is the number of the order statistic and n is the sample size. The confidence level associ-
ated with expected percentages varies with the sample size, whereas it is constant for
median percentages.
As mentioned previously, the percentage survival values given in Table 8
are median values; they are based on a "confidence level" of 50 per cent. 14
Percentage survival values corresponding to higher confidence levels, such
as 95 or 99 per cent, are given in Table 25 for a single sample when m = 1.
For example, if three specimens are tested at an applied stress level, 79
per cent of the population are expected to survive N\ cycles (50 per cent
confidence level), but the statement that at least 37 per cent of the popula-
tion will survive N\ cycles may be made with greater confidence (confidence
level = 95 per cent). If estimates of the population percentage are made
from a series of samples tested at one applied stress level and the statement
is made that at least 79 per cent of the population will survive N\ cycles,
50 per cent of such statements are expected to be incorrect. If the statement
is made each time that at least 37 per cent will survive N\ cycles, only 5
per cent of such statements are expected to be incorrect. However, S-N
curves corresponding to a 50 per cent confidence level are usually shown.
The effect of fitting a curve to the same order statistics at several stress
levels probably increases the confidence level; how much is not known. If
S-N curves are based on other confidence levels, the fact should be plainly in-
dicated on the chart.
(Z>) Several Samples, or Groups, at Each Stress Level. —If it is not possible
to test all the specimens in a sample simultaneously and if stopping the tests
before all the specimens have failed is desirable to save time, the required
sample may be divided, at random, into two or more groups (see references
17 and 18). Then the median of the particular order statistics (the first,
second, and so forth) for the several groups may be used for constructing
the S-N curve. Table 25 gives values of percentage survival for several num-
bers of groups and several confidence levels.
EXAMPLE.—With five testing machines available, 15 specimens were tested at a
constant applied stress level in three groups of 5 each. For each group, all machines
were assumed to be stopped after the second failure. (Actually, all machines were
allowed to run until fracture occurred or until 10 million cycles of fatigue stressing
had been applied, so that the time saved could be estimated for this particular set
of tests.)
The test data are:
Group Life, kilocycles
1 162, 229, (261, 668, 2 281)
2 105, 131, (140, 245, 10 000+)
3 275, 373, (5 503, 8 695, 10 000+)
From these data we have:
Lowest Ranking Points Median
162, 105, 275 162
229, 131, 373 229
2Vi .
#2
Entering Table 25, under "Lowest Ranking Points," in the column for m = 3
groups, opposite n = 5 in each group, and at a confidence level of 50 per cent,
14 Technically speaking, the S-N curves based on order statistics are "nonparametric
tolerance limits," which are described by Murphy (16). The probability that at least p
per cent of the population lies above Ni cycles, where Ni is the ith order statistic of the
sample, is properly called a "tolerance level"; but the term confidence level appears to
have been used more frequently.
read 87.05 per cent. This value is an estimate of the percentage of the population
from which the original 15 specimens were selected that will survive 162 kilocycles.
Similarly, at a confidence level of 95 per cent, 67.03 per cent or more of the popula-
tion are estimated to survive the 162 kilocycles. Again, for the "Second Ranking
Points," at a confidence level of 50 per cent, 68.61 per cent of the population are
estimated to survive 229 kilocycles and, at a 95 per cent confidence level, 45.40 per
cent or more of the population are estimated to survive 229 kilocycles.
Additional information can be obtained from the preceding test data by con-
sidering all 15 specimens as one "group" and determining the percentage of the
population expected to survive 105 kilocycles, which is the lowest ranking point
for m = 1 and n = 15 in Table 25. For a 95 per cent confidence level, straight-line
interpolation between 74.11 per cent for n = 10 and 86.09 per cent for n — 20
TABLE 9.—CONFIDENCE INTERVALS
FOR THE MEDIAN."
Confidence Level =£ 0.95.
Confidence Limits
Sample Size, n
Lower Upper
6 # #6
7 #saa #7
8 # #8
9 #2 N8
10 #2 #9
11 #2 N10
12 N2 #W
13 #3 #11
14 #3 #12
15 N4 N12
20 #
N6 N15
25 # #W
30 #1 »
35 #1 1 #24
40. #1 . #27
45 #1 S #30
50 N13 » #3
" Based on a table in Nair (19).
gives about 80 per cent. From this, it is estimated that at a 95 per cent confidence
level about 80 per cent of the population will survive 105 kilocycles.
2. Estimates of Parameters—Single Stress Level:
(a) Median Fatigue Life:
1. Point Estimate. —A point estimate of the population median is the
sample median, described above in n V Al(a).
2. Confidence Interval Estimate. —A confidence interval for the median
that does not assume a particular frequency distribution for the population
may be computed if the sample size is larger than five.
The n observed values offatigue life, N, are arranged in order of magnitude
as follows:
Ni£N*gNf- £ N.
ANALYSIS OF FATIGUE DATA 27
The confidence limits corresponding to a confidence level of at least 0.95 are
given by the order statistics designated in Table 9, p. 26.
EXAMPLE. —Assume that ten specimens are tested at a particular stress level and
the observed values of fatigue life in kilocycles are 201, 224, 226, 230, 232, 238, 24
244, 245, and 248. The point estimate of median fatigue life is the average of the
two middlemost values, namely 235 kilocycles. The interval estimate is defined by
A/2 and Ng (see Table 9), which are 224 and 245 kilocycles, respectively.
The population median may be above or below the sample median—235
kilocycles —but the chances are at least 95 in 100 that the statement, "the
median lies between 224 and 245 kilocycles," is correct if the sample came from
one population.
(b} Mean Fatigue Life:
1. Point Estimate. —A point estimate of the population mean is the
sample average.
TABLE 1 0—APPROXIMATE CONFIDENCE INTERVALS FOR THE MEAN.

Confidence Level = 0.95.
Sample Size, n Procedure". 6 Length of Interval
3 ...... add the range of the observed values to the largest 3 X range
value and subtract it from the smallest value:
that is, Ni - (N3 - JVi) and N3 + (N3 - Ni).
4 ...... add (range) /4 to the largest value and subtract it lj£ X range
from the smallest value:
5 ...... use the range: N\ and Ns I X range

0
6 See Youden (20) for n = 3.
Private correspondence from W. J. Youden, for values of n greater than 3.
2. Approximate Confidence Interval Estimate. —An approximate confidence

interval estimate for the mean that does not assume a particular frequency
distribution for the population may be computed as shown in Table 10, if
the sample size is 3, 4, or 5.
(c) Per Cent Survival for a Stated Value of Fatigue Life:
1. Point Estimate. —A point estimate of the percentage of the population
that has fatigue life values equal to or above a stated value is the sample
percentage of observed values equal to or above the same stated value.
2. Confidence Interval Estimate. — Confidence limits corresponding to
possible values of sample percentage, p, for four sample sizes are given in
Table 11. Values for other sample sizes may be read from a chart from Dixon
and Massey (9), p. 415, from which many values in Table 11 were taken.
EXAMPLE. —Using the data given in the above example of this Section and 230
kilocycles as the stated value of fatigue life, the following estimates of the popu-
lation value of per cent survival are obtained: (1) point estimate: 70 per cent and
(2) interval estimate: 34 to 94 per cent.
A.larger sample size will give a shorter interval estimate (see Table 11).
(d) Fatigue Life for a Staled Value of Per Cent Survival:
1. Point Estimate. —A point estimate of the population value of fatigue
life for a stated value of per cent survival is based on order statistics as
TABLE 1 1 .—CONFIDENCE INTERVALS FOR PERCENTAGES."
Confidence Level = 0.95.
Sample Size
»=s n = 10 » = 20 » = 40
Limits Limits Limits Limits

#,per P, per P, per P, per
cent cent cent cent
Lower Upper Lower Upper Lower Upper Lower Upper
100 48 100 100 68 100 100 82 100 100 91 100

97.5 87 100
95 75 100 95 83 99
92.5 79 98
90 54 100 90 68 . 98 90 77 97
87.5 73 96
85 62 97 85 70 94
82.5 67 93
80 29 99 80 43 98 80 56 94 80 64 91
77.5 61 89
75 51 92 75 58 87
72.5 56 85
70 34 94 70 45 88 70 53 83
67.5 51 82
65 40 85 65 48 79
62.5 46 77
60 15 95 60 25 89 60 36 81 60 43 *75
57.5 41 73
55 32 77 55 38 71
52.5 36 68
50 18 82 50 27 73 50 34 66
Where: p = sample percentage (for example, percentage surviving). Confidence lim-
its corresponding to (100 — p) per cent are: lower: 100 — (tabular value for upper
limit corresponding to p, per cent); upper: 100 — (tabular value for lower limit cor-'
responding
0
to p, per cent).
Based on chart from Dixon and Massey (9), p. 415, and, for n = 40, on chart from
Pearson and Hartley (2), p. 204.
outlined in the Section on S-N curves: "One Group at Each Stress Level"
(Section V A). A particular value is e median, corresponding to 50 per
cent survival.
Another point estimate may be derived from the. cumulative frequency
distribution of the observed values. In general, the two point estimates would
not be exactly equal.
2. Confidence Interval Estimate. —Interval estimates for medians '(50 per
cent survival) are described in Section V A2(a). Interval estimates for fatigue
life values corresponding to other percentage points may be computed by
using reference (21) ,15
(a) Differences of Group Medians—Single Stress Level.—-If two or more
groups of specimens are tested, the question of whether the observed differ-
ences in the values are due to chance or to some differences in the popula-
tions from which the groups were drawn often arises. The observed differ-
ences, for example, could arise because of differences in material lots or
differences in the characteristics of the testing machines.
The rank tests given in this section assume that the several groups are in-
dependently and randomly drawn from populations that are of the same
shape but may differ with respect to their medians. All the observed values
in one group are assumed to come from one population. Since the populations
are assumed to be of the same (though unknown) shape, only those groups
that are tested at the same stress level should be compared, since the form
of the distribution tends to change with change in stress level.
1. Rank Test for Two Groups. —In the rank test for two groups the rank
of each observation in the two groups combined is determined. The lowest
value is given the rank of 1, the next higher observed value is given the rank
of 2, and so forth. If one value appears several times, that is, there is a tie,
the average of the ranks for those numbers is assigned to each one. For ex-
ample, if the llth, 12th, 13th, and 14th values are all equal, they are each
given the rank of (11 + 12 + 13 + 14)/4 = 12.5. The ranks for the two
groups are totaled separately and the total for one of the groups (the one
with the smaller number of observations if the group sizes are unequal) is
compared with the critical values given in Table 26 for sample sizes equal to
the group sizes.
If the observed value falls within the range of values given in Table 26 for
the chosen significance level (5 or 1 per cent), the groups may be considered
to have come from one population. If the observed value falls outside the
range of values given in the table, the two groups are said to be significantly
different, that is, to have come from two populations with different medians.
The use of the 1 per cent significance level gives a smaller risk of calling the
15 The interval Nk to N may be computed as follows:
m
(1) k is chosen so that
(2) m is chosen so that
where
« = 1 — (confidence level), and
p = (stated value of per cent survival)/100.
groups significantly different when they are actually drawn from one popu-
lation and the observed difference is due to chance.
EXAMPLE.—To compare two machines, the rank test was applied to the data
from 27 specimens randomly assigned to two testing machines. (See Table 12.)
According to Table 26, the rank total for Machine A in Table 12, which has "the
smaller number of measurements," should be between 101 and 179 (Ni = 10,
NZ = 17) for the 5 per cent level of significance, and between 89 and 191 for the
1 per cent level of significance. This means that the actual total, 87, would not be
expected to occur as often as once in a hundred samples due to chance alone, if the
two machines were completely interchangeable. Thus, on the average, the machines
give significantly different fatigue life values.
TABLE 12.—FATIGUE TEST DATA.
Machine A Machine B
Rank Kilocycles Rank Kilocycles

1 624 3 667
2 662 7 715
4 681 11 811
5 688 12 822
6 99 13 833
8 732 14 841
9 774 15.5 842
10 781 15.5 842
18 865 17 849
24 948 19 869
20 892
87 Total 21 903
22 944
23 946
25 1 , 032
26 1,067
27 1,092
291 Total
2. Rank Test for More than Two Groups. —The method of assigning ranks
is the same as for the two-group test, ranking the observations for all the
groups combined. The ranks are totaled separately for each group and the
following test-statistic, H, is computed from the rank totals (22):
where:
k =
number of groups,
Hi =
number of observations in the ith group,
N =
y^, ni, the number of observations in all groups combined, and
Ri =
sum of the ranks in the ith group.
The test-statistic H is distributed approximately as x2 with k — 1 de-
grees of freedom if each Ui is at least five. (For a discussion of x2, see refer-
ence (9)). Thus, the value of H calculated from the observed data may be
compared with the values of x2 given in Table 27 to determine whether
there may be a significant difference among the populations from which
the groups were drawn or not. If H is greater than the x2 value for k — 1
degrees of freedom and the chosen significance level, the populations are
said to be different; that is, the groups may be said to have been drawn
from two or more populations. Inspection of the rank totals will usually
show which groups are different from the others if the difference is signifi-
cant.
EXAMPLE.—To compare five machines, the rank test was applied to the data
from 25 specimens, randomly assigned to the five machines (see Table 13).
TABLE 13.—FATIGUE TEST DATA.
Machines A B c D E Total
(5)596 (6)599 (3)539 (2)530 (1)477
(10)640 (13)661 (12)651 (8)624 (4)568
(11)646 (21)760 (14)662 (9)638 (7)607
(18)733 (22)774 (15)675 (16)684 (17)719
(24)807 (23)781 (19)744 (25)889 (20)757
Sum of ranks, Ri. . . 68 85 63 60 49
R* 4624 7225 3969 3600 2401
Bf 924.8 793.8 720.0 480.2 4363.8
m 1445.0
H = 80.56 - 78 = 2.56
Entering Table 27 with 4 degrees of freedom, one less than the number of groups,
gives x2 = °.49, corresponding to a 5 per cent significance level or a percentile of
95. Since the computed value of H, 2.56, is very much smaller than 9.49, the ob-
served values of fatigue life may be considered to be from one population; the ma-
chines may be considered to be interchangeable.
(ft) Differences of Two or More Percentages (for example, per cent survival
values). —The test-statistic used to test the significance of the differences
among percentage values computed from observed data is x2- The formula
for x2 may be written in two ways; the second one is usually better for
computation purposes.
1. When the sample sizes are unequal:
where:
k = number of samples;
sum over k samples;
»,• = size of ith sample (i = 1, 2, • • • k);
Xi = observed number of "events" in the ith sample; an event may be a
failure, a survival, etc. ;
pi = observed fraction for the ith sample: pi = #*/«;; and
P = Z_, Xi/£_, ni = average fraction for all samples combined.
2. When the sample sizes are equal the formula reduce to
where n = sample size and x = ^ Xi/k (23, pp. 175-178). The other terms
were defined previously.
The computed value of x2 may be compared with the tabular values given
in Table 27 for k — 1 degrees of freedom (d.f.). If the computed value of
X2 is larger than the tabular value corresponding to: percentile =
100 — (chosen significance level), the percentages are said to be signifi-
cantly different; that is, the samples were drawn from different populations.
If the computed value of x2 is smaller than the tabular value, the samples
may be considered to have come from one population.
Another use of the x2 test is to test whether or not the observed per-
centage values are significantly different from an arbitrary value, such as
50 per cent. The method of computation is the same as that given previously,
except that: (1) the first way of writing the formula for x2 is used for the
computations, (2) the arbitrary value, which may be called p', replaces
p, and (3) d.f. = k.
EXAMPLE.—To compare six lots of phosphor-bronze 16strip, the x2 test was applied
to the data given in Table 14, using a significance level of 10 per cent.
TABLE 14.—PERCENTAGES SURVIVING 108 CYCLES.
Stress = ±25,000 psi.
Lot Sample Size, nt- Per Cent Number Sur- Xi*
Surviving, V)0pi viving, Xi »i
1 15 60.0 9 5.40
2 20 40.0 8 3.20
3 17 58.8 10 5.88
4 25 48.0 12 5.76
5 19 57.9 11 6.37
6 14 50.0 7 3.50
Total 110 = 57 = 3 0.1 1 =
16
Significance levels commonly used are 10 and 5 per cent.
The tabular value of x2 corresponding to degrees of freedom = 5 and percentile
= 90 per cent is 9.24. Since the computed value, 2.28, is much smaller than 9.24,
no significant difference among samples is indicated by these data and the samples
may be considered to have come from one population. If the computed value of x2
were larger than 9.24, the lots from which the samples were drawn would be con-
sidered to be significantly different.
4. Response Curves:
Instead of drawing S-N curves when specimens are tested at several
levels of applied stress, response curves may be constructed from the data
if the applied stress levels have been properly chosen. A response curve is
constructed for a stated value of the number of cycles-to-failure, such as
107, 108, and so forth. It is based on the values of per cent survival at N
cycles that are observed for several (at least 3) values of applied stress.
This method of analyzing fatigue data is especially useful when some of
the specimens "run out," that is, survive the duration of the test.
While a response curve may be drawn on any type of graph paper, ex-
perience has shown that the per cent survival values tend to lie along a
straight line when the data are plotted on Normal probability paper. 17
(a) Fitting a Response Curve. — If the observed per cent survival values lie
along a straight line when plotted on Normal probability paper, a straight
line may be fitted to the points by eye or by the method of least squares.
The latter method is, of course, more precise and not subject to the biases
that may be introduced by a person fitting a line by eye.
The equations for the slope and intercept of a line fitted by the method
of least squares to per cent survival values having equal weight are: 18
where :
X = applied stress value (usually coded to reduce the size of the number) ,
p = per cent survival in each group of specimens tested,
Y = z, a value of the Normal deviate obtained from Table 28. The value
of 0 corresponding to p is obtained by entering the column headed
"Area" with 1 — (/>/100) and reading the corresponding value of z.
Interpolation between tabular values may be required. Y is called
the transformed value of p,
X — ^L, X/k = average of . X values,
Y = ^^ Y/k = average of Y values,
k = total number of groups tested, or total number of observed
p values, and
2^ = sum from 1 to k.
Intercept, a = Y
17
No. 3127, Codex Book Co., Inc., Norwood, Mass; or No. 358-23, Keuffel & Esser
Co.,18 New York, N. Y.
If the specimens have been allocated as indicated in Table 1, the per cent survival
values will have approximately equal weight. If this has not been done, references (1),
(2), or (3) should be consulted for a method of analyzing the data.
34 FATIGUE TESTING AND STATISTICAL ANALYSIS OP DATA
The equation for the fitted line isf Y= a + b (X — X} JThe subscript, /,
is used to denote the fitted line values.
TABLE 15.— "PROBIT" TEST DATA.
AppliedStress, Ksi Applied Stress, ksi Number^ Specimens Numberof Specimens Per Cent Survival, ,
40.0. 15 14 93. 33
41. 5. 8 6 75. 00
43.0. 5 3 60.00
44.5. 8 2 25. 00
46.0. 15 1 6.67
TABLE 16.—COMPUTATIONS FOR FITTING A RESPONSE CURVE

BY METHOD OF LEAST SQUARES.
Applied Stress, Coded0 Per Cent

ksi Value, X Survival, p
Transformed
Values, Y „ XY
Fitted Values
F/ Pf
40.0 1 93.33 -1.50 1 -1 .50 -1.51 93.4

41.5 2 75.00 -0.67 4 -1 .34 -0.78 78.2
43.0 3 60.00 -0.27 9 -0.75 +0.05 52.0
44.5 4 25.00 + 0.67 16 •> +2.68 +0.68 24.8
46.0 5 6.67 + 1.50 25 +7.50 + 1.41 7.9
Sum 15 -0.25 55 + 6.59
X = 3 F - -0.05
" The stress values are coded by assigning numbers 1 to 5 to the successive applied
stress values.
EXAMPLE.—Fatigue tests of specimens of a certain steel gave the data in Table

15 for five applied stress levels. The number of specimens tested at each stress level
was chosen in accordance with the procedure outlined in Section III Al, so that the
per cent survival values would have approximately equal weight.
In order to fit a straight line to these data, the p values must be transformed to
Y values, as outlined above and demonstrated in Table 16; also it is convenient to
code the applied stress values when the increments between successive values of
applied stress are equal.
Slope,
Intercept, a = 5
-0.05 + 0.730T - X)
Yf =
pf = 100 (1 - "Area" value)
where an Area value corresponds to a value of z = F/ in Table 28.

