Nonparametric Testing in Excel

The Excel Statistical Master

Nonparametric Testing
in Excel
The Excel Statistical Master
By Mark Harmon
Copyright 2011 Mark Harmon
No part of this publication may be reproduced
or distributed without the express permission
of the author.
[email protected]
www.ExcelMasterSeries.com
ISBN: 978-0-9833070-4-4

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 1

Nonparametric Testing in Excel

The Excel Statistical Master

Table of Contents
(Click On Chapters To Go To Them)
Nonparametric Tests - When To Use Them.6
The Mann-Whitney U Test Done in Excel..7
Step 1 - Combine all samples into 1 group.8
Step 2 - Rank All Samples...9
Step 3 - Put Samples Back In Original Groups.9
Step 4 - Count the Rankings For Each Group.10
Step 5 - Calculate the U Statistic.10
Step 6 - Calculate the Mean, U , and Standard Deviation, U..11
Step 7 - Calculate the Z Score..11
Step 8 - Compare this Z Score to Z Critical..11
The Kruskal-Wallis Test - Done in Excel..12
Step 1 - Combine all samples into 1 group..13
Step 2 - Rank All Samples.14
Step 3 - Put samples back in original groups.15
Step 4 Calculate the Following Metrics.16
Step 6 Calculate H Critical..17
The Wilcoxon Rank Sum Test.18
Test Procedures in Excel..19
Wilcoxon Rank Sum Test for Small Samples in Excel..20
Step 1 - Count the samples in Each Group..21
Step 2 - Rank all samples according to sample size in Excel.21

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 2

Nonparametric Testing in Excel

The Excel Statistical Master

Step 3 - Combine All Samples and Rank Them..22

Step 4 - Re-sort the Rows According To Day..22
Step 5 - Calculate the Sum of the Ranks (N1 and N2) For Each Sample Group23
Step 6 - Determine whether R = N1 or N2.23
Step 7 - Use the Wilcoxon Rank Sum Small Sample Chart to Evaluate R24
Wilcoxon Rank Sum Test For Large Samples in Excel.26
Step 1 - Count the Samples in Each Group.27
Step 2 - Rank All Samples According To Sample Size in Excel.27
Step 3 - Combine All Samples and Rank Them..28
Step 4 - Re-sort the Rows According to the Day of the Week29
Step 5 - Calculate the Sum of the Ranks (N1 and N2) For Each Sample Group30
Step 6 - Determine Whether R = N1 or N2.31
Step 7 - Calculate Z Score and Z Critical to Determine Whether the Sample Groups Come From Different
Populations...31
Calculate the Z Score.32
Calculate Z Critical..32
The Spearman Correlation Coefficient Test Done in Excel.33
Step 1 - Rank the Data34
Step 2 - Calculate the Difference Between Ranks 1 and 2, and Then Square That Difference..34
Step 3 Calculate the Spearman Rank Correlation Coefficient.35
The Sign Test (Nonparametric) Done in Excel39
Step 1 - Take the difference between each Before and After Data Point41
Step 2 - Record Whether That Difference Is Positive or Negative41
Step 3 - Record the Total Number of Positive and Negative Results41
Large Samples - 25 or More Samples43
Wilcoxon Signed-Rank Test in Excel for Small Samples.44

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 3

Nonparametric Testing in Excel

The Excel Statistical Master

Step 1 - Calculate Differences Between Before and After Data For Each Pair...45
Step 2 - Take the Absolute Value of that Difference..46
Step 3 - Rank of the Differences and Record the Sign of Each Difference.47
Step 4 - Combine the Rank of Each Difference With the Sign of the Difference48
Step 5 - Calculate the Test Statistic W..48
Wilcoxon Signed-Rank Test in Excel For Large Samples50
1) Before and After Sales Data52
2) Calculate Differences Between Before and After Data For Each Pair..53
3) Calculate the Absolute Value of the Difference and Then the Rank and Sign of that Difference.54
4) Combine the Rank and Sign of Each Difference55
5) Calculate the Sums of All Ranks56
6) Calculate the Test Statistic W.57
Calculate the Z Score for Test Statistic T.57
Calculate the p Value That Corresponds With That Z Score...59
Statistical Mistakes You Dont Want To Make.61
1) Assuming that correlation equals causation..61
2) Not graphing and eyeballing the data prior to performing regression analysis61
3) Not doing correlation analysis on all variables prior to performing regression...61
4) Adding a large number of new input variables into a regression analysis all at once62
5) Applying input variables to a regression equation that are outside of the value of the original input
variables that were used to create the regression equation62
6) Not examining the residuals in regression..62
7) Only evaluating r square in a regression equation62
8) Not drawing a representative sample from a population.63
9) Drawing a conclusion without applying the proper statistical analysis..63
10) Drawing a conclusion before a statistically significant result has been reached..63

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 4

Nonparametric Testing in Excel

The Excel Statistical Master

11) Analyzing non-normal data with the normal distribution..63

12) Not removing outliers prior to statistical analysis..64
13) Not controlling or taking into account other variables..64
14) Using the wrong t test.64
15) Attempting to apply the wrong type of hypothesis test.64
16) Not using Excel.65
17) Always requiring 95% certainty65
18) Thinking it is impossible to get a statistically significant sample if your target market is large65
19) Not taking steps to ensure that your sample is normally distributed when analyzing with the normal
distribution65
20) Using covariance analysis instead of correlation analysis.66
21) Using a one-tailed test instead of a two-tailed test when accuracy is needed..66
22) Not using nonparametric tests when analyzing small samples of unknown distribution..66
Correctable Reasons Why Your Data Is Not Normally Distributed67
What can you do if your data does not appear to be normally distributed?.....................................................67
The Biggest 7 Correctable Causes of Non-Normality in Data Samples..67
1) Outliers.67
2) Data has been affected by more than one process..67
3) Not enough data.68
4) Measuring devices that have poor resolution68
5) A different distribution describes the data.68
6) Data approaching zero or a natural limit.68
7) Only a subset of process output is being analyzed69

