J Clin Epidemiol Vol. 52, No. 3, pp.

229–235, 1999 0895-4356/99/ $–see front matter

Copyright © 1999 Elsevier Science Inc. All rights reserved. PII S0895-4356(98)00168-1

Increasing Physicians’ Awareness of the Impact of

Statistics on Research Outcomes: Comparative Power of
the t-test and Wilcoxon Rank-Sum Test in Small
Samples Applied Research
Patrick D. Bridge1,* and Shlomo S. Sawilowsky2
1Department of Family Medicine, Wayne State University School of Medicine and 2Department of Theoretical and
Behavioral Foundations, College of Education, Wayne State University, Detroit, Michigan

ABSTRACT. To effectively evaluate medical literature, practicing physicians and medical researchers must
understand the impact of statistical tests on research outcomes. Applying inefficient statistics not only increases
the need for resources, but more importantly increases the probability of committing a Type I or Type II error.
The t-test is one of the most prevalent tests used in the medical field and is the uniformally most powerful
unbiased test (UMPU) under normal curve theory. But does it maintain its UMPU properties when assumptions
of normality are violated? A Monte Carlo investigation evaluates the comparative power of the independent
samples t-test and its nonparametric counterpart, the Wilcoxon Rank-Sum (WRS) test, to violations from
population normality, using three commonly occurring distributions and small sample sizes. The t-test was more
powerful under relatively symmetric distributions, although the magnitude of the differences was moderate.
Under distributions with extreme skews, the WRS held large power advantages. When distributions consist of
heavier tails or extreme skews, the WRS should be the test of choice. In turn, when population characteristics are
unknown, the WRS is recommended, based on the magnitude of these power differences in extreme skews, and
the modest variation in symmetric distributions. J CLIN EPIDEMIOL 52;3:229–235, 1999. © 1999 Elsevier Science Inc.

KEY WORDS. Research methods; t-test; Wilcoxon Rank-Sum test; nonparametric statistics; parametric statistics;
power

INTRODUCTION under the assumption of normality and is therefore the uni-

formally most powerful unbiased test (UMPU) when data
The use of statistics in medical research has increased con-
are normally distributed. (UMPU means that by definition,
siderably in past 60 years, and the types of statistics used
when the data are normally distributed, no other test has
have become much more complex [1]. Concomitant with
greater ability to detect true differences for a given sample
this influx in the use of statistics, unfortunately, was an in-
size.) But, does it maintain its UMPU properties when vio-
crease in statistical errors [2–12]. For a practicing physician
lations from normality occur for small samples typical of ap-
trying to stay current with the literature, or a medical re-
plied research?
searcher adding new knowledge to the field, it is important
In medicine and social and behavioral science, normal
to understand the application and efficiency of statistical
theory tests (e.g., t-test, ANOVA) have been used more
tests and their impact on the outcomes of research.
extensively than nonparametric statistics (which do not ap-
Many researchers view the application of statistical tests
peal to the population shape as part of their derivation)
as a simple and clear-cut process. Nevertheless, in many sit-
[13–15, 17–19]. Yet, statisticians and researchers question
uations the appropriate application of statistical tests may
the frequency of normally distributed data in real world
be unclear, and in other situations, controversial. For exam-
problems [17, 20–24]. For example, Micceri [17] examined
ple, consider the independent samples t-test, which is one
440 psychometric and ability measures and found that all of
of the most prevalent statistics used in medicine, psychol-
the data sets were nonnormal according to the Kolomog-
ogy, and education research [13–16]. The t-test is derived
orow-Smirnov test for normality at the 0.01 alpha level.
Only 3% were even remotely similar to the normal curve
(i.e., smooth symmetric with light tails). The problem of
*Address for correspondence: Patrick D. Bridge, University Health Center,
4201 St. Antoine, Room 4J, Wayne State University, Detroit, MI 48201. non-normality in applied data is also prevalent in medical
Accepted for publication on 6 November 1998. research. For example, a literature search was conducted to
230
identify studies in which data violated normal distribution
theory. The various disciplines where this problem was preva-
lent included surgery [25], epidemiology [26–28], emergency
medicine [29], rehabilitation [30], chemistry [31], pediatrics
[32], biomedical science [33], anesthesiology [34], and den-
tistry [35], among others.
