Designation: G16 − 13

Standard Guide for

Applying Statistics to Analysis of Corrosion Data1
This standard is issued under the fixed designation G16; the number immediately following the designation indicates the year of original
adoption or, in the case of revision, the year of last revision. A number in parentheses indicates the year of last reapproval. A superscript
epsilon (´) indicates an editorial change since the last revision or reapproval.

1. Scope of materials are under test, but statistical methods provide a

1.1 This guide covers and presents briefly some generally rational approach to this problem.
accepted methods of statistical analyses which are useful in the 3.2 Modern data reduction programs in combination with
interpretation of corrosion test results. computers have allowed sophisticated statistical analyses on
1.2 This guide does not cover detailed calculations and data sets with relative ease. This capability permits investiga-
methods, but rather covers a range of approaches which have tors to determine if associations exist between many variables
found application in corrosion testing. and, if so, to develop quantitative expressions relating the
variables.
1.3 Only those statistical methods that have found wide
acceptance in corrosion testing have been considered in this 3.3 Statistical evaluation is a necessary step in the analysis
guide. of results from any procedure which provides quantitative
information. This analysis allows confidence intervals to be
1.4 The values stated in SI units are to be regarded as estimated from the measured results.
standard. No other units of measurement are included in this
standard. 4. Errors
4.1 Distributions—In the measurement of values associated
2. Referenced Documents
with the corrosion of metals, a variety of factors act to produce
2.1 ASTM Standards:2 measured values that deviate from expected values for the
E178 Practice for Dealing With Outlying Observations conditions that are present. Usually the factors which contrib-
E691 Practice for Conducting an Interlaboratory Study to ute to the error of measured values act in a more or less random
Determine the Precision of a Test Method way so that the average of several values approximates the
G46 Guide for Examination and Evaluation of Pitting Cor- expected value better than a single measurement. The pattern
rosion in which data are scattered is called its distribution, and a
IEEE/ASTM SI 10 American National Standard for Use of variety of distributions are seen in corrosion work.
the International System of Units (SI): The Modern Metric
4.2 Histograms—A bar graph called a histogram may be
System
used to display the scatter of the data. A histogram is
3. Significance and Use constructed by dividing the range of data values into equal
intervals on the abscissa axis and then placing a bar over each
3.1 Corrosion test results often show more scatter than interval of a height equal to the number of data points within
many other types of tests because of a variety of factors, that interval. The number of intervals should be few enough so
including the fact that minor impurities often play a decisive that almost all intervals contain at least three points; however,
role in controlling corrosion rates. Statistical analysis can be there should be a sufficient number of intervals to facilitate
very helpful in allowing investigators to interpret such results, visualization of the shape and symmetry of the bar heights.
especially in determining when test results differ from one Twenty intervals are usually recommended for a histogram.
another significantly. This can be a difficult task when a variety Because so many points are required to construct a histogram,
it is unusual to find data sets in corrosion work that lend
1
This guide is under the jurisdiction of ASTM Committee G01 on Corrosion of themselves to this type of analysis.
Metals and is the direct responsibility of Subcommittee G01.05 on Laboratory 4.3 Normal Distribution—Many statistical techniques are
Corrosion Tests.
Current edition approved Dec. 1, 2013. Published December 2013. Originally based on the normal distribution. This distribution is bell-
approved in 1971. Last previous edition approved in 2010 as G16–95 (2010). DOI: shaped and symmetrical. Use of analysis techniques developed
10.1520/G0016-13.
2
for the normal distribution on data distributed in another
For referenced ASTM standards, visit the ASTM website, www.astm.org, or
contact ASTM Customer Service at [email protected]. For Annual Book of ASTM
manner can lead to grossly erroneous conclusions. Thus, before
Standards volume information, refer to the standard’s Document Summary page on attempting data analysis, the data should either be verified as
the ASTM website. being scattered like a normal distribution, or a transformation

