Statistical Primer for Cardiovascular Research

Rank Score Tests

Lisa M. LaVange, PhD; Gary G. Koch, PhD

N onparametric statistical methods are useful tools for data

analysis when there is reason to believe that the outcome
variables of interest may fail certain distributional assump-
bution of difference values {di}, then the null hypothesis is
that ⌬⫽0. That is, the null hypothesis is the hypothesis that
the true, underlying difference between the 2 conditions is
tions required for parametric methods. Variables may be zero. The following steps describe the calculation of the test
ordered categories in nature and thereby not suitable for statistic to assess the null hypothesis against the alternative of
analysis methods that assume normally distributed variables, a shift in location associated with one of the conditions or
such as t tests or analyses of variance and covariance. time points:
Variables may also be metric or continuous but subject to
excessive variability or the presence of outliers. When the 1. Differences {di} are formed for each pair of observations
research hypothesis involves comparing a sample of subjects on each subject.
under 2 conditions or at 2 time points or comparing 2 samples 2. The ranks of the absolute values of the nonzero differences
of subjects with respect to an outcome variable of interest, are computed, with ties assigned the average of the
then univariate nonparametric methods based on rank score applicable ranks (called midranks); zero differences are
tests can be invoked. A study design feature such as random ignored.
3. The signs of the differences are computed as ⫺1 or 1.
assignment of conditions or treatments is typically all that is
4. Signed ranks are then computed by multiplying the sign of
required for these methods to be valid. Furthermore, the the difference by the corresponding rank.
methods can be quite powerful under a number of alterna- 5. The test statistic is calculated by dividing the sum of the
tives, particularly those involving shifts in the median. signed ranks U by the square root of the sum of squares of
The Wilcoxon signed rank test, the Spearman rank corre- the signed ranks S to form U/S; S is the standard deviation
lation coefficient, and the Wilcoxon rank sum test are among of U.
the most commonly used nonparametric tests and cover a
variety of research questions. These tests are described here. The null hypothesis that the median difference is zero is
Downloaded from http://ahajournals.org by on July 16, 2024

Although the focus is on hypothesis testing, related methods assessed by comparing the test statistic U/S to critical values
for estimation of confidence intervals are also presented. of the standard normal distribution for large sample sizes (eg,
Extensions of nonparametric methods to handle stratification nⱖ20) or by tabulating the exact critical region for small
and covariate adjustment are also described. Scenarios in sample sizes. Computation of P values from critical values
which nonparametric methods may be most useful and the for both the normal approximation and the exact distribution
power they can be expected to yield are discussed. The is available through commercial statistical software packages
methods are illustrated with data from a clinical trial assess- (eg, SAS Proc Univariate3 and StatXact4).
ing the impact of exposure to low levels of carbon monoxide When there are no ties among the observed differences and
on exercise capacity in patients with ischemic heart disease. no differences equal to zero, then the signed rank test statistic
simplifies to a commonly used form. If T denotes the sum of
Wilcoxon Signed Rank Test the positive ranks, then U⫽2T⫺n(n⫹1)/2 and
When the response variable of interest is a metric measure-
ment that follows a symmetrical distribution with substantial S⫽ 冑n共n⫹1兲共2n⫹1兲/6.
variability or outliers, then the Wilcoxon signed rank test1,2 is The significance of U/S is determined by comparison with a
a useful test for differences between paired samples of standard normal distribution or by computation of the exact
subjects or paired conditions on a sample of subjects. For a critical region, as above.
sample of size n, let i⫽1,. . .,n identify the subjects in the If it is assumed that the distribution of the {di} is contin-
sample, let Xi and Yi denote the observed values of the uous and symmetrical, a point estimate of the median differ-
outcome variable under the 2 conditions or at the 2 time ence ⌬ is given by
points of measurement for the ith subject, and let di denote the
difference (Xi⫺Yi). If ⌬ represents the median of the distri- ⌬ˆ ⫽Median关 fii⬘⫽共di⫹di⬘兲/2兴 for 1ⱕiⱕi⬘ⱕn.