In consideration of how rarely normality occurs in ap-
plied research practice, the persistence and prevalence in
the use of the t-test is questionable. Does the t-test, despite
its property of being the UMPU test, maintain its superior
power properties (i.e., the ability to detect a treatment ef-
fect if it exists) when normal theory assumptions are vio-
lated? Blair [36] stated, “One might assume that because the
t-test is the uniformly most powerful (UMP) unbiased test
under normal theory, it will naturally be more powerful
than other tests in the non-normal situation, provided that
its normal theory power is preserved. But this is fallacious
reasoning because the optimal power properties associated
with the t-test under normal theory are no longer in force
once the normality stipulation has been abandoned.”
STATISTICAL ISSUES
A Type I error (alpha) occurs when the null hypothesis is
rejected, when in fact it is true. A Type II error (beta) oc-
curs by failing to reject a null hypothesis when it is actually
false. Robustness with respect to Type I error means that if
nominal alpha was set to 0.05, then the actual Type I error
rate is reasonably close to this rate even if normality (or
other underlying assumptions) is not met.
Robustness with respect to Type II error means that a
test achieves approximately the same rejection rate (i.e.,
ability to detect a difference) when normality (or other un-
derlying assumptions) is not met. Statistical power, com-
monly referred to as the Pittman Efficiency, is the test’s
ability to detect a false null hypothesis. Note, however, that
even if a test is robust with respect to Type II errors, there
still may be a competitor that is more powerful when nor-
mality is not met [37].
Traditionally, statisticians’ first criterion of the quality of
a test is its robustness (i.e., insensitivity to violation of as-
sumptions). The t-test’s robustness properties have been ex-
amined carefully over the past few decades. The current
thinking on the subject was provided by Sawilowsky and
Blair [38], who examined the robustness of the independent
samples t-test when normality was violated under condi-
tions using real world data. The t-test was found to be ro-
bust when sample sizes are nearly equal, fairly large (25–30),
and a two-tailed test rather than a one-tailed test is used.
Despite the t-test’s remarkable robustness powers, some
statisticians caution readers to refrain from their hurried use
of the t-test, because a nonparametric competitor, such as
the Wilcoxon Rank-Sum (WRS) test, may be a more pow-
erful statistic. Robustness with respect to Type I errors (i.e.,
false positives) is assured with the nonparametric Wilcoxon
P. D. Bridge and S. S. Sawilowsky
statistic. As noted by Mansfield [39], “The hallmark of
these tests is that they avoid the assumption of normality.”
Thus, Sawilowsky [40] concluded, “the real issue of the ef-
fects of nonnormality is on the comparative power, not ro-
bustness of the t-test.”
The researcher’s recognition of the power properties of a
test enhances the ability to detect treatment differences,
promotes replication of the study under the same condi-
tions, and diminishes the wasting of resources through the
use of efficient statistical tests. Remarkably, however, until
recently, the power of a statistic (or power analysis) has
been a neglected methodological tool in medical research.
(An anonymous reviewer pointed out that work in clinical
trials is a notable exception, and power analysis has been
considered a requirement for funded research in clinical tri-
als for quite some time.) For example, Friedman et al. [41]
reviewed 71 negative clinical trials and found that 67 of the
trials had a greater than 10% risk of missing a 25% thera-
peutic improvement and 50 trials may have missed a 50%
therapeutic improvement. Mengel [42] canvassed three ma-
jor family practice journals in 1988, and found that only
5 of 86 studies calculated statistical power. Although it
turned out that 80% of the studies had sufficient power to
detect medium and large effect sizes, very few could detect
small effect sizes. In some situations, even large effect sizes
could not be detected. Similarly, Silagy [11] reviewed 55
randomized control trials in four peer reviewed family med-
icine journals from 1987–1991, and found that statistical
power was reported in only five studies. Williams [12] re-
viewed 44 negative clinical trials in dermatology and found
that all but one had a 10% chance of missing a 25% treat-
ment difference; and 31 of the 44 negative trials had such a
small sample size that they had a greater than 10% chance
of missing a 50% treatment difference. Edlund [43] echoed
similar results.