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

1
 G16 − 13
should be used to obtain a data set which is approximately nonlinearity, a transformation may be used to obtain a new,
normally distributed. Transformed data may be analyzed sta- transformed data set that may be normally distributed. Al-
tistically and the results transformed back to give the desired though it is sometimes possible to guess at the type of
results, although the process of transforming the data back can distribution by looking at the histogram, and thus determine the
create problems in terms of not having symmetrical confidence exact transformation to be used, it is usually just as easy to use
intervals. a computer to calculate a number of different transformations
4.4 Normal Probability Paper—If the histogram is not and to check each for the normality of the transformed data.
confirmatory in terms of the shape of the distribution, the data Some transformations based on known non-normal
may be examined further to see if it is normally distributed by distributions, or that have been found to work in some
constructing a normal probability plot as described as follows situations, are listed as follows:
(1).3 y = log x y = exp x
y5 œx y = x2
4.4.1 It is easiest to construct a normal probability plot if y = 1/x y5sin21 œx/n
normal probability paper is available. This paper has one linear
axis, and one axis which is arranged to reflect the shape of the where:
cumulative area under the normal distribution. In practice, the y = transformed datum,
“probability” axis has 0.5 or 50 % at the center, a number x = original datum, and
approaching 0 percent at one end, and a number approaching n = number of data points.
1.0 or 100 % at the other end. The marks are spaced far apart Time to failure in stress corrosion cracking usually is best
in the center and close together at the ends. A normal fitted with a log x transformation (2, 3).
probability plot may be constructed as follows with normal
Once a set of transformed data is found that yields an
probability paper.
approximately straight line on a probability plot, the statistical
NOTE 1—Data that plot approximately on a straight line on the procedures of interest can be carried out on the transformed
probability plot may be considered to be normally distributed. Deviations data. Results, such as predicted data values or confidence
from a normal distribution may be recognized by the presence of
deviations from a straight line, usually most noticeable at the extreme ends
intervals, must be transformed back using the reverse transfor-
of the data. mation.
4.4.1.1 Number the data points starting at the largest nega- 4.6 Unknown Distribution—If there are insufficient data
tive value and proceeding to the largest positive value. The points, or if for any other reason, the distribution type of the
numbers of the data points thus obtained are called the ranks of data cannot be determined, then two possibilities exist for
the points. analysis:
4.4.1.2 Plot each point on the normal probability paper such 4.6.1 A distribution type may be hypothesized based on the
that when the data are arranged in order: y (1), y (2), y (3), ..., behavior of similar types of data. If this distribution is not
these values are called the order statistics; the linear axis normal, a transformation may be sought which will normalize
reflects the value of the data, while the probability axis location that particular distribution. See 4.5 above for suggestions.
is calculated by subtracting 0.5 from the number (rank) of that Analysis may then be conducted on the transformed data.
point and dividing by the total number of points in the data set. 4.6.2 Statistical analysis procedures that do not require any
NOTE 2—Occasionally two or more identical values are obtained in a specific data distribution type, known as non-parametric
set of results. In this case, each point may be plotted, or a composite point methods, may be used to analyze the data. Non-parametric tests
may be located at the average of the plotting positions for all the identical do not use the data as efficiently.
values.
4.4.2 If normal probability paper is not available, the 4.7 Extreme Value Analysis—In the case of determining the
location of each point on the probability plot may be deter- probability of perforation by a pitting or cracking mechanism,
mined as follows: the usual descriptive statistics for the normal distribution are
4.4.2.1 Mark the probability axis using linear graduations not the most useful. In this case, Guide G46 should be
from 0.0 to 1.0. consulted for the procedure (4).
4.4.2.2 For each point, subtract 0.5 from the rank and divide 4.8 Significant Digits—IEEE/ASTM SI 10 should be fol-
the result by the total number of points in the data set. This is lowed to determine the proper number of significant digits
the area to the left of that value under the standardized normal when reporting numerical results.
distribution. The cumulative distribution function is the
number, always between 0 and 1, that is plotted on the 4.9 Propagation of Variance—If a calculated value is a
probability axis. function of several independent variables and those variables
4.4.2.3 The value of the data point defines its location on the have errors associated with them, the error of the calculated
other axis of the graph. value can be estimated by a propagation of variance technique.
See Refs (5) and (6) for details.
4.5 Other Probability Paper—If the histogram is not sym-
metrical and bell-shaped, or if the probability plot shows 4.10 Mistakes—Mistakes either in carrying out an experi-
ment or in calculations are not a characteristic of the population
3
The boldface numbers in parentheses refer to a list of references at the end of and can preclude statistical treatment of data or lead to
this standard. erroneous conclusions if included in the analysis. Sometimes

2
 G16 − 13
mistakes can be identified by statistical methods by recogniz- d = the difference between the average and the measured
ing that the probability of obtaining a particular result is very value,
low. n − 1 = the degrees of freedom available.
4.11 Outlying Observations—See Practice E178 for proce- Variance is a useful measure because it is additive in systems
dures for dealing with outlying observations. that can be described by a normal distribution; however, the
dimensions of variance are square of units. A procedure known
5. Central Measures as analysis of variance (ANOVA) has been developed for data
5.1 It is accepted practice to employ several independent sets involving several factors at different levels in order to
(replicate) measurements of any experimental quantity to estimate the effects of these factors. (See Section 9.)
improve the estimate of precision and to reduce the variance of 6.3 Standard Deviation—Standard deviation, σ, is defined
the average value. If it is assumed that the processes operating as the square root of the variance. It has the property of having
to create error in the measurement are random in nature and are the same dimensions as the average value and the original
as likely to overestimate the true unknown value as to measurements from which it was calculated and is generally
underestimate it, then the average value is the best estimate of used to describe the scatter of the observations.
the unknown value in question. The average value is usually
6.3.1 Standard Deviation of the Average—The standard
indicated by placing a bar over the symbol representing the
deviation of an average, Sx̄, is different from the standard
measured variable.
deviation of a single measured value, but the two standard
NOTE 3—In this standard, the term “mean” is reserved to describe a deviations are related as in (Eq 2):
central measure of a population, while average refers to a sample.
S
5.2 If processes operate to exaggerate the magnitude of the Sx̄ 5 (2)
=n
error either in overestimating or underestimating the correct
measurement, then the median value is usually a better where:
estimate. n = the total number of measurements which were used to
5.3 If the processes operating to create error affect both the calculate the average value.
probability and magnitude of the error, then other approaches When reporting standard deviation calculations, it is impor-
must be employed to find the best estimation procedure. A tant to note clearly whether the value reported is the standard
qualified statistician should be consulted in this case. deviation of the average or of a single value. In either case, the
5.4 In corrosion testing, it is generally observed that average number of measurements should also be reported. The sample
values are useful in characterizing corrosion rates. In cases of estimate of the standard deviation is s.
penetration from pitting and cracking, failure is often defined 6.4 Coeffıcient of Variation—The population coefficient of
as the first through penetration and in these cases, average variation is defined as the standard deviation divided by the
penetration rates or times are of little value. Extreme value mean. The sample coefficient of variation may be calculated as
analysis has been used in these cases, see Guide G46. S/x̄ and is usually reported in percent. This measure of
5.5 When the average value is calculated and reported as the variability is particularly useful in cases where the size of the
only result in experiments when several replicate runs were errors is proportional to the magnitude of the measured value
made, information on the scatter of data is lost. so that the coefficient of variation is approximately constant
over a wide range of values.
6. Variability Measures 6.5 Range—The range is defined as the difference between
6.1 Several measures of distribution variability are available the maximum and minimum values in a set of replicate data
which can be useful in estimating confidence intervals and values. The range is non-parametric in nature, that is, its
making predictions from the observed data. In the case of calculation makes no assumption about the distribution of
normal distribution, a number of procedures are available and error. In cases when small numbers of replicate values are
can be handled with computer programs. These measures involved and the data are normally distributed, the range, w,
include the following: variance, standard deviation, and coef- can be used to estimate the standard deviation by the relation-
ficient of variation. The range is a useful non-parametric ship:
estimate of variability and can be used with both normal and w
other distributions. S. , n,12 (3)
=n
6.2 Variance—Variance, σ2, may be estimated for an experi-
where:
mental data set of n observations by computing the sample
estimated variance, S2, assuming all observations are subject to S = the estimated sample standard deviation,
the same errors: w = the range, and
n = the number of observations.
S2 5 (d2
(1) The range has the same dimensions as standard deviation. A
n21
tabulation of the relationship between σ and w is given in Ref
where: (7).