From the Department of Biostatistics, School of Public Health, University of North Carolina, Chapel Hill.
The online-only Data Supplement, consisting of a technical appendix, is available with this article at http://circ.ahajournals.org/cgi/content/
full/114/23/2528/DC1.
Correspondence to Lisa M. LaVange, PhD, Collaborative Studies Coordinating Center, Department of Biostatistics, CB 8030, School of Public Health,
University of North Carolina, Chapel Hill, NC 27514-4145. E-mail lisa lavange@unc.edu
(Circulation. 2006;114:2528-2533.)
© 2006 American Heart Association, Inc.
Circulation is available at http://www.circulationaha.org DOI: 10.1161/CIRCULATIONAHA.106.613638

2528
 LaVange and Koch Rank Score Tests 2529

The n(n⫹1)/2 quantities involved here are the n differences TABLE 1. Example Data From Ischemic Heart Disease Clinical
and their n(n⫺1)/2 pairwise averages. Furthermore, a confi- Trial: Duration of Exercise in Seconds After Exposure to Air
dence interval can be constructed about ⌬ via the methods of and CO
Hodges and Lehmann on the basis of the exact distribution of Duration After Duration After
T.1,5,6 Patient Order Baseline, s Exposure to Air, s Exposure to CO, s
The asymptotic relative efficiency (ARE) is a useful way to 1 1 240 270 285
compare a nonparametric test with its parametric counterpart.
2 1 270 220 220
Briefly, the ARE can be defined as the ratio of sample sizes
3 1 450 406 370
required by the 2 statistics to achieve the same power under
a certain distributional assumption.7 For a test of paired 4 1 540 540 480
samples or paired conditions on a sample of subjects, the 5 1 1020 1020 1020
paired t test would be the parametric test of choice. The ARE 6 1 560 540 540
of the Wilcoxon signed rank test relative to the paired t test is 7 1 240 180 180
at least 0.864 in the entire class of continuous symmetrical 8 1 810 780 780
distributions and at least 0.955 when the differences {di} 9 1 780 780 780
follow a normal distribution.5
10 1 750 868 840
11 1 300 505 330
Example Dataset
The methods described above are illustrated with an example 12 1 190 540 360
from cardiovascular research. Example data are from a 13 1 360 510 322
clinical trial designed to assess the impact of low levels of 14 1 1002 1020 1020
exposure to carbon monoxide (CO) on exercise tolerance in 15 1 475 462 540
patients with ischemic heart disease.8 A total of 42 nonsmok- 16 1 900 798 720
ing patients with documented obstructive coronary artery 17 2 390 370 300
disease and a history of exercise-induced ischemia were
18 2 405 325 300
enrolled in this 2-period crossover study. The study period
19 2 285 280 280
consisted of 3 days, a training day and 2 exposure days during
which patients were exposed to either air or CO in an 20 2 640 795 720
environmentally controlled chamber. On all 3 days, patients 21 2 400 345 320
Downloaded from http://ahajournals.org by on July 16, 2024