The failure to recognize the need for power analysis has
mystified many. Kraemer and Thiemann [44] attributed the
neglect to over-training in dealing with significant levels
and under-training in the use of power. Cohen [45] stated,
“One possible reason for the continued neglect of statistical
power analysis in research in the behavioral sciences is the
inaccessibility of or difficulty with the standard material.”
In addition, cost and resources associated with conducting
research and obtaining a large sample size has also attrib-
uted to the low power in the medical literature. With
readily available resources, such as Cohen’s [45, 46] refer-
ences, the failure to conduct a power analysis is no longer
acceptable.
PURPOSE
Small samples research (i.e., repeated sampling Monte
Carlo results) conducted on the comparative power of the
t-test and the Wilcoxon test indicate the Wilcoxon test
holds promise for increased statistical power when the as-
Impact of Statistics on Research Outcomes
sumption of normality is violated [47–53]. However, these
studies were conducted using fabricated data, which was
distributed according to well known mathematical curves
(i.e., half-normal, Laplace, exponential, mixed normal).
Micceri [17] raised the question of the generalizability of
such studies because the real data sets found to occur in
practice are far less “tame” and “mathematically expedient”
than theoretical curves. Thus, despite the plethora of re-
search on the comparative power of the t-test and the Wil-
coxon test, the question remains as to how these two proce-
dures fare with real data sets.
The purpose of this study, therefore, is to assess the com-
parative power of the independent samples t-test and the
Wilcoxon Rank-Sum test to violations from normality us-
ing real world data sets, and to guide physicians and medi-
cal researchers in the appropriate use of these tests. Hope-
fully, when conducting research or evaluating the medical
literature for clinical practice, the impact of this study will
have educated physicians and medical researchers in under-
standing the effect statistical tests have on research out-
comes.
METHODS
Using a Gateway 2000 4DX2-66V and Microsoft FOR-
TRAN 5.1 program, Monte Carlo techniques were used to
analyze the comparative power properties of the indepen-
dent samples t-test with the Wilcoxon Rank-Sum test. For
ease of computation, the Wilcoxon Rank-Sum test was
conducted by applying the independent samples t-test on
the ranks of the original scores. Zimmerman and Zumbo
[54], Conover and Iman [55], and others have shown the
Wilcoxon Rank-Sum test on original scores is “alpha equiv-
alent” to conducting an independent samples t-test on the
ranks of those scores. Thus, the results obtained by applying
the Wilcoxon Rank-Sum test on original scores would be
the same as those results obtained by applying the t-test on
the ranks of those original scores.
The observations were randomly sampled with replace-
ment from three real data sets from Micceri [17] using the
International Mathematical and Statistical Libraries [56]
RNSET, RNUND, and RNKSM, subroutines. Sample sizes
(n1, n2) 5 (10,10), (15,25), (30,30), and (15,45) were used,
with nominal alpha set at 0.05. Four treatment alternatives
of varying sizes were analyzed for each distribution and sam-
ple size to measure shift in location parameters. As is cus-
tomary in Monte Carlo studies where treatments are mod-
eled as a shift in location, effect sizes (c) were a function of
a distribution’s population standard deviation, in multiples
of 0.25s. This is appropriate in the current study due to
Micceri’s [17] proposition that the data sets obtained in his
study were sufficiently large to be taken as representative of
the population. (Another common, but subjective and less
thorough approach is to model small, medium, and large
231
FIGURE 1. Multimodal and lumpiness distribution.
treatment effects, instead of multiples of the population
standard deviation [45].) The simulated treatment effect
(c) was then added to each n1, producing the shift in loca-
tion in comparison with the n2 scores.