3
 G16 − 13
6.6 Precision—Precision is closeness of agreement between 7.2 Degrees of Freedom—The degrees of freedom of a
randomly selected individual measurements or test results. The statistical test refer to the number of independent measure-
standard deviation of the error of measurement may be used as ments that are available for the calculation.
a measure of imprecision. 7.3 t Test—The t statistic may be written in the form:
6.6.1 One aspect of precision concerns the ability of one
investigator or laboratory to reproduce a measurement previ- t5
? x̄ 2 µ ? (4)
S ~ x̄ !
ously made at the same location with the same method. This
aspect is sometimes called repeatability. where:
6.6.2 Another aspect of precision concerns the ability of x̄ = the sample average,
different investigators and laboratories to reproduce a measure- µ = the population mean, and
ment. This aspect is sometimes called reproducibility. S(x̄) = estimated standard deviation of the sample average.
6.7 Bias—Bias is the closeness of agreement between an The t distribution is usually tabulated in terms of significance
observed value and an accepted reference value. When applied levels and degrees of freedom.
to individual observations, bias includes a combination of a 7.3.1 The t test may be used to test the null hypothesis:
random component and a component due to systematic error. m5µ (5)
Under these circumstances, accuracy contains elements of both
precision and bias. Bias refers to the tendency of a measure- For example the value m is not significantly different than µ,
ment technique to consistently under- or overestimate. In cases the population mean. The t test is then:
where a specific quantity such as corrosion rate is being ? x̄ 2 m ?
t5 (6)
estimated, a quantitative bias may be determined.
6.7.1 Corrosion test methods which are intended to simulate
service conditions, for example, natural environments, often
S~x! Π1
n

are more severe on some materials than others, as compared to The calculated value of t may be compared to the value of t
the conditions which the test is simulating. This is particularly for the degrees of freedom, n, and the significance level.
true for test procedures which produce damage rapidly as 7.3.2 The t statistic may be used to obtain a confidence
compared to the service experience. In such cases, it is interval for an unknown value, for example, a corrosion rate
important to establish the correspondence between results from value calculated from several independent measurements:
the service environment and test results for the class of material ~ x̄ 2 t S ~ x̄ !! ,µ, ~ x̄1t S ~ x̄ !! (7)
in question. Bias in this case refers to the variation in the
acceleration of corrosion for different materials. where:
6.7.2 Another type of corrosion test method measures a tS(x̄) = one half width confidence interval associated with the
characteristic that is related to the tendency of a material to significance level chosen.
suffer a form of corrosion damage, for example, pitting 7.3.3 The t test is often used to test whether there is a
potential. Bias in this type of test refers to the inability of the significant difference between two sample averages. In this
test to properly rank the materials to which the test applies as case, the expression becomes:
compared to service results. Ranking may also be used as a
qualitative estimate of bias in the test method types described t5
? x̄ 1 2 x̄ 2 ? (8)
in 6.7.1. S ~x! =1/n 1 11/n 2
where:
7. Statistical Tests
x̄1 and x̄2 = sample averages,
7.1 Null Hypothesis Statistical Tests are usually carried out n1 and n2 = number of measurements used in calculating x̄1
by postulating a hypothesis of the form: the distribution of data and x̄2 respectively, and
under test is not significantly different from some postulated S(x) = pooled estimate of the standard deviation from
distribution. It is necessary to establish a probability that will both sets of data.
be acceptable for rejecting the null hypothesis. In experimental i.e.:
work it is conventional to use probabilities of 0.05 or 0.01 to
reject the null hypothesis.
7.1.1 Type I errors occur when the null hypothesis is
S~x! 5 Π~ n 1 2 1 ! S 2~ x 1! 1 ~ n 2 2 1 ! S 2~ x 2!
n 1 1n 2 2 2
(9)

rejected falsely. The probability of rejecting the null hypothesis 7.3.4 One sided t test. The t function is symmetrical and can
falsely is described as the significance level and is often have negative as well as positive values. In the above
designated as α. examples, only absolute values of the differences were dis-
7.1.2 Type II errors occur when the null hypothesis is cussed. In some cases, a null hypothesis of the form:
accepted falsely. If the significance level is set too low, the
µ.m (10)
probability of a Type II error, β, becomes larger. When a value
of α is set, the value of β is also set. With a fixed value of α, or
it is possible to decrease β only by increasing the sample size
assuming no other factors can be changed to improve the test. µ,m

4
 G16 − 13
may be desired. This is known as a one sided t test and the then no sign is recorded. The total number of plus signs, P, and
significance level associated with this t value is half of that for minus signs, N, is computed. Significance is determined by the
a two sided t. following test:
7.4 F Test—Labeling the variable with the larger observed ? P 2 N ? .k =P1N (13)
variance as x1, the F statistic is used to test whether the
where k = a function of significance level as follows:
variance associated with that variable is significantly larger
than the variable associated with variable x2. The F statistic is k Significance Level
__ ________________
then: 1.6 0.10
2.0 0.05
S 2~ x 1! 2.6 0.01
Fx1 , x 2 5 (11)
S 2~ x 2!
The sign test does not depend on the magnitude of the
The F test is an important component in the analysis of difference and so can be used in cases where normal statistics
variance used in experimental designs. Values of F are tabu- would be inappropriate or impossible to apply.
lated for significance levels and degrees of freedom for both 7.7 Outside Count—The outside count test is a useful
variables. In cases where the data are not normally distributed, non-parametric technique to evaluate whether the magnitude of
the F test approach may falsely show a significant effect one of two data sets of approximately the same number of
because of the non-normal distribution rather than an actual values is significantly larger than the other. The details of the
difference in variances being compared. procedure may be found elsewhere (8).
7.5 Correlation Coeffıcient—The correlation coefficient, r, 7.8 Corner Count—The corner count test is a non-
is a measure of a linear association between two random parametric graphical technique for determining whether there
variables. Correlation coefficients vary between −1 and +1 and is correlation between two variables. It is simpler to apply that
the closer to either −1 or +1, the better the correlation. The sign the correlation coefficient, but requires a graphical presentation
of the correlation coefficient simply indicates whether the of the data. The detailed procedure may be found elsewhere
correlation is positive (y increases with x) or negative (y (8).
decreases as x increases). The correlation coefficient, r, is given
by: 8. Curve Fitting—Method of Least Squares