followed a bicycle exercise protocol in which exercise was 22 2 1100 1280 1170
conducted at increasing work loads until angina, fatigue, or 23 2 840 720 490
hypertension occurred. On the first or training day of the 24 2 780 740 720
study, patients became familiar with the environmentally 25 2 510 780 780
controlled chamber and conducted a training exercise on the
26 2 540 510 550
bicycle. Patients demonstrating exercise-induced ischemia
27 2 395 1020 780
during training were then randomly assigned to one of 2
exposure sequences, exposure to air followed by carbon 28 2 855 614 750
monoxide (Air:CO) and the reverse (CO:Air). Cardiac func- 29 2 1200 840 840
tion and exercise capacity were measured on each day after 30 2 540 720 759
exposure in the chamber.
A total of 30 patients (8 women and 22 men) successfully
completed training and were randomized to exposure se- ranked across the 30 patients. Differences of zero are ignored,
quence. The outcome variable used for illustration here is and midranks are used in the case of ties. Signs are then
duration of exercise (seconds) after the exposure condition,
applied to indicate which differences are ⬍0 or ⬎0, corre-
provided in Table 1 for each patient. This outcome variable
sponding to a decrease and an increase in exercise time,
typically has a somewhat skewed distribution and can be
respectively. Table 2 provides the rank matrix for the exam-
subject to outliers, and therefore the use of nonparametric
ple data. In this example, the sum of the positive ranks is
methods for hypothesis testing is particularly appealing. The
order variable groups the patients according to order of T⫽166.5, the sum of the signed ranks is U⫽123, the square
exposure (1⫽CO first and 2⫽Air first). Therefore, 16 pa- root of the sum of squares of the signed ranks (the standard
tients were exposed to CO on the first day and exposed to Air deviation of U) is S⫽53.56, and the test statistic is
on the second day, and 14 patients were exposed to Air on the U/S⫽2.296. The exact P⫽0.0198, and the approximate
first day, followed by CO on the second day. The baseline P⫽0.0217, both indicating that the null hypothesis of a zero
measure corresponds to the duration of exercise recorded on median for the difference between exposure conditions in
the training day before randomization. exercise times is rejected in favor of a significant difference.
The Wilcoxon signed rank test can be used to test for Subjects were able to exercise for significantly longer periods
differences between duration of exercise under the 2 expo- of time after exposure to Air than after exposure to CO. The
sures. First, differences are formed between the 2 exercise Hodges-Lehmann point estimate for the median difference is
times, and the absolute values of the nonzero differences are 54.0 seconds with a 95% confidence interval of 15.5 to 110.0.
 2530 Circulation December 5, 2006

TABLE 2. Rank Matrix for the Wilcoxon Signed Rank Test A test of significance for the association between the 2
Difference Air Signed
variables of interest (X and Y) is given by (n⫺1)rS2, which is
Patient Order vs CO, s Ranks* approximately ␹2 distributed with 1 degree of freedom, when
the 2 variables are independent and thereby have no associ-
2 1 0 䡠䡠䡠
ation (ie, the null hypothesis is true).9
5 1 0 䡠䡠䡠 The Spearman rank correlation coefficient is appropriate
6 1 0 䡠䡠䡠 for both ordered categorical and continuous variables. The
7 1 0 䡠䡠䡠 computations are valid with the use of midranks, and there-
8 1 0 䡠䡠䡠 fore ties with respect to either variable can be accommodated.
9 1 0 䡠䡠䡠 Critical values of the test statistic can be computed with the
14 1 0 䡠䡠䡠 large-sample ␹2 approximation when sample sizes are large
19 2 0 䡠䡠䡠
(eg, nⱖ40) and through tabulation of the critical regions of
25 2 0 the exact distribution, when sample sizes are small. The
䡠䡠䡠
29 2 0
statistical procedures SAS Proc FREQ3 and StatXact4 both
䡠䡠䡠
provide exact probability levels for the Spearman rank
1 1 ⫺15 ⫺1.0
correlation test.
24 2 20 2.0
18 2 25 3.5 Example Dataset, Continued
21 2 25 3.5 In the example study, it may be of interest to determine
10 1 28 5.0 whether the differences in exercise times under the 2 expo-
3 1 36 6.0 sure conditions vary with baseline values. If baseline appears
30 2 ⫺39 ⫺7.0 to be correlated with differences in exercise times, then an
26 2 ⫺40 ⫺8.0 analysis that adjusts for baseline differences among subjects
4 1 60 9.0
may be warranted. The Spearman rank correlation coefficient
can be applied to assess this correlation. To compute the
17 2 70 10.0
correlation coefficient, the differences in exercise times
20 2 75 11.0
between the 2 conditions require ranking, irrespective of zero
15 1 ⫺78 ⫺12.5 values. Midranks are again used in the case of ties. The
16 1 78 12.5 ranked values are provided in Table 3.
Downloaded from http://ahajournals.org by on July 16, 2024