Figure 7 shows the observed values of per cent survival at 107 cycles
(a) Confidence band for per cent survival values.

(b) Confidence band for fatigue strength values.
FIG. 7.—Response Curve for a Particular Type of Steel.
(open circles) plotted on Normal probability paper. To draw in the line
fitted to the observed values by the method of least squares, select at least
two values of applied stress and the corresponding values of pf from Table
16; connect these plotted values by a straight line. (The confidence limits
are discussed in the next section.)
In fatigue,«tudies the derived estimate, ^(F), corresponding to a stated
value of p (or F) is often required. The equation for the derived estimate,
.called X, ajfunction of F, is:
where F = the transformed value of the stated per cent survival value. This
derived estimate is called the fatigue strength for p per cent survival at N
cycles.
TABLE 17.—COMPUTATION OF STANDARD DEVIATION, s.
Applied Stress, ksi Y Yf (F - F/)*
40.0 — 1 50 — 1 51 - 0 0001
41. 5 — 0 67 — 0 78 0 0121
43.0 -0. 25 — 0.05 0.0400
44.5 + 0.67 + 0 68 0.0001
46 . 0 + 1 . 50 + 1 41 0.0081
Sum 0.0604
(V) Confidence Limits for a Response Curve. 19—The method of computing

symmetrical20 confidence limits fora response curve plotted as a straight line
on Normal probability paper when the test specimens have been allocated
according to Table 1 is presented here. While no assumption about the distri-
bution of fatigue life values is required, the computation of confidence
limits by the method outlined here assumes that the Normal transformation
of the per cent survival values is suitable, that is, that the transformed
percentages are Normally distributed about the fitted line.
Two types of confidence limits may be computed for a response curve
that has been fitted by the method of least squares:
1. Confidence limits for per cent survival at N cycles for a stated value
of applied stress (Fig. 7(a)). (This method is an alternate for the one de-
scribed in Section V A2(c) for a single stress level.)
2. Confidence limits for fatigue h at N cycles for a stated value of
per cent survival (Fig. 7(6)).
The two types of confidence limits are computed differently, but the
resulting confidence bands are numerically equivalent (Fig. 7).
For either type of confidence limits, the standard deviation of the ob-
19
20
Much of the material in this section is based on Chapter 18 of Hald (24).
The risk, equal to 1 — (confidence level), is evenly divided between the two limits.
served Y values about the fitted line is needed. Provided that all Y values
have equal weight (see Section III Al), the following equation may be used:
The symbols have been defined previously; the computation procedure is

given in the following example.
EXAMPLE.—Table 17 shows the method of computing the standard deviation
of F values about the fitted line, using the values of F and F/ given in Table 16.
1. Confidence Limits for Per Cent Survival Values at N Cycles. —In com-
puting such confidence limits for a stated value of applied stress, the trans-
TABLE 18.—METHOD OF COMPUTING 95 PER CENT CONFIDENCE
LIMITS FOR PER CENT SURVIVAL VALUES.
Confidence Limits,
Applied _ Coded
Stress, ksi Value, Y Values p Values
X
i
Lower Upper Lower Upper
40.0 1 4 0.40 0.77 0.47 -1.51 -1 .98 -1 .04 97.6 85.1

41 .5 2 1 0.1 0 0.55 0.34 -0.78 -1 .1 2 -0.44 86.9 67.0
43.0 3 0 0.0 0.45 0.27 -0.05 -0. 3 2 + 0.22 62. 6 41 .3
44 5 4 1 0.1 0 0.55 0.34 + 0.68 + 0.34 + 1.02 36.7 1 5.4
46.0 5 4 0.40 0.77 0.47 + 1.41 + 0.94 + 1.88 1 7.4 3.0
Sum . . . 10
formed values of per cent survival are used. The formula for symmetrical
confidence limits is:
where:
X' = the stated value of applied stress (coded),
X = average of X values for the data used in fitting the line,
^(X — X) 2 is computed for the X values for the data, and
FZ , k -2 = a factor obtained from Table 32 for 2 and k — 2 degrees
of freedom for numerator and for denominator, respec-
tively. 21
The other symbols have been defined previously.
The confidence limits should be computed for all the applied stress levels
21
Hald (24) uses t = (Fi, k - 2)1/2 instead of (2F2, k - 2)1/2. The wider limits given here
ensure the desired confidence level when several confidence statements are made using
the same fitted line.
used in the experiment and may be computed for any other value within the
range of the applied stress values. The method of computation is given in the
following example.
EXAMPLE.—Table 18 shows the method of computing 95 per cent confidence
limits for per cent survival, at 107 cycles, using the value of 5 computed in Table
17 and the Yf values from Table 16, where:
n = 5, the number of observed p values,
^2,3 = 9.55, the value of F0.95 in Table 32 for 2 d.f. for numerator and 3
d.f. for denominator,
(2F2, 3 )1/2 = (19.10)1 /2 = 4.37.
s(2F2, s) 1/2 = 0.14 (4.37) = 0.61, and
p = 100 (1 — Area value); an Area value corresponds to a z = Y value in
Table 28.
The confidence limits computed in Table 18 are plotted in Fig. 7(a) and
connected by dotted lines to form a confidence band. The band applies to
TABLE 19.—METHOD OF COMPUTING 95 PER CENT CONFIDENCE
LIMITS FOR FATIGUE STRENGTH VALUES.
Confidence Limit! •CJ- Es(2F*c i)"t
Coded
p F X(Y)
Values Coded Values Fatigue
Strength, ksi°
Lower Upper Lower Upper
90 -1 .28 1 .322.82 0.282 0.68 0.62 1. 19 0.57 1.81 39.36 41 .22

75 -0.67 2.1 5 0.72 0.072 0.51 0.46 2. 09 1.63 2.55 40.95 42.32
50 0 3.07 0.00 0.000 0.43 0.39 3. 07 2.68 3.46 42.52 43.69
25 + 0.67 3.99 0.98 0.098 0.53 0.48 4. 06 3.58 43.87 45.31
4.54
10 + 1.28 4.82 3.31 0.331 0.72 0. 66 4. 96 4.30 5.62 44.95 46.93
" 38.50 + (1.5 X coded value); 38.50 corresponds to zero on the coded scale, see Table
16. For example: lower confidence limit for fatigue strength corresponding to p = 90 is
38.50 + (1.5 X 0.57) = 38.50 + 0.86 = 39.36 ksi.
any value of applied stress within the range used in the experiment. 22 For
example, for an applied stress of 41 ksi, the point estimate of the population
of per cent survival at 107 cycles is 85 per cent. The confidence interval
estimate, corresponding to a confidence level of 0.95, is 74 to 92 per cent.
2. Confidence Limits for Fatigue Strength at N Cycles. —These are confi-
dence limits on the derived estimate, X(Y}. The exact formula for the
confidence limits is:
where:
F = the transformed value of the stated per cent survival value and
The other symbols have been defined previously.

22
Caution should be used in drawing conclusions from extrapolated portions of the
response curve or the confidence band.
In many cases C deviates only slightly from b and the approximate
confidence limits will be close to the exact ones:
EXAMPLE.—Table 19 illustrates the computation of 95 per cent confidence

limits for fatigue strength values at 107 cycles, using the exact formula. In table 19:
p = stated value of per cent survival;
Y = transformed value of p (See Section V A3 (a);
The confidence limits computed in Table 19 are plotted in Fig. 7(6) and
connected by dotted lines to form a confidence band. The band applies to
any value of per cent survival within the range of observed values, that is,
the extent of the fitted line. It will be noted that this confidence band differs
from the one in Fig. 7(a) only at the ends and that, numerically, the bands
are equivalent. Thus it is recommended that confidence bands be computed
according to the simpler method illustrated in Table 18 and the confidence
limits for fatigue strength be read from the intersections of the confidence
band and the stated value of per cent survival. For example, for 50 per cent
survival, the point estimate of the population value of fatigue strength is
43.1 ksi. The confidence interval estimate, corresponding to a confidence
level of 0.95, is 42.5 to 43.7 ksi. Likewise, for 80 per cent survival, the point
estimate is 41.4 ksi and the confidence interval estimate is 40.5 to 42.0 ksi.
The method illustrated in Table 19 should be used when a confidence
band is not presented or when the s are needed to more decimal
places than can be read from a chart.
B. LITE DISTRIBUTION SHAPE ASSUMED
1. Normal Distribution of Fatigue Life:
The preceding analyses, which made no specific assumptions about the
shape of the fatigue life distributions, are always applicable but, because of
their generality, may not give results that are precise enough for all pur-
+40 FATIGUE TESTING AND STATISTICAL ANALYSIS or DATA
poses. In order to sharpen the inferences which may be made from the data,
particular functional forms can be assumed for these distributions and the
statistical analyses carried out relative to these forms. It should be pointed
out, however, that the precision so gained is real only when the particular
distribution chosen honestly reflects the random process being studied.
(a) Choice of Distribution Shape. —In the literature are several specific
fatigue life distributions which have some basis in theory. However, em-
pirical evidence gathered to date has not allowed a clear-cut choice to be
made among them. In this position, it is sensible to choose a form for the
fatigue life distribution according to the following two criteria:
1. There should be satisfactory agreement between the data gathered and
the results predicted by the "theoretical" distribution.
2. The distribution form should be mathematically tractable or, at worst,
one on which extensive studies have already been made so that laborious
computations can be kept at a minimum.
These two conditions can be fulfilled by finding a transformation which,
when applied to a set of fatigue life data, results in new data which closely
follow the Normal or Gaussian distribution. Once this has been done, all
the results of Normal distribution theory are available to analyze the trans-
formed data.
The particular transformation which will do the job mentioned above is
not necessarily the same for all cases. It has been found that replacing the
observed fatigue life, N, by log N will often give a distribution which is
approximately Normal. Some situations may arise, however, where this
transformation is inappropriate. The joint efforts of the test engineer and
experimental statistician will be needed to make a specific choice which is
suitable for the problem being studied. If usable inferences are to be made,
the data transformation should not be so variable as to be a function of tjje
individual sets of data, but the problem of what transformation to use is
not a cut-and-dried statistical one.
In Section B, all random variables, unless otherwise noted, are assumed
to have been transformed so that they are Normally distributed.
(b) Estimates of Parameters. —On the assumption that the variable being
studied is Normally distributed, the information in the sample can be used
to obtain estimates of the parameters for the particular population. In
many situations it is sufficient to obtain point estimates of n and a. More
meaningful estimates, however, may be gotten by the use of confidence
intervals, since these utilize not only the point estimates, but also depend on
the amount of variation expected in these estimates from experiment to
experiment. A 95 per cent confidence interval on /*, for example, can be
described in the following way:
Once the data from a particular experiment have been collected, an inter-
val which is presumed to include the true parameter value, p, can be calcu-
lated from the sample values in a prespecified way. Since this interval is a
function of the observations, it is itself a random quantity, both as to posi-
tion and length, so that it will not be certain to include /x in any single
experiment. The process for constructing the interval from the sample can
be determined, however, in such a way that the chance of \L being covered
by the interval in a given experiment is exactly 0.95. It is then said,, "With
a confidence level of 0.95, the true value, >, is included in the interval so
computed."
The mean, /*, of the Normal population, which is equal to the median of
the population, may be estimated either by the sample mean or by the sample
median. The sample median is the less efficient of the two, in the sense that
it experiences a larger variation from experiment to experiment but has the
advantage of being simpler to compute. The standard deviation, a, is
estimated by the sample standard deviation. The sample quantities—the
mean, x, and the standard deviation, s—are summarized below, where
Xi is the tth sample value from a sample of size n:
where ^ is the sum from I ton.

These quantities as they stand serve as point estimates of the population
parameters. If these estimates are in terms of transformed variables, they
should be converted to cycle life values for presentation. The construction
of appropriate confidence intervals is described in Section V B3.
2. S-N Curves:
Suppose the cycle life (or some transformation of cycle life) of specimens
from a certain population is a Normally distributed random variable, the
applied stress being fixed. If the parameters of this distribution, /x and a,
were known, then for any preassigned percentage, p, a number, K, could be
determined so that p per cent of the population would have cycle lives
exceeding /j, — Ka. In general, however, n and a are unknown and can only
be estimated by information obtained from a sample drawn from this popu-
lation. What is done in this case is to determine a number k such that the
probability of the random variable x — ks not exceeding /x — Ka is exactly
7, where 7 is a confidence level chosen in advance. It is then said, with a
confidence level of 7, that at least p per rcent of the population is greater
than x — ks, where x and s have been computed from a sample of size n as
described in Section V Bl. The numbers, k, which are called "one-sided
tolerance factors," are functions of p, 7, and n. Table 33 lists some values
for k.
Consider the problem of constructing a specific S-N curve, say, for 90
information:
If the applied stress is Si, then 90 per cent of the specimens to be tested
will survive Ni cycles. Since the parameters of the fatigue life distribution
are not known, the above defined S-N curve cannot be constructed. One
can, however, construct a curve whereon any point (£2, ^2) has the fol-
lowing meaning: if the applied stress is Sz, then with a confidence level of
7, at least 90 per cent of the specimens to be tested will survive Nz cycles.
The confidence level, 7, may be chosen in advance in accordance with the
specific purpose of the S-N curve.
An S-N curve is determined by fitting a curve of appropriate shape to
points computed from data at several stress levels as described in the next
paragraph. The confidence level associated with the fitted curve cannot be
specified precisely but is probably larger than the confidence level used at
each value of applied stress.
The mechanics of constructing a point for fitting an S-N curve may be
described as follows:
Given a sample of n cycle lives for a fixed stress level S, compute the
mean, x, and standard deviation, s, of the transformed cycle lives. From
Table 33, read the value of k corresponding to the per cent survival, p,
the confidence level, 7, and the sample size, n, that are being considered.
The value x — ks is then the appropriate abscissa for the ordinate, S, on
the S-N curve. As in other instances, for presentation this value may be
transformed back to read cycle life directly.
Other sources discuss tolerance limits, which are different from those
given here. These are basically of two types:
(a) Two-sided Tolerance Intervals. —These are discussed in reference (9).
Tables are given there for two-sided tolerance factors. These factors are not
applicable directly to fatigue data analysis, nor is it possible to transform
them in an elementary way to obtain one-sided factors.
(b) One-sided Tolerance Intervals of the Form (-co, x + ks).—These
are completely symmetric with the intervals of the form (x — ks, <») as
given here. The tolerance factors, k, are the same for both types of one-
sided intervals.
3. Confidence Intervals:
(a) Confidence Interval for the Mean*. —The statistics, x and 5, in addition
to serving as point estimates of n and a, are also used in constructing con-
fidence intervals for these parameters.
The construction of a confidence interval for n is done as follows:
1. Select a confidence level, say 7 (where 7 is a proportion, 0 < 7 < 1,
not a percentage), which is appropriate. Keep in mind that there is a risk
of 1 — 7 that the interval being constructed will not contain p. Also re-
member that the greater the confidence, the wider the interval will be.
2. Having chosen 7, compute fti = (1 — 7)/2 and j82 = (1 + 7)/2. From
Table 29, where t$ is tabulated as a function of degrees of freedom (d.f.) for
several values of ft, read fa and fa , taking d.f. = n — 1. (Note that fa is
negative.)
3. The desired confidence interval s the limits:
(b) Confidence Interval for the Standard Deviation. —A confidence interval

for a has analogous meaning to that for /x and is similarly constructed:
1. Select a confidence level, 7. The remarks about 7 made before are still
pertinent.
ANALYSIS OF FA 43
2. Compute ft\ and (82 as before. From Table 30, read:
using percentiles 100 ft and 100 ft respectively, with d.f. = n — 1.

3. The desired confidence interval has the limits
EXAMPLE.—This example will illustrate the construction of confidence intervals.