@ ( ~ x i 2 x̄ !~ y i 2 ȳ ! # 8.1 It is often desirable to determine the best algebraic

r5 1 expression to fit a data set with the assumption that a normally
@ ( ~ x i 2 x̄ ! 2 ( ~ y i 2 ȳ ! #
2 2
distributed random error is operating. In this case, the best fit
~ ( x i y i ! 2 nx̄ȳ will be obtained when the condition of minimum variance
5 1 (12) between the measured value and the calculated value is
$ @ ( ~ x i ! 2 n x̄ 2 # @ ( ~ y i 2 !
2
2 n ȳ 2 # % 2
obtained for the data set. The procedures used to determine
where: equations of best fit are based on this concept. Software is
xi = observed values of random variable x, available for computer calculation of regression equations,
yi = observed values of random variable y, including linear, polynomial, and multiple variable regression
x̄ = average value of x, equations.
ȳ = average value of y, and 8.2 Linear Regression—2 Variables—Linear regression is
n = number of observations. used to fit data to a linear relationship of the following form:
Generally, r2 values are preferred because they avoid the y 5 mx1b (14)
problem of sign and the r2 values relate directly to variance.
Values of r or r2 have been tabulated for different significance In this case, the best fit is given by:
levels and degrees of freedom. In general, it is desirable to m 5 ~n 2 2
( xy 2 ( x ( y ! / @ n ( x 2 ~ ( x ! # (15)
report values of r or r2 when presenting correlations and
1
regression analyses. b 5 @( x 2 m( y# (16)
n
NOTE 4—The procedure for calculating correlation coefficient does not where:
require that the x and y variables be random and consequently, some
investigators have used the correlation coefficient as an indication of y = the dependent variable
goodness of fit of data in a regression analysis. However, the significance x = the independent variable,
test using correlation coefficient requires that the x and y values be m = the slope of the estimated line,
independent variables of a population measured on randomly selected b = the y intercept of the estimated line,
samples. ∑x = the sum of x values and so forth, and
7.6 Sign Test—The sign test is a non-parametric test used in n = the number of observations of x and y.
sets of paired data to determine if one component of the pair is This standard deviation of m and the standard error of the
consistently larger than the other (8). In this test method, the expression are often of interest and may be calculated easily (5,
values of the data pairs are compared, and if the first entry is 7, 9). One problem with linear regression is that all the errors
larger than the second, a plus sign is recorded. If the second are assumed to be associated with the dependent variable, y,
term is larger, then a minus sign is recorded. If both are equal, and this may not be a reasonable assumption. A variation of the

5
 G16 − 13
linear regression approach is available, assuming the fitting y = the observed dependent
equation passes through the origin. In this case, only one variable.
adjustable parameter will result from the fit. It is possible to use Because of the complexity of this problem, it is generally
statistical tests, such as the F test, to compare the goodness of handled with the help of a computer. One strategy is to
fit between this approach and the two adjustable parameter fits compute the value of all the “b’s,” together with standard
described above. deviation for each “b.” It is usually necessary to run several
8.3 Polynomial Regression—Polynomial regression analy- regression analysis, dropping variables, to establish the relative
sis is used to fit data to a polynomial equation of the following importance of the independent variables under consideration.
form: 9. Comparison of Effects—Analysis of Variance
2
y 5 a1bx1cx 1dx and so forth 3
(17) 9.1 Analysis of variance is useful to determine the effect of
a number of variables on a measured value when a small
where:
number of discrete levels of each independent variable is
a, b, c, d = adjustable constants to be used to fit the data set, studied (5, 7, 9, 10, 11). This is best handled by using a
x = the observed independent variable, and factorial or similar experimental design to establish the mag-
y = the observed dependent variable. nitude of the effects associated with each variable and the
The equations required to carry out the calculation of the magnitude of the interactions between the variables.
best fit constants are complex and best handled by a computer. 9.2 The two-level factorial design experiment is an excel-
It is usually desirable to run a series of expressions and lent method for determining which variables have an effect on
compute the residual variance for each expression to find the the outcome.
simplest expression fitting the data. 9.2.1 Each time an additional variable is to be studied, twice
as many experiments must be performed to complete the
8.4 Multiple Regression—Multiple regression analysis is
two-level factorial design. When many variables are involved,
used when data sets involving more than one independent
the number of experiments becomes prohibitive.
variable are encountered. An expression of the following form 9.2.2 Fractional replication can be used to reduce the
is desired in a multiple linear regression: amount of testing. When this is done, the amount of informa-
y 5 a1b 1 x 1 1b 2 x 2 1b 3 x 3 and so forth (18) tion that can be obtained from the experiment is also reduced.
where: 9.3 In the design and analysis of interlaboratory test
programs, Practice E691 should be consulted.
a, b1, b2, b3, and so forth = adjustable constants used to ob-
tain the best fit of the data set 10. Keywords
x1, x2, x3, and so forth = the observed independent vari-
10.1 analysis of variance; corrosion data; curve fitting;
ables
statistical analysis; statistical tests

APPENDIX

(Nonmandatory Information)