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 5

Nonparametric Testing in Excel

The Excel Statistical Master

Nonparametric Tests - When To Use Them

Statistical procedures are either parametric or nonparametric. Parametric statistical tests require assumptions
about the population from which the samples are drawn. For example, many tests such as the t Test, ChiSquare tests, z Tests, and F tests, and many types of hypothesis tests require the underlying population to be
normally distributed. Some tests require equal variances of both populations.
These assumptions cannot be always be assumed. Examples of this would be if the population is highly
skewed or if the underlying distribution or variances were entirely unknown. Creating a quick Histogram of the
data is fast way to check whether the distribution is skewed or not according any specific distribution. This is
one brief step that would always be run on available population or sample data.
Nonparametric tests have no assumptions regarding distribution of underlying populations or variance. Most
nonparametric tests are very easy to perform but they are not usually as precise as parametric tests and their
Null Hypotheses usually requires more evidence to be rejected than would be the case in a parametric test.

Nonparametric tests are often used as shortcut replacements for more complicated parametric tests. You can
usually get a quick answer that requires little calculation by running a nonparametric test.
Nonparametric tests are often used when the data is ranked but cannot be quantified. For example, how would
you quantify consumer rankings such as very satisfied, moderately satisfied, just satisfied, less than satisfied,
dissatisfied?
Nonparametric tests can be applied when there are a lot of outliers that might skew the results. Nonparametric
tests often evaluate medians rather than means and therefore if the data have one or two outliers, the outcome
of the analysis is not affected.
Nonparametric tests come in especially handy when dealing with non-numeric data, such as having customers
rank products or attributes according to preference.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 6

Nonparametric Testing in Excel

The Excel Statistical Master

The Mann-Whitney U Test

Done in Excel
- The Mann-Whitney U Test is a great substitute for the two-sample t test when it can't be determined that both
samples come from Normally-distributed populations.
- Is used to decide if whether or not there is a difference between two groups or, equivalently, whether or not
both groups come from the same population.

Procedures
Objective: To determine if these two groups come from different populations

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 7

Nonparametric Testing in Excel

The Excel Statistical Master

Step 1 - Combine All Samples Into 1 Group

Create a mechanism to Keep Track of Each Sample's Group. In this case, each sample from the 1st group
was labeled with an "A" and each sample from the 2nd group was labeled with a "B."

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 8

Nonparametric Testing in Excel

The Excel Statistical Master

Step 2 - Rank All Samples. Sort the sample by the value of each sample and rank them
in order of sample value.

Step 3 - Put Samples Back In Original Groups Sort by the A and B column

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 9

Nonparametric Testing in Excel

The Excel Statistical Master

Step 4 - Count the Rankings For Each Group

N1 = Count of Samples in 1st Group = 7

N2 = Count of Samples in 2nd Group = 7

Step 5 - Calculate the U Statistic by the following formula:

U = N1*N2 + [ ( N1 * (N1 + 1) ) / 2 ] - R1 = 28

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 10

Nonparametric Testing in Excel

The Excel Statistical Master

Step 6 - Calculate the Mean, U , and Standard Deviation, U, by the following

formulas:

U = N1 * N2 / 2 = 24.5
U = SQRT [ ( N1 * N2 * ( N1 + N2 + 1) ) / 12 ] = 7.826

Step 7 - Calculate the Z Score With the Following Formula:

Z Score = ( U - U ) / U = 0.447

Step 8 - Compare this Z Score to Z Critical for the Given

For a one-tailed test, Z Critical = NORMSINV()
For a two-tailed test, Z Critical = NORMSINV(/2)
If We Require a 95% Degree of Certainty, Then = 0.05
For a one-tailed test, Z Critical = NORMSINV() = NORMSINV(0.05) = - 1.645
For a two-tailed test, Z Critical = NORMSINV(/2) =NORMSINV(0.025) - 1.960

Rule
The Result if Significant If | Z Score | > | Z Critical |

In This Case, | Z Score | (0.447) > | Z Critical |

(1.96 for two-tailed and 1.645 for one-tailed)

So The Result Is Not Statistically Significant and We Cannot State With 95% Certainty That There is a
Difference Between The Two Groups for Either a One-Tailed Test Or a Two-Tailed Test.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 11

Nonparametric Testing in Excel

The Excel Statistical Master

The Kruskal-Wallis Test - Done in Excel

- Is used to decide whether or not there is a difference between more than two groups or, equivalently, whether
or not all groups come from the same population.
- Is a good substitute for Single-Factor (One-Way) ANOVA when it cannot be determine that the samples are
drawn from Normally-distributed populations. It is almost identical to the one-way (single-factor) ANOVA except
that the data is replaced by the ranks of each piece of data.

Procedures
Objective: To determine if these three groups come from different populations

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 12

Nonparametric Testing in Excel

The Excel Statistical Master

Step 1 - Combine all samples into 1 group.

Create a mechanism to keep track of each sample's group. In this case, each sample in the 1st group was
labeled "A," each sample in the 2nd group was labeled "B," and each sample in the 3rd group was labeled "C."
This step is taken so that the data can be sorted by this column to return all records to their original groups.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 13

Nonparametric Testing in Excel

The Excel Statistical Master

Step 2 - Rank All Samples

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 14

Nonparametric Testing in Excel

The Excel Statistical Master

Step 3 - Put samples back in original groups. Do this by sorting by the A B C

column.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 15

Nonparametric Testing in Excel

The Excel Statistical Master

Step 4 Calculate the Following Metrics

N1 = Count of Samples in 1st Group = 7
N2 = Count of Samples in 2nd Group = 7
N3 = Count of Samples in 1st Group = 7
n = Total Number of Samples = 21
k = Number of Groups = 3
df = Degrees of Freedom = k - 1 = 2

Step 5 - Calculate the H Statistic by the following formula:

H Statistic =
[ 12 / { n*(n + 1) } ] * [ ( R1^2)/N1 + (R2^2)/N2 + + (RN^2)/NN ] 3*(n + 1)
H Statistic = -33.818

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 16

Nonparametric Testing in Excel

The Excel Statistical Master

Step 6 Calculate H Critical

Test Statistic H is very nearly distributed as the Chi-Square Distribution With k - 1 Degrees of Freedom
as long as the number of samples in each group Is at least 5.
Therefore the Critical Value Can Be Found By The Following Excel Formula:
H Critical = CHIINV( , df )
If we wish to be at least 95% of whether or not the result is statistically significant, then = 0.05
H Critical = CHIINV( , df ) = 5.991

Rule
The Result if Significant If H Statistic > H Critical
In This Case, H Statistic > H Critical

So The Result Is Statistically Significant and We Can State Within 95% That There is Difference Between
The Groups, Or, Equivalently, They Come From Different Populations.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 17

Nonparametric Testing in Excel

The Excel Statistical Master

The Wilcoxon Rank Sum Test

Done in Excel
The Wilcoxon Rank Sum Test has the same purpose as the t Test; to determine whether there is a difference
between two groups. Specifically, we are trying to determine whether the two samples came from the same
populations or not. The Wilcoxon Rank Sum Test can be used in place of the t Test when either of the sample
groups does not appear to be normally distributed.
The Wilcoxon Rank Sum Test is a hypothesis test and has the following Null Hypothesis and an Alternate
Hypothesis.
H0 - Null Hypothesis states that the two samples came from the same populations. In other words, there is no
difference between the two groups.
H1 - Alternate Hypothesis states that the two samples do not come from the same populations.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 18

Nonparametric Testing in Excel

The Excel Statistical Master

Test Procedures in Excel

Large and Small sampling require different procedures for the Wilcoxon Rank Sum Test. The first couple steps
of this test will be the same for both large and small samples:
We will collect two groups of samples and then calculate the following characteristics of each group:

n1 = Sample size of the 1st group - Number of samples in the group

n2 = Sample size of the 2nd group - Number of samples in the group
N1 = Sum of the ranks of all samples in the 1st group
N2 = Sum of the ranks of all samples in the 2nd group
R = either N1 or N2, whichever comes from the group with the smaller sample size
On the next pages we will show how to calculate N1 and N2 in Excel.
First, we have to determine whether we are dealing with small samples or large samples. The small sample
procedure is used if n1 and n2 are both each less than 10. Use the large sample procedure is there are more
than 20 samples total ( n1 + n2 > 20).
The small sample procedure and the large sample procedure calculate n1, n2, N1, N2, and R the same way.
Small samples and large samples use different procedures beyond this point.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 19

Nonparametric Testing in Excel

The Excel Statistical Master

Wilcoxon Rank Sum Test

for Small Samples In Excel
"Small samples" occur when the sample size of both groups is less than 10. In this case, n1 and n2 are both
less than 10.
Problem: We have collected sales data from the last 6 Tuesdays and the last 7 Wednesdays. We want to
determine if Tuesday sales have been different from Wednesday sales over the last month and a half. The data
in either group does not appear to be normally distributed. In this case we will have to use a nonparametric test
to determine whether the groups are different. The Wilcoxon Rank Sum Test for Small Samples is ideal for this
situation.
Here are the data samples for each of the two groups:

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 20

Nonparametric Testing in Excel

The Excel Statistical Master

Step 1 - Count the samples in Each Group

Sample Size =
n1 = 7
n2 = 6

Step 2 - Rank all samples according to sample size in Excel

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 21

Nonparametric Testing in Excel

The Excel Statistical Master

Step 3 - Combine All Samples and Rank Them. If any samples have the same
values, assign them the rank that would be the average of the two ranks if they weren't equal.

Step 4 - Re-sort the Rows According To Day

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 22

Nonparametric Testing in Excel

The Excel Statistical Master

Step 5 - Calculate the Sum of the Ranks (N1 and N2) For Each
Sample Group

Step 6 - Determine whether R = N1 or N2

R is taken from the sample with the smaller size.
n1 = 7 and n2 = 6 so sample group 2 is smaller
Therefore R = N2 = 40

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 23

Nonparametric Testing in Excel

The Excel Statistical Master

Step 7 - Use the Wilcoxon Rank Sum Small Sample Chart to Evaluate R to
Determine Whether R Falls Within or Outside the Lower and Upper Critical
Values for the Given n1 and n2:
Below is the chart that is referenced. The green cells represent the n1, n2, and upper and lower limits for that
n1 and n2.

The chart on the previous page is a standard Critical Value chart for the Wilcoxon Small Sample Rank Sum
Test.
If R = 40, N1 = 7, and N2 = 6, then R is between the upper and lower critical values for both = 0.025 (Critical
Values = 28 and 56) and = 0.05 (Critical Values = 30 and 54). That is shown on the above chart.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 24

Nonparametric Testing in Excel

The Excel Statistical Master

We therefore do not reject the Null Hypothesis, which states that there is no difference between the two sample
groups.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 25

Nonparametric Testing in Excel

The Excel Statistical Master

Wilcoxon Rank Sum Test

for Large Samples in Excel
"Large samples" imply that both n1 and n2 add up to more than 20 and that each is greater than 10.

Problem: We have collected sales data from the last 14 Tuesdays and the last 12 Wednesdays. We want to
determine if Tuesday sales have been different from Wednesday sales over the last month and a half. Calculate
with at least 95% certainty whether there is a difference between the two groups.
The data in either group does not appear to be normally distributed. In this case we will have to use a
nonparametric test to determine whether the groups are different. The Wilcoxon Rank Sum Test for Large
Samples is ideal for this situation.