Three data sets from Micceri [17] were considered proto-
typical and representative of real behavioral science data.
They are based on measures such as the Minnesota Mul-
tiphasic Personality Inventory (MMPI), and other well
known instruments purporting to measure “anger, anxiety,
curiosity, locus of control, masculinity/femininity, satisfac-
tion, sociability, visual hallucinations, and so forth” [38].
Micceri [17] labeled the distribution shapes most deviant
from normality as multimodal lumpy, discrete mass at zero
with gap, and extreme asymmetry.
Multimodal lumpy distributions are common where two
(or more) dominant sub-populations exist in the same pa-
tient pool. For example, a study on developmental disabili-
ties might produce two sub-populations, where one has a
physical disability and the other has a mental disability.
The discrete mass at zero with gap occurs with “onset” or
“first use” variables. For example, a survey of adolescents
might be conducted to determine the age of first suicide at-
tempt or age when first began smoking. The extreme asym-
metry shape is also prolific, such as the purported decline in
attention deficient hyperactivity disorder from infancy into
adulthood [57,58]. (Note that asymmetry is not synony-
mous with exponentiality [59].) Histograms of the distribu-
tions studied are presented in Figures 1–3, and descriptive
statistics are compiled in Table 1.
The independent samples t-test and Wilcoxon Rank-
Sum test were performed on each sample. Power levels (and
hence, advantages of one procedure over the other) were
obtained by comparing the rejection rate for each test. Ten
thousand repetitions were conducted for each distribution,
sample size, and treatment effect interval.
RESULTS
Results of the comparative power analysis are shown in Ta-
bles 2–4. Columns 1 and 2 identify the sample size (n) and
232
FIGURE 2. Mass at zero with gap distribution.
test statistic. The remainder of the table depicts the effect
size (c) examined, the resulting power level for each statis-
tic, and the difference in comparative power (D) between
both statistics for each condition studied.
Multimodal and Lumpiness
Comparative power results for the t-test and the WRS for
the multimodal and lumpy distribution are compiled in Ta-
ble 2. This distribution is somewhat symmetric (as is the
normal curve), and therefore, as expected, the t-test held a
slight power advantage for most of the comparisons. The
average power difference, however, was only 0.03, with the
largest advantage of 0.05 occurring when (n1, n2) 5 5, 15.
It should also be noted that neither test was particularly
powerful in detecting small treatment effects.
Mass at Zero with Gap
The WRS produced considerable power advantages in 94%
(15 of 16) of the conditions studied when the data were
sampled from the mass at zero with gap data set. The WRS
held an impressive average power advantage of 0.49, as
noted in Table 3. Power advantages were predominant for
the smaller effect sizes (e.g., c 5 .25s) for all sample sizes
considered. The most remarkable power advantage oc-
curred for (n1, n2) 5 15, 45, with the Wilcoxon test dis-
playing a power advantage of 0.803. The power advantages
remained substantial for moderate effect sizes (c 5 .50 s),
averaging approximately 0.52.
TABLE 1. Summary data for the three data sets from Micceri [17]
Distribution Mean Median
Multimodal and lumpiness 21.15 18.00
Discrete mass at zero with
gap 1.85 0.00
Extreme asymmetry 13.67 11.00
P. D. Bridge and S. S. Sawilowsky
FIGURE 3. Extreme asymmetry distribution.
Extreme Asymmetry
As noted in Table 4, the WRS demonstrated advantages for
all 16 conditions studied, with power differences over the
t-test ranging from 0.029 to 0.618. The power advantages
were at consistent levels favoring the Wilcoxon Rank-Sum
test for the smaller sample sizes (i.e., n1, n2 5 5, 15 and 10,
10). The comparative power converged for the two statis-
tics for the larger sample sizes (i.e., n1, n25 15, 45 and 30,
30) when the effect sizes were large (i.e., c 5 1.0 s).