X1. SAMPLE CALCULATIONS

X1.1 Calculation of Variance and Standard Deviation The standard deviation is:
X1.1.1 Data—The 27 values shown in Table X1.1 are s ~ x ! 5 ~ 0.0135! 1/2 5 0.116 (X1.3)
calculated mass loss based corrosion rates for copper panels in
a one year rural atmospheric exposure. The coefficient of variation is:
X1.1.2 Calculation of Statistics: 0.116
3 100 5 5.75 % (X1.4)
X1.1.2.1 Let xi = corrosion rate of the ith panel. The average 2.016
corrosion rate of 27 panels, x̄:
The standard deviation of the average is:
x̄ 5 (
x 54.43
i
5 5 2.016 (X1.1) 0.116
n 27 s ~ x̄ ! 5 1 (X1.5)
2
~ 27! 2 5 0.0223
The variance estimate based on this sample, s (x):
The range, w, is the difference between the largest and
2
s ~x! 5
(x i
2
2 nx̄ 2
5 (X1.2) smallest values:
n21
w 5 2.21 2 1.70 5 0.41 (X1.6)
110.085 2 27 3 ~ 2.016! 2 0.350
5 5 0.0135 The mid-range value is:
26 26

6
 G16 − 13
TABLE X1.1 Copper Corrosion Rate—One-Year Exposure is 100r/n + 1 (see Guide G46). The median is the corrosion rate at the
Plotting Position 50 % plotting position and is 2.03 for panel 142.
Panel CR (mm/yr) Rank
(%)
121 2.16 25 90.74 X1.3 Probability Paper Plot of Data: See Table X1.1
122 2.21 27 98.15
123 2.15 24 87.04
X1.3.1 The corrosion rate is plotted versus plotting position
131 2.05 16.5 59.26 on probability paper, see Fig. X1.1.
132 2.06 19 68.52
133 2.04 15 53.70 X1.3.2 Normal Distribution Plotting Position Reference:
141 1.90 5 16.67 X1.3.2.1 In order to compare the data points shown in Fig.
142 2.03 14 50.00 X1.1 to what would be expected for a normal distribution, a
143 2.06 19 68.52
151 2.02 12.5 44.44 straight line on the plot may be constructed to show a normal
152 2.06 19 68.52 distribution.
153 1.92 6.5 22.22 (1) Plot the average value at 50 %, 2.016 at 50 %.
161 2.08 21 75.93
162 2.05 16.5 59.26 (2) Plot the average +1 standard deviation at 84.13 %, that
163 1.88 3.5 11.11 is, 2.016 + 0.116 = 2.136 at 84.13 %.
211 1.99 9.5 33.33 (3) Plot the average −1 standard deviation at 15.87 %, that
212 2.01 11 38.89
213 1.86 2 5.56 is, 2.016 − 0.116 = 1.900 at 15.87 %.
411 1.70 1 1.85 (4) Connect these three points with a straight line.
412 1.88 3.5 11.11
413 1.99 9.5 33.33
X11 1.93 8 27.78
X1.4 Evaluation of Outlier
X12 2.20 26 94.44 X1.4.1 Data—See X1.1, Table X1.1, and Fig. X1.1.
X13 2.02 12.5 44.44
071 1.92 6.5 22.22 X1.4.2 Is the 1.70 result (panel 411) an outlier? Note that
072 2.13 22.5 81.48
073 2.13 22.5 81.48
this point appears to be out of line in Fig. X1.1.
X1.4.3 Reference Practice E178 (Dixon’s Test)—We choose
α = 0.05 for this example, that is, the probability that this point
2.2111.70
could be this far out of line based on normal probability is 5 %
2
5 1.955 (X1.7) or less.
X1.4.4 Number of data points is 27:
X1.2 Calculation of Rank and Plotting Points for Prob-
x3 2 x 1 1.88 2 1.70
ability Paper Plots r 22 5 5 5 0.391 (X1.9)
x n22 2 x 1 2.16 2 1.70
X1.2.1 The lowest corrosion rate value (1.70) is assigned a
rank, r, of 1 and the remaining values are arranged in ascending The Dixon Criterion at α = 0.05, n = 27 is 0.393 (see
order. Multiple values are assigned a rank of the average rank. Practice E178, Table 2).
For example, both the third and fourth panels have corrosion X1.4.4.1 The r22 value does not exceed the Dixon Criterion
rates of 1.88 so that the rank is 3.5. See the third column in for the value of n and the value of α chosen so that the 1.70
Table X1.1. value is not an outlier by this test.
X1.4.4.2 Practice E178 recommends using a T test as the
X1.2.2 The plotting positions for probability paper plots are best test in this case:
expressed in percentages in Table X1.1. They are derived from
the rank by the following expression: x̄ 2 x 1 2.016 2 1.70
T1 5 5 5 2.72 (X1.10)
s 0.116
Plotting position 5 100 ~ r 2 1/2 ! /expressed as percent (X1.8)
Critical value T for α = 0.05 and n = 27 is 2.698 (Practice
See Table X1.1, fourth column, for plotting positions for this E178, Table 1). Therefore, by this criterion the 1.70 value is an
data set. outlier because the calculated T1 value exceeds the critical T
NOTE X1.1—For extreme value statistics the plotting position formula value.

7
 G16 − 13

FIG. X1.1 Probability Plot for Corrosion Rate of Copper Panels in a 1-Year Rural Atmospheric Exposure

X1.4.5 Discussion: 0.101

s ~ x̄ ! 5 5 0.0198
X1.4.5.1 The 1.70 value for panel 411 does appear to be out =26
of line as compared to the other values in this data set. The T
test confirms this conclusion if we choose α = 0.05. The next Median 5 2.035
step should be to review the calculations that lead to the
determination of a 1.70 value for this panel. The original and w 5 2.21 2 1.86 5 0.35
final mass values and panel size measurements should be
2.2111.86
checked and compared to the values obtained from the other Mid range 5 5 2.035
2
panels.
X1.4.5.2 If no errors are found, then the panel itself should The average, median, and mid range are closer together
be retrieved and examined to determine if there is any evidence excluding the 1.70 value, as expected, although the changes are
of corrosion products or other extraneous material that would relatively small. In cases where deviations occur on both ends
cause its final mass to be greater than it should have been. If a of the distribution, a different procedure is used to check for
reason can be found to explain the loss mass loss value, then outliers. Please refer to Practice E178 for a discussion of this
the result can be excluded from the data set without reserva- procedure.
tion. If this point is excluded, the statistics for this distribution
become: X1.5 Confidence Interval for Corrosion Rate
x̄ 5 2.028 (X1.11)
X1.5.1 Data—See X1.1, Table X1.1, and X1.4.1, excluding
2
s ~ x ! 5 0.0102 the panel 411 result.
Significance level α 5 0.05 (X1.12)
s ~ x ! 5 0.101
Confidence interval calculation:
0.101
Coefficient of variation 5 3 100 5 4.98 %
2.028 Confidence interval 5 x̄6ts~ x̄ !