22 2 110 14.0 The Spearman rank correlation coefficient for baseline by

28 2 ⫺136 ⫺15.0 differences in exercise times is ⫺0.1093 with P⫽0.5564,
11 1 175 16.0 indicating no significant association between these 2 vari-
12 1 180 17.0 ables. A logical follow-up question is whether baseline is
13 1 188 18.0 correlated with exercise times after either exposure condition.
The Spearman rank correlation coefficient between the base-
23 2 230 19.0
line value and exercise time after exposure to Air is 0.7843
27 2 240 20.0
(P⬍0.0001) and between baseline and exercise time after
*Signed ranks reflect ranks of absolute values of differences, with zeroes exposure to CO is 0.8234 (P⬍0.0001). Baseline values are
ignored and midranks used in the case of ties.
therefore strongly associated with postexposure exercise
times, regardless of the condition. The difference between
Spearman Rank Correlation Coefficient
conditions with respect to exercise duration does not, how-
The Spearman rank correlation coefficient is a measure of
association between 2 variables that is particularly useful ever, appear to vary with baseline.
when 1 or both variables are either ordered categorical or are
continuous but from a highly skewed distribution. The Wilcoxon Rank Sum Test
coefficient can be used to test the null hypothesis of no When the research question of interest involves comparing 2
association between the 2 variables versus the alternative samples or groups of subjects with respect to a response
hypothesis of an association. To compute the Spearman rank variable, such as comparing disease outcomes among patients
correlation coefficient, the values of each of the 2 variables randomized to receive a test versus control treatment, the
are first ranked, with the use of midranks in the case of ties, Wilcoxon rank sum test has utility. The null hypothesis in this
and the standard Pearson correlation coefficient is then setting is that of no association between the 2 groups of
computed with the use of these ranks. Let i⫽1,. . .,n identify subjects and the response variable, and the alternative hy-
the n subjects in the sample, and let Xi and Yi denote the pothesis is that of a location shift for the population repre-
variable values for the ith subject. Let Ri and Si denote the sented by one group versus the other, eg, relatively more
ranks of Xi and Yi, respectively. Then the Spearman rank higher values of the response variable as a result of the test
correlation coefficient is given by the following equation: treatment. The Wilcoxon rank sum test can be applied
冘共R ⫺R៮ 兲共S ⫺S៮ 兲 regardless of whether the response variable is metric or

冑冘共R ⫺R៮ 兲 冘共S ⫺S៮ 兲

i i
rS ⫽ ordered categorical, and, as is the case for the signed rank
2 2
i i test, the methods of Hodges and Lehmann can be applied to
 LaVange and Koch Rank Score Tests 2531