For large samples, you will calculate n1, n2, N1, N2, and R as before.
Here are the data samples for each of the two groups:

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 26

Nonparametric Testing in Excel

The Excel Statistical Master

Step 1 - Count the Samples in Each Group

Sample Size =
n1 = 14
n2 = 12

Step 2 - Rank All Samples According To Sample Size in Excel

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 27

Nonparametric Testing in Excel

The Excel Statistical Master

Step 3 - Combine All Samples and Rank Them. If any samples have the same
values, assign them the rank that would be the average of the two ranks if they weren't equal.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 28

Nonparametric Testing in Excel

The Excel Statistical Master

Step 4 - Re-sort the Rows According to the Day of the Week

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 29

Nonparametric Testing in Excel

The Excel Statistical Master

Step 5 - Calculate the Sum of the Ranks (N1 and N2) For Each
Sample Group

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 30

Nonparametric Testing in Excel

The Excel Statistical Master

Step 6 - Determine Whether R = N1 or N2

R is taken from the sample with the smaller size.
n1 = 14 and n2 = 12 so sample group 2 is smaller
Therefore R = N2 = 178

Step 7 - Calculate Z Score and Z Critical to Determine Whether the Sample

Groups Come From Different Populations.
When n1 and n2 are both > 10, the Normal distribution can approximate the distribution of R.
If the Z Score of R is greater than Z Critical, we reject the Null Hypothesis and state that the two sample groups
come from different populations. If the Z Score of R is less than Z Critical, we do not reject the Null Hypothesis
and state that both sample groups came from the same populations.

Z Score Calculation
Calculate r and r

r = [ n1 * ( n1 + n2 + 1 ) ] / 2
r = SQRT [ ( n1 * n2 * ( n1 + n2 + 1 ) ) / 12 ]
r = [ 14 * ( 14 + 12 + 1 ) ] / 2 = 189
r = SQRT [ ( 14 * 12 * ( 14 + 12 + 1 ) ) / 12 ]

Copyright 2011

= 19.44

http://excelmasterseries.com/New_Manuals.php

Page 31

Nonparametric Testing in Excel

The Excel Statistical Master

Calculate the Z Score

Z = ( R - r ) / r = -0.5658

Calculate Z Critical
Zcrit = NORMSINV ( 1 - / 2 ) For example, for = 0.05 and two-tailed test

Zcrit = NORMSINV ( 1 - 0.05 / 2 ) = NORMSINV (0.975) = 1.96

Rule
Reject H0 if | Z | > | Z Critical | for the given and correct number of tails
In this case, |Z| is less than | Z Critical | so we do not reject the Null Hypothesis. We state that we are 95%
certain that the two sample groups were taken from the same population.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 32

Nonparametric Testing in Excel

The Excel Statistical Master

The Spearman Correlation Coefficient Test

Done in Excel
- Used to determine if the strength and direction of a relationship between 2 sets of numbers.
- Is sometimes considered to be similar to the Pearson Correlation Coefficient between the ranks of numbers.

Procedures
Objective: To determine if there is a correlation between these two sets of numbers:

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 33

Nonparametric Testing in Excel

The Excel Statistical Master

Step 1 - Rank the Data.

Step 2 - Calculate the Difference Between Ranks 1 and 2, and Then Square
That Difference.

(d^2) = 18
n = Number of Samples = 7

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 34

Nonparametric Testing in Excel

The Excel Statistical Master

Step 3 Calculate the Spearman Rank Correlation Coefficient

Spearman Rank Correlation Coefficient, rs, = 1 [ 6*(d^2) ] / [n*(n^2 1)]

0.6786

Rule:
rs is statistically significant if the following p Value is less than :

p Value = FDIST [ ( df * rs^2) / ( 1 - rs^2), 1, df ]

df = degrees of freedom = n - 2 = 5
p Value = 0.001635228
If We Require a 95% Degree of Certainty, Then = 0.05
In This Case, We Can Be At Least 95% Certain That There Is a Relationship Between The Variables.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 35

Nonparametric Testing in Excel

The Excel Statistical Master

Here Is a Quick Way To Assign Ranking To Samples In a Data Set

Such As This:

Step 1 - Assign the Count to Each Sample According Sample Size

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 36

Nonparametric Testing in Excel

The Excel Statistical Master

Step 2 - Sort Both Columns By the Original Data Column (Right Column)
and Assign a Numerical Rank (New Left Yellow-Highlighted Column)

Step 3 - Sort All 3 Columns By the Middle Column

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 37

Nonparametric Testing in Excel

The Excel Statistical Master

Step 4 Delete the Middle Column, Leaving the Original Data

Column (Right) and the Rank of Record (Left Column)

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 38

Nonparametric Testing in Excel

The Excel Statistical Master

The Sign Test (Nonparametric)

Done in Excel
1) Used to determine whether a population median is equal to, less than, or greater than a specific number
2) Used to determine whether there is a real difference between pairs of data, such as before and after
measurements
3) Has wide application but is not as statistically accurate as the parametric tests such as the paired samples t
test.

Procedures
Objective: To determine whether Before and After data are different.
Small samples - less than 25 samples
For Case 1 above, in the event of testing to determine whether a population median is equal to, less than, or
greater to a specific number, subtract the sample from the fixed number.
For Case 2 above, in the event of testing to determine whether there is a difference between before and after
measurements, subtract the before-measurement from the after-measurement.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 39

Nonparametric Testing in Excel

The Excel Statistical Master

Small Sample Exercise: Determine Whether the Before and After Data is Different:
Here is the original Before and After Data:

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 40

Nonparametric Testing in Excel

The Excel Statistical Master

Step 1 - Take the Difference Between Each Before and After Data Point.
Step 2 - Record Whether That Difference Is Positive or Negative.
Step 3 - Record the Total Number of Positive and Negative Results.