8
 G16 − 13
t for α 5 0.05, DF 5 25; is 2.060 X1.6.4.3 Calculation of t statistic:
x̄ h 2 x̄ p
95 % confidence interval for the average corrosion rate. t5 (X1.17)

x̄6 ~ 2.060!~ 0.0198! 5 x̄60.041 or 1.987 to 2.069

F
s p~ x !
1
1
np nh
1
G 1/2

Note that this interval refers to the average corrosion rate. If 2.55 2 2.24 0.31
t5 5 5 2.83 (X1.18)
one is interested in the interval in which 95 % of measurements
of the corrosion rate of a copper panel exposed under those =0.018 Π1 1
1
3 3
0.110

identical conditions will fall, it may be calculated as follows: DF 5 212 5 4 (X1.19)

x̄6ts~ x̄ ! (X1.13) X1.6.5 Conclusion—The critical value of t for α = 0.05 and
x̄6 ~ 2.060!~ 0.101! 5 x̄60.208 or 1.820 to 2.236
DF = 4 is 2.132. The calculated value for t exceeds the critical
value and therefore the null hypothesis can be rejected, that is,
X1.6 Difference Between Average Values the helices are corroding at a significantly higher rate than the
panels. Note that the critical t value above is listed for α = 0.1
X1.6.1 Data—Triplicate zinc flat panels and wire helices
most tables. This is because the tables are set up for a two-sided
were exposed for a one year period at the 250 m lot at Kure
t test, and this example is for a one-sided test, that is, is xh > xp?
Beach, NC. The corrosion rates were calculated from the loss
in mass after cleaning the specimens. The corrosion rate values X1.6.6 Discussion—Usually the α level for the F test shown
are given in Table X1.2. in X1.6.4.1 should be carried out at a more stringent signifi-
cance level than in the t test, for example, 0.01 rather than 0.05.
X1.6.2 Statistics:
In the event that the F test did show a significant difference
Panel Average x̄p = 2.24 then a different procedure must be used to carry out the t test.
Panel Standard Deviation = 0.18
Helix Average: x̄h = 2.55 It is also desirable to consider the power of the t test. Details on
Helix Standard Deviation = 0.066 these procedures are beyond the scope of this appendix but are
X1.6.3 Question—Are the helices corroding significantly covered in Ref (10).
faster than the panels? The null hypothesis is therefore that the X1.7 Curve Fitting—Regression Analysis Example
panels and helices are corroding at the same or lower rate. We
will choose α = 0.05, that is, the probability of erroneously X1.7.1 The mass loss per unit area of zinc is usually
rejecting the null hypothesis is one chance in twenty. assumed to be linear with exposure time in atmospheric
exposures. However, most other metals are better fitted with
X1.6.4 Calculations: power function kinetics in atmospheric exposures. An exposure
X1.6.4.1 Note that the standard deviations for the panels program was carried out with a commercial purity rolled zinc
and helices are different. If they are not significantly different alloy for 20 years in an industrial site. How can the mass loss
then they may be pooled to yield a larger data set to test the results be converted to an expression that describes the results?
hypothesis. The F test may be used for this purpose.
X1.7.2 Experimental—Forty panels of 16 gauge rolled zinc
s 2~ x p! ~ 0.18! 2 strips were cut to approximately 4 in. to 6 in. in size (100 by
F5 2 5 5 7.438 (X1.14)
s ~ x h ! ~ 0.066! 2 150 m). The panels were cleaned, weighed, and exposed at the
The critical F for α = 0.05 and both numerator and denomi- same time. Five panels were removed after 0.5, 1, 2, 4, 6, 10,
nator degrees of freedom of 2 is 19.00. The calculated F is less 15, and 20 years exposure. The panels were then cleaned and
than the critical F value so that the hypothesis that the two reweighed. The mass loss values were calculated and con-
standard deviations are not significantly different may be verted to mass loss per unit area. The results are shown in Table
accepted. As a consequence, the standard deviations may be X1.3 below:
pooled. X1.7.3 Analysis—Corrosion of zinc in the atmosphere is
X1.6.4.2 Calculation of pooled variance, s2p(x): usually assumed to be a constant rate process. This would
~ n p 2 1 ! s 2~ x p! 1 ~ n h 2 1 ! s 2~ x h! imply that the mass loss per unit area m is related to exposure
s 2p~ x ! 5 (X1.15) time T by:
~n p 2 1!1~nh 2 1!
m 5 k 1T (X1.20)
substituting:
2 ~ 0.18! 2 12 ~ 0.066! 2 where:
s 2p~ x ! 5 5 0.018 (X1.16) k 1 = is the corrosion rate.
212
Most other metals are better fitted by a power function such
as:
TABLE X1.2 Corrosion Rate Values
m 5 kTb (X1.21)
Corrosion rates, CR, of zinc alloy after one year of
atmospheric exposure at the 250 m lot at Kure Beach, µm/year
where:
Panel ID CR Helix ID CR
I3A111P 2.04 I3A111H 2.49 k = is the mass loss coefficient and b is the time exponent.
I3A112P 2.30 I3A112H 2.54
I3A113P 2.38 I3A113H 2.62 The data in Table X1.3 may be handled in several ways.
Linear regression can be applied to yield a value of k1 that

9
 G16 − 13
TABLE X1.3 Mass Loss per Unit Area, Zinc in the Atmosphere (All values in mg/cm2)
Exposure
Panel Designation
Duration
Years 1 2 3 4 5 Avg Std Dev