DISCUSSION
The t-test is based on the assumption of normality and is
one of the most prevalent tests used in the medical field, as
well as other professions. This test assumes normally distrib-
uted data, but it is now obvious that the occurrence of “bell
curve” data is extremely uncommon. The t-test has been
found to be robust with respect to violations from normality
to Type I and Type II errors, but this does not preclude the
use of a nonparametric alternative that may be more power-
ful under these conditions.
This study assessed the comparative power of the Wil-
coxon Rank-Sum test and the independent samples t-test
in measuring for shift in location parameters for three real
data sets, and demonstrated the impact nonparametric sta-
tistics can have on research outcomes when data are non-
normally distributed. The results indicate that the t-test
was more powerful only under a distribution that was rela-
tively symmetric, although the magnitude of the differences
was trivial. In contrast, the Wilcoxon Rank-Sum test held
Standard
deviation Skew Kurtosis
11.90 0.19 1.80
3.80 1.65 3.98
5.75 1.64 4.52
Impact of Statistics on Research Outcomes 233

TABLE 2. Rejection rates and power comparison of the Wil- TABLE 4. Rejection rates and power comparison of the
coxon Rank-Sum (WRS) test (t-test on ranks) with the inde- Wilcoxon Rank-Sum (WRS) test (t-test on ranks) with the
pendent samples t-test for the multimodal and lumpiness independent samples t-test for the extreme asymmetry dis-
distribution [17], a 5 0.05; 10,000 repetitions tribution [17], a 5 0.05; 10,000 repetitions
Effect size Effect size
Sample size Test 0.25s 0.50s 0.75s 1.0s Sample size Test 0.25s 0.50s 0.75s 1.0s

5,15 t 0.065 0.145 0.265 0.426 5,15 t 0.092 0.142 0.232 0.463
WRS 0.060 0.140 0.246 0.377 WRS 0.266 0.391 0.491 0.661
D= 0.005 0.005 0.019 0.049 D = 20.174 20.249 20.259 20.198
10,10 t 0.073 0.173 0.338 0.543 10,10 t 0.088 0.167 0.295 0.580
WRS 0.073 0.179 0.331 0.502 WRS 0.368 0.496 0.593 0.735
D= 0.000 20.006 0.007 0.041 D = 20.280 20.329 20.298 20.155
15,45 t 0.128 0.372 0.691 0.912 15,45 t 0.136 0.275 0.526 0.913
WRS 0.114 0.352 0.648 0.863 WRS 0.716 0.887 0.955 0.993
D= 0.014 0.020 0.043 0.049 D = 20.580 20.612 20.429 20.080
30,30 t 0.153 0.473 0.817 0.971 30,30 t 0.165 0.358 0.659 0.964
WRS 0.137 0.450 0.768 0.933 WRS 0.783 0.914 0.963 0.993
D= 0.016 0.023 0.049 0.038 D = 20.618 20.556 20.304 20.029

Note: D = power difference. Note: D = power difference.

huge power advantages for data sets which presented skew-
ness or heavy tails.
If the characteristics of a population are known to be rel-
atively symmetric with light tails, the t-test should be ap-
plied. However, if the population characteristics are un-
known, which is generally the case in applied research, and
the hypothesis being tested is one of shift in means (or
other location parameter) the Wilcoxon Rank-Sum is rec-
ommended. The logic motivating this suggestion is
straightforward. There is little to lose in terms of statistical
power (average of 0.03) should the population turn out to
be normally distributed, but there is considerable to gain (as
much as 0.803) if the population is skewed or heavy tailed.