The entries in the following table are logarithms of fatigue life values as observed
in a test on eight specimens:
i x{ = log Ni
1 4.8388
2 4.9243
3 4.9445
4 4.9542
5 4.9731
6 4.9777
7 5.0334
8 5.0828
The estimates are readily determined to be: x = 4.9661 and s = 0.0726.
These are obtained using the denning formulas from Section V Bl. It is
now desired to have 90 per cent confidence intervals on both n and <r. The
following quantities are relevant (see above):
The desired confidence interval for /* has the limits
or
4.9175, 5.0147
Similarly, the desired confidence interval for <r has the limits:
or
0.0512, 0.1304
Note that the point estimates and confidence interval estimates are log-
arithms and may be converted to cycle life values for presentation.
Sample median = antilog X = 92 thousand cycles. (This value may be
found from the original, imtransformed observations.) Confidence limits
for the population median are derived directly from the confidence limits
for JLI: antilog 4.9175 = 82,700 cycles and antilog2 5.0147 = 103,400 cycles.
Sample average, N = antilog (X + 1.1513<r ).23 Since o-2 is unknown,
an approximate value of the sample average may be found by replacing
er2 by s2. Then:
sample average, N = antilog (X ± 1.1513 s2) = 93,792 cycles
where = means "approximately equal to".
Sample standard deviation:
where N' is the population mean in terms of cycle life. Replacing N' by
N gives:
Sample standard deviation
Approximate confidence limits for the population standard ^deviation are

derived from the confidence limits for v.
If cycle life data are collected at several stress levels for the purpose of
estimating the life distributions of these several levels, the techniques of
analysis just described are applicable. If, however, the principal interest is
a comparison of distributions in the sense of deciding whether their paranr-
eters are the same, then a different set of analyses is needed in order to use
the information in the data more effectively. A summary of these techniques,
classified according to the questions they are designed to investigate, is
given in this section. Except for the t on "Differences Among k Means,"
this section is concerned with a comparison between two sets of data.
(a) Difference Between Two Standard Deviations. —Before testing whether
or not the means of two samples are significantly different, it is important
to investigate whether the standard deviations are significantly different or
not. Let $i be the standard deviation for a sample of size n\ from the first
population and let s 2 be similarly defined for the second. Compute the ratio
23 See Hald (24), p. 161, and Soyero and Olds (25), p! 671.
of the squares of the standard deviations, V, putting the larger standard
deviation in the numerator; if
then
In order to decide if the difference is statistically significant, a signifi-

cance level must have been chosen in advance. If the chosen significance
level (as a proportion) is denoted by a, compute 0 = 1 — (a/2) (see Note)
and, if /3 = 0.95 or 0.975, read from Table 32 the value of Fp corresponding
to n\ — 1 ''Degrees of Freedom for Numerator" and n% — 1 "Degrees of
Freedom for Denominator." (If s-? > Si2, the degrees of freedom are re-
versed.)
If V > Fp , then the standard deviations are considered to be signifi-
cantly different. If V ^ F$ , the standard deviations are considered to be
not significantly different.
NOTE. —-In "two-sided" tests, one-half of the critical region of the distribution curve of
the statistic is in the upper tail, while the other half is in the lower tail; consequently, when
using tables to obtain Fp and //j values, these values are obtained for significance levels of
one-half of the level stated.
(Z>) Difference Between Two Means:
1. Standard Deviations Not Significantly Different. — -If V ^ Fp (see (a)
above) , the hypothesis that the population means are equal may be tested
by the following procedure: compute s2, an estimate of the common vari-
ance, by
Next compute the test-statistic
where x\ and £ 2 are the sample means for the first and second samples,
respectively.
With a preassigned significance , a, compute 0 = 1 — (a/2) and
find tp , from Table 29, corresponding to d.f. = n\ , + n z — 2. If \t \ > t$,
conclude that the populations are different in mean or in variance or both.
(On the average, identical populations will be erroneously judged different
about 100 a per cent of the time.) If the samples are large enough so that
the test for the difference between two standard deviations would have
probably detected any important difference in the variances, a value
\ 1 1 > fy can be attributed to different population means. With the same
reservation about sample size, if I / 1 < tp, one may conclude that the
population means are not significantly different.
2. Sample Standard Deviations Significantly Different. —Ifthe sample
standard deviations are significantly different, the hypothesis that the24
population means are equal may be tested by the following procedures.
Compute the numbers:
and
From Table 29 read a value /j», corresponding to
If this value is not an integer, use the nearest smaller integer.

If I /' I < tp, the population means are judged equal and if I /' I > tp,
they are judged unequal; a = 2(1 — /3) is the approximate proportion of
the time when the means are actually equal but will be incorrectly judged
unequal.
TABLE 20.—COMPUTATIONS FOR SIGNIFICANCE TESTS.
First Surface Finish Second Surface Finish
Xi = log Ni *,- - x (*,- - 5)* *< = log Ni Xi — X (*< -i)«
4.8388 -0.1273 162.1 X 10~4 4.5315 -0.1180 139.2 x io-4
4.9243 -0.0418 17.5 4.6232 -0.0263 6.9
4.9445 -0.0216 4.7 4.6232 -0.0263 6.9
4.9542 -0.0119 1.4 4.6435 -0.0060 0.4
4.9731 +0.0070 0.5 4.6435 -0.0060 0.4
4.9777 +0.0116 1.3 4.6532 +0.0037 0.1
5.0334 +0.0673 45.3 4.6721 + 0.0226 5.1
5.0828 +0.1167 136.2 4.6902 + 0.0407 16.6
4.6902 +0.0407 16.6
4.7243 +0.0748 56.0
39.7288 369.0 46.4949 248.2
EXAMPLE.—Assume that, after testing specimens with one surface finish, another
lot is fabricated with a different finish and tested at the same stress level. The tests
are to be analyzed to determine whether the change in surface finish significantly
affects the fatigue life at the stress level used for the tests. The distribution of
24 This is an approximate test; see Hald (24), pp. 397-398.
fatigue lives at this stress level is roughly log-normal; therefore, the computations
are in terms of log N rather than N. (See Table 20.)
We now have the mean, x, and the variance, s2, of each sample. To test them
for significant differences, use the F-ratio test and then the /-test, choosing a
significance level of 0.05.
From Table 32, F0. m (d.f. numerator = 7) (d.f. denominator = 9) = 4.20.

The variances of the two samples are considered to be not significantly different.
Thus the test given in Section V B4(6)l may be used:
From Table 29, t0. 976 (d.f. = 16) = 2.12. Since t = 10.8 is larger than t0.y, 5 =
2.12, the mean of the first sample is judged to be significantly larger than the mean
of the second sample. In the matter of fatigue life at the test stress level, the
second surface finish appears to be inferior to the first.
(c) Difference Among k Means. :
If k (k > 1) sets of data have been obtained, each of which is a random
sample from a Normal population, these populations being taken as having
a common standard deviation, then one can test the hypothesis that these
populations have a common mean. Let:% be the/th observation from the iih
group, Xi 2be the mean of the ith group, and Ui be the size of the tth group.
Define sw , average variance within samples, and sb*, variance among sam-
ples, by
and
where x is the mean of the numbers xt . Compute the ratio

From Table 32 read the value of F$ (/3 = 1 — a), corresponding to k — 1
"Degrees of Freedom in the Numerator" and T^ is
w,- — jfe "Degrees of Prec-
dom in the Denominator." The hypothesis that the means of the k popula-
tions are equal is accepted only if V < Fp .
It is recommended that a statistician be consulted if a more complete
exploration of this situation is considered necessary.
5. The Staircase Method:
As pointed out in Section III B2, the staircase method has been found
useful for estimating the mean fatigue strength at a preassigned cycle life,
N. The procedure for experimentation and some discussion of the ad-
vantages and limitations are given in Section III.
(a) OtUline of A nalytical Method:
1. Determine whether failures or runouts are the less frequent event.
Only the less frequent event is used in the analysis.
2. Number the stress levels and make a table as follows:
Stress i Ni iNi
Sz 2 N2 2N2
Si 1 Ni N!
So 0 No 0
Sum: A
where:
i = 0 is assigned to the lowest stress, S0, on which the less frequent
event occurs;
i = 1 is, assigned to the stress level for i0 + d, etc.;
Ni = number of the less frequent events at the corresponding levels;
iNj = product of i and Ni at each level; and
A = "£iNi.
3. Estimate the mean fatigue strength by
where:
N = total number of the less frequent events,
d = preselected stress increment, an
S0 = first stress level.
And use +1 if the less frequent event is a run out, or — ^, if the less frequent
event is a failure.
6. The Modified Staircase Method:
(a) For Small Sample Sizes. —For sample sizes less than 15, use the fol-
lowing simplified procedure:
ANALYS h
1. Start the staircase procedure as before, but use a step size, d, equal to
twice the preselected stress increment until the first pair of opposite results
is obtained. Subsequently, use a step size, d, equal to the preselected stress
increment.
2. In the analysis, disregard the specimens used up to the first pair
which give opposite results. Let n = total number of specimens tested,
starting with the first pair of opposite results. Let C = sum of the stresses
used on the last n — 1 tests plus the stress which would have been used
on the next test had it been run. Then,
C
m ——
n
or merely the average stress used on the last n — 1 tests plus that which
would have been used on the (n + l) th test.
EXAMPLE.—The following data were obtained in a modified staircase fatigue
test:
Stress, ksi » = 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
62 h h h
h h h h
61 h h h h h h h
60 h h h
59
58 h
Where x = specimen has failed and o = specimen has not failed (as in Fig. 3).
The n is taken as 1 with the first specimen of the pair which gave opposite
results. The (n + l) th, or 17th, test would have been run at 61.0 ksi. The last
n — 1 tests started at n = 2. Therefore, the total C is composed of the stresses
from n = 2 (62.0), to n = 17 (61.0).
C 980
m — — = -— = 61.25 ksi
n 16
(V) For Reducing Testing Time—For staircase runs involving more than
30 specimens, the time required to complete the staircase method is long
because of the sequential nature of the procedure. This long time can be
reduced by dividing one long staircase program into several shorter, inde-
pendent staircases and conducting these simultaneously.
1. Proceed by dividing the total r of specimens, T, into r groups
of n each, so that rn = T. Test each group as a separate small-sample stair-
case program, as described previously. Test as many groups simultaneously
as desired.
2. The best estimate of m for the whole set of T specimens is then:
s iMii
m = h
= arithmetic average of r values of m.
EXAMPLE.—The step size is 5 ksi; and the data arose from testing five groups
of 15 specimens each.
Group m
No. 1 101
No. 2 102
No. 3 97
No. 4 99
No. 5 99
TABLE 21.—R. R. MOORE ROTATING BEAM; STEP0 TESTS OF
42 SPECIMENS.
SAE 4330 Steel: Tensile Strength, 140 ksi (Nominal).
Failure Stress, Kilocycles
Failure at
Stress Failure Stress, Kilocycles
Failure Stressat
ksi ksi
928 950
585 692
2 082 1 077
665 65 1 177
969 5 184
60 < 1 567 7 661
1 522
490 997
1 085 1 524
1 285 566
10 390 630
108
289 70 815
595 3 850
384 1 166
479 681
1 039 967
65 < 973 606
329
1 489 680
1 265 h 886
2 076 735
(. Continued, nex ! column) 1 479
0 Starting stress = 55 ksi. Steps = 5 ksi. Run = 107 cycles or to failure.
The estimate of the mean fatigue strength of the population is given by:
7. The Step Method:

The data from the step method described in Section III can be analyzed
by the use of arithmetic or logarithmi probability paper. This is illustrated
by analysis of the data presented in Table 21.
In this test, all 42 specimens were first tested at an applied stress of 55
ksi, at which stress level all survived 107 cycles. The applied stress was then
raised to 60 ksi and the test repeated; the 31 surviving specimens were
tested at 65 ksi; and so forth. In each case, the surviving specimens are
subjected to an applied stress that is 5 ksi higher than the preceding value.
This is repeated until all specimens have failed. The mean between the last
TABLE 22—ANALYSIS OF DATA IN TABLE 21.
Estimated Fatigue Number of Specimens Per Cent of 42 with This Per Cent of 42 Having at
Strength, ksi with This Fatigue Fatigue Strength Least This Fatigue
hhh Strength
57.5 11 26.2 100
62.5 16 38.1 73.8
67.5 11 26.2 35.7
72.5 4 9.5 9.5
(77.5) 0 0 0
stress survived and the failure stress is the estimated fatigue strength for
each specimen. The data are then tabulated as indicated in Table 22. The
data in the last column can be plotted on arithmetic probability paper, as
shown in Fig. 8.
FIG. 8.—Per Cent of Specimens Having At Least the Indicated Fatigue Strength at
107 Cycles.
The stress levels of 100 per cent and 0 per cent survivals are indicated
on Fig. 8 by short arrows. It is interesting to note that the dash-line ex-
trapolations of the "reasonably straight line" suggest that, if a very much
larger number of specimens had been tested, some of them would probably
have a fatigue strength higher than 72.5 ksi and some might not have
survived a stress of 55.0 ksi. Extrapolation on probability paper of the type
used for Fig. 8, however, depends critically on the assumption of a Normal
population; this assumption is almost certainly not safe. G. W. Snedecor
(26) says that extrapolation is mostly a guessing game.
If a straight line does not reasonably represent the plotted points, the
population distribution is probably not nearly Normal enough to warrant
analysis by logarithmic or arithmetic probability paper. In such cases, a
statistician should be consulted to determine whether some transformation
will Normalize the data. If the straight line is acceptably close to the plotted
points, the mean fatigue strength is approximately the value of the ab-
scissa at P = 0.50. Otherwise, that value is the median, that is, the value
above which half of the specimens survived, but it is not necessarily the
mean. If the straight line is acceptably close, the approximate fatigue
strength corresponding to any given mortality ratio (per cent failure) can
be estimated by reading directly from the line, but only between the first
and last plotted points. Extrapolation to very low or very high probabilities
is entirely unjustified and may be very misleading.
TABLE 23.—PROT TEST COMPUTATIONS.
Fracture Change in
Group Stress, S « (Mean),
psi per cycle Log a Line'i Log a
(Mean), psi (A Log a)
C 92 700 0.208 9, .3181 - 3
B 80 600 0.0592 1 ,7723 - 3 C -B 0.5458
A 71 800 0.00705 0 ,8482 - 3 B—A 0.9241
C 92 700 0.208 9, ,3181 - 3 C -A 1.4699
0 See Fig. 11.
8. The Prot Method:
The Prot method is based upon the assumption that the curve relating
the failure stresses and some power of the rate of increase of stress per
cycle is a straight line, provided that (1) the power is properly selected and
(2) this straight line will intersect the axis of failure stresses at the fatigue
limit. This may be expressed by the equation:
where:
,5 = fracture stress (final breaking stress), psi,
E = fatigue limit, psi,
K = a constant,
a = rate of increase of stress, psi per cycle, and
n — exponent of a that makes S a linear function of an.
A mathematical solution for the value of the fatigue strength and the
optimum value of n, the power of a that will place the plotted test values
on a straight line, and for the standard deviation of the fatigue strength is
proposed by Corten et al (5). The following "cut-and-try" method provides
an alternative to the mathematical solution.
The arithmetic mean of the individual values of a in each of the three
groups is determined, as well as the arithmetic mean of the individual values
of S—the fracture stress—in each group. (Strictly, the geometric means
should be used instead of arithmetic means, but the difference would usually
be unimportant.) This calculation gives three pairs of values, such as those
included for illustration in the second and third columns of Table 23. They
represent the entire set of test data fairly well. The test data may also be
plotted as shown in Fig. 9.
Stress Cycles
FIG. 9.—Prot Test: Stress as Linear Function of Stress Cycles.
TABLE 24.—PROT TEST COMPUTATIONS.
Assumed Log Change in Slope, n =
E, psi Group S-E (S-E) Line Log (S — E) A log (S - E)
[A log (S - E)] -5- A log a
f° 21 700 4.3365 C -B 0.3542 0.3542/0.5458 = 0.65

71 000. . . B 9 600 3.9823 B -A 1.0792 1.0792/0.9241 = 1.17
800 2.9031 1.4334 1.4334/1.4699 = 0.98
ic 21 700 4.3365 C -A
f°B 23 700 4.3747 C -B 0.3102 0.3102/0.5458 = 0.57

69 000 . . . 11 600 4.0645 B -A 0.6173 0.6173/0.9241 = 0.67
h 2 800 3.4472 C -A 0.9275 0.9275/1.4699 = 0.63
1C 23 700 4.3747
f°B 25 700 4.4099 C -B 0.2764 0.2764/0.5458 = 0.506
67 000. . . 13 600 4.1335 B -A 0.4523 0.4523/0.9241 = 0.49
h 4 800 3.6812
c 25 700 4.4099 C -A 0.7287 0.7287/1.4699 = 0.496
The cut-and-try method depends upon the fact that Prot's equation can
be written thus:
log (S-E) = log K + n log a
This plots as a straight line through points C, B, and A, for the correct
value of E, with slope n. The procedure is to assume values of E, as shown
in Table 24. The slopes of lines C-B, B-A, and C-A are computed. Since
C, B, and A, for the correct value of E in the logarithmic equation, all lie
on a straight line, the slopes of the three lines must come out the same
when the correct value of E has been determined. 25 Table 24 is a suggested
form for systematizing the trial computations.
A value for E close to but a little less than the value of the mean failure
stress for A must first be assumed. The quantities indicated in Table 24
are then computed, using values of S and A log a given in Table 23. If the
(a) Incorrect assumed value of fatigue limit.