0.5 0.387 0.392 0.362 0.423 0.319 0.3766 0.0388

1 0.829 0.759 0.801 0.738 0.780 0.7814 0.0355
2 1.793 1.667 1.585 1.727 1.642 1.6828 0.0800
4 3.688 3.406 3.297 3.280 3.297 3.3936 0.1720
6 5.825 5.257 5.391 5.280 5.333 5.4172 0.2338
10 10.759 9.772 9.653 9.966 9.835 9.9970 0.4407
15 15.440 14.557 15.102 14.910 Lost 15.0022 0.3689
20 20.700 19.507 18.963 19.336 18.658 19.4326 0.7812

minimizes the variance for the constant rate expression above, X1.7.4 Calculations—The values in Table X1.4 were used
or any linear expression such as: to calculate the following:
m 5 a1k 2 T (X1.22) ∑x = 23.11056
∑y = 20.92232
where:
n = 39
a = is a constant.
∑x2 = 24.742305
Alternatively, a nonlinear regression analysis may be used
∑y2 = 24.159116
that yields values for k and b that minimize the variance from
the measured values to the calculated value for m at any time ∑xy = 24.341352
using the power function above. All of these approaches ~(x!2
( 'x 2 5 ( x 2 2 n
assume that the variance observed at short exposure times is
comparable to variances at long exposure times. However, the x̄ = 0.592758
data in Table X1.3 shows standard deviations that are roughly ȳ = 0.536470
proportional to the average value at each time, and so the ~ 23.11056! 2
assumption of comparable variance is not justified by the data ( 'x 2 524.7423052 39
511.047485
2
at hand. ~ 20.92232 !
Another approach to handle this problem is to employ a ( 'y 2 524.1591162 39 512.934924
logarithmic transformation of the data. A transformed data set ~ 23.11056!~ 20.92232!
is shown in Table X1.4 where x = log T and y = log m. These
( 'xy524.3413522 39
511.943236
2
data may be handled in a linear regression analysis. Such an ~ ( 'xy!
analysis is equivalent to the power function fit with the k and
( 'C 2 5 ( 'x 2 512.911616
b values minimizing the variance of the transformed variable, ∑∑'yi2 = 0.017810
y. ( 'xy 5 11.943236 51.08108
b5
The logarithmic transformation becomes: ( 'x 2 11.047485
logm 5 logk1blogT (X1.23) a = ȳ − bx̄ = 0.53647 − 1.08108(0.592578) = −0.10416
or k = 0.7868
y 5 a1bx (X1.24) X1.7.5 Analysis of Variance—One approach to test the
adequacy of the analysis is to compare the residual variance
where:
from the regression to the error variance as estimated by the
a = log k. variance found in replication. The null hypothesis in this case
Note that the standard deviation values, s(yi), in Table X1.4 is that the residual variance from the calculated regression
are approximately constant for both short and long exposure expression is not significantly greater than the replication
times. variance.

TABLE X1.4 Log of Data from Table X1.3

i xi yi1 yi2 yi3 yi4 yi5 ȳi Std Dev
1 −0.30103 −0.41229 −0.40671 −0.44129 −0.37366 −0.49621 −0.42603 0.04600
2 0.00000 −0.08144 −0.11976 −0.09637 −0.13194 −0.10790 −0.10748 0.01968
3 0.30103 0.25358 0.22194 0.20003 0.23729 0.21537 0.22564 0.02056
4 0.60206 0.56679 0.53224 0.51812 0.51587 0.51812 0.53023 0.02144
5 0.77815 0.76530 0.72074 0.73167 0.72263 0.72697 0.73346 0.01829
6 1.00000 1.03177 0.98998 0.98466 0.99852 0.99277 0.99954 0.01870
7 1.17609 1.18865 1.16307 1.17903 1.17348 Lost 1.17606 0.01069
8 1.30103 1.31597 1.29019 1.27791 1.28637 1.27086 1.28826 0.01721

10
 G16 − 13
TABLE X1.5 Analysis of Variance
Item Expression Symbol SOS DF MS
xy regression (^'xy)2/^'x2 ^'C2 12.911616 1 12.911616
Residual from xy line ^'y2 − ^'C 2 − ^ ^ 'y2 ^'ŷ2 0.005498 6 0.000916
Replication variance ... ^ ^'y2 0.017810 31 0.000575
SOS = sum of squares value
DF = degrees of freedom
MS = mean squares value
F test on hypothesis in X1.7.5:
0.000916
F5 5 1.59
0.000575

The critical F value at an α value of 0.05 and 6/31 degrees of freedom is 2.41. Therefore, the residual variance from the regression expression appears to be ho-
mogeneous with the replication error variance when tested at the 5 % level. Thus, the regression model estimated may be used to describe the results for the test
time period.

X1.7.6 Comparison of Power Function to Linear CI 5 .71072 to .75620

Kinetics—Is the expression found better than a linear expres-
sion? A linear kinetics function would have a slope of one after Converting y to m: CI 5 5.137 to 5.704 mg/cm2
transformation to a log function. Therefore, this question can Calculation of the confidence interval for the regression
be reformulated to: Is the b value significantly different from expression at exposure time T = 6, α = 0.05, DF = 6, t = 2.45,
one? The null hypothesis is therefore that the b value is not i = 5:
different from 1.000 with α = 0.05. A t test will be used:
b 5 1.08108 (X1.25) s 2 ~ ŷ i ! 5 s 2 ~ y ! F 1 ~ x̄ 2 x i ! 2
n
1
( 'x i 2
G (X1.27)