X = Lesser of (+) or (-) Numbers = 1

N = Sum of (+) and (-) Numbers = 3 + 5 = 9

X = Lesser of (+) or (-) Numbers = 1

N = Sum of (+) and (-) Numbers = 3 + 5 = 9

p Value = -2*BINOMDIST(X, N, 0.5, 1) = 0.0390625

The p value is the total area under the outer 2 tails in a 2-tailed test

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 41

Nonparametric Testing in Excel

The Excel Statistical Master

Rule
The Result is Significant if p Value <
If 95% certainty is required, then = 0.05

In This Case, the Result Is Significant and We Can State That There Is a Difference Between the Before
and After Numbers.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 42

Nonparametric Testing in Excel

The Excel Statistical Master

Large Samples - 25 or More Samples

Plug the X and N into the Following Formula
Z Score = [ (X + 0.5 ) (N 2) ] / [ SQRT(N) / 2 ]

Then Compare this Z Score to Z Critical for the Given

For a one-tailed test, Z Critical = NORMSINV()

For a two-tailed test, Z Critical = NORMSINV(/2)

Rule
The Result if Significant If | Z Score | > | Z Critical |

For Example:
X=3
N=9
Z Score = [ (X + 0.5 ) (N 2) ] / [ SQRT(N) / 2 ] = -2.33
If We Require a 95% Degree of Certainty, Then = 0.05
For a one-tailed test, Z Critical = NORMSINV() = -1.64
For a two-tailed test, Z Critical = NORMSINV(/2) = -1.96
In this case, we can state with at least 95% certainty that the result is significant and that there is a
difference between Before and After data for both the one-tailed and two-tailed tests.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 43

Nonparametric Testing in Excel

The Excel Statistical Master

Wilcoxon Signed-Rank Test in Excel

for Small Samples
The Wilcoxon Signed-Rank Test is used as an alternate to the Paired Student's t Test when the underlying
population cannot be determined to be normally distributed. This test is used to determine whether there is a
difference between paired samples.
Unlike another similar nonparametric test - the Paired Sample Sign Test - this test considers the magnitude of
differences between the paired samples.
The Wilcoxon Signed-Rank Test is performed differently for small samples than it is for large samples. Small
samples mean sample groups with less than 30 pairs of sample data.
The Null Hypothesis for the Wilcoxon Signed-Rank Test states that there is no difference between the paired
samples. For example, this Null Hypothesis might state that there is no difference between before and after
data, if each pair represents a before and after measurement of one sampled object.
The Alternate Hypothesis states that there is a difference between the paired samples.

Procedures For the Wilcoxon Signed-Rank Test

In this test, we are trying to determine if landing pages produce greater average sales when a bonus eBook is
added to the offering. We wish to be at least 95% certain whether average sales do increase when an eBook is
added to the offering on each landing page.
A landing page is that web page on a web site that a visitor is directed to (lands on) when the visitor clicks on
an ad that is external to the web site. The ad can be any type of ad, such as a banner ad in another web site, a
Google AdWords ad, or a link in a promotional email.
Bonus eBooks were added to the offering that appeared on 9 separate landing pages. Average daily sales from
each landing page were recorded for one month before and one month after the bonus eBook was added to
each landing page's offering. All 9 landing pages were very similar and were the destination URLs (landing

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 44

Nonparametric Testing in Excel

The Excel Statistical Master

pages) of similar Google AdWords ads.

We do not know whether sales results from the landing pages are normally distributed so we need to use a
nonparametric test. The most accurate nonparametric test for paired data is the Wilcoxon Signed-Rank Test.
Here is how it's done in Excel:

Here is the Original Before and After Sales Data:

Step 1 - Calculate Differences Between Before and After Data For

Each Pair

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 45

Nonparametric Testing in Excel

The Excel Statistical Master

Step 2 - Take the Absolute Value of that Difference

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 46

Nonparametric Testing in Excel

The Excel Statistical Master

Step 3 - Rank of the Differences and Record the Sign of Each

Difference
When the differences are the same, assign to each the average rank between them.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 47

Nonparametric Testing in Excel

The Excel Statistical Master

Step 4 - Combine the Rank of Each Difference With the Sign of the
Difference

Calculate the following two numbers:

a) The absolute value of the sum of all positive sign ranks.
b) The absolute value of the sum of all negative sign ranks.
The Test Statistics W equals the lesser of the two absolute values.
| Sum of all positive sign ranks | = | 5 + 5 + 2.5 + 7.5 + 9 + 7.5 + 2.5 + 5 | = | 44 | = 44
| Sum of all negative ranks | = | -1 | = 1
The Test Statistic W therefore equals 1.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 48

Nonparametric Testing in Excel

The Excel Statistical Master

Rule
Reject the Null Hypothesis if Test Statistic W is greater than or equal to W Critical
Below is a standard chart containing W Critical Values we need:

W Critical for n = 9 (9 pairs of sample data) for a one-tailed test in which alpha = 0.05 equals 8.
Since Test Statistics W (1) is less than W Critical (8), we reject the Null Hypothesis and state with at least 95%
certainty that adding a bonus eBook to the offering on each landing page does increase the average sales from
each landing page.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 49

Nonparametric Testing in Excel

The Excel Statistical Master

Wilcoxon Signed-Rank
Test in Excel for
Large Samples
The Wilcoxon Signed-Rank Test is used as an alternate to the Paired Student's t Test when the underlying
population cannot be determined to be normally distributed. This test is used to determine whether there is a
difference between paired samples. The most widely used application of the Wilcoxon Signed-Rank Test is to
determine whether there really is a difference between before and after data. In this article, we will show how to
use the Wilcoxon Signed-Rank Test in Excel to determine whether sales really did increase across a group of
landing pages after a free bonus was added to the offer on each landing page.