TABLE 3. Rejection rates and power comparison of the Wil-
coxon Rank-Sum (WRS) test (t-test on ranks) with the inde-
pendent samples t-test for the mass at zero with gap distri-
bution [17], a 5 0.05; 10,000 repetitions
Effect size
Sample size Test 0.25s 0.50s 0.75s 1.0s
5,15 t 0.101 0.264 0.467 0.681
WRS 0.849 0.877 0.879 0.880
D = 20.748 20.613 20.412 20.199
10,10 t 0.036 0.140 0.428 0.660
WRS 0.700 0.720 0.811 0.826
D = 20.664 20.580 20.383 20.066
15,45 t 0.161 0.475 0.820 0.974
WRS 0.964 0.963 0.965 0.964
D = 20.803 20.488 20.145 0.010
30,30 t 0.200 0.605 0.914 0.993
WRS 0.996 0.997 0.997 0.997
D = 20.796 20.392 20.083 20.004
Note: D = power difference.
To emphasize this point, consider the relationship be-
tween statistical power and sample size. Suppose a medical
researcher conducted an experiment on a treatment and
control group, where the sample size was (n1, n2) 5 15,45;
the treatment effect was small (0.25s), as is common in
medical research; and the construct of interest was an “on-
set” variable. As noted in Table 3, the power of the Wil-
coxon Rank-Sum test was 0.803 under these conditions.
According to Cohen [45], the t-test would require an ap-
proximate sample size of (n1, n2) 5 77, 231 to obtain simi-
lar power levels.
To summarize, a small treatment effect that could be de-
tected by the Wilcoxon Rank-Sum having a sample size
with a harmonic mean of 22.5 (i.e., n1, n25 15, 45) would
require a sample size having a harmonic mean of 115.5 (i.e.,
n1, n25 77, 231) before the t-test could detect the same
treatment.1 The expense, in terms of time, effort, and cost
of having to increase the sample size more than five-fold
(115.5/22.5) could be avoided by simply using the nonpara-
metric Wilcoxon Rank-Sum test instead of the classical
parametric t-test. Finally, and most important, smaller sam-
ple sizes represent a reduction in the number of patients
who must face the risk of participating in an experimental
study.
1 As pointed out by an anonymous reviewer, this illustration assumes the
distribution is known (and Gaussian), which is rarely the case in applied
research. The point was to illustrate the potential deleterious effects of
ignoring power properties in terms of sample size, a frame of reference
appreciated by applied researchers. Also, another anonymous reviewer
stated that the use of the harmonic mean is “alien to readers of this jour-
nal.” We point out that using 0.5(n1 1 n2) would result in too many
degrees of freedom. For example, samples of size (15,45) would result in
n1 1 n2 2 2 5 58 df using the arithmetic mean, whereas the harmonic
mean results in 2mh-2 5 43 df, where mh 5 harmonic mean. Clearly, the
latter is more indicative of the unbalanced samples in the layout.
234
References
1. Altman DG, Goodman SN. Transfer of technology from sta-
tistical journals to the biomedical literature. JAMA 1994; 272
(2): 129–132.
2. Altman DG. Statistics in medical journals. Stat Med 1982; 1:
59–71.
3. Altman DG. The scandal of poor medical research. BMJ
1994; 308: 283–284.
4. Altman DG. Statistics in medical journals: Developments in
the 1980’s. Stat Med 1991; 10: 1897–1913.
5. Avram MJ, Shanks CA. Statistical methods in anesthesia arti-
cles: An evaluation of two American journals during two six-
month periods. Anesth Analg 1985; 64: 607–611.
6. Glantz SA. Biostatistics: How to detect, correct and prevent
errors in the medical literature. Circulation 1980; 61(1): 1–7.]
7. Lent V, Langenbach A. A retrospective quality analysis of 102
randomized trials in four leading urological journals form
1984–1989. Urol Res 1996; 24(2): 119–122.
8. Mosteller JP, Gilbert JP. Reporting standards and research
strategies for controlled trials. Control Clin Trials 1980; 1:
37–58.
9. Mulrow CD. The medical review article: State of the science.
Ann Intern Med 1987; 106: 485–488.
10. Pocock PJ, Huges MD, Lee RJ. Statistical problems in the re-
porting of clinical trials. A survey of three medical journals. N
Engl J Med 1987; 317(7): 426–487.
11. Silagy CA, Jewell D, Mant D. An analysis of randomized con-
trolled trials published in the US family medicine literature,
1987–1991. J Fam Pract 1994; 39(3): 236–242.