TABLE 3. Rank Matrix for the Wilcoxon Rank Sum Test and when sample sizes are large (eg, ⱖ20 per group). When
Spearman Correlation sample sizes are small and subjects are randomly allocated
to groups (either by the design of the study or as implied
Difference Air vs
by the null hypothesis), the significance level is calculated
Patient Order CO, s Ranks*
by comparing the test statistic to the critical region of the
1 1 ⫺15 5.0 exact distribution.10
2 1 0 10.5
3 1 36 20.0 If there are no ties among the ranks, then the test statistic
4 1 60 21.0
simplifies to a commonly used form. Let T be the sum of the
ranks in group 1. Then the rank sum test statistic is given by
5 1 0 10.5
6 1 0 10.5 T⫺n 1共n⫹1兲/ 2
.
7 1 0 10.5 冑n n 共n ⫹n ⫹1兲/12
1 2 1 2
8 1 0 10.5
The statistical procedures SAS Proc NPAR1WAY3 and
9 1 0 10.5
StatXact4 both provide exact probability levels for the Wil-
10 1 28 19.0
coxon rank sum test.
11 1 175 26.0 If it is assumed that metric distributions for the 2 groups
12 1 180 27.0 have the same shape and scale, Hodges-Lehmann estimates
13 1 188 28.0 for the difference in medians between the 2 groups of
14 1 0 10.5 patients, ⌬, and confidence limits about ⌬ are available. The
15 1 ⫺78 2.0 point estimate corresponds to the median of all pairwise
16 1 78 24.0 differences between observations in one group versus those in
17 2 70 22.0
the other group. There are n1n2 such differences.
The ARE for the Wilcoxon rank sum test relative to the t
18 2 25 17.5
test for comparing 2 independent samples is at least 0.864
19 2 0 10.5 when the alternative hypothesis is a location shift in the
20 2 75 23.0 distributions of the 2 samples and all continuous distributions
21 2 25 17.5 are considered. When the distributions are normal, the ARE is
22 2 110 25.0 at least 0.955. Note that when the distributions of the
Downloaded from http://ahajournals.org by on July 16, 2024

23 2 230 29.0 response variables are highly skewed, with long tails at either
24 2 20 16.0 end, then the ARE can exceed 1.0, indicating that the
25 2 0 10.5
Wilcoxon rank sum test will be more powerful than a t test in
this instance.11
26 2 ⫺40 3.0
27 2 240 30.0 Example Dataset, Continued
28 2 ⫺136 1.0 The association between order of exposure and difference in
29 2 0 10.5 exposure times can be assessed by applying the Wilcoxon
30 2 ⫺39 4.0 rank sum test because order of exposure defines 2 groups of
*Ranks of differences incorporate all values, including zeroes, and midranks patients for comparison. The sum of the ranks for subjects
are used in the case of ties. exposed to CO first is 245.5, and the sum of the ranks for
patients exposed to Air first is 219.5. Expected values of
compute point estimates and confidence intervals for the these 2 quantities under the null hypothesis of no association
difference in medians between the 2 samples. are 248.0 and 217.0, respectively, where expected value is
defined as the average value of the sum of the ranks across all
Let n1 denote the sample size in the first group and n2
possible randomizations of n1 subjects to the first group and
denote the sample size in the second group. The total sample
n2 subjects to the second group. The Wilcoxon rank sum test
size is n⫽n1⫹n2. The Wilcoxon rank sum test is computed as
statistic⫽219.5, and the exact P⫽0.9250, compatible with no
follows:
difference between the 2 groups with respect to differences in
exercise duration. The approximate P⫽0.9178. The Spear-
1. Ranks are assigned to all observations of the response
man rank correlation coefficient for order of exposure by
variable, pooling across groups of subjects and using
midranks for ties. differences in exercise times is ⫺0.0197, and P⫽0.9178,
2. The test statistic can most easily be expressed in the identical to that associated with the Wilcoxon rank sum test,
manner previously described for the Spearman rank cor- which is compatible with no correlation between these 2
relation coefficient by letting S have the value 1 for the n1 variables.
subjects in group 1, and the value 0 for the n2 subjects in For this example dataset, the fact that the order of exposure
group 2 (where n⫽n1⫹n2), and letting R correspond to the conditions did not appear to be related to the response
ranks of the response variable. variable validates the use of a signed ranks analysis of
3. The significance level is calculated by comparing the test exercise times under the 2 conditions, ignoring the order of
statistic to the ␹2 distribution with 1 degree of freedom, exposure. Had order been related to response, then a proper
 2532 Circulation December 5, 2006

