Guidelines and Decision Tables For Selecting The Appropriate Statistical Procedure

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Guidelines and Decision Tables for Selecting

the Appropriate Statistical Procedure


Tables 1.15-1.18 are designed to facilitate the selection of the appropriate statistical test.
These tables list the major inferential statistical procedures described in the book, based on the
level ofmeasurement the data being evaluated represent. Specifically, Table 1.15 lists inferential
statistical tests employed with intervallratiodata, Table 1.16 lists inferential statistical tests employed with ordinallrank-order data,* and Table 1.17 lists inferential statistical tests employed
with categorical/nominal data. Table 1.18 lists the measures of correlation/associationthat are
described in the book. Using the aforementionedtables, the following guidelines should be employed in selecting the appropriate statistical test.
1. Determineifthe analysis involves computing a correlation coefficientJmeasureofassociation
and, if it does, go to Table 1.18. The selection of the appropriate measure in Table 1.18 is
based on the level of measurement represented by each of the variables for which the
measure of correlation/association is computed.
2. If the analysis does not involve computing a measure of correlation/association, it will be
assumed that the data will be evaluated through use of an inferential statistical test. To
select the appropriate inferential statistical test, the following protocol should be employed.
a) State the general hypothesis that is being evaluated.
b) Determine if the study involves a single sample or more than one sample.
c) If the study involves a single sample, the appropriate test will be one of the tests for a
single sample in Tables 1.15,I. 16, or 1.17. In order to determine which table to employ,
determine the level of measurement represented by the data that are being evaluated. If
the level of measurement is intervallratio, Table 1.15 is employed. If the level of
measurement is ordinallrank-order, Table 1.16 is appropriate. If the level of
measurement is categoricallnominal, Table 1.17 is utilized.
d) If there is more than one sample, determine how many samples/treatmentsthere are and
whether they are independent or dependent. Determine the level of measurement represented by the data that are being evaluated (which represents the dependent variable in
the study).
1) Ifthe level ofmeasurement is intervallratio, go to Table 1.15. Identifythe test or tests
that are appropriate for that level of measurement with respect to the number and
types of samples employed in the study.
2) If the level of measurement is ordinallrank-order, go to Table 1.16. Identify the test
or tests that are appropriate for that level ofmeasurementwith respect to the number
and types of samples employed in the study.
3) Ifthe level ofmeasurement is categoricallnominal, go to Table 1.17. Identify the test
or tests that are appropriate for that level ofmeasurement with respect to the number
and types of samples employed in the study.

* In the case of the following three tests listed in Table 1.16, the dependent variable will be intervallratio
data which are converted into a format in which the resulting scores are rank-ordered: The Wilcoxon
signed-ranks test (Test 6), the Moses test for equal variability (Test IS), and the Wilcoxon matchedpairs signed-ranks test (Test 18).
Copyright 2004 by Chapman & Hal/CRC

Handbook of Parametric and Nonparametric Statistical Procedures

114

Table 1.15 Decision Table for Inferential Statistical Tests Employed with IntewalIRatio Data
Number of samples
Hypothesis evaluated

One independent
variable

Hypothesis about a population


mean

Single sample

Test

The single-samplez test (Test 1) (o known)


The single-sample t test (Test 2) (o unknown)

The single-sample chi-square test for a population


variance (Test 3)
The single-sample test for evaluating population
skewness (Test 4)
Hypothesisabout a
The single-sample test for evaluating population
parameterlcharacteristic other than
kurtosis (Test 5 )
the mean
The mean sauare successive difference test (for serial
randomness) (Test log)
The D'Agostino-Pearson test of normality (Test 5a)
Procedures for identifyingoutliers (Test 1If)

1
about difference
between two independent
Two
independent population means
samples

The t test for two independent samples (Test 11)


The z test for two independent samples ( ~ e s 1t ie)
The singlefactor between-subjects analysis of
variance (Test 2 1)
The singlefactor between-subjects analysis of
covariance (Test 2 lj)