12. Williams HC, Seed P. Inadequate size of “negative” clinical
trials in dermatology. Br J Dermatol 1993; 128: 317–326.
13. Emerson JD, Colditz GA. Use of statistical analysis in the
New England Journal of Medicine. N Engl J Med 1983;
309(12): 709–713.
14. Fromm BS, Snyder VL. Research design and statistical proce-
dures used in the Journal of Family Practice. J Fam Pract
1986; 23(6): 564–566.
15. Longnecker DE. Support versus illumination: Trends in medi-
cal statistics. J Anesthesiol 1982; 57(2): 73–74.
16. Sawilowsky SS. Nonparametric tests of interaction in experi-
mental design. Rev Educ Res 1990; 60(1): 91–126.
17. Micceri T. The unicorn, the normal curve, and other improb-
able creatures. Psychol Bull 1989; 105(1): 156–166.
18. Harwell MR. A general approach to hypothesis testing for
nonparametric tests. J Exp Educ 1990; 58(2): 143–156.
19. Hunter MA, May RB. Some myths concerning parametric
and nonparmetric tests. Can Psychol 1993; 34(4): 384–389.
20. Pearson K. Contributions to the mathematical theory of evo-
lution. II. Skew variation in homogeneous material. Philo
Transac R Soc 1895; Series A: 343–414.
21. Geary RC. Testing for normality. Biometrika 1947; 34: 209–
242.
22. Nunnally J. Psychometric Theory, 2nd ed. New York:
McGraw-Hill; 1978.
23. Pearson ES, Please NW. Relation between the shape of popu-
lation distribution and the robustness of four simple test statis-
tics. Biometrika 1975; 62(2): 223–241.
24. Tan WY. Sampling distributions and robustness of t, f, and
variance-ratio in two samples and ANOVA models with re-
spect to departures from normality. Commun Stat 1982; A11:
2485–2511.
25. Dobi RA, Wilson MJ. A comparison of t-test, F-test, and co-
herence methods of detecting steady-state auditory-evoked
potentials, distortion-product otoacoustic emission, or other
sinusoids. J Acoust Soc Am 1996; 100(4/1): 2236–2246.
26. Bohning D, Hempfling A, Schelp FP, Schlattmenn P. The
P. D. Bridge and S. S. Sawilowsky
area between curves (ABC) measure in nutritional anthro-
pometry. Stat Med 1992; 11(10): 1289–1304.
27. Gilbert GH, Longmate J, Branch LG. Factors influencing the
effectiveness of mailed health surveys. Public Health Rep
1992; 107(5): 576–584.
28. Elashoff JD, Cantor RM, Shain S. Power and validity of meth-
ods to identify variability genes. Genet Epidemiol 1991; 8(6):
381–388.
29. Lucke JF. Students t test and the glasgow coma scale. Ann
Emerg Med 1996; 28(4): 408–413.
30. Maxfield M, Schweitzer J, Gouvier WD. Measures of central
tendency, variability, and relative standing in nonnormal dis-
tributions: Alternatives to the mean and standard score. Arch
Phys Med Rehabil 1988; 69(6): 406–409.
31. Zellner D, Frankewitsch T, Simon S, Keller F. Statistical anal-
ysis of hetrogenious pharmacokinetic data from the literature.
Eur J Clin Chem Clin Biochem 1996; 34(7): 585–589.
32. Niklasson A, Ericson A, Fryer JG. An update of the Swedish
reference standards for weight, length and head circumference
at birth for given gestational age (1977–1981). Acta Paediat-
rica Scandinavica 1991; 80(8-9): 756–762.
33. Valenta HL, Fischer SK. A monte carlo simulation of range
for an invasive impedance respiration monitor. Biomed Sci
Instrum 1990; 26: 181–184.
34. Colliver JA, Manchikanti L, Markwell SJ. Evaluation and
comparison of the distributions of gastric pH and hydrogen
ion concentration. Anesthesiol 1987; 67(3): 391–394.
35. Sullivan LM, D’Agostino RB. Robustness of the t test applied
to data distorted from normality by floor effects. J Dent Res
1992; 71(12): 1938–1943.
36. Blair RC. A reaction to “consequences of failure to meet as-
sumptions underlying the fixed effects analysis of variance and
covariance.” Rev Educ Res 1981; 51(4): 499–507.
37. Scheffe H. The Analysis of Variance. New York: Wiley;
1958.
38. Sawilowsky SS, Blair RC. A more realistic look at the robust-
ness and type II error properties of the t test to departures from
population normality. Psychol Bull 1992; 111(2): 352–360.
39. Mansfield E. Basic Statistics with Applications. New York:
W. W. Norton; 1986.
40. Sawilowsky SS. Comments on using alternatives to normal
theory statistics in social and behavioural science. Can Psy-
cho 1993; 34(4): 432–439.
41. Friedman JA, Chalmers TC, Smith H, Kuebler RR. The im-
portance of beta, the type II error and sample size in the de-
sign and interpretation of the randomized control trial: Survey
of 71 “negative” trials. N Engl J Med 1978; 299: 690–694.
42. Mengel MB, Davis AB. The statistical power of family prac-
tice research. Fam Pract Res J 1993; 13(2): 105–111.
43. Edlund MJ, Overall JE, Rhoades HM. Beta, or type II error in
psychiatric controlled clinical trials. J Psychiat Res 1985; 19
(4): 563–567.
44. Kraemer HC, Thiemann S. How Many Subjects?: Statistical
Power Analysis in Research. Newbury Park, CA: Sage Publi-
cations; 1987.
45. Cohen J. Statistical Power Analysis for the Behavioral Sci-
ences, 2nd ed. Hillsdale, NJ: Erlbaum; 1988.
46. Cohen JA. Power primer. Psychol Bull 1992; 112(1): 155–159.
47. Dixon WJ. Power under normality of several nonparametric
tests. Ann Mathemat Stat 1954; 25: 610–614.
48. Chernoff H, Savage IR. Asymptotic normality and efficiency
of certain nonparametric test statistics. Ann Mathemat Stat
1958; 29: 972–999.
49. Neave HR, Granger CWJ. A Monte Carlo study comparing
various two-sample tests for differences in mean. Technomet-
rics 1968; 10: 509–522.
Impact of Statistics on Research Outcomes
50. Lehmann EL. Nonparametrics. San Francisco, CA: Holden-
Day; 1975.
51. Randles RH, Wolfe DA. Introduction to the Theory of Non-
parametric Tests. New York: John Wiley; 1979.
52. Blair RC, Higgins JJ. A comparison of the power of the Wil-
coxon’s rank-sum statistic to that of Student’s t statistic under
various non-normal distributions. J Educ Stat 1980; 5(4):
309–335.
53. Blair RC, Higgins JJ. A note on the asymptotic relative effi-
ciency of the Wilcoxon rank-sum test relative to the indepen-
dent means t test under mixtures of two normal distributions.
Br J Math Stat Psychol 1981; 31: 124–128.
54. Zimmerman DW, Zumbo BD. Effects of outliers on the rela-
235
tive power of parametric and nonparametric statistical tests.
Percep Motor Skills 1990; 71: 339–349.
55. Conover WJ, Iman RL. Rank transformations as a bridge be-
tween parametric and nonparametric statistics. Am Stat 1981;
35: 124–129.
56. International Mathematical and Statistical Libraries. IMSL
Library Reference Manual. 10th ed. Houston, TX: 1987.
57. Hill JC, Schoener EP. The age-dependent decline of ADHD.
ADHD Report 1998; 6(1): 4–5.
58. Barkley RA. Age-dependent decline in ADHD: A final re-
joinder. ADHD Report 1998; 6(1): 7–8.
59. Sawilowsky S, Musial JL. Modeling ADHD as exponential de-
cay. ADHD Report 1998; 6(1): 10–11.