crossover analysis of the study data would be required that ence in exercise times, the fact that the covariate adjustment
accounted for the order of exposure in assessing the impact of had little impact on the results of the unadjusted Wilcoxon
CO versus air on exercise times.12 rank sum test is not surprising.
To illustrate the extension of these methods for stratifica-
Extensions of the Rank Sum Test tion, subjects were stratified according to whether their
The Wilcoxon rank sum test can be extended to allow for baseline value was below the median of 540 seconds versus
covariate adjustment in a nonparametric analog to ANCOVA. equal to or above the median. A stratified Wilcoxon rank sum
Rank ANCOVA13 can be performed through the following test (ie, van Elteren test) was then performed. The probability
steps: value for testing the null hypothesis of no association in all
strata is 0.8810 and therefore is compatible with no associa-
1. Ranks are computed for the response variable, ignoring tion between order of exposure and differences in exercise
groups and using midranks in the case of ties. duration after stratifying on baseline value (below versus
2. Ranks are then computed for the covariate, ignoring above the median).
groups and using midranks in the case of ties.
3. A linear regression model is fit, regressing the ranked Discussion
response variable onto the ranked covariates (with groups
Nonparametric methods such as those described here are
ignored), and the residuals are output.
4. A test of the association between group and the response most useful in situations in which their use is prespecified
variable, adjusting for the covariate, is provided by apply- before data analysis. When response variables are ordered
ing computations like those described for the Spearman categories or when it is known in advance that assumptions
rank correlation coefficient (the formula for rs), but with R about metric response variables will likely fail, then such
equal to the residuals, S⫽1 for subjects in group 1, and prespecification is possible at the time the statistical analysis
S⫽0 for subjects in group 2.9 plan is prepared. In these situations, nonparametric methods
will have good power properties, with power as high as 93%
The method in step 4 for comparing the residuals in the 2 of that expected for standard parametric methods (eg, t tests)
groups is an extension of the Wilcoxon rank sum test to applied under ideal circumstances.
provide covariate adjustment. Rank ANCOVA can provide The methods described here all address a shift in location
additional power through the variance reduction typically as the alternative hypothesis, in which location corresponds to
associated with a baseline covariate adjustment, even when the median of the response variable distribution. If trans-
the response variable does not follow a normal distribution.14 formed data are expected to be normally distributed (eg, the
Downloaded from http://ahajournals.org by on July 16, 2024

Stratification may also be an important aspect of the study response variable follows a log-normal distribution), then
design resulting from patients being sorted into subsets before nonparametric methods will be at least 95% efficient, and
the conduct of the study, (eg, male and female strata or strata their use precludes identifying the most optimal
consisting of patients from different clinical centers in a transformation.
multicenter study). Extensions for the Wilcoxon rank sum test When the response variable is so highly skewed that the
and the test for the Spearman rank correlation coefficient that distribution appears to have an “L” or “J” shape, the Wilcox-
account for stratification are available.9,10 The stratified on tests will not have good power, and Savage (or log-rank)
extension for the Wilcoxon rank sum test is referred to as the tests will be better.16 With highly skewed distributions, both
van Elteren statistic.15 Computations for this method are groups of subjects will tend to have ranked values on one side
shown in the technical appendix in the online-only Data of the median, but only one group will have ranked values on
Supplement and are available through SAS Proc FREQ.3 the other or tail side of the distribution. In this case, only the
rank values on the tail side are informative, and against this
Example Dataset, Continued alternative, the Wilcoxon rank sum test will not be the most
Rank ANCOVA can be applied to the example dataset to appropriate test. Because the Wilcoxon tests address shifts in
assess the impact of order of exposure on the differences in location only, ranked values from both groups are expected to
exercise times between exposure conditions while controlling occur to the left and to the right of the median, and both are
for baseline values. Following the steps outlined above, the informative. Under the alternative hypothesis of a shift in
difference scores between air and CO are ranked as for the location, one group will tend to have ranks on one side of the
Wilcoxon rank sum test (Table 3). Baseline values are also overall median, whereas the other will tend to have ranks on
ranked, and a linear regression model is fit with the use of the opposite side. This is precisely the setting in which the
SAS Proc GLM.3 The residuals from the model are output, Wilcoxon tests are most useful and nearly as powerful as
and the Pearson correlation coefficient between the residuals parametric methods applied when all assumptions hold.
and the order variable is computed as ⫺0.0372. The proba-
bility value from the exact test is 0.8437 and from the Disclosures
asymptotic test is 0.8412, both showing little association None.
between order of exposure and difference scores, after adjust-
ment for baseline values. These results are similar to the References
1. Hollander M, Wolfe DA. Nonparametric Statistical Methods. New York,
unadjusted test of association computed above. Because the
NY: John Wiley & Sons; 1973.
Spearman rank correlation coefficient showed no evidence of 2. Conover WJ. Practical Nonparametric Statistics. New York, NY: John
association between baseline exercise times and the differ- Wiley & Sons; 1971.
 LaVange and Koch Rank Score Tests 2533