Unlike another similar nonparametric test - the Paired Sample Sign Test - this test considers the magnitude of
differences between the paired samples.
The Wilcoxon Signed-Rank Test is performed very differently for large samples than it is for small
samples. Large samples mean sample groups with more than 30 pairs of sample data.
The Null Hypothesis for the Wilcoxon Signed-Rank Test states that there is no difference between the paired
samples. For example, this Null Hypothesis might state that there is no difference between before and after
data, if each pair represents a before and after measurement of one sampled object.
The Alternate Hypothesis states that there is a difference between the paired samples.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 50

Nonparametric Testing in Excel

The Excel Statistical Master

Procedures For the Wilcoxon Signed-Rank Test

In this test, we are trying to determine if landing pages produce greater average sales when a bonus eBook is
added to the offering. We wish to be at least 95% certain whether average sales do increase when an eBook is
added to the offering on each landing page.
A landing page is that web page on a web site that a visitor is directed to (lands on) when the visitor clicks on
an ad that is external to the web site. The ad can be any type of ad, such as a banner ad in another web site, a
Google AdWords ad that came up during a Google search, or a link in a promotional email.
Bonus eBooks were added to the offering that appeared on 35 separate landing pages. Average daily sales
from each landing page were recorded for one month before and one month after the bonus eBook was added
to each landing page's offering. All 35 landing pages were very similar and were the destination URLs (landing
pages) of similar Google AdWords ads.
We do not know whether sales results from the landing pages are normally distributed so we need to use a
nonparametric test. The most accurate nonparametric test for paired data is the Wilcoxon Signed-Rank Test.
Here is how it's done in Excel for large samples (sample size is greater than 30):

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 51

Nonparametric Testing in Excel

The Excel Statistical Master

1) Before and After Sales Data

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 52

Nonparametric Testing in Excel

The Excel Statistical Master

2) Calculate Differences Between Before and After Data For Each

Pair On Next Page

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 53

Nonparametric Testing in Excel

The Excel Statistical Master

3) Calculate the Absolute Value of the Difference and Then the

Rank and Sign of that Difference When the differences are the same, assign to each the
average rank between them:

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 54

Nonparametric Testing in Excel

The Excel Statistical Master

4) Combine the Rank and Sign of Each Difference

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 55

Nonparametric Testing in Excel

The Excel Statistical Master

5) Calculate the Sums of All Ranks

Sum of All Ranks = 224.75 - 216.25 = 8.5

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 56

Nonparametric Testing in Excel

The Excel Statistical Master

6) Calculate the Test Statistic W

Test Statistic W equals the sum of the signed ranks, which was calculated above:
W = 8.5

Since sample size is greater than 30, the distribution of Test Statistic T can be approximated by the
Normal Distribution.

7) Calculate the Z Score for Test Statistic T

With large sample sizes (sample size, n, greater than 30), the distribution of Test Statistic W can approximated
with the Normal Distribution. We can therefore calculate a Z Score for Test Statistic W.
The Z Score is the number of standard errors that Test Statistic T is from the mean of the Normal Distribution
curve that was use to approximate the distribution of Test Statistic T.
There are 2 pieces of information that are needed in order to construct the specific Normal Distribution curve
that is used to approximate the distribution of Test Statistic W. These 2 pieces of information are the 2
parameters necessary to construct any Normal Distribution curve. They are:

- Mean (of Test Statistic W) = W

- Standard Deviation (of Test Statistic W) = W

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 57

Nonparametric Testing in Excel

The Excel Statistical Master

The Mean of the Normal Distribution curve used to estimate the distribution of Test Statistic W (W) is
determined by the Null Hypothesis. The Null Hypothesis states that there is no difference between the before
and after sales. Put another way, the Null Hypothesis states that the difference between the before and after
sales equals zero. Therefore the mean of Test Statistic W and the Normal Distribution curve that is used to
approximate its distribution also equals zero. So;

W = 0
The standard deviation of Test Statistic W and therefore of the Normal Distribution curve that is used to
approximate the distribution of Test Statistic W is found by the following formula:

= SQRT [ ( n * (n+1) * (2n + 1) ) / 6 ]

n equals the number of before-and-after pairs of sample data whose difference does not equal zero. In this
case n = 32.

The above formula for

W can be derived from the fact that for any particular value of n, it can be shown that

the standard deviation of the sampling distribution of W is equal to:

= SQRT [ ( n * (n+1) * (2n + 1) ) / 6 ]

A number of sources incorrectly state that the above formula is as follows:

= SQRT [ ( n * (n+1) * (2n + 1) ) / 24 ]

Since n = 32

= 106.96 which can be rounded off to W = 107

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 58

Nonparametric Testing in Excel

The Excel Statistical Master

We can now map the Normal Distribution curve that is used to approximate the distribution of Test Statistic W
as follows:

The Z Score of Test Statistic W can be determined by the following formula:

Z Score = ( W -

0.5)

/ W

The Z Score above includes a continuity correction, to account for the fact that we use a continuous probability
function (the Normal Distribution) and our test value is based on discrete (ordinal, ranked) data. If the difference
W - W is negative the continuity correction is +0.5 (this is the case when W is less than; W). If the difference
W - W is positive the correction is -0.5 (this is the case when W is greater than W).
Z Score = ( W -

0.5)

/ W

Z Score = ( 8.5 - 0 - 0.5) / 107 = 0.0748

8) Calculate the p Value That Corresponds With That Z Score

In Excel, the formula is for calculating the p value when it is in the right outer tail is:

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 59

Nonparametric Testing in Excel

The Excel Statistical Master

p Value = 1 - NORMDIST ( X, X, X , Cumulative ? )

Which in this case is:

p Value = 1 - NORMDIST ( Test Statistic W, W, W , Cumulative ? )

p Value = 1 - NORMDIST ( 8.5, 0, 107, TRUE) = 1 - 0.53 = 0.47

Rule:
Reject the Null Hypothesis if the p Value Test Statistic W is less than , which is the Level of
Significance.
Since we require 95% certainty in our test, = 0.05.
We do not reject the Null Hypothesis in this case because the p Value of Test Statistic W is significantly greater
than .

Here is a graphical representation of the answer:

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 60

Nonparametric Testing in Excel

The Excel Statistical Master

Statistical Mistakes You Dont Want To Make

1) Assuming that correlation equals causation
This is, of course, not true. However, if you find a correlation, you should look hard for links between the two.
The correlation may be pure chance, but then again, it may not be. A correlation is a reason to look for
underlying causes behind the behavior. Correlation is often a symptom of a larger issue.