Hypothesis about two independent Hartley's F-, test for homogeneity of variance1
population variances
F test for two population variances (Test 1la)
samples

dependent
samples

The t test for two dependent samples (Test 17)


Sandier's A test (Test 17d)
Hypothesis about difference
between two dependent population The z test for two dependent samples (Test 17e)
means
The single-factor within-subjectsanalysis of variance
(Test 24)

Hypothesis about two dependent


population variances

Hypothesis about difference


between
two or more independent
Two or more
independent 1Ipopulation means
Two
or more
samples
- - --- - - -

Hypothesis about two or more


independentpopulation variances
Hypothesis about difference
between two or more dependent
Two or more Domiation means
dependent
samples '
H
or more ~
I dependent population variances

1
Two independent
variables

Copyright 2004 by Chapman & Hal/CRC

Hypothesis about difference


between two or more population
means

The t test for homogeneity of variance for two


dependent samples (Test 17a)

The singlefactor between-subjects analysis of


variance (Test 2 1)
The singlefector between-subjects analysis of
covariance (Test 2 1j)
Hartley's Fm test for homogeneity of variance1
Ftest for two population variances (Test 1la)
The singlefactor within-subjects analysis of variance
(Test 24)
See discussion of sphericity assumption under the
single-factor within-subjects
~
analysis
~
of variance ~

,(Test 24)
l ~ h between-subjects
e
factorial analysis of variance
(Test 27)
The factorial analysis of variance for a mixed design
(Test 27i)
The within-subjects factorial analysis of variance
(Test 27j)

Guidelines and Decision Tables


Table 1.16 Decision Table for Inferential Statistical Tests Employed
with OrdinallRank-Order Data
Hypothesis evaluated

Number of simples

Test

The Wilcoxon signed-ranks test (Test 6)


Hypothesis about a population
The Kolmogorov-Smimov goodness-of-fit test for a
median or the distribution of data in single sample (Test 7)
The Lilliefors test for normality (Test 7a)
a single population
The single-sample test for the median (Test 9b)

Single sample

Hypothesis about two independent


w~ulationmedians. or some other
characteristic
(other than
Two
independent variability) of two independent
populations
samples
Two
samples

The Mann-Whitney U test (Test 12)


The randomization test for two independent samples
(Test l2a)
The bootstrap (Test 12b) (The bootstrap can also be
employed for evaluating hypotheses concerning
variability, as well as for evaluating various
hypotheses for single sample designs and designs
involving two or more independent or dependent
samples.)*
The iackknife (Test 12b) (The iackknife can also be
employed for evaluating' hypotheses concerning
variability, as well as for evaluating various
hypotheses for single sample designs and designs
involving two or more independent or dependent
samples.)*
The Kolmogorov-Smimov test for two independent
samples (Test 13)
The median test for independent samples
(Test 16e)
The van der Waerden normal-scores test for k
independent samples (Test 23)

Hypothesis about variability in two The SiegeI-Tukey test for equal variability (Test 14)
independent populations
The Moses test for equal variability (Test 15)
Hypothesis about the ordering of
data in two dependent populations

samples

he

Two
or more
samples

The Wilcoxon matched-pairs signed-ranks test (Test


18)
The binomial sign test for two dependent samples
(Test 19)

Kruskal-Wallis one-way- analysis


. of variance by
ranks (Test 22)
Hypothesis about two or more
The Jonckheere-Terpstra test for ordered alternatives
Two or more
independent population medians, or
independent some 0 t h characteristicof two or (Test 22a)
samples
The van der Waerden normal-scores test fork
more independent populations
independent samples (Test 23)
test for independent samples (Test 16e)

he median

IWo
Or
samples

~ypothesisabout two or molt


dependent populatiOn

The Friedman two-way analysis of variance by ranks


nest 25)
The Page test for ordered alte~natives(Test 25.1

*Although discussed under procedures for evaluating ordinal/rank-order data, the bootstrap and jackknife can be employed
to evaluate data representing any level of measurement.