3. SAS Institute Inc. SAS/STAT User’s Guide, Version 9. Cary, NC: SAS 11. Moses L. Wilcoxon-Mann-Whitney test. In: Armitage P, Colton T, eds.
Institute Inc; 2004. Encyclopedia of Biostatistics. Vol 6. West Wessex, England: John Wiley
4. Cytel Software Corporation. StatXact 7 On-line User Manual. Cam- & Sons Ltd; 1998:4742– 4745.
bridge, Mass: Cytel Software Corporation; 2005. 12. Tudor G, Koch GG. Review of non-parametric methods for the analysis
5. Woolson RF. Wilcoxon signed-rank test. In: Armitage P, Colton T, eds. of crossover studies. Stat Methods Med Res. 1994;3:345–381.
Encyclopedia of Biostatistics. Vol 6. West Wessex, England: John Wiley
13. Koch GG, Carr GJ, Amara IA, Stokes ME, Uryniak TJ. Categorical data
& Sons Ltd; 1998:4739 – 4740.
analysis. In: Berry DA, ed. Statistical Methodology in the Pharmaceutical
6. Hodges JL, Lehmann EL. Rank methods for combination of independent
experiments in analysis of variance. Ann Math Stat. 1962;33:482– 497. Sciences. New York, NY: Marcel Dekker; 1990:389 – 473.
7. DasGupta A. Encyclopedia of Biostatistics. Vol 1. Armitage P, Colton T, 14. LaVange LM, Durham TA, Koch GG. Randomization-based nonpara-
eds. West Wessex, England: John Wiley & Sons Ltd; 1998:210 –215. metric methods for the analysis of multicentre trials. Stat Methods Med
8. Adams KF, Koch GG, Chaterjee B, Goldstein GM, O’Neil JJ, Bromberg Res. 2005;14:281–301.
PA, Sheps DS, McAllister S, Price CJ, Bissette J. Acute elevation of 15. Lehmann EL. Nonparametrics: Statistical Methods Based on Ranks. San
blood carboxyhemoglobin to 6% impairs exercise performance and Francisco, Calif: Holden-Day; 1975.
aggravates symptoms in patients with ischemic heart disease. J Am Coll 16. Koch GG, Sen PK, Amara I. Log-rank scores, statistics, and tests. In:
Cardiol. 1988;12:900 –909. Kotz S, Johnson NL, eds. Encyclopedia of Statistical Sciences. Vol 5.
9. Stokes ME, Davis CS, Koch GG. Categorical Data Analysis Using the New York, NY: John Wiley & Sons; 1985:136 –142.
SAS System. 2nd ed. Cary, NC: SAS Institute Inc; 2000.
10. Landis RJ, Sharp TJ, Kuritz SJ, Koch GG. Mantel-Haenszel methods. In:
Armitage P, Colton T, eds. Encyclopedia of Biostatistics. Vol 3. West
Essex, England: John Wiley & Sons Ltd; 1998:2378 –2391. KEY WORDS: statistics, nonparametric 䡲 probability 䡲 variable distributions
Downloaded from http://ahajournals.org by on July 16, 2024