2) Not graphing and eyeballing the data prior to performing

regression analysis
Always graph the data before you do regression analysis. Youll know immediately whether youre dealing with
linear regression, non-linear regression, or completely unrelated data that cant be regressed.

3) Not doing correlation analysis on all variables prior to

performing regression
Youll save yourself a lot of time if you can remove any input variables that have a low correlation with the
dependent
(output > Y) variable or that have a high correlation with another input variable (this error is called
multicollinearity). In the 2nd case, you would want to remove the input variable from the highly correlated pair of
input variables that has the lowest correlation with the output variable.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 61

Nonparametric Testing in Excel

The Excel Statistical Master

4) Adding a large number of new input variables into a regression

analysis all at once
You always want to add new input variables one at a time and run a separate regression each time a new input
variable is added. The changes you observe to the output of the regression will tell you whether the new input
variable adds to the predictive power of the regression equation. Adjusted r squared only increases when a new
variable adds greater predictive power to the regression equation.

5) Applying input variables to a regression equation that are

outside of the value of the original input variables that were used
to create the regression equation
Here is an example to illustrate why this might produce totally invalid results. Suppose that you created a
regression equation that predicted a childs weight based upon the childs age, and then you provided an adult
age as an input. This regression equation would predict a completely incorrect weight for the adult, because
adult data was not used to construct the original regression equation.

6) Not examining the residuals in regression

You should always at least eyeball the residuals. If the residuals show a pattern, your regression equation is not
explaining all of the behavior of the data.

7) Only evaluating r square in a regression equation

In the output of regression performed in Excel, there are actually four very important components of the output
that should be looked at. There is an article in my blog: http://blog.excelmasterseries.com/2010/03/how-toquickly-read-output-of-excels.html that covers this topic in a lot more detail than could be done in this bullet
point.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 62

Nonparametric Testing in Excel

The Excel Statistical Master

8) Not drawing a representative sample from a population

This is usually solved by taking a larger sample and using a random sampling technique such as nth-ing
(sampling every nth object in the population).

9) Drawing a conclusion without applying the proper statistical

analysis
This occurs quite often when people simply eyeball the results instead of performing a hypothesis test to
determine if the observed change has at least an 80% chance (or whatever level of certainty you desire) of not
being pure chance.

10) Drawing a conclusion before a statistically significant result

has been reached
This is often caused by choosing a statistical test requiring a lot of samples but depending on a low sample
rate. A common occurrence of this would be performing multivariate testing on a web site that does not have
sufficient traffic. Such a test is likely to be concluded prematurely. A better solution might be to perform a
number of successive A/B split-tests in place of multivariate analysis. You get a lot more testing done a lot
faster, and correctly.

11) Analyzing non-normal data with the normal distribution

Data should always be eyeballed in a Histogram and then and analyzed for normality before using the normal
distribution-based testing techniques. If the data is not normally distributed, you must use data fitting techniques
to determine which statistical distribution most closely fits the data.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 63

Nonparametric Testing in Excel

The Excel Statistical Master

12) Not removing outliers prior to statistical analysis

A couple of outliers can skew results badly. Once again, eyeball the data and determine what belongs and what
doesnt. Another solution is to test the data using nonparametric analysis. Many nonparametric tests are not
significantly affected by a few outliers.

13) Not controlling or taking into account other variables besides

the one(s) being testing when using the t test, ANOVA, or
hypothesis tests.
Other variables that not part of the test need to be held as constant as possible during the above tests or your
answer might be invalid without you realizing it.

14) Using the wrong t test

The t-test to be applied depends upon factors such as whether or samples have the same size and variance. It
is important to pick the right t-test before starting. Read the t-Test chapters in this e-manual.

15) Attempting to apply the wrong type of hypothesis test

There are 4 ways that the data must be classified before the correct hypothesis test can be selected. Another
article in this blog discusses this. Also, Chapters 8 and 9 of the Excel Statistical Master:
http://excelmasterseries.com/ClickBank/Download_ESM_1234_Site_3.php provide clear, detailed instructions
on how to analyze your data prior to hypothesis test selection. You probably wouldnt get far into a hypothesis
test if you have incorrectly classified the data and selected the wrong hypothesis test.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 64

Nonparametric Testing in Excel

The Excel Statistical Master

16) Not using Excel

This point may sound a little self-serving, but knowing how to do this stuff in Excel is a real time-saver,
particularly if you are in marketing, and especially if youre an Internet marketer. Youll never need to pick up
another thick confusing statistics text book or figure how to work those confusing statistics tables ever again.
Ive actually thrown out all of my statistics text books (well, not quite, I sold them on eBay).

17) Always requiring 95% certainty

This could really slow you down. For example, if Im A/B split-testing keywords or ads in an AdWords campaign,
I will typically pick a winner when my split-tester tells me that it is 80% sure that one result is better than the
other. Achieving 95% certainty would often take too long.

18) Thinking it is impossible to get a statistically significant

sample if your target market is large
The sample size you need from a large population is probably quite a bit smaller than you think. Nationwide
surveys are normally within a percentage point or two from real answer after only several thousand interviews
have been conducted. That of course depends hugely on obtaining a representative sample to interview.

19) Not taking steps to ensure that your sample is normally

distributed when analyzing with the normal distribution
One way to ensure that you have a normally distributed sample for analysis is to take a large number of
samples (at 30) with each sample consisting of several random and simultaneously-chosen data points and
then take the mean from each sample. Make that mean the sample. Your samples will now be Normally
distributed. You can take as few as 2 data points per sample, but the more data points per sample, the fewer
data points it will take for your samples (each sample is the mean of the data points collected for one sample) to
appear to be Normally distributed. If you are taking only 3 data points per sample, you may have to collect over
samples (that would be a total of 300 data points) for your samples to appear to be Normally distributed.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 65

Nonparametric Testing in Excel

The Excel Statistical Master

A statistical theory called the Central Limit Theory states that the means of samples (at least 30 samples and
each sample having at least 2 data points that are averaged to get a mean, which will be the value of the
sample) will be Normally distributed, no matter how the underlying population is distributed. You can then
perform statistical analysis on that group using the normal distribution-based techniques.

20) Using covariance analysis instead of correlation analysis

The output of covariance analysis is dependent upon the scale used to measure the data. Different scales of
measurement can produce completely different results on the same data if covariance analysis is used.
Correlation analysis is completely independent of the scale used to measure the data. Different scales of
measurement will produce the same results on a data set using correlation analysis, unlike covariance analysis.

21) Using a one-tailed test instead of a two-tailed test when

accuracy is needed
If accuracy it what you are seeking, it might be better to use the two-tailed when performing, for example, a
hypothesis test. The two-tailed test is more stringent than the one-tailed test because the outer regions (I refer
to them the regions of uncertainty) are half the size in a two-tailed test than in a one-tailed test. The two-tailed
test tells you merely that the means are different. The one-tailed test tells you that the means are different in
one specific direction.

22) Not using nonparametric tests when analyzing small samples

of unknown distribution
The t Distribution should only be used in small sample analysis if the population from which the samples were
drawn was Normally distributed. Nonparametric tests are valid when the population distribution is not known, or
is known not to be Normally distributed. Using the t distribution in either of these cases for small sample
analysis is invalid.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 66

Nonparametric Testing in Excel

The Excel Statistical Master

Correctable Reasons Why Your

Data Is Not Normally Distributed
In the ideal world, all of your data samples are normally distributed. In this case you can usually apply the wellknown parametric statistical tests such as ANOVA, the t Test, and regression to the sampled data.

What can you do if your data does not appear to be normally

distributed?
You can either:
- Apply nonparametric tests to the data. Nonparametric tests do not rely on the underlying data to have any
specific distribution
- Evaluate whether your non-normal data was really normally- distributed before it was affected by one of the
seven correctable causes listed below:

The Biggest 7 Correctable Causes of Non-Normality in Data

Samples
1) Outliers Too many outliers can easily skew normally-distributed data. If you can identify and remove
outliers that are caused by error in measurement or data entry, you might be able to obtain normally-distributed
data from your skewed data set. Outliers should only be removed if a specific cause of their extreme value is
identified. The nature of the normal distribution is that some outliers will occur. Outliers should be examined
carefully if there are more than would be expected.

2) Data has been affected by more than one process It is very important to
understand all of the factors that can affect data sample measurement. Variations to process inputs might skew
what would otherwise be normally-distributed output data. Input variation might be caused by factors such as
shift changes, operator changes, or frequent changes in the underlying process. A common symptom that the
output is being affected by more than one process is the occurrence of more than one mode (most commonly

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 67

Nonparametric Testing in Excel

The Excel Statistical Master

occurring value) in the output. In such a situation, you must isolate each input variation that is affecting the
output. You must then isolate the overall effect which that variation had on the output. Finally, you must remove
that input variations effect from output measurement. You may find that you now have normally-distributed
data.

3) Not enough data A normal process will not look normal at all until enough samples have been
collected. It is often stated that 30 is the where a large sample starts. If you have collected 50 or fewer
samples and do not have a normally-distributed sample, collect at least 100 samples before re-evaluating the
normality of the population from which the samples are drawn.

4) Measuring devices that have poor resolution Devices with poor resolution
may round off incorrectly or make continuous data appear discrete. You can, of course, use a more accurate
measuring device. A simpler solution is to use a much larger sample size to smooth out sharp edges.

5) A different distribution describes the data Some forms of data inherently

follow different distributions. For example, radioactive decay is described by the exponential distribution. The
Poisson distribution describes events event that tend to occur at predictable intervals over time, such as calls
over a switchboard, number of defects, or demand for services. The lengths of time between occurrences of
Poisson-distributed processes are described by the exponential distribution. The uniform distribution describes
events that have an equal probability of occurring. Application of the Gamma distribution often based on
intervals between Poisson-distributed events, such as queuing models and the flow of items through a
manufacturing process. The Beta distribution is often used for modeling planning and control systems such are
PERT and CPM. The Weibull distribution is used extensively to model time between failure of manufactured
items, finance, and climatology. It is important to become familiar with the applications of other distributions. If
you know that the data is described by a different distribution than the normal distribution, you will have to apply
the techniques of that distribution or use nonparametric analysis techniques.

6) Data approaching zero or a natural limit If the data has a large number of
value than are near zero or a natural limit, the data may appear to be skewed. In this case, you may have to
adjust all data by adding a specific value to all data being analyzed. You need to make sure that all data being
analyzed is raised to the same extent.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 68

Nonparametric Testing in Excel

The Excel Statistical Master

7) Only a subset of process output is being analyzed If you are sampling

only a specific subset of the total output of a process, you are likely not collecting a representative sample from
the process and therefore will not have normally distributed samples. For example, if you are evaluating
manufacturing samples that occur between 4 and 6AM and not an entire shift, you might not obtain the
normally-distributed sample that a whole shift would provide. It is important to ensure that your sample is
representative of an entire process.
If you are unable to obtain a normally-distributed data sample, you can usually apply non-parametric tests to the
data.

Copyright 2011

http://excelmasterseries.com/New_Manuals.php

Page 69

Meet the Author

Mark Harmon is a master number cruncher. Creating overloaded Excel spreadsheets
loaded with complicated statistical analysis is his idea of a good time. His profession as
an Internet marketing manager provides him with the opportunity and the need to
perform plenty of meaningful statistical analysis at his job.
Mark Harmon is also a natural teacher. As an adjunct professor, he spent five years
teaching more than thirty semester-long courses in marketing and finance at the AngloAmerican College in Prague and the International University in Vienna, Austria. During
that five-year time period, he also worked as an independent marketing consultant in the
Czech Republic and performed long-term assignments for more than one hundred clients.
His years of teaching and consulting have honed his ability to present difficult subject
matter in an easy-to-understand way.
Harmon received a degree in electrical engineering from Villanova University and MBA
in marketing from the Wharton School.