s 2~b! 5 ( 'ŷ 2
5
.005498
5 0.0000829 ( 'x 5 11.0475
i
2

~ n 2 2 ! ( 'x 2 11.04749

~ ŷ ! 5 ( 5
2
2
'ŷ 0.005498
s ~ b ! 5 0.00911 s 5 0.000916
n22 6
Note that a pooled estimate of this variance could have been
1.08108 2 1.00000
t5 5 8.90 used.
0.00911
where:
The critical t at 6 degrees of freedom and α = 0.05 is 2.45, x̄ = 0.59258 x5= log 6 = .77815
which is smaller than the calculated t value above. Therefore
the null hypothesis may be rejected, and the slope is signifi-
s 2 ~ ŷ 5 ! 5 0.000916 F 1 ~ 0.59258 2 0.77815! 2
6
1
11.0475 G 5 .0001556
cantly different from one.
where:
X1.7.7 Confidence Interval for Regression: s(ŷ5) = = 0.01247
X1.7.7.1 A confidence interval calculated from the replicate ŷ5 = −0.10416 + 1.08108 x 5 = −0.10416 + 1.08108
information at each exposure time represents an interval in (.77815) = .73708
which the unknown mean mass loss value will be located CIyi = ŷi 6 ts(ŷi) = .737086 2.45(.01247) = .70653 to
unless a 1 in 20 chance has occurred in the sampling of this .76763
experiment. On the other hand, a confidence interval calculated CIm = 5.088 to 5.856
from the regression results represents an interval that will cover
Note that the confidence interval calculated from the regres-
the unknown regression line unless a 1 in 20 chance has
sion is slightly larger than that calculated from the replicate
occurred in the sampling of this experiment.
values at that exposure time.
X1.7.7.2 The confidence interval for each exposure time is
equally spaced around the average of the log values. This will X1.7.7.3 Fig. X1.2 is a log-log plot showing the regression
also be true for the regression confidence interval. However, equation with the 95 % confidence interval for the regression
when these intervals are plotted on linear coordinates the shown as dashed lines and the averages and corresponding
interval will appear to be unsymmetrical. An example of a confidence intervals shown as bars. Fig. X1.3 shows the same
confidence interval calculation is shown as follows: information on linear coordinates.
Calculation of the confidence interval, CI, for the average X1.7.8 Other Statistics from the Regression—Standard error
value: of estimate, s(ŷ) for the logarithmic expression.
Exposure time, T 5 years, α 5 0.05, DF 5 4, t 5 2.78
s ~ ŷ ! 5 =.000916 5 .03026 (X1.28)
(X1.26)
Correlation Coefficient, R, for the logarithmic expression.
ȳ 5 5 0.73346 s ~ y 5 ! 5 0.01829
~ ( 'xy! 2 ~ 11.943236! 2
R2 5 5 5 .998198
CI 5 ȳ6
ts
5 ȳ6
2.78 3 0.01829
5 ȳ6.02274 ( 'x ( 'y 2 2 11.047485 3 12.934924
=n =5 (X1.29)

11
 G16 − 13

FIG. X1.2 Log Plot of Mass Loss versus Exposure Time for Replicate Rolled Zinc Panels in an Industrial Atmosphere

R 5 .9991 (X1.30) be a reasonable estimate of mass loss performance for rolled

2
Note that R or R is often quoted as a measure of the quality zinc in this atmosphere. However, neither the linear nor a
of fit of a regression expression. However, it should be noted nonlinear power function regression analysis will yield a
that the correlation coefficient calculated for the logarithmic confidence interval that matches as closely the replicate data
expression is not comparable to a correlation coefficient confidence intervals as the logarithmic transformation shown
calculated for a nontransformed regression. above.
X1.7.9 Discussion: X1.7.9.3 The regression expression can be used to project
X1.7.9.1 The use of a log transformation to obtain a power future results by extrapolation of the results beyond the range
function fit is convenient and simple but has some limitations. of data available. This type of calculation is generally not
The log transformation tends to depress the calculated values to advisable unless there is good information indicating that the
the low side of the linear average. It also produces a nonlinear procedure is valid, that is, that no changes have occurred in any
error function. In the example above the use of a log transfor- of the environmental and surface conditions that govern the
mation produces an almost constant standard deviation over the kinetics of the corrosion reaction.
range of exposure times.
X1.7.9.2 A linear regression analysis also may be used with
these mass loss results, and the corresponding expression may

12
 G16 − 13

FIG. X1.3 Linear Plot of Mass Loss versus Exposure Time for Replicate Rolled Zinc Panels in an Industrial Atmosphere

13
 G16 − 13

REFERENCES

(1) Tufte, E. R., The Visual Display of Quantitative Information, Graphic Mathematics in Chemical Engineering, 2nd ed., McGraw-Hill, New
Press, Cheshire, CT, 1983. York, NY, 1957, pp. 46–99.
(2) Booth, F. F., and Tucker, G. E. G., “Statistical Distribution of (7) Snedecor, G. W., Statistical Methods Applied to Experiments in
Endurance in Electrochemical Stress-Corrosion Tests,” Corrosion, Agriculture and Biology, 4th Edition, Iowa State College Press, Ames,
CORRA, Vol 21, 1965, pp. 173–177. IA, 1946.
(3) Haynie, F. H., Vaughan, D. A., Phalen, D. I., Boyd, W. K., and Frost, (8) Brown, B. S., “6 Quick Ways Statistics Can Help You,” Chemical
P. D., A Fundamental Investigation of the Nature of Stress-Corrosion Engineers Calculation and Shortcut Deskbook, X254, McGraw-Hill,
Cracking in Aluminum Alloys, ARML-TR-66-267, June 1966. New York, NY, 1968, pp. 37–43.
(4) Aziz, P. M., “Application of Statistical Theory of Extreme Values to (9) Freeman, H. A., Industrial Statistics, John Wiley & Sons, Inc., New
the Analysis of Maximum Pit Depth Data for Aluminum,” Corrosion, York, NY, 1942.
CORRA, Vol 12, 1956, pp. 495–506t. (10) Davies, O. L., ed., Design and Analysis of Industrial Experiments,
(5) Volk, W., Applied Statistics for Engineers, 2nd ed., Robert E. Krieger Hafner Publishing Co., New York, NY, 1954.
Publishing Company, Huntington, NY, 1980. (11) Box, G. E. P., Hunter, W. G., and Hunter, J. S. Statistics for
(6) Mickley, H. S., Sherwood, T. K., and Reed, C. E., Editors, Applied Experimenters, Wiley, New York, NY, 1978.

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

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

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

14