(b) Correct assumed value of fatigue limit.
FIG. 10.—Log-Log Plot of Prot Data.
slopes are not within 1 or 2 per cent of one another (see Fig. 10), another
value of E must be assumed and the calculation repeated. The value of E
that makes the slopes sufficiently close to one another so that the three
points lie on the same straight line is the estimate of the fatigue limit as*
given by the Prot method.
25
Although theoretically only three groups of specimens need be used, in some cases
more groups with fewer specimens per group may be advisable. The "cut-and-try" pro-
cedure described for three groups can easily be adapted to four or more groups. Some ma-
terials have been found—for example, some case-hardened steels—for which the Prot
method gave no solution. Whether or not this means that they have no true fatigue limit
is not known.
APPENDIX I
MISCELLANEOUS REFERENCE TABLES

hhhhh
Minimum Per Cent of Popu-
Number Confi- Lowest Ranking Points Second Ranking Points
of Points dence
in Each Level,
Group, n per cent Number of Groups, m Number of Groups, m
1 3 5 7 9 15 25 1 3 5 7 9 15 25
50 50.00 50.00 50.00 50.00 50.00 50.00 50.00

75 25.00 32.64 35.94 37.88 39.20 41.50 43.35
90 10.00 19.58 24.66 27.86 30.10 34.15 37.51
1 95 5.00 13.53 18.94 22.53 25.14 30.00 34.14
98 2.00 8.40 13.52 17.27 20.10 25.61 30.48
99 1.00 5.89 10.56 14.23 17.10 23.14 28.14
99.9 0.10 1.84 4.76 7.67 10.25 16.12 22.06
50 79.37 79.37 79.37 79.37 79.37 79.37 79.37 50.00 50.00 50.00 50.00 50.00 50.00 50.00
75 63.00 68.84 71.10 72.35 73.18 74.58 75.68 32.64 38.20 40.51 41.84 42.74 44.30 45.55
90 46.42 58.06 62.71 65.31 67.01 69.89 72.12 19.58 28.37 32.38 34.76 36.39 39.26 41.59
3 95 36.84 51.34 57.41 60.85 63.11 66.94 69.89 13.53 23.09 27.82 30.73 32.73 36.32 39.26
98 27.14 43.80 51.33 55.68 58.57 63.50 67.29 8.40 17.83 23.08 26.43 28.79 33.09 36.67
99 21.54 38.90 47.27 52.20 55.50 61.15 65.53 5.89 14.75 20.16 23.73 26.28 31.00 34.97
99.9 10.00 26.38 36.22 42.47 46.80 54.41 60.42 1.84 8.04 13.18 16.97 19.84 25.43 30.36
50 84.09 84.09 84.09 84.09 84.09 84.09 84.09 61.42 61.42 61.42 61.42 61.42 61.42 61.42
75 70.71 75.58 77.42 78.45 79.12 80.26 81.14 45.63 50.91 53.04 54.25 55.06 56.45 57.55
90 56.23 66.52 70.47 72.65 74.06 76.44 78.26 32.05 41.40 45.38 47.68 49.22 51.90 54.02
4 95 47.29 60.65 65.95 68.89 70.80 74.00 76.43 24.86 35.91 40.85 43.77 45.73 49.16 51.88
98 37.61 53.84 60.64 64.46 66.95 71.13 74.30 17.94 30.04 35.90 39.43 41.83 46.07 49.48
99 31.62 49.26 57.01 61.41 64.30 69.15 72.83 14.09 26.38 32.71 36.59 39.27 44.03 47.87
99.9 17.78 36.81 46.69 52.61 56.58 63.36 68.53 6.40 17.41 24.41 29.04 32.34 38.38 43.40
50 87.05 87.05 87.05 87.05 87.05 87.05 87.05 68.61 68.61 68.61 68.61 68.61 68.61 68.61
75 75.79 79.93 81.49 82.35 82.91 83.87 84.60 54.58 59.38 61.28 62.35 63.07 64.29 65.26
90 63.10 72.17 75.58 77.44 78.65 80.66 82.19 41.61 50.65 54.35 56.46 57.85 60.26 62.15
5 95 54.93 67.03 71.68 74.22 75.86 78.59 80.65 34.26 45.40 50.13 52.85 54.67 57,79 60.25
98 45.74 60.93 67.02 70.38 72.55 76.15 78.85 26.71 39.60 45.40 48.78 51.05 54.98 58.09
99 39.81 56.75 63.79 67.70 70.23 74.44 77.60 22.21 35.84 42.26 46.07 48.64 53.10 56.63
99.9 25.12 44.95 54.37 59.82 63.41 69.41 73.91 12.20 26.10 33.78 38.58 41.90 47.79 52.52
50 89.09 89.09 89.09 89.09 89.09 89.09 89.09 73.55 73.55 73.55 73.55 73.55 73.55 73.55
75 79.37 82.97 84.32 85.06 85.54 86.36 86.99 61.05 65.38 67.08 68.03 68.67 69.76 70.61
90 68.13 76.20 79.19 80.81 81.86 83.60 84.92 48.97 57.45 60.84 62.75 64.01 66.17 67.86
6 95 60.70 71.65 75.77 78.00 79.44 81.81 83.60 41.82 52.56 56.97 59.48 61.13 63.96 66.16
98 52.10 66.18 71.64 74.62 76.53 79.68 82.03 34.17 47.04 52.56 55.72 57.82 61.42 64.23
99 46.42 62.37 68.75 72.25 74.49 78.20 80.95 29.43 43.38 49.59 53.19 55.58 59.70 62.91
99.9 31.62 51.36 60.19 65.17 68.41 73.76 77.73 18.14 33.54 41.35 46.06 49.25 54.80 59.17
50 93.30 93.30 93.30 93.30 93.30 93.30 93.30 83.77 83.77 83.77 83.77 83.77 83.77 83.77
75 87.06 89.40 90.27 90.74 91.05 91.58 91.98 75.26 78.28 79.45 80.09 80.52 81.25 81.82
90 79.43 84.95 86.93 88.00 88.68 89.81 90.66 66.31 72.68 75.11 76.46 77.34 78.82 79.97
10 95 74.11 81.87 84.66 86.15 87.10 88.65 89.81 60.59 69.07 72.33 74.14 75.32 77.30 78.82
98 67.62 78.06 81.86 83.89 85.17 87.26 88.79 53.98 64.81 69.06 71.42 72.95 75.52 77.48
99 63.10 75.33 79.86 82.28 83.80 86.28 88.09 49.64 61.87 66.79 69.54 71.31 74.30 76.57
99.9 50.12 67.05 73.74 77.34 79.63 83.31 7 37.63 53.40 60.19 64.02 66.53 70.73 73.92
50 96.59 96.59 96.59 96.59 96.59 96.59 96.59 91.75 91.75 91.75 91.75 91.75 91.75 91 75
75 93.30 94.55 95.01 95.26 95.43 95.69 95.90 87.10 88.77 89.41 89.77 90.00 90.40 90.70
90 89.12 92.17 93.24 93.81 94.17 94.76 95.21 81.90 85.63 87.01 87.76 88.25 89.07 89.70
20 95 86.09 90.48 92.01 92.81 93.32 94.15 94.76 78.39 83.53 85.43 86.46 87.13 88.23 89.07
98 82.23 88.35 90.48 91.59 92.29 93.41 94.23 74.12 80.99 83.53 84.90 85.78 87.24 88.33
99 79.43 86.79 89.36 90.71 91.54 92.88 93.85 71.12 79.19 82.19 83.81 84.84 86.55 87.83
99.9 70.79 81.88 85.87 87.94 89.23 91.27 92.72 62.24 73.74 78.14 80.52 82.03 84.51 86.33
0 m Schuette (27).
n evaluating fatigue curves faired through test data, subtract three from the number of
groups before entering table.
56
T hhhhhhhhh
lation Exceeding Median of:
Third Ranking Points Fourth Ranking Points
Number of Groups, m Number of Groups, m
1 3 5 7 9 15 25 1 3 5 7 9 15 25
20.63 20.63 20.63 20.63 20.63 20.63 20.63

9.14 12.33 13.79 14.67 15.28 16.36 17.25
3.45 7.00 9.00 10.31 11.25 13.00 14.50
1.70 4.73 6.75 8.15 9.20 11.20 12.99
0.67 2.88 4.72 6.12 7.20 9.39 11.41
0.33 2.00 3.65 4.98 6.05 8.29 10.42
0.03 0.61 1.60 2.62 3.54 5.69 7.97
38.57 38 . 57 38.57 38.57 38.57 38.57 38.57 15.91 15.91 15.91 15.91 15.91 15.91 15.91
24.30 28. 7& 30.63 31.72 32.46 33.76 34.80 6.94 9.40 10.53 11.22 11.69 12.54 13.24
14.22 20.96 24" . 09 25. 99" 27.29 29.61 31.51 2.60 5.30 6.83 7.83 8.56 9.92 11.09
•9.76 16.90 20.54 22.80 24.38 27.23 29.60 1.30 3.56 5.11 6.18 6.97 8.52 9.91
6.01 12.94 16.90 19.46 21.29 24.66 27.51 0.51 2.16 3.56 4.62 5.45 7.1 2 8.68
4.18 10.65 14.70 17.39 19.35 23.01 26.1 5 0.25 1.50 2.74 3.76 4.58 6.28 7.93
1.31 5.75 9.50 12.30 14.45 18.69 22.51 0.03 0.46 1.21 1.97 2.67 4.29 6.03
50.00 50.00 50.00 50.00 50.00 50.00 50.00 31.38 31.38 31.38 31.38 31.38 31.38 31.38
35.94 40.51 42.38 43.46 44.18 45.44 46.44 19.38 23.06 24.63 25.55 26.17 27.26 28.15
24.66 32.36 35.73 37.70 39.04 41.37 43.25 11.21 16.64 19.21 20.77 21.84 23.78 25.37
18.93 27.81 31.91 34.35 36.03 38.98 41.37 7.64 13.35 16.30 18.14 19.44 21.80 23.77
13.53 23.04 27.80 30.71 32.73 36.32 39.26 4.69 10.17 13.35 15.42 16.90 19.67 22.03
10.56 20.12 25.19 28.37 30.58 34.58 37.87 3.27 8.35 11.58 13.75 15.33 18.31 20.90
4.76 13.11 18.57 22.25 24.90 29.85 34.05 1.02 4.48 7.43 9.66 11.38 14.80 17.91
57.85 57.85 57.85 57.85 57.85 57.85 57.85 42.14 42.14 42.14 42. 14 42.14 42.14 42.14
44.68 49.06 50.82 51.84 52.51 53.67 54.60 29.71 33.67 35.32 36.27 36.91
38.03 38.93
33.32 41.17 44.47 46.38 47.65 49.88 51.65 20.09 26.61 29.50 31.22 32.38 34.43 36.09
27.13 36.58 40.71 43.13 44.76 47.60 49.87 15.32 22.73 26.22 28.32 29.76 32.33 34.42
20.93 31.61 36.57 39.52 41.53 45.04 47.87 10.88 18.74 22.73 25.20 26.92 30.02 32.57
17.31 28.45 33.88 37.15 39.39 43.35 46.54 8.47 16.31 20.54 23.21 25.09 28.51 31.36
9.40 20.44 26.74 30.75 33.57 38.65 42.83 3.79 10.54 15.02 18.07 20.29 24.47 28.06
74.14 74.14 74.1 4 74.14 74.14 74.14 74.14 64.49 64.49 64.49 64.49 64.49 64.49 64.49
64.46 67.80 69.11 69.85 70.33 71.18 71.84 54.23 57.69 59.07 59.86 60.38 61.28 61.99
55.04 61.67 64.29 65.76 66.74 68.41 69.71 44.83 51.40 54.06 55.58 56.59 58.34 59.71
49.31 57.87 61.30 63.24 64.52 66.70 68.40 39.33 47.60 51.02 52.98 54.29 56.55 58.33
42.95 53.51 57.86 60.33 61.95 64.74 66.90 33.43 43.34 47.59 50.05 51.68 54.52 56.76
38.82 50.58 55.53 58.35 60.21 63.41 65.89 29.71 40.54 45.30 48.08 49.94 53.16 55.71
28.15 42.41 48.92 52.73 55.26 59.61 63.0 20.46 32.94 38.98 42.59 45.04 49.33 52.74
86.84 86.84 86.84 86.84 86.84 86.84 86.84 81.94 81.94 81.94 81.94 81.94 81.94 81.94
81.33 83.27 84.03 84.45 84.73 85.20 85.58 75.79 77.93 78.76 79.23 79.55 80.08 80.50
75.52 79.66 81.22 82.10 82.66 83.62 84.37 69.58 73.98 75.68 76.63 77.26 78.32 79.14
71.74 77.31 79.43 80.60 81.36 82.64 83.62 65.63 71.48 73.74 75.00 75.83 77.23 78.31
67.26 74.53 77.31 78.84 79.83 81.49 82.76 61.04 68.54 71.47 73.10 74.17 75.97 77.36
64.17 72.59 75.83 77.62 78.77 80.70 82.17 57.93 66.52 69.91 71.80 73.03 75.11 76.71
55.26 66.86 71.48 74.02 75.66 78.40 80.45 49.13 60.64 65.36 68.00 69.73 72.63 74.85
57
TABLE 26.—UNPAIRED
Critical lower and upper rank totals for the 5 per cent and 1 per cent levels of significance. Values
h
4 5 6 7 8 9 10 11 12 13
Ni
4 0.05 11 25
0.01
5 0.05 12 28 18 37
0.01 15 40
6 0.05 13 31 19 41 26 52
0.01 10 34 16 44 23 55
7. 0.05 13 35 20 45 28 56 37 68
0.01 11 37 17 48 24 60 33 72
8. 0.05 14 38 21 49 29 61 39 73 49 87
0.01 11 41 18 52 25 65 34 78 44 92
9. 0.05 15 41 22 53 31 65 41 78 51 93 63 108
0.01 12 44 19 56 27 69 36 83 46 98 57 114
10 0.05 16 44 24 56 33 69 43 83 54 98 66 114 79 131
0.01 12 48 19 61 28 74 37 89 47 105 59 121 71 189
11. 0.05 16 48 25 60 34 74 44 89 56 104 68 121 82 138 96 157
0.01 13 51 20 65 29 79 38 95 49 111 61 128 74 146 88 165
12. 0.05 17 51 26 64 36 78 46 94 58 110 71 127 85 145 100 164 116 184
0.01 13 55 21 69 30 84 40 100 51 117 63 135 76 154 90 174 106 194
13. 0.05 18 54 27 68 37 83 48 99 61 115 74 133 88 152 104 171 120 192 137 214
0.01 14 58 22 73 31 89 41 106 53 123 65 142 79 161 93 182 109 203 126 226
14. 0.05 19 57 28 72 39 87 50 104 63 121 77 139 91 159 107 179 124 200 142 222
0.01 14 62 23 77 32 94 43 111 55 129 67 149 81 169 96 190 112 212 129 235
15 0.05 30 75 41 91 52 109 65 127 79 146 95 165 111 186 128 208 146 231
0.01 23 82 33 99 44 117 56 136 70 155 84 176 99 198 116 220 133 244
16 0.05 42 96 54 114 68 132 82 152 98 172 114 194 132 216 150 240
0.01 35 103 46 122 58 142 72 162 87 183 102 206 119 229 137 253
17. 0.05 56 119 70 138 85 158 101 179 118 201 136 224 155 248
0.01 47 128 60 148 74 169 89 191 105 214 122 238 141 262
18 0.05 73 143 88 164 104 186 121 209 140 232 159 257
0.01 62 154 76 179 92 198 108 222 126 246 144 272
19 0.05 91 170 107 193 125 216 144 240 163 266
0.01 79 182 94 206 111 230 129 255 147 282
20 0.05 111 199 129 223 148 248 168 274
0.01 97 213 114 238 133 263 151 291
21. 0.05 132 231 152 256 172 283
0.01 117 246 136 272 155 300
22. 0.05 156 264 177 291
0.01 139 281 158 310
23. 0.05 181 300
0.01 162 319
24. 0.05
0.01
25 0.05
0.01
26 0.05
0.01
27 0.05
0.01
28. 0.05
0.01
29. 0.05
0.01
30. 0.05
0.01
"Adapted from work of Dr. Frank Wilcoxon (28).

MISCELLANEOUS REFERENCE TABLES 59
RANK TEST. 0
in the body of the table refer to the group with the smaller number of measurements, NI.
14 15 16 17 18 19 20 21 22 23 24 25
160 246
147 259
165 255 185 280
151 269 171 294
170 264 190 290 212 316
156 278 175 305 196 332
175 273 196 299 218 326 241 354
159 289 179 316 200 344 223 372
179 283 201 309 223 337 247 365 271 395
163 299 184 326 205 355 228 384 251 415
184 292 206 319 229 347 253 376 278 407 303 438
167 309 188 337 210 366 233 396 257 427 282 459
189 301 211 329 234 358 259 387 284 418 310 450 338 482
171 319 193 347 215 377 238 408 263 439 288 472 315 505
194 310 216 339 240 368 265 398 290 430 317 462 345 495 374 529
175 329 197 358 220 388 244 419 268 452 294 485 321 519 349 554
199 319 222 348 246 378 271 409 297 441 324 474 352 508 381 543 412 579
179 339 202 368 225 399 249 431 274 464 300 498 328 532 356 568 385 605
203 329 227 358 251 389 277 420 303 453 331 486 360 521 389 556 420 592 451 630
184 348 206 379 230 410 254 443 280 476 306 511 334 546 363 582 392 620 423 658
208 338 232 368 257 399 283 431 310 464 338 498 367 533 397 569 428 606 460 644 493 683
188 358 211 389 235 421 260 455 286 489 313 524 341 559 370 596 400 634 431 673 463 713
237 378 263 409 289 442 316 476 345 510 374 546 405 582 436 620 469 658 502 698 536 739
215 400 239 433 265 466 291 501 319 536 347 573 377 611 407 649 438 689 471 729 505 771
268 420 295 453 323 487 352 522 382 558 412 596 444 634 477 673 511 713 546 754
244 444 270 478 297 513 325 549 354 586 383 625 414 664 446 704 479 745 513 787
301 464 329 499 359 534 389 571 420 609 452 648 486 687 520 728 555 770
275 490 303 525 331 562 360 600 390 639 422 678 454 719 487 761 522 803
336 510 366 546 396 584 428 622 461 661 494 702 529 743 565 785
308 538 337 575 367 613 397 653 429 693 462 734 495 777 530 820
373 558 404 596 436 635 469 675 503 716 538 758 575 800
343 588 373 627 404 667 436 708 469 750 504 792 539 836
411 609 444 648 477 689 512 730 547 773 584 816
380 640 411 681 444 722 477 765 512 808 547 852
TABLE 27.—PERCENTILES OF THE x2 DISTRIBUTION."

Percentiles
Degrees 6of
freedom
0.5 1 2.5 5 10 90 95 97.5 99 99.5
1 0.000039 0.00016 0.00098 0.0039 0.0158 2.71 3.84 5.02 6.63 7.88
2 0.0100 0.0201 0.0506 0.1026 0.2107 4.61 5.99 7.38 9.21 10.60
3 0.0717 0.115 0.216 0.352 0.584 6.25 7.81 9.35 11.34 12.84
4 0.207 0.297 0.484 0.711 1.064 7.78 9.49 11.14 13.28 14.86
5 0.412 0.554 0.831 1.15 1.61 9.24 11.07 12.83 15.09 16.75
6 0.676 0.872 1.24 1.64 2.20 10.64 12.59 14.45 16.81 18.55
7 0.989 1.24 1.69 2.17 2.83 12.02 14.07 16.01 18.48 20.28
g 1.34 1.65 2.18 2.73 3.49 13.36 15.51 17.53 20.09 21.96
9 1.73 2.09 2.70 3.33 4.17 14.68 16.92 19.02 21.67 23.59
10 2.16 2.56 3.25 3.94 4.87 15.99 18.31 20.48 23.21 25.19
11 2.60 3.05 3.82 4.57 5.58 17.28 19.68 21.92 24.73 26.76
12 3.07 3.57 4.40 5.23 6.30 18.55 21.03 23.34 26.22 28.30
13 3.57 4.11 5.01 5.89 7.04 19.81 22.36 24.74 27.69 29.82
14 4.07 4.66 5.63 6.57 7.79 21.06 23.68 26.12 29.14 31.32
15 4.60 5.23 6.26 7.26 8.55 22.31 25.00 27.49 30.58 32.80
16 5.14 5.81 6.91 7.96 9.31 23.54 26.30 28.85 32.00 34.27
18 6.26 7.01 8.23 9.39 10.86 25.99 28.87 31.53 34.81 37.16
20 7.43 8.26 9.59 10.85 12.44 28.41 31.41 34.17 37.57 40.00
24 9.89 10.86 12.40 13.85 15.66 33.20 36.42 39.36 42.98 •45.56
30 13.79 14.95 16.79 18.49 20.60 40.26 43.77 46.98 50.89 53.67
40 20.71 22.16 24.43 26.51 29.05 51.81 55.76 59.34 63.69 66.77
60 35.53 37.48 40.48 43.19 46.46 74.40 79.08 83.30 88.38 91.95
120 83.85 86.92 91.58 95.70 100.62 140.23 146.57 152.21 158.95 163.64
a
From Dixon and Massey (9) by permission of the publishers.
6 For large values of degrees of freedom the approximate formula may be used:
X«2 = Yz\. Za + (2n - l)i/»p
where Za is the normal deviate and n is the number of degrees of freedom. For example, hhhh
Yl (2.326 + 10.909)2 = 87.6 for the 99th percentile, for 60 d.f.
hhh 61
TABLE 28.—AREAS OF THE "NORMAL" CURVE."

z X Area z X Area
-3.0 — 3. 0 ff 0.0013 0.1 At + 0.1 ff 0.5398
-2.9 — 2 9 ff 0.0019 0.2 At + 0.2 ff 0.5793
— 2.8 — 2 8 <r 0.0026 0.3 At + 0.3 ff 0.6179
-2.7. — 2 7 ff 0.0035 0.4 At + 0.4 ff 0.6554
— 2.6 — 2 6 ff 0.0047 0.5 At + 0.5 ff 0.6915
-2.5 n - 2. 5 0.0062 0.6 At + 0.6 ff 0.7257
-2.4 h — 2.4 ff 0.0082 0.7 At + 0.7 ff 0.7580
-2.3 h — 2 3 ff 0.0107 0.8 At + 0.8 ff 0.7881
-2.2 h — 2.2 ff 0.0139 0.9 At + 0.9 ff 0.8159
— 2.1 h — 2 1 <r 0.0179 1.0 At + 1.0 ff 0.8413
—2. 0 h — 2 0 «r 0.0228 1.1 At + 1.1 ff 0.8643
— 1 .9 h — 1 9 <r 0.0287 1.2 At + 1.2 ff 0.8849
— 1.8 h — 1 8 ff 0 0359 1.3 At + 1.3 ff 0.9032
— 1.7 h — 1 7 ff 0 0446 1.4 H + 1 4 ff 0.9192
— 1.6 h — 1 6 ff 0 0548 15 At + 1 5 ff 0.9332
-1 .5 u - 1 .5 0.0668 1.6 At + 1.6 ff 0.9452
— 1 .4 u — 1.4 ff 0.0808 1.7 At + 1.7 ff 6.9554
— 1.3 h — 1.3 ff 0.0968 1.8 /t + 1.8 f 0.9641
— 1.2 h — 1 2 ff 0.1151 1.9 At + 1.9 ff 0.9713
— 1.1 M - 1 .1 0.1357 2.0 M + 2.0 ff 0.9772
— 1.0 A* — 1.0 ff 0.1587 2.1 At + 2.1 ff 0.9821
— 0.9 h — 0.9 ff 0.1841 2.2 At + 2.2 ff 0.9861
— 0.8 h — 0.8 ff 0.2119 2.3 At + 2.3 ff 0.9893
— 0.7 hh — 0 7 ff 0.2420 2.4 At + 2.4 ff 0.9918
— 0.6 h — 0 6 ff 0.2743 2.5 At + 2.5 ff 0.9938
— 0.5 h — 0 5 ff 0.3085 2.6 At + 2.6 ff 0.9953
— 0.4 h — 0 4 ff 0.3446 2.7 At + 2.7 ff 0.9965
— 0.3 /* — 0.3 ff 0.3821 2.8 At + 2.8 ff 0.9974
— 0.2 h — 0.2 ff 0.4207 2.9 At + 2.9 ff 0.9981
— 0.1 . h — 0.1 ff 0.4602 3.0 At + 3.0 ff 0.9987
0 h
h 0.5000
-3.090 A* - 3.090 0.001 + 3.090 At + 3.090 0.999
-2.576 M - 2.576 0.005 + 2.576 At + 2.576 0.995
-2.326 M - 2.326 0.010 + 2.326 At + 2.326 0.990
-1 .960 A* - 1 .960 0.025 + 1.960 At + 1 .960 0.975
-1.645 M - 1 .645 0.050 + 1 .645 At + 1 .645 0.950
-1 .282 M - 1 .282 0.100 + 1 .282 At + 1 .282 0.900
-1.036 A* - 1 .036 0.150 + 1 .03 At + 1.036 0.850
-0.842 At - 0.842 0.200 + 0.842 At + 0.842 0.800
-0.674 M - 0.674 0.250 + 0.674. At + 0.674 0.750
-0.524 A* - 0.524 0.300 +0.524 At + 0.524 0.700
-0.385 At - 0.385 0.350 0.385. At + 0.385 0.650
-0.253 At - 0.253 0.400 + 0.253. At + 0.253 0.600
-0.1 26 M - 0.1 26 0.450 + 0.126. At + 0.1 26 0.550
0 n 0.500
From Dixon and Massey (9) by permission of the publishers.
TABLE 29.—VALUES 0 - 5 OF t.
Degrees of Freedom to. 9i to .976 /0.9875 to. nt ^0.9975
1 6.31 12.7 25.5 63.7 127

2 2.92 4.30 6.21 9.92 14.1
3 2.35 3.18 4.18 5.84 7.45
4 2.13 2.78 3.50 4.60 5.60
5 2.01 2.57 3.16 4.03 4.77
6 1.94 2.45 2.97 3.71' 4.32
7 1.89 2.36 2.84 3.50 4.03
8 1.86 2.31 2.75 3.36 3.83
9 1.83 2.26 2.69 3.25 3.69
10 1.81 2.23 2.63 3.17 3.58
11 1.80 2.20 2.59 3.11 3.50
12 1.78 2.18 2.56 3.05 3.43
13 1.77 2.16 2.53 3.01 3.37
14 1.76 2.14 2.51 2.98 3.33
15 1.75 2.13 2.49 2.95 3.29
16 1.75 2.12 2.47 2.92 3.25
17 1.74 2.11 2.46 2.90 3.22
18 1.73 2.10 2.45 2.88 3.20
19 1.73 2.09 2.43 2.86 3.17
20 1.72 2.09 2.42 2.85 3.15
21 1.72 2.08 2.41 2.83 3.14
22 1.72 2.07 2.41 2.82 3.12
23 1.71 2.07 2.40 2.81 3.10
24 1.71 2.06 2.39 2.80 3.09
25 1.71 2.06 2.38 2.79 3.08
26 1.71 2.06 2.38 2.78 3.07
27 1.70 2.05 2.37 2.77 3.06
28 1.70 2.05 2.37 2.76 3.05
29 1.70 2.05 2.36 2.76 3.04
30 1.70 2.04 2.36 2.75 3.03
40 1.68 2.02 2.33 2.70 2.97
60 1.67 2.00 2.30 2.66 2.91
120 1.66 1.98 2.27 2.62 2.86
00 1.64 1.96 2.24 2.58 2.81
Degrees of Freedom 'O.OB <0.026 to. oit* /0.006 /0.0025
When the table is read from the foot, the tabled values are to be prefixed with a
negative sign. Interpolation should be performed using the reciprocals of the degrees of
freedom.
0 From Dixon and Massey (9) by n of the publisher.
6 The values in this table were computed from percentiles of the distribution. See
F
Pearson and Hartley (2).
MISCELLANEOUS REFERENCE TABLES 63
TABLE 30.—PERCENTILES OF THE xVd.f. DISTRIBUTION. 0 - 6

h
Degrees of
Freedom
0.5 1 2.5 5 10 90 95 97.5 99 99.5
1 0.000039 0.00016 0.00098 0.0039 0.0158 2.71 3.84 5.02 6.63 7.88
2 0.00501 0.0101 0.0253 0.0513 0.1054 2.30 3.00 3.69 4.61 5.30
3 0.0239 0.0383 0.0719 0.117 0.195 2.08 2.60 3-12 3.78 4.28
4 0.0517 0.0743 0.121 0.178 0.266 1.94 2.37 2.79 3.32 3.72
5 0.0823 0.111 0.166 0.229 0.322 1.85 2.21 2.57 3.02 3.35
6 0.113 0.145 0.206 0.273 0.367 1.77 2.10 2.41 2.80 3.09
7 0.141 0.177 0.241 0.310 0.405 1.72 2.01 2.29 2.64 2.90
8 0.168 0.206 0.272 0.342 0.436 1.67 1.94 2.19 2.51 2.74
9 0.193 0.232 0.300 0.369 0.463 1.63 1.88 2.11 2.41 2.62
10 0.216 0.256 0.325 0.394 0.487 1.60 1.83 2.05 2.32 2.52
11 0.237 0.278 0.347 0.416 0.507 1.57 1.79 1.99 2.25 2.43
12 0.256 0.298 0.367 0.436 0.525 1.55 1.75 1.94 2.18 2.36
13 0.274 0.316 0.385 0.453 0.542 1.52 1.72 1.90 2.13 2.29
14 0.291 0.333 0.402 0.469 0.556 1.50 1.69 1.87 2.08 2.24
15 0.307 0.349 0.417 0.484 0.570 1.49 1.67 1.83 2.04 2.19
16 0.321 0.363 0.432 0.498 0.582 1.47 1.64 1.80 2.00 2.14
18 0.348 0.390 0.457 0.522 0.604 1.44 1.60 1.75 1.93 2.06
20 0.372 0.413 0.480 0.543 0.622 1.42 1.57 1.71 1.88 2.00
24 0.412 0.452 0.517 0.577 0.652 1.38 1.52 1.64 1.79 1.90
30 0.460 0.498 0.560 0.616 0.687 1.34 1.46 1.57 1.70 1.79
40 0.518 0.554 0.611 0.663 0.726 1.30 1.39 1.48 1.59 1.67
60 0.592 0.625 0.675 0.720 0.774 1.24 1.32 1.39 1.47 1.53
120 0.699 0.724 0.763 0.798 0.839 1.17 1.22 1.27 1.32 1.36
00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
0Interpolation should be performed using reciprocals of .the degrees of freedom.
6 From Dixon and Massey (9) by permission of the publisher.
The values in the table were computed from percentiles of the F distribution. See
Pearson and Hartley (2).
TABLE 31°.-«o.os6 AND «o.975 FOR RUNS AMONG ELEMENTS IN SAMPLES OF SIZES Ni AND
K0.02B MO, 975
Nl 2 3 4 5 6 7 8 9 10 11 12 13 14 IS 16 17 18 19 20 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
tf
2 4
3 5 6
4 5 7 8
5 2 2 5 7 g 9
6 2 2 3 3 S 7 g 9 10
7 2 2 3 3 3 s 7 9 10 11 1?
g 2 3 3 3 4 4 s 7 9 10 11 1? 13
9 2 3 3 4 4 5 5 5 7 9 11 12 13 13 14
10 2 3 3 4 s 5 5 6 5 7 9 11 12 13 14 15 15
11 2 3 4 4 5 5 6 6 7 5 7 9 11 12 H 14 15 16 16
12 2 2 3 4 4 s 6 6 7 7 7 S 7 9 11 12 1S 15 15 16 17 18
13 2 2 3 4 5 5 6 6 7 7 g g S 7 9 11 H 14 IS 16 17 18 18 19
14 2 2 3 4 s 5 6 7 7 g g 9 9 5 7 9 11 13 14 15 16 17 18 19 19 20
15 2 3 3 4 5 6 6 7 7 g g 9 9 10 S 7 9 11 13 14 15 17 17 18 19 20 21 21
16 2 3 4 4 5 6 6 7 8 8 9 9 10 10 11 S 7 9 11 13 15 16 17 18 19 20 20 21 22 22
17 2 3 4 4 5 6 7 7 g 9 9 10 10 11 11 ii s 7 9 11 13 IS 16 17 18 19 20 21 22 22 23 74
18 2 3 4 5 5 6 7 g g 9 9 10 10 11 11 12 12 s 7 9 11 13 15 16 17 18 19 20 21 22 23 24 24 25
19 2 3 4 s 6 6 7 g g 9 10 10 11 11 12 12 13 H 5 7 9 11 13 15 16 17 19 20 21 22 22 23 24 25 25 76
20 2 3 4 s 6 6 7 8 9 9 10 10 11 12 12 n n n 14 S 7 9 11 H IS 16 17 19 20 71 22 23 24 ?4 7S 26 76 77
The values listed are such that a number less than or equal to the Wo.o26
Nl = #2 «0 .025 MO .976 Ni = Nz Mo .025 «0. 975
value will occur not more than 2.5 per cent of the time and a number greater
20 14 31 50 than Mo.975 will occur not more than 2.5 per cent of the time.
21 15
27
28
40
42 33 52 For values of Ni and -/V2 larger than 20, a Normal approximation may be
22
23
16
16
29
31
44
46
35
37
54
56 used- The mean is
24 17 32 48 38 59
25 18 33 50 40 61
26 19 34 55 45 66
27 20 35 60 49 72
28
29
21
22
36
37
65
70
54
58
77
83
and the variance is
30 22 39 75 63 88
32 24 41 80 68 93
34 26 43 85 72 99
36 28 45 77 104
38.. 30 47
90
95 82 109 . For example, for Ni = N2 - 20, the mean is 21 and the variance is 9.74.
100 86 115
The 2.5 1and 97.5 percentiles are 21 + 1.96 (9.74)1 /2 = 27.1 and 21 -
0 With permission from Eisenhart and Swed (29). 1.96 (9.74) / = 14.9.
2
TABLE 32.— F DISTRIBUTION. 0
Degrees of Freedom for Numerator
1 2 3 4 5 6 7 8 9 10 12 IS 20 24 30 . 40 60 120 00
Upper 5 per cent Points (^0.95) •

1 161 200 216 225 230 234 237 239 241 242 244 246 248 249 250 251 252 253 254
2 18.5 19.0 19.2 19.2 19.3 19.3 19.4 19.4 19.4 19.4 19.4 19.4 19.4 19.5 19.5 19.5 19.5 19.5 19.5
3 10.1 9.55 9.28 9.12 9.01 8.94 8.89 8.85 8.81 8.79 8.74 8.70 8.66 8.64 8.62 8.59 8.57 8.55 8.53
4 7.71 6.94 6.59 6.39 6.26 6.16 6.09 6.04 6.00 5.96 5.91 5.86 5.80 5.77 5.75 5.72 5.69 5.66 5.63
5 6.61 5.79 5.41 5.19 5.05 4.95 4.88 4.82 4.77 4.74 4.68 4.62 4.56 4.53 4.50 4.46 4.43 4.40 4.36
6 5.99 5.14 4.76 4.53 4.39 4.28 4.21 4.15 4.10 4.06 4.00 3.94 3.87 3.84 3.81 3.77 3.74 3.70 3.67
7 5.59 4.74 4.35 4.12 3.97 3.87 3.79 3.73 3.68 3.64 3.57 3.51 3.44 3.41 3.38 3.34 3.30 3.27 3.23
8 5.32 4.46 4.07 3.84 3.69 3.58 3.50 3.44 3.39 3.35 3.28 3.22 3.15 3.12 3.08 3.04 3.01 2.97 2.93
9 5.12 4.26 3.86 3.63 3.48 3.37 3.29 3.23 3.18 3.14 3.07 3.01 2.94 2.90 2.86 2.83 2.79 2.75 2.71
10 4.96 4.10 3.71 3.48 3.33 3.22 3.14 3.07 3.02 2.98 2.91 2.85 2.77 2.74 2.70 2.66 2.62 2.58 2.54
11 4.84 3.98 3.59 3.36 3.20 3.09 3.01 2.95 2.90 2.85 2.79 2.72 2.65 2.61 2.57 2.53 2.49 2.45 2.40
12 4.75 3.89 3.49 3.26 3.11 3.00 2.91 2.85 2.80 2.75 2.69 2.62 2.54 2.51 2.47 2.43 2.38 2.34 2.30
13 4.67 3.81 3.41 3.18 3.03 2.92 2.83 2.77 2.71 2.67 2.60 2.53 2.46 2.42 2.38 2.34 2.30 2.25 2.21
14 4.60 3.74 3.34 3.11 2.96 2.85 2.76 2.70 2.65 2.60 2.53 2.46 2.39 2.35 2.31 2.27 2.22 2.18 2.13
15 4.54 3.68 3.29 3.06 2.90 2.79 2.71 2.64 2.59 2.54 2.48 2.40 2.33 2.29 2.25 2.20 2.16 2.11 2.07
16 4.49 3.63 3.24 3.01 2.85 2.74 2.66 2.59 2.54 2.49 2.42 2.35 2.28 2.24 2.19 2.15 2.11 2.06 2.01
17 4.45 3.59 3.20 2.96 2.81 2.70 2.61 2.55 2.49 2.45 2.38 2.31 2.23 2.19 2.15 2.10 2.06 2.01 1.96
18 4.41 3.55 3.16 2.93 2.77 2.66 2.58 2.51 2.46 2.41 2.34 2.27 2.19 2.15 2.11 2.06 2.02 1.97 1.92
19 4.38 3.52 3.13 2.90 2.74 2.63 2.54 2.48 2.42 2.38 2.31 2.23 2.16 2.11 2.07 2.03 1.98 1.93 1.88
20 4.35 3.49 3.10 2.87 2.71 2.60 2.51 2.45 2.39 2.35 2.28 2.20 2.12 2.08 2.04 1.99 1.95 1.90 1.84
21 4.32 3.47 3.07 2.84 2.68 2.57 2.49 2.42 2.37 2.32 2.25 2.18 2.10 2.05 2.01 1.96 1.92 1.87 1.81
22 4.30 3.44 3.05 2.82 2.66 2.55 2.46 2.40 2.34 2.30 2.23 2.15 2.07 2.03 1.98 1.94 1.89 1.84 1.78
23 4.28 3.42 3.03 2.80 2.64 2.53 2.44 2.37 2.32 2.27 2.20 2.13 2.05 2.01 1.96 1.91 1.86 1.81 1.76
24 4.26 3.40 3.01 2.78 2.62 2.51 2.42 2.36 2.30 2.25 2.18 2.11 2.03 1.98 1.94 1.89 1.84 1.79 1.73
25 4.24 3.39 2.99 2.76 2.60 2.49 2.40 2.34 2.28 2.24 2.16 2.09 2.01 1.96 1.92 1.87 1.82 1.77 1.71
30 4.17 3.32 2.92 2.69 2.53 2.42 2.33 2.27 2.21 2.16 2.09 2.01 1.93 1.89 1.84 1.79 1.74 1.68 1.62
40. 4.08 3.23 2.84 2.61 2.45 2.34 2.25 2.18 2.12 2.08 2.00 1.92 1.84 1.79 1.74 1.69 1.64 1.58 1.51
60 4.00 3.15 2.76 2.53 2.37 2.25 2.17 2.10 2.04 1.99 1.92 1.84 1.75 1.70 1.65 1.59 1.53 1.47 1.39
120 3.92 3.07 2.68 2.45 2.29 2.17 2.09 2.02 1.96 1.91 1.83 1.75 1.66 1.61 1.55 1.50 1.43 1.35 1.25
00 3.84 3.00 2.60 2.37 2.21 2.10 2.01 1.94 1.88 1.83 1.75 1.67 1.57 1.52 1.46 1.39 1.32 1.22 1.00
NOTE.—Interpolation T hhhhhhhhhhhhhhhhhhhhhhhhhhh
0 This table is taken, with permission of the a Trustees, from Table 18 of Pearson & Hartley (2). (Continued next page)
TABLE 32— F DISTRIBUTION— Continued
Degrees of Freedom for Numerator
1 2 3 4 s 6 7 8 9 10 12 IS 20 24 30 40 60 120 00
Upper 2.5 per cent Points (^0.975)

1 648 800 864 900 922 937 948 957 963 969 977 985 993 997 1001 1006 1010 1014 1018
2 38.5 39.0 39.2 39.2 39.3 39.3 39.4 39.4 39.4 39.4 39.4 39.4 39.4 39.5 39.5 39.5 39.5 39.5 39.5
3 17.4 16.0 15.4 15:1 14.9 14.7 14.6 14.5 14.5 14.4 14.3 14.3 14.2 14.1 14.1 14.0 14.0 13.9 13.9
4 12.2 10.6 9.98 9.60 9.36 9.20 9.07 8.98 8.90 8.84 8.75 8.66 8.56 8.51 8.46 8.41 8.36 8.31 8.26
5 10.0 8.43 7.76 7.39 7.15 6.98 6.85 6.76 6.68 6.62 6.52 6.43 6.33 6.28 6'. 23 6.18 6.12 6.07 6.02
6 8.81 7.26 6.60 6.23 5.99 5.82 5.70 5.60 5.52 5.46 5.37 5.27 5.17 5.12 5.07 5.01 4.96 4.90 4.85
7 8.07 6.54 5.89 5.52 5.29 5.12 4.99 4.90 4.82 4.76 4.67 4.57 4.47 4.42 4.36 4.31 4.25 4.20 4.14
8 7.57 6.06 5.42 5.05 4.82 4.65 4.53 4.43 4.36 4.30 4.20 4.10 4.00 3.95 3.89 3.84 3.78 3.73 3.67
9 7.21 5.71 5.08 4.72 4.48 4.32 4.20 4.10 4.03 3.96 3.87 3.77 3.67 3.61 3.56 3.51 3.45 3.39 3.33
10 6.94 5.46 4.83 4.47 4.24 4.07 3.95 3.85 3.78 3.72 3.62 3.52 3.42 3.37 3.31 3.26 3.20 3.14 3.08
11 6.72 5.26 4.63 4.28 4.04 3.88 3.76 3.66 3.59 3.53 3.43 3.33 3.23 3.17 3.12 3.06 3.00 2.94 2.8,8
12 6.55 5.10 4.47 4.12 3.89 3.73 3.61 3.51 3.44 3.37 3.28 3.18 3.07 3.02 2.96 2.91 2.85 2.7,9 2.72
13 6.41 4.97 4.35 4.00 3.77 3.60 3.48 3.39 3.31 3.25 3.15 3.05 2.95 2.89 2.84 2.78 2.72 2.66 2.60
14 6.30 4.86 4.24 3.89 3.66 3.50 3.38 3.29 3.21 3.15 3.05 2.95 2.84 2.79 2.73 2.67 2.61 2.55 2.49
15 6.20 4.77 4.15 3.80 3.58 3.41 3.29 3.20 3.12 3.06 2.96 2.86 2.76 2.70 2.64 2.59 2.52 2.46 2.40
16 6.12 4.69 4.08 3.73 3.50 3.34 3.22 3.12 3.05 2.99 2.89 2.79 2.68 2.63 2.57 2.51 2.45 2.38 2.32
17 6.04 4.62 4.01 3.66 3.44 3.28 3.16 3.06 2.98 2.92 2.«2 2.72 2.62 2.56 2.50 2.44 2.38 2.32 2.25
18 5.98 4.56 3.95 3.61 3.38 3.22 3.10 3.01 2.93 2.87 2.77 2.67 2.56 2.50 2.44 2.38 2.32 2.26 2.19
19 5.92 4.51 3.90 3.56 3.33 3.17 3.05 2.96 2.88 2.82 2.72 2.62 2.51 2.45 2.39 2.33 2.27 2.20 2.13
20 5.87 4.46 3.86 3.51 3.29 3.13 3.01 2.91 2.84 2.77 2.68 2.57 2.46 2.41 2.35 2.29 2.22 2.16 2.09
21 5.83 4.42 3.82 3.48 3.25 3.09 2.97 2.87 2.80 2.73 2.64 2.53 2.42 2.37 2.31 2.25 2.18 2.11 2.04
22 5.79 4.38 3.78 3.44 3.22 3.05 2.93 2.84 2.76 2.70 2.60 2.50 2.39 2.33 2.27 2.21 2.14 2.08 2.00
23 5.75 4.35 3.75 3.41 3.18 3.02 2.90 2.81 2.73 2.67 2.57 2.47 2.36 2.30 2.24 2.18 2.11 2.04 1.97
24 5.72 4.32 3.72 3.38 3.15 2.99 2.87 2.78 2.70 2.64 2.54 2.44 2.33 2.27 2.21 2.15 2.08 2.01 1.94
25 5.69 4.29 3.69 3.35 3.13 2.97 2.85 2.75 2.68 2.61 2.51 2.41 2.30 2.24 2.18 2.12 2.05 1.98 1.91
30 5.57 4.18 3.59 3.25 3.03 2.87 2.75 2.65 2.57 2.51 2.41 2.31 2.20 2.14 2.07 2.01 „ 1.94 1.87 1.79
40 5.42 4.05 3.46 3.13 2.90 2.74 2.62 2.53 2.45 2.39 2.29 ' 2 . 1 8 2.07 2.01 1.94 1.88 1.80 1.72 1.64
60 5.29 3.93 3.34 3.01 2.79 2.63 2.51 2.41 2.33 2.27 2.17 ^"2.06 1.94 1.88 1.82 1.74 1.67 1.58 1.48
120 5.15 3.80 3.23 2.89 2.67 2.52 2.39 2.30 2.22 2.16 2.05 1.94 1.82 1.76 1.69 1.61 1.53 ,L-43 1.31
00 5.02 3.69 3.12 2.79 2.57 2.41 2.29 ^.19 2.11 ' 2.05 1.94 1.83 1.71 1.64 1^57 1.48 1.39 1.27 1.00
S
NOTE.— Interpolation should be performe using reciprocals of the degrees of freedom.
TABLE 33.—k FACTORS0 FOR S-N CURVES (NORMAL DISTRIBUTION ASSUMED).
P 75 90 95 99 99.9 75 90 95 99 99.9
n T = 0.50 •y = 0.75
3 0.773 1.498 1.939 2.765 3.688 1.464 2.501 3 152 4.396 5.805
4 0.739 1.419 1.830 2.601 3.464 1.256 2.134 2 680 3.726 4.910
5 0.722 1.382 1.780 2.526 3.362 1.152 1.961 2 463 3.421 4.507
6 0.712 1.360 1.750 2.483 3.304 1.087 1.860 2 336 3.243 4.273
7 0.705 1.346 1.732 2.455 3.265 1.043 1.791 2 250 3.126 4.118
8 0.701 1.337 1.719 2.436 3.239 1.010 1.740 2 190 3.042 4.008
9 0.698 1.329 1.709 2.421 3.220 0.984 1.702 2 141 2.977 3.924
10 0.694 1.324 1.702 2.411 3.205 0.964 1.671 2 103 2.927 3.858
11 0.693 1.320 1.696 2.402 3.193 0.947 1.646 2 073 2.885 3.804
12 0.691 1.316 1.691 2.395 3.183 0.933 1.624 2 048 2.851 3.760
13 0.690 1.313 1.687 2.388 3.175 0.919 1.606 2 026 2.822 3.722
14 0.689 • 1.311 1.684 2.384 3.168 0.909 1.591 2 007 2.796 3.690
15 0.688 1.308 1.680 2.379 3.163 0.899 1.577 1 991 2.776 3.661
16 0.686 1.307 1.678 2.376 3.157 0.891 1.566 1 977 2.756 3.637
17 0.686 1.305 1.676 2.373 3.153 0.883 1.554 1 964 2.739 3.615
18 0.685 1.303 1.674 2.370 3.150 0.876 1.544 1 951 2.723 3.595
19 0.684 1.302 1.672 2.367 3.146 0.870 1.536 1 942 2.710 3.577
20 0.684 1.301 1.671 2.366 3.143 0.865 1.528 1 933 2.697 3.561
21 0.683 1.300 1.670 2.364 3.140 0.859 1.520 1 923 2.686 3.545
22 0.683 1.299 1.668 2.361 3.138 0.854 1.514 1 916 2.675 3.532
23 0.683 1.299 1.668 2.360 3.136 0.849 1.508 1 907 2.665 3.520
24 0.682 1.298 1.667 2.358 3.134 0.845 1.502 1 901 2.656 3.509
25 0.682 1.297 1.666 2.357 3.132 0.842 1.496 1 895 2.647 3.497
•y = 0.90 y = 0.95
3 2.602 4.258 5.310 7.340 9.651 3.804 6.158 7 655 10.552 13.857
4 1.972 3.187 3.957 5.437 7.128 2.619 4.163 5 145 7.042 9.215
5 1.698 2.742 3.400 4.666 6.112 2.149 3.407 4 202 5.741 7.501
6 1.540 2.494 3.091 4.242 5.556 1.895 3.006 3 707 5.062 6.612
7 1.435 2.333 2.894 3.972 5.201 1.732 2.755 3 399 4.641 6.061
8 1.360 2.219 2.755 3.783 4.955 1.617 2.582 3 188 4.353 5.686
9 1.302, 2.133 2.649 3.641 4.772 1.532 2.454 3 031 4.143 5.414
10 1.257 2.065 2.568 3.532 4.629 1.465 2.355 2 911 3.981 5.203
11 1.219 2.012 2.503 3.444 4.515 1.411 2.275 2 815 3.852 5.036
12 1.188 1.966 2.448 3.371 4.420 1.366 2.210 2 736 3.747 4.900
13 1.162 1.928 2.403 3.310 4.341 1.329 2.155 2 670 3.659 4.787
14 1.139 1.895 2.363 3.257 4.274 1.296 2.108 2 614 3.585 4.690
15 1.119 1.866 2.329 3.212 4.215 1.268 2.068 2 566 3.520 4.607
16 1.101 1.842 2.299 3.172 4.164 1.242 2.032 2 523 3.463 4.534
17 1.085 1.820 2.272 3.136 4.118 1.220 2.001 2 486 3.415 4.471
18 1.071 1.800 2.249 3.106 4.078 1.200 1.974 2 453 3.370 4.415
19 1.058 1.781 2.228 3.078 4.041 1.183 1.949 2 423 3.331 4.364
20 1.046 1.765 2.208 3.052 4.009 1.167 1.926 2 396 3.295 4.319
21 1.035 1.750 2.190 3.028 3.979 1.152 1.905 2 371 3.262 4.276
22 1.025 1.736 2.174 3.007 3.952 1.138 1.887 2 350 3.233 4.238
23 1.016 1.724 2.159 2.987 3.927 1.126 1.869 2 329 3.206 4.204
24 1.007 1.712 2.145 2.969 3.904 1.114 1.853 2 309 3.181 4.171
25 0.999 1.702 2.132 2.952 3.882 1.103 1.838 2 292 3.158 4.143
0 In which:
n = sample size,
p = per cent survival, and
7 = confidence level.
Table 33 was originally prepared by D. H. Shaffer of the Westinghouse Research
Laboratories. After this table was prepared, Technical Report No. 34 of Nov. 1 1957,
on "Tables from One-Sided Statistical Tolerence Limits" by G. S. Lieberman was re-
ceived from the Office of Naval Research. This report, prepared at the Applied Mathe-
matics and Statistics Laboratory of Stanford University under Contract N6 ONR-
6 (NR-042-002), contains similar tabular values for sample sizes ranging from 3 to
with an explanation on the construction of the tables.
67
APPENDIX II
ADDITIONAL TECHNIQUE FOR DISTRIBUTION SHAPE

NOT ASSUMED
RUN TESTS
In addition to the rank test for two groups given in Section V A3, there is
another simple test called the run test that may be used, although it is probably
not as sensitive as the rank test.
First arrange all individuals of both samples in ascending or descending order.
Then count the number of "runs" from each sample. For example, suppose the
observations from the two samples are identified by A and B, and the ordered
series gives
ABBAAABABB
Here the number ofruns is six: A - BB - AAA - B - A - BB. From Table 31,
can be established whether this number of runs is too small for the observations
of both samples to have been drawn from one population. If the number of runs
is too small, it may be concluded that the samples are probably different.
EXAMPLE—Inspection of the data in the rank test example, Table 12, shows a total
of ten "runs." Ranks 1 and 2 count as a run, rank 3 counts as a run, and, similarly, 4 to
6, 7, 8 to 10, and so on. According to Table 31, eight or less runs should not occur, fof*
NI = 10 and Nz = 17, more than once in 40 times (2.5 per cent), on the average. In the
example used, the run test does not indicate clearly that the machines are not inter-
changeable. However, the fact that the ten runs found are close to the critical number of
eight runs implies the desirability of another check, such as already has been provided
by the rank test, for example.
68
APPENDIX in
ANALYSIS OF CORRELATION BETWEEN TWO VARIABLES

"QUADRANT SUM" CORRELATION TEST
, Although statistical analysis of correlation between two variables is not dis-
cussed in this Guide, the following test is very useful in determining whether or
not a correlation between two variables probably exists. This simple, quickly
applied test is described by Wilcoxon (28) and credited by him to Olmstead and
Tukey, referring to their paper, "A Corner Test for Association," (30). After the
scatter diagram has been plotted, the following procedure is to be followed: 1
"Two median lines are drawn so as to divide the points into two groups of equal num-
bers, horizontally and vertically. The plotted points now lie in four quadrants, the lower
left and upper right being taken as plus (or minus) quadrants while the upper left and
lower right are taken as minus (or plus) quadrants. Commencing at the right side of the
diagram, lay a ruler or pencil parallel to the vertical axis and move it to the left, counting
the plotted points passed over until the next point lies on the other side of the y-median.
Next lay the ruler at the bottom of the diagram parallel to the horizontal axis and move
it upwards, counting the plotted points passed over before a point is reached on the other
side of the z-median. In a similar manner, move in from the left, and down from the top.
Four values are thus obtained to which are attached the signs of the quadrants in which
they lie. The algebraic sum of these four values is called the quadrant sum and its expected
value is zero if there is no association of the two plotted quantities."
Table 34 gives critical values of the quadrant sum, indicating significant associa-
tion at various probability levels. These critical values are almost independent
of the number of points.
TABLE 34—WORKING SIGNIFICANCE LEVELS FOR QUADRANT SUM.
Significance Levels,
per cent
Quadrant
Sums
Significant
per centLevels, Quandrant
Sums
10 ± 9 2 ±13
5 ±11 1 ±14 to 15°
a
Use 14 for 14 or more points, 15 for fewer points.
When it is necessary to deal with an odd number of points, one of the co-
ordinates of a point will lie on the ^-median and one of the coordinates of another
point will lie on the j-median. In this case, substitute for these points a new
point, with those coordinates taken from the original points which do not involve
the medians. Then proceed in the usual manner.
Another difficulty arises when tied values are encountered. Some of the tied
points may be on the side of the median favorable to being included in the quad-
rant sum, and one or more of the other members of the tied group may be on the
1 Quoted with permission of Frank Wilcoxon (28).
69
other side of the median. In this case treat the tied group as if the number of its
points before crossing the median were:
number favorable for inclusion in quadrant sum
1 + number unfavorable
FIG. 11.—Scatter Diagram.

EXAMPLE.—The scatter diagram of Fig. 11 shows inclusion size plotted against speci-
men life to failure and suggests a possible correlation. The above test was applied and
shows a quadrant sum of —16, as indicated at the right of the figure. The conclusion is
that a definite correlation probably exists between the two variables.
APPENDIX IV
THE WEIBULL DISTRIBUTION FUNCTION FOR FATIGUE LIFE*

On the assumption that fatigue failures are initiated at the "weakest link,"
the fatigue lives of a group of specimens tested under a given set of conditions
may be represented by one of a family of frequency distribution functions:
where:
N = specimen life,
No ^ 0 = minimum life parameter,
Na = characteristic life parameter occurring at the 63.2 per cent
failure point for the population [63.2 = 100(e - \/e), e = 2.718],
and
b > 0 = Weibull shape (or "slope") parameter.
This function is a simple exponential distribution function when 6 = 1; the
Rayleigh distribution function when b = 2; and a good approximation of the
Normal distribution function when b = 3.57, that is, when the mean and the
median values are equal.
The curve representing this function (Fig. 12) is usually skewed to the right,
going on to infinity, and, for b > 1, reaches zero frequency (touches the life axis)
to the left of the mode, which is the life value where the highest number of failures
occur.
The distribution is said to have a nonzero minimum life if the curve touches
the life axis at a value of life greater than 0. In other words, any specimen from
the population represented by such a distribution will have zero probability of
* This description of the Weibull distribution function, as an addition to ASTM
STP 91-A, was originally prepared by a Task Group in Subcommittee VI on Statistical
Aspects of Fatigue of ASTM Committee E-9 on Fatigue, composed of: C. A. Moyer,
chairman, Physical Laboratories, Timken Roller Bearing Co.; J. J. Bush, General Motors
Research Laboratories; and B. T. Ruley, New Departure Div., General Motors Corp. It
has been revised, prior to publication in its present form, by another Task Group in
Subcommittee VI, composed of: John K. H. Kao, chairman, New York University;
Robert A. Heller, Columbia University; B. T. Ruley, New Departures Div., General
Motors Corp.; J. M. Holt, Applied Research Laboratories, U. S. Steel Corp.; M. P.
Semenek, International Harvester Co.; and . R. Gohn and Miss M. N. Torrey, Bell
Telephone Laboratories, Inc.
There has been a demand from the roller bearing industry for the inclusion of an
additional section covering the use of the extreme-value distribution originally proposed
for the analysis of fatigue data by W. Weibull (31,32). Since Fisher and Tippett (33) are
often credited with first showing that this distribution was one of three limiting types of
the extreme-value distribution, it is sometimes referred to as "Fisher-Tippett Type III
for smallest values." As pointed out by Freudenthal and Gumbel (34), this distribution
has some theoretical basis, assuming that fatigue failures are examples of extreme values,
that is, they are smallest-strength or weakest-link values. It has also been used by others
in the analysis of life test data.
71
FIG. 12.—Typical Weibull Distribution Curves, from Kao (35).

TABLE 35.—ORE)INATE LOCATIO NS CORRESPON DING TO PER CENT
FAIL!]D VALUES.
1 1
F(N) X 100 log 1 - PUt) F(N) X 100 Iog 1 - F(N)
2 0.0088 52 0 3188
4 0.0177 54 0 3372
5 0 0223 55 0 3468
6 0 0269 56 0 3565
8 0 0362 58 0 3768
10 0.0458 60 0 . 3979
12 0.0555 62 0 4202
14 0.0655 63 2 0 4341
15 0.0706 64 0 4437
16 0.0757 65 0 4559
18 0.0862 66 0.4685
20 0.0969 68 0.4949
22 0.1079 70 0.5229
24 0.1192 72 0.5528
25 0.1249 74 0 5850
26 0.1308 75 0 6021
28 0 1427 76 0 6198
30 0 1549 78 0 6576
32 0.1675 80 0.6990
34 0.1805 82 0.7447
35 0.1871 84 0.7959
36 0.1938 85 0.8239
38 0.2076 86 0.8539
40 0.2218 88 0 9208
42 0.2366 90 . 1 . 000
44 0.2518 92 1.097
45 0.2596 94 1.222
46. 0.2676 95 1.301
48 0.2840 96 1.398
50 0.3010 98 1.699
NOTE.—All logs are to the base 10.
failure prior to N0 life. Later it will be n how to test for N0 values greater
than zero, but if it is reasonable to assume N0 = 0, the frequency distribution
function is simplified.
Since the data are usually obtained in an ordered manner in fatigue testing, it
is easy to fit a cumulative distribution function to fatigue life. The cumulative
function for the fraction of population failed prior to life N is
WEIBULL DISTRIBUTION FUNCTION FOR FATIGUE LIFE 73
This function can be transformed into the straight-line relationship
which allows a simple graphical method for fitting the Weibull distribution to the
data and the subsequent graphical estimation of the parameters (b, N0, and Na)
in the formula.
95
FIG. 13.—Construction of Weibull Probability Paper from Log-Log Paper.

Construction of Probability Paper
Although Weibull probability paper can be purchased from a source such as
Cornell University, Ithaca, N. Y., Columbia University, New York, N. Y., or
Technical and Engineering Aids for Management, 104 Belrose Ave,, Lowell,
Mass., it can be constructed rather simply from square log-log paper, that is,
log-log paper in which the cycles are the same size in both directions. The paper
is prepared by the marking off on the vertical logarithmic scale of the probability
percentages F(N) corresponding to the values of
given in Table 35. For example, in Fig. 13, the ordinate of the 90 per cent failure
value is 1.000 on the vertical logarithmic scale. Similarly, the ordinate for the 20
TABLE 36.— MEAN -RANI C ESTI MATE 3° OF THE P ER CE NT PO PULArDION iÂILEI) COR]RESPO NDINCJ TO F AILUEJE ORIDER IJf SAM!PLE.
Samp' e Size, n.
Order No., q 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
No. 1 50.00 33.33 25.00 20.00 16.67 14.29 12.50 11.11 10.00 9.09 8.33 7.69 7.14 6.67 6.25 5.88 5.56 5.26 5 00 4 76
No 2 66.67 50.00 40.00 33.33 28.57 25.00 22.22 20.00 18.18 16.67 15.38 14.29 13.33 12.50 11 76 11 11 10 53 10 00 9 52
No 3 75.00 60.00 50.00 42.86 37.50 33.33 30.00 27.27 25.00 23.08 21.43 20.00 18.75 17 65 16 67 15 79 15 00 14 29
No 4 80.00 66.67 57.14 50.00 44.44 40.00 36.36 33.33 30.77 28.57 26.67 25.00 23 53 22 22 21 05 20 00 19 05
No 5 83.33 71.43 62.50 55.56 50.00 45.45 41.67 38.46 35.71 33.33 31.25 29.41 27.78 26.32 25 00 23 81
No 6 85.72 75.00 66.67 60.00 54.55 50.00 46.15 42.86 40.00 37.50 35 29 33 33 31 58 30 00 28 57
No. 7 87.50 77.78 70.00 63.64 58.33 53.85 50.00 46.67 43.75 41.18 38.89 36.84 35 00 33 33
No 8 88.89 80.00 72.73 66.67 61.54 57.14 53.33 50.00 47.06 44.44 42.11 40 00 38 10
No 9 90.00 81.82 75.00 69 . 23 64.29 60.00 56.25 52.94 50 00 47.37 45 00 42 86
No 10 90.91 83.33 76.92 71.43 66.67 62.50 58 82 55 56 52 63 50 00 47 62
No 1 91.67 84.62 78.57 73.33 68.75 64.71 61.11 57.89 55 00 52 38
No 12 92.31 85.71 80.00 75.00 70.59 66.67 63.16 60 00 57 14
No 1 92.86 86.67 81.25 76.47 72.22 68.42 65.00 61 90
No 1 4 93.33 87.50 82.35 77.78 73.68 70 00 66 67
No 15 93.75 88.24 83.33 78.95 75 00 71 43
No 1 6 94.12 88.89 84.21 80.00 76 19
No 17 94.44 89.47 85.00 80 95
No 18 94.74 90.00 85.71
No 19 95.00 90.48
No 20 95.24
0 Mean-rank estimates = 100
WEIBULL DISTRIBUTION FUNCTION FOR FATIGUE LIFE 75
per cent failure line is 0.0969 on the logarithmic scale. On such paper, the tangent
of the angle 6 is an estimate of the Weibull "slope," b, for the population line.
The angle 0 may be measured with a protractor, or the slope of the line may be
computed.
Plotting Positions on Probability Paper:
The fatigue data for any one sample are first ordered from shortest to longest
life, each specimen being given an order number, q, from 1 through n. The hori-
zontal plotting position is its individual life value. All runouts are assumed to
have longer lives than the last ordered specimen that failed, but such data are
treated separately below under "Estimates of the Distribution Function Pa-
rameters."
The vertical plotting position of the per cent failed (Fig. 13) is the estimate of
the per cent of the population failed, F(N), based upon the specimen order num-
ber. Mean-rank estimates of the percentages of the population failed at successive
TABLE 37.—TYPICAL FATIGUE TABL,E 38.—TYPI CAL FATIGUE
TEST DATA. 1PEST DATA, WITHOUT
RUNG UTS.
Number of Revolutions to Failure Plot of N Versus 1?(N) Nonlinear
Order, q
Lot 1 Lot 2 Order, q Specimen Number of Revolu-
tions to Failure
No. 1 . . 1.1 X 106 2 0 X 106
No. 2 . . 2.3 3 7 No. 1. . . No. 4 4.0 X 105
No. 3 . . 4.0 5 0 No. 2 . . . No. 2 5.0
No. 4. . 6.5 8 0 No. 3 . . . No. 5 6.0
No. 5. . 8.6 11 5 No. 4 . . . No. 8 7.3
No. 6 13 0 No. 5. . . No. 1 8.0
No. 7 20 0 No. 6 . . . No. 7 9.0
No 8 23 5 No. 7. .. No. 6 10.6
No. 8. . . No. 3 13.0
order numbers are given in Table 36 for sample sizes ranging from 1 through 20.
Mean rank, q/(n + 1), is an unbiased estimate of F(N); such estimates are recom-
mended by Gumbel (36) and Weibull (37). Blom(38) suggests modified mean-rank
estimates. For the data given in Table 37 for the sample taken from lot 1, the
abscissa for the first specimen is plotted at its life value of N = 1.1 X 106 revo-
lutions and the ordinate at F(N) X 100 = 16.67, the plotting position for the first
of a sample of five based upon mean ranks given in Table 36.
Estimates of the Distribution Function Parameters:
1. An estimate of the population cumulative distribution that corresponds to
the data plotted in Fig. 13 can be fqund quickly by drawing a line by eye through
the failed points. More refined techniques for calculating this line can be found
by referring to Gumbel (36), Lieblein , or Kao (35). It is possible to calculate
this line by the method of least squares, as illustrated in Section V A4 of this
guide. For example:
and
Comparisons using these methods as against the graphic method sho.w,,however,
that the latter is usually adequate for small samples.
2. An estimate of the characteristic life, Na , is obtained from Fig. 13 by reading
off the life value corresponding to the intersection of the fitted line and a hori-
zontal line corresponding to F(N) X 100 = 63.2 per cent.
3. An estimate of the median life is obtained by reading off the life value corre-
sponding to the intersection of the straight line of Fig. 13 and a horizontal line
corresponding to F(N) X 100 = 50 per cent.
FIG. 14.—Estimation of Weibull Distribution Function Parameters for Data in Table 38.
4. In Fig. 13, the minimum life, N0 , is assumed to equal zero, since the plot
of the fatigue data is approximately linear. The plotted data from Table 38
result in a line which curves downward (Fig. 14(a)); thus the existence of a finite
minimum life value greater than 0 would be suspected. To find an estimate of
minimum life, N0 : (1) note the life value which the curve approaches asymp-
totically, (2) obtain the quantity N — N0 for each point by subtracting the N0
value from each individual specimen life, and (3) plot this life difference on Weibull
paper versus the same per cent failed values as before. Thus, by trial and error,
the best estimate of N0 will be found so that the data shown in Fig. 14(o) will,
when transformed, plot as a straight line, as shown in Fig. 14(6).
5. The slope parameter, b, is equal to the tangent of the angle 6 shown in Fig.
13. Another estimate of b can be made by computing the tangent of 6 from the
logarithms of the ordinates and abscissas of two widely separate points, NI and
Nz, on the fitted line. Thus
estimate of b
6.The skewness of the Weibull distribution varies with the shape parameter,
b; and the Weibull mean, in general, may occur at various per-cent-failed values;
WEIBULL DISTRIBUTION FUNCTION FOR FATIGUE LITE 77
TABLE 39.—TYPICAL FATIGUE TEST

DATA, WITH RUNOUTS.
Order, Specimen Number of Revolutions
q
to Failure
No.'l No. 2 1.30 X 106
No. 2 No. 5 1.60
No. 3 No. 4 1.75
No. 4 No 1 2.10
No. 5 No 6 2.35
No. 6 No 3 2.70
No. 7 No. 7 runout
No. 8 No. 8 runout
FIG. 15.—Per Cent Failed at Weibull Mean.
FIG. 16. —Êstimation of Weibull Distribution Function Parameters for Data in Table 39.
that is, the mean does not coincide with the median. Using the estimated Weibull
slope, b, it is possible to read from Fig. 15 an estimate of the per cent failed at the
Weibull mean and then refer back to the estimated population line on Weibull
probability paper, as in Fig. 13, to read off the estimated mean life from the1
curve. Gumbel (36) and Kao (35) give methods for calculating the Weibull mean
when the characteristic life Na and the slope b are known.
7. For data containing run-out specimens (Table 39), the n' broken specimens
(6 in the example, Fig. 16), out of a total of n specimens tested, are plotted on
probability paper at the mean-rank plotting positions, corresponding to a sample
size n (8 in the example, Fig. 16(a)). The line drawn through these points will
approach a horizontal asymptote, F/racture , which is equal to the ratio of the
first plotting positions corresponding to sample sizes n and n', respectively (Fig.
16(a)).
The parameters of this distribution may be obtained graphically by plotting
only the n' broken specimens at mean-rank plotting positions, corresponding to a
sample size n' versus N — N0 , where N0 is again the estimate of the vertical
asymptote approached by the curve. The slope of the resulting straight line (Fig-
16(6)), tan 6 = b, can be obtained as described in this Section. Na , at the proba-
bility level of 63.2 per cent, is taken directly from the plotted line. The estimated
equation of the probability function for the complete sample of size n will then
become
where F/ = rfrac ture .

The curve of Fig. 16(a) may now be replotted by using, as ordinates, Fracture
times the ordinates of the straight line and, as abscissas, N0 plus the abscissas
of the straight line. Note that Na is, in this case, no longer the estimate of the?
characteristic life parameter of the complete distribution, F(N). The value of N
at the 63.2 per cent probability of failure level may be obtained from the plot in
Fig. 16(o).
1 Weibull mean:
where T — the gamma function; and for Weibull variance:

FATIGUE TESTING AND STATISTICAL ANALYSIS OF DATA 79
REFERENCES
(1) D. J. Finney, Probit Analysis, Cambridge University Press, 1952.
(2) E. S. Pearson and H. O. Hartley, Biometrika Tables for Statisticians, Cambridge Uni
versity Press, 1954.
(3) R. A. Fisher and F. Yates, Statistical Tables for Biological, Agricultural and Medical
Research, Fourth Edition, Oliver & Boyd, London, 1953.
(4) T. S. Dolan, "Certain Mechanical Strength Properties of Aluminum Alloys 25S-T and
X76S-T," NACA TN914, October, 1943.
(5) H. T. Corten, Todor Dirnoff, T. J. Dolan, and Masaki Sugi, "An Appraisal of the Prot
Method of Fatigue Testing, Part II," Technical Report No. 35 on the Behavior of
Metals under Repeated Stress, ONR Contract N6-ori 071(04), University of Illinois.
June, 1953.
(6) E. Prot, "Fatigue Testing Under Progressive Loading, A New Technique for Testing
Materials," translated by Edward J. Ward, Captain, USAF, WADC TR 52-148,
September, 1952.
(7) E. J. Ward and D. C. Schwartz, "Investigation of Prot Accelerated Fatigue Test,"
WADC TR 52-234, November, 1952.
(8) A. P. Boresi and T. J. Dolan, "An Appraisal of the Prot Method of Fatigue Testing,"
Technical Report No. 34 on the Behavior of Metals under Repeated Stress, ONR Con-
tract N6-ori-71, T.O. IV, University of Illinois, January, 1953.
(9) W. J. Dixon and F. J. Massey, Jr., Introduction to Statistical Analysis, McGraw
Hill Book Co., 1957.
(10) E. L. Crow, F. A. Davis, and M. W. Maxfield, Statistical Manual, Dover Publications,
Inc., New York, N.Y., 1960.
(11) The Design and Analysis of Industrial Experiments, edited by O. L. Da vies, Hafner
Publishing Co., New York, N.Y., 1956.
(12) D. B. Owen, Handbook of Statistical Tables, Addison Wesley Publishing Co., Inc.,
Reading, Mass., 1962.
(13) J. A. Greenwood and M. M. Sandomire, "Sample Size Required for Estimating the
Standard Deviation as a Per Cent of Its True Value," Journal, Am. Statistical Assn.,
Vol. 45, 1950, p. 258.
(14) C. P. Ferris, F. E. Grubbs and C. L. Weaver, "Operating Characteristics for the
Common Statistical Tests of Significance," Annals of Mathematical Statistics, Vol.
17, 1946, p. 178.
(15) A. M. Freudenthal and E. J. Gumbel, "Minimum Life in Fatigue," Journal, Am.
Statistical Assn., September, 1954.
(16) R. B. Murphy, "Non-Parametric Tolerance Limits," Annals of Mathematical Sta-
tistics. Vol. XIX, 1948, pp. 581-589.
(17) E. H. Schuette, "A Simplified Procedure for Obtaining Design-Level Fatigue Curves,"
Proceedings, Am. Soc. for Testing Mats., Vol. 54, 1954.
(18) E. H. Schuette, "The Prediction of Exceedances in Limit-Value Testing," Statistical
Methods in Materials Research, proceedings for a short course conducted by the
Pennsylvania State University, June, 1956.
(19) I£. R. $air, "Table of Confidence Intervals for the Median in Samples from Any
Continuous Population," Sankhya, Vol. 4, 1940, pp. 551-558.
(20) W. J. Youden, "Systematic Error in Physical Constants," Physics Today, Vol. 14,
September, 1961, p. 32.
(21) "Tables of the Binomial Probability " Applied Mathematics Series 6,
Nat. Bureau Standards, U. S. Government Printing Office, Washington, D. C., 1949.
(22) W. H. Kruskal and W. A. Wallis, "Use of Ranks in One Criterion Variance Analy-
sis," Journal, Am. Statistical Assn., Vol. 47, 1952, pp. 583-621.
(23) P. G. Hoel, Introduction to Mathematical Statistics, Second Edition, John Wiley &
Sons, Inc., New York, N. Y., 1954.
(24) A. Hald, Statistical Theory with Engineering Applications, John Wiley & Sons, Inc.
New York, 1952, pp. 550-551.
(25) N. C. Severe and E. G. Olds, "A Comparison of Tests on the Mean of a Logarithico-

Normal Distribution with Known Variance," Annals of Mathematical Statistics,
Vol. 27, No. 3, September, 1956, p. 670.
(26) G. W. Snedecor, Statistical Methods, Fifth Edition, The Iowa State College Press,
1946.
(27) E. H. Schuette, "The Significance of Test Results from Small Groups of Specimens,"
Proceedings, Am. Soc. Testing Mats., Vol. 57, 1957.
(28) F. Wilcoxon, Some Rapid Approximate Statistical Procedures, American Cyanamid
Co., New York, N. Y., 1949.
(29) C. Eisenhart and F. Swed, "Tables for Testing Randomness of Grouping in a Se-
quence of Alternatives," Annals of Mathematical Statistics, Vol. 14, 1943, p. 66.
(30) P. S. Olmstead and J. W. Tukey, "A Corner Test for Association," Annals of Mathe-
matical Statistics, Vol. 18, 1947, pp. 495-513.
(31) W. Weibull, "A Statistical Distribution Function of Wide Applicability," Transac-
tions, Am. Soc. Mechanical Engrs.; and Journal of Applied Mechanics. Vol. 73,
September, 1951, pp. 293-297.
(32) W. Weibull, Fatigue Testing and the Analysis of Results, Pergamon Press, New York
N. Y., 1961.
(33) R. A. Fisher and L. H. C. Tippett, "Limiting Forms of the Frequency Distribution
of the Largest or Smallest Member of a Sample," Proceedings, Cambridge Philosophi-
cal Soc., Vol. 24, Part 2, 1928, p. 180. Reprinted in Fisher's Contributions to Mathe
matical Statistics, John Wiley and Sons, Inc., New York, N. Y., 1950.
(34) A. M. Freudenthal and E. J. Gumbel, "Physical and Statistical Aspects of Fatigue,"
Advances in Applied Mechanics, Vol. 4, 1956, pp. 117-158.
(35) J. H. K. Kao, "A Summary of Techniques on Reliability Studies of Components
Using Weibull Distribution," Proceedings, Sixth Symposium on Reliability and
Quality Control, Cornell University, Ithaca, N. Y., January, 1960.
(36) E. J. Gumbel, "Statistical Theory of Extreme Values and Some Practical Applica-
tions," Applied Mathematics Series 33, Nat. Bureau Standards, Feb. 12, 1954.
(37) W. Weibull, "A Statistical Representation of Fatigue Failures in Solids," Acta
Polytechnica, Mechanical Engineering Series, Vol. 1, No. 9, 1949.
(38) G. Blom, Statistical Estimates and Transformed Beta Variables, John Wiley and
Sons, Inc., New York, N. Y., 1958.
(39) J. Lieblein, "A New Method of Analyzing Extreme-Value Data," Technical Note
3053, National Advisory Committee for Aeronautics, Washington, D. C., January,
1954.
SUPPLEMENTARY READING FOR APPENDIX IV
(40) J. H. K. Kao, "The Design and Analysis of Life-Testing Experiments," Transac'
tions, 1958 Middle Atlantic Conference, Am. Soc. Quality Control; and Reliability
Training, Inst. Radio Engrs., 2nd Edition, Chapter II, March, 1960.
(41) E. J. Gumbel, Statistics of Extremes, Columbia University Press,. New York, N. Y.,
1958.
(42) E. J. Gumbel, "Probability Tables for the Analysis of Extreme-Value Data," Applied
Mathematics Series 22, Nat. Bureau Standards, July 6, 1953.
(43) W. Weibull, "New Methods for Computing Parameters of Complete or Truncated
Distributions," FFA Report 58, Aeronautical Research Inst. of Sweden, February,
1955.
INDEX
Abbreviations, p. 7 Estimates of parameters, single stress level, pp.
Analysis of correlation between two variables, 18, 40, 76
Appendix III (p. 69) Estimates, mean rank—Weibull distribution
Allocation of test specimens—probit method, function, Table 36 (p. 74)
Table 1 (p. 11) Estimation, p. 4, Fig. 14 (p. 76)
Analysis of fatigue data, p. 22 Estimation, Weibull distribution function pa-
Areas of the Normal curve, Table 28 (p. 61) rameters, Fig. 14 (p. 76)
Arithmetic mean, p. 5 Fatigue, p. 2
Average, sample, p. 5 Fatigue data (see analysis of), p. 22
X2/d.f. distribution, percentiles of, Table 30 (p. Fatigue life, pp. 2, 6, 27, 39
63) Fatigue life for a stated value of per cent sur-
X22 distribution, percentiles of, Table 27 (p. 60) vival, p. 28
X ,P-3 2 Fatigue life for p per cent survival, p. 6
Choice of distribution shape, p. 40 Fatigue limit, p. 3
Computations for fitting a response curve by Fatigue limit for p per cent survival, p. 6
method of least squares, p. 34, Table 16 (p. 34) Fatigue notch factor, p. 4
Computation of significance tests, Table 20 (p. Fatigue notch sensitivity, p. 4
46) Fatigue strength, p. 6, Fig. 8 (p. 51)
Computation of standard deviation of values Fatigue strength for p per cent survival at N
about fitted line, Table 17 (p. 36) cycles, p. 6
Confidence coefficient, p. 5 Fatigue test data, Table 12 (p. 30), Table 13 (p.
Confidence interval, pp. 5, 26, 27, 28, 42, Table 31), Table 37 (p. 75), Table 38 (p. 75), Table
9 (p. 26), Table 10 (p. 27), Table 11 (p. 28) 39 (p. 77)
Confidence level, p. 5 Fatigue tests, p. 1
Confidence limits (see confidence interval) F-distribution, Table 32 (p. 65)
Constant amplitude tests, pp. 1, 9-13 F-ratio test, pp. 45, 47
Constant life fatigue diagram, p. 14 Frequency distribution, p. 4
Construction of Weibull probability paper from Gaussian distribution curve, Fig. 6 (p. 22)
log-log paper, Fig. 13 (p. 73) Group, p. 4
Correlation between two variables, Appendix III Increasing amplitude tests, pp. 1, 13, Fig. 4 (p.
(p. 69) 14), Fig. 5 (p. 15)
Cycle ratio, p. 4 Interval, pp. 5, 9
Definitions, p. 2 Interval estimate, p. 5
Definitions relating to fatigue tests and test ^-factors for S-N curves, Table 33 (p. 67)
methods, p. 2 Least squares, method for fitting a response
Definitions relating to statistical analysis, p. 4 curve, Table 16 (p. 34)
Definitions relating to statistical analysis of Level, confidence, p. 5
fatigue data, p. 6 Level, significance, p. 6
Difference among k means, p. 47 Level, tolerance, p. 5
Difference between two means, pp. 21, 45 Limits, confidence, p. 5
Difference between two standard deviations, Limits, fatigue strength at N cycles, p. 38
pp. 20, 44 Limits, method of computing, Table 19 (p. 38)
Distribution, p. 4, Table 27 (p. 60), Table 30 Limits, tolerance, p. 5
(p. 63), Table 32 (p. 65), Fig. 6 (p. 22) Maximum stress, p. 3
Distribution curves, Fig. 12 (p. 72), Fig. 6 (p. 22) Mean, confidence interval for, p. 19
Distribution shape, choice of, pp. 22, 40 Mean, definition, p. 5
Estimate, p. 5 Mean, confidence limits for, Table 10 (p. 27)
, interval (see confidence interval) Mean fatigue life, p. 27
, point, p. 5 Mean, sample, p. 21
81
Mean rank estimates: per cent of population "Quadrant sum" correlation test, Appendix III
failed corresponding to failure order in sample, (p- 69)
Table 36 (p. 74) "Quadrant sum," working significance level, Ta-
Means, confidence interval for, p. 42, Table 2 ble 34 (p. 69)
(p-1 9) Range of stress, p. 3
Means, differences between two, p. 45 Rank test, pp. 25, 30, 31, Table 25 (p. 56), Table
Means, differences among k, p. 47 26 (p. 58)
Mean stress, p. 3 References, p. 79
Mean, Weibull, Table 36 (p. 74), Fig. 15 (p. 77) Response curves, pp. 7, 18, 33, 34, 36, 38, Fig. 7
Median, confidence limits for, Table 9 (p. 26) (p. 35), Table 16 (p. 34)
Median fatigue life, pp. 6, 26 Response or survival tests, pp. 1, 10, 12, 13,
Median fatigue strength at N cycles, p. 6 Fig. 2 (p. 11)
Median percentage of survivors for the popula- Rotating beam tests, R. R. Moore, Table 21
tion, Table 8 (p. 24) (p. 50)
Median, sample, p. 43 Run test, Appendix II (p. 68)
Medians, differences of group, p. 29 Runs along elements in samples of sizes NI and
Method of least squares, pp. 34, 35, Table 16 N2, Table 31 (p. 64)
(p. 34) Sample, p. 4
Minimum per cent of population exceeding me- Sample average, p. 5
dian of low ranking points, Table 25 (p. 56) Sample means, p. 21
Minimum stress, p. 3 Sample median, p. 5
Modified staircase test method, pp. 1, 13, 48, 49 Sample standard deviation, p. 5
Moore rotating beam step test, Table 21 (p. 50), Sample percentage, p. 5
Table 22 (p. 51), Fig. 8 (p. 51) Sample variance, p. 5
Mo.025 and MO.QTS for runs among elements in sam- Scatter diagram, Fig. 11 (p. 70)
ples of sizes NI and N%, Table 31 (p. 64) Selection of test specimens, pp. 16, 17, 19
Nominal stress, p. 2 Significance level, p. 6, Table 34 (p. 69)
Normal curve, areas of, Table 28 (p. 61) Significance level, for "quadrant sum," Table 34
Normal distribution curve, Fig. 6 (p. 22) (p. 69)
Normal distribution of fatigue hie, p. 39 Significant, p. 5
Normal distribution, ^-factors for S-N curves, S-N curve for 50 per cent survival, p. 6
Table 33 (p. 67) S-N curve for p per cent survival, p. 7
Number of test specimens, minimum, pp. 16-21, S-N curves, pp. 6, 7, 17, 19, 23, 25, 40, 41, Table
Table 2 (p. 19), Table 3 (p. 19), Table 4 (p. 8 (p. 24), Table 25 (p. 56), Table 33 (p. 67)
20), Table 5 (p. 20), Table 6 (p. 21), Table 7 S-N diagrams, p. 3
(p. 21) Staircase test method, pp. 1, 48, Fig. 3 (p. 12)
Parameter, pp. 4, 18, 26, 40, Fig. 14 (p. 76), Fig. Standard deviation, pp. 19, 44-46, Table 2 (p.
16 (p. 77) 19)
Per cent of specimens having at8 least the indi- Standard deviation, confidence interval for, p. 42
cated fatigue strength at 10 cycles, Fig. 8 Standard tests, pp. 1, 9, 23, 25
(p. 51) Statistic, p. 4
Per cent survival for a stated value of fatigue Steady component of stress, p. 3
life, p. 27 Step test method, pp. 1, 13, 50, Fig. 4 (p. 14)
Per cent survival values at N cycles, confidence Step tests, R. R. Moore rotating beam specimens,
limits for, p. 37 Table 21 (p. 50), Table 22 (p. 51)
Percentiles of the x2 distribution, Table 27 (p. Stress, p. 3
60) Stress amplitude, p. 3
Percentiles of the x2/d.f. distribution, Table 30 Stress concentration factor, p. 4
(p. 63) Stress cycle, p. 2
Point estim_ate, pp. 5, 26-29 Stress cycles endured, p. 3
Population, p. 4 Stress ratio, p. 3
Probability-stress-cycle curve, Fig. 1 (p. 10) Survival tests, Fig. 2 (p. 11)
Probability paper, pp. 33, 73 Symbols, p. 7
Probit test—allocation of test specimens, Table Test of significance, pp. 6, 18, 20, 29, 30, 31, 45,
1 (p- 11) 46, 47, Table 20 (p. 46)
Probit test data, Table 15 (p. 34) Test procedures, pp. 8, 9, 10, 12, 13, 15, Fig. 3
Probit test method, p. 10 (p. 12), Fig. 4 (p. 14T, Fig. 5 (p. 15)
t test, pp. 1, 15, 52, Table 23 (p. 52), Fig. 9 Test specimens, minimum number, p. 16
(p. 53) Test specimens, selection of, p. 16
INDEX 83
Test-statistic, p. 6 Universe, p. 4
Theoretical stress concentration factor, p. 4 Unpaired rank test, p. 30, Table 26 (p. 58)
Tolerance interval, pp. 5, 41, 42 Values of t, Table 29 (p. 62)
Tolerance level, p. 5 Variable component of stress, p. 3
Tolerance limits, p. 5 Weibull distribution, p. 71
2-test, p. 45 Weibull mean, Fig. 15 (p. 77), Footnote p. 78
/-values, Table 29 (p. 62) Wohler test method, p. 1
This page intentionally left blank
T HIS PUBLICATION is one of many
issued by the American Society for Testing and Materials
in connection with its work of promoting knowledge
of the properties of materials and developing standard
specifications and tests for materials. Much of the data
result from the voluntary contributions of many of the
country's leading technical authorities from industry,
scientific agencies, and government.
Over the years the Society has published many tech-
nical symposiums, reports, and special books. These may
consist of a series of technical papers, reports by the
ASTM technical committees, or compilations of data
developed in special Society groups with many organiza-
tions cooperating. A list of ASTM publications and
information on the work of the Society will be furnished
on request.

STP 91-A-1963

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

STP 91-A-1963

Uploaded by

Copyright:

Available Formats

F ft O F E R T Y O F

Reg. U. S. Pat. Off.

ASTM Special Technical No. 91-A* (Second Edition}

Price: $5.00; to Members: $4.00

Printed in Baltimore, Md.

RELATED ASTM PUBLICATIONS

Abstracts of Articles on Fatigue (STP 9)

GUIDE FOR FATIGUE TESTING AND

Copyright© 1 964 by ASTM International www.astm.org

Theoretical Stress Concentration Factor (or Stress Concentration Factor),

NOTE 1.—This is a special case of the more general definition 51.

LIST OF SYMBOLS AND ABBREVIATIONS

FIG. 1.—Probability-Stress-Cycle (P-S-N) Curve for Phosphor-Bronze Strip.

FIG. 2.—Response or Survival Tests.

FIG. 3.—Illustration of Staircase Method.

FIG. 4.—Representation of "Step" Testing of Single Specimen.

FIG. 5.—Graphical Illustration of Prot Data.

of the estimate found from constant amplitude methods. It is not certain,

0 2<r ... X± 0. 1 <7 384 0.14<7 385

is compared with £0.975 in Table 29. e Section V B4(6).) No F-ratio test

FIG. 6.—"Normal" or Gaussian Distribution Curve.

TABLE 1 0—APPROXIMATE CONFIDENCE INTERVALS FOR THE MEAN.

Sample Size, n Procedure". 6 Length of Interval

5 ...... use the range: N\ and Ns I X range

2. Approximate Confidence Interval Estimate. —An approximate confidence

Limits Limits Limits Limits

100 48 100 100 68 100 100 82 100 100 91 100

(2) m is chosen so that

Rank Kilocycles Rank Kilocycles

2. When the sample sizes are equal the formula reduce to

TABLE 16.—COMPUTATIONS FOR FITTING A RESPONSE CURVE

Applied Stress, Coded0 Per Cent

40.0 1 93.33 -1.50 1 -1 .50 -1.51 93.4

EXAMPLE.—Fatigue tests of specimens of a certain steel gave the data in Table

where an Area value corresponds to a value of z = F/ in Table 28.

(a) Confidence band for per cent survival values.

Applied Stress, ksi Y Yf (F - F/)*

(V) Confidence Limits for a Response Curve. 19—The method of computing

The symbols have been defined previously; the computation procedure is

40.0 1 4 0.40 0.77 0.47 -1.51 -1 .98 -1 .04 97.6 85.1

90 -1 .28 1 .322.82 0.282 0.68 0.62 1. 19 0.57 1.81 39.36 41 .22

The other symbols have been defined previously.

EXAMPLE.—Table 19 illustrates the computation of 95 per cent confidence

where ^ is the sum from I ton.

(b) Confidence Interval for the Standard Deviation. —A confidence interval

2. Compute ft\ and (82 as before. From Table 30, read:

using percentiles 100 ft and 100 ft respectively, with d.f. = n — 1.

EXAMPLE.—This example will illustrate the construction of confidence intervals.

The desired confidence interval for /* has the limits

Approximate confidence limits for the population standard ^deviation are

In order to decide if the difference is statistically significant, a signifi-

Next compute the test-statistic

From Table 29 read a value /j», corresponding to

If this value is not an integer, use the nearest smaller integer.

From Table 32, F0. m (d.f. numerator = 7) (d.f. denominator = 9) = 4.20.

where x is the mean of the numbers xt . Compute the ratio

7. The Step Method:

f° 21 700 4.3365 C -B 0.3542 0.3542/0.5458 = 0.65

f°B 23 700 4.3747 C -B 0.3102 0.3102/0.5458 = 0.57

(a) Incorrect assumed value of fatigue limit.

Copyright© 1 964 by ASTM International www.astm.org

50 50.00 50.00 50.00 50.00 50.00 50.00 50.00

20.63 20.63 20.63 20.63 20.63 20.63 20.63

"Adapted from work of Dr. Frank Wilcoxon (28).