Copyright 2004 by Chapman & Hal/CRC

Handbook of Parametric and Nonparametric Statistical Procedures


Table 1.17 Decision Table for Inferential Statistical Tests Employed
with CategoricaVNominal Data

Number of samples

Hypothesis evaluated

Test

The chi-square goodness-of-fit test (Test 8)


The binomial sign test for a single sample (Test 9)
The z test for a population proportion (Test 9a)
The single-sample runs test (Test 10)
Hypothesis about distribution of The runs test for serial randomness ( ~ e s tlOa)
The frequency test (for randomness) (Test lob)
data in a single population
The gap test (for randomness) (Test lOc)
The poker test (for randomness)(Test lOd)
The maximum test (for randomness) ( ~ e slOe)
t
The coupon collector's test (for randomness) (Test lOf
The chi-square test of independence (Test 16b)
Hypothesis about distribution of The chi-square test for homogeneity (Test 16a)
Two
independent data in two independent
The Fisher exact test (Test 16c)
samples populations
The z test for two independent proportions (Test 16d)

1 wo
'"mples

Two

l ~ h McNanar
e
test (Test 20)
Hvoothesis about distribution of The Cart test for order eSects (Test 20at
The Bowker test of internal symmetry (Test 20b)
The Stuart-Maxwell test of marginal homogeneity
(Test 20c)

The chi-square test for homogeneity (Test 16a)


Two
or more
samples
The Cochran Q test (Test 26)

Copyright 2004 by Chapman & Hal/CRC

Guidelines and Decision Tables

117

Table 1.18 Decision Table for Measures of Correlation/Association


Level of measurement
Bivariate
Interval/ratio
data

OrdinaVrank
order data

Categorical/
nominal data

More than two


sets of scores

Test

I he Pearson product-moment correlation coefficient (Test 28)

The intraclass correlation coefficient (Test 24i)


-

--

multiple correlation coefficient (Test 28k)


partial correlation coefficient 281)
The semipartial correlation coefficient (Test 28m)
Spearman's rank-order correlation coefficient (Test 29)
Kendall's tau (Test 30)
Goodman and Kruskal's gamma (for ordered contingency tables) (Test 32)
More than two
sarnpledsets of
ranks

Kendall's coefficient of concordance(Test 3 1)

The contingency coefficient (Test 160


The phi coefficient (Test 16g)
Fwo dichotomous Yule's Q (Test 16i)
variables*
The odds ratio (Test 16j)
Cohen's kappa (Test 16k)
Binomial effect size display (Test 28r)

nOndichOtOmOus

The contingency coefficient (Test 16f)


Cramer's phi coefficient (Test 16h)
The odds ratio (Test 16j)
Cohen's kappa (Test 16k)

Omega squared (One variable, interval/ratio data; second variable, two or more
nominal levels) (Tests 11d17d2 1g/24g/27g)
Eta squared (One variable, interval/ratio data; second variable, two or more
nominal levels) (Test 1Id (two nominal levels); Test 2 1h)
Cohen's d index (and g index sample analogue) (Test 1l bIl7b) (One variable,
interval/ratio data; second variable, two nominal levels). (with
Test 2a for one
.
variable)
kher bivariate correlational
ieasures for which interval/ratio Cohen'sfindex (One variable, interval/ratio data, second variable, two or more
ata are employed or implied for at nominal levels) (Test2 1i724hi27h)
a s t one of the variables
The point-biserial correlation coefficient (One variable, interval/ratio data;
second variable represented by dichotomous categories) (Test 28h)
The biserial correlation coefficient(One variable, intervallratio data; second
variable, an interval/ratio variable expressed in form of dichotomous categories)
(Test 28i)
The tetrachoric correlation coefficient (Two interval/ratio variables, both of
which are expressed in the form of dichotomouscategories) (Test 28j)

*A dichotomous variable is comprised of two mutually exclusive categories.

Copyright 2004 by Chapman & Hal/CRC